Datasets:
Mardiyyah
/
TAPT_CeLLaTe2.0

Dataset card Data Studio Files
xet
 Community
1
PMCID
string
Title
string
FullText
string
PMC11075828
Direct and indirect effects of CYTOR lncRNA regulate HIV gene expression
The implementation of antiretroviral therapy (ART) has effectively restricted the transmission of Human Immunodeficiency Virus (HIV) and improved overall clinical outcomes. However, a complete cure for HIV remains out of reach, as the virus persists in a stable pool of infected cell reservoir that is resistant to therapy and thus a main barrier towards complete elimination of viral infection. While the mechanisms by which host proteins govern viral gene expression and latency are well-studied, the emerging regulatory functions of non-coding RNAs (ncRNA) in the context of T cell activation, HIV gene expression and viral latency have not yet been thoroughly explored. Here, we report the identification of the Cytoskeleton Regulator (CYTOR) long non-coding RNA (lncRNA) as an activator of HIV gene expression that is upregulated following T cell stimulation. Functional studies show that CYTOR suppresses viral latency by directly binding to the HIV promoter and associating with the cellular positive transcription elongation factor (P-TEFb) to activate viral gene expression. CYTOR also plays a global role in regulating cellular gene expression, including those involved in controlling actin dynamics. Depletion of CYTOR expression reduces cytoplasmic actin polymerization in response to T cell activation. In addition, treating HIV-infected cells with pharmacological inhibitors of actin polymerization reduces HIV gene expression. We conclude that both direct and indirect effects of CYTOR regulate HIV gene expression.The introduction of antiretroviral therapy (ART) has successfully limited the spread of Human Immunodeficiency Virus (HIV) and improved patient clinical outcomes. However, a complete cure for HIV infection remains out of reach, as the transcriptionally silent but replication-competent provirus that is integrated into the host genome persists in long-lived cellular reservoirs, which are comprised of memory-resting CD4 T cells, as well as cells of myeloid lineages . These reservoirs are highly stable and are resistant to both ART and the effects of the host immune surveillance, thus posing a significant obstacle to eradicating the HIV reservoirs. Consequently, in most people living with HIV, interrupting ART leads to rapid viral load rebound, usually within weeks after treatment cessation [3–6]. As T cell stimulation triggers activation of proviral transcription, one strategy that has been proposed to eliminate the HIV reservoirs is a “Shock-and-Kill” approach, which utilizes latency-reversing agents (LRAs) to first activate dormant HIV-infected T cells and facilitate cell death by viral cytopathic effects or immune-mediated killing. This step is done in the presence of ART, so there are no further rounds of HIV replication. [7–9]. Alternatively, a “Block and Lock” approach frees infected individuals from ART by silencing HIV transcription and inducing a deep state of latency. Nevertheless, despite promising therapeutic options, these strategies and others have regretfully failed to achieve significant clinical efficacy. These failures highlight our lack of knowledge of the molecular mechanisms that govern latency establishment and reversal and the need for alternative therapies capable of eliminating the viral reservoirs [10–15]. Epigenetic constraints that suppress proviral gene transcription are essential for establishing HIV latency . Low levels of basal and elongating transcription factors in the infected T cell, together with the absence of the viral trans-activator of transcription (Tat), ensure that proviral transcription remains below detectable thresholds . Within the infected T cells, gene transcription of the integrated provirus and the host genome are synchronized . Both display key steps of gene transcription, which include initiation, promoter arrest, and elongation. HIV-Tat orchestrates transcription elongation of the provirus by binding to TAR RNA and recruiting P-TEFb and Super Elongation Complex (SEC) to the viral promoter [22–26]. However, despite extensive efforts to elucidate the mechanisms of metazoan transcriptional control and its role in the regulation of HIV gene transcription, the knowledge of how HIV latency is established is still incomplete . Long non-coding RNAs (lncRNAs) are transcripts with longer than 200 nucleotides that lack protein-coding capacity. To date, over 200,000 cell type-specific lncRNAs have been identified and display critical regulatory functions of many processes within cells [28–31]. However, the functions of most of these transcripts remain poorly understood. In the context of HIV, roles for several cellular lncRNAs have been documented [32–40]. Moreover, significant gaps still remain in our knowledge about the mechanistic roles that lncRNAs play in CD4 T cell activation and HIV latency. In this study, we monitored changes in gene expression in an HIV-infected Jurkat-derived T cell line (J-Lat 6.3) upon response to T cell stimulation with Phorbol 12-myristate 13-acetate—PMA/Ionomycin (P/I). We documented RNA expression in stimulated J-Lat 6.3 cells that carry either active or cells latent HIV, and among identified ncRNA, Cytoskeleton Regulator RNA (CYTOR) exhibited a profound change in expression in cells that expressed active HIV following T cell stimulation. CYTOR directly binds the HIV promoter and activates viral gene transcription and latency reversal by recruiting P-TEFb to the viral promoter. CYTOR also exerts its effects indirectly by controlling global gene expression along with actin dynamic pathways, thereby affecting T cell activation and HIV infection. In search for novel host regulators of HIV gene expression and viral latency, we employed RNA-Sequencing analysis to monitor changes in the transcriptome of Jurkat-derived J-Lat 6.3. These cells carry a transcriptionally repressed intact copy of HIV-1 proviral DNA with a GFP reporter under the control of the HIV promoter that is inserted in the nef gene. As expected, in response to T-cell stimulation, HIV gene expression in J-Lat 6.3 cells was enhanced, as exhibited by elevated expression levels of GFP . J-Lat 6.3 cells were stimulated with the PKC activator PMA and Ionomycin (P/I), which potently activate CD4+ T lymphocytes. Stimulated J-Lat 6.3 cells were then sorted by FACS based on their HIV-GFP expression and divided into two distinct populations: Stimulated cells that expressed HIV genes (GFP+; P1) and stimulated cells that carried latent provirus (GFP-) (Fig 1A). RNA from both cell groups was isolated, and libraries were generated for transcriptome analysis by next generation sequencing (RNA-Seq). As expected, a pronounced change in cellular gene expression, including mRNAs, miRNAs, snoRNAs, snRNAs and lncRNAs was observed in stimulated cells that expressed active HIV or latent HIV (Fig 1B). Subsequent RNA-Seq from HIV expressing cells that carry active (GFP+) or latent (GFP-), indicated different transcriptional profiles in cells where HIV is activated versus cells where the virus remained latent. A total of 3490 annotated transcripts were identified whose expression was changed in cells that carried transcriptionally active HIV relative to latent HIV. Of these, 2400 transcripts corresponded to protein-coding genes, while 843 were lncRNAs. Upon T-cell stimulation, 468 lncRNA transcripts were upregulated (enriched in cells expressing active HIV; GFP+), and 375 were downregulated (enriched in cells carrying latent HIV provirus; GFP-) (Fig 1C). We further assessed the relative expression levels of the highly ranked lncRNAs in CD4+ T primary T cells by RT-qPCR. Analysis was performed under resting or stimulating conditions of primary CD4+ T cells. For most tested lncRNAs, a shift in expression levels was confirmed when comparing primary CD4+ T cells under resting or stimulated conditions, where HIV was latent or active, respectively (Fig 1D). Notably, ncRNAs with reported effects on HIV replication and latency, including HEAL and NRON were identified via our RNA-Seq analysis, demonstrating the potential of this screening approach (Fig 1D). Also indicated are ncRNAs that are currently under investigation like IL21R-AS; PCBP3-AS; APOBEC3B-AS; IER3-AS (Fig 1D). Similarly, mRNAs genes, including HSP90 , ESR-1 , and IFI16 , that were previously reported to control HIV replication were also identified by our screening approach (S1 Table; GSE254771). Finally, we confirmed surface expression of CD25 and CD69 activation markers in stimulated primary CD4+ T cells that were infected with HIVGKO and carry active or latent HIV (S1 Fig). (A) FACS histogram analysis of PMA/Ionomycin (P/I)-stimulated J-Lat 6.3 cells. GFP(+) cells carrying active HIV (P1 region) were sorted from GFP (-) cells carrying latent HIV (GFP-). Cells were sorted and sent to RNA-Seq (n = 4). (B) Heatmap of differential transcript expression pattern (FC≥ a 2-fold change and above between cells carrying active versus latent HIV with an adjusted p value of ≤0.05. (C) Pie chart corresponding to the numbers of differentially expressed mRNAs and lncRNAs, up and downregulated in cells where HIV was reactivated. (D) RNA levels of selected ncRNA in primary CD4+ T cells. Analysis of expression levels of selected ncRNA based on the RNA-Seq analysis in primary CD4+ T cells that were either under resting conditions (-) or stimulated with P/I (+). RNA levels were analyzed by RT-qPCR. Data were normalized to gapdh levels. Data are from 2 healthy donors. Among the lncRNAs that were strongly induced upon T cell stimulation, we focused our work on Cytoskeleton Regulator RNA (CYTOR)—also known as lincRNA00152. Elevated RNA levels of CYTOR upon T cell stimulation were confirmed in J-Lat 6.3 and in primary CD4+ T cells (Fig 1D). To monitor the effects of HIV infection on CYTOR RNA levels, Jurkat T cells were infected with HIV and levels of CYTOR RNA were determined by RT qPCR relative to non-infected cells. Our analysis confirmed that CYTOR RNA levels were not affected by HIV infection (S2 Fig). CYTOR is an intergenic 828 nucleotide lncRNA located on chromosome 2p11.2. It is highly conserved in primates and rodents but less so in lower organisms. CYTOR is mainly present in the cytoplasm. However, previous reports show that it is also localized to the nucleus. Within the nucleus, CYTOR functions as an oncogene and is upregulated in multiple human malignancies . CYTOR also acts as an “endogenous sponge” for several micro-RNAs by binding to them, inhibiting their activity, and promoting malignancy. Interestingly, CYTOR reportedly regulates cellular actin dynamics and cytoskeletal reorganization in fibroblasts by controlling the expression of genes of the actin polymerization machinery . However, the functional importance of CYTOR in CD4 T cells and in the context of HIV infection has not been studied. We next conducted gain and loss-of-function studies in J-Lat 6.3 T cells to determine the role of CYTOR in regulating HIV gene expression. To achieve CYTOR over-expression, cells were transduced with a lentivirus that drives the expression of CYTOR—exons 1, 4, and 5, the most abundantly expressed form in humans . Following antibiotic selection, resistant J-Lat 6.3 T stable cells were subjected to RT-qPCR and exhibited a significant increase in CYTOR RNA levels relative to control cells (Fig 2A; blue bar versus grey bar). Reducing CYTOR expression (knockdown; KD) was also achieved by transducing J-Lat 6.3 T cells with a lentivirus encoding a CYTOR-targeting small-hairpin RNA (shRNA), resulting in a significant decrease of CYTOR RNA levels relative to control cells, expressing a scrambled shRNA as measured by RT-qPCR (Fig 2A; red bar versus grey bar). Parallel FACS-based analysis of GFP expression in HIV-infected J-Lat 6.3 cells, as a measure of viral gene transcription, revealed that in the absence of T-cell stimulation, no effects on HIV gene expression were observed upon modulation of CYTOR expression. However, following T cell stimulation with P/I, CYTOR over-expression led to a relatively small 2-3-fold increase of HIV GFP expression over control cells (Fig 2B; compare blue to grey bars). In contrast, depletion of CYTOR led to a 5-fold decrease in HIV gene expression over control cells (Fig 2B; compare red to grey bars). HIV GFP expression in control cells expressing scrambled shRNA was unaffected (Fig 2B; grey bar). (A). Modulation of CYTOR RNA levels in J-Lat 6.3 cells. RT-qPCR analysis measuring CYTOR RNA levels in J-Lat 6.3 T cells, where CYTOR expression is knockdown (KD; red bar) or overexpressed (light blue bar). RNA levels were normalized to gapdh and presented relative to control cells expressing scrambled shRNA (grey bar). Statistical significance is based on calculating ±SD of data points from four independent experiments using two-way ANOVA. ***p≤0.05. (B) Effects of CYTOR on HIV gene expression. FACS quantification analysis of the percentage of cells that express HIV-GFP in P/I stimulated J-Lat 6.3 cells expressing control scramble shRNA (grey bar), or in which CYTOR was overexpressed (blue bar) or knockdown (KD; red bar). Statistical significance is based on calculating mean ± SD from three independent experiments using two-way ANOVA. ***p≤0.05. (C) Kinetics of latency establishment in the context of CYTOR expression. 2D10 latency model Jurkat T cells that carry a mini-Tat-Rev GFP under the regulation of the HIV LTR promoter and express either scrambled shRNA (grey), CYTOR KD (red), or cells over-expressing CYTOR (blue) were reactivated and sorted to obtain a pure cell population that expresses GFP. GFP expression was then followed over time as a measurement of entry into latency. Statistical significance is based on calculating mean ± SD from three independent experiments using two-way ANOVA. ***p≤0.05 and ns: not significant. We also followed the establishment of HIV latency post-T cell activation, documenting HIV-GFP expression in control 2D10 cells that expressed scramble shRNA or in cells where CYTOR expression levels were modulated. Like J-Lat 6.3 cells, 2D10 cells serve as a Jurkat-based latency cell model that carries a minimal Tat-Rev cassette in the context of a GFP reporter under the regulation of the HIV promoter. Upon T-cell stimulation of control 2D10 Jurkat cells, we confirmed that the expression of HIV-GFP was significantly induced. Control and CYTOR-modulated stimulated 2D10 Jurkat cells were sorted based on their HIV-GFP expression, obtaining a relatively pure cell population with 100% HIV-GFP expression levels. We then monitored latency establishment by following GFP expression in the context of control or CYTOR-modulated cells (Fig 2C). Our FACS analysis revealed that lower CYTOR levels were associated with a rapid establishment of latency relative to control cells (Fig 2C; grey versus red lines). Conversely, CYTOR-over-expression enhanced latency reversal, as determined by the elevated levels of HIV-GFP expression that remained relatively high for an extended period following T cell stimulation (Fig 2C; blue line). These results suggest that CYTOR expression activates HIV gene expression, significantly reversing latency in 2D10 cells. Next, we shifted our analysis to CD4+ primary T cells isolated from healthy donors and the natural target cells for HIV infection. Depletion of CYTOR in primary human CD4+ T cells was achieved by first stimulating purified cells with anti CD3/CD28 beads and IL2 and then transducing them with a lentivirus encoding a CYTOR-specific shRNA. Lentivirus driving the expression of scrambled shRNA was used as a control (Fig 3A; n = 3). RT qPCR confirmed a significant decrease in CYTOR expression RNA levels relative to control cells that expressed the scramble shRNA (Fig 3B). The next day, CYTOR-depleted CD4+ primary stimulated cells (KD) or control cells were transduced with HIVGKO, which codes for a codon-optimized GFP reporter under the control of the HIV-1 promoter and in the context of expression of all viral proteins, and a mKO2 reporter under the control of the constitutive promoter EF10 α . HIVGKO transduction can be analyzed by FACS two days later, monitoring parallel transduction efficiency (mKO2+), as well as HIV gene expression (GFP+). Upon transduction of stimulated primary CD4+ T cells, KD of CYTOR led to decreased HIV gene expression, as monitored by reduced levels of HIV-GFP-expressing cells. Conversely, the proportion of cells that expressed EF10 α -mKO2 was slightly elevated in control or CYTOR KD-expressing cells, implying that transduction efficiencies were not affected due to CYTOR depletion but rather specifically drove HIV into a latency state (Fig 3C and 3D). (A). Experimental workflow overview for isolating primary CD4+ T cells. See methods for a detailed description. The figure was generated by Biorender. (B). Depletion of CYTOR in stimulated primary CD4+ T cells using lentivirus encoding CYTOR shRNA. Data were measured by RT-qPCR, normalized to GAPDH, and presented relative to cells expressing scrambled shRNA—set to 1. Statistical significance is based on calculating mean ± SD from three independent experiments using two-way ANOVA. ***p≤0.05; n = 3. (C). FACS analysis presenting effects of CYTOR knockdown (KD) on HIVGKO infection in primary CD4+ T cells. Cells were stimulated and then transduced with HIVGKO before being analyzed by FACS for mKO2 and GFP expression. (D). Quantification of quadrate percentage from three independent experiments of FACS analysis for HIVGKO transduction in CD4+ primary T cells, where CYTOR is KD. Statistical significance is based on the calculation of mean ±SD from three independent experiments (n = 3) using Two-way ANOVA. ***p≤0.05. Towards direct effects of CYTOR on HIV gene transcription, this would require its localization within the cell nucleus. We therefore monitored the subcellular distribution of CYTOR lncRNA between the nucleus and cytoplasm in resting or activated conditions of primary CD4+ T cells by cell fractionation and subsequent RT-qPCR analysis (Fig 4A). Levels of CYTOR RNA were compared to those of the abundant 7SK lncRNA, which is known to interact with inactive P-TEFb. RNA levels were normalized to the 7SL RNA, which does not bind to the HIV promotor and is commonly used as a specificity control for these experiments . Our analysis showed that CYTOR is localized to the cytoplasm and the nucleus. Notably, upon T cell stimulation, levels of nuclear CYTOR were elevated relative to the levels of nuclear 7SK, which were decreased (Fig 4A). To further extend our understanding of the mechanism by which CYTOR activates HIV gene expression, we tested whether CYTOR binds to the HIV promoter, thereby regulating HIV gene transcription. We monitored CYTOR occupancy on the HIV promoter by employing Chromatin Isolation by RNA Purification (ChIRP) analysis in J-Lat 6.3 cells. In vitro-transcribed biotinylated CYTOR RNA was synthesized, purified, and then incubated with ChIP material isolated from unstimulated or P/I stimulated HIV J-Lat 6.3 cells. RNA-protein complexes were then specifically pulled down with streptavidin beads, and pulled down levels of CYTOR on the HIV promoter were monitored by RT-qPCR with specific primers that target the viral promoter (Fig 4B). Our analysis showed that CYTOR binds to the HIV promoter even in unstimulated conditions. Significantly, CYTOR occupancy on the viral promoter was further elevated following T-cell stimulation (Fig 4B). CYTOR occupancy on gene promoters was also demonstrated for cellular genes that are known to be regulated by P-TEFb, such as NF-κB, IL21Ra, myc. Of note, the binding of CYTOR to HIV downstream reverse transcriptase sequences was not observed, suggesting that the specificity of CYTOR within the HIV genome lies within the HIV promoter (S3 Fig). (A). CYTOR is localized to the nucleus and its levels are elevated upon T-cell stimulation. Resting or stimulated primary CD4+ T cells were subjected to cell fractionation, separating the samples into a nuclear fraction (grey bar) or cytoplasmic fraction (light blue bar). Samples were then subjected to RT-qPCR and monitored for CYTOR or 7SK ncRNA levels. Data were normalized to 7SL RNA in each of the cellular fractions and conditions. Data are presented relative to cytoplasmic fraction in each condition—set to 1. (B). ChIRP-qPCR analysis for CYTOR binding to the HIV promoter. CYTOR-specific (black bar) or control lacZ (red bar) antisense biotinylated probes were incubated with lysates isolated from unstimulated or P/I stimulated J-Lat 6.3 cells. Biotinylated RNA was pulled down with streptavidin beads, and following washing, associated DNA was eluted and analyzed by qPCR with primers for the HIV promoter. Statistical significance was calculated between the two probes and between unstimulated and stimulated states. IgG served as a non-specific antibody for IP control (grey bar). The analysis is based on calculating mean ± SD from three independent experiments using two-way ANOVA. ***p≤0.05. **0.05≤p≤ 0.1; n.s—not significant. (C, D) CYTOR affects the phosphorylation state of RNAPII CTD and histone landscape. ChIP qPCR analysis in control or CYTOR KD J-Lat 6.3 cells. ChIP material from cells was immune-precipitated (IP) with antibodies targeting RNAPII-Ser2P or RNAPII Ser5P (C); or for H3K4me3 and H3K27Ac histone activation marks (D). IP fraction was analyzed for enrichment of the indicated modifications on the HIV promoter by qPCR with specific primers. Non-specific IgG served as a control (grey bar). Percentage of input are means ±SD; n = 3; *** p≤0.05 calculated between scrambled and KD cells for each antibody. n.s—not significant. To further obtain insights into the mechanisms of action of CYTOR, we performed Chromatin immunoprecipitation (ChIP) qPCR from J-Lat 6.3 cells, where CYTOR expression was manipulated. We monitored the levels of phosphorylated C-terminal domain (CTD) of RNA Polymerase II (RNAPII) at Ser2 (Ser2P) or Ser5 (Ser5P) residues on control or CYTOR KD expressing cells, using specific antibodies that target the CTD phosphorylation states of RNAPII (Fig 4C). CDK9/P-TEFb phosphorylates Ser2 and marks RNAPII pause-release and elongation of transcription [50–52]. CDK7/TFIIH phosphorylates Ser5P on the CTD and catalyzes transcription initiation and promoter clearance. Our analysis demonstrated that in the context of CYTOR depletion, levels of Ser2P on the HIV promoter were decreased without affecting those of Ser5P, implying the involvement of P-TEFb in CYTOR-mediated HIV gene activation (Fig 4C). In addition, ChIP-qPCR was employed using antibodies that target the histone activation markers, H3K27Ac or H3K4me3. Our analysis confirmed that CYTOR mediates its activation properties by modifying the histone landscape around the HIV. Upon CYTOR depletion, levels of these histone activation markers were reduced (Fig 4D). These results further imply that CYTOR activates HIV gene expression via P-TEFb, affecting transcription elongation. To expand the above results on the mechanism by which CYTOR enhances HIV gene expression, we employed RNA-precipitation (RNA-IP; RIP) followed by RT-qPCR in J-Lat 6.3 cells under resting or stimulated conditions (Fig 5). As our above results indicate that CYTOR promotes the Ser2 phosphorylation on the CTD, which is mediated by CDK9, we monitored CYTOR association with P-TEFb. Lysates isolated from nuclei from resting or stimulated J-Lat 6.3 cells were incubated with antibodies that target CDK9 or CYCLIN T1, and samples were IP followed by RT-qPCR to detect CYTOR RNA levels by using specific primers. We show that upon T cell stimulation, the levels of CYCLIN T1 and CDK9 that were associated with CYTOR RNA increased. As expected, levels of 7SK that are associated with P-TEFb were reduced upon T cell stimulation. We also followed the association of P-TEFb with 7SL, which served as control. As expected, P-TEFb was not associated with 7SL in each of the tested conditions. These results indicate that CYTOR associates with P-TEFb in cells (Fig 5A). (A) RIP analysis demonstrates the association of CYTOR with P-TEFb. Isolated ChIP material from resting or stimulated J-Lat 6.3 CD4+ T cells was subjected to immune precipitation with antibodies targeting CDK9 or CYCLIN T1 of P-TEFb, followed by RT-qPCR with primers for the relevant lncRNA (7SK or CYTOR). Non-specific IgG served as a control for the IP step. 7SL ncRNA served as a control for an RNA that does not associate with P-TEFb and, therefore, not precipitated with CDK9 or CYCLIN T1 antibodies. Statistical significance is based on the calculation of mean ±SD from three independent experiments using two-way ANOVA. ***p≤0.05. ** 0.05≤p≤0.1; ns: not significant. (B) CYTOR associates with P-TEFb in cells. RNA pull-down followed by western blotting where lysates from J-Lat 6.3 cells were incubated with an in-vitro transcribed biotinylated CYTOR anti-sense probe and reactions were pulled down with streptavidin beads. Eluted RNP complexes were subjected to western blotting with indicated antibodies. Non-specific IgG served as a control for non-relevant IgG. Scramble RNA served as RNA that does not associate with P-TEFb. 7SK probe confirmed association with P-TEFb. Input is 5% of the total cell lysate . Next, we performed RNA pull-down experiments combined with western blotting to detect P-TEFb subunits (CYCLIN T1/CDK9) that are associated with CYTOR. Lysates from J-Lat 6.3 cells were incubated with an in-vitro transcribed biotinylated anti-sense CYTOR probe, and reactions were pulled down with streptavidin beads. Eluted RNP complexes were then subjected to western blotting with antibodies that target CYCLIN T1 or CDK9, demonstrating the association of CYTOR lncRNA with P-TEFb within cells. Non-specific IgG was used as a specificity control for the IP step, while a non-specific scramble RNA probe served as a control for RNA-protein association. In addition, a 7SK RNA probe confirmed the association with P-TEFb (Fig 5B). These results establish that CYTOR binds to the HIV promoter and suggest that its activation effects are mediated by association with P-TEFb. Since CYTOR has been previously recognized as a regulator of cytoskeleton-regulating genes in fibroblasts , we assessed if it could also affect HIV gene expression by indirect mechanisms through regulation of its downstream targets. For this, we performed RNA-Seq analysis in stimulated primary CD4+ T cells, where CYTOR expression was depleted or over-expressed (n = 3). CYTOR modulation of expression did not affect the activation state of cells as monitored by staining with T cell activation markers (S4 Fig). Analyzing changes in the cellular transcriptome of stimulated CD4+ T cells upon depletion of CYTOR revealed a modest change in the cell gene expression program (S2 Table). Additional gene GO analysis identified significant enrichment scores in various cellular pathways, including those of gene expression, signal transduction as well as actin dynamics and T-cell activation (Fig 6A and 6B). (A). Volcano plot of the expression pattern of genes from an RNA-Seq analysis upon CYTOR depletion following T cell stimulation. -Log10 P is shown on the y-axis, and Log2FC is on the x-axis. RNA was isolated from 3 biological replicates (n = 3). The fold of change cutoff is defined as FC ≥2. FDR of p≤0.05 was used as a cutoff for significance. (B). Gene Ontology analysis for enriched CYTOR gene targets. For enrichment analysis, the DAVID program was employed to identify enriched pathways and terms associated with the selected genes. (C). Experimental flow for microscopy-base analysis of cell morphology and formation of F-actin rich structures. The figure was generated by Biorender. (D). Representative confocal images of F-actin organization for control and CYTOR KD Jurkat cells after contact with anti-CD3/28 coated surfaces. Cells were stained with fluorescent phalloidin and DAPI to visualize F-actin and cell nuclei. Shown are merged images of both channels, scale bar = 10 μM). (E). Relative frequency of cells with circumferential F actin ring (AR) in control or CYTOR KD cells with proper cell spreading and circumferential F-actin relative to control cells (mean± SD, 100 cells per experiment/condition, n = 3). (F). Relative CYTOR RNA levels in CYTOR KD Jurkat cells relative to control cells of the cells analyzed in (E). (G). Representative images of the different morphotypes observed for Jurkat cells after anti-CD3/28 surface stimulation (analyzed as in D), (H). Quantifying the frequency of the morphotypes defined in (G) for control and CYTOR KD Jurkat cells (mean± SD, 100 cells per experiment/condition, n = 3). ** 0.05≤p≤0.1. (I). Inhibition of actin remodeling disrupts HIV gene expression upon T cell activation. 2D10 cells carrying an integrated HIV-GFP provirus where CYTOR expression was either depleted or over-expressed were treated with an actin polymerization inhibitor, Latrunculin B (LanB) for 1 hour, followed by T cell stimulation with anti-CD3/CD28 for an additional 3 hours. Cells were harvested 24 hours later, and the percentage of cells expressing HIV GFP was monitored by FACS. Data are presented as fold of activation relative to untreated cells and activated with the indicated T cell activator. Statistical significance is based on calculating mean ± SD from three independent experiments using two-way ANOVA. ***p≤0.05. ** 0.05≤p≤0.1. T cell activation elicits complex and highly dynamic signaling cascades that ultimately lead to the activation of transcription factors, including NF-κB and NF-AT, to increase the expression of T cell receptor target genes . The involvement of these transcription factors in HIV gene expression, at least in part, explains the beneficial effects of T cell activation on HIV gene expression . Since many of the downstream signaling events elicited by TCR engagement depend on the immediate polymerization of cortical actin, we tested if CYTOR affects the actin polymerization response to TCR engagement of Jurkat T cells. Scramble control or CYTOR KD Jurkat cells were placed on a cell stimulatory surface coated with anti-CD3/CD28 antibodies, fixed, and stained for F-actin. Control cells displayed the typical cell spreading and formation of circumferential F-actin-rich rings (actin ring; AR) (Fig 6C, 6D, 6E and 6F). Although CYTOR expression was only moderately reduced in KD cells (Fig 6F), fewer cells responded to TCR stimulation (approx. 40% less cells with AR in CYTOR KD than in control cells; Fig 6E). Detailed analysis of the different cell morphologies revealed that the CYTOR KD particularly resulted in a significant increase in the fraction of cells that were unable to both spread as well polymerize actin into an F-actin ring in response to T cell activation. In contrast, the frequency of cells that failed to spread despite efficient actin polymerization was unaffected (Fig 6H). Increasing CYTOR levels by overexpression did not further increase the frequency of cells that formed ARs, did not alter the morphology of F-actin structures formed in response to TCR activation, and did not result in the formation of ARs in the absence of TCR stimulation (S5 Fig). We conclude that CYTOR is an important regulator of TCR-induced actin polymerization in CD4+ T cells, but its normal endogenous expression levels are not limiting for this response. To assess whether TCR-induced actin remodeling affects HIV gene expression in our experimental system, we measured the induction of HIV gene expression by TCR engagement in 2D10 cells, a CD4+ T cell line that carries a latent GFP cassette under the control of the LTR promoter. Experiments were performed in the absence or presence of the actin polymerization inhibitor, Latrunculin B—an inhibitor that interferes with actin polymerization and is reversible upon washout (Fig 6I). Stimulation with anti-CD3/anti-CD28 resulted in a marked induction of GFP expression and, as observed before, silencing CYTOR expression reduced this induction. Notably, interfering with actin polymerization during the first 3 hours of TCR stimulation in control cells limited the induction of HIV gene expression to the levels observed upon CYTOR KD, and interference with actin dynamics in CYTOR KD did not result in an additional reduction of GFP expression. Finally, overexpressing CYTOR rendered the TCR-mediated induction of GFP expression insensitive to Latrunculin B (Fig 6I) . Together, these results reveal that the regulation of host cell actin dynamics is necessary but not sufficient for the regulation of gene expression of latent HIV. In search of regulators of HIV latency, we profiled changes in the expression of ncRNAs by employing RNA-Seq analysis in resting and stimulated HIV-infected J-Lat 6.3 T cells, comparing RNA expression levels in cells that carry active HIV (GFP+) or latent HIV (GFP-). Our analysis show that different transcriptional profiles exist in cells where HIV is activated versus cells where it remains latent. CYTOR lncRNA was identified as one of these RNAs, and its expression is elevated upon T cell stimulation, where HIV is active. These observations were further confirmed in primary CD4+ T cells (Fig 1). Functional analyses show that following T cell stimulation, over-expression of CYTOR activates HIV gene expression, while its depletion inhibits viral gene expression. Significantly, upon T cell stimulation, depletion of CYTOR promoted entry of HIV into a latent state, while its over-expression delayed entry into latency and enhanced latency reversal (Fig 2). Effects of CYTOR on HIV infection and latency establishment were also confirmed in stimulated primary CD4+ T cells (Fig 3). We are aware that the model of stimulated CD4+ primary cells does not recapitulate the actual state of the reservoir, which is mainly comprise of resting CD4+ T cells that do not support HIV infection. As this is a limitation of the current study, we are trying to adopt a recently developed gene editing approach to lncRNAs to deplete CYTOR in this unique cell population and monitor the effects of latency kinetics without altering its activation . Mechanistically, our observations show that CYTOR directly binds to the HIV promoter and enhances the phosphorylation of the Ser2 CTD of RNAPII through association with P-TEFb to activate viral gene expression (Figs 4 and 5). Changes in histone activation marks around the viral promoter in CYTOR-depleted cells also imply that CYTOR activates the proviral gene expression (Fig 4). In addition to the direct effects of CYTOR on HIV gene expression, we also demonstrate that CYTOR controls global gene expression. CYTOR is recruited to other gene promoters that are regulated by P-TEFb, like myc, NF-κB, and IL2Ra (S3 Fig). Among the identified enriched pathways that potentially are regulated by CYTOR are those that are involved in actin dynamics. Consistently, reduced levels of CYTOR expression are associated with reduced polymerization of cortical actin in response to TCR engagement (Fig 6). In turn, elevated levels of CYTOR do not further increase actin polymerization in response to T cell stimulation and cannot induce morphological responses of T cells in the absence of stimulation (S5 Fig). Thus, CYTOR is an important regulator of TCR-induced actin polymerization in T cells. However, its normal endogenous expression levels are sufficient for a proper response. To test a mechanistic link between actin remodeling, CYTOR levels, and HIV gene expression, we inhibited actin dynamics with specific inhibitors (Fig 6I). Effects of inhibition of actin polymerization phenocopied the effect of CYTOR depletion on HIV gene expression, suggesting that CYTOR may affect HIV gene expression by the regulation of genes that control cellular actin dynamics (Fig 6I). Accordingly, we propose a model where CYTOR exerts its effects on global gene expression and promotes HIV gene expression by both direct and indirect effects (Fig 7). CYTOR directly binds the HIV promoter and recruits the elongation transcription machinery to enhance RNAPII CTD phosphorylation and deposition of active histone markers around the HIV promoter, ultimately activating HIV gene expression. Indirectly, CYTOR controls gene targets that regulate actin dynamics in the nucleus and at the plasma membrane to optimize the response to T cell activation, presumably via the regulation of cellular gene expression. Following T cell activation, levels of CYTOR are elevated in the nucleus. CYTOR is recruited to the HIV promoter and binds to P-TEFb, leading to the activation of viral gene expression. Cellular genes regulated by CYTOR include actin remodeling genes that promote actin polymerization and the indirect activation of HIV gene expression. Like CYTOR, other lncRNAs have been reported to occupy the HIV promoter and modulate its activity at either transcriptional or posttranscriptional levels . Most act as scaffolds that associate with other transcriptional activators or repressors to control HIV gene expression [35–39,58–60]. In the case of CYTOR, its effects on gene expression occur by recruiting the transcription elongation machinery to activate gene expression, either from the viral promoter or other cellular promoters. It will be essential to identify other partners that are associated with CYTOR lncRNA and control HIV promoter activity. As we also aim to dissect the role of CYTOR in gene expression control, specifically for HIV gene regulation, it will be essential to define how events within the nucleus are regulated by CYTOR and translated to the control of downstream effector functions of stimulated T cells. Future studies will further identify the downstream targets of CYTOR that control actin dynamics upon T-cell activation. As additional pathways were identified by our RNA-seq analysis in CYTOR-depleted cells, we visualize that future work will identify novel downstream targets of CYTOR and elucidate their mechanisms of function in regulating HIV gene expression and latency. These may open new ways for developing novel therapeutic tools that will be integrated or substitute current strategies to successfully eliminate the HIV reservoir. Jurkat J-Lat 6.3 T cells are immortalized human T lymphocytes that serve as a model for studying HIV latency, as it harbors a transcriptionally silent integrated HIV provirus that encodes for a GFP reporter instead of Nef, which reactivated following T cell stimulation. 2D10 cells are also Jurkat-based T cells, carrying a mini—HIV cassette coding for Tat and rev and a 2dGFP reporter gene. Jurkat T cells were maintained in RMPI medium (GIBCO) containing with 10% fetal bovine serum (FBS), 2mg/ml L-glutamine, penicillin-streptomycin, and non-essential amino acids (Sigma, M7145). Cells were cultured at 37°C with 5% CO2. Human Embryonic Kidney HEK293T, this cell line was mainly used for the production of viral-like particles were maintained in DMEM complete medium (GIBCO). Cells were cultured at 37°C with 5% CO2. For the isolation of primary human CD4+ T cells, human Buffy Coats from anonymous healthy donors were obtained from the Soroka Medical Center Hospital Blood Bank. At day 0, PBMCs were isolated over a Ficoll gradient (Millipore). PBMCs were maintained at 2 x10 cells/ml overnight at 37°C. CD4+ T cells were isolated by negative selection with the RosetteSep Human CD4+ T Cell Enrichment Cocktail Stemcell Technologies), resulting in homogenous populations of CD4+ T cells with a purity of 90–95% as assured by flow cytometry. CD4+ T cells were cultured in complete RPMI media containing recombinant human IL2 at 20 U/ml (Roche) to a final concentration 10 cells/ml. Cells were then stimulated with anti-CD3/CD28 dynabeads (Invitrogen) and further cultured for 48 hour. The level of activation was monitored by FACS measuring staining with APC anti human CD25 (Biolegend #302609) and Pacific Blue anti-human CD69 (Biolegend #310919). At day two, stimulated cells were counted, centrifuged for 5 minutes at 1500 rpm, and resuspended in fresh RPMI to a final concentration 0.5x10 cell/ml and IL-2 before transduction with high titter HIV carrying CYTOR shRNA at an MOI of 10. 24 hr later (day three), cells were further transduced with HIVGKO lentivirus at MOI of 10. Transduced cells were cultured in complete RPMI media containing recombinant human IL2 and dynabeads at a ratio of 25 μl human beads per 10 million cells and analyzed by FACS at day five. For the IP of P-TEFb, we used the following antibodies: anti-CDK9 (Abcam ab6544) or anti-CYCLIN T1 antibodies (Abcam; ab176702). For ChIP-qPCR for the detection of histone marks activation markers, we used anti-H3K27Ac (ab4729) and anti-H3K4me3 (ab8580). For detecting the different states of the phosphorylation of RNAPII CTD, we used the phosphorylated serine 2 antibody (Ser2P; ab238146) and phosphorylated serine 5 (Ser5P; ab5131). To monitor T cell activation following stimulation, the following antibodies were used: APC anti human CD25 (Biolegend; 302609); Pacific Blue anti human CD69 (Biolegend; 310919). Actin remodeling in response to T cell receptor (TCR) engagement was monitored by forming circumferential F-actin rings as previously described . In brief, stimulatory coverslips were prepared by coating with a 0.01% poly-L-lysine (PLL; Sigma) solution for 10 minutes at room temperature, followed by wet-chamber incubation for 3 hours at 37°C with 7 μg/ml anti-CD3 antibody (50 μl per coverslip, clone HIT3a against CD3E; BD Biosciences) in phosphate-buffered saline (PBS). Stimulatory coverslips were subsequently washed in PBS and stored at 4°C in PBS until use. 5x10 cells per anti-CD3-coated coverslip, respectively) were used to seed coverslips for 4 minutes to allow TCR-mediated actin ring formation. Cells were subsequently fixed in 3% paraformaldehyde for 15 minutes, permeabilized for 2 minutes in 0.1% TritonX-100, and blocked for 30 minutes in 1% Fetal Calf Serum (FCS) in PBS. F-actin was visualized with tetramethyl rhodamine isothiocyanate (TRITC)-conjugated phalloidin (1:1,000, 1 hour, room temperature; Sigma). Samples were mounted on glass slides and analyzed by epifluorescence (Olympus IX81 S1F-3, cellM software) and confocal (spinning-disc PerkinElmer UltraView VoX, Velocity software) microscopes. For quantification of phenotype frequencies, at least 100 transfected cells were counted. Pseudotyped viruses were generated in HEK293T cells as described . Briefly, the plasmid driving the expression of the shRNA transgene was transfected into cells using Lipofectamine 2000 (Invitrogen) together with other lentiviral packaging plasmids coding for Gag, Pol Tat Rev, and the VSV-G envelope. Transfections were done in a 10cm format, and the supernatant containing the virus was harvested 72 hours post-transfection, filtered through a 0.45 μm filter spun at 2000 rpm for 5 min to remove cells debrides and stored at -80°C. 2x10 Jurkat T cells were transduced with the pseudotyped particles for transduction. 16 hours later, the medium containing lentiviral particles was changed. Following transduction, cells were cultured in a medium supplemented with 2 μg/ml of puromycin to eliminate non-transduced cells that did not express shRNA. Upon the death of all the control cells, the medium was changed, and surviving cells were propagated for future experiments. For transducing CD4 primary cells, we used HIVGKO (a gift from Eric Verdin), which codes for a codon-optimized GFP reporter under the control of the HIV-1 promoter and in the context of expression of all viral proteins and a mKO2 reporter under the control of the constitutive promoter EF1α . For knockdown (KD) of CYTOR expression, J-Lat 6.3 cells were transduced with lentiviruses that drive the expression of shRNA that specifically targets CYTOR. Cells were next selected on puromycin, and polyclonal stable cells were monitored for CYTOR expression by RT-qPCR. To achieve CYTOR over-expression, cells were transduced with a lentivirus that drives the expression of CYTOR—exons 1, 4 and 5, the most abundantly expressed form in humans. Following antibiotic selection, resistant J-Lat 6.3 T cells were subjected to RT-qPCR to confirm CYTOR over-expression or knockdown. CYTOR RNA levels were normalized to the gapdh gene. Although we are aware that GAPDH expression is elevated following TCR stimulation, we did analyze several cellular genes in search of a better marker, but concluded that, e.g., genes for actin polymerization machinery are all affected more strongly by T cell activation than gapdh . We therefore used gapdh for normalization. To obtain CYTOR knockdown in stimulated primary CD4+ T cells, cells were isolated from health donors (n = 3) and stimulated with anti-CD3/CD28 beads (1:1 ratio of beads to cell number). Cells were cultured on stimulation media (RPMI+IL2), and on day 3 post isolation and stimulation, cells were subjected to transduction with lentivirus expressing shRNA against CYTOR. The following day cells were transduced with HIVGKO and 48 hour later were analyzed by FACS for HIV-GFP and mKO2 expression. To monitor the effects of CYTOR in promoting HIV latency, we followed the kinetics of entry of stimulated Jurkat 2D10 T cells that express a cassette of the HIV provirus, expressing 2dGFP reporter. Cells where CYTOR is depleted or over-expressed and control cells that express scramble shRNA were activated with P/I and then sorted by FACS to isolate those that express GFP. Cells were then grown, during which their HIV GFP expression was followed by FACS. Control cells expressing scramble shRNA or cells where CYTOR expression was depleted (KD) were cross-linked with 1% formaldehyde for 10 minutes and then washed with PBS and reverse cross-linked with glycine (125mM; 5 minutes). Cells were then lysed for 10 minutes on ice in 130μl sonication buffer (20 mM Tris pH-7.8, 2 mM EDTA, 0,5% SDS, 0.5 mM phenylmethylsulfonyl fluoride (PMSF), and 1% protease inhibitor cocktail), and the nuclear pellets were collected. DNA was fragmented by sonication at the following settings: amplitude 20% for 30 cycles at 10 seconds on/10 seconds off. Samples were centrifuged (15 minutes, 14,000 rpm, 4°C). The soluble chromatin fraction (25 μg) was collected and immunoprecipitated (IP) overnight at 4°C on a rotating wheel in IP buffer (0.5% Triton X-100, 2 mM EDTA, 20 mM Tris pH-7.8, 150 mM NaCl and 10% glycerol) with 2.5 μg of one of the indicated antibodies. The next day, the IP material was incubated with 25 μl dynabeads protein G for two hours to ensure the binding of the antibody to the magnetic beads. DNA was eluted with freshly prepared elution solution (1% SDS and 0.1 M NaHCO3) and heated at 65°C overnight to reverse-crosslink the samples. Precipitated DNA fragments were then extracted using a ChIP DNA clean and concentrator kit (ZYMO Research), and HIV DNA levels were quantified by qPCR with the primers specifically located on the NFκB region at the HIV-LTR promoter. All signals were normalized relative to input DNA. ChIP assays were also performed with an anti-rabbit or mouse IgG as negative control. 3x10 cells were cross-linked with freshly made 1% formaldehyde in PBS for 10 minutes at room temperature while shaking. Crosslinking was quenched with 125 μM glycine for 5 minutes at room temperature. Cells were centrifuged at 1200 rpm for 5 minutes at 4°C and washed twice with PBS on ice. The pellet was re-suspended in 300 μl of sonication buffer (50mM Tris 7.0, 10mM EDTA, 1% SDS, DTT, PMSF, protease inhibitors (Roche), and RNase inhibitor (NEB). Cells were then incubated on ice for 10 minutes and sonicated in Bioruptor at high settings of 3 rounds each of 10 cycles 40 seconds ON/40 seconds OFF. Water was changed to ice-cold between the rounds. Sociated samples were centrifuged at max speed for 10 minutes at 4°C, and then chromatin material was kept at -80°C. For IP, chromatin was diluted in twice the volume of hybridization buffer (500 mM, NaCl, 1% SDS, 100 mM Tris pH-8, 10 mM EDTA, 15% Formamide, protease inhibitors (Roche) and RNase inhibitor (NEB).2 μg of biotinylated RNA was added to 0.5 ml diluted chromatin and mixed by end-to-end rotation at 37°C for 4 hours. Streptavidin-magnetic beads were washed three times in sonication buffer, blocked with 500 ng/μl yeast total RNA and 1 mg/ml BSA for 1 hour at room temperature before resuspended in their original volume. 40 μl of beads were added, and the reaction was incubated for 30 minutes at 37°C. Beads were captured by a magnet and washed five times with the wash buffer (2x SSC, 0.5% SDS, supplemented with fresh DTT and PMSF). Beads were resuspended in 3-times of the original volume with the DNA elution buffer (50 mM NaHCO3, 1% SDS, 200 mM NaCl), and DNA was eluted with a cocktail of 100 μg/ml RNaseA (Sigma) and 0.1 U/μl RNase H (Epicenter). Chromatin was reverse-cross-by treatment with 0.2 U/μl proteinase K at 65°C for 45 minutes. DNA was then extracted with an equal volume of phenol:chloroform: isoamyl alcohol(Invitrogen) and precipitated with ethanol at -80°C overnight. For probe in-vitro transcription of linear RNA synthesis, 1 μg of RNA was transcribed and biotinylated using AmpliScribe-T7-Flashbiotin-RNA transcription kit (Epicentre) according to the manufacturer’s instructions. Eluted DNA was analyzed by qPCR with primers specific to the HIV promoter. 10 Jurkat J-Lat 6.3 cells were washed twice with PBS and resuspended in 800 μl of RNA-IP buffer (0.5% NP-40, 20 mM HEPES pH 7.8, 100 mM KCl, 0.2 mM EDTA supplemented with RNase inhibitor (NEB) and protease inhibitor (Sigma). Cells were cross-linked, and cell lysate was incubated on ice for 10 minutes before isolating nuclei through centrifugation at 2500g for 15 minutes. The supernatant was collected and resuspended in freshly prepared RIP buffer. ChIP material was then sonicated, and the pelleted nuclear membrane and debris were removed by centrifugation at 13,000 rpm for 10 minutes. Isolated ChIP material was incubated with 2.5 μg of indicated antibodies overnight at 4°C. Then, 20 μl of pre-blocked BSA protein A beads were added and incubated for an additional 2 hours at 4°C. 50 μl of cell lysate was collected as input samples. Beads were washed 4 times with washing buffer (0.5% NP-40, 20 mM HEPES pH 7.8, 100 mM KCl, 0.2 mM EDTA supplemented with RNase inhibitor (NEB) and protease inhibitor (Sigma) to remove unbound material. The pellet was resuspended in 100 μl of the lysis buffer and extracted using a TRIZOL reagent (Sigma). RNA was reverse transcribed using qPCRBIO kit (PCPbiosystems), and qPCR was performed using indicated primers against CYTOR or 7SK ncRNA. The amplification of 7SL RNA served as a control RNA that is not associated with P-TEFb. Input RNA was extracted and reverse-transcribed the same way. Dilutions of input were used for standard curve and calculations. 10 Jurkat cells were washed twice with PBS and resuspended in 800 μl of RNA-pull-down buffer (0.5% NP-40, 20 mM HEPES pH 7.8, 100 mM KCl, 0.2 mM EDTA supplemented with RNase inhibitor (NEB) and protease inhibitor (Sigma). Lysates were incubated with an in-vitro transcribed biotinylated CYTOR anti-sense probe (synthesized by IDT), and reactions were pulled down with streptavidin beads. Beads were resuspended in 3 times of their original volume of DNase buffer (100 mM NaCl and 0.1% NP-40), and protein was eluted with a cocktail of 100 μg/ml RNaseA (Sigma) and 0.1 U/μl RNaseH (Epicenter) and 100 U/ml DNase I (Invitrogen) at 37°C for 30 minutes. Eluted proteins were subjected to western blotting with indicated antibodies. Non-specific IgG served as control. Biotinylated scrambled RNA was used as a control for RNA-IP. 7SK RNA confirmed association with P-TEFb. Input is 5% of the total cell lysate . The cytosolic extracts were prepared by resuspending 3x10 cells in 500μl of Buffer A (10 mM KCl, 10 mM MgCl2, 10 mM HEPES, 1 mM EDTA, 1 mM DTT, 0.1% PMSF, and EDTA-free complete protease inhibitor cocktail (Roche) with 0.5% NP-40 for 10 minutes on ice. The nuclei were spun down at 5,000 g for 5 minutes, and the supernatant was saved as the cytosolic extract (CE). The nuclei were washed once with 200 μl of Buffer A with 0.5% NP-40 and re-pelleted. The nuclei were resuspended in 100 μl of Buffer B (450 mM NaCl, 1.5 mM MgCl2, 20 mM HEPES, 0.5 mM EDTA, 1 mM DTT, 0.1% PMSF, and EDTA-free complete protease inhibitor cocktail (Roche) and incubated on ice for 60 minutes. The lysates were clarified by centrifugation at 20,000g for 10 minutes to prepare nuclear extract (NE). RNA from nuclear or cytoplasmic was extracted with Trizol, and RNA was reverse transcribed using a qPCRBIO kit (PCPbiosystems), and qPCR was performed using indicated primers against CYTOR or 7SK ncRNA. CYTOR and 7SK RNA levels were normalized to 7SL RNA in each of the fractions and conditions. For identifying ncRNAs that are differentially expressed upon T cell stimulation in cells that carry active HIV (GFP+) versus latent HIV (GFP-), J-Lat 6.3 cells were stimulated with P/I and sorted based on their GFP expression (n = 4). For analysis of transcriptome upon CYTOR depletion cells, primary CD4+ T cells were stimulated with CD3/CD28 beads and then subjected to CYTOR KD by transducing cells with lentivirus that drive the expression of shRNA that target CYTOR (n = 3). CYTOR overexpression was obtained by transducing stimulated cells with lentivirus that drive CYTOR expression. RNA was purified utilizing the RNeasy Mini kit (QIAGEN) according to the manufacturer’s instructions. The integrity of the isolated RNA was tested using the Agilent High Sensitivity RNA Kit and Tapestation 4200 at the Genome Technology Center at the Faculty of Medicine Bar-Ilan University. Total RNA was used for mRNA enrichment by using the NEBNext mRNA polyA Isolation Module (NEB; E7490L), and libraries for Illumina sequencing were performed using the NEBNext Ultra II RNA kit (NEB; E7770L). Quantification of the library was performed using dsDNA HS Assay Kit and Qubit 2.0 (Molecular Probes, Life Technologies), and qualification was done using the Agilent D1000 Tapestation Kit and Tapestation 4200. 150 nM of each library was pooled together and was diluted to 4nM according to NextSeq manufacturer’s instructions. 1.6 pM was loaded onto the Flow Cell with 1% PhiX library control. Libraries were sequenced on an Illumina NextSeq 500 instrument, 75 cycles of single-read sequencing. For analysis: Quality Control was conducted by evaluating the quality of the FASTQ files using ’Fastqc’ (v0.12.1). Subsequently, the samples underwent quality trimming via the ’Fastp’ software (v0.23.3). To identify potential contaminations, ’fastq-screen’ software was applied. Read alignment involved aligning the reads to the Human GRCh38 genome (Ensembl release 110) using ’STAR’ software (v2.7.10b). Read assignments to coding regions were determined using the ’SubRead’ package (’FeatureCounts’ v2.0.6). Finally, BAM files were sorted using ‘samtools’. Differential expression analysis was conducted in DAVID. PCA was performed to evaluate data dispersion, revealing a batch effect among samples collected on different dates. A correction was implemented in the DESeq2 model to minimize this batch effect. Finally, the indicated groups were compared (without including the control in the model), identifying 90 DEGs. The selected cutoff values were an adjusted p-value <0.05 and a fold-change > 2. For enrichment analysis, the DAVID program was employed to identify enriched pathways and terms associated with the selected genes . These quality control and data analysis steps ensure the reliability and accuracy of the RNA-seq analysis. Primers on the HIV promoter: NFκB forward: 5’ - AGGTTTGACAGCCGCCTA -3’ NFκB Reverse: 5’ - AGAGACCCAGTACAGGCAAAA -3’ gapdh Forward: 5’ - AGCCACATCGCTCAGACAC -3’ gapdh Reverse: 5’ - GCCCAAACGACCAAATCC -3’ Primers for CYTOR: Forward: 5’- AACTTGCCAGCCTCCATC; Reverse: 5’- GAGCTTCCTGTTTCATCTCCC Primers for 7SK: Forward; 5‘- GAGGGCGATCTGGCTGCGACAT Reverse: 5‘- ACATGGAGCGGTGAGGGAGGAA Statistical evaluation was performed with GraphPad Prism 7 using two-way ANOVA with no correction for multiple comparisons. Number of independent data points refers to biological replicates. Each data point, as mentioned in the figure legends, represents the mean of 3–4 independent experiments with the errors calculated based on mean ± SD. Differences were considered statistically significant and denoted as ***p≤0.05; n.s., not significant. We would also like to thank people in the Taube lab who read the manuscript. Also, we thank Dr. Liron Levin, Dr. Aviad Sivan, and Mr. Yehuda Baruch for helping with the bioinformatic analysis. All relevant data can be found at NIH GEO - accession number GSE254771. All other data are within the manuscript and its Supporting information files. This work is supported by the Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 508136175 – FA 378/25-1 to OTF and RT) and the German Centre for Infection Research (DZIF) (TTU 04.820 – HIV reservoir to OTF). OTF is a member of the CellNetworks cluster of excellence (EXC81). Additional funding for RT is from the Bi-National Science Foundation (BSF) - 2021273 to RT and JS and the National Institute of Health -NIH-R21 – 5R21AI170195 for RT and KF. KF is also supported by NIH R01AI167778, and Gilead Mentored Scientist Award from the UCSF AIDS Research Institute (ARI) and a Boost award from the NIH-funded UCSF-Bay Area Center for AIDS Research (P30 AI027763). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. All relevant data can be found at NIH GEO - accession number GSE254771. All other data are within the manuscript and its Supporting information files.
PMC12118871
PCYT2 overexpression induces mitochondrial damage and promotes apoptosis in hepatocellular carcinoma cells
Phosphatidylethanolamine cytidyltransferase 2 (PCYT2) is commonly regarded as the rate-limiting enzyme in eukaryotic phosphatidylethanolamine synthesis. However, the role of PCYT2 in the development of hepatocellular carcinoma (HCC) unknow. In this study, the role of PCYT2 overexpression in the development of HCC was examined by culturing HepG2 cells. We compared the expression levels of PCYT2 in L02 cells and HepG2 cells. Then, the HepG2 cells were infected with the lentivirus, establishing PCYT2 overexpression cell models. The proliferation, migration, and apoptotic abilities of PCYT2 overexpression in HepG2 cells was observed using western blotting, CCK-8 assay, Transwell assay, wound healing, and plate cloning methods. Based on this overexpression model, we determined the mitochondrial function and lipid content of HepG2 cells using lipidomics. CDP-ethanolamine (CDP-Etn), a downstream product of PCYT2, was added to the HepG2 cells, inhibiting their proliferation and migration. BALB/c female nude mice inoculated with subcutaneously transplanted tumors were used to explore the role of PCYT2. The results of the in-vitro experiments, shown that the expression of PCYT2 in normal hepatocytes was higher than that in HCC cells, and addition of CDP-Etn and PCYT2 overexpression inhibited the proliferation and migration of HCC cells, promoted the apoptosis of HCC cells, and caused mitochondrial damage. The results of in vivo experiments demonstrated that the tumor volume in the PCYT2 overexpression group was significantly smaller than that in the blank control group. Thus, PCYT2 overexpression inhibits the development of HCC, and its mechanism may be related to the impairment of mitochondrial function.Hepatocellular carcinoma (HCC) is an exceedingly fatal malignancies, with mortalities that approximate the incidence rates worldwide . HCC is usually diagnosed at advanced stages owing to late symptom manifestations with limited therapeutic options, leading to ineffective intervention and poor prognosis . An increasing number of studies have focused on the progression, pathological features, and prognosis of liver cancer [3–5]. HCC epidemiology is rapidly evolving, one of the most common causes is non-alcoholic fatty liver disease , further proving that lipid metabolism plays a crucial role in HCC occurrence. Therefore, the identification of novel therapeutic targets is urgently needed to improve the treatment of patients with HCC . Phosphatidylethanolamine (PE), also known as cephalin, is the most abundant lipid in the cytoplasmic layer of cell membranes and is involved in cellular processes such as membrane fusion , autophagy and apoptosis [9–11]. For eukaryotic PE in vivo, two main synthetic pathways exist : including the Kennedy pathway of CDP-ethanolamine(CDP-Etn) and mitochondrial phosphatidylserine decarboxylation pathway . Phosphatidylethanolamine cytidyltransferase 2(PCYT2) is the rate-limiting enzyme of the CDP-Etn pathway. Previous studies have shown that PCYT2 is highly specific, and is present only in the rough endoplasmic reticulum of eukaryotes . P-Eth is then catalyzed by PCYT2 to form CDP-Etn, leading to PE synthesis . Generally, PCYT2 expression is reduced in various epithelial-derived cancer cell lines compared to normal cells . Compared epithelial-derived cancer cell lines PCYT2 activity with that of breast epithelial cells MCF-10A showed that its PCYT2 activity was inhibited in breast cancer cells (MCF-7) . PCYT2 expression was significantly reduced in invasive human metastatic colon cancer cells compared to that in primary tumor cells , and previous studies have shown that PCYT2 knockdown under nutrient-rich conditions significantly facilitated the proliferation of HeLa and T98G cells and promoted in vivo tumor growth. The inhibition of PCYT2 increases P-Etn levels in cancer cells and stimulates tumor growth . However, in organoid models, PCYT2 knockdown inhibits cell growth . Despite these findings, no evidence exists that suggests that PCYT2 expression is associated with HCC occurrence and development. This study aimed to investigate the role of PCYT2 in human hepatocellular carcinoma cells (HepG2) by inhibiting their proliferation, invasion, and migration abilities and promoting cell apoptosis. CDP-Etn (90756) was purchased from Sigma (Darmstadt, Germany).The ATP assay kit (S0026) and BCA protein assay kit (P0010S) were purchased from Beyotime (Shanghai, China), and RPMI 1640 medium (SH30027.01) was purchased from Gibco (Waltham, MA,USA). Fetal bovine serum(FBS #11012–8611) was purchased from TIANHANG (Zhejiang, China). PCYT2 antibody (ab135290) was provided by Abcam (Cambridge, UK); BAX (#5023), Bcl-2 (#3498), cleaved caspase-3 (#9661) and β-actin (#3700) were purchased from Cell Signaling Technology (Danvers, MA, USA); goat anti-rabbit (E-AB-1034) and goat anti-mouse (E-AB-1035) secondary antibodies were provided by Elabscience (Shanghai, China). L02 cells (normal human liver cells) were cultured in RPMI1640 medium, and HepG2 were cultured in DMEM containing 10% FBS and 1% penicillin-streptomycin. The above two cell types of cells were cultured at 37°C in a humidified incubator atmosphere containing 5% CO2. Lentiviral plasmids overexpressing PCYT2 (LV/In) and negative control (NC/In) were purchased from GENECHEM (Shanghai, China). HepG2 cells were infected with LV/In or NC/In, 48 h later, the cells were screened with DMEM medium containing puromycin (2 µg/mL) for approximately 2 weeks. The expression level of PCYT2 was detected using western blotting after 3–4 generations to confirm PCYT2 expressing up-regulation. RIPA buffer was used to extract total cellular protein. Protein samples were electrophoresed using 9–11% SDS-PAGE, then transferred to PVDF membranes (Merck, Darmstadt, Germany) that were blocked with 5% skimmed milk in TBST, and incubated overnight utilizing primary antibodies, including anti-PCYT2 (1:250), anti-β-actin (1:3000), anti-BAX (1:1000), and anti-Bcl-2 (1:1000). Subsequently, samples were incubated 1–2 h using an appropriate peroxidase-linked secondary antibodies. Consequently, employing β-actin we normalized the protein levels. Images were visualized using the Chemi-Doc MP system (Bio-Rad). HepG2 cells were seeded into 96-well plates and incubated with a Cell Counting kit (CCK-8) solution for 40 min at 37°C. Absorbance was measured at 450 nm using a Spectra MAX M5 microplate spectrophotometer to detect cell viability. To determine cell migration, the Transwell assay was performed by adding 500 µL of complete medium to a 24-well plate, then placing the Transwell chamber on the plate. The cell suspension was prepared by adding basal medium and 200 µL (3 × 10 cells/200 µL) to each top chamber. Subsequently, the cells were incubated at 37°C for 1 d. The cells that successfully migrated and attached to the surface of the underlying membrane were fixed with paraformaldehyde and stained with 0.1% crystal violet. Five to six fields of view (original magnification: 200×) were randomly selected for cell counting under a light microscope. Dilute the Matrigel with basal medium (1:8), lay it flat on a Transwell membrane, and incubate at 37°C for 4 h. The subsequent experimental steps were identical to those used for the migration assay, including cell culture, paraformaldehyde fixation and 0.1% crystal violet staining. Two milliliter of cell suspension was added to a six-well plate (3 × 10 cells/well) and cultured at 37°C in a humidified incubator atmosphere at 5% CO2; 24 h later, the center of each well was scratched with a 100 µL plastic micropipette and the medium was substituted with basal medium. Each well was imaged under a 200 × light microscope at three randomly selected identical locations after culturing for 24 and 48 h. Finally, we measured cell migration ability by comparing the distance between the edges of each wound at 24 and 48 h. A cell suspension (2 mL, 5 × 10 cells/mL) was added to each well of a six-well plate and incubated for 2 weeks. When the clones were visible to the naked eye, they were washed twice with phosphate-buffered saline (PBS), and the cells were fixed using –4% paraformaldehyde for 10 min. Paraformaldehyde was washed away and 0.1% crystal violet staining was performed for 15 min. The clones were washed twice with PBS, dried at room temperature and the number of colonies was statistically analyzed. Cellular ATP levels were measured using an ATP kit (Beyotime). Sample tubes containing 20 μL of sample or standards were placed in a luminometer (SuPerMax 3100) and rapidly mixed with a micropipette. After > 2 s, relative light unit values were measured. The tiled cells (1 × 10) were removed, immersed in PBS, and the cell surface was rinsed. Then 2.5% of pre-cooled glutaraldehyde was added to the tiled cells at 4°C, fixed at 4°C for 2 h or overnight, aspirated, and soaked in PBS twice for 10 min each. Finally, ethanol gradient dehydration (30, 50, 70, 80, 90, 100%), critical point drying, coating, and electron microscopy was performed to observe mitochondrial morphology. Five-week-old BALB/c female nude mice were purchased from Jiangsu Jicui Pharmaceutical Biotechnology Co, passed SPF level training and assessment, and were routinely reared using standard SPF conditions. PCYT2 overexpressed HepG2 cells and control cells were subcutaneously injected into the left and right sides of the nude mice (approximately 1 × 10 cells on each side, n = 6 tumors in each group). Following 20–25 d, the tumors achieved a certain size and the micewere anesthetized with isopentane and sacrificed; the tumors were collected for follow-up evaluation. The animal experiments were approved by the Animal Ethics Committee of Anhui Medical University and conducted in accordance with the guidelines for the care and use of laboratory animals. Data were analyzed using GraphPad Prism software (version 8.0), and the results were expressed as mean ± standard deviation (SD). T-test was performed to determine the significant of differences between two groups. A one-way analysis of variance was used for comparison across groups. Statistical significance was set at P < 0.05. The Cancer Genome Atlas (TCGA) database, Cancer Research Project, which is a collaboration between the National Cancer Institute and National Human Genome Research Institute, provides a large, free reference database for cancer research by collecting and collating cancer-related data. By analyzing the TCGA database, we attempted to determine the relationship between the PCYT2 expression level and HCC. Our findings demonstrated that overall survival of patients with HCC was closely correlated with PCYT2 expression. We observed that the survival percentage of patients gradually decreased over time; however, patients with high PCYT2 expression levels had a higher survival percentage than those with low PCYT2 expression (Fig 1A). This indicates that high PCYT2 expression levels are beneficial for the overall survival of patients with HCC. Western blotting was conducted on cultured L02 and HepG2 cells to verify their PCYT2 expression level (Fig 1B and 1C), we found that PCYT2 expression was higher in L02 cells than in HepG2 cells. To verify the effect of PCYT2 on HCC development, we transfected HepG2 cells with a lentivirus and performed western blotting, successfully establishing PCYT2 overexpression in HepG2 cells (Fig 2A). The expression levels of Bcl-2 (an anti-apoptotic factor), Bax (an apoptogenic factor) and cleaved-caspase-3(an apoptogenic factor)(S1 Fig.) were detected using western blotting, and the results revealed that the LV group exhibited increased HepG2 cell apoptosis (Fig 2B–2D). The CCK-8 assay indicated that PCYT2 overexpression significantly reduced HepG2 cells viability (Fig 2E). Utilizing the Transwell assay, we discovered that the HepG2 cells invasion and migratory ability was suppressed post-lentiviral transfection (Fig 2F–2H). The scratch wound healing assay also determined that the migratory ability of the LV group was inhibited (Fig 2I). Additionally, PCYT2 overexpressing HepG2 cells formed less colonies than the negative control group (Fig 2J and 2K). In summary, we demonstrated that PCYT2 overexpression inhibited the proliferation, invasion, and migration of HepG2 cells. (A–D) The level of PCYT2 (A) as well as the protein expression of Bax and Bcl-2 was measured per group (NC, normal control; LV, lentivirus transfection to over-express PCYT2) using western blotting and representative protein quantification (n = 3 per group). (E) The proliferation ability of each group was detected using CCK-8 (n = 6 per group). (F–H) A Transwell assay was conducted per group to investigate the effects of cellular migration and invasion. Magnification: 100 × . (I–K) The scratch wound healing and colony formation assays of cells post-PCYT2 overexpression. The magnification for scratch wound healing assay is 100 × . Data are presented as means ± SD. *P < 0.05, **P < 0.01, ***P < 0.001. During cell apoptosis, the mitochondria undergoes many changes, such as respiratory chain depolymerization, oxidative phosphorylation, decreased ATP synthesis, and increased reactive oxygen species. We measured the concentration of ATP in the cells of the NC and LV groups (Fig 3A), and photographed the mitochondria using transmission electron microscopy (Fig 3B). The results showed that the LV group exhibited mitochondrial damage and its ATP content was lower than that in the NC group. Compared to the NC group, in the LV group, the mitochondria was swollen and its number was significantly reduced (Fig 3B), further suggesting that PCYT2 overexpression promotes apoptosis. (A) The ATP levels of each group (NC, normal control; LV, lentivirus transfection to over-express PCYT2) were detected using commercial reagent kits (n = 3 per group). (B) Images of mitochondrial damage post-PCYT2 overexpression were captured via transmission electron microscopy. Data are presented as means ± SD. **P < 0.01. PCYT2’s inhibition of cellular proliferation and migration was mediated by altering downstream metabolite levels, CDP-Etn was introduced to HepG2 cells, and cell viability was measured using CCK-8. Therefore, HepG2 + CDP-Etn cells have lower cell viability than HepG2 cells (Fig 4A). In addition, TUNEL staining was performed to observe apoptosis in HepG2 and HepG2 + CDP-Etn cells; the results indicated that CDP-Etn promoted apoptosis in HepG2 cells, affirming those of the CCK-8 assay (Fig 4B and 4C). Furthermore, Transwell assay (Fig 4D) and scratch experiment (Fig 4E) was conducted to detect cell migration, suggesting that supplementation with CDP-Etn inhibited cell migration. Subsequently, plate cloning showed that fewer colonies formed in HepG2 + CDP-Etn cells than in HepG2 cells (Fig 4F and 4G). Thus, we confirm that CDP-Etn inhibited the proliferation and migration ability of HepG2; that is, PCYT2 inhibited proliferation and migration by altering downstream PE metabolism. (A) The proliferation ability of HepG2 cells with or without CDP-Etn supplementation were detected using CCK-8 (n = 3 per group). (B–C) HepG2 cells underwent TUNEL staining, and the nucleus underwent DAPI staining. Magnification: 100 × (D–G) Transwell (D), scratch wound healing (E) and colony formation assays (F–G) were performed post-CDP-Etn supplementation. Magnification for Transwell and scratch wound healing assay is 100 × . Data are presented as means ± SD. *P < 0.05, **P < 0.01. To investigate whether PCYT2 affects hepatocarcinogenesis and development in vivo, we subcutaneously implanted PCYT2-overexpressing HepG2 cells and blank control HepG2 cells into 12 nude BALB/c mice. The tumor volume was measured at 2 d intervals(Fig 5C). After 20 d, the mice were anesthetized with isoflurane to observe the tumor volume using an in vivo imaging system (Fig 5A). The nude mice were sacrificed and tumors were removed for imaging (Fig 5B) and weighing (Fig 5D). The diameter of the largest tumor was 1.9 cm. The results showed that the fluorescence intensity of the tumors in the LV group was weaker than that in the NC group and the tumor volume and weight in the NC group were greater than that in the LV group, suggesting that PCYT2 overexpression inhibit tumor growth in vivo. (A–B) Representative image of in vivo imaging system (A) and tumors isolated from mice xenograft model (B), which were established by subcutaneously implanting PCYT2-overexpressing HepG2 cells (LV) and blank control HepG2 cells (NC). (C, D) Cancer volumes were measured every other day, and the average tumor weights in each group were measured. n = 6 per group, data are presented as means ± SD. ***P < 0.001, ****P < 0.0001. PCYT2 is a rate-limiting enzyme in PE synthesis that is commonly used in the study of obesity-related diseases, such as non-alcoholic fatty liver disease and type 2 diabetes [12,20–22]. Recently, interest in the role of PCYT2 in cancer has been growing . Previous studies show that PCYT2 has different roles in various cancers and cancer settings. For instance, in metastatic colorectal cancer (CRC), PCYT2 is significantly downregulated and functions as a tumor metastasis inhibitor . In human breast cancer cells (MCF-7), the level of PCYT2 in cancer cells is elevated in response to the stressful environment . Our findings show that PCYT2 expression is abnormally downregulated in HepG2 cells, which is consistent with that of previous studies where in PCYT2 expression was downregulated in MCF-7 and invasive human metastatic CRC . Although PCYT2 regulates several human cancers , its role in HCC cells remains unknown. Based on the literature and our findings, PCYT2 in human cancers appears to have a consistent expression profile in human cancers, irrespective of tumor origin or location . Regarding the mechanism whereby PCYT2 influences cancer development, a previous report showed that PCYT2 downregulation-induced phosphoethanolamine (PEtn) accumulation correlated with tumor growth under nutrient starvation, thereby PCYT2 overexpression reduced PEtn levels and tumor growth . However, in the present, we found that CDP-Etn supplementation inhibited HCC migration, invasion and proliferation. In our previous study, we showed that the levels of BAX and cleaved caspase-3 were significantly increased whereas Bcl-2 was significantly reduced in the livers of type 2 diabetic mice and L02 cells after stimulation with high glucose and free fatty acids (HG&FFA)(12). CDP-Etn (100 μM) protected cells from HG&FFA-induced apoptosis by reducing BAX and cleaved caspase-3 levels as well as and increasing Bcl-2 levels . Whether PCYT2 downregulation induced PEtn accumulation contributes to HCC development further study. Increasing evidence suggests that PCYT2 is aberrantly expressed in various models of liver disease and may predict clinical outcomes in patients. PCYT2 is instrumental in the deregulation of these processes leading to the development of obesity, insulin resistance, liver steatosis and dyslipidemia . CDP-Etn supplementation has been reported to alleviate PCYT2 deficiency engendering age-dependent and insulin-resistant non-alcoholic steatohepatitis to improve patient prognosis [20,29–31]. Chronic administration of peroxisome proliferators can increase the content of hepatic PC and PE for hepatomegaly and proliferation as well as cause liver cancer in rodents . Based on the current studies, we hypothesized that PCYT2 may be involved in the regulation of cellular processes in HCC. Therefore, we utilized in vivo and in vitro validation methods to assess the expression and mechanistic roles of PCYT2 in liver cancer cells. Herein, PCYT2 overexpression was determined to inhibit HCC cell proliferation, migration and invasion both in vitro and in vivo. And, the number and morphology of mitochondria in HCC cells overexpressing PCYT2 were significantly different from those in HCC cells without any treatment, such as a decrease in the number of mitochondria and swelling of the mitochondria. These changes suggest that PCYT2 affects the mitochondrial function of cells. Notably, when the HepG2 cells received CDP-Etn supplementation, their proliferation, migration, and invasion were inhibited in vitro. Notably, the ATP level decreased in HCC cells following the overexpression of PCYT2, and the cells were found to be accompanied by mitochondrial damage using transmission electron microscopy. However, previous reports have indicated that PCYT2 is present only in the endoplasmic reticulum of hepatocytes . Additionally, the phenotype of liver PCYT2 knockout mice showed no signs of liver injury, however, they experienced massive accumulation of liver triglycerides (TAG) . Therefore, we hypothesized that the influence of PCYT2 on mitochondrial function is mediated by metabolites such as TAG, DAG, and PE. However, further studies are needed to clarify whether the PCYT2 exerted inhibition of HCC cells alleviates mitochondrial damage. As understanding, the mechanism whereby PCYT2 overexpression causes mitochondrial damage will deepen our understanding of PCYT2 regulation in HCC cells. In conclusion, this study provided compelling data demonstrating the aberrant expression and functional role of PCYT2 in HepG2 cells. PCYT2 expression levels were lower in HepG2 than in L02. Furthermore, PCYT2 overexpression in HepG2 cells induced mitochondrial damage; inhibited proliferation, invasion, and migration; and promoted cell apoptosis.
PMC1462997
Derivation of normal macrophages from human embryonic stem (hES) cells for applications in HIV gene therapy
Many novel studies and therapies are possible with the use of human embryonic stem cells (hES cells) and their differentiated cell progeny. The hES cell derived CD34 hematopoietic stem cells can be potentially used for many gene therapy applications. Here we evaluated the capacity of hES cell derived CD34 cells to give rise to normal macrophages as a first step towards using these cells in viral infection studies and in developing novel stem cell based gene therapy strategies for AIDS. Undifferentiated normal and lentiviral vector transduced hES cells were cultured on S17 mouse bone marrow stromal cell layers to derive CD34 hematopoietic progenitor cells. The differentiated CD34 cells isolated from cystic bodies were further cultured in cytokine media to derive macrophages. Phenotypic and functional analyses were carried out to compare these with that of fetal liver CD34 cell derived macrophages. As assessed by FACS analysis, the hES-CD34 cell derived macrophages displayed characteristic cell surface markers CD14, CD4, CCR5, CXCR4, and HLA-DR suggesting a normal phenotype. Tests evaluating phagocytosis, upregulation of the costimulatory molecule B7.1, and cytokine secretion in response to LPS stimulation showed that these macrophages are also functionally normal. When infected with HIV-1, the differentiated macrophages supported productive viral infection. Lentiviral vector transduced hES cells expressing the transgene GFP were evaluated similarly like above. The transgenic hES cells also gave rise to macrophages with normal phenotypic and functional characteristics indicating no vector mediated adverse effects during differentiation. Phenotypically normal and functionally competent macrophages could be derived from hES-CD34 cells. Since these cells are susceptible to HIV-1 infection, they provide a uniform source of macrophages for viral infection studies. Based on these results, it is also now feasible to transduce hES-CD34 cells with anti-HIV genes such as inhibitory siRNAs and test their antiviral efficacy in down stream differentiated cells such as macrophages which are among the primary cells that need to be protected against HIV-1 infection. Thus, the potential utility of hES derived CD34 hematopoietic cells for HIV-1 gene therapy can be evaluated.Human embryonic stem cells (hES cells) show great promise for many novel cellular therapies due to their pluripotent nature . These cells have the capacity to give rise to mature cells and tissues that arise from all three germ layers during embryonic development [2-4]. Several pluripotent hES cell lines have so far been derived from the inner cell mass of human blastocysts and can be cultured indefinitely in an undifferentiated state [5-7]. Thus, these cells provide a renewable source of pluripotent stem cells from which many types of differentiated cells could be produced for experimental and therapeutic purposes. Cell differentiation protocols currently exist for the derivation of neurons, cardiomyocytes, endothelial cells, hematopoietic progenitor cells, keratinocytes, osteoblasts, and hepatocytes to name a few . In addition to providing for potential cellular replacement therapies, opportunities exist in programming hES cells to correct a genetic defect and/or to express a therapeutic transgene of interest. Using such approaches, many possibilities exist for treating a number of genetic and immune system disorders . Many novel applications can be foreseen for hES cells in infectious disease research. AIDS is a potential disease that can benefit from exploiting hES cells for cell replacement therapy as they have the capacity to differentiate into various hematopoietic cells. HIV continues to be a major global public health problem with infections increasing at an alarming rate . Given the present lack of effective vaccines and the ineffectiveness of drug based therapies for a complete cure, new and innovative approaches are essential. Gene therapy through intracellular immunization offers a promising alternative approach and possible supplement to current HAART therapy [12-14]. HIV mainly targets cells of the hematopoietic system, namely, T cells, macrophages, and dendritic cells . As infection progresses, the immune system is rendered defenseless against other invading pathogens and succumbs to opportunistic infections. There is a great deal of progress in the area of stem cell gene therapy for AIDS . A primary goal of many ongoing studies is to introduce an effective anti-HIV gene into hematopoietic stem cells [16-18]. As these cells possess the ability to self renew, they have the potential to continually produce HIV resistant T cells and macrophages in the body thus providing long term immune reconstitution. These approaches use CD34 hematopoietic stem cells for anti-HIV gene transduction via integrating viral vectors such as lentiviral vectors [16-18]. Lentiviral vectors have several advantages over conventional retroviral vectors since higher transduction efficiencies can be obtained and there is less gene silencing. The CD34 cells currently used for many therapies are primarily obtained from bone marrow or mobilized peripheral blood . Thus, CD34 progenitor cells are an essential ingredient for HIV gene therapy. In view of the need for CD34 cells for HIV gene therapy as well as for other hematopoietic disorders, if one can produce these cells in unlimited quantities from a renewable source, it will overcome the limitations of securing large numbers of CD34 cells for therapeutic purposes. In this regard, progress has been made in deriving CD34 cells from hES cells (hES-CD34). Different methods currently exist to derive CD34 cells from hES cells with varying efficiencies [20-27]. Recent reports have indicated the capacity of hES cell derived CD34 cells to give rise to lymphoid and myeloid lineages thus paving the way for utilization of these cells for hematopoietic cell therapy [20,27-29]. For the effective utilization of hES-CD34 cells for HIV gene therapy, a number of parameters need to be examined. First, one has to demonstrate that hES-CD34 cells can give rise to macrophages and helper T cells which are the main cells that need to be protected against HIV infection. Recent evidence has shown that hES-CD34 cells can give rise to myelomonocytic cells . However, thorough phenotypic or functional characterization of these cells is lacking. It is also not clear if these cells are susceptible to HIV infection. Similarly, although the hES-CD34 cells were shown to have lymphoid progenitor capacity, only B cell and natural killer (NK) cell differentiation has been examined so far . The capacity to generate T cells remains to be evaluated. With this background, as a first step, our primary goal in these studies is to examine the capacity of hES-CD34 cells to give rise to phenotypically and functionally normal macrophages and whether such cells are susceptible to productive HIV infection. Since lentiviral vectors have been shown to successfully transduce hES cells [30-33], we further investigated the ability of transduced hES cells to differentiate into transgenic macrophages that can support HIV-1 infection. Demonstration of HIV-1 productive infection in these cells will permit future efficacy evaluations of anti-HIV genes in this system. Here we show that normal and lentiviral vector transduced hES-CD34 cells can give rise to phenotypically and functionally normal macrophages that support HIV infection thus paving the way for many novel approaches to evaluate their potential for HIV gene therapy. Undifferentiated hES cell colonies grown in media supplemented with 4 ng/ml bFGF displayed normal morphology of pluripotent human embryonic stem cells with tight and discreet borders on the MEF feeder layers (Fig 1A). Similarly, lentiviral vector transduced hES cell colonies, also displayed normal morphology and growth characteristics (Fig 1A). As expected, the vector transduced colonies displayed green fluorescence due to the presence of the GFP reporter gene. When cultured on irradiated S17 mouse bone marrow stromal cells, both nontransduced and transduced hES cells developed into embryonic cystic bodies (Fig 1A). FACS analysis of single cell suspensions of the cystic bodies showed levels of CD34 cells which ranged from 7–15%. Figure 1B displays a representative FACS profile of hES-CD34 cells. Purified CD34 cells were later cultured in semi-solid methylcellulose medium to derive myeloid colonies. Both nontransduced (denoted as ES in figures) and vector transduced (denoted as GFP ES in figures) hES cell derived CD34 cells gave rise to normal myelomonocytic colonies similar to human fetal liver derived CD34 cells (denoted as CD34 in figures) (Fig 1A). When pooled colonies were cultured further in liquid cytokine media for 12–15 days for differentiation, the cells developed into morphologically distinct macrophages (Fig 1A). When compared, the morphology of macrophages derived from all stem cell progenitor populations appeared similar. These results were found to be consistent in replicative experiments. The transgene GFP expression was also maintained during the differentiation of hES cells into mature macrophages. GFP expression in cystic body derived CD34 cells was around 80% (data not shown) with similar levels seen in differentiated macrophages (Fig 2). Derivation of macrophages from lentiviral vector transduced and normal hES cells. A) Transduced and non-transduced H1 hES cells were cultured on mouse S17 bone marrow stromal cell layers to derive cystic bodies. Cystic body derived CD34 cells were purified by positive selection with antibody conjugated magnetic beads and placed in methocult media to obtain myelomonocytic colonies. Pooled colonies were cultured in liquid cytokine media supplemented with GM-CSF and M-CSF to promote macrophage growth. For comparison, fetal liver derived CD34 cells were cultured similarly to derive macrophages. Representative ES cell colonies, cystic bodies, methocult colonies, and derivative macrophages are shown with GFP expressing cells fluorescing green under UV illumination. B) Representative FACS profile of hES cell derived CD34 cells stained with PE conjugated antibodies. Percent positive CD34 cells are shown with isotype control shown in the left panel. Phenotypic FACS analysis of hES cell derived macrophages. A) Macrophages derived from transduced and nontransduced hES CD34 and fetal liver CD34 cells were stained with antibodies to CD14, HLA-DR, CD4, CCR5, and CXCR4 and the expression of these surface markers was analyzed by FACS. B) Isotype controls for PE and PE-CY5 antibodies. Percent positive cells are displayed in the plots for each respective cell surface marker staining. Dot plots are representative of triplicate experiments. Macrophages play a critical role in immune system function and are also major target cells for many viral infections including HIV-1. Distinct surface phenotypic markers exist on these cells and, thus far, there has been no thorough evaluation of hES cell derived macrophages. Therefore we analyzed hES cell derived macrophages for the presence of characteristic cell surface markers and compared these to the phenotypic profile displayed on fetal CD34 cell derived macrophages. The surface markers analyzed were CD14, a monocyte/macrophage specific marker, HLA-DR (a class II MHC molecule found on antigen presenting cells), CD4, the major receptor for HIV-1 infection, and CCR5 and CXCR4, chemokine receptors which are critical coreceptors essential for HIV-1 entry. EGFP expression was also analyzed to determine the levels of transduction and any transgene silencing that may occur during differentiation. Fetal liver (CD34), nontransduced (ES), and vector transduced (GFP ES) hES cell derived macrophages were all positive for the monocyte/macrophage marker CD14 (99.3%, 88.7%, and 99.2%, respectively) (Fig 2A). However, the mean fluorescent intensity (MFI) was found to be lower on hES cell derived macrophages. Surface expression of HLA-DR was observed at similar levels between macrophages derived from fetal liver CD34 cells (99.6%), nontransduced hES cells (92.8%), and transduced hES cells (98.2%) (Fig 2A). CD4 levels were comparable for all stem cell derived macrophages (99.2%, 83.3%, and 88.7%, respectively) (Fig 2A). CCR5 and CXCR4 cell surface expression was also observed for fetal liver CD34 cell (99.6% and 99.3%), nontransduced hES cell (91.9% and 92.6%), and transduced hES cell (98.9% and 99.3%) derived macrophages (Fig 2A). As compared to fetal liver CD34 cell derived macrophages, hES cell derived macrophages displayed a higher level of expression of CXCR4. Isotype controls for both PE and PECY5 stains are shown in Fig 2B. The above phenotypic data are representative of triplicate experiments. The antigen presenting cell surface specific marker HLA-DR (MHC II) on normal macrophages is critical for presenting antigen to CD4 T cells. A second co-stimulatory molecule, B7.1 is present at low basal levels on resting macrophages and is necessary to activate T cells. Its expression is elevated upon activation with certain stimuli such as LPS. Our results of LPS stimulation of respective macrophages have shown upregulation of B7.1 with values for fetal liver CD34 cell (CD34) (27.9% to 75.4%) nontransduced (ES) (17.8% to 49.4%) and transduced (GFP ES) (35.6% to 65.7%) hES cell derived macrophages (Fig 3A). These values represent a significant upregulation of B7.1 for all three macrophage populations. Functional analysis of hES cell derived macrophages for B7.1 costimulatory molecule upregulation and phagocytosis of E. coli particles: A) Mature macrophages were stimulated with LPS to determine B7.1 upregulation. Twenty-four hours post-stimulation, macrophages were labeled with a PE-CY5 conjugated anti-B7.1 antibody and analyzed by FACS. B7.1 upregulation data are representative of triplicate experiments. Isotype control is shown in the left panel. B) To assess phagocytic function, E. coli Bioparticleswere added directly to the cultured macrophages. Twenty four hours post-addition, cells were analyzed by FACS. Percent positive cells are displayed in the plots for each experiment. These data are representative of triplicate experiments. Another important function of macrophages is their ability to phagocytose foreign material and present antigenic peptides on their cell surface. To evaluate phagocytic function, fluorescently labeled E. coli Bioparticleswere added to macrophage cultures followed by FACS analysis. Nontransduced (94.6%) as well as lentiviral vector transduced (98.7%) hES cell derived macrophages were found to be capable of phagocytosing the Bioparticlesin comparison to fetal liver CD34 cell derived macrophages (95.8%) (Fig 3B). These values are representative of triplicate experiments. Magi-CXCR4 cells with no phagocytic capacity were used as non-phagocytic cell controls and similarly exposed to E. coli Bioparticles(Fig 3B). No uptake of the bacteria could be seen. Thus, uptake of E. coli Bioparticlesby macrophages is indicative of active ingestion. Macrophages, as effector cells, play a key role in the inflammatory response. Activated macrophages secrete various cytokines, two of the major ones being IL-1 and TNF-α. To determine if hES cell derived macrophages have such a capacity, cells were stimulated with LPS. On days 1, 2, and 3 post-stimulation, culture supernatants were analyzed by ELISA to detect IL-1 and TNF-α. As seen in figure 4A, there were no significant differences in IL-1 secretion between the three sets of macrophages. Similarly, nontransduced and transduced hES cell derived macrophages were also capable of TNF-α secretion upon LPS stimulation. However, levels of the respective cytokines detected were slightly lower than those from fetal liver CD34 cell derived macrophages (Fig 4B). The values of cytokine secretion levels represent triplicate experiments. Cytokine IL-1 and TNFα secretion by stimulated hES cell derived macrophages: Macrophages derived from transduced and nontransduced hES and fetal liver CD34 cells were stimulated with 5 μg/ml LPS. On days 1, 2, and 3 post-stimulation, supernatants were collected and assayed by ELISA for (A) IL-1 and (B) TNFα. Experiments were done in triplicate. The above data have shown that hES cell derived macrophages are very similar to normal human macrophages based on phenotypic and functional analysis. In addition to being important cells of the immune system, macrophages are among the major target cells for certain viral infections, particularly for HIV-1. We wanted to determine if hES cell derived macrophages were susceptible to HIV-1 infection compared to standard macrophages. In these studies, we only used an R5-tropic strain of HIV-1 since macrophages are natural targets for this virus. Our results from challenge studies of these cells clearly indicated the capacity of hES cell derived macrophages in supporting a productive infection. Levels of virus increased up to 15 days similar to non-hES derived macrophages showing that the initial viral input was amplified in productive viral infection. However, the levels of viral yield were found to be slightly lower for the ES cell derived macrophages. In the case of GFP-ES macrophages, there was a decline in viral titer. This could be due to possible lower numbers of cells present in the initial cultures. As a first step towards the use of hES cells for hematopoietic stem cell and HIV gene therapies, we have shown here that phenotypically and functionally normal macrophages could be derived from hES-CD34 cells. Both non transduced and lentiviral vector transduced hES cells were found to be capable of generating CD34 cells that give rise to macrophages which could support productive HIV-1 infection. Current sources of CD34 cells consist of human bone marrow, cytokine mobilized peripheral blood, fetal liver, and cord blood . However, the number of cells that can be obtained for manipulations is not unlimited. Therefore, deriving CD34 cells for therapeutic and investigative purposes from hES cells with unlimited growth potential has the advantage of a consistent and uniform source. The ability to obtain phenotypically normal and functionally competent macrophages from hES cells is important to evaluate their potential therapeutic utilities in the future. Additionally, testing of transgenic hES cells derived via lentiviral vector gene transduction is also helpful to determine the stability of the transgene expression and their capacity for differentiation into end stage mature cells such as macrophages. Based on these considerations, both non- transduced and lentiviral vector transduced hES cells were evaluated for their capacity to give rise to CD34 progenitor cells. In colony forming assays using semisolid methylcellulose medium, the morphology of myelomonocytic colonies derived from hES CD34 cells appeared similar to that of fetal liver CD34 cells. When subsequently cultured in cytokine media that promotes macrophage differentiation, morphologically normal macrophages were obtained with hES-CD34 cells similar to that of fetal liver CD34 cells. At higher magnification, the macrophages displayed flat projecting cellular borders with fried egg appearance with distinct refractory lysosomal granules in the cytoplasm (data not shown). Lentiviral vector transduced hES cells also did not display any abnormal growth or differentiation characteristics as compared to nontransduced hES-CD34 cells indicating no adverse effects due to vector integration and expression. Transduced cells gave rise to cystic bodies with similar CD34 cell content and profiles upon development. The transduced hES-CD34 cells also gave rise to apparently normal macrophages that expressed the transgene as shown by GFP expression. These results are consistent with those of others that showed normal differentiation of hES cells to other cell types following lentiviral transduction . A requirement for successful cellular and HIV-1 gene therapy is that mature end stage cells derived from CD34 progenitor cells be phenotypically and functionally normal to maintain and restore the body's immunological function. Accordingly, hES cell derived macrophages were evaluated to determine if they met these criteria. Macrophages display distinct cell surface markers upon end stage differentiation. To determine whether hES cell derived macrophages display these surface markers, FACS analysis was performed to detect the presence of CD14, HLA-DR (MHCII), CD4, CCR5, and CXCR4. As observed in Fig 2A, both nontransduced and transduced hES cell derived macrophages expressed all of these markers with some differences in their levels of expression. HLA-DR, CD4, and CCR5 expression profiles were comparable between all cell types analyzed. Even though all cell types analyzed stained positive for CD14, relative expression of CD14 was slightly lower on hES cell derived macrophages compared to fetal liver CD34 cell derived macrophages. On the contrary, the levels of CXCR4, a chemokine receptor involved in cellular homing, were found to be higher on hES-CD34 cell derived macrophages. This may be due to inherent differences in the cell types and/or due to their physiological state at the time of harvest . Additional hES cell lines need to be evaluated in the future to establish if these differences are consistent. A major functional role of macrophages in vivo is their ability to serve as professional antigen presenting cells. During this process macrophages present antigen peptide fragments complexed with both classes of MHC molecules and deliver a costimulatory signal through the expression of B7 molecules. Upon stimulation with LPS, hES-CD34 cell derived macrophages had shown upregulation of the costimulatory molecule B7.1 similar to cells derived from fetal liver. Furthermore, the hES-CD34 cell derived macrophages also showed a normal capacity to ingest foreign particles in phagocytosis assays using E.coli Bioparticles. In addition to antigen presentation and phagocytosis, macrophages also play a critical role in inflammation and secrete cytokines in response to external stimuli. When exposed to LPS, the hES-CD34 cell derived macrophages secreted two important cytokines IL-1 and TNF-α similar to that of fetal liver derived cells. The above data has established that phenotypically and functionally normal macrophages could be derived from hES-CD34 cells. Macrophages in addition to playing important physiological roles are also major cell targets for certain viral infections, particularly HIV-1. Here we evaluated the susceptibility of hES-CD34 cell derived macrophages to be productively infected with HIV-1. Similar to that of fetal liver CD34 cell derived cells, the hES-CD34 macrophages also supported HIV-1 infection although the levels of viral yield differed somewhat. However this should not be a major concern for testing anti-HIV genes in these cells. In all the above experiments, the vector transduced transgenic macrophages also behaved similarly to that of nontransduced cells showing that they were also physiologically normal. The lack of vector toxicity on cellular maturation is encouraging for future work with transduced hES-CD34 cells to derive other important differentiated cells like T cells and dendritic cells relevant for HIV studies. Although there are numerous studies on hES cell differentiation into many important end stage mature cells, systematic work on hES cell hematopoietic differentiation and thorough characterization of end stage mature cells that participate in critical immune responses has just begun [21,27-29]. Our current results established that physiologically normal macrophages could be derived from hES cells and that these cells have the potential for use in cellular and gene therapies. To our knowledge this is the first demonstration that hES cell derivatives can be used for infectious disease research. Due to the extensive ability for hES cells to self-renew, large numbers of differentiated cells can be derived so that infection studies and evaluation tests can be carried out in a more standardized way. Our results showing that both normal and transgenic derivative macrophages support HIV-1 infection points out to their utility for testing anti-HIV constructs transduced into hES-CD34 cells and pave the way for their application in stem cell based HIV gene therapy. So far a number of studies including our own have tested many gene therapeutic constructs in CD34 cells from conventional sources. These constructs include anti-HIV ribozymes, RNA decoys, transdominant proteins, bacterial toxins, anti-sense nucleic acids, and most recently siRNAs [36-50]. In addition, a number of cellular molecules that aid in HIV-1 infection such as cellular receptors and coreceptors CD4, CCR5 and CXCR4 have also been successfully tested in CD34 cell derived macrophages and T cells . Some of these approaches have progressed into clinical evaluations as well . Based on our current results, many of these novel anti-HIV constructs can also be tested in hES-CD34 cells for their potential application. Although there are advantages of using hES cell derived CD34 cells for potential cellular therapies, transplantation of these cells constitutes an allogenic source with immune rejection as a major issue. However, a recent study using human leukocyte reconstituted mice suggested that hESCs and their derivative cell types were less prone to invoking an allogeneic response . Other recent studies demonstrated successful engraftment of primary and secondary recipients with hES cell derived hematopoietic cells in both immunodeficient mice and in vivo fetal sheep models adding further support that any obstacles could be overcome . Moreover, multiple novel strategies to avoid immune-mediated rejection of hES cell-derived cells have been proposed . It is not too far in the future that even autologous hES cells may be derived from specific individuals for deriving CD34 cells which can be used for cell replacement therapy. Phenotypically normal and functionally competent macrophages could be derived from hES-CD34 cells. Since these cells are susceptible to HIV-1 infection, they provide a uniform source of macrophages for viral infection studies. Based on these results, it is also now feasible to transduce hES-CD34 cells with anti-HIV genes such as inhibitory siRNAs and test their antiviral efficacy in down stream differentiated cells such as macrophages which are among the primary cells that need to be protected against HIV-1 infection. Thus, the potential utility of hES derived CD34 hematopoietic cells for HIV-1 gene therapy can be evaluated. The NIH approved human ES H1 cell line was obtained from WiCell (Madison, Wisconsin). hES cell colonies were cultured on mouse embryonic fibroblasts (MEF) (Chemicon, Temecula, CA) in the presence of DMEM-F12 (Invitrogen, Carlsbad, CA) supplemented with 20% KNOCKOUT serum replacement with 1 mM L-glutamine, 1% Nonessential Amino Acids, 0.1 mM β-mercaptoethanol, 0.5% penicillin/streptomycin, and 4 ng/ml human basic fibroblast growth factor. Culture medium was replaced daily with fresh complete DMEM-F12. Mature colonies were subcultured weekly by digesting with collagenase IV as previously described . A VSV-G pseudotyped lentiviral vector (SINF-EF1a-GFP) containing a GFP reporter gene (kindly supplied by R. Hawley, George Washington University) was used for hES cell transductions as previously described (30, 58). Generation of the pseudotyped vector in 293T cells and its concentration by ultracentrifugation were described previously . For vector transduction, the undifferentiated hES cells were prepared into small clumps of 50–100 cells with enzyme digestion as done for routine passaging of cells. The cell clumps were incubated with the vector for 2 hrs in the presence of polybrene 6 ug/ml. A secondary cycle of transduction was done by adding fresh vector and incubating for another 2 hrs. The general vector titers were 1 × 10and the multiplicity of infection was 10. The transduction efficiency was about 50%. The transduced colonies were cultured on MEF like above. Undifferentiated hES cells were cultured on S17 mouse bone marrow stromal cell monolayers to derive cystic bodies containing CD34+ hematopoietic progenitor stem cells. hES cell cultures were treated with collagenase IV(1 mg/ml) for 10 minutes at 37°C and subsequently detached from the plate by gentle scraping of the colonies. The hES cell clusters were then transferred to irradiated (35 Gy) S17 cell layers and cultured with RPMI differentiation medium containing 15% FBS (HyClone, Logan, UT), 2 mM L-glutamine, 0.1 mM β-mercaptoethanol, 1% MEM-nonessential amino acids, and 1% penicillin/streptomycin. Media was changed every 2 to 3 days during 14–17 days of culture on S17 cells . After allowing adequate time for differentiation, hES cystic bodies were harvested and processed into a single cell suspension by collagenase IV treatment followed by digestion with trypsin/EDTA supplemented with 2% chick serum (Invitrogen, Carlsbad, CA) for 20 minutes at 37°C. Cells were washed twice with PBS and filtered through a 70 uM cell strainer to obtain a single cell suspension. To assess the levels of CD34 cells in the bulk cell suspension, cells were labeled with PE conjugated anti-CD34 antibody (BD Biosciences, San Jose, CA) and analyzed by FACS. To purify the CD34 cells, Direct CD34 Progenitor Cell Isolation Kit (Miltenyi Biotech, Auburn, CA) was used following the manufacturer's protocol. Isolated CD34 hematopoietic progenitor stem cells were then analyzed by FACS as mentioned above to determine cell purity. For comparative experiments, human CD34 hematopoietic progenitor cells were also purified from fetal liver tissue as described above. CD34 cells were cultured initially in semisolid media to derive myelomonocytic colonies followed by liquid culture in cytokine supplemented media as described below. Purified CD34+ progenitor cells (~2.5 × 10to 4.0 × 10) were placed directly into Methocult semisolid medium (Stem Cell Technologies, Vancouver, BC), mixed, and cultured in 35 mm plates. Myeloid colonies were allowed to develop for 12–15 days. Upon differentiation and proliferation, myelomonocytic colonies were harvested by the addition of 5 ml DMEM containing 10% FBS, 10 ng/ml each GM-CSF and M-CSF. Cells (~10) were placed in a 35 mm well and allowed to adhere for 48 hours. At two and four days post-harvest, medium was replaced with fresh complete DMEM supplemented with 10 ng/ml GM-CSF and M-CSF. By 4–5 days, cells developed into mature macrophages which were used for subsequent phenotypic and functional characterization. To determine if nontransduced and lentiviral vector transduced hES cell derived macrophages display normal macrophage surface markers, FACS analysis was performed using respective fluorochrome conjugated antibodies. Fetal liver derived CD34+ cells as well as nontransduced and transduced hES cell derived macrophages were evaluated in parallel. Cells were scraped from their wells, washed two times with PBS, and stained with the following antibodies: PE-CD14, PE-HLA-DR, PECY5-CD4, PECY5-CCR5, PECY5-CXCR4 (BD Biosciences, San Jose, CA). A blocking step was first performed by incubating the cells with the respective isotype control for 30 minutes at 4C before staining with the respective cell surface marker antibodies. Isotype control staining was used to determine background levels. FACS analysis was performed on a Beckman-Coulter EPICS XL-MCL flow cytometer with data analysis using EXPO32 ADC software (Coulter Corporation, Miami, FL). A minimum of 8,000 cells were analyzed in each FACS evaluation. Physiological roles of macrophages include phagocytic and immune related functions. To determine if hES cell derived macrophages were functionally normal, a stimulation assay to determine upregulation of the costimulatory molecule B7.1 was performed. Activated macrophages upregulate the expression of B7.1 upon activation with various stimuli. Accordingly, fetal liver CD34, nontransduced hES, and GFP-alone transduced hES cell derived macrophages were stimulated by the addition of LPS (5 ug/ml) to the cell culture medium. Twenty-four hours post-stimulation, cells were stained with an anti-B7.1 antibody labeled with PE-Cy5 (BD Biosciences, San Jose, CA) and analyzed by FACS. To assess the hES cell derived macrophages' phagocytic function, 5 ug/ml of fluorescently labeled E. coli Bioparticles(Invitrogen, Carlsbad, CA) were added directly to the cell culture medium. Four hours later, macrophages were washed six times with PBS and fresh medium with 10 ng/ml GM-CSF and M-CSF was added. Twenty-four hours later, cells were analyzed by FACS for the presence of ingested Bioparticleswhich can be detected in the PE (FL2) channel. Lentiviral vector transduced Magi-CXCR4 cells, a HeLa cell derivative with no phagocytic capacity, were used as non-phagocytic cell controls and similarly exposed to E. coli Bioparticles Human ES cell derived macrophages were also analyzed for their ability to secrete two major cytokines, IL-1 and TNF-α, upon external stimulation. Accordingly, macrophages were stimulated with 5 ug/ml of LPS during culture. On days 1, 2, and 3 post-stimulation, cell culture supernatant samples were collected and analyzed by a QuantikineELISA kit (R&D Systems, Minneapolis, MN). Non-stimulated supernatants were also analyzed for basal levels of cytokine secretion. To determine if hES cell derived macrophages can be infected with HIV-1 and support viral replication, cells were challenged with a macrophage R5-tropic BaL-1 strain of HIV-1. An m.o.i. of 0.01 in the presence of 4 ug/ml polybrene was used. At different days post-infection, culture supernatants were collected and assayed for p24 antigen by ELISA. To quantify viral p24 levels, a Coulter-p24 kit (Beckman Coulter, Fullerton, CA) was used. The author(s) declare that they have no competing interests. JA and SB contributed equally to this work. SB was responsible for deriving CD34 cells from the hESC and culturing macrophages. JA performed the phenotypic, functional and infection assays on the differentiated macrophages. DSK provided hES cell protocols and supplied lentiviral vector transduced cells. RA was responsible for the overall experimental design and implementation of the project. hES cell derived macrophages support productive HIV-1 infection: Macrophages derived from transduced and nontransduced hES CD34 and fetal liver CD34 cells were infected with macrophage R5-tropic HIV-1 BaL-1 strain at an m.o.i. of 0.01. Culture supernatants were collected on different days post infection and assayed for viral p24 antigen by ELISA. Data is representative of triplicate experiments.
PMC1160574
Derivation of Multipotent Mesenchymal Precursors from Human Embryonic Stem Cells
Human embryonic stem cells provide access to the earliest stages of human development and may serve as a source of specialized cells for regenerative medicine. Thus, it becomes crucial to develop protocols for the directed differentiation of embryonic stem cells into tissue-restricted precursors. Here, we present culture conditions for the derivation of unlimited numbers of pure mesenchymal precursors from human embryonic stem cells and demonstrate multilineage differentiation into fat, cartilage, bone, and skeletal muscle cells. Our findings will help to elucidate the mechanism of mesoderm specification during embryonic stem cell differentiation and provide a platform to efficiently generate specialized human mesenchymal cell types for future clinical applications.Embryonic stem (ES) cells are pluripotent cells derived from the inner cell mass of the blastocyst that can be maintained in culture for an extended period of time without losing differentiation potential. The successful isolation of human ES cells (hESCs) has raised the hope that these cells may provide a universal tissue source to treat many human diseases. However, directed differentiation of hESCs into specific tissue types poses a formidable challenge. Protocols are currently available for only a few cell types, mostly of neural identity [1–3], and differentiation into many of the cell types derived from the paraxial mesoderm has not been reported, with the exception of a recent study indicating osteoblastic differentiation . Mesenchymal stem cells (MSCs) have been isolated from the adult bone marrow , adipose tissue , and dermis and other connective tissues . Harvesting MSCs from any of these sources requires invasive procedures and the availability of a suitable donor. The number of MSCs that can be obtained from a single donor is limited, and the capacity of these cells for long-term proliferation is rather poor. In contrast, hESCs could provide an unlimited number of specialized cells. In this study, we present techniques for the generation and purification of mesenchymal precursors from hESCs and their directed differentiation in vitro into various mesenchymal derivatives, including skeletal myoblasts. Our isolation method for mesenchymal precursors is the first example, to our knowledge, of efficiently deriving structures of the paraxial mesoderm from ES cells, and further highlights the potential of hESCs for basic biology and regenerative medicine. Undifferentiated hESCs, H1 (WA-01, XY, passages 40–65) and H9 (WA-09, XX, passages 35–45), were cultured on mitotically inactivated mouse embryonic fibroblasts (Specialty Media, Phillipsburg, New Jersey, United States) and maintained under growth conditions and passaging techniques described previously . OP9 cells were maintained in alpha MEM medium containing 20% fetal bovine serum (FBS) and 2 mM L-glutamine. Mesenchymal differentiation was induced by plating 10 × 10 to 25 × 10 cells/cm on a monolayer of OP9 cells in the presence of 20% heat-inactivated FBS in alpha MEM medium. Flow-activated cell sorting (FACS) (CD73-PE; PharMingen, San Diego, California, United States) was performed on a MoFlo (Cytomation, Fort Collins, Colorado, United States). All human ES cell–derived mesenchymal precursor cell (hESMPC) lines in this study are of polyclonal origin. Primary human bone marrow–derived MSCs and primary human foreskin fibroblasts (both from Poietics, Cambrex, East Rutherford, New Jersey, United States) were grown in alpha MEM medium containing 10% FBS and 2 mM L-glutamine. hESMPCs are grown to confluence followed by exposure to 1 mM dexamethasone, 10 μg/ml insulin, and 0.5 mM isobutylxanthine (all from Sigma, St. Louis, Missouri, United States) in alpha MEM medium containing 10% FBS for 2–4 wk. Data were confirmed in hESMPC-H1.1, -H1.2, -H1.3, and -H9.1 (hESMPC-H1.4 was not tested). Differentiation of hESMPCs was induced in pellet culture by exposure to 10 ng/ml TGF-β3 (R & D Systems, Minneapolis, Minnesota, United States) and 200 μM ascorbic acid (Sigma) in alpha MEM medium containing 10% FBS for 3–4 wk. Data were confirmed in hESMPC-H1.1, -H1.3, and -H9.1 (hESMPC-H1.2 and -H1.4 were not tested). hESMPCs were plated at low density (1 × 10 to 2.5 × 10 cells/cm) on tissue-culture-treated dishes in the presence of 10 mM β-glycerol phosphate (Sigma), 0.1 μM dexamethasone, and 200 μM ascorbic acid in alpha MEM medium containing 10% FBS for 3–4 wk. Data were confirmed in hESMPC-H1.1, -H1.3, and -H9.1 (hESMPC-H1.2 and -H1.4 were not tested). Confluent hESMPCs were maintained for 2–3 wk in alpha MEM medium with 20% heat-inactivated FBS. More rapid induction was observed in the presence of medium conditioned for 24 h by differentiated C2C12 cells. Coculture of hESMPCs and C2C12 cells was carried out in alpha MEM with 3% horse serum and 1% FBS . Data were confirmed in hESMPC-H1.3, -H1.4, and -H9.1 (hESMPC-H1.1 and -H1.2 were not tested). Immunocytochemistry for all surface markers was performed on live cells. Monoclonal antibodies VCAM, STRO-1, ICAM-1(CD54), CD105, CD29, and MF20 were from Developmental Studies Hybridoma Bank (University of Iowa, Iowa City, Iowa, United States); CD73, CD44, and ALCAM(CD166) were from BD Biosciences Pharmingen (San Diego, California, United States). All other immunocytochemical analyses were performed after fixation in 4% paraformaldehyde and 0.15% picric acid, followed by permeabilization in 0.3% Triton X100. Polyclonal antibodies used were MyoD (Santa Cruz Biotechnology, Santa Cruz, California, United States) and nestin (gift from R. McKay); monoclonal antibodies were vimentin, alpha smooth muscle actin, fast-switch myosin, pan-cytokeratin (all from Sigma), and human nuclear antigen (Chemicon, Temecula, California, United States). Alkaline phosphatase reaction was performed using a commercially available kit (Kit-86; Sigma) and the mineral was stained with silver nitrate according to the von Kossa method. Fat granules were visualized by Oil Red O staining solution (Sigma). Alcian Blue (Sigma) was used to detect extracellular matrix proteoglycans in chondrogenic cultures. Total RNA was extracted by using the RNeasy kit and DNase I treatment (Qiagen, Valencia, California, United States). Total RNA (2 μg each) was reverse transcribed (SuperScript; Invitrogen, Carlsbad, California, United States). PCR conditions were optimized and linear amplification range was determined for each primer by varying annealing temperature and cycle number. PCR products were identified by size, and identity was confirmed by DNA sequencing. Primer sequences, cycle numbers, and annealing temperatures are provided in Table S1. Total RNA (5 μg) from primary MSCs, from hESMPC-H9.1, hESMPC-H1.2, and three samples of undifferentiated hESCs (H1; passages 42–46), were processed by the Memorial Sloan-Kettering Cancer Center Genomics Core Facility and hybridized on Affymetrix (Santa Clara, California, United States) U133A human oligonucleotide arrays. Data were analyzed using MAS5.0 (Affymetrix) software. Transcripts selectively expressed in each of the mesenchymal cell populations (MSC, hESMPC-H9.1, and hESMPC-H1.2) were defined as those called “increased” by the MAS5.0 algorithm in each of three comparisons with independent samples of undifferentiated hESCs. A Venn diagram was generated to visualize overlap in gene expression. Further statistical analyses were performed as described below. Mesenchymal differentiation of hESCs (lines H1 [WA-01] and H9 [WA-09]) was induced by plating undifferentiated hESCs on a monolayer of murine OP9 stromal cells , in the presence of 20% heat-inactivated FBS in alpha MEM medium. OP9 cells have been previously shown to induce blood cell differentiation from mouse ES cells . After 40 d of coculture, cells were harvested and sorted by FACS for CD73, a surface marker expressed in adult MSCs (Figure 1A). An average of 5% CD73+ cells was obtained from the mixed culture of OP9 and differentiated hESC progeny. CD73+ cells were replated in the absence of stromal feeders on tissue culture plates and expanded in alpha MEM medium with 20% FBS for 7–14 d. We next established the membrane antigen profile of the resulting population of flat spindle-like cells. The H1- and H9-derived CD73+ cells expressed a comprehensive set of markers that are considered to define adult MSCs, including CD105(SH2), STRO-1, VCAM (CD106), CD29(integrin β1), CD44, ICAM -1(CD54), ALCAM(CD166), vimentin, and alpha smooth muscle actin (Figure 1B and 1C). The cells were negative for hematopoietic markers such as CD34, CD45, and CD14. They were also negative for neuroectodermal, epithelial, and muscle cell markers including nestin, pancytokeratin, and desmin (data not shown). The human identity of these presumed mesenchymal cells (termed hESMPC-H1.1, -H1.2, -H1.3, -H1.4, and -H9.1) was confirmed for all experiments by immunocytochemistry for human nuclear antigen to rule out the possibility of contamination with OP9 cells (Figure S1). (A) FACS (MoFlo, Cytomation) for the isolation of CD73+ precursors (right) and isotype control (left). (B) Flow cytometry analysis of the CD73+ hESMPC population for various markers characteristic of MSCs, including CD44, CD73, CD105, CD166, VCAM, ICAM-1, CD29, and STRO-1. (C) Immunocytochemistry of hESMPCs for MSC markers (VCAM, STRO-1, CD73, and CD105). The cells also express vimentin and alpha smooth muscle actin. Scale bar = 50 μm. (D) Venn diagram presenting the overlap among transcripts selectively expressed in hESMPC-H1.2, hESMPC-H9.1, and primary adult human MSCs. To further characterize hESMPCs, we performed genome-wide expression analysis using oligonucleotide arrays (Affymetrix U133A). The expression profiles of hESMPC-H1.2 and hESMPC-H9.1 were compared with that of human primary adult MSCs. Housekeeping genes for each of the mesenchymal cell populations were eliminated by subtracting those transcripts also expressed in at least one of three independent samples of undifferentiated hESCs. Based on this analysis, 1,280 transcripts were selectively expressed in hESMPC-H1.2, 932 transcripts in hESMPC-H9.1, and 1,218 transcripts in primary adult MSCs. A remarkable overlap of 579 transcripts shared among the three mesenchymal populations was observed (Figure 1D). Using the genes that were selected in the initial filter, we performed a statistical analysis on the expression levels to determine whether the genes were expressed significantly differently in the two cell types. We used a Bayesian extension to the standard t-test to assess this difference. Of the 579 genes, 412 of them were significantly different, at a false discovery rate cutoff of 0.05. The relative fold changes were also extremely large in many of the cases. We also looked at the variance of the expression levels within the cell types. For the MSCs, 94% had a coefficient of variation less than 20% for the expression (log transformed); for the ES-derived cells, 72% had a coefficient of variation less than 20%. Numerous known MSC markers were included in the list of 412 genes, such as the mesenchymal stem cell protein DSC54 (13.9-fold increase, p < 0.001), neuropilin 1 (30.4-fold increase, p < 0.001), hepatocyte growth factor (48.1-fold increase, p < 0.001), forkhead box D1 (14.8-fold increase, p < 0.001), and notch homolog 2 (2.9-fold increase, p < 0.001) . Table S2 lists the p-values from the test, the mean and standard deviation of the expression levels, and the relative fold change of all 412 genes between the two types. Known markers of MSCs, such as mesenchymal stem cell protein DSC54, were all included within the 579 shared transcripts. These findings support the immunocytochemical data and suggest that hESMPCs and primary MSCs are highly related. MSCs are characterized functionally by their ability to differentiate into mesenchymal tissues, such as fat, cartilage, and bone. Therefore, we tested whether hESMPCs have the same potential (Figure 2). (A) Adipocytic differentiation in the presence of dexamethasone, insulin, and isobutylxanthine. Adipocytic characterization by Oil Red O staining and RT-PCR analysis for PPARγ. (B) Chondrocytic differentiation in the presence of TGF-β3 and ascorbic acid. Chondrocytic characterization by Alcian Blue staining and RT-PCR for aggrecan and collagen II. (C) Osteogenic differentiation in the presence of β-glycerolphosphate, dexamethasone, and ascorbic acid. Osteocytic characterization by von Kossa staining and RT-PCR for bone-specific alkaline phosphatase (ALP) and bone sialoprotein (BSP). (D) Phase-contrast image of hESMPCs and RT-PCR for the ES cell markers Nanog and Oct-4 in hESMPC-H1.1 and -H9.1 compared with undifferentiated H1 hESCs. Scale bar = 50 μm for all panels. Adipocytic differentiation of hESMPCs was induced under conditions described previously for primary adult MSCs . Appearance of cells harboring fat granules was observed after 10–14 d in culture. After 3 wk of induction, more than 70% of the cells displayed Oil Red O+ fat granules, and PPARγ, a marker of adipocytic differentiation, was detected by RT-PCR. (Figure 2A). Chondrocytic differentiation was achieved using the pellet culture system . After 28 d in culture, more than 50% of all cells exhibited robust staining for Alcian Blue, a marker specific for extracellular matrix proteoglycans. Chondrocytic differentiation was confirmed by the gene expression of collagen II and aggrecan, two components of extracellular matrix selectively expressed by chondrocytes, using RT-PCR (Figure 2B). Osteogenic differentiation was induced in the presence of β-glycerolphosphate . Osteogenesis was demonstrated by specific staining for calcium deposition in the matrix (von Kossa, Figure 2C; or Alizarin Red, Figure S2A) and increased expression of bone-specific alkaline phosphatase and bone sialoprotein at day 28 of treatment (Figures 2C and S2B). At day 28, Alizarin Red staining was detected in approximately 70% of all cells. Throughout these studies, human adult MSCs and foreskin fibroblasts were used as positive and negative controls, respectively. In addition to adipocytic, chondrocytic, and osteogenic differentiation, reports suggested that adult MSCs can form skeletal muscle . Although generation of skeletal muscle cells from adult MSCs remains controversial, we tested whether hESMPCs exhibit this potential. Under the conditions previously described , hESMPC-H1.1 and -H9.1 did not yield significant numbers of MyoD+ cells after 15–20 d in culture. However, when confluent cells were maintained in culture in the presence or absence of 5-AzaC without passage for more than 21 d, expression of specific skeletal muscle markers such as MyoD and fast-switch myosin was observed (Figure 3A). More rapid myogenic differentiation was obtained in the presence of 24-h-conditioned medium from the murine myoblastic cell line C2C12 previously induced to form myotubes . Direct coculture of hESMPCs with C2C12 cells led to the formation of hESMPC-derived myotubes, as visualized by expression of human nuclear antigen (Figure 3B), similar to those formed by host C2C12 cells. After 1 wk of coculture, myotubes composed of human nuclei accounted for more than 10% of the total number of human cells present, and each human myotube was composed of up to ten human nuclei. Human cell contribution to myotubes in coculture was confirmed by expression of human muscle-specific transcripts such as MyoD, myosin heavy chain IIa, and myogenin (data not shown). These data demonstrate that hESMPCs can give rise to mesenchymal derivatives typically obtained from primary adult MSCs, as well as to cells expressing markers of skeletal muscle. (A) Immunocytochemistry for MyoD (red) and fast-switch myosin (green). RT-PCR for MyoD in human skeletal muscle as a positive control (hSM), and in hESMPC-H9.1 cells differentiated for 10 d in the presence of C2C12-conditioned medium (hESMPC). (B) Myotube formation induced at high cell densities in the presence of C2C12 cells. Myotube characterization by immunocytochemistry for MF20 against sarcomeric myosin (green) and human nuclear antigen (hNA, red). Left panel: Control undifferentiated hESCs (H9) do not fuse with C2C12. Right panel: Under identical culture conditions, hESMPCs (line 9.1) efficiently fuse with C2C12 cells, forming myotubes containing human nuclei. RT-PCR for human specific muscle transcripts myosin heavy chain IIa (MYHC-2) and MyoD in C2C12 cells, in human skeletal muscle as positive control (huSM), and in hESMPC-H9.1 cells cocultured with C2C12 cells. One concern for the clinical application of hESC-derived progeny in regenerative medicine is the risk of teratoma formation due to the presence of residual undifferentiated ES cells among the differentiated progeny. We did not detect markers of undifferentiated hESCs, such as Nanog or Oct-4 , in any of the hESMPCs by RT-PCR (see Figure 2D) and immunocytochemistry (data not shown), suggesting the lack of any undifferentiated ES cells in hESMPC cultures. However, future in vivo studies are required to rule out the potential of these cells for teratoma formation. Previous studies have demonstrated the derivation of neural cells [1–3], hematopoietic and endothelial lineages , and cardiomyocytes from hESCs. This study presents the induction of paraxial mesoderm with the generation of multipotent mesenchymal precursors. We calculate that under these conditions a single undifferentiated hESC yields an average of one CD73+ cell at day 40 of differentiation, suggesting a balance between cell proliferation and cell selection. There were no obvious differences in marker and gene-expression profile or in differentiation behavior among the five hESMPC lines generated. However, some of the lines (e.g., hESMPC9.1) exhibited a tendency of spontaneous osteogenic differentiation after long-term propagation. Directed differentiation of hESCs into somatic stem-cell-like precursors represents a substantial advancement in harnessing the developmental potential of hESCs. The high purity, unlimited availability, and multipotentiality of hESMPCs will provide the basis for future therapeutic efforts using these cells in preclinical animal models of disease. Such in vivo studies will also be required to properly assess the safety profile of these cells. Furthermore, our system also offers a novel platform to study basic mechanisms of mesodermal induction and differentiation during early human development. The Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo) accession number for all raw microarray data used in this study is GSE2248. The Unigene (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=unigene) accession numbers for the gene products discussed in this paper are aggrecan (Hs.2159 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=2159; bone sialoprotein (Hs.518726 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=518726; bone-specific alkaline phosphatase (Hs.75431 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=75431; collagen II (Hs.408182 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=408182; forkhead box D1 (Hs.519385 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=519385; hepatocyte growth factor (Hs.396530 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=396530; mesenchymal stem cell protein (DSC54, Hs.157461 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=157461; MyoD (Hs.520119 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=520119; myogenin (Hs.2830 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=2830; myosin heavy chain IIa (Hs.513941 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=513941; Nanog (Hs.329296 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=329296]) ; neuropilin 1 (Hs.131704 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=131704; notch homolog 2 (Hs.549056 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=549056; Oct-4 (Hs.504658 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=504658; and PPARγ (Hs.162646 [http://www.ncbi.nlm.nih.gov/UniGene/clust.cgi?ORG=Hs&CID=162646]). The discovery and isolation of human embryonic stem cells (cells that are capable of renewing themselves and turning into the many different cell types that make up the human body) has the potential to revolutionize the treatment of many diseases that require the replacement of abnormal or missing cells. In particular, it would be very valuable to be able to replace tissues that are derived from one particular tissue type—mesenchyme—which bone, cartilage, fat and muscle develop from. However, before such treatments can happen, it will be necessary to work out exactly how embryonic cells become other cells, and whether it is possible to make these changes happen in the laboratory. They took two lines of completely undifferentiated human embryonic stem cells and by culturing them in the presence of mouse cells stimulated them to turn into mesenchymal cells. They then treated these cells with compounds to make them change into specialized bone, cartilage, fat, and muscle cells. They were able to confirm that these cells were all human (important because the early part of the experiment is done in the presence of mouse cells) and also that there was no evidence that the cells became cancerous. It is theoretically possible to produce lines of bone, cartilage, fat, and muscle cells from human embryonic stem cells. However, the process will need more refinement before the cell lines could be used for treatment; ideally, for example, all the culturing would be done without any mouse cells. The United States National Institutes of Health has a group of Web pages on stem cells: http://stemcells.nih.gov/info/faqs.asp The International Society for Stem Cell Research has a list of frequently asked questions about stem cells: http://www.isscr.org/science/faq.htm We thank R. McKay for nestin antibody; P. Song and the Sloan-Kettering Genomics and Flow Cytometry Core Facilities for technical assistance; and R. Stan, V. Tabar, M. Tomishima, Y. Elkabetz, and S. Desbordes for critical review of the manuscript. This work was supported in part by the Kinetics Foundation. The funder had no role in the study design, data analysis, decision to publish, or manuscript preparation and content. embryonic stem flow-activated cell sorting fetal bovine serum human embryonic stem cell human embryonic stem cell–derived mesenchymal precursor cell mesenchymal stem cell Citation: Barberi T, Willis LM, Socci ND, Studer L (2005) Derivation of multipotent mesenchymal precursors from human embryonic stem cells. PLoS Med 2(6): e161.
PMC12396968
Multimodal profiling reveals tissue-directed signatures of human immune cells altered with age
The immune system comprises multiple cell lineages and subsets maintained in tissues throughout the lifespan, with unknown effects of tissue and age on immune cell function. Here we comprehensively profiled RNA and surface protein expression of over 1.25 million immune cells from blood and lymphoid and mucosal tissues from 24 organ donors aged 20–75 years. We annotated major lineages (T cells, B cells, innate lymphoid cells and myeloid cells) and corresponding subsets using a multimodal classifier and probabilistic modeling for comparison across tissue sites and age. We identified dominant site-specific effects on immune cell composition and function across lineages; age-associated effects were manifested by site and lineage for macrophages in mucosal sites, B cells in lymphoid organs, and circulating T cells and natural killer cells across blood and tissues. Our results reveal tissue-specific signatures of immune homeostasis throughout the body, from which to define immune pathologies across the human lifespan.The immune system leverages a dynamic network of specialized cells spread across the body to defend against infections and cancer, regulate inflammation and repair tissue damage. Myeloid cells—macrophages, monocytes and dendritic cells (DCs)—initiate innate immunity at mucosal and barrier sites, while adaptive immunity is mediated by antigen-specific T and B lymphocytes in lymphoid organs. Immune memory is established following antigen-driven activation and differentiation of T cells and B cells, resulting in heterogeneous subsets of circulating and tissue resident memory T cells (TRM) and B cells that persist in diverse tissues. With age, immune memory accumulates, although responses can become dysregulated, increasing susceptibility to infections, cancer and autoimmunity. As human immune cells and the effect of age are mostly studied in blood, we lack a comprehensive understanding of the effect of age on the majority of innate and adaptive immune cells that are maintained in tissues. Investigating human tissue immunity across diverse ages has been difficult to achieve. Obtaining tissues from organ donors enables the acquisition of blood and multiple tissues from individual donors and the isolation of viable immune cells across lineages for phenotypic, functional and multimodal single-cell profiling. Previous studies of lymphocytes from organ donors showed that T lymphocyte, natural killer (NK) cell and innate lymphoid cell (ILC) subset composition, tissue residence and certain functional attributes are specific to the tissue, indicating that localization has a dominant role in determining the maintenance and functional responses of lymphocytes. Whether these tissue-specific effects on human T cells are exhibited by other immune cell lineages, such as B cells and myeloid cells, and whether aging exerts general or tissue-specific effects have not been established. Here, we present a comprehensive analysis of human immune cells using cellular indexing of transcriptomes and epitopes (CITE-seq) to simultaneously profile transcriptomes and >125 surface proteins of myeloid and lymphoid-lineage cells in 14 tissue sites of 24 organ donors aged 20–75 years. We identified a crucial role of tissue on immune cell composition, function, homing and differentiation across myeloid and lymphocyte lineages, including signatures specific for the gut, lungs and lymphoid organs. Across age, site-specific immune cell composition was largely maintained, although age-associated changes in function, signaling and metabolism were identified in certain subsets and sites, including macrophages in the lung, B cells in lymphoid organs and CD8 T cells across sites. Together, our findings reveal the complex interplay between tissue, lineage, subset and age in immune homeostasis that is important for defining immune dysfunctions in disease. We isolated mononuclear cells (MNCs) from blood, multiple lymphoid organs, lungs, airways, intestines and other sites using established protocols from 24 donors (10 females and 14 males, aged 20–75 years) (Fig. 1a). Organ donors originated from New York City (USA) and Cambridge (UK) and were free of chronic infection, cancer and overt disease (Supplementary Table 1). We performed single-cell RNA sequencing (scRNA-seq) on tissues from all donors, including CITE-seq with 127 proteins from 22 donors (Supplementary Table 2). We obtained 1.28 million immune cell events from 10 sites with >75,000 cells per site, including blood, bone marrow (BM), spleen, different lymph nodes (LNs) including lung-associated LN (LLN), mesenteric LN (MLN) and inguinal LN (ILN), lungs, comprising bronchoalveolar lavage (BAL) and lung parenchyma, and jejunum (JEJ), divided into the intraepithelial layer (JEL) and lamina propria (JLP) (Fig. 1a). We also purified low numbers of immune cells from liver, skin, colon epithelium and colon lamina propria from 9 donors (4 females, aged 25–75 years; 5 males, aged 20–55 years). These sites were therefore not included in the annotated dataset below and are provided as a separate reference (Extended Data Fig. 1a).Fig. 1A multi-lineage human immune cell atlas encompasses tissues and age.a, Plot of cell numbers across 24 donors and 10 tissue sites (blood (BLO); BM; spleen (SPL); LNs, including ILN, LLN and MLN; lung (LNG), comprising BAL and parenchyma (PAR); and JEJ (divided into JEL and JLP)) and donor metadata including 10× Genomics sequencing chemistry (3′, n = 2; 5′, n = 22), sex (male, n = 14; female, n = 10), CMV serostatus (positive, n = 16; negative, n = 8), age (range, 20–75 years) and location of tissue acquisition (US, n = 12; UK, n = 12). Bottom bars depict the number of cells profiled from each donor, and right bars depict the number of cells in each tissue across all donors. b–d, UMAP embeddings colored by donor (b), tissue site (c) or immune lineage classified by MMoCHi (d) in the BLO, BM, SPL, ILN, LLN, MLN, BAL, PAR, JEL and JLP. Lymph., lymphocytes. a, Plot of cell numbers across 24 donors and 10 tissue sites (blood (BLO); BM; spleen (SPL); LNs, including ILN, LLN and MLN; lung (LNG), comprising BAL and parenchyma (PAR); and JEJ (divided into JEL and JLP)) and donor metadata including 10× Genomics sequencing chemistry (3′, n = 2; 5′, n = 22), sex (male, n = 14; female, n = 10), CMV serostatus (positive, n = 16; negative, n = 8), age (range, 20–75 years) and location of tissue acquisition (US, n = 12; UK, n = 12). Bottom bars depict the number of cells profiled from each donor, and right bars depict the number of cells in each tissue across all donors. b–d, UMAP embeddings colored by donor (b), tissue site (c) or immune lineage classified by MMoCHi (d) in the BLO, BM, SPL, ILN, LLN, MLN, BAL, PAR, JEL and JLP. Lymph., lymphocytes. For data integration, we leveraged multi-resolution variational inference (MrVI), which is designed for cohort studies. MrVI harmonizes variation between cell states (for unified annotation of cell states across samples) and accounts for differences between samples. Visualization with uniform manifold approximation and projection (UMAP) showed similar results across US and UK donors, sequencing technologies and other donor covariates such as sex, cytomegalovirus (CMV) and Epstein–Barr virus serostatus (Fig. 1b and Extended Data Fig. 1b–g). Although cells from blood and lymphoid organs (BM, spleen, LNs) clustered similarly, cells from mucosal sites (lung, JEJ) clustered distinctly (Fig. 1c). For annotation, we used MultiModal Classifier Hierarchy (MMoCHi), which leverages both surface protein and gene expression to hierarchically classify cells into predefined categories (Supplementary Fig. 1 and Supplementary Table 3). MMoCHi defined six major immune lineages found across all tissues (Fig. 1d) comprising 13 T cell, 5 NK/ILC, 6 B cell and 7 myeloid subsets (Extended Data Fig. 1h). These results showed broad and consistent representation of major immune lineages in our dataset. We next analyzed the subset composition and heterogeneity for each immune cell lineage across tissues based on the MMoCHi annotations above. T lymphocytes (610,429 cells) comprised low-frequency γδ T cells, which develop early in ontogeny, and predominant αβ T cells (Fig. 2a,b and Supplementary Fig. 2). CD4 and CD8 T cells (αβ TCR) were subdivided into naive (TN), terminal effector (TEMRA) and memory subsets, including effector-memory (TEM), central memory (TCM) and TRM, along with CD4 regulatory T cells (Treg) (Fig. 2a,b). Surface proteins were essential for identifying T cell subsets that were not fully resolved by scRNA-seq (Fig. 2b and Supplementary Fig. 3), as shown before. For example, surface CD45RA expression was required to distinguish CD45RA TN cells from CD45RA TCM cells and CD45RA TEM cells from CD45RA TEMRA cells, and surface γδ or αβ T cell receptor (TCR) expression to accurately identify γδ T cells from CD8 T cells, which can express TRDC (Fig. 2b). In addition, TRM cells were distinguished from TEM cells based on surface expression of CD69, CD103 and/or CD49a (Fig. 2b and Supplementary Fig. 3). T cell subsets were differentially distributed across sites; CD4 TN, CD8 TN and CD4 TCM cells were enriched in blood and multiple LN, CD4 Treg cells were enriched in LN, while CD4 and CD8 TRM cells predominated in JEJ and were present at lower frequencies in lungs, spleen and LN (Fig. 2a,b). TEMRA cells were mostly CD8 and enriched in the BM and spleen and, to a lesser extent, in lungs, while TEM cells were distributed across most sites (Fig. 2a,b). Mucosal-associated invariant T (MAIT) cells, distinguished by TRAV1.2 expression, CD161 and other markers, were predominantly found in the spleen, BM and lungs (Fig. 2a,b). TCR clonal analysis provided additional correlative evidence for subset delineation and tissue distribution (for example, with the highest clonality observed in the TEMRA subset, as previously described) (Supplementary Fig. 2).Fig. 2Multimodal classification reveals immune subset heterogeneity and distribution across sites.a–h, MMoCHi classification of immune cell subsets among T cells (a and b), NK/ILCs (c and d), B cells (e and f) and myeloid cells (g and h) in the BLO, BM, SPL, LNs (ILN, LLN, MLN), lung (BAL, PAR) and JEJ (JEL and JLP) shown as UMAP embeddings (a, c, e and g), with immune cell profiles colored by MMoCHi subset classification (left) or tissue of origin (right), or as heatmaps (b, d, f and h) showing percentage positive surface protein expression (red dot plot), averaged gene expression (blue dot plot) and relative frequency distribution (rows sum to 100%) across tissues (yellow heatmap) for subsets. Bars on the right depict cell numbers for each subset. i, Violin plots of immune subset composition in BLO, BM, SPL, LNs (including ILN, LLN and MLN), lung (including PAR and BAL) and JEJ (including JEL and JLP). Dots represent the frequency of a subset within each donor (frequencies sum to 100% for each donor). Av., average; ex., expression; pos., positive; GC B, germinal center B cell; cMono, classical monocyte; ncMono, non-classical monocyte; cDC, classical dendritic cell. a–h, MMoCHi classification of immune cell subsets among T cells (a and b), NK/ILCs (c and d), B cells (e and f) and myeloid cells (g and h) in the BLO, BM, SPL, LNs (ILN, LLN, MLN), lung (BAL, PAR) and JEJ (JEL and JLP) shown as UMAP embeddings (a, c, e and g), with immune cell profiles colored by MMoCHi subset classification (left) or tissue of origin (right), or as heatmaps (b, d, f and h) showing percentage positive surface protein expression (red dot plot), averaged gene expression (blue dot plot) and relative frequency distribution (rows sum to 100%) across tissues (yellow heatmap) for subsets. Bars on the right depict cell numbers for each subset. i, Violin plots of immune subset composition in BLO, BM, SPL, LNs (including ILN, LLN and MLN), lung (including PAR and BAL) and JEJ (including JEL and JLP). Dots represent the frequency of a subset within each donor (frequencies sum to 100% for each donor). Av., average; ex., expression; pos., positive; GC B, germinal center B cell; cMono, classical monocyte; ncMono, non-classical monocyte; cDC, classical dendritic cell. Innate lymphocytes (130,414 cells) were predominantly mature CD56CD16 NK cells expressing cytolytic markers (KLRF1, GZMB) and enriched in blood, BM and lungs (Fig. 2c,d). Immature CD56CD16 NK cells were present at lower frequencies across most tissues (Fig. 2c,d). We detected low frequencies of ILCs consisting largely of CD16NCR2IL7R ILC1s with high expression of tissue residency markers (CD69, CD49a, CD103) enriched in JEJ, and IL7RKITRORC ILC3 (ref. ) found in LN, spleen and JEJ (Fig. 2c,d). Putative CD127CD62LTCF7 NK/ILC precursors, resembling CD56CD16 NK and ILC3s, were enriched in blood but present in all tissues (Fig. 2c,d). B cells (272,162 cells) were classified into 6 subsets largely confined to lymphoid organs (Fig. 2e,f), including IgD naive B cells, CD27 memory B cells, germinal center B cells expressing AICDA (encoding activation-induced adenosine deaminase, which mediates somatic mutation and class switch recombination), plasma cells expressing immunoglobulin genes and SDC1 (CD138) and plasmablasts expressing proliferation markers (MKI67, TOP2A) (Fig. 2f). CD11c memory B cells expressing TBX21 (encoding the transcription factor T-BET) resembled ‘atypical B cells’ and were found at low frequencies in spleen and BM (Fig. 2e,f). Variable frequencies of memory B cells in LN and spleen expressed CD69 (Fig. 2e,f), denoting tissue residency. Plasma cells expressing IgA were enriched in the JLP, while IgG plasmablasts were enriched in lymphoid organs (Fig. 2e,f). B cell receptor (BCR) analysis indicated that plasmablasts exhibited the highest clonal expansion across lymphoid sites, while memory B cells and plasma (but not naive) cells expressed mutated BCR (Supplementary Fig. 4). Myeloid lineage cells (225,268 cells) comprised C1QAMS4A7 macrophages, FCN1CD14 classical and FCN1FCGR3A non-classical monocytes and DC subsets, including CLEC9A DC1, CLEC10A DC2, CCR7 migratory DCs and CD123LILRA4 plasmacytoid DCs (pDCs) (Fig. 2g,h). Classical monocytes were most abundant in blood, while non-classical monocytes were found mainly in BM and lung (Fig. 2g,h). DCs were found at low frequency across all tissues; pDCs were enriched in the BM, migratory DCs in the LNs, while DC2s were found mainly in the LNs and JEJ (Fig. 2g,h). Macrophages were found predominantly in the lung and at low frequencies in the BM, spleen, LNs and JEJ (Fig. 2h). Immune cell subset composition for the lineages above in blood, lymphoid organs and mucosal sites was specific to the site and conserved across donors (Fig. 2i and Extended Data Fig. 2). We defined additional subset heterogeneity based on RNA expression, identifying proliferating cells across lineages and functional subsets for CD8 TEM cells and CD8 TEMRA cells, also with tissue-specific distribution (Supplementary Figs. 5 and 6). This comprehensive, annotated map of immune cells across tissues can serve as a reference for future analysis, and we provided pre-trained models for label transfer of our cell-type annotation in the popV framework (Supplementary Fig. 7). This subset analysis showed consistent, tissue-specific composition between sites. To understand the influence of tissue localization on gene expression, we performed a two-step differential expression (DE) analysis. First, we compared the major immune lineages (CD8 T cells, CD4 T cells, γδ/ΜΑΙΤ cells, myeloid cells, NK/ILC and B cells) within each site (for example, blood, BM, spleen, LN, lung, JEJ) versus the other sites using pseudobulked linear mixed modeling (LMM), controlling for covariates (for example, sex, CMV serostatus) (Fig. 3a and Supplementary Tables 4 and 5). We identified 13 clusters from significantly differentially expressed genes (DEGs) (C1–13) grouped by lineage and/or tissue (Fig. 3b and Supplementary Table 6). We then conducted a similar across-tissue DE analysis for each subset and evaluated whether these signatures were expressed by individual subsets within a lineage using gene set enrichment analysis (GSEA) (Fig. 3a). To visualize effect sizes, determine the contribution of compositional differences and identify tissue-specific signatures across lineages, we integrated this analysis with subset frequencies within a tissue, fold changes (FC) compared to the other sites and the average expression of gene clusters by subset (Fig. 3c–h, Extended Data Fig. 3, and Supplementary Tables 7 and 8).Fig. 3Differential expression analysis identifies tissue localization as a major determinant of immune cell identity across lineages.a, Schematic showing cell lineage-level and cell subset-level DE analysis using linear mixed models (evaluated by dreamlet) in which cell lineage-level analysis compares a lineage in one tissue (for example, B cells from JEJ) against the same lineage in all other tissues, and cell subset-level analysis compares an immune cell subset in one tissue (for example, plasma cells in JEJ) against the same immune cell subset in all other tissues, followed by pre-ranked GSEA to assess gene cluster enrichment in subsets across tissues. b, Heatmap of lineage-level DEGs (adjusted (adj.) P value (false discovery rate, FDR) < 0.05, log2(FC) > 1, average mean expression >2) showing z-score average gene expression across tissue–cell lineage combinations, labeled by gene cluster with selected genes. c–h, Evaluation of gene clusters in cell subset-level DE analysis for myeloid cells (c and d), T cells and NK/ILCs (e and f), and B cells (g and h), shown as bar plots (c, e and g) with normalized enrichment scores (NES) of indicated gene clusters in cell subset-level DE by pre-ranked GSEA and heatmaps (d, f and h) showing cell subset frequencies as a percentage of their lineage within each tissue, FC in subset frequency in indicated tissue versus other tissues and split violin plots showing average gene cluster expression in the indicated tissue versus other tissues. Plotted bars denote adj. P value (FDR) < 0.05; n.s., not significant. AMP, adenosine monophosphate; O-glyco., O-glycosylation; prot., protein; circ., circulating; unconv., unconventional. a, Schematic showing cell lineage-level and cell subset-level DE analysis using linear mixed models (evaluated by dreamlet) in which cell lineage-level analysis compares a lineage in one tissue (for example, B cells from JEJ) against the same lineage in all other tissues, and cell subset-level analysis compares an immune cell subset in one tissue (for example, plasma cells in JEJ) against the same immune cell subset in all other tissues, followed by pre-ranked GSEA to assess gene cluster enrichment in subsets across tissues. b, Heatmap of lineage-level DEGs (adjusted (adj.) P value (false discovery rate, FDR) < 0.05, log2(FC) > 1, average mean expression >2) showing z-score average gene expression across tissue–cell lineage combinations, labeled by gene cluster with selected genes. c–h, Evaluation of gene clusters in cell subset-level DE analysis for myeloid cells (c and d), T cells and NK/ILCs (e and f), and B cells (g and h), shown as bar plots (c, e and g) with normalized enrichment scores (NES) of indicated gene clusters in cell subset-level DE by pre-ranked GSEA and heatmaps (d, f and h) showing cell subset frequencies as a percentage of their lineage within each tissue, FC in subset frequency in indicated tissue versus other tissues and split violin plots showing average gene cluster expression in the indicated tissue versus other tissues. Plotted bars denote adj. P value (FDR) < 0.05; n.s., not significant. AMP, adenosine monophosphate; O-glyco., O-glycosylation; prot., protein; circ., circulating; unconv., unconventional. Myeloid cells from all tissues exhibited transcriptional profiles not expressed by other immune lineages (clusters C1–C4), which also varied by tissue (Fig. 3b). C1 (chemokine, complement, lipid transport) and C2 (PPAR signaling associated with alveolar macrophages) were enriched in the JEJ and lung, respectively, particularly in macrophages (Fig. 3c,d). By contrast, C4 (anti-microbial peptide production and cell signaling) enrichment was explained by increased frequencies of BM and spleen monocytes relative to macrophages rather than transcriptional changes within and across subsets (Extended Data Fig. 3a,b). Tissue-associated genes within clusters C5–C9 were expressed primarily by T cells and innate lymphocytes (Fig. 3b). C5 genes encoding stem-like transcription factors and markers (TCF7, LEF1, ITGA6) and LN homing receptors (CCR7, SELL) were enriched in CD4 and CD8 TRM cells and TEM cells in LN (Fig. 3e,f). Conversely, C6 genes encoding molecules for mucosal residency (ITGA1, CXCR6, ITGAE) and gut homing (CCR9) were enriched in CD4 and CD8 TRM cells, γδ T cells, MAIT cells and NK cells in the JEJ (Fig. 3e,f). CD8 TEMRA cells, γδ cells, MAIT cells and NK/ILCs in the JEJ showed reduced expression of C9 genes associated with cytolytic effector function (GZMB, PRF1, IFNG, NKG7) compared to other sites, while C9 was enriched in the lung (Fig. 3b and Extended Data Fig. 3c,d). C7 genes associated with protein metabolism were enriched in NK/ILCs and innate T cells in the JEJ, while C8 genes encoding chemokines (CCL4, XCL1) and innate lymphocyte functions (KLRC1, NCR1) were expressed by T cell and NK/ILC subsets across all sites (Fig. 3b and Extended Data Fig. 3e,f). Therefore, this analysis identified shared gene expression profiles across T and innate lymphocyte subsets that varied by site. Tissue-enriched signatures for B cells (C10–C13) were mostly associated with specific subsets (Fig. 3b,g,h and Extended Data Fig. 3g,h). C10 included cell cycle genes and was enriched in B cells and plasma cells in the lungs (Fig. 3g,h), possibly because of higher plasmablast abundance. JEJ B cells showed increased expression of C12 (B cell differentiation and protein transport) and C13 (B cell activation) genes (Fig. 3b), which were derived from plasma cells comprising the majority of the B cell lineage in intestinal sites (Fig. 3g,h). We also applied consensus single-cell hierarchical Poisson factorization (scHPF) to identify gene co-expression patterns common to different immune lineages or sites (considering the JEJ, lung and LN) (Extended Data Fig. 4a and Supplementary Table 9). We found a proliferation module enriched in the lungs, a lymphoid-specific module in the LNs, an intestine-specific residency module and modules associated with effector and cytolytic functions prominent in the lungs but not JEJ or LNs (Extended Data Fig. 4a–d). The lymphoid module included genes associated with stemness (KLF7, LEF1, SOX4) and lymphoid homing (CCR7, SELL, ITGA6) markers expressed mostly in T cells, some NK cells and ILCs, and not in B cells or myeloid cells (Extended Data Fig. 4b–e and Supplementary Table 10). The JEJ residency module included intestinal tissue residency genes (CCR9, ITGAE, CD101, CD160) enriched in T cells, NK cells and ILCs (Extended Data Fig. 4f–h). These signatures were also reflected at the surface protein level (Extended Data Fig. 4e,i and Supplementary Table 11). Thus, tissue-specific gene expression modules spanned multiple cell types, suggesting shared tissue adaptations. The tissue environment poses unique requirements for resident immune populations such as TRM cells, plasma cells and macrophages for maintaining homeostasis. We integrated the DE analysis shown in Fig. 3 with surface protein expression to define site-specific signatures for resident immune cells (Fig. 4 and Supplementary Table 12). CD4 and CD8 TRM cells expressed genes and/or surface proteins for tissue residency (CD103, CD101, CD49a), gut homing and localization (CCR9, CCR5 for CD4 TRM cells) and reduced PD-1 (for CD8 TRM cells) in intestines relative to lungs and lymphoid organs (Fig. 4a–f). Lung TRM cells showed increased expression of effector or cytotoxicity (IFNG, GZMH, GZMA, PRF1) and regulatory (CTLA4) genes relative to the JEJ and lymphoid organs; TRM cells in LN, spleen and BM had higher expression of stem-like markers (TCF7, KLF2) and certain integrins and costimulatory markers (ITGB2, CD27, CD28, ICOS) relative to the JEJ and lungs (Fig. 4a–f). These site-specific profiles for TRM cells showed adaptations related to migration, localization and function.Fig. 4Cross-tissue analysis of resident immune cell subsets highlights site-specific signatures.a–l, Expression of tissue-enriched genes from cell subset-level DE analysis and corresponding surface proteins for CD4 TRM cells (a–c), CD8 TRM cells (d–f), plasma cells (g–i) and macrophages (j–l), shown as dot plots with the percentage of cells expressing selected genes enriched in one tissue versus the other sites (a, d, g and j) or surface protein expression (c, f, i and l) and heatmaps showing median log2(FC) surface marker expression across tissues (b, e, h and k). Dot size represents the frequency of gene expression in group (a, d, g and j) or frequency (% positive) (c, f, i and l); color (red, blue) intensity indicates average gene expression. a–l, Expression of tissue-enriched genes from cell subset-level DE analysis and corresponding surface proteins for CD4 TRM cells (a–c), CD8 TRM cells (d–f), plasma cells (g–i) and macrophages (j–l), shown as dot plots with the percentage of cells expressing selected genes enriched in one tissue versus the other sites (a, d, g and j) or surface protein expression (c, f, i and l) and heatmaps showing median log2(FC) surface marker expression across tissues (b, e, h and k). Dot size represents the frequency of gene expression in group (a, d, g and j) or frequency (% positive) (c, f, i and l); color (red, blue) intensity indicates average gene expression. Tissue plasma cells and macrophages also exhibited distinct tissue signatures. For plasma cells, we found several JEJ-enriched genes, including IGHA2 (consistent with predominant IgA plasma cells in the gut), the non-classical HLA molecule CD1d, plasma cell transcription factors (RUNX2, ID3) and the tissue residency marker CD69 (Fig. 4g–i). Plasma cells in lungs, and to a lesser extent, LN and spleen, expressed higher levels of integrins (CD11c, CD18) (Fig. 4g–i). JEJ macrophages had increased expression of integrins (ICAM1, ITGA4), chemokines (CCL24, CCL3L1) and regulators of macrophage activation (for example, SLAMF7), whereas lung macrophages had higher expression of CD11c (ITGAX), markers of efferocytosis (MRC1, MARCO) and lipid metabolism (PPARG, TREM2, FABP4) (Fig. 4j–l), consistent with alveolar macrophages. Lastly, macrophages in the spleen expressed markers of red pulp macrophages (SPIC, VCAM1) (Fig. 4j). These observations revealed site-specific signatures for activation, migration, metabolism and cell–cell interactions involved in tissue residency. We investigated age-associated effects across immune lineages and tissues. A global analysis of transcriptional variance within each major lineage revealed that tissue site explained the majority of variance, while age accounted for a much smaller fraction (Fig. 5a and Supplementary Table 13). Most of the top age-dependent genes in each lineage (Fig. 5b) also exhibited tissue-specific variation. Immune cell composition in the different sites was largely maintained across age, except for significantly decreased frequencies of CD8 TN cells in the blood and LNs, concomitant increases in TEM cells in the blood and TRM cells in the LNs and lower frequencies of classical monocytes in the BM (Fig. 5c and Supplementary Table 14). These results indicate that tissue-driven immune cell composition and profile are largely maintained with age.Fig. 5Immune aging manifests across lineage, subset and tissue of origin.a, Variance decomposition analysis (dreamlet), with box plots showing percentage of variance in gene expression explained by age (<40 or >40 years old), CMV serostatus (positive or negative), sex (male or female), processing site (US or UK), 10× Genomics chemistry (5′ or 3′) and tissue group (BLO, BM, SPL, LN, LNG or JEJ) for CD4 T cells, CD8 T cells, B cells, myeloid cells and NK/ILCs. Box plots show the median (center), interquartile range (IQR; box) and whiskers extending to 1.5× IQR. b, Variance decomposition analysis with stacked bar plots showing selected genes from the top (up with age) and bottom (down with age) 50 DEGs by cell lineage-level linear mixed model (dreamlet) across age for CD4 T cells, CD8 T cells, B cells, myeloid cells and NK/ILCs. c, Heatmaps of change in subset composition by age (<40 or >40 years old) across BLO, BM, SPL, LN, LNG and JEJ, shown as log2(FC) calculated by generalized linear model (two-sided Wald test). Gray denotes insufficient cell numbers for comparison. d, t-SNE of immune cell subsets based on age-associated DEGs (log2(FC) > 0.1, unadjusted (unadj.) P < 0.05) by subset-level linear mixed model (dreamlet). e–g, Bar-and-dot plots of top DEGs by age (adj. P value (FDR) < 0.05 for at least 1 tissue; among top 50) for classical monocytes in BLO, BM, LNG and SPL (e), non-classical monocytes in BM, LNG and SPL (f), and macrophages in JEJ, LN, LNG and SPL (g). Solid bars show median log2(FC) across tissues; error bars, 95% CI. Statistically significant genes are indicated by circles outlined in black. h, Box plot showing the percentage of CD95 macrophages in the lung in donors <40 or >40 years old. Statistical significance determined by generalized linear model (two-sided Wald test). i, Consensus scHPF macrophage APOE–TREM2 signature with dot plots showing gene rank (left) and raw and covariate-adjusted (partial reg.) cell scores by linear mixed model in individuals <40 or >40 years old (right). j, Heatmap of enrichment of top 200 factor genes in cell subset-level DE by GSEA in cMono, ncMono and macrophages in BLO, BM, SPL, LN, LNG and JEJ. k, APOE–TREM2 signature cell scores from a human lung dataset (two-sided Wilcoxon rank-sum test) (left) and enrichment of the factor in macrophage DE by GSEA across age (right). For all panels, q denotes adj. P value (FDR); *adj. P value (FDR) < 0.05. a, Variance decomposition analysis (dreamlet), with box plots showing percentage of variance in gene expression explained by age (<40 or >40 years old), CMV serostatus (positive or negative), sex (male or female), processing site (US or UK), 10× Genomics chemistry (5′ or 3′) and tissue group (BLO, BM, SPL, LN, LNG or JEJ) for CD4 T cells, CD8 T cells, B cells, myeloid cells and NK/ILCs. Box plots show the median (center), interquartile range (IQR; box) and whiskers extending to 1.5× IQR. b, Variance decomposition analysis with stacked bar plots showing selected genes from the top (up with age) and bottom (down with age) 50 DEGs by cell lineage-level linear mixed model (dreamlet) across age for CD4 T cells, CD8 T cells, B cells, myeloid cells and NK/ILCs. c, Heatmaps of change in subset composition by age (<40 or >40 years old) across BLO, BM, SPL, LN, LNG and JEJ, shown as log2(FC) calculated by generalized linear model (two-sided Wald test). Gray denotes insufficient cell numbers for comparison. d, t-SNE of immune cell subsets based on age-associated DEGs (log2(FC) > 0.1, unadjusted (unadj.) P < 0.05) by subset-level linear mixed model (dreamlet). e–g, Bar-and-dot plots of top DEGs by age (adj. P value (FDR) < 0.05 for at least 1 tissue; among top 50) for classical monocytes in BLO, BM, LNG and SPL (e), non-classical monocytes in BM, LNG and SPL (f), and macrophages in JEJ, LN, LNG and SPL (g). Solid bars show median log2(FC) across tissues; error bars, 95% CI. Statistically significant genes are indicated by circles outlined in black. h, Box plot showing the percentage of CD95 macrophages in the lung in donors <40 or >40 years old. Statistical significance determined by generalized linear model (two-sided Wald test). i, Consensus scHPF macrophage APOE–TREM2 signature with dot plots showing gene rank (left) and raw and covariate-adjusted (partial reg.) cell scores by linear mixed model in individuals <40 or >40 years old (right). j, Heatmap of enrichment of top 200 factor genes in cell subset-level DE by GSEA in cMono, ncMono and macrophages in BLO, BM, SPL, LN, LNG and JEJ. k, APOE–TREM2 signature cell scores from a human lung dataset (two-sided Wilcoxon rank-sum test) (left) and enrichment of the factor in macrophage DE by GSEA across age (right). For all panels, q denotes adj. P value (FDR); *adj. P value (FDR) < 0.05. To interrogate specific effects of aging on immune cells across lineages and sites, we conducted a separate DE analysis for each tissue group and immune subset with sufficient representation, comparing younger (<40 years) versus older (>40 years) individuals, while controlling for sex, CMV status and other covariates (Supplementary Tables 15–17). An embedding of similarities between age-related DEGs revealed that some immune cells exhibited changes specific to tissue site (for example, T cells in the LN, lymphocytes in the JEJ), while others showed subset-specific changes independent of site (for example, monocytes in blood, BM and lung, and memory B cells in LNs and lung) (Fig. 5d). We identified genes regulated with age in each subset and tissue by integrating DE results with GSEA (Supplementary Fig. 8a). For myeloid lineage cells, mucosal sites had the most age-related DEGs, with similar trends in lymphoid organs (Fig. 5e–g). Age-related changes in classical monocytes included increased expression of genes associated with proliferation and inflammation (KRAS, CALM1) and decreased expression of genes for macrophage differentiation (for example, RAB44) (Fig. 5e); non-classical monocytes showed age-associated increase in expression of genes for cell–cell interactions (LGALS1, ITGA9) and decreased expression of metabolism and mitochondrial regulation genes (LYRM7, SARS2) (Fig. 5f). Macrophages had decreased expression of genes associated with metabolism and mitochondrial fitness (MARC1, MTOR) and increased expression of genes associated with M2 macrophages (ID1, ADGRE1 NPR2) and interferon signaling (MX2) (Fig. 5g) along with increased CD95 (Fas) at the protein level (Fig. 5h). Together, these age-related changes in monocytes and macrophages were subset-specific, enriched in mucosal sites and indicated decreased overall fitness. We used scHPF to identify age-associated gene signatures for myeloid cells and GSEA to assess their expression in different subsets and sites (Supplementary Tables 18 and 19). We uncovered an APOE–TREM2 signature, including apolipoprotein genes (APOC1, APOC2 and APOE) and TREM2, a triggering receptor expressed on myeloid cells that binds ApoE and facilitates macrophage functions, such as phagocytosis and chemotaxis, and induces metabolic changes (Fig. 5i). This APOE–TREM2 signature was significantly downregulated with age in monocytes and macrophages in the lungs, lymphoid organs and blood (Fig. 5j), and in an independent published dataset from human lungs (n = 29; see Methods) (Fig. 5k). Overall, this analysis showed subset and site-specific features of macrophage aging involving a major functional and metabolic pathway. We applied the above approaches to identify age-associated signatures in T cells and B cells. For CD4 T cells, CD8 T cells and B cells, there were relatively few genes (for example, those associated with oxidation and inflammation: SOD1, IL18BP, IL15) that changed over age in two or more subsets within each lineage (Extended Data Fig. 5a–d). Pathway analysis revealed increased inflammation, apoptosis and reduced TCR signaling across multiple T cell subsets with age (Supplementary Fig. 8b). Despite the paucity of age-associated gene expression changes across subsets, CD8 TEMRA cells exhibited multiple age-associated changes conserved across sites; these included increased expression of NK cell genes (NCAM1, KLRF1, GNLY) and the NK cell marker CD56, consistent with findings in blood, and reduced expression of genes associated with signaling (CD6, JAK3), proliferation and metabolism (TCF7, RPTOR) (Fig. 6a,b).Fig. 6Immune aging in adaptive lymphocytes varies across tissues.a, Bar-and-dot plots of DEGs from individuals <40 or >40 years old in CD8 TEMRA cells in BLO, BM, SPL, LN and LNG by linear mixed model (dreamlet). Bars show median log2(FC) across tissues; error bars, 95% CI; dot color indicates tissue, with dots for significant genes (adj. P value (FDR) < 0.05) encircled in black. b, Box plot (center, median; box, IQR; whiskers, 1.5× IQR) showing percentage of CD8 TEMRA cells expressing CD56 in the BM of donors <40 or >40 years old. Statistical significance determined by generalized linear model (two-sided Wald test). c, Consensus scHPF CD8 T cell cytokine/chemokine signature, with dot plots showing gene rank (left) and raw and covariate-adjusted (partial reg.) cell scores by linear mixed model in donors <40 or >40 years old (right). d, Heatmap of enrichment of top 200 genes in the cytokine/chemokine signature in cell subset-level DE by GSEA in CD8 T cell subsets from BLO, BM, SPL, LN, LNG and JEJ. e, CD8 cell cytokine/chemokine signature cell scores in CD8 T cells from a human lung dataset in donors <40 or >40 years old (two-sided Wilcoxon rank-sum test) (left) and enrichment of factor in CD8 T cells DE across age by GSEA (right). f, Consensus scHPF CD8 T cell GZMK signature with dot plots showing gene rank (left) and raw and covariate-adjusted (partial reg.) cell scores by linear mixed model in donors <40 or >40 years old (right). g, Heatmap of enrichment of top 200 genes from the GZMK signature in cell subset-level DE by GSEA in subsets of CD8 T cells in BLO, BM, SPL, LN, LNG and JEJ. h, Violin plot showing frequency of GZMKCD8 TEMRA cells in BM, SPL and BLO in donors <40 or >40 years old. Statistical significance determined by two-sided Wilcoxon rank-sum test. i, Scatterplot of clonality score (1 − Pielou’s evenness index) for CD8 TEMRA cells with >70% GZMB, >70% GZMK or a mix of GZMB and GZMK expression in BM, SPL and BLO of CMV or CMV donors, with statistical significance determined by linear model. j, Bar-and-dot plot of DEGs in donors <40 or >40 years old in memory B cells as in a. k, Box plot (center, median; box, IQR; whiskers, 1.5× IQR) showing frequency of CXCR3 and IgM memory B cells in the LN of donors <40 or >40 years old. Statistical significance determined by generalized linear model (two-sided Wald test). l, Box plot of BCR showing the frequency of IgA, IgM, IgG or IgD memory B cells in LN of donors <40 or >40 years old with statistical significance determined by generalized linear model (two-sided Wald test). m, Consensus scHPF revealing RAS signature in memory B cells, with dot plots showing gene rank (left) and raw and covariate-adjusted (partial reg.) cell scores by linear mixed model in donors <40 or >40 years old (right). n, Heatmap of enrichment of top 200 genes in the RAS signature in cell subset-level DE by GSEA in naive B cells, memory B cells, plasmablasts and plasma cells in BM, SPL, LN, LNG and JEJ. o, B cell RAS signature cell score in B cells from a human BM atlas in donors <40 or >40 years old (left) and enrichment of factor in B cells DE across age by GSEA (right). For all panels, q denotes adj. P value (FDR); *adj. P value (FDR) < 0.05. a, Bar-and-dot plots of DEGs from individuals <40 or >40 years old in CD8 TEMRA cells in BLO, BM, SPL, LN and LNG by linear mixed model (dreamlet). Bars show median log2(FC) across tissues; error bars, 95% CI; dot color indicates tissue, with dots for significant genes (adj. P value (FDR) < 0.05) encircled in black. b, Box plot (center, median; box, IQR; whiskers, 1.5× IQR) showing percentage of CD8 TEMRA cells expressing CD56 in the BM of donors <40 or >40 years old. Statistical significance determined by generalized linear model (two-sided Wald test). c, Consensus scHPF CD8 T cell cytokine/chemokine signature, with dot plots showing gene rank (left) and raw and covariate-adjusted (partial reg.) cell scores by linear mixed model in donors <40 or >40 years old (right). d, Heatmap of enrichment of top 200 genes in the cytokine/chemokine signature in cell subset-level DE by GSEA in CD8 T cell subsets from BLO, BM, SPL, LN, LNG and JEJ. e, CD8 cell cytokine/chemokine signature cell scores in CD8 T cells from a human lung dataset in donors <40 or >40 years old (two-sided Wilcoxon rank-sum test) (left) and enrichment of factor in CD8 T cells DE across age by GSEA (right). f, Consensus scHPF CD8 T cell GZMK signature with dot plots showing gene rank (left) and raw and covariate-adjusted (partial reg.) cell scores by linear mixed model in donors <40 or >40 years old (right). g, Heatmap of enrichment of top 200 genes from the GZMK signature in cell subset-level DE by GSEA in subsets of CD8 T cells in BLO, BM, SPL, LN, LNG and JEJ. h, Violin plot showing frequency of GZMKCD8 TEMRA cells in BM, SPL and BLO in donors <40 or >40 years old. Statistical significance determined by two-sided Wilcoxon rank-sum test. i, Scatterplot of clonality score (1 − Pielou’s evenness index) for CD8 TEMRA cells with >70% GZMB, >70% GZMK or a mix of GZMB and GZMK expression in BM, SPL and BLO of CMV or CMV donors, with statistical significance determined by linear model. j, Bar-and-dot plot of DEGs in donors <40 or >40 years old in memory B cells as in a. k, Box plot (center, median; box, IQR; whiskers, 1.5× IQR) showing frequency of CXCR3 and IgM memory B cells in the LN of donors <40 or >40 years old. Statistical significance determined by generalized linear model (two-sided Wald test). l, Box plot of BCR showing the frequency of IgA, IgM, IgG or IgD memory B cells in LN of donors <40 or >40 years old with statistical significance determined by generalized linear model (two-sided Wald test). m, Consensus scHPF revealing RAS signature in memory B cells, with dot plots showing gene rank (left) and raw and covariate-adjusted (partial reg.) cell scores by linear mixed model in donors <40 or >40 years old (right). n, Heatmap of enrichment of top 200 genes in the RAS signature in cell subset-level DE by GSEA in naive B cells, memory B cells, plasmablasts and plasma cells in BM, SPL, LN, LNG and JEJ. o, B cell RAS signature cell score in B cells from a human BM atlas in donors <40 or >40 years old (left) and enrichment of factor in B cells DE across age by GSEA (right). For all panels, q denotes adj. P value (FDR); *adj. P value (FDR) < 0.05. By scHPF analysis, we identified two prominent age-associated transcriptional signatures shared across multiple CD8 T cell subsets. A cytokine signature, containing genes for effector cytokines and chemokines (for example, CCL3, CCL4, XCL1, IFNG, TNF), was increased with age in all CD8 T cell subsets and γδ T cells across all sites examined (Fig. 6c,d). This aging signature was also detected in published datasets from human lungs (Fig. 6e) and in peripheral blood mononuclear cells (PBMCs) from the Sound Life cohort (n = 96, age 26–65 years) within the Human Immune Health Atlas (Extended Data Fig. 5e). A second signature contained GZMK encoding the cytolytic molecule granzyme K, the transcription factor EOMES and activation or signaling markers (PDCD1, HLA-DR, FCRL3) (Fig. 6f). This signature aligned with a granzyme K-containing, age-associated signature identified in T cells in aged mice and human blood and in PBMCs from the Human Immune Health Atlas (Extended Data Fig. 5f). This GZMK signature was increased with age in CD8 TEMRA cells, γδ T cells and CD8 TEM cells across tissues but not in CD8 TRM cells in lungs and JEJ (Fig. 6g). The frequency and clonal expansion of CD8 TEMRA cells enriched in the GZMK signature was higher in older than in younger donors (Fig. 6h,i). These results show distinct age-associated signatures in mucosal resident T cells compared to circulation. For B cells, there were more age-related DEGs in memory compared to naive B cells in LNs (Fig. 6j and Extended Data Fig. 5d). Memory B cells from older donors had increased expression of genes associated with inflammatory cytokines (IL18, IL18BP), cell adhesion (CD58, LGALS1) and cell death or autophagy (FAS, ITM2A), along with reduced expression of proliferation (CDCA7, IL2RA), lipid metabolism (ACSM3, PNPLA7) and differentiation markers (Fig. 6j). Select IL-18 pathway-associated genes in B cells and CD4 T cells in blood were validated in the Human Immune Health Atlas cohort (Supplementary Fig. 9a,b). Transcript and/or surface expression of IgM (IGHM) and IGHD were reduced in older compared to younger B cells, and the frequency of IgM B cells decreased with age (Fig. 6k,l and Supplementary Table 20). We identified by scHPF that a gene signature related to RAS signaling (RASA4B, RASGRF1, GAB2) downstream of the BCR was downregulated with age in naive and memory B cells across all sites (Fig. 6m,n). We validated this age-associated signature in BM-derived B cells from an independent dataset (n = 39, age 2–84 years) and in PBMCs from the Human Immune Health Atlas (Fig. 6o and Extended Data Fig. 5g). Pathway analysis further revealed increased inflammation and reduced BCR signaling in LN B cells with age (Supplementary Fig. 8c). These results showed that B cells exhibited diminished signaling and functional dysregulation across tissues over age. CMV infection drives immune cell alterations, including increased accumulation of TEMRA cells with age in blood, spleen and lungs. We investigated the impact of CMV serostatus on cell composition and immune aging and found no significant associations with CD4 T cell, CD8 T cell or B cell frequencies (Supplementary Fig. 10a,b). Two CD8 T cell signatures were associated with CMV serostatus after regression of other covariates: the GZMK signature and a GNLY signature (GNLY, FGFBP2 and CX3CR1) (Supplementary Fig. 10c–f). The GNLY signature was enriched across all CD8 T cell subsets in CMV donors, while the GZMK signature was variably enriched in different sites and subsets of CMV donors by GSEA (Supplementary Fig. 10d,f and Supplementary Tables 21 and 22). Therefore, CMV infection drives T cell gene signature changes that overlap with, but are distinct from, age-related immune alterations. CD4 T cells are highly heterogeneous and exhibit functional and phenotypic continuums, suggesting that age effects could differentially manifest within or across subsets. We applied an annotation-independent analysis of aging in CD4 T cells in the lung, JEJ and LN, leveraging a per-cell estimation of age effects using counterfactual analysis with MrVI, which separately considers each cell and controls for covariates. This analysis identified groups of cells with similar predicted age-associated changes in gene expression (‘modules’), which we interrogated by DE analysis across age (Fig. 7, Extended Data Fig. 6, and Supplementary Tables 23 and 24). In the lung, a fraction of CD4 T cells (~25%, comprising TEM cells, TRM cells and TEMRA cells) exhibited decreased cytotoxicity (GZMH, GNLY, GZMA) and increased cytokine receptor (IL18R1, IFNGR1) genes with age (Fig. 7a–c and Extended Data Fig. 6a,b). Similar upregulation of cytokine responsiveness with age occurred in some CD4 T cells in blood, BM, LN and spleen, while decreased cytotoxicity was unique to the lungs (Fig. 7c,d and Extended Data Fig. 6b). CD4 T cells in the JEJ (mainly TRM cells) exhibited an age-related decline in TH17-associated genes (IL17A, IL17F, IL22, RORC, CCR6) and increase in pro-inflammatory cytokines (IFNG, TNF) (Fig. 7e–g and Extended Data Fig. 6c,d); age-associated downregulation of IL-17-associated genes was also observed in CD4 T cells in lung and blood (Fig. 7g–h and Extended Data Fig. 6d). CD4 T cells in the LN (also in spleen and lungs) exhibited reduced expression of genes associated with regulation (IL10, TIGIT, CTLA4, CD27) and increased expression of inflammation and activation markers (IL18BP, TNFRSF4, TNF) with age (Fig. 7i–l and Extended Data Fig. 6e,f). These results revealed age-associated transcriptional changes in tissue CD4 T cells associated with site-specific functions.Fig. 7Integrated analysis reveals tissue-dependent signatures of aging in CD4 T cells.a–l, MrVI analysis shows modules of genes changing over age for CD4 T cells in the lungs (a–d), gut (e–h) and LNs (i–l), depicted as UMAP embeddings (a, e and i), colored by subset (left) and tissue module score (right). Age-associated DEGs identified by linear mixed model (dreamlet) in donors <40 or >40 years old are shown in volcano plots for module-positive cells in each tissue (b, f and j) and box-and-dot plots (c, g and k) in BLO, BM, LN, LNG and SPL (c), BLO, LN, LNG and JEJ (g), and BM, LN, LNG and SPL (k), and heatmaps (d, h and l) showing enrichment of manually selected module genes across BLO BM, LN, LNG and SPL (d), BLO, LN, LNG and JEJ (h), and BM, LN, LNG and SPL (l) in tissue-specific DE of module-positive cells by age using GSEA. In a, e and i, module score is computed as the sum of inferred log2(FC) for genes upregulated in donors >40 years old minus the sum of log2(FC) for genes downregulated in individuals >40 year old. In b, f and j, selected genes with high effect size (by estimated log2(FC), unadj. P < 0.05) are labeled. Bars show median log2(FC) across tissues; dot color indicates tissue with trending DEGs (unadj. P < 0.05) outlined in black. In c, g and k, genes with known effector or activation functions are shown, selected from the module or trending DEG list. Tissues are included only if sufficient signature-positive cells were detected. Separate analyses (columns) were conducted for upregulated and downregulated genes. *Unadj. P < 0.1. a–l, MrVI analysis shows modules of genes changing over age for CD4 T cells in the lungs (a–d), gut (e–h) and LNs (i–l), depicted as UMAP embeddings (a, e and i), colored by subset (left) and tissue module score (right). Age-associated DEGs identified by linear mixed model (dreamlet) in donors <40 or >40 years old are shown in volcano plots for module-positive cells in each tissue (b, f and j) and box-and-dot plots (c, g and k) in BLO, BM, LN, LNG and SPL (c), BLO, LN, LNG and JEJ (g), and BM, LN, LNG and SPL (k), and heatmaps (d, h and l) showing enrichment of manually selected module genes across BLO BM, LN, LNG and SPL (d), BLO, LN, LNG and JEJ (h), and BM, LN, LNG and SPL (l) in tissue-specific DE of module-positive cells by age using GSEA. In a, e and i, module score is computed as the sum of inferred log2(FC) for genes upregulated in donors >40 years old minus the sum of log2(FC) for genes downregulated in individuals >40 year old. In b, f and j, selected genes with high effect size (by estimated log2(FC), unadj. P < 0.05) are labeled. Bars show median log2(FC) across tissues; dot color indicates tissue with trending DEGs (unadj. P < 0.05) outlined in black. In c, g and k, genes with known effector or activation functions are shown, selected from the module or trending DEG list. Tissues are included only if sufficient signature-positive cells were detected. Separate analyses (columns) were conducted for upregulated and downregulated genes. *Unadj. P < 0.1. We present a comprehensive analysis of the human immune system across tissues and ages through multimodal profiling of blood and tissues from organ donors spanning six decades of adult life. We found that tissue localization was a dominant driver of the immune landscape, determining immune cell composition, cell states and functional capacity. With age, these tissue-specific properties were largely maintained, although certain subsets and sites showed altered function, migration and regulation. Our results reveal that the human immune system is highly specialized for diverse tissue environments to maintain homeostasis and mount effective immune responses. We demonstrated that each tissue imposed site-specific immune cell compositions and adaptations that varied by lineage, and these tissue effects were conserved across donors. Although we realized and reinforced site-specific features for TRM cells at barrier sites and lymphoid organs, whether these adaptations applied to other immune cells remained unknown. Here, we found that site-specific signatures for T cells in the gut (high tissue residency, low cytotoxicity), lungs (high effector function, increased regulation) and lymphoid organs (stem-like features) were not exclusive to the canonical resident populations, were shared across NK cell and ILC subsets and were absent from B cell and myeloid lineages. The enhanced expression of stem-like transcription factors TCF-1 and LEF-1 by LN memory T cells suggests that lymphoid organs may serve as reservoirs for long-lived memory cells, given these factors’ requirement for memory T cell generation. Macrophages and plasma cells also exhibited site-specific features in the gut, lungs and lymphoid organs through distinct subset-specific pathways, such as alveolar macrophages in the lungs and red pulp macrophages in the spleen. These lineage-dependent tissue adaptations probably reflect niche localization and interactions with distinct structural and immune cells within each tissue. Age-associated gene signatures identified for macrophages, T cells and B cells were intrinsic to the subset and site. The APOE–TREM2 gene signature, essential for crucial macrophage functions, was reduced with age by lung macrophages. APOE–TREM2 expression in microglia is associated with neurodegeneration in Alzheimer’s disease and in other macrophage types with cardiovascular diseases. TREM2 can have different effects on macrophage functions; promoting anti-inflammatory ‘M2-like’ function in some contexts and phagocytosis and sustained inflammation in others. The age-associated loss of TREM2 in lung macrophages could thus account for compromised immunity to respiratory pathogens and increased lung cancer susceptibility observed in the aging population. TREM agonists that enhance phagocytic function are being tested in clinical trials in Alzheimer’s disease and could be considered in the rejuvenation of aging macrophages in other sites. Other age-associated features were specific to lymphocyte lineages. T cells in circulation expressed higher levels of genes associated with inflammation, cytotoxicity and NK-like markers with age, as previously reported. Circulating TEMRA cells and TEM cells upregulated GZMK and other markers, similar to senescent GzmKCD8 T cells found in mice and human blood. TRM cells in the lungs and intestines did not exhibit this age-associated gene signature, suggesting that the tissue environment may insulate them from signals that promote cellular aging or that cellular aging is tissue-specific. However, both circulating (TEM, TEMRA) and TRM cells had increased expression of genes for pro-inflammatory cytokines and chemokines with age, consistent with inflammaging implicated in cardiovascular diseases and metabolic dysregulation. Our findings suggest that human T cells may be more prone to innate functions such as cytokine-driven activation (for example, via IL-18) with age. We also identified an age-associated increase in IL-18 expression and reduced BCR-mediated signaling within tissue B cells, which is a feature of NK-like B cell subsets identified in disease contexts. Thus, aging may reflect a broader age-related shift to innate-like functions in both T cells and B cells. Our findings have important implications for immune monitoring, therapeutic modulation and clinical advancement. The compartmentalization of immune subsets across tissues emphasizes the importance of site-specific immune monitoring in disease states, as exemplified in severe COVID-19, in which immune dynamics in the respiratory tract rather than blood correlated with infection outcome. The distinctness of gut-specific subsets provides rationale for targeted intestinal interventions, as demonstrated by rotavirus vaccines. The identification of stem-like profiles (marked by TCF7 and LEF1 expression) in LN T cells and NK cells has direct relevance to adoptive CAR-T immunotherapies, in which stemness is associated with remission. LNs may thus represent an optimal source of NK and T cells for engineering adoptive cell therapies against cancer, infections and autoimmunity. Our study has several limitations. The low frequency of certain immune subsets in tissues, including DCs, macrophages in lymphoid organs and hematopoietic progenitors, precluded aging analysis and will require sorting for future studies. Similarly, an in-depth analysis of TCR and BCR across sites and age would require isolating memory T cells and B cells from each site. Finally, although we identified age-associated changes in 24 donors, additional donors would increase power and probably reveal additional aging signatures. In conclusion, this dataset, along with the models and analyses presented, can serve as a valuable and actionable resource, informing targeted immune modulation by site and age in future treatments for infectious, neoplastic and inflammatory diseases. This study uses samples obtained from deceased organ donors and does not qualify as human subjects research in the USA, given that the donors are deceased and not living, as confirmed by the Columbia University Institutional Review Board. In the UK, samples were collected and analysed under an ethically approved research protocol (REC 15/EE/0152). The study analyzed immune cells from multiple tissue samples obtained from 24 organ donors. No statistical method was used to predetermine sample size. No data were excluded from the analysis. Investigators were not blinded to allocation during experiments and outcome assessment, as this is a profiling study. Tissues were obtained from deceased organ donors (Supplementary Table 1) at the time of organ acquisition for clinical transplantation. In the USA, this was done through an approved protocol and material transfer agreement via LiveOnNY, the organ procurement organization for the New York metropolitan area. In the UK, tissues were obtained through the Cambridge Biorepository for Translational Medicine (CBTM), REC 15/EE/0152, as previously described. Owing to the different amounts of tissues and some distinct samples (for example, skin, liver and colon) obtained at each location, protocols for processing may differ, as described below. Each tissue was subjected to a tissue-specific protocol to maximize MNC recovery and viability across a diversity of sites. Detailed, step-by-step protocols for immune cell isolation from blood, BM, spleen, LNs, lungs (parenchyma and airway or BAL) and JEJ (JLP and JEL) are presented elsewhere. All single-cell suspensions from each site were centrifuged (400g, 10 min at 4 °C) and washed twice with PBS containing 5% (v/v) FBS and 2 mM EDTA. Cells were counted using the NC-2000 Cell Counter (Chemometec), and 50 million viable cells from each site were treated with TruStain FcX (BioLegend) and FcR Blocking Reagent (Miltenyi). Cells were subsequently labeled for 30 min at 4 °C with biotinylated anti-CD66B, anti-CD235ab and anti-CD326 to remove granulocytes, red blood cells and epithelial cells, respectively, by streptavidin-coated magnetic particles and negative selection (Bangs Laboratories). All single-cell suspensions were subjected to dead cell removal using a Dead Cell Removal Kit (Miltenyi). Each single-cell suspension was hashtagged to allow pooling of samples for loading on the 10× Genomics Chromium instrument. MNCs from each site (10 per site) were transferred into 4 ml flow cytometry tubes, pelleted by centrifugation as above and resuspended in PBS containing 5% (v/v) FBS and 2 mM EDTA and then incubated with TruStain FcX (BioLegend) and FcR Blocking Reagent (Miltenyi) at 4 °C for 10 min to reduce background labeling. Each hashtag was spun at 14,000g for 10 min, added to each sample (1 µl hashtag per tube), incubated at 4 °C for 30 min, pelleted and washed 3 times with PBS containing 5% (v/v) FBS and 2 mM EDTA. For CITE-seq antibody staining, 200,000 cells from each sample were resuspended in reconstituted TotalSeq-A Universal Cocktail (BioLegend) (donors 496 and 503) and TotalSeq-C Universal Cocktail (BioLegend) (remaining 10 donors) in PBS containing 5% (v/v) FBS and 2 mM EDTA, incubated at 4 °C for 30 min and washed 3 times with PBS containing 5% (v/v) FBS and 2 mM EDTA before resuspension in a final volume of 1 ml. CITE-seq antibody panels are listed in Supplementary Table 1. Each tissue was subjected to a tissue-specific protocol to generate a single-cell suspension of immune cells that has been published in detail elsewhere. Immune cells were isolated from blood, BM aspirates (sternum), spleen, LNs, lungs, liver, JEJ (JEL and JLP) and skin. Each single-cell suspension was hashtagged to allow pooling of samples for loading on the 10× Genomics Chromium instrument. Approximately 500,000 MNCs per tissue were transferred into 1.5 ml Lo-Bind DNA tubes. Cells were centrifuged at 400g for 5 min, the supernatant removed and resuspended in 50 μl PBS containing 0.04% BSA. Cells were treated with 5 μl TruStain FcX (BioLegend) to reduce background labeling and incubated at 4 °C for 10 min, then each hashtag was added to the sample (1 µl hashtag per tube). Samples were incubated at 4 °C for 30 min, washed three times with PBS containing 0.04% BSA and equal numbers of cells from each tissue were pooled based on the number processed per donor. Cells were incubated with TotalSeq-C Human Universal Cocktail (BioLegend) (Supplementary Table 2) for 30 min at 4 °C and subsequently washed 3 times with PBS containing 0.04% BSA. Cells were resuspended in 500 µl PBS containing 0.04% BSA and passed through a 40 µm Flowmi pipette tip filter to remove any clumps of cells. For scRNA-seq experiments, single cells were loaded onto the channels of a Chromium chip (10× Genomics). cDNA synthesis, amplification and sequencing libraries were generated using either the Single Cell 5′ Reagent (v1 and v2) or 3′ Reagent (v3) Kits. TCRαβ and BCR paired VDJ libraries were prepared from samples made with the 5′ Reagent Kit. All libraries were sequenced on either an Illumina HiSeq 4000, NextSeq or NovaSeq 6000 instrument. Alignment was performed using Cell Ranger (v6.0.0) from 10× Genomics with the appropriate chemistry option (fiveprime or SC3Pv3). We added the cell hashing antibody and the protein antibody fastqs to a single call of cellranger count. Immune receptors (TCR and BCR) were aligned using cellranger vdj. TCR and BCR alignment results from Cell Ranger were used for quality control and filtering of low-quality cells (individual cells with both TCR and BCR detected). In cases where a single cell had both TCR and BCR reads, the immune receptor data were discarded, and the cell was labeled as a multiplet. For all alignments, we used reference genome refdata-gex-GRCh38-2020-A and immune receptor reference refdata-cellranger-vdj-GRCh38-alts-ensembl-5.0.0. Samples were demultiplexed by hashtag expression using hashsolo with default parameters. Cells that were not uniquely assigned to an individual sample were removed from downstream analysis. Filtering was performed to remove cells with fewer than 50 unique genes detected. Mitochondrial counts were quantified, summing all genes starting with ‘MT-’, and ribosomal counts were quantified using all genes starting with ‘RPS’ and ‘RPL’. For erythrocyte-related counts, all genes starting with ‘HB’ as well as ALAS2 and EPOR (to detect erythrocyte precursors) were quantified. Cells with more than 20% mitochondrial counts were flagged as potentially low-quality for later filtering. Counts for mitochondrial genes and MALAT1 were subsequently removed from the gene expression object and downstream analysis. To exclude contamination from ambient RNA, we processed the data using DecontX. Two samples (one from liver and one from skin) with abnormally high ambient counts were removed, as DecontX could not correct the ambient counts (for example, plasma cell genes like ALB in all immune cells). All downstream analysis was performed on uncorrected counts, as we found few ambient counts in other samples. We used a CellTypist model (at https://cog.sanger.ac.uk/celltypist/models/Red_Blood_CZI/v1/Red_Blood_CZI.pkl) to detect erythrocytes. Doublets were additionally detected using Scrublet with a sim_doublet_ratio of ten. For each unique tissue site, we performed an initial integration across all samples by training a single-cell variational inference (scVI) model on the gene expression with following parameters: 10,000 highly variable genes using the seurat_v3 option in Scanpy, early stopping enabled and 50 epochs, 10 epochs for n_epochs_kl_warmup, two layers in encoder and decoder, nb gene likelihood and a mini-batch size of 256. To perform filtering of low-quality events, we used the following quality metrics: the probability of a doublet predicted by Scrublet, the probability of a doublet from HashSolo, the percentage of erythrocyte genes as described above, whether a cell contained both TCR and BCR, whether the CellTypist erythrocyte model predicted a cell to be an erythrocyte, as well as cells with a total count below 2,000 unique molecular identifiers, 1,200 unique genes or 200 protein counts. All scores were added to generate a per-cell quality metric. To perform filtering, we argued that cells that group together and have evidence of low quality should be removed from downstream analysis. We first used Louvain clustering on the coordinates from scVI latent space using 15 nearest neighbors to cluster the per-tissue integrated data with a resolution of 5.0. Every cluster with a median low-quality score (described above) of at least one was removed from downstream analysis. Although some low-quality events were retained with this filtering, their frequency was drastically reduced. We additionally established tissue-specific cut-offs to remove additional events and removed clusters with a mean low-quality score of 0.3 from all tissues, except for the lung LN and JEJ, for which the threshold was manually increased to recover higher-quality cells. Using a course cell-type annotation based on manual annotation of clusters, we identified cell types that were consistently filtered out, even though their quality did not appear to be spuriously low by manual inspection. We retained mast cells and hematopoietic stem cells from all tissues, all macrophages from LNs and spleen, all erythrocytes and platelets from BM and all monocytes from liver. We concatenated cells from all tissues and computed the 10,000 top highly variable genes using the seurat_v3 option in Scanpy and used the same parameters as described above, but with a mini-batch size of 1,024 to accelerate the training process. We used this integrated latent space to assign initial cell types and removed all cell types that were not labeled as immune cells. Additionally, we removed all cells for which manual labeling and automatic labeling using MMoCHi (see below) were inconclusive about coarse cell-type identity (for example, B cell, myeloid, T cell). These events were of low quality by manual inspection. We performed post hoc manual removal of these and other clusters of low-quality cells after integrating all cells. To identify canonical immune cell subsets, we used a recently reported, supervised machine learning algorithm, MMoCHi) (v0.2.1). We first normalized the gene expression (GEX) count matrix using log(10,000Cg,i / TG,i + 1), where Cg,i represents the counts for GEX feature g in cell i, and TG,i is the total counts for all GEX features in cell i. Similarly, we normalized the antibody-derived tag (ADT) count matrix using log(1,000Ca,i / TA,i + 1), where Ca,i represents the counts for ADT feature a in cell i and TA,i is the total counts for all ADT features in cell i. We applied landmark registration (MMoCHi) to batch-correct the ADT expression across experimental batches. In brief, we applied automatic detection of landmarks (peaks) in the expression distributions for each ADT feature in a given sample, applying manual adjustments as needed using the graphical user interface, then performed curve registration and warping to align the positive and negative peaks for each ADT feature across batches. We provided MMoCHi with a hierarchy of immune cell subsets and their canonical surface protein-level and RNA-level markers (Supplementary Fig. 2) and used the markers to identify high-confidence members (cells) of each subset for training. For each classification level, automatic thresholds for high-confidence ADT or GEX marker-positive and marker-negative cells were manually adjusted as needed using the supplied GUI (Supplementary Table 3). Following MMoCHi’s internal training data refinement, we applied an 80–20 train–test split and trained a random forest classifier, sklearn.ensemble.RandomForestClassifier, on both gene and protein expression. For 2 of the 24 organ donors and a subset of samples from a third donor, we did not perform CITE-seq and only had scRNA-seq profiles. Thus, these samples were excluded from the MMoCHi classification described here. However, we used a k-nearest neighbors approach to transfer the classifier labels to individual cells profiled from these two organ donors. Specifically, we used the sklearn.neighbors.KNeighborsClassifier with n_neighbors = 10 to construct a k-nearest neighbors graph in the mrVI embedding of the dataset (see below) and classify the remaining cells. Of the subsets, pDCs were identified using two separate nodes on the hierarchy (Supplementary Fig. 1), as pDCs shared expression with both B cells and myeloid cells. Once classified, the two subsets were merged into a single population of pDCs. The MMoCHi annotation was used at two separate levels throughout the paper, defined as either one of the 34 fine-grained subsets or grouped into CD4 T cells, CD8 and unconventional T cells (including γδ T cells and CD8 MAIT cells), B cells, NK cells and ILCs or myeloid cells (including monocytes, macrophages, cDCs, migratory DCs and pDCs). Owing to the breadth of tissues and human subjects sampled in our dataset and high-resolution annotation of immune subsets, we anticipated that our immune atlas would be useful to the research community as a reference for performing cell-type label transfer. To facilitate this application, we trained a model using popV, a tool developed for cell-type label transfer that uses several annotation algorithms and consensus voting to determine annotations and evaluate their confidence. popV also calculates joint embeddings of the query and reference datasets, which can be used for visualization of the query data and other analysis tasks. A popV model was trained using the tissues and MMoCHi annotations (Fig. 2) as the reference dataset. Label transfer performance was evaluated using the Human Lung Cell Atlas as a query dataset. To visualize the data, we computed UMAP embeddings as described above on joint scVI embeddings, which were calculated as part of the popV pipeline. To evaluate the importance of ADT information in the MMoCHi classification performance, we additionally applied a pre-trained CellTypist model (Immune_All_Low; https://celltypist.cog.sanger.ac.uk/models/Pan_Immune_CellTypist/v2/Immune_All_Low.pkl) using default settings to the tissue immune cells (Fig. 2). To integrate scRNA-seq profiles of immune cells in our study, we first used scVI, which did not yield a fully integrated latent space and clustered by site of collection (for example, US or UK). We next leveraged MrVI, which uses a mixture-of-Gaussian as a prior and enforces stronger separation of true cell state and effect of donors on gene expression, as has been recently demonstrated. MrVI takes advantage of a prior based on a multimodal variational mixture of posteriors (similar to a VampPrior), which have been shown to outperform Gaussian priors for scRNA-seq integration in benchmarking studies. In brief, MrVI finds a sample-agnostic latent space, U, and computes a sample-specific embedding. A second latent space, Z, is defined by adding an attention-based concatenation between U and the sample embedding space to the original U-space. Another layer of attention is used to incorporate an embedding of 10× Genomics chemistry and experimental site (Cambridge, UK versus Columbia, NY), and this third latent space is decoded using a linear decoder to yield the rate of a negative binomial distribution. We use a cell-type-aware Gaussian mixture prior in U-space. To introduce cell type awareness, we use a bias to the mixture proportions that makes it likely for cells of the same type to be sampled from the same Gaussian. For the latent embedding highlighted throughout the article and used for manual cell-type curation, we used the donor identities as the sample keys and used the output of MMoCHi classification (see above) as the cell-type prior in MrVI. We used default parameters except n_epochs_kl_warmup of 25, n_latent_u of 20, n_latent in Z-space of 200, dropout in qz as well as pz of 0.03 (adopted from a previous publication). To visualize cells (either the total immune component or individual major lineages), we computed nearest neighbors (scanpy.pp.neighbors) on the MrVI U latent space and calculated UMAP embeddings (scanpy.tl.umap) using the 15 nearest neighbors, a minimum distance of 0.4, a spread of 1.0 and initialization with PAGA after running scanpy.tl.paga. To identify additional heterogeneity in cell states within samples in addition to the cell-type annotation provided by MMoCHi, we performed manual annotation. For each MMoCHi annotated population, a new scVI model was trained with donor as the batch key, then Leiden clustering (scanpy.tl.leiden) was performed on the lineage-specific neighbors graph at an appropriate resolution, selected to minimize over-clustering (ranging from 1 to 15). Markers for each cluster were computed by scanpy.tl.rank_genes_groups, and clusters with similar marker expression were merged. To annotate proliferating cells, scanpy.tl.score_genes_cell_cycle was run, and the output was used in combination with the gene expression of MKI67 and TOP2A. We focused our DE analysis on immune lineages and cell types with sufficient representation across experimental sites, tissues and donor ages. This included six tissue groups: blood, BM, spleen, gut (JLP and JEL), LNs (ILN, LLN and MLN) and lungs (consisting of BAL and parenchyma); six immune lineages: myeloid, CD4 T cells, CD8 T cells, invariant T cells (that is, γδ T cells and MAIT cells), B cells and ILC/NK cells; and 26 individual cell types within all lineages. Covariates included 10× Genomics chemistry (3′ versus 5′), sex (male versus female), laboratory (Cambridge, UK versus Columbia, NY) and CMV status (positive versus negative). For aging analyses, donors were categorized as being <40 or >40 years of age. Variance decomposition and pseudobulk DE analysis were performed using LMM through the dreamlet R package (v1.4.1). Depending on the resolution of the analysis, DE was performed separately either for each immune lineage (for example, myeloid cells, B cells, and so on) or for each immune subset (for example, macrophages, naive B cells and so on) using the cluster_id parameter in dreamlet. The raw GEX count matrix was pseudobulked across samples, and each tissue in each donor was treated as a separate sample. Before performing DE, samples and genes with poor representation were filtered using dreamlet::processAssays. Samples with fewer than 50 cells and genes not represented in at least 40% of the samples with at least five counts were excluded. To confirm findings by MrVI counterfactual analysis (see below), these thresholds were reduced to a minimum of ten cells for a sample to be included, and at least 10% of samples with at least five counts. DE for a subset was not performed when fewer than three or four samples (for tissue and age analysis, respectively) met the minimum cell thresholds. Variance decomposition was performed for age analysis for each lineage using dreamlet::fitVarPart with sex, sequencing chemistry, CMV serostatus, age group, processing site and tissue as covariates (Supplementary Table 13). LMM was performed using dreamet::dreamlet with eBayes estimation enabled. Tissue effects (Figs. 3 and 4, Extended Data Figs. 3 and 4, and Supplementary Tables 4, 7 and 10) were modeled by comparing each lineage/subset in one tissue group to the same lineage/subset in the remaining tissue groups, with donor identity encoded as a random effect. Age effects (Figs. 5–7, Extended Data Fig. 5, and Supplementary Tables 15 and 24) were modeled across each tissue-group–age-group combination while controlling for CMV serostatus and sex as fixed effects and with sequencing chemistry and processing site as random effects. Age effects within each tissue group were then measured using the contrasts parameter in dreamlet::dreamlet between old and young for each tissue group (for example, the effect of age in the gut was computed as ‘old-gut − young-gut’). CMV effects (Supplementary Fig. 10 and Supplementary Table 21) were modeled across each tissue-group–CMV serostatus combination while controlling for age and sex as fixed effects and with sequencing chemistry and processing site as random effects. CMV effects within each tissue group were then measured using the contrasts parameter in dreamlet::dreamlet between CMV and CMV for each tissue group. For identifying cross-tissue and cross-donor gene signatures for each major immune lineage, we constructed probabilistic factor models directly from scRNA-seq count matrices using scHPF. The output of scHPF includes two matrices: an M × K gene score matrix containing weights for each of M genes in each of K factors and a K × N cell score matrix containing weights for each of N cells in each of K factors. In the original report of scHPF, the algorithm required a user-supplied value of K, the number of factors in the model. Here, we use a new consensus factorization implementation of scHPF, in which the user specifies a broad range of K values from which many scHPF models are generated. The gene score matrices for these models are then clustered to identify K recurrent factors, which are combined to seed a final round of training to construct a final consensus model with K factors. We constructed two types of scHPF models: a tissue-level model (Extended Data Fig. 4), in which the number of cells from each of three tissue groups was balanced by random sub-sampling (gut: JEL and JLP; lung parenchyma; and LNs: MLN and LLN), and a donor-level model (Figs. 5 and 6), in which the number of cells from each organ donor was balanced. We constructed both types of models for the major immune lineages: CD4 T cells, CD8 T cells (including all invariant T cells), NK cells, ILCs, B cells and macrophages. For donor models, donors with fewer than 300 cells for a given lineage were removed. In both models, the count matrices were randomly downsampled such that the average number of transcripts per cell was the same for each donor to avoid coverage bias. scHPF models considered only protein-coding genes (excluding TCR and immunoglobulin cassettes) detected in at least 1% of cells across the final subsampled and downsampled training matrix. For all consensus scHPF models, we ran scHPF five times for each of 16 values of K (15–30), from which we selected the top three models for each value of K based on convergence criteria for clustering. We applied walktrap clustering to identify recurrent clusters, which we required to form clusters with factors from at least two different models from which we trained the final consensus model. Immune subset composition within each lineage across tissues was visualized by violin plots or box plots (using seaborn). Tissue-specific enrichment of immune subset frequencies in specific tissues was also assayed within each major lineage using scCODA for Bayesian inference. Significant enrichment of an immune subset in one tissue over the rest was determined using sccoda.util.comp_ana.CompositionalAnalysis to detect credible effects, and was run sequentially, selecting each cell type as the reference. Majority voting was then used to identify cell types that are credibly changing more than half the time with automatic reference-subset selection and the default false discovery rate of 0.05. To determine tissue-specific gene expression signatures across immune lineages (Fig. 3), significant DEGs were defined as adjusted P < 0.05 and log2(FC) > 1 by pseudobulk DE across tissues at the lineage level (see above). Mean z-score gene expression was calculated for each pairing of tissue group and lineage. Genes and samples were both hierarchically clustered using scipy.cluster.hierarchy.linkage with Ward’s method and Euclidean distance. Discrete clusters of genes with similar expression patterns were calculated using scipy.cluster.hierarchy.fcluster with the ‘maxclust’ method (Supplementary Table 6). For each gene cluster, association with specific tissue groups or lineages could arise from DE within one or more specific subsets of that lineage or from compositional shifts in subsets across tissues. To disentangle these possibilities, we first used pre-ranked GSEA to compare the gene clusters identified via lineage-level DE to the effect size (that is, log(FC)) of DE across tissues in the subset-level DE. To visualize potential effects caused by compositional shifts across tissues, we computed the average frequency of the subset (as a proportion of the total cells within that lineage group) within a tissue, the FC of that frequency over the frequency in the remaining tissue groups and the average expression of the gene cluster. To assess whether differential transcript expression was reflected in the surface protein profiling (Supplementary Tables 11 and 12), we selected ADTs corresponding to DEGs in at least one tissue. To identify enrichment in one tissue group over the other tissue groups, we used scanpy.tl.rank_genes_groups on the normalized expression with Wilcoxon and tie-correction enabled. To minimize the influence of technical staining artifacts or donor covariates, analysis was conducted separately within each donor. Donors with fewer than 50 cells for a particular tissue-group–lineage combination or tissue-group–lineage combinations with fewer than four suitable donors were excluded from analysis. Before DE analysis, the ADT count matrix was subsampled to equalize cell numbers and randomly downsampled such that the average number of transcripts per cell was the same for each group to avoid coverage bias. We next sought to identify factors from the tissue-level scHPF models of each major immune lineage that were shared across cell types. As described above, we first constructed consensus scHPF models for CD4 T cells, CD8 T cells, macrophages, NK cells, ILCs and B cells with equal representation of cells from each of three major tissue groups (gut, lung and LNs). From each model, we removed probable nuisance factors containing heat shock protein-encoding genes (common dissociation artifact, >1 gene), ribosomal protein-encoding genes (common coverage artifact, >10 genes), genes from the highly inducible metallothionein cluster (>1 gene), hemoglobin transcripts (red blood cell contamination, >0 genes) and genes in a previously published signature of dissociation-induced cell stress in scRNA-seq (>7 genes) among the 30 top-weighted genes. Next, we computed the average cell score for each factor in each of the three major tissue groups and identified all factors with an average tissue-group score that was at least 80% higher in one tissue group than the average of the remaining two. Thus, the resulting set of 53 scHPF factors from across all 6 lineage-specific models exhibits some degree of tissue specificity. To compare these factors to each other, we computed the Pearson correlation between the gene score vectors for each pair of factors. We then identified factors with a pairwise correlation that was greater than the 95% confidence threshold with at least two other factors, which yielded 31 scHPF factors from across the six major immune lineages. Finally, we performed hierarchical clustering of the Pearson correlation matrix for these 31 factors (seaborn.clustermap using Euclidean distances) to identify modules containing factors with similar gene signatures that originated from different, lineage-specific scHPF models (Extended Data Fig. 4). Modules of genes were further interrogated by average gene expression and validated in specific immune subsets using pseudobulk GEX DE and ADT DE as described above. To detect shifts in the subset composition of specific lineages across the age groups and CMV serostatus (Supplementary Table 14), we performed generalized linear modeling by fitting a statsmodels.GLM model for each tissue subset, considering sex, sequencing chemistry, CMV serostatus and processing site as additional covariates. Donors with fewer than 50 cells for a particular tissue-group–lineage combination or tissue-group–lineage combinations with fewer than four suitable donors were excluded from analysis. The estimated coefficients were used to calculate a covariate-aware log2(FC) for visualization. We used the statsmodels.multipletests function to adjust P values for multiple comparisons (Benjamini–Hochberg method), and subset–tissue combinations with adjusted P < 0.05 were considered significantly changing across age. To depict age-associated effects on the immune system, we visualized the similarity of trending DEGs by age on immune subsets across tissues using t-distributed stochastic neighbor embedding (t-SNE) (Fig. 5d). We first calculated trending DEG (unadjusted P < 0.05; <40 or >40 years old) pairwise similarities by summing the intersection of positively regulated genes (log2(FC) > 0.1) and negatively regulated genes (log2(FC) < −0.1), divided by the overall union of both. The similarity or distance (1 − similarity) was applied to cell types containing more than 70 DEGs (unadjusted P < 0.05; mean log-normalized expression, >0.05) and present in at least 3 donors per tissue and age group. The similarity levels of cell types and tissues with more than 200 DEGs were further clustered using the Ward.D2 method and projected into a distance-based t-SNE illustration. To investigate the effect of age on specific genes within each immune subset, we plotted genes that were significant in at least one tissue (adjusted P < 0.05) and within the top 50 significant genes. Although our power to detect age effects by DE was limited, genes that were significantly DE in one subset were often trending in the same direction across multiple tissue groups. To assess the effect of age on surface protein expression (Supplementary Table 17), we used landmark-registered protein expression data (by MMoCHi, see above) to account for donor-to-donor batch effects in ADT staining quality. Although landmark registration preserves the separation between positive-expressing and negative-expressing cells for thresholding, this non-parametric normalization can obscure changes in overall expression intensity between samples. Therefore, we focused on shifts in percent positivity for a marker in each tissue subset. We performed automatic thresholding by MMoCHi, followed by manual adjustment (as described above), on the landmark-registered expression of all ADTs corresponding to a DEG by age. The percentage of cells with expression of a given ADT above the positive threshold was calculated for each donor–tissue-group–subset combination. Donors with fewer than 50 cells for a particular tissue-group–lineage combination or tissue-group–lineage combinations with fewer than four suitable donors were excluded from analysis. The percent positivity was used as the response variable in the same linear regression model used to detect shifts in composition across age groups. We adjusted P values for multiple comparisons as above, and ADTs with adjusted P < 0.05 were considered significant. We constructed donor-level scHPF models for each major immune lineage with uniform representation of cells from each donor to identify age-associated gene signatures, as described above. For each scHPF model, we performed LMM to account for covariates and identify age associations. Each LMM contained six categorical covariates as fixed effects: age group, sequencing chemistry, sex, processing site and CMV serostatus. We also considered three tissue types: mucosal (BAL, lung parenchyma, JLP, JEL), LNs (ILN, MLN and LLN) and blood-rich, including blood, BM and spleen), which required us to select one category (blood-rich) as a held-out variable. Thus, we have two categorical variables for tissue, which effectively represent mucosal versus blood-rich and LN versus blood-rich. We encoded donor identity as a random effect. LMM coefficients and P values were computed for each factor in a given scHPF model using the cell scores as response variables by fitting a statsmodels.MixedLM model and using the statsmodels.multipletests function to adjust P values for multiple comparisons (Benjamini–Hochberg method). To cross-validate age-associated scHPF factors in other datasets, we further analyzed a bone marrow atlas containing 36 age-annotated donors with good B cell representation for a B cell aging factor, a lung atlas containing 29 age-annotated donors with good macrophage and CD8 T cell coverage and PBMC data from the Sound Life cohort (age 25–65 years, n = 96) from the Human Immune Health Atlas (Figs. 5 and 6 and Extended Data Fig. 5). Using the published cell-type annotations from each atlas, we extracted the appropriate scRNA-seq profiles and projected them into the corresponding donor-level scHPF models generated from the data reported here using the scHPF project function. This resulted in cell scores for cells from the external data sets for the same factors that were generated from this data set, allowing us to compare the average cell scores for young versus older donors from the external data. As an orthogonal approach, we also performed pseudobulk DE analysis between older and younger donors (using an age cutoff of 40 years) from the external data sets, ranked the genes by FC and used GSEA to analyze the statistical enrichment of age-associated factors among young versus old donors (Supplementary Table 19). We used the top 200 genes (ranked by scHPF gene score) in each age-associated factor as gene sets for GSEA. For the DE analysis described in Fig. 7, we subsetted the MrVI sample embeddings to each tissue group and modeled the predicted ε in MrVI by a linear model adjusting for covariates in sex, CMV serostatus and age group as fixed effects, considering sequencing chemistry and processing site as site covariates in MrVI. A ridge regression parameter of 0.1, owing to collinearity of cofactors, was added. This decomposition of ε was performed for every single cell. This yields an estimated effect in Z-space for each covariate. The effect vector was added to the mean cell embedding in Z-space, and DEGs were computed based on the modified and mean embedding for each cell. For downstream analysis, this matrix of estimated log(FC) for each cell and gene was further processed for each immune subset. First, all cells that were represented only in fewer than three samples were filtered out. Second, for each cell type, we excluded genes with less than 0.01 raw average expression or an estimated log(FC) across age groups with a 95th percentile below 0.1, retaining only genes that might be affected by age in a group of cells. To dissect predicted gene effects into modules, neighborhood smoothing was performed using 15 nearest neighbors in U-space and multiplying two times the normalized affinity matrix by the predicted gene effects. Spectral co-clustering was performed with four gene clusters and four cell clusters, with mini-batch enabled using sklearn.cluster.SpectralCoclustering. Marker genes for each module were identified by averaging the predicted log(FC) across all cells from the corresponding cell module, and the top 50 genes for each module were identified. We used decoupleR-py to compute a module score of log(FC) scores using weighted means of the signs of those marker genes (Supplementary Table 23). For the lung, we isolated a gene module in CD4 T cells that contained TRM cells. To detect similar cells in other tissues, we computed the best cutoff for the module score to identify cells in a specific cell module based on Youden’s J statistic, computed the module score for all cells from other tissues as described above and applied the same cutoff to all other tissues as the tissue of interest. Given that the gut contained TRM cells with a TH17 phenotype and all other tissues had no module-positive cells, we selected all cells with a MrVI predicted negative log2(FC) of IL17A below −0.05. To confirm our findings on a per-gene level, we selected module-positive cells and used pseudobulk estimates of DE using dreamlet (Supplementary Table 24). Samples with fewer than 5 module-positive cells or 1,000 total counts and genes with fewer than 3 total counts were removed. Aging DE was performed using the contrasts method, as described above. Genes within a shared functional group were manually selected from the MrVI signature for visualization. Pseudobulk DE analysis was performed on the classifier-predicted cells in other tissues using the same settings as above in this cell subset. Enrichment of module or selected marker genes in the pseudobulk DE analysis was performed using GSEA implemented in decoupler.run_gsea (Supplementary Table 23). Cell Ranger-mapped TCR and BCR contigs contained in ‘all_contigs.fasta’ and ‘all_contig_annotations.csv’ output files were re-annotated using the Dandelion preprocessing pipeline. This pipeline includes the following steps: (1) sample suffix or prefix assignment to each sample barcode; (2) re-annotation of contigs with IgBLAST (v1.19.0) against IMGT (international ImMunoGeneTics) reference sequences (last downloaded on 24/04/2023); (3) re-annotation of D and J genes separately using blastn to enable the annotation of contigs without the V gene present; and (4) identification and recovery of nonoverlapping individual J gene segments. For BCRs, three additional steps were also performed: (1) additional re-annotation of heavy-chain constant (C) region calls using blastn (v2.13.0) against curated sequences from CH1 regions of respective isotype class; (2) heavy-chain V gene allele correction using TIgGER (v1.0.1); and (3) BCR mutation calling. Cell-level quality control was performed using Dandelion’s ‘filter_contigs’ function, which only considers productive VDJ contigs, asserts that a single cell should only have one VDJ and one VJ pair or only an orphan VDJ chain and explicitly removes contigs that fail these checks (except for IgM/IgD and TRB/TRD extra pairs). Contigs that did not match any cell barcodes in the gene expression data were also removed at this step. TCRs and BCRs were then grouped into clones or clonotypes. The following default sequential criteria, which apply to both chain contigs, were applied: (1) identical V and J genes usage; (2) identical junctional CDR3 amino acid length; and (3) CDR3 sequence similarity: 100% nucleotide sequence identity at the CDR3 junction for TCRs and 85% amino acid sequence similarity (based on Hamming distance) for BCRs. TCR or BCR data were then transferred into the corresponding AnnData object. Cells without receptor data or that presented more than one receptor were discarded from further immunoreceptor analysis. For T cell analysis, cells annotated as MAIT cells or ɣδ T cells were also discarded. Clonality of the different populations was calculated as 1 − Pielou’s evenness index, varying from zero (more diverse) to one (less diverse), with the Pielou’s evenness corresponding Hs / Hmax, where Hs is the Shannon entropy of sample s and Hmax = log2C, where C is the number of unique clonotypes in s. All clonality scores were calculated on a subsample of 100 cells for each donor, cell type, tissue or cell type and tissue. To detect shifts in the BCR isotype composition of specific B cell lineages across the age groups (Supplementary Table 20), we performed generalized linear modeling by fitting a statsmodels.GLM model for each tissue subset, considering sex, sequencing chemistry, CMV serostatus and processing site as additional covariates. Donors with fewer than 50 cells for a particular tissue-group–lineage combination or tissue-group–lineage combinations with fewer than for suitable donors were excluded from analysis. The estimated coefficients were used to calculate a covariate-aware log2(FC) for visualization. We used the statsmodels.multipletests function to adjust P values for multiple comparisons (Benjamini–Hochberg method), and subset–tissue combinations with adjusted P < 0.05 were considered significantly changing across age. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41590-025-02241-4.
PMC12435838
Organ-specific features of human kidney lymphatics are disrupted in chronic transplant rejection
Lymphatic vessels maintain tissue fluid homeostasis and modulate inflammation, yet their spatial organization and molecular identity in the healthy human kidney, and how these change during chronic transplant rejection, remain poorly defined. Here, we show that lymphatic capillaries initiate adjacent to cortical kidney tubules and lack smooth muscle coverage. These vessels exhibit an organ-specific molecular signature, enriched for CCL14, DNASE1L3, and MDK, with limited expression of canonical immune-trafficking markers found in other organ lymphatics, such as LYVE1 and CXCL8. In allografts with chronic mixed rejection, lymphatics become disorganized and infiltrate the medulla, with their endothelial junctions remodeling from a button-like to a continuous, zipper-like, architecture. Lymphatics in rejecting kidneys localize around and interconnect tertiary lymphoid structures at different maturation stages, with altered intralymphatic and perilymphatic CD4 T cell distribution. The infiltrating T cells express IFN-γ, which upregulates coinhibitory ligands in lymphatic endothelial cells, including PVR and LGALS9. Simultaneously, lymphatics acquire HLA class II expression and exhibit C4d deposition, consistent with alloantibody binding and complement activation. Together, these findings define the spatial and molecular features of human kidney lymphatics, revealing tolerogenic reprogramming accompanied by structural perturbations during chronic transplant rejection. Keywords: Immunology, Nephrology, Vascular biology Keywords: Adaptive immunity, Lymph, Organ transplantationLymphatics are blind-ended vessels lined by lymphatic endothelial cells (LECs); they are responsible for clearing fluid and macromolecules from the microenvironment and play a critical role in maintaining tissue homeostasis (1–3). During inflammation, lymphatics expand through lymphangiogenesis to facilitate leukocyte efflux. While their role and therapeutic potential in lymphedema (4), cardiovascular disease (5–7), cancer (8), and neuropathology or neuroinflammation (9, 10) are becoming increasingly recognized, these functions rely on organ-specific structural and molecular specialization. In the kidney, lymphatic vasculature is considered an important entity in physiology and disease (11–14). Epithelial nephrons and their associated blood capillary networks (15) underlie plasma ultrafiltration, fluid homeostasis, and acid-base balance. In contrast, although lymphatics appear in the human fetal kidney by the end of the first trimester (16), their precise spatial organization, molecular identity, and cellular interactions remain poorly defined. Kidney transplantation, the most common solid organ transplant, has excellent short-term outcomes but is limited long-term by chronic rejection, a major cause of late allograft failure (17, 18). Chronic rejection, driven by both T cell–mediated injury and/or donor-specific antibodies targeting HLA, results in microvascular injury, interstitial fibrosis, and tubular atrophy (19, 20). Lymphatics serve as a route for trafficking of antigens and leukocytes, but their role in transplant immunity remains controversial (11, 21, 22). Lymphangiogenesis in kidney transplants (23–26) has been associated with promoting resolution of inflammation and improving allograft function (27–29) but also with alloantigen presentation and fibrosis (30, 31). These contrasting findings underscore the need to better understand lymphatic remodeling in transplant rejection and how it diverges from other kidney pathologies featuring lymphangiogenesis (32–34). Here, we used 3D microscopy of optically cleared and immunolabeled tissue, in addition to single-cell RNA-Seq (scRNA-Seq), to map the spatial organization and molecular profile of lymphatics in the healthy human kidney. We identified blind-ended lymphatic capillaries around cortical nephron segments, with a distinct molecular signature compared with LECs from other organs. In allografts with chronic mixed rejection, lymphatic vessels expanded into the medulla and were structurally disorganized, with disrupted LEC-cell junctions. In this setting, lymphatics surrounded tertiary lymphoid structures, with altered intralymphatic and perilymphatic CD4 T cell distribution. However, our molecular analyses suggest that LECs are tolerogenic and respond to T cell–derived IFN-γ by upregulating immune-inhibitory molecules. Critically, lymphatics in rejecting allografts also expressed HLA class II and exhibited complement 4d (C4d) deposition, indicative of alloantibody binding (35). Thus, our findings revealed that kidney lymphatics in chronic rejection adopt potentially compensatory tolerogenic changes but are concurrently structurally perturbed, better defining their contributions to alloimmunity. To characterize lymphatic architecture in the healthy human kidney, we analyzed tissue samples from 4 deceased organ donors with minimal chronic damage (<10% interstitial fibrosis or tubular atrophy, Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/JCI168962DS1) (36). Intact tissue samples (<3 mm) were immunolabeled using a D2-40 monoclonal antibody targeting podoplanin (PDPN) (37) and imaged using confocal or light-sheet fluorescence microscopy (LSFM) (16, 38). PDPN vessel networks were visualized in the human kidney cortex (Figure 1A and Supplemental Video 1), and antibody-omitted controls displayed minimal autofluorescence or nonspecific binding (Supplemental Figure 1A). Mapping vessel radius revealed a hierarchical network, with small lymphatics (radius ~3.5 μm) initiating in the cortex and converging into larger vessels (radius ~50 μm) at the corticomedullary junction (Figure 1B). The cells lining these vessels expressed prospero homeobox protein 1 (PROX1) (Figure 1C), a canonical LEC transcription factor (39, 40) but showed sparse expression of lymphatic vessel endothelial hyaluronan receptor 1 (LYVE1) (Figure 1D), a glycoprotein important for leukocyte entry into lymphatics (41). (A) Representative maximum intensity z-projection, from low-resolution confocal tile scans, of n = 3 human kidney tissues labeled for PDPN and UMOD, demonstrating PDPN lymphatics (arrowheads). Scale bar: 2,000 μm. (B) Segmented and rendered light-sheet imaging of lymphatics from the same kidney tissue in A, representative of n = 3 images. 3D color renderings represent vessel branch radii, with blue representing the smallest radius (<3.5 μm, asterisks) and red representing the largest radii (>50 μm, arrowheads). (C and D) Representative 3D reconstruction of cortical regions from n = 2 human kidney tissues labeled for PDPN and either PROX1 or LYVE1. The PROX1 signal and LYVE1 signal are masked to only include expression from within the vessel, demonstrating expression of PDPN cells. Sparse membrane localization of LYVE1 is demonstrated (arrowheads). Representative of 5 regions of interest imaged. Scale bars: 30 μm. (E–G) Regional localization of lymphatics (arrowheads) in the human kidney using LTL (cortex), UMOD (medulla), and UAE-I (with dotted lined delineating the capsule). Regional structures are indicated with asterisks, including proximal tubules in E, loops of Henle in F, and glomeruli in G. Scale bars: 70 μm (E), 150 μm (F), 100 μm (G). (H–K) Spatial relationships of lymphatics (arrowheads) relative to UAE-I renal arterioles (RA) and glomeruli (G) in H, LRP2 proximal tubules (PT) in I, CALB1 distal nephron tubules (DT) in J, and DBA collecting ducts (CD) in K. Scale bars: 50 μm (H), 80 μm (I and J), 300 μm (K). (L) Schematic depicting the spatial relationships of lymphatics (arrowheads) to nephron segments. All imaging from E–K representative of 5 regions of interest imaged across n = 2 kidneys. To elucidate the microanatomical localization of lymphatics in the human kidney, autofluorescent tissue signals were captured alongside PDPN labeling. Large caliber lymphatics were observed adjacent to arteries at the corticomedullary junction branching into smaller cortical vessels (Supplemental Video 2). Colabeling with Lotus tetragonolobus lectin (LTL, proximal tubules) and uromodulin (UMOD, loop of Henle) revealed PDPN blind-ended lymphatics in the renal cortex (Figure 1E and Supplemental Video 3) and their absence from the medulla (Figure 1F). Despite previous reports of subcapsular lymphatics (42, 43), these were not detected in 3D reconstructions (Figure 1G) or optical z-sections (Supplemental Figure 1B), even with the kidney capsule intact. In the cortex, lymphatics followed UAE-I arterioles toward glomeruli (Figure 1H), extending terminal branches near megalin (LRP2) proximal tubules (Figure 1I) and calbindin 1 (CALB1) distal nephron epithelium (Figure 1J and Supplemental Video 4). Lymphatics converged toward the kidney hilum, adjacent to medullary Dolichos biflorus agglutinin (DBA) collecting ducts (Figure 1K) and UMOD medullary tubules (Supplemental Figure 1C). A model summarizing these findings is presented in Figure 1L. Because of the rarity of lymphatics in the human kidney relative to other cell types, isolating sufficient LECs for molecular profiling is challenging. To surmount this, we leveraged published scRNA-Seq data from 59 kidneys, supplemented with 5 new samples (Supplemental Figure 2A). This integrated dataset comprised 217,411 human kidney cells, with 151,038 control samples (living donor biopsies or unaffected regions of tumour nephrectomies) and 66,373 cells from diseased samples (chronic kidney disease, CKD) and kidney allograft injury; covering both chronic rejection and non-alloimmune etiologies) (Supplemental Figure 2B). All cell types were manually annotated (Supplemental Figure 2C), revealing 38 clusters (Figure 2A), including a transcriptionally distinct LEC cluster containing 700 cells. (A) Uniform manifold approximation and projection (UMAP) of an integrated atlas of 217,411 cells, including 151,058 control cells from live biopsies or nephrectomies, 46,540 cells from different etiologies of graft injury, and 19,813 from chronic kidney disease. TREM, triggering receptor expressed on myeloid cells 2. (B) Dot plot of top 20 markers of lymphatic profiles across all control cell types in the atlas. Groupings for each cell type are shown on the right. (C–F) Analysis of nonlymphatic expression of PROX1 and LYVE1 using 3D imaging. Arrowheads show the expression of each marker relative to CDH1 medullary tubules (C), CD31 vasa recta (D), or CD68 macrophages (E) and peritubular capillaries (F). Scale bars: 50 μm (C, E, and F), 30 μm (D). (G and H) Examination of ACTA2 expression relative to PDPN lymphatics (arrowheads) in the renal hilum (G) and cortex (H). Scale bars: 50 μm (G), 100 μm (H). (I) Subclustering analysis of n = 452 lymphatic endothelial cells (LECs) derived from human control kidney datasets reveals 2 transcriptionally distinct clusters, which we term LEC1 and LEC2. (J) Feature plots demonstrating expression of markers of all LECs (PROX1, PDPN), lymphatic capillaries (CCL21, LYVE1), and lymphatic collecting vessels (GATA2, FOXC2). (K) Volcano plot showing differentially expressed genes (DEGs) between the 2 lymphatic subclusters, with each point representing a gene. The x axis represents average log-fold change (log2FC), whereas the y axis represents –log10 of the adjusted P value of the Wilcoxon rank-sum test for differential expression. Blue dots represent genes that meet significance. Selected marker genes for each cluster are shown in boxes. From control samples, we curated a transcriptional signature of healthy kidney lymphatics (Supplemental Data 1), comprising 227 differentially expressed genes (DEGs) from 295 LECs. These genes were enriched for gene ontology (GO) terms associated with lymphatic fate commitment (GO:0060838, fold-enrichment > 100, FDR = 1.66 × 10) and lymphangiogenesis (GO:0001946, fold-enrichment = 67.4, FDR = 8.37 × 10). Canonical LEC markers were identified, including PROX1 (log2FC = 2.97), PDPN (log2FC = 2.65), neuropilin 2 (NRP2, log2FC = 2.73), and C-C motif ligand CCL21 (log2FC = 7.23) (Figure 2B). We also identified genes previously linked to kidney disease (44–46), such as fatty acid binding protein 4 (FABP4, log2FC = 5.69), trefoil factor 3 (TFF3, log2FC = 5.58), and angiopoietin 2 (ANGPT2, log2FC = 2.46) (Figure 2B). Given the frequent use of PROX1 and LYVE1 to identify or target kidney lymphatics in preclinical studies (11, 14), we examined their expression within the human kidney in more detail. PROX1 was detected not only in LECs, but also in loop of Henle and distal convoluted tubule clusters (Figure 2B), a finding validated by PROX1 and E-cadherin (CDH1) immunolabeling of medullary tubules (Figure 2C) (47). In contrast to mouse data (38, 48), PROX1 was not detected in vasa recta at both the transcript (Figure 2B) and protein level (Figure 2D). LYVE1, meanwhile, was expressed by macrophages (Figure 2E) as reported in mouse (49) and human kidneys (50), and it was also detected in glomerular (Figure 2F) and peritubular capillary endothelium (Supplemental Figure 1, D–E). To probe the phenotype of human kidney lymphatics, and whether the vessels we detected included smooth muscle–lined collecting vessels (51) as in the lungs (52) and skin and mesentery (53), we costained kidneys for PDPN and α-smooth muscle actin (ACTA2). Kidney lymphatics in both hilum and cortex lacked smooth muscle coverage (Figure 2, G and H). We corroborated this using subclustering analysis of our scRNA-Seq atlas, combining 295 LECs from healthy kidney with 157 additional cells from a recent study (54). This revealed 2 transcriptionally distinct LEC subclusters (Figure 2I), which expressed LEC capillary markers PROX1, PDPN, and CCL21 (Figure 2J) with sparse LYVE1 expression, consistent with our imaging data (Figure 1D). Only rare cells, not specific for either subcluster, expressed molecular markers of lymphatic valve endothelial cells, GATA2 and FOXC2 (Figure 2J). Differential expression analysis between the 2 capillary subclusters identified 129 DEGs (Supplemental Data 2). One subcluster was enriched for adipose signaling peptide neurotensin (NTS, log2FC = 3.60) (55), and the other expressed CCL2 (log2FC = 3.30), CXCL2 (log2FC = 4.22), and ICAM1 (log2FC = 4.15) (Figure 2K), indicative of capillaries involved in immune cell egress. To investigate if, akin to their blood vascular counterparts, kidney lymphatics possess an organ-specific signature (56, 57), we created a multiorgan human LEC atlas by integrating our 452 kidney LECs with scRNA-Seq data of LECs from other organs, including skin (n = 4,765 cells) (58), breast (n = 4,991) (59), heart (n = 432) (60, 61), lung (n = 1,891) (62), and small and large intestine (n = 462 and 471, respectively) (63). The final dataset encompassed 13,454 LECs from 19 anatomical sites (Supplemental Figure 3A). We resolved 5 transcriptionally distinct subclusters (Figure 3A and Supplemental Data 3). Four subclusters expressed lymphatic capillary markers CCL21 and LYVE1, while the fifth expressed lymphatic valve markers FOXC2 and integrin alpha 9 (ITGA9) (Supplemental Figure 3B) (64). Visceral organ–derived LECs (kidney, heart, lung, intestines) were predominantly grouped within 1 subcluster (LEC1), whereas breast lymphatics were found in LEC1, LEC2, and LEC3, and skin lymphatics in LEC2, LEC3, and LEC4 (Figure 3B). This spatial segregation was reflected in predicted transcription factor activity (Supplemental Figure 3C). (A) Integrated UMAP featuring 13,454 cells from a total of 7 human organs incorporating kidney, skin, breast, heart, lung, small intestine, and large intestine. Unsupervised clustering resulting in 5 transcriptionally distinct clusters of lymphatic cells, which we designate LEC1, LEC2, LEC3, and LEC4, all of which have capillary identity, and a fifth cluster representing valve LECs. (B) UMAPs highlighting the cells corresponding to each organ and where they are represented within the dataset. Based on this analysis, LEC1 and LEC2 are dominated by cells from visceral organs, including kidney, heart, lung, and intestines. Conversely, LEC3 and LEC4 are dominated by cells from superficial organs, the skin and breast tissue. All organs show cells mapping to valve LECs. (C) Heatmap showing the top 35 differentially expressed genes (DEGs) enriched in kidney lymphatic cells versus top 35 genes that have low expression by kidney lymphatics compared with other organs. (D) Dot plot of differentially expressed chemokines, interleukins, and immune trafficking receptors across lymphatics of different organs. (E) Expression of DNASE1L3 and MDK (F) at the RNA level in the tubulointerstitium of patients within the publicly available NephroSeq database. Number of patients per condition are shown as follows for DNASE1L3: healthy (n = 8), diabetic kidney disease (DKD, n = 11), focal segmental glomerulosclerosis (FSGS, n = 22), lupus nephritis (n = 31, **P = 0.0013), minimal change disease (MCD, n = 9), and MDK: healthy (n = 14), DKD (n = 10, ****P < 0.0001), FSGS (n = 18, ****P < 0.0001), lupus nephritis (n = 31, ****P < 0.0001), MCD (n = 5). For both genes, significance values represent increase relative to healthy samples. Comparative analysis (Supplemental Data 4) identified 118 DEGs upregulated in kidney LECs compared with other organs (Figure 3C). The top kidney lymphatic-enriched genes included DNASE1L3 (log2FC = 3.77, P = 3.24 × 10), the chemokine CCL14 (log2FC = 3.03, P = 7.00 × 10), the netrin receptor UNC5B (log2FC = 2.26, P = 9.65 × 10), the growth factor midkine (MDK, log2FC = 1.98, P = 5.56 × 10), and the anti-protease α2 macroglobulin (A2M, log2FC = 1.80, P = 4.00 × 10). Most of these genes were also expressed by blood endothelia in the kidney, whereas A2M was also expressed by stromal cells (Supplemental Figure 3D). Among the 251 DEGs with lower expression in kidney LECs compared with those from other organs (Figure 3D) were LYVE1 and the major neutrophil chemoattractant CXCL8 (65), the latter of which was also absent from heart, lung, and intestinal LECs. Conversely, LECs in these visceral organs expressed the alarmin cytokine IL33 (66), which was reduced in lymphatics of the skin and breast (Figure 3D and Supplemental Figure 3E). To provide pathological context to the kidney lymphatic DEGs, we examined their expression in NephroSeq, a gene expression database of kidney diseases. DNASE1L3 was significantly upregulated in the tubulointerstitium of patients with lupus nephritis (n = 31) compared with controls (n = 8, mean difference in log2 expression = 1.1 ± 0.34–1.9, P = 0.0013) (Figure 3E). Conversely, MDK was significantly upregulated in several inflammatory and metabolic kidney diseases, except for minimal change disease (Figure 3F). Collectively, our analyses demonstrated that kidney LECs have an organ-specific molecular profile, enriched for DNASE1L3, MDK, and CCL14, with reduced expression of canonical immune trafficking markers such as LYVE1 and CXCL8. Lymphangiogenesis has been observed during transplant rejection in both rodent models (23, 25, 31, 67) and humans (24, 26–28), but whether this is protective or promotes alloimmunity remains unclear. To investigate this in the human context, we profiled kidney transplants with chronic mixed rejection, a setting in which both donor-specific antibodies and T cells target HLA molecular expressed on tubular epithelial and blood endothelial cells. We analyzed 3 allografts with histological features consistent with chronic mixed rejection, including T cell– and antibody-mediated injury (Supplemental Table 2), and compared them with control kidneys obtained from nontransplanted donor organs. In rejecting allografts, the lymphatic vascular network exhibited marked disorganization, with loss of the hierarchical structure observed in controls (Figure 4A). Quantitative analysis revealed a 7-fold increase in mean lymphatic vessel density (95.12 ± 49.21 vs. 690.3 ± 121.6 vessels/mm, P = 0.0014), accompanied by reductions in the distribution of vessel lengths (median difference = 132 vs. 68.4 μm, P = 0.0001), vessel radius (9.05 vs. 4.9 μm, P < 0.0001), and branching angle (112 versus 103, P < 0.0001) (Figure 4B and Supplemental Videos 5 and 6). Notably, lymphatic vessels also infiltrated the allograft medulla, a region devoid of lymphatics in healthy kidneys (Figure 4, C and D). (A) 3D renderings of segmented lymphatic networks from donor kidneys and rejecting kidney allografts using LSFM; n = 3 samples per group. Vessel branch radii are color-coded: blue is smallest radius (<3.5 μm; asterisks) and red the largest (>18 μm; arrowheads). (B) Quantitative analysis of lymphatic branching architecture. Vessel metrics are shown per kidney (scatterplot, n = 3 per group) and pooled across vessels (violin plots, n = 75,036 vessels). Vessel density was significantly increased in rejection (95.12 ± 49.21 vs. 690.3 ± 121.6 vessels/mm, **P = 0.0014, unpaired t test). Vessel length, radius, and branching angle distributions were significantly shifted in rejection (****P < 0.0001 for each; Kolmogorov–Smirnov tests). (C and D) Confocal imaging of PDPN lymphatic vessels (arrowheads) in cortex adjacent to DBA tubules (C) and medulla adjacent to UMOD tubules (D), showing lymphatic expansion in cortex and infiltration into medulla. Representative of 6 regions across n = 3 kidneys/group. Scale bars: 200 μm (C), 100 μm (D). (E and F) 3D reconstruction of CDH5 lymphatic endothelial junctions in control (E) and rejecting (F) kidneys (n = 2 kidneys/group). Junctions identified within PDPN lymphatics using surface rendering in Imaris. Scale bars: 30 μm. Below: surface-rendered high-magnification views of lymphatic vessel blind ends from E and F, showing discontinuous CDH5 “button-like” junctions in controls and continuous “zipper-like” junctions in rejection. Scale bars: 4 μm (control), 10 μm (rejection). (G and H) Quantification of total PDPN lymphatic vessel volume per field (G) and density of discontinuous CDH5 junctions per mm³ of vessel volume (H). Each point represents a single image; circles, Repeat 1 and squares, Repeat 2. Rejecting kidneys showed increased lymphatic volume (mean difference = 2.01 × 10 ± 6.83 × 10 mm) and reduced density of discontinuous junctions (mean difference = 265,674 ± 73,557 discontinuous junctions per mm). LEC-cell junctions are key regulators of immune cell trafficking. In homeostasis, these junctions form discontinuous “button-like” structures that facilitate leukocyte entry into lymphatics, whereas during chronic inflammation, they transition into continuous “zipper-like” formations that impair lymphatic drainage (68–70). Given the accumulation of infiltrating lymphocytes in chronically rejecting grafts (71–73), we hypothesized that altered lymphatic junctional architecture might be a feature of rejection. To assess this, we immunostained for vascular endothelial cadherin (CDH5), a key component of endothelial junctions (Supplemental Video 7), and used PDPN to distinguish lymphatics from blood vessels (Supplemental Figure 4). Discontinuous CDH5 LEC junctions were quantified in both control (Figure 4E) and chronic rejection (Figure 4F) samples, and values were normalized to total lymphatic network volume (Figure 4G). We observed a reduction in disconnected (button-like) junctions in rejecting allografts compared with controls (Figure 4H, mean difference = 2.7 × 10 ± 7.3 × 10 CDH5 junctions per mm lymphatic vessel), consistent with a shift toward a zipper-like configuration. Given the structural perturbation of kidney lymphatics in rejecting allografts, we next examined their spatial relationship to organized immune responses within chronic rejection. A hallmark of alloimmunity is the formation of tertiary lymphoid structures (TLSs), ectopic lymph node–like aggregations of T cells and B cells, where follicular DCs and high endothelial venules (HEVs) also develop. TLSs facilitate local antigen presentation and lymphocyte activation, and they have been associated with progressive graft injury and dysfunction (74–79). Using triple immunolabeling, we found PDPN lymphatics were observed close to CD4 T cell– and CD20 B cell–rich aggregates (Figure 5A) in 3 rejecting allografts, consistent with previous reports (26–28). To assess the relationship between lymphatics and TLS maturation, we examined PDPN lymphatics relative to CD21 follicular DCs and peripheral lymph node addressin (PNAd) HEVs, the latter serving as a marker of mature TLS (67, 80, 81). The lymphatic network in rejecting allografts interconnected multiple mature TLSs containing HEVs (Figure 5B and Supplemental Video 8). Such connections were not detected between CD31 vessels (Supplemental Figure 5A). Spatiotemporal analysis revealed that all identified TLSs were near lymphatic vessels (Figure 5C) (n = 9/9, 100%), whereas only half contained HEVs (n = 5/9, 55.6%, P = 0.023). In mature TLSs with HEVs, PDPN LECs were significantly closer to the TLS core than HEVs (Figure 5D, mean distance = 49.53 ± 23.83 μm vs. 109.6 ± 25.13 μm, 95% CI = 24.33–95.76, P = 0.0047), suggesting that lymphatics are an early feature of TLS organization. (A) Representative segmented confocal images of PDPN lymphatics (white arrowhead), CD20 B cells, and CD4 T cells in regions with evidence of ectopic lymphoid aggregation. A tertiary lymphoid structure (TLS) is shown (white asterisk). Representative image of 4 T cell– and B cell–rich TLSs taken from n = 2 rejecting allografts. Scale bar: 40 μm. (B) 3D rendering of TLS interconnected by lymphatics. Such interconnections (white arrowhead) were observed between TLSs in all (n = 3) rejecting allografts imaged. (C) Representative segmented confocal images of TLS, containing PDPN lymphatics (white arrow), CD21 follicular DCs (FDCs) and peripheral lymph node addressin (PNAd) high endothelial venules (HEVs). Nine TLSs were imaged across n = 3 rejecting allografts. Each image represents TLSs at different stages, with either HEVs absent (early stage; top image), scant (mid-stage; middle image), or present (late-stage, bottom image). Scale bar: 50 μm. (D) Comparison of distance between the CD21 FDC core and lymphatic vessel (green) or HEVs (orange), with each data point representing an individual TLS imaged. Circles represent Repeat 1, squares Repeat 2, and triangles Repeat 3. Lymphatic vessels were significantly closer to CD21 FDCs than HEVs (mean difference = 60.04, 95% CI = 24.33–95.76, **P = 0.0047, unpaired t test). To explore lymphatic-lymphocyte relationships beyond defined TLS regions, we performed 3D imaging and spatial quantification of PDPN lymphatics relative to CD20 B cells and CD4 T cells (Figure 6, A and B, and Supplemental Video 9). Intraluminal CD20 B cell density was reduced by half in rejecting allografts compared with controls (Figure 6C), although total B cell numbers were equivocal, suggesting that this reflects increased lymphatic volume rather than changes in B cell abundance. In contrast, total intraluminal CD4 T cells increased in rejecting kidneys, with a 3-fold increase in CD4 T cell density (Figure 6D) relative to controls, a markedly higher density than was detected in the surrounding allograft parenchyma. (A and B) Segmented (A) and rendered (B) confocal images of PDPN lymphatics (white arrow), CD20 B cells (yellow asterisk), and CD4 T cells (white asterisk). In B, the transparency of rendered lymphatics is increased to visualize intraluminal B cells and T cells. Scale bars: 30 μm. (C and D) Number of intraluminal CD20 B cells (C) or CD4 T cells (D), normalized by volume, was quantified and compared with that of the tissue parenchyma. Each point represents 1 volume of interest imaged, with circles representing Repeat 1 and squares representing Repeat 2. Luminal CD20 B cell density was higher than that of the tissue parenchyma in both control kidneys and rejecting allografts. A similar trend was observed for intraluminal CD4 T cells, with a greater magnitude in increase in density within rejection. (E and F) Spatial point-pattern of perilymphatic CD20 cell (E) or CD4 cell (F) density, where lymphatic branch points represent gray dots and CD20 cells are color-coded according to their density around the lymphatic network. (G and H) Histograms of CD20 cell (G) or CD4 T cell (H) frequency as a function of distance from the nearest lymphatic vessel. P values demonstrate whether lymphocytes are clustered around lymphatics greater than would be expected under complete spatial randomness. The only significant association observed was between CD4 T cells and lymphatics in donor kidneys (*P = 0.029). All imaging data are representative of n = 5 imaging volumes, each acquired from n = 2 allografts with chronic mixed rejection and n = 2 donor controls. To further assess how lymphocyte position relative to lymphatics is altered in rejection, we performed spatial statistical analysis, by computing a normalized distance metric for each B cell (Figure 6E) and T cell (Figure 6F) to its nearest lymphatic vessel, and comparing this to a null model of random spatial distribution (82). CD20 B cells showed no significant spatial association with lymphatics in either control kidneys (n = 703 cells; P = 0.631) or rejecting allografts (n = 2,963 cells; P = 0.326) (Figure 6G). However, CD4 T cells (n = 2,149 cells across 2 controls) had a peak distribution within 0–100 μm from the nearest lymphatic vessel and were significantly enriched near lymphatic vessels compared with a random distribution (P = 0.029). This association was lost in rejecting allografts (n = 4,382 cells, P = 0.699) (Figure 6H), indicating disrupted T cell–lymphatic proximity in the context of chronic rejection. Having established that lymphatics are structurally perturbed and spatially associated with immune aggregates in chronic rejection, we next investigated whether LECs in this setting exhibit an altered molecular profile. To do this, we first performed comparative transcriptomic analysis of LECs from healthy kidneys, rejection, and CKD (Supplemental Data 5–7). GO revealed that LECs from rejecting allografts were enriched for pathways related to the negative regulation of viral process (GO:0048525, fold-enrichment = 90.26, FDR = 5.95 × 10), including IFN-induced transmembrane proteins IFITM2 (log2FC = 1.76, P = 5.89 × 10) and IFITM3 (log2FC = 1.62, P = 6.86 × 10) (Figure 7A). IFN-γ was specifically enriched in T cells and NK cells in our scRNA-Seq dataset (Figure 7B), whereas other IFN types were not detected. We then examined an IFN-γ response signature — including levels of IFITM2, IFITM3, and the IFN-γ receptor subunits IFNGR1 and IFNGR2 — which was prominent in LECs and in blood endothelial cells and macrophages from rejecting allografts (Figure 7C). To contextualize this response, we compared the LEC profile in chronic rejection with that of HEVs, identified by enrichment for PNAd (NTAN1) and downregulation of Notch pathway genes RBPJ and JAG1 (Supplemental Figure 5B) (83, 84). Unlike LECs, HEVs lacked lymphatic markers PROX1 and PDPN (Supplemental Figure 5C). Instead, they expressed transcripts involved in leukocyte recruitment, activation, and regulation, such as CXCL16, fractalkine (CX3CL1), CD40, and IL-32 (Supplemental Figure 5D and Supplemental Data 8), highlighting a distinct immune regulatory profile compared with LECs. (A) Violin plots showing upregulation of IFN-inducible genes IFITM2 and IFITM3 in LECs from rejecting allografts. (B) UMAP of the scRNA-Seq data showing enrichment of IFN-γ within the T/NK cell cluster. (C) UMAP showing enrichment of an IFN-γ signature, including IFNGR1, IFNGR2, IFITM2, and IFITM3. (D) CellPhoneDB interaction map depicting predicted lymphatic-CD4 T cell crosstalk in rejection. Inhibitory interactions (blue) include PVR and LGALS9; stimulatory interactions (red) are also shown. Node size reflects expression frequency; line intensity indicates interaction strength. Ligands of interest, PVR and LGALS9, are highlighted. (E) Heatmap of immune checkpoint interactions between LECs and effector CD4 T cells across disease states. Color indicates normalized CellPhoneDB interaction score. All scores were normalized for each ligand-receptor pair. (F) Immunofluorescence validation of PVR expression on PDPN lymphatics (arrowhead) in rejecting kidneys (n = 2); CD4 T cell shown in contact (asterisk). Scale bar: 30 μm. (G) IFN-γ stimulation of cultured human LECs increases LGALS9 levels at 24 and 48 hours (qPCR; ***P = 0.0002, **P = 0.0093, respectively) relative to HPRT. (H) LGALS9 protein secretion increased at 48 and 72 hours (ELISA; ***P = 0.0002, ****P < 0.0001, respectively) after IFN-γ stimulation of cultured human LECs. qPCR and ELISA experiments were repeated 3 times, and all assays were performed in duplicate, with each dot on the graph representing the mean data obtained for each repeat. We next explored potential ligand-receptor interactions between LECs and lymphocytes using CellPhoneDB (85). Predicted cell-cell communication was highest in rejecting kidneys compared with CKD or healthy controls (Supplemental Figure 6A), with most interactions occurring between LECs and T cell subsets (Supplemental Figure 6B). These included IFN-γ–IFNGR signaling from CD8 T cells to LECs across both control and rejecting kidneys (Supplemental Figure 6C). Chemokine-based interactions included established axes such as CCL21, CCL2, and ACKR2 (Supplemental Figure 7A), although CCL14/ACKR2 signaling with CD4 effector T cells was reduced in rejection. Many chemokine receptors for ACKR2 ligands, including CCR2, CCR5, and CCR7, were expressed by T cells (Supplemental Figure 7B). Notably, most of the remaining predicted interactions were coinhibitory in nature. These included LEC expression of poliovirus receptor (PVR) and galectin 9 (LGALS9), which suppress effector T cell responses via TIGIT and HAVCR2 signaling, respectively (86) (Figure 7D). While also present in CKD and non-alloimmune graft injury (Supplemental Figure 8, A–C), these interactions had higher signaling scores in chronic rejection (Figure 7E). Immunostaining confirmed PVR expression on PDPN lymphatics in direct contact with CD4 T cells in rejecting allografts (Figure 7F). When stimulated by IFN-γ, blood endothelia express PVR and LGALS9 to dampen T cell responses (87, 88). To examine whether this was the case for LECs, we stimulated a human LEC line with recombinant IFN-γ. LGALS9 transcripts were significantly upregulated after 24 hours (mean FC = 9.05, 95% CI = 5.37–12.73, adjusted P = 0.0002) and remained elevated at 48 hours (mean FC = 5.10, 95% CI = 1.42–8.78, adjusted P = 0.0093) (Figure 7G). Corresponding increases in LEC-secreted LGALS9 protein were observed at 48 hours (difference in mean concentration = 5.54 ng/mL, 95% CI = 3.26–7.83, adjusted P = 0.0002) and 72 hours (difference in mean concentration = 16.87 ng/mL, 95% CI = 14.58–19.16, adjusted P < 0.0001) (Figure 7H), confirming that LECs can acquire a coinhibitory profile in response to IFN-γ exposure. However, in solid organ transplantation, IFN-γ–induced expression of HLAs on endothelial cells can facilitate alloantigen presentation and antibody binding to donor vasculature (89, 90). Similarly, we found rejected allograft LECs also expressed HLA-DP and HLA-DR (Figure 8A). To determine whether lymphatics were of donor or recipient origin, we assessed genotype using single-nucleotide variant calling, and found a majority of LECs were donor derived, with a small recipient cell contribution (n = 3/247, 1.2%) (Figure 8B), consistent with a previous study of sex-mismatched renal allografts (91). Immunostaining for HLA-DR in chronic rejection (Figure 8C) demonstrated its expression on CD31 blood endothelial cells (Figure 8D), CD68 macrophages (Figure 8E) (92, 93), and PDPN lymphatics (Figure 8, F and G). Importantly, we detected complement factor C4d deposition, a histological hallmark of alloantibody-mediated complement activation, on PDPN lymphatic vessels in 2 rejecting allografts from patients with de novo donor-specific antibodies (Figure 8H). These HLA-DR lymphatic regions were surrounded by CD3 T cells (Supplemental Video 10), suggesting coordinated alloantibody and T cell engagement. Together, these data demonstrate that LECs in chronic rejection acquire an IFN-γ–responsive, immune-inhibitory transcriptional phenotype, marked by coinhibitory ligand expression, HLA class II upregulation, and evidence of complement activation. (A) Dot plot of the expression of transcripts encoding MHC class II molecules within lymphatics in the dataset. (B) Single nucleotide variant–based analysis of the origin of lymphatics in allograft tissues from the scRNA-Seq atlas. Cells are grouped by control, chronic rejection, or alternative causes of graft injury. (C) Representative optical z-sections from control and chronically rejecting renal tissue stained for HLA-DR. Isolated, discrete HLA-DR cells are shown with asterisks in both conditions, whereas in rejection there is also vascular staining (white arrowheads). Representative of 3 nonoverlapping fields of view per kidney, imaged across n = 2 kidneys per group. (D–F) 3D confocal images of HLA-DR expression (arrowheads) in CD31 endothelia (D), CD68 macrophages (E), and PDPN lymphatics (F). Images are representative of 5 regions imaged across n = 2 kidneys with chronic transplant rejection. All scale bars: 30 μm. (G) Representative 3D reconstructions of n = 2 transplant donor kidney tissues and n = 2 allograft tissues with chronic rejection stained using D2-40 and HLA-DR antibody. The HLA-DR signal is masked by D2-40 expression, such that only the signal inside lymphatics is visible. HLA-DR expression is observed in rejection (see white arrowheads). Three nonoverlapping fields of view per kidney were imaged. Scale bar: 50 μm. (H) 3D confocal images of C4d deposition, representative of 5 regions imaged across n = 2 kidneys with chronic transplant rejection. C4d deposition is observed in PDPN lymphatics (arrowheads) and presumptive blood capillaries (asterisks). Scale bar: 30 μm. Lymphatic vessels play a central role in maintaining fluid balance and immune homeostasis, yet their structural and molecular features in the human kidney remain underexplored. This gap is clinically relevant, as lymphangiogenesis occurs across a range of kidney diseases (11–14), and augmenting lymphatic function confers therapeutic benefit in preclinical models of kidney disease (94–96), hypertension (97–99), and acute kidney transplant rejection (29). Here, we combined 3D imaging of optically cleared tissue with scRNA-Seq to resolve the spatial architecture and molecular identity of lymphatics in the healthy human kidney and to interrogate their remodeling in chronic transplant rejection. Although previous studies have identified lymphatics in the kidney hilum and cortex (11–14), our 3D imaging approach yielded potentially new spatial insights, including a hierarchical arrangement of kidney lymphatics and the initiation of blind ends near proximal and distal tubular nephron segments, key sites of reabsorption and solute exchange between the urinary filtrate and blood. Using scRNA-Seq, we defined a transcriptional census of human kidney LECs, identifying expression of molecules previously characterized in other lymphatic beds but not in human kidney LECs, such as FABP4 (100, 101) and ANGPT2 (102–104). A recent analysis has transcriptionally profiled a population of LECs in the lymph node (105). Our findings further extend the evidence for organ-specific heterogeneity of human lymphatics. Compared with lymphatics from barrier tissues such as skin, lung, and intestines, kidney LECs displayed reduced expression of genes encoding classical immune trafficking molecules like CXCL8 and LYVE1, the latter confirmed at the protein level and also recently corroborated in mouse kidneys (106). Instead, kidney LECs express a repertoire of other molecules, including DNASE1L3, a molecule involved in extracellular DNA clearance and deficiency of which is implicated in lupus nephritis (107–109). Such findings could suggest tissue-specific adaptations of the lymphatic regulation of immunity and may inform future studies of immune-mediated kidney disease. Although lymphatic valve markers were sparsely detected, unlike in mouse kidneys (110), we identified transcriptional heterogeneity among kidney LECs, including a subpopulation enriched for CCL2 and CXCL2. This is reminiscent of molecularly distinct and immune-interacting LEC subsets in the nasal mucosa (111, 112) and dermis (113). This heterogeneity may arise, in part, from microenvironment signals, such as IFN-γ, which drive context-dependent reprogramming of LECs in inflammation or cancer (114–116). We show that LECs upregulate PVR and LGALS9 in response to IFN-γ, echoing responses in the blood endothelium (87, 88) and supporting a paradigm in which the behavior of lymphatics is actively shaped by their surrounding milieu. In kidney transplantation, lymphatics have been associated with improved graft survival, possibly through increased leukocyte clearance (27–29), but also with immune activation and fibrosis (23, 25, 31, 117, 118). Our findings challenge the notion that lymphangiogenesis is uniformly pathogenic. Although we observed lymphangiogenesis and proximity of these vessels to TLSs in rejecting allografts, we showed that allograft LECs acquire a tolerogenic transcriptional program driven by IFN-γ. LEC-derived immune-inhibitory ligands dampen effector T cell function in cancer (119, 120), neuroinflammation (121), and infection (69), and we confirmed the expression of 2 exemplar molecular candidates, PVR and LGALS9, at both the transcript and protein level. However, this tolerogenic molecular program coincides with structural perturbations to allograft lymphatics. In rejection, lymphatics exhibited loss of hierarchical organization, infiltration into the medulla, and transformation of cell-cell junctions from button- to zipper-like morphology, changes known to impair fluid and cell transport (68–70). Building on previous studies in kidney (26, 27) and other inflammatory contexts (80, 122, 123), we identified TLSs of varying maturity positioned along lymphatic networks. Given the potential for in situ antigen presentation and T cell activation within the TLS (75, 77, 78, 124–127), and given the observed altered localization of CD4 T cells within and around lymphatic vessels, it is tempting to speculate that lymphatic perturbation may contribute to CD4 T cell retention within allografts, heralding the formation and maintenance of the TLS in chronic rejection. Additionally, we demonstrate that allograft LECs express HLA class II and show C4d deposition in patients with de novo donor-specific antibodies, consistent with alloantibody targeting and complement activation. Analogous injury to the blood vasculature (19) is well-characterized in transplant pathology (24), and donor lymphatics may thus represent a previously underappreciated target of alloimmune responses. This study has several limitations. First, our 3D imaging was cross-sectional and included a small number of fixed samples, restricting inference of dynamic events during transplant rejection. Second, and common to all scRNA-Seq studies of human tissues, our control tissues were derived from nontransplanted kidneys and tumour nephrectomies and are thus likely subject to inflammatory changes. We attempted to mitigate this by using samples with histological evidence of minimal chronic damage. Third, although we identified expression of coinhibitory ligands and evidence of alloantibody binding of kidney lymphatics, the downstream consequences on alloimmunity and graft function require further mechanistic study, which is challenging given the absence of an animal model that mimics the long-term sequalae of chronic mixed rejection, which occurred in our cohort of patients over decades to years, while enabling simultaneous genetic or pharmacological manipulation of LECs in a targeted manner. Together, our data provide a comprehensive and multimodal view of the lymphatic vasculature in human kidney health and rejection. We propose that lymphatics acquire a tolerogenic, IFN-γ–driven phenotype during chronic rejection, but this is accompanied by structural disorganization and immune-associated perturbations. These findings point to a potentially new perspective on the role of lymphatic remodeling in transplantation, featuring a tolerogenic profile yet subject to alloimmune injury. This work lays the foundation for future studies exploring kidneys in health and disease and opens new avenues for therapeutic targeting of the lymphatic vasculature to improve the longevity of kidney transplants. Given the exploratory nature of 3D imaging and scRNA-Seq performed in this study and the limited kidneys available for 3D imaging analysis, sex was not considered as a biological variable. Human kidney tissue was fixed in 4% paraformaldehyde in PBS at 4°C overnight and stored in PBS with 0.02% sodium azide. A modified SHANEL protocol (128) was used for whole-mount immunolabeling, followed by optical clearing in benzyl alcohol/benzyl benzoate (1:2). Imaging was performed using an LSM880 upright confocal microscope (Zeiss) or custom-built mesoscale selective plane illumination microscope (mesoSPIM) (129). Image segmentation and 3D reconstruction were carried out in Imaris and Amira. Binarized lymphatic networks were skeletonized in Fiji using BoneJ (130). CD4 T cell and CD20 B cell counts, centroids, and areas were obtained using 3D Objects Counter with no further preprocessing (131). The mean distance of each cell from the nearest point of the lymphatic network (d) was calculated using the cross-product 3D point-line distance: (Equation 1) where x1 and x2 are the 2 closest adjacent nodes from the lymphatic 3D skeleton, found by minimizing cross–nearest neighbor distances, and x0 is the centroid of the cell of interest. To evaluate whether the cell distances were different from what would be expected by chance, within each region of interest, the CD4 T cell and CD20 B cell populations were randomly redistributed under complete spatial randomness for 20 simulations. A comparison was then made as to whether the measured mean cell-lymphatic distances fell within the 95% CIs obtained through the simulations under complete spatial randomness. Single-cell suspensions from fresh kidney explants were processed using the 10x Genomics Chromium 5′v2 kit and sequenced on an Illumina NovaSeq. Data were mapped to GRCh38 and processed using Scanpy and Seurat, using scVI (132) or Harmony (133) for integration. Cell identity was assigned via marker gene expression and assisted by CellTypist prediction. Differential expression was assessed using Wilcoxon rank-sum tests and GO term enrichment using PANTHER. To infer putative cell-cell interactions in scRNA-Seq data, the CellPhoneDB resource (85) was used. To generate the human lymphatic cell atlas, LECs were extracted from publicly available single-cell datasets across multiple organs and integrated using Harmony. SCENIC (134) was used to infer transcription factor activity across clusters. The NephroSeq database (v5, RRID:SCR_019050) was used to examine candidate genes by pulling data from its online browser. Adult human dermal LECs (PromoCell, C-12217) were cultured in MV2 medium and treated with recombinant human IFN-γ (50 ng/mL) or unstimulated control medium for 24, 48, or 72 hours. LGALS9 transcript levels were quantified by qRT-PCR and normalized to HPRT using the 2 method. Secreted LGALS9 protein in conditioned media was measured by ELISA (R&D Systems). Data are shown as fold-change relative to untreated controls. Assays were performed across 2 independent cell lines in triplicate. Statistical analyses were performed using GraphPad Prism unless otherwise specified. Data normality and variance were assessed using Shapiro-Wilk and Brown-Forsythe tests, respectively. For normally distributed data, comparisons between 2 groups used 2-tailed Student’s t test and 1-way ANOVA with Bonferroni’s post hoc tests for multiple groups. A P value less than 0.05 was considered significant. Data are presented as mean ± SD, with SEM shown for graphical error bars. Statistical methods for scRNA-Seq and spatial analyses are described separately. Use of human tissue was approved by NHS Blood & Transplant (NHSBT), the National Research Ethics Committee in the UK (21/WA/0388, NC.2018.010, NC.2018.007, REC 16/EE/0014), and the Royal Free London NHS Foundation Trust-UCL Biobank Ethical Review Committee (RFL B-ERC/B-ERC-RF, NC.2018.010; IRAS 208955). Written informed consent for research use of donated organs was obtained via NHSBT. Ethical approvals for public datasets are detailed in the original studies. Raw sequencing data for the 5 new human kidney scRNA-Seq samples have been made publicly accessible via the European Genome-phenome Archive (accession EGAD00001015631). Processed Seurat and h5ad files are available at Zenodo (https://doi.org/10.5281/zenodo.7566982). Code for data analysis is available at GitHub (https://github.com/daniyal-jafree1995/). Imaging data are available upon reasonable request. All raw data used to plot graphs, except for scRNA-Seq analyses, are provided within the Supporting Data Values file. Full experimental details are provided in the Supplemental Methods, including reagents and protocols, in addition to the steps involved in computational analysis. The authors thank Mark Lythgoe (UCL), Alan Salama (UCL), René Hägerling (Charité Universitätsmedizin Berlin), Joaquim Vieira (Kings College London), and Jeremy Hughes (University of Edinburgh) for helpful discussions and support. We also thank Kelvin Tuong (University of Queensland) for advising on ligand-receptor interaction analysis. Explantation was performed by the Renal Transplant Surgery team at the Royal Free Hospital, London, UK. Confocal imaging was supported by the Light Microscopy Core Facility at UCL GOSICH and LSFM imaging by the UK Research and Innovation (UKRI) Dementia Research Institute (DRI) through UK DRI Ltd (UKDRI-1208), principally funded by the UK Medical Research Council (MRC). This work was supported by Kidney Research UK (IN_012_20190306), a Rosetrees Trust PhD Plus Award (PhD2020\100012), a Foulkes Foundation Fellowship, and a Wellcome Trust Accelerator Award (314710/Z/24/Z) to DJJ. DJJ is also supported by the Specialised Foundation Programme at the East of England NHS Deanery. MRC was supported by an NIHR Professorship (RP-2017-08-ST2-002), by the NIHR Blood and Transplant Research Unit in Organ Donation (NIHR203332), by the NIHR Cambridge Biomedical Research Centre (NIHR203312) and a Wellcome Investigator Award (220268/Z/20/Z). MAB is supported by a UKRI Future Leaders Fellowship (MR/X011038/1). ASW acknowledges support from the MRC-National Institute for Health and Care Research (NIHR) Rare Disease Research Platform MR/Y008340/1 and MRC project grant APP14742. DAL was also supported by a Wellcome Trust Investigator Award (220895/Z/20/Z), the NIHR Biomedical Research Centre at Great Ormond Street Hospital for Children NHS Foundation Trust. In-Press Preview Electronic publication Facilities and equipment LSFM imaging by the UKRI Dementia Research Institute Accelerator Award to DJ Project grant to ASW Conflict of interest: The authors have declared that no conflict of interest exists. Copyright: © 2025, Jafree et al. This is an open access article published under the terms of the Creative Commons Attribution 4.0 International License. Reference information: J Clin Invest. 2025;135(18):e168962.https://doi.org/10.1172/JCI168962. Daniyal J. Jafree, Email: daniyal.jafree.13@ucl.ac.uk. Benjamin J. Stewart, Email: bs567@cam.ac.uk. Karen L. Price, Email: k.price@ucl.ac.uk. Maria Kolatsi-Joannou, Email: m.joannou@ucl.ac.uk. Camille Laroche, Email: camille.laroche@umontreal.ca. Barian Mohidin, Email: barian.mohidin.23@ucl.ac.uk. Benjamin Davis, Email: b.m.davis@ucl.ac.uk. Hannah Mitchell, Email: H.Mitchell@qub.ac.uk. Lauren G Russell, Email: lauren.russell.19@ucl.ac.uk. Chun Jing Wang, Email: c.wang@ucl.ac.uk. William J Mason, Email: w.mason@ucl.ac.uk. Byung Il Lee, Email: byungil.lee@ucl.ac.uk. Lauren Heptinstall, Email: lauren.heptinstall@nhs.net. Ayshwarya Subramanian, Email: as3894@cornell.edu. Gideon Pomeranz, Email: gideon.pomeranz.16@ucl.ac.uk. Dale Moulding, Email: d.moulding@ucl.ac.uk. Laura Wilson, Email: laura-wilson@ucl.ac.uk. Tahmina Wickenden, Email: tahmina.aktar.19@ucl.ac.uk. Saif N. Malik, Email: saif.malik.17@ucl.ac.uk. Natalie Holroyd, Email: natalie.holroyd.16@ucl.ac.uk. Claire L. Walsh, Email: c.walsh.11@ucl.ac.uk. Jennifer C. Chandler, Email: j.chandler@ucl.ac.uk. Kevin X. Cao, Email: k.cao@ucl.ac.uk. Paul J.D. Winyard, Email: pwinyard@ich.ucl.ac.uk. Adrian S. Woolf, Email: adrian.woolf@manchester.ac.uk. Marc Aurel Busche, Email: m.busche@ucl.ac.uk. Simon Walker-Samuel, Email: simon.walkersamuel@ucl.ac.uk. Lucy S.K. Walker, Email: lucy.walker@ucl.ac.uk. Tessa Crompton, Email: t.crompton@ucl.ac.uk. Peter J. Scambler, Email: p.scambler@ucl.ac.uk. Reza Motallebzadeh, Email: r.motallebzadeh@ucl.ac.uk. Menna R. Clatworthy, Email: mrc38@cam.ac.uk. David A. Long, Email: d.long@ucl.ac.uk. Raw sequencing data for the 5 new human kidney scRNA-Seq samples have been made publicly accessible via the European Genome-phenome Archive (accession EGAD00001015631). Processed Seurat and h5ad files are available at Zenodo (https://doi.org/10.5281/zenodo.7566982). Code for data analysis is available at GitHub (https://github.com/daniyal-jafree1995/). Imaging data are available upon reasonable request. All raw data used to plot graphs, except for scRNA-Seq analyses, are provided within the Supporting Data Values file. Full experimental details are provided in the Supplemental Methods, including reagents and protocols, in addition to the steps involved in computational analysis.
PMC11578878
An integrated transcriptomic cell atlas of human neural organoids
Human neural organoids, generated from pluripotent stem cells in vitro, are useful tools to study human brain development, evolution and disease. However, it is unclear which parts of the human brain are covered by existing protocols, and it has been difficult to quantitatively assess organoid variation and fidelity. Here we integrate 36 single-cell transcriptomic datasets spanning 26 protocols into one integrated human neural organoid cell atlas totalling more than 1.7 million cells. Mapping to developing human brain references shows primary cell types and states that have been generated in vitro, and estimates transcriptomic similarity between primary and organoid counterparts across protocols. We provide a programmatic interface to browse the atlas and query new datasets, and showcase the power of the atlas to annotate organoid cell types and evaluate new organoid protocols. Finally, we show that the atlas can be used as a diverse control cohort to annotate and compare organoid models of neural disease, identifying genes and pathways that may underlie pathological mechanisms with the neural models. The human neural organoid cell atlas will be useful to assess organoid fidelity, characterize perturbed and diseased states and facilitate protocol development.Human neural organoids, self-organizing three-dimensional human neural tissues grown in vitro, are becoming powerful tools for studying the mechanisms of human brain development, evolution and disease. They can be generated using external patterning factors (for example, morphogens) to guide their development towards certain brain regions or to drive the emergence of specific cell types (guided protocols). Conversely, unguided protocols rely on the self-patterning capacity of organoids to generate diverse cell types and states. Single-cell RNA sequencing (scRNA-seq) is a powerful technology to characterize cell type heterogeneity in complex tissues, and has illuminated a remarkable heterogeneity of diverse progenitor, neuronal and glial cell types that can develop within neural organoids, as well as differentiation trajectories of certain neural lineages. The data also enable the comparison of human neural organoid cells to those in the primary human brain, and most analyses have revealed strong similarity in molecular signatures. Substantial differences have also been reported, including differential gene expression linked to media components and perturbed metabolic signatures associated with glycolysis. Nevertheless, analysis of organoid tissues supports a useful recapitulation of early brain development, and scRNA-seq methods have been applied to study the molecular basis of neural cell type fate determination, evolutionary differences in primates and pathological changes in neural disorders. However, it is unclear which portions of the developing central nervous system can be generated with existing protocols and which ones are still lacking. It has also remained challenging to systematically quantify the transcriptomic fidelity of neural organoid cells compared to their primary counterparts. In this study, we address these challenges by combining 36 scRNA-seq datasets covering numerous human neural organoid protocols into an integrated transcriptomic cell atlas. We establish an analytical pipeline that allows for the comprehensive and quantitative comparison of the organoid atlas to reference atlases of the developing human brain. We harmonize annotations of cell populations in the primary and organoid systems, estimate the capacity and precision of different neural organoid protocols to generate different brain regions, and identify primary cell populations that are under-represented in neural organoids. We estimate transcriptomic fidelity of neurons in neural organoids, and identify previously described cell stress as a universal factor distinguishing metabolic states of in vitro neurons from primary neurons without strongly affecting core identities of neuronal cell types. We map the data of a neural organoid morphogen screen to the integrated atlas to assess regional specificity and generation of new states. We also collect 11 scRNA-seq datasets modelling 10 different neural diseases, and map the integrated data to the neural organoid atlas for cell type annotation and differential expression (DE) analysis. Finally, we show that the atlas can be expanded by projecting new data to the current atlas. Together, our work provides a rich resource and a new framework to assess the fidelity of neural organoids, characterize perturbed and diseased states and streamline protocol development. To build a transcriptomic human neural organoid cell atlas (HNOCA), we collected scRNA-seq data and detailed, harmonized technical and biological metadata from 36 datasets, including 34 published and two as yet unpublished ones (Supplementary Table 1), accounting for 1.77 million cells after consistent preprocessing and quality control (Fig. 1a). The HNOCA represents cell types and states generated with 26 distinct neural organoid differentiation protocols, including three unguided and 23 guided ones, at time points ranging from 7 to 450 days (Fig. 1b). To remove batch effects, we implemented a three-step integration pipeline. First, we projected the HNOCA to a single-cell atlas of the developing human brain using reference similarity spectrum (RSS). Then, we developed snapseed (Methods) to perform preliminary marker-based hierarchical cell type annotation. Last, we used scPoli for label-aware data integration based on the snapseed annotations. Evaluation of different integration approaches using a previously established benchmarking pipeline showed that scPoli had the best performance for these datasets (Extended Data Fig. 1). We performed clustering on the basis of the scPoli representation and annotated clusters on the basis of canonical marker gene expression, organoid sample age and the auto-generated cell type labels. A uniform manifold approximation and projection (UMAP) embedding highlighted three neuronal differentiation trajectories corresponding to dorsal telencephalic, ventral telencephalic and non-telencephalic populations as well as trajectories leading from progenitors to glial cell types such as astrocytes and oligodendrocytes precursors (Fig. 1c–e and Extended Data Fig. 2). Cells from both unguided and guided protocols were distributed across all trajectories (Fig. 1f).Fig. 1Integrated HNOCA.a, Overview of HNOCA construction pipeline. b, Metadata of biological samples included in HNOCA. c–f, UMAP of the integrated HNOCA, coloured by level 2 cell type annotations (c), gene expression profiles of selected markers (d), sample ages (e) and differentiation protocol types (f). g, Proportions of cells assigned to different cell types in the HNOCA. Every stacked bar represents one biological sample, grouped by datasets and ordered by increasing sample ages. Top bars show 36 datasets, organoid differentiation protocols, protocol types. Bottom bars show the sample age. h, UMAP of the integrated HNOCA coloured by top-ranked diffusion component (DC1) on the real-time-informed transition matrix between cells. The stream arrows visualize the inferred flow of cell states toward more mature cells. i, Marker gene expression profiles along cortical pseudotime. j, UMAP of non-telencephalic neurons, coloured and labelled by clusters. k, Heatmap showing relative expression of selected genes across different non-telencephalic neuron clusters. Coloured dots show cluster identities as shown in j. Cb, cerebellum; ChP, choroid plexus; CP, choroid plexus; Hy, hypothalamus; max., maximum; MB, midbrain; MH, medulla; min., minimum; Oligo, oligodendrocyte; OPC, oligodendrocyte progenitor cell; PSC, pluripotent stem cell; telen., telencephalon; Th, thalamus; vTelen, ventral telencephalon. a, Overview of HNOCA construction pipeline. b, Metadata of biological samples included in HNOCA. c–f, UMAP of the integrated HNOCA, coloured by level 2 cell type annotations (c), gene expression profiles of selected markers (d), sample ages (e) and differentiation protocol types (f). g, Proportions of cells assigned to different cell types in the HNOCA. Every stacked bar represents one biological sample, grouped by datasets and ordered by increasing sample ages. Top bars show 36 datasets, organoid differentiation protocols, protocol types. Bottom bars show the sample age. h, UMAP of the integrated HNOCA coloured by top-ranked diffusion component (DC1) on the real-time-informed transition matrix between cells. The stream arrows visualize the inferred flow of cell states toward more mature cells. i, Marker gene expression profiles along cortical pseudotime. j, UMAP of non-telencephalic neurons, coloured and labelled by clusters. k, Heatmap showing relative expression of selected genes across different non-telencephalic neuron clusters. Coloured dots show cluster identities as shown in j. Cb, cerebellum; ChP, choroid plexus; CP, choroid plexus; Hy, hypothalamus; max., maximum; MB, midbrain; MH, medulla; min., minimum; Oligo, oligodendrocyte; OPC, oligodendrocyte progenitor cell; PSC, pluripotent stem cell; telen., telencephalon; Th, thalamus; vTelen, ventral telencephalon. To elucidate the dynamics and transitions of cell states and types, we reconstructed a real-age-informed pseudotime of HNOCA cells on the basis of neural optimal transport using moscot (Fig. 1h). Focusing on the dorsal telencephalic neural trajectory, we observed consistent pseudotemporal expression profiles of marker genes such as SOX2 (neural progenitor cells (NPCs)), BCL11B (deeper layer cortical neurons) and SATB2 (upper layer cortical neurons) (Fig. 1i). To further resolve heterogeneity among non-telencephalic neurons, we performed subclustering of this population, which revealed numerous neuronal populations characterized by distinct marker gene expression (Fig. 1j,k). To assess our cell type annotation, and more precisely annotate the heterogeneous non-telencephalic neuronal populations, we compared the HNOCA to a recently published single-cell transcriptomic atlas of the developing human brain (Fig. 2a). We applied scVI and scANVI to the primary reference atlas, and used scArches to project the HNOCA to the same latent space. The shared latent space allowed us to reconstruct a bipartite weighted k-nearest-neighbour (wkNN) graph between cells in the HNOCA and the primary reference atlas, which was used to transfer the ‘CellClass’ and ‘Subregion’ labels, as well as the neurotransmitter transporter (NTT) information of neuroblasts and neurons to the HNOCA. The transferred labels are strongly consistent with our assigned labels (Extended Data Fig. 3) and allowed us to refine the regional annotation of HNOCA non-telencephalic NPCs and neurons, as well as the NTT annotation of the non-telencephalic neurons (Fig. 2b), resulting in the final hierarchical HNOCA cell type annotation (Extended Data Fig. 3).Fig. 2Projection of HNOCA to primary developing human brain cell atlases assists organoid neural cell type annotation and estimation of primary cell type representation.a, UMAP of a human developing brain cell atlas, coloured by NTT subtypes (left), region (middle) and annotated cell classes (right). b, UMAP of HNOCA, coloured by the mapped neuron NTT subtypes (left) and regional labels of NPCs, intermediate progenitor cells (IP) and neurons. c, Heatmap showing proportions of cells from organoids of different ages matched to cells from different primary developmental (dev.) stages. d, Percentages of neural cells representing different regions (telencephalon, diencephalon, midbrain and hindbrain) in different datasets. The x axes show datasets, descendingly ordered by the total proportions (bar height). Datasets based on unguided differentiation protocols are marked by dots underneath. The bars at the bottom of each panel show organoid protocol types. e, UMAP of the human developing brain cell atlas coloured by cell population presence within HNOCA datasets (max presence score). A low score denotes under-representation of cell state in HNOCA datasets. f, Distribution of max presence scores of different cell classes in the human reference atlas. Eryt., erythrocyte; Imm., immune; Vas., vascular; G-blast, glioblast; F-blast, fibroblast; NC, neural crest; Plac., placodes; RG, radial glia; IPC, intermediate progenitor cell; N-blast, neuroblast; N, neuron. g, Box plots showing distribution of max presence scores in different primary reference cell clusters. Bottom side bars show neuronal versus non-neuronal, cell class, region information of primary reference. h, UMAP of human developing brain atlas showing primary neural cell types or states under-represented in HNOCA (in red). Hippo, hippocampus; HyTh, hypothalamus; d, dorsal; v, ventral; CB, cerebellum. a, UMAP of a human developing brain cell atlas, coloured by NTT subtypes (left), region (middle) and annotated cell classes (right). b, UMAP of HNOCA, coloured by the mapped neuron NTT subtypes (left) and regional labels of NPCs, intermediate progenitor cells (IP) and neurons. c, Heatmap showing proportions of cells from organoids of different ages matched to cells from different primary developmental (dev.) stages. d, Percentages of neural cells representing different regions (telencephalon, diencephalon, midbrain and hindbrain) in different datasets. The x axes show datasets, descendingly ordered by the total proportions (bar height). Datasets based on unguided differentiation protocols are marked by dots underneath. The bars at the bottom of each panel show organoid protocol types. e, UMAP of the human developing brain cell atlas coloured by cell population presence within HNOCA datasets (max presence score). A low score denotes under-representation of cell state in HNOCA datasets. f, Distribution of max presence scores of different cell classes in the human reference atlas. Eryt., erythrocyte; Imm., immune; Vas., vascular; G-blast, glioblast; F-blast, fibroblast; NC, neural crest; Plac., placodes; RG, radial glia; IPC, intermediate progenitor cell; N-blast, neuroblast; N, neuron. g, Box plots showing distribution of max presence scores in different primary reference cell clusters. Bottom side bars show neuronal versus non-neuronal, cell class, region information of primary reference. h, UMAP of human developing brain atlas showing primary neural cell types or states under-represented in HNOCA (in red). Hippo, hippocampus; HyTh, hypothalamus; d, dorsal; v, ventral; CB, cerebellum. We also sought to compare organoid cells to stages of human brain development beyond the first trimester. Focusing on dorsal telencephalon, we compared the transcriptomic profile of HNOCA NPCs and neurons with cells in a primary atlas of human cortex development spanning the first trimester to adolescence. We observed a transition from cell states observed in the first trimester to more mature states observed in the second-trimester cortex (Fig. 2c), and did not detect substantial matching to later stages. We extended the comparison to other brain regions using two primary atlases representing the first and second trimester, respectively. We confirmed increased similarity to second-trimester cell states in older organoids for other brain regions (Extended Data Fig. 3). We evaluated the capacity of each neural organoid protocol to generate neural cells of different brain regions (Fig. 2d, Extended Data Figs. 3 and 4 and Supplementary Table 2). Datasets of unguided neural organoids contain cells across all brain regions with proportions varying across datasets, indicating the capacity of unguided protocols to generate many brain regions but with high variability. By contrast, datasets derived from guided organoid protocols are strongly enriched for cells of the targeted brain region, but often show an increased proportion of cells of the brain regions neighbouring the targeted regions. For example, several datasets derived from midbrain organoid protocols also show high proportions of hindbrain neurons, indicating an imprecision of morphogen guidance. To comprehensively evaluate how well organoid protocols represented by the HNOCA generate primary brain cell types, we estimated presence scores for every primary cell type in each HNOCA dataset (Methods). A large presence score indicates high frequency and likelihood that cells of a similar type are observed in the HNOCA dataset. By normalizing the scores per organoid dataset (Extended Data Fig. 5 and Supplementary Table 3), we obtained a metric to describe how well each primary cell type is represented in at least one HNOCA dataset (Fig. 2d). This analysis confirmed the absence of erythrocytes, immune cells and vascular endothelial cells in the HNOCA, all of which are derived from non-neuroectodermal germ layers (Fig. 2e). As expected, telencephalic cell types are most strongly represented in HNOCA. By contrast, cell types of the thalamus, midbrain and cerebellum are least represented, including thalamic reticular nucleus GABAergic neurons, dorsal midbrain m1-derived GABAergic neurons and m1/m2-derived glutamatergic neurons, and cerebellar Purkinje cells (Fig. 2f,g). It is worth noting that, even though these cell types are less abundant in HNOCA datasets than in the primary atlas, certain organoid protocols can generate some of these under-represented cell types (for example, Purkinje cells in posterior brain organoid protocols). We next aimed to understand the transcriptomic similarities and differences between organoids generated by distinct differentiation protocols as well as between organoids and primary brain tissue. We identified differentially expressed genes (DEGs), comparing neural cell types in the HNOCA with their primary counterparts (Fig. 3a and Supplementary Table 4). We found that for most neural cell types, more than one-third (mean 34.4%, standard deviation 12.1%) of DEGs were shared across at least half of the protocols (protocol-common DEGs), suggesting that many transcriptomic differences between organoid and primary cells were independent of organoid protocol (Fig. 3b). We verified our results using an extra primary human cortex scRNA-seq dataset (Extended Data Fig. 6 and Supplementary Table 5). We next assessed differential transcriptomic programmes that were shared across regional neural cell types, and identified a total of 920 ubiquitous, protocol-common DEGs (uDEGs) that were differentially expressed in at least 14 out of the 16 neural cell types (Fig. 3c). These uDEGs showed consistent fold changes (r > 0.8) across neuron types and protocols (Fig. 3d), and represent consistent molecular differences between neurons in organoids and those in primary tissues regardless of protocol or neuronal cell type. Out of all 920 uDEGs, 363 genes were consistently upregulated and 673 genes were consistently downregulated, with only 59 genes (6%) inconsistently differentially expressed across subtypes or protocols (Fig. 3e).Fig. 3Transcriptomic comparison between organoid neurons and their primary counterpart reveals universal cell stress in organoids.a, Schematic of DE analysis comparing neural cell types in different protocols in HNOCA to their primary counterparts. b, Proportions of expressed genes in different neural cell types that show DE in certain fractions of protocols that generate the corresponding subtypes. Top left, glutamatergic neurons; bottom right, GABAergic neurons. Colour shows the brain region. c, Numbers of protocol-common DEGs (DE in at least half of protocols), grouped by the number of neural cell types in which a gene is DE. d, Distribution of expression log-fold-change (logFC) correlation of ubiquitous DEGs among different neuron subtype*protocol (that is, each of the neural cell types generated by each of the different protocols). e, Numbers of DEGs per category. f, Gene ontology enrichment analysis of downregulated (upper, blue) and upregulated (lower, red) ubiquitous DEGs. Sizes of the squares correlate with −log-transformed adjusted P values. g,h, Distribution of the mitochondrial ATP synthesis-coupled electron transport module scores (g), canonical glycolysis module scores (h, left) and the Molecular Signatures Database hallmark glycolysis module scores (h, right), in primary neural cell types (upper, dark) and organoid counterparts (lower, light). P values, significance of a two-sided Wilcoxon test. i, Heatmap shows pairwise correlation (corr.) of the three module scores. j, Hallmark glycolysis score of dorsal telencephalic excitatory neurons (dTelen VGLUT-N), split by the three primary developing human brains and 27 organoid datasets with at least 20 dTelen VGLUT-N. The lower panel shows selected features of differentiation protocols that may be relevant to cell stress. The protocol and publication indices are shown in Extended Data Fig. 1. Mat. media, maturation media. k, Spearman correlations between gene expression profiles of neural cell types in HNOCA and those in the human developing brain atlas, across the variable transcription factors (TFs). Datasets are in the same order as in Supplementary Table 1. a, Schematic of DE analysis comparing neural cell types in different protocols in HNOCA to their primary counterparts. b, Proportions of expressed genes in different neural cell types that show DE in certain fractions of protocols that generate the corresponding subtypes. Top left, glutamatergic neurons; bottom right, GABAergic neurons. Colour shows the brain region. c, Numbers of protocol-common DEGs (DE in at least half of protocols), grouped by the number of neural cell types in which a gene is DE. d, Distribution of expression log-fold-change (logFC) correlation of ubiquitous DEGs among different neuron subtype*protocol (that is, each of the neural cell types generated by each of the different protocols). e, Numbers of DEGs per category. f, Gene ontology enrichment analysis of downregulated (upper, blue) and upregulated (lower, red) ubiquitous DEGs. Sizes of the squares correlate with −log-transformed adjusted P values. g,h, Distribution of the mitochondrial ATP synthesis-coupled electron transport module scores (g), canonical glycolysis module scores (h, left) and the Molecular Signatures Database hallmark glycolysis module scores (h, right), in primary neural cell types (upper, dark) and organoid counterparts (lower, light). P values, significance of a two-sided Wilcoxon test. i, Heatmap shows pairwise correlation (corr.) of the three module scores. j, Hallmark glycolysis score of dorsal telencephalic excitatory neurons (dTelen VGLUT-N), split by the three primary developing human brains and 27 organoid datasets with at least 20 dTelen VGLUT-N. The lower panel shows selected features of differentiation protocols that may be relevant to cell stress. The protocol and publication indices are shown in Extended Data Fig. 1. Mat. media, maturation media. k, Spearman correlations between gene expression profiles of neural cell types in HNOCA and those in the human developing brain atlas, across the variable transcription factors (TFs). Datasets are in the same order as in Supplementary Table 1. Using gene ontology enrichment analysis on the uDEGs, we found downregulated uDEGs enriched in neurodevelopmental processes including neuron cell–cell adhesion and synapse organization (Fig. 3f). Upregulated uDEGs were enriched in many metabolism-associated terms including mitochondrial ATP synthesis-coupled electron transport (electron transport in short) and canonical glycolysis (Fig. 3f). An enrichment of energy-associated pathways has previously been associated with metabolic changes caused by the limitations of current culture conditions. Also, the Molecular Signatures Database gene set hallmark glycolysis has previously been used to define metabolic states in neural organoids. Scoring mitochondrial electron transport, canonical glycolysis and hallmark glycolysis gene sets across the HNOCA and the primary reference atlas, we found that all three terms showed significant separation of organoid and primary cells (Fig. 3g,h). Using the datasets from refs. and as representative examples, we identified a similar distribution of glycolysis scores across all neural cell types with an overall increased score in organoid cells (Extended Data Fig. 7). Focusing on dorsal telencephalic neurons, we compared the distribution of glycolysis scores across organoid differentiation protocols and identified several protocol features that correlated with metabolic cell stress. For instance, the usage of maturation media, slicing or cutting of organoids and, to a lesser extent, shaking or spinning of organoids led to overall lower glycolysis scores (Fig. 3h). Mean glycolysis score and transcriptomic similarity of organoid and primary reference cell types across differentiation protocols were negatively correlated. The correlation was significantly reduced when considering only variable transcription factors, indicating that the metabolic changes in organoids have limited impact on the core molecular identity of neuronal cell types (Extended Data Fig. 7). This observation is consistent with previous studies of distinct metabolic states of cells in neural organoids relative to the primary tissue, which were shown to not affect neuron fate specification and maturation. Next, we focused on the expression of 366 variable transcription factors to calculate the correlation between corresponding neuronal cell types in the HNOCA datasets and the primary reference atlas. We found that both guided and unguided organoid differentiation protocols generated neuronal cell types with comparable similarity to the corresponding primary reference cell types. However, we observed brain region-dependent differences in transcriptomic similarity. For example, organoid neurons from the dorsal parts of most brain regions showed higher similarity to their primary counterparts across organoid datasets than cell types derived from the ventral parts of most brain regions (Fig. 3i). To identify molecular features other than metabolic state that decreased organoid fidelity, we incorporated dorsal telencephalic glutamatergic neurons from four different primary developing human brain atlases as an integrated primary reference, and identified neuron subtype and maturation state heterogeneity (Extended Data Fig. 8). Projection of dorsal telencephalic neurons in the HNOCA to the primary atlases revealed the corresponding heterogeneity in neural organoids. Considering metabolic state, maturation state and cell subtype as covariates during DE analysis significantly reduced the number of DEGs, supporting the idea that these are the major factors differentiating organoid and primary brain cells (Extended Data Fig. 8 and Supplementary Table 6). We observed enriched biological processes that included synaptic vesicle cycle and negative regulation of high voltage-gated calcium channel activity (Extended Data Fig. 8), suggesting that organoids are deficient in these processes. Of note, these differences are observed across organoid protocols, and highlight areas of consistent transcriptomic divergence between in vitro and primary counterparts. The HNOCA, as well as the analytical pipeline we established, provides a framework to query new neural organoid scRNA-seq datasets not included in the HNOCA. To showcase this application, we retrieved scRNA-seq data from a recently published multiplexed neural organoid morphogen screen and projected them to the HNOCA and primary reference latent spaces (Fig. 4a, Extended Data Fig. 9 and Supplementary Table 7). We transferred regional labels and found high consistency with the provided regional annotation, but with higher resolution within each of the broad brain sections of forebrain, midbrain and hindbrain (Fig. 4b). Our transferred annotation therefore allowed a more comprehensive assessment of the effects of different morphogen conditions on generating neurons of different brain regions (Fig. 4c). We further calculated presence scores for reference cells in each screen condition and compared the data of the different screen conditions with the 36 HNOCA datasets. Using hierarchical clustering on average presence scores revealed distinct presence score profiles for many screen conditions (Fig. 4d), suggesting regional cell type composition distinct from the HNOCA datasets. Next, we summarized the max presence scores for the whole morphogen screen data (Fig. 4e), and compared them to those for the HNOCA data to identify primary reference cell types with increased presence in the screen (Fig. 4f). This analysis highlighted several reference cell clusters with significant abundance increase under certain screen conditions (Fig. 4g) such as LHX6/ACKR3/MPPED1 triple-positive GABAergic neurons in the ventral telencephalon and dopaminergic neurons in ventral midbrain. In summary, the projection of the morphogen screen query data to HNOCA and primary reference allowed a refined annotation of the morphogen screen data, as well as a comprehensive and quantitative evaluation of the value of new differentiation protocols to generate neuronal cell types previously under-represented or lacking in neural organoids.Fig. 4Projection of neural organoid morphogen screen scRNA-seq data to HNOCA and human developing brain atlas allows cell type annotation and organoid protocol evaluation.a, Schematic of projecting neural organoid morphogen screen scRNA-seq data to the HNOCA, and a human developing brain reference atlas. UMAPs show screen condition groups (left, using morphogens SAG (sonic hedgehog signaling agonist), CHIR, BMP and FGF) and regional annotation of screen data (right). b, Comparison of regional annotation of screen data (rows) and scArches-transferred regional labels from the primary reference. c, Proportions of cells assigned to different regions on the basis of reference projection. Every stacked bar represents one screen condition. d, Clustering of HNOCA datasets with conditions in the screen data on the basis of average presence scores of clusters in the primary reference. The heatmap shows average presence scores per cluster in the primary reference (columns). e, UMAP of primary reference coloured by the dissected regions (right) and the maximum presence scores across the screen conditions (left). f, Gain of cell cluster coverage of screen conditions relative to HNOCA datasets, with negative values trimmed to zero. The grey horizontal line shows the threshold (0.3) to define gained clusters in screen data. g, UMAP of the primary reference, with gained clusters highlighted in shades of blue. Dashed circles highlight two clusters with highest gain of coverage in telencephalon and midbrain, respectively. h, Coexpression scores of cluster marker genes of the two clusters highlighted in g, in the primary reference (upper) and screen dataset (lower). DA, dopaminergic. a, Schematic of projecting neural organoid morphogen screen scRNA-seq data to the HNOCA, and a human developing brain reference atlas. UMAPs show screen condition groups (left, using morphogens SAG (sonic hedgehog signaling agonist), CHIR, BMP and FGF) and regional annotation of screen data (right). b, Comparison of regional annotation of screen data (rows) and scArches-transferred regional labels from the primary reference. c, Proportions of cells assigned to different regions on the basis of reference projection. Every stacked bar represents one screen condition. d, Clustering of HNOCA datasets with conditions in the screen data on the basis of average presence scores of clusters in the primary reference. The heatmap shows average presence scores per cluster in the primary reference (columns). e, UMAP of primary reference coloured by the dissected regions (right) and the maximum presence scores across the screen conditions (left). f, Gain of cell cluster coverage of screen conditions relative to HNOCA datasets, with negative values trimmed to zero. The grey horizontal line shows the threshold (0.3) to define gained clusters in screen data. g, UMAP of the primary reference, with gained clusters highlighted in shades of blue. Dashed circles highlight two clusters with highest gain of coverage in telencephalon and midbrain, respectively. h, Coexpression scores of cluster marker genes of the two clusters highlighted in g, in the primary reference (upper) and screen dataset (lower). DA, dopaminergic. We next tested whether the integrated HNOCA can serve as a control cohort for assessing organoid models of neural disease. We collected 11 scRNA-seq datasets from 10 neural organoid disease models and their respective controls (microcephaly, amyotrophic lateral sclerosis, Alzheimer’s disease, autism, fragile-X syndrome (FXS), schizophrenia, neuronal heterotopia, Pitt–Hopkins syndrome, myotonic dystrophy and glioblastoma) (Fig. 5a, Extended Data Fig. 10 and Supplementary Table 8). We projected the data to the HNOCA and the primary reference atlas to transfer annotations (Fig. 5b–f). We found differences in cell type and brain regional composition between disease model organoids and their respective, study-specific control organoids for most studies (Fig. 5g,h). These differences might represent disease phenotypes, but could also be the consequence of cell line variability. It is therefore important to properly annotate the cell type and regional composition of disease and control organoids to identify disease phenotypes, particularly when analysing disease-associated transcriptomic alterations in a given cell type.Fig. 5The HNOCA as a control cohort to facilitate cell type annotation and transcriptomic comparison for neural organoid disease-modelling data.a, Overview of disease-modelling neural organoid atlas construction, and projection to primary atlas and HNOCA for downstream analysis. b–f, UMAP of integrated disease-modelling neural organoid atlas coloured by predicted cell type annotation (b), predicted regional identities of NPCs, intermediate progenitor cells and neurons (c), publications (d), disease status (e) and marker gene expression (f). g,h, Proportions (prop.) of cells assigned to different cell classes (g) and regions (h). Every stacked bar represents one biological sample. Side bars show disease status and publication. i, Schematic of reconstructing matched HNOCA metacell for each cell in the disease-modelling neural organoid atlas. j, UMAP of disease-modelling neural organoid atlas, coloured by transcriptomic similarity with the matched HNOCA metacells. k, Violin plot indicates distribution of estimated transcriptomic similarities, split by publication. Left, distribution in control cells and right, distribution in disease cells. l, Heatmap showing expression of top DEGs between the AQP4 population in the GBM-2019 dataset and their matched HNOCA metacells. Rows show DEGs with the ten strongest decreased and increased expressions. Columns show average expression in the AQP4 population of disease-modelling samples (first panel), the matched HNOCA metacells per sample (second panel), all predicted control astrocytes and all astrocytes in HNOCA. m, Volcano plot shows DE analysis between dorsal telencephalic cells in the FXS-2021 dataset and their matched HNOCA metacells. DEGs coloured in red (increased in FXS) and blue (decreased in FXS). Encircled dots show DEGs annotated in SFARI database. Top bars show the log-transformed odds ratio of SFARI gene enrichment in the increased (red) and decreased (blue) DEGs. GBM, glioblastoma. a, Overview of disease-modelling neural organoid atlas construction, and projection to primary atlas and HNOCA for downstream analysis. b–f, UMAP of integrated disease-modelling neural organoid atlas coloured by predicted cell type annotation (b), predicted regional identities of NPCs, intermediate progenitor cells and neurons (c), publications (d), disease status (e) and marker gene expression (f). g,h, Proportions (prop.) of cells assigned to different cell classes (g) and regions (h). Every stacked bar represents one biological sample. Side bars show disease status and publication. i, Schematic of reconstructing matched HNOCA metacell for each cell in the disease-modelling neural organoid atlas. j, UMAP of disease-modelling neural organoid atlas, coloured by transcriptomic similarity with the matched HNOCA metacells. k, Violin plot indicates distribution of estimated transcriptomic similarities, split by publication. Left, distribution in control cells and right, distribution in disease cells. l, Heatmap showing expression of top DEGs between the AQP4 population in the GBM-2019 dataset and their matched HNOCA metacells. Rows show DEGs with the ten strongest decreased and increased expressions. Columns show average expression in the AQP4 population of disease-modelling samples (first panel), the matched HNOCA metacells per sample (second panel), all predicted control astrocytes and all astrocytes in HNOCA. m, Volcano plot shows DE analysis between dorsal telencephalic cells in the FXS-2021 dataset and their matched HNOCA metacells. DEGs coloured in red (increased in FXS) and blue (decreased in FXS). Encircled dots show DEGs annotated in SFARI database. Top bars show the log-transformed odds ratio of SFARI gene enrichment in the increased (red) and decreased (blue) DEGs. GBM, glioblastoma. We developed a wkNN-based strategy to generate matched HNOCA metacells for every cell in each disease model organoid scRNA-seq dataset (Fig. 5i), and quantified their transcriptomic similarity (Fig. 5j). The dataset of glioblastoma organoids showed substantially lower similarity to their primary counterpart than the other disease models (Fig. 5k). To assess these transcriptomic differences, we performed DE analysis between glioblastoma and matched control metacells. Focusing on the AQP4 population (Extended Data Fig. 10), we identified 1,951 DEGs in glioblastoma cells compared to matched HNOCA metacells (Supplementary Table 9) and found increased expression of genes such as RBM25 (ref. ) CALD1 (ref. ), HNRNPU and SPARC (Fig. 5l), all of which have been reported to be relevant to glioblastoma. Next, we focused on the organoid model of FXS, in which NPCs and neurons in the control organoids were of non-telencephalic identities whereas the disease model organoids mainly contained telencephalic cells (Fig. 5h and Extended Data Fig. 10). The integrated HNOCA provides the opportunity to perform DE analysis for FXS neocortical neurons with matched HNOCA metacells, which identified 444 DEGs. DEGs higher expressed in FXS cells (122 genes) were enriched for autism-associated genes annotated in the Simons Foundation Autism Research Initiative (SFARI) database. One such gene, CHD2, was reported in the original publication as a key regulator of FXS with increased protein level, but its expression change on messenger RNA (mRNA) level change could not be detected in a bulk RNA-seq experiment. We also detected decreased expression of FMR1, whose loss-of-function mutation causes FXS. New scRNA-seq datasets of human neural organoids continue to be generated, and it will be important to continuously extend and update the HNOCA with this extra data. We therefore established a computational toolkit to project new scRNA-seq data to the HNOCA (Fig. 6a). We demonstrate the use of the toolkit by incorporating scRNA-seq data from six more studies into the HNOCA (HNOCA-extended; Fig. 6b and Supplementary Table 10), using query-to-reference mapping. We harmonized cell type annotations using wkNN-based label transfer, and placed the cells in the context of the existing organoid single-cell transcriptomic landscape as represented by the HNOCA (Fig. 6c–e). Mapping further datasets to the HNOCA using our approach enhances the atlas by increasing its coverage over existing neural organoid protocols and neural cell types generated in organoids.Fig. 6Extending the HNOCA by means of projection of extra datasets.a, Schematic of projecting further scRNA-seq data by the community to extend the HNOCA. b, UMAP shows the dataset composition of the current extended HNOCA. c, UMAP shows the projected cell type annotation of cells in the five extended datasets. NE, neuroepithelium; NC-D, neural crest derivatives; MC, mesenchymal cell; EC, endothelial cell. d, Dot plot shows the expression of selected cell type and regional markers across projected cell types in the extended HNOCA datasets. e, Dot plot shows cell type composition and average similarity to the matched HNOCA metacells of the extended datasets. f, Schematic shows the analytical pipelines and varied interfaces to facilitate analysing scRNA-seq data of neural organoids for the community. a, Schematic of projecting further scRNA-seq data by the community to extend the HNOCA. b, UMAP shows the dataset composition of the current extended HNOCA. c, UMAP shows the projected cell type annotation of cells in the five extended datasets. NE, neuroepithelium; NC-D, neural crest derivatives; MC, mesenchymal cell; EC, endothelial cell. d, Dot plot shows the expression of selected cell type and regional markers across projected cell types in the extended HNOCA datasets. e, Dot plot shows cell type composition and average similarity to the matched HNOCA metacells of the extended datasets. f, Schematic shows the analytical pipelines and varied interfaces to facilitate analysing scRNA-seq data of neural organoids for the community. To enable researchers to use the HNOCA in their own analysis, we provide various options for exploration and interaction with the atlas (Fig. 6f). The HNOCA can be browsed through an online portal, enabling visualization of gene expression and discovery of marker genes. We also provide the HNOCA through an online interface (http://www.archmap.bio/) for the interactive mapping of new datasets, enabling label transfer, presence score computation and metabolic scoring of cell states. Finally, we have developed HNOCA-tools, a Python package implementing all central analysis approaches presented in this paper, such as annotation, reference mapping, label transfer and DE testing methods. In this study, we built a large-scale integrated cell atlas of human neural organoids, the HNOCA, by integrating 1.8 million cells spanning 36 scRNA-seq datasets generated by 15 different laboratories worldwide using 26 different differentiation protocols as well as diverse scRNA-seq technologies. The resulting atlas revealed the high complexity of neuronal, glial and non-neural cell types that can develop in neural organoids grown under existing protocol conditions. Mapping the HNOCA data to various human developing brain cell reference atlases allowed comprehensive evaluation of neural organoid protocols to generate cell types of different brain regions. We found that organoids in the first 3 months of culture best match to first-trimester primary data, whereas organoids around 3 months of culture and older best match second-trimester primary cell states. We did not observe significant neuronal maturation and diversification signatures matching older developmental stages, suggesting a limitation of neuronal maturation in current neural organoid protocols. We performed DE analysis between organoid neuron types and their primary counterparts to evaluate transcriptomic fidelity, and identified metabolic changes related to the glycolysis pathway as a main factor that distinguishes organoid and primary cell states, consistent with previous reports. Despite the negative effects of metabolic stress on overall transcriptomic fidelity, the molecular identity of regional cell types is maintained as evidenced by transcription factor coexpression patterns that are highly consistent with primary counterparts. We showcased the mapping of query data, a recently published single-cell transcriptomic neural organoid morphogen screen, to the HNOCA and the primary reference, which enabled a refined cell type annotation, as well as a compositional comparison with existing neural organoid datasets. Our powerful framework will facilitate quantitative and comparative analysis of scRNA-seq data of human neural organoids, and for the benchmarking of new neural organoid protocols. Consistent with earlier reports, we find that unguided protocols generate neural cells with high brain regional variability, which is useful when studying broader fate determination during neurodevelopment. Guided protocols resulted in a strong enrichment of the targeted brain regions. We also note that some guided protocols, particularly those targeting midbrain, show relatively low specificity and generate neural cells from the nearby brain regions. This issue may be due to a differential response of neural stem cells in the organoid to the same morphogen cue, or to the lack of a full understanding of the timing, concentration and combinations of morphogens required to precisely define cells of the deeper regions in the central nervous system. The integrated HNOCA is also an excellent resource for analysis of disease-modelling neural organoid data. It facilitates cell type annotation and provides a large control cohort of single-cell transcriptomes for comparison. For example, we observed discrepancy of cell type and regional composition between control and disease model samples in many studies. At the same time, the HNOCA provides the opportunity to identify disease-specific molecular features against a multi-line multi-protocol large-scale control cohort. We demonstrate how the HNOCA can be extended and updated by projecting extra single-cell transcriptomic data of neural organoids to the atlas. Further, we have developed a computational toolkit, HNOCA-tools, which will enable other researchers to recapitulate the analytic framework applied in our study. Together, we imagine that the HNOCA will be kept up to date and continue to reflect the landscape of human neural cell states generated in organoids in vitro, serving as a living resource for the neural organoid community that enables the assessment of organoid fidelity, the characterization of perturbed and diseased states and the development of new protocols. We included 33 human neural organoid data from a total of 25 publications plus three unpublished datasets in our atlas (Supplementary Table 1). We curated all neural organoid datasets used in this study through the sfaira framework (GitHub dev branch, 18 April 2023). For this, we obtained scRNA-seq count matrices and associated metadata from the location provided in the data availability section for every included publication or directly from the authors in case of unpublished data. We harmonized metadata according to the sfaira standards (https://sfaira.readthedocs.io/en/latest/adding_datasets.html) and manually curated an extra metadata column organoid_age_days, which described the number of days the organoid had been in culture before collection. We next removed any non-applicable subsets of the published datasets: diseased samples or samples expressing disease-associated mutations (refs. ), fused organoids (ref. ), primary fetal data (refs. ), hormone-treated samples (ref. ), data collected before neural induction (refs. ) and share-seq data (ref. ). We harmonized all remaining datasets to a common feature space using any genes of the biotype ‘protein_coding’ or ‘lncRNA’ from ensembl release 104 while filling any genes missing in a dataset with zero counts. On average, 50% of the full gene space (36,842 genes) was reported in each of the constituent datasets. We then concatenated all remaining datasets to create a single AnnData object. All processing and analyses were carried out using scanpy (v.1.9.3) unless indicated otherwise. For quality control and filtering of HNOCA, we removed any cells with fewer than 200 genes expressed. We next removed outlier cells in terms of two quality control metrics: the number of expressed genes and percentage mitochondrial counts. To define outlier cells on the basis of each quality control metric, z-transformation is first applied to values across all cells. Cells with any z-transformed metric less than −1.96 or greater than 1.96 are defined as outliers. For any dataset collected using the v.3 chemistry by 10X Genomics, which contains more than 500 cells after the filtering, we fitted a Gaussian distribution to the histogram denoting the number of expressed genes per cell. If a bimodal distribution was detected, we removed any cell with fewer genes expressed than defined by the valley between the two maxima of the distribution. We then normalized the raw read counts for all Smart-seq2 data by dividing it by the maximum gene length for each gene obtained from BioMart. We next multiplied these normalized read counts by the median gene length across all genes in the datasets and treated those length-normalized counts equivalently to raw counts from the datasets obtained with the help of unique molecular identifiers in our downstream analyses. As a next step we generated a log-normalized expression matrix by first dividing the counts for each cell by the total counts in that cell and multiplying by a factor of 1,000,000 before taking the natural logarithm of each count + 1. We computed 3,000 highly variable features in a batch-aware manner using the scanpy highly_variable_genes function (flavor = ‘seurat_v3’, batch_key = ‘bio_sample’). Here, bio_sample represents biological samples as provided in the original metadata of the datasets. On average, 72% of the 3,000 highly variable genes were reported in each of the constituent HNOCA datasets. We used these 3,000 features to compute a 50-dimensional representation of the data using principal component analysis (PCA), which in turn we used to compute a k-nearest-neighbour (kNN) graph (n_neighbors = 30, metric = ‘cosine’). Using the neighbour graph we computed a two-dimensional representation of the data using UMAP and a coarse (resolution 1) and fine (resolution 80) clustering of the unintegrated data using Leiden clustering. Snapseed is a scalable auto-annotation strategy, which annotates cells on the basis of a provided hierarchy of cell types and the corresponding cell type markers. It is based on enrichment of marker gene expression in cell clusters (high-resolution clustering is preferred), and data integration is not necessarily required. In this study, we used snapseed to obtain initial annotations for label-aware integration. First, we constructed a hierarchy of cell types including progenitor, neuron and non-neural types, each defined by a set of marker genes (Supplementary Data 1). Next, we represented the data by the RSS to average expression profiles of cell clusters in the recently published human developing brain cell atlas. We then constructed a kNN graph (k = 30) in the RSS space and clustered the dataset using the Leiden algorithm (resolution 80). For both steps, we used the graphical processing unit (GPU)-accelerated RAPIDS implementation that is provided through scanpy. For all cell type marker genes on a given level in the hierarchy, we computed the area under the receiver operating characteristic curve (AUROC) as well as the detection rate across clusters. For each cell type, a score was computed by multiplying the maximum AUROC with the maximum detection rate among its marker genes. Each cluster was then assigned to the cell type with the highest score. This procedure was performed recursively for all levels of the hierarchy. The same procedure was carried out using the fine (resolution 80) clustering of the unintegrated data to obtain cell type labels for the unintegrated dataset that were used downstream as a ground-truth input for benchmarking integration methods. This auto-annotation strategy was implemented in the snapseed Python package and is available on GitHub (https://github.com/devsystemslab/snapseed). Snapseed is a light-weight package to enable scalable marker-based annotation for atlas-level datasets in which manual annotation is not readily feasible. The package implements three main functions: annotate() for non-hierarchical annotation of a list of cell types with defined marker genes, annotate_hierarchy() for annotating more complex, manually defined cell type hierarchies and find_markers() for fast discovery of cluster-specific features. All functions are based on a GPU-accelerated implementation of AUROC scores using JAX (https://github.com/google/jax). We performed integration of the organoid datasets for HNOCA using the scPoli model from the scArches package. We defined the batch covariate for integration as a concatenation of the dataset identifier (annotation column ‘id’), the annotation of biological replicates (annotation column ‘bio_sample’) as well as technical replicates (annotation column ‘tech_sample’). This resulted in 396 individual batches. The batch covariate is represented in the model as a learned vector of size five. We used the top three levels of the RSS-based snapseed cell type annotation as the cell type label input for the scPoli prototype loss. We chose the hidden layer size of the one-layer scPoli encoder and decoder as 1,024, and the latent embedding dimension as ten. We used a value of 100 for the ‘alpha_epoch_anneal’ parameter. We did not use the unlabelled prototype pretraining. We trained the model for a total of seven epochs, five of which were pretraining epochs. To quantitatively compare the organoid atlas integration results from several tools, we used the GPU-accelerated scib-metrics Python package (v.0.3.3) and used the embedding with the highest overall performance for all downstream analyses. We compared the data integration performance across the following latent representations of the data: unintegrated PCA, RSS integration, scVI (default parameters except for using two layers, latent space of size 30 and negative binomial likelihood) integration, scANVI (default parameters) integrations using snapseed level 1, 2 or 3 annotation as cell type label input, scPoli (parameters shown above) integrations using either snapseed level 1, 2 or 3 annotation or all three annotation levels at once as the cell type label input, scPoli integrations of metacells aggregated with the aggrecell algorithm (first used as ‘pseudocell’) using either snapseed level 1 or 3 annotation as the cell type label input to scPoli. We used the following scores for determining integration quality (each described in ref. ): Leiden normalized mutual information score, Leiden adjusted rand index, average silhouette width per cell type label, isolated label score (average silhouette width-scored) and cell type local inverse Simpson’s index to quantify conservation of biological variability. To quantify batch-effect removal, we used average silhouette width per batch label, integration local inverse Simpson’s index, kNN batch-effect test score and graph connectivity. Integration approaches were then ranked by an aggregate total score of individually normalized (into the range of ) metrics. Before we carried out the benchmarking, we iteratively removed any cells from the dataset that had an identical latent representation to another cell in the dataset until no latent representation contained any more duplicate rows. This procedure removed a total of 3,293 duplicate cells (0.002% of the whole dataset) and was required for the benchmarking algorithm to complete without errors. We used the snapseed level 3 annotation computed on the unintegrated PCA embedding as ground-truth cell type labels in the integration. To infer a global ordering of differentiation state, we sought to infer a real-time-informed pseudotime on the basis of neural optimal transport in the scPoli latent space. We first grouped organoid age in days into seven bins ((0, 15], (15, 30], (30,60], (60, 90], (90, 120], (120, 150], (150, 450]). Next, we used moscot to solve a temporal neural problem. To score the marginal distributions on the basis of expected proliferation rates, we obtained proliferation and apoptosis scores for each cell with the method score_genes_for_marginals(). Marginal weights were then computed with[12pt] $$ (4 (}-}))$$(4×(prolif_score−apoptosis_score)) The optimal transport problem was solved using the following parameters: iterations = 25,000, compute_wasserstein_baseline = False, batch_size = 1,024, patience = 100, pretrain = True, train_size = 1. To compute displacement vectors for each cell in age bin i, we used the subproblem corresponding to the [i, i + 1] transport map, except for the last age bin, where we used the subproblem [i − 1,i]. Displacement vectors were obtained by subtracting the original cell distribution from the transported distribution. Using the velocity kernel from CellRank we computed a transition matrix from displacement vectors and used it as an input for computing diffusion maps. Ranks on negative diffusion component 1 were used as a pseudotemporal ordering. The cell ranger-processed scRNA-seq data for the primary atlas were obtained from the link provided on its GitHub page (https://storage.googleapis.com/linnarsson-lab-human/human_dev_GRCh38-3.0.0.h5ad). For further quality control, cells with fewer than 300 detected genes were filtered out. Transcript counts were normalized by the total number of counts for that cell, multiplied by a scaling factor of 10,000 and subsequently natural-log transformed. The feature set was intersected with all genes detected in the organoid atlas and the 2,000 most highly variable genes were selected with the scanpy function highly_variable_genes using ‘Donor’ as the batch key. An extra column of ‘neuron_ntt_label’ was created to represent the automatic classified neural transmitter transporter subtype labels derived from the ‘AutoAnnotation’ column of the cell cluster metadata (https://github.com/linnarsson-lab/developing-human-brain/files/9755350/table_S2.xlsx). To compare our organoid atlas with data from the primary developing human brain, we used scArches to project it to the above mentioned primary human brain scRNA-seq atlas. We first pretrained a scVI model on the primary atlas with ‘Donor’ as the batch key. The model was constructed with following parameters: n_latent = 20, n_layers = 2, n_hidden = 256, use_layer_norm = ‘both’, use_batch_norm = ‘none’, encode_covariates = True, dropout_rate = 0.2 and trained with a batch size of 1,024 for a maximum or 500 epochs with early stopping criterion. Next, the model was fine-tuned with scANVI using ‘Subregion’ and ‘CellClass’ as cell type labels with a batch size of 1,024 for a maximum of 100 epochs with early stopping criterion and n_samples_per_label = 100. To project the organoids atlas to the primary atlas, we used the scArches implementation provided by scvi-tools. The query model was fine-tuned with a batch size of 1,024 for a maximum of 100 epochs with early stopping criterion and weight_decay = 0.0. With the primary reference and query (HNOCA) data projected to the same latent space, an unweighted bipartite kNN graph was constructed by identifying 100 nearest neighbours of each query cell in the reference data with either PyNNDescent or RAPIDS-cuML (https://github.com/rapidsai/cuml) in Python, depending on availability of GPU acceleration. Similarly, a reference kNN graph was also built by identifying 100 nearest neighbours of each reference cell in the reference data. For each edge in the reference-query bipartite graph, the similarity between the reference neighbours of the two linked cells, defined as A and B, respectively, is represented by the Jaccard index:[12pt] $$J(A,B)=.$$(A,B)=∣A∩B∣∣A∪B∣. The square of Jaccard index was then assigned as the weight of the edge, to get the bipartite weighted kNN graph between the reference and query datasets. Given the wkNN estimated between primary reference and query (HNOCA), any categorical metadata label of reference can be transferred to query cells by means of majority voting. In brief, for each category, its support was calculated for each query cell as the sum of weights of edges that link to reference cells in this category. The category with the largest support was assigned to the query cell. To get the final regional labels for the non-telencephalic NPCs and neurons, as well as the NTT labels for non-telencephalic neurons, constraints were added to the transfer procedure. For regional labels, only the non-telencephalic regions, namely diencephalon, hypothalamus, thalamus, midbrain, midbrain dorsal, midbrain ventral, hindbrain, cerebellum, pons and medulla, were considered valid categories to be transferred. The label-transfer procedure was only applied to the non-telencephalic NPCs and neurons in HNOCA. Before any majority voting was done, the support scores of each valid category across all non-telencephalic NPCs and neurons in HNOCA were smoothed with a random-walk-with-restart procedure (restart probability alpha, 85%). Next, a hierarchical label transfer, which takes into account the structure hierarchy, was applied. First, the considered regions were grouped into diencephalon, midbrain and hindbrain, with a support score of each structure as its score summed up with scores of its substructures. Majority voting was applied to assign each cell to one of the three structures. Next, a second majority voting was applied to only consider the substructures under the assigned structure (for example, hypothalamus and thalamus for diencephalon). For NTT labels, we first identified valid region-NTT label pairs in the reference on the basis of the provided NTT labels in the reference neuroblast and neuron clusters and their most common regions. Here, the most common regions were re-estimated in a hierarchical manner to the finest resolution mentioned above. Next, when transferring NTT labels, for each non-telencephalic neuron with the regional label transferred, only NTT labels that were considered valid for the region were considered during majority voting. To match telencephalic NPCs and neurons in HNOCA to developmental stages, we used the recently published human neocortical development atlas as the reference. The processed single nucleus RNA-seq data were obtained from its data portal (https://cell.ucsf.edu/snMultiome/). Given the ‘class’, ‘subclass’ and ‘type’ labels in the provided metadata as annotations, and ‘individual’ as the batch label, scPoli was applied for label-aware data integration. Next, data representing different developmental stages were split. For each stage, Louvain clustering based on the scPoli latent representation (resolution, 5) was applied. Clusters of all stages were pooled, and highly variable genes were identified on the basis of coefficient of variations as described in this page: https://pklab.med.harvard.edu/scw2014/subpop_tutorial.html. Finally, every one of HNOCA telencephalic NPCs and neurons were correlated to each cluster across the identified highly variable genes. The stage label of the best-correlated cluster was assigned to the query HNOCA cell. To extend the analysis to other neuronal cell types, the second-trimester multiple-region human brain atlas was also introduced. The processed count matrices and metadata were obtained from the NeMO data portal (https://data.nemoarchive.org/biccn/grant/u01_devhu/kriegstein/transcriptome/scell/10x_v2/human/processed/counts/). Given the ‘cell_type’ label of the provided metadata as the annotation and ‘individual’ as the batch label, scPoli was run for label-aware data integration. Louvain clustering was applied to the scPoli latent representation to identify clusters (resolution, 20). Similarly, Louvain clustering with a resolution of 20 was also applied to the first-trimester multiple-region human brain atlas on the basis of the scANVI latent representation we generated earlier. Average expression profiles were calculated for all the clusters, and highly variable genes were identified using the same procedure as above for clusters of the two primary atlases combined. Next, every NPC and neuron in HNOCA was correlated to the average expression profiles of those clusters. The best-correlated first- and second-trimester clusters, as well as the correlations, were identified. The differences between the two correlations were used as the metrics to indicate the stage-matching preferences of NPCs and neurons in HNOCA. Given a reference dataset and a query dataset, the presence score is a score assigned to each cell in the reference, which describes the frequency or likelihood of the cell type or state of that reference cell appearing in the query data. In this study, we calculated the presence scores of primary atlas cells in each HNOCA dataset to quantify how frequently we saw a cell type or state represented by each primary cell in each of the HNOCA datasets. Specifically, for each HNOCA dataset, we first subset the wkNN graph to only HNOCA cells in that dataset. Next, the raw weighted degree was calculated for each cell in the primary atlas, as the sum of weights of the remaining edges linked to the cell. A random-walk-with-restart procedure was then applied to smooth the raw scores across the kNN graph of the primary atlas. In brief, we first represented the primary atlas kNN graph as its adjacency matrix (A), followed by row normalization to convert it into a transition probability matrix (P). With the raw scores represented as a vector s0, in each iteration t, we generated st as[12pt] $$_= }}_+(1- )^_$$=αs0+(1−α)PTst−1 This procedure was performed 100 times to get the smooth presence scores that were subsequently log transformed. Scores lower than the 5th percentile or higher than the 95th percentile were trimmed. The trimmed scores were normalized into the range of as the final presence scores in the HNOCA dataset. Given the final presence scores in each of the HNOCA datasets, the max presence scores in the whole HNOCA data were then easily calculated as the maximum of all the presence scores for each cell in the primary atlas. A large (close to one) max presence score indicates a high frequency of appearance for the cell type or state in at least one HNOCA dataset whereas a small (close to zero) max presence score suggests under-representation in all the HNOCA datasets. To test the cell type compositional changes on admission of certain morphogens from different organoid differentiation protocols, we used the pertpy implementation of the scCODA algorithm. scCODA is a Bayesian model for detecting compositional changes in scRNA-seq data. For this, we have extracted the information about the added morphogens from each differentiation protocol and grouped them into 15 broad molecule groups on the basis of their role in neural differentiation (Supplementary Table 1). These molecule groups were used as a covariate in the model. The region labels transferred from the primary atlas were used as labels in the analysis (cell_type_identifier). For cell types without regional identity, the cell type labels presented in Fig. 1c were used. Pluripotent stem cells and neuroepithelium cells were removed from the analysis because they are mainly present in the early organoid stages. We used bio_sample as the sample_identifier. We ran scCODA sequentially with default parameters, using No-U-turn sampling (run_nuts function) and selecting each cell type once as a reference. We used a majority vote-based system to find the cell types that were credibly changing in more than half of the iterations. To complement the composition analysis conducted with scCODA, we devised an alternative approach to test for differential composition using regularized linear regression. We fit a generalized linear model with the region composition matrix as the response Y and molecule usage as independent variables X:[12pt] $$Y X}$$~Xβ The model was fit with lasso regularization (alpha = 1) using Gaussian noise and an identity link function. The regularization parameter lambda was automatically determined through cross-validation as implemented in the function cv.glmnet() from the glmnet R package. All non-zero coefficients β were considered as indications of enrichment and depletion. To study the transcriptomic differences between organoid and primary cells, we subset HNOCA using the final level 1 annotation to cells labelled ‘Neuron’. We furthermore subset the human developing brain atlas to cells that had been assigned a valid label in the neuron_ntt_label annotation column. We added an extra two datasets of fetal cortical cells from ref. and ref. . For the data from ref. , we subset the data to cells labelled ‘fetal’ and estimated transcripts per million reads for each gene in each cell using RSEM given the STAR mapping results. We then computed a PCA, a kNN graph, UMAP and Leiden clustering (resolution 0.2) using scanpy. We then selected the cluster with the highest STMN2 and NEUROD6 expression as the cortical neuron cluster and used only those cells. For the data from ref. we subset the datasets to cells annotated as ‘Neuronal’ in Supplementary Table 5 (‘Cortex annotations’) of their publication and computed a PCA, neighbourhood graph and UMAP to visualize the dataset. We found that only samples from the individuals CS14_3, CS20, CS22 and CS20 contained detectable expression of STMN2 and NEUROD6 so we subset the dataset further to only cells from those individuals. To compute DE between HNOCA cells and their primary counterparts, we first aggregated cells of the same regional neural cell type into pseudobulk samples by summing the counts for every sample (annotation columns, ‘batch’ for HNOCA; ‘SampleID’ for the human developing brain atlas; ‘sample’ for ref. and ‘individual’ for ref. ) using the Python implementation of decoupler (v.1.4.0) while discarding any samples with fewer than ten cells or 1,000 total counts. We then subsetted the feature space to the intersection of features of all datasets and removed any cells with fewer than 200 genes expressed. We further removed any genes expressed in less than 1% of neurons in HNOCA and any genes located on the X and Y chromosomes. Out of the remaining 11,636 genes, on average, 99% were reported in each of the constituent HNOCA datasets. For each regional neural cell type, we removed any sample from the pseudobulk data that was associated with an organoid differentiation assay with fewer than two total samples or fewer than 100 total cells. We next used edgeR to iteratively compute DE genes between each organoid differentiation protocol and primary cells of the matching regional neural cell types for every regional neural cell type while correcting for organoid age in days, number of cells per pseudobulk sample, median and standard deviation of the number of detected genes per pseudobulk sample. We used the data from ref. (the human developing brain atlas mentioned above), ref. and ref. as primary data for the DE comparison in the cell type ‘Dorsal Telencephalic Neuron NT-VGLUT’, whereas for all other cell types we used the human developing brain atlas as the fetal dataset. We used the edgeR genewise negative binomial generalized linear model with quasi-likelihood F-tests. We deemed a gene significantly DE if its false-discovery rate (Benjamini–Hochberg) corrected P value was smaller than 0.05 and it had an absolute log2-fold change above 0.5. We used the GSEApy Python package to carry out functional enrichment analysis in our DE results using the ‘GO_Biological_Process_2021’ gene set. To evaluate the effect of different primary datasets on the DE results, we computed the DE between Dorsal Telencephalic Neuron NT-VGLUT from the HNOCA subset generated with the protocol from ref. and the matching cell type from the Braun et al. primary dataset as well as the data from ref. . To prevent technology effects to affect this analysis, we only used cells generated with the 10X Genomics 3′ v.2 protocol in this comparison. We generate pseudobulk samples as described above and corrected organoid age in days and number of cells per pseudobulk sample in the DE comparison. We used the same edgeR-based procedure and cut-offs as described above. We used the scipy fcluster method to cluster genes on the basis of their log-fold changes in the two primary datasets. We grouped clusters to represent consistently upregulated, consistently downregulated and three different inconsistently regulated groups of genes. We computed functional enrichment of each gene group as described above. To evaluate the effect of different organoid datasets on the protocol-based DE analysis, we computed DE between Dorsal Telencephalic Neuron NT-VGLUT of every organoid publication (further split by protocol, where more than one protocol was used in a publication) and the matching cell type in the dataset from ref. . We computed pseudobulk samples and carried out the DE analysis using the same procedure and cut-offs as in the protocol-based DE analysis. To estimate the transcriptomic similarity between neurons in HNOCA and the human developing brain atlas, we first summarized the average expression of each neural cell type in the primary reference, as well as in each dataset of HNOCA. For each HNOCA dataset, only neural cell types with at least 20 cells were considered. Highly variable genes were identified across the neural cell types in the primary reference using a Chi-squared test-based variance ratio test on the generalized linear model with Gamma distribution (identity link), given coefficient of variance of transcript counts across neural cell types as the response and the reciprocal of average transcript count across neural cell types as the independent variable. Genes with Benjamini–Hochberg adjusted P values less than 0.01 were considered as highly variable genes. Similarity between one neural cell type in the primary atlas and its counterpart in each HNOCA dataset was then calculated as the Spearman correlation coefficient across the identified highly variable genes. To estimate the similarity of the core transcriptomic identity, which is defined by the coexpression of transcription factors, the highly variable genes were subset to only transcription factorsfor calculating Spearman correlations. The list of transcription factors was retrieved from the AnimalTFDB v.4.0 database. To identify metabolically stressed cells in the datasets, we used the scanpy score_genes function with default parameters to score the ‘canonical glycolysis’ gene set obtained from the enrichR GO_Biological_Process_2021 database across all neuronal cells from HNOCA and refs. . To estimate the significance of the difference between the correlation of glycolysis scores and whole transcriptomic similarities, and the correlation of glycolysis scores and core transcriptomic identity similarities, we generated 100 subsets of highly variable genes, each with the same size as the highly variable transcription factor. Transcriptomic similarities were calculated on the basis of those subsets, and then correlated with the glycolysis scores. To characterize heterogeneity of telencephalic NPCs and neurons in HNOCA, we first transferred the cell type labels (as indicated as the ‘type’ label in the given metadata) from the human neocortical development atlas to the HNOCA telencephalic NPCs, intermediate progenitor cells and neurons, on the basis of transcriptomic correlation. In brief, each primary atlas cluster we obtained as mentioned above was assigned to a cell type as the most abundant cell type among cells in the cluster. The label of the best-correlated primary cluster was then transferred to every query cell. Given the transferred label, together with the level 2 cell type annotation shown in Fig. 1c, as the annotation label, scPoli was applied to the telencephalic subset of HNOCA for data integration. To benchmark how well different integration strategies recover the neuron subcell type heterogeneity, we generated four different clustering labels: (1) Louvain clustering (resolution, 2) with the original scPoli latent representation; (2) Louvain clustering (resolution, 2) with the updated scPoli representation; (3) Louvain clustering (resolution, 2) with PCA of HNOCA telencephalic subset (based on scaled expression of 3,000 highly variable genes of the telencephalic subset with flavor = ‘seurat’) and (4) Louvain clustering (resolution, 1) for each sample separately (each with 3,000 highly variable genes identified with flavor = ‘seurat’, followed by data scale and PCA). Next, for each sample with at least 500 dorsal telencephalic neurons, the adjusted mutual information scores were calculated between each of those four clustering labels with the transferred cell type label mentioned above as the gold standard, across the dorsal telencephalic neurons as annotated as the level 2 annotation. To create a comprehensive primary atlas of dorsal telencephalic neurons for DE analysis between neural organoids and primary tissues, we subset dorsal telencephalic neurons or neocortical neurons from four different primary atlases. For ref. , cells in five author-defined clusters (60, 57, 79, 45, 65) with high expression of MAP2, DCX and NEUROD6 were selected. For ref. , cells with the following ‘clusterv2 - final’ labels were selected: ‘Neuron_28’, ‘Neuron_34’, ‘GW19_2_29NeuronNeuron’, ‘Neuron_30’, ‘Neuron_66Neuron’, ‘GW18_2_42NeuronNeuron’, ‘Neuron_33’, ‘Neuron_39Neuron’, ‘Neuron_35’, ‘Neuron_63Neuron’, ‘Neuron_9’, ‘Neuron_11’, ‘Neuron_20’, ‘Neuron_22’, ‘Neuron_5Neuron’, ‘Neuron_21’, ‘Neuron_18’, ‘Neuron_101Neuron’, ‘Neuron_17’, ‘Neuron_19’, ‘Neuron_16’, ‘Neuron_50Neuron’, ‘Neuron_12’, ‘Neuron_13’, ‘Neuron_68Neuron’, ‘Neuron_100Neuron’, ‘Neuron_25’, ‘Neuron_27’, ‘Neuron_53Neuron’, ‘Neuron_23’, ‘Neuron_26’, ‘Neuron_24’, ‘Neuron_102Neuron’, ‘Neuron_72Neuron’, ‘Neuron_15’, ‘Neuron_29’ and ‘Neuron_35Neuron’ on the basis of their high expression of NEUROD6 and FOXG1. For ref. , cells dissected from dorsal telencephalon that were annotated as neurons with and only with the VGLUT NTT label were selected. For ref. , cells annotated as excitatory neurons were selected. The curated clusters of the Wang et al. primary atlas, as described earlier, were also subset to those with excitatory neuron labels. The selected dorsal telencephalic neuron subsets of the atlases were merged into the joint neocortical neuron atlas. Next, cells in the joint neocortical neuron atlas were correlated with the average expression profile of each excitatory neuron cluster of the Wang et al. atlas. The cluster label of the best-correlated cluster was assigned to each cell in the joined neocortical neuron atlas, so that cell cluster labels were harmonized for all cells in the atlas. Label-aware data integration was then performed using scPoli. On the basis of the scPoli latent representation, Louvain clustering was performed on the joint neocortical neuron atlas (resolution, 1). This cluster label was transferred to the dorsal telencephalic neurons in HNOCA with max-correlation manner across highly variable genes defined on average transcriptomic profiles of clusters in the joint neocortical neuron atlas. We used scArches to map scRNA-seq data from the neural organoid morphogen screen to both the scANVI model of the human developing brain atlas and the scPoli model of the HNOCA. In both cases, the ‘dataset’ field of the screen data was used as the batch covariate, which indicates belonging to one of the three categories: ‘organoid screen’, ‘secondary organoid screen’ or ‘fetal striatum 21pcw’. For mapping to the primary reference, we used the scvi-tools implementation of scArches without the use of cell type annotations and trained the model for 500 epochs with weight_decay of 0 and otherwise default parameters. For mapping to HNOCA we used scArches through scPoli and trained the model for 500 epochs without unlabelled prototype training. We included 11 scRNA-seq datasets of neural organoids, which were designed to model 10 different neural diseases including microcephaly, amyotrophic lateral sclerosis, Alzheimer’s disease, autism, FXS, schizophrenia, neuronal heterotopia, Pitt–Hopkins syndrome, myotonic dystrophy and glioblastoma. Count matrices and metadata were directly downloaded for the ten datasets with processed data provided in the Gene Expression Omnibus or ArrayExpress. For the dataset with only FASTQ files available, we downloaded the FASTQ files and used Cell Ranger (v.4.0) to map reads to the human reference genome and transcriptome retrieved from Cell Ranger website (GRCh38 v.3.0.0) for gene expression quantification. All datasets were concatenated together with anndata in Python (join = ‘inner’). For each dataset, samples were grouped into either ‘disease’ or ‘control’ as their disease status, with ‘disease’ representing data from patient cell lines, mutant cell lines with disease-related alleles, cells carrying targeting guide RNAs (gRNAs) in CRISPR-based screen and tumour-derived organoids. and ‘control’ representing data from healthy cell lines, mutation-corrected cell lines and cells carrying only non-targeting gRNAs in a CRISPR-based screen. To compare the disease-modelling atlas with the integrated HNOCA, we used scArches to project it to the HNOCA as well as the first-trimester primary human brain scRNA-seq atlas. For projecting to the primary atlas, the same implementation as mentioned above to map HNOCA to the atlas was used. For projecting to HNOCA, the query model was based on the scPoli model pretrained with the HNOCA data, and fineturned with a batch size of 16,384 for a maximum of 30 epochs with 20 pretraining epochs. A nearest neighbour graph was created for the disease-modelling atlas on the basis of the projected latent representation to HNOCA with scanpy (default parameters), with which a UMAP embedding was created with scanpy (default parameters). Next, for both HNOCA and the disease-modelling atlas, cells were represented by the concatenated representation of HNOCA-scPoli and primary-scANVI models. A bipartite wkNN graph was then reconstructed as mentioned above, by identifying 50 nearest neighbours in HNOCA for each disease-modelling atlas cell. On the basis of the bipartite wkNN, the majority voting-based label transfer was applied to transfer the four levels of hierarchical cell type annotation and regional identity to the disease-modelling atlas. For each cell in the disease-modelling atlas, a matched HNOCA metacell was reconstructed on the basis of the above mentioned bipartite wkNN. In brief, for a query cell i and a gene j measured in HNOCA, its matched metacell expression of j, denoted as , is calculated as:[12pt] $$_}^ }=__}_}_}}__}_}}$$′=∑k⊆Niwikekj∑k⊆NiwikHere, Ni represents all HNOCA nearest neighbours of the query cell ci, wik represents the edge weight between query cell i and reference cell k, and ekj represents expression level of gene j in reference cell k. Given the matched HNOCA metacell transcriptomic profile, the similarity between a query cell and its matched cell state in HNOCA is then calculated as the Spearman correlation between the query cell transcriptomic profile and its matched HNOCA metacell transcriptomic profile. To analyse the glioblastoma organoid dataset (GBM-2019), cells from the publication were subset from the integrated disease-modelling atlas. Using scanpy, highly variable genes were identified with default parameters. The log-normalized expression values of the highly variable genes were then scaled across cells, the truncated PCA was performed with the top 20 principal components used for the following analysis. Next, harmonypy, the Python implementation of harmony, was applied to integrate cells from different samples. On the basis of the harmony-integrated embeddings, the neighbour graph was reconstructed. UMAP embeddings and Louvain clusters (resolution, 0.5) were created on the basis of the nearest neighbour graph. Among the 12 identified clusters, cluster-7 and cluster-0, the two clusters with the highest AQP4 expression, were selected for the following DE analysis. To analyse the FXS dataset (FXS-2021), cells from the publication were subset from the integrated disease-modelling atlas. The same procedure of highly variable gene identification, data scaling and PCA as the GBM-2019 dataset was applied. Next, the nearest neighbour graph was created directly on the basis of the top 20 principal components. UMAP embeddings and Louvain clusters (resolution, 1) were then created on the basis of the reconstructed nearest neighbour graph. Among the 30 clusters, cluster-17 and cluster-23, which express EMX1 and FOXG1 and were largely predicted to be dorsal telencephalic NPCs and neurons according to the transferred labels from HNOCA, were selected for the following DE analysis. To compare expression levels of two groups of paired cells, the expression difference per gene of each cell pair is first calculated on the basis of the log-normalized expression values. Next, for each gene to test for DE, its variance over the calculated expression difference per cell pair (σ) is compared with the sum of squared of expression differences (di for gene i) normalized by the number of cell pairs:[12pt] $$_^=_^_}.$$02=∑i=1ndin.Here, an F-test is applied for the comparison, with f = σ/s0, d.f.1 = n − 1 and d.f.2 = n. To construct the HNOCA-CE, we first collected raw count matrices and associated metadata of five more neural organoid studies. For two publications, we obtained them from the sources listed in the ‘Data availability’ section of the paper. For the remaining three publications, count matrices and associated metadata were provided directly by the authors. We subset each dataset to the healthy control cells and removed any cells with fewer than 200 genes expressed. We subset the gene space of every dataset to the 3,000 HVGs of HNOCA while filling the expression of missing genes in the community datasets with zeros. On average, 23% of genes with zero expression were added per dataset. We instantiated a mapping object from the HNOCA-tools package (at commit fe38c52) using the saved scPoli model weights from the HNOCA integration. Using the map_query method of the mapper instance, we projected the community datasets to HNOCA. We used the following training hyperparameters: retrain = ‘partial’, batch_size = 256, unlabeled_prototype_training = False, n_epochs = 10, pretraining_epochs = 9, early_stopping_kwargs = early_stopping_kwargs, eta = 10, alpha_epoch_anneal = 10. We computed the wkNN graph using the compute_wknn method of the mapper instance with k = 100. We transferred the final level_2 cell type labels from HNOCA to the community datasets using this neighbour graph. To obtain the combined representation of HNOCA-CE, we projected HNOCA together with the added community datasets through the trained model and computed a neighbour graph and UMAP from the resulting latent representation. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-024-08172-8.
PMC12595314
Establishment and comprehensive characterization of a subline with highly bone-metastatic propensity derived from the lung adenocarcinoma A549 cell line
Lung cancer remains a leading cause of cancer-related mortality worldwide , with bone metastases representing one of the most common and devastating complications in non-small cell lung cancer (NSCLC). At the time of diagnosis, 20–30 % of NSCLC patients already exhibit bone involvement, and during disease progression, 35–60 % will develop bone metastases [, , ]. These skeletal lesions frequently result in severe skeletal-related events (SREs), including pain, pathological fractures, spinal cord compression, and hypercalcemia, all of which markedly diminish patients' quality of life and worsen clinical outcomes . Although bone-targeted therapies such as denosumab have shown benefits in reducing skeletal complications , their efficacy remains limited and is often accompanied by adverse events . More importantly, the molecular mechanisms that govern skeletal colonization in lung adenocarcinoma remain poorly understood. A major limiting factor is the lack of lung adenocarcinoma cell lines with intrinsic bone-tropic potential, which restricts both mechanistic insights and the development of clinically relevant preclinical models. Disseminated tumor cells (DTCs) reaching bone tissue must undergo complex interactions with various bone microenvironment cells at different stages, including the perivascular niche, the osteogenic niche, and the formation of the vicious cycle. Given the complexity of these interactions, selecting appropriate modeling strategies and suitable cell lines is a critical foundation for advancing research on bone metastasis. Among current strategies, methods such as left ventricular injection , intra-tibial or femur injection , intra-iliac artery injection , and tail artery injection have provided powerful approaches for quantitatively investigating various stages of the bone metastasis cascade. However, these methods often suffer from high experimental variability—an issue particularly pronounced in lung cancer bone metastasis research due to the lack of cell lines with bone-tropic potential, leading to high rates of non-bone metastases and inconsistent tumor engraftment. These limitations underscore the continued need for lung cancer models that are not only stable and reproducible, but also faithfully recapitulate the biological features of bone metastasis. Although a few bone-metastatic lung cancer cell lines have been developed, their number remains limited, and each exhibits specific shortcomings in stability, bone tropism, or biological relevance—issues that constrain their broader applicability. This study aims to address these challenges by establishing a more reliable model. In the present study, we established a highly bone-metastatic subline, A549-BM5, derived from the lung adenocarcinoma A549 cell line through an in vivo selection strategy. We systematically characterized its biological behavior both in vitro and in vivo, demonstrating enhanced bone-homing ability and metastatic tropism. Furthermore, integrative transcriptomic and proteomic analyses revealed key molecular alterations associated with skeletal colonization. This subline offers a robust and biologically relevant platform for dissecting the mechanisms of lung cancer bone metastasis and for evaluating targeted therapies in a preclinical setting. Human lung adenocarcinoma cell line A549 was obtained from the American Type Culture Collection (ATCC). The A549 cells and their derived subline cells were cultured using Ham’s F-12 K medium (Basal Media, L450KJ). All media were supplemented with 10 % fetal bovine serum (FBS) (YeaSen, 40130ES76) and 1 % penicillin–streptomycin solution (Basal Media, S110jv). Cells were incubated at 37 °C in a humidified atmosphere containing 5 % CO2. Female NOD/SCID or BALB/c nude mice aged 6 to 8 weeks were housed in standard specific pathogen-free (SPF) laboratory animal facilities. In the left ventricular injection experiments, 1 × 10 A549-Parental or A549-BM5 cells were injected into the left ventricle under ultrasound guidance. In vivo bioluminescent imaging (BLI) was used to monitor tumor burden. After 4 to 5 weeks, the mice were humanely euthanized, and bone tissues were collected for subsequent experiments. In the intra-iliac artery injection experiments, following previously reported methods , briefly, 0.5 × 10 A549-Parental or A549-BM5 cells were injected into the iliac artery of female BALB/c nude mice aged 6 to 8 weeks. At the end of the eighth day, the mice were humanely euthanized, and bone tissues were collected for subsequent experiments. Micro-CT analysis of excised femora and tibiae was performed. Raw projection images were reconstructed into cross-sectional slices using NRecon (Bruker). Quantitative bone morphometric parameters, including bone integrity, bone volume fraction (BV/TV), trabecular thickness (Tb.Th), trabecular number (Tb.N), and trabecular spacing (Tb.Sp), were calculated using CTAn software (Bruker). Three-dimensional reconstructions of bone microarchitecture and osteolytic lesions were generated with CTvox software (Bruker). All analyses followed ASBMR guidelines. For the cell viability assay, 1 × 10 cells per well were seeded into 96-well plates (Corning, 3599) with three replicate wells per group and incubated overnight to allow for cell attachment. The following day, cell viability was assessed using the CCK8 reagent kit (YeaSen, 40203ES80) according to the manufacturer’s instructions. For the colony formation assay, 6 × 10 cells per well were seeded into 6-well plates (Corning, 3516) and cultured for 10 to 14 days to enable colony development. Surviving colonies were fixed with methanol for 30 min and subsequently stained with 0.1 % crystal violet solution (Beyotime, C0121) for another 30 min. After thorough washing and air-drying, the number of colonies was counted. Cells were collected from 6-cm cell culture dishes, washed twice with phosphate-buffered saline (PBS), and then fixed in 70 % ethanol at − 20 °C overnight. Following centrifugation and removal of the supernatant, the cells were washed once with PBS and stained with 500 µL of propidium iodide (PI) staining solution (BD Biosciences, 550825) for 10 min. The DNA content was analyzed using a flow cytometer (CytoFLEX S; Beckman Coulter). Migration and invasion assays were performed using 8 μm pore chambers (Corning, 353097). For the migration assay, 5 × 10 cells in 200 µL of serum-free culture medium were seeded into the upper chamber. For the invasion assay, the upper chamber membranes were pre-coated with Matrigel (Corning, 356234), and 1 × 10 cells in 200 µL of serum-free culture medium were seeded into the upper chamber. In both assays, culture medium containing 10 % fetal bovine serum (FBS) was added to the lower chamber. After incubation for 16–20 h, the cells that had migrated or invaded through the membrane were fixed with methanol for 30 min and stained with 0.1 % crystal violet for 15–20 min. Nine random fields were photographed under a microscope (Olympus, IX73) at a magnification of 100×. Cells were washed with prechilled PBS triplicately, then lysed in buffer (Thermo, 78510) supplemented with protease inhibitors (YeaSen, 20123ES50) and phosphatase inhibitors (YeaSen, 20109ES20). Cells were scraped and thoroughly lysed on ice for 30 min. Lysates were centrifuged at 12,000 rpm for 15–20 min, and the supernatants were collected. Protein concentrations were measured using a BCA assay kit (Thermo Fisher Scientific, 23227), and samples were normalized to equivalent concentrations. Following the addition of 5× protein loading buffer (Yeasen, cat. no. 20315ES20), samples were denatured at 100 °C for 10 min and stored at −80 °C. Proteins were then subjected to sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) and transferred onto polyvinylidene fluoride (PVDF) membranes (Bio-Rad, 1620177). Membranes were blocked at room temperature for 2 h prior to incubation with primary antibodies. The primary antibodies used in this study were as follows: anti-N-Cadherin(13116, CST), anti-E-Cadherin(3195, CST), anti-ZO-1(8193, CST), anti-ZEB1(3396, CST), anti-Vimentin(5741, CST), anti-Claudin-1(13255, CST), anti-β-Catenin(8480, CST), anti-MMP2(RM8377, Biodragon), anti-MMP3(RM8188, Biodragon), anti-MMP9(RM3763, Biodragon), anti-MMP13(RM8226, Biodragon), anti-CCL3(RM2987, Biodragon), anti-BMP2(BD-PT5651, Biodragon), anti-BMP3(BD-PT0499, Biodragon), anti-BMP6(BD-PT5655, Biodragon), anti-BMP7(BD-PT0503, Biodragon), anti-OPG(ab183910, Abcam), anti-RANKL(BD-PT5404, Biodragon), anti-OPN(RM4342, Biodragon), anti-COL1A1(RM0325, Biodragon), anti-ICAM-1(BD-PT2269, Biodragon), anti-RANK(BD-PT5881, Biodragon), anti-SP7(BD-PN0332, Biodragon),anti-Beta-Actin(HRP-66009, Proteintech) and anti-Vinculin(26520-1-AP, Proteintech). Bands were visualized by enhanced chemiluminescence (Share-Bio, sb-wb011), and densitometry measurements of the bands were acquired with Quantity One software (Bio-Rad). Bone marrow mesenchymal stromal cells (BMSCs) and bone marrow–derived monocytes/macrophages (BMMs) were isolated from 6 to 8-week-old C57BL/6J mice. Under sterile conditions, femurs and tibias were dissected, and the ends of the bones were cut. Bone marrow cells were collected by centrifugation at 10,000×g for 5 s, and the pellet was subjected to red blood cell lysis for 5 min on ice. Cells were then centrifuged at 450 × g for 3 min, washed once with PBS, and resuspended in α-MEM complete medium. The suspension was plated in 10-cm dishes, and after 24 h, adherent cells were predominantly BMSCs, while non-adherent cells in the supernatant contained mainly monocytes. Bone microenvironment cell recruitment assays were performed using 8 μm pore Transwell chambers (Corning, 353097). A549-Parental and A549-BM5 cells were seeded at appropriate densities in 24-well plates. Bone microenvironment cells (5 × 10 cells in 200 µL of serum-free culture medium) were seeded into the upper chamber. After incubation for 16–20 h, the cells that migrated through the membrane were fixed with methanol for 30 min and stained with 0.1 % crystal violet for 15–20 min. Nine random fields per chamber were photographed under a microscope (Olympus, IX73) at 100× magnification. Statistical analysis was performed using GraphPad Prism 9. Under sterile and light-protected conditions, osteogenic induction medium was prepared by mixing 45 mL α-MEM medium, 5 mL fetal bovine serum, 500 µL penicillin–streptomycin solution, 50 µL of 50 mg/mL vitamin C solution, 5 µL of 1 mM dexamethasone solution, and 500 µL of 1 M β-glycerophosphate solution. The medium was stored at 4 °C for future use. Bone marrow mesenchymal stromal cells (BMSCs) or MC3T3-E1 subclone 14 cells were cultured to sufficient quantities and digested with trypsin. Cells were seeded at a density of 5 × 104 cells per well in 24-well plates. After cell adhesion, the medium was replaced with osteogenic induction medium, with medium changes every other day under light-protected conditions. According to the experimental design, conditioned medium (CM) derived from A549-Parental or A549-BM5 cells was mixed at a 1:1 ratio with osteogenic induction medium (final concentration 50 % CM) during the culture period. After 14 days of induction, Alkaline Phosphatase (ALP) staining was performed. Osteoclast differentiation induction medium was prepared by adding cytokines RANKL (50 ng/mL) and M−CSF (10 ng/mL) to α-MEM complete medium. Bone marrow macrophages (BMMs) were cultured to sufficient quantities. Adherent BMMs were digested with Versene dissociation solution and seeded into 96-well plates at a density of 1 × 10 cells per well. After cell adhesion, the medium was replaced with osteoclast differentiation induction medium to induce differentiation, with medium changes every other day. According to the experimental design, conditioned medium (CM) derived from A549-Parental or A549-BM5 cells was mixed at a 1:1 ratio with osteoclast differentiation medium (final concentration 50 % CM) during the culture period. MC3T3-E1 subclone 14 cells were seeded at an appropriate density in 6-well plates. After the cells adhered, 200,000 cells in 1000 μL suspension were added to each well and incubated at 37 °C for 30 min. The supernatant was discarded, and the wells were washed with phosphate-buffered saline (PBS). Cell images were captured using a fluorescence microscope at 100 × magnification. Five random fields were selected per well. The number of cells was counted, and the average cell number for each sample was calculated. All experiments were performed in triplicate. Total RNA was extracted using the TRIzol reagent (Invitrogen, CA, USA) according to the manufacturer’s protocol. RNA purity and quantification were evaluated using the NanoDrop 2000 spectrophotometer (Thermo Scientific, USA). RNA integrity was assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Then the libraries were constructed using VAHTS Universal V6 RNA-seq Library Prep Kit according to the manufacturer’s instructions. The transcriptome sequencing and analysis were conducted by OE Biotech Co., Ltd. (Shanghai, China). The libraries were sequenced on an llumina Novaseq 6000 platform and 150 bp paired-end reads were generated. About 50 M raw reads for each sample were generated. Raw reads of fastq format were firstly processed using fastp and the low quality reads were removed to obtain the clean reads. Then about 48 M clean reads for each sample were retained for subsequent analyses. The clean reads were mapped to the reference genome using HISAT2 . FPKM of each gene was calculated and the read counts of each gene were obtained by HTSeq-count . PCA analysis were performed using R (v 3.2.0) to evaluate the biological duplication of samples. Differential expression analysis was performed using the DESeq2 . Q value <0.05 and foldchange >2 or foldchange <0.5 was set as the threshold for significantly differential expression gene (DEGs). Hierarchical cluster analysis of DEGs was performed using R (v 3.2.0) to demonstrate the expression pattern of genes in different groups and samples. The radar map of top 30 genes was drew to show the expression of up-regulated or down-regulated DEGs using R packet ggradar. Based on the hypergeometric distribution, GO , KEGG pathway, Reactome and WikiPathways enrichment analysis of DEGs were performed to screen the significant enriched term using R (v 3.2.0), respectively. R (v 3.2.0) was used to draw the column diagram, the chord diagram and bubble diagram of the significant enrichment term. Gene Set Enrichment Analysis (GSEA) was performed using GSEA software . The analysis was used a predefined gene set, and the genes were ranked according to the degree of differential expression in the two types of samples. Then it is tested whether the predefined gene set was enriched at the top or bottom of the ranking list. Nanoflow reversed-phase chromatography was performed on an EASY-nLC 1200 system (Thermo Fisher Scientific). Peptides were separated in 90 min at a flow rate of 300 nL/min on a 25 cm × 75 μm column (1.6 μm C18, ionopticks). Mobile phases A and B were 0.1 vol% formic acid solution and 80:20:0.1 vol% ACN: water: formic acid, respectively. The total run was 90 min (0 ∼ 66 min, 3–27 % B；66 ∼ 73 min, 27–46 % B；73 ∼ 84 min, 46–100 % B；84 ∼ 90 min, 100 % B) or 60 min (0 ∼ 45 min, 5–27 % B；45 ∼ 50 min, 27–46 % B；50 ∼ 55 min, 46–100 % B；55 ∼ 60 min, 100 % B). Liquid chromatography was coupled online to a hybrid TIMS quadrupole TOF mass spectrometer (Bruker timsTOF Pro) via a CaptiveSpray nano-electrospray ion source. Capillary voltage was 1.5 kV, dry gas temperature was 180 °C, and dry gas flow rate was 3.0 L/min. The dual TIMS analyzer was operated at a fixed duty cycle close to 100 % using equal accumulation and ramp times of 100 ms. We performed DDA in PASEF mode with 10 PASEF scans per topN acquisition cycle. The full MS scan range was set from 100 to 1700 m/z. The ion mobility range was 0.75–1.4 vs/cm2, and the collision energy range was 20–59 ev. The LC-MS/MS raw data were imported in Maxquant (Version1.6.17.0) for labeling free quantification analysis and the search engine was Andromeda. The database was offered by researchers. All experiments were performed at least in triplicate unless otherwise specified. Data are represented as mean ± standard deviation (SD). P values were calculated using the unpaired Student’s t test or Mann–Whitney U test, depending on the normality and homoscedasticity of the data. Statistical significance was defined as follows: n.s., P > 0.05; ∗, P < 0.05; ∗∗, P < 0.01; ∗∗∗, P < 0.001; ∗∗∗∗, P < 0.0001. Graphs were generated using GraphPad Prism 9, with bar charts displaying individual data points together with mean ± SD. To establish a highly bone-metastatic subline derived from the lung adenocarcinoma A549 cell line, we utilized an in vivo selection strategy based on left ventricle injection (Fig. 1A). A549 cells labeled with GFP and luciferase were injected into the left ventricles of NOD/SCID mice to establish a lung cancer bone metastasis mouse model. In-vivo bioluminescent imaging (BLI) was immediately used to confirm successful injections, and BLI was then continuously utilized to monitor the progression of bone metastatic lesions. Approximately one month after injection, detectable bone metastatic lesions had emerged. At this point, the mice were humanely euthanized, and the bone metastatic tissues were collected, minced, and subjected to primary cell culture for subsequent in-vivo passaging. This sequence of left ventricle injection, primary culture, and reinjection was repeated for five consecutive cycles. During these cycles, both the incidence and the onset time of bone metastases significantly increased, thus validating the efficacy of this selection strategy. The cells obtained after five rounds of in-vivo selection were designated as A549-BM5. Fluorescence microscopy verified that all cells expressed GFP, demonstrating the successful derivation of a highly purified population of A549-derived cells from the xenografts (Fig. 1B). Bright-field microscopy showed that, in contrast to the typical adherent epithelial-like morphology of parental A549 cells, A549-BM5 cells adopted a more elongated, spindle-shaped appearance (Fig. 1B).Fig. 1The in vivo selection of a highly bone-metastatic A549 subline. (A) Schematic diagram illustrating the establishment and characterization of the A549-BM5 subline. (B) Morphological characteristics of A549-Parental and A549-BM5 cells observed under bright-field microscopy and fluorescence microscopy (GFP channel) at 100× and 200× magnifications. The in vivo selection of a highly bone-metastatic A549 subline. (A) Schematic diagram illustrating the establishment and characterization of the A549-BM5 subline. (B) Morphological characteristics of A549-Parental and A549-BM5 cells observed under bright-field microscopy and fluorescence microscopy (GFP channel) at 100× and 200× magnifications. To comprehensively clarify the in vitro characteristics of the A549-BM5 cell line, we conducted a detailed comparison between it and its parental cells with respect to proliferation, migration, and epithelial-mesenchymal status. Cell Counting Kit-8 (CCK-8) assays indicated that the A549-BM5 cell line did not show a proliferative advantage over the parental A549 cells in vitro (Fig. 2A). Additionally, the colony formation ability of the A549-BM5 cells was comparable to that of the parental cells (Fig. 2B). Cell cycle analysis revealed a reduction in the proportion of cells in the G0/G1 phase and an elevation in the S-phase population, while the proportion of cells in the G2/M phase remained largely unchanged when compared to the parental cells (Fig. 2C). Transwell assays clearly demonstrated that the A549-BM5 cells exhibited significantly enhanced migration and invasion capabilities compared to their parental counterparts (Fig. 2D). Western blot analysis showed a clear reduction of E-cadherin in A549-BM5 compared with parental A549 cells. Minor changes were also observed in N-cadherin, β-catenin, and Snail (Fig. 2E). Notably, we observed that E-cadherin is significantly downregulated, while N-cadherin, β-catenin, and Snail are upregulated. This pattern suggests the cells are undergoing epithelial-mesenchymal transition (EMT), becoming more migratory and invasive.Fig. 2The in vitro functions and epithelial-mesenchymal status of A549-BM5 cell subline (A) Cell counting kit-8 assay: in vitro proliferation analysis of A549-Parental and A549-BM5 cells (n = 3). (B) Colony formation ability of A549-Parental and A549-BM5 cells (n = 3). (C) Flow cytometry was used to detect changes in the DNA content distribution in the A549-Parental and A549-BM5 cells (n = 3). (D) Transwell assay assessing the migration and invasion abilities of A549-Parental and A549-BM5 (n = 9). (E) Western blot analysis of E-Cadherin, N-Cadherin, β-Catenin and Snail in A549-Parental and A549-BM5 cells Data are represented as mean ± SD. P values were calculated using the unpaired Student’s t test or Mann-Whitney U test (Depending on the normality and homoscedasticity of the data). n.s., P > 0.05, ∗, P < 0.05; ∗∗, P < 0.01, ∗∗∗, P < 0.001, ∗∗∗∗, P < 0.0001. The in vitro functions and epithelial-mesenchymal status of A549-BM5 cell subline (A) Cell counting kit-8 assay: in vitro proliferation analysis of A549-Parental and A549-BM5 cells (n = 3). (B) Colony formation ability of A549-Parental and A549-BM5 cells (n = 3). (C) Flow cytometry was used to detect changes in the DNA content distribution in the A549-Parental and A549-BM5 cells (n = 3). (D) Transwell assay assessing the migration and invasion abilities of A549-Parental and A549-BM5 (n = 9). (E) Western blot analysis of E-Cadherin, N-Cadherin, β-Catenin and Snail in A549-Parental and A549-BM5 cells Data are represented as mean ± SD. P values were calculated using the unpaired Student’s t test or Mann-Whitney U test (Depending on the normality and homoscedasticity of the data). n.s., P > 0.05, ∗, P < 0.05; ∗∗, P < 0.01, ∗∗∗, P < 0.001, ∗∗∗∗, P < 0.0001. Tumor progression during bone metastasis relies on intricate interactions with a variety of cells in the bone microenvironment . To investigate the interaction capacity of A549-BM5 cells with these cellular constituents, we examined their impacts on osteoclasts, osteoblasts, and their progenitor cells. Using Transwell assays, we found that A549-BM5 cells exhibited a significantly enhanced ability to recruit osteoclast precursors compared with parental cells. Both RAW264.7 cells and bone marrow–derived macrophages (BMMs) were used as osteoclast precursor models, and in both conditions A549-BM5 cells demonstrated stronger recruitment (Fig. 3A). For osteoclast differentiation experiments, we specifically used BMMs isolated from C57BL/6J mice. After 14 days of culture under osteoclastogenic induction, A549-BM5 CM significantly promoted osteoclast formation, as confirmed by TRAP staining and per-well quantification (Fig. 3B). At the transcriptional level, RT-qPCR further revealed significant upregulation of osteoclast differentiation markers including Tnfrsf11a, Ctsk, Car2, Calcr, Tm7sf4, Acp5, and Nfatc1 in the A549-BM5 group (Fig. 3C).Fig. 3The capacity of A549-BM5 cells to interact with bone microenvironment cells. (A) Transwell assays showing the recruitment capacity of A549-Parental and A549-BM5 cells toward osteoclast precursors. Both RAW264.7 cells and bone marrow–derived macrophages (BMMs) were used as osteoclast precursor models. Quantification indicates significantly stronger recruitment in the A549-BM5 group (n = 6). (B) Osteoclast differentiation assay using BMMs. After 14 days of induction with conditioned medium (CM) from A549-Parental or A549-BM5 cells, TRAP staining and per-well quantification confirmed markedly enhanced osteoclast formation in the A549-BM5 group (n = 3). (C) RT-qPCR analysis of osteoclast differentiation markers (Tnfrsf11a, Ctsk, Car2, Calcr, Tm7sf4, Acp5, and Nfatc1) in BMMs cultured for 14 days with conditioned medium (CM) from A549-Parental or A549-BM5 cells. Gene expression was normalized to β-actin and calculated using the 2 method (n = 6). (D) Transwell assays showing recruitment capacity of A549-Parental and A549-BM5 cells toward osteoblast progenitors (MC3T3-E1 subclone 14 and BMSCs). Quantification demonstrates enhanced recruitment by A549-BM5 cells (n = 6). (E) Osteoblast differentiation assay. MC3T3-E1 subclone 14 cells and BMSCs were induced for 14 days with osteogenic medium containing CM from A549-Parental or A549-BM5 cells. ALP staining demonstrated stronger osteogenic differentiation in the A549-BM5 group (n = 3). (F) Conditioned media (CM) from A549-Parental and A549-BM5 cells were used to treat MC3T3 E1 sub14 cells for 14 Days, and the expression levels of CCL3, BMP6, COL1A1 and ICAM-1 were detected using Western blot analysis. (G) RT-qPCR analysis of osteoblast differentiation markers (Alpl, Runx2, and Bglap) in MC3T3-E1 subclone 14 cells cultured for 14 days with conditioned medium (CM) from A549-Parental or A549-BM5 cells. Gene expression was normalized to β-actin and calculated using the 2 method (n = 6). (H) Adhesion assay of GFP-labeled A549-Parental and A549-BM5 cells to MC3T3-E1 subclone 14 cells. Quantification showed that A549-BM5 cells exhibited significantly greater adhesion than parental cells (n = 9). (I) GSEA analysis shows that the GO:0007155 gene set (heterophilic cell–cell adhesion via plasma membrane cell adhesion molecules) is enriched in A549-BM5 cells compared with the A549-Parental group. The heatmap below displays the expression of leading-edge genes in the corresponding samples.Data are represented as mean ± SD. P values were calculated using the unpaired Student’s t test or Mann-Whitney U test (Depending on the normality and homoscedasticity of the data). n.s., P > 0.05, ∗, P < 0.05; ∗∗, P < 0.01, ∗∗∗, P < 0.001, ∗∗∗∗, P < 0.0001. The capacity of A549-BM5 cells to interact with bone microenvironment cells. (A) Transwell assays showing the recruitment capacity of A549-Parental and A549-BM5 cells toward osteoclast precursors. Both RAW264.7 cells and bone marrow–derived macrophages (BMMs) were used as osteoclast precursor models. Quantification indicates significantly stronger recruitment in the A549-BM5 group (n = 6). (B) Osteoclast differentiation assay using BMMs. After 14 days of induction with conditioned medium (CM) from A549-Parental or A549-BM5 cells, TRAP staining and per-well quantification confirmed markedly enhanced osteoclast formation in the A549-BM5 group (n = 3). (C) RT-qPCR analysis of osteoclast differentiation markers (Tnfrsf11a, Ctsk, Car2, Calcr, Tm7sf4, Acp5, and Nfatc1) in BMMs cultured for 14 days with conditioned medium (CM) from A549-Parental or A549-BM5 cells. Gene expression was normalized to β-actin and calculated using the 2 method (n = 6). (D) Transwell assays showing recruitment capacity of A549-Parental and A549-BM5 cells toward osteoblast progenitors (MC3T3-E1 subclone 14 and BMSCs). Quantification demonstrates enhanced recruitment by A549-BM5 cells (n = 6). (E) Osteoblast differentiation assay. MC3T3-E1 subclone 14 cells and BMSCs were induced for 14 days with osteogenic medium containing CM from A549-Parental or A549-BM5 cells. ALP staining demonstrated stronger osteogenic differentiation in the A549-BM5 group (n = 3). (F) Conditioned media (CM) from A549-Parental and A549-BM5 cells were used to treat MC3T3 E1 sub14 cells for 14 Days, and the expression levels of CCL3, BMP6, COL1A1 and ICAM-1 were detected using Western blot analysis. (G) RT-qPCR analysis of osteoblast differentiation markers (Alpl, Runx2, and Bglap) in MC3T3-E1 subclone 14 cells cultured for 14 days with conditioned medium (CM) from A549-Parental or A549-BM5 cells. Gene expression was normalized to β-actin and calculated using the 2 method (n = 6). (H) Adhesion assay of GFP-labeled A549-Parental and A549-BM5 cells to MC3T3-E1 subclone 14 cells. Quantification showed that A549-BM5 cells exhibited significantly greater adhesion than parental cells (n = 9). (I) GSEA analysis shows that the GO:0007155 gene set (heterophilic cell–cell adhesion via plasma membrane cell adhesion molecules) is enriched in A549-BM5 cells compared with the A549-Parental group. The heatmap below displays the expression of leading-edge genes in the corresponding samples.Data are represented as mean ± SD. P values were calculated using the unpaired Student’s t test or Mann-Whitney U test (Depending on the normality and homoscedasticity of the data). n.s., P > 0.05, ∗, P < 0.05; ∗∗, P < 0.01, ∗∗∗, P < 0.001, ∗∗∗∗, P < 0.0001. To investigate osteoblast-related interactions, we employed MC3T3-E1 subclone 14 cells and primary BMSCs. Transwell assays showed that A549-BM5 cells significantly enhanced recruitment of osteoblast progenitors (Fig. 3D). Under osteogenic induction with CM for 14 days, osteoblast differentiation was markedly increased in the A549-BM5 group, as demonstrated by ALP staining (Fig. 3E). We next evaluated the expression of potential mediators in MC3T3-E1 subclone 14 cells treated with conditioned medium (CM) for 14 Days by Western blot. A549-BM5 CM induced a clear increase in CCL3, with BMP6, COL1A1 and ICAM-1 showing a tendency to be higher (Fig. 3F). At the transcriptional level, RT-qPCR analysis of MC3T3-E1 subclone 14 cells cultured for 14 days under CM treatment further confirmed upregulation of osteoblast differentiation markers Alpl, Runx2, and Bglap in the A549-BM5 group (Fig. 3G). Finally, adhesion assays showed that A549-BM5 cells displayed superior attachment to MC3T3-E1 cells compared with parental A549 (Fig. 3H). Consistently, GSEA revealed enrichment of the GO:0007155 gene set (heterophilic cell–cell adhesion via plasma membrane cell adhesion molecules) in A549-BM5 cells compared with the A549-Parental group, and the accompanying heatmap shows the expression of the leading-edge genes, including VCAM1, ICAM1, NECTIN4, and CRTAM, in the corresponding samples (Fig. 3I). To investigate the bone metastatic potential of A549-BM5 cells in vivo, we established mouse models using both left ventricular and intra-iliac artery injections. Following left ventricular injection, the progression of skeletal metastasis was dynamically monitored by bioluminescent imaging (BLI). All mice injected with A549-BM5 cells developed substantial bone metastatic lesions within 14 days, whereas parental A549 cells showed only limited colonization. At the endpoint, A549-BM5–bearing mice exhibited a significantly elevated bone metastatic burden, which was further confirmed by gross anatomy, X-ray imaging, micro-CT reconstruction, and histological examination (Fig. 4A–C). Quantitative micro-CT analysis revealed pronounced osteolytic bone destruction in A549-BM5 mice, including reduced bone integrity, decreased bone volume fraction (BV/TV), trabecular thickness (Tb.Th), and trabecular number (Tb.N), along with increased trabecular spacing (Tb.Sp) (Fig. 4D–H).Fig. 4The in vivo metastatic capacity of A549-BM5 (A) An intra-cardiac injection was used to establish bone metastasis CDX models of A549-Parental and A549-BM5 cells. In vivo bioluminescent imaging (BLI) was employed to detect the bone metastatic burden (n = 6).(B) Representative gross anatomy, X-ray imaging and H&E staining of bone lesions from A549-Parental and A549-BM5 groups. Extensive osteolytic destruction was observed in A549-BM5–bearing mice. (C) Representative three-dimensional micro-CT images from A549-Parental and A549-BM5 groups. (D) Quantitative analysis of bone integrity, showing significantly reduced integrity in the A549-BM5 group compared with A549-Parental (n = 6). (E) Quantitative analysis of bone volume fraction (BV/TV), revealing markedly decreased bone mass in the A549-BM5 group (n = 6). (F) Quantitative analysis of trabecular thickness (Tb.Th), showing thinner trabeculae in A549-BM5 mice (n = 6). (G) Quantitative analysis of trabecular number (Tb.N), indicating fewer trabeculae in the A549-BM5 group (n = 6). (H) Quantitative analysis of trabecular spacing (Tb.Sp), showing significantly increased spacing in the A549-BM5 group (n = 6). (H) Intra-iliac artery injection model of hindlimb bone metastasis. Representative longitudinal BLI monitoring shows significantly higher tumor burden in the A549-BM5 group compared to A549-Parental. Quantification of bone metastasis burden over time (left panel) and ex vivo BLI of hindlimb bone tissues at day 8 (right panel) are shown (n = 10). Data are represented as mean ± SD. P values were calculated using the unpaired Student’s t test or Mann-Whitney U test (Depending on the normality and homoscedasticity of the data). n.s., P > 0.05, ∗, P < 0.05; ∗∗, P < 0.01, ∗∗∗, P < 0.001, ∗∗∗∗, P < 0.0001. The in vivo metastatic capacity of A549-BM5 (A) An intra-cardiac injection was used to establish bone metastasis CDX models of A549-Parental and A549-BM5 cells. In vivo bioluminescent imaging (BLI) was employed to detect the bone metastatic burden (n = 6).(B) Representative gross anatomy, X-ray imaging and H&E staining of bone lesions from A549-Parental and A549-BM5 groups. Extensive osteolytic destruction was observed in A549-BM5–bearing mice. (C) Representative three-dimensional micro-CT images from A549-Parental and A549-BM5 groups. (D) Quantitative analysis of bone integrity, showing significantly reduced integrity in the A549-BM5 group compared with A549-Parental (n = 6). (E) Quantitative analysis of bone volume fraction (BV/TV), revealing markedly decreased bone mass in the A549-BM5 group (n = 6). (F) Quantitative analysis of trabecular thickness (Tb.Th), showing thinner trabeculae in A549-BM5 mice (n = 6). (G) Quantitative analysis of trabecular number (Tb.N), indicating fewer trabeculae in the A549-BM5 group (n = 6). (H) Quantitative analysis of trabecular spacing (Tb.Sp), showing significantly increased spacing in the A549-BM5 group (n = 6). (H) Intra-iliac artery injection model of hindlimb bone metastasis. Representative longitudinal BLI monitoring shows significantly higher tumor burden in the A549-BM5 group compared to A549-Parental. Quantification of bone metastasis burden over time (left panel) and ex vivo BLI of hindlimb bone tissues at day 8 (right panel) are shown (n = 10). Data are represented as mean ± SD. P values were calculated using the unpaired Student’s t test or Mann-Whitney U test (Depending on the normality and homoscedasticity of the data). n.s., P > 0.05, ∗, P < 0.05; ∗∗, P < 0.01, ∗∗∗, P < 0.001, ∗∗∗∗, P < 0.0001. To further assess whether A549-BM5 cells could effectively mimic early biological events of bone colonization, we established a unilateral hindlimb metastasis model via intra-iliac artery injection. Notably, differences in metastatic burden became apparent as early as day 5–7, with BLI and ex vivo analyses showing markedly higher tumor load in A549-BM5 compared to parental A549 controls (Fig. 4I). We conducted a comprehensive characterization of A549-BM5 cells at the transcriptomic and proteomic levels. mRNA sequencing was performed on A549-BM5 cells and their parental counterparts. Using a filtering threshold of a fold change greater than 2 and an adjusted p-value less than 0.05, we identified 1,371 differentially expressed genes (DEGs), including 589 upregulated and 782 downregulated genes (Fig. 5A). Principal component analysis (PCA) revealed a clear separation between A549-BM5 cells and their parental counterparts (Fig. 5B). Gene Ontology (GO) enrichment analysis identified multiple gene sets associated with epithelial-mesenchymal transition (EMT), cell–cell adhesion, the WNT signaling pathway, the NF-κB signaling pathway, and the transforming growth factor (TGF) pathway (Fig. 5C), all of which known to play key roles in bone metastasis. Additionally, bone-related gene sets, including bone tissue morphogenesis and development, were significantly enriched (Fig. 5D), which may indicate that A549-BM5 cells acquire osteogenic-like transcriptional features. Gene Set Enrichment Analysis (GSEA) further revealed enrichment of gene sets related to cell migration, adhesion, neuron differentiation, and cytokine signaling (Fig. 5E).Fig. 5Analysis of the transcriptomic and proteomic characteristics of A549-BM5. (A) The volcano plot generated from transcriptome analyses of RNA sample from A549-Parental and A549-BM5 (n = 3). (B) Principal component analysis (PCA) of RNA sequencing data from A549-Parental and A549-BM5 cells. (C) The bubble chart illustrates the Gene Ontology (GO) enrichment analysis of differentially expressed genes (DEGs) between A549-Parental and A549-BM5 cells. (D) The bubble chart illustrates the enrichment of bone-related gene sets in the Gene Ontology (GO) enrichment analysis of differentially expressed genes (DEGs) between A549-Parental and A549-BM5 cells. (E) Gene Set Enrichment Analysis (GSEA) using the GO database in the transcriptomic analysis. (F) The volcano plot generated from proteomic analyses of A549-Parental and A549-BM5 (n = 3). (G) Principal component analysis (PCA) of proteomic data from A549-Parental and A549-BM5 cells. (H) The bubble chart illustrates the Gene Ontology (GO) enrichment analysis of differentially expressed proteins between A549-Parental and A549-BM5 cells; (I) Gene Set Enrichment Analysis (GSEA) using the GO database in the proteomic analysis. (J) The Nine-square Grid displays genes that have common changing trends between proteomic and transcriptomic analyses. (K–L) Using the LUAD cohort from TCGA, two random forest models were constructed to evaluate the contribution of genes to metastasis (M1). The ROC curves respectively display the classification performance of models established using commonly altered genes (K) and characteristic genes selected from the transcriptomic data (L). (M) Mean Decrease Gini of genes exhibiting concurrent changes in transcriptomic and proteomic profiles. (N) Mean Decrease Gini of characteristic genes in transcriptomic profiles. Analysis of the transcriptomic and proteomic characteristics of A549-BM5. (A) The volcano plot generated from transcriptome analyses of RNA sample from A549-Parental and A549-BM5 (n = 3). (B) Principal component analysis (PCA) of RNA sequencing data from A549-Parental and A549-BM5 cells. (C) The bubble chart illustrates the Gene Ontology (GO) enrichment analysis of differentially expressed genes (DEGs) between A549-Parental and A549-BM5 cells. (D) The bubble chart illustrates the enrichment of bone-related gene sets in the Gene Ontology (GO) enrichment analysis of differentially expressed genes (DEGs) between A549-Parental and A549-BM5 cells. (E) Gene Set Enrichment Analysis (GSEA) using the GO database in the transcriptomic analysis. (F) The volcano plot generated from proteomic analyses of A549-Parental and A549-BM5 (n = 3). (G) Principal component analysis (PCA) of proteomic data from A549-Parental and A549-BM5 cells. (H) The bubble chart illustrates the Gene Ontology (GO) enrichment analysis of differentially expressed proteins between A549-Parental and A549-BM5 cells; (I) Gene Set Enrichment Analysis (GSEA) using the GO database in the proteomic analysis. (J) The Nine-square Grid displays genes that have common changing trends between proteomic and transcriptomic analyses. (K–L) Using the LUAD cohort from TCGA, two random forest models were constructed to evaluate the contribution of genes to metastasis (M1). The ROC curves respectively display the classification performance of models established using commonly altered genes (K) and characteristic genes selected from the transcriptomic data (L). (M) Mean Decrease Gini of genes exhibiting concurrent changes in transcriptomic and proteomic profiles. (N) Mean Decrease Gini of characteristic genes in transcriptomic profiles. To complement the transcriptomic data, we performed quantitative proteomics analysis on A549-BM5 and parental cells. Based on filtering criteria of a fold change greater than 1.5 and an adjusted p-value less than 0.05, we identified 144 upregulated proteins and 193 downregulated proteins (Fig. 5F). PCA of the proteomic data also demonstrated a clear separation between A549-BM5 cells and their parental counterparts (Fig. 5G). GO enrichment analysis and GSEA of the proteomic data revealed significant enrichment of pathways involved in cell migration, adhesion, Ras signaling, mitochondrial function, histone modification, EMT, and various metabolism − related pathways (Fig. 5H-I). Integrated analysis of transcriptomic and proteomic data identified 21 genes that were concurrently upregulated and 8 genes that were concurrently downregulated at both levels (Fig. 5H–J). To explore the potential roles of these key genes in the distant metastasis of lung adenocarcinoma (LUAD) patients, we retrieved LUAD patient data from the TCGA database. Patients were stratified into M0 (no distant metastasis) and M1 (distant metastasis) groups according to their metastatic status. By utilizing a random forest model, we assessed the contribution of target genes to the metastatic status (Fig. 5K–L). Potential biomarkers for LUAD metastasis were identified based on their Mean Decrease Gini scores obtained from the random forest model (Fig. 5M–N). Recent single-cell analyses have revealed substantial intra-line heterogeneity in gene expression and metastatic potential within cancer cell lines, supporting the rationale for selecting organ-tropic sublines via in vivo selection [, , ]. Given the unique physicochemical properties and cytokine milieu of the bone and bone marrow microenvironment—characterized by hypoxia , high extracellular calcium levels , abundant matrix proteins such as collagen and osteopontin , and soluble factors like TGF-β, CXCL12, and RANKL —in vivo circulation selection enables effective isolation and enrichment of tumor subpopulations with a propensity for bone metastasis. The A549 cell line is a widely used model for studying lung adenocarcinoma . However, our previous study revealed that although A549 cells possess the capacity to metastasize to bone after intracardiac injection, their metastatic efficiency is limited, with low and variable engraftment and a dispersed pattern of skeletal colonization. These features indicate the absence of strong bone-tropic preference and reduce their suitability for generating consistent and biologically relevant bone metastasis models . Such characteristics make A549 suboptimal for quantitative studies specifically focused on bone metastasis. In the following analysis, we examined the key features of currently available lung cancer cell models with high bone-metastatic potential. Hung et al. obtained results similar to our previous findings when establishing a bone metastasis model by injecting A549 cells via the left ventricle . While bone lesions were observed, notable metastatic spread also occurred in other organs, including the lungs and abdominal viscera, potentially complicating the interpretation of bone-specific metastatic mechanisms. Tan et al. established a bone metastasis model via intra-tibial injection of A549 cells . While this approach efficiently induces bone lesions, it bypasses key early steps of the metastatic cascade, limiting its relevance for mechanistic studies. Moreover, the invasive procedure introduces tissue injury and allows cancer cells to enter circulation, increasing the risk of systemic spread. Recent findings also suggest that injury-induced bone remodeling involving NG2 stromal cells may artificially promote tumor colonization, further compromising model specificity . This limitation was partially addressed by Cai et al., who established the A549L6 subline through in vivo selection, which preferentially metastasizes to the spine. While A549L6 partially mimics the clinical pattern of lung cancer bone metastasis—typically initiating in the axial skeleton—it is less amenable to in vivo monitoring and is suboptimal for studying metastases to appendicular bones. In contrast, our A549-BM5 subline exhibits a broader and more consistent bone-tropic pattern, including frequent involvement of limb bones, thereby offering a more versatile and observable model for investigating skeletal colonization and therapeutic interventions . However, given the unique biological features of vertebral bone—such as its distinct skeletal stem cell populations and specialized metastatic niche—this model may be more appropriate for investigating spine-specific metastasis rather than broadly representative skeletal dissemination . Collectively, these limitations underscore the urgent need for a lung adenocarcinoma model that not only exhibits stable and reproducible bone tropism but also accurately reflects the general biological characteristics of skeletal metastasis—criteria that A549-BM5 is well positioned to fulfill. Given the clinical characteristics of bone metastasis, identifying biomarkers and therapeutic targets associated with early biological events of skeletal colonization holds significant research value. In this study, we evaluated the performance of the A549-BM5 subline in combination with intra-iliac artery injection to model this process. Remarkably, A549-BM5 exhibited a clear proliferative advantage as early as five days post-injection, suggesting that it effectively recapitulates several key niches involved in the early stages of bone metastasis, including the perivascular niche and the osteogenic niche. Consistent with the notion that organ-specific microenvironments can drive phenotypic divergence even among genetically similar cancer cells , A549-BM5 exhibited not only stable proliferation in vitro but also acquired distinct metastatic traits—including reduced cell adhesion, elevated EMT marker expression, and significantly enhanced migratory and invasive capacities. These features, coupled with its robust and early skeletal colonization in vivo, suggest that repeated exposure to the bone microenvironment exerted strong selective pressure, enriching for a tumor subpopulation with specialized bone-homing potential. Mechanistically, several components of the bone niche may have contributed to this phenotypic evolution. Hypoxic conditions, characteristic of specific niches within the bone marrow microenvironment , are known to activate HIF-1α signaling, which in turn promotes EMT and enhances metastatic capacity . TGF-β, abundantly released during osteolytic bone remodeling, has also been shown to induce mesenchymal transition and support metastatic seeding . Additionally, extracellular matrix proteins such as collagen and osteopontin, along with mechanical cues like increased matrix stiffness, can activate integrin and focal adhesion kinase (FAK) signaling pathways to further promote tumor cell motility and survival . Moreover, accumulating evidence indicates that even subtle environmental cues can induce long-lasting transcriptional and epigenetic reprogramming in tumor cells during in vivo selection . Therefore, the stable acquisition of bone-adaptive traits in A549-BM5 may reflect not only selective outgrowth of pre-existing subclones, but also microenvironment-driven plasticity that enhances tumor cell fitness in the skeletal niche. Importantly, the clinical relevance of the A549-BM5 model was further substantiated by transcriptomic and proteomic analyses, which not only revealed classical molecular hallmarks of bone metastasis but also aligned closely with clinical observations. Previous studies in prostate cancer have shown that once tumor cells colonize the bone marrow microenvironment, they acquire osteomimicry features—such as the expression of alkaline phosphatase, osteocalcin, osteopontin, and bone morphogenetic proteins (BMPs)—to support their survival and progression within bone tissue . Interestingly, A549-BM5 displayed a similar trend, with a transcriptomic profile resembling that of osteogenic cells and robust activation of the WNT signaling pathway—a critical regulator of both osteogenesis and bone metastasis—further reinforcing its bone-adaptive phenotype . In addition, A549-BM5 showed marked enrichment of neuron-associated gene sets, suggesting a possible contribution of neurobiological processes to its metastatic behavior. Although the role of neural regulation in lung cancer bone metastasis has been relatively underexplored, emerging evidence is beginning to highlight its importance. For instance, one study revealed that sensory nerves expressing calcitonin gene–related peptide (CGRP) are enriched at bone metastatic sites and promote tumor progression via the CGRP/CRLR/p38/HSP27 signaling axis . These findings provide compelling support for leveraging A549-BM5 as a preclinical model to explore mechanisms underlying bone metastasis. Taken together, A549-BM5 offers a stable, bone-tropic, and clinically relevant model for lung cancer bone metastasis, characterized by distinct transcriptomic and proteomic alterations compared to the parental A549 line and enhanced interactions with key components of the bone microenvironment. These features make it well-suited for mechanistic studies and therapeutic exploration targeting the skeletal niche. The immune system plays an important role in the process of bone metastasis . Due to the limitations of Lewis lung cancer cells, we chose A549 cells to establish a highly bone-metastatic subline, which limits the possibilities for immune-related studies. Moreover, the in vivo selection strategy based on intracardiac injection bypasses the early steps of the metastasis cascade, such as pre-metastatic alterations in bone , the rise of bone-tropic metastatic seeds in primary tumors , and the perivascular niche and metastasis dormancy . This study did not generate new unique reagents. There is no original code in this study. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Yujian Xu: Writing – original draft, Visualization, Methodology, Investigation, Data curation, Conceptualization. Yahan Qin: Software, Methodology, Investigation. Wenjun Chai: Resources, Methodology, Formal analysis. Ke Xue: Software, Data curation. Xiaoli Liu: Supervision. Jing Li: Supervision. Yue Cao: Validation. Lei Sun: Supervision, Methodology. Hongyu Pan: Writing – review & editing, Supervision, Conceptualization. Mingxia Yan: Writing – review & editing, Supervision, Methodology, Funding acquisition, Conceptualization. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
PMC8809252
Spatial proteogenomics reveals distinct and evolutionarily conserved hepatic macrophage niches
The liver is the largest solid organ in the body, yet it remains incompletely characterized. Here we present a spatial proteogenomic atlas of the healthy and obese human and murine liver combining single-cell CITE-seq, single-nuclei sequencing, spatial transcriptomics, and spatial proteomics. By integrating these multi-omic datasets, we provide validated strategies to reliably discriminate and localize all hepatic cells, including a population of lipid-associated macrophages (LAMs) at the bile ducts. We then align this atlas across seven species, revealing the conserved program of bona fide Kupffer cells and LAMs. We also uncover the respective spatially resolved cellular niches of these macrophages and the microenvironmental circuits driving their unique transcriptomic identities. We demonstrate that LAMs are induced by local lipid exposure, leading to their induction in steatotic regions of the murine and human liver, while Kupffer cell development crucially depends on their cross-talk with hepatic stellate cells via the evolutionarily conserved ALK1-BMP9/10 axis. Keywords: liver, atlas, spatial transcriptomics, CITE-seq, proteogenomic, across species, multi-omic, lipid-associated macrophage, Kupffer cell, NAFLDThe immense advances in single-cell transcriptomics have enabled a better understanding of the cellular composition of different organs across species. However, we still lack information regarding how these cells are organized in their distinct microenvironmental niches. Moreover, the specific cell-cell interactions determining the identity of individual cells within tissues remain to be defined (Guilliams and Scott, 2017; Lindeboom et al., 2021). While the spatial organization of hepatocytes within the liver is understood (Halpern et al., 2017), that of non-parenchymal liver cells remains unclear. This is the case for the mouse liver, but even more so for the human liver, where the identity and the precise localization of most hepatic cells is unknown. Moreover, the link between the transcriptome and the proteome has not been studied, resulting in a lack of reliable surface markers to identify these cells by flow cytometry and confocal microscopy. Here, we used proteogenomic techniques including cellular indexing of transcriptomes and epitomes by sequencing (CITE-seq) and spatial approaches to identify all cells and their specific locations within the healthy and obese livers of mice and humans. By doing so, we have developed strategies for the identification and further study of hepatic cells. Demonstrating the usefulness of this approach, we identify the evolutionary conserved and spatially restricted signals driving the distinct hepatic macrophage phenotypes. To generate a proteogenomic atlas of the liver, we first examined the optimal method for retrieving all hepatic cells. Using the murine liver, we compared single-cell RNA sequencing (scRNA-seq) using cells isolated via ex vivo or in vivo enzymatic digestion with single-nuclei RNA sequencing (snRNA-seq) (Figures S1A–S1C). We did not observe any differences in the number of genes/cell between the two digestion methods (Figures S1D and S1E), but snRNA-seq typically yielded a lower number of genes/cell (Figures S1D and S1F). This did not prevent distinct cell types from being identified in the snRNA-seq dataset as both scRNA-seq and snRNA-seq identified highly expressed genes in each population. However, expression was often higher in the scRNA-seq (Figure S1F). Additionally, we observed a signature of digestion-associated genes and snRNA-seq-associated genes across cell types (Figure S1F). While in terms of genes/cell snRNA-seq is inferior to scRNA-seq, this method best recapitulated the cell frequencies observed in vivo (Figures S1D–S1N). As each method has advantages and disadvantages, the optimal method to use depends on the biological question being addressed. Here, as we sought to generate a proteogenomic atlas of all hepatic cells, we thus opted to use a combination of all protocols. To investigate mRNA and protein expression at single-cell resolution, we used CITE-seq (Stoeckius et al., 2017), staining a selection of the scRNA-seq samples with 107–161 oligo-conjugated antibodies (Figure 1A). Data were pooled together for a single analysis where, with TotalVI (Gayoso et al., 2021), both the protein and mRNA profiles were considered for clustering (Figure 1B). Analysis of the differentially expressed genes and proteins (DEGs/DEPs; Figures S2A and S2B; Table S1) identified 17 cell types (Figure 1B), which were differentially represented with each isolation method (Figure S2C). Addition of antibodies in the CITE-seq analysis enabled surface markers for all cells to be identified, including VSIG4 and FOLR2 for Kupffer cells (KCs) (Figure S2B), without affecting the quality of the transcriptomic data (Figure 2D). This analysis identified one subset of KCs. However, 2 subpopulations termed KC1 and KC2 have recently been described (Blériot et al., 2021; Simone et al., 2021). As our CITE-seq analysis identified that the markers used to identify KC2s, namely CD206 and ESAM, are largely expressed by liver sinusoidal endothelial cells (LSECs) (Figure S2E), we next sought to determine if we had previously removed any potential KC2s in our initial QC steps as LSEC-KC doublets. To examine this, we generated a UMAP of samples containing CD206 and ESAM in the CITE-seq panel and performed a minimal QC filtering on number of genes and percentage of mitochondrial genes. In this UMAP, we found multiple subpopulations including 3 populations of KCs and 2 populations of B cells (Figure S2F). To determine if any of these populations could be KC2s, we harnessed the power of CITE-seq to recreate the gating strategy recently proposed to identify KC2s (Blériot et al., 2021). Converting the CITE-seq data into an FCS file and analyzing this in FlowJo identified 2 of the KC populations to be KC2s expressing CD206 and ESAM, and these cells were present in a similar ratio among KCs as reported by Blériot et al. (Figure S2G). However, employing the same gating strategy with the B cells also enabled a B cell2 population to be identified (Figure S2G). The KC2 and B cell2 populations also expressed other protein markers associated with LSECs including CD26, CD31, and CD38, suggesting that these may be doublets (Figure S2H). Consistent with this, we did not uncover any DEGs specifically expressed in the KC2s or B cell2s, rather these cells had an intermediate profile between either KC1s or B cell1s and LSECs (Figure S2I), as would be expected for doublets. Finally, we perfused the livers with antigen fix to inflate the LSECs to be able to distinguish more readily between KCs and LSECs and performed confocal microscopy. 3D reconstruction of these images indicates that CD206 expression is observed primarily in the LSECs with which the KCs are intertwined. These microscopy images also show that KCs express some CD206, however, consistent with our CITE-seq analysis, this was observed across the KC population rather than in a KC subset further suggesting the existence of only one-KC population (Figure S2J). With this in mind, we decided to continue to apply our initial strict QC controls eliminating these potential doublets from further analyses. Cell types identified in transcriptomic studies depend upon cell/nuclei isolation technique used, related to Figure 1 Cells were isolated from livers of healthy C57B/l6 mice by either ex vivo or in vivo enzymatic digestion. Alternatively, livers were snap frozen and nuclei subsequently isolated following tissue homogenization by a sucrose gradient (3 mice per isolation method). Live cells/intact nuclei were identified and purified using flow cytometry. For the cells, either live CD45, live CD45 or live hepatocytes were sorted. 1 ex vivo digested sample and 1 in vivo digested sample were also stained with a panel of 107 (ex vivo cells) or 161 (in vivo cells) oligo-conjugated antibodies for CITE-seq analysis. FACS-purified cells/nuclei were loaded onto the 10× chromium platform and scRNA-seq, CITE-seq, or snRNA-seq performed. Following clean up and QC, cells from the same mice were pooled together in the same ratios (CD45:CD45:Heps) as found in the tissue as a whole before sorting, different mice were then pooled together and the data were analyzed using scVI. (A–C) UMAPs showing annotations of cell types and proportions of each cell type as a % of total cells in the UMAP isolated using (A) ex vivo digestion; 13,144 cells, (B) in vivo digestion; 24,014 cells and (C) nuclei; 8,583 nuclei. (D) Average number of genes/cell in the annotated mac, B cell, hepatocyte, endothelial, and stromal cell populations following each isolation method. p < 0.05 one-way ANOVA with Bonferroni post-test per cell type. (E) Correlation plot showing genes captured within the mac population when the liver is digested with the in vitro versus the in vivo digestion protocol. (F) Correlation plots showing genes captured within the mac, endothelial cell, and hepatocyte populations when cells are isolated using the in vivo digestion protocol or nuclei are isolated. (G–L) Confocal microscopy images to determine true abundance of (G) stromal cells and cholangiocytes (H) endothelial cells, (I) macs, (J) dendritic cells, (K) B cells, and (L) T cells in vivo. Scale bars, 200 μm. (M) The percentage of each population was calculated based on the percentage of a given population divided by the total number of nuclei. A threshold was applied to the DAPI channel (picture 1) in ImageJ (picture 2) and nuclei were automatically counted based on the ImageJ “analyze particles” plugin (size ). Due to the density of some liver zones, some nuclei were not automatically counted (arrow, picture 3). Those were then manually counted and added to the total number of nuclei. For the populations of interest, cells were counted manually based on specific markers (for example, CD3 for T cells, picture 4). Counting was performed blinded prior to analysis of the sequencing results. (N) Proportion of indicated cell types as a % of total cells identified in confocal microscopy images. Data are from 3–7 images per cell type taken from 2–4 mice. A proteogenomic atlas of the healthy murine liver (A) Hepatic cells were isolated from healthy C57B/l6 mice by ex vivo (5 mice, 15 samples) or in vivo (5 mice, 19 samples) enzymatic digestion. Alternatively, nuclei were isolated by tissue homogenization (4 mice, 12 samples). Live cells/intact nuclei were FACS-purified. For cells, total live, live CD45, live CD45, live hepatocytes, or myeloid cells (live CD45, CD3, CD19, B220, NK1.1) were sorted. 18 samples (7 ex vivo, 11 in vivo) were also stained with a panel of 107–161 barcode-labeled antibodies for CITE-seq analysis. All datasets were pooled together and after QC 185,894 cells/nuclei were clustered using TotalVI. (B) UMAP of sc/snRNA-seq data. (C) Tissue and capsule images from Visium analysis with clusters overlaid. (D) UMAP of zonation of Visium spots (left) and origin of the cells (right). (E) Zonation pattern mapped onto tissue slice. (F and G) Indicated cell signatures from sc/snRNA-seq mapped onto the Visium zonation data. (H) mRNA zonation pattern in Visium highly multiplexed protein analysis and VSIG4-ADT expression pattern (left) and zonated expression patterns of indicated antibodies (right). (I) MICS analysis of indicated proteins. (J) Molecular Cartography of indicated genes and cell types. (K) mRNA (Xcr1, Flt3l, Mafb, and Clec10a) and protein (MHCII and F4/80) expression in the same tissue slice. Scale bars, 50 μm. PV, portal vein; CV, central vein. Arrows indicate specific cell types, colors correspond to cell type/markers. Images are representative of 2–4 mice. See also Figure S2 and Tables S1, S2, S3, and S5. Combination of CITE-seq, scRNA-seq, snRNA-seq, and spatial analyses enables identification of all hepatic cell types including bona fide cell doublets, related to Figure 1 (A and B) Top DEGs (A) and DEPs (B) for cell types from Figure 1B. (C) Distinct profiles of cells or nuclei within the UMAP depending on isolation protocols; 71,162 cells from ex vivo digestions, 96,066 cells from in vivo digestions, and 18,666 nuclei. Numbers on plots represent numbers of cells/nuclei per population. (D) Correlation plots showing genes captured within the KC, B cell and neutrophil populations with and without addition of CITE-seq antibodies. (E) Expression of VSIG4, CD206, and ESAM (protein, top) and Vsig4, Mrc1, and Esam (mRNA, bottom). (F) UMAP showing clusters of cells when only minimal QC for gene number and % mitochondrial genes is performed; 17,669 cells pooled from 3 samples. Expression of Cd5l, Cd19, and Kdr by the clusters facilitating identification of cell types per annotation. (G) CITE-seq data from (F) in Flow-Jo showing expression of CD206 and ESAM in total KCs (left) and total B cells (middle). Numbers represent % of entire KC or B cell population. Identified populations were then mapped back onto the original UMAP (right). (H) Expression of CD31, CD26, and CD38 by indicated populations. (I) Heatmaps showing expression of top DEGs between KC1s and LSECs (left), KC2s and KC1s + LSECs (middle) and B cell2s and B cell1s + LSECs (right). (J) 3D reconstruction of murine liver following perfusion with antigen fix to inflate endothelial cells and staining with antibodies against CD31, CD206, and F4/80. (K) UMAP showing clusters generated from Visium analysis of liver tissue (4 samples) and liver capsule (1 sample). (L) Top unbiased genes defining zonation trajectory from portal to central vein in Visium. (M) Expression of Glul and Epcam by confocal microscopy (left), annotation of portal, periportal, mid, and central regions on same tissue section (middle) and overlay of both datasets (right). (N) Identification of cholangiocyte (left) and cDC (right) signatures on zonated Visium spots. (P) Molecular Cartography showing expression of indicated zonated hepatocyte mRNAs in liver tissue. Data are representative of 2 mice. (O) Expression of Itgae (encoding CD103) in the UMAP of the total liver (left) and flow cytometric analysis of total cDC1s for CD103 and MHCII expression in the healthy murine liver (right). A population of macrophages reside around the bile duct in the healthy murine liver (A) UMAP of murine myeloid cells (71,261 cells/nuclei) isolated from Figure 1B and re-clustered with TotalVI. (B and C) Top DEGs (B) and DEPs (C) between cell types. (D) Expression of Gpnmb and Cd207. (E) Expression of VSIG4 and F4/80 (left) or MHCII, CD11c, and DAPI (right) by confocal microscopy. Capsule macs identified by white arrows. Scale bars, 50 μm. (F) Molecular Cartography of indicated genes at liver capsule. (G) Expression of VSIG4, F4/80, GLUL, and DAPI (left) or F4/80 or CCR2 (right, inset) by confocal microscopy. Scale bars, 100 μm. (H) Molecular Cartography of indicated genes at portal triad. PV, portal vein; CV, central vein; HA, hepatic artery; BD, bile duct. Arrows indicate specific cell types, where color corresponds to markers. Images are representative of 2–4 mice. (I and J) Top GO terms for KCs (I) and bile-duct LAMs (J). (K) Representative image showing expression of VSIG4 (red) CD19 (yellow) and CD3E (magenta) by MICS analysis (left) and % of B or T cells found with/without a KC per field of view (right). Data are pooled from multiple fields of view in 2 mice. p < 0.001 Student’s t test. (L) Mice (29-week-old) were treated with 3.5 mg/kg LPS or PBS and 2 h later, livers were harvested without the capsule. KCs and LAMs were FACS-purified and expression of Il1b, Tnf, IL10, and Il18 was examined by qPCR, compared with b-actin. p < 0.05, p < 0.01, p < 0.001, p < 0.0001, one-way ANOVA with Bonferroni post-test. See also Figure S3 and Table S2. To locate the cells identified we performed spatial transcriptomics analysis using Visium. For this, we cut the liver in two distinct orientations to profile both the liver tissue and the capsule (Figures 1C and S2K). We ordered each Visium spot along a spatial trajectory, and annotated portal, periportal, mid, and central zones based on known hepatocyte zonation markers (Halpern et al., 2017; Figures 1D, 1E, and S2L) and confirmed this annotation using confocal microscopy (Figure S2M). By using the reference sc/snRNA-seq data, we then deconvolved each spot into its constituent cell types and investigated how cell abundance changed with zonation (Figures 1F, 1G, and S2N). Validating this approach, cholangiocytes mapped specifically to the portal zones (Aizarani et al., 2019), while KCs were preferentially located in periportal and mid zones (Bonnardel et al., 2019; Gola et al., 2021). While KC location is zonated, we did not identify a strong zonation pattern in the gene expression profiles of KCs (data not shown). We further identified B cells, T cells, endothelial cells (ECs), and stromal cells (SCs) across all zones, while conventional dendritic cells (cDCs) were found at the portal vein (PV), with a minor presence at the central vein (CV) (Figures 1F, 1G, and S2N). To validate these locations at single-cell resolution, we next sought to identify the best cell-specific surface markers that would also work by confocal microscopy. As the fixation step utilized for confocal microscopy often affects the integrity of protein epitopes, it is not possible to predict which antibodies will work spatially on fixed tissue slices. Therefore, to simultaneously screen multiple antibodies to identify those working by microscopy, we performed a second Visium analysis which we complemented with 100 oligo-conjugated antibodies, chosen based on the CITE-seq results (Figure 1H). The antibodies identified to work spatially were then validated at single-cell resolution, using MACSima™ Imaging Cyclic Staining (MICS) technology and a 60-plex antibody panel (Figure 1I). Unfortunately, we could not identify useful surface markers for all populations. For example, we did not identify enough discriminatory surface markers that worked by confocal microscopy to distinguish the cDC subsets. Notably, while CD103 has previously been proposed to distinguish hepatic cDC1s from cDC2s (Eckert et al., 2015), our data demonstrate that this is only expressed by a fraction of these cells (Figure S2O). To confirm the locations of the cDC subsets, we thus turned to Molecular Cartography™ (Resolve BioSciences) that allows for 100-plex spatial mRNA analysis. Genes were selected based on the DEGs from the sc/snRNA-seq data that were also spatially resolved according to Visium. We also identified the portal-central trajectory in this dataset using cholangiocytes genes (Epcam and Spp1) and known zonated hepatocyte genes (Glul, Cyp2e1, Hal, and Sds; Figure S2P). Using expression of Xcr1, Clec9a (cDC1s) and Cd209a, Mgl2, and Clec10a (cDC2s), we confirmed that both cDC1s and cDC2s were localized primarily at the PV (Figure 1J). As cDC2s shared a number of genes with monocyte-derived cells (Figure S2A), we also examined the expression of general monocyte/mac (Cd14, Adgre1, Axl, Mafb, Cx3cr1, and C5ar1) and KC-specific genes (Cd5l and Vsig4) to further validate their identification as bona fide cDC2s (Figure 1J). The punctate nature of mRNA expression in these analyses combined with the dendritic shape of myeloid cells renders it difficult to convincingly determine cell boundaries and to conclude these cDC2s and macs were distinct cells. To validate this, we therefore developed a protocol that combines mRNA detection (RNAScope) with surface protein detection. Examining expression of cDC- or mac-specific mRNAs combined with protein surface markers confirmed the presence of PV cDC1, cDC2s, and non-KC macs (Figure 1K). Taken together, by combining multiple spatial transcriptomic and proteomic approaches, we located all the cells within the murine liver and identified additional heterogeneity within the myeloid cells, not revealed when examining the sc/sn-RNA-seq dataset in isolation. This highlights the power of combining single-cell and spatial proteogenomic techniques to investigate cellular heterogeneity. To better understand the non-KC macs, we zoomed in on myeloid cells (cDCs, KCs, monocytes, and monocyte-derived cells) in our sc/snRNA-seq analysis defining 11 populations (Figures 2A–2C; Table S2). This included KCs, 3 populations of non-KC macs and cells that had a profile intermediate between monocytes and patrolling monocytes or macs, termed transitioning monocytes. Closer inspection of the non-KC macs identified cluster6 as peritoneal macs (Figure 2B). The DEGs between the remaining populations suggested that cluster7 likely resembles capsule macs (Sierro et al., 2017), expressing Cd207 and Cx3cr1 while cluster8 resembles GpnmbSpp1 lipid-associated macrophages (LAMs) we recently described in the fatty liver (Remmerie et al., 2020; Figures 2A–2D). Conversion of the CITE-seq data into an FCS file allowed an in silico gating strategy to be defined (Figure S3A). Validating this, we utilized the strategy to FACS-purify the populations and assess gene expression (Figures S3B–S3D). Fitting with a recent report (Jin et al., 2021), washing the liver prior to digestion enriched the peritoneal macs in the wash fraction, demonstrating these were contaminants on the liver surface rather than being present in the liver tissue itself (Figure S3E). While the CITE-seq markers did not discriminate between cluster7 and cluster8, adding CD207 to the panel enabled the non-KCs to be divided into CD207 and CD207 macs (Figure S3F). Fitting with their designation as capsule macs, the relative abundance of CD207 macs was increased if we dissected and digested the capsule (Figure S3F). However, although Molecular Cartography confirmed the presence of Cd207 macs in the capsule, it also revealed Cd207 macs at the CV, which were rarely found at the PV (Figures 2E–2H and S3G–S3J). Thus, cluster7 consists of both capsule and CV CD207 macs. This finding further demonstrates the need for spatial approaches to confirm cell identities. Molecular Cartography also identified macs at the PVs and CVs expressing Ccr2 and Chil3 (Figures 2G, 2H, S3H, and S3J), resembling transitioning monocytes (cluster11). Finally, a population of Gpnmb-expressing macs were found to be enriched around the bile ducts (Figures 2G, 2H, S3H, S3K, and S3L). As Gpnmb expression is cluster8 specific (Figure 2B), and these cells resemble LAMs (Remmerie et al., 2020), we termed these cells bile-duct LAMs. Validated flow cytometry gating strategy for murine myeloid cells, related to Figure 2 (A) CITE-seq data from the murine myeloid cells in Figure 2A were exported as an FCS file and an in silico gating strategy identified in FlowJo software. (B) Application of the in silico gating strategy with a 21-color flow cytometry panel. Myeloid cells were pre-gated as live CD45 lineage cells (Ly6GCD19NK1.1B220CD3). Data are representative of 3 experiments with 3–6 mice per experiment. (C) cDC1s, cDC2s, migratory cDCs (Mig. cDCs), peritoneal macs (Peri. Macs), KCs, and non-KC macs (non-KCs) were FACS-purified using gating strategy in (B), mRNA was isolated and qPCR performed to examine expression of indicated genes defining each population to validate their identity. Data are representative of 2 experiments with n = 3–6. (D) Putative peritoneal macs were FACS-purified using gating strategy in (B) and expression of Gata6 was examined by qPCR compared with other hepatic myeloid populations. Data are from a single experiment with n = 6. (E) Peritoneal macs as a % of total macs recovered from the liver using different digestion techniques (in vivo, ex vivo, or capsule) or in supernatants in which livers were washed following removal from the mouse but prior to digestion (wash). Data are from a single experiment with n = 4. p < 0.05, p < 0.01 one-way ANOVA with Bonferroni post-test compared with wash data. (F) Expression of CD14 and CD207 within the non-KC mac population from (B) (left) and % of CD207 and CD207 populations among total macs in livers digested using the ex vivo or in vivo protocols or in dissected and digested liver capsule (right). Data are representative of two experiments with n = 4–5 mice per experiment. p < 0.0001 mixed effects analysis with Tukey’s multiple comparison test. (G) Expression of VSIG4, F4/80, GLUL, and DAPI by confocal microscopy. Insets represent zones featured in Figures 2E, 2G, and S3I. (H) Molecular Cartography of indicated genes and cell types. Insets represent zones featured in Figures 2F, 2H, and S3J. (I) Expression of VSIG4, F4/80, GLUL, and DAPI by confocal microscopy at the central vein. Scale bars, 50 μm. (J) Molecular Cartography of indicated genes and cell types at central vein. (K) Expression of F4/80, EPCAM, CCR2, GPNMB, and DAPI by confocal microscopy at a portal vein (top) or F4/80 or GPNMB alone (bottom). Scale bars, 25 μm. (L) Quantification of % of Gpnmb & Trem2 counts over Adgre1 counts in indicated regions of tissue as assessed using Molecular Cartography data. Each dot represents an individual region. p < 0.05, p > 0.0001 one-way ANOVA with Bonferroni post-test. (M) Expression of DESMIN and F4/80 at the liver capsule and underlying parenchyma (left) or EPCAM, DESMIN and F4/80 at the bile duct by confocal microscopy. PV, portal vein; CV, central vein; HA, hepatic artery; BD, bile duct. Arrows indicate specific cell types, where color corresponds to cell type/markers. All images are representative of 2–6 mice. We next sought to investigate the differences between KCs and LAMs. Analysis of GO terms associated with biological processes for these cells suggested that KCs may play a role in regulating humoral responses, while LAMs were more broadly associated with immune responses (Figures 2I and 2J). Consistent with this, in the 100-plex protein microscopy data we noted that a significant proportion of the B cells present were interacting with KCs, which was not observed with T cells (Figure 2K). This suggests cross-talk between these two populations, potentially linked to the high expression of the B cell chemokine Cxcl13 by murine KCs (Table S2). To assess the inflammatory nature of LAMs compared with KCs, we FACS-purified the cells and performed qPCR analysis to examine expression of various cytokines. To enable LAM purification, we eliminated the capsule prior to digesting the tissue. Fitting with the GO analysis, LAMs expressed more Il1b at steady state compared with KCs (Figure 2L). However, despite this, upon in vivo TLR4 stimulation, they were less responsive than KCs, both in terms of pro- and anti-inflammatory cytokines (Figure 2L), possibly indicative of LPS tolerance. This may result from their location at the PV and hence exposure to blood from the intestine, although this remains to be tested. Taken together this highlights the distinct nature and functions of these cells; however, further research is required to determine the precise roles of these cells. As all the mac populations are in close contact with CD45 cells in their local environment (Bonnardel et al., 2019; Figure S3M), we further analyzed the CD45 cells, identifying multiple subsets of ECs and SCs and a gating strategy to distinguish them (Figures 3A–3C and S4A–S4C; Table S3). ECs could be further subdivided into 4 distinct clusters and analysis of their locations allowed them to be identified as CV ECs (cluster10), LSECs (cluster9), PV ECs (cluster11), and lymphatic ECs (LECs; cluster12) (Figures 3D, 3E, S4D, and S4E). As Visium found fibroblasts at both the PVs and CVs (Figure 3D), and as a previous report has suggested the presence of distinct subsets within these cells (Dobie et al., 2019), we further zoomed in on the SCs to better assess their heterogeneity (Figures 3F, 3G, S4A–S4C, and S4F; Table S4). This revealed subsets of mesothelial cells and fibroblasts restricted to the capsule (Figures 3H and 3I). Myh11 vascular smooth muscle cells (VSMCs) were localized around hepatic arteries, PVs and CVs (Figures 3F–3H, 3J, and S4G) and Mfap4Svep1Clic5Reln fibroblasts were found to be CV fibroblasts (Figures 3G, 3H, 3J, and S4G). Finally, we identified a subset of Clic5Reln fibroblasts (cluster3), which were localized around the cholangiocytes, that we termed bile-duct fibroblasts (Figures 3J and S4G). Taken together, the presence of these spatially distinct subsets of ECs and SCs highlights the uniqueness of the specific microenvironments in which the distinct mac populations reside. Hepatic macrophage populations reside in distinct niches (A) UMAP of murine CD45 cells (83,410 cells/nuclei) isolated from Figure 1B and re-clustered with TotalVI. (B and C) Top DEGs (B) and DEPs (C) between cell types. (D) Indicated cell signatures from sc/snRNA-seq mapped onto the Visium zonation data. (E) Molecular Cartography of indicated genes at central vein (left) and 2 different portal triads (center and right). (F) UMAP of murine stromal cells (5,430 cells/nuclei) isolated from the UMAP in Figure 3A and re-clustered with scVI. (G) Top DEGs between different cell types identified. (H) Identification of mesothelial cell (top) and VSMC (bottom) signatures on zonated Visium data. (I and J) Molecular Cartography of indicated genes at the liver capsule (I) or the central vein (J; left) and portal triad (J; right). PV, portal vein; CV, central vein; HA, hepatic artery; BD, bile duct. Arrows indicate specific cell types, where color corresponds to markers. Images are representative of 2–4 mice. See also Figure S4 and Tables S3 and S4. Protein markers of murine CD45 cell subsets, related to Figure 3 (A) CITE-seq data from the murine CD45 cells in Figure 3A were exported as an FCS file and an in silico gating strategy identified in FlowJo. (B) Gated cell overlay of populations identified using strategy in (A). (C) Expression of CD90, CD204, CD73, and CD29 markers by indicated cell types. (D) Expression of indicated protein markers in 60-plex MICS analysis in endothelial cells. (E) Expression of DESMIN, EPCAM, LYVE1, and CD31 at a portal triad (left) with inset (right). (F) Expression of indicated protein markers in 60-plex MICS analysis in stromal cells. (G) Molecular Cartography of indicated genes and cell types at portal vein. PV, portal vein; CV, central vein. Arrows indicate specific cell types, where color corresponds to cell type/markers. All images are representative of 2–6 mice. Finally, in addition to providing gating strategies for myeloid and CD45 cells, we also wanted to investigate if the CITE-seq data would allow us to develop similar strategies for the lymphoid populations. To examine this, we re-clustered the T cells, NK cells, ILCs, B cells, and pDCs from Figure 1B identifying 12 distinct populations (Figure S5A). Analysis of the DEGs highlighted that while B cells, NK cells, ILC1s, and pDCs were distinct populations, there was considerable overlap between the transcriptomic profiles of the T cells (Figure S5B; Table S5). However, by analyzing the DEPs we were able to define distinct subsets including naive CD4 and CD8 T cells, TRegs, TH1s, CTLs, and TH17s (Figures S5A and S5C; Table S5). Moreover, we were able to design a gating strategy to isolate the distinct populations (Figure S5D). Combination of CITE-seq, scRNA-seq, snRNA-seq, and spatial analyses enables generation of a human liver atlas, related to Figure 4 (A) Murine lymphoid cells (B cells, T cells, NK cells, ILC1s, pDCs; 27,398 cells) were isolated from Figure 1B and re-clustered with TotalVI. (B and C) Top DEGs (B) and DEPs (C) for the cell types from Figure S5A. (D) CITE-seq data from Figure S5A were exported as an FCS file and an in silico gating strategy identified in FlowJo. (E) Human lymphoid cells (B cells, T cells, NK cells, ILC1s, pDCs; 105,790 cells) were isolated from Figure 4B and re-clustered with TotalVI. (F and G) Top DEGs (F) and DEPs (G) for the cell types from Figure S5E. (H) CITE-seq data from Figure S5E were exported as an FCS file and an in silico gating strategy identified in FlowJo. (I) Proportion of indicated cell types arising from patients with <10% (purple) or >10% steatosis (yellow). (J) Hepatic cells were isolated from 22 C57B/l6 mice fed either a standard diet (SD) or a western diet (WD) for 24 or 36 weeks to induce NAFLD and NASH by ex vivo (10 samples) or in vivo (12 samples) enzymatic digestion. Alternatively, livers were snap frozen and nuclei isolated by tissue homogenization (14 samples). Live cells/intact nuclei were purified using FACS. For cells, total live, live CD45, live CD45, live hepatocytes or myeloid cells (live CD45, CD3, CD19, B220, NK1.1) were sorted. 10 samples were also stained with a panel of 107–161 barcode-labeled antibodies for CITE-seq analysis. All datasets were pooled together and after QC 121,980 cells/nuclei were clustered using TotalVI. (K) Murine lymphoid cells (B cells, T cells, NK cells, ILC1s, pDCs; 21,322 cells) from mice fed the SD or WD for 24 or 36 weeks were isolated from Figure S5J and re-clustered with TotalVI. (L) Proportion of indicated cell types arising from mice fed the SD (purple) or WD (yellow). (M) Top DEGs between CTLs isolated from mice fed the SD (purple) or WD (yellow). To determine the degree of conservation between the mac subsets and their different microenvironmental niches between the mouse and the human liver, we next generated a proteogenomic atlas of the human liver using sc/snRNA-seq and CITE-seq on 19 liver biopsies (Figures 4A, 4B, S6A, and S6B; Tables S1 and S5). Of these, most were histologically healthy with only 5 patients showing >10% hepatic steatosis in the absence of any significant fibrosis (Table S6). Cellular proportions varied according to the isolation technique used, and while there was some variability between patients, this was not linked to the surgery (Figures S6C–S6E). Further confirming the lack of fibrosis in the steatotic livers, we did not detect any increase in CTLs (Figures S5E–S5I; Table S5), which have been shown to correlate with non-alcoholic steatohepatitis (NASH) (Haas et al., 2019). A significant increase in CTLs was detected in the setting of murine NASH induced by feeding a western diet (WD) for up to 36 weeks, demonstrating that this is not due a limitation in detecting these differences using CITE-seq (Figures S5J–S5M; Table S7). As Visium reliably located murine hepatic cells, we used this to locate the cells of the human liver in 4 biopsies (Figure 4C). However, the Visium spots from patients with >10% steatosis were found to cluster separately from the healthy samples (<10% steatosis; Figures 4D and 4E). We therefore used the healthy samples to calculate a baseline zonation and then transferred this trajectory onto the steatotic samples (Figures 4F and S6F). This identified the steatosis to be predominantly present in regions expressing peri-central zonation genes like CYP2E1. This zonation pattern was further validated using Molecular Cartography (Figure S6G). This fits with previous clinical studies demonstrating peri-central steatosis to be most common in non-alcoholic fatty liver disease (NAFLD) patients, especially in early disease (Chalasani et al., 2008; Kleiner and Makhlouf, 2016). However, as these peri-central regions are larger than in the healthy controls, it could also imply that the presence of steatosis alters expression of the zonated hepatocyte genes, but this remains to be tested. Notably, the overall cellular distribution was not impacted by the presence of steatosis, although neutrophils and monocytes and monocyte-derived cells were preferentially localized peri-centrally in the steatotic patients correlating with the presence of steatosis (Figure 4G). MICS 100-plex protein analysis further validated the cellular distributions predicted by Visium and confirmed the increase in neutrophils in the steatotic livers (Figures S6H and S6I). Identification of bona fide Kupffer cells across species (A) Cells/nuclei were isolated from liver biopsies (∼1–2 mm; 14 cells, 5 nuclei) from patients undergoing either liver resection, cholecystectomy or gastric bypass. Live cells/intact nuclei were FACS-purified. Either total live, live CD45, and live CD45 or live CD45, CD3, and CD19 cells were sorted. 7 cell samples were stained with a panel of 198 barcode-labeled antibodies for CITE-seq analysis. All datasets were pooled together and after QC, 167,598 cells/nuclei were analyzed using TotalVI. (B) UMAP of sc/snRNA-seq data. (C) UMAP of Visium data from 4 patient biopsy samples. (D) Split of Visium spots based on % steatosis. (E) Healthy and steatotic Visium liver tissue with clusters overlaid and H+E staining to identify steatotic zones. (F) Zonation of Visium data (top) with zonation pattern mapped onto liver tissue (bottom). (G) Indicated cell signatures from sc/snRNA-seq mapped onto Visium zonation trajectory, healthy (top), steatotic (bottom). (H) Myeloid cells (40,821 cells) were isolated from Figure 4B and re-clustered with TotalVI. (I) Expression of VSIG4 protein (top) and CD5L mRNA (bottom). (J) Expression of VSIG4, F4/80, FOLRB, and GLUL combined with Cd5l/CD5L on murine (left) and human (H25; right) livers. Scale bars, 50 μm. Inset in bottom panels. Scale bars, 20 μm. Images are representative of 2–4 livers. (K) Livers (2/species) were isolated from healthy macaque, pig, chicken, hamster, and zebrafish. Cells were isolated by ex vivo digestion for CITE-seq (pig; 198 human antibodies) or scRNA-seq (hamster, chicken, and zebrafish), or nuclei were isolated for snRNA-seq (macaque). Total live cells (hamster, chicken, and pig), DsRedGFP cells (zebrafish) or nuclei (macaque) were FACS-purified. Following QC, 8,483 nuclei (macaque) or 21,907 (pig), 5,965 (hamster), 7,457 (chicken), and 4,957 (zebrafish) cells were analyzed using TotalVI (pig) or scVI (macaque, hamster, chicken, and zebrafish) (top). KCs were identified using the human-murine KC signature and the signature finder algorithm (Pont et al., 2019) (bottom). See also Figures S6 and S7 and Tables S1, S2,S3, S5, S6,S8, and S9. Combination of CITE-seq, scRNA-seq, snRNA-seq, and spatial analyses enables generation of a human liver atlas and identification of bona fide human KCs, related to Figure 4 (A and B) Top DEGs (A) and DEPs (B) for the cell types from Figure 4B. (C) Distinct profiles of cells or nuclei within the UMAP depending on isolation protocol used; 152,535 cells from ex vivo digestions and 15,063 nuclei. (D) Proportion of each cell type per patient profiled. (E) Proportion of indicated cell types as a % of total CD45 cells calculated from ex vivo digested samples per surgery type. Ch; cholecystectomy, Re; resection, GB; gastric bypass. p < 0.05; one-way ANOVA with Bonferroni post-test. (F) Mapping of Visium UMAP zonation patterns onto tissue sections from patient H35 and H37. (G) Expression of indicated zonation genes in patients H35–H38 assessed by Molecular Cartography. (H and I) Expression of indicated proteins by MICS 100-plex protein analysis in the healthy (H) and steatotic (I) human liver. (J) Murine myeloid cells (cDC1s, cDC2s, Mig. cDCs, Macs, monocytes, and monocyte-derived cells; 42,922 cells) from mice fed the SD or WD for 24 or 36 weeks were isolated from Figure S5J and re-clustered with TotalVI. (K) Distribution of cells in UMAP originating from SD- (purple) or WD- (yellow) fed mice. (L) Proportion of indicated cell types arising from mice fed the SD (purple) or WD (yellow). (M and N) Flow cytometry analysis of indicated cell populations in SD and WD-fed mice (24 weeks). Representative gating strategies (M) and absolute number of indicated populations (N). p < 0.05, p < 0.01 Student’s t test. Data are from 2 independent experiments with n = 5–6 per diet. (O and P) Top DEGs (O) and DEPs (P) for cell types from Figure 4H. (Q) Top 25 Murine KC genes as expressed by the human myeloid cell clusters. (R) Mapping of KC signature onto Visium trajectory for healthy (purple) and steatotic (orange) livers. (S) Expression of VSIG4 mRNA within human myeloid cells. (T) Expression of VSIG4 (red) and CD163 (gray, top) or CD169 (gray, bottom) by MICS analysis in healthy human liver. (U) Representative images showing KC location (red) as assessed by MICS analysis in the healthy (left) and steatotic (right) human liver. PV, portal vein; CV, central Vein, dashed line indicates zones of steatosis. (V) Representative image of CD68 and CD163 staining in 10–15-year-old human liver paraffin sections. Image is representative of 6 different patients. (W) In silico gating strategy to isolate distinct myeloid cell populations identified from CITE-seq data. (X) Expression of VSIG4 and FOLR2 by live CD45 cells also expressing CD14 in indicated human liver biopsies by flow cytometry. Data are representative of 21 biopsy samples analyzed. To date, no validated markers of bona fide human KCs have been described. Explaining the difficulty to accurately define human KCs, we found monocytes and macs formed a single continuum in the human sc/snRNA-seq data, preventing a simple definition of human KCs (Figure 4B). Notably, a similar continuum from monocyte to KCs was also observed in the NASH murine liver (Figures S5J and S6J–S6N, Table S8). Consistent with our previous report (Remmerie et al., 2020), we observed both long-term resident Timd4-expressing KCs and recently recruited Timd4 monocyte-derived KCs (moKCs; Figures S6J–S6N). As the presence of such a continuum in the human liver suggests that there may also be monocyte contribution to the KC pool in the healthy human liver, we next zoomed in on myeloid cells to examine this, identifying 10 clusters (Figures 4H, S6O, and S6P; Table S2). To define the KCs, we examined expression of the top 25 murine KC genes by these clusters, which identified cluster10 to be the genuine human KCs (Figure S6Q). Unlike in mice, these were preferentially located in the mid zone (Figure S6R). Cluster9 also expressed many of these genes but lacked TIMD4 (Figure S6Q), suggesting that these cells may be recently recruited moKCs. The presence of moKCs in the liver is consistent with reports that host-derived macs are identified in transplanted donor livers (Bittmann et al., 2003; Pallett et al., 2020) and suggests that the KC population may be a mix of embryonic and monocyte-derived cells. Although not the case at mRNA level, VSIG4 was found to be the best human KC protein marker in the CITE-seq data, while FOLR2, CD163, and CD169 were also identified as useful markers of these cells for flow cytometry and confocal microscopy on frozen and paraffin sections (Figures 4I and S6S–S6X). Co-staining human livers for VSIG4 protein and KC-specific CD5L mRNA and MICS 100-plex protein analysis also confirmed the mid-zonal localization of KCs (Figures 4J and S6U). To assess if KC identity was further conserved in evolution, we profiled macaque, pig, hamster, chicken, and zebrafish livers (Figure 4K). We identified the KCs in an unbiased manner by mapping the conserved human-mouse KC signature onto the datasets (Figures 4K and S7A–S7C). We then examined the main features of each KC population identified (Figures S7D–S7H; Table S9). A strong overlap in transcriptomes across species was observed likely due to the conserved expression of core KC transcription factors (Figure S7I). However, each species also harbored a number of unique KC genes (Figure S7J; Table S9). VSIG4 protein expression was also conserved in pig and macaque KCs (Figures S7K–S7M). Similarly, we were also able to identify most of the other hepatic cells across species on the basis of conserved genes (Figures S7N and S7O). cDC2s were the main exception to this, as specific cDC2 marker genes were not conserved across all species (Figure S7O). Conserved and unique features of KCs across species, related to Figure 4 (A and B) Expression of human-murine KC signature genes across cell types in mouse (A) and human (B). (C) Unbiased identification of KCs in mouse and human using the human-murine KC signature and the signature finder algorithm (Pont et al., 2019). (D–H) Annotated UMAPs from indicated species and expression of top KC-specific genes compared with other cells per species. (I) Expression of previously identified core murine transcription factors (Bonnardel et al., 2019) by KCs across species. (J) Venn diagram showing convergence and divergence of expression of top 50 KC genes per species across species, see Table S9 for genes lists per species. (K) Top DEPs (identified with cross reactive human antibodies) in the pig CITE-seq data. (L) Expression of VSIG4 in the porcine liver by confocal microscopy. (M) Expression of VSIG4, CD68 (protein), and CD5L (mRNA) in macaque liver. PV, portal vein; HA, hepatic artery; BD, bile duct. All images are representative of 2 livers. (N and O) Conserved expression of indicated genes across CD45 (N) and CD45 (O) cell types and species. Alongside KCs, we also identified distinct clusters of macs in the human myeloid cells (Figure 4H). To better understand the nature of these clusters we performed confocal microscopy to examine the specific locations of CD68VSIG4 macs in the human liver (Figure S8A). This identified CD68VSIG4 macs in the liver capsule, in close proximity to central and PVs as well as at bile ducts (BDs) (Figures 5A–5C and S8A). Similar populations were also observed at the PVs and CVs and at the BDs in the healthy macaque liver (Figure S7M). Examination of the scRNA-seq data and comparison with murine signatures identified immature and mature LAMs, with immature LAMs expressing some monocyte genes (Figures 4H, 5D, S8B, and S8C). Although recently suggested to be specific to fibrotic human livers (Ramachandran et al., 2019), we identified LAMs in all patients profiled with scRNA-seq, but there was a trend toward increased proportions of LAMs in the livers with >10% steatosis (Figure S8D) consistent with the increased population of LAMs in murine NAFLD (Figures S6J–S6N). As in the healthy mouse, Visium identified human LAMs in portal zones of non-steatotic livers. However, in steatotic human livers, LAMs were primarily located peri-centrally, in zones with steatosis (Figure 5E), suggesting that monocytes are recruited to distinct locations in the healthy and obese liver where they then differentiate into LAMs. This altered location of LAMs was further validated by confocal microscopy and Molecular Cartography (Figures 5F–5H). However, this analysis did not identify any capsule macs, suggesting that these cells may be absent from our UMAP, likely as a result of the small amount of capsule tissue on a biopsy. The Mac1 population expressing IGSF21 was present in very low numbers throughput the tissue (Figures 5G and 5H). This coupled with their similar transcriptomic profile to moKCs could suggest that these are moKC precursors as observed in the mouse, but this requires further study. Focusing on the LAMs, the change in their location in the steatotic human liver was also observed in the murine NAFLD model. Here, LAMs were found across portal, periportal, and mid zones (Figures 5I, S8E, and S8F), fitting with the presence of steatosis in these regions and consistent with our previous report (Remmerie et al., 2020). Comparison of DEGs between LAMs in standard diet (SD) and WD-fed mice identified that LAMs had a more mature phenotype in WD-fed mice, downregulating their expression of some monocyte genes and increasing their expression of prototypical mac markers, consistent with the presence of both immature and mature LAMs in the human liver (Figure 5J). Fitting with a more mature phenotype, WD-derived LAMs also expressed lower levels of Il1b, Tnf, and Il10 compared with SD LAMs (Figure 5K). While further studies are required to assess the precise functions of these cells in NAFLD, this could further suggest a protective rather than a pathogenic role for LAMs (Daemen et al., 2021). Evolutionarily conserved signals regulate LAM and KC development, related to Figures 5 and 7 (A) Confocal microscopy of healthy human liver showing expression of indicated markers. Scale bars, 200 μm. (B and C) Expression of conserved human-murine bile-duct LAM signature in human (B) and mouse (C) hepatic myeloid cells. (D) Proportion of indicated myeloid cell populations as a % of total myeloid cells in human liver biopsies profiled by scRNA-seq when divided based on presence of steatosis. (E) Mice were fed a western diet (WD) or standard diet (SD) for 36 weeks to induce NAFLD and Visium analysis was performed. Analysis is pooled from 1 liver slice from the SD condition and 3 liver slices from the WD condition. Shown are cluster and sample annotations. (F) Zonation of all cell types from Figure S5J in murine NAFLD map (SD&WD). (G) Differential NicheNet highlighting prioritized conserved (human-mouse) ligand-receptor (LR) pairs between indicated macs and their niche cells. LR pairs are grouped according to the niche cell type with highest ligand expression. (H) Expression of ALK1 (ACVRL1), BMP9 (GDF2), and BMP10 in human, mouse, and macaque livers where both KCs and stellate cells were profiled. (I) Livers were harvested from Clec4f-CrexAcvrl1 mice or Acvrl1 controls and KCs examined (top) and quantified (middle) using VSIG4 expression. Expression of indicated KC markers by mac populations in Clec4f-CrexAcvrl1 or Acvrl1 control mice (bottom). Data are pooled from 2 independent experiments with n = 14 per group. Student’s t test. p < 0.0001. (J) Expression of CD31 (ECs), DESMIN (stromal cells), F4/80 (Macs), and EPCAM (cholangiocytes) by confocal microscopy in Fcgr1-CrexAcvrl1 mice and Acvrl1 controls. PV, portal vein; CV, central vein. Images are representative of 2 mice per group. LAMs are found at the bile duct in the healthy liver, but at zones of steatosis in the obese liver (A) Expression of indicated markers at the capsule of the healthy human liver (H14) by confocal microscopy. Scale bars, 20 μm. Arrowheads indicate capsule macs. (B and C) Representative images showing expression of indicated markers at the portal triad by confocal microscopy. Insets on right. Scale bars, 100 μm. Insets scale bars, 50 μm (B). Scale bars, 150 μm. Insets scale bars, 75 μm (C). (D) Human bile-duct LAMs identified using the murine bile-duct LAM gene signature and the signature finder algorithm (Pont et al., 2019). (E) Human LAM signature from scRNA-seq mapped onto the Visium zonation data, healthy (purple), steatotic (yellow). (F) Expression of indicated markers by confocal microscopy. Insets on right. Scale bars, 150 μm. Insets scale bars, 75 μm. (G and H) Expression of indicated genes in healthy (G) and steatotic (H) human liver. Insets on right. Images are representative of 2 patients per condition. (I) Mice were fed a western diet (WD) or standard diet (SD) for 36 weeks to induce NAFLD and Visium analysis was performed. Analysis is pooled from 1 liver slice from the SD condition and 3 liver slices from the WD condition. Zonation pattern and H&E staining (left) and LAM and KC location (right). (J) Heatmap showing DEGs between LAMs from SD (purple) and WD (yellow) fed mice (24 + 36 weeks pooled). (K) LAMs were FACS-purified from the liver of mice fed the SD or WD for 24 weeks (with removal of capsule prior to digestion), RNA was isolated and expression of indicated genes was assessed by qPCR relative to β-actin. p < 0.05, p < 0.0001, Student’s t test. Data are pooled from 2 independent experiments with n = 7–10 mice per group. Images from (A–C and F) are representative of 4–5 patients per condition. PV, portal vein; CV, central vein. See also Figure S8 and Table S2. Given the altered localization of KCs in the healthy murine versus human liver and LAMs in the healthy versus steatotic liver of mice and humans, we next sought to investigate how the cells of the mac niches differed in these settings. Analysis of CD45 cells in the human liver identified similar (sub)populations of ECs, SCs, and hepatocytes as observed in the healthy mouse liver (Figures 6A and 6B; Table S3). However, while the 100-plex MICS analysis detected a second subset of portal fibroblasts expressing DESMIN (Figure S6H), we did not detect these cells in our UMAP, likely due to difficulties detecting DESMIN with snRNA-seq. No significant differences were found in terms of localization of the identified CD45 cells that could explain the altered location of KCs compared with the mouse (Figures 6C and 6D). With this in mind, we next utilized NicheNet to examine potential ligand-receptor pairs between KCs and CD45 cells present only in the human that could regulate KC location. This analysis identified CCL23 and CCL14 expression by LSECs and CCL16 by hepatocytes that binds to CCR1 expressed by human KCs but not murine KCs (Figure 6E). Crucially, these ligands were preferentially located peri-centrally (Figure 6F) and thus represent interesting targets for further study regarding the potential regulation of KC location in the human liver. Macrophage niche cells may regulate macrophage locations in healthy versus steatotic liver (A) UMAP of human CD45 cells (15,481 cells/nuclei) isolated from Figure 4B and re-clustered with scVI. (B) Top DEGs between cell types identified. (C) Indicated cell signatures from sc/snRNA-seq mapped onto the Visium zonation data, healthy (top), steatotic (bottom). (D) Molecular Cartography localizing distinct CD45 cells. (E) Circos plot showing NicheNet predicted ligand-receptor pairs between KCs and LSECs, HSCs and hepatocytes uniquely expressed in the human liver (left) and UMAPs showing normalized expression of indicated chemokines in the human liver (right) and CCR1/Ccr1 in the human and murine NAFLD liver (bottom). (F) Zonation of indicated genes in the healthy and steatotic human liver. (G) UMAP of murine NAFLD (SD and WD) stromal cells (4,025 cells/nuclei) isolated from Figure S5J and re-clustered with scVI (left) and proportions of each cell type in SD- and WD-fed mice (right). (H) Top DEGs between cell types. (I) Mice were fed a western diet (WD) or standard diet (SD) for 36 weeks to induce NAFLD, and Visium analysis was performed. Analysis is pooled from 1 liver slice from the SD condition and 3 liver slices from the WD condition. Zonation of Ccl2 fibroblasts and fibroblasts in SD- and WD-fed mice. See also Tables S3 and S10. We next aimed to investigate the role of CD45 cells in the regulation of LAM location in the healthy versus steatotic liver. Given the low number of steatotic human samples in our sc/snRNA-seq analysis and the variation between patients, we turned to the murine NASH model to investigate this. Zooming in on SCs (Figure 6G; Table S10), we identified that there was an increase in fibroblasts in the WD-fed mice. There was also a considerable overlap between fibroblasts and HSCs in terms of their gene expression profiles suggesting there may be a differentiation trajectory between the HSCs and the fibroblasts, although this remains to be validated in vivo. We also noted a sub-population of fibroblasts expressing Ccl2, a known ligand for CCR2 which is expressed on monocytes recruited to the liver during NAFLD (Remmerie et al., 2020), as well as Cd44 and Vcam1, two genes involved in monocyte recruitment and adhesion (Johnson and Ruffell, 2009; Meerschaert and Furie, 1995). Both fibroblast populations were enriched in the periportal steatotic regions (Figure 6I), in which we also observe enrichment of the LAMs (Figure 5I). We also found high CCL2 and Cd44 expression in human fibroblasts (Table S3), potentially recruiting bile-duct LAMs. Together, this suggests that fibroblasts may play an important role in recruiting LAMs. Having identified a potential role for the mac niche cells in regulating mac subset location, we next sought to determine the involvement of these cells in regulating mac phenotypes. To assess the roles of conserved cell-cell interactions in driving mac heterogeneity across species, we performed a differential NicheNet (Browaeys et al., 2019) analysis between the distinct hepatic macs and the CD45 cells present in their respective niches focusing on ligands and receptors conserved in both human and mouse. This revealed very few specific ligand-receptor pairs for LAMs (Figure S8G; data not shown), hinting that local factors such as metabolites rather than unique cell-cell interactions may drive the LAM phenotype. Indeed, this would be consistent with the presence of LAM-like cells in multiple tissues including the obese adipose tissue, the brain, the lung, and the heart (Jaitin et al., 2019; Keren-Shaul et al., 2017; Liao et al., 2020; Rizzo et al., 2020). In line with this, BM monocytes cultured with acetylated low-density lipoprotein expressed LAM-associated genes (Figure 7A), demonstrating a dominant role for lipids in inducing the LAM phenotype. Conversely, for KCs, we found multiple ligand-receptor pairs to be conserved between human and mouse (Figures 7B and S8G). One of these, an activin receptor-like kinase (ALK1)-bone morphogenic protein (BMP)9/10 circuit between KCs (ALK1; encoded by Acvrl1) and stellate cells (BMP)9/10 encoded by Gdf2/Bmp10 respectively) was found to be conserved in all 7 species and was predicted to control the expression of a number of the conserved KC genes (Figures 7B, 7C, and S8H). To validate a role for this axis in KCs, we generated Fcgr1-Cre × Acvrl1 mice, eliminating ALK1 specifically from CD64-expressing macs (Scott et al., 2018). This led to an almost complete loss of VSIG4 KCs (Figures 8D–8F), demonstrating that evolutionarily conserved ALK1 signaling is crucial for KCs. To determine if ALK1 is required for KC maintenance we generated Clec4f-Cre × Acvrl1 mice, eliminating ALK1 only from differentiated KCs (Scott et al., 2018). This revealed a relatively similar phenotype than observed in the Fcgr1-Cre mice, suggesting that ALK1 is also required for KC maintenance (Figure S8I). To directly test the need for ALK1 in KC development, we generated BM chimeras whereby CD45.1 Clec4f-Dtr mice were irradiated with their livers shielded to avoid any radio damage. These mice were then reconstituted with either CD45.2 Acvrl1 or Fcgr1-CrexAcvrl1 BM. 4 weeks later, mice were given a single i.p. injection of DT to deplete the KCs and 7 or 13 days thereafter chimerism within the KC population was determined (Figure 7G). We observed that, already by day 7, KO BM-derived cells were almost completely outcompeted by WT BM-derived cells within the KC pool. This was not observed in monocytes but a similar pattern was observed in VSIG4 macs demonstrating that ALK1 is critically required for early KC development (Figure 7H). Finally, while NicheNet predicts that BMP9/10 from stellate cells would signal through ALK1 to induce KC development and maintenance, a prediction in line with our previous NicheNet analysis (Bonnardel et al., 2019), another recent study has suggested that transforming growth factor (TGF)β signaling would be important for KCs (Sakai et al., 2019). This prediction was made on the basis of SMAD4 signaling, which is a common downstream effector of both TGF-β Receptor and ALK1-induced signaling. As TGF-β was also found to participate in an ALK1-TGF-β receptor containing signaling complex (Goumans et al., 2003), we thus examined whether TGF-β signaling is important for KCs. To this end we utilized an ALK1-Fc trap or TGF-β type II receptor (TGFβRII)-Fc trap alongside appropriate isotype controls. ALK1-Fc selectively sequesters BMP9/10 (Desroches-Castan et al., 2021), while TGFβRII-Fc selectively interferes with TGF-β1/TGF-β3 receptor binding (Komesli et al., 2017). Clec4f-Dtr mice were depleted of KCs and simultaneously treated with either receptor-Fc traps or isotype controls and KC development was examined 7 days later (Figure 7I). In line with the chimera study, treatment with ALK1-Fc significantly abrogated KC development; however, blocking TGFβ signaling had only a minor effect on the proportion of VSIG4 macs (Figure 7J). Altogether this validates the NicheNet prediction and demonstrates that the evolutionarily conserved ALK1-BMP9/10 axis is crucial for the development and maintenance of KCs. ALK1-BMP9/10 axis regulates KC development (A) Mouse BM monocytes were cultured in the presence of CSF1 and indicated concentrations of human ac-LDL, prior to analyzed for expression of indicated genes by qPCR. Data are pooled from 2 experiments. One-way ANOVA with Bonferroni post-test compared with 0 ng/mL. (B) NicheNet circos plot highlighting conserved ligand-receptor pairs and induced target genes between KCs and indicated niche cells in human and mouse. (C) Feature plots showing expression of ALK1 (Acvrl1) in human myeloid cells (left) and GDF2/BMP10 in CD45 cells (right). (D) Livers were harvested from Fcgr1-CrexAcvrl1 mice or Acvrl1 or Acvrl1 controls and KCs examined (left) and quantified (right) using VSIG4 expression. (E) Expression of indicated KC markers by mac populations in Fcgr1-CrexAcvrl1 or Acvrl control mice. Data are pooled from 3 independent experiments with n = 9 per group. Student’s t test. (F) Expression of indicated markers in livers of Fcgr1-CrexAcvrl1 or Acvrl1 or Acvrl1 control mice by confocal microscopy. Scale bars, 50 μm. Images are representative of 2 mice per group. (G) Schematic of chimera experiment setup. (H) % chimerism normalized to levels in blood Ly6C monocytes in Clec4f-Dtr mice 7 or 13 days after DT administration following partial irradiation and receiving either Acvrl or Fcgr1-CrexAcvrl BM 4 weeks earlier. Data are pooled from 2 independent experiments with n = 10–12 mice per group. (I) Schematic of Fc trap experiment setup. (J) Representative FACS plots showing VSIG4 and CLEC2 expression by total macs. Numbers represent % of total mac population in the indicated gate (left) and % of VSIG4 and VSIG4 macs among total CD45 cells in the different treatment conditions. One-way ANOVA with Bonferroni post-test. p < 0.05, p < 0.01, p < 0.001, p < 0.0001. See also Figure S8. To generate a practical cellular atlas of any human tissue and unravel the cell-cell circuits essential for the identities of cells inhabiting that tissue, four key pieces of information are required: (1) an inventory of all cells present, (2) the location of the different cells within the tissue to identify interactions between neighboring cells, (3) an alignment between the human and animal models allowing for any predicted cell-cell interactions to be perturbed, and (4) the identification of reliable antibody-based panels for the efficient screening of different patients and/or transgenic animals. Here, by integrating single-cell and spatial transcriptomic and proteomic data, we provide these 4 pieces of information for the liver and uncover evolutionarily conserved microenvironmental circuits controlling the development of hepatic macs. Unraveling the spatial localization of all hepatic cells, we identify LAMs around the bile ducts in the healthy mouse, human, and macaque liver. However, when steatosis is present, LAMs are preferentially recruited to the steatotic regions of the liver. This spatial information at least partially invalidates the hypothesis that LAM identity is specifically induced by fibrotic SCs (Ramachandran et al., 2019). Rather, our data suggest that LAMs are induced by local lipid exposure. We also provide an alignment of the liver atlas across seven species. This reveals the conserved and unique transcriptomic programs of steady-state KCs and uncovers the spatially restricted and conserved ligand-receptor pairs between KCs and the cells constituting their niche. Underlining the need to first characterize the healthy tissue before attempting to understand how disease perturbs the cells, we identify the DLL-NOTCH interaction to be an evolutionarily conserved cross-talk between homeostatic LSECs and KCs and therefore not unique to hepatocellular carcinoma or fibrosis, as proposed (Ramachandran et al., 2019; Sharma et al., 2020). Similarly, we find that FOLR2 expression is not specific to tumor-associated hepatic macs (Sharma et al., 2020) but is expressed by KCs in the healthy mouse and human liver. Finally, we apply a proteogenomic pipeline starting from broad oligo-conjugated antibody panels for both single-cell and spatial profiling. This is crucial as transcriptomic profiling does not always correspond with the ability to detect proteins by flow cytometry or microscopy. By screening broadly, we identify the best surface markers for the isolation and localization of hepatic macs and their respective niche cells. This allows both the validation of the spatial location at the single-cell level, and the efficient screening of transgenic mouse models for the loss of KCs. Characterization of Fcgr1-CrexAcvrl1 mice using our defined panel readily demonstrates the cruciality of the ALK1-BMP9/10 axis in KC development emphasizing that mac-stromal-cell cross-talk goes much further than the exchange of growth factors (Guilliams et al., 2020; Zhou et al., 2018). Moving forward, applying these relatively cheap antibody panels to large patient cohorts or multiple transgenic mouse models should enable any perturbations disturbing liver homeostasis to be efficiently identified. The current study has two main limitations. First, the analysis of the human liver remains restricted by the number of human patients included (19 patients for the sc/snRNA-seq, 4 patients for Visium, and 15 patients by microscopy). While this study provides markers to cheaply and efficiently screen large patient cohorts allowing this analysis to be extended, given the heterogeneity between patients, multiple studies will need to be integrated in a single analysis if we are to be able to interrogate transcriptomic differences in a particular cell subset related to age, sex, ethnicity, or pathological parameters. Second, this study highlights that more research will be needed to fully characterize human SCs. As we could only retrieve these cells through snRNA-seq this means we have not been able to identify good surface markers through CITE-seq. Additionally, for small populations of fibroblasts we did not recover enough nuclei to truly probe their heterogeneity. Better isolation protocols are required to retrieve and enrich these cells from the human liver in order to run a broad panel of CITE-seq markers to identify the different subsets and their corresponding surface markers. Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Martin Guilliams (martin.guilliams@irc.vib-ugent.be). This study did not generate new unique reagents. WT C57Bl/6J mice (Janvier) were used for this study. Male and Female mice were used for all experiments, with the exception of NAFLD experiments were only male mice were used. For NAFLD experiments, mice were put on the SD or WD at 5 weeks of age and sacrificed after either 24 or 36 weeks on the diet as indicated. Fcgr1-Cre mice (Scott et al., 2018) were obtained from Prof. Bernard Malissen, CIML, Marseille and Clec4f-Cre mice (Scott et al., 2018) were crossed with Acvrl1 mice (Park et al., 2008) obtained from Paul Oh, Barrow Neurological Institute, Florida, USA. Clec4f-Dtr mice (Scott et al., 2016) were crossed to CD45.1 mice (Janvier) for KC depletion and development experiments All mice were used between 6 and 12 weeks of age unless otherwise stated. All mice maintained at the VIB (Ghent University) under specific pathogen free conditions. All animals were randomly allocated to experimental groups. All experiments were performed in accordance with the ethical committee of the Faculty of Science, UGent and VIB. Piglets (female, 10 weeks old) were purchased at a local farm and transported to the animal facilities of the Faculty of Veterinary Medicine. The animals were housed in isolation units as blood donors and had access to feed and water ad libitum. At 30 weeks of age, the animals were euthanized by intravenous injection of sodium pentobarbital 20% (60mg/2.5kg) and livers were collected. The animal study was reviewed and approved by the Ethical Committee of the Faculty of Veterinary Medicine (EC2018/55). Study animals were clinically healthy Leghorn hens of approximately 58 weeks old collected from a commercial farm. The hens were housed at the Faculty of Veterinary Medicine according to acceptable welfare standards and were observed at least twice daily for health problems. Feed and water was offered ad libitum. The chickens were euthanized through intravenous injection (in the wing) with sodium pentobarbital. The EC approval number of this trial was EC2019/015. Male Cynomolgus macaques (≥2 years) were sourced from China and supplied by Guangzhou Xiangguan Biotech Co., Ltd and confirmed healthy before being assigned to the study. Animal handling, husbandry and euthanasia was performed by WuxiAppTec Co., Ltd., China according to local ethical guidelines (AAALAC accredited 2010). Study animals consists of 4 groups, orally dosed once with a Janssen proprietary immune modulator or vehicle control. Vehicle control animals only were used in this study. Liver tissue samples were snap-frozen immediately after euthanasia. Samples were thawed once before shipping to Ghent for snRNA-seq analysis. Female syrian hamsters (Janvier) were housed per one or two in ventilated isolator cages at a temperature of 21°C, humidity of 55% and 12:12 dark/light cycles, with access to food and water ad libitum and cage enrichment. All hamsters had SPF status at arrival and manipulations were performed in a laminar flow cabinet. Housing conditions and experimental procedures were approved by the ethical committee of KU Leuven (license P015-2020). Animals were euthanized at 6-8 weeks of age by intraperitoneal injection of 200 mg/mL sodium pentobarbital and livers were collected for analysis. Zebrafish were maintained under standard conditions, according to FELASA guidelines (Alestrom et al., 2019). All experimental procedures were approved by the ethical committee for animal welfare (CEBEA) from the ULB (Université Libre de Bruxelles) (Protocol #594N). The following transgenic lines at 6 months of age were used: Tg(mpeg1:EGFP) (Ellett et al., 2011); Tg(kdrl:Cre) (Bertrand et al., 2010); Tg(actb2:loxP-STOP-loxP-DsRed) (Bertrand et al., 2010) enabling macs to be sorted for sequencing as DsRed, GFP double positive cells. Patient studies were run in collaboration with Ghent University Hospital. Liver biopsies (1–2mm) were isolated with informed consent from patients undergoing cholecystectomy or gastric bypass. In addition, liver biopsies were isolated from healthy adjacent tissue removed during liver resection due to colorectal cancer metastasis. In most cases, a second biopsy was also taken to evaluate liver histology. A full overview of all patient samples used in this study can be found in Table S6. Paraffin-embedded human liver samples were obtained through collaboration with Dr. Jan Lerut (Université Catholique de Louvain, UCL). All studies were performed in accordance with the ethical committee of the Ghent University Hospital (study numbers: 2015/1334 and 2017/0539). Liver cells were isolated by either ex vivo digestion (all species, except zebrafish) or in vivo liver perfusion (mice only) and digestion as described previously (Bonnardel et al., 2019; Scott et al., 2016). Briefly, for ex vivo digestion, livers were isolated, cut into small pieces and incubated with 1mg/ml Collagenase A and 10U/ml DNAse at 37C for 20 mins with shaking. For in vivo digestion, after retrograde cannulation, livers were perfused for 1-2mins with an EGTA-containing solution, followed by a 5min (6ml/min) perfusion with 0.2mg/ml collagenase A. Livers were then removed, minced and incubated for 20mins with 0.4mg/ml collagenase A and 10U/ml DNase at 37°C. All subsequent procedures were performed at 4°C. Samples were filtered over a 100μm mesh filter and red blood cells were lysed. Samples were again filtered over a 40μm mesh filter. At this point in vivo digestion samples only were subjected to two centrifugation steps of 1 min at 50g to isolate hepatocytes. Remaining liver cells (leukocytes, LSECs and HSCs; in vivo protocol) and total cells from the ex vivo digests were centrifuged at 400g for 5mins before proceeding to antibody staining for flow cytometry. A combination of Collagenase A and DNase were used to digest livers in both protocols to minimize cleavage of surface epitopes. Dissected livers from 6 months old transgenic zebrafish were triturated and treated with Liberase TM at 33°C for 20 min. Cells were then filtered through 40μm nylon mesh and washed with 2% FBS in PBS by centrifugation. Sytox Red was then added to the samples at a final concentration of 5nM to exclude nonviable cells before proceeding to flow cytometry. DsRedGFP cells were then FACS-purified. Nuclei were isolated from snap frozen liver tissue with a sucrose gradient as previously described (Habib et al., 2016). Briefly, frozen liver tissue is homogenized using Kimble Douncer grinder set in 1ml homogenization buffer with RNAse inhibitors. Homogenised tissue is then is then subjected to density gradient (29% cushion – Optiprep) ultracentrifugation (7700rpm, 4C, 30 mins). After resuspension, nuclei are stained with DAPI and intact nuclei were FACS-purified from remaining debris. To induce NAFLD and NASH, mice were fed a western diet (WD) high in fat, sugar and cholesterol for 24 or 36 weeks as described previously (Remmerie et al., 2020). This consisted of 58% fat, 1% cholesterol (Research Diets; D09061703i) and drinking water was supplemented with 23.1g/L fructose (MPBio) and 18.9g/L sucrose (VWR). Control mice were fed a standard diet with 11kcal% fat with corn starch (D12328i; Research Diets). Cells were pre-incubated with 2.4G2 antibody (Bioceros) to block Fc receptors and stained with appropriate antibodies at 4°C in the dark for 30-45 minutes. Cell viability was assessed using Fixable Viability dyes (eFluor780 or eFluor506; Thermo Fischer) and cell suspensions were analyzed with a BD FACSymphony or purified using a BD Symphony S6, BD FACSAria II or III. Nuclei were sorted on basis of DAPI positivity and size. Analysis was performed with FlowJo software (BD). Intracellular staining for CD207 was performed by fixing and permeabilizing extracellularly stained cells according to the manufacturer’s instructions using the FoxP3 Fixation/Permeabilization Kit (Thermo Fischer). Confocal staining was performed as described previously (Bonnardel et al., 2019). Immediately after sacrificing mice with CO2, inferior vena cava were cannulated and livers were perfused (4 mL/min) with Antigenfix (Diapath) for 5 min at room temperature. After excision, 2-3 mm slices of livers were fixed further by immersion in Antigenfix for 1h at 4°C, washed in PBS, infused overnight in 34% sucrose and frozen in Tissue-Tek OCT compound (Sakura Finetek). 20 μm-thick slices cut on a cryostat (Microm HM 560, Thermo Scientific) were rehydrated in PBS for 5 min, permeabilized with 0,5% saponin and non-specific binding sites were blocked for 30 min with 2% bovine serum albumin, 1% fetal calf serum and 1% donkey or goat serum for 30 minutes. Tissue sections were labeled overnight at 4°C with primary antibodies followed by incubation for 1h at room temperature with secondary antibodies. When two rat antibodies were used on the same section, the directly conjugated rat antibody was incubated for 1h after staining with the unconjugated and anti-rat secondary and after an additional blocking step with 1% rat serum for 30 minutes. Slides were mounted in ProLong Diamond, imaged with a Zeiss LSM780 confocal microscope (Carl Zeiss, Oberkochen, Germany) with spectral detector and using spectral unmixing and analyzed using ImageJ and QuPath software. Experiments were performed using the RNAScope Multiplex Fluorescent V2 Assay kit (ACDBio 323100). Probes targeting intronic regions for Hs-Cd5l (ACDBio 850511), Mfa-Cd5l (ACDBio 873211), Mm-Cd5l (ACDBio 573271), Mm-Flt3 (ACDBio 487861), Mm-Xcr1 (ACDBio 562371), Mm-Mafb (ACDBio 438531) and Mm-Mgl2-O1 (ACDBio 822901) were custom-designed and synthesized. They were then labelled with TSA opal 520 (PerkinElmer FP1487001KT), TSA opal 540 (PerkinElmer FP1494001KT), TSA opal 570 (PerkinElmer FP1488001KT), TSA opal 620 (PerkinElmer FP1495001KT) or TSA opal 650 (PerkinElmer FP1496001KT). Tissues were fixed for 16 hours in AntigenFix (Diapath P0016), dehydrated and embedded in OCT as described above. Slices were pre-treated with hydrogen peroxide for 10 min and protease III for 20 min. The recommended Antigen retrieval step was not performed in order to preserve epitope integrity. Probes were hybridized and amplified according to the manufacturer’s instructions. Slides were then stained for protein markers as described above. Mice were euthanized by means of carbon dioxide (CO2) overdose. The liver was excised and consequently trimmed, on ice, to smaller tissue pieces fitting the 10X Visium capture area. Trimmed tissue pieces were embedded in Tissue-Tek® O.C.T.™ Compound (Sakura) and snap frozen in isopentane (Sigma) chilled by liquid nitrogen. Embedded tissue pieces where stored at -80°C until cryosectioning. A 10X Visium Spatial Gene expression slide was placed in the cryostat (Cryostar NX70 Thermo Fisher) 30 minutes prior to cutting. 10 μm sections where cut and placed within the capture area. Single 10X Visium Spatial Gene expression slides were stored in an airtight container at -80°C until further processing. 10X Visium cDNA libraries were generated according the manufacturer’s instructions. In short: Tissue sections where fixed in chilled Methanol. A H&E staining was performed to assess tissue morphology and quality. Tissue was lysed and reverse transcription was performed followed by second strand synthesis and cDNA denaturation. cDNA was transferred to a PCR tube and concentration was determined by qPCR. Spatially barcoded, full length cDNA was amplified by PCR. Indexed sequencing libraries where generated via End Repair, A-tailing, adaptor ligation and sample index PCR. Full length cDNA and indexed sequencing libraries were analyzed using the Qubit 4 fluorometer (Thermo Fisher) and Agilent 2100 BioAnalyzer. Liver slices were prepared as described above for the classical Visium protocol. Slices were dried for 1 min at 37°C and subsequently fixed using 1% paraformaldehyde in PBS. Next, slices were blocked for 30 min (2% BSA, 0.1ug/ul Salmon Sperm, 0.5% Saponin, 1 U/μl protector RNase inhibitor (Roche) in 3X SSC) and incubated with the oligo-conjugated antibody staining mix (2% BSA, 0.1μg/μl Salmon Sperm, 0.5% Saponin, 1 U/μl protector RNase inhibitor, 10uM polyT-blocking oligo (TTTTTTTTTTTTTTTTTTTTTTTTT/3InvdT/), in 3X SSC) for 1h at 4°C. Slides were mounted (90% glycerol, 1 U/μl protector RNase inhibitor) and imaged on Zeiss Axioscan Z1 at 20X magnification. Samples were then processed for a transcriptomic experiment as per manufacturer’s instructions (Visium, 10X Genomics) with modifications to also capture antibody tags. In short, tissue was permeabilized using Tissue Removal Enzyme (Tissue Optimization kit, 10x Genomics) for 9 minutes, as determined by a tissue optimization experiment (10X Genomics, Visium Spatial Tissue Optimization). After reverse transcription, 2 μl of 100 μM FB additive primer (CCTTGGCACCCGAGAATTCCA) per sample was added to the second strand synthesis mix. During cDNA amplification 1 μl of 0,2 μM FB additive primer (CCTTGGCACCCGAGAATTCCA) was added. After cDNA amplification, antibody products and mRNA derived cDNA were separated by 0.6X SPRI select. The purified full-length cDNA fraction was quantified by qRT-PCR using KAPA SYBR FAST-qPCR kit on a PCR amplification and detection instrument. After enzymatic fragmentation indexed sequencing libraries were generated via End Repair, A-Tailing, adaptor ligation and sample index PCR. The supernatant containing antibody product was cleaned up by two rounds of 1.9X SPRI select. Next, 45 μl of the purified antibody fraction was amplified with a 96 deep well reaction module: 95°C for 3 min; cycled 8 times: 95°C for 20 s, 60°C for 30 s, and 72°C for 20 s; 72°C for 5 min; end at 4°C. ADT libraries were purified once more with 1.6X SPRI select. Full length cDNA, indexed cDNA libraries and antibody libraries were analyzed using the Qubit 4 fluorometer (Thermo Fisher) and Agilent 2100 Bioanalyzer. The separation of the cDNA and ADT libraries were performed according to the manufacturer’s instructions (10X genomics). The MACSima™ Imaging System is a fully automated instrument combining liquid handling with widefield microscopy for cyclic immunofluorescence imaging. In brief, staining cycles consisted of the following automated steps: immunofluorescent staining, sample washing, multi-field imaging, and signal erasure (photobleaching or REAlease). Cryosectioned slices on slides were taken out of the -80°C storage and the appropriate MACSWell™ imaging frame was mounted immediately on the slide. An appropriate volume of ice-cold 4% PFA solution was added (according to the MACSWell™ imaging frames datasheet) and incubated for 10 minutes at room temperature. The slide was washed three times with MACSima Running Buffer. After washing the appropriate initial sample volume of MACSima Running Buffer was added (according to the MACSWell™ imaging frames datasheet). Right before the start of the MACSima ™ instrument a DAPI pre-staining was performed: the MACSima Running Buffer was removed from the sample to be analysed and stained for 10 min with a 1:10 dilution of a DAPI staining solution (volume depends on working volume for the different MACSwell™ formats, see datasheet). The DAPI staining solution was removed and 3 washing steps were performed (MACSima Running Buffer). Finally, the initial sample volume of MACSima Running Buffer was added. Details of the antibodies used can be found in the key resources table. Liver was frozen and sectioned as described above for Visium analysis and liver slices were placed within capture areas on Resolve BioScience slides. Samples were then sent to Resolve BioSciences on dry ice for analysis. Upon arrival, tissue sections were thawed and fixed with 4% v/v Formaldehyde (Sigma-Aldrich F8775) in 1x PBS for 30 min at 4 °C. After fixation, sections were washed twice in 1x PBS for two min, followed by one min washes in 50% Ethanol and 70% Ethanol at room temperature. Fixed samples were used for Molecular Cartography™ (100-plex combinatorial single molecule fluorescence in-situ hybridization) according to the manufacturer’s instructions (protocol 3.0; available for download from Resolve’s website to registered users), starting with the aspiration of ethanol and the addition of buffer BST1 (step 6 and 7 of the tissue priming protocol). Briefly, tissues were primed followed by overnight hybridization of all probes specific for the target genes (see below for probe design details and target list). Samples were washed the next day to remove excess probes and fluorescently tagged in a two-step color development process. Regions of interest were imaged as described below and fluorescent signals removed during decolorization. Color development, imaging and decolorization were repeated for multiple cycles to build a unique combinatorial code for every target gene that was derived from raw images as described below. The probes for 100 genes were designed using Resolve’s proprietary design algorithm. Briefly, the probe-design was performed at the gene-level. For every targeted gene all full-length protein-coding transcript sequences from the ENSEMBL database were used as design targets if the isoform had the GENCODE annotation tag ‘basic’ (Frankish et al., 2019; Yates et al., 2020). To speed up the process, the calculation of computationally expensive parts, especially the off-target searches, the selection of probe sequences was not performed randomly, but limited to sequences with high success rates. To filter highly repetitive regions, the abundance of k-mers was obtained from the background transcriptome using Jellyfish (Marçais and Kingsford, 2011). Every target sequence was scanned once for all k-mers, and those regions with rare k-mers were preferred as seeds for full probe design. A probe candidate was generated by extending a seed sequence until a certain target stability was reached. A set of simple rules was applied to discard sequences that were found experimentally to cause problems. After these fast screens, every kept probe candidate was mapped to the background transcriptome using ThermonucleotideBLAST (Gans and Wolinsky, 2008) and probes with stable off-target hits were discarded. Specific probes were then scored based on the number of on-target matches (isoforms), which were weighted by their associated APPRIS level (Rodriguez et al., 2018), favoring principal isoforms over others. A bonus was added if the binding-site was inside the protein-coding region. From the pool of accepted probes, the final set was composed by greedily picking the highest scoring probes. Probe details are included in the key resources table. Samples were imaged on a Zeiss Celldiscoverer 7, using the 50x Plan Apochromat water immersion objective with an NA of 1.2 and the 0.5x magnification changer, resulting in a 25x final magnification. Standard CD7 LED excitation light source, filters, and dichroic mirrors were used together with customized emission filters optimized for detecting specific signals. Excitation time per image was 1000 ms for each channel (DAPI was 20 ms). A z-stack was taken at each region with a distance per z-slice according to the Nyquist-Shannon sampling theorem. The custom CD7 CMOS camera (Zeiss Axiocam Mono 712, 3.45 μm pixel size) was used. For each region, a z-stack per fluorescent color (two colors) was imaged per imaging round. A total of 8 imaging rounds were done for each position, resulting in 16 z-stacks per region. The completely automated imaging process per round (including water immersion generation and precise relocation of regions to image in all three dimensions) was realized by a custom python script using the scripting API of the Zeiss ZEN software (Open application development). The algorithms for spot segmentation were written in Java and are based on the ImageJ library functionalities. Only the iterative closest point algorithm is written in C++ based on the libpointmatcher library (https://github.com/ethz-asl/libpointmatcher). As a first step all images were corrected for background fluorescence. A target value for the allowed number of maxima was determined based upon the area of the slice in μm multiplied by the factor 0.5. This factor was empirically optimized. The brightest maxima per plane were determined, based upon an empirically optimized threshold. The number and location of the respective maxima was stored. This procedure was done for every image slice independently. Maxima that did not have a neighboring maximum in an adjacent slice (called z-group) were excluded. The resulting maxima list was further filtered in an iterative loop by adjusting the allowed thresholds for (Babs-Bback) and (Bperi-Bback) to reach a feature target value (Babs: absolute brightness, Bback: local background, Bperi: background of periphery within 1 pixel). This feature target values were based upon the volume of the 3D-image. Only maxima still in a z-group of at least 2 after filtering were passing the filter step. Each z-group was counted as one hit. The members of the z-groups with the highest absolute brightness were used as features and written to a file. They resemble a 3D-point cloud. To align the raw data images from different imaging rounds, images had to be corrected. To do so the extracted feature point clouds were used to find the transformation matrices. For this purpose, an iterative closest point cloud algorithm was used to minimize the error between two point-clouds. The point clouds of each round were aligned to the point cloud of round one (reference point cloud). The corresponding point clouds were stored for downstream processes. Based upon the transformation matrices the corresponding images were processed by a rigid transformation using trilinear interpolation. The aligned images were used to create a profile for each pixel consisting of 16 values (16 images from two color channels in 8 imaging rounds). The pixel profiles were filtered for variance from zero normalized by total brightness of all pixels in the profile. Matched pixel profiles with the highest score were assigned as an ID to the pixel. Pixels with neighbors having the same ID were grouped. The pixel groups were filtered by group size, number of direct adjacent pixels in group, number of dimensions with size of two pixels. The local 3D-maxima of the groups were determined as potential final transcript locations. Maxima were filtered by number of maxima in the raw data images where a maximum was expected. Remaining maxima were further evaluated by the fit to the corresponding code. The remaining maxima were written to the results file and considered to resemble transcripts of the corresponding gene. The ratio of signals matching to codes used in the experiment and signals matching to codes not used in the experiment were used as estimation for specificity (false positives). Final image analysis was performed in ImageJ using genexyz Polylux tool plugin from Resolve BioSciences to examine specific Molecular Cartography™ signals. 40000-160000 cells of interest from livers of the different species were FACS-purified and pelleted by centrifugation at 400g for 5 mins. To ensure sufficient numbers of all cell types were present in our analyses, depending on the sample distinct populations of cells including Live CD45, Live CD45, Hepatocytes, Myeloid cells and Stromal cells were FACS-purified. When CITE-seq was to be performed, cells were then stained with 2.4G2 antibody to block Fc receptors and CITE-seq antibodies for 20mins at 4°C, before being washed in excess PBS with 2% FCS and 2mM EDTA. Antibody details are included in the key resources table. 40000-100000 nuclei were also FACS-purified based on DAPI expression. These were sorted into BSA coated tubes and pelleted by centrifuging for 3 mins at 400g and 5 mins at 600g sequentially. Cells/Nuclei were then resuspended in PBS with 0.04%BSA at ∼1000 cells/ml. Cell suspensions (target recovery of 8000-10000 cells) were loaded on a GemCode Single-Cell Instrument (10x Genomics, Pleasanton, CA, USA) to generate single-cell Gel Bead-in-Emulsions (GEMs). Single-cell RNA-Seq libraries were prepared using GemCode Single-Cell 3ʹGel Bead and Library Kit (10x Genomics, V2 and V3 technology) according to the manufacturer’s instructions. Briefly, GEM-RT was performed in a 96-Deep Well Reaction Module: 55°C for 45min, 85°C for 5 min; end at 4°C. After RT, GEMs were broken down and the cDNA was cleaned up with DynaBeads MyOne Silane Beads (Thermo Fisher Scientific, 37002D) and SPRIselect Reagent Kit (SPRI; Beckman Coulter; B23318). cDNA was amplified with 96-Deep Well Reaction Module: 98°C for 3 min; cycled 12 times : 98°C for 15s, 67°C for 20 s, and 72°C for 1 min; 72°C for 1 min; end at 4°C. Amplified cDNA product was cleaned up with SPRIselect Reagent Kit prior to enzymatic fragmentation. Indexed sequencing libraries were generated using the reagents in the GemCode Single-Cell 3ʹ Library Kit with the following intermediates: (1) end repair; (2) A-tailing; (3) adapter ligation; (4) post-ligation SPRIselect cleanup and (5) sample index PCR. Pre-fragmentation and post-sample index PCR samples were analyzed using the Agilent 2100 Bioanalyzer. RNA was extracted from 10000 sorted cells (gated using strategies shown) from livers of C57BL/6 mice using a RNeasy Plus micro kit (QIAGEN). Sensifast cDNA synthesis kit (Bioline) was used to transcribe total RNA to cDNA. Real-time RT-PCR using SensiFast SYBR No-Rox kit (Bioline) was performed to determine gene expression, therefore a PCR amplification and detection instrument LightCycler 480 (Roche) was used. Gene expression was normalized to β-actin gene expression. Primers used in the study can be found in the key resources table. sc/snRNA-seq libraries were loaded on an Illumina HiSeq or Illumina NovaSeq 6000 with sequencing settings recommended by 10X Genomics (26/8/0/98 – 2.1pM loading concentration, ADT and cDNA libraries were pooled in a 25:75 ratio). Visium sequencing libraries were loaded on an Illumina NovaSeq 6000 with sequencing settings recommended by 10X Genomics (28/10/10/75 – 2.1pM loading concentration). Sequencing was performed at the VIB Nucleomics Core (VIB, Leuven). The demultiplexing of the raw data was performed using CellRanger software (10x – version 3.1.0; cellranger mkfastq which wraps Illumina’s bcl2fastq). The reads obtained from the demultiplexing were used as the input for ‘cellranger count’ (CellRanger software), which aligns the reads to the mouse reference genome (mm10) using STAR and collapses to unique molecular identifier (UMI) counts. The result is a large digital expression matrix with cell barcodes as rows and gene identities as columns. To remove ambient RNA, the FastCAR R package (v0.1.0) with a contamination chance cutoff of 0.05 was run on the samples separately before merging them. The UMI cut off was determined individually for the different samples, using the CellRanger web_summary output plot (see GitHub). The Scater R package (v1.14.6) was used for the preprocessing of the data. The workflow to identify the outliers, based on 3 metrics (library size, number of expressed genes and mitochondrial proportion) described by the Marioni lab (Lun et al., 2016) was followed. As a first step cells with a value x median absolute deviation (MADs) higher or lower than the median value for each metric were removed. This value was determined individually for the different datasets (see github). Secondly, the runPCA function (default parameters) of the Scater R package was used to generate a principal component analysis (PCA) plot. The outliers in this PCA plot were identified by the R package mvoutlier. By creating the Seurat object, genes that didn’t have an expression in at least 3 cells were removed. To normalize, scale and detecting the highly variable genes, the R package SCTransform (v0.2.1) was used. If batch correction (on sample level) was needed, the NormalizeData (log2 transformation), FindVariableFeatures and ScaleData functions of the Seurat R package (v3.1.2) were used in combination with the Harmony R package (v1.0). The Seurat pipeline was followed to find the clusters and create the UMAP plots. The number of principal components used for the clustering and the resolution were determined individually for the different datasets (see GitHub). On these initial UMAP plots we did multiple rounds of cleaning by removing proliferating and contaminating (e.g. doublets) cells. For non CITE-seq datasets the count data for the clean cells acquired by the previous steps were further processed with the scVI model (scvi Python package v0.6.7) (Lopez et al., 2018). Datasets including Cite-seq samples were further processed with the TotalVI model (Gayoso et al., 2021). The workflows described on scvi-tools.org were followed to generate new UMAPs, DEGs and DEPs. This information was further processed with the pheatmap R package (v1.0.12) to create heatmaps using the normalized values (denoised genes) calculated in the scVI/TotalVI workflow. The plots showing the expression of certain genes or proteins are created with the ggplot2 R package (v3.2.1) with a quantile cut off of 0.01. For mouse all the ABs from the whitelist (181 ABs) were loaded into TotalVI, while for the other species only the added ABs were loaded into TotalVI. For the ‘human liver-pool of techniques and patients’ we noticed that the batch correction (between samples) faced difficulties for the hepatocytes and stellate cells as the cells all originated from snRNA-Seq samples, while the other cell types originated from both snRNA-seq and scRNA-seq samples. To overcome this issue we randomly allocated 30% of the hepatocytes to scRNA-seq samples which were not CITE-seq samples. We did the same for 30% of the stellate cells. Heatmaps were made by scaling the normalized values (denoised values; calculated in the scVI/TotalVI workflow) using the scale_quantile function of the SCORPIUS R package (v1.0.7) and the pheatmap R package (v1.0.12). The plots showing the expression of certain genes or proteins were created based on the normalized values (denoised values) using a quantile cutoff of 0.99 and via either the ggplot2 R package (v3.2.1) or the scanpy.pl.umap function of the Scanpy Python package (v1.5.1). To find the conserved human and mouse KC markers we started by identifying the human KC markers. We mapped the annotation of the human myeloid UMAP on the human pool of techniques/patients UMAP to identify the real KCs in this last UMAP. The real KCs were identified as the top part of the mac cluster. Using this new annotation we then calculated the DE genes and DE proteins for each cluster. Some genes are listed as marker for multiple clusters, only for the cluster where the gene had the highest score (raw_normalized_mean1/raw_normalized_mean2lfc_mean), the gene was kept as marker. This way we found 110 potential human KC markers. We then created a heatmap of these 110 genes (using denoised gene values scaled between 0 and 1) and filtered this heatmap by removing the genes where the scaled normalized value was higher than 0.50 in more than 30% of the cells of a certain cell type other than KCs. Except for the macs, we only removed a gene when it had a scaled normalized value higher than 0.50 in more than 70% of the macs. After this filtering we ended up with 36 human KC markers. Next we converted these human gene symbols into MGI IDs via the BioMart tool on the HGNC website (https://biomart.genenames.org/martform/#!/default/HGNC?datasets=hgnc_gene_mart). We found a MGI ID for 30 genes. We then converted these MGI IDs into mouse gene symbols via the MGI webtool (http://www.informatics.jax.org/batch/). To identify the mouse KC markers we similarly mapped the annotation of the mouse myeloid UMAP on the mouse pool of techniques UMAP to identify the real KCs in this last UMAP. The real KCs matched with the mac cluster. Similarly as in human, the DE genes for each cluster was calculated and genes listed as marker for multiple clusters were dealt with in a similar way. This way we found 264 potential mouse KC markers. We then removed the genes that had a score (raw_normalized_mean1/raw_normalized_mean2lfc_mean) lower than 10 and ended up with 214 genes. We then created a heatmap of these 214 genes (using denoised gene values scaled between 0 and 1) and filtered this heatmap by removing the genes where the scaled normalized value was higher than 0.50 in more than 30% of the cells of a certain cell type other than KCs. After this filtering we ended up with 68 mouse KC markers. Next we converted these mouse gene symbols into MGI IDs via the MGI webtool (http://www.informatics.jax.org/batch/). We then converted these MGI IDs into human gene symbols via the BioMart tool on the HGNC website (https://biomart.genenames.org/martform/#!/default/HGNC?datasets=hgnc_gene_mart) and ended up with 60 genes. At this point we found 30 human KC markers and 60 mouse KC markers. In a next step, we only kept the human KC markers that we identified as a Highly Variable Gene (HVG) in the mouse pool of techniques UMAP (20 genes) and the mouse KC markers that were identified as HVGs in the human pool of the techniques UMAP (30 genes). We next put these 20 mouse KC markers in SingleCellSignatureExplorer (Pont et al., 2019) to see where these genes are enriched in the mouse pool of techniques UMAP. In order to only get an enrichment in the KCs we decided to only use top 10 mouse KC markers (ordered on score), together with Slc40a1 and Hmox1. We then started to add the top human KC markers as long as we keep the enrichment solely in the KCs. This way we ended up with final list of 15 human-mouse conserved KC markers. We next converted these KC markers into the monkey, pig, chicken or zebrafish orthologs by looking up the human gene symbol on NCBI (https://www.ncbi.nlm.nih.gov/search/) and checking if there is an ortholog of the species of interest listed under the ‘Ortholog’ tab. The found orthologs were then used as input for the SingleCellSignatureExplorer tool. The protein normalized values (denoised values; calculated in the TotalVI workflow) were converted into an FCS file using the write.FCS function of the flowCore R package (v1.50.0). We first removed per sample all spots that were clear outliers compared to the location of the tissue. Each sample was then normalized individually using the SCTransform function of the Seurat R package (v3.2.3) with default parameters. All samples were then merged with the merge function of the Seurat R package (v3.2.3) with default parameters. Next, we determined the HVGs, created a PCA plot, performed clustering and created an UMAP plot as described in the spatial workflow available on the Seurat website (https://satijalab.org/seurat/articles/spatial_vignette.html). Clusters which showed high mitochondrial gene expression were removed. Spots located at the darker parts of the tissue were also removed as these parts are considered to be dead tissue or of bad quality. For modelling the cell type composition and zonation, spatial CITE-seq and transcriptomics data were analyzed using probabilistic graphical models, similar to what is used in tools such as cell2location and scVI. In brief, transcriptomics data was modelled as a NegativeBinomial distribution, parameterized with a mean and dispersion , the latter optimized as a free parameter for each gene. Visium Highly Multiplexed Protein data was modelled as a mixture of NegativeBinomials, with a and and a shared dispersion . The actual foreground/background signal within a modality was modelled as a that depends on the latent space, and which is multiplied with the empirical library size to get . For Visium Highly Multiplexed Protein, was modelled as a latent variable specific for each gene. for Visium Highly Multiplexed Protein and for RNA-seq were modelled as deterministic functions depending on the use case as described in the following paragraphs. The posterior of the probabilistic graphical model was inferred using black-box variational inference (Ranganath et al., 2013), in which the variational distribution was specified as a diagonal Normal distribution, transformed into the correct domain using transforms (, , ). Free parameters within this model were optimized using gradient descent, with the ELBO as loss function and Adam as optimizer as implemented in Pytorch (Paszke et al., 2019) (pytorch.org). We used a learning rate of 0.01 for variational parameters, and 0.001 for parameters of the amortization functions. To calculate the average expression of each gene within a cell type, we used a linear model in which both and were modelled as a latent variable specific for each gene and cell type. The for nuclei were multiplied with a gene-specific correction factor (optimized as a latent variable) that corrected for differences between scRNA-seq and snRNA-seq. Given that spatial transcriptomics data sequences the whole cell, the uncorrected values were used for spatial deconvolution. To infer the proportions of each cell type within a spot, we used a model in which the gene expression is modelled as a linear combination of cell type proportions and average expression in each cell type: For we adapted the values from the reference, but included - A capture bias per gene, which corrects for technical and biological differences between spatial and sc/sn-RNA-seq. The capture bias was modelled as a latent variable with prior - A red blood cell cell type, which was not included in the reference dataset but nonetheless had a dominant presence in the spatial data. The of this cell type was set to zero for all genes except Hbb-bt, Hbb-bs, Hba-a1, Hba-a2 for mouse and HBB, HBA1, HBA2 for human, which were modelled as free parameters. - Similarly, the expression of complement factors (C3, C2, C4B/C4b) within hepatocytes was modelled as free parameters. A background signal shared for all spots was also modelled as follows: With a latent variable specific to each spot and a latent variable specific to each gene. A likelihood ratio test was used to assess whether a cell type was significantly present in a spot. Specifically, if is the gene expression of all genes at a particular spot, we used Monte Carlo samples from the posterior to estimate: A cell type was deemed significantly present if the log-likelihood was higher than 10. The zonation of spots was modelled as a univariate latent variable specific to each spot. This latent variable influenced the gene expression using a spline function by using a gaussian basis function () with 10 knots at uniform fixed positions. The coefficients of this spline were modelled as a latent variable specific for each gene, with prior a Gaussian random walk distribution, and the step . was determined empirically as 2 times the standard deviation of the log1p transformed expression values in the whole dataset. The variational parameters of the zonation and were not optimized directly but were estimated using an amortization function. This amortization function used the count matrix as input, and estimated the variational parameters using the following layers: Linear (with 100 output dimensions), BatchNorm, ReLU, Linear (again with 100 output dimensions), ReLU, and a final Linear layer. This amortization function was used to transfer the zonation onto a different dataset, i.e., 1) to transfer the zonation trained on mouse spatial transcriptomics onto mouse Visium highly multiplexed protein and 2) to transfer the zonation trained on human low steatosis (<10%) onto human high steatosis (>30%). To determine the differential abundance of a cell type across zonation, the significant presence of a cell type within a spot was modelled using a spline function with the zonation of a cell type as input. The coefficients of this spline function were modelled as a latent variable with the step size . To determine differences in abundance between patients with high and low steatosis, we first modelled the zonation on human data on patients with steatosis < 10%. Potential interaction effects between zonation and steatosis status were then modelled using a spline function as before, but with a separate set of coefficients for both high and low steatosis. A likelihood ratio test was then used to determine whether this interaction was present significantly, by comparing the likelihood of this model with a model with shared coefficients. To analyze cell-cell communication in the hepatic mac niches, we applied Differential NicheNet, which is an extension of the default NicheNet pipeline to compare cell-cell interactions between different niches and better predict niche-specific ligand-receptor (L-R) pairs. It uses a flexible prioritization scheme that allows ranking L-R pairs according to several properties, such as niche- and region-specific expression of the L-R pair, ligand activity, and level of database curation. This in contrast to the default NicheNet pipeline which prioritizes expressed L-R pairs solely based on ligand activity predictions. All analyses were conducted according to the Differential NicheNet tutorial (https://github.com/saeyslab/nichenetr/blob/master/vignettes/differential_nichenet.md). As input to the Differential NicheNet pipeline, we used the data after normalization via SCTransform and integration of scRNA-seq and snRNA-seq according to the Seurat procedure for integration (Stuart et al., 2019). For the mouse analyses, Differential NicheNet was first performed for the following 3 niche comparisons: 1) KCs versus central vein macs; 2) KCs versus capsule macs; 3) KCs versus LAMs. Following sender cell types were considered for these niches: KC niche: periportal hepatocytes, periportal LSECs, and periportal stellate cells; Central vein mac niche: central vein ECs and central vein fibroblasts; Capsule mac niche: mesothelial cells and capsule fibroblasts; LAM niche: cholangiocytes and bile duct fibroblasts. Because of the preferentially periportal localization of KCs in the mouse liver, we also included a ‘region specificity' factor in the Differential NicheNet prioritization framework. This was done to increase the ranking of ligands that are more strongly expressed in periportal than pericentral niche cells. Periportal sender cells were determined after subclustering based on the following markers: Hal and Sds for hepatocytes; Mecom, Msr1, and Efnb2 for LSECs; Ngfr, Igfbp3, and Dach1 for stellate cells. In the heatmap (Figure S8G), we show the prioritization scores of the top 40 ligands (and their highest scoring receptor) in the KC niche (score averaged over the 3 analyses), and of all the non-KC niche L-R pairs with a prioritization score ≥ the score of the lowest scoring KC L-R pair of this top 40. For each L-R pair/niche combination, we only displayed the score of the sender cell with the highest score (e.g. for the Csf1-Csf1r interaction in the KC niche, the score is shown for the LSEC-KC interaction because that score was higher than for Stellate–KC and Hepatocyte–KC; in the LAM niche, the score of Csf1-Csf1r is shown for the bile duct fibroblast – LAM interaction and not for the cholangiocyte–LAM interaction, etc.). Because of the strong concordance between the top-ranked L-R pairs in these 3 non-KC mac niches, it was decided to also conduct a subsequent analysis in which the KC niche is compared against all non-KC hepatic mac niches combined. For this final ‘KC versus all non-KC mac analysis’, KCs were compared to central vein macs, capsule macs, and LAMs together, with the same sender cell types as described here above (but now analyzed together). For the human analyses, Differential NicheNet was performed to compare the KC niche with the non-KC mac niches (similarly as the final analysis in mouse). For the KC niche, all hepatocytes, LSECs, and stellate cells were selected as sender cells; and KCs as receiver cells. For the non-KC mac niche, cholangiocytes, fibroblasts, and central vein ECs were considered as the sender cells; Mat. LAMs, Imm. LAMs, and Mac1s as the receiver cells (Figure 4H). To find KC-niche-specific L-R pairs that are conserved across mouse and human, the individual mouse and human prioritization scores were averaged to form a ‘conservation score’. The 40 ligands (and maximally 3 of their highest scoring receptors) with the highest conservation score were selected for further analysis (note: the L-R pair should be expressed by the same sender-receiver pair in both species). In the circos plot (Figure 6C; Gu et al., 2014), only a subset of these top L-R pairs is shown to keep the figure clearly interpretable. Following ligands were not shown: ITGA9, SEMA6D, JAM3, ITGB1 (stellate cells); ITGA9, F8, CD274, HSP90B1 (LSECs); C5, F9, F2, FGA, TF, TTR, COL18A1, COL5A3, SERPINA1, SERPINC1 (hepatocytes). The depicted target genes are KC-specific in both mouse and human, and a top-predicted target according to the NicheNet ligand-target regulatory potential scores. NR1H3 was manually added as a NOTCH2 target based on recent studies (Bonnardel et al., 2019). Resected human liver was fixed in 4% formalin for 24-48h and subsequently embedded in paraffin. Samples were stored for 10-15 years at RT before analysis. Sections of 6 μm thick were cut using a Microm HM360 and mounted on a polarized glass slide. These sections were deparaffinized in xylene and rehydrated in a graded ethanol series. Antigen retrieval was performed by immersing the samples for 5 min in pH 8.3 TRIS-EDTA at 98°C. Slides were then cooled to RT and washed in PBS. Confocal staining was performed as described above. BM was isolated from the tibia and femur of mice by centrifugation. Red blood cells were lysed and single cell suspensions were stained with antibodies for flow cytometry. BM monocytes were sorted as live CD45+ CD11b+ Ly6G- Ly6C+ CD115+ cells using a BD FACSAria III. Monocytes were resuspended in DMEM/F12 media supplemented with 10% FCS, 30ng/ml CSF1, 2mM Glutamine and 100U/ml penicillin and streptomycin. 150,000 monocytes were seeded in each well of an adherent 24-well plate pre-coated with bovine collagen type I and cultured overnight (37C, 5% CO2). The following day 0, 25 or 50ng/ml of ac-LDL was added. Ac-LDL was kindly provided by Sophie Janssens, Ghent, Belgium who received the material from Wilfried Le Goff, Paris, France. 14 hours later cells were harvested and live F4/80+ cells were FACS-purified in RLT plus buffer containing 1% β-mercaptoethanol. RNA isolation, cDNA synthesis and qPCR were performed as described above. Bone marrow chimeras were generated as described previously (Scott et al., 2016). Briefly, 6-12 week old Clec4f-Dtr mice (CD45.1) were anaesthetized by intraperitoneal administration of Ketamine (150 mg/kg) and Xylazine (10 mg/kg). Mice were lethally irradiated with 8 Gy, with the livers being protected with a 3-cm-thick lead cover. Once recovered from the anesthesia, mice were reconstituted by intravenous administration of 5-10×10 BM cells from CD45.2 Acvrl1 or Fcgr1-CrexAcvrl1l mice. 4 weeks after reconstitution mice were administrated a single dose of 500ng DT via intraperitoneal injection to deplete KCs. Chimerism was assessed 7 or 13 days later by flow cytometry and compared with chimerism levels in blood Ly6C monocytes. Clec4f-Dtr mice were administered 10mg/kg ALK1Fc, TGFβRIIFc or appropriate isotype controls (hIgG1 and mIgG2a; Acceleron Pharma) by intraperitoneal injection on days -1, 2, 3 and 5. On Day 0 mice were also administered a single dose of 500ng DT i.p. to deplete KCs. Livers were harvested at day 7 to assess KC development. In all experiments, data are presented as mean ±SEM and/or individual data points are presented unless stated otherwise. Statistical tests were selected based on appropriate assumptions with respect to data distribution and variance characteristics. Details of the precise test used for each analysis can be found in the figure legends. Statistical significance was defined as p<0.05. Sample sizes were chosen according to standard guidelines. Number of animals/patients is indicated as ‘‘n’’. The investigators were not blinded to the group allocation, unless otherwise stated. The sc/snRNA-sequencing, CITE-seq FCS files and spatial transcriptomics datasets will be made available for visualization, analysis and download at www.livercellatlas.org. We thank all patients and their families for participating in this study. We also thank Janssen for providing the macaque samples; Acceleron Pharma for the kind gift of the ALK1-Fc, TGFBRII-Fc, and isotype controls; the IRC-VIB Flow and Bioimaging core facilities for assistance; VIB Tech Watch and the VIB single-cell accelerator program for their help benchmarking technologies; and 10X Genomics for their help setting up the Visium highly multiplexed protein analysis. Finally, we thank the VIB-UGent animal house staff. BioRender was used to generate some figures. Funding: Chan Zuckerberg initiative; liver seed atlas grant (M.G. and C.L.S.), FWO SBO; iPSC LiMics (C.L.S., M.G., and Y.S.), ERC consolidator grant; KupfferCellNiche, 725924 (M.G.), GOA; BOF18-GOA-024 (M.G. and Y.S.), ERC starting grant; MyeFattyLiver, 851908 (C.L.S.), FWO project grant; 3G000519, (C.L.S. and M.G.), FWO PhD fellowship; 1129521N (B.H.), 1181318N (R.B.), 11L2122N (F.F.D.P.), MSCA IF fellowships; MACtivate 101027317 (C.Z.), LiverMacRegenCircuit 844301 (F.R.S.). The authors declare no competing interests. Published: January 11, 2022 Supplemental information can be found online at https://doi.org/10.1016/j.cell.2021.12.018. Martin Guilliams, Email: martin.guilliams@ugent.be. Charlotte L. Scott, Email: charlotte.scott@ugent.be. The datasets generated during this study have been deposited in the Gene Expression Omnibus public database under accession number GSE192742. The datasets generated during this study have been deposited in the Gene Expression Omnibus public database under accession number GSE192742.
PMC11699618
Bivalent SMAC mimetic APG-1387 reduces HIV reservoirs and limits viral rebound in humanized mice
Latent viral reservoirs (VRs) represent a main barrier to HIV cure. Thus, developing new approaches that can purge and eliminate VRs paves the path toward achieving an HIV-1 cure. APG-1387, a bivalent SMAC mimetic (SM), efficiently reactivates latent HIV expression in T cell line models and enhances active caspase 3 expression, a condition that typically leads to apoptosis. In primary CD4 T cells infected with a dual reporter-encoded HIV, APG-1387 decreases latently infected cells without a notable effect on productively infected cells. In virally suppressed humanized (hu)-BLT mice, APG-1387 augments cell-associated viral RNA and potently reduces HIV DNA-containing cells without modulating T cell activation or proliferation. Upon antiretroviral therapy (ART) interruption, HIV rebound was decreased in APG-1387-treated humanized mice (hu-mice), and the viremia maintained at levels below that of pre-ART. Thus, the ability of APG-1387 to affect VRs and decrease viral rebound highlights the potential of bivalent SMs in HIV cure strategies.Antiretroviral therapy (ART) has been highly effective at suppressing human immunodeficiency virus (HIV) replication, thus significantly reducing disease progression and mortality in infected individuals. However, ART is not curative and does not meaningfully impact persistent viral reservoirs (VRs) of latently infected cells that comprise CD4 T cells and myeloid cells. Indeed, upon antiviral treatment interruption (ATI), the virus quickly rebounds after several weeks, highlighting how the presence of VRs obviates an HIV cure. Therefore, eliminating or controlling VRs remains an unwavering priority for HIV cure research. In this context, the aim of the “shock-and-kill” strategy is to reactivate latently infected cells with latency reversal agents (LRAs) and render them vulnerable to cell death or clearance by the immune system. Previously reported approaches to HIV reactivation, including protein kinase C (PRKC) agonists, histone deacetylase (HDAC) inhibitors (HDACi), and toll-like receptor (TLR) agonists, have been highly effective in in vitro models of latency, but their efficacy is only moderate when tested in vivo. Moreover, by broadly activating cellular pathways, these compounds elicit significant proinflammatory effects or alter the function and fate of specific immune effector cells, hence limiting their use in clinical settings. Similarly, promising candidates such as bryostatin-1 and analogs, phorbol esters, and phosphatidylethanolamine binding protein 1 (PEBP1) agonists reactivate HIV via activation of the canonical nuclear factor kappa B (NFKB) pathway, inadvertently leading to uncontrolled cytokine release and overt T cell activation. Consequently, there is a need to develop alternative approach(es) that can reactivate latent reservoirs and eliminate infected cells without triggering systemic activation, inflammation, or impairment of immune clearance mechanisms. In this regard, small molecules known as mimetics of the second mitochondrial activator of caspases (SMAC mimetics [SMs]) were originally developed as cancer therapeutics. They have received increasing recognition because they specifically activate the non-canonical NFKB pathway, which naturally exhibits higher functional selectivity and more confined proinflammatory effects compared to the canonical NFKB pathway. Engagement of the non-canonical NFKB pathway by SMAC leads to degradation of cellular inhibitor of apoptosis (cIAP), accumulation of NFKB-inducing kinase (NIK), and activation of component of inhibitor of NFKB kinase complex (CHUK) homodimer, culminating in cleavage of inactive NFKB2 p100 to active p52. An association between RELB and p52 induces expression of target genes and, in the context of HIV, the pathway reactivates latently infected VRs. The non-canonical NFKB pathway can also be triggered by signaling intermediates of the apoptosis cascade. In fact, cleavage of SMAC/DIABLO exposes the N-terminal motif Ala-Val-Pro-Ile, which binds specifically to baculovirus intermediate repeat domains 2 and 3 of IAP proteins. These proteins in turn trigger downstream events that ultimately lead to degradation of baculoviral IAP repeat containing 2 (BIRC2, also known as cIAP1) and baculoviral IAP repeat containing 3 (BIRC3, also known as cIAP2) and potentiation of apoptosis. The unique ability of SMAC to degrade IAPs and activate apoptosis pathway(s) makes SMs interesting candidates in the field of HIV cure research because latently infected CD4 T cells display aberrant expression of cell survival factors, including XIAP, BIRC2 and BCL2. Pharmacological activation of the non-canonical NFKB pathway by SMs was recently found to induce on-ART plasma viremia in animal models of HIV latency, underscoring the potential of this class of molecules as LRAs. However, it remains unclear whether induction of HIV expression by SMs leads to a reduction of VRs in lymphoid and non-lymphoid tissues of animal models. In this study, we show that bivalent APG-1387, currently in clinical development in the oncology field, activates the non-canonical NFKB pathway and hence, is a potent LRA. By degrading IAPs, this compound also induces the expression of active caspase 3 (CASP3), a key component of the execution phase of apoptosis, in latently infected cells. Likewise, in primary CD4 T cells infected with a dual reporter-encoded HIV, APG-1387 reduces the level of latent cells without notably affecting the productively infected pool. Accordingly, in vivo treatment with APG-1387 could reactivate expression of latent viruses and was found to meaningfully reduce the integrated HIV-DNA level in tissues of ART-suppressed humanized bone marrow liver thymus (BLT) mice, without assessable immunotoxicity. Upon ART interruption, APG-1387-treated mice rebounded more slowly and to a lower set point. Overall, the study demonstrates that APG-1387 has the capacity to not only reverse HIV latency but also potentiate cell apoptosis, thus supporting the notion that bivalent SMs could be harnessed to reduce VRs without causing generalized T cell activation. APG-1387 is a bivalent SM with demonstrable antitumor activity. However, it is not known whether this compound can reactivate latent HIV and/or sensitize latently infected or reactivated cells to death. Hence, we first assessed HIV reactivation by APG-1387 in 2D10 cells, a latently infected Jurkat T cell line carrying a lentiviral vector that expresses Tat H13L and Rev in cis and a short-lived green fluorescent protein in place of Nef, and compared it to that of other SMs known to have latency-reversing activity. Indeed, bivalent SMs such as APG-1387 and Birinapant were more effective at inducing viral reactivation in 2D10 cells at equivalent concentrations compared to monovalent SMs GDC-0152, AT-406, and LCL-161 (Figures 1A and S1A). Subsequently, we performed the same analysis in J-lat 10.6 cells, a subclone of Jurkats that carries a single copy of an integrated full length Env-defective GFP (in place of Nef)-marked HIV. When J-Lat 10.6 cells were treated with the same SMs, we observed significant viral intracellular reactivation by the bivalent SMs, thus extending the reactivation potential of APG-1387 to another CD4 T cell line model of HIV latency (Figure S1B).Figure 1SMs reactivate HIV-latently infected model T cells via activation of the non-canonical NFKB pathway and lead to degradation of IAPs(A) Left panel: Chemical structure of APG-1387 with the IUAPC nomenclature. Right panel: HIV reactivation in latently infected 2D10 cells, as measured by the percentage of cells expressing GFP, 48 h after treatment with increasing concentrations (0.1 nM–10 μM) of monovalent and bivalent SMs. Data are represented as mean ± SD from two experiments (n = 2). The dotted line is the level of reactivation in vehicle control (DMSO)-treated cells.(B) Immunoblots for 2D10 cells treated with vehicle, TNF (an activator control of the canonical NFKB pathway, 10 ng/mL), or different concentrations of APG-1387. Expression of the indicated markers of non-canonical NFKB (p100 and p52) and canonical NFKB (IKBA) pathways was determined. Shown is a representative blot of three experiments (n = 3). Numbers shown underneath the blots were obtained from densitometry analysis of immunoblot bands of protein of interest. Quantification was done separately for p100, p52, and IKBA by normalizing to the loading control actin beta (ACTB) using ImageJ software.(C) Immunoblots of 2D10 cells treated with vehicle control, TNF, or APG-1387 probed for BIRC2 (cIAP1) expression. Shown is a representative blot from two independent experiments (n = 2). Quantification was done as described for Panel B.(D) Evaluation of HIV reactivation in NIK-knockout 2D10 cells treated with AGP-1387, as was done in Panel A. Shown are results from three independent experiments (n = 3) using GraphPad Prism 8.0 software (data are represented as mean ± SD).(E) Flow cytometry analysis showing the frequency of 2D10 cells expressing active (cleaved) CASP3 after 24-h treatment with APG-1387. Vehicle (DMSO only) was used as a negative control. See also Figures S1–S3. SMs reactivate HIV-latently infected model T cells via activation of the non-canonical NFKB pathway and lead to degradation of IAPs (A) Left panel: Chemical structure of APG-1387 with the IUAPC nomenclature. Right panel: HIV reactivation in latently infected 2D10 cells, as measured by the percentage of cells expressing GFP, 48 h after treatment with increasing concentrations (0.1 nM–10 μM) of monovalent and bivalent SMs. Data are represented as mean ± SD from two experiments (n = 2). The dotted line is the level of reactivation in vehicle control (DMSO)-treated cells. (B) Immunoblots for 2D10 cells treated with vehicle, TNF (an activator control of the canonical NFKB pathway, 10 ng/mL), or different concentrations of APG-1387. Expression of the indicated markers of non-canonical NFKB (p100 and p52) and canonical NFKB (IKBA) pathways was determined. Shown is a representative blot of three experiments (n = 3). Numbers shown underneath the blots were obtained from densitometry analysis of immunoblot bands of protein of interest. Quantification was done separately for p100, p52, and IKBA by normalizing to the loading control actin beta (ACTB) using ImageJ software. (C) Immunoblots of 2D10 cells treated with vehicle control, TNF, or APG-1387 probed for BIRC2 (cIAP1) expression. Shown is a representative blot from two independent experiments (n = 2). Quantification was done as described for Panel B. (D) Evaluation of HIV reactivation in NIK-knockout 2D10 cells treated with AGP-1387, as was done in Panel A. Shown are results from three independent experiments (n = 3) using GraphPad Prism 8.0 software (data are represented as mean ± SD). (E) Flow cytometry analysis showing the frequency of 2D10 cells expressing active (cleaved) CASP3 after 24-h treatment with APG-1387. Vehicle (DMSO only) was used as a negative control. See also Figures S1–S3. Further, APG-1387 treatment led to an accumulation of p52 and simultaneous reduction of its precursor p100 (Figure 1B; Figure S1C), strongly indicating that the non-canonical NFKB pathway was activated. Unsurprisingly, stimulation with tumor necrosis factor (TNF), a quintessential activator of the canonical NFKB pathway, led to a near complete disappearance of total levels of the NFKB transcription factor inhibitor, IKBA, when compared to the untreated controls. However, no changes were noted in response to APG-1387 treatment, suggesting that the canonical NFKB pathway was not affected (Figure 1B and quantitative analysis of IKBA levels; Figure S1C). To further confirm engagement of the non-canonical NFKB pathway by APG-1387, we examined expression levels of host factors known to be downregulated upon activation of the non-canonical NFKB signaling cascade. As shown, BIRC2 was significantly depleted in APG-1387-treated cells in a concentration-dependent manner (Figure 1C, upper panel) with a complete depletion observed starting at 10 nM. Activation of the non-canonical NFKB pathway hinges on the stability of the NIK, which is downregulated by BIRC2/3. The fact that APG-1387 treatment of NIK-knock-out 2D10 cells with APG-1387 did not lead to measurable viral reactivation, unless it was at a very high concentration (10 μM) and even then, the reactivation was drastically reduced, highlights the importance of the non-canonical NFKB pathway in latency reversal induced by SMs (Figure 1D). Given that SMAC are activators of caspases, it is not surprising that SMs can induce apoptosis in cancer cells. In the context of our study, we assessed whether APG-1387 could sensitize latently infected cells to apoptosis. Indeed, APG-1387 treatment led to a dose-dependent increase in the level of CASP3 in both 2D10 (Figure 1E) and J-Lat 10.6 (Figure S1D), signifying potential activation of apoptosis pathway(s). Taken together, APG-1387 is a bivalent SM, capable of efficiently reactivating HIV expression via the non-canonical NFKB pathway and enhancing the expression of active CASP3 in CD4 T cell line models of viral latency. We next assessed whether the findings obtained in the cell lines could also be replicated in primary CD4 T cells. In uninfected cells, APG-1387 indeed significantly induced BIRC2 and BIRC3 degradation, enhanced p100 processing into p52 while keeping IKBA at comparable levels, indicating a primary activation of the non-canonical NFKB pathway (Figure S2). As well, APG-1387 treatment led to a marked increase in the level of cleaved caspase 3 in both uninfected and HIV-infected CD4 T cells, and the enhancement was more pronounced in infected cells compared to the uninfected (Figure S3). To study whether APG-1387 could reactivate latent virus in primary CD4 T cells and/or sensitize them to death, we took advantage of the single-cycle HI.fate.E dual reporter virus with which latent cells (i.e., not expressing viral genes) or cells supporting viral replication (or replicating infected cells), can be readily identified by flow cytometry as only ZS-Green-positive cells or E2-Crimson positive, ZS-Green -positive/negative cells, respectively (Figure 2A). As originally reported, in this dual reporter virus, HIV LTR-directed gene expression was used as a marker for cells supporting viral replication although the virus is not replication-competent and viral particles released are not assessed. As expected, pretreatment of CD4 T cells with the reverse transcriptase inhibitor Efavirenz blocked infection. Phorbol-12-myristate-13-acetate (PMA) and ionomycin, which activate the canonical NFKB pathway, reduced the frequency of latently infected cells, and increased the level of cells supporting replicating virus as compared to the control. At 10 μM APG-1387, the latently infected cell population was decreased by at most∼2-fold but the pool of cells supporting HIV LTR-directed gene expression was not significantly affected (Figures 2A and 2B), implying that APG-1387 might be more effective at targeting the latent cell population in this primary cell model. Indeed, compared to vehicle-treated cells, the decrease in the frequency of latent cells was statistically significant at 10 μM APG-1387, and such a reduction was also statistically meaningful between 0.1 μM and 10 μM (Figure 2B). In this context, our finding is consistent with that reported by others, which have shown that SMs can selectively eliminate latently infected primary cells or latent reservoirs from T cells of HIV-infected, ART-suppressed individuals, without inducing detectable viral reactivation. We did not observe any sex-related differences, albeit there was only one male out of six subjects studied. Taken together, our data demonstrate that APG-1387 is an effective LRA in T cell lines but in primary CD4T cells, it tends to decrease the frequency of latently infected cells without noticeably causing a detectable increase in the level of cells undergoing LTR-directed gene expression.Figure 2APG-1387 reduces the frequency of latently infected primary CD4 T cells in vitro(A) Flow cytometry analysis of CD4 T cells infected with the HI.fate.E dual-fluorescent reporter virus, which identifies replicating and latently infected cell populations. HI.fate.E construct contains an E2 Crimson (E2-CRMZ) reporter under the control of the HIV LTR that serves as a marker for cells undergoing productive proviral expression (E2-CRMZ positive) and an EF1α promoter driving constitutive expression of a green fluorescent protein (ZS-Green) reporter. Cells harboring a latent provirus are identified as ZS-green positive/E2-CRMZ negative cells while replicating infected cells are positive for E2-CRMZ and ZS-green. Primary CD4 T cells were transduced with HI.fate.E virus for 3 days and treated with DMSO, PMA/io (positive control), and APG-1387 for 24 h. Frequencies of replicating and latently infected cells were determined by flow cytometry. Efavirenz (EFV) was used to confirm the authenticity of infection.(B) Fold change over vehicle control (VC) in replicating and latently infected cell populations. Each symbol represents a donor (n = 6 donors; M, male; F, female as indicated on the Figure). Results are shown at 24h post-APG-1387 treatment. Data are represented as median. Statistical analysis: non-parametric Friedman test followed by Dunn’s multiple comparisons test ∗p = 0.026, ∗∗p = 0.007, ∗∗∗p = 0.0004 (GraphPad Prism 8.0). Statistical significance results are as indicated on the Figure. All other pairings are not statistically significant. See also Figure S2. APG-1387 reduces the frequency of latently infected primary CD4 T cells in vitro (A) Flow cytometry analysis of CD4 T cells infected with the HI.fate.E dual-fluorescent reporter virus, which identifies replicating and latently infected cell populations. HI.fate.E construct contains an E2 Crimson (E2-CRMZ) reporter under the control of the HIV LTR that serves as a marker for cells undergoing productive proviral expression (E2-CRMZ positive) and an EF1α promoter driving constitutive expression of a green fluorescent protein (ZS-Green) reporter. Cells harboring a latent provirus are identified as ZS-green positive/E2-CRMZ negative cells while replicating infected cells are positive for E2-CRMZ and ZS-green. Primary CD4 T cells were transduced with HI.fate.E virus for 3 days and treated with DMSO, PMA/io (positive control), and APG-1387 for 24 h. Frequencies of replicating and latently infected cells were determined by flow cytometry. Efavirenz (EFV) was used to confirm the authenticity of infection. (B) Fold change over vehicle control (VC) in replicating and latently infected cell populations. Each symbol represents a donor (n = 6 donors; M, male; F, female as indicated on the Figure). Results are shown at 24h post-APG-1387 treatment. Data are represented as median. Statistical analysis: non-parametric Friedman test followed by Dunn’s multiple comparisons test ∗p = 0.026, ∗∗p = 0.007, ∗∗∗p = 0.0004 (GraphPad Prism 8.0). Statistical significance results are as indicated on the Figure. All other pairings are not statistically significant. See also Figure S2. To assess whether APG-1387 could be a potential LRA in vivo, we first performed a pharmacological analysis in uninfected hu-BLT mice to gauge any off-target effects or immune-mediated toxicity. In brief, mice were administered with APG-1387 (20 mg/kg) or vehicle control every third day for 4 weeks (Figure 3A). This dosage was chosen because APG-1387 anti-tumor activity was observed at this concentration in a xenograft mouse model. T cell frequency along with their activation status were monitored in the blood and other tissues before and after treatment (Figure 3B; Figure S4). As indicated, APG-1387 treatment did not lead to notable, global changes in the level of T cells. No increase in the frequency of CD4 T or CD8 T cells expressing the activation marker HLA-DR was observed in tissues of APG-1387-treated mice. Importantly, APG-1387 treatment did not increase the levels of inflammatory cytokines TNF or interleukin 6 (IL6) (Figure 3C). Taken together, these results demonstrate that APG-1387 treatment does not result in overt immunotoxicity in vivo.Figure 3APG-1387 does not cause overt immunotoxicity in hu-BLT mice(A) Experimental setup to assess effects of APG-1387 in hu-BLT mice. Uninfected mice were left untreated (UN) or treated with either vehicle control (VC) or APG-1387 (SM) for 4 weeks. Untreated mice were used to gauge the effect of the vehicle control.(B) Shown are frequencies of HLA-DR cells among CD4 and CD8 T cells (gated on human CD45 cells) before and after APG-1387 treatment. Data were represented as median with range. Statistical analysis: Wilcoxon matched pairs signed rank test for samples in the blood; non-parametric two-tailed Mann-Whitney unpaired rank test for the other tissues; ∗p = 0.0286; ns, not significant. Exceptionally, the % of HLA-DR CD8 T cells in the spleen was statistically significant between untreated (UN) and APG-1387 (SM) treated mice (p = 0.0286), but not between vehicle (VC) and APG-1387 treated animals.(C) Plasma from untreated (UN) mice or those treated with vehicle (VC) or APG-1387 (SM) was analyzed by ELISA for TNF and IL6. In all panels, there were n = 4 mice per experimental group. Data were represented as median with 95% CI. Statistical analysis: non-parametric Friedman test followed by Dunn’s multiple comparisons test (GraphPad Prism 8.0); ns, not significant. See also Figure S4. APG-1387 does not cause overt immunotoxicity in hu-BLT mice (A) Experimental setup to assess effects of APG-1387 in hu-BLT mice. Uninfected mice were left untreated (UN) or treated with either vehicle control (VC) or APG-1387 (SM) for 4 weeks. Untreated mice were used to gauge the effect of the vehicle control. (B) Shown are frequencies of HLA-DR cells among CD4 and CD8 T cells (gated on human CD45 cells) before and after APG-1387 treatment. Data were represented as median with range. Statistical analysis: Wilcoxon matched pairs signed rank test for samples in the blood; non-parametric two-tailed Mann-Whitney unpaired rank test for the other tissues; ∗p = 0.0286; ns, not significant. Exceptionally, the % of HLA-DR CD8 T cells in the spleen was statistically significant between untreated (UN) and APG-1387 (SM) treated mice (p = 0.0286), but not between vehicle (VC) and APG-1387 treated animals. (C) Plasma from untreated (UN) mice or those treated with vehicle (VC) or APG-1387 (SM) was analyzed by ELISA for TNF and IL6. In all panels, there were n = 4 mice per experimental group. Data were represented as median with 95% CI. Statistical analysis: non-parametric Friedman test followed by Dunn’s multiple comparisons test (GraphPad Prism 8.0); ns, not significant. See also Figure S4. Humanized mice (hu-mice) were inoculated with HIV-1 NL4.3-ADA-GFP and at peak viremia, infected mice were treated with subcutaneous ART daily for 6 weeks. To gauge the effect of repeated administration of APG-1387 on HIV RNA and DNA levels in the context of uninterrupted ART, we treated ART-suppressed mice with at least four doses of APG-1387 (or vehicle control) over the course of two weeks (Figure 4A). Here, one mouse in the APG-1387-treated group had a viremia level above the background after the two-week treatment while the rest displayed an undetectable viremia throughout the follow-up (Figure 4B). Of note, when virally suppressed mice received a single dose of APG-1387 upon cessation of ART, plasma viral load became detectable in three out of six animals, suggesting that the activity of APG-1387 as a latency reversal agent is more apparent in this context (Figure S5).Figure 4APG-1387 reduces the frequency of latently infected cells and modestly increases HIV RNA detection in ART-suppressed hu-BLT mice(A) Hu-BLT mice were infected with NL4.3-ADA-GFP virus and virally suppressed with ART for 6 weeks. Thereafter, mice received four doses of either vehicle or APG-1387 in the presence of ART.(B) Plasma viral load in mice receiving vehicle (VC, n = 6, blue lines) or APG-1387 (n = 7, red lines). Pink shading depicts the period of ART administration. The dotted line represents the limit of detection (LOD) of the assay.(C–E) HIV-RNA (C), total HIV DNA (D) and integrated HIV-DNA (E) were analyzed in different tissues by RT-qPCR or qPCR, respectively. Box and whisker graphs indicate median, min and max values. The number on the top of the horizontal line shows the fold change in median levels of HIV RNA or HIV DNA in APG-1387- treated mice over vehicle (VC)-treated mice. Samples were analyzed in technical duplicates and there were five to seven mice per group (n = 5 to 7). For each tissue, the number of mice tested per group was between five to seven except for the bone marrow for the vehicle-treated group when there were three available. Statistical analysis: Two-tailed Mann-Whitney unpaired rank test. ∗p = 0.0317; ∗∗p = 0.0043; ns, not significant (GraphPad Prism 8.0). In (E), p = 0.1255 for the lung. See also Figures S5 and S6. APG-1387 reduces the frequency of latently infected cells and modestly increases HIV RNA detection in ART-suppressed hu-BLT mice (A) Hu-BLT mice were infected with NL4.3-ADA-GFP virus and virally suppressed with ART for 6 weeks. Thereafter, mice received four doses of either vehicle or APG-1387 in the presence of ART. (B) Plasma viral load in mice receiving vehicle (VC, n = 6, blue lines) or APG-1387 (n = 7, red lines). Pink shading depicts the period of ART administration. The dotted line represents the limit of detection (LOD) of the assay. (C–E) HIV-RNA (C), total HIV DNA (D) and integrated HIV-DNA (E) were analyzed in different tissues by RT-qPCR or qPCR, respectively. Box and whisker graphs indicate median, min and max values. The number on the top of the horizontal line shows the fold change in median levels of HIV RNA or HIV DNA in APG-1387- treated mice over vehicle (VC)-treated mice. Samples were analyzed in technical duplicates and there were five to seven mice per group (n = 5 to 7). For each tissue, the number of mice tested per group was between five to seven except for the bone marrow for the vehicle-treated group when there were three available. Statistical analysis: Two-tailed Mann-Whitney unpaired rank test. ∗p = 0.0317; ∗∗p = 0.0043; ns, not significant (GraphPad Prism 8.0). In (E), p = 0.1255 for the lung. See also Figures S5 and S6. We observed that APG-1387 treatment modestly increased the level of cell-associated HIV RNA transcripts in the lung and spleen by up to 1.7- and 2.2-fold, respectively, but not in the bone marrow or the liver (Figure 4C). Regarding the effect of APG-1387 on the level of total HIV DNA-harboring cells, we observed a trend toward a reduction in the spleen, liver and lung but not in the bone marrow (Figure 4D). For the integrated HIV DNA, the difference was noticeably more pronounced in the spleen, liver, and lung of APG-1387-treated mice (Figure 4E), and the reduction reached statistical significance (by two-tailed Mann-Whitney tests) for the spleen (p = 0.0043) and liver (p = 0.0317) tissues, but not for the lung (p = 0.1255). APG-1387 does not markedly affect T cells in uninfected mice (Figure 3B; Figure S4) but augments the level of active CASP3 in latent T cell lines (Figure 1E; Figure S1D). In a dual reporter virus system, it preferentially reduces the frequency of latently infected primary CD4 T cells (Figure 2). Thus, it is tempting to hypothesize that the decreased abundance of integrated HIV DNA-containing cells is due to a potential reduction in the frequency of persistent latent VRs by APG-1387. Previously, it was reported that AZD5582, a bivalent SM, might induce activation and proliferation of T cells. Thus, we evaluated whether APG-1387 altered the expression of marker of proliferation Ki-67 MKI67 or activation markers HLA-DR and CD69 on T cells. Here, we found no significant changes in the level of HLA-DR and CD69 in any of the tissues tested (Figures S6A and S6B). MKI67 expression was enhanced in certain organs, similar to what was reported previously. However, plasma levels of proinflammatory cytokines IL6 and TNF were comparable between the two groups (Figure S6C). We found no differences in the frequency of BCL2 CD3CD8 T cells (i.e., CD4 T lymphocytes) in tissues of mice treated with APG-1387 (vs. vehicle control). However, consistent with earlier data obtained in the latent T cell lines (Figures 1E and S1C), the frequency of CASP3-positive CD8-T cells (i.e., CD4 T cells) was significantly higher in the bone marrow and liver cells of APG-1387-treated mice compared to vehicle-treated animals (Figure S6D). Having observed that APG-1387 reduced the frequency of HIV DNA-containing cells in several tissues in ART-suppressed hu-BLT mice, we next explored how APG-1387 might have affected different immune cell subsets in these animals. Various SMs have been shown to inhibit Th17 function or decrease Treg differentiation, and because both Th17 and Tregs are known to be affected during HIV infection, we assessed whether in vivo APG-1387 treatment could affect these populations. First, we found that ex vivo stimulation of combined PMA and ionomycin greatly improved the detection of IL17A-producing CD4 T (Th17-like cells) in splenocytes of ART-suppressed mice previously treated or not with APG-1387 (Figure 5A). However, cell activation with PMA/ionomycin did not affect the identification of forkhead box P3 (FOXP3)-expressing CD4 T cells (Figure 5B). In vivo treatment with APG-1387 modestly decreased the frequency of splenic Th17-like cells in ART-suppressed mice by 45% (mean 4.5 ± 1.8% in VC-treated group vs. 2.5 ± 1.3% in APG-1387-treated group) but the difference was not statistically significant (p = 0.11), possibly because of the small number of mice and variations between them (Figure 5A). In the case of FOXP3-expressing CD4 T cells, we observed no significant increase (p = 0.41) in the proportion of these cells in the APG-1387 treated group (mean 7.3 ± 1.1% in VC-treated vs. 10.7 ± 4.5% in APG-1387 treated group) (Figure 5B).Figure 5Effects of APG-1387 on Th17-like and FOXP3-expressing CD4 T cells in vivo and ex vivo(A and B) Spleen cells from virally suppressed, vehicle-treated (VC) or APG-1387-treated hu-BLT mice were left unstimulated (Unstim) or stimulated ex vivo with combined PMA and ionomycin (Stim) for up to 24h and analyzed for IL17A-producing (Th17-like) (Panel A) or FOXP3-expressing CD4 T cells (Panel B). Representative flow graphs depicting the two cell subsets (gated on human CD45 CD19CD14CD3 populations) are shown on the top. Bottom panels are summary graphs showing results from multiple mice for each condition. Data are represented as median. Statistical analysis: two-tailed Mann-Whitney tests; ns, not significant (GraphPad Prism 8.0). The p values for the ‘Stim' conditions were 0.11 for (A) and 0.41 for (B). In all relevant panels, each dot is one mouse (n = 5 for vehicle and n = 4 or 5 for APG-1387).(C and D) Splenocytes from HIV-infected, ART-naïve mice were analyzed by flow cytometry for viral Gag p24 in Th-17-like and FOXP3-expressing CD4 T cells. The dot plot flow graphs in Panel C show the gating strategy to identify Th17-like (CD3CD8IL17A) or FOXP3-expressing (CD3CD8FOXP3) from human CD45 CD19CD14 cells in one mouse as an example. The graphs illustrate HIV-1 Gag p24-positive IL17A CD8T cells or p24-positive FOXP3 CD8T cells expressing dim or no CD4. In Panel D, splenocytes isolated from HIV-infected, ART-naive mice (n = 5) were treated with APG-1387 (1 μM) ex vivo for up to 48h and stained for HIV-1 Gag p24. Splenocytes treated with vehicle control (VC) were used as a negative control. Depicted is the p24 cell count in total CD4 T cells, Th17-like and FOXP3-expressing T cells. Statistical analysis: two-tailed Wilcoxon paired rank tests; ns, not significant (GraphPad Prism 8.0). In (D), p = 0.062 for total CD4 T, p > 0.99 for Th17-like cells and p = 0.62 for FOXP3 cells. See also Figure S6. Effects of APG-1387 on Th17-like and FOXP3-expressing CD4 T cells in vivo and ex vivo (A and B) Spleen cells from virally suppressed, vehicle-treated (VC) or APG-1387-treated hu-BLT mice were left unstimulated (Unstim) or stimulated ex vivo with combined PMA and ionomycin (Stim) for up to 24h and analyzed for IL17A-producing (Th17-like) (Panel A) or FOXP3-expressing CD4 T cells (Panel B). Representative flow graphs depicting the two cell subsets (gated on human CD45 CD19CD14CD3 populations) are shown on the top. Bottom panels are summary graphs showing results from multiple mice for each condition. Data are represented as median. Statistical analysis: two-tailed Mann-Whitney tests; ns, not significant (GraphPad Prism 8.0). The p values for the ‘Stim' conditions were 0.11 for (A) and 0.41 for (B). In all relevant panels, each dot is one mouse (n = 5 for vehicle and n = 4 or 5 for APG-1387). (C and D) Splenocytes from HIV-infected, ART-naïve mice were analyzed by flow cytometry for viral Gag p24 in Th-17-like and FOXP3-expressing CD4 T cells. The dot plot flow graphs in Panel C show the gating strategy to identify Th17-like (CD3CD8IL17A) or FOXP3-expressing (CD3CD8FOXP3) from human CD45 CD19CD14 cells in one mouse as an example. The graphs illustrate HIV-1 Gag p24-positive IL17A CD8T cells or p24-positive FOXP3 CD8T cells expressing dim or no CD4. In Panel D, splenocytes isolated from HIV-infected, ART-naive mice (n = 5) were treated with APG-1387 (1 μM) ex vivo for up to 48h and stained for HIV-1 Gag p24. Splenocytes treated with vehicle control (VC) were used as a negative control. Depicted is the p24 cell count in total CD4 T cells, Th17-like and FOXP3-expressing T cells. Statistical analysis: two-tailed Wilcoxon paired rank tests; ns, not significant (GraphPad Prism 8.0). In (D), p = 0.062 for total CD4 T, p > 0.99 for Th17-like cells and p = 0.62 for FOXP3 cells. See also Figure S6. To assess whether APG-1387 targets specifically infected cells that actively express viral proteins, we performed a complementary experiment in which splenocytes from HIV-infected, ART-naïve mice were treated with APG-1387 ex vivo (Figures 5C and 5D). Here, exposure to APG-1387 led to a moderate decrease in the number of p24-expressing total CD4 T cells (p = 0.062) in all mice tested (n = 5). For the Th17-like and FOXP3 CD4 T cell subsets, this trend was observed in 3 of 5 mice for Th17-like CD4 T cells (p > 0.99) and 4 of 5 for the FOXP3 CD4 T cell subset (p = 0.62). Taken together, our data suggest that APG-1387 can reduce the proportion of infected CD4 T cells that are actively producing viral protein(s). Given the effect of APG-1387 on the pool of latently infected cells in previous experiments with the dual reporter virus (Figure 2) and in virally suppressed hu-mice (Figure 4), we sought to perform an ART-treatment interruption (ATI) to assess the impact of APG-1387 on the VRs. To test our hypothesis that mice treated with APG-1387 would display a slower viral rebound upon ATI, we administered four doses of APG-1387 (or vehicle control) alongside ART to virally suppressed mice as shown in Figure 4. Two weeks thereafter, ART was removed and the mice were sacrificed after 4 weeks off ART (Figure 6A). As shown in Figure 6B, plasma viral loads in the three APG-1387-treated mice were at least 66-fold lower than that in the control group two weeks after ART cessation (Table S1). At the end of the experiment, the viremia remained lower in APG-1387-treated mice by more than 6-fold (Table S1). Importantly, a longitudinal analysis of the viremia showed that in APG-1387-treated mice the viral rebound plateaued at a level notably lower than that pre-ART, suggesting that the size of the reservoir was affected. On this note, the data overall support the notion that APG-1387 reduces viral rebound in HIV-infected hu-mice upon ART interruption by decreasing the pool of latently infected cells that persists during ART.Figure 6Sequential treatment of virally suppressed hu-BLT mice with APG-1387 reduced the level of viral rebound upon ART interruption(A) Hu-BLT mice were infected with NL4.3-ADA-GFP and viral replication was suppressed by ART for 6 weeks. Mice were then treated with either APG-1387 or vehicle control for 2 weeks (4 doses) in the presence of ART. These mice were then subjected to a 4-week antiviral treatment interruption (ATI).(B) Viremia of virally suppressed mice treated with VC (n = 3, blue lines) or APG-1387 (n = 3; red lines) was analyzed at regular intervals. Just before the sacrifice (sac) at 20 weeks post infection, we lost one mouse in each group. The orange background denotes the period of ART administration. The dotted line represents the limit of detection (LOD). Statistical analysis: non-parametric Kruskal-Wallis test (p = 0.057) followed by Dunn’s multiple comparison test (p > 0.999), as analyzed using GraphPad Prism 8.0 software. Sequential treatment of virally suppressed hu-BLT mice with APG-1387 reduced the level of viral rebound upon ART interruption (A) Hu-BLT mice were infected with NL4.3-ADA-GFP and viral replication was suppressed by ART for 6 weeks. Mice were then treated with either APG-1387 or vehicle control for 2 weeks (4 doses) in the presence of ART. These mice were then subjected to a 4-week antiviral treatment interruption (ATI). (B) Viremia of virally suppressed mice treated with VC (n = 3, blue lines) or APG-1387 (n = 3; red lines) was analyzed at regular intervals. Just before the sacrifice (sac) at 20 weeks post infection, we lost one mouse in each group. The orange background denotes the period of ART administration. The dotted line represents the limit of detection (LOD). Statistical analysis: non-parametric Kruskal-Wallis test (p = 0.057) followed by Dunn’s multiple comparison test (p > 0.999), as analyzed using GraphPad Prism 8.0 software. Several LRAs have shown their biological activity both in vitro and in vivo but are clinically unsafe for further evaluations. The most potent LRAs, including HDACi and protein kinase C (PRKC) agonist bryostatin-1, activate the canonical NFKB pathway, causing explicit T cell activation and broad cytotoxicity, and hence eliciting significant collateral damage on host cells. Reactivation of latent HIV by PRKC agonists has recently been demonstrated to induce resistance to apoptosis, a phenomenon often associated with phosphorylation and activation of the antiapoptotic protein BCL2. Thus, the failure of common LRAs such as PRKC agonists and HDACi to safely purge HIV and effectively reduce VRs in people living with HIV necessitates the development of clinical approaches to achieve an HIV cure. In this study, we demonstrate that APG-1387, a bivalent SM initially developed as a cancer therapy, can efficiently reactivate HIV in CD4 T cell line models of HIV-1 latency via a process that involves activation of the noncanonical NFKB pathway. Although this IAP antagonist modestly increases HIV RNA detection in virally suppressed hu-mice, it has the capability to reduce both the frequency of latently infected cells and the level of viral rebound upon ART treatment interruption. Indeed, APG-1387 treatment enhances the expression of caspase-3, a marker of apoptosis, in latently infected cells from T cell lines or in T cells from certain tissues of virally suppressed hu-mice. In the context of productive infection, in vitro stimulation with APG-1387 also enhanced cleavage of CASP3. IAPs are overly expressed in various cancers, enabling prolonged survival of cancerous cells [as reviewed by]. Consequently, their antagonists are thought to either directly induce or sensitize cancerous cells to death by triggering proapoptotic pathways. Interestingly, IAPs such as BIRC2 have recently been shown to be negative regulators of HIV transcription, and their expression is found to be correlated with viral latency. Indeed, both BIRC2/3 and XIAP are overexpressed in HIV latently infected CD4 T and myeloid cells, and the SMs which induce the degradation of these IAPs can reactivate HIV. We demonstrate herein that various monovalent and bivalent SMs can induce efficient viral reactivation in different T cell models of latency and that some are more potent than others. Bivalent SMs are more effective than their monovalent counterparts, probably because of the presence of dimers that may contribute to a more stable and enhanced activation via interactions with the two adjacent binding domains of IAPs. Among the bivalent SMs we examined, APG-1387 is most effective at degrading IAPs, facilitating a conversion of NFKB p100 to p52, and reactivating HIV expression through an NIK-dependent process, a hallmark of an activated non-canonical NFKB pathway. In CD4 T cell models of HIV latency, APG-1387 treatment is associated with remarkable viral reactivation. However, in primary CD4 T cells infected with HI.fate.E dual reporter virus, exposure to APG-1387 is accompanied by a reduction in the frequency of latently infected cells (Figure 2) without a detectable change in the level of cells supporting LTR-directed transcription. The data suggest that the latent cells might be preferentially targeted for elimination without reactivation in contrast to PMA and ionomycin stimulation. This said, given the intrinsic properties of SMs, we cannot completely exclude the possibility that there was no change in the frequency of reactivated cells because latently infected cells rapidly die after reactivation. Consistent with previous works with other SMs including AZD5582 and ciapavir, APG-1387 induces detection of viremia, albeit modestly, in ART-suppressed hu-mice as early as 48 h after treatment. Importantly, in mice treated with multiple doses of APG-1387, the proportion of cells carrying the integrated HIV DNA was meaningfully reduced, suggesting that the bivalent APG-1387 might preferentially target latently infected cells directly or indirectly for death (Figure 4). In addition, in APG-1387-treated mice the fact that the viral rebound was consistently lower throughout the ATI and plateaued at a level below pre-ART further strengthens the notion that APG-1387 impacts negatively the pool of VRs (Figure 6). It is conceivable that such an impact is significant considering the modest effect of APG-1387 on latency reversal in this experimental condition and the limited functional immune clearance mechanisms present in hu-BLT mice. This said, further experimentation with a larger group of animals analyzed at endpoint is needed to confirm the modulatory role of APG-1387 on VRs. A more detailed analysis gauging the effect of APG-1387 on different CD4 T cell subsets revealed a potential modulation of splenic Th17-like cells and FOXP3-expressing CD4 T cells. We observe a trend toward a reduction in Th17-cell frequency in both ART-suppressed mice treated in vivo with APG-1387 and in ART-naïve infected mice whose splenocytes were stimulated ex vivo with APG-1387 (Figure 5). Since Th17 cells have been proposed to be an important source of HIV latent reservoirs [reviewed in], a decrease in Th17 cells might suggest a reduction in the level of integrated HIV DNA-harboring cells, an observation that was made with APG-1387 treated mice (Figure 4). Regarding FOXP3-expressing CD4 T cells in the spleen, there was no significant difference between ART-suppressed mice treated with APG-1387 or with the vehicle control. However, ex vivo stimulation of splenocytes from ART-naïve, HIV-infected mice with APG-1387 modestly decreased the number of p24-expressing FOXP3 CD4 T cells, although the difference was not statistically significant. Whether the non-canonical NFKB pathway is more functional in certain memory CD4 T cell subsets remains to be fully elucidated. The fact that the central memory subset has been shown to require strong TCR-mediated signaling for maintenance suggests a more important role of the canonical NFKB pathway in this context [reviewed in]. Taken together, these results highlight the importance of evaluating the potential effects of SMs on various immune subsets known to be susceptible to HIV. Our study shows that APG-1387 has the capability to reactivate HIV, activate markers of apoptosis in latently infected cells, and reduce VRs without causing global T cell activation. However, the findings also underscore the need to combine bivalent SMs with other therapeutics to improve the reactivation and elimination of VRs. Indeed, recent findings have shown that combined panobinostat and pegylated interferon alpha 2 can transform the VR landscape through latency reversal and innate immune activation. Variations in the level of human cell reconstitution among mice, a common occurrence in this small animal model, may have contributed to different levels of effect by APG-1387. In addition, although the bivalent APG-1387 can decrease the frequency of virus-expressing Th17 and FOXP3-expressing CD4 T cells in hu-mice following ex vivo stimulation, the small number of mice per group and intrinsic differences between mice prevented us from reaching an irrefutable conclusion about the impact of APG-1387 on Th17 and FOXP3-expressing CD4 T cells in ART-suppressed hu-mice. In addition, potential differences in APG-1387 levels in different tissues might explain some of the disparities in the results. Also, the availability of the sex of the mice but not the sex of the fetal tissues prevents us from adequately assessing the potential sex-based effects with APG-1387 treatment. Lastly, in the experiments involving the use of CD4 T cells from healthy human donors, we did not have enough males and females studied, thus limiting the generalizability of the research findings. As well, an analysis of the influence of gender, ancestry, and ethnicity on the results of the study could not be established given the anonymity clause stipulated in the consent forms (human participants). Another limitation of the study is the small number of mice available at endpoint to evaluate the modulation of VRs by APG-1387. Although hu-BLT mice remain the gold standard small animal model to conduct HIV cure research, it may not be the right system to investigate the effect of APG-1387 on microglia and long-lived resident macrophages that are derived from precursor cells in the yolk sac and thought to be potential sites of viral reservoirs. In addition, the immune system in hu-BLT mice does not fully recapitulate that of a human, thus limiting our ability to assess how infected cells are cleared in the context of APG-1387 treatment. Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Éric A. Cohen (eric.cohen@ircm.qc.ca). This study did not generate new unique reagents. •Data reported in this paper will be shared by the lead contact upon request.•This paper does not report the original code.•Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Data reported in this paper will be shared by the lead contact upon request. This paper does not report the original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. The following reagents were obtained through the NIH AIDS Reagent Program, Division of AIDS, NIAID, and NIH: ACH-2 cells from Dr. Thomas Folks (Cat # 349–443), TZM-bl cells, ARP-8129, contributed by Dr. John C. Kappes, Dr. Xiaoyun Wu and Tranzyme Inc. and J-Lat 10.6, ARP-9849, contributed by Dr. Eric Verdin. 2D10 was obtained from Dr. Jonathan Karn’s lab (School of Medicine, Case Western Reserve University), and 2D10 NIK knock-out cells were a kind gift from Dr. Sumit K. Chanda (The Scripps Research Institute). We appreciate the technical support of the staff from the IRCM and CHU-Ste Justine Research Center animal facilities. We are grateful for the technical support provided by the IRCM flow cytometry platform. We are thankful to Robert Lodge, Isa Munoz Arias, and Mariana Bego for helpful discussions. We thank Mélanie Laporte and Olga Volodina for assistance with the animal work. Editorial assistance was provided by Ashutosh Pathak, Stephen Gutkin, Ndiya Ogba, and Paul Fletcher from Ascentage Pharma Group Inc. (Rockville, MD). The graphical abstract was created in BioRender. Cai, C. (2025) https://BioRender.com/m63y853. We thank all the blood donors and the IRCM clinic staff for their valuable contributions. This study was supported by the Canadian Institutes of Health Research (CIHR) grants FDN-154324 and HB2-164064 (Canadian HIV Cure Enterprise-CanCURE) to É.A.C. This work was also supported in part by the 10.13039/501100000156Fonds de Recherche du Québec - Santé-supported Réseau SIDA/Maladies Infectieuses. J.J. received postdoctoral fellowships from the CIHR HIV/AIDS priority program and FRQS. M.C.R.-M. is a recipient of a FRQS postdoctoral fellowship. S.B. was awarded a Richard and Edith Strauss postdoctoral fellowship in Medicine from McGill University. É.A.C. is the recipient of the IRCM-Université de Montréal Chair of Excellence in HIV Research. We dedicate this paper to the memory of co-author Dr. Natasha Patey, who passed away after the paper was accepted. Y.Z. is a full-time employee of Ascentage Pharma and an equity shareholder of Ascentage Pharma Group International, the parent company of Ascentage Pharma. All other authors declare no competing interests. We, the authors, have a patent application related to this work. REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesMouse anti-human CD3 (Clone OKT3)BioLegendCat# 317330; RRID: AB_2563507Mouse anti-human CD4 (Clone OKT4)BioLegendCat # 317428; RRID: AB_21186122Mouse anti-human CD8 (Clone SK1)BioLegendCat# 344729; RRID: AB_2564509Mouse anti-human CD8 (Clone SK1)BioLegendCat# 344724; RRID: AB_2562790Mouse anti-human HLA-DR (Clone L243)BioLegendCat#307639; RRID: AB_11219187Mouse anti-human CD14 (Clone M5E2)BioLegendCat# 301833; RRID: AB_11126983Mouse anti-human CD14 (Clone M5E2)BioLegendCat# 301851; RRID: AB_2629575Mouse anti-human CD14 (Clone M5E2)BioLegendCat# 301822; RRID: AB_493747Mouse anti-human IL17A (Clone eBio64DEC17)Thermo Fisher ScientificCat# 25-7179-41; RRID: AB_11042972Mouse anti-human FOXP3 (Clone 236A/E7)Thermo Fisher ScientificCat# 12-4777-41; RRID: AB_1944448Mouse anti-human CD45 (Clone 2D1)BioLegendCat# 368526; RRID: AB_2687377Rat anti-mouse CD45 (Clone 30-F11)BioLegendCat# 103146; RRID: AB_2564003Mouse anti-human CD69 (Clone FN50)BioLegendCat# 310912; RRID: AB_314847Mouse anti-human MKI67 (Clone B56)BD BiosciencesCat# 563757; RRID: AB_2688008Rabbit anti-NFKB2 p100/p52 (18D10, Clone N/A)Cell Signaling TechnologyCat# 3017S; RRID: AB_10697356Rabbit anti-IKBA (Clone 44D4)Cell Signaling TechnologyCat# 4812S; RRID: AB_10694416Rabbit anti-BIRC2 (cIAP1) (Clone EPR4673)AbcamCat# ab108361; RRID: AB_10862855Rabbit anti-BIRC3 (cIAP2) (Clone E40)AbcamCat# ab32059; RRID: AB_726890Anti-actin beta HRP (Clone AC-15)AbcamCat# ab49900; RRID: AB_867494Rabbit anti-active CASP3 (caspase 3) (Clone C92-605)BD BiosciencesCat# 560626; RRID: AB_1727414Mouse anti-human BCL2 (Clone 100)BioLegendCat# 658708; RRID: AB_2563282Mouse anti-human HIV-1 core antigen, KC57-FITC (Clone FH190-1-1)Beckman CoulterCat# 6604665; RRID: AB_1575989Goat anti-rabbit IgG H&L (HRP)AbcamCat# ab205718; RRID: AB_2819160Goat anti-mouse IgG (H + L)Life TechnologiesCat# A-11001; RRID: AB_2534069Bacterial and virus strainsHIV: pNL4.3-ADA-GFPDave et al.N/AHI.fate.E. dual reporter virusRatnapriya et al.N/AChemicals, peptides, and recombinant proteinsAPG-1387Ascentage PharmaN/ACollagenase ISigma-AldrichCat# C0130-1GCollagenase XISigma-AldrichCat# C7657-1GDNase (bovine pancreas)Sigma-AldrichCat# D4513-1VLHyaluronidaseSigma-AldrichCat# H3506-1GBirinapant (TL32711)AdooQ BioscienceCat# A12738-25GDC-0152Cayman ChemicalCat# 17810AT-406Cayman ChemicalCat# 19929LCL-161Cayman ChemicalCat# 22420Cremophor ELMillipore SigmaCat# 238470Polyethylene glycol 400Sigma-AldrichCat# 8074851000Ionomycin calcium ionophoreSTEMCELL TechnologiesCat# 73724PMASigma-AldrichCat# P8139-1MGPHA-LSigma-AldrichCat# 11249738001Human IL2Thermo Fisher ScientificCat# 200-02-100UGRaltegravir potassiumAPIChemCat# AC-2062EmtricitabineAPIChemCat# AC-392TenofovirAPIChemCat# AC-5262EfavirenzSigma-AldrichCat# SML0536-10MGPercollSigma-AldrichCat# GE17-0891-01SuperScript II Reverse Transcriptase 10,000 UThermo Fisher ScientificCat# 18064014Critical commercial assaysCOBAS AmpliPrep/COBAS TaqMan HIV-1 test (Version 2)Rochehttps://diagnostics.roche.com/BD Cytofix/Cytoperm™ Fixation/Permeabilization Solution KitBD BiosciencesCat# 554722BD Perm/Wash BufferBD BiosciencesCat# 554723FOXP3 transcription factor staining buffer setThermo Fisher ScientificCat# 00-5523-00TaqMan Fast Advanced Master MixThermo Fisher ScientificCat# 4444556QIAzol Lysis ReagentQIAGEN SciencesCat# 79306Taq DNA polymerase with Standard Taq BufferNew England BiolabsCat# M0273LLipofectamine 3000Thermo Fisher ScientificCat# L3000015ELISA MAX Standard Set Human TNFBioLegendCat# 430201ELISA MAX Deluxe Set Human IL6BioLegendCat# 430504Experimental models: Cell linesHuman: ACH-2 cells (Sex of cells: Female)NIH AIDS Reagent ProgramCat# ARP-349-443; RRID: CVCL_0138Human: HEK293T (Sex of cells: Female)ATCCATCC Cat# CRL-3216; RRID: CVCL_0063Human: 2D10 (Sex of cells: Male)Pearson et al.N/AHuman: 2D10 NIK knock-out (Sex of cells: Male)Pache et al.N/AHuman: J-Lat 10.6 (Sex of cells: Male)Jordan et al.Cat# ARP-9849; RRID: CVCL_8281Human: Jurkat E6.1 (Sex of cells: Male)ATCCATCC Cat# TIB-152; RRID: CVCL_0367Human: TZM-bl (HeLa-derived)(Sex of cells: Female)NIH AIDS Reagent ProgramCat#ARP-8129; RRID: CVCL_B478Experimental models: Organisms/strainsMouse: NOD-scid IL2RgammanullNOD.Cg-PrkdcIl2rg/SzJThe Jackson Laboratoryhttps://www.jax.org/strain/005557#;RRID: IMSR JAX:005557OligonucleotidesPrimer ULF1: ATG CCA CGT AAG CGA AAC TCT GGG TCT CTC TDG TTA GACVandergeeten et al.N/APrimer UR1: CCA TCT CTC TCC TTC TAG CVandergeeten et al.N/APrimer HCD3OUT5’: ACT GAC ATG GAA CAG GGG AAGVandergeeten et al.N/APrimer HCD3OUT3’: CCA GCT CTG AAG TAG GGA ACA TATVandergeeten et al.N/APrimer Alu1: TCC CAG CTA CTG GGG AGG CTG AGGVandergeeten et al.N/APrimer Alu2: GCC TCC CAA AGT GCT GGG ATT ACA GVandergeeten et al.N/APrimer Lambda T: ATG CCA CGT AAG CGA AAC TVandergeeten et al.N/APrimer UR2: CTG AGG GAT CTC TAG TTA CCVandergeeten et al.N/APrimer HCD3IN5’: GGC TAT CAT TCT TCT TCA AGG TVandergeeten et al.N/APrimer HCD3IN3’: CCT CTC TTC AGC CAT TTA AGT AVandergeeten et al.N/ARecombinant DNApsvCMV-VSV-GLodge et al.N/ASoftware and algorithmsFACS DivaBD Bioscienceshttps://www.bdbiosciences.com/en-us/products/software/instrument-software/bd-facsdiva-softwareFlowJo (Versions 9.9.3 and 10.1)TreeStarhttps://www.flowjo.com/solutions/flowjoChemDrawRevvity Signalshttps://revvitysignals.com/products/research/chemdrawPrism (Version 8.0)GraphPadhttps://www.graphpad.comImageJSchneider et al.https://github.com/imagej/ImageJOtherPrecision count beadsBioLegendCat # 424902 The use of hu-BLT mice as a model to study HIV persistence and test therapeutic interventions was approved by the Center Hospitalier Universitaire (CHU) Sainte-Justine institutional (CER#2126) and the CSSS Jeanne-Mance review boards (Montreal, Canada), and applied in accordance with federal and provincial laws. Human fetal tissues were obtained following written informed consent to participate in this study. The sex of the fetal tissues was not available because it was either unknown to the donors or kept anonymous to experimenters as stipulated in the consent forms. Hu-BLT mice were generated as previously reported. In brief, male NOD-scid IL2Rgammanull (NSG, RRID:IMSR_JAX:005557) mice were purchased from the Jackson Laboratory (Bar Harbor, ME, USA) and housed in pathogen-free environment at CHU Sainte-Justine Research Center. NSG mice were subjected to total body irradiation with a sublethal dose (2.5 Gy) at 6–10 weeks of age. Irradiated mice were implanted with pieces of human fetal thymus fragments under the kidney capsule and received autologous fetal liver CD34 hematopoietic stem cells. Humanized mice were housed in germ-free facilities at CHU Sainte-Justine Research Center and Institut de recherches cliniques de Montréal (IRCM) under 12-h light-dark cycles, at room temperature with food and water ad libitum. All animal experiments were approved by the IRCM Animal Care Committee (IRCM 2018-11) and by the CHU Sainte-Justine Animal Care and Use Committee (CIBPAR#2021–2961) following Good Laboratory Practices for Animal Research. Experimental procedures were performed in the BSL-3 Laboratory at the IRCM and conformed to the relevant regulatory standards in accordance with institutional and national guidelines. Fifty-six mice were used and data reported in this study (Table S2). The humanization age was between 10 and 15 weeks and the extent of human cell reconstitution (i.e., percentage of human CD45 cells in the blood at the start of the experiments) ranged between 24.6% and 97.3%. No randomization was performed; variations in the level of human CD45 cells between mice were evenly distributed among experimental groups. Exact details about the number of mice or tissues tested (sample size) in each experimental group were provided in appropriate Figure legends. No inclusion or exclusion criteria were applied. Peripheral blood from anonymous healthy human donors were obtained after the participants had given written inform consent to participate in the study in accordance with the Declaration of Helsinki under research protocols approved by the IRCM Human Research Ethics Review Board (IRCM 2012-16). As per the terms of the consent forms, the blood donors remained anonymous to experimenters; thus, no information on the identity, age, ancestry, race, ethnicity, socioeconomic status or a combination of these factors could be obtained. Peripheral blood from 7 females and 3 males were used in the study, as described below and indicated on Figures or in the corresponding Figure legends. These 10 healthy subjects were used in three experiments and information related to the sample size and specific details for each experiment could be found in the Figure legends. The absence of sex-based analyses limits the generalizability of some of the findings in this study and this is highlighted under Section ‘Limit of Study’. An analysis of the influence of gender, ancestry, ethnicity on the results of the study could not be performed given the anonymity clause of the consent forms. The HEK 293T cell line (RRID:CVCL_0063), obtained from the American Type Culture Collection (ATCC) is a derivative of the human embryonic kidney (HEK) 293 cell line which was generated from the kidney of an aborted female fetus. The TZM-bl (RRID: CVCL_B478), obtained from the NIH AIDS Reagent Program, was generated from a female with cervical carninoma. The ACH-2 cell line (RRID:CVCL_0138), obtained from the NIH AIDS Reagent Program, was from a four-year-old female with T acute lymphoblastic leukemia. The Jurkat E6.1 cell line (RRID:CVCL_0367), obtained from the ATCC, came from a 14-year-old male with acute T cell leukemia. The 2D10, 2D10 NIK-knockout and J-Lat 10.6 (RRID:CVCL_8281) cell lines were all derived from the parental Jurkat E6.1 cell line mentioned above. All cell lines were cultivated in DMEM (HEK 293T and TZM-bl) or RPMI-1640 media (ACH-2 and all Jurkat-based cell lines) supplemented with 10% fetal bovine serum (FBS) at 37C in incubators supplied with 5% CO2 in air atmosphere. All cell lines were routinely checked in our laboratory and confirmed to be mycoplasma-free. No other information related to cell authentication was available. Primary cells were isolated from blood and tissues of hu-mice (all males) described in the preceding Section entitled ‘Hu-mice’. Cells were used in experiments approved by the IRCM Animal Care Committee (IRCM 2018-11) and by the CHU Saint-Justine Animal Care and Use Committee (CIBPAR#2021–2961). All experimental procedures conformed to the relevant regulatory standards in accordance with the institutional and national guidelines. The sex of human cells in hu-mice was not available since the sex of human fetal tissues used to generate the animals was either unknown to the donors or kept confidential in accordance with the terms of the consent forms. Cells isolated from tissues of hu-mice were cultivated in RPMI-1640 media supplemented with 10% FBS at 37C in incubators supplied with 5% CO2 in air atmosphere. CD4 T cells were purified from peripheral blood of 7 female and 3 male healthy subjects whose blood was drawn after the participants had given written inform consent in accordance with the Declaration of Helsinki under research protocols approved by the IRCM Human Research Ethics Review Board (IRCM 2012-16). CD4 T cells were used in the experiments shown in Figures 2, S2, and S3. Precise usage of the donors was indicated on the Figures or in the corresponding Figure legends. CD4 T cells, grown in RPMI-1640 media supplemented with 10% FBS, were cultivated at 37C in incubators supplied with 5% CO2 in air atmosphere. No randomization and/or stratification was performed. Hu-mice with variable levels of human CD45 cells were equally distributed among experimental groups. All efforts were made to have equal or comparable number of mice or samples (i.e., number of n) per experimental group or condition. The exact number of n was provided in relevant Figure legends. Experimenters were blinded during processing biological samples, conducting experiments and analyzing samples in different assays. No inclusion or exclusion criteria were applied to the study. No statistical methods were used to pre-determine strategies for randomization and/or stratification, population size, inclusion and exclusion of any data or subjects, or whether the data met assumptions of the statistical approach. Hu-BLT mice were left untreated or treated via intraperitoneal route (IP) with either vehicle- (10% sterile Cremophor dissolved in 5% polyethylene glycol-400 and 85% PBS) or 20 mg/kg (100 APG-1387 (Ascentage Pharma, China) every third day (maximum 100 μL injection volume) for up to 4 weeks. Plasma was collected at different intervals and white blood cells isolated by treating whole blood with red blood cell lysis buffer (Invitrogen, U.S.A). Cells from blood and tissues were analyzed by flow cytometry as described in Section ‘flow cytometry’ below for the effect of APG-1387 on cell proliferation and activation. Proinflammatory cytokines were evaluated in plasma of healthy untreated, APG-1387- and vehicle-treated hu-BLT mice using Legend Max enzyme-linked immunosorbent assay (ELISA) kits for human TNF and IL6 (both from BioLegend, U.S.A) as per the manufacturer’s protocol. Data analysis was performed using GraphPad Prism software (Version 8.0). Stocks of HIV-1 NL4.3-ADA-GFP were prepared and titered as previously reported. In brief, HEK293T cells (5x10) were transfected with 25 μg CCR5-tropic HIV pNL4.3-ADA-GFP using the calcium phosphate method. Culture supernatant was collected 48h later and virus was concentrated by ultracentrifugation over a 20% sucrose gradient. The pelleted virus was resuspended in DMEM media and titrated in TZM-bl cells to determine 50% tissue culture infectious units (TCID50). TCID50 values were calculated using the Spearman-Karber method. Hu-BLT mice were inoculated by IP (in 100 μL volume) twice (24-h apart) with 100,000 TCID50 each. Plasma HIV viral load was determined every 1–2 weeks using the quantitative COBAS AmpliPrep/COBAS TaqMan HIV-1 test, Version 2.0 (detection limit, 20 copies/ml; Roche Diagnostics, U.S.A). After viral peak was reached (6–8 weeks post infection), mice were subjected to a daily ART regimen containing emtricitabine (100 mg/kg), tenofovir disoproxil fumarate (50 mg/kg), and raltegravir (68 mg/kg) via subcutaneous injections (maximum volume 120 μL). The three compounds were obtained from APIChem Technology (Zhejiang, China). Subsequently, virally suppressed mice were treated for up to 2 weeks with APG-1387 (20 mg/kg, every third day) or vehicle via the IP route as described in Section ‘Pharmacological and toxicity analysis’ above. In the experiment where mice received more than one doses of APG-1387, ART was maintained during the two-week treatment with APG-1387. In some cases, virally suppressed mice, treated or not with APG-1387, were subjected to a 4-week ATI after APG-1387 treatment. For experiments which aimed to assess the effect of APG-1387 ex vivo, infected mice were not treated with ART or APG-1387 at anytime leading up to the endpoint of the experiment (euthanasia). Hu-BLT mice were euthanized by gas (isoflurane) anesthesia overdose and intracardiac puncture according to institutional protocols approved by the IRCM Animal Care Committee. Euthanasia was done between 4 and 20 weeks after initiation of the experiment. Cardiac perfusion was performed before tissues were harvested using PBS containing 20 IU/mL heparin. Bone marrow cells were isolated from the femur as described. Briefly, femurs were removed and residual tissues excised from the bones before the cavities were flushed with PBS-2% FBS to recover bone marrow cells. Cells from the spleen, lung and liver were isolated as reported previously. In brief, spleen was crushed and passed through a 40-μm cell strainer to obtain single-cell suspensions (BD Biosciences). Lung and liver tissues were digested with a mixture of enzymes (all from Sigma-Aldrich) containing 1350 U collagenase I, 37.5 U collagenase XI, 18 U hyaluronidase and 7.2 U DNase in 2 mL HBSS for 1h at 37°C. Cell suspension was filtered through a sterile 70-μm cell strainer, and cells purified by centrifugation (870 x g) over a 40%–80% Percoll (Sigma-Aldrich) gradients. In all cases, cells were washed with PBS and contaminating red blood cells (RBCs) were removed using RBC lysis buffer (PBS containing 0.8% ammonium chloride). In certain experiments, spleen cells from virally suppressed, vehicle-treated or APG-1387-treated hu-mice were seeded in a 96 U-bottom well plate (0.5–1 million cells) with 200 μL RPMI-1640 media (+10% FBS) and stimulated with PMA (5 ng/mL) and ionomycin (500 ng/mL). Cells treated with vehicle control (DMSO) were used as negative controls. Twenty-four hours later, they were analyzed for IL17A-producing (Th17-like) or FOXP3-expressing CD4 T cells using flow cytometry as described below. Alternatively, spleen cells from HIV-infected, ART-naïve mice were treated APG-1387 (1 μM) or vehicle (DMSO) for 48h and analyzed by flow cytometry as described below for the frequency of total CD4 T-, Th17-like- or FOXP3 CD4 T-expressing the HIV Gag p24 protein. HEK293T cells (5×10) were transfected with 20 μg of HI.fate.E dual reporter proviral DNA and pseudotyped with 8 μg psvCMV-VSV-G, using Lipofectamine 3000 Transfection Reagent as per manufacturer’s instructions (Thermo Fisher Scientific). The HI.fate.E dual reporter lentiviral vectors were concentrated by ultracentrifugation over a 20% sucrose cushion and titered in Jurkat E6.1 cells. The proportion of infected cells supporting viral replication, designated replicating (E2-CRMZ-positive cells) was used to estimate infectious units. CD4 T cells were isolated from peripheral blood mononuclear cells using CD4 T cell isolation kit (Miltenyi Biotec, USA) and activated with phytohemagglutinin-L (10 μg/mL, Sigma-Aldrich) in the presence of IL2 (100 U/ml, Peprotech, USA). Thereafter, cells were transduced with HI.fate.E dual reporter virus at multiplicity of infection-1 (MOI-1). To confirm the authenticity of transduction, CD4 T cells were treated with 5 μM efavirenz (EFV, Sigma-Aldrich) for 2 h before viral inoculation. At 72 h post transduction, cells were treated with varying concentrations of APG-1387 (0.1–10 μM) or vehicle (DMSO) for 24 h. Transduced primary CD4 T cells treated with PMA/ionomycin, which activates the canonical NFKB pathway, were used as controls. Frequencies of cells supporting viral replication (replicating) and latently infected CD4 T cells were determined by flow cytometry. Jurkat-based cell lines J-Lat 10.6 , 2D10 and 2D10 NIK knock-out were seeded at 20,000 cells (100 μL) in a 96-well plate and treated for up to 48 h with APG-1387 or other SMs including GDC-0152, AT-406, LCL-161 and Birinapant from as low as 0.01 pM up to 10 μM. Cells treated with vehicle (DMSO) were used as controls. Samples were fixed and acquired on SA3800 Spectral Analyzer (Sony Biotechnology) or BD LSRFortessa Cell Analyzer with FACS Diva software (BD Bioscience). These Jurkat-based cell lines contain an HIV-1 genome that expresses GFP upon activation. Thus, reactivation of CD4 T cell models of latency was determined by measuring the percentage of GFP cells after treatment. Results were analyzed using FlowJo (Versions 9.9.3 and 10.1; FlowJo LLC, BD Life Sciences). Cells from blood and/or tissues were stained with fluorescently labeled Abs against mouse CD45 along with those specific for human CD45, CD3, CD4, CD8, CD14, CD69 and HLA-DR as required for the experiments. When appropriate, surface-stained cells were fixed and permeabilized using the BD Cytofix/Cytoperm solution kit and BD Perm/Wash buffer (BD Biosciences) as per the manufacturer’s instructions and intracellularly stained for HIV-Gag p24 using anti-human HIV-1 core antigen, KC57-FITC (Beckman Coulter, USA), and/or cytoplasmic proteins of interests using appropriate Abs (e.g., anti-active CASP3, anti-human BCL2). In some cases, permeabilization was done using a FOXP3/Transcription Factor Staining kit (eBioscience, USA) to stain for ILT17 using anti-human IL17A Ab or nuclear-like proteins including FOXP3 and marker of proliferation marker Ki-67, MKI67 using anti-human FOXP3 and anti-human MKI67 Abs, respectively. The full Ab list used for flow cytometry analysis can be found in the key resources table. When appropriate, precision counting beads (BioLegend, USA) were used to determine the cell count. Samples were acquired on a BD LSRFortessa Cell Analyzer (BD Bioscience) using FACSDiva software and analyzed by FlowJo (Versions 9.9.3 and 10.1). Cells were washed with phosphate-buffered saline (PBS) and then lysed using radioimmunoprecipitation assay RIPA buffer (1% NP-40, 140 mM NaCl, 0.05% SDS, 5 g/L Na-Deoxycholate, 8 mM Na2HPO4 and 2 mM NaH2PO4, pH 7.2) for 20 min at 4°C. Immunoblotting was performed according to standard protocols. Primary Abs such as α-NFKB2 p100/p52, α-IKBA, α-BIRC2, α-BIRC3 and α-ACTB, were added to protein immunoblots and incubated overnight at 4 °C. Protein signals were detected with appropriate secondary Abs including an anti-rabbit Ab. Quantification of protein signals on Western blots was done using ImageJ software. Information about the Abs used in Western blotting can be found in key resources table. RNA from all lymphoid and non-lymphoid tissues were extracted using QIAzol Lysis Reagent (QIAGEN Sciences, USA) according to manufacturer’s instructions and analyzed using previously published protocols. RNA was reverse transcribed using SuperScript II RT (Thermo Fisher Scientific, U.S.A). Total DNA from tissue-derived cells was extracted using the QIAamp DNA mini kit (QIAGEN Sciences, U.S.A) according to manufacturer’s instructions. HIV DNA was quantitated as described previously. In brief, preamplification of total DNA was done using 4 primers (300 nM each) ULF1, UR1, HCD3OUT5′ and HCD3OUT3′, while that for integrated DNA was done using ULF1 (150 nM) and 300 nM each of Alu1, Alu2, HCD3OUT5′ and HCD3OUT3’. The second round PCR were performed in real time using ViiA 7 Real-Time PCR system using TaqMan Fast Advanced Master Mix (Thermo Fisher Scientific, U.S.A). The primers are Lambda T and UR2 (both at 1250 nM) for total and integrated HIV DNA. Primers HCD3IN5′ and HCD3IN3′ are for the amplification of the CD3 gene (2 copies per cell) and used to determine the exact number of cells in the reaction. DNA from serially diluted ACH-2 cells (NIH HIV Reagent Program) was extracted and amplified in parallel to generate a standard curve from which unknown samples were enumerated. Flow cytometry data were analyzed using FlowJo (Versions 9.9.3 and 10.1). Quantification of Western blots was performed using ImageJ software. Data analysis and presentation was done using GraphPad Prism (Version 8.0). Experimenters were blinded during data analysis of samples. Descriptive measures (mean, median, minimum/maximum range, 95% confidence intervals, and percent) were used to summarize the data and illustrate in graphical presentations. All statistical analysis was done using GraphPad Prism (Version 8.0). Nonparametric (unpaired) Mann-Whitney’s U-test (two-tailed) was conducted to compare ranks between two experimental groups (e.g., treated with vehicle or with APG-1387). Nonparametric (paired) Wilcoxon test was performed to compare the ranks of two matched samples (e.g., before and after treatment with APG-1387). When comparing multiple groups, non-parametric tests Kruskal-Wallis or Friedman were performed and followed by Dunn’s multiple comparison test (e.g., vehicle-treated group compared to those treated with different concentrations of APG-1387 or combined PMA and ionomycin; effect of APG-1387 on viral rebound at different time points post treatment interruption). A p value of less than 0.05 was considered statistically significant. ns, ∗, ∗∗, ∗∗∗, signify not significant, <0.05, <0.01, and <0.001, respectively. Accordingly, in the non-parametric Friedman test followed by Dunn’s multiple comparisons test shown in Figure 2, the exact p values were ∗p = 0.026, ∗∗p = 0.007, ∗∗∗p = 0.0004. In the two-tailed Mann-Whitney unpaired rank test shown in Figure 3, the ∗p value was 0.0286 while in Figure 4, the p values were as follows: ∗p = 0.0317 and ∗∗p = 0.0043. For Figure S6, the ∗p value was <0.05. No statistical methods were used to pre-determine strategies for randomization and/or stratification, population size, inclusion and exclusion of any data or subjects, or whether the data met assumptions of the statistical approach. All software used in data analysis along with statistical parameters were mentioned in the appropriate Figure legends. Details about the statistical tests, exact values of n and definitions of the n were as indicated in the legend for each relevant Figure. When applicable, definition of asterisks and descriptive measures were indicated in the Figure legends or Results. Drawing of the chemical structure of APG-1387 shown in Figure 1 was done using the ChemDraw software.
PMC12479362
A single-cell and spatial genomics atlas of human skin fibroblasts reveals shared disease-related fibroblast subtypes across tissues
Fibroblasts sculpt the architecture and cellular microenvironments of various tissues. Here we constructed a spatially resolved atlas of human skin fibroblasts from healthy skin and 23 skin diseases, with comparison to 14 cross-tissue diseases. We define six major skin fibroblast subtypes in health and three that are disease-specific. We characterize two fibroblast subtypes further as they are conserved across tissues and are immune-related. The first, F3: fibroblastic reticular cell-like fibroblast (CCL19CD74HLA-DRA), is a fibroblastic reticular cell-like subtype that is predicted to maintain the superficial perivascular immune niche. The second, F6: inflammatory myofibroblasts (IL11MMP1CXCL8IL7R), characterizes early human skin wounds, inflammatory diseases with scarring risk and cancer. F6: inflammatory myofibroblasts were predicted to recruit neutrophils, monocytes and B cells across multiple human tissues. Our study provides a harmonized nomenclature for skin fibroblasts in health and disease, contextualized with cross-tissue findings and clinical skin disease profiles.Fibroblasts are crucial cells for shaping tissue architecture and immune cell niches. Studying the heterogeneity of fibroblast subtypes has been challenging due to the scarcity of unique surface markers and their tendency to adopt activated phenotypes during in vitro culture. Single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics technologies have overcome these challenges, enabling the dissection of fibroblast heterogeneity in human tissues. While recent studies have described fibroblast states in human skin, they have not spatially resolved their tissue microanatomical location. Very few, if any, have interrogated fibroblasts in diverse disease conditions in the skin and across human tissues. Consequently, the fibroblast composition and function in human skin; how it changes across a range of diseases (inflammatory, cancer and fibrosis/scarring); and how these populations relate to other human tissues is still unclear. In this study, we integrated published large-scale scRNA-seq datasets of healthy human skin and 23 skin diseases and generated spatial transcriptomics data from two different modalities to construct a high-resolution spatially resolved atlas of more than 350,000 adult human skin fibroblasts. We provide a consensus annotation of skin fibroblasts based on gene expression profiles and spatial locations, and contextualize these findings with fibroblast data from other healthy and diseased human tissues. Our scRNA-seq and spatial datasets resources are freely available for download and interactive data exploration at https://cellatlas.io/studies/skin-fibroblast. We re-processed and integrated 2.1 million cells from scRNA-seq data of adult human skin, comprising 32 datasets and 251 donors (Fig. 1a and Supplementary Table 1) using single-cell variational inference (scVI) (Methods). After quality control, 357,276 high-quality fibroblasts were selected based on canonical marker gene expression (Fig. 1a and Extended Data Fig. 1a).Fig. 1Identification of fibroblast subtypes in healthy skin.a, Overview of study methodology, including skin atlas integration to delineate fibroblasts, construction of a healthy/nonlesional reference, mapping of 23 diseases to the healthy reference atlas and downstream analysis for cross-tissue comparison. b, Uniform Manifold Approximation and Projection (UMAP) of healthy and nonlesional skin fibroblasts colored by fibroblast subtype. DS, dermal sheath; DP, dermal papilla. c, Dotplot of marker gene expression for healthy fibroblasts. ‘All’ indicates a marker for a general population, but which contains subtypes. Supplementary Data Fig. 1a provides additional differentially expressed genes for fibroblast subtypes. d, Summary of skin fibroblast subtypes in healthy steady-state tissue. Illustrations in a and d were partly created using BioRender.com. a, Overview of study methodology, including skin atlas integration to delineate fibroblasts, construction of a healthy/nonlesional reference, mapping of 23 diseases to the healthy reference atlas and downstream analysis for cross-tissue comparison. b, Uniform Manifold Approximation and Projection (UMAP) of healthy and nonlesional skin fibroblasts colored by fibroblast subtype. DS, dermal sheath; DP, dermal papilla. c, Dotplot of marker gene expression for healthy fibroblasts. ‘All’ indicates a marker for a general population, but which contains subtypes. Supplementary Data Fig. 1a provides additional differentially expressed genes for fibroblast subtypes. d, Summary of skin fibroblast subtypes in healthy steady-state tissue. Illustrations in a and d were partly created using BioRender.com. In healthy skin, we identified six major fibroblast subtypes based on differential gene expression (Supplementary Data Fig. 1a and Supplementary Table 2) and pathway enrichment analysis (Extended Data Fig. 2a and Methods). The six fibroblast subtypes were observed across different covariates (Extended Data Fig. 1b–g and Supplementary Note 1). Complementary spatial transcriptomic methods validated the presence of each of the six fibroblast subtypes and revealed their distinct microanatomical locations (Fig. 2a–c, Extended Data Figs. 3 and 4 and Supplementary Fig. 2).Fig. 2Skin fibroblasts occupy unique spatial and functional niches.a, Spatial location of fibroblast subtypes in microenvironments (cell2location abundance predictions (10x Genomics Visium)) in a single section of healthy human skin (left). Histopathological annotation of tissue microenvironments (right). b, Spatial location of fibroblasts at single-cell resolution (10x Genomics Xenium 5000-gene panel) for skin sections from nonlesional skin of atopic dermatitis (noninflamed (left) and noninflamed post-treatment (right)), colored by cell type. c, Summary of fibroblast niches: Xenium cell types overlying H&E-stained image (manual approximations). a, Spatial location of fibroblast subtypes in microenvironments (cell2location abundance predictions (10x Genomics Visium)) in a single section of healthy human skin (left). Histopathological annotation of tissue microenvironments (right). b, Spatial location of fibroblasts at single-cell resolution (10x Genomics Xenium 5000-gene panel) for skin sections from nonlesional skin of atopic dermatitis (noninflamed (left) and noninflamed post-treatment (right)), colored by cell type. c, Summary of fibroblast niches: Xenium cell types overlying H&E-stained image (manual approximations). Two of the six fibroblast populations (F1: superficial (papillary) and F2: universal (reticular)) were uniformly present throughout skin at different tissue depths. F1: superficial (papillary) fibroblasts localized adjacent to the skin epithelium in the papillary dermis (Fig. 2b,c) and expressed genes encoding superficial dermal collagens (COL13A1, COL18A1 and COL23A1) and Wnt signaling inhibitors (APCDD1, WIF1 and NKD2) (Fig. 1c). A Wnt-mediated synergistic interplay between superficial dermal fibroblasts and basal epithelial cells has been reported to reciprocally maintain cellular identity. F2: universal (reticular) fibroblasts were located deeper in the skin, interspersed between large collagen fibers in the reticular dermis (Fig. 2b,c). This population was characterized by high expression of marker genes of universal PI16 fibroblasts (PI16, CD34 and MFAP5), a fibroblast subtype found in many human tissues and postulated to represent a precursor fibroblast cell state. Transcription factor activity inference identified KLF5 in F2: universal fibroblasts (Extended Data Fig. 2b), which has been reported to drive the universal Pi16 state. As fascial fibroblasts (F_Fascia) are proposed as a potential progenitor cell in mouse skin, we included these cells in an additional integration, identifying that F_Fascia formed a subset of F2: universal (Extended Data Fig. 1i,j and Supplementary Note 2). The remaining fibroblast subsets were more focal in localization, being associated with vascular or adnexal structures. We thus used hematoxylin and eosin (H&E) staining to illustrate these microenvironments. F3: fibroblastic reticular cell (FRC)-like fibroblasts were located predominantly in the superficial perivascular region in proximity to immune cells (Fig. 2b and Extended Data Figs. 3a,b and 4a,b). F3: FRC-like fibroblasts transcriptomically resembled FRCs, which are specialized fibroblasts found in lymphoid organs/structures that maintain immune niches (Extended Data Fig. 1h), expressing genes that attract and compartmentalize immune cells (CCL19, CXCL12 and CH25H), maintain immune cell survival and function (IL33, IL15, TNFSF13B and VCAM1) and enable antigen presentation (CD74 and major histocompatibility complex (MHC)-II molecules). F2/3: perivascular fibroblasts also localized with immune cells but, unlike F3: FRC-like fibroblasts, were additionally enriched in deep perivascular regions and other sites (Fig. 2a–c and Extended Data Fig. 4b,c). A fraction of F2/3: perivascular fibroblasts showed elevated expression of PPARG (Fig. 1c) and pathway analysis suggested a role in adipocyte differentiation (Extended Data Fig. 2a). The capability to differentiate into adipocytes is characteristic of the reticular fibroblast (equivalent to F2: universal) lineage. F2/3: perivascular fibroblasts shared select gene expression with both F2: universal and F3: FRC-like fibroblasts (Fig. 3c).Fig. 3Prototype meta-learning to identify disease-adapted and disease-specific populations.a, Overview of reference-mapping approach used for integration, where lesional/diseased data were mapped using a pretrained model. b, UMAP of scPoli embeddings colored by predicted cell-type labels and relabeled populations after re-clustering. c, Dotplot of marker gene expression for disease-adapted and disease-specific fibroblast populations. Supplementary Data Fig. 1c provides additional differentially expressed genes for disease-associated fibroblast subtypes. d, Density of cells in embedding by site status. e, Gene expression in F1 and F3 fibroblasts from health and disease, including differentially expressed genes in lesional/diseased states. f, Summary of disease-adapted and disease-specific populations. g, Feature maps for genes associated with myofibroblast subtypes. Color bars indicate expression (log1P norm). h, PROGENy pathway scores for fibroblasts from lesional and healthy samples. Illustrations in f were partly created using BioRender.com. a, Overview of reference-mapping approach used for integration, where lesional/diseased data were mapped using a pretrained model. b, UMAP of scPoli embeddings colored by predicted cell-type labels and relabeled populations after re-clustering. c, Dotplot of marker gene expression for disease-adapted and disease-specific fibroblast populations. Supplementary Data Fig. 1c provides additional differentially expressed genes for disease-associated fibroblast subtypes. d, Density of cells in embedding by site status. e, Gene expression in F1 and F3 fibroblasts from health and disease, including differentially expressed genes in lesional/diseased states. f, Summary of disease-adapted and disease-specific populations. g, Feature maps for genes associated with myofibroblast subtypes. Color bars indicate expression (log1P norm). h, PROGENy pathway scores for fibroblasts from lesional and healthy samples. Illustrations in f were partly created using BioRender.com. F4: hair follicle-associated fibroblasts (ASPNCOL11A1) encompassed three subclusters that were associated with specific regions of the hair follicle (Fig. 2b,c). The first is a well-characterized dermal sheath (DS) population (F4: DS_DPEP1) that wraps around the lower/mid hair follicle (Fig. 2b). The second is a novel F4: TNNCOCH subtype, expressing tendon-associated genes (MKX and TNMD) and observed at the isthmus (mid-hair shaft) (Fig. 2b and Extended Data Fig. 4d). The third F4: DP_HHIP subtype uniquely expressed dermal papilla marker genes (CORIN, HHIP, RSPO3 and LEF1). F5: Schwann-like fibroblasts (SCN7A, FMO2, FGFBP2 and OLFML2A) contained two subclusters (F5: NGFR and F5: RAMP1) (Extended Data Fig. 1k). F5: RAMP1 fibroblasts were enriched near innervated eccrine glands and expressed genes encoding the receptor complex for the neuropeptide CGRP (Fig. 2b,c and Extended Data Figs. 1l, 3c,d and 4c), suggesting a possible interface with the nervous system. F5: NGFR colocalized with Schwann cells, suggesting that they are a nerve-associated population. Fibroblasts have been described in the endoneurium and perineurium of nerve fibers from imaging studies, and ‘Schwann-like fibroblasts’ have recently been reported in human skin scRNA-seq data. We confirmed that our six fibroblast subtypes were distinct from Schwann cells and pericytes (Extended Data Fig. 1j,k and Supplementary Note 2). In addition, we harmonized our skin fibroblast annotation with a previous classification (Supplementary Data Fig. 1b). Overall, we provide a new framework for healthy human skin fibroblast annotation based on gene expression profiles (Fig. 1) and spatial location (Fig. 2) that integrates previous fibroblast descriptions in skin and across tissues. Our findings of transcriptionally defined fibroblast subtypes in distinct microanatomical locations suggest a role for regional fibroblasts in supporting distinct niche functions. We next sought to identify how fibroblast states change in diseased skin. We used scPoli, a deep-learning model for integration and identification of novel cell states in single-cell transcriptome data (Methods) (Fig. 3a). We mapped fibroblasts from skin diseases to our healthy/nonlesional F1–F5 fibroblast reference. Out of 190,756 fibroblasts from diseased states, 121,167 diseased cells were confidently assigned existing F1–F5 cell labels (Extended Data Fig. 5a,b). The remaining 69,589 fibroblasts from the disease data were classified as uncertain (unlabeled) by scPoli (Fig. 3b). Manual annotation based on differential gene expression (Supplementary Data Fig. 1c and Supplementary Table 3) and pathway analysis (Extended Data Fig. 5c) revealed two ‘disease-adapted’ and three ‘disease-specific’ fibroblast subtypes (Fig. 3b–f). ‘Disease-adapted’ fibroblasts resembled a healthy fibroblast subtype counterpart (Fig. 3e) and were expanded in disease settings (Fig. 3d). The first disease-adapted fibroblast subtype resembled F1: superficial fibroblasts in healthy skin (Fig. 3e). The F1-like disease population upregulated genes suggestive of regenerative function (CRABP1, CYP26B1 and WNT5A). CRABP1 and CYP26B1 are markers of superficial/upper wound fibroblasts in mice, which are thought to be the source of wound-induced hair follicle neogenesis, and involved in retinoic acid degradation. CRABP1 fibroblasts are also associated with regeneration in reindeer skin and early-gestational human skin. The second disease-adapted fibroblast subtype resembled F3: FRC-like fibroblasts and upregulated CXCL9 and/or ADAMDEC1 (Fig. 3e). CXCL9 is a chemoattractant for CXCR3 cells and has been reported as an activation marker for FRCs in lymphoid tissues. ‘Disease-specific’ fibroblasts (F6: inflammatory myofibroblasts, F7: myofibroblasts and F8: fascia-like myofibroblasts) did not have a healthy skin fibroblast counterpart and highly expressed a myofibroblast gene signature. This myofibroblast signature included contractility (ACTA2), extracellular matrix (ECM) (COL3A1, COL5A1, COL8A1, POSTN and CTHRC1) and other myofibroblast-associated genes (LRRC15, SFRP4, ASPN, RUNX2 and SCX) (Fig. 3c,g and Extended Data Fig. 5d,e). F6: inflammatory myofibroblasts additionally expressed immune-related genes such as interleukins (IL11 and IL24), chemokines (CXCL5, CXCL8, CXCL13 and CCL11) and matrix metalloproteinases that can remodel tissue to facilitate immune cell infiltration (MMP1) (Fig. 3c). JAK–STAT and hypoxic signaling genes were also elevated (Fig. 3h). F7: myofibroblasts and F8: fascia-like myofibroblasts were distinguished by a higher expression of ECM and TGFβ signaling genes, as well as the mechanotransducer PIEZO2 (Fig. 3c,h). F8: fascia-like myofibroblasts were distinguished by expression of F_Fascia-associated genes (Fig. 3c). Overall, our results indicate that healthy fibroblasts can acquire a regenerative phenotype in F1: superficial fibroblasts (CRABP1CYP27B1), a distinct polarization in F3: FRC-like fibroblasts (CXCL9/ADAMDEC1) and potentially give rise to myofibroblast states (ACTA2COL8A1SFRP4) in diseased skin. We next leveraged the diverse clinical profiles of skin diseases to assess whether fibroblast subtypes provide molecular insights into disease endotypes with respect to scarring. We assigned the 23 skin diseases into three clinically determined risk of scarring groups: low scarring risk, moderate scarring risk, and established scarring/fibrosis (see Methods) (Fig. 4a). We excluded neurofibroma from this analysis as it was the only case of benign neoplasia, consisting primarily of F5: Schwann-like and F2/3: perivascular fibroblasts (Extended Data Fig. 6a).Fig. 4Fibroblast compositional signatures characterize the stroma of distinct skin diseases and scarring risk categories.a, Proportion of fibroblast populations by individual disease. Labels overlying each bar indicate the disease category. b, Proportion of disease-adapted and disease-specific fibroblast subtypes by disease category (mean ± s.e.m.). Scarring risk group was based on clinical profiles (Methods). c, Immunofluorescence of LRRC15 (green) and ADAM12 (magenta) showing myofibroblast populations only in inflamed hidradenitis suppurativa skin (right-most) (from two representative atopic dermatitis and hidradenitis suppurativa inflamed and noninflamed samples). Scale bar, 100 µm. d, Xenium 5k data for lesional/inflamed atopic dermatitis skin, with cells colored by cell type. e, Xenium 5k data for cutaneous melanoma, with cells colored by cell type. f, Proportion of fibroblast populations by disease status for Xenium 5k data. g, Gene module scores for each disease-associated fibroblast subtype across diseases with row normalization (0–1). VE, vascular endothelium; SCLE, subacute cutaneous lupus erythematosus; DLE, discoid lupus erythematosus. a, Proportion of fibroblast populations by individual disease. Labels overlying each bar indicate the disease category. b, Proportion of disease-adapted and disease-specific fibroblast subtypes by disease category (mean ± s.e.m.). Scarring risk group was based on clinical profiles (Methods). c, Immunofluorescence of LRRC15 (green) and ADAM12 (magenta) showing myofibroblast populations only in inflamed hidradenitis suppurativa skin (right-most) (from two representative atopic dermatitis and hidradenitis suppurativa inflamed and noninflamed samples). Scale bar, 100 µm. d, Xenium 5k data for lesional/inflamed atopic dermatitis skin, with cells colored by cell type. e, Xenium 5k data for cutaneous melanoma, with cells colored by cell type. f, Proportion of fibroblast populations by disease status for Xenium 5k data. g, Gene module scores for each disease-associated fibroblast subtype across diseases with row normalization (0–1). VE, vascular endothelium; SCLE, subacute cutaneous lupus erythematosus; DLE, discoid lupus erythematosus. We identified distinct fibroblast compositions for each scarring risk category (Fig. 4b). Low scarring risk diseases were characterized by a high prevalence of F1: superficial (CRABPCYP27B1) and F3: FRC-like fibroblasts (CXCL9/ADAMDEC1) (Fig. 4b), without notable F6–F8 myofibroblast populations. This finding agrees with the regenerative-associated gene profile of disease-associated F1: superficial fibroblasts and a role for F3: FRC-like fibroblasts in maintaining immune niches. Diseases with scarring risk were characterized by a uniquely high prevalence of F6: inflammatory myofibroblasts, which was not observed in low scarring risk or established fibrosis (Fig. 4b). F7: myofibroblasts were observed at a similar prevalence in diseases with scarring risk and established fibrosis. These data point toward F6: inflammatory myofibroblast as a population influencing scarring risk, but which are largely absent in established fibrosis. F8: fascia-like myofibroblasts were also elevated in established fibrosis but were predominantly observed in Dupuytren contracture, a fibroproliferative disease of the palmar fascia (Fig. 4a). We used two further approaches to demonstrate the role for distinct fibroblast subtypes predicting scarring risk. First, we trained a random forest classifier and identified that F6: inflammatory myofibroblasts and F7: myofibroblasts were the most important fibroblast subtypes for predicting scarring risk category (Extended Data Fig. 6b). Second, we profiled a well-recognized myofibroblast marker (LRRC15) at the protein level. LRRC15 was evident in inflammation with scarring risk (inflamed hidradenitis suppurativa skin) but not in noninflamed skin or inflamed skin without scarring risk (atopic dermatitis skin) (Fig. 4c). Having established that disease-associated fibroblasts are enriched in distinct scarring categories, we next used spatial transcriptomics to validate these fibroblast populations in distinct scarring risk stroma (Fig. 4d–f and Supplementary Fig. 3). In keeping with scRNA-seq data (Fig. 4a), F3: FRC-like fibroblasts were expanded in inflamed atopic dermatitis skin (low risk), without major myofibroblasts (Fig. 4d,f and Extended Data Fig. 6c). We localized the F3: FRC-like population to the superficial perivascular immune niche (Fig. 4d), which we further validated using 10x Visium data (Extended Data Fig. 6d–f). In melanoma (scarring risk), aside from F1, the entire stroma comprised F6: inflammatory myofibroblasts and F7: myofibroblasts (Fig. 4e,f and Extended Data Fig. 6g). F7: myofibroblasts showed a matrix-producing phenotype (COL1A1, COL3A1 and POSTN) that characterizes myofibroblastic cancer-associated fibroblasts (CAFs) (myoCAFs). F6: inflammatory myofibroblasts demonstrated high expression of inflammatory CAF (iCAF) marker genes (MMP1, MMP3, CXCL8 and IL24), which was observed in both cancer and inflammatory diseases with scarring risk (Extended Data Fig. 6h,i). Finally, to complement our analysis of fibroblast proportions by disease, we assessed transcriptomic variability of disease-associated fibroblast subtypes by calculating gene module scores for each disease using defined marker genes to define transcriptomic variability across different disease conditions (Fig. 4g and Methods). The F6: inflammatory myofibroblast signature score was highest in hidradenitis suppurativa, acne and keratinocytic skin cancers. Overall, our findings support distinct stromal composition in skin diseases associated with differential scarring risk. F6: inflammatory myofibroblasts were observed in diseases with scarring risk but relatively infrequently observed in established fibrosis, raising the possibility that they may be an intermediate differentiation state toward F7: myofibroblasts. The differentiation process of healthy fibroblasts into myofibroblasts remains poorly understood in human tissues despite its clinical relevance. Fibroblasts are tissue resident, and thus intermediate states of myofibroblast differentiation are likely to be captured in the molecular snapshots of skin diseases analyzed. We therefore performed trajectory analysis of fibroblasts in diseased skin to gain further insights into myofibroblast differentiation, before utilizing time-resolved human wound data as a validation of dynamic changes in stromal composition. We first included all fibroblast subtypes in a partition-based graph abstraction (PAGA) analysis (Extended Data Fig. 7a), and then focused further analyses on fibroblast populations found across diseases on hair-bearing and hairless skin (Methods). F7: myofibroblasts were a terminally differentiated myofibroblast state (Fig. 5a–c), consistent with their presence in established fibrosis. We observed two potential sources for F7: myofibroblasts in skin across analyses (Fig. 5b,c and Extended Data Fig. 7b). One trajectory arose directly from the F2: universal lineage. A second trajectory originated from F1: superficial fibroblasts, transitioning to F7: myofibroblasts via an intermediate F6: inflammatory myofibroblast state (Fig. 5a–c and Extended Data Fig. 7b). These two inferred trajectories are consistent with in vivo lineage tracing studies in mice.Fig. 5Origin of skin disease-specific fibroblast subtypes.a–c, Velocity pseudotime (a), directed PAGA overlaid on UMAP (b) and velocity kernel from CellRank2 for lesional fibroblasts (c). For further details see Methods. d, UMAP visualization of fibroblast subtypes from human skin wounds data colored by cell type (left) and MKI67 (encodes Ki-67) expression (bottom right). Proportions of fibroblast populations by time point (top right), where each bar represents a donor at a given time point. e, Schematic of predicted trajectories. Dashed arrows indicate predictions with multiple lines of evidence. Fibroblast populations are colored by the predominant scarring/fibrosis risk observed in an earlier analysis: green (prevalent in low-risk scarring stroma), orange (prevalent in scarring risk stroma and cancer), red (prevalent in established scarring/fibrotic disorders). Gray boxes indicate signaling pathways identified in our gene expression/pathway analysis. Schematic in e was partly created using BioRender.com. a–c, Velocity pseudotime (a), directed PAGA overlaid on UMAP (b) and velocity kernel from CellRank2 for lesional fibroblasts (c). For further details see Methods. d, UMAP visualization of fibroblast subtypes from human skin wounds data colored by cell type (left) and MKI67 (encodes Ki-67) expression (bottom right). Proportions of fibroblast populations by time point (top right), where each bar represents a donor at a given time point. e, Schematic of predicted trajectories. Dashed arrows indicate predictions with multiple lines of evidence. Fibroblast populations are colored by the predominant scarring/fibrosis risk observed in an earlier analysis: green (prevalent in low-risk scarring stroma), orange (prevalent in scarring risk stroma and cancer), red (prevalent in established scarring/fibrotic disorders). Gray boxes indicate signaling pathways identified in our gene expression/pathway analysis. Schematic in e was partly created using BioRender.com. To investigate these predicted trajectories in real time, we leveraged a human skin wound dataset of 58,823 cells (Methods). Skin tissue had been collected from healthy human volunteers at baseline (pre-wound) and subsequently from healing wounds. At baseline, myofibroblasts were not present, but on day 1 post-wounding, a small number of F6: inflammatory myofibroblasts were observed (Fig. 5d and Extended Data Fig. 7c). By day 7, F6: inflammatory myofibroblasts were the predominant population. By day 30, F7: myofibroblasts had become the predominant population, consistent with a role in established fibrosis/scarring. Overall, our results point toward F6: inflammatory myofibroblasts as an intermediate differentiation state toward F7: myofibroblasts in human skin, with potential plasticity in fibroblast origin (Fig. 5d,e and Extended Data Fig. 7e). We next investigated if the skin fibroblast subtypes we identified were conserved across other human tissues. Previous studies have reported fibroblast states that are found across human tissues. However, because these studies each defined fibroblast subtypes with different nomenclature and gene markers, it is unclear how these populations relate to each other and to the skin fibroblast populations reported in our study. To answer this and perform an overarching analysis across tissues and diseases, we undertook two approaches. The first was to assess the expression of marker genes for cross-tissue fibroblast subtypes in our skin data. This identified that reported cross-tissue populations from previous studies are likely present in human skin, consistent with F2: universal, F3: FRC-like, F6: inflammatory myofibroblast, and F7: myofibroblast subtypes (Fig. 6a and Extended Data Fig. 8a).Fig. 6Human cross-tissue disease fibroblast populations.a, Human tissues previously included in cross-tissue fibroblast studies and the fibroblast subtypes they have identified (above heatmap), colored by study (Buechler et al. in green, Korsunsky et al. in orange and Gao et al. in blue). Heatmap shows gene expression of marker genes previously reported for cross-tissue fibroblast populations in our lesional skin fibroblast subtypes. Immediately above the heatmap we show the skin fibroblast subtype with most similar gene expression to reported cross-tissue populations. b, UMAP visualization for cross-tissue integration (left) and for fibroblasts specifically (right), colored by tissue. Color bars indicate expression (log1P norm). c, UMAP visualization for fibroblasts colored by re-annotated clusters (Methods). d, Dotplots of expression of marker genes we previously used for skin fibroblasts in cross-tissue atlas clusters by tissue type. Note that not all genes were available as the endometrial dataset contained ~17,000 genes. Illustrations in a were partly created using BioRender.com. a, Human tissues previously included in cross-tissue fibroblast studies and the fibroblast subtypes they have identified (above heatmap), colored by study (Buechler et al. in green, Korsunsky et al. in orange and Gao et al. in blue). Heatmap shows gene expression of marker genes previously reported for cross-tissue fibroblast populations in our lesional skin fibroblast subtypes. Immediately above the heatmap we show the skin fibroblast subtype with most similar gene expression to reported cross-tissue populations. b, UMAP visualization for cross-tissue integration (left) and for fibroblasts specifically (right), colored by tissue. Color bars indicate expression (log1P norm). c, UMAP visualization for fibroblasts colored by re-annotated clusters (Methods). d, Dotplots of expression of marker genes we previously used for skin fibroblasts in cross-tissue atlas clusters by tissue type. Note that not all genes were available as the endometrial dataset contained ~17,000 genes. Illustrations in a were partly created using BioRender.com. The second approach was to integrate published datasets of ~5.8 million cells from human skin, lung, intestine, synovium, endometrium, heart and nasal mucosa (Methods and Fig. 6b). This approach uses the whole transcriptome profile, instead of restricted marker genes, and thus more comprehensively defines cell state similarity. Approximately 1 million fibroblasts were selected for downstream analysis based on expression of canonical marker genes (Fig. 6b). In the cross-tissue integrated dataset, we were able to discern shared fibroblast states across tissues, as well as fibroblasts that were unique to certain tissues (Fig. 6b,c and Extended Data Fig. 8b). In addition to known cross-tissue populations identified above (Fig. 6a), we found evidence for F2/3: perivascular (CXCL12, APOC1 and PPARG) and F5: NGFR (Schwann-like; SCN7A, NGFR, ITGA6 and EBF2) fibroblasts across human tissues, including in lung, gut and nasal mucosa (Fig. 6c,d and Extended Data Fig. 8b,c). Further interrogation of labeled intestine and lung datasets supported these results (Supplementary Note 3). F1: superficial showed variable gene expression by tissue, which may reflect distinct epithelia patterning across sites (Extended Data Fig. 8d). Overall, our results point toward the presence of previously reported cross-tissue fibroblast states in skin, despite major differences in the biophysical properties of different human tissues. We additionally suggest F2/3: perivascular and F5: NGFR (nerve-associated) fibroblasts as novel cross-tissue populations. Fibroblast-mediated processes such as fibrosis and maintenance of immune cell niches are observed across multiple human tissues. We therefore asked whether disease-associated fibroblast states identified in skin were similarly enriched by disease category in non-skin tissues (Methods). We focused on F3: FRC-like and F6: inflammatory myofibroblasts based on their conserved states across tissues and potential immune-interacting roles from pathway enrichment analysis (Extended Data Figs. 2a and 5c). Then, to predict functional interactions with immune cells, we utilized our skin data to perform cell–cell communication analysis. Across tissues, we observed that F3: FRC-like fibroblasts were present in both inflammatory disorders and fibrotic processes with immune-mediated pathology, including lung (COVID-19 and interstitial lung disease) and intestine (inflammatory bowel disease (IBD)) (Fig. 7a,b). While FRC-like fibroblasts were not reported in the Human Lung Cell Atlas (HLCA), following re-clustering we identified an F3: FRC-like population in lung (Extended Data Fig. 8e,f), consistent with a previous report. We also confirmed equivalence of F3: FRC-like fibroblasts to T reticular cells (an FRC subset) in IBD (Supplementary Note 3 and Extended Data Fig. 8g). Receptor–ligand analysis suggested interactions of skin F3: FRC-like fibroblasts with migrating dendritic cells (MigDCs) (CCL19-CCR7) and T cell subsets (CXCL12-CXCR4) (Extended Data Fig. 9a), suggesting that F3: FRC-like fibroblasts maintain T lymphoid populations and facilitate T cell–dendritic cell interactions analogous to T reticular cells in lymphoid tissue. We corroborated these findings using NicheCompass for niche identification in inflamed atopic dermatitis skin, which revealed CCR7 MigDCs and CXCR4 T cells within the F3 superficial perivascular niche (Fig. 7c and Extended Data Fig. 9b).Fig. 7Cross-tissue F3: FRC-like fibroblasts and F6: inflammatory myofibroblasts regulate skin immune niches.a, Proportion of disease-associated F3: FRC-like fibroblasts, F6: inflammatory myofibroblasts and F7: myofibroblasts in non-skin tissues by disease (left) and disease category (right). b, Dotplot of F3: FRC-like expression across diseases (skin and non-skin) (left). Dotplot of F6: inflammatory myofibroblasts gene expression across diseases (skin and non-skin) (right). Diseases with a minimum of 50 cells. c, H&E slide of lesional atopic dermatitis skin with annotation of perivascular infiltrate regions (top left). Niche identification and proportion of cells in the perivascular superficial niche (bottom left). Composition of the perivascular superficial niche in 10x Genomics Xenium. Insert: zoomed in version of perivascular niche cluster. d, Cell–cell communication analysis for F6: inflammatory myofibroblasts and skin immune cells (Methods). TCM, T central memory. e, Proportion of F6: inflammatory myofibroblasts in IBD by intestinal tissue inflammation status and linear regression of proportion with inflammation scores with 95% CI. f, Schematic summary of cell–cell interactions for F3: FRC-like fibroblasts and F6: inflammatory myofibroblasts. Schematic in f were created using BioRender.com. a, Proportion of disease-associated F3: FRC-like fibroblasts, F6: inflammatory myofibroblasts and F7: myofibroblasts in non-skin tissues by disease (left) and disease category (right). b, Dotplot of F3: FRC-like expression across diseases (skin and non-skin) (left). Dotplot of F6: inflammatory myofibroblasts gene expression across diseases (skin and non-skin) (right). Diseases with a minimum of 50 cells. c, H&E slide of lesional atopic dermatitis skin with annotation of perivascular infiltrate regions (top left). Niche identification and proportion of cells in the perivascular superficial niche (bottom left). Composition of the perivascular superficial niche in 10x Genomics Xenium. Insert: zoomed in version of perivascular niche cluster. d, Cell–cell communication analysis for F6: inflammatory myofibroblasts and skin immune cells (Methods). TCM, T central memory. e, Proportion of F6: inflammatory myofibroblasts in IBD by intestinal tissue inflammation status and linear regression of proportion with inflammation scores with 95% CI. f, Schematic summary of cell–cell interactions for F3: FRC-like fibroblasts and F6: inflammatory myofibroblasts. Schematic in f were created using BioRender.com. F6: inflammatory myofibroblasts were abundant in cancer and inflammation but relatively uncommon in established fibrosis (Fig. 7a,b), consistent with skin data (Fig. 4b). Inflammatory myofibroblasts are well described in IBD, and we confirmed equivalence of these cells to skin F6: inflammatory myofibroblasts (Supplementary Note 3 and Extended Data Fig. 8g). To further assess the clinical relevance of F6 in IBD, we used an scRNA-seq dataset with clinical metadata. F6: inflammatory myofibroblasts were significantly elevated in inflamed tissue, compared to non-inflamed tissue, and their prevalence correlated with clinical inflammation severity scores (Fig. 7e). Receptor–ligand analysis suggested that F6: inflammatory myofibroblasts recruit and maintain neutrophils (CXCL5/6/8-CXCR2 and CSF3-CSF3R), macrophages/monocytes (CCL5/26-CCR1 and CSF3-CSF3R) and B cells (CXCL13/CXCR5) in the skin (Fig. 7c and Methods). These genes were highly expressed in the skin during wound healing, acne and hidradenitis suppurativa, as well as in IBD and lung cancer (Fig. 7b), suggesting a similar mechanism for recruitment of immune cells across tissues. Overall, our results suggest that F3: FRC-like (CCL19CD74TNFRSF13B and IL33/IL15) and F6: inflammatory myofibroblasts (IL11MMP1CXCL5IL7R) mediate distinct immune niches driving pathology in the skin and other tissues (Fig. 7f). Given the similar transcriptomic profiles of adult skin F3: FRC-like and intestinal T reticular cells, we hypothesized that these fibroblasts had similar origins in their respective tissues. Intestinal FRCs are found in the Peyer’s patch and are thought to arise from prenatal lymphoid tissue organizer (LTo) cells; however, the skin does not harbor the equivalent of Peyer’s patch and the origin of F3: FRC-like cells remains unknown. To explore F3 ontogeny from a developmental perspective, we first integrated adult and prenatal skin fibroblasts and identified the corresponding fibroblast populations (Supplementary Note 4). This identified that adult F3: FRC-like fibroblasts correlated with prenatal skin CCL19 fibroblasts (Fig. 8a,b).Fig. 8Adult skin F3: FRC-like fibroblasts potentially arise from prenatal skin LTo cells.a, UMAP visualization for prenatal human skin and adult skin colored by cell type and (insert) age group. b, Heatmap of adult skin gene module scores applied to prenatal skin fibroblasts. c, UMAP visualization for prenatal human skin and intestine. Insert: cluster of LTo-like cells. (m)LTo: (mesenchymal) lymphoid tissue organizer. Dotplot of marker gene expression for LTo-like cells by tissue. d, Dotplot of expression of F3: FRC-like fibroblasts and LTo-associated marker genes in healthy adult skin F3: FRC-like fibroblasts, prenatal skin CCL19 fibroblasts and intestinal mLTo cells. e, Dotplot of expression of marker genes for adult skin F3: FRC-like fibroblasts and LTo-associated marker genes in diseased adult skin, by disease, for diseases with a minimum of 100 F3: FRC-like fibroblasts. DRESS, drug reaction with eosinophilia and systemic symptoms. f, Dotplot of expression of F3: FRC-like fibroblast marker genes in mouse steady-state cross-tissue atlas (Buechler et al.). g, Proportion of Ccl19 fibroblasts (labels from original study) in mouse in different healthy tissues (including mouse flank skin) and of F3: FRC-like fibroblasts in healthy human skin. Illustrations in f and g were partly created using BioRender.com. a, UMAP visualization for prenatal human skin and adult skin colored by cell type and (insert) age group. b, Heatmap of adult skin gene module scores applied to prenatal skin fibroblasts. c, UMAP visualization for prenatal human skin and intestine. Insert: cluster of LTo-like cells. (m)LTo: (mesenchymal) lymphoid tissue organizer. Dotplot of marker gene expression for LTo-like cells by tissue. d, Dotplot of expression of F3: FRC-like fibroblasts and LTo-associated marker genes in healthy adult skin F3: FRC-like fibroblasts, prenatal skin CCL19 fibroblasts and intestinal mLTo cells. e, Dotplot of expression of marker genes for adult skin F3: FRC-like fibroblasts and LTo-associated marker genes in diseased adult skin, by disease, for diseases with a minimum of 100 F3: FRC-like fibroblasts. DRESS, drug reaction with eosinophilia and systemic symptoms. f, Dotplot of expression of F3: FRC-like fibroblast marker genes in mouse steady-state cross-tissue atlas (Buechler et al.). g, Proportion of Ccl19 fibroblasts (labels from original study) in mouse in different healthy tissues (including mouse flank skin) and of F3: FRC-like fibroblasts in healthy human skin. Illustrations in f and g were partly created using BioRender.com. We next queried whether prenatal CCL19 fibroblasts were equivalent to LTo-like cells by jointly integrating human prenatal skin and intestinal data. Notably, prenatal skin CCL19 cells and prenatal intestinal mesenchymal LTo cells clustered together (Fig. 8c). Prenatal skin CCL19 cells expressed known mesenchymal LTo markers, including CCL21, CXCL13, MADCAM1, FDCSP and TNFSF11 (RANKL) (Fig. 8c), suggesting that prenatal skin CCL19 cells may give rise to adult skin F3: FRC-like cells in a manner analogous to intestinal LTo cells. We next investigated the LTo gene program in adult skin F3: FRC-like fibroblasts. The LTo gene program (including CXCL13 and FDCSP) was not expressed in healthy adult skin but could be upregulated in specific skin diseases (Fig. 8d,e), particularly hidradenitis suppurativa. Tertiary lymphoid structures (TLS), for which CXCL13 is an important chemokine, are not well recognized in adult human skin, but have recently been reported in hidradenitis suppurativa specifically, suggesting that the LTo gene program contributes to this process. We next asked whether F3: FRC-like fibroblasts were unique to adult human skin, as LTo-like fibroblasts have not been reported in mouse embryonic skin. Our comparative analysis showed that adult F3: FRC-like cells correspond to mouse Ccl19 fibroblasts (Fig. 8f). Mouse Ccl19 fibroblasts were found predominantly in lymphoid organs (Extended Data Fig. 10a), but were also present in other tissues such as lung (Fig. 8g), whereas F3: FRC-like fibroblasts were relatively abundant in healthy human skin, the equivalent Ccl19 fibroblasts were notably rarer in healthy mouse skin (Fig. 8g). In summary, we suggest F3: FRC-like fibroblasts are enriched in human skin and not observed or absent in murine skin. We harmonize skin fibroblast subtype nomenclature in health and disease, spatially resolve distinct fibroblast anatomical niches, and identify conserved fibroblast subtypes in human diseases affecting multiple tissues. FRCs are the paradigm of immunomodulatory fibroblasts, maintaining discrete immune structures and facilitating specific immune cell interactions in lymphoid organs, with increasing evidence that they are present across human tissues. Our data suggest that F3: FRC-like fibroblasts are located in the superficial perivascular niche in human skin and have an analogous role to T zone reticular FRCs, mediating T-DC interactions. In keeping with previous murine studies, we find that F3: FRC-like fibroblasts show a uniquely high prevalence in human skin. This enrichment of F3: FRC-like fibroblasts in human skin would explain both the absence of fibroblasts in murine skin TLS-like structures and why the prominent perivascular infiltrate structures that characterize many human inflammatory skin diseases are not reported in mice. Inflammatory myofibroblasts were recently reported in a large-scale integration of predominantly CAFs and have been independently described in IBD. We identify that the same inflammatory myofibroblast phenotype (IL11MMP1CXCL8IL7R) can be observed in early human skin wounds, skin cancer and inflammatory skin diseases with scarring risk. Consistent with the immune milieu in early wounds, acne and hidradenitis suppurativa, we suggest that these fibroblasts recruit immune cells such as neutrophils and monocytes. Neutrophils are also reported to be recruited by inflammatory myofibroblasts in IBD. Our study also suggests that F6: inflammatory myofibroblasts are an intermediate myofibroblast differentiation state in human skin, which agrees with recent work in mouse skin and lung. Further work validating myofibroblast trajectories in human skin is needed as current trajectory inference methods are limited for predicting multiple cell states converging to a final phenotype, and lineage plasticity is likely in fibroblasts, with both tissue-specific and universal populations suggested to give rise to myofibroblasts. Notably, skin could serve as an exemplar tissue to further investigate myofibroblast development in humans in vivo given the ability to sample tissue temporally with low morbidity. A limitation of our study is that we relied on the uncertainty mechanism incorporated in scPoli to identify disease-associated populations. Our cross-tissue analysis using semi-supervised integration may underestimate tissue-specific differences between fibroblasts, and further investigation using methods such as contrastive analysis may be valuable. LTo-like cells in prenatal skin were rare and more definitive lineage tracing methods are required to understand prenatal to adult fibroblast transitions. In summary, our annotated skin fibroblast dataset of 357,276 cells provides a foundational resource for fibroblast transcriptomic states in health and across distinct disease categories in skin tissue. Further comments on annotation can be made via the centralized community annotation platform (https://celltype.info/project/388). Raw scRNA-seq data were downloaded and aligned using STARsolo (GRCh38-2020-A reference) unless already available in local storage. We included publicly available data generated from fresh skin biopsies using the 10x Genomics Chromium platform. We collected essential metadata (sample ID, dataset ID, site status, patient status, sex and anatomic location) as more extensive metadata are being collected as part of the Human Skin Cell Atlas. CellBender v.0.3 was used to correct for ambient messenger RNA. To remove low-quality cells, we included only cells with >200 genes, >1,000 and <300,000 total unique molecular identifiers and a mitochondrial gene percentage of <15%. We calculated doublet scores using scrublet on a per sample basis using scrublet and removed cells with a doublet score of >0.3 (ref. ). Integration of all cells was performed using scVI using raw counts. For feature (highly variable gene (HVG)) selection, we did not consider the following genes: mitochondrial genes; cell cycle genes, from https://github.com/haniffalab/skin_fibroblast_atlas/blob/main/misc/cc_genes.csv); hypoxic genes; and ribosomal genes. We selected 6,000 HVGs as features, and batch-aware HVG selection was performed by setting the batch key to sample ID. In future analyses post-integration, all genes were considered. The following hyperparameters were used for the scVI model: number of layers: 2; number of latent dimensions: 30; gene likelihood: zero-inflated negative binomial distribution, and dispersion: gene-batch. An early stopping patience of five epochs was used. The batch key was ‘sampleID’ and no other covariates were passed to the model for correction. We constructed a k-nearest neighbors (k-NN) graph (k = 30) using the scVI embedding and performed community detection (Leiden algorithm) with resolution 0.1 for the dataset with all skin cells. Visualization in two dimensions was performed using UMAP with the initialized positions from PAGA implemented in scanpy. We then selected the fibroblast cluster for further analysis based on canonical marker gene expression, including PDGFRA, DCN and LUM) (Extended Data Fig. 1a). We did not include a distinct stress response cluster (MT2A, MT1M, MT1X, HSP90AA1, JUNB, GADD45B and IER3) as this population was not evident on Xenium analysis and thus likely related to cell dissociation. For a sensitivity analysis with additional cell types, we also selected Schwann and pericyte clusters. We repeated integration using scVI for healthy and phenotypically normal (nonlesional) skin fibroblasts only. We used the same workflow as above. Visualization in two dimensions was performed using UMAP with positions initialized from PAGA. We show gene expression values post-normalization. Normalization is a two-step procedure involving depth normalization and variance stabilization. We used the shifted logarithm with a scaling factor of 10,000 based on strong performance in a recent benchmarking paper. For each cluster, we calculated differentially expressed genes (DEGs) using the t-test with the scanpy rank_genes_groups function. The top DEGs for each population are shown in Supplementary Fig. 1a and Supplementary Table 2. For selecting marker genes to present for each population in Fig. 1c, we selected genes with the highest specificity of expression for that cluster from visualization. We selected our nomenclature for fibroblasts based on our previous report of F1–F3 fibroblasts. We refined the names based on spatial location and our in-depth characterization of cell states, including across tissues. We renamed F1 as F1: superficial, F2 to F2: universal and F3 to F3: FRC-like. In this work, we defined novel fibroblast subtypes. F2/3 fibroblasts expressed select F2 (CD34 and PI16) and F3 markers (CXCL12, PLA2G2A and C7), but additionally expressed PPARG. F2/3 fibroblasts also showed a distinct location enriched in both superficial and deep dermal perivascular structures, compared to F2 (deep dermis but not around vasculature) and F3 (superficial perivascular regions) (Extended Data Fig. 4c). We therefore named this population F2/F3: Perivascular. We additionally identified novel F4 and F5 populations. We grouped three subtypes as F4 based on transcriptomic (ASPN/COL11A1) and spatial similarity (localization around the hair follicle). We grouped two subtypes as F5 based on shared expression of genes also expressed by Schwann cells, including SCN7A, FMO2 and OFLML2A, and spatial association with nerve structures. We took two approaches to minimize the possibility of missing a rare distinct cluster. First, we repeated unsupervised clustering at a higher resolution and assessed whether any clusters showed distinct gene expression. Second, we reviewed reported marker genes for fibroblast clusters in previously reported scRNA-seq data for human skin to identify if a novel distinct population has been reported. Transcription factor activity inference was performed using decoupler, using the get_collectri (human) and run_ulm functions. This analysis was performed using the data subset of 6,000 HVGs. Gene set enrichment analysis was implemented using GSEAPY. We used the top 500 DEGs per fibroblast subtype and the GO_Biological_Process_2023 gene set. A cutoff statistical significance of 0.01 was used. We generated new spot-based spatial transcriptomic data using the 10x Genomics Visium Spatial Gene Expression platform from frozen OCT-embedded human adult skin tissue (n = 1 lesional atopic dermatitis and n = 2 nonlesional atopic dermatitis). All research ethics committee and regulatory approvals were in place for the collection of research samples at Newcastle and for their storage at the Newcastle Dermatology Biobank (REC reference no. 19/NE/0004). Skin samples were sectioned at 15-µm thickness and the optimal tissue permeabilization time was determined as 14 min. H&E images were taken using a Zeiss AxioImager with apotome microscope (Carl Zeiss Microscopy) and Brightfield imaging (Zeiss Axiocam 105 48-color camera module) at ×20 magnification. The ZEN blue edition v.3.1 (Carl Zeiss Microscopy) software was used to acquire the H&E images following z-plane and light balance adjustment and image tile stitching. Spatial gene expression libraries were sequenced using an Illumina NovaSeq 6000 to achieve a minimum number of 50,000 read pairs per tissue covered spot. The 10x Genomics Visium data were mapped using Spaceranger v.1.3.0 using GRCh38-2020-A reference. Visium provides whole transcriptome coverage over a 55-μm diameter spot area. We therefore used the cell2location (v.0.1.3) to deconvolute the cell types predicted to be present in a given spot. We constructed a reference signature using sample as batch_key. We included our fibroblast atlas and other skin cell types from Reynolds et al.. Then we performed deconvolution with the following parameters: detection_alpha = 20,N_cells_per_location = 30 andmax_epochs 50_000. Values of all other parameters were kept default. Following the cell2location tutorial, we used 5% quantile of posterior distribution (q05_cell_abundance_w_sf) as predicted cell-type abundances. Nonlesional (n = 2; one sample nonlesional pre-treatment, one sample nonlesional post-treatment) and lesional (n = 1; inflamed) atopic dermatitis human adult skin tissue was used to generate in situ gene expression data using the 10x Genomics Xenium in situ 5k-plex platform. All research ethics committees and regulatory approvals were in place for the collection and storage of research samples at St John’s Institute of Dermatology, Guy’s Hospital, London (REC reference no. EC00/128). The 10x Genomics Xenium data were filtered to exclude cells with <10 genes per cell. Integration of all cells was performed using scVI using raw counts. Batch-aware HVG selection was performed by setting the batch key to sample ID. We selected 2,000 HVGs as features for the integration. As Xenium data are more sparse than scRNA data, we used a simpler encoder (number of layers: 1; number of latent dimensions: 10) than scRNA. The batch key was ‘sampleID’ and no other covariates were passed to the model for correction. For normalization, we again used the shifted log transformation for count normalization. We constructed a k-NN graph (k = 20, given the smaller dataset size) using the scVI embedding. We performed community detection (Leiden algorithm) with resolution 0.1 and used the same markers as scRNA data to select the fibroblast cluster. We used marker genes from scRNA-seq data to label fibroblast populations through manual annotation. Cell coordinates colored by cell type were visualized using squidpy (sq.pl.spatial_scatter). We used the same integration strategy for newly generated atopic dermatitis data and publicly available cutaneous melanoma data. Cutaneous melanoma data were downloaded from https://www.10xgenomics.com/datasets/xenium-prime-ffpe-human-skin. NicheCompass was run after selecting 1,024 spatially variable genes. The number of neighbors selected per cell was 8, and otherwise default settings were used. To calculate neighborhood enrichment scores, we first constructed an adjacency matrix for indicating which cells were connected using the spatial_neighbors functions in squidpy. We determined the neighborhood as a cell as cells within 20 µm of an index cell. We then applied neighborhood enrichment analysis (nhood_enrichment function in squidpy) to quantify which cell types were more frequently colocalized than expected by chance. Heatmaps of enrichment scores were visualized using sq.pl.nhood_enrichment. To identify fibroblast clusters in disease, we mapped lesional data to the healthy/nonlesional reference described above. We used a state-of-the-art deep meta-learning model (scPoli). This approach first trains a model using the healthy/nonlesional (reference) to generate centroids (Fig. 3a). Then, cells in the query that are distinct to centroids generated from the reference are marked as uncertain, which facilitates the discovery of new cell types/states in the query data (Fig. 3a). This permits automated cell annotation while highlighting cells that could not be mapped to the reference through prototypical learning. Due to the proposed role of hypoxia in myofibroblast differentiation, we included hypoxic genes for consideration in feature selection. To ensure this selection did not bias results, we repeated an scVI integration using the same methodology as for healthy fibroblasts (Supplementary Fig. 4a,b). Our definition of ‘uncertain’ cells was derived from scPoli. scPoli utilizes Euclidean distances of query cells from prototypes (or centroids) generated from the reference to yield an uncertainty associated with each cell. To manually annotate cells labeled as uncertain, we calculated DEGs for each Leiden cluster and also assessed expression of healthy fibroblast marker genes in each population. The top DEGs for each cluster considered as disease-associated or disease-specific are shown in Supplementary Fig. 1c and Supplementary Table 3. We also considered which populations were enriched in disease. To validate our reported fibroblast subtypes, we used two approaches. First, we used skin scRNA-seq datasets not used in the original integration. Using the scPoli model to generate embeddings and transfer labels to these populations, we identified expected populations from earlier analysis (Supplementary Fig. 4c). Second, we used Xenium data to validate the existence of the same clusters in which cell gene expression profiles are generated in situ, without tissue dissociation. For CellDISECT, we used 6,000 HVGs and raw counts as input. Model architecture and training schedule is listed in the provided code. We used the top 500 human genes per pathway and default settings for the multivariate linear model (decoupler.run_mlm). We broadly grouped individual diseases into disease categories (inflammatory + low scarring risk, inflammatory + high scarring risk, cancer, established scarring/fibrosis) based on clinical disease features. Of note, clear separation of diseases is not possible because of a well-recognized link between inflammation and fibrosis. For example, an inflammatory component to systemic sclerosis is well recognized and first-line treatments typically include immunosuppressants, but this disease also features established fibrosis. Additionally, we included sarcoidosis and granuloma annulare (both disorders of granulomatous inflammation) in ‘inflammatory high scarring risk’ as fibrosis can occur within granulomas. Additionally, pulmonary sarcoidosis is a well-recognized cause of pulmonary fibrosis; however, most cases of cutaneous sarcoidosis and granuloma annulare do not scar. Prurigo nodularis was classified as low scarring risk as it is unclear whether scarring arises secondary to scratching. We considered cancer as moderate scarring risk as fibrosis can occur within lesions (desmoplasia) and self-resolving melanoma can result in scarring. For fibroblast proportions by scarring category, we calculated the s.e.m. for each category using the mean proportion for each donor. The s.e.m. for each disease category was derived from the s.d. of donor-level proportions divided by the square root of the number of donors in that category. All research ethics committees and regulatory approvals were in place for the collection and storage of atopic dermatitis skin samples at the St John’s Institute of Dermatology, Guy’s Hospital, London (REC reference no. EC00/128) and hidradenitis suppurativa skin samples at Newcastle Dermatology Biobank (REC reference no. 19/NE/0004). Fresh-frozen OCT-embedded skin samples were sectioned at 10-µm thickness directly onto superfrost microscope slides and stored at −80 °C. Slides were air dried at room temperature for 10 min and then fixed using 4% PFA for 10 min. Next, a blocking solution of 5% normal goat serum with 0.01% Triton X-100 was applied to the tissue sections and incubated for 1 h at room temperature. Slides were then incubated with primary antibodies overnight at 4 °C. The next day, slides were washed with 1× PBS and incubated with secondary antibodies for 1 h at room temperature. Then, 4,6-diamidino-2-phenylindole (DAPI) was used to demarcate nuclei and slides were mounted with DAKO mounting medium before applying coverslips and leaving slides to dry overnight. Skin sections were imaged using a Leica SP8 confocal microscope. To calculate gene scores, we used the score_genes() function in scanpy. We used arguments of ctrl_size = 1,000 and n_bins = 25. We used the marker genes reported for each population, rather than more extensive gene lists, based on the rationale that larger gene programs would be more likely to include tissue-specific gene expression and thus underestimate transcriptomic similarity across tissues. We trained a random forest classifier using fibroblast subtype composition as input to predict scarring risk group. We evaluated performance in terms of average F1 score, a widely used metric for evaluating classification performance, computed using classification_report from scikit-learn. We applied fivefold stratified cross-validation to train and evaluate a RandomForestClassifier (100 estimators, random_state = 42). To identify cell types that are most predictive of scarring status, we extracted the final trained model’s feature importances. For trajectory inference, we used both velocity-based (RNA velocity (scVelo and CellRank2 Velocity kernel)) and graph-based (PAGA and Monocle 3) approaches. We used data for which we could calculate RNA velocity using velocyto and scvelo. We first used PAGA (as implemented in scanpy), plotted using a threshold of 0.1 and applied to the whole dataset. In future analyses, we excluded F5: Schwann-like fibroblasts, which seemed to be distinct (Extended Data Fig. 8a) and F_Fascia, which were observed in few diseases (Fig. 4a). As healing and scarring is observed on non-hair-bearing sites, we also did not include F4: hair follicle-associated fibroblasts. We generated new scVI embeddings for the lesional fibroblasts and re-calculated the k-NN graph using the top 2,000 HVGs, followed by UMAP visualization. Velocity pseudotime was calculated using scvelo. We re-calculated the PAGA plot using only lesional fibroblasts, again using a threshold of 0.1. For Monocle 3 (v.1.3.7), the expression count matrix along with the corresponding cell and gene metadata from the processed anndata object in scanpy was used to create Monocle object (cell_data_set object). The cell_data_set object was then pre-processed using default settings and aligned to correct for batch effects based on the ‘dataset_id’. Dimensionality reduction was performed using ‘UMAP’ as the reduction method. Cells were clustered with a resolution of 1 × 10 − 6 and ‘UMAP’ as the reduction method. A trajectory graph was learned by adjusting parameters such as geodesic distance ratio (0.5) and minimal branch length (10) to optimize for large datasets. Finally, cells were arranged in pseudotime by manually selecting root nodes from the F2: universal population. We used F2: universal as the root state in Monocle 3 based on velocity pseudotime results and previous work. The ordered and learned graph object was then used to plot the pseudotime trajectory plots. The RNA velocity kernel was calculated using CellRank2. We obtained post-alignment skin wound data from the authors of ref. and processed these using CellBender as previously described for skin fibroblast data. To annotate skin wound fibroblasts, we integrated the unlabeled skin wound cells with our labeled integrated skin dataset using scanVI. We used the same model architecture (30 latent dimensions, two layers) and the same number of input HVGs (n = 6,000). We selected only fibroblasts for further analysis, using the same downstream strategy as used for skin fibroblasts previously. For cross-tissue marker gene comparisons, we selected reported populations from Korsunsky et al., Buecher et al. and Gao et al. As Gao et al. reported six universal and five shared populations, we show matching populations in the main figure and other populations in the extended figure. For cross-tissue integration, we concatenated our labeled skin data with other tissues (raw count data). HLCA data, gut atlas data and Human Endometrial Cell Atlas data were available locally. We downloaded nasal tissue data from https://ngdc.cncb.ac.cn/gsa-human/browse/HRA000772 (ref. ), heart data from https://data.humancellatlas.org/explore/projects/e9f36305-d857-44a3-93f0-df4e6007dc97, rheumatoid arthritis from https://www.immport.org/shared/study/SDY998 (ref. ) and additional intestinal data from the Gene Expression Omnibus (GEO) under accession code GSE282122. We used the same number of input HVGs (n = 6,000). We integrated data in a semi-supervised manner using scANVI, where skin cell types were labeled and cells from other tissues were unlabeled. We used the same scANVI hyperparameters as for wound data but with a smaller number of maximum epochs (n = 10) due to the larger dataset size. We then calculated k-NN (k = 30) and performed low-resolution Leiden clustering (resolution 0.1). We selected a fibroblast cluster based on canonical marker genes, which also contained the labeled skin fibroblasts. To annotate clusters, we labeled clusters by the majority skin fibroblast population (Fig. 6c and Extended Data Fig. 8b) (for example if F3: FRC-like was the predominant skin fibroblast subtype, we labeled the cluster as F3: FRC-like). We then assessed gene expression markers for each cluster, excluding skin fibroblasts, to ensure that skin fibroblasts did not drive the gene expression signature for that cluster. We also plotted gene expression for each cluster by tissue using our previously reported marker genes for each cluster. To assess F3: FRC-like fibroblasts in HLCA data, we performed the same clustering strategy as previously described for skin fibroblasts and then plotted F3: FRC-like marker genes by cluster. Inflammation severity scores were obtained from GEO under accession code GSE282122. A linear regression model was fitted using ordinary least squares with the F6 proportion as the dependent variable and inflammation severity score as the independent variable using seaborn’s regplot function. We used CellPhoneDB v.5 (method 2) for cell–cell communication analysis. We combined our fibroblast data with skin immune cells from our previously published scRNA-seq data from skin with more granular immune cell annotations (Reynolds et al.). We restricted interactions to marker genes for F3: FRC-like fibroblasts and F6: inflammatory myofibroblasts. We visualized the results using ktplotspy. For prenatal skin and prenatal intestine, we concatenated raw count adata objects from intestine with our prenatal skin data from a previous publication. We ran scVI using the same parameters as for the healthy/nonlesional integration. We used the same strategy for the adult skin and prenatal skin integration. For mouse comparisons, we downloaded the mouse steady-state atlas from https://www.fibroxplorer.com/download. We loaded the data as an adata object using pandas2ri in rpy2. We used labels for Ccl19 fibroblast from the original study. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41590-025-02267-8.
PMC2211323
A large-scale proteomic analysis of human embryonic stem cells
Much of our current knowledge of the molecular expression profile of human embryonic stem cells (hESCs) is based on transcriptional approaches. These analyses are only partly predictive of protein expression however, and do not shed light on post-translational regulation, leaving a large gap in our knowledge of the biology of pluripotent stem cells. Here we describe the use of two large-scale western blot assays to identify over 600 proteins expressed in undifferentiated hESCs, and highlight over 40 examples of multiple gel mobility variants, which are suspected protein isoforms and/or post-translational modifications. Twenty-two phosphorylation events in cell signaling molecules, as well as potential new markers of undifferentiated hESCs were also identified. We confirmed the expression of a subset of the identified proteins by immunofluorescence and correlated the expression of transcript and protein for key molecules in active signaling pathways in hESCs. These analyses also indicated that hESCs exhibit several features of polarized epithelia, including expression of tight junction proteins. Our approach complements proteomic and transcriptional analysis to provide unique information on human pluripotent stem cells, and is a framework for the continued analyses of self-renewal.Human embryonic stem cells (hESCs) are pluripotent cells isolated from the inner cell mass of the blastocyst . They can be maintained for prolonged periods in culture and differentiate to representatives of the three germ layers as well as trophoblasts and germ cells. This differentiation potential may be used to model certain aspects of human embryogenesis, including the development and differentiation of pluripotent and other stem cell types during the processes of gastrulation, neurogenesis and organogenesis. Thus, hESCs provide a unique and powerful system to study otherwise intractable aspects of human development. Furthermore, these approaches have the potential to provide differentiated cell types for cell replacement therapies of degenerative disorders such as Parkinson's disease and Type I diabetes . Before these cell therapy applications are developed, an understanding of the molecular and cellular mechanisms that drive self-renewal and differentiation is required. Fundamental to this understanding is the elucidation of the transcriptome and proteome of hESCs, using approaches that lay a framework for functional analyses of the unique properties of these cells. Large-scale gene expression analyses such as microarray, massive parallel signature sequencing (MPSS), expressed sequenced tag (EST) enumeration, and serial analysis of gene expression (SAGE) have been used to compare multiple hESC lines [4-7]; hESCs to germ cell tumors ; or to differentiated derivatives in embryoid bodies [9-11] or neural populations . These approaches have highlighted an expanded set of transcripts that mark the pluripotent state , cross-species commonalities in the molecular profile of ESCs , prominent receptors expressed by hESCs and pathways that may play a role in the regulation of pluripotency . Nevertheless, cataloguing the cellular transcriptome is only predictive of protein expression and typically does not shed light on post-transcriptional regulation. For example, while tens of thousands of transcripts can be followed simultaneously with SAGE, microarrays and MPSS, these methods do not routinely detect differences in transcript splice variants, or polyadenylation status. These differences may have profound effects on translation, as well as the isoform and function of the protein produced. Finally, numerous post-translational modifications are known to regulate protein function, including enzymatic cleavage, covalent coupling to other molecules, glycosylation, phosphorylation and ubiquitination. These issues all highlight potential shortfalls in our understanding of the hESC proteome. Several practical approaches for proteomic analyses are currently available, the most established of which is the 2-dimensional (2D) separation of proteins by polyacrylamide gel electrophoresis (PAGE). HPLC-tandem mass spectrometry (HPLC-MS/MS) based technology is rapidly evolving and has recently been used to detect protein expression in multiple cell types. An alternate approach is the recent large-scale adaptation of standard western blotting . In this procedure, a large well is used to separate the sample by PAGE and lanes are created on the membrane containing immobilized protein with the use of a manifold. Compatible combinations of primary antibodies are predetermined, with the criterion of being able to identify proteins that do not co-migrate. Different combinations of primary antibodies are added to each well, with appropriate dilutions of each primary antibody so that expressed proteins are detected in a single condition. The scalability of the system depends on defining suitable combinations of primary antibodies, with up to 1000 antibodies in 200 lanes being used in the largest screens thus far. Detection software is used to identify proteins based on their expected and observed gel mobility. Unlike 2D PAGE and HPLC-MS/MS, large-scale western blotting only identifies proteins for which antibodies are already available. While this is not an appropriate screen for identifying uncharacterized proteins, it greatly simplifies the verification and functional analyses of proteins that are detected. In addition, this approach is highly flexible, and if desired can be focused to particular sets of proteins or protein function, such as cell signaling molecules. Importantly, the foundation of this approach is the large amount of data on individual antibodies, which are already available and characterized in the literature. More recently, two research groups have conducted proteomic analyses of hESCs using MS [19-22]. In the present study, we used two large-scale western blot systems to examine the expression of > 1000 proteins in hESCs and detected > 600 proteins that were grouped into 18 functional classes. In addition, we identified 42 examples of multiple bands for a single protein, likely to be protein isoforms and/or post-translational modifications, and 22 phosphorylation events in cell signaling molecules. We correlated the expression of members of key active pathways in our transcriptional and proteomic databases and confirmed the validity of this approach. Using these approaches we identified new markers for undifferentiated hESCs and highlighted unrecognized epithelial characteristics of hESCs. Our data confirm the importance of proteomic analyses in complementing transcriptional profiling and provide a framework for continued analyses of the molecular and cellular biology of pluirpotent hESCs. We first employed a large-scale western blot screen, the PowerBlot system, to profile protein expression in undifferentiated hESCs. This system used 934 antibodies toward proteins representing 22 diverse classes of function, such as transcription factors, the MAP kinase (MAPK) pathway, and apoptosis, among others. To expand a large-scale culture of BG01 cells for this assay, a collagenase- and trypsin- based passaging method was used . While these conditions have been associated with the accumulation of trisomies of chromosomes 12, 17 and X , the ease of use of these cultures and similarity in gene expression and differentiation potential to karyotypically normal BG01 hESCs make them suitable for such large scale applications. For the PowerBlot screen, whole cell lysate from BG01 hESCs was separated on five 4–15% gradient gels. Each blot contained size markers and 39 lanes. Each lane was screened with 1–8 antibodies in combinations that had been predetermined to enable accurate identification of well-separated proteins (Fig. 1A–E). The gels and blots were performed in duplicate and expressed proteins were identified by their predicted size and verified by visual inspection. PowerBlot analysis of undifferentiated BG01 hESCs. This large-scale western blot consisted of five gels run in duplicate and probed with 934 antibodies. (A-E) One set of blots is shown at a contrast that highlights most bands. (F) A representative lane (gel C, lane 24) aligned with protein markers used for band identification. (G) Scatter-plot of the normalized average intensity (i.u.) values for each protein indicating a linear relationship between duplicate blots. Datasets for this analysis are in Additional Tables 1 and 2. A total of 545 antibodies detected bands of appropriate size, which could be compressed to 529 proteins with unique SwissProt identification numbers (Fig. 1A–E and Additional File 1). An enlargement of a representative lane (lane 24 of Blot C) alongside protein markers is shown in Fig. 1F. Thirteen proteins including AKT, caveolin1 and ERK1 were detected in multiple lanes using the same or different antibodies. Information on the antibody catalogue number and dilution, band intensity for each repeat and the averaged value, description of protein function, and Entrez gene and SwissProt database identification numbers is shown in Additional File 1. Three hundred and eighty three antibodies did not detect bands in this screen, indicating lack of expression, or possibly technical issues with detection under standard conditions (Additional File 1). The size of the detected proteins ranged from 15 kD (GS15) to 280 kD (ABP-280). The average intensity of the detected proteins ranged from 195 to 117926 normalized intensity units (i.u.), with an average of 5367 i.u. The proteins with the highest band intensity were the B2 Bradykinin Receptor (117926 i.u.), Karyopherin α (80698 i.u.), and BiP (74922 i.u.), whilst the proteins with the lowest intensity that could be verified by visual inspection were Inhibitor 2 (247 i.u.), Caspase 8 (201 i.u.), and OXA1Hs (195 i.u.). Finally, the consistency of this assay was demonstrated by plotting the normalized average intensity values for each protein, which revealed a linear relationship between the duplicate samples (Fig. 1G). A more focused screen was used to profile expression of protein kinases, phosphatases and phosphorylated sites in cell signaling molecules in hESCs. The Kinexus assays contained 140 antibodies to these related classes of proteins and phospho-sites. Karyotypically normal BG03 hESCs grown on a fibronectin matrix in MEF-CM were used for this analysis, and whole cell lysate was separated on four 12.5% gels for western blotting. Eighty five immunoreactive bands were identified, representing 38 protein kinases and 16 phosphatases, their isoforms, and 22 phosphorylated sites in signaling molecules (Fig. 2A–D, Additional File 1). Sixty-four antibodies did not detect their corresponding antigen (Additional File 1). Kinexus blots of undifferentiated BG03 cells. Four blots were used to probe BG03 lysate with (A, B) 76 antibodies for protein kinases, (C) 27 antibodies for phosphatases and (D) 37 antibodies for phosphoylated sites in cell signaling molecules. Identified bands are indicated (*). Datasets for this analysis are in Additional Tables 1 and 2. The PowerBlot and Kinexus assays identified a diverse range of proteins expressed in hESCs. To further annotate these data, the detected proteins were ordered into 18 subgroups based on protein function (Additional File 2). For example, 16 factors with known or implied roles in the regulation of self-renewal or pluripotency of mESCs or hESCs, such as Oct4 , STAT3 , members of the FGF , PI3 kinase , Src or MAPK pathways , and phosphorylated isoforms of GSK3, STAT3 and p38 MAPK, were grouped under "Pluripotency" (Fig. 3A and Additional File 2). Another functional group (Cell surface) consisted of 20 transmembrane or cell surface proteins (Additional File 2). This included several receptors for peptides and growth factors, such as neurotensin receptor 3, the B2 bradykinin, endothelin 1, and thrombin receptors, and the glial derived neurotrophic factor receptor α (Fig. 3B). These molecules may be useful as targets for cell sorting experiments, and expression of these receptors could identify bioactive peptides or growth factors that may influence hESC self-renewal or differentiation. Functional classification and mobility variants of proteins detected in hESCs. (A) Proteins with known or suggested roles in self-renewal are shown, including Oct4, STAT3, Smad2/3 and FGF2 (Additional Table 2, "Pluripotency"). Isoforms of FGF2, and phospho-GSK3 are indicated (*). (B) Cell surface proteins are shown, including Connexin 43, E-Cad and GDNFRα (Additional Table 2, "Cell Surface"). Other functional classes of proteins are indicated in Additional Table 2. (C) A total of 42 proteins, including FGF2, HSP70 and ERK1, were found to have multiple bands in either the PowerBlot or Kinexus blots. These bands migrated closely but were sufficiently separated from other detected proteins. Bands predicted to be isoforms of the indicated protein are highlighted in some panels (*). Other functional classification of the proteins detected by the PowerBlot screen included: transcription factors (71 proteins), nucleus and nuclear transport (144), cytoskeleton (75), cell adhesion (45), MAP kinase pathway (24), protein kinase A (13), protein kinase C (20), tyrosine kinases (15), adaptors and tyrosine kinase substrates (51), protein phosphatases (17), GTPases and regulators (42), calcium signaling (23), cell cycle (87), apoptosis (61), membrane research (62), and other functions (51) (Additional File 1). Some proteins were included in multiple functional categories due to overlapping properties, such as AIM-1, which was included in the cell cycle as well as in the nucleus/nuclear transport categories. The Kinexus expression data was organized separately into cell signaling-related functional groups (Additional File 1). In addition, 35 proteins were detected by both the PowerBlot and Kinexus systems (Table 1). Proteins detected by both PowerBlot and Kinexus systems Unlike many cDNA-based gene expression assays, western blotting has the capacity to detect multiple protein isoforms due to translation of different mRNA splice variants, as well as post-translational modifications such as enzymatic cleavage, glycosylation, or phosphorylation. Examination of the blots described here identified 42 examples of multiple banding for a single target antigen (Fig. 3C). These candidates exhibited closely migrating multiple bands, which were close to their predicted size but were sufficiently separated from other proteins. For example, four closely migrating bands were observed for FGF2 (Fig. 3C, top panel), which may represent known glycosylation variants of this growth factor . Other known examples of post-translational modifications included those of HSP70, IKKgamma and ERK1. The PowerBlot and Kinexus assays identified proteins based on their expected and observed molecular weight, using combinations of antibodies that had been predetermined to detect proteins of sufficiently different sizes. Proteins known to be expressed by hESCs and also identified by these assays, included Oct4, E-CAD, Connexin 43 and Hsp70. To verify expression using a complementary approach, we performed immunoflurorescent staining for 10 proteins not previously reported to be expressed in hESCs by immunocytochemistry, using karyotypically normal BG01 cultures (Fig. 4A–K). These included ABP-280, a homodimeric actin-binding protein often associated with membrane glycoproteins; CtBP1 and CtBP2, two C terminal binding proteins that are a class of transcription corepressors; GS-28, a golgi protein; HDJ-2, a member of the DnaJ-related Hsp40 (heat shock protein 40) subfamily; L-Caldesmon, a cytoplasmic actin-binding protein; Rabaptin, a GTP-binding protein; phosphorylated-p130 Cas, a docking protein with an amino-terminal SH3 domain that may function as a molecular switch that regulates CAS (Crk-associated substrate) tyrosine phosphorylation; Ras-GAP and phosphorylated Ras-GAP (p-Y460), a protein that down-regulates the signal transducer p21; and ShcC, a protein with an N-terminal phosphotyrosine-binding domain. These proteins were all expressed by hESCs, with the expected subcellular localization (Fig. 4A–K). Oct4 was used as a positive control (Fig. 4L). These results suggested that most of the bands in the PowerBlot and Kinexus assays were likely to be correctly identified. Verification of protein expression using immunocytochemistry. (A-K) Ten proteins that were detected in undifferentiated hESCs by western blotting were also detected by immunofluorescence of BG01 cells grown in MEF-CM. Ras-GAP (pY460) is a phosphorylated form of Ras-GAP. The same antibodies were used in this analysis as in the PowerBlot assay, except phospho-p130 Cas (Tyr165). (L) Oct4 was used as a positive control. (M-R) Oct4, TNIK and p130 Cas as markers of undifferentiated hESCs. BG01 cultures were partially differentiated by exposure to 10% fetal bovine serum for 3 days. (M) Oct4 was expressed uniformly in undifferentiated cells, (P) but was downregulated in morphologically differentiated areas after 3 days in serum (arrowhead). (N) TNIK expression was localized to the cytoplasm, and (N, Q) expression appeared to be restricted to morphologically undifferentiated cells (arrowhead). (O) p130 Cas was detected in a membrane/peripheral-cytoplasmic pattern in undifferentiated cells, (R) but this distribution was substantially altered in differentiating cells with a flattened morphology, which exhibited a general cytoplasmic, or perinuclear profile. Scale bar for A-L: (A, L) 200 μm; (C, D, F, H, I, J, K) 100 μm; (B, E, G) 50 μm. Scale bar for M-R: (M, N, P, Q): 100 μm ; (O, R): 50 μm. Preliminary analyses also indicated that expression of some of these proteins was downregulated in differentiated cells, including p130 Cas and the Traf2- and Nck-interacting kinase (TNIK). TNIK is known to be involved in the inhibition of cell spreading via disruption of F-actin . Immunofluorescence was used to examine the expression of TNIK and p130 Cas during early differentiation of hESCs. BG01 cultures were partially differentiated by growth in serum containing media for 3 days. This condition generated heterogeneous populations containing Oct4cells with characteristic hESC morphology and less tightly packed, and morphologically differentiated areas, lacking expression of Oct4 (Fig 4M, P). TNIK was expressed highly in undifferentiated hESCs, and in the undifferentiated areas at day 3, but was downregulated in areas undergoing morphological differentiation (Fig 4N, Q). This may indicate that TNIK is active in hESCs and degraded rapidly upon differentiation. p130 Cas was detected in a membrane/peripheral-cytoplasmic pattern in hESCs (Fig 4O). The distribution of p130Cas was substantially altered in differentiating cells with a flattened morphology, exhibiting a general cytoplasmic, or perinuclear profile (Fig 4R). This could indicate an alteration in the function of p130 Cas as pluripotent cells differentiate. These analyses suggested that the change in expression or distribution of these proteins could be used as markers for undifferentiated hESCs. We have previously employed the Illumina Bead Array system for the large-scale profiling of gene expression in hESCs using 24,000 transcript probes . To compare proteomic and transcriptional analyses of hESCs, the levels of > 600 proteins detected using large scale blotting were correlated with the levels of transcripts detected with the Illumina platform (Additional File 3). In general, a close match between the expression level of transcript and protein was observed: transcripts for nearly all the detected proteins were also identified in the Illumina analysis, and most proteins expressed at high levels also exhibited high mRNA levels. We reasoned that a focused comparison of specific signaling pathways using a combination of proteomic and transcriptional data was likely to be much more informative than a global interrogation of hESCs. Several major signal pathways that have been suggested to be involved in self-renewal were examined to test this approach. These included the FGF, TGFβ, GSK3β/Wnt/β-catenin and Jak/Stat pathways [17,29,36-39], as well as the more recently suggested MAPK/ERK and Gap junction pathways . Correlating transcriptional and proteomic data provided direct confirmation that these pathways were present and likely functional in hESCs (Table 2). For example, FGF2 protein was expressed highly in hESCs and expression of key members of the TGFβ, Wnt, Jak/Stat and Gap junction pathways, namely Stat1, SMADs, GSK3β, β-catenin and Connexin 43, were detected in both transcriptional and proteomic databases. Signal pathways that may be active in hESCs Protein expression level: > 10,000: ++++; 5,000–10,000: +++; 1,000–5,000:++; 100–1,000: + mRNA gene expression level: > 5,000:++++; 1,000–5,000: +++; 100–1,000: ++; 30–100: + *: not included in the gene expression array This independent confirmation of known networks led us to examine other pathways that showed a similar correlation but have not been identified as key regulators of either self-renewal or differentiation, or suggest unappreciated characteristics of hESCs. Four signaling pathways (IGF, ERBB2, GPCR, and GDNF) and the tight junction complex were highlighted by this analysis (Table 2), and expression of key proteins in these pathways was confirmed. A detailed study demonstrating the importance of the IGF and ERBB2 pathways in hESC self-renewal has been performed and enabled the development of a defined medium for hESC maintenance (TCS and AJR, submitted). Tight junctions are apical cell-cell junctions found in epithelia that establish a barrier to the extracellular environment and a border for apical-basolateral polarity. While hESCs grow in colonies that are highly reminiscent of epithelia, and have been shown to be coupled by gap junctions , the formation of tight junction complexes has not been described. hESCs expressed the ZO1 and occludin tight junction proteins along cell borders as expected in polarized epithelia. The distribution of ZO1 expression changed dramatically as hESCs proliferated in culture. When tight junction complexes were disrupted by disaggreagation to single cells, only a subset of cells showed ZO1 staining 4 days after plating (Fig. 5). Continued proliferation to a confluent monolayer on day 7 was accompanied by widespread expression of ZO1, suggesting the formation of a general tight junction barrier. These cultures were undifferentiated and retained uniform expression of Oct4 protein (not shown). ERBB2 and 3 are members of the epidermal growth factor (EGF)-receptor family, which regulate epithelial proliferation via EGF-family ligands. ERBB2 and 3 transcripts are expressed by hESCs , are known to function as a heterodimer , and transmit a strong proliferative signal for hESCs by Heregulin 1β (an EGF-family ligand) (TCS and AJR, submitted). Immunofluorescence revealed general cell surface expression of ERBB2 on hESCs. Conversely, ERBB3 was highly localized to a concentrated area, and observed in cells that also expressed ZO1. Epithelial cells are known to localize ERBB receptors to the basolateral side of tight junctions, which serves to functionally separate receptors from ligands . This is a basic epithelial wound healing mechanism, whereby disruption of the tight junction barrier by injury immediately exposes receptors to extracelluar ligands . These staining patterns are also suggestive of basolateral sorting of ERBB3 in hESCs. The pathways and complexes identified by these analyses lay a framework for future functional analyses of signaling networks in hESCs. Tight junction proteins and ERBB2/3 expression in hESCs. BG01 hESCs were disaggregated to single cells using accutase and cultured in defined conditions. (A) ZO1 expression four and (B) seven days after plating, indicating progressive tight junction formation. (C) Occludin expression 5 days after plating. (D) General cell surface expression of ERBB2, in the same field of view as (A). (E) Localized expression of ERBB3, in the same field of view as (B). (F) Higher magnification of ERBB3 localization in ZO1 expressing BG01 cells, 5 days after plating. Nuclei were stained with DAPI. Attempts to harness the potential of hESCs for models of human embryogenesis and cell therapy applications will be greatly enhanced by a detailed understanding of their molecular characteristics. This includes definition of the transcripts, splice variants, and protein isoforms expressed by these cells. Post-translational modifications such as phosphorylation and glycosylation, and the receptors and signaling pathways active in the pluripotent state, or during early differentiation, also need to be determined. This should also be complemented by an understanding of epigenetic characteristics of pluripotency, including methylation, imprinting and chromatin conformation. Such a comprehensive definition of the molecular state of hESCs will enable more accurate prediction and testing of the conditions used for growth and differentiation of hESCs, by precise genetic modification or application of specific growth factor cocktails and reagents. For example, a scalable, fully defined and GMP-certified culture system will need to be developed for the eventual development of hESC-based cellular therapies. Progress has been made in defining growth factor conditions that support self-renewal [44-46], and hESC lines have been isolated in the absence of mouse embryonic fibroblasts and in animal protein free culture conditions . A more refined understanding of the biology of hESCs has contributed the development of a defined medium utilizing ligands for IGF1R and ERBB2/3 receptors to promote in self-renewal (TCS and AJR, submitted). We and others have performed transcriptional analyses of hESCs, using cDNA and oligonucleotide microarrays, SAGE, MPSS and EST enumeration. These techniques have enabled the collation and comparison of transcriptional profiles from multiple hESC lines and their differentiated derivatives and have highlighted an expanded set of hESC specific markers and signaling pathways that may regulate self-renewal or differentiation. Using pathway analysis we were also able to identify key pathways that are active in ESCs (reviewed in ). While these efforts have been highly valuable in defining the transcriptional profile of undifferentiated hESCs, they are only predictive of translation and do not shed light on post-translational events in this unique cell type. These processes may also be highly regulated, which could contribute significantly to the overall conversion of genetic information to actual protein function. We report here a proteomic analysis of pluripotent hESCs by using two large-scale western blotting systems and highlight post-translational events in undifferentiated hESCs. The expression of 545 bands was detected, potentially representing 529 proteins, or their migratory isoforms. In addition, one hundred and forty phospho-specific antibodies were used to identify 85 different phosphorylated sites, on 76 proteins in these cells. The detected proteins were annotated into functional classes representing diverse cellular processes. For example, multiple proteins were detected that have been suggested to regulate the pluirpotent state in mouse ESCs or hESCs. Defining the interplay of these multiple signaling pathways will be critical in understanding the self-renewal versus differentiation decisions of hESCs. Therefore, our data provide a powerful framework for the functional analysis of specific proteins, protein classes, or molecular pathways. In particular, the availability of antibodies for candidate proteins is a major benefit of this approach compared to 2D-gel or HPLC-MS/MS based proteomics. Although these western blotting approaches are currently more limited in scope than most large-scale cDNA based assays, detecting up to 1000 proteins compared to tens of thousands of transcripts, they have the potential to highlight translational events and post-translational modifications. By comparison, SAGE and MPSS are limited to detecting short sequence "tags" adjacent to the poly-A tail of transcripts, and may not distinguish splice variants with the same 3' exon. We detected 42 proteins with multiple closely migrating bands (Fig. 3C), suggestive of closely related isoforms or post-translational modifications such as phosphorylation. These focused proteomic approaches are therefore likely to be highly complimentary to transcriptional analyses in investigating the functional expression of the genome in hESCs and during cellular differentiation. One potential issue with this approach is that multiple antibodies are included in each lane, which could possibly lead to misidentification of bands. To demonstrate that identified proteins were expressed in hESCs, the same antibodies used in the PowerBlot assay were used to confirm expression of 10 representative proteins by immunofluorescence (Fig. 4). Furthermore, 13 proteins were detected with multiple different antibodies, and 35 proteins (Table 1) were detected in both the PowerBlot and Kinexus assays. This provided internal, or independent, confirmation of expression of these proteins. Other studies have also demonstrated the expression of several of the proteins we detected in hESCs. These include Oct4, a key marker of the pluripotent state, Connexin 43 and GSK3β, confirming the reliability of large-scale western blotting. Finally, several proteins detected by our assays were also detected in hESCs by MS approaches including Karyopherin α . Additionally, the PowerBlot assay was performed in duplicate, and was shown to be highly reproducible. This suggested that this approach should be informative when comparing hESCs to their differentiated derivatives. Two candidate proteins, TNIK and p130 Cas, were downregulated, or exhibited altered localization upon spontaneous differentiation of hESCs, respectively. This indicated that they were novel markers of undifferentiated cells and molecules that could be functionally involved with self-renewal. It is impossible in an initial manuscript to analyze and rigorously test all the predictions that could be made from comparing transcriptional and proteomic data sets. However, we did examine key features to illustrate the power of this methodology. Potential new markers for hESCs were identified, the expression and activation of proteins in key self-renewal pathways were confirmed, and a diverse range of proteins were detected and expression correlated with transcriptional analyses. In addition, we highlighted several candidate signaling pathways that may be relevant to self-renewal. Examination of tight junction protein expression indicated that undifferentiated hESCs could form polarized epithelia, which has also been recently suggested by ultrastructural analyses . Discrete localization of ERBB3 may also suggest basolateral separation of this receptor from soluble ligand. These analyses highlight that predictions from a combination of transcriptional and proteomic approaches will serve to focus the investigation of hESCs in the future. In summary, we generated a focused proteome of hESCs using large-scale western blotting and sorted the detected proteins according to function and signaling pathways. This characterization provides important basic information on expressed proteins, their isoforms and post-translational modifications, and tools for the continued investigation of the underlying molecular characteristics of hESCs. Importantly, we provide a list of tools, in the form of commercially available antibodies, which can be used to interrogate the function of these molecules in self-renewal or differentiation. For the PowerBlot analysis, enzymatically passaged BG01 hESCs were grown as described previously . These conditions were necessary to scale up the culture to generate the milligram amounts of protein lysate required for this analysis. These conditions maintain cell populations that express the appropriate markers of pluripotency and can differentiate to representatives of all three germ layers, but may lead to eventual accumulation of trisomies for chromosomes 12, 17 or X . For the Kinexus assays, BG03 hESCs were maintained in MEF-conditioned medium (MEF-CM) without the accumulation of karyotypic abnormalities as described previously . hESCs were also maintained in a defined medium as indicated. These conditions are described in detail elsewhere (TCS, AJR, submitted). Briefly, the media consisted of DMEM/F12 (Invitrogen), 2% fatty acid-free Cohn's fraction V BSA (Serologicals), 1× nonessential amino acids, 50 U/ml penicillin/streptomycin, 50 μg/ml ascorbic acid, 10 μg/ml bovine transferrin, 0.1 mM β-mecaptoethanol (all from Invitrogen), 1× Trace Elements A, B & C (Mediatech), 10 ng/ml hergulin1β (Peprotech), 10 ng/ml activinA (R&D Systems), 200 ng/ml LR-IGF1 (JRH Biosciences), and 8 ng/ml FGF2 (R&D Systems). Cultures were passaged using Collagenase IV and plated on growth factor depleted Matrigel (BD Biosciences) diluted 1:200. These cultures were karyotypically normal. To partially differentiate hESC cultures for immunostaining analysis, karyotypically normal BG01 cells were plated on matrigel and grown for three days in DMEM/F12 containing 10% fetal calf serum (HyClone), 1× nonessential amino acids, 20 mM L-glutamine, 50 U/ml penicillin/streptomycin, and 0.1 mM β-mecaptoethanol. BG01 hESC lysate was prepared in 10 mM Tris-HCl pH 7.4, 1 mM sodium orthovanadate and 1% SDS, and the PowerBlot assays were performed by BD Biosciences (BD Biosciences). Briefly, 200 μg of protein lysate was loaded in a single, gel-wide well, on a SDS-4–15% gradient polyacrylamide gel. The full PowerBlot screen consisted of five gels, which were blotted and probed with 934 antibodies, and was performed in duplicate with the same cell lysate. The gel dimensions were 130 × 100 × 0.5 mm, and proteins were separated at 150 volts for 1.5 hours, and transferred to an Immobilon-P membrane (Millipore). The membranes were blocked and clamped in a manifold that created 40 lanes across each membrane. A mix of 1 to 8 mouse monoclonal primary antibodies was added to each lane, in dilutions and combinations that had been predetermined to enable accurate identification of well-separated proteins. The predicted sizes of detectable proteins in the blots ranged from 10–540 kD, and the dilutions of the primary antibodies ranged from 1:250 to 1:15,000. The blots were removed from the manifolds, washed and incubated with goat anti-mouse secondary antibody conjugated to the Alexa680 fluorophore (Molecular Probes). The membranes were scanned using the Odyssey Imaging System (LI-COR). Molecular weight standards were generated by adding a cocktail of antibodies to P190 (190 kD), Adaptin beta (106 kD), STAT-3 (92 kD), PTP1D (72 kD), Mek-2 (46 kD), RACK-1 (36 kD), GRB-2 (24 kD) and Rap2 (21 kD) to lane 40 of gels A-D. Molecular standards for gel E were generated by adding a cocktail of antibodies to Exportin-1/CRM1 (112 kD), MCM (83 kD), Nucleoporin p62 (62 kD), α-tubulin (55 kD), Actin (42 kD), KNP-1/HES1 (28 kD) and NTF2 (15 kD) to lane 16, and antibodies to p190 (190 kD), Hip1R (120 kD), Transportin (101 kD), Calreticulin (60 kD), Arp3 (50 kD), eIF-6 (27 kD) and Rap2 (21 kD) to lane 17. Bands were detected and raw signal intensity captured automatically using the PDQuest software (Bio-Rad). To normalize the signal intensities, the total raw quantity of each band was divided by the average intensity value of the molecular standards in that image and the normalized values for the duplicate samples were averaged and expressed as normalized intensity units (i.u.). These values represent the relative signal intensity observed for each identified protein band, rather than relative expression levels of different proteins, due to differences in the efficiencies of antibody binding and dilution of the primary antibodies used. Proteins were identified based on the similarity of expected and observed band migration profiles and bands that could not be identified were excluded from the analysis. All identified proteins were verified by visual inspection, and proteins exhibiting a low signal intensity, with an averaged signal of < 1000 i.u., were verified by visual inspection using contrast enhancement in Adobe Photoshop. Bands with > 800 i.u. could typically be observed without additional image enhancement. Microsoft Excel files were generated that contained information on: gel number, lane number, antibody catalogue number (BD Biosciences), protein name, expected size, observed size, repeat 1 i.u. value, repeat 2 i.u. value, averaged i.u. value, antibody dilution, outline of protein function, Entrez gene and SwissProt identification numbers. These tables were used to list expressed proteins (Additional File 1). Preparation of the BG03 cell lysate and western blotting was performed according to published protocols . Briefly, cell lysate was prepared in 20 mM MOPS pH 7.0, 2 mM EGTA, 5 mM EDTA, 30 mM sodium fluoride, 40 mM β-glycerolphosphate pH 7.2, 20 mM sodium pyrophosphate, 1 mM sodium orthovanadate, 1 mM PMSF, 3 mM benzamidine, 5 μM pepstatin, 10 μM leupeptin, 0.5% nonidet P-40, with the final pH adjusted to 7.2. The Kinexus assays for protein kinases (KPKS-1.2A and B [76 antibodies]), phosphatases (KPPS-1.2 [27 antibodies]) and phosporylated sites in cell signaling molecules (KPSS-3.1 [37 antibodies]) were performed by Kinexus. The Bio-Rad Mini-PROTEAN 3 electrophoresis system was used to separate proteins by SDS-PAGE. For each assay, 250 μg of cell lysate was loaded in a single well spanning the width of the stacking gel, then separated through a 12.5% SDS-Polyacrylamide gel and transferred to a PVDF membrane. A 20-lane manifold was placed over the membrane and a different mixture of up to 3 primary antibodies was added to each well. The combinations of primary antibodies had been predetermined to detect well-separated proteins, avoiding crossreaction to different proteins that co-migrate. The primary antibodies were rabbit and goat polyclonal, and mouse monoclonal antibodies, diluted 1:1000. After incubation with the primary antibodies, the membranes were removed from the manifolds, washed and incubated with a mix of the appropriate secondary antibodies. The secondary antibodies were donkey anti-rabbit (at 1:5000), sheep anti-mouse (at 1:10,000) and bovine anti-goat (at 1:10,000), all conjugated with horse radish peroxidase. The membranes were washed and immunoreactive bands detected by enhanced chemiluminescence (Amersham-Pharmacia) using a FluorS Max Multi-imager (Bio-Rad). Prestained size markers (201.5, 156.8, 106, 79.7, 48.4, 37.8, 23.3, and 18.2 kD) and predetermined human-specific protein migration profiles were used to accurately identify proteins using the Kinexus immuno-reactivity identification system (IRIS) software. Detected proteins were verified by visual inspection. Immunocytochemistry and staining procedures were as described previously . Briefly, cells were fixed with 4% paraformaldehyde for half an hour, blocked in blocking buffer (5% goat serum, 1% BSA, 0.1% Triton X-100) for 1 hour followed by incubation with the primary antibody at 4°C overnight. Appropriately coupled secondary antibodies (Molecular Probes) were used for single and double labeling. All secondary antibodies were tested for cross reactivity and non-specific immunoreactivity. The following antibodies were used: ABP-280 (1:250, BD Biosciences 610798), CtBP1 (1:1000, BD Biosciences 612042), CtBP2 (1:1000, BD Biosciences 612044), GS-28 (1:2000, BD Biosciences 611184), HDJ-2 (1:100, BD Biosciences 611872), L-Caldesmon (1:2000, BD Biosciences 610660), Rabaptin-5 (1:500, BD Biosciences 611080), phospho-p130 Cas (Tyr165) (1:50, Cell Signaling Technology 4015), phospho-Ras-GAP (pY460) (1:250, BD Biosciences 612736), Ras-GAP (1:250, BD Biosciences 610043), Shc-C (1:1000, BD Biosciences 610642), Oct-4 (Santa Cruz biotechnology, 1:200 SC-8628), TNIK (1:100, BD Biosciences, 612250), p130 Cas (1:100, BD Biosciences, 610272), ERBB2 (1:100, Lab Vision, 9G6.10), ERBB3 (1:100, R&D Systems, MAB348), ZO1 (1:100, Invitrogen, 61–7300), or Occludin (1:100, Invitrogen, 71–1500). Hoechst (Invitrogen) or DAPI (Sigma) were used to identify nuclei, and Triton X-100 was omitted when staining for extracellular antigens (ZO1, occludin, ERBB2/3). Images were captured on an Olympus or Nikon fluorescence microscope. Expression levels of proteins detected by the PowerBlot assay were compared to our previous published database of multiple hESC lines examined using the Illumina bead array platform (Liu et al., 2006). Averaged transcript expression signals from the BG01, BG02 and BG03 cell lines were converted to a +/- format, based on the following criteria: A mean transcript detection level of > 5,000 was designated as ++++; 1,000–5,000 as +++; 100–1,000 as ++; 30–100 as +; and signals < 30 was represented as -. In parallel, the protein expression levels were converted to a +/- format based on these criteria: i.u. > 10,000 as ++++; 5,000–10,000 as +++; 1,000–5,000 as ++; 100–1,000 as +. In addition, genes were categorized into the same functional/signaling pathways as per the western blot database. We thank Alex Wright and Amanda McLean for technical assistance and Dr. David Madden for comments on the manuscript. This research was supported in part by the National Institute of Research Resources (9R24RR021313-04, TCS), the Intramural Research Program of the National Institute of Drug Abuse (XZ), and the Larry L Hillblom Foundation (XZ). Thomas C Schulz, Email: tschulz@novocell.com. Anna Maria Swistowska, Email: amswisto@buckinstitute.org. Ying Liu, Email: ying.liu1@invitrogen.com. Andrzej Swistowski, Email: aswistowski@buckinstitute.org. Gail Palmarini, Email: tschulz@novocell.com. Sandii N Brimble, Email: sbrimble@novocell.com. Eric Sherrer, Email: tschulz@novocell.com. Allan J Robins, Email: arobins@novocell.com. Mahendra S Rao, Email: vzeq2tcr@verizon.net. Xianmin Zeng, Email: xzeng@buckinstitute.org.
PMC2238795
Adherent Self-Renewable Human Embryonic Stem Cell-Derived Neural Stem Cell Line: Functional Engraftment in Experimental Stroke Model
Human embryonic stem cells (hESCs) offer a virtually unlimited source of neural cells for structural repair in neurological disorders, such as stroke. Neural cells can be derived from hESCs either by direct enrichment, or by isolating specific growth factor-responsive and expandable populations of human neural stem cells (hNSCs). Studies have indicated that the direct enrichment method generates a heterogeneous population of cells that may contain residual undifferentiated stem cells that could lead to tumor formation in vivo. We isolated an expandable and homogenous population of hNSCs (named SD56) from hESCs using a defined media supplemented with epidermal growth factor (EGF), basic fibroblast growth factor (bFGF) and leukemia inhibitory growth factor (LIF). These hNSCs grew as an adherent monolayer culture. They were fully neuralized and uniformly expressed molecular features of NSCs, including nestin, vimentin and radial glial markers. These hNSCs did not express the pluripotency markers Oct4 or Nanog, nor did they express markers for the mesoderm or endoderm lineages. The self-renewal property of the hNSCs was characterized by a predominant symmetrical mode of cell division. The SD56 hNSCs differentiated into neurons, astrocytes and oligodendrocytes throughout multiple passages in vitro, as well as after transplantation. Together, these criteria confirm the definitive NSC identity of the SD56 cell line. Importantly, they exhibited no chromosome abnormalities and did not form tumors after implantation into rat ischemic brains and into naïve nude rat brains and flanks. Furthermore, hNSCs isolated under these conditions migrated toward the ischemia-injured adult brain parenchyma and improved the independent use of the stroke-impaired forelimb two months post-transplantation. The SD56 human neural stem cells derived under the reported conditions are stable, do not form tumors in vivo and enable functional recovery after stroke. These properties indicate that this hNSC line may offer a renewable, homogenous source of neural cells that will be valuable for basic and translational research.To date there have been no effective treatments for improving residual structural and functional deficits resulting from stroke. Current therapeutic approaches, such as the use of thrombolytics, benefit only 1 to 4% of patients . Consequently, the majority of stroke patients experience progression of ischemia associated with debilitating neurological deficits. Recent evidence has suggested that the transplantation of cells derived from cord blood, bone marrow or brain tissue (fetal and adult) enhances sensorimotor function in experimental models of stroke , . However, the normal human-derived somatic stem cells used in these studies have a limited capacity to differentiate into the diverse neural cell types optimal for structural and physiological tissue repair and are not amenable for large-scale cell production. Unlike other sources of stem cells, hESC lines possess a nearly unlimited self-renewal capacity and the developmental potential to differentiate into virtually any cell type of the organism. As such, they constitute an ideal source of cells for regenerative medicine. The successful derivation of hESC lines from the inner cell mass of preimplantation embryos and their long-term maintenance in vitro over multiple passages has been demonstrated and standardized. Differentiation and enrichment processes that direct hESCs towards a neural lineage have also been achieved. To promote neuralization, ESCs were cultured in a defined media supplemented with morphogens or growth factors , , or cultured under conditions that promote “rosettes”, structures morphologically similar to the developing neural tube , . This neuralization process has proven invaluable in understanding the specification of hESC-derived neural tissue , , . However, the enriched neural progeny derived displayed overgrowth and limited migration after grafting into normal newborn mice and lesioned adult rat striatum , , , . The inner cores of these grafts contained tumorigenic precursor cells (reviewed in ). These findings suggest that neural cells generated by acute exposure to growth factors and/or morphogens may still be heterogeneous and potentially tumorigenic. We report an alternative method for the isolation and the perpetuation of a multipotent hNSC line from the hESCs with a primitive mode of self-renewal. We also demonstrate their long-term expansion, non-tumorigenic properties and functional engraftability in an experimental model of stroke. The hESCs were maintained and expanded on mouse feeder layer in media supplemented with bFGF (Figure 1A). After cell dissociation, a portion of the hESCs was cultured in serum free medium containing EGF, bFGF and LIF. These factors are known to stimulate the proliferation of human fetal-derived NSCs , . After 3 days in vitro (DIV), there was selective survival and growth of cells that aggregated in clusters or spheres (Figure 1B). These primary spheres were harvested and replated in fresh media. During the following week, the spheres attached to the flask and a fibroblast-like cell population began to migrate out (Figure 1C). Secondary spheres (2° spheres) were generated from these cultures and lifted off by the end of the week leaving a hollow in the middle of the attached cells (Figure 1D). The floating 2° spheres were collected and replated in fresh growth medium for 2 weeks. The cultures were then passaged by collagenase cell dissociation every 7 DIV for an additional 4 passages (Figure S1). At the 5 and 6 passages, spheres were dissociated into a single-cell suspension using trypsin-EDTA. At this stage there was a change in the hNSCs' adherent properties, and the cells began to grow as a monolayer with multiple foci of cells throughout the culture (Figure 1F). The adherent hNSC culture stained uniformly for nestin (Figure 1K), vimentin (Figure 1L) and with the radial glial marker 3CB2 (Figure 1M) indicating their homogeneity and NSC identity. Under these culture conditions, it is noteworthy that we did not observe the formation of rosettes which has been previously reported to occur under certain conditions during neuralization of hESCs , , . RT-PCR analysis confirmed that these hNSCs did not express the pluripotency transcripts Oct-4 and Nanog (Figure 1I). Furthermore, the hNSCs did not express transcripts for brachyury and foxa2, marker genes for mesoderm and endoderm respectively (negative result, data not shown). The hESCs were grown on a mouse feeder layer (A). Primary neurospheres (B) were isolated and replated to eliminate other non-neural cells. The selectively harvested secondary neurospheres (arrow in C), left behind hollow cores in the surface area (star in D) where they attached earlier. They were perpetuated for an additional 5 passages (E). These 2° spheres were then passaged using a single cell dissociation protocol (F). Arrow in F shows an example of a focus of proliferating cells. (G, H) The hNSCs were passaged every 5–7 days, as described in the Methods section. Starting from an initial population of 1 million cells, the cumulative cell number was calculated at each passage as the fold of increase×the total cell number and plotted as logarithm with base 2 in function of time (G). The cell perpetuation (G) and population doubling (H) analysis demonstrated the continuous and stable growth of the hNSCs. (I) RT-PCR analysis showing the down regulation of the pluripotency transcripts Oct4 and Nanog in secondary neurospheres and in expanded hNSCs at passage 8 (P8). (J) Cytogenetic evaluation of the SD56 hNSCs line at passage 12 by standard G-banding was performed. Twenty metaphase cells were analyzed and showed a normal female chromosome complement (46,XX). Isolated and expanded hNSCs expressed the neural precursor cell markers nestin (K), Vimentin (L) and the radial glial cell marker 3CB2 (M) in virtually all the progeny. (N-P) Clonal self-renewal ability of the isolated hNSCs through symmetrical cell division. Single (N), two-cell stage (O) and four-cell stage (P) of a hNSC proliferating over a 3-day culture period. Note the symmetrical segregation of BrdU and nestin in the progeny. Bars: (A, B, C, D) 200 µm; (E, F) 100 µm; (K–M) 20 µm; (N–P) 10 µm. To ascertain self-renewal ability under clonal conditions, a single cell suspension was plated at clonal density (1–2 cell/10 µl). To determine if the hNSCs divide symmetrically, we pulsed cultures with the thymidine analog, bromodeoxyuridine (BrdU), after plating and looked for the expression of nestin in the progeny. Cultures were fixed after 1, 2 or 3 DIV (Figure 1N–P). After 2 days, plated single cells first underwent a symmetric cell division and gave rise to daughter cells that were both positive for BrdU and nestin. The clone of cells continued to grow over the 3 DIV and all the progeny expressed nestin. BrdU labeling demonstrated that it was rare for only one daughter cell to inherit the BrdU and thus had undergone asymmetric segregation of the chromatids (Figure S2). G-band karyotyping of these hNSCs confirmed the normal, non-transformed nature of cells after 12 passages (Figure 1J). We named the derived hNSCs SD56 (intermittently referred to as SD56 hNSCs or hNSCs). Under these defined growth conditions, the hNSCs showed stable growth with a 2.7±0.2 fold increase every 5 to 7 days (Figure 1G). The population doubling at each passage averaged at 1.4±0.1 (Figure 1H). The viability of hNSCs at each passage was consistent at the approximate value of 98%. The projected cumulative cell numbers demonstrated that trillions of cells could be generated over a period of 5 months (Figure 1G). We expanded the isolated hNSCs lines for over 20 passages with a stable phenotype. An initial cell bank of 75 vials containing 2 to 5 million cells each was generated and cryopreserved. Upon removal of the mitogenic factors and plating on a coverslip pre-coated with poly-L-ornithine (PLO) substrate, the hNSCs spontaneously differentiated into neurons, astrocytes and oligodendrocytes, a property that is consistent with normal multipotent hNSCs (Figure 2). After 2 DIV, hNSCs expressed transcripts for the neural-specific genes nestin, Notch1 and neural cell adhesion molecule (N-CAM) (Figure 2A) and for the lineage specific markers β-tubulin class III, medium-size neurofilament (NF-M) and microtubule-associated protein 2 (MAP-2) for neurons, GFAP for astrocytes and myelin basic protein (MBP) for oligodendrocytes (Figure 2A). Transcripts for the GABAergic cell marker glutamic acid decarboxylase (GAD) were expressed, but transcripts for the tyrosine hydroxylase (TH), a marker for dopaminergic neurons, were undetectable. Immunocytochemical analysis (Figure 2B–F) of 10 day-old cultures demonstrated that the proportion of nestin+ cells was 36.6±2.7%, 62.5±2.8% expressed the neuronal marker TuJ1, 1.9±0.3% expressed the astrocytic marker GFAP and 7.1±0.4% were oligodendrocytes and expressed galactocerebrocide (GC) (Figure 2F). A subset (9.8±1.6%) of the GFAP+ astrocytes co-expressed nestin. Dissociated hNSCs were washed free of growth factors and plated on poly-L-onithine coated glass coverslips. Differentiated cultures were either harvested after 2 DIV for total RNA extraction and RT-PCR analysis or fixed after 10 DIV and processed for indirect immunocytochemistry. (A) Differentiated hNSCs expressed the neural-specific transcripts nestin and Notch1 as well as transcripts: for neurons (β-tubulin class III, MAP-2, NCAM and medium-size neurofilament, NF-M), for astrocytes (GFAP) and for oligodendrocytes (MBP). The hNSCs expressed transcripts for GAD, but not for TH. B, C & D, morphology of NSC-derived progeny differentiated into GFAP+ astrocytes (B), GC-expressing oligodendrocytes (C) and TuJ1+ neurons (D), DAPI (blue) show life cell nuclei. (E) Photo showing cultures double-immunostained for TuJ1 (green) and nestin (red) (DAPI, blue). (F) Quantitative analysis of immunostained 10 day-old cultures for the 3 neural cell types. Results are mean±s.e.m. of three independent experiments, each performed in duplicate. Bars: (c) 20 µm; (d, e) 10 µm. The self-renewal and pluripotent abilities of the hESCs are also associated with tumorigenic properties. Therefore, the first critical step toward developing therapeutic hNSCs is to verify that they are non-tumorigenic. The monolayer culture of SD56 hNSCs clearly demonstrated contact inhibition of growth, a normal karyotype and did not express the pluripotency transcripts Oct-4 and Nanog. Removal of mitogenic factors and continued culture on plastic resulted in cell senescence that is characteristic of non-transformed cells. To determine whether SD56 hNSCs form tumors in vivo, we transplanted them at high density (see Methods) into the forebrain and subcutaneously into the flank of nude rats. The animals were kept for a two-month post-transplant survival period. To label mitotically active cells in vivo during S-phase, the rats were injected IP with BrdU (50 mg/kg) every 8 hours during the last 24 hours before euthanasia. The transit amplifying endogenous precursors located in the subventricular zone (SVZ) were labeled (Figure S3); however, we were unable to detect grafted SD56 hNSCs co-localizing the human-specific nuclear marker hNuc and BrdU (Figure S3). No surviving SD56 hNSCs were detected in the flank of the transplanted animals suggesting that the grafted cells are not tumorigenic. To investigate the survival and functional engraftment in an injury environment, hNSCs (4×10) were transplanted into the ischemic boundary zone in the rat striatum one week after the middle cerebral artery occlusion (MCAO) was performed. Animals were euthanized two months later and the brains processed for histo-pathology and immunocytochemistry. Grafted SD56 hNSCs, identified with hNuc, demonstrated a 37.0±15.8% survival rate and a remarkable dispersion toward the stroke-damaged tissue with no sign of overgrowth or tumorigenesis. The majority of grafted cells (61.2±4.7%) migrated at least 200 µm away from the injection site and penetrated an average distance of 806.4±49.3 µm into the stroke-damaged tissue (Figure 3A–C). Immunostaining with the blood vessel marker, GluT1, revealed dilated vessels in the infarcted striatum and a close association between vessels and the grafted hNSCs (Figure 3B, 3C). The grafted cells rarely expressed the proliferation marker Ki67 (Figure 3D), 29.8±4.4% expressed nestin (Figure 3E), 6.5±0.9% expressed doublecortin (DCX) and 60.8±8.1% were TuJ1+ (Figure 3F, G). Grafted cells rarely co-expressed the astroglial marker GFAP (Figure 3H) or differentiated into CNPase-expressing oligodendrocytes (Figure 3I). Immunostaining for GAD demonstrated that 25.1±2.3% of grafted hNSCs differentiated into GABAergic neurons while less than 2% were positive for glutamate (Figure 3J, K). (A) Schematic drawing of a frontal section through the striatum illustrating the dispersion of grafted hNSCs in the focal ischemia-lesioned parenchyma (shaded area). (B, C) Photos show frontal sections through the graft in the striatum immunostained with the human specific antibodies: anti-hNuc (green in B & C) and anti-GluT1 (red, B & C) showing blood vessels and dispersed hNSCs in the graft zone. C: higher magnification of the inset in B. (D–I) Photos taken from frontal sections through the graft in the striatum double immunoprocessed for cell proliferation and neural lineage markers. (D) Note the endogenous Ki67+ cells (red cells, arrow) in the stroke damaged area and the hNuc+ (green)/Ki67- grafted hNSCs (arrowheads). (E) Examples of grafted SD56 hNSCs showing co-expression of hNuc (green) and nestin (red). (F) Confocal 3D reconstructed orthogonal images of the hNuc+(green)/DCX+(red) NSCs (arrowheads) viewed in the x-z plan on the top and y-z plan on the right. (G) Examples show the majority of grafted NSC progeny co-expressing hNuc (red) and the neuronal marker TuJ1 (green). Grafted NSCs rarely differentiate into GFAP+ astrocytes (H). In I, rare example of grafted NSC progeny becoming an oligodendrocyte identified by the expression of CNPase (green). Grafted NSCs expressed the GABAergic marker GAD65/67 (J) and rarely expressed glutamate (K). (Abbreviations: Cx: cortex, Str: striatum). Bars: (B, C) 100 µm; (D, F) 20 µm; (E, G–K) 10 µm. We asked whether transplanted SD56 hNSCs could enhance the recovery of sensorimotor function that is compromised in the stroke-injured rats. We used the cylinder test to measure the sensorimotor asymmetry in forelimb use during spontaneous exploration . To establish the baseline of the stroke-induced sensorimotor deficit, spontaneous behavior of rats in a transparent cylinder was videotaped one week after stroke (pre-transplant, Figure 4). Tests were then conducted 4 and 8 weeks after vehicle and SD56 hNSCs transplantation. Stable asymmetry in forelimb use was observed 7 days post-stroke (pre-transplant, Figure 4). Ischemic rats used their impaired forelimbs (contralateral to lesion) during lateral exploration less than they did before stroke. Transplantation of SD56 hNSCs significantly enhanced the independent use of the impaired contralateral forelimb 4 weeks post transplantation (P<0.05 vs pre-transplant). Eight weeks after transplantation the improvement in the use of the impaired forelimb was stable and significant when compared to the pre-transplant group and significantly improved in comparison to vehicle treated group at 8 weeks (Figure 4). In the vehicle treated group, the independent use of the contralateral forelimb remained impaired 4 and 8 weeks post-injection. In an independent study and using the same MCAO rat animal model, we found that transplantation of dermal fibroblasts did not improve the stroke-induced motor deficits (unpublished data). Forelimb use during spontaneous lateral exploration was measured in the cylinder test (see Method and Results sections for details). Groups of vehicle injected (n = 7) and hNSCs (n = 10) transplanted are represented. The animals were tested as described in Method section. Note the significant increase in the independent use of the impaired contralateral forelimb at 4 and 8 weeks post transplantation (P<0.05 vs pre-transplant group). The contralateral forelimb remained impaired in the vehicle treated group at 4 and 8 weeks post-injection. Bars represent percentages±s.e.m. of steps taken by the ipsilateral, contralateral and both forelimbs simultaneously. *P<0.05 vs pre-transplant group; P<0.05 vs vehicle groups. Our results indicate that a self-renewable and homogenous population of hNSCs, SD56, was derived from hESCs using defined media supplemented with a specific combination of growth factors. The SD56 hNSCs grew as an adherent monolayer culture, uniformly expressed molecular features of hNSCs including nestin, vimentin and the radial glial marker 3CB2, and did not express the pluripotency markers Oct4 or Nanog. The self-renewal property of the hNSCs was characterized by a predominant symmetrical mode of cell division. They exhibited no chromosomal abnormalities and demonstrated non-tumorigenic properties after implantation into ischemic brains and into naïve nude rat brains and flanks. Furthermore, the transplanted SD56 hNSCs migrated toward the stroke-damaged adult brain parenchyma, engrafted and improved the independent use of the stroke-impaired forelimb. Maintenance of stem cells requires symmetrical and asymmetrical cell divisions to both expand and to give rise to specialized progeny of a specific tissue (reviewed in ). In vivo, a complex cellular micro-environment or niche ensures the self-maintenance property of NSCs , , , . In vitro, defined growth factors and extracellular matrices support stem cell self-renewal , . The embryonic stem cells can propagate in a predominantly proliferative symmetrical mode, leading to homogeneous cell cultures growing relatively quickly with minimal cell differentiation , , , , . Tissue specific stem cells, however, self-renew in a predominant asymmetric mode to maintain them selves and compensate for the loss of differentiated cells due to disease or injury. Thus, NSCs isolated from developing or adult brain grow as a mixture of undifferentiated and differentiated cells due the predominant asymmetrical mode of cell division , , , , , , . A recent study has reported that a murine ESC-derived NSC line (LC1) is propagated as an adherent homogenous culture with a dominant mode of symmetrical self-renewal . A combination of EGF and FGF2 was sufficient to propagate these NSCs as an adherent monolayer. However, the SD56 hNSC line described here required the combination of EGF, bFGF and LIF for self-maintenance. Although there are morphological and molecular similarities between our hNSCs and the NSCs previously described , the methods of isolation and growth are different. In addition to the different combination of growth factors used, the hNSC line we have isolated did not go through the rosette-structure stage. The in vitro analysis of BrdU incorporation and nestin expression indicated that our hNSCs divide predominantly symmetrically. This type of growth pattern is characteristic of primitive normal stem cells undergoing mostly symmetric cell division to increase the stem cell pool at the early stage of development or during tissue regeneration after injury . RT-PCR and immunocytochemistry analysis demonstrated that these undifferentiated SD56 cells did not express any pluripotency, endodermal or mesodermal markers. Furthermore, the SD56 hNSCs described here exhibited the multipotential characteristic to differentiate into neurons, astrocytes and oligodendrocytes both in vitro and upon transplantation. Together these findings suggest that the hNSC line we isolated are appropriately programmed and share similar characteristics with the definitive NSCs of the developing brain. The SD56 hNSCs demonstrated a remarkable ability to migrate toward the stroke-damaged parenchyma of the adult rat brain. This directed migration by the majority of the grafted cells could be due to their uniform cellular composition, which results in an equal response to the host microenvironment. In previous studies, subpopulations of transplanted hESCs that were enriched in neural cells migrated in host microenvironments conducive to cell migration, such as the developing brain or in structures such as the rostral migratory stream , . In the adult lesioned brain, the grafted hESC-derived neural cells proliferated and formed rosettes , teratomas , or a cellular mass that induced a gliotic host response whereby local astrocytes demarcated the grafts . Enriched neural cultures derived from mouse and monkey ESCs have produced behavioral improvements when transplanted into animal models of stroke and brain injury. However, in these cases, the transplanted non-human ESCs also formed a mass with signs of overgrowth in the core, as well as deformations , , . ESCs plated at low density acquire neural identity within few hours after plating . Interestingly, nearly all viable cells expressed nestin, the early neural fate marker Sox1, and the pluripotency marker Oct4. Together, these studies are seminal and suggest that complete neuralization may not be achieved through certain enrichment processes, consequently the neural cells could revert to a pluripotent stage . The dispersion of the grafted hNSCs within host parenchyma may allow for more graft-host interactions that could stabilize differentiation, inhibit growth and prevent gliotic host response. In the present study, SD56 hNSC-transplanted animals demonstrated a stable improvement in the sensorimotor function when evaluated for spontaneous exploratory activity in the cylinder test that detects long-lasting deficits in forelimb use in the experimental models of stroke . The transplantation of hNSCs significantly enhanced the independent use of the impaired contralateral forelimb 8 weeks post transplantation. This is the first report demonstrating that the transplantation of hNSCs derived from hESCs can improve neurologic behavior after experimental stroke. Together, these findings are encouraging and suggest that these cells are promising for future development. In addition to their therapeutic application, the hNSCs isolated under the reported conditions offer a means to interrogate host environments and unravel mechanistic features of self-renewal, non-tumorigenicity and functional engraftability in animal models of neurological disorders. The hESC line H9 (WiCell Research Institute) was propagated every 5 to 7 days on irradiated mouse embryonic fibroblasts. The cell culture media consisted of a 1∶1 mixture of Dulbecco's modified Eagle's medium (DMEM) and F12 nutrient, 20% serum replacement (Invitrogen), 0.1 mM β-mercaptoethanol, 2 µg/ml heparin and 4 ng/ml bFGF (R&D Systems). To generate the NSCs, dissociated hESCs were cultured in a chemically defined medium composed of DMEM/F12 (1∶1) including glucose (0.6%), glutamine (2 mM), sodium bicarbonate (3 mM), and HEPES buffer (5 mM) [all from Sigma except glutamine (Invitrogen)]. A defined hormone mix and salt mixture (Sigma), including insulin (25 mg/ml), transferrin (100 mg/ml), progesterone (20 nM), putrescine (60 mM), and selenium chloride (30 nM) was used in place of serum. The medium was supplemented with EGF (20 ng/ml), bFGF (10 ng/ml) and LIF (10 ng/ml). Dissociated hNSCs were plated at a density of 100,000 cell/ml in Corning T75 (Invitrogen) culture flasks in the defined media together with the growth factors. After 5–7 DIV, the adherent culture was incubated in 0.025%trypsin/0.01% EDTA (w/v) for 1 min followed by the addition of trypsin inhibitor (Invitrogen) then gently triturated to achieve single cell suspension. The cells were then washed twice with fresh medium and reseeded in fresh growth factor-containing media at 100,000 cells/ml. This procedure was performed for 21 passages and the fold of increase and population doubling were calculated at each passage. For clonal analysis, single spheres or confluent hNSC cultures were single cell dissociated and serially diluted to yield 1–2 cell/10 µl. A 10-µl-cell suspension was then added to each of 96 or 24 well plates containing 200–300 µl of growth media. Wells containing one viable cell were marked the next day and re-scored 5 to 7 days later for cell proliferation. The differentiation of the hNSCs was performed as previously described . Dissociated hNSCs were plated at a density of 10 cells/ml in control (media/hormone mix) medium devoid of any growth factors and supplemented with 1% fetal bovine serum (FBS) on poly-L-ornithine-coated (15 mg/ml; Sigma) glass coverslips in 24-well Nunclon culture dishes with 0.5 ml/well. After 2, 7–15 DIV cultures were fixed and processed for immunocytochemistry or used for RT-PCR analysis. Long-term cultures of hNSCs were incubated at 37°C and harvested for metaphase chromosomes when the cultures were 75% confluent. Metaphase chromosomes were obtained by standard chromosome harvest methods by exposure to Colcemid at 0.1 µg/ml for 1 hour at 37°C, a 2-minute exposure to trypsin/EDTA, hypotonized with 0.057 M KCl and fixed with 3∶1 methanol:acetic acid. Slide preparations were made by dropping the fixed cell pellet onto cold, wet slides and air-dried. After incubating the slides at 90°C for 30 minutes, chromosomes were trypsin banded and then Wright/Giemsa stained for G-banding analysis. Twenty metaphase cells were completely analyzed and a normal female chromosome complement was found (46,XX). All animal experiments were conducted according to the National Institute of Health (NIH) guidelines and approved by the IACUC. Normal adult NIH-Nude rats (n = 5, 8 week-old, Taconic, Germantown, New York, United States) were used to test the tumorigenic potential of the SD56 hNSCs. Undifferentiated hNSCs from passage 9 were single cell dissociated using trypsin-EDTA and suspended at the concentration of 125,000 cell/µl in preparation for cell transplantation. Two µl of the cell suspension were stereotaxically transplanted into 4 sites within the striatum at the following coordinates: AP: +1.0 mm, ML: +3.2 mm, DV: −5.0; AP: +0.5 mm, ML: +3.0 mm, DV: −5.0; AP: −0.5 mm, ML: +3.0 mm, DV: −5.0; AP: −1.0 mm, ML: +3.5 mm, DV: −5.0 mm with the incisor bar set at 3.4 mm. The injection rate was 1 µl/min, and the cannula was left in place for an additional 5 min before retraction. For the flank tumor assay, 2×10 cells (125,000 cell/µl) were injected subcutaneously to the side of the adult nude rats. To label mitotically active cells in vivo during S-phase, the rats were injected IP with the BrdU (50 mg/kg, Sigma) every 8 hours during the last 24 hours before euthanasia. After 2-month survival time, rats were euthanized and a postmortem examination for tumor formation was performed. All animal experimentations were conducted according to the National Institute of Health (NIH) guidelines and approved by the IACUC. Sprague Dawley adult male rats (n = 17, 275g–310g, Charles River Laboratories, Wilmington, Massachusetts, United States) were subjected to one and a half hour suture occlusion of the middle cerebral artery (MCAO), as previously described and immunosuppressed 2 days before cell transplantation and daily thereafter for one week with i.p. injections of cyclosporine A (20 mg/ml, Sandimmune, Novartis Pharmaceuticals). Thereafter oral cyclosporine was used at 210 µg/ml in drinking water until euthanasia. Undifferentiated SD56 hNSCs from passages between P9 and P13 were single cell dissociated using trypsin-EDTA in preparation for cell transplantation. One week after the stroke lesion, 2 µl of the hNSCs, at a concentration of 50,000 cell/µl, were stereotaxically transplanted into 4 sites within the lesioned striatum (n = 10) at the following coordinates: AP: +1.0 mm, ML: +3.2 mm, DV: −5.0; AP: +0.5 mm, ML: +3.0 mm, DV: −5.0; AP: −0.5 mm, ML: +3.0 mm, DV: −5.0; AP: −1.0 mm, ML: +3.5 mm, DV: −5.0 mm with the incisor bar set at 3.4 mm. The injection rate was 1 µl/min, and the cannula was left in place for an additional 5 min before retraction. As a control group, we used rats subjected to ischemia and injected with the vehicle (n = 7). All animals underwent baseline motor behavioral assessment before and after the ischemic lesion, and 4 & 8 weeks after cell transplantation. The animals were killed after 2-month survival time by transcardial perfusion with phosphate buffered saline (PBS) followed by 4% paraformaldehyde. The brains were cryoprotected in an increasing gradient of 10, 20 and 30% sucrose solution and cryostat sectioned at 40 µm and processed for immunocytochemistry. Cultures were fixed with 4% paraformaldehyde for 15 min. Both cells and brain sections were rinsed in PBS for 3×5 min then incubated for 2 hrs (cultures) or overnight (brain sections) with the appropriate primary antibodies for multiple labeling. Secondary antibodies raised in the appropriate hosts and conjugated to FITC, RITC, AMCA, CY3 or CY5 chromogenes (Jackson ImmunoResearch) were used. Cells and sections were counterstained with the nuclear marker 4′,6-diamidine-2′-phenylindole dihydrochloride (DAPI). Positive and negative controls were included in each run. Immunostained sections were coverslipped using fluorsave (Calbiochem) as the mounting medium. The following antibodies were used: Anti-human Nuclei (hNuc, monoclonal 1∶100, Chemicon), Anti-TuJ1 (monoclonal 1∶100, Covance; Polyclonal 1∶200, Aves Labs); anti-GAD65/67 (polyclonal 1∶1000, Chemicon); Anti-glial fibrillary acidic protein (GFAP, monoclonal 1∶1000, Chemicon; polyclonal 1∶200, Aves Labs); Anti-galactocerebrocide (GC, monoclonal 1∶250, Chemicon); Anti-CNPase (polyclonal 1∶200, Aves Labs); Anti-Glucose Transporter type 1 (Glut-1 polyclonal, 1∶500, Chemicon); Anti-Nestin (polyclonal 1∶1000, Chemicon); Anti-vimentin (monoclonal 1∶500, Calbiochem); Anti-3CB2 (monoclonal 1∶500, Developmental Studies Hybridoma Bank); Anti-doublecortin (DCX, polyclonal 1∶250, SantaCruz Biotechnology); Anti-Ki67 (polyclonal 1∶250, SantaCruz Biotechnology). Fluorescence was detected, analyzed and photographed with a Zeiss LSM550 laser scanning confocal photomicroscope. For each animal, quantitative estimates of the total number of grafted cells were stereologically determined using the optical fractionator procedure . A computer-assisted image analysis system was performed using Stereo Investigator software (MicroBrightField, Inc.). The rostral and caudal limits of the reference volume were determined by first and last frontal sections containing grafted cells. The striatum and cortex were accurately outlined at low magnification (2.5× objective). The optical fractionator probe was selected to perform systematic sampling of the immunoreactive cell population distributed within the serial sections to estimate the population number in the volume of tissue. The counting frame of the optical fractionator was defined at 50×50 µm squares and the systematic sampling was performed by translating a grid with 200×200 µm squares onto the sections of interest using the Stereo Investigator software. The sample sites were systematically and automatically generated by the computer and examined using a 60× objective of a Nikon Eclipse TE 300 microscope. The counting frame displayed inclusion and exclusion lines and only immunoreactive cell bodies falling within the counting frame with no contact with the exclusion lines were counted. The cell dispersion was measured by counting the number of cells within 200 µm distance from the graft site. The number and distance in µm of cells dispersed beyond 200 µm was also measured. An average of 2,000 cells was counted per animal. Double labeling was determined using the confocal laser scanning microscope by random sampling of 100 or more cells per marker for each animal, scoring first for hNuc+, followed by DAPI+ nuclei and then the marker of choice. The double labeling was always confirmed in x-z and y-z cross-sections produced by the orthogonal projections of z-series. Total RNA was extracted from cultured cells using RNAeasy kit (Quiagen). Aliquots (1 µg) of total RNA from the cells were reverse transcribed in the presence of 50 mM Tris-HCl, pH 8.3, 75 mM KCl, 3 mM MgCl2, 10 mM DTT, 0.5 µM dNTPs, and 0.5 µg oligo-dT(12–18) with 200 U Superscript RNase H-Reverse Transcriptase (Invitrogen). PCR amplification was performed using standard procedure with Taq Polymerase. Aliquots of cDNA equivalent to 50 ng of total RNA were amplified in 25 µl reactions containing 10 mM Tris-HCl, pH 8.3, 50 mM KCl, 1.5 mM MgCl2 , 50 pmol of each primer, 400 µM dNTPs, and 0.5 U AmpliTaq DNA polymerase (Perkin-Elmer). PCR was performed using the following thermal profile: 4 min at 94°C; 1 min at 94°C, 1 min at 60°C, 1.5 min at 72°C, for 30–40 cycles; 7 min at 72°C, and finally a soak at 4°C overnight. The following day, 10 µl aliquots of the amplified products were run on a 2% agarose Tris–acetate gel containing 0.5 mg/ml ethidium bromide. The products were visualized through a UV transilluminator, captured in a digital format using Quantify One Gel Analysis software (Bio-Rad Laboratories) on a Macintosh G4 computer. The PCR primers specific to each transcript were as follows: GFAP, forward (F), 5′-TCATCGCTCAGGAGGTCCTT–3′ Reverse (R), 5′-CTG TTGCCAGAGATGGAGGTT–3′; MAP2 (F) 5′-GAAGACTCGCATCCGAATGG–3′, (R) 5′-CGCAGGATAGGAGGAAGAGACT–3′; MBP (F) 5′-TTAGCTGAATTC GCGTGTGG–3′, (R) 5′-GAGGAAGTGAATGAGCCGGTTA-3′ were deigned using the Primer Designer software, Version 2.0 (Scientific and Educational Software) . 18S, β-tubulin class III, N-CAM, Nestin, NF-M, Notch-1 primers . Oct4, Nanog primers . FOXa2 (HNF3B), Brachyury primers . The cylinder test was used to assess the spontaneous forelimb use during lateral exploration movement . Rats were placed in a transparent acrylic cylinder (20 cm diameter) for 5 minutes. The cylinder encourages use of the forelimbs for vertical exploration. A mirror was placed behind the cylinder so that the forelimbs could be viewed at all times. Testing sessions were videotaped and forelimb use was scored by a blinded operator. Movements scored were the independent use of the left or right forelimb or simultaneous use of both the left and right forelimb to contact the wall of the cylinder during a full rear, to initiate a weight-shifting movement, or to regain center of gravity while moving laterally in a vertical posture along the wall. Animals were tested for their baselines after stroke and 4 and 8 weeks after cell transplantation. Outcome measurement for each experiment was reported as mean±SEM. All data were analyzed using SPSS 11 for Mac OS X (SPSS Inc.). Significance of inter-group differences was performed by applying Student's t-test where appropriate. The One-Way ANOVA analysis was used to compare group differences for the forelimb use as the dependant variable and groups as the single independent factor variable. Differences between the groups were determined using Bonferroni's post hoc test. A P-value of less than 0.05 was considered to be statistically significant. We are indebted to Drs. Theo Palmer, Julie Baker, Eric Chiao and Pak Chan. We thank members of Drs. Steinberg, Palmer and Chan laboratories for their help, support and constructive comments, Dr. Athena Cherry for the karyotype analysis of the cells, Dr. Bruce Schaar for comments on the manuscript, David Kunis for laboratory support, Guo Hua Sun and Dr. Sang Hyung Lee for their outstanding technical expertise in the stroke animal model and help with the cyclosporine injections and brain sectioning and Beth Hoyte for preparation of the figures. Competing Interests: Stanford University has filed a patent application on this method of derivation and expansion of neural stem cells. Funding: This work was supported in part by Russell and Elizabeth Siegelman, Bernard and Ronni Lacroute, the William Randolph Hearst Foundation, John and Dodie Rosekrans, the Edward E. Hills Fund, Gerald and Marjorie Burnett, and NIH NINDS grants RO1 NS27292 and P01 NS37520.
PMC12115102
Stilbene Glycosides in Pinus cembra L. Bark: Isolation, Characterization, and Assessment of Antioxidant Potential and Antitumor Activity on HeLa Cells
Stilbenes are plant secondary metabolites with remarkable antidiabetic, anti-inflammatory, antimicrobial, antioxidant, antitumor, and neuroprotective properties. As these compounds are valuable constituents in healthcare products and promising drug candidates, exploring new sources of stilbenes is essential for therapeutic advancement. The present study reports the isolation of two stilbene glycosides, resveratroloside and pinostilbenoside, from Pinus cembra L. bark. Their antioxidant activity and cytotoxic effects against HeLa cells were evaluated in comparison to the raw bark extract. The structures of resveratroloside and pinostilbenoside were confirmed by nuclear magnetic resonance (NMR) and mass spectrometry (MS) data analyses. Antioxidant activity was assessed by 2,2-diphenyl-1-picrylhydrazyl (DPPH) radical scavenging and reducing power assays. Cell viability, apoptosis, cell proliferation, and cell cycle assays were used to evaluate the cytotoxic potential against HeLa cells. Resveratroloside and pinostilbenoside exhibited lower activity as free radical scavengers and reducing agents. However, they showed greater efficacy in reducing viability and suppressing proliferation in human cervical carcinoma HeLa cells. Given the promising findings of our study, the therapeutic potential of resveratroloside and pinostilbenoside should be further investigated. Keywords: Pinus cembra L., bark extract, resveratroloside, pinostilbenoside, HeLa cells, antioxidant activity, antitumor activityStilbenes are secondary metabolites with a 1,2-diphenylethylene (C6-C2-C6) structure biosynthesized in plants as a response to various stress conditions, such as bacterial, fungal, and viral infections, insect attacks, UV radiation, and heat. Most of them act as phytoalexins, playing a key role in plant defense against various phytopathogens . According to Teka et al. , 459 stilbenes from 45 plant families and 196 plant species have been identified to date. The plant families Cyperaceae, Dipterocarpaceae, Euphorbiaceae, Fabaceae, Gnetaceae, Moraceae, Orchidaceae, Pinaceae, Polygonaceae, and Vitaceae are recognized for their high stilbene content . Stilbenes exhibit significant structural diversity arising from hydroxylation, methoxylation, prenylation, glycosylation, isomerization, and oligomerization. Due to the ethylene moiety, stilbenes exist in two stereoisomeric forms: trans (E)-stilbene and cis (Z)-stilbene, the former being more stable and more common in nature . The vast structural diversity endows stilbenes with remarkable bioactivity and versatility . Resveratrol (3,5,4′-trihydroxy-stilbene), the most prominent stilbene, acts on multiple pathways involved in oxidative stress, inflammation, and cell death, such as nuclear erythroid 2-related factor 2 (Nrf2), nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB), forkhead box O (FOXO), signal transducer and activator of transcription (STAT) 1/3, phosphatidylinositol 3-kinase/protein kinase B (PI3K/Akt), c-Jun NH 2-terminal kinase (JNK), adenosine monophosphate-activated protein kinase (AMPK), insulin-like growth factor 1 receptor (IGF-1R)/Akt/Wnt, and p53 pathways . Additionally, resveratrol targets various enzymes, cytokines, chemokines, and adhesion molecules, exerting a positive impact on superoxide dismutase (SOD), glutathione peroxidase (GPx), glutathione S-transferase (GST), glutathione reductase (GR), and sirtuin 1/3 (SIRT1/3) while suppressing matrix metalloproteinase (MMP)-2, -9, myeloperoxidase (MPO), nicotinamide adenine dinucleotide phosphate oxidase 4 (NOX4), cyclooxigenases (COX), inducible nitric oxide synthase (iNOS), interleukin (IL)-1, -6, -8, tumor necrosis factor (TNF)-α, monocyte chemoattractant protein-1 (MCP-1), C-X-C motif chemokine ligand 10 (CXCL10), chemokine (C-C motif) ligand 3 (CCL3), intercellular adhesion molecule 1 (ICAM-1), and vascular cell adhesion molecules (VCAMs) . In vitro and in vivo studies have revealed the wide range of biological activities of resveratrol (anti-aging, antidiabetic, anti-inflammatory, anti-obesity, anti-osteoporosis, antioxidant, antitumor, cardioprotective, and neuroprotective) . Human clinical trials conducted in recent years have provided evidence for the benefits of resveratrol in diabetes as well as neurological and cardiovascular diseases . Pterostilbene (3,5-dimethoxy-4′-hydroxystilbene), a dimethoxy analog of resveratrol, is another promising candidate for clinical use due to its anti-inflammatory, antioxidant, and antitumor effects. Some of its targets (AMPK, PI3K/Akt, Nrf2, STAT3, SIRT1, NF-κB, TNF- α, IL-1β, -6, MMP-2, -9, COX-2, SOD) are similar to those of resveratrol. Pterostilbene inhibits transforming growth factor (TGF)-1β (involved in fibrotic diseases). In tumor cells, pterostilbene induces autophagy and modulates metastasis-associated protein 1 (MTA1)/hypoxia-inducible factor 1 α (HIF1α) and phosphatase and tensin homolog (PTEN)/Akt pathways (involved in cell proliferation, angiogenesis, and cell growth, respectively), microRNAs (miRNAs), endoplasmic reticulum stress (a limiting factor in tumor development), and epithelial–mesenchymal transition (involved in cell invasion) . Piceatannol (3,3′,4,5′-tetrahydroxy-stilbene), a hydroxylated analog of resveratrol, is another stilbene with pleiotropic effects that interacts with various pathways (Janus kinase (JAK)/STAT, Nrf2, NF-κB, FOXO, PI3K/Akt) and molecular/cellular targets (mammalian target of rapamycin (mTOR), epidermal growth factor receptor (EGFR), activator protein-1 (AP-1), p38-mitogen-activated protein kinase (MAPK), SIRT1, STAT3, TGF-β) . Pinosylvin (3,5-dihydroxy-stilbene) possesses a broad spectrum of biological activities, e.g., anti-inflammatory, antimicrobial, antioxidant, antitumor, and neuroprotective, due to its ability to interact with several targets associated with various diseases [Nrf2/antioxidant response element (ARE), PI3K/Akt-glycogen synthase kinase-3β (GSK-3β), focal adhesion kinase (FAK)/cellular Src (c-Src)/extracellular signal-regulated kinase (ERK), and p38 signaling pathways, COX-2, MMP-2, -9, iNOS, IL-6] . Another stilbene of considerable interest is isorhapontigenin (3,5,4′-trihydroxy-3′-methoxystilbene), which is notable for its anti-inflammatory, antioxidant, and antitumor activities mainly attributed to the modulation of EGFR-PI3K-Akt, NF-κB, and Nrf2 pathways . Many other stilbenes, both monomers and oligomers, have been investigated for their biological activities and the mechanisms supporting their bioactivities. According to Teka et al. , the bioactivity of 116 stilbenes has been investigated to date; the antidiabetic, anti-inflammatory, antimicrobial, antioxidant, antitumor, and neuroprotective effects are the frequently reported activities . Overall, the stilbene scaffold has shown an outstanding biological potential. Nowadays, stilbenes are valuable ingredients in dietary supplements, functional foods, and cosmetic products and some of them (resveratrol, pterostilbene) are undergoing clinical trials to evaluate their benefits in severe diseases . Therefore, the exploration of novel stilbene sources is crucial for therapeutic progress. Some of the above-mentioned stilbenes have also been identified in the bark of conifer species, for example, resveratrol in the bark of Picea abies (L.) Karst. , Picea mariana (Mill.) Britton, Sterns & Poggenb. , and Pinus koraiensis Siebold & Zucc. , piceatannol in the bark of Picea abies (L.) Karst. , pinosylvin and its monomethyl and dimethyl ethers in the bark of Picea glauca (Moench) Voss., Pinus resinosa Sol. ex Aiton, and Pinus banksiana Lamb. , and isorhapontigenin in the bark of Picea abies (L.) Karst. and Picea mariana (Mill.) Britton, Sterns & Poggenb. . To the best of our knowledge, the presence of stilbenes in the bark of Pinus cembra L. has not been investigated before. Stilbene derivatives (pinostilbene, pinosylvin and its monomethyl and dimethyl ethers, dihydropinosylvin and its monomethylether) have been reported only in the knotwood and heartwood of this species . In the present study, we report for the first time the presence of two stilbene glycosides, resveratroloside and pinostilbenoside, in Pinus cembra L. bark as well as their antioxidant activity and cytotoxicity on human cervical carcinoma HeLa cells. The isolated compounds were identified as trans-resveratroloside (trans-resveratrol 4′-O-β-D-glucopyranoside, compound 1) and trans-pinostilbenoside (trans-pinostilbene 4′-O-β-D-glucopyranoside, compound 2) (Figure 1) by nuclear magnetic resonance (NMR) spectroscopy, notably, H NMR and C NMR, and by comparison of their H-NMR and C-NMR data with the literature data . The molecular weights of compounds 1 and 2 were determined using high-resolution electrospray ionization mass spectrometry in positive ion mode (HRESIMS). Stilbene glycosides isolated from Pinus cembra L. bark [resveratroloside (1) R = H, pinostilbenoside (2) R = Me]. Trans-resveratroloside (trans-3,5,4′-trihydroxystilbene 4′-O-β-D-glucopyranoside, compound 1): amorphous, white powder; HRESIMS m/z 391.1383 [M + H] (calculated 391.1387 for C20H23O8); H NMR (400 MHz, CD3OD) δ 7.45 (2H, d, J = 8.8 Hz, H-2′,6′), 7.08 (2H, d, J = 8.8 Hz, H-3′,5′), 7.00 (1H, d, J = 16.4 Hz, H-8), 6.88 (1H, d, J = 16.4 Hz, H-7), 6.47 (2H, d, J = 2.0 Hz, H-2,6), 6.18 (1H, dd, J = 2.0 Hz, H-4), glucose 4.91 (1H, d, J = 7.6 Hz, H-1″), 3.91 (1H, dd, J = 12.0,1.6 Hz, H-6″a), 3.71 (1H, dd, J = 12.0, 5.2 Hz, H-6″b), 3.38–3.50 (4H, overlapped peaks as multiplet, H-2″,3″,4″,5″); C NMR (100 MHz, CD3OD) δ 159.9 (C-3,5), 158.8 (C-4′), 141.2 (C-1), 133.4 (C-1′), 129.0 (C-8), 128.73 (C-2′,6′), 128.7 (C-7), 118.1 (C-3′,5′), 106.1 (C-2,6), 103.1 (C-4), glucose 102.4 (C-1″), 78.3 (C-3″), 78.2 (C-5″), 75.1 (C-2″), 71.5 (C-4″), 62.7 (C-6″). Trans-pinostilbenoside (trans-3-methoxy-5,4′-dihydroxystilbene 4′-O-β-D-glucopyranoside, compound 2): amorphous, white powder; HRESIMS m/z 405.1538 [M + H] (calculated 405.1544 for C21H25O8; H NMR (400 MHz, CD3OD) δ 7.47 (2H, d, J = 8.8 Hz, H-2′,6′), 7.09 (2H, d, J = 8.8 Hz, H-3′,5′), 7.04 (1H, d, J = 16.4 Hz, H-8), 6.93 (1H, d, J = 16.4 Hz, H-7), 6.58 (1H, d, J = 1.6 Hz, H-2), 6.57 (1H, d, J = 1.6 Hz, H-6), 6.27 (1H, dd, J = 2.0 Hz, H-4), glucose 4.93 (1H, d, J = 7.2 Hz, H-1″), 3.91 (1H, dd, J = 12.0,1.6 Hz, H-6″a), 3.78 (3H, s, 3-OMe), 3.71 (1H, dd, J = 12.0, 5.2 Hz, H-6″b), 3.39–3.48 (4H, overlapped peaks as multiplet, H-2″,3″,4″,5″); C NMR (100 MHz, CD3OD) δ 162.7 (C-3), 159.9 (C-5), 158.9 (C-4′), 141.2 (C-1), 133.3 (C-1′), 129.3 (C-8), 128.8 (C-2′,6′), 128.6 (C-7), 118.1 (C-3′,5′), 107.0 (C-6), 104.7 (C-2), 101.8 (C-4), glucose 102.4 (C-1″), 78.3 (C-3″), 78.2 (C-5″), 75.1 (C-2″), 71.5 (C-4″), 62.7 (C-6″), 55.8 (3-OMe). The NMR and HRESIMS spectra are provided in the Supplementary Materials. Resveratroloside (1) and pinostilbenoside (2) were less active than the raw bark extract and positive control, catechin, in 2,2-diphenyl-1-picrylhydrazyl (DPPH) radical scavenging and reducing power assays. The antioxidant effects of the raw bark extract and catechin have been reported elsewhere . At 166.67 μg/mL, resveratroloside (1) and pinostilbenoside (2) scavenged DPPH radical by 19.88 ± 0.97% and 14.67 ± 0.51%, respectively, while the scavenging effects of the raw bark extract and catechin were significantly higher (72.32 ± 0.69% and 95.74 ± 0.05%, respectively). Similarly, at 50 μg/mL, resveratroloside (1) and pinostilbenoside (2) had weaker reducing effects (0.49 ± 0.00 and 0.09 ± 0.01, respectively) in comparison with the raw bark extract and catechin (0.90 ± 0.00 and 0.53 ± 0.00, respectively). It is noteworthy that, at 50 μg/mL, the reducing capacity of resveratroloside (1) was slightly lower than that of catechin. After 48 h incubation, resveratroloside (1) and pinostilbenoside (2) strongly reduced HeLa cell viability as compared to the control and raw bark extract. At 25 μg/mL, resveratroloside (1) and pinostilbenoside (2) decreased the percentage of viable cells to 75.67 ± 4.68% and 68.36 ± 1.14%, respectively. HeLa cell viability dropped to near 50% in the case of exposure to 50 μg/mL of resveratroloside (1) or pinostilbenoside (2). At 100 μg/mL, both resveratroloside (1) and pinostilbenoside (2) caused significant reductions in HeLa cell viability (24.82 ± 4.20% and 29.73 ± 0.41%, respectively). Approximately 90% of the HeLa cells were viable in the control and cultures treated with the raw bark extract at 25 and 50 μg/mL. HeLa cell viability was slightly reduced to 82.91 ± 0.47% by the raw bark extract at 100 μg/mL (Figure 2). Viability of HeLa cells after 48 h exposure to the raw bark extract, resveratroloside (1), and pinostilbenoside (2), assessed by 7-amino-actinomycin (7-AAD) staining ((A)—histograms; (B)—HeLa cell viability); (a) p < 0.001, (c) p > 0.05. Annexin V-fluorescein isothiocyanate (FITC)/7-amino-actinomycin (7-AAD) staining indicated no apoptosis-inducing effects for the raw bark extract. The percentages of early and late apoptotic cells after 48 h exposure to the raw bark extract (at 25 μg/mL: 0.01 ± 0.01% and 0.11 ± 0.01%, respectively; at 50 μg/mL: 0.02 ± 0.01% and 0.09 ± 0.02%, respectively) were similar to the control (0.01 ± 0.01% and 0.10 ± 0.03%, respectively). On the other hand, resveratroloside (1) triggered apoptosis in HeLa cells, the effect being more pronounced at 50 μg/mL (4.06 ± 0.58% early apoptotic cells and 7.25 ± 1.46% late apoptotic cells vs. 0.09 ± 0.03% early apoptotic cells and 0.54 ± 0.17% late apoptotic cells in HeLa cells exposed to resveratroloside (1) at 25 μg/mL). The pro-apoptotic effects of pinostilbenoside (2) were negligible (at 25 μg/mL: 0.54 ± 0.37% early apoptotic cells and 1.22 ± 0.52% late apoptotic cells; at 50 μg/mL: 0.45 ± 0.30% early apoptotic cells and 0.37 ± 0.17% late apoptotic cells). In addition, resveratroloside (1) and pinostilbenoside (2) dose-dependently augmented the percentage of dead cells (21.53 ± 4.42% and 36.52 ± 3.53% in HeLa cells exposed to 25 and 50 μg/mL of resveratroloside (1), respectively; 28.41 ± 1.63% and 44.32 ± 4.28% in HeLa cells exposed to 25 and 50 μg/mL of pinostilbenoside (2), respectively) (Figure 3). Percentage distribution of viable, dead, early apoptotic, and late apoptotic HeLa cells after 48 h exposure to the raw bark extract, resveratroloside (1), and pinostilbenoside (2) at 25 μg/mL (A,B) and 50 μg/mL (A,C), assessed by annexin V–fluorescein isothiocyanate (FITC)/7-amino-actinomycin (7-AAD) staining ((A)—cytograms; (B,C)—HeLa cell distribution); (a) p < 0.001, (b) p < 0.05; (c) p > 0.05. As shown in Figure 4, in the control, HeLa cells were distributed in high percentage in the G0/G1 phase (63.93 ± 0.94%) and, to a lesser extent, in the S and G2/M phases (17.53 ± 0.78% and 11.67 ± 0.35%, respectively); a small percentage of cells (6.03 ± 0.55%) were in the sub-G1 phase. At 25 μg/mL, the raw bark extract induced a slight increase in the percentage of HeLa cells in the S phase (19.96 ± 0.15% vs. 17.53 ± 0.78% in the control) whereas, at 50 μg/mL, modest accumulations of cells in the G2/M and sub-G1 phases were observed (15.25 ± 0.62% vs. 11.67 ± 0.35% in the control and 11.03 ± 0.75% vs. 6.03 ± 0.55% in the control, respectively). At 25 μg/mL, resveratroloside (1) caused a marked increase in the percentages of HeLa cells in the S and sub-G1 phases (29.39 ± 2.07% vs. 17.53 ± 0.78% in the control and 30.88 ± 1.86% vs. 6.03 ± 0.55% in the control, respectively). A similar behavior was noticed for pinostilbenoside (2). Following 48 h exposure to 25 μg/mL pinostilbenoside (2), the percentages of HeLa cells in the S and sub-G1 phases increased in comparison to the control (29.61 ± 1.62% and 15.56 ± 1.35%, respectively, vs. 17.53 ± 0.78% and 6.03 ± 0.55%, respectively). A strong accumulation in the sub-G1 phase was observed in HeLa cells after 48 h treatment with 50 μg/mL resveratroloside (1) and pinostilbenoside (2) (49.96 ± 0.36% and 46.56 ± 1.31%, respectively, vs. 6.03 ± 0.55% in the control). Cell cycle analysis in HeLa cells after 48 h exposure to the raw bark extract, resveratroloside (1), and pinostilbenoside (2) at 25 μg/mL (A,B) and 50 μg/mL (A,C), assessed by nuclear isolation medium—4′,6-diamidino-2-phenylindole dihydrochloride (NIM-DAPI) staining ((A)—histograms; (B,C)—cell cycle distribution); (a) p < 0.001; (b) p < 0.05. The raw bark extract (25 and 50 µg/mL) had negligible effects on HeLa cell proliferation as indicated by minor increases in the mean fluorescence intensity (X) of the 48 h-treated cells compared to the control cells (107.30 ± 0.79 and 110.91 ± 1.78, respectively, vs. 104.06 ± 1.11). In contrast to the raw extract, exposure to both concentrations of resveratroloside (1) caused significant reductions in HeLa cell proliferation (X values of 132.81 ± 0.96 and 148.65 ± 3.13, respectively, vs. 104.06 ± 1.11 in the control). In the case of pinostilbenoside (2), only the low dose (25 µg/mL) exerted an antiproliferative effect (X value of 154.96 ± 12.35 vs. 104.06 ± 1.11 in the control) (Figure 5). Mean fluorescence intensity in HeLa cells after 48 h exposure to the raw bark extract, resveratroloside (1), and pinostilbenoside (2), assessed by carboxyfluorescein succinimidyl ester (CFSE) staining; (a) p < 0.001, (c) p > 0.05. Conifer bark, the medicinal use of which dates back more than 2000 years, contains bioactive alkaloids, flavonoids, lignans, phenolic acids, proanthocyanidins, stilbenes, and terpenoids that contribute to its therapeutic potential. Various conifer bark extracts are used nowadays in the nutraceutical, food, and pharmaceutical industries because of their health-promoting effects in numerous ailments and diseases . Such extracts are Pycnogenol, Flavangenol, and Oligopin derived from Pinus pinaster Ait. bark , Enzogenol produced from Pinus radiata D. Don bark , and Abigenol originating from Abies alba Mill. bark . Pinus cembra L. (Pinaceae, Swiss stone pine, Arolla pine, cembran pine, cedar pine) is a coniferous tree growing in the Alps and Carpathian Mountains . The bark has been scarcely investigated for its chemical composition and biological activity. We have previously assessed the antioxidant potential of the raw bark extract (80% methanolic bark extract) and found EC50 values of 71.1 ± 0.5 and 26.0 ± 0.3 μg/mL in the DPPH radical scavenging and reducing power assays, respectively. The antioxidant potential of the raw bark extract is strongly associated with the total phenolic content, quantified as 299.3 ± 1.4 mg/g. In the same assays, catechin was more effective (EC50 = 5.56 ± 0.05 and 3.70 ± 0.03 μg/mL, respectively) . Other polar conifer bark extracts (80% methanolic, aqueous) scavenged the DPPH radical with EC50 values ranging from 6.46 ± 0.36 to 100.1 ± 0.1 μg/mL . In the reducing power assay, polar conifer bark extracts exhibited EC50 values ranging from 9.17 ± 0.13 to 25.32 ± 0.62 μg/mL . Overall, the EC50 values of the raw extract of cembran pine bark in the DPPH and reducing power assays fall within the range of values reported for other polar conifer bark extracts. The cytotoxic potential of the raw bark extract was further investigated. HeLa cells were used for this purpose as they have advantages over other tumor cell lines, for example, high adaptive capacity and proliferation rate . The study revealed moderate or weak cytotoxicity. At 25 and 50 μg/mL, the extract had an insignificant impact on the viability of HeLa cells, lacked apoptosis-inducing effects, and induced slight increases (≤5%) in the HeLa cell percentages in the S, G2/M, and sub-G1 phases in comparison with the control. Other polar conifer bark extracts exhibited higher activity on HeLa cells. The aqueous extract of Pinus massoniana Lamb. bark significantly inhibited HeLa cell viability and caused a substantial increase in the proportions of HeLa cells in the sub-G1 and G2/M phases . Accumulation of HeLa cells in the sub-G1 phase is considered an indicator of a pro-apoptotic effect . In other studies, the extract significantly inhibited the migration and invasion of HeLa cells, respectively, the latter effect being attributed to cathepsin B down-regulation . Pro-apoptotic effects in HeLa cells were also reported for the 80% methanolic extract of Pinus sylvestris L. bark , ethanolic extract of Pinus merkusii Jung. & de Vriese bark , and procyanidin-rich extract of Pinus koraiensis Siebold & Zucc. bark . The pro-apoptotic effects of conifer bark extracts were found to be mediated by activation of caspase-9 and -3, up-regulation of the pro-apoptotic protein Bax, and down-regulation of the anti-apoptotic protein Bcl-2 and survivin . Purification of the cembran pine raw bark extract resulted in the isolation of two stilbene glycosides, namely resveratroloside (1) and pinostilbenoside (2), the structures of which were confirmed through spectroscopic techniques. To the best of our knowledge, this is the first report on the presence of resveratroloside (1) and pinostilbenoside (2) in Pinus cembra L. bark. Both compounds were previously isolated from other conifer barks: resveratroloside from Pinus sibirica R. Mayr bark and pinostilbenoside from Pinus sibirica R. Mayr bark and Pinus koraiensis Siebold & Zucc. bark . The evaluation of the antioxidant potential of resveratroloside (1) and pinostilbenoside (2) demonstrated weaker effects than the raw bark extract, indicating a potential synergistic interaction among the components of the extract. Previous studies have reported similar synergistic interactions in pine bark extracts. Pycnogenol, a standardized extract obtained from the bark of French maritime pine (Pinus maritima Lam.), exhibits stronger biological effects than its components when tested individually . In contrast to our findings, Dar et al. (2016) reported a strong antioxidant potential for resveratroloside in the DPPH assay (IC50 = 14.0 μg/mL) . The explanation lies in the fact that Dar et al. used another experimental protocol. Resveratroloside (1) exhibited higher activity than pinostilbenoside (2) in both assays. The findings align with previous studies reporting a higher antioxidant capacity (evaluated as oxygen radical absorbance capacity, ORAC) for resveratroloside (4.01 ± 0.71 Trolox equivalents/μM) than pinostilbenoside (1.89 ± 0.25 Trolox equivalents/μM). In the same assay, the aglycones, resveratrol and pinostilbene, were more active than the corresponding glycosides, showing ORAC values of 5.26 ± 0.26 and 5.01 ± 0.27 Trolox equivalents/μM, respectively . Glycosylation and methylation negatively impact the antioxidant capacity of stilbenes by blocking the free phenolic hydroxyl groups responsible for the antioxidant activity . On the other hand, glycosylation and methylation of the stilbene hydroxyl groups might enhance other bioactivities such as tyrosinase inhibitory activity and anticancer activity, respectively . Glycosylation enhances the stability of stilbenes, while methylation increases their lipophilicity, leading to improved bioavailability . In this study, resveratroloside (1) and pinostilbenoside (2) demonstrated promising cytotoxic activity against HeLa cells. The activity was evaluated after 48 h of incubation with 25 or 50 μg/mL of each compound (equivalent to 64 or 128 μM of resveratroloside (1) and 62 or 124 μM of pinostilbenoside (2)). The selection of the concentrations to be tested and incubation time was based on previous studies investigating the cytotoxicity of resveratrol in HeLa cells . In addition, this study revealed pronounced cytotoxicity (less than 30% cell viability) for both compounds at 100 μg/mL. This served as additional support for selecting lower doses (25 and 50 μg/mL) in cell-based assays. To the best of current knowledge, this is the first study evaluating the effects of resveratroloside (1) and pinostilbenoside (2) on human cervical carcinoma HeLa cells. Resveratroloside (1) has been scarcely investigated for its antitumor potential. Only its antiproliferative effects on H2452 malignant pleural mesothelioma cells (approximately 30% inhibition at 200 μM) were reported so far . To the best of our knowledge, the antitumor potential of pinostilbenoside (2) has not been investigated before. Cytotoxic therapies eliminate cancer cells by triggering various pathways of cell death. Induction of apoptosis (programmed cell death) has been a primary objective in cancer therapy for more than 30 years . In recent years, many drugs, including natural compounds, have been reported to trigger other types of death in cancer cells such as autophagy, ferroptosis, necroptosis, pyroptosis, paraptosis, lysosome-dependent cell death, oncosis, and necrosis . Resveratroloside (1) and pinostilbenoside (2) are not the sole stilbenes that cause tumor cell death by triggering non-apoptotic mechanisms. Resveratrol was reported to induce tumor cell death by apoptosis, autophagy, necroptosis, and necrosis . Pterostilbene was found to activate apoptosis, autophagy, and necrosis in cancer cells, apoptosis being the major mechanism involved in cancer cell death . Combrestatins (diaryl stilbenoids) are effective promoters of tumor necrosis . Dysregulation of the cell cycle, a process involving cell growth, DNA replication, and cell division, is a hallmark of cancer. An important strategy in cancer therapy is the induction of cell cycle arrest. Flavopiridol, abemaciclib, and palbociclib are a few examples of antitumor drugs that suppress the cell cycle via inhibition of enzymes/proteins (cyclin-dependent kinases/cyclins) responsible for driving the progression of the cell cycle from one phase to the next one . According to this study, resveratroloside (1) and pinostilbenoside (2) impeded HeLa cell proliferation, with pinostilbenoside (2) being more active than resveratroloside (1) at 25 μg/mL. This result aligns with previous studies showing that stilbenes inhibit the proliferation of tumor cell lines. Resveratrol , piceatannol , pterostilbene , and polydatin (piceid) were reported to suppress the proliferation of various cancer cell lines (lung, prostate, breast, colorectal, liver, pancreatic, cervical, ovarian, bladder, leukemia, multiple myeloma, bone, oral, esophageal, head and neck). Regarding the impact of resveratroloside (1) and pinostilbenoside (2) on the HeLa cell cycle, both compounds induced a significant dose-dependent increase in the sub-G1 phase population. In addition, both compounds (25 μg/mL) induced cell cycle arrest at the S phase, indicating a blockage of DNA replication . As mentioned earlier, the sub-G1 population is a hallmark of apoptosis . In fact, not only apoptotic cells accumulate in the sub-G1 phase, but this phase consists of cells showing DNA fragmentation, a process observed in both apoptosis and necrosis . When exploring a potential pro-apoptotic effect, only resveratroloside (1) showed activity (approximately 11% increase in the early and late apoptotic HeLa cells following 48 h treatment with resveratroloside (1) at 50 μg/mL). The results of this study indicate that resveratroloside (1) and pinostilbenoside (2) impact the viability and proliferation of HeLa cells by triggering mainly non-apoptotic (highly likely necrotic) cell death, as well as S-phase cell cycle arrest. Similar results have been reported for other stilbene derivatives. Resveratrol was found to induce apoptosis and block cell cycle progression in the S phase in human SW480 colon carcinoma, MCF7 breast carcinoma, HCE7 esophageal squamous carcinoma, HL60 promyelocytic leukemia cells , and neuro-2a cells derived from C1300 murine neuroblastoma . Piceatannol caused apoptosis and G0/G1 phase arrest in T24 and HT1376 human bladder cancer cells . Pterostilbene was reported to induce apoptosis and S-phase arrest in MOLT4 human leukemia cells , Jurkat and Hut-78 T-cell leukemia/lymphoma cells , and diffuse large B-cell lymphoma cells , apoptosis and G1 phase arrest in HT-29 colon cancer cells and human gastric carcinoma AGS cells , and autophagy and S phase arrest in HCCC-9810 and RBE human cholangiocarcinoma cells . To conclude the cytotoxicity assays, resveratroloside (1) and pinostilbenoside (2) reduced viability (mostly via non-apoptotic routes) and proliferation (via sub-G1- and S-phase arrest) in HeLa cells. The results are consistent with previous studies on the antitumor potential of stilbenes. Resveratrol, the basic scaffold of resveratroloside (1) and pinostilbenoside (2), was reported to promote cell cycle arrest at the S phase, apoptosis, and autophagy in HeLa cells . Polydatin, a glycoside of resveratrol, namely resveratrol-3-O-β-mono-D-glucoside, reduced proliferation and induced apoptosis in HeLa cells . In this study, resveratroloside (1) and pinostilbenoside (2) exhibited comparable activity in arresting the HeLa cell cycle at the S and sub-G1 phases (at 25 and 50 μg/mL, respectively). On the other hand, pinostilbenoside (2) exhibited higher activity than resveratroloside (1) in increasing the number of dead cells through non-apoptotic mechanisms (at 25 and 50 μg/mL) and in reducing HeLa cell proliferation (at 25 μg/mL). The latter findings are consistent with earlier studies reporting increased cytotoxic activity for the methoxylated analogs of resveratrol compared to resveratrol itself . The two compounds (1 and 2) isolated in this study are stilbene glycosides. Glycosylation is known to positively impact the water solubility, intestinal absorption, and bioactivity of stilbenes . A notable example is polydatin, one of the main compounds in the roots of Polygonum cuspidatum Sieb. et Zucc., identified in other plant species across the Liliaceae, Fabaceae, and Vitaceae families. Based on its anti-inflammatory, antioxidant, and apoptosis-modulating potential, polydatin displays diverse biological activities (anticancer, antidiabetic, antimicrobial, cardioprotective, hepatoprotective, and neuroprotective effects, as well as protective effects on the gastrointestinal, renal, respiratory, and skeletal systems). A large number of studies conducted on polydatin has revealed versatility in modulating numerous targets related to oxidative stress (Nrf2 and Akt pathways, glutathione, catalase (CAT), SOD, GPx, GST, MPO), inflammation (NF-κB, phospholipase A2 (PLA2), COX-2, iNOS, TNF-α, IL-1β, IL-6, ICAM-1, MAPKs, ERK1/2, JNK1/2), and apoptosis (p53/MAPK/JNK and PI3K/Akt/mTOR pathways, B-cell lymphoma 2 (Bcl-2), Bcl-2-associated x (Bax), D-cyclins, caspase-3, cytochrome c). Clinical trials support the benefits of polydatin in chronic pelvic pain, liver diseases, inflammatory bowel syndrome, and EGFR-tyrosine kinase inhibitor (TKI)-related ashes. Moreover, various drug delivery systems (liposomes, micelles, nanoparticles, polymeric nanocapsules) have been developed to improve the bioavailability, biocompatibility, and efficacy of polydatin . The results of the present study, along with the remarkable biological potential of the stilbene scaffold and the broad bioactivity of polydatin, a resveratrol glycoside, indicate that the bioactive properties of resveratroloside (1) and pinostilbenoside (2) require further in-depth investigation. Future studies should explore the ability of resveratroloside (1) and pinostilbenoside (2) to modulate cellular signaling pathways, enzymes, and other molecules involved in the antioxidant defense and oxidative damage repair. Research on the antitumor potential (mechanisms underlying cytotoxic activity in HeLa cells, cytotoxicity against other tumor cell lines) should also continue. Exploration of additional bioactivities and development of appropriate delivery systems are crucial for the therapeutic valorization of resveratroloside (1) and pinostilbenoside (2). Diethyl ether and ethyl acetate were purchased from Sigma-Aldrich Laborchemikalien GmbH (Seelze, Germany). Acetone, (+)-catechin, deuterated methanol (CD3OD), dimethylsulfoxide (DMSO), disodium hydrogen phosphate, DPPH radical, iron (III) chloride, methanol, polyamide 6 (50–160 µm), potassium ferricyanide, tetramethylsilane, and trichloroacetic acid were acquired from Sigma-Aldrich (Steinheim, Germany). Methanol for HPLC LiChrosolv and monosodium phosphate were from Merck KGaA (Darmstadt, Germany) while n-butanol was from Chimopar SA (Bucharest, Romania). Amphotericin B, Dulbecco’s Modified Essential Medium (DMEM), ethylenediaminetetraacetic acid (EDTA), fetal bovine serum, penicillin, phosphate-buffered saline (PBS), streptomycin, and trypsin were purchased from Biochrom AG (Berlin, Germany). The CellTrace carboxyfluorescein succinimidyl ester (CFSE) cell proliferation kit was obtained from Invitrogen (Waltham, MA, USA). The annexin V-FITC/7-AAD apoptosis kit and nuclear isolation medium—4′,6-diamidino-2-phenylindole dihydrochloride (NIM-DAPI) were purchased from Beckman Coulter (Fullerton, CA, USA). Ultrapure water was obtained using the SG Water Ultra Clear TWF water purification system (Barsbüttel, Germany). The source of plant material as well as drying and storage conditions have already been described elsewhere . The dried bark fragments (150 g) were powdered and extracted with 80% aqueous methanol (1.5 L) by stirring with a magnetic stirrer for 1 h (500 rpm). The extraction was repeated twice. The combined extracts were filtered, evaporated under reduced pressure at 40 °C (Büchi R-210 rotary evaporator system, Büchi Labortechnik AG, Flawil, Switzerland), and freeze-dried (Unicryo TFD 5505 freeze-dryer, UniEquip GmbH, Munich, Germany), resulting in 23.09 g of raw bark extract (yield: 15.39%). The raw bark extract (21.16 g) was suspended in 210 mL of ultrapure water and extracted successively with diethyl ether (14 × 200 mL), ethyl acetate (10 × 200 mL), and n-butanol (8 × 200 mL). The resulting extracts were combined, evaporated under reduced pressure at 40 °C, and weighed to yield diethyl ether (8.16 g), ethyl acetate (3.45 g), and n-butanol (7.71 g) extractive fractions. The remaining aqueous phase was lyophilized yielding 1.80 g. The ethyl acetate extractive fraction (EAF, 1.5 g) was purified by open column chromatography (39 × 2.4 cm) using polyamide 6 as the stationary phase. Four separate fractions (EAF-1-4) were collected following elution with methanol–water 1:1 (v/v, 450 mL), methanol–water 7:3 (v/v, 300 mL), methanol (1100 mL), and acetone–water 7:3 (v/v, 1200 mL). Each fraction was evaporated under reduced pressure at 40 °C and lyophilized yielding 852.90, 135.30, 120.40, and 131.30 mg, respectively. Fraction EAF-1 (846.40 mg) was dissolved in methanol–water mixture (2:8, v/v, 14.5 mL) and further purified by semipreparative reversed-phase high-performance liquid chromatography (RP-HPLC) using an Agilent Technologies 1200 Series HPLC system (Agilent Technologies, Santa Clara, CA, USA) equipped with degasser (G1322A), quaternary pump (G1311A), thermostat (G1316A), and diode array detector (G1315B). The chromatographic separation was performed as follows: column: Discovery BIO Wide Pore C18 (250 × 100 mm, 5 μm), mobile phase: ultrapure water (A) and methanol (B), elution gradient: 0–5 min: 0% B, 5–15 min: 0–30% B, 15–45 min: 30–100% B, detection wavelength λ = 280 nm, flow rate 1 mL/min, and injection volume 1 mL. Twelve runs were performed and two fractions (EAF-1-1 and EAF-1-2) were collected corresponding to the retention times of 46–50 min and 59–61 min, respectively. The two fractions were evaporated under reduced pressure (40 °C) and lyophilized yielding 171 and 162.2 mg, respectively. EAF-1-1 (160 mg) was dissolved in methanol–water 1:1 (v/v, 12 mL) and purified by semipreparative RP-HPLC using the same conditions as previously mentioned except that the elution gradient was as follows: 0–35 min: 25% B, 35–45 min: 25–100% B, 45–55 min: 100% B. Compound 1 was collected in ten runs (retention time 13–31 min); the corresponding eluates were evaporated and freeze-dried yielding 58 mg. Another gradient (0–35 min: 40% B, 35–45 min: 40–100% B, 45–50 min: 100% B) applied to EAF-1-2 (160 mg dissolved in 9 mL of methanol–water 3.5:1.5, v/v) allowed the isolation of compound 2. The eluates of eight runs, corresponding to the retention time of 16–30 min, were evaporated and lyophilized producing 84 mg of compound 2. Structure elucidation of compounds 1 and 2 was performed by spectroscopic techniques including NMR spectroscopy and HRESIMS. H-NMR and C-NMR spectra were recorded on a Bruker Avance DRX 400 MHz NMR spectrometer (Bruker BioSpin, Rheinstetten, Germany). H- and C-NMR experiments were performed at 400 and 100 MHz, respectively. CD3OD was used to dissolve compounds 1 and 2. Tetramethylsilane was used as an internal standard. The chemical shift values (δ) were expressed in ppm relative to tetramethylsilane. HRESIMS spectra were acquired in the positive ion mode on an LTQ Orbitrap with LTQ MS Mass Spectrometer (Thermo Fischer Scientific, Waltham, MA, USA). Resveratroloside (1) and pinostilbenoside (2) were dissolved in DMSO to achieve a concentration of 10 mg/mL and subjected to DPPH assay as previously described . The assay is based on the ability of antioxidant agents to reduce the DPPH free radical (violet) to diphenylpicrylhydrazyl (DPPH-H, yellow) which causes a reduction in absorbance at 517 nm . Briefly, each compound (10 mg/mL in DMSO, 0.05 mL) was mixed with a solution of DPPH radical in methanol (2.95 mL, A517nm = 1.00 ± 0.05). The absorbance of the latter was determined before the compound was added (Astart) and after a 5 min reaction time (Aend). The DPPH radical scavenging activity (%) of each compound was calculated as 100 × (Astart − Aend)/(Astart) . Resveratroloside (1) and pinostilbenoside (2) were dissolved in DMSO (8.25 mg/mL). The reducing power assay was performed as previously reported . The assay evaluates the ability of antioxidant agents to reduce potassium ferricyanide to potassium ferrocyanide, the latter being quantified as Perl’s Prussian Blue after reaction with ferric chloride (λ = 700 nm) . In brief, each compound (8.25 mg/mL in DMSO, 0.1 mL) was mixed with 0.2 M phosphate buffer (pH = 6.6, 2.4 mL) and 1% potassium ferricyanide (2.5 mL) followed by 20 min incubation at 50 °C. The reaction mixture was treated with 10% trichloroacetic acid and centrifuged (3000 rot/min, 10 min). An aliquot of the upper layer (2.5 mL) was mixed with ultrapure water and 0.1% ferric chloride (2.5 and 0.5 mL, respectively). The absorbance at 700 nm was recorded after 90 s. Higher absorbance values indicate stronger reducing activity . For the experiments, stabilized cultures of human cervical cancer cells (HeLa, ATCC CCL2), uncontaminated with Mycoplasma sp., were used. HeLa cells were grown in DMEM containing fetal bovine serum (10%), streptomycin (100 µg/mL), penicillin (100 UI/mL), and amphotericin B (50 µg/mL) in a humidified incubator (5% CO2) at 37 °C . When HeLa cells reached confluence, they were detached from the plate using a solution containing 0.25% trypsin and 0.02% EDTA, centrifuged at 1800 rpm for 2 min (Sigma Sartorius 2–16 PK centrifuge, Gottingen, Germany), and resuspended in DMEM to provide an optimal cell density (1.5 × 10 cells/mL). HeLa cells were further seeded in cell culture wells (0.3 mL/well) and stored in the incubator at 37 °C . To assess cell viability, HeLa cells were stained with 7-AAD, a fluorescent dye able to penetrate only the dead cells and intercalate between guanine and cytosine bases of DNA . After 24 h, when the HeLa cell monolayer was formed, the culture medium (0.3 mL) was discarded and replaced with medium containing either the raw bark extract, resveratroloside (1), or pinostilbenoside (2) (at 25, 50, and 100 µg/mL), or with medium containing the sample solvent (control). After 48 h of treatment, HeLa cells were quickly detached by trypsinization, centrifuged, washed twice with cold PBS, resuspended in the binding buffer (50 µL), and stained with 7-AAD (10 µL/sample). After 30 min cooling on ice in the dark, a volume of 250 µL of the binding buffer was added to each sample; the sample was immediately analyzed by flow cytometry using the blue laser for fluorochrome excitation. The fluorescence was collected using the 670 LP filter (FL3 detector). The flow cytometry data were collected as LMD files and analyzed using Flowing Software (Cell Imaging Core, Turku Centre for Biotechnology, Åbo Akademi University, Turku, Finland). To investigate whether the decrease in HeLa cells’ viability was related to apoptosis, a similar protocol was used except that HeLa cells were successively stained with annexin V-FITC (5 µL/sample) and 7-AAD (10 µL/sample). The fluorescence was collected using a 525 BP filter (FL1 detector) for annexin V-FITC-stained HeLa cells and a 670 LP filter (FL3 detector) for 7-AAD-stained HeLa cells . In contrast to 7-AAD, annexin V binds to phosphatidylserine expressed on the surface of apoptotic cells. Therefore, annexin V-FITC and 7-AAD staining discriminates viable (annexin V-FITC negative, 7-AAD negative), dead (annexin V-FITC negative, 7-AAD positive), early apoptotic (annexin V-FITC positive, 7-AAD negative), and late apoptotic (annexin V-FITC positive, 7-AAD positive) cells . The raw bark extract, resveratroloside (1), and pinostilbenoside (2) were tested at 25 and 50 µg/mL. HeLa cells exposed to the sample solvent were used as control. The data were collected and analyzed as described in Section 4.6.2. To investigate the effects on the cell cycle, DNA content in HeLa cells was quantified using NIM-DAPI staining. DAPI, a fluorescent dye, binds to DNA regions rich in adenine and thymine, the fluorescence intensity being proportional to the DNA amount in cells. An increase in the DNA content in a cell cycle phase indicates cell accumulation and phase arrest . After 48 h treatment with the raw bark extract or stilbene glycosides (at 25 and 50 µg/mL), HeLa cells were collected by trypsinization, resuspended in DMEM supplemented with 10% fetal bovine serum, and pelleted by centrifugation (1800 rpm, 4 min). The pellets were washed twice with cold PBS, resuspended in NIM-DAPI, and allowed to stain overnight at 4 °C. For the HeLa culture exposed to the sample solvent (control), bark extract, or stilbene glycosides, 20,000 cells were analyzed by flow cytometry, using a 100 W mercury arc lamp, a 355/37 exciter, and a 460 BP filter for fluorescence collection and linear amplification . The data were collected and analyzed as described in Section 4.6.2. The effects of the raw bark extract and resveratrol derivatives on HeLa cell proliferation were monitored by flow cytometry using staining with a fluorochrome (CFSE). The assay is based on the progressive decrease in fluorescence of CFSE-stained cells as a consequence of successive divisions and equal distribution of CFSE among daughter cells. Higher fluorescence of treated cells compared to the control indicates cell division blockage . HeLa cells were grown in monolayer culture (2 × 10 cells/mL). The cell monolayer was detached by trypsinization, washed twice with cold PBS, centrifuged at 1800 rpm for 4 min, and resuspended in PBS to a density of 1 × 10 cells/mL. CFSE was added to the cell suspension to a final concentration of 1.5 µM with further incubation at 37 °C for 10 min. The staining was blocked by the addition of 100% fetal bovine serum. After another incubation (37 °C, 10 min), HeLa cells were centrifuged (1800 rot/min, 4 min), washed three times with DMEM supplemented with 10% fetal bovine serum, and resuspended in DMEM . The stained HeLa cells were further seeded in 24-well plates (5 × 10 cells/well) and incubated for 24 h followed by 48 h treatment with the raw bark extract or each isolated compound (at 25 and 50 µg/mL). Following the treatment, CFSE-stained HeLa cells were collected by trypsinization and analyzed on a Beckman Coulter Cell Lab Quanta SC–MPL flow cytometer (Beckman Coulter, Brea, CA, USA), using a 488 nm (blue) laser (CFSE excitation) and 525 nm bandpass filters (fluorescence collection). The control consisted of HeLa cells treated with the sample solvent. The data were collected and analyzed as described in Section 4.6.2. Antioxidant assays were performed in triplicate, and the results were expressed as mean ± standard deviation (SD). HeLa cell-based assays were performed in triplicate; the results were expressed as mean ± standard error (SE). The differences between the results were tested using one-way ANOVA with Tukey’s HSD test (SPSS version 18.0); p < 0.05 was considered statistically significant. In this study, resveratroloside (1) and pinostilbenoside (2) were first isolated from Pinus cembra L. bark. This is the first report of these compounds in this species. Their structures were confirmed by H-NMR, C-NMR, and HRESIMS. Compared to the raw bark extract, resveratroloside (1) and pinostilbenoside (2) showed lower activity as free radical scavengers and reducing agents. However, they were more effective in reducing the viability and suppressing the proliferation of human cervical carcinoma HeLa cells. At 25 µg/mL, both compounds induced S-phase cell cycle arrest in HeLa cells. At 25 and 50 µg/mL, they significantly reduced the viability of HeLa cells, mainly through non-apoptotic mechanisms. Glycosylated stilbene scaffolds have great potential for therapeutic applications, so further studies are needed to assess the bioactive potential of resveratroloside (1) and pinostilbenoside (2). C.L. and A.M. gratefully acknowledge Pincu Rotinberg, Institute of Biological Research, Iasi, Romania, for the support provided in conducting the cytotoxicity studies on HeLa cells. L.N. gratefully acknowledges the support from the European Regional Development Fund—Project ENOCH (No. CZ.02.1.01/0.0/0.0/16_019/0000868) and The Czech Agency Grants—Project 23-05474S and Project 23-05389S. The original contributions presented in this study are included in the article/Supplementary Materials. Further inquiries can be directed to the corresponding author. Author Cosmin-Teodor Mihai was employed by the company Medical Investigations Praxis SRL. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The study benefited from partial support through a grant from the Core Program, developed with the support of the Romanian Ministry of Research, Innovation and Digitization, contract no. 7N/2023, project PN 23020402. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. The original contributions presented in this study are included in the article/Supplementary Materials. Further inquiries can be directed to the corresponding author.
PMC1885650
Real-time PCR mapping of DNaseI-hypersensitive sites using a novel ligation-mediated amplification technique
Mapping sites within the genome that are hypersensitive to digestion with DNaseI is an important method for identifying DNA elements that regulate transcription. The standard approach to locating these DNaseI-hypersensitive sites (DHSs) has been to use Southern blotting techniques, although we, and others, have recently published alternative methods using a range of technologies including high-throughput sequencing and genomic array tiling paths. In this article, we describe a novel protocol to use real-time PCR to map DHS. Advantages of the technique reported here include the small cell numbers required for each analysis, rapid, relatively low-cost experiments with minimal need for specialist equipment. Presented examples include comparative DHS mapping of known TAL1/SCL regulatory elements between human embryonic stem cells and K562 cells.Mapping the location of DNaseI-hypersensitive sites (DHSs) remains central to developing our understanding of transcriptional regulation. Elements with a range of transcriptional regulatory functions have been identified initially as DHSs. These include transcriptional enhancers (1,2) and repressors (3,4) as well as chromatin insulators and barrier elements (5,6). A number of techniques have been published recently that permit the mapping of DHSs without the need for Southern blotting (7–13). These include high-throughput sequencing of cloned DNA libraries derived from DNaseI-digested chromatin (8,9), and a number of different approaches that use genomic tiling path arrays to map DHSs (7,11,13). While these approaches have the advantage of covering large genomic regions with a limited number of experiments, they are inherently costly and less applicable to the rapid DHS mapping of specific genomic sites in a range of cell types. An alternative approach that does permit the targeted semi-quantitative DHS mapping of specific loci has used quantitative real-time PCR to map sites from digested DNA (10). This technique depends on the quantification of relative loss of PCR signal observed when PCR primers amplify across regions of digested DNA compared with amplification of undigested DNA. This approach has been reported as showing good sensitivity, but a limitation is the large number of PCR reactions required to quantify the calculated loss of signal with significant certainty. In this article, we present an alternative method to identify DHS using real-time PCR by adapting the protocol we have used to map DHS using genomic tiling path arrays (11). We have previously demonstrated the high sensitivity and specificity of the basic protocol with regard to identifying DHS across large genomic regions (11). Here, we detail the laboratory protocol that permits the rapid comparative mapping of known and candidate DHS between different cell types using real-time PCR. As examples, we include comparative DHS mapping of regulatory elements located across the extended TAL1 (T-cell acute lymphocytic leukaemia-1, also known as SCL (stem cell leukaemia)) locus in human embryonic stem (hES) cells and the leukaemia cell line K562. The basic protocol is outlined in Figure 1. Nuclei are extracted from living cells, then digested on ice for 1 h with a range of DNaseI concentrations, as detailed in the ‘Materials and methods’ sections. Following RNaseA and proteinaseK treatment, the DNA is extracted and run on a 1% agarose gel to check for the size of digested DNA. The gel in Figure 1A shows the DNA from mouse thymocyte nuclei digested with 0, 40 and 120 units of DNaseI. Maximal enrichment at DHS is usually observed with samples that are not over-digested. In the experiment shown, maximal enrichment at a known control DHS was seen with 40 units DNaseI (real-time PCR quantification shown in Figure 2B). The DNA is then blunt-ended using T4 polymerase (Figure 1B) and ligated with an asymmetric double-stranded linker. After extraction, DNA is amplified using a biotinylated linker-specific primer and 35 thermal polymerase cycles, as detailed in ‘Materials and methods’ section. As the linker will ligate to both ends of digested DNA, the amplification will represent a mix of primer extension and PCR, depending on the length of DNA amplified. As has been previously reasoned (13), one double strand of DNA in a region of DNaseI hypersensitivity is more likely to be digested twice within a short distance than non-hypersensitive DNA. This will lead to the preferential amplification of DNA from regions of DHS. The amplified DNA is then extracted using para-magnetic streptavidin beads, which provides a DNA library representative of whole-genome DHS. Agarose gel electrophoresis of the DNA recovered from the beads confirms that the vast majority of these products are between 300 and 500 base pairs in length (Figure 1B). (A) Nuclei are isolated from cells, and aliquots are digested with a range of DNaseI as detailed in ‘Materials and methods’ section. The DNA is extracted and run on 1% agarose gels using gel electrophoresis (right panel). With the example shown, the 40-units sample gave maximal enrichment at a housekeeping promoter using the complete protocol. (B) The digested DNA is blunt-ended with T4 polymerase and ligated to the LP21–25 linker as detailed in ‘Materials and methods’ section. Following amplification with the biotinylated LP25 primer, the extracted DNA template represents a library of whole-genome DHS. When samples of this library are visualised using gel electrophoresis (bottom panel), the majority of products are between 300 and 500 bp in size. (A) The left-hand panels show the Sybr-green real-time PCR profiles (fluorescence versus cycle number) of 5-fold dilutions of quantified mouse genomic DNA standards for two primer pairs within (ppA) and 3′ (ppB) of the Stil promoter. The right-hand panel shows that DNA template derived from the 40 units DNaseI-treated sample amplified nearly six PCR cycles in advance of the 0 DNaseI-treated sample with ppA. No difference in amplification kinetics was seen between the 0 and 40 samples with ppB. (B) Quantification of samples relative to genomic standards using primers for the ‘housekeeping’ Hmbs promoter shows maximal enrichment with 40 units DNaseI treatment for primary mouse thymocytes (left panel) compared with maximal enrichment with 120 units DNaseI for the mouse T-cell line, BW5147 (right panel). Figure 2 documents the quantification of specific DNA sites from different mouse-cell-derived DNA libraries. Figure 2A shows the quantification of material from two primer sets, which are within (primer pair A) and 3′ (primer pair B) to the mouse Stil promoter. The left-hand panels of Figure 2A show the Sybr-green real-time fluorescence profiles using serial 5-fold dilutions of quantified mouse genomic DNA standards with primers A and B. A calculated standard curve then permits the quantification of DNA from this sequence in library samples. The right-hand panels of Figure 2A show the quantification of samples from 0 and 40 units DNaseI-treated material using primers A and B. While there is no difference in amplification between the samples using primer pair B, with primer pair A, the 40 units DNaseI sample amplifies ∼5.5 PCR cycles before the 0 units sample. This equates to greater than 40-fold enrichment at the Stil promoter compared with no observed enrichment downstream of the promoter. Figure 2B shows the quantifiable differences in enrichment at the porphobilinogen deaminase (Hmbs) promoter between primary mouse thymocytes and the mouse T-cell line BW5147. This representative primary mouse cell experiment was performed with the DNA shown in Figure 1A, which confirms that with these cells, maximal enrichment is observed from DNA that is not over-digested. The comparison of the thymocytes with the BW5147 cells illustrates another common finding that, in our experience of primary cell experiments, maximal enrichment is often obtained using lower amounts of DNaseI compared with cell lines. As the protocol generates template by primer extending away from a DNaseI-digestion site, as well as PCR amplification between DNaseI-digestion sites, there is the potential for the genomic size of the real-time DHS to be larger than sites revealed by Southern blotting. There is also potential for signal to be lost at the most ‘open’ stretch of DNA, due to over-digestion with DNaseI. These issues are addressed in Figure 3. The upper panel shows a Southern blot of the human STIL (SCL/TAL1 interrupting locus) promoter in K562 cells. The BglII restricted fragment is probed from the 3′ end, as shown in the upper panel. This reveals a central DHS ∼500 bp wide, with a suggestion of weaker hypersensitivity for a few hundred base pairs 5′ and 3′ to the central region. The lower panel shows real-time PCR data from K562 material using 10 primer sets, each generating an amplicon ∼120 bp long. This permits the quantification of enrichments over 1200 bp, centred around the STIL transcription start site. The lower panel represents the mean ± SD enrichment from three independent biological replicates from K562 cells. The black bar denotes the location of the ‘core’ hypersensitive site as defined by Southern blotting. There is good correlation between the location of the DHS between the two techniques. The 5′ extension of enrichment seen in the lower panel appears to reflect the weaker 5′ extension seen in the Southern blot in the upper panel. A dip in enrichment is observed in the lower panel over the previously mapped transcription-factor-binding sites (14), which may represent over-digestion of the most accessible core region of the DHS. The upper panel shows a Southern blot DHS map of the human STIL promoter in K562 cells using a 3′ probe (grey box labelled ‘P’) to analyse the BglII-restricted fragment (cut −4 kb and +4.2 kb relative to the STIL transcription start site). This revealed a core DHS ∼500 bp wide. The lower panel represents the mean ± SD real-time PCR quantification of enrichments from three biological replicates of K562 material relative to known genomic standards, using 10 separate primer sets that together amplify ∼1200 bp centred around the STIL transcription start site. The black bar represents the approximate location of the DHS, as identified by the Southern blot experiment in the upper panel. The real-time PCR amplification profile closely represents the Southern blot profile, although a central dip in real-time PCR signal is observed over the site of transcription factor binding. As examples of relative enrichments at different regulatory elements, we present a comparison of enrichments across the extended TAL1 locus from human embryonic stem (hES) cells and K562 cells in Figure 4B. Relative quantification of mRNA expression using real-time PCR shows that both of these cell types express STIL, but only K562 cells express TAL1 (Figure 4A). TAL1 expression is critical to the establishment of haematopoiesis in a developing embryo (15). Post embryogenesis, TAL1 expression is maintained in the non-lymphoid haematopoietic system, although in addition, expression is observed in a range of non-haematopoietic tissues, including endothelium and brain (16–18). Over the past 10 years, we and others have dissected the regulatory elements that direct the expression of TAL1 to different tissues (11,19–31). These include the 3′ haematopoietic stem cell enhancer (+19) (24,30), the 5′ endothelial-haematopoietic enhancer (−4) (31), the 3′ erythroid enhancer (+40 in mouse, +50 in human) (28), and a number of neuronal elements (25,26). The location of the STIL promoter has been previously mapped (14), although no other STIL regulatory elements have yet been identified. (A) Quantitative reverse transcriptase PCR quantification of STIL and TAL1 expression in hES and K562 cells. Expression was normalised to β-actin and plotted as detailed in ‘Materials and methods’ section. Both cell types express STIL, but TAL1 transcripts were not detectable (U) in hES cells. (B) Quantitative real-time PCR DNaseI profiles from hES and K562 cells using nine primer sets across the TAL1 locus. Primer set numbers refer to approximate kilobases (kb) relative to the start of TAL1 exon 1a. Primer sets corresponding to known regulatory elements are highlighted in red. Profiles represent the mean ± SD enrichment of material derived from two independent hES cell biological replicates, and two independent K562 cell cultures. Each bar represents the mean ratio of enrichment obtained from DNaseI-treated samples versus DNaseI-untreated samples for each cell type. With both cell types, no enrichment is seen at the four control sites (−16, +5, +22, +70 kb). K562 cells show maximal enrichments at the TAL1 regulatory elements and the STIL promoter, while hES cells show the highest enrichments at the STIL promoter and +50 enhancer, with minimal enrichment at the other TAL1 regulatory elements. Figure 4B shows the mean ± SD ratio (DNaseI-treated enrichment/DNaseI-untreated enrichment) of quantitative enrichments from two independent hES cell biological replicates and two independent K562 biological replicates, using primer sets from different locations as indicated in the figure. Primers indicated in red correspond to known regulatory elements. The hES cell cultures were maintained as detailed in methods. Both cell types show significant enrichment at the STIL promoter. Four control regions were selected: −16 kb, upstream of TAL1; +5 and +22 kb, within the TAL1 locus; +70 kb, downstream of TAL1. These sequences were chosen as control regions as they show limited homology between a number of species (26) and were considered highly unlikely to represent regulatory elements. All four regions showed no enrichments between DNaseI-treated and -untreated samples in either the hES cells or K562 cells. The most striking differences between hES cells and K562 cells were found at the TAL1 regulatory elements. With K562 cells, there is prominent enrichment at the TAL1 promoter 1a, and the −4, +19 and +50 enhancers. These cells express high levels of TAL1 transcripts, and were originally derived from a patient with blastic transformation of chronic myeloid leukaemia. K562 cells are relatively undifferentiated, although they have a partial erythroid phenotype. This potentially explains the marked enrichment seen at the +19 stem cell enhancer and the +50 erythroid enhancer. This is similar to data obtained from primary mouse erythroid cells (day 14.5 foetal liver), which show the highest enrichments at the mouse +40 enhancer (homologous to the human +50) (data not shown). In contrast, the hES cells show markedly reduced enrichment at the TAL1 regulatory elements when compared with K562 cells. Although there is enrichment at the +50 enhancer, there is minimal enrichment at the −4, promoter 1A and +19 elements. Reduced/absent accessibility of regulatory elements to DNaseI digestion is consistent with the lack of expression of TAL1 in hES cells (Figure 4A). As ES cells differentiate to form embryoid bodies, expression of TAL1 is rapidly switched on, appearing from day 3 of mouse ES cell differentiation (32). The relative accessibility of the +50 enhancer may reflect a poised chromatin state in hES cells, which permits a rapid response to the changing transcription factor environment that accompanies differentiation. The development of techniques that permit the rapid comparative mapping of DHS between different cell types will greatly facilitate the study of transcriptional regulation in both normal and diseased cells. Recently published high-throughput techniques that map DHS sites using high-throughput sequencing (8,9) and genomic array tiling paths (7,11,13) have clear advantages of scale over more targeted approaches. Major disadvantages of these approaches include cost and lack of focus. This makes them less suitable for many laboratories that want to assess the chromatin accessibility of a number of defined or presumed regulatory elements in a range of cell types. A real-time-PCR-based approach to DHS mapping has, therefore, a number of potential advantages for researchers interested in specific regulatory questions at defined loci. Real-time PCR is relatively inexpensive compared with the large-scale techniques, and permits a rapid, focused DHS analysis of defined regions of DNA from multiple cell types. It also provides flexibility, as any genomic region can be analysed from the DNA library derived from the DNaseI-treated samples by designing further real-time PCR primer sets. We have previously shown that our basic technique for amplifying a library of DHS generates a template representative of known DHS with excellent sensitivity and specificity (11). Although the experiments presented in this paper were each performed using 5 million cells/digestion condition, we have obtained reproducible data using 5-fold fewer cells as starting material. We feel our technique can deliver acceptable specificity and sensitivity for DHS mapping with small numbers of cells, and will therefore be of use to those researchers working with limited numbers of primary cells. An alternative approach using real-time PCR to define DHS has been previously published (10). The two approaches differ in that Dorscher et al. quantify the DHS through the loss of PCR signal obtained from DHS when DNA is digested, whereas our approach uses real-time PCR to quantify a gain of signal observed from DHS. Dorscher et al. report excellent sensitivity using their approach to map DHS. However, the technique depends on large numbers of comparative quantitative real-time PCR reactions across a region in both digested and undigested material, in order to quantify the loss of enrichment. One advantage of our technique published here is that data can be obtained using far fewer quantitative PCR reactions. The technique is highly reproducible, with relatively little variation in quantifiable enrichments observed between different biological replicates. Moreover, we demonstrate tissue specificity, with variable enrichment at known regulatory elements between different cell types. The technique published here permits the rapid comparative analysis of DHS between different cell types from relatively small numbers of cells. It will have potential use for researchers across a broad spectrum of biology for the study of transcriptional regulation in both healthy and diseased tissues. Cell lines were maintained in culture as previously described (11). Care was taken to ensure maximum viability of cells when taken for experiments. The primary thymocytes used for Figure 2 were prepared by gentle physical disassociation of a whole thymus gland into PBS supplemented with 2% FCS. Cells were filtered to ensure that a single-cell suspension was taken forward for nuclei isolation. Human embryonic stem cells (hES) were grown in chemically defined media in the presence of Activin and FGF2, as detailed previously (33). In these conditions, hES cells remain homogenously undifferentiated. Up to 3 × 10 cells were washed in ice-cold PBS and resuspended in 2 ml of ice-cold cell lysis buffer [300 mM sucrose, 10 mM Tris pH 7.4, 15 mM NaCl, 5 mM MgCl, 0.1 mM EGTA, 60 mM KCl, 0.2% NP-40, 0.5 mM DTT, 0.5 μM spermidine, 1× protease inhibitor (complete, Roche)]. After 5 min, the lysed cells were spun at 500 g for 5 min at 4°C with brakes off. After careful removal of supernatant, the nuclei were gently resuspended in 200 μl of ice-cold reaction buffer (20 μl 10× DNaseI buffer, 4 μl glycerol, 176 μl water) using pipette tips with cut off ends. The nuclei were spun again at 500 g for 5 min at 4°C and, following supernatant removal, were resuspended in 30 μl of reaction buffer per DNaseI condition. For example, if 2 × 10 cells were taken to look at four different conditions (e.g. 0, 20, 60, 120 units DNaseI), the nuclei were resuspended in 120 μl of reaction buffer. Separate 30 μl aliquots were then taken and gently mixed with 70 μl of DNaseI mix (see Table 1) on ice. This made a final digestion volume of 100 μl for each sample, which was left to incubate for 1 h on ice in the cold room. DNaseI mixes of 70 μl aliquots were prepared as detailed The resuspended nuclei were added as 30 μl aliquots to make a final digestion volume of 100 μl. After 1 h, 700 μl of nuclei lysis buffer (100 mM tris HCL pH 8, 5 mM EDTA pH 8, 200 mM NaCl, 0.2% SDS) was added to each sample with 50 μg proteinase K. Following gentle mixing with inversion, the lysed samples were incubated at 55°C for 1 h. RNaseA (10 μg (Ambion)) was then added to each sample and further incubated at 37°C for 30 min. DNA was then extracted using standard phenol–chloroform techniques. Care was taken to use cut-off tips and very gentle pipetting to reduce non-specific DNA sheering. Following precipitation, DNA was resuspended in 200 μl of 0.1 TE and quantified using spectophotometry. Samples (1 μg) were analysed using gel-electrophoresis, as shown in Figure 1A. Following quantification, 7.5 μg of DNA was taken for each sample condition. This was blunt-ended using T4 polymerase (10 μl 10× buffer, 0.5 μl 25 mM dNTP, 3 μl BSA, 1 μl T4 polymerase (3 U/μl NE Biolabs), 85.5 μl of water/DNA in 0.1 TE). The samples were mixed gently on ice, then put at 12°C for 16 min. The reaction was stopped with excess EDTA (4 μl 0.5 M) followed by 75°C for 10 min. The DNA was further extracted using phenol–chloroform and precipitated with ethanol. Glycogen carrier (5 μg) was used at this stage, and the pellet resuspended in 80 μl water. The blunt-ended DNA was then ligated to the LP21–25 linker. A stock of linker was prepared by mixing 80 μl LP21 (100 μM: GAATTCAGATCTCCCGGGTCA) with 80 μl LP25 (100 μM: GCGGTGACCCGGGAGATCTG AATTC) and 240 μl water. The mix was then placed on a 95°C hot block and the power supply turned off. When the block had cooled to room temperature, the LP21–25 linker was aliquoted and frozen. The 80 μl DNA sample was split into two samples for ligation (5 μl 10× buffer, 3 μl LP21–25 linker, 3 μl ligase (NE biolabs 400 U/μl and 39 μl DNA). This was mixed well with a pipette and left at 16°C overnight. Following ligation, the DNA was precipitated with ethanol, and resuspended in 42 μl water. The sample was amplified using Vent exo-polymerase and a biotinylated LP25 primer. The mix (5 μl ThermoPol 10× buffer, 0.5 μl B-LP25 (200 μM), 2 μl Vent exo-(NE Biolabs 2 U/μl), 1 μl dNTP (25 mM) and 41.5 μl DNA) was amplified with the following thermal cycle: (95°C 3 min, 95°C 30 s, 61°C 30 s, 72°C 30 s) × 35 cycles. Following amplification, the biotinylated products were extracted using Dynal streptavidin beads. For each sample, 30 μl of beads (Dyabeads M-270; Dynal biotech) were washed twice in 2× binding buffer (10 mM Tris HCL pH7.5, 1 mM EDTA, 2 M NaCl) using a magnet, and then resuspended in 50 μl of 2× binding buffer/sample. Each post amplification sample (50 μl) was then mixed with 50 μl of resuspended beads and incubated at room temperature on a rotator for 1 h. After binding, the samples were washed twice with 2× binding buffer, then once with 1× TE. After the last wash, each sample was resuspended in 30 μl 0.1 TE. The paired samples that were split before ligation were then pooled, making a 60 μl final aliquot. This was heated at 95°C for 10 min to free the DNA from the beads, and then stored at 4°C. We have used Stratagene Brilliant SYBR Green QPCR Master Mix as our standard kit for quantification of genomic controls and samples. We have used a range of primer sets. The HMBS primers are: human—Forward; ATGCTGCCTATTTCAAGGTTGT, Reverse; GAATT GGAACATTGCGACAGT, and mouse—Forward; CGGAGTCATGTCCGGTAAC, Reverse; CGACCAA TAGACGACGAGAA. Primers from the TAL1 locus (human and mouse) are available on request. Amplification conditions and PCR mixes were as recommended by Stratagene. Genomic standard curves were calculated for each primer using serial 5-fold dilutions from 50 ng genomic DNA/PCR reaction. For each sample PCR, a 45 μl stock was made using 1.3 μl of sample DNA + 43.7 μl water. This stock was used as 5 μl per real-time PCR reaction, and following amplification, quantified relative to the genomic DNA standard curve, for each primer set. The quantifications shown in Figure 3 represent the absolute quantification of DNA at each primer set for the 5 μl PCR sample. The mRNA expression levels of TAL1 and STIL in Figure 4A were normalised relative to β-actin and plotted relative to mRNA expression from normal human donor CD34 + cells, as described previously (11). GAF is a Leukaemia Research Fund (LRF) Bennett Fellow. Work in BG's laboratory is supported by the LRF and ARG's laboratory is supported by the LRF and the Wellcome Trust. Funding to pay the open Access publication charge was provided by the Wellcome Trust. Conflict of interest statement. None declared.
PMC12559006
MIRO2-mediated mitochondrial transfer from cancer cells induces cancer-associated fibroblast differentiation
Cancer-associated fibroblasts (CAFs) are key components of the tumor microenvironment that commonly support cancer development and progression. Here we show that different cancer cells transfer mitochondria to fibroblasts in cocultures and xenograft tumors, thereby inducing protumorigenic CAF features. Transplantation of functional mitochondria from cancer cells induces metabolic alterations in fibroblasts, expression of CAF markers and release of a protumorigenic secretome and matrisome. These features promote tumor formation in preclinical mouse models. Mechanistically, the mitochondrial transfer requires the mitochondrial trafficking protein MIRO2. Its depletion in cancer cells suppresses mitochondrial transfer and inhibits CAF differentiation and tumor growth. The clinical relevance of these findings is reflected by the overexpression of MIRO2 in tumor cells at the leading edge of epithelial skin cancers. These results identify mitochondrial transfer from cancer cells to fibroblasts as a driver of tumorigenesis and provide a rationale for targeting MIRO2 and mitochondrial transfer in different malignancies.Mitochondria are central to energy conversion and signaling events and engage in metabolism and cell fate decisions in health and disease. Once an alphaproteobacterial species that evolved into an organelle, mitochondria persist as functionally specialized units, carrying the ability to move between cells. This phenomenon, known as mitochondrial transfer, has emerged as a powerful strategy for tissue revitalization and rejuvenation in injured or diseased organs. Recent studies have identified important roles of mitochondrial dynamics in cancer, revealing how mitochondrial transfer contributes to metabolic heterogeneity among tumor cells and influences disease outcomes and treatment responses. Mitochondrial transfer can occur through gap junctions, extracellular vesicles, direct mitochondrial release and uptake or tunneling nanotubes (TNTs), which are thin membranous structures that form dynamic connections between cells. Transfer of mitochondria from stromal or immune cells in the tumor microenvironment to cancer cells has also been reported, which promoted tumor growth. For example, mitochondrial transfer from CD8 T cells, mesenchymal stem cells or cancer-associated fibroblasts (CAFs) into cancer cells has been described for different tumors, resulting in enhanced cancer cell proliferation, motility and lactate metabolism. However, the opposite process—mitochondrial transfer from cancer cells to stromal cells, including fibroblasts—has not been reported, although this may have important consequences for the fibroblast phenotype. Here, we identify mitochondrial transfer from cancer cells to fibroblasts as a key regulator of CAF differentiation. Given the association of the CAF phenotype with metabolic alterations, we tested whether cancer cells transfer their mitochondria to fibroblasts using cocultures of early-passage human primary skin fibroblasts (HPFs) with highly malignant A431 vulvar carcinoma cells. A431 cells stably expressing fluorescently labeled actin (LifeAct A431) were incubated with MitoTracker green, which stains mitochondria in living cells (Fig. 1a). This approach was chosen because of the strong fluorescence signal of MitoTracker green. Only some HPFs in close proximity to A431 cells became positive for MitoTracker green after a 24-h coculture, indicating that they received cancer cell mitochondria (Fig. 1b). They were clearly discernible against the weak background fluorescence, which may have resulted from dye leakage—a previously reported limitation of MitoTracker dyes that often produces false-positive results.Fig. 1Cancer cells transfer mitochondria to fibroblasts through TNTs.a, Coculture setup with LifeAct A431 cells (red) stained with MitoTracker green and unstained HPFs. This image was created with BioRender.com. b, Immunofluorescence images of the cocultures, counterstained with Hoechst. c, Representative photomicrographs of cocultures of A431 cells prestained with MitoTracker green and HPFs immunostained for COLI (white) and counterstained with phalloidin (red) and Hoechst (blue). TNT-like structures are indicated by white rectangles together with their length (n = 3 A431–HPF cocultures). d, Percentage of MitoTracker-high HPFs in direct or transwell coculture with A431 cells (n = 3 cocultures per setup). e, Percentage of MitoTracker-high HPFs after coculture with A431 cells in the presence of carbenoxolone (CBX) or vehicle (n = 3 cocultures per treatment group). f, RT–qPCR for GJB2 (encoding connexin 26) relative to RPL27 using RNA from A431 cells transfected with control (scrambled) or connexin 26 (Cx26) siRNA, and percentage of MitoTracker-high HPFs after coculture of siCtrl or siCx26 A431 cells (n = 3 cultures per group). g, Holotomographic imaging showing mitochondrial transfer (white arrows) from A431 LifeAct–MitoTracker green cells to HPFs (unstained) (Supplementary Video 1). h, Representative image of a coculture of A431 cells stained with MitoTracker green and HPFs, immunostained for COLI and counterstained with phalloidin and Hoechst (n = 3 A431–HPF cocultures). White arrows point to TNT-like structures. i, Percentage of MitoTracker-high HPFs after coculture with A431 cells in the presence of nocodazole (Noc), dihydrocytochalasin B (Cyto B) or vehicle (n = 3 cocultures per treatment group). j, Western blot analysis for SEC3 and SEC5 using lysates from A431 cells transfected with siCtrl, siSEC3 or siSEC5 (n = 2 cultures per group). Graph shows the percentage of MitoTracker-high HPFs after coculture with MitoTracker-stained control or SEC3–SEC5-knockdown A431 cells (n = 3 cocultures per group). k, Percentage of MitoTracker-high HPFs after coculture with HaCaT or A431 cells (n = 3 cocultures per cell line). l, Representative immunofluorescence images depicting cocultures of MDA-MB-231 and PANC1 cells prestained with MitoTracker green and HPFs immunostained for COLI (white) and counterstained with Hoechst (blue) (n = 3 cocultures per cell line). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (d–f,i,k) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (j) was used to determine statistical significance. Scale bars, 50 μm (b,g), 20 μm (c) or 25 μm (h,l).Source data a, Coculture setup with LifeAct A431 cells (red) stained with MitoTracker green and unstained HPFs. This image was created with BioRender.com. b, Immunofluorescence images of the cocultures, counterstained with Hoechst. c, Representative photomicrographs of cocultures of A431 cells prestained with MitoTracker green and HPFs immunostained for COLI (white) and counterstained with phalloidin (red) and Hoechst (blue). TNT-like structures are indicated by white rectangles together with their length (n = 3 A431–HPF cocultures). d, Percentage of MitoTracker-high HPFs in direct or transwell coculture with A431 cells (n = 3 cocultures per setup). e, Percentage of MitoTracker-high HPFs after coculture with A431 cells in the presence of carbenoxolone (CBX) or vehicle (n = 3 cocultures per treatment group). f, RT–qPCR for GJB2 (encoding connexin 26) relative to RPL27 using RNA from A431 cells transfected with control (scrambled) or connexin 26 (Cx26) siRNA, and percentage of MitoTracker-high HPFs after coculture of siCtrl or siCx26 A431 cells (n = 3 cultures per group). g, Holotomographic imaging showing mitochondrial transfer (white arrows) from A431 LifeAct–MitoTracker green cells to HPFs (unstained) (Supplementary Video 1). h, Representative image of a coculture of A431 cells stained with MitoTracker green and HPFs, immunostained for COLI and counterstained with phalloidin and Hoechst (n = 3 A431–HPF cocultures). White arrows point to TNT-like structures. i, Percentage of MitoTracker-high HPFs after coculture with A431 cells in the presence of nocodazole (Noc), dihydrocytochalasin B (Cyto B) or vehicle (n = 3 cocultures per treatment group). j, Western blot analysis for SEC3 and SEC5 using lysates from A431 cells transfected with siCtrl, siSEC3 or siSEC5 (n = 2 cultures per group). Graph shows the percentage of MitoTracker-high HPFs after coculture with MitoTracker-stained control or SEC3–SEC5-knockdown A431 cells (n = 3 cocultures per group). k, Percentage of MitoTracker-high HPFs after coculture with HaCaT or A431 cells (n = 3 cocultures per cell line). l, Representative immunofluorescence images depicting cocultures of MDA-MB-231 and PANC1 cells prestained with MitoTracker green and HPFs immunostained for COLI (white) and counterstained with Hoechst (blue) (n = 3 cocultures per cell line). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (d–f,i,k) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (j) was used to determine statistical significance. Scale bars, 50 μm (b,g), 20 μm (c) or 25 μm (h,l). Source data We observed elongated, thin bridges with a length of 10–100μm between cancer cells and HPFs (Fig. 1c), suggesting that the transfer of mitochondria occurs through TNTs. Consistently, we did not observe MitoTracker-high HPFs in a transwell assay, which does not allow communication through TNTs and gap junctions (Fig. 1d). Treatment of the cocultures with the gap junction inhibitor carbenoxolone even increased the transfer, as shown by flow cytometry analysis of MitoTracker-high cells, and knockdown of connexin 26, a major connexin in skin cancer cells, had no effect (Fig. 1e,f). These data further suggest that the transfer occurs through TNTs. This was confirmed by real-time holotomographic imaging (Fig. 1g and Supplementary Video 1). Phalloidin combined with MitoTracker staining revealed actin-containing TNTs transferring mitochondria from A431 cancer cells to HPFs (Fig. 1h). Because TNTs include actin and, in some cases, also microtubules, we explored the requirement of these cytoskeletal components for mitochondrial transfer. Treatment of the cocultures with the microtubule polymerization inhibitor nocodazole even enhanced the transfer, while the actin polymerization inhibitor dihydrocytochalasin B had a strong inhibitory effect (Fig. 1i). Knockdown of the exocyst complex components SEC3 (EXOC1) and SEC5 (EXOC2), which have a documented role in the regulation of the actin cytoskeleton and in TNT formation, also reduced the transfer (Fig. 1j). Mitochondrial transfer was also observed in cocultures of HPFs with immortalized but nontumorigenic human keratinocytes (HaCaT cells). However, their transfer efficiency was significantly lower compared to A431 cancer cells (Fig. 1k). In addition, MDA-MB-231 breast cancer and PANC1 pancreatic cancer cells transferred mitochondria to fibroblasts, demonstrating that this process occurs in different types of cancer cells (Fig. 1l). The selective effects of actin polymerization inhibitors and of SEC3–SEC5 small interfering RNAs (siRNAs) strongly suggest that the bright signal observed in some fibroblasts adjacent to cancer cells resulted from mitochondrial transfer rather than from dye leakage. Nevertheless, we performed additional controls to further confirm the specificity. Fluorescence analysis of fluorescence-activated cell sorting (FACS)-sorted HPFs showed that MitoTracker was stably incorporated into their mitochondrial network after serial passages and MitoTracker staining was not detectable after culture of HPFs in the conditioned medium (CM) of MitoTracker-treated A431 cells (Extended Data Fig. 1a,b). As an alternative, we used cocultures of human A431 cells and MitoTracker green-positive mouse fibroblasts. After 24 h, we detected human mitochondrial DNA in the fibroblasts by PCR (Fig. 2a). They exclusively expressed murine fibronectin 1 (Fn1), confirming their murine origin (Fig. 2b). Therefore, human cancer cells also transfer their mitochondria into mouse fibroblasts, although the transfer efficiency was significantly lower than in the human–human cocultures (Fig. 2c). We confirmed the transfer by making use of species-specific single-nucleotide polymorphisms (SNPs). Unique sequence variants within the 16S ribosomal RNA (rRNA) gene region of A431 cell mitochondria were detected in the mitochondria of the recipient mouse fibroblasts (Fig. 2d).Fig. 2Cancer cells transfer mitochondria to fibroblasts in vitro and in vivo.a, qPCR for the human mtDNA encoding tRNA-Leu(UUR) relative to mouse nuclear DNA encoding beta2-microglobulin (B2m) using DNA from MitoTracker-positive and MitoTracker-negative mouse fibroblasts sorted from n = 3 cocultures. Total mtDNA content was calculated on the basis of Ct values. b, RT–qPCR for human and mouse FN1 and Fn1 relative to RPL27 or Rps29, respectively, using RNA from MitoTracker-high and MitoTracker-low mouse fibroblasts sorted from n = 3 cocultures. c, Transfer efficiency in human–human and human–mouse cocultures (n = 3 cocultures per group). d, Comparison of SNPs within the 16S rRNA gene region of A431 cell mitochondria with those of control and recipient mouse fibroblasts. SNPs from A431 cells in recipient fibroblasts are indicated with rectangles (n > 300,000 cells pooled from three independent cocultures). e, Experimental setup and fluorescence images of HPF Su9–RFP and A431 Su9–GFP cocultures. This image was created with BioRender.com. Bottom, (1) colocalization of A431 and HPF mitochondria (orange), (2) HPF periphery with own mitochondrial network (red) and (3) nonrecipient HPFs (red). f, Section of a xenograft tumor formed by Su9–RFP A431 cells, showing Su9–RFP fluorescence in mitochondria of keratin 14 (K14)-negative cells and cultured fibroblasts from these tumors showing Su9–RFP fluorescence (red) and vimentin expression. g, qPCR for the mtDNA-encoded human tRNA-Leu gene relative to the mouse B2m gene using total DNA from cultured mouse fibroblasts isolated from noninjected ear skin (NS) or A431 xenograft tumors (n = 3 normal skin samples and n = 3 tumor samples from different mice). h, Representative immunofluorescence staining of A431 xenograft tumors for COLI (green) and human mitochondria (red). Costaining (yellow) of stromal cells adjacent to tumors was confirmed by colocalization analysis (site indicated with an asterisk). The white arrow indicates the line along which the intensity values of the different fluorescence signals were measured, starting from the initial position at the base of the arrow and ending at the arrowhead. Separate channels of zoomed-in regions are displayed (n = 3 sections from different tumors). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (a,c,g) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (b) was used to determine statistical significance. One control value was set to 1. Scale bars, 25 μm (e,f,h).Source data a, qPCR for the human mtDNA encoding tRNA-Leu(UUR) relative to mouse nuclear DNA encoding beta2-microglobulin (B2m) using DNA from MitoTracker-positive and MitoTracker-negative mouse fibroblasts sorted from n = 3 cocultures. Total mtDNA content was calculated on the basis of Ct values. b, RT–qPCR for human and mouse FN1 and Fn1 relative to RPL27 or Rps29, respectively, using RNA from MitoTracker-high and MitoTracker-low mouse fibroblasts sorted from n = 3 cocultures. c, Transfer efficiency in human–human and human–mouse cocultures (n = 3 cocultures per group). d, Comparison of SNPs within the 16S rRNA gene region of A431 cell mitochondria with those of control and recipient mouse fibroblasts. SNPs from A431 cells in recipient fibroblasts are indicated with rectangles (n > 300,000 cells pooled from three independent cocultures). e, Experimental setup and fluorescence images of HPF Su9–RFP and A431 Su9–GFP cocultures. This image was created with BioRender.com. Bottom, (1) colocalization of A431 and HPF mitochondria (orange), (2) HPF periphery with own mitochondrial network (red) and (3) nonrecipient HPFs (red). f, Section of a xenograft tumor formed by Su9–RFP A431 cells, showing Su9–RFP fluorescence in mitochondria of keratin 14 (K14)-negative cells and cultured fibroblasts from these tumors showing Su9–RFP fluorescence (red) and vimentin expression. g, qPCR for the mtDNA-encoded human tRNA-Leu gene relative to the mouse B2m gene using total DNA from cultured mouse fibroblasts isolated from noninjected ear skin (NS) or A431 xenograft tumors (n = 3 normal skin samples and n = 3 tumor samples from different mice). h, Representative immunofluorescence staining of A431 xenograft tumors for COLI (green) and human mitochondria (red). Costaining (yellow) of stromal cells adjacent to tumors was confirmed by colocalization analysis (site indicated with an asterisk). The white arrow indicates the line along which the intensity values of the different fluorescence signals were measured, starting from the initial position at the base of the arrow and ending at the arrowhead. Separate channels of zoomed-in regions are displayed (n = 3 sections from different tumors). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (a,c,g) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (b) was used to determine statistical significance. One control value was set to 1. Scale bars, 25 μm (e,f,h). Source data Next, we stably expressed mitochondria-targeted red fluorescent protein (Su9–RFP) in HPFs and green fluorescent protein (Su9–GFP) in A431 cells. After 24 h coculture, we detected HPFs with mitochondria that appeared orange in close proximity to the GFP-labeled cancer cells (Fig. 2e), confirming the uptake of mitochondria from cancer cells and suggesting their fusion with mitochondria of recipient fibroblasts. A431 cells expressing Su9–RFP were then injected intradermally into the ears of immunocompromised NOD scid mice. Fluorescence microscopy analysis of the resulting tumors showed mitochondrial structures of stromal fibroblasts that were positive for the mitochondrial protein expressed by cancer cells. We also detected human mitochondrial DNA (mtDNA) in cultured primary mouse fibroblasts from the tumors (Fig. 2f,g). Lastly, we stained sections from skin cancer xenograft tumors formed by A431 cells with an antibody specific for human mitochondria. In addition to the expected strong staining of the tumor cells, adjacent stromal cells showed clear staining, which overlapped with staining for the pan-fibroblast markers collagen type I (COLI) or platelet-derived growth factor alpha (PDGFRα) (Fig. 2h and Extended Data Fig. 1c). The transfer of mitochondria into fibroblasts in vivo was verified with breast and pancreatic cancer cells by costaining of respective xenograft tumors with antibodies to human mitochondria and COLI (Extended Data Fig. 1d). To assess the functional relevance of the mitochondrial transfer, we sorted viable HPFs with high and low MitoTracker Green fluorescence intensity (Extended Data Fig. 2a) and analyzed them by RNA sequencing (RNA-seq). Control fibroblasts were subjected to the sorting procedure but not maintained in cocultures. E-cadherin mRNA was not detected in the sorted fibroblasts, confirming the efficient sorting. Principal component analysis (PCA) showed distinct clustering (Extended Data Fig. 2b). There were significant differences in gene expression between HPFs in the coculture (MitoTracker-high and MitoTracker-low) versus control HPFs in monoculture and also between MitoTracker-high and MitoTracker-low HPFs (Fig. 3a–c), although the two latter populations were exposed to the same cancer cell secretome. This finding suggests a strong impact of cancer-cell-derived mitochondria on the recipient HPFs. Genes significantly upregulated in the MitoTracker-high versus MitoTracker-low group were predominantly involved in pathways related to inflammation, immune response, cellular metabolism and stress responses (Extended Data Fig. 2c,d). Activation of the interferon pathway was reflected by the increased expression of several interferon response genes (ISGs) (Extended Data Fig. 2e).Fig. 3Transferred cancer cell mitochondria induce a CAF phenotype.a–c, Volcano plots displaying differentially expressed genes in MitoTracker-low versus control (a), MitoTracker-high versus control (b) and MitoTracker-high versus MitoTracker-low HPFs (c) sorted from n = 3 cocultures with A431 cells. d,e. RNA-seq data from sorted HPFs depicting expression of INHBA, IL6, ACTA2 and COL1A1 (d) or PDGFRA, PDGFRB, S100A4, FAP and CD74 (e) in MitoTracker-high, MitoTracker-low and control groups sorted from n = 3 cocultures. f–h, Comparative analysis of gene signatures in MitoTracker-high versus MitoTracker-low HPFs with published CAF datasets, showing similarities of MitoTracker-high HPFs with myCAFs and iCAFs. i, Volcano plot displaying differentially abundant proteins in MitoTracker-high versus MitoTracker-low HPFs sorted from n = 4 cocultures with A431 cells. j, Correlation analysis of gene and protein expression in MitoTracker-high versus MitoTracker-low HPFs. Significantly regulated pathways (q < 0.1) are highlighted (blue, Hallmarks of Cancer pathways; purple, Wikipathways). k, Percentage of Ki67-positive HPFs and relative levels of intracellular ATP and MitoSOX in sorted HPFs (n = 3 cocultures for Ki67 and 6 cocultures for ATP and MitoSOX). l, Percentage of Ki67-positive A431 cells in spheroids cultured with CM from control HPFs and sorted MitoTracker-high and MitoTracker-low HPFs (n = 3 spheroids per treatment group). m, Transwell migration of A431 cells in CM from sorted MitoTracker-low and MitoTracker-high and control HPFs (n = 3 transwell cultures per treatment group). n. Relative colony size of A431 cells plated on dECM from sorted HPFs (n = 3 cultures). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (k (right and middle),n) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (d,e,k (left),l,m) was used to determine statistical significance. One control value was set to 1.Source data a–c, Volcano plots displaying differentially expressed genes in MitoTracker-low versus control (a), MitoTracker-high versus control (b) and MitoTracker-high versus MitoTracker-low HPFs (c) sorted from n = 3 cocultures with A431 cells. d,e. RNA-seq data from sorted HPFs depicting expression of INHBA, IL6, ACTA2 and COL1A1 (d) or PDGFRA, PDGFRB, S100A4, FAP and CD74 (e) in MitoTracker-high, MitoTracker-low and control groups sorted from n = 3 cocultures. f–h, Comparative analysis of gene signatures in MitoTracker-high versus MitoTracker-low HPFs with published CAF datasets, showing similarities of MitoTracker-high HPFs with myCAFs and iCAFs. i, Volcano plot displaying differentially abundant proteins in MitoTracker-high versus MitoTracker-low HPFs sorted from n = 4 cocultures with A431 cells. j, Correlation analysis of gene and protein expression in MitoTracker-high versus MitoTracker-low HPFs. Significantly regulated pathways (q < 0.1) are highlighted (blue, Hallmarks of Cancer pathways; purple, Wikipathways). k, Percentage of Ki67-positive HPFs and relative levels of intracellular ATP and MitoSOX in sorted HPFs (n = 3 cocultures for Ki67 and 6 cocultures for ATP and MitoSOX). l, Percentage of Ki67-positive A431 cells in spheroids cultured with CM from control HPFs and sorted MitoTracker-high and MitoTracker-low HPFs (n = 3 spheroids per treatment group). m, Transwell migration of A431 cells in CM from sorted MitoTracker-low and MitoTracker-high and control HPFs (n = 3 transwell cultures per treatment group). n. Relative colony size of A431 cells plated on dECM from sorted HPFs (n = 3 cultures). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (k (right and middle),n) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (d,e,k (left),l,m) was used to determine statistical significance. One control value was set to 1. Source data Because the pathways enriched in the MitoTracker-high fibroblasts are often activated in CAFs, we analyzed the dataset for skin CAF marker genes. Many of them were indeed overexpressed in the MitoTracker-high and, to a lesser extent, in the MitoTracker-low population. Among them were INHBA (encoding the protumorigenic cytokine activin A), IL6 (encoding interleukin-6), ACTA2 (encoding α smooth muscle actin) and COL1A1 (encoding COLI, alpha 1 subunit) (Fig. 3d). This was verified by reverse transcription (RT)–qPCR in independent coculture experiments (Extended Data Fig. 2f). PDGFRA and the CAF markers PDGFRB, S100A4 (encoding FSP1), FAP and CD74 (ref. ) were also upregulated in the MitoTracker-high population (Fig. 3e). To determine whether the MitoTracker-high cells (Mito-CAFs), correspond to inflammatory CAFs (iCAFs), myofibroblastic CAFs (myCAFs) or antigen-presenting CAFs (apCAFs), we compared their expression profile to those of published CAF datasets. Mito-CAFs overexpressed genes characteristic of both iCAFs and myCAFs (Fig. 3f–h). A proteomic analysis using the same fibroblast populations revealed clear clustering of the groups (Extended Data Fig. 3a), increased abundance of CAF markers and ISG-encoded proteins and activation of proinflammatory pathways in the MitoTracker-high population (Fig. 3i, Extended Data Fig. 3b–d and Supplementary Table 1). Many of the observed gene expression changes were reflected by protein abundance changes (Fig. 3j). MitoTracker-high fibroblasts also exhibited functional characteristics of protumorigenic CAFs, including increased proliferation and higher concentrations of intracellular adenosine triphosphate (ATP) and mitochondrial reactive oxygen species (ROS) (Fig. 3k). Increased proliferation and CAF marker expression in sorted fibroblasts were also observed when HPFs in cocultures received mitochondria from A431 cells expressing Su9–RFP (Extended Data Fig. 3e–g), further confirming the reliability of the MitoTracker approach in our setting. Lastly, the CM from MitoTracker-high HPFs promoted cancer cell proliferation and transwell migration more efficiently than CM from MitoTracker-low HPFs. This was observed for A431 cells (Fig. 3l,m), primary skin squamous cell carcinoma (SCC) cells and SCC13 (ref. ) and HA–Ras-transformed HaCaT cells (HaCaT-Ras) (Extended Data Fig. 3h–j). Furthermore, the decellularized extracellular matrix (dECM) produced by MitoTracker-high fibroblasts induced the formation of larger A431 colonies (Fig. 3n). To specifically test the role of cancer cell mitochondria in fibroblast reprogramming, we isolated and purified mitochondria from cancer cells and transplanted them directly into HPFs using MitoCeption. While mitochondria directly move from the cytoplasm of the donor cells to the cytoplasm of recipient cells during TNT-mediated transfer, MitoCeption induces the rapid uptake of purified mitochondria through the plasma membrane, most likely through an endocytic pathway. The uptake using MitoCeption was confirmed by detection of MitoTracker green fluorescence and by an increase in mtDNA content in the ‘MitoCepted’ fibroblasts (Fig. 4a,b), which was in a similar range to that described for MitoCepted endothelial cells (13%). The MitoTracker staining likely overestimates the uptake because cancer cell mitochondria fuse with the mitochondria of recipient HPFs, as seen after transplantation of MitoTracker green-labeled or Su9–RFP-expressing mitochondria from A431 cells into HPFs prestained with MitoTracker red or expressing TOM20–GFP, respectively (Fig. 4c). We found a substantial colocalization of the MitoCepted and the endogenous mitochondria using confocal microscopy. As expected, it was more pronounced with MitoTracker green because the dye labels the entire mitochondria. Although an additional effect of dye leakage cannot be excluded, the findings obtained with MitoTracker-labeled and, in particular, with genetically labeled mitochondria demonstrate that the MitoCepted mitochondria were released into the cytoplasm.Fig. 4CAF reprogramming through transplantation of cancer cell mitochondria.a, Representative fluorescence images of HPFs MitoCepted with MitoTracker green-stained A431 mitochondria (MitoCepted HPFs) or mock treatment, counterstained with Hoechst (blue) (n = 3 cultures per group). b, qPCR for the mtDNA encoding tRNA-Leu(UUR) relative to the nucDNA encoding B2M using DNA from MitoCepted (MC) or mock-treated (Ctrl) HPFs. Relative mtDNA content (based on Ct values) is indicated (n = 3 cultures per group). c, Representative confocal image in the xyz plane showing HPFs prestained with MitoTracker red and MitoCepted with MitoTracker green-labeled A431 mitochondria (left) and TOM20–GFP-expressing HPFs MitoCepted with mitochondria from A431 Su9–RFP cells (right). Yellow staining indicates mitochondrial fusion. d, Percentage of Ki67 MitoCepted or control HPFs among all cells (n = 3 cultures per group). e, RT–qPCR for INHBA, IL6, ACTA2 and COL1A1 relative to RPL27 using RNA from MitoCepted or mock-treated HPFs (n = 3 cultures per group). f, FluidFM experimental setup, adapted from a previous study. g, Image of FluidFM-mediated injection of mitochondria into HPFs. h, Percentage of Ki67 fibroblasts in HPFs injected with A431-derived mitochondria using FluidFM or mock treatment (n = 3 cultures per group). i, RT–qPCR for INHBA using RNA from HPFs subjected to MitoCeption with mitochondria from HaCaT, HaCaT-Ras or A431 cell lines (n = 3 cultures per cell line). j, RT–qPCR for INHBA using RNA from (1) mock-treated HPFs or HPFs subjected to MitoCeption with mitochondria (2) from keratinocytes of a healthy individual, (3) from normal keratinocytes of a person with SCC or (4) malignant cancer cells of the same person with SCC. Right, representative FN1–COLI immunofluorescence stainings with quantification of staining intensity in the dECM produced by HPFs after MitoCeption with mitochondria from the different primary donor cells (n = 3 MitoCeptions per cell type). k, Percentage of Ki67 HPFs subjected to MitoCeption with different amounts of mitochondria isolated from A431 cells. Numbers on the x axis show the ratio of donor A431 cells (used for mitochondrial isolation) and recipient HPFs (n = 3 cultures per group). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (b,d,e,h) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (i–k) was used to determine statistical significance. Scale bars, 100 μm (a,j), 10 μm (c) and 25 μm (g).Source data a, Representative fluorescence images of HPFs MitoCepted with MitoTracker green-stained A431 mitochondria (MitoCepted HPFs) or mock treatment, counterstained with Hoechst (blue) (n = 3 cultures per group). b, qPCR for the mtDNA encoding tRNA-Leu(UUR) relative to the nucDNA encoding B2M using DNA from MitoCepted (MC) or mock-treated (Ctrl) HPFs. Relative mtDNA content (based on Ct values) is indicated (n = 3 cultures per group). c, Representative confocal image in the xyz plane showing HPFs prestained with MitoTracker red and MitoCepted with MitoTracker green-labeled A431 mitochondria (left) and TOM20–GFP-expressing HPFs MitoCepted with mitochondria from A431 Su9–RFP cells (right). Yellow staining indicates mitochondrial fusion. d, Percentage of Ki67 MitoCepted or control HPFs among all cells (n = 3 cultures per group). e, RT–qPCR for INHBA, IL6, ACTA2 and COL1A1 relative to RPL27 using RNA from MitoCepted or mock-treated HPFs (n = 3 cultures per group). f, FluidFM experimental setup, adapted from a previous study. g, Image of FluidFM-mediated injection of mitochondria into HPFs. h, Percentage of Ki67 fibroblasts in HPFs injected with A431-derived mitochondria using FluidFM or mock treatment (n = 3 cultures per group). i, RT–qPCR for INHBA using RNA from HPFs subjected to MitoCeption with mitochondria from HaCaT, HaCaT-Ras or A431 cell lines (n = 3 cultures per cell line). j, RT–qPCR for INHBA using RNA from (1) mock-treated HPFs or HPFs subjected to MitoCeption with mitochondria (2) from keratinocytes of a healthy individual, (3) from normal keratinocytes of a person with SCC or (4) malignant cancer cells of the same person with SCC. Right, representative FN1–COLI immunofluorescence stainings with quantification of staining intensity in the dECM produced by HPFs after MitoCeption with mitochondria from the different primary donor cells (n = 3 MitoCeptions per cell type). k, Percentage of Ki67 HPFs subjected to MitoCeption with different amounts of mitochondria isolated from A431 cells. Numbers on the x axis show the ratio of donor A431 cells (used for mitochondrial isolation) and recipient HPFs (n = 3 cultures per group). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (b,d,e,h) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (i–k) was used to determine statistical significance. Scale bars, 100 μm (a,j), 10 μm (c) and 25 μm (g). Source data We found that, 24 h after MitoCeption with MitoTracker-labeled A431 mitochondria, the proliferation rate was significantly higher compared to mock-treated HPFs (Fig. 4d) and the MitoCepted cells showed higher expression of CAF genes (Fig. 4e and Extended Data Fig. 4a). By contrast, expression of three selected ISGs was not increased after MitoCeption (Extended Data Fig. 4b), suggesting that mitochondrial transfer alone is not sufficient to activate these genes. To validate the effect of purified mitochondria on CAF differentiation, we used fluid force microscopy (FluidFM) to inject purified MitoTracker Green-stained A431 mitochondria directly into the cytoplasm of HPFs (Fig. 4f,g). The injected HPFs also showed increased proliferation (Fig. 4h), again demonstrating that the effect of mitochondria on the induction of CAF features is independent of the mode of uptake. We next compared the effect of mitochondria isolated from the HaCaT keratinocyte cell line, their malignant counterpart HaCaT-Ras and the highly malignant A431 cell line on HPFs in MitoCeption experiments using mitochondria isolated from the same number of cells. This normalization is justified because of the similar protein content of the mitochondrial isolates from all cell lines (Extended Data Fig. 4c). The expression of INHBA in the recipient HPFs increased in accordance with the malignancy of the donor cell line (Fig. 4i). Consistently, MitoCepted fibroblasts, which received mitochondria from primary donor-derived SCC cells, displayed increased expression of INHBA compared to fibroblasts, which received mitochondria from adjacent nontransformed keratinocytes or from keratinocytes of a healthy individual, and deposited more FN1 and COLI (Fig. 4j). The effect of mitochondria on HPFs was concentration dependent. MitoCeption of HPFs with A431 mitochondria at a 1:1 ratio (same number of donor A431 cells and recipient HPFs) caused a significant increase in HPF proliferation, while lower amounts of mitochondria had only a minor effect. A further increase in the amount of MitoCepted mitochondria did not further promote HPF proliferation (Fig. 4k). To gain mechanistic insight into the alterations in HPFs that occur upon mitochondrial transfer from cancer cells, we measured their oxygen consumption rate (OCR) using Seahorse analysis. Transplantation of A431-derived but not HaCaT cell-derived mitochondria promoted basal respiration and proton leak in the recipient fibroblasts (Fig. 5a,b and Extended Data Fig. 4d). The values observed in HPFs after MitoCeption with A431 mitochondria almost reached those observed in A431 cells (Fig. 5a,b). Mitochondria from HaCaT-Ras cells also promoted proton leak but had no effect on basal respiration (Extended Data Fig. 4e). These findings provide a possible explanation for the higher proliferation of recipient HPFs because increased oxidative phosphorylation (OxPhos) in cultured fibroblasts was shown to promote their proliferation. Consistently, inhibition of OxPhos in HPFs by oligomycin prevented the increase in CAF marker expression and proliferation after transplantation of A431 mitochondria (Extended Data Fig. 4f,g). High OCR and ATP levels in CAFs are also important for their release of protumorigenic factors. Consistently, CM from fibroblasts, which received A431 mitochondria using MitoCeption, promoted proliferation and transwell migration of A431 cells to a significantly higher extent compared to CM from control (mock-treated) fibroblasts (Fig. 5c).Fig. 5Functional cancer cell mitochondria are required for CAF reprogramming.a, Basal respiration of HPFs subjected to MitoCeption with A431 or HaCaT mitochondria or mock treatment in comparison to A431 cells (n = 5 independent MitoCeptions per cell type). b, Proton leak in the same cultures as in a. c, Percentage of Ki67 A431 cells and transwell migration of A431 cells cultured in CM of MitoCepted or mock-treated HPFs (n = 3 cultures per group). d, RT–qPCR for INHBA and IL6 using RNA from MitoCepted (mitochondria from MDA-MB-231 breast cancer cells) or mock-treated HPFs (n = 3 cultures per group). e, Clonogenicity of MDA-MB-231 cells cultured in CM from HPFs subjected to MitoCeption with MitoTracker green-stained mitochondria from MDA-MB-231 cells or mock treatment (n = 3 cultures per treatment group). f, Representative image of 3-week-old ear xenograft tumors (arrowheads) following intradermal coinjection of A431 cells and MitoCepted (with A431 mitochondria) or mock-treated HPFs and tumor volume at various time points (n = 5 tumors per group from different mice). g, Representative immunofluorescence stainings of tumors formed by A431 cells and MitoCepted or mock-treated HPFs for E-cadherin and FN1 (green) and MECA32 (red), counterstained with Hoechst (blue) (n = 5 tumors per group from different mice). h, Percentage of Ki67 HPFs subjected to MitoCeption with A431 lmt mitochondria or mock treatment and RT–qPCR for INHBA using RNA from HPFs subjected to MitoCeption with A431 lmt mitochondria or mock treatment (n = 9 Ki67 or n = 3 RT–qPCR cultures per treatment group). i, Percentage of Ki67 A431 cells (left) or transwell migration of A431 cells (right) cultured in CM from HPFs subjected to MitoCeption with A431 lmt mitochondria or mock treatment (n = 9 Ki67 or n = 3 transwell migration cultures per treatment group). j, Left, tumor volume at various time points during tumor development by A431 cancer cells coinjected with MitoCepted HPFs, which received mitochondria from control or lmt A431 cells (n = 4 tumors per group from different mice). Right, Histological stainings of a tumor from each group. Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (c–e,h,i) or two-sided one-way (a,b) or two-way (f,j) ANOVA with Bonferroni post hoc multiple comparison test was used to determine statistical significance. Scale bars, 200 μm (g) and 1 mm (j).Source data a, Basal respiration of HPFs subjected to MitoCeption with A431 or HaCaT mitochondria or mock treatment in comparison to A431 cells (n = 5 independent MitoCeptions per cell type). b, Proton leak in the same cultures as in a. c, Percentage of Ki67 A431 cells and transwell migration of A431 cells cultured in CM of MitoCepted or mock-treated HPFs (n = 3 cultures per group). d, RT–qPCR for INHBA and IL6 using RNA from MitoCepted (mitochondria from MDA-MB-231 breast cancer cells) or mock-treated HPFs (n = 3 cultures per group). e, Clonogenicity of MDA-MB-231 cells cultured in CM from HPFs subjected to MitoCeption with MitoTracker green-stained mitochondria from MDA-MB-231 cells or mock treatment (n = 3 cultures per treatment group). f, Representative image of 3-week-old ear xenograft tumors (arrowheads) following intradermal coinjection of A431 cells and MitoCepted (with A431 mitochondria) or mock-treated HPFs and tumor volume at various time points (n = 5 tumors per group from different mice). g, Representative immunofluorescence stainings of tumors formed by A431 cells and MitoCepted or mock-treated HPFs for E-cadherin and FN1 (green) and MECA32 (red), counterstained with Hoechst (blue) (n = 5 tumors per group from different mice). h, Percentage of Ki67 HPFs subjected to MitoCeption with A431 lmt mitochondria or mock treatment and RT–qPCR for INHBA using RNA from HPFs subjected to MitoCeption with A431 lmt mitochondria or mock treatment (n = 9 Ki67 or n = 3 RT–qPCR cultures per treatment group). i, Percentage of Ki67 A431 cells (left) or transwell migration of A431 cells (right) cultured in CM from HPFs subjected to MitoCeption with A431 lmt mitochondria or mock treatment (n = 9 Ki67 or n = 3 transwell migration cultures per treatment group). j, Left, tumor volume at various time points during tumor development by A431 cancer cells coinjected with MitoCepted HPFs, which received mitochondria from control or lmt A431 cells (n = 4 tumors per group from different mice). Right, Histological stainings of a tumor from each group. Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (c–e,h,i) or two-sided one-way (a,b) or two-way (f,j) ANOVA with Bonferroni post hoc multiple comparison test was used to determine statistical significance. Scale bars, 200 μm (g) and 1 mm (j). Source data Together, these results suggest that the transfer or transplantation of epithelial cancer cell-derived mitochondria alone is sufficient to reprogram fibroblasts. This is not limited to SCC cells, as transfer of mitochondria from the breast and pancreatic cancer cell lines MDA-MB-231 and PANC1, respectively, also increased the expression of CAF markers and their CM promoted clonogenic growth of MDA-MB-231 and PANC1 cells (Fig. 5d,e and Extended Data Fig. 5a,b). At day 5 after seeding, the proliferation rate of HPFs declined, particularly after MitoCeption (Extended Data Fig. 5c versus Fig. 4d). Concomitantly, the number of β-galactosidase-positive fibroblasts increased among the MitoCepted fibroblasts, indicating senescence (Extended Data Fig. 5d). This is supported by their increased expression of the senescence markers CDKN1A and CDKN2B, while expression of most CAF markers was no longer increased at this time point (Extended Data Fig. 5e,f). The only exception was IL6, which is also a senescence marker. Nevertheless, the recipient HPFs may still exert protumorigenic effects because senescent cells often have a protumorigenic senescence-associated secretory phenotype. Indeed, when we coinjected A431-MitoCepted or mock-treated HPFs with A431 cells into the ears of immunodeficient mice, the tumors that formed in the presence of fibroblasts containing A431-derived mitochondria were significantly larger and showed increased deposition of FN1 and more blood vessels (Fig. 5f,g). This was associated with the long-term presence of the MitoCepted fibroblasts as determined in a separate experiment with HPFs, which received Su9–RFP-expressing A431 cancer cell mitochondria. Then, 2 weeks after injection, these tagged fibroblasts were still detectable (Extended Data Fig. 5g). To test whether disruption of mitochondrial function in cancer cells prevents CAF differentiation upon mitochondrial transfer, we used FluidFM to extract mitochondria from A431 Su9–RFP cells, which were depolarized using carbonyl cyanide m-chlorophenylhydrazone (CCCP). HPFs, which received CCCP-treated mitochondria, had a mildly but significantly lower proliferation rate compared to HPFs, which received mock-treated mitochondria (Extended Data Fig. 6a). In addition, HPFs MitoCepted with mitochondria from A431 cells, which were pretreated with CCCP, had a strongly reduced OCR compared to HPFs that received mitochondria from vehicle-treated cancer cells (Extended Data Fig. 6b). We next generated A431 cancer cells with a 40% reduction in the amount of mtDNA (termed low mtDNA (lmt) cells) using extended low-dose ethidium bromide treatment (Extended Data Fig. 6c). Mitochondria from lmt cells had a similar protein content to those from control cells and the viability and proliferation of lmt A431 cells were not reduced. However, their mitochondrial respiration was significantly impaired (Extended Data Fig. 6d–g). Upon transplantation of these mitochondria, the proliferation rate of the recipient cells was even reduced and expression of most CAF markers was not significantly altered (Fig. 5h and Extended Data Fig. 6h). The CM of HPFs, which received mitochondria from lmt cancer cells, did not promote proliferation and migration of cancer cells (Fig. 5i). In xenograft experiments, tumors formed by A431 cells coinjected with HPFs containing mitochondria from lmt A431 cells were significantly smaller compared to those formed with HPFs containing mitochondria from control A431 cells (Fig. 5j). This finding underscores the critical role of mitochondrial DNA, which encodes important components of the respiratory chain, in the induction of a protumorigenic CAF phenotype in MitoCepted fibroblasts. Additional experiments using only lmt A431 cells showed that mice injected with these cells did not develop tumors within 15 days (Extended Data Fig. 6i,j). To identify potential regulators of the mitochondrial transfer from skin cancer cells to fibroblasts, we used published single-cell RNA (scRNA)-seq data of tumors and site-matched normal skin from persons with cutaneous SCCs. We used preprocessed annotations to identify keratinocyte populations and focused on genes with a documented function in mitochondrial trafficking, including MIRO1 (RHOT1), MIRO2 (RHOT2), TRAK1 and TRAK2 (ref. ). We defined a gene as ‘highly expressed’ when its expression level exceeded the mean expression level observed across all cell populations examined. Expression of MIRO2 was significantly elevated in malignant versus nonmalignant epithelial cells (Fig. 6a). Tumor-specific keratinocytes (TSKs), a cluster exclusively present in tumor samples, exhibited particularly high MIRO2 expression (Fig. 6b). This was confirmed with another scRNA dataset from skin SCCs (Extended Data Fig. 7a,b). Analysis of spatial transcriptomics data revealed elevated MIRO2 expression in invasively growing cells at tumor margins (Fig. 6c). To delineate MIRO2 mRNA localization relative to CAF subtypes, we used coexpression analysis with PDGFRA and specific markers for iCAFs (MMP11), myCAFs (ACTA2) and adipose CAF (adiCAF; CFD). We detected substantial colocalization of MIRO2 mRNA with these CAF subtypes, particularly with MMP11-positive iCAFs (Extended Data Fig. 7c). We further applied Cell2Location spatial deconvolution using skin SCC data as a single-cell (ref. ) to estimate cell type distributions per spot. We computed CAF scores using fibroblast-specific expression inferred by Cell2Location and identified MIRO2 spots on the basis of keratinocyte-specific expression. CAF scores were consistently higher in MIRO2 spots and neighboring regions compared to non-MIRO2 spots, although only minor differences were observed among the four CAF subtypes (Extended Data Fig. 7d).Fig. 6MIRO2 is overexpressed at the leading edge of SCCs.a, Dot plot showing expression of mitochondrial trafficking genes; violin plot showing expression of MIRO2 in different cell types in SCCs (n = 5,799 myeloid cells, 4,644 tumor cells, 1,495 epithelial cells, 584 fibroblasts, 413 lymphoid cells, 169 endothelial cells and 129 melanocytes). b, Violin plot showing expression of MIRO2 in tumor cell subpopulations in SCCs based on scRNA-seq data (n = 296 tumor-specific keratinocytes (TSKs), 1,385 basal tumor keratinocytes (KC), 725 cycling tumor keratinocytes and 2238 differentiating tumor keratinocytes). c, Feature plots showing spatial distribution of MIRO2 transcripts in human skin SCC; violin blots showing MIRO2 transcripts at the tumor leading edge versus the total tumor and its microenvironment (TME) (n = 2 tumors from different patients; P2 and P6). d, Western blot of lysates from HPFs, HaCaT, HaCaT-Ras and A431 cells for MIRO2 and GAPDH. e, MIRO2 and K14 immunofluorescence stainings of sections from 3D organotypic skin cultures with HPFs and HaCaT or A431 cells and quantification of the MIRO2-positive area (n = 3 3D cultures per epithelial cell line). Scale bar, 100 μm. f, Forest plot showing the 5-year disease-specific survival (DSS) associated with MIRO2 expression across solid cancers based on TCGA. Hazard ratios (HRs) and 95% confidence intervals (CIs) based on Cox proportional hazard model are shown. The last point represents the estimate from the random-effects meta-analysis (n = 8,941 patients). g, Pearson correlation coefficient (ρ) and 95% CIs between the enrichment score of the leading edge (LE) signature and MIRO2 expression across the different solid cancers in TCGA. The last point represents the estimate from the random-effects meta-analysis (n = 10,238 patients). h, Dependency of different cancers on MIRO2 expression as documented in the DepMap Portal. Gene effect scores are derived from DEMETER2 or CERES. A lower score denotes a greater dependency on expression. Violin plots in a–c show the median (center line), 25th and 75th percentiles (box bounds) and whiskers extending to the most extreme data points within 1.5 times the interquartile range from the box. Points outside this range are plotted as outliers. The graph in e shows the mean ± s.e.m. A Mann–Whitney U-test for comparison between two groups (a,b) or unpaired two-sided Student’s t-test (e) was used to determine statistical significance.Source data a, Dot plot showing expression of mitochondrial trafficking genes; violin plot showing expression of MIRO2 in different cell types in SCCs (n = 5,799 myeloid cells, 4,644 tumor cells, 1,495 epithelial cells, 584 fibroblasts, 413 lymphoid cells, 169 endothelial cells and 129 melanocytes). b, Violin plot showing expression of MIRO2 in tumor cell subpopulations in SCCs based on scRNA-seq data (n = 296 tumor-specific keratinocytes (TSKs), 1,385 basal tumor keratinocytes (KC), 725 cycling tumor keratinocytes and 2238 differentiating tumor keratinocytes). c, Feature plots showing spatial distribution of MIRO2 transcripts in human skin SCC; violin blots showing MIRO2 transcripts at the tumor leading edge versus the total tumor and its microenvironment (TME) (n = 2 tumors from different patients; P2 and P6). d, Western blot of lysates from HPFs, HaCaT, HaCaT-Ras and A431 cells for MIRO2 and GAPDH. e, MIRO2 and K14 immunofluorescence stainings of sections from 3D organotypic skin cultures with HPFs and HaCaT or A431 cells and quantification of the MIRO2-positive area (n = 3 3D cultures per epithelial cell line). Scale bar, 100 μm. f, Forest plot showing the 5-year disease-specific survival (DSS) associated with MIRO2 expression across solid cancers based on TCGA. Hazard ratios (HRs) and 95% confidence intervals (CIs) based on Cox proportional hazard model are shown. The last point represents the estimate from the random-effects meta-analysis (n = 8,941 patients). g, Pearson correlation coefficient (ρ) and 95% CIs between the enrichment score of the leading edge (LE) signature and MIRO2 expression across the different solid cancers in TCGA. The last point represents the estimate from the random-effects meta-analysis (n = 10,238 patients). h, Dependency of different cancers on MIRO2 expression as documented in the DepMap Portal. Gene effect scores are derived from DEMETER2 or CERES. A lower score denotes a greater dependency on expression. Violin plots in a–c show the median (center line), 25th and 75th percentiles (box bounds) and whiskers extending to the most extreme data points within 1.5 times the interquartile range from the box. Points outside this range are plotted as outliers. The graph in e shows the mean ± s.e.m. A Mann–Whitney U-test for comparison between two groups (a,b) or unpaired two-sided Student’s t-test (e) was used to determine statistical significance. Source data Western blot analysis showed strong expression of MIRO2 in A431 and HaCaT-Ras cells but it was hardly detectable in the parental HaCaT cells and in fibroblasts (Fig. 6d). The predominant expression of MIRO2 in the epithelium was confirmed by immunostaining of three-dimensional (3D) organotypic cultures (Fig. 6e). Increased MIRO2 expression did not consistently correlate with poor survival across 30 different tumors (Fig. 6f and Extended Data Fig. 8). However, the cancer expression data are based on bulk cancer tissue, whereas the expression of MIRO2 was mainly upregulated at the tumor edge. Indeed, analysis of the leading edge signature from oral SCC showed a positive correlation of this signature with MIRO2 expression across most of the 30 cancer types (Fig. 6g). This is of likely functional relevance, because Cancer Dependency Map (DepMap) analysis showed an important role of MIRO2 in the proliferation and survival of cancer cells (Fig. 6h). We next investigated the impact of MIRO2 knockdown on intercellular mitochondrial transfer by establishing cocultures of HPFs and LifeAct A431 cells, which were transfected with MIRO2 or scrambled siRNAs and stained with MitoTracker green (Fig. 7a,b). Transfection with fluorescently labeled siRNA showed no detectable transfer of siRNA to cocultured HPFs (Extended Data Fig. 9a). Consistently, HPFs isolated from cocultures of A431 cells transfected with siMIRO2 or siCtrl showed no significant difference in MIRO2 expression (Extended Data Fig. 9b).Fig. 7MIRO2 is required for mitochondrial transfer.a, RT–qPCR for MIRO2 using RNA from siCtrl or siMIRO2 A431 cells; Western blot of total and mitochondrial lysates from these cells for MIRO2, vinculin or HSP60 (loading controls) (n = 3 cultures per group). b, Fluorescence images of LifeAct A431 cells (red) stained with MitoTracker green and transfected with siCtrl or siMIRO2 in coculture with HPFs, counterstained with Hoechst. White arrowheads indicate A431 cells. c, Percentage of MitoTracker-high HPFs in cocultures with siMIRO2 or siCtrl A431 cells, normalized to the number of cancer cells (n = 3 cocultures per group). d, qPCR for mtDNA encoding tRNA-Leu(UUR) relative to nucDNA encoding B2M using DNA from A431 cells transfected with siCtrl or siMIRO2. Total mtDNA content was calculated on the basis of Ct values (n = 3 cultures per group). e, Mitochondrial mass in siCtrl and siMIRO2 A431 cells based on MitoTracker green mean fluorescence intensity (MFI) (n = 3 per group). f, Confocal images of siCtrl or siMIRO2 A431 cells incubated with MitoTracker green. The dashed line marks the outer edge of the cell (n = 3 cultures per group). g, RT–qPCR for INHBA and IL6 using RNA from HPFs incubated with CM of siCtrl or siMIRO2 A431 cells (n = 3 cultures per treatment group). h, RT–qPCR for INHBA and IL6 using RNA from sorted HPFs cocultured with siCtrl or siMIRO2 A431 cells (n = 3 cocultures per group). DC, direct culture. i, OCR of siCtrl or siMIRO2 A431 cells. The time of drug injection is indicated (n = 3 cultures per group). j, RT–qPCR for MIRO2 using RNA from A431 cells transfected with control or MIRO2 overexpression vectors (OE-MIRO2) (n = 3 cultures per group). Western blot of lysates from control or MIRO2-overexpressing A431 cells for MIRO2 or GAPDH (n = 2 cultures per group). k, Percentage of MitoTracker-high HPFs after coculture with MitoTracker-stained control or MIRO2-overexpressing A431 cells (n = 3 cocultures per group). l, Percentage of Ki67 HPFs after coculture with control or MIRO2-overexpressing A431 cells (n = 9 cocultures per group). m, Percentage of MitoTracker-high HPFs after coculture with MitoTracker-stained control or MIRO2-overexpressing A431 cells, with or without treatment with dihydrocytochalasin B (n = 3 cocultures per group). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (a,c–e,g,h,l) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (m) was used to determine statistical significance. Scale bars, 100 μm (b) and 25 μm (f).Source data a, RT–qPCR for MIRO2 using RNA from siCtrl or siMIRO2 A431 cells; Western blot of total and mitochondrial lysates from these cells for MIRO2, vinculin or HSP60 (loading controls) (n = 3 cultures per group). b, Fluorescence images of LifeAct A431 cells (red) stained with MitoTracker green and transfected with siCtrl or siMIRO2 in coculture with HPFs, counterstained with Hoechst. White arrowheads indicate A431 cells. c, Percentage of MitoTracker-high HPFs in cocultures with siMIRO2 or siCtrl A431 cells, normalized to the number of cancer cells (n = 3 cocultures per group). d, qPCR for mtDNA encoding tRNA-Leu(UUR) relative to nucDNA encoding B2M using DNA from A431 cells transfected with siCtrl or siMIRO2. Total mtDNA content was calculated on the basis of Ct values (n = 3 cultures per group). e, Mitochondrial mass in siCtrl and siMIRO2 A431 cells based on MitoTracker green mean fluorescence intensity (MFI) (n = 3 per group). f, Confocal images of siCtrl or siMIRO2 A431 cells incubated with MitoTracker green. The dashed line marks the outer edge of the cell (n = 3 cultures per group). g, RT–qPCR for INHBA and IL6 using RNA from HPFs incubated with CM of siCtrl or siMIRO2 A431 cells (n = 3 cultures per treatment group). h, RT–qPCR for INHBA and IL6 using RNA from sorted HPFs cocultured with siCtrl or siMIRO2 A431 cells (n = 3 cocultures per group). DC, direct culture. i, OCR of siCtrl or siMIRO2 A431 cells. The time of drug injection is indicated (n = 3 cultures per group). j, RT–qPCR for MIRO2 using RNA from A431 cells transfected with control or MIRO2 overexpression vectors (OE-MIRO2) (n = 3 cultures per group). Western blot of lysates from control or MIRO2-overexpressing A431 cells for MIRO2 or GAPDH (n = 2 cultures per group). k, Percentage of MitoTracker-high HPFs after coculture with MitoTracker-stained control or MIRO2-overexpressing A431 cells (n = 3 cocultures per group). l, Percentage of Ki67 HPFs after coculture with control or MIRO2-overexpressing A431 cells (n = 9 cocultures per group). m, Percentage of MitoTracker-high HPFs after coculture with MitoTracker-stained control or MIRO2-overexpressing A431 cells, with or without treatment with dihydrocytochalasin B (n = 3 cocultures per group). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (a,c–e,g,h,l) or two-sided one-way ANOVA with Bonferroni post hoc multiple comparison test (m) was used to determine statistical significance. Scale bars, 100 μm (b) and 25 μm (f). Source data Mitochondrial transfer was significantly reduced when HPFs were cocultured with siMIRO2 versus control A431 cells (Fig. 7c), while knockdown of MIRO1, TRAK1 and TRAK2 even increased the transfer (Extended Data Fig. 9c,d). MIRO2 knockdown did not lead to a decrease in mitochondrial DNA copy number or mitochondrial mass (Fig. 7d,e) and did not impede the activation of CAF marker expression through MitoCeption (Extended Data Fig. 9e), suggesting that MIRO2 is mainly responsible for the mitochondrial transfer rather than the effect on the recipient fibroblasts. The impaired mitochondrial transfer by MIRO2-knockdown cells correlated with perinuclear clustering of mitochondria (Fig. 7f). This is relevant, because proper distribution of mitochondria is important for protein secretion and cell migration. It is consistent with the role of MIRO family proteins in the regulation of mitochondrial distribution. The depletion of mitochondria at the periphery of cancer cells is likely to impact the mitochondrial transfer to fibroblasts. Incubation of fibroblasts with the CM of siMIRO2 versus control A431 cells did not significantly affect CAF marker gene expression, in contrast to the effect of MIRO2 knockdown in direct coculture (Fig. 7g,h). These findings again demonstrate that direct contact between cancer cells and fibroblasts, which allows mitochondrial transfer through TNTs, is necessary for the induction of a CAF phenotype. Depletion of MIRO2 did not significantly impact the metabolic activity of A431 cells (Fig. 7i). Therefore, the perinuclear clustering of mitochondria upon MIRO2 depletion is not attributed to major deficiencies in mitochondrial respiration or metabolism. Instead, it is likely a consequence of altered mitochondrial motility and distribution within the cell. Overexpression of MIRO2 promoted transfer activity of A431 and SCC13 cells and a mild effect was also seen for HaCaT cells, as shown by flow cytometry analysis of MitoTracker-high HPFs (Fig. 7j,k and Extended Data Fig. 9f). However, the increase in mitochondrial transfer from A431 cells to HPFs did not further promote HPF proliferation (Fig. 7l). Together with the results from MitoCeption studies with different amounts of mitochondria (Fig. 4k), these data suggest a threshold for CAF reprogramming, beyond which further mitochondrial transfer to fibroblasts has no additional effect. Inhibition of actin polymerization nearly abolished the elevated mitochondrial transfer from MIRO2-overexpressing A431 cells to fibroblasts (Fig. 7m). The selective effect of MIRO2 knockdown or overexpression on the number of MitoTracker-positive fibroblasts further confirms the suitability of MitoTracker staining under our experimental conditions, as dye leakage would not be affected by these treatments. Next, we tested the impact of MIRO2 depletion on the malignant features of A431 cells. A 24-h knockdown of MIRO2 significantly reduced their proliferation without impacting their viability (Fig. 8a,b). MIRO2-knockdown cells also migrated more slowly than control cells in a transwell assay and formed smaller, less developed 3D spheroids (Fig. 8c,d). Similar results were obtained with SCC13 cells (Fig. 8e–h). Lastly, NOD scid mice injected with siMIRO2 A431 cells failed to develop tumors, whereas control cells rapidly formed large tumors (Fig. 8i,j). Microscopic examination of the ear tissue from the siMIRO2 A431 group revealed small, undeveloped cancer cell colonies with undetectable MIRO2 expression (Fig. 8j, inset). Therefore, even a transient reduction of MIRO2 levels during the early phase of tumor formation is sufficient to prevent tumorigenesis.Fig. 8MIRO2 depletion in cancer cells reduces mitochondrial transfer and tumor growth.a, Percentage of Ki67 A431 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). b, Relative viability of A431 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). c, Transwell migration of A431 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). d, Relative spheroid area of a single hanging drop formed by siCtrl or siMIRO2 A431 cells and representative images of the spheroids (n = 9 spheroids per group). e, Percentage of Ki67 SCC13 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). f, Relative viability of SCC13 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). g, Transwell migration of SCC13 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). h, Relative spheroid area of a single hanging drop formed by siCtrl or siMIRO2 SCC13 cells and representative images of the spheroids (n = 15 spheroids per group). i, Photo of 5-week-old ear xenograft tumors (indicated by arrowheads) formed after injection of 200,000 A431 cells transfected with siMIRO2 or siCtrl and tumor volume at different time points (n = 3 tumors per group from different mice). j, Representative photomicrographs of Herovici-stained tumors (left) formed by siMIRO2 or siCtrl A431 cells and immunofluorescence staining of sections from these tumors for K14 (green) and MIRO2 (red), counterstained with Hoechst (blue). Inset, a tumor cell island with persistent MIRO2 knockdown (n = 3 sections from independent tumors per group). k, Normalized cell count of A431 LifeAct–RFP cells transfected with siCtrl or siMIRO2, cocultured with or without HPFs in spheroids and analyzed by FACS after 5 days (n = 3 spheroids per group). l, Tumor volume during tumor development following coinjection of HPFs (with or without A431-derived mitochondria, introduced using MitoCeption) and A431 cells transfected with either siMIRO2 or siCtrl (n = 5 tumors per group). m, Representative H&E stainings of tumors from each group (n = 5 mice). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (a–h) or two-sided one-way (k) or two-way (i,l) ANOVA with Bonferroni post hoc multiple comparison test (i,k,l) was used to determine statistical significance. Scale bars, 100 μm (d), 200 μm (j) and 1 mm (m).Source data a, Percentage of Ki67 A431 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). b, Relative viability of A431 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). c, Transwell migration of A431 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). d, Relative spheroid area of a single hanging drop formed by siCtrl or siMIRO2 A431 cells and representative images of the spheroids (n = 9 spheroids per group). e, Percentage of Ki67 SCC13 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). f, Relative viability of SCC13 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). g, Transwell migration of SCC13 cells 24 h after transfection with siCtrl or siMIRO2 (n = 3 cultures per group). h, Relative spheroid area of a single hanging drop formed by siCtrl or siMIRO2 SCC13 cells and representative images of the spheroids (n = 15 spheroids per group). i, Photo of 5-week-old ear xenograft tumors (indicated by arrowheads) formed after injection of 200,000 A431 cells transfected with siMIRO2 or siCtrl and tumor volume at different time points (n = 3 tumors per group from different mice). j, Representative photomicrographs of Herovici-stained tumors (left) formed by siMIRO2 or siCtrl A431 cells and immunofluorescence staining of sections from these tumors for K14 (green) and MIRO2 (red), counterstained with Hoechst (blue). Inset, a tumor cell island with persistent MIRO2 knockdown (n = 3 sections from independent tumors per group). k, Normalized cell count of A431 LifeAct–RFP cells transfected with siCtrl or siMIRO2, cocultured with or without HPFs in spheroids and analyzed by FACS after 5 days (n = 3 spheroids per group). l, Tumor volume during tumor development following coinjection of HPFs (with or without A431-derived mitochondria, introduced using MitoCeption) and A431 cells transfected with either siMIRO2 or siCtrl (n = 5 tumors per group). m, Representative H&E stainings of tumors from each group (n = 5 mice). Graphs show the mean ± s.e.m. An unpaired two-sided Student’s t-test (a–h) or two-sided one-way (k) or two-way (i,l) ANOVA with Bonferroni post hoc multiple comparison test (i,k,l) was used to determine statistical significance. Scale bars, 100 μm (d), 200 μm (j) and 1 mm (m). Source data To determine whether the poor spheroid growth of MIRO2-knockdown cells and their failure to form tumors in mice is simply a consequence of their cell-autonomous defect in proliferation or migration or whether it involves non-cell-autonomous effects, such as impaired mitochondrial transfer to fibroblasts, we set up coculture spheroid assays. The difference in proliferation between siCtrl and siMIRO2 cancer cells was even more pronounced in the presence of fibroblasts (Fig. 8k), indicating an important non-cell-autonomous role of MIRO2. In xenograft experiments, coinjection of siMIRO2 A431 cells with control HPFs already caused a mild stimulation of tumor growth but the tumor-promoting effect was much stronger with MitoCepted HPFs. This combination compensated for the deficiency of siMIRO2 cancer cells in tumor formation and the rate of tumor growth was almost comparable to that of the control group, in which siCtrl A431 cells were coinjected with control fibroblasts (Fig. 8l,m). These findings highlight the notable influence of fibroblasts with mitochondria from cancer cells on tumor formation and further suggest an important role of MIRO2 in this transfer. We identified mitochondrial transfer through TNTs as a strategy of cancer cells to promote CAF differentiation. Because cancer cells and CAFs often have direct contact in the tumor, particularly at its periphery, this transfer is likely to contribute to the increased invasiveness of cancer cells. We further show that CAF differentiation through mitochondrial transfer from cancer cells is supported by two mechanisms. First, cancer cells at the invasive front overexpress MIRO2, which promotes the mitochondrial transfer. Second, mitochondria from malignant cells but not from nontumorigenic epithelial cells can induce a CAF phenotype. This could be explained by alterations in the proteome of mitochondria from cancer cells and the associated metabolic alterations. In support of this hypothesis, uptake of mitochondria from A431 cancer cells altered the expression of several metabolic proteins and promoted OxPhos and ATP production in the recipient fibroblasts. These features were shown to promote proliferation, matrix production and protein secretion by fibroblasts and CAFs. Consistent with an important role of OxPhos in the fibroblast reprogramming by mitochondrial transfer, inhibition of this metabolic pathway in the recipient fibroblasts blocked the induction of important CAF features. Therefore, metabolic alterations induced by mitochondria from cancer cells contribute to the CAF phenotype. The induction of a CAF phenotype was associated with significant changes in the expression of genes and proteins associated with inflammation, immune response, cellular metabolism and stress response. In the cocultures, we found increased expression of many ISGs. This is consistent with the sensing of genomic damage of cancer cells by fibroblasts, which resulted from transcytosis of cytoplasm from cancer cells into neighboring fibroblasts and activation of the stimulator of interferon genes–interferon regulatory factor 3 pathway. However, ISG expression was not upregulated in the MitoCepted fibroblasts, demonstrating that they are not notably affected by cancer cell mitochondria. By contrast, several classical CAF markers were upregulated in the MitoTracker-high population and in MitoCepted fibroblasts, suggesting that coculture with cancer cells alters the expression of different sets of genes in fibroblasts through distinct mechanisms. In a search for the mechanistic underpinning of the mitochondrial transfer, we identified MIRO2. This Rho guanosine triphosphatase links mitochondria to the cellular trafficking machinery and is responsible for intracellular mitochondrial positioning. It also has important cell-autonomous functions that are important for cancer cell proliferation and migration, as also shown in this study. These features and its overexpression in invasive cancer cells at the edge of different tumors make it an interesting target for cancer treatment. This is not restricted to skin cancer because we also observed mitochondrial transfer from vulvar, breast and pancreatic cancer cells to fibroblasts, which promoted a CAF phenotype. Given the high expression of MIRO2 in metastatic prostate cancer, it will be of interest to study the role of intercellular mitochondrial transfer in metastasis. In conclusion, we show that cancer cells transfer their mitochondria to fibroblasts and thereby reprogram them into protumorigenic CAFs. We also discovered the mechanism underlying this transfer and identified MIRO2 as a potential target for cancer treatment. These findings offer promising therapeutic opportunities for skin cancer and also for malignancies with high mortality, such as pancreatic cancer, which has a large stromal component. Lastly, the data obtained in this study suggest that mitochondrial transfer from epithelial to stromal cells is an important mechanism of cell–cell communication, which may also be relevant for development, homeostasis and tissue repair and for the pathogenesis of nonmalignant diseases. The work performed in this study complies with all relevant ethical regulations. Mouse maintenance and experimentation were approved by the veterinary authorities of Zurich (Kantonales Veterinäramt Zurich, 32060, 35555, 36338 and 33866). Human skin and tumor samples, which were used for the isolation of primary cells, were obtained anonymously from the Department of Dermatology, University Hospital of Zurich (in the context of the biobank project). Written informed consent for use in research was obtained from all donors (in case of foreskin from the parents). All experiments with human samples were approved by the local and cantonal Research Ethics Committees (Kantonale Ethikkommission Zurich, BASEC no. 2017-00684), adhering to the Declaration of Helsinki Principles. NOD scid (NOD.CB17-Prkdc/NCrCrl) mice were bred in the ETH Zurich EPIC facility and kept under specific pathogen-free conditions in a 12 h dark–light cycle at 21–23 °C and 40–60% humidity. They received food and water ad libitum. HaCaT, HaCaT-Ras cells and SCC13 cells were provided by P. Boukamp. A431 cells were from Merck (85090402). LifeACT cell lines were generated by transduction with the lentiviral vector rLV-Ubi-LifeAct–RFP-Tag (Vitaris). HaCaT cells are spontaneously immortalized but nontumorigenic human keratinocytes. HaCaT-Ras cells were obtained by transfection of HaCaT cells with a c-HA–RAS oncogene. SCC13 cells were derived from a human cutaneous SCC. The metastatic MDA-MB-231 triple-negative breast cancer cell line and the metastatic PANC1 pancreatic cancer cell line were obtained from the American Type Culture Collection (HTB-26, CRL-1469). LM2 cells, a lung metastatic variant of MDA-MB-231, were kindly provided by J. Massagué. Cell lines expressing fluorescent mitochondrial proteins were generated by lentiviral transduction. Primary fibroblasts expressing TOM20–GFP were generated by lentiviral transduction with pLenti-X1-blast-GFP-TOM20-MTS, provided by J. Corn (ETH Zurich). Immortalized mouse fibroblasts were isolated from PDGFRα–eGFP transgenic mice and spontaneously immortalized by serial passaging. Authentication of HaCaT, HaCaT-Ras, SCC13 and A431 cells was performed by Microsynth using highly polymorphic short tandem repeat loci (most recently in February 2025). Absence of Mycoplasma was confirmed monthly using the PCR Mycoplasma test kit I/C (PromoKine) or the MycoStrip kit (InvivoGen). Cells were cultured in DMEM supplemented with 10% FBS and 1% penicillin–streptomycin (complete DMEM), unless indicated otherwise. HPKs from skin of adult healthy volunteers or from the edges of skin SCCs of adult participants (diagnosed by an experienced dermatopathologist) were from H.-D. Beer (University Hospital Zurich). Foreskin HPFs were obtained from foreskin of healthy boys. Human skin and tumor samples, which were used for the isolation of primary cells, were obtained anonymously from the Department of Dermatology, University Hospital of Zurich (in the context of the biobank project). Written informed consent for use in research was obtained from all donors (in case of foreskin from the parents). All experiments with human samples were approved by the local and cantonal Research Ethics Committees (Kantonale Ethikkommission Zurich, BASEC no. 2017-00684), adhering to the Declaration of Helsinki principles. HPKs were cultured in keratinocyte serum-free medium with epidermal growth factor and bovine pituitary extract (all from Thermo Fisher Scientific). HPFs were cultured in complete DMEM. Cancer cells were transfected with siRNAs (Microsynth AG) using Lipofectamine RNAiMAX (Thermo Fisher Scientific) and incubated for 24–72 h or retransfected after 72 h and incubated for an additional 24 h. The following siRNAs were used: Connexin 26 siRNA (72 h): 5′-CCCAGUUGUUAGAUUAAGATT-3′ MIRO1 siRNA (72 h + 24 h): 5′-UAACCAAAUCGUCGAAGCACAGUCCTT-3′ MIRO2 siRNA (24 h): 5′-GCGUGGAGUGUUCGGCCAATT-3′ SEC3 siRNA (72 h): 5′-AGAUGAAUACCAAGAGUUA-dTdT-3′ SEC5 siRNA (72 h): 5′-GGGUGAUUAUGAUGUGGUUdTdT-3′ TRAK1 siRNA (72 h): 5′-GGAAACGAUGAGCGGAGUATT-3′ TRAK2 siRNA (72 h): 5′-GGAUAGAUAUGCACUGAAATT-3′ Negative control siRNA: 5′-AGGUAGUGUAAUCGCCUUG-3′ First, 1 μg of pRK5-myc-MIRO2 expression vector (Addgene, 47891) or empty vector were used for transfection using Lipofectamine 2000 (11668019, Thermo Fisher Scientific). DNA, Lipofectamine and Opti-MEM reduced-serum medium (31985062, Thermo Fisher Scientific) were incubated at room temperature for 20 min. Cells were incubated with the mixture for 6 h, followed by incubation in complete DMEM for 24 h. Xenograft skin tumorigenesis assays were performed as described by intradermal injection of 2 × 10 cancer cells or 10 cancer cells together with 10 fibroblasts into the ear of male NOD scid mice at the age of 8–12 weeks. Breast and pancreatic cancer xenografts were established by orthotopic injection of LM2 breast cancer cells into the mammary fat pad (glands 2–3) of adult female NOD scid mice or by direct injection of PANC1 cells into the pancreas of adult mice. The maximal tumor size permitted by the ethics committee (1-cm diameter for skin cancer, 2.8-cm volume for breast cancer) or the end point for wellbeing (hunching, piloerection or decreased activity for pancreatic cancer) was never reached in our experiments. Spheroid assays were performed using the experimental parameters proposed by The MISpheroID Consortium. A total of 2,000 cancer cells in 20 μl of CM from HPFs, which were cultured in complete DMEM, were placed on the lids of 6-cm culture plates using a hanging-drop method. Then, 5 ml of PBS was added to the bottom. Equal numbers of HaCaT or cancer cells and fibroblasts were seeded to reach 80–100% confluency. Before coculture, cells were stained with MitoTracker or PKH67 cell linker (PKH67GL, Sigma-Aldrich) and washed with PBS. For imaging, cells were cultured on glass coverslips in complete DMEM and imaged using an Axio Imager M2 microscope equipped with an Axiocam MR camera and ZEN 2 software or using an Axioscan 7 microscope slide scanner equipped with a color Axiocam 705 color complementary metal–oxide–semiconductor camera and a fluorescence Axiocam 712 mono camera (all from Carl Zeiss). Image processing and analysis were performed using Fiji ImageJ (https://imagej.net/Fiji) or QuPath. For mechanistic studies, cocultures were treated with dihydrocytochalasin B (100 nM), nocodazole (10 μM) or carbenoxolone (100 μM) (all from Sigma-Aldrich) for 24 h. Injection of mitochondria was performed using a FluidFM setup as previously described. Imaging was carried out using a spinning disc confocal microscope (Visitron Systems) with a Yokogawa CSUW1 scan head and an electron-multiplying charge-coupled device camera system (Andor Oxford Instruments). A total of 100,000 fibroblasts were seeded in two-well microinsets 6 h before injection. The microfluidic probe was positioned over individual fibroblasts and mitochondria were inserted using a cantilever system. Z stacks were taken to identify successful transfer of mitochondria. Consequently, 100 fibroblasts were injected and fixed in 4% paraformaldehyde (PFA) for 24 h after transfer and analyzed by immunofluorescence staining. Mitochondria were extracted from 20,000,000 epithelial cells using a mitochondria isolation kit (89874, Thermo Fisher Scientific). MitoCeption was performed as described using preseeded cells in six-well culture plates. A total of 100,000 HPFs were seeded the day before MitoCeption. Mitochondria isolated from 100,000 cells of various cell lines or primary cells, which included similar amounts of total protein (Extended Data Figs. 4c and 6d), were added to the bottom, ensuring even distribution. Culture plates were centrifuged at 1,500g for 15 min at 4 °C and incubated at 37 °C and 5% CO2 for 2 h. The centrifugation procedure was repeated and cells were cultured for 24 h before further processing. For OxPhos inhibition, recipient fibroblasts were treated with 1 μM oligomycin (O4876, Sigma-Aldrich) for 24 h. All experiments were performed with consistent exposure time of cells to mitochondria, uniform mitochondrial uptake across experiments and standard post-transplantation conditions. Data were normalized to the number of donor cells. Cells were stained with Sytox Blue (S34857, Thermo Fisher Scientific) before acquisition or analysis. Live MitoTracker-high and MitoTracker-low cells or Su9–RFP-high and Su9–RFP-low cells were sorted using a FACSAria Fusion sorter (BD Biosciences) with a 100-μm nozzle and 20 psi pressure on the basis of their fluorescence intensity. The number of cells was normalized after each FACS run. To analyze proliferation, cells were fixed and permeabilized using a Foxp3 transcription factor staining buffer set (00-5523-00, eBioscience) before intracellular staining with PE-conjugated (sc-7846, Santa Cruz) or PE–Cyanine7-conjugated (25-5698-82, eBioscience) anti-Ki67 for 1 h at room temperature. For mitochondrial content analysis, cells were stained with MitoTracker green FM (M46750, Thermo Fisher Scientific). Flow cytometry was performed using an LSR Fortessa or a FACSAria Fusion cell analyzer (both from BD Biosciences). Data were analyzed using FlowJo version 10.10 software (BD Biosciences). Chemotactic transwell migration was assessed as described previously and cancer cells were allowed to migrate for 24 h toward CM from fibroblasts. Transwell coculture assays were previously described. A total of 100,000 cells per well were seeded on XF96 Seahorse plates in full medium. The medium was then switched to Seahorse XF base medium (103335-100, Agilent Technologies) supplemented with 10 mmol L glucose, 1 mmol L sodium pyruvate and 2 mmol L glutamine (assay concentration, https://www.agilent.com/cs/library/usermanuals/public/XF_Cell_Mito_Stress_Test_Kit_User_Guide.pdf) and incubated in a CO2-free incubator for 1 h. Oligomycin, carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (FCCP) and antimycin A (AA) + rotenone were prepared in the XF assay medium with final concentrations of 1, 1.5 and 1/0.1 μmol L, respectively, and provided by the Seahorse XF cell mito stress test kit (103015-100; Agilent Technologies). The compounds were injected to assess the OCR of cells in an XF96 plate. Metabolic flux data were normalized to cell count, which was determined using Hoechst staining and analysis in a fluorescence reader (Agilent Technologies, BioTek Cytation 1) on the day of analysis. ATP levels were determined using the CellTiter-Glo assay kit (Promega). Mitochondrial ROS were measured as previously described using MitoSOX-Red (M36008, Thermo Fisher Scientific). ECM was decellularized with 0.5% Triton X-100 in 20 mM NH4OH for 1–3 min and washed with PBS before fixation with 4% PFA for 15 min at room temperature. Immunostaining was performed to analyze the expression of ECM proteins in the dECM. Hoechst staining was performed to assess the efficiency of decellularization. Histological analyses and immunostainings were performed as described previously using the antibodies listed in Supplementary Table 2. Immunofluorescence images were analyzed using Fiji and staining intensity was normalized to cell number with at least nine microscopic fields of view for each condition analyzed. Mitochondrial networks were analyzed with MiNA as previously described. The relative distance is indicated with values from 1 to 10; at least 100–200 intensity profiles were measured. Colony area and dECM area were measured after thresholding the images using ImageJ (Fiji). Colocalization analysis was adapted from Delaunay et al.. Fluorescent intensity profile on the specified line was measured using ImageJ (Fiji) and normalized by the highest intensity value. All images from the same experiment were processed in an identical way by adjusting brightness and contrast and subtracting background signal to identify cell edge and contour or thresholding with the same values using a wide-field microscope. The length of TNTs between cancer cells and fibroblasts was measured using the line or Polygon tool in QuPath. Holotomographic imaging was performed with a Tomocube HT-X1 microscope (Tomocube). Cells were seeded on glass-bottomed dishes (P06-1.5H-N, Cellvis) at a density of 1 × 10 cells per dish. The laser module was aligned for optimal illumination of the sample. The imaging process captured the cells’ refractive index and immunofluorescence labeling. Videos were acquired using a high-speed camera for 6 h with a time interval of 5 min. Finally, Tomocube software was used to process the acquired images. RNA isolation and RT–qPCR were performed as described previously using the primers listed in Supplementary Table 3. Values obtained for the first control were set to 1. Western blot analysis was performed using standard procedures and antibodies to MIRO2 (H00089941-B01P, Novus Biologicals; 1:1,000 diluted), MIRO1 (NBP1-59021, Novus Biologicals; 1:500 diluted), TRAK1 (PA5-70029, Invitrogen; 1:500 diluted), TRAK2 (PA5-21858, Invitrogen; 1:500 diluted), EXOC1 (ab251853, Abcam; 1:500 diluted), EXOC2 (ab140620, Abcam; 1:500 diluted), vinculin (V4505, Sigma-Aldrich; 1:2,000 diluted), GAPDH (5G4, Hytest; 1:10,000 diluted) and HSP60 (Ab59457, Abcam, 1:500 diluted). Secondary antibodies were anti-rabbit or anti-mouse IgG (W4011 and W4021, Promega; 1:10,000 diluted) conjugated with horseradish peroxidase. Band intensities were quantified with ImageJ. Cells were collected with a cell scraper, transferred into Eppendorf tubes, and centrifuged at 6,500 rpm for 5 min. The pellet was resuspended in 200 µl of lysis buffer supplemented with 10 µl of proteinase K (10 mg ml; AppliChem). Samples were incubated overnight at 55 °C and then for 10 min at 95 °C to inactivate proteinase K and centrifuged at 17,000g for 10 min at room temperature. The supernatant, containing crude DNA, was retained for qPCR analysis. DNA was isolated from approximately 3 × 10 immortalized fibroblasts from PDGFRα–eGFP transgenic mice and from A431 cells using a QIAamp DNA mini kit (51306, Qiagen). Both cell types were cocultured for 24 h, followed by FACS isolation of mouse fibroblasts. DNA was isolated and analyzed by PCR using primers that amplify the gene encoding human mitochondrial 16S RNA. PCR products were visualized on a 2% agarose gel. The PCR product was purified using a QIAquick PCR purification kit (28104, Qiagen) and sequenced. Species-specific and cell-type-specific SNPs were determined by comparing individual chromatograms using SnapGene software (GSL Biotech). scRNA-seq and spatial transcriptomics 10X Visium analysis was performed on the basis of published SCC datasets. For scRNA-seq, prefiltered data were used according to the quality control procedures of each paper. The logged count per 10k (CP10k) was used for normalization. The distribution of expression of MIRO2 was compared using a Wilcoxon rank-sum test. For spatial transcriptomics, the spots were normalized using the logged CP10k expression. To analyze the colocalization of MIRO2 mRNA positive spots with CAF subtypes, we first assigned spots to fibroblasts or CAFs (positive for PDGFRA), to an iCAF subtype (positive for PDGFRA and MMP11), to a myofibroblast CAF subtype (positive for PDGFRA and ACTA2), to an adiCAF subtype (positive for PDGFRA and CFD) or to INHBA+ fibroblasts (expression of INHBA detected). To assess statistical significance of MIRO2 and CAF subtype spots (INHBA), we computed the six nearest neighbors of each spot and compared, using Fisher’s exact test, the enrichment of neighboring spots expressing both MIRO2 and the CAF subtype marker or INHBA versus noncolocalized spots. For spatial transcriptomics data, the leading edge annotations were obtained from the authors. The effect of MIRO2 knockdown in genome-wide knockdown screens was analyzed using DepMap (DepMap.org). TCGA clinical, survival and RNA-seq data from primary tumors of 8,911 participants across 30 solid cancer types were downloaded from the UCSC Xena data hub (https://xena.ucsc.edu) using the UCSCXenaTools R package (version 1.4.8). Gene expression values were downloaded as log2-transformed RSEM normalized counts. The continuous MIRO2 gene expression was used for survival analysis censored at 5 years of follow-up. Hazard ratios were computed using the Cox proportional hazard model implemented in the ‘coxph’ function from the R package survival (version 3.5-7). For visualization, gene expression was divided into terciles and Kaplan–Meier survival curves were computed using the R package ggsurvfit (version 0.3.1). The leading edge expression signature was composed by 91 genes upregulated at the leading edge compared to the tumor core. For each TCGA sample, a signature enrichment score was computed using the ‘gsva’ method from the gene set variation analysis R Bioconductor package (version 1.46.0). The correlation between the leading edge signature score and MIRO2 expression was assessed using Pearson’s correlation. Random-effects meta-analysis across all cancer types was conducted with the ‘metagen’ function from the R package meta (version 7.0-0). Total RNA was isolated as described above. RNA-seq was performed after poly(A) enrichment and True-Seq library preparation on a Novaseq 6000 sequencer (Illumina). FastQ files were trimmed using Trimmomatic (version 0.36), and processed using Salmon (version 1.10.2) using default parameters. The count matrix was processed using tximport to obtain gene-level counts and transcripts per million (TPM) estimates. PCA was computed on the standardized TPM expression. Differential expression among control, MitoTracker-high and MitoTracker-low fibroblasts was computed using PyDESeq2 (https://github.com/owkin/PyDESeq2), a Python variant of DESeq2 (ref. ). Significance levels were cut at 10. To compare the markers of MitoTracker-high fibroblasts with known CAF subtypes, we defined a gene signature for MitoTracker-high fibroblasts using genes that were significantly differentially expressed compared to both MitoTracker-low fibroblasts (adjusted P < 0.01) and control populations (adjusted P < 0.01) and overexpressed in the MitoTracker-high population (fold change (FC) > 1.25). We computed the enrichment of this signature in known CAF subtype signatures with Fisher’s exact test, using genes quantified in the RNA-seq experiment as background. Single-cell data from human SCCs were used as a reference to deconvolve Visium 10X spots. A negative binomial model was trained on the discovery cohort with default parameters to estimate cell-type-specific average gene expression profiles. Next, the Cell2Location model was applied, setting the prior to n = 15 average cells per spot and α = 20 for relaxed regularization, which produced estimated counts of each cell type per spot. We next sampled from the model’s posterior distribution to determine cell-type-specific gene expression per spot. Fibroblast-specific expression estimates per spot were used to score CAF subtypes, leveraging marker genes from human SCCs. Additionally, we assessed MIRO2 expression using the estimated raw counts for keratinocytes. Samples were prepared using SP3 technology. Briefly, cell pellets were lysed in radioimmunoprecipitation assay buffer (50 mM Tris, 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate and 0.1% SDS) and lysates were sonicated in an ultrasonic bath. Protein amount was quantified with a bicinchoninic acid assay (PIER23225, Pierce Biotechnology, Thermo Fisher Scientific). Then, 10 µg of protein was used for downstream analyses. After reduction of proteins with 5 mM dithiothreitol (DTT), they were alkylated with 5.5 mM iodoacetamide and quenched with 5 mM DTT. SP3 beads (Sera-Mag SpeedBeads, GE Healthcare, 45152105050250 and 65152105050250) were added to protein lysates in a 10:1 ratio. Binding to the beads was induced by the addition of 100% ethanol. After rinsing of beads with 80% ethanol, samples were digested overnight with trypsin (1:25) in 100 mM ammonium bicarbonate, pH 8. Peptides were desalted using STAGE tips and adjusted to a concentration of 100 ng µl in 0.1% formic acid. Peptides were analyzed by LC–MS/MS on a Vanquish Neo ultrahigh-performance LC instrument coupled to an Orbitrap Exploris 480 (both from Thermo Fisher Scientific) as previously described. Briefly, samples were applied to fused silica C18 column tips (inner diameter: 75 μm; New Objective), produced in house with ReproSil-Pur 120 C18 AQ (1.9 μm, length: 20 cm; Dr. Maisch) using a mixture of water (solvent A) and 80% acetonitrile in water (solvent B), both acidified with of 0.1% formic acid. Samples were separated at a flow rate of 250 nl min within 85 min (5–30% solvent B). Data were acquired by data-independent acquisition (DIA; full MS, 350–1,200 m/z; 120,000 resolution; maximum injection time, 60 ms; 28 MS/MS scans with a width of 30.4 m/z; 1-Da overlap). A normalized automatic gain control target value of 300%, resolution of 30,000 and normalized stepped collision energy of 25.5%, 27% and 30% were used. The MS raw files were processed with Spectronaut 17, direct DIA+, using a full-length Homo sapiens database (UniProt, January 2022) and common contaminants, such as trypsin and keratins, as reference. Data analysis was performed using Perseus 2.0.9.0. Values below 5 after log2 transformation (result of matching across runs in Spectronaut 17) were transformed to nonvalid values. To determine significant differences in protein abundance, each condition was first compared to the control by standard t-test. Only proteins that were significantly more or less abundant in minimal one condition were further analyzed. Missing values were replaced by normally distributed random values. Width and down shift were used separately for each column according to default settings. After grouping replicates, significant changes were determined using a permutation-based false discovery rate (FDR ≤ 0.05). Hallmarks of Cancer and Wikipathways were downloaded from the Molecular Signature Database (MSigDB) website (https://www.gsea-msigdb.org/gsea/msigdb/). GSEApy (https://gseapy.readthedocs.io/en/latest/introduction.html) was used to quantify the enrichment of pathways in transcriptomic and proteomic data, using a ranked gene list based on log2FC as input, with a minimum gene set size of 5, maximum size of 1,000 and 500 permutations. Pathways that were significant at FDR P < 0.1 were reported. No statistical method was used to predetermine sample size. Sample sizes were determined on the basis of previous studies by us and others using similar technologies and approaches. For mouse experiments, the number was chosen to comply with 3R principles. No animals or data points were excluded from the analyses. Randomization was used for animal experiments and mice were blindly selected before injection. All statistical data are based on biological replicates. Statistical analysis was performed with PRISM software, version 9 for Mac OS X or Windows (GraphPad Software). All experiments were performed at least twice with similar results. An unpaired two-sided Student’s t-test was used for the comparison of two groups, assuming a normal distribution, which was, however, not formally tested. A Mann–Whitney U-test was used for data that were not normally distributed. For comparisons involving more than two groups, we used a two-sided one-way or two-way analysis of variance (ANOVA) with Bonferroni post hoc multiple comparison test. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
PMC12141184
Curcumin inhibits IFN-γ induced PD-L1 expression via reduction of STAT1 Phosphorylation in A549 non-small cell lung cancer cells
Immune evasion in non-small cell lung cancer (NSCLC) is largely mediated by programmed death-ligand 1 (PD-L1), which is upregulated by interferon-gamma (IFN-γ)-induced STAT1 activation. Targeting this pathway may improve immunotherapy outcomes. Curcumin, a natural polyphenol, has been reported to modulate various oncogenic signaling pathways, but its role in inhibiting IFN-γ-driven PD-L1 expression in NSCLC remains unclear. The NSCLC cell line A549 were treated with curcumin (50 µM) for 2 h before stimulation with IFN-γ (500 U/ml). Western blot, qRT-PCR, and immunofluorescence microscopy were used to evaluate STAT1 phosphorylation, PD-L1 expression, and the localization of phosphorylated STAT1 (p-STAT1). The expression of interferon-stimulated genes (ISGs), including SOCS1 and ISG15, was also examined. Additionally, the Resazurin assay was performed to assess cell viability. IFN-γ significantly induced STAT1 phosphorylation, leading to a time-dependent upregulation of PD-L1 expression. Immunofluorescence confirmed that p-STAT1 is translocated to nucleus. Curcumin treatment inhibited STAT1 phosphorylation by 68% (p < 0.001), leading to a marked reduction in PD-L1 expression. Moreover, curcumin suppressed IFN-γ-induced SOCS1 (63%) and ISG15 (54%) expressions, indicating a broader effect on STAT1-mediated immune evasion. Finally, curcumin enhanced IFN-γ-mediated growth inhibition, reducing cell viability by 47% at 48 h (p < 0.01). Curcumin effectively inhibits IFN-γ-induced STAT1 phosphorylation and PD-L1 expression, downregulates ISGs, and enhances IFN-γ-mediated tumor suppression. These findings suggest that curcumin may serve as a therapeutic adjuvant in NSCLC, potentially improving immune checkpoint inhibitor (ICI) efficacy.Lung cancer continues to be the most prevalent cancer-related mortality globally, responsible for more than 1.8 million deaths per year (Siegel et al. 2024). Non-small cell lung cancer (NSCLC) accounts for ~ 85% of all lung cancer cases, an issue that is of significant public health importance and also a target for oncological research (Salih et al. 2025). Over the last decade, the development of immune checkpoint inhibitors (ICIs) targeting the programmed death- 1/programmed death-ligand 1 (PD- 1/PD-L1) axis has shifted the treatment landscape for NSCLC, significantly improving survival rates in a particular group of patients (Rizvi et al. 2015). Nevertheless, with these advancements, a large proportion of NSCLC patients either show a lack of responsiveness to ICIs (primary resistance) or develop resistance ("acquired resistance") over time that significantly restricts their clinical efficacy (Sharma et al. 2017; Rotte 2019). Immune evasion of NSCLC is significantly attributed to the expression of PD-L1, a transmembrane molecule that binds to PD- 1 receptors located on cytotoxic T cells and induces immune suppression and tumor tolerance. PD-L1 overexpression is often linked to poor prognosis, tumor aggression, and poor response to ICIs (Rizvi et al. 2015; Kim and Chen 2016). One of the major regulators of PD-L1 expression in the tumor microenvironment is IFN-γ, a key cytokine modulating the immune system. Although IFN-γ is classically described to have tumor-suppressive effects, paradoxically, it also plays a pro-tumorigenic role by inducing PD-L1 expression, enabling tumor cells to escape immune surveillance (Garcia-Diaz et al. 2017). IFN-γ signaling is largely achieved via Janus Kinase—Signal Transducer and Activator of Transcription (JAK-STAT) pathway and, in particular, via STAT1. Binding of IFN-γ to its receptors, JAK1 and JAK2, phosphorylates STAT1 on Tyr701. Upon phosphorylation, STAT1 dimerizes, translocate to the nucleus, and binds to gamma-activated sequence (GAS) motifs of the PD-L1 promoter, in turn increasing the level of PD-L1 transcription (Sumitomo et al. 2022). As confirmed by several studies, IFN-γ-induced activation of STAT1 is one of the most potent inducers of PD-L1 expression in NSCLC, which in turn leads to adaptive immune resistance that prevents T-cell-mediated tumor destruction (Drake et al. 2006; Shin et al. 2017). Since IFN-γ levels are typically increased in the tumor microenvironment of NSCLC, strategies that can inhibit IFN-γ-driven upregulation of PD-L1 promise to be valuable for improving the efficacy of ICIs (Spranger et al. 2013). Despite extensive research regarding PD-L1 regulation in NSCLC, therapeutic approaches that specifically aim to silence IFN-γ-induced PD-L1 expression are still limited. Strategies presently are based on ICIs in synergy with kinase inhibitors, epigenetic modulators, or chemotherapy, but such pharmacologic efforts are typically associated with toxicity and short-term failure (Zhou and Yang 2023). Accordingly, the discovery of non-toxic, naturally derived compounds that effectively and specifically suppress IFN-γ-mediated PD-L1 expression but do not negatively affect the host immune function is of significant interest. Curcumin, a polyphenolic compound isolated from the extract of Curcuma longa (turmeric), has attracted attention as a multi-targeted therapeutic agent due to its anti-inflammatory, antioxidant, and anti-cancer functions (Allegra et al. 2017). It has been shown to suppress a variety of oncogenic signaling pathways such as nuclear factor-kappa B (NF-κB), cyclooxygenase- 2 (COX- 2), STAT3, and AKT, leading to growth inhibition and apoptosis in a variety of cancers (Kumar et al. 2021). Several preclinical studies have suggested that curcumin can downregulate PD-L1 expression in different cancer models, including breast and colon cancer, by inhibiting STAT1 phosphorylation and preventing its nuclear translocation (Midura-Kiela et al. 2012). Nevertheless, its specific involvement in downregulating IFN-γ-induced PD-L1 expression in NSCLC, especially in A549 cells, is unknown. As high PD-L1 expression is associated with poor response to ICIs, understanding natural products (like curcumin) that could regulate IFN-γ-induced PD-L1 expression might provide new strategies for upgrading the outcome of NSCLC treatment. Although these encouraging results are promising, there are some deficiencies in our knowledge about how curcumin functions in regulating IFN-γ/STAT1 signaling in lung cancer. Firstly, while previous reports have shown that curcumin can block the phosphorylation of STAT1, they have not specifically evaluated the effect of curcumin on the expression of PD-L1 in NSCLC. Second, the underlying mechanism of curcumin's inhibitory action on STAT1 is still unknown, especially whether curcumin directly inhibits STAT1 phosphorylation or inhibits its nuclear translocation. Third, the ability of curcumin to enhance the efficacy of ICIs by decreasing PD-L1 expression has not been extensively studied, and this should be explored. Given these gaps in knowledge, the present study aims to investigate the effect of IFN-γ on STAT1 phosphorylation and PD-L1 expression in A549 cells and determine whether curcumin can inhibit IFN-γ-induced STAT1 activation and subsequent PD-L1 upregulation. In addition, explore the impact of curcumin on IFN-γ-induced expression of interferon-stimulated genes (ISGs), such as SOCS1 and ISG15, which are involved in immune signaling and tumor immune escape, and evaluate whether curcumin enhances the anti-proliferative effects of IFN-γ, suggesting a potential therapeutic benefit beyond immune modulation. Human non-small cell lung cancer (NSCLC) cell, A549, was obtained from American Type Culture Collection (ATCC) and cultured in Dulbecco's Modified Eagle Medium (DMEM) containing 10% fetal bovine serum (FBS) (Life Technologies), 1% penicillin–streptomycin, and 2 mM L-glutamine. Cells were cultured in a humidified incubator at 37 °C and 5% CO₂ with regular 2–3-day passaging in a 0.25% trypsin–EDTA suspension. All experiments were performed on cells between passage 5 and passage 15, thereby permitting reproducibility. Recombinant human IFN-γ was purchased from PeproTech (EC Ltd PeproTech, London, UK) and stored in sterile phosphate-buffered saline (PBS) at 10 U/ml stock concentration. Curcumin ≥ 95% was purchased from Santa Cruz (Heidelberg, Germany) and dissolved in dimethyl sulfoxide (DMSO) (≥ 99.9% purity) to make a 50-mM solution. The working dilutions of curcumin were freshly prepared in the complete DMEM prior to each experiment. Cells were pretreated with curcumin (50 µM) for 2 h prior to the IFN-γ (500 U/ml) stimulation for studied time points. Control groups were treated with vehicle (DMSO < 0.1%). Cell viability was assessed using the Resazurin assay (Sigma-Aldrich, #R7017 - 1G). A549 cells were seeded in 96-well plates at a density of 3 × 10 cells per well (final volume of 100 μL/well) and allowed to adhere overnight. Following 24 h treatment with IFN-γ, curcumin, or a combination of both, 20 µl of Resazurin reagent (0.15 mg/ml) was then added to each well, and incubation was performed for 2–4 h at 37 °C. The conversion of Resazurin to resorufin was measured by fluorescence using a microplate reader (excitation 560 nm, emission 590 nm) or by absorbance at 570 nm with 600 nm as a reference wavelength. Cell viability was determined as a percentage compared to untreated control cells. Cell Viability (%) = (Absorbance of control cells of treated cells​) × 100. Total protein lysates were extracted from treated A549 cells using ice-cold radioimmunoprecipitation assay (RIPA) buffer supplemented with protease and phosphatase inhibitors and centrifuged at 10,000 × g for 10 min at 4◦C. The total protein concentrations were measured with the bicinchoninic acid (BCA) protein assay kit (Pierce, Thermo Fisher Scientific, USA). Equal amounts of protein (20 μg per sample) were separated on 10% sodium dodecyl sulfate–polyacrylamide gels (SDS-PAGE) and transferred onto 0.45 μM nitrocellulose membranes (Merck). Membranes were blocked with 5% BSA in Tris-buffered saline with 0.1% Tween- 20 (TBST) for 1 h at room temperature to prevent non-specific binding and incubated overnight at 4 °C with primary antibodies against phospho-STAT1 (Tyr701), total STAT1, PD-L1, and β-actin (Santa Cruz Biotechnology, Dallas, TX, USA). After washing with TBST, membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary anti-mouse IgG antibody or anti-rabbit IgG, HRP-linked antibody, for 1 h at room temperature. Protein bands were detected using enhanced chemiluminescence (ECL) reagent (GE Healthcare, USA) and visualized using a ChemiDoc imaging system (Bio-Rad, USA). Analysis of protein bands was performed using ImageJ software (NIH, Bethesda, MD, USA). The primary and secondary antibodies are listed in Table 1.Table 1The description of primary and secondary antibodiesAntibodySpeciesCloneDilutionCa#SourcePD-L1RabbitMonoclonal1:1000ab213524AbcamSTAT1MouseMonoclonal1:10009176Cell Signaling TechnologypSTAT1RabbitMonoclonal1:10009177SCell Signaling Technologyβ-actinMouseMonoclonal1:500047,778Santa Cruz Biotechnology, Dallas, TX, USAAnti-mouseHorse-1:10,0007076Cell Signaling, Danvers, MA, USAAnti-RabbitGoat-1:10,0007074Cell Signaling, Danvers, MA, USA The description of primary and secondary antibodies Total RNA was extracted from A549 cells using the RNeasy Mini-Kit (Qiagen, #74,104) according to the manufacturer’s instructions, and the eluted RNA purity and concentration were assessed using a NanoDrop One spectrophotometer (Thermo Scientific, USA). For cDNA synthesis, 500 ng of RNA was reverse transcribed using the RevertAid First Strand cDNA Synthesis Kit to cDNA as per the manufacturer’s instructions. Quantitative real-time PCR (qRT-PCR) was conducted using the SYBR Green PCR Master Mix (Applied Biosystems™) on a QuantStudio 5 Machine (Thermo Fisher Scientific, Inc.). The relative expression levels of PD-L1, STAT1, and ISGs (ISG15 and SOCS1) were normalized to the housekeeping gene GAPDH and analyzed using the 2^ − ΔΔCt method. The PCR reactions were carried out in duplicate with 40 cycles of denaturation (15 s at 95 °C), annealing (20 s at 65 °C), and elongation (20 s at 72 °C) after an initial enzyme activation (15 min at 95 °C). The primer sequences used are presented in Table 2.Table 2List of PCR primers designed using NCBI/Primer-BLAST programPrimerPrimer sequencesForwardReverseCD274 ()5′-TGGCATTTGCTGAACGCATTT- 3′5′-AGTGCAGCCAGGTCTAATTGT- 3′ISG155′-ATCACCCAGAAGATCGGCGT- 3′5′-TCGCATTTGTCCACCACCAG- 3′SOCS15’- TTCGCCCTTAGCGTGAAGATGG- 3′5’- TAGTGCTCCAGCAGCTCGAAGA- 3′GAPDH5′-GGAAGGTGAAGGTCGGAGTC- 3′5′-TGAAGGGGTCATTGATGGCA- 3′ List of PCR primers designed using NCBI/Primer-BLAST program A549 cells were seeded onto sterile coverslips in 12-well plates and treated as described. After incubation, cells were fixed with 4% paraformaldehyde for 15 min and permeabilized with 0.1% Triton X- 100 for 10 min. After blocking with 1% bovine serum albumin (BSA) for 30 min, cells were incubated overnight at 4 °C with primary antibodies against PD-L1 and phospho-STAT1. Following PBS washes, cells were incubated with Alexa Fluor 488- or 594-conjugated secondary antibodies for 1 h at room temperature. Nuclei were counterstained with DAPI, and images were captured using a fluorescence microscope (Zeiss Axio Observer, Germany). All experiments were performed in triplicate, and data are presented as mean ± standard deviation (SD). Statistical analyses were conducted using GraphPad Prism 10 (GraphPad Software, USA). One-way analysis of variance (ANOVA) followed by Tukey’s post-hoc test was used to compare multiple groups, and an unpaired Student’s t-test was used for pairwise comparisons. Differences were considered statistically significant at p < 0.05. STAT1 phosphorylation at tyrosine 701 (Tyr701) is a key regulatory event in IFN-γ-mediated signaling, leading to STAT1 dimerization, nuclear translocation, and transcriptional activation of ISGs, including PD-L1. To confirm the activation of the JAK-STAT1 pathway in A549 cells, we analyzed phospho-STAT1 (Tyr701) levels by Western blotting after stimulation with 500 U/ml of IFN-γ for different time points (0, 1, 2, 6, 12, and 24 h). Our results demonstrated a time-dependent increase in STAT1 phosphorylation. A significant induction observed as early as 30 min post-treatment (p < 0.01), peaking at 2 h (p < 0.001), and remaining elevated up to 24 h. Total STAT1 protein levels remained unchanged across all time points, indicating that the increase in phospho-STAT1 was due to phosphorylation rather than upregulation of STAT1 expression, as shown in Fig. 1.Fig. 1Western blot analysis showing STAT1 phosphorylation and PD-L1 expression in A549 NSCLC cells following IFN-γ treatment. A549 cells were treated with 500 U/mL of IFN-γ for various time points (0, 30 min, 1 h, 2 h, and 24 h). Protein lysates were collected and analyzed by Western blot using antibodies against phosphorylated STAT1 (Tyr701), total STAT1, and PD-L1. β-Actin was used as a loading control Western blot analysis showing STAT1 phosphorylation and PD-L1 expression in A549 NSCLC cells following IFN-γ treatment. A549 cells were treated with 500 U/mL of IFN-γ for various time points (0, 30 min, 1 h, 2 h, and 24 h). Protein lysates were collected and analyzed by Western blot using antibodies against phosphorylated STAT1 (Tyr701), total STAT1, and PD-L1. β-Actin was used as a loading control To further investigate the effect of IFN-γ on STAT1 activation, we performed immunofluorescence microscopy to visualize the cellular localization of phosphorylated STAT1 (p-STAT1) in A549 cells following a 2-h treatment with IFN-γ (500 U/ml). In untreated control cells, STAT1 was primarily localized in the cytoplasm, with minimal nuclear fluorescence detected. However, upon IFN-γ stimulation, a marked increase in nuclear accumulation of p-STAT1 was observed, as shown in Fig. 2, indicating its activation and translocation to the nucleus, where it functions as a transcription factor. This translocation pattern was confirmed through co-staining with DAPI, a nuclear marker, which showed strong co-localization of p-STAT1 within the nucleus.Fig. 2Immunofluorescence microscopy showing p-STAT1 localization in A549 cells treated with IFN-γ. A549 cells were treated with 500 U/mL of IFN-γ for 2 h, then analyzed by immunofluorescence microscopy to assess p-STAT1 localization. In untreated cells, p-STAT1 was primarily detected in the cytoplasm. Following IFN-γ treatment, increased nuclear localization of p-STAT1 was observed. DAPI was used for nuclear staining, and β-actin staining was used to visualize the cytoskeleton Immunofluorescence microscopy showing p-STAT1 localization in A549 cells treated with IFN-γ. A549 cells were treated with 500 U/mL of IFN-γ for 2 h, then analyzed by immunofluorescence microscopy to assess p-STAT1 localization. In untreated cells, p-STAT1 was primarily detected in the cytoplasm. Following IFN-γ treatment, increased nuclear localization of p-STAT1 was observed. DAPI was used for nuclear staining, and β-actin staining was used to visualize the cytoskeleton Studies reveal that IFN-γ dramatically increases the expression of PD-L1 on cancer cells, mainly by activating the STAT1 signaling pathway. To validate this association, we examined PD-L1 upregulation upon IFN-γ treatment in A549 cells. Western blot analysis revealed a time-dependent increase in PD-L1 protein levels, with minimal expression in untreated control cells. Upon IFN-γ stimulation, PD-L1 levels began to increase at 30 min and 1 h, corresponding with the early activation of STAT1 phosphorylation (p-STAT1, Tyr701). By 2 h, PD-L1 expression was noticeably upregulated, and by 24 h, it reached its maximum induction (p < 0.01). This trend correlated with sustained STAT1 phosphorylation, indicating that IFN-γ-driven STAT1 activation plays a key role in PD-L1 upregulation in A549 cells (Fig. 1). To further assess the effect of IFN-γ on PD-L1 regulation, we performed quantitative real-time PCR (qRT-PCR) to measure PD-L1 mRNA expression levels in A549 cells treated with IFN-γ (500 U/ml) for different time points. The results demonstrated a time-dependent increase in PD-L1 mRNA expression. At 6 h, PD-L1 mRNA levels were significantly upregulated by approximately 1.8-fold compared to untreated control cells (p < 0.05). By 12 h, PD-L1 expression increased further to 2.9-fold (p < 0.01), and by 24 h, the induction reached a peak with a 3.8-fold increase (p < 0.001). These findings confirm that IFN-γ robustly upregulates PD-L1 expression at the transcriptional level over time, supporting its role in immune evasion mechanisms (Fig. 3).Fig. 3qRT-PCR analysis showing time-dependent upregulation of PD-L1 mRNA expression in A549 cells following IFN-γ treatment. A549 cells were treated with 500 U/mL of IFN-γ for 6, 12, and 24 h. Total RNA was extracted and analyzed by qRT-PCR to measure PD-L1 mRNA expression levels. Data are presented as mean ± SD relative to untreated controls. Statistical significance is indicated (*p < 0.05, **p < 0.01, ***p < 0.001) qRT-PCR analysis showing time-dependent upregulation of PD-L1 mRNA expression in A549 cells following IFN-γ treatment. A549 cells were treated with 500 U/mL of IFN-γ for 6, 12, and 24 h. Total RNA was extracted and analyzed by qRT-PCR to measure PD-L1 mRNA expression levels. Data are presented as mean ± SD relative to untreated controls. Statistical significance is indicated (*p < 0.05, **p < 0.01, ***p < 0.001) To investigate whether curcumin inhibits IFN-γ-induced STAT1 phosphorylation, A549 cells were pretreated with curcumin (50 µM) for 2 h before stimulation with IFN-γ (500 U/ml) for 24 h. Western blot analysis revealed a significant reduction in phosphorylated STAT1 (p-STAT1, Tyr701) levels in curcumin-pretreated cells, with a 68% decrease compared to IFN-γ-treated cells alone (p < 0.001). Notably, total STAT1 protein levels remained unchanged, confirming that the decrease in p-STAT1 was due to inhibition of phosphorylation rather than reduced STAT1 expression. Additionally, PD-L1 expression, which is regulated by STAT1 activation, was also markedly reduced in curcumin-treated cells, with 50 µM curcumin reducing PD-L1 levels by 72% compared to IFN-γ-treated cells (p < 0.01). These findings indicate that curcumin effectively suppresses IFN-γ-induced STAT1 activation and its downstream signaling, including PD-L1 upregulation, in A549 cells (Fig. 4).Fig. 4Western blot analysis showing the effect of curcumin on IFN-γ-induced STAT1 phosphorylation and PD-L1 expression in A549 cells. A549 cells were pretreated with curcumin (50 µM) for 2 h, followed by stimulation with IFN-γ (500 U/mL). Protein lysates were collected and analyzed by Western blot for phosphorylated STAT1 (Tyr701), total STAT1, and PD-L1 expression. β-Actin was used as a loading control Western blot analysis showing the effect of curcumin on IFN-γ-induced STAT1 phosphorylation and PD-L1 expression in A549 cells. A549 cells were pretreated with curcumin (50 µM) for 2 h, followed by stimulation with IFN-γ (500 U/mL). Protein lysates were collected and analyzed by Western blot for phosphorylated STAT1 (Tyr701), total STAT1, and PD-L1 expression. β-Actin was used as a loading control Since STAT1 regulates the expression of multiple ISGs involved in immune evasion, we next evaluated the effect of curcumin on the expression of ISG15 and SOCS1. qRT-PCR analysis showed that IFN-γ significantly upregulated ISG15 and SOCS1 mRNA expression by 3.2-fold and 4.5-fold, respectively, compared to untreated controls (p < 0.001). However, pretreatment with curcumin (50 µM) led to a suppression of ISG15 (54% reduction) and SOCS1 (63% reduction) expression, further supporting the role of curcumin in inhibiting IFN-γ-driven STAT1 signaling (Fig. 5).Fig. 5qRT-PCR analysis showing the effect of curcumin on IFN-γ-induced expression of interferon-stimulated genes (ISGs) in A549 cells. A549 cells were treated with IFN-γ (500 U/mL) for 24 h, with or without curcumin pretreatment (50 µM for 2 h). Total RNA was extracted and analyzed by qRT-PCR for SOCS1 and ISG15 mRNA expression. Data are presented as mean ± SD. Statistical significance is indicated (*p < 0.05, **p < 0.01, ***p < 0.001) qRT-PCR analysis showing the effect of curcumin on IFN-γ-induced expression of interferon-stimulated genes (ISGs) in A549 cells. A549 cells were treated with IFN-γ (500 U/mL) for 24 h, with or without curcumin pretreatment (50 µM for 2 h). Total RNA was extracted and analyzed by qRT-PCR for SOCS1 and ISG15 mRNA expression. Data are presented as mean ± SD. Statistical significance is indicated (*p < 0.05, **p < 0.01, ***p < 0.001) To evaluate the time-dependent effects of curcumin on IFN-γ-induced growth suppression, cell viability was assessed using the Resazurin assay at 0, 12, 24, and 48 h following treatment. At 12 h, a modest reduction in cell viability (10%) was observed in response to IFN-γ and curcumin co-treatment. By 24 h, cell viability had decreased further (21% reduction), consistent with previous findings on IFN-γ-mediated cytotoxicity in A549 cells. Notably, at 48 h, the combination of curcumin and IFN-γ resulted in a 47% reduction in cell viability, demonstrating a significant time-dependent enhancement of IFN-γ’s anti-proliferative effects (p < 0.01). These findings suggest that curcumin sensitizes NSCLC cells to IFN-γ-induced growth suppression in a time-dependent manner, highlighting its potential as a therapeutic adjuvant (Fig. 6).Fig. 6Resazurin assay showing the effect of curcumin and IFN-γ co-treatment on cell viability in A549 NSCLC cells. A549 cells were treated with curcumin (50 µM) and IFN-γ (500 U/mL) for 12, 24, and 48 h. Cell viability was assessed using the Resazurin assay. Data are presented as mean ± SD, showing a time-dependent decrease in viability with combination treatment. Statistical significance is indicated (*p < 0.05, **p < 0.01, ***p < 0.001) Resazurin assay showing the effect of curcumin and IFN-γ co-treatment on cell viability in A549 NSCLC cells. A549 cells were treated with curcumin (50 µM) and IFN-γ (500 U/mL) for 12, 24, and 48 h. Cell viability was assessed using the Resazurin assay. Data are presented as mean ± SD, showing a time-dependent decrease in viability with combination treatment. Statistical significance is indicated (*p < 0.05, **p < 0.01, ***p < 0.001) Immune evasion remains a significant challenge in the treatment of NSCLC, with PD-L1 upregulation being one of the primary mechanisms by which tumors escape immune surveillance (Cui et al. 2024). The IFN-γ/STAT1 signaling pathway plays a crucial role in PD-L1 regulation, enabling tumor cells to suppress T-cell-mediated immune responses and resist immune checkpoint blockade therapy (Padmanabhan et al. 2022). In this study, we investigated how IFN-γ-induced STAT1 phosphorylation leads to PD-L1 upregulation in A549 cells and explored the potential of curcumin, a bioactive polyphenol with known anti-inflammatory and anti-cancer properties, as a therapeutic agent capable of modulating this pathway. Our findings provide strong evidence that curcumin inhibits IFN-γ-induced STAT1 activation, thereby reducing PD-L1 expression and enhancing the anti-proliferative effects of IFN-γ in NSCLC. Consistent with previous studies, we observed that IFN-γ induces a robust phosphorylation of STAT1 at Tyr701 in A549 cells, leading to its nuclear translocation and subsequent activation of target genes, including PD-L1. The kinetics of STAT1 phosphorylation followed a time-dependent pattern, with phosphorylation detected as early as 2 h, peaking at 6 h, and remaining elevated up to 24 h. These results are in agreement with earlier reports that demonstrated a similar pattern of IFN-γ-induced STAT1 activation in various cancer models, including melanoma (Schmitt et al. 2012), colorectal cancer (Zhao et al. 2020), and lung adenocarcinoma (Gao et al. 2018). STAT1 phosphorylation is a prerequisite for its dimerization and nuclear translocation, which is required for the transcriptional activation of ISGs (Wang et al. 2017). Immunofluorescence analysis confirmed that IFN-γ treatment led to a marked accumulation of p-STAT1 in the nucleus, reinforcing the notion that STAT1 plays a crucial role in IFN-γ-mediated transcriptional regulation. These findings align with studies showing that sustained STAT1 activation promotes an immunosuppressive tumor microenvironment by inducing PD-L1 expression and other immune-regulatory genes. The upregulation of PD-L1 in response to IFN-γ was confirmed at both the mRNA and protein levels, as demonstrated by qRT-PCR and Western blotting. The increased surface expression of PD-L1 following IFN-γ treatment highlights the functional significance of this regulation, as surface PD-L1 interacts with PD- 1 on T cells to inhibit anti-tumor immune responses (Arak et al. 2021). These results are in accordance with previous reports showing that IFN-γ is one of the most potent inducers of PD-L1 in NSCLC, facilitating immune escape and tumor progression (Pawelczyk et al. 2019). Additionally, the use of fludarabine, a STAT1 inhibitor, significantly attenuated IFN-γ-induced PD-L1 expression, confirming that STAT1 is the primary mediator of this regulatory axis. This finding corroborates prior studies demonstrating that STAT1-deficient cells fail to upregulate PD-L1 in response to IFN-γ, emphasizing the centrality of STAT1 in this pathway. One of the most significant findings of this study is the ability of curcumin to inhibit IFN-γ-induced STAT1 activation and PD-L1 expression in A549 cells. Western blot analysis revealed that curcumin suppressed STAT1 phosphorylation in a dose-dependent manner, with the highest concentration (50 µM) reducing phosphorylation by 68%. This effect was not due to a decrease in total STAT1 protein levels, indicating that curcumin selectively inhibits STAT1 activation rather than its expression. Previous studies have reported that curcumin can interfere with JAK-STAT signaling in other cancer types. Curcumin directly inhibits the phosphorylation of STAT3, a key component of the JAK-STAT signaling pathway in breast cancer (Golmohammadi et al. 2024) and prostate cancer (Li et al. 2024), as well as the downregulation of the STAT1 in melanoma (Xu et al. 2018), but its specific effect on STAT1 in IFN-γ-stimulated NSCLC cells had not been previously explored. Our findings extend these observations by demonstrating that curcumin effectively blocks STAT1 activation in lung cancer cells, preventing the downstream induction of PD-L1. The suppression of PD-L1 expression by curcumin was observed at both the transcriptional and translational levels, as evidenced by qRT-PCR and Western blotting. This finding is particularly relevant in the context of NSCLC, where high PD-L1 expression correlates with poor prognosis and resistance to immunotherapy. Previous studies have reported that curcumin downregulates PD-L1 in other cancer models, such as melanoma (Xu et al. 2018) and hepatocellular carcinoma (Guo et al. 2021), but the specific inhibition of IFN-γ-induced PD-L1 expression in NSCLC had not been thoroughly investigated. Our study provides the first evidence that curcumin can effectively suppress IFN-γ-mediated PD-L1 upregulation in lung cancer cells via STAT1 pathway, highlighting its potential as an immune-modulatory agent. In addition to PD-L1, STAT1 regulates the expression of multiple ISGs involved in immune evasion, including SOCS1 (Ilangumaran et al. 2024) and ISG15 (Desai 2015). Our results demonstrated that IFN-γ significantly upregulated both SOCS1 and ISG15, reinforcing the notion that IFN-γ signaling contributes to an immunosuppressive tumor microenvironment. Curcumin pretreatment, however, led to a significant reduction in both SOCS1 and ISG15 expression, further supporting its ability to interfere with IFN-γ-driven STAT1 signaling. SOCS1 is known to act as a feedback inhibitor of JAK-STAT signaling (Liau et al. 2018), but paradoxically, its overexpression in tumors has been associated with immune escape mechanisms. By suppressing SOCS1 expression, curcumin may enhance the responsiveness of tumor cells to immune-mediated clearance. Similarly, ISG15 has been implicated in tumor progression and resistance to therapy (Meng et al. 2024), suggesting that its downregulation by curcumin may have additional therapeutic benefits. Finally, we observed that curcumin enhances the anti-proliferative effect of IFN-γ in A549 cells. While IFN-γ alone resulted in a modest reduction in cell viability (21%), the combination of IFN-γ and curcumin led to a significantly greater reduction (47%), suggesting a synergistic effect. These findings align with previous reports that curcumin enhances the anti-tumor activity of cytokines by modulating cell cycle regulators and apoptotic pathways (Hu et al. 2018). The precise mechanism by which curcumin sensitizes NSCLC cells to IFN-γ-induced growth suppression remains to be elucidated, but it may involve inhibition of survival pathways downstream of STAT1 activation. Given that STAT1 has been implicated in both pro-apoptotic and pro-survival signaling, the net effect of its inhibition may depend on the cellular context and additional regulatory factors. In conclusion, our study provides novel evidence that IFN-γ induces PD-L1 expression in NSCLC cells via STAT1 activation and that curcumin effectively inhibits this process by suppressing STAT1 phosphorylation and nuclear translocation. Furthermore, curcumin downregulates IFN-γ-induced ISGs and enhances IFN-γ-mediated tumor cell growth suppression, highlighting its potential as a therapeutic adjuvant in NSCLC. These findings suggest that curcumin could be used to improve the efficacy of immune checkpoint inhibitors by reducing tumor immune evasion. Future studies should focus on elucidating the precise molecular mechanisms underlying curcumin’s effects on STAT1 signaling and investigating its potential synergistic effects with existing immunotherapies in preclinical and clinical settings.
PMC11419360
Release of P-TEFb from the Super Elongation Complex promotes HIV-1 latency reversal
The persistence of HIV-1 in long-lived latent reservoirs during suppressive antiretroviral therapy (ART) remains one of the principal barriers to a functional cure. Blocks to transcriptional elongation play a central role in maintaining the latent state, and several latency reversal strategies focus on the release of positive transcription elongation factor b (P-TEFb) from sequestration by negative regulatory complexes, such as the 7SK complex and BRD4. Another major cellular reservoir of P-TEFb is in Super Elongation Complexes (SECs), which play broad regulatory roles in host gene expression. Still, it is unknown if the release of P-TEFb from SECs is a viable latency reversal strategy. Here, we demonstrate that the SEC is not required for HIV-1 replication in primary CD4+ T cells and that a small molecular inhibitor of the P-TEFb/SEC interaction (termed KL-2) increases viral transcription. KL-2 acts synergistically with other latency reversing agents (LRAs) to reactivate viral transcription in several cell line models of latency in a manner that is, at least in part, dependent on the viral Tat protein. Finally, we demonstrate that KL-2 enhances viral reactivation in peripheral blood mononuclear cells (PBMCs) from people living with HIV (PLWH) on suppressive ART, most notably in combination with inhibitor of apoptosis protein antagonists (IAPi). Taken together, these results suggest that the release of P-TEFb from cellular SECs may be a novel route for HIV-1 latency reactivation.The persistence of replication-competent, but transcriptionally inhibited HIV-1 proviral DNA in long-lived, latent cellular reservoirs is a significant barrier to the development of a functional cure Even after long-term, suppressive antiretroviral therapy (ART), spontaneous reactivation of proviral gene expression from the latent reservoir is sufficient to initiate viral rebound shortly after ART cessation, thus requiring life-long adherence [3–5]. Multifaceted and heterogenous blocks to viral gene expression establish and maintain HIV-1 latency at the epigenetic, transcriptional, and post-transcriptional levels . Several strategies to either reactivate latent proviruses and clear infected cells (i.e., “shock and kill”) or reinforce latency to prevent spontaneous reactivation (i.e., “block and lock”) are currently under investigation . While many small molecule latency reversing agents (LRAs) have been described that reactivate latent proviruses ex vivo, they have had little success in clinical trials . This failure is partly due to the latent reservoir’s heterogenous nature, such that treatment by a single agent that acts to target a specific block may only ever reactivate a small fraction of proviruses in vivo . Even if transcriptional reactivation is achieved, it is unlikely this is sufficient to result in the clearance of these infected cells without additional immune augmentation such as the administration of antibodies, vaccines, or immunotherapies to prime the immune system . Given these limitations, much research is now focused on combinatorial approaches to trigger more widespread and robust reactivation . For example, a recent study reported synergistic reactivation potential between an activator of non-canonical NF-kB signaling (AZD5582) and BET bromodomain inhibitors that act to lift blocks to transcriptional initiation and elongation, respectively . Blocks to transcriptional elongation are major contributors to establishing and maintaining HIV-1 latency . After integration of the proviral DNA, RNA Polymerase II (RNA Pol II) is recruited to the transcription start site by transcription factors that bind cis-elements in the HIV-1 promoter region. After transcriptional initiation, RNA Pol II synthesizes 20–60 nucleotides before stalling through a well-conserved process known as promoter-proximal pausing . Pausing is enforced by several negative elongation factors, including negative elongation factor (NELF), DRB Sensitivity Inducing Factor (DSIF), and the RNA Polymerase II Associated Factor 1 (PAF1) complex [16–18]. Pause release is regulated by positive transcription elongation factor-b (P-TEFb), a heterodimeric protein complex composed of cyclin-dependent kinase 9 (CDK9) and cyclin T1 (CCNT1) [19–21]. P-TEFb phosphorylates the C-terminal tail of RNA Pol II and several negative elongation factors, which collectively license transcriptional elongation [22–24]. Recruitment of P-TEFb to sites of nascent transcription is a highly regulated process mediated by several cellular complexes. The majority of cellular P-TEFb is sequestered in an inactive state by the 7SK ribonucleoprotein (RNP) complex . Diverse extracellular stimuli and intracellular signals can induce the release of P-TEFb from the 7SK complex [27–29] where it can be recruited to sites of nascent transcription by transcription factors (i.e., NF-kB and c-MYC) [30–32], epigenetic regulators (i.e., BRD4) , or super elongation complexes (SECs) composed of an ARF4/FMR2 (AFF) family scaffold protein in complex with AF9, ENL, an eleven-nineteen Lys-rich leukemia (ELL) family protein, and an ELL-associated factor (EAF) protein . To circumvent this regulatory step, HIV-1 encodes a trans-activator protein (Tat) that binds to and recruits P-TEFb specifically to sites of nascent proviral transcription through recognition of a transactivation response (TAR) RNA stem loop produced at the immediate 3’ end of all viral RNA transcripts [36–38]. The distribution of and competition for P-TEFb binding among different complexes is an area of active investigation, with several strategies to enhance the biogenesis or availability of P-TEFb showing promise for HIV-1 latency reversal. For example, BET bromodomain inhibitors (such as JQ1) have been shown to be potent LRAs in ex vivo models by releasing P-TEFb from BRD4 . Likewise, the release of P-TEFb from the 7SK RNP complex has been shown to reactivate latent proviruses in ex vivo models . That being said, the release of P-TEFb from the 7SK RNP complex has been shown to directly correlate with increased BRD4 binding, suggesting that release from any one complex will not necessarily increase the amount of unbound P-TEFb or the amount recruited to specific sites of transcription . Furthermore, post-translational modification of P-TEFb (i.e., through phosphorylation of CDK9 at Serine 175) has been shown to influence P-TEFb distribution in certain regulatory complexes, again highlighting the unique properties of release from each complex . While the release of P-TEFb from the 7SK RNP complex and BET proteins such as BRD4 have been explored as strategies for HIV-1 latency reversal, the release of P-TEFb from SECs has not been explored. Previous work has demonstrated that HIV-1 Tat biochemically co-purifies with several SEC proteins , though the reason for this is unclear as they seemingly have functionally redundant purposes in P-TEFb recruitment. In this study, we test the hypothesis that the SEC is not necessary for HIV-1 viral transcription and that the release of P-TEFb from cellular SEC complexes can serve as a novel strategy to reactivate latent HIV-1 proviruses. While biochemical purification of HIV-1 Tat from cell lines has been shown to pull down other SEC members besides P-TEFb (including AFF1, AFF4, ELL2, ENL and AF9), the role of the SEC in HIV-1 replication in primary CD4+ T cells is unclear. To determine whether the SEC is required for HIV-1 replication in primary CD4+ T cells, we used multiplexed CRISPR-Cas9 RNPs to knock-out expression of each SEC component in cells from multiple healthy donors (Fig 1A) [43–45]. Each multiplexed reaction consisted of 4 or 5 independent guide RNA targeting the same gene . A non-targeting (NT) guide RNA was used as a negative control whereas a previously validated guide RNA targeting the HIV-1 co-receptor gene, CXCR4, was used as a positive control . Immunoblots of protein lysates collected 72 hours after CRISPR-Cas9 RNP electroporation demonstrated depletion of each target (Fig 1B). Visualization of AFF1 and AFF4 depletion required CCNT1 immunoprecipitation, likely due to their low steady-state levels in CD4+ T cells and low-affinity antibodies (Fig 1C). Notably, the depletion of some targets had secondary effects on other complex members. For example, knock-out of CCNT1 decreased CDK9 steady state levels and knock-out of AFF1, AFF4, and AF9 each decreased ENL steady-state levels. None of the knockouts were found to decrease cell viability significantly at this time point (S1A and S1B Fig). A) Experimental schematic of multiplex CRISPR-Cas9 gene editing of primary CD4+ T cells. B) Immunoblots to assess CCNT1, CDK9, ELL2, AF9, and ENL knock-out efficiency in primary CD4+ T cells relative to a non-targeting (NT) control (1 representative donor). C) Immunoblots to assess AFF1 and AFF4 knock-out efficiency in primary CD4+ T cells relative to a NT control following immunoprecipitation of CCNT1 (1 representative donor). D) Percent infected (GFP+) primary CD4+ T cells (normalized to the donor-matched NT control) 48 hours after challenge with HIV-1 NL4.3 Nef:IRES:GFP. Each dot represents the average of technical triplicates; the black line represents the mean of means ± standard error of means. n = 12 donors for NT, CXCR4, CCNT1, CDK9, AFF1, AFF4, and ELL2; n = 6 donors for AF9; and n = 9 donors for ENL. Statistics were calculated by two-way ANOVA with Dunnet’s Multiple Comparison Test; significant p-values (p < 0.05) are shown. To determine the impact of SEC component knock-out on HIV-1 replication, we challenged each cell population with replication-competent HIV-1 NL4.3 containing an IRES-driven GFP reporter inserted after Nef (HIV-1 NL4.3 Nef:IRES:GFP) in technical triplicate. The percentage of infected cells was quantified at two-days post-challenge by flow cytometry (S1C Fig) and normalized to the NT control (Fig 1D). Knock-out of CXCR4 ablated CXCR4-tropic NL4.3 strain infection as expected. Knock-out of each P-TEFb component (CCNT1, CDK9) significantly decreased infection. However, knock-out of the other SEC components either did not alter HIV-1 infection (AFF4, ELL2, AF9) or resulted in a very slight, but statistically significant increase in infection (AFF1, ENL). These data suggest that, while P-TEFb is required, the rest of the SEC is dispensable for HIV-1 replication in primary CD4+ T cells. Seeing that knock-out of several SEC components had no effect on HIV-1 replication in primary CD4+ T cells, we next wanted to determine the impact of chemical perturbation of this complex using the previously described SEC inhibitor, KL-2 . KL-2 inhibits SEC function by disrupting the interaction between CCNT1 and AFF1/AFF4 without altering overall protein levels of P-TEFb (Fig 2A) . Given that the SEC is dispensable for HIV-1 replication, we hypothesized that KL-2 treatment may increase the availability of P-TEFb for recruitment to sites of viral transcription. Activated CD4+ T cells from three independent donors were treated over a range of KL-2 concentrations for 24 hours and then challenged with HIV-1 NL4.3 Nef:IRES:GFP for 48 hours in technical triplicate. Higher concentrations of KL-2 dramatically decreased cell viability, with 3.125 μM being the highest tolerated dose with minimal toxicity (Fig 2B). This dose was sufficient to inhibit the CCNT1:AFF4 interaction in primary CD4+ T cells within 24 hours as determined by CCNT1 immunoprecipitation (Fig 2C). Compared to DMSO-treated control cells, KL-2 treatment resulted in a dose-dependent increase in infection with significant increases observed at 1.56 μM and 3.125 μM (Fig 2D). A) Model of Tat-mediated recruitment of P-TEFb to paused RNA Pol II at sites of nascent HIV-1 proviral transcription. B) Percent viable primary human CD4+ T cells (normalized to the donor-matched DMSO control) after 72 hours of treatment with increasing concentrations of KL-2 as measured by amine dye staining and flow cytometry. Data represent the average ± standard deviation of technical triplicates; the green arrow indicates the concentration chosen for downstream experiments in primary CD4+ T cells. C) Immunoblots for AFF4 and CCNT1 following CCNT1 immunoprecipitation from primary CD4+ T cells treated for 24-hours with either DMSO or 3.125 μM KL-2 (1 representative donor). Quantification of immunoprecipitated AFF4 levels normalized to immunoprecipitated CCNT1 is shown on the right. D) Percent infected (GFP+) primary CD4+ T cells (normalized to the donor-matched DMSO control) 48 hours after challenge with HIV-1 NL4.3 Nef:IRES:GFP in the presence of increasing concentrations of KL-2 (24 hours pre-treatment before challenge). Data represent the average ± standard deviation of technical triplicates (n = 3 donors); statistics were calculated relative to the DMSO control by two-way ANOVA and Sidak’s Multiple Comparison test with significant p-values (p < 0.05) shown. E) Percent of activated (CD25+) primary CD4+ T cells following treatment with DMSO, 3.125 μM KL-2, or 1μg/mL PHA for 48 hours as measured by immunostaining and flow cytometry. Data represent the average ± standard deviation of technical triplicates (n = 3 donors); statistics were calculated by two-way ANOVA and Sidak’s Multiple Comparison test with p-values shown. F) Mean fluorescence intensity (MFI) of CD4 and CXCR4 on primary CD4+ T cells following treatment with DMSO or 3.125 μM KL-2 for 48 hours as measured by immunostaining and flow cytometry. Data represent the average ± standard deviation of technical triplicates (n = 1 donor); statistics were calculated by Student’s t-test with p-values shown. G) Relative transcript levels of HIV-1 TAR and H) long LTR to human β-Actin in activated primary CD4+ T cells after 48-hours of challenge with HIV-1 in the presence or absence of a 24-hour pretreatment with KL-2. Data represent the mean of means of 3 biological replicates in technical duplicate ± standard error; statistics were calculated by two-way ANOVA with Sidak’s Multiple Comparison test. Given that receptor and co-receptor expression can alter HIV-1 susceptibility, we next tested the impact of KL-2 treatment on CD4 and CXCR4 cell surface expression. Activated, primary CD4+ T cells from one representative donor were treated with 3.125 μM KL-2 or DMSO for 48 hours prior to immunostaining and flow cytometry (S2A and S2B Fig). KL-2 treatment resulted in a slight, but significant increase in CXCR4 expression (mean fluorescence intensity) and a slight, but significant decrease in CD4 expression (Fig 2E). To determine if KL-2 could alternately impact CD4+ T cell activation, we treated unstimulated CD4+ T cells from three independent donors with DMSO, 3.125 μM KL-2, or the T cell mitogen Phytohemagglutinin (PHA) for 48 hours. Unlike PHA, which resulted in robust activation, KL-2 treatment did not impact T cell activation as measured by CD25 cell surface staining (Fig 2F). These results suggest that increased infection in the presence of KL-2 is likely not driven by changes in receptor expression. To assess whether this increase in infection was due to enhanced viral transcription, activated CD4+ T cells from three healthy donors were treated with 3.125 μM KL-2 or DMSO for 24 hours and then challenged with HIV-1 NL4.3 Nef:IRES:GFP in technical triplicate. After 48 hours, RNA from the infected cultures was isolated and the expression of viral transcripts was measured using quantitative reverse transcription PCR (qRT-PCR). Quantification of TAR and long LTR transcripts were used to measure viral transcriptional initiation and elongation, respectively, relative to the human housekeeping gene, β-Actin. We found that KL-2 treatment significantly increased the expression of both TAR and long LTR transcripts (Fig 2G and 2H). To determine if KL-2 has similar effects on HIV-1 replication across different cell types, CHME microglial cells engineered to express CXCR4 (CHME-4X4 cells), were pretreated with KL-2 over a range of concentrations for 24 hours then challenged with replication-competent HIV-1 NL4.3 containing a nano-luciferase reporter in place of Nef (Nef:Nano-Luc). As in the primary CD4+ T cells, we observed a dose-dependent increase in infection, though higher concentrations again resulted in viability defects (S2C and S2D Fig). Taken together, these data support the genetic data above that the interaction between P-TEFb and the larger SEC is not required for HIV-1 replication in primary CD4+ T cells and that SEC disruption can enhance viral transcription. The release of P-TEFb from sequestration by BRD4 or the 7SK complex has proven effective in latency reactivation . Given that the SEC is not required for HIV-1 replication, we hypothesized that releasing P-TEFb from cellular SECs may also reactivate latent proviruses. To test this hypothesis, we treated J-Lat 5a8 cells (a Jurkat subclone harboring a silent, integrated non-replicative full-length provirus and a GFP reporter gene ) over a range of KL-2 concentrations for 48 hours. As in primary CD4+ T cells, higher concentrations of KL-2 decreased J-Lat 5A8 cell viability (Fig 3A). However, treatment with KL-2 alone did not significantly increase proviral reactivation (GFP+ cells as measured by flow cytometry) until 6.25 μM, at which dose the cells are only about 50% viable (Fig 3A and 3B). This dose was sufficient to inhibit the CCNT1:AFF4 interaction in J-Lat 5A8 cells within 24 hours, as determined by CCNT1 immunoprecipitation (Fig 3C). A) Percent viable J-Lat 5A8 cells (normalized to the DMSO control) after 48 hours of treatment with increasing concentrations of KL-2 as measured by amine dye staining and flow cytometry. Data represent the average ± standard deviation of technical triplicates; the green arrow indicates the concentration chosen for downstream experiments in cell lines. B) Percent reactivated (GFP+) J-Lat 5A8 cells (normalized to the DMSO control) after 48 hours of treatment with increasing concentrations of KL-2. Data represent the average ± standard deviation of technical triplicates; statistics were calculated relative to the DMSO control by two-way ANOVA and Sidak’s Multiple Comparison test. C) Immunoblots for AFF4 and CCNT1 following CCNT1 immunoprecipitation from J-Lat 5A8 cells treated for 24 hours with either DMSO or 6.25 μM KL-2. Quantification of immunoprecipitated AFF4 levels normalized to immunoprecipitated CCNT1 is shown on the right. D) Percent reactivated (GFP+) J-Lat 5A8 cells, E) J-Lat 6.3 cells, or F) J-Lat 11.1 cells (normalized to the DMSO control) after 48 hours of treatment with the indicated compounds in the presence or absence of KL-2. Data represent the average ± standard deviation of technical triplicates; statistics were calculated by two-way ANOVA with Sidak’s Multiple Comparison test. Relative transcript levels of HIV-1 TAR (G) and long LTR (H) to human β-Actin in J-Lat 5A8 cells after 48 hours of treatment with the indicated compounds in the presence or absence of KL-2 (normalized to the DMSO control). Data represent the mean of means of 3 biological replicates in technical duplicate ± standard error; statistics were calculated by two-way ANOVA with Sidak’s Multiple Comparison test. While KL-2 was not sufficient to reactivate latent proviruses in J-Lat 5A8 cells, we hypothesized that it could enhance the activity of other latency reversing agents (LRAs) that act through different mechanisms, similar to the PAF1 complex inhibitors we reported previously . To test this hypothesis, we treated J-Lat 5a8 cells with several well-characterized LRAs in the presence or absence of KL-2, including JQ1 (a BRD4 inhibitor that enhances P-TEFb availability and relieves chromosomal repression), Phorbol 12-myristate 13-acetate (PMA, a protein kinase C activator that activates canonical NF-kB transcription), and AZD5582 (an IAP antagonist that activates non-canonical NF-kB transcription). While KL-2 alone did not significantly increase reactivation compared to the DMSO control, it significantly enhanced reactivation in the presence of the other three LRAs (Fig 3D). A similar pattern was observed in two other J-Lat clones with different proviral integration sites: J-Lat 6.3 cells (Fig 3E) and J-Lat 11.1 cells (Fig 3F). Paired cell viability data is provided in S3A–S3C Fig. To assess whether the effect of KL-2 in combination with other LRAs was synergistic, we calculated excess over Bliss scores for KL-2 in combination with JQ1 and AZD5582. Excess over Bliss calculations measure whether the observed combinatorial effects at given concentrations are above or below predicted additivity, where a score of 0 is additive, scores above 0 are considered synergistic, and scores less than 0 are considered antagonistic. At the fixed 6.25 μM concentration of KL-2 used in the previous assays in combination with ascending doses of AZD5582, we observed positive excess over Bliss scores ranging from 0.46 to 0.94, suggestive of synergistic effects (S3D Fig; top). Notably, at a fixed concentration of AZD5582 (10 nM), we observe strictly additive effects of KL-2 until the effective concentration of 6.25 μM is reached. Similar trends are observed with JQ1. At the fixed 6.25 μM concentration of KL-2 with ascending doses of JQ1, we again saw positive excess over Bliss scores ranging from 0.54 to 0.94 (S3D Fig; bottom). Likewise, at a fixed concentration of JQ1 (1 μM), we observe strictly additive effects of KL-2 until the effective concentration of 6.25 μM is reached. This data would indicate that at concentrations of KL-2 lower than 6.25 μM we are likely not seeing optimal disruption of the SEC, but upon disruption at the effective dose of KL-2, the release of P-TEFb acts synergistically with the LRAs tested. Given the role of KL-2 in P-TEFb release from the SEC, we hypothesized that treatment with KL-2 would enhance transcriptional elongation. To test this, we repeated our combinatorial LRA treatments with and without KL-2 in J-Lat 5A8 cells and extracted RNA at 48 hours post-treatment to quantify viral transcripts using qRT-PCR. HIV-1 long LTR transcripts, made after the start of transcriptional elongation, were not statistically increased upon KL-2 treatment alone, but were enhanced by KL-2 addition to every other tested LRA (Fig 3H). HIV-1 TAR transcripts, made immediately after transcriptional initiation, were not statistically increased upon KL-2 addition either alone or in the presence of PMA, however, KL-2 treatment did significantly increase transcriptional initiation in the presence of JQ1 and AZD5582 (Fig 3G). Taken together, these results suggest that KL-2 can enhance the latency reactivation activity of other LRAs in J-Lat models by increasing the availability of P-TEFb and promoting transcriptional elongation. However, additional impacts on transcriptional initiation cannot be ruled out as also observed in our primary CD4+ T cell data (Fig 2G). Inhibition of P-TEFb binding to SECs by KL-2 enhanced the HIV-1 latency reactivation potential of several distinct classes of LRAs, but is likely to also cause broad transcriptional changes at SEC-regulated genes. To better understand the effect of KL-2 on the transcriptome, we performed bulk mRNA sequencing (RNA-Seq) on J-Lat 5a8 cells treated with DMSO, AZD5582, KL-2, or AZD5582 plus KL-2 for 48 hours. The transcriptome of cells treated with AZD5582 alone closely resembled that of the DMSO-treated cells (Fig 4A) with only 50 differentially expressed genes (DEGs) (40 downregulated and 10 upregulated; Fig 4B). Treatment with KL-2 resulted in more broad transcriptional changes with more downregulated than upregulated DEGs (764 versus 546, respectively). The cells treated with AZD5582 plus KL-2 exhibited the most transcriptional changes (765 downregulated and 6012 upregulated DEGs), a majority of which were shared with the KL-2 alone. However, the combination of AZD5582 and KL-2 also resulted in the unique upregulation (n = 275) and downregulation (n = 159) of a subset of genes, similar to what was observed with the integrated provirus. Functional enrichment analysis of gene ontology (GO) terms revealed that KL-2 suppressed genes were largely involved in biosynthetic processes corresponding to Myc-regulated pathways as previously reported (Fig 4C) . KL-2 in combination with AZD5582 suppressed a similar subset of genes, but also increased expression of several genes involved in the innate immune response to infection, though it remains unclear if this is a direct result of the drug treatment or an indirect effect of proviral reactivation. A) Heatmap of transcript levels after variance stabilization transformation for each differentially expressed gene following treatment of J-Lat 5A8 cells with DMSO, KL-2, AZD5582, or a combination of AZD5582 and KL-2 for 48 hours (3 biological replicates per treatment). Values are scaled by gene. B) Venn diagrams illustrating the number of significantly downregulated (top) and upregulated (bottom) genes that are unique or shared between treatments as compared to the DMSO control. C) Gene Ontology (GO) analysis of significantly upregulated (left) and downregulated (right) genes in each treatment condition relative to the DMSO control. The size of the circle indicates the proportion of differentially expressed genes found in each functional category and the color of the circle indicates adjusted p-value. D) Sequencing read depth normalized to an equally bounded internal control over the integrated HIV-1 provirus in J-Lat 5A8 cells following CCNT1 or E) RNA Pol II ChIP-Seq. Viral open reading frames are indicated below with the location of the integrated GFP reporter indicated in green. Reads from two independent biological replicates are overlaid. The downregulation of several Myc-regulated pathways is consistent with the reported function of KL-2 in disrupting SEC-mediated gene transcription. We hypothesize that the release of P-TEFb from these complexes would allow for its recruitment to sites of proviral transcription where it would facilitate RNA Pol II pause release into the gene body. To test this directly, we performed CCNT1 and RNA Pol II Chromatin Immunoprecipitation Sequencing (ChIP-seq) in J-Lat 5a8 cells treated with either DMSO, KL-2, AZD5582, or KL-2 plus AZD5582 and analyzed occupancy along the provirus relative to an equally bounded internal control (Fig 4D and 4E). The CCNT1 CHIP-Seq data revealed very little P-TEFb along the proviral genome in the DMSO control cells with only slight increases in P-TEFb occupancy at the promotor and in the gene body upon KL-2 or AZD5582 treatment (Fig 4D). Similarly, the RNA Pol II CHIP-Seq data revealed very little RNA Pol II along the proviral genome in the DMSO control and AZD5582 treated cells (Fig 4E). In contrast, treatment with KL-2 alone resulted in a more substantial increase in RNA Pol II at the promoter as well as a slight increase throughout the proviral gene body. Only the combination of KL-2 and AZD5582, however, resulted in a striking increase in both P-TEFb and RNA Pol II throughout the proviral gene body (Fig 4D and 4E). Taken together, these data are consistent with a model in which SEC disruption promotes P-TEFb recruitment to sites of nascent proviral transcription to promote RNA Pol II pause release. However, the impact of KL-2 on transcriptional initiation and RNA Pol II recruitment to the proviral promoter are suggestive of a potential secondary mechanism that may be related to its broader impacts on the cellular transcriptome. The SEC is dispensable for HIV-1 replication in primary CD4+ T cells, and release of P-TEFb from the SEC promotes latency reversal in J-Lat cells, suggesting that P-TEFb is recruited to paused RNA Pol II at proviral integration sites in an SEC-independent manner in these models. This is most likely through the viral accessory protein Tat, which can recruit P-TEFb directly to sites of nascent viral transcription through an interaction with the TAR stem-loop on the 5’ end of viral transcripts (though alternate mechanisms for P-TEFb recruitment have been described). To assess the Tat-dependency of latency reactivation, we first turned to two cell line models of latency that lack a functional Tat-TAR axis: U1 cells and ACH-2 cells. The U1 cell line is a U937-based monocytic cell line that harbors two copies of the HIV-1 provirus, one of which expresses no Tat due to the lack of a start codon and the second of which encodes a Tat mutant with suboptimal P-TEFb affinity . The ACH-2 cell line is a T cell line with one integrated provirus that encodes a fully functional Tat, but that lacks a functional TAR stem loop . Both cell lines were treated with the same LRA panels as before in the presence and absence of KL-2 for 48 hours. These cell lines lack a fluorescent reporter; therefore, reactivation was monitored by intracellular p24 immunostaining and flow cytometry (viability data in S4A and S4B Fig). JQ1 and PMA treatment resulted in strong reactivation in both cell line models while AZD5582 had minimal effects (Fig 5A and 5B). Treatment with KL-2 alone likewise induced negligible reactivation. In contrast to what was observed in J-Lat lines, however, combinatorial treatment with KL-2 significantly decreased the reactivation potential of JQ1 and PMA in both cell lines (Fig 5A and 5B). A) Percent reactivated (KC57-FITC+) U1 cells after 48 hours of treatment with indicated compounds in the presence or absence of KL-2. B) Percent reactivated (KC57-FITC+) ACH-2 cells after 48 hours of treatment with indicated compounds in the presence or absence of KL-2. C) Percent reactivated (GFP+) J-Lat 5a8 cells after 48 hours of treatment with indicated compounds in the presence or absence of KL-2 and the Tat inhibitors Spironolactone and Triptolide. D) Percent reactivated (KC57-FITC+) Lenti-Tat U1 cells after 48 hours of treatment with indicated compounds in the presence or absence of KL-2 and with Tat induction at three different concentrations of doxycycline (0, 4, and 8 ng/mL). For all panels, the data represent the average ± standard deviation of technical triplicates; statistics were calculated by two-way ANOVA with Sidak’s Multiple Comparison test. These data suggest that while the SEC may be dispensable for Tat-dependent transcription and reactivation, it may play an essential role in P-TEFb recruitment without functional Tat. If so, the delivery of exogenous Tat to U1 cells might be sufficient to circumvent the need for the SEC, reversing the KL-2 treatment phenotype. To test this hypothesis, we transduced U1 cells with a doxycycline (Dox)-inducible lentiviral construct encoding full-length HIV-1 NL4.3 Tat (referred to as Lenti-Tat), selecting for a pure polyclonal population of transduced cells in puromycin. The U1 + Lenti-Tat cells were treated with a range of Dox concentrations, resulting in a dose-dependent increase in proviral reactivation (S4C Fig). U1 + Lenti-Tat cells were then treated with the same panel of LRAs in the presence or absence of KL-2 in the presence of either 0, 4, or 8 ng/mL Dox (Fig 5C; viability data in S4D Fig). Without Tat induction, KL-2 suppressed JQ1- and PMA-mediated reactivation as before (Fig 5A). Tat induction by Dox was sufficient to induce a basal level of reactivation, which was almost entirely reversed by KL-2 treatment (Fig 5C). JQ1 and PMA still induced reactivation above the Tat-induced baseline, but retained sensitivity to KL-2, while AZD5582 failed to reactivate above baseline even in the presence of Tat. These data suggest that the presence of Tat alone is not sufficient to dictate the impact of KL-2 on latency reversal, which is likely additionally dependent on other factors including P-TEFb availability and distribution. Indeed, immunoblotting for P-TEFb (CDK9 and CCNT1) and SEC (AFF4) components revealed that U1 cells have lower steady-state levels of P-TEFb and AFF4 compared to J-Lat cells at baseline, which might alter the efficacy and secondary effects of an SEC disrupter (S4E Fig). Given these considerations, we next asked if Tat was necessary for the enhanced reactivation phenotype of KL-2 in the J-Lat 5A8 cell line model using two previously reported Tat-dependent transcription inhibitors: Spironolactone and Triptolide (Fig 5D; viability data in S4F Fig). Spironolactone induces the degradation of the XBP helicase, a component of the TFIIH initiation complex, which inhibits Tat-dependent transcription , while Triptolide promotes proteasomal degradation of the Tat protein itself . In the presence of the DMSO control, KL-2 alone had little effect, but enhanced the latency reactivation activity of JQ1, PMA, and AZD5582 as observed previously (Fig 3D). Both Tat inhibitors dramatically reduced the reactivation potency of each LRA, in some cases to near baseline levels. Treatment with KL-2, however, still enhanced reactivation in combination with JQ1 and PMA, though not in combination with AZD5582, suggesting that this LRA may be uniquely dependent on Tat and/or the SEC for P-TEFb recruitment (Fig 5D). Spironolactone and Triptolide do not directly target Tat and are known to have several additional impacts on the cellular transcriptional machinery, including degradation of RNA Polymerase II and inhibition of NF-kB signaling [56–59]. While didehydro-cortistatin A has been reported to be a more specific inhibitor of Tat, we were unable to source or synthesize the compound for testing . Therefore, as an alternate approach, we next tested the impact of KL-2 on latency reversal in a pair of J-Lat cell lines that have an integrated HIV-1 LTR-driven GFP reporter that either do (A2 cells) or do not (A72 cells) express Tat . Each cell line was treated with the same panel of LRAs as before in the presence and absence of KL-2 for 48 hours. In the Tat-expressing J-Lat A2 cells, KL-2 alone had no effect while each LRA resulted in robust reactivation (S4G Fig). As expected, KL-2 enhanced JQ1 and PMA activity, but surprisingly decreased AZD5582 activity slightly. In contrast, J-Lat A72 cells had a high level of basal GFP expression, responded only mildly to any of the LRAs, and exhibited a slight increase in reactivation in the presence of KL-2 in all conditions (S4H Fig). Taken together, these data suggest that while SEC disruption may enhance viral transcription and latency reactivation in most models, it may also promote latency in certain cellular contexts that is not fully explained by the presence or absence of Tat expression. Blocks to transcriptional elongation have been shown to contribute to HIV-1 latency maintenance in cells from virally suppressed people living with HIV (PLWH) . The release of P-TEFb sequestration from BRD4 using bromodomain inhibitors (such as JQ1) has proven effective at reactivating viral gene expression in peripheral blood mononuclear cells (PBMCs) from these patients . We hypothesized that KL-2 would likewise reactivate viral gene expression in patient PBMCs. To test this, cryopreserved PBMCs were obtained from five PLWH enrolled in the Northwestern University Clinical Research Site for the MACS/WIHS Combined Cohort Study (MWCCS) (Fig 6A). The selected PLWH had been virally suppressed with ART for more than five years with undetectable HIV-1 plasma levels at the time of blood draw (<50 copies/ml) (Fig 6B). Cells from these patients were treated for 48 hours with DMSO, JQ1, or AZD5582 alone or in combination with KL-2. RNA was isolated from the treated cells and reactivation of viral gene expression was measured by qRT-PCR for HIV-1 gag (relative to the human housekeeping gene LDHA, Fig 6C). A) Experimental schematic of the latency reversal assay using PBMCs from five HIV-1 patients with undetectable viral loads. Intracellular gag transcript levels were measured by qRT-PCR after treatment with JQ1, PMA, or AZD5582 in the presence or absence of KL-2. B) Overlaid line graphs depicting the CD4 count (cells/mL, blue) and levels of viral RNA (copies/mL, red) in each subject over time. The dotted line shows the time of antiretroviral therapy initiation, and the time of the blood draw used for this assay is depicted by an arrow. C) Relative transcript levels of HIV-1 gag relative to human LDHA in patient PBMCs (n = 5 donors) after 48 hours of treatment with the indicated compounds in the presence or absence of KL-2. Data represent the average 2^-ΔΔCt [(Ctgag-CtLDHA)LRAs— (Ctgag-CtLDHA)DMSO] ± SEM of technical triplicates. Statistics were calculated using a two-way ANOVA with Sidak’s Multiple Comparison Test. The fold change in mean 2^-ΔΔCt is shown above for relevant comparisons in green text with corresponding p-values shown below in black text. D) Model of cellular P-TEFb reservoirs whose disruption have demonstrated activity in HIV-1 latency reversal. Similar to what was observed in our cell line models, treatment with KL-2 alone resulted in minimal latency reactivation with only a slight increase in gag expression in three out of the five donors, with the mean transcript levels increasing 1.5-fold (not statistically significant, Fig 6C). Treatment with JQ1 alone was not sufficient to induce gag expression in these donors, though the addition of JQ1 and KL-2 together resulted in a 2.88-fold increase over JQ1 treatment alone (not significant, Fig 6C). AZD5582 treatment alone resulted in a roughly 3.36-fold increase in gag expression over the DMSO control. However, KL-2 significantly enhanced the ability of AZD5582 to reactivate HIV-1 transcription across all donors with a marked 53-fold increase over the DMSO only control and a 16-fold increase compared to AZD5582 alone (Fig 6C). Taken together, these data demonstrate that the release of P-TEFb from cellular SECs using the small molecule KL-2 can enhance the effects of a range of LRAs in cell line models of HIV-1 latency as well as improve transcriptional reactivation by AZD5582 in primary PBMCs from HIV-1 patients. In this study, we demonstrate a potential new strategy for enhancing HIV-1 latency reversal through the release of P-TEFb from the cellular pool of SECs. We show that KL-2, a small molecule inhibitor of the interaction between the SEC and P-TEFb, is sufficient to enhance viral transcription in primary CD4+ T cells and can synergistically enhance the activity of other LRAs in certain cell line models of latency. Finally, we demonstrate that KL-2 can increase HIV-1 gag expression in PBMCs from PLWH on suppressive ART, most notably in combination with the non-canonical NF-kB agonist, AZD5582. We propose a model in which KL-2 release of P-TEFb from the cellular pool of SECs enhances transcriptional elongation of integrated proviruses, akin to BET bromodomain inhibitors and 7SK RNP inhibitors (Fig 6D). These results have several implications for our understanding of viral transcription and future directions. Latency reversal through the release of P-TEFb from cellular SECs had not been previously explored, likely due to the perceived dependency of viral transcription on the SEC. This effect has been supported by biochemical purifications of HIV-1 Tat from human cell lines that revealed interactions with a larger SEC , as well as by genetic knock-down experiments in cell lines showing a decrease in Tat-dependent transcription upon SEC component depletion . However, given that both the SEC and Tat recruit P-TEFb to sites of nascent transcription, they share some functional redundancy. We found that knock-out of SEC components from activated, primary CD4+ T cells from 12 independent donors did not inhibit viral replication, suggesting that the SEC is not required in this cellular context. This result was independently verified using KL-2, which inhibits the interaction between CCNT1 (P-TEFb) and AFF1/4 (of the larger SEC). Notably, disruption of the SEC using KL-2 resulted in significant increases in HIV-1 replication in primary CD4+ T cells whereas genetic knockout of most SEC members had minimal to no impact on replication. One potential explanation for this difference is that genetic knockout results in the ablation of SEC assembly and relocalization of P-TEFb into other complexes at steady-state whereas chemical perturbation by KL-2 results in the release of P-TEFb from SECs that are continually being formed. Understanding the dynamics of P-TEFb distribution and relocalization upon different types of perturbation is an ongoing area of investigation. While we expected KL-2 to enhance the transcriptional elongation of integrated proviruses due to the release of P-TEFb from cellular SECs, we also saw increases in transcriptional initiation as measured by qRT-PCR for TAR transcript levels. Likewise, in J-Lat 5A8 cells, we saw increases in transcriptional elongation, transcriptional initiation, and in RNA Pol II recruitment to the proviral promoter when KL-2 was added, even though KL-2 addition alone was not sufficient for reactivation as measured by GFP positivity in this model. This finding suggests that either KL-2 has secondary effects not mediated by P-TEFb or that the redistribution of P-TEFb away from SECs has secondary effects that could impact transcriptional initiation at proviral integration sites. BET bromodomain inhibitors have also been reported to increase HIV-1 transcriptional initiation , though these effects have been suggested to occur through modulation of the epigenetic regulatory functions of these proteins . Other reports have indicated that P-TEFb mediated release of paused RNA Pol II can result in enhanced transcriptional initiation and even RNA Pol II recruitment simply by increasing the number of transcribing polymerases . Still, this connection between transcriptional elongation and initiation has yet to be fully understood in the context of HIV-1 transcription. While KL-2 was sufficient to boost viral replication in activated, primary CD4+ T cells, in and of itself it displayed minimal reactivation potential in both cell line models of latency and in PBMCs from PLWH on suppressive ART. This finding is similar to our recent report of a novel inhibitor of the PAF1 complex (iPAF1C) that had minimal activity on its own, but greatly enhanced the reactivation potential of other LRAs . Both cases highlight the multifaceted nature of the blocks to viral gene expression that underlie the latent state as well as the limitations to single agent drug screening to identify promising, next-generation LRAs. Combinatorial approaches to dissect the genetic underpinnings of HIV-1 latency and discover new, synergistic drug interactions should be prioritized. While KL-2 alone failed to significantly increase HIV-1 gag transcript levels in patient PBMCs, in combination with AZD5582 it resulted in a 16-fold increase over AZD5582 treatment alone and a 53-fold increase over the DMSO control. Crosswise dose titrations in J-Lat 5A8 cells showed a strong synergistic potential between AZD5582 and KL-2. This is consistent with reports of robust synergy between AZD5582 and P-TEFb release through BET bromodomain inhibition . Notably, the BET bromodomain inhibitor JQ1 showed minimal reactivation activity in our patient PBMCs, even in the presence of KL-2, in contrast to our cell line data. This finding could reflect stochastic differences driven by variations in patient characteristics, integration site, chromatin state, transcription factor availability, etc. . Future work will compare P-TEFb release from SECs to release from other cellular reservoirs, such as BRD4 or the 7SK RNP. Recent studies have shown that post-translational modifications of P-TEFb, most notably phosphorylation of CDK9 Serine 175 and Threonine 186, can drive inclusion into different complexes and may strongly influence bioavailability and activity [40, 68–70]. Therefore, it is possible that disruption of complexes housing ‘active’ P-TEFb is a more direct route to redirecting P-TEFb activity. This is not to say that the SEC is never required for viral transcription. In latency model cell lines that lacked a functional Tat (U1 cells) or TAR stem loop (ACH-2 cells), KL-2 inhibited the reactivation potential of several LRAs, suggesting that viral transcription may be more dependent on the SEC when Tat is either defective or not expressed. We attempted to test this by inhibiting Tat in the J-Lat 5A8 model cell line using two previously described Tat-dependent transcription inhibitors, Triptolide and Spironolactone. While both compounds reduced LRA efficacy, KL-2 still boosted the activity of JQ1 and PMA, but not AZD5582. This suggests that P-TEFb can be recruited to proviral integration sites in a Tat and SEC-independent manner upon PMA or JQ1 treatment, potentially through a transcription factor such as NF-kB. The inability of AZD5582 to do so suggests that non-canonical NF-kB activation does not recruit the same milieu of transcription factors, making it uniquely Tat or SEC dependent. This is consistent with the complete lack of activity of AZD5582 in cell line models lacking functional Tat/TAR activity and may underlie the remarkable synergy between non-canonical NF-kB agonists and agents that release P-TEFb . Additionally, triptolide has been characterized outside of HIV-1 transcription in its ability to prevent RNA Pol II reinitiation following pausing through inhibition of xeoderma pigmentosum group B-complementing protein (XPB) and is often used as a tool compound for measuring the fate of paused RNA Pol II at different time points . With this in mind, it is possible that compounds JQ1 and PMA result in de novo recruitment of RNA Pol II thereby increasing transcriptional initiation whereas AZD5582 may be more reliant on RNA Pol II pause-release. To further explore the Tat dependency of KL-2, we tried to rescue Tat function in the U1 cell line using a Dox-inducible system. We hypothesized that by providing Tat, the SEC would no longer be required for viral gene expression such that SEC disruption by KL-2 would enhance reactivation as seen in the J-Lat models. While Tat induction itself was sufficient for reactivation, this reactivation was completely abolished by the addition of KL-2. Even when Tat was minimally induced and other LRAs were added, KL-2 still inhibited reactivation, suggesting that additional factors—such as steady-state levels of P-TEFb, integration site, epigenetic factors, or transcription factor availability—may drive SEC dependency besides just the presence or absence of Tat. Indeed, SEC disruption by KL-2 in the original report of the inhibitor demonstrated an outsized impact on Myc-dependent transcription , suggesting that additional factors driving the SEC dependency of proviral transcription have yet to be described. Regardless, the dual-acting nature of KL-2 in enhancing latency reactivation in some circumstances (i.e., if Tat is present) and inhibiting latency reactivation in others (i.e., when the cell state dictates SEC dependency) presents a unique opportunity to leverage the heterogenous nature of the latent reservoir to both reverse and promote latency. Taken together, our results indicate that release of P-TEFb from cellular SECs is a novel mechanism for promoting HIV-1 viral transcription during both active and latent infection. We demonstrated the enhancement of latency reversal in multiple latent cell line models and in primary PBMCs from PLWH on suppressive ART. This work demonstrates the importance of increasing the production or availability of free P-TEFb for recruitment to viral loci as a powerful strategy for bolstering current LRAs, most notably non-canonical NF-kB agonists. Due to the heterogeneity of blocks to viral replication in the latent reservoir, it is likely that combinatorial LRA treatments will be the best strategy for potent latency reversal moving forward. Further efforts are needed to understand the intracellular distribution of active P-TEFb to characterize the most critical reservoir to target to enhance transcription. Additionally, our work demonstrates that disruption of SECs could enhance latency reversal or promote the maintenance of latency depending on the cellular context. Understanding the mechanism that controls this molecular switch would aid in understanding whether SEC disruptors could be a viable dual-acting molecule to aid in finding a functional cure for HIV-1 infection. Primary human CD4+ T cells from anonymous, healthy donors were isolated from leukoreduction chambers provided by a commercial vendor (STEMCELL Technologies). These were provided without any identifying information and did not require we attain written informed consent. PBMCs from people living with HIV were provided from study subjects enrolled in the Northwestern University Clinical Research Site for the MACS/WIHS combined cohort study (MWCCS). All of these participants provided written informed consent. The Institutional Review Board of Northwestern University approved the study (STU00022906-CR0008) with most recent approval date of May 16, 2022. Primary human CD4+ T cells from healthy donors were isolated from leukoreduction chambers after Trima apheresis (STEMCELL Technologies). PBMCs were isolated by Ficoll centrifugation. Bulk CD4+ T cells were subsequently isolated from PBMCs by magnetic negative selection using an EasySep Human CD4+ T cell isolation kit (STEMCELL Technologies; per the manufacturer’s instructions). Isolated CD4+ T cells were suspended in RPMI 1640 (Sigma-Aldrich) supplemented with 5 mM HEPES (Corning), 1% penicillin-streptomycin (50 mg/ml; Corning), 5 mM sodium pyruvate (Corning), and 10% HI FBS (Gibco). Media were supplemented with interleukin-2 (IL-2; 20 IU/ml; Miltenyi) immediately before use. For activation, bulk CD4+ T cells were immediately plated on anti-CD3–coated plates coated for 2 hours at 37 °C with anti-CD3 (20 mg/ml) (UCHT1; Tonbo Biosciences) in the presence of soluble anti-CD28 (5 mg/ml; CD28.2; Tonbo Biosciences). Cells were stimulated for 72 hours at 37 °C and 5% CO2 before treatment with KL-2. Lyophilized crRNA and tracrRNA (Dharmacon) was resuspended at a concentration of 160 μM in 10 mM Tris-HCL (7.4 pH) with 150 mM KCl. Cas9 ribonucleoproteins (RNPs) were made by incubating 5 μL of 160 μM crRNA (Horizon) with 5 μL of 160 μM tracrRNA for 30 minutes at 37 °C, followed by incubation of the gRNA:tracrRNA complex product with 10 μL of 40 μM Cas9 (UC Berkeley Macrolab) to form RNPs. Five 3.5 μL aliquots were frozen in Lo-Bind 96-well V-bottom plates (E&K Scientific) at −80 °C until use. For synthesis of multiplexed RNPs, four independent crRNA targeting the same gene were mixed at a 1:1:1:1 ratio prior to addition of the tracrRNA as above. crRNA was ordered from Horizon Discovery using either the catalog numbers for predesigned guide sequences or custom guide sequences as indicated in Table 1 below. Following CD4+ T cell isolation and stimulation, as above, cells were counted, centrifuged at 400 × g for 5 minutes, the supernatant was removed by aspiration, and the pellet was resuspended in 20 μL of supplemented room-temperature P3 electroporation buffer (Lonza) per reaction. Each reaction consisted of 1 × 10 cells, 3.5 μL of RNP, and 20 μL of electroporation buffer. The cell suspension was then gently mixed with thawed RNP and aliquoted into a 96-well electroporation cuvette for electroporation with the 4D 96-well shuttle unit (Lonza) using pulse code EH-115. Immediately after electroporation, 80 μL of prewarmed media without IL-2 was added to each well and cells were allowed to rest for at least 30 minutes in a 37 °C cell culture incubator. Subsequently, cells were moved to 96-well flat-bottom culture plates prefilled with 100 μL of warm complete media with IL-2 at 40 IU/mL (for a final concentration of 20 IU/mL) and anti-CD3/anti-CD2/anti-CD28 beads (T cell Activation and Stimulation Kit, Miltenyi) at a 1:1 bead:cell ratio per the manufacturer’s instructions. Whole cell lysates were prepared by suspension of cell pellets (Typically ~150,000 cells) directly in 2.5x Laemmli Sample Buffer followed by denaturization at 98 °C for 30 minutes. 10 million cells were pelleted, washed with PBS, and subsequently resuspended in 1 mL of lysis buffer (0.5% NP40, 50 mM Tris–HCl pH 7.4; 150 mM NaCl, 1 mM EDTA, cOmplete protease (Roche) and PhosSTOP phosphatase (Roche) inhibitors). Samples were then rotated at 4 °C for 30 minutes and transferred to -80 °C for at least 30 minutes to complete cell lysis. Following lysis, samples were thawed on ice and centrifuged in a prechilled microcentrifuge at 3500 x g for 20 minutes to remove cellular debris. Lysates were then precleared by incubation for 1 hour at 4 °C while rotating with 50 μL protein A agarose beads (Cell Signaling Technologies, Cat #9863). CCNT1 antibody (Cell Signaling Technologies, Cat #81464) was then added to precleared lysates at 1:100 and incubated with rotation overnight at 4 °C. After incubation, 50 μL of fresh Protein A agarose beads were added to each sample and incubated with rotation for 1–3 hours at 4 °C. Samples were then washed twice with IP buffer (50 mM Tris–HCl pH 7.4; 150 mM NaCl, 1 mM EDTA) and supernatant was removed. 100 μL 2.5X Laemmli Sample Buffer was then added to the washed beads and denatured at 98 °C for 30 minutes. Samples were run on 4–20% Tris-HCl SDS-PAGE gels (BioRad Criterion) at 90 V for 40 minutes followed by separation at 150 V for 85 minutes. Proteins were transferred to PVDF membranes by electrotransfer (BioRad Criterion Blotter) at 90 V for 2 hours. Membranes were blocked in 5% milk in DPBS, 0.1% Tween-20 or 5% BSA in DPBS, 0.1% Tween-20 for 1 hour prior to primary antibody incubation overnight at 4 °C. Anti-rabbit or anti-mouse IgG horseradish peroxidase (HRP)-conjugated secondary antibodies (1:20000, polyclonal, Jackson ImmunoResearch Laboratories, Cat. Nos. 111-035-003 and 115-035-003) were detected using Pierce ECL Western Blotting Substrate (ThermoFisher) on iBright (Thermofisher) blot scanner. Blots were incubated in a 1xPBS, 0.2 M glycine, 1.0% SDS, 1.0% Tween-20, and pH 2.2 stripping buffer before reprobing. Details for each primary antibody used in this study are provided in Table 2 below, including the target protein, manufacturer, animal, catalog number, and URL. Replication-competent reporter virus stocks were generated from an HIV-1 NL4.3 molecular clone in which GFP had been cloned behind an IRES cassette following the viral nef gene (NIH AIDS Reagent Program, catalog no. 11349). Briefly, 10 μg of the molecular clone was transfected (PolyJet; SignaGen) into 5 × 10 human embryonic kidney (HEK) 293T cells (ATCC, CRL-3216) according to the manufacturer’s protocol. 25 mL of the supernatant was collected at 48 and 72 hours and then combined. The virus-containing supernatant was filtered through 0.45-mm polyvinylidene difluoride filters (Millipore) and precipitated in 8.5% polyethylene glycol [average molecular weight (Mn), 6000; Sigma-Aldrich] and 0.3 M NaCl for 4 hours at 4 °C. Supernatants were centrifuged at 3500 rpm for 20 minutes, and the concentrated virus was resuspended in 0.25 ml of PBS for a 100X effective concentration. Aliquots were stored at −80 °C until use. Edited primary CD4+ T cells were plated into a 96-well, round-bottom plate at a cell density of 1 × 10 cells per well and cultured overnight in 200 μL of complete RPMI 1640 as described above in the presence of IL-2 (20 IU/ml) and 2.5 μL of concentrated virus stock. Cells were cultured in a dark humidified incubator at 37 °C and 5% CO2. On days 2 and 5 after infection, 75 μL of each culture was removed and mixed 1:1 with freshly made 2% formaldehyde in PBS (Sigma-Aldrich) and stored at 4 °C for analysis by flow cytometry. Cultures were supplemented with 75 ml of complete IL-2–containing RPMI 1640 medium and returned to the incubator. Activated primary CD4+ T cells were plated into a 96-well, round-bottom plate at a cell density of 1 × 10 cells per well and cultured overnight in 200 μL of complete RPMI 1640 as described above in the presence of IL-2 (20 IU/mL) with different concentrations of KL-2 or equivalent volumes of DMSO. The next day, 2.5 μL of concentrated virus stock was added to each well. Cells were cultured in a dark humidified incubator at 37 °C and 5% CO2. On days 2 and 5 after infection, 75 μL of each culture was removed and mixed 1:1 with freshly made 2% formaldehyde in PBS (Sigma-Aldrich) and stored at 4 °C for analysis by flow cytometry. Cultures were supplemented with 75 mL of complete IL-2–containing RPMI 1640 medium and returned to the incubator. CHME3-4x4 cells were plated into a 96-well, flat-bottom plate at a cell density of 10,000 cells per well and cultured overnight in 200 uL of DMEM + 10% FBS. After overnight seeding, the media was removed and replaced with DMEM + 10% FBS and different concentrations of KL-2 or equivalent volumes of DMSO. The next day, the cells were infected with 2.5 uL of HIV-1 NL4.3 containing a nano-luciferase reporter in place of Nef (Nef:Nano-Luc). Two days following infection, the cell culture media was removed and Nano-Luc production was measured using the Nano-Glo Luciferase Assay System (Promega, Cat# N1110). In parallel, an identical plate of CHME3-4x4 cells treated with KL-2 was cultured as described above. On the same day as the infection analysis, the identical plate was analyzed for viability using the CellTiter Glo Luminescent Viability Assay System (Promega, Cat# G7572). J-Lat 5A8, J-Lat 11.1, J-Lat 6.3, J-Lat A72, U1, and ACH-2 cell lines were plated in 96-well flat bottom plates at a density of 50,000 cells/200 μL supplemented RPMI 1640. Cells were DMSO-treated or treated with KL-2 (6.25 μM), JQ1, PMA, and AZD5582 for 48 hours at the concentrations indicated below in Table 3. Cells were then washed in PBS and resuspended in PBS + 1% formaldehyde and fixed for 30 minutes. Analysis was performed by flow cytometry gating on GFP-positive cells for J-Lat cell lines. Extracellular staining was performed on uninfected, live primary T cell populations with anti-CD4-PE (Miltenyi Biotec, 130-113-225), anti-CXCR4-APC (Miltenyi Biotec, Cat#130-120-708), and anti-CD25-APC (Miltenyi Biotec, Cat#1130-115-535) antibodies according to the manufacturer’s instructions. Briefly, cells were pelleted, and media was removed. Cells were then washed once with DPBS and resuspended in a 1:50 dilution of the appropriate antibodies in MACS buffer (DPBS + 0.5% bovine serum albumin (BSA) and 2 mM EDTA) and incubated for 15 minutes at 4 °C. Cells were then pelleted, washed with MACS buffer, and suspended in DPBS + 1% formaldehyde for 30 minutes prior to flow cytometry. Viability staining was performed on various cell populations using the amine-reactive dye Ghost Red 710 (Tonbo, Cat#13-0871-T100) according to manufacturer’s instructions. Briefly, cells were pelleted and media was removed. Cells were then washed once with DPBS and resuspended in a ghost red dye solution consisting of 1 μL of ghost red dye per 1 mL of DPBS and incubated for 30 minutes at 4 °C. Cells were then pelleted, washed with MACS buffer, and suspended in DPBS + 1% formaldehyde for 30 minutes prior to flow cytometry. Intracellular HIV-1 p24 staining was performed on fixed U1 and ACH-2 cells with KC57-FITC (Beckman Coulter, Cat#6604665). After allowing at least 30 minutes for fixation, cells were pelleted and washed with DPBS. Subsequently, cells were suspended in DPBS + 1% BSA, 0.1% w/v saponin and incubated at room temperature for 20 minutes to block and permeabilize. Cells were then spun down to remove the supernatant and incubated for 30 minutes at room temperature in the dark with KC57-FITC at a concentration of 1:50 in DPBS plus 1% BSA, 0.1% saponin w/v. Cells were then pelleted again, washed with PBS plus 1% BSA, and suspended in 1% formaldehyde DPBS for fixation prior to flow cytometry. Flow cytometry analysis was performed on an Attune NxT acoustic focusing cytometer (Thermo Fisher Scientific), recording all events in a 40-μL sample volume after one 150 μL of mixing cycle. Data were exported as FCS3.0 files using Attune NxT Software v3.2.0 and analyzed with a consistent template on FlowJo. Briefly, cells were gated for lymphocytes by light scatter followed by doublet discrimination in both side and forward scatter. Cells with equal fluorescence in the BL-1 (GFP) channel and the VL-2 (AmCyan) channel were identified as autofluorescent and excluded from the analysis. A consistent gate was then used to quantify the fraction of remaining cells that expressed the target of interest. Total RNA isolation from cells (typically 100,000 cells) treated with our LRA panel in the presence or absence of KL-2 was carried out with an RNeasy kit (Qiagen), with the optional on-column deoxyribonuclease I digestion step. The isolated total RNA was eluted in ribonuclease-free water and RNA concentrations were subsequently quantified using a NanoDrop One (Thermo Fisher). cDNA was synthesized from extracted RNA (typically 100 ng) using the SuperScript IV first strand synthesis system (Invitrogen, Cat#18091050) according to the manufacturer’s instructions. Briefly, a 13 μL reaction mixture was made with random hexamers (1 μL; 50 ng/μL), 10 mM dNTP (1 μL), template RNA (100 ng) and DEPC-treated water. The RNA-hexamer mix was then incubated at 65 °C for 5 minutes before incubating on ice for another minute. Then a mix of 5x SSIV Buffer (4 μL), 100 mM DTT (1 μL), Ribonuclease Inhibitor (μL), and SuperScript IV Reverse Transcriptase (1 μL, 200 U/μL) was added to each sample. The combined reaction mixture was then incubated at 23C for 10 minutes, followed by 50 °C for 10 minutes, and the reaction was inactivated by incubation at 80 °C for 10 minutes. Following complete reverse transcription, residual RNA was cleared by the addition of 1 μL E. Coli RNase H to each reaction mix and incubated at 37 °C for 20 minutes. The RT products were then stored at -20 °C. Following cDNA synthesis, viral transcripts were assessed by qRT-PCR as described previously . For HIV-1 TAR, the following primers were used: PF: 5′-GTCTCTCTGGTTAGACCAG-3′; PR: 5′-TGGGTTCCCTAGYTAGCC-3′; and probe: 5′-AGCCTGGGAGCTC-3′. HIV-1 Long LTR levels were assessed using the following primers: PF: 5′-GCCTCAATAAAGCTTGCCTTGA-3′; PR: 5′-GGGCGCCACTGCTAGAGA-3′; and probe: 5′-CCAGAGTCACACAACAGACGGGCACA-3. For expression level normalization, β-Actin was used (Thermo Fisher Scientific, catalog no. 4331182). Notably, β-Actin is an RNA Pol II controlled gene. To ensure that β-Actin levels were not being altered by KL-2 induced SEC disruption, we compared β-Actin expression to that of the 18S ribosomal subunit, an RNA Pol I controlled gene (Thermo Fisher Scientific, catalog no. 4331182). No statistically significant changes were noted between the two housekeeping genes upon KL-2 treatment, so β-Actin was used as the normalization gene for subsequent qPCR analyses. The reaction was performed using TaqMan Fast Advanced Master Mix (Thermo Fisher Scientific, catalog no. 4444553) according to manufacturer’s instructions. Briefly, 10 μL of reaction was mixed using 5 μL of Taqman Master Mix, 0.5 μL of 20x primer probe mix (18 μM of primers and 5 μM probe), 2.5 μL water, and 2 μL of template cDNA. The PCR cycles were as follows: 50 °C for 2 minutes, 95 °C for 20 seconds, followed by 40 cycles of 95 °C for 1 second and 60 °C for 20 seconds. ChIP-seq was performed according to a previously published protocol . Briefly, culture media were aspirated and about 20 million to 50 million of cells were washed twice with ice-cold 1× phosphate-buffered saline (PBS; Thermo Fisher Scientific, catalog no. 14190250) and then cross-linked with 1% paraformaldehyde (Thermo Fisher Scientific, catalog no. 28908) for 10 min while shaking at room temperature. The reaction was quenched using 0.2 M glycine (Fisher Scientific, catalog no. BP381-5) for 5 minutes at room temperature followed by centrifugation at 1350 RPM for 5 minutes. Subsequently, cells were gently washed with ice-cold DPBS and centrifuged at 1350 RPM for 5 minutes. For RNA Pol II and CCNT1 ChIP-seq, the chromatin was sonicated using Covaris E220 for 4 minutes using the following sonication conditions: 10% duty cycle, 140 peak intensity power, and 200 cycles per burst. Pulldown of chromatin was carried out overnight at 4°C using the specific antibodies [Rpb1 NTD (D8L4Y, CST catalog no. 14958) rabbit monoclonal antibody for RNA Pol II and CCNT1 (D1B6G, CST catalog no. 81464) rabbit monoclonal antibody]. The next day, Dynabeads protein G (Invitrogen, catalog no. 10004D) were added to the immunoprecipitation mix and incubated at 4 °C for 4 hours. Nonspecific proteins were washed away, and bead-bound proteins were digested using proteinase K (400 mg/ml; Roche, catalog no. 3115828001). Reverse cross-linking was performed at 65 °C overnight while shaking at 1200 RPM. Immunoprecipitated DNA was extracted using phenol-chloroform (Thermo Fisher Scientific, catalog no. 17909) followed by ethanol precipitation and washing. DNA was dissolved in 10 mM tris-HCl (pH 8) and quantified using Qubit. DNA libraries were prepared by the HTP Library Preparation Kit for Illumina (KAPA) and sequenced on NovaSeq 6000 or NovaSeq X (Illumina) in the single-end (SE) mode. In order to analyze RNA Pol II and CCNT1 occupancy across the proviral gene body, we used all Chip-Seq sequences attained and performed HIV-1 detection and assembly using the Burrows-Wheeler Aligner (BWA) v.0.7.17 ). To this end, we used the mem algorithm of BWA and the HXB2 HIV-1 reference genome (K03455.1). Subsequently, samtools v.1.17 was used to generate sorted.bam files and to calculate the depth of coverage with its coverage function. Coverage metrics where then analyzed and plotted using R v.4.3.2 in-house scripts and ggplot2 v.3.5.1 R package. In order to control for different inputs or sequencing depth that could bias viral quantification, we normalized the viral transcriptional quantification with equally bounded regions of the human genome under the different tested conditions. With this goal, we performed alignment, peak detection, and differential binding using the Chip-Seq human reads. To complete this analysis, we first performed sequencing data trimming using Trimmomatic v0.36 to remove adapters and low-quality reads. Trimmed reads were then aligned to the Homo sapiens reference genome GRCh38 using the Hisat2 v2.2.1 disabling the splice alignment option. Subsequently, peaks were detected with MACS v3.0.1 with a minimum FDR cutoff for peak detection of 0.01 using the input of each experiment as control. Finally, differential binding was tested using DiffBind v3.12.0 R package with the DESeq2 method. Two peaks corresponding with transcription start sites identified using ChIPpeakAnno v3.0.0 R package that were equally bound in the DMSO and KL2 conditions in the CCNT1 Chip-Seq or in the RNA Pol II ChIP-Seq [as determined by having some of the highest FDR values (FDR>0.99)] were selected as internal controls to normalize the HIV-1 reads. For the CCNT1 ChIP-Seq, we selected the ALG6 transcription start site while for the RNA Pol II ChIP-Seq we used the FNIP2 transcription start site. Reads quantified for these peaks by DiffBind were used as normalizing factors for the HIV-1 depth of coverage analysis. RNA samples from J-Lat 5A8 cells were submitted to Novogene for next-generation sequencing. Libraries were prepared using NEBNext Ultra II kit with poly(A) selection according to manufacturer protocols (New England Biolabs). Samples were sequenced on an Illumina NovaSeq X Plus with 150 bp paired-end reads. Sequencing data was demultiplexed and trimmed using Trimmomatic v0.36 to remove adapters and low-quality reads. Trimmed reads were aligned to the Homo sapiens reference genome GRCh38 and transcripts quantified using the Hisat2-StringTie pipeline . Differential gene expression analysis of the quantified gene transcripts was performed with DESeq2 v.1.42.0 R package using R v.4.3.2. After retaining genes with nonzero total read count and with more than 10 reads in total between all samples, we fitted a model that included all treatments to account for overall variability and identified differentially expressed genes (DEGs) within that model between all tested conditions against DMSO-treated cells (i.e. KL2 vs DMSO, AZD5582 vs DMSO, and AZD5582+KL2 vs DMSO). To define DEGs, we used as cut-offs an absolute log2 fold change > 1 and a false discovery rate (FDR) < 0.05 using the Benjamin-Hochberg procedure. Gene enrichment analyses for each comparison were subsequently performed using gene set enrichment analysis (GSEA) to identify specific Gene Ontologies (GO), KEGG, and REACTOME pathways associated with CO-iMs and/or HIV infection and ART treatment. We performed GSEA using clusterProfiler v.4.10.0 in R with all lists of genes ranked by the corresponding log2 fold change and compared the different treatments with the compareCluster function. For these analyses all genes whose gene symbols could be mapped to ENTREZ Ids using the org.Hs.eg.db v.3.18.0 Bioconductor annotation package were included. Codon optimized HIV-1 NL4.3 Tat with a C-terminal 2xStrep(TagII)-TEV-3xFlag tag was ordered as a GeneBlock (IDT) and inserted into BamHI/ECORI linearized pLVX-TetOne-Puro by Gibson Assembly and subsequently sequence verified. Briefly, 10 μg of the molecular clone was transfected (PolyJet; SignaGen) into 5 × 10 human embryonic kidney (HEK) 293T cells (ATCC, CRL-3216) according to the manufacturer’s protocol. Twenty-five mL of the supernatant was collected at 48 and 72 hours and then combined. The virus-containing supernatant was filtered through 0.45-mm polyvinylidene difluoride filters (Millipore) and precipitated in 8.5% polyethylene glycol [average molecular weight (Mn), 6000; Sigma-Aldrich] and 0.3 M NaCl for 4 hours at 4 °C. Supernatants were centrifuged at 3500 rpm for 20 minutes, and the virus was resuspended in 0.25 ml of PBS for a 100X effective concentration. Aliquots were stored at −80 °C until use. Aliquots were thawed and added to cultures of U1 cells at a concentration of 1:100 in complete RPMI and cells were cultured for 48 hours. Following transduction, the cells were pelleted, and media was removed and replaced with fresh complete RPMI + 10 μg/mL of puromycin to select for successfully transduced cells. Cells were cultured in selective growth media for 7 days. After selection, cells were plated in 96-well flat-bottom plates according to previous methods and treated with latency reversing agents and doxycycline for 48 hours before analysis of p24 expressing cells by flow cytometry. We selected five study subjects enrolled in the Northwestern University Clinical Research Site for the MWCCS who were well-suppressed [undetectable plasma HIV-1 (<50 copies/ml)] and had received antiretroviral drugs for at least 5 years. We obtained patient PBMCs from cryostorage. Laboratory procedures for clinical sample management are described previously . The Institutional Review Board of Northwestern University approved the study (STU00022906-CR0008) with most recent approval date of May 16, 2022. All participants provided written informed consent. Total RNA was isolated from cultures of 1 x 10 PBMCs that were treated with for 48 hours with JQ1, or AZD5582 in the presence or absence of KL-2 using an RNeasy kit (Qiagen), with the optional on-column deoxyribonuclease I digestion step. The isolated total RNA was eluted in ribonuclease-free water, cleaned, and concentrated using the RNA Clean and Concentration kit (Zymo) and assessed for quantity and quality by Qubit (Thermo Fisher Scientific) and 4200 TapeStation (Agilent), respectively before qRT-PCR. We performed real-time qRT-PCR using an HIV-1-gag–specific primers-probe (FAM) set: HIV-1-gagF, 5′-GGTGCGAGAGCGTCAGTATTAAG-3′; HIV-1-gagR, 5′-AGCTCCCTGCTTGCCCATA-3′; HIV-1-gagProbe, 6FAM-5′-TGGGAAAAAATTCGGTTAAGGCCAGGG-3′-QSY. We used the lactate dehydrogenase A (LDHA) gene for the internal normalization primers-probe (VIC) set (VIC-MGB: assay ID Hs03405707_g1; TaqMan Gene Expression Assay, Thermo Fisher Scientific). Briefly, a 10-μL RT-PCR mixture contained TaqMan Fast Virus 1-Step Master Mix, 400 nM forward and reverse HIV-1-gag primers, 0.3 μL of LDHA Gene Expression Assay (Thermo Fisher Scientific), 250 nM each of the probes, and 5 μL of extracted RNA or water for the no template controls. We programmed the 7900HT real-time PCR system (Applied Biosystems) for 20 minutes at 50 °C and 20 seconds at 95 °C, followed by 40 cycles of 15 seconds at 95 °C and 60 seconds at 60 °C. The qRT-PCR data were analyzed in technical triplicate. We calculated the fold change in gene expression using the standard 2^-ΔΔCT method. All statistical analysis was performed using GraphPad Prism version 10.2.0 (392) for Windows 64-bit, GraphPad Software, Boston, Massachusetts, USA (www.graphpad.com).
PMC1523200
Genome wide profiling of human embryonic stem cells (hESCs), their derivatives and embryonal carcinoma cells to develop base profiles of U.S. Federal government approved hESC lines
In order to compare the gene expression profiles of human embryonic stem cell (hESC) lines and their differentiated progeny and to monitor feeder contaminations, we have examined gene expression in seven hESC lines and human fibroblast feeder cells using Illuminabead arrays that contain probes for 24,131 transcript probes. A total of 48 different samples (including duplicates) grown in multiple laboratories under different conditions were analyzed and pairwise comparisons were performed in all groups. Hierarchical clustering showed that blinded duplicates were correctly identified as the closest related samples. hESC lines clustered together irrespective of the laboratory in which they were maintained. hESCs could be readily distinguished from embryoid bodies (EB) differentiated from them and the karyotypically abnormal hESC line BG01V. The embryonal carcinoma (EC) line NTera2 is a useful model for evaluating characteristics of hESCs. Expression of subsets of individual genes was validated by comparing with published databases, MPSS (Massively Parallel Signature Sequencing) libraries, and parallel analysis by microarray and RT-PCR. we show that Illumina's bead array platform is a reliable, reproducible and robust method for developing base global profiles of cells and identifying similarities and differences in large number of samples.Embryonic stem cells (ESCs), derived from the inner cell mass of pre-implantation embryos, have been recognized as the most pluripotent stem cell population. Human ES cells (hESCs) can be maintained and propagated on mouse or human fibroblast feeders for extended periods in media containing basic fibroblast growth factor (bFGF) [1-4] while retaining the ability to differentiate into ectoderm, endoderm and mesoderm as well as trophoectoderm and germ cells. Gene expression in hESC has been investigated by a variety of techniques including massively parallel signature sequencing (MPSS), serial analysis of gene expression (SAGE), expressed sequence tag (EST) scan, large scale microarrays, focused cDNA microarrays, and immunocytochemistry [5-7]. Markers for hESCs that may also contribute to the "stemness" phenotype have been established and markers that distinguish ESCs from embryoid bodies (EB) have been developed. Novel stage-specific genes that distinguish between hESCs and EBs have been identified and allelic differences between ESC have begun to be recognized [8-10]. As the potential of hESCs and their derivatives for regenerative medicine is being evaluated, it has become clear that the overall state of the cells, degree of contamination and comparisons of the more than a hundred different newly derived lines will need to be performed. It will be necessary to develop methods to monitor and assess hESC and their derivatives on a routine basis. Since differentiated cells are often scattered within or at the edge of colonies and the differentiation is so subtle that morphological characteristics and even immunohistochemistry are insufficient to detect it, larger scale methods of analysis need to be developed. Our strategy was to compare a variety of different hESC lines that were derived and expanded by three different institutions (WiCell Research Institute, BresaGen, Inc., and Technion-Israel Institute of Technology), and cultured in two separate laboratories (Burnham Institute and NIA) to a baseline set of data against which cell samples can be compared. By using cells grown in different conditions we expected to be able to identify core commonalities and by comparing feeders and embryoid bodies (EB) with hESC identify measures of contamination and early markers of differentiation. Further, by comparing embryonal carcinoma cell (EC) and karyotypically variant lines with hESC, we would be able to directly assess their utility as surrogates (for quality control purposes) for hESC. We employed a pre-commercial prototype of the Illumina HumanRef-8 BeadChip , a genome-scale bead based array technology that combines the sensitivity and low cost of a focused array with the coverage of a large scale array, while requiring much smaller sample sizes than MPSS, EST scan or SAGE. We show that the Illumina bead based array correctly identified blinded duplicates as the closest related samples and readily distinguished between hESC lines, as well as between ESCs and EBs derived from them. This array allowed us to estimate the degree of feeder contamination present in the cultures. Similarities and differences between EC line NTera2 and hESC lines could be determined and verified, and the database comparisons allowed us to identify core self-renewal pathways that regulate hESC propagation. Forty-eight samples were selected from multiple laboratories and gene expression profiles were examined using a bead array containing 24,131 transcripts derived from the Human RefSeq database that included full length and splice variants. Each gene was represented by sequences containing an average of thirty beads to provide an internal measure of reliability. Samples included 7 hESC lines BG01, BG02, BG03, I6, H1, H7 and H9, EBs that were differentiated from hESCs of the three BG lines, human fibroblast feeder HS27 (ATCC), hESC-derived fibroblasts, karyotypically abnormal hESC line BG01 Variant (BG01V) and EC line NTera2 . Samples were blinded and biological and technical repeats were examined at the same time. A single slide contained eight replicates and six such slides were used for the present set of samples. Results were normalized to average following Illumina Beadstudio manual and the quality of each sample was assessed by immunocytochemitsry and RT-PCR prior to subjecting them for analysis (data not shown). Results from the entire sample set are available for download as an excel spreadsheet (Additional file 1) and a CD of the results is available upon request. The total number of genes identified as expressed at >0.99 confidence is summarized in Table 1. Intensity results are reported in arbitrary units and ranged from 10 to 20,000 (a two thousand fold range). Although the sensitivity of the array has been reported to be high, in the present report we have restricted our analysis to expression of at least 100 units in any one sample. Using this cutoff, on average cells expressed approximately 8,000 transcripts (Table 1, 2), a number similar to the number detected by SAGE, MPSS and EST analysis [5-7,10,15,16]. As with other analysis, genes with the highest abundance were housekeeping genes, ribosomal genes and structural genes (Table 2 and Additional file 1). These genes were similar in most samples though relative levels varied. Correlation coefficients of paired samples in this bead array In order to test the reproducibility and reliability of the bead array, duplicate samples of hESC lines H9, I6, and EC line NTera2 and human fibroblast feeders (HS27) were run at the same time and correlation coefficients (R) of duplicates were generated using the entire data of all genes with expression level >0 (§), or genes with detection confidence >0.99 (*), or genes with detection confidence >0.99 and expression level > 100 arbitrary units (#). Note that the correlation coefficients are in the range of 0.9382–0.9761 and the number of genes was in the range of 10,000–14,000. Distribution of genes with expression levels <50 and >50–10,000 as detected by Illumina bead array in 8 hESC populations All human ESC samples were hybridized in one experiment and the relative detection levels of genes were binned to obtain a global overview of transcription, approximately 8, 000 genes (~50%) were greater than 100 arbitrary units. The numbers are similar to results obtained by other large scale analysis such as MPSS. One of the advantages of the Illumina arrays is the ability of running multiple samples simultaneously thus allowing multiple pairwise comparisons to be performed readily. To show the similarity of relative gene expression between samples, we have used Illumina Beadstudio and clustering software packages Pcluster and TreeView to generate a heat-map (Figure 1) and a dendrogram (Figure 2). Based on their properties, we classified some of our samples into four groups, (A) undifferentiated hESCs (including a sample from karyotypically abnormal variant, designated as "ES", n = 11); (B) differentiated ES cells and EBs (designated as "EB", n = 6); (C) hESC derived neural cells (designated as "NS", n = 3); and (D) hESC derived mesenchyme and human fibroblast feeder cells (designated as "FB", n = 5) and these groups were shown in the heat-map. Comparing the overall pattern of expression, we made several important observations: 1) Duplicates clustered close to each other and were more related to each other than to any other sample; 2) ESCs appeared more similar to each other than to EBs; 3) NTera2 cells appeared more similar to ESCs while differentiated NTera2 and EBs can be readily distinguished from their parent populations (Figure 2); 4) BG01V appeared similar to undifferentiated BG01 cells; 5) In general ESC lines grown in one laboratory appeared more similar than samples grown in other laboratories, suggesting that culture conditions affected gene expression but that this effect was much smaller than the effect of differentiation. Unsupervised two-way hierarchical cluster analysis of differentially expressed genes illustrated in a heat-map. Each row represents the relative levels of expression of a single gene. Each column represents a sample. The samples include four groups of cells, ES designates 11 samples of hESCs, EB contains 6 samples of differentiated ESCs and EBs, NS consists of 3 hESC derived neural cells and FB is a collection of hESC derived mesenchyme and fibroblasts. High expressions relative to mean are colored red. Low expressions are colored green. Black represents no significant change in expression level between mean and sample. Samples cluster closer within their own group than samples from other groups. Dendrogram of unsupervised one-way hierarchical clustering analysis of relative expression of genes in selected samples. The clustering analysis was based on the average linkage and Euclidean distances as the similarity metric using differentially expressed genes identified by ANOVA (p < 0.05). hESCs clustered together and BG lines cultured in the same laboratory shared the largest similarities. EBs were separated from hESCs from which they were derived. EC line NTera2 and feeder cells can be distinguished from hESCs respectively. The global analysis suggested that the bead arrays used were sufficiently sensitive such that individual subsets of genes could be analyzed, different populations of cells could be readily distinguished and that a subset of candidate genes could be sufficient to distinguish between groups of cells. The comparison across multiple samples will allow a set of core stem cell markers to be identified. In subsequent sections we have performed such analysis. Readers are urged to analyze the expression of desired genes directly as it is impossible to test every gene given the large body of data generated. We have previously used EST scan and MPSS to analyze pooled samples of ESCs and EBs from three different WiCell lines (H1, H7 and H9) . Comparison between the two methodologies indicated that while there is good concordance for genes expressed at high levels, this does not hold for genes expressed at lower levels. As a test of the quality of the data generated in these experiments and to evaluate whether comparisons can be made across different methodologies, we re-ran the identical samples on the bead array platform. The complete comparison of gene expression is shown in Additional file 2 and is summarized in Tables 3 and Table 4. Overall, concordance in Illumina array was better than that evident between EST scan and MPSS datasets , but clearly showed much wider differences than that seen with running duplicates in the same assay format. Nevertheless, this comparison provides an independent verification of the data and suggests that if a sample is detected in more than one large-scale analysis, the reliability of the gene expression detection is high, which also reduces the number of individual genes needed to be verified. Caution should be observed in comparing different samples run on different platforms, especially when there has not been rigorous bioinformatic matching of the source sequences used to identify genes in the platforms. Often genes called by the same symbol originate from different database records, which may originate from different splice variants or contain sequence differences due to polymorphisms or outright error . Expression of hESC specific markers in pooled hESC sample as detected by Illumina bead array The expression of previously identified hESC markers was examined in all hESC samples (the values displayed represent the expression level of pooled H1, H7 and H9). Most of the genes were also identified using Illumina bead array in all 8 hESC populations in this study (1*), the gene CER1 was detected in all except one duplicate of H9 (2*), Nanog was not detected in all populations (3*) and Sox2, Lin41, NR6A1 and FoxD3 were not detected in the array although they were present in the chips for hybridization (4*). Comparison of MPSS and Illumina bead array results The samples were analyzed by MPSS and bead array. The number of genes detected by each method and the degree of overlap is summarized. Note much higher degree of overlap when the top 2000 hits were compared. *: Most of the genes detected by MPSS were novel genes not included in the bead array. For all samples, we conducted an unsupervised one-way hierarchical clustering analysis. The clustering analysis was based on the average linkage and Euclidean distances as the similarity metric using differentially expressed genes identified by ANOVA (P < 0.05). The analysis revealed the underlying features and variation patterns of gene expression in each cell types. Figure 2 shows results of the cluster analysis of relative gene expression in selected samples. As one of our purposes of this study was to distinguish between human fibroblast feeders cells and hESCs and hEBs, wishing to readily detect feeder contamination in hESCs, we included one of the human feeder cells HS27 (ATCC) in this study. We have been using HS27 as feeder cells for H9 hESCs for more than two years and all hESCs grown on HS27 had normal karyotype, expressed all undifferentiated markers, and made teratomas with all germ layers (data not show). The global pairwise comparison clearly showed that human feeders were far more dissimilar to hESCs than hESCs grown in different laboratories, hESCs compared to their differentiated EBs that contained mesodermal tissue, and hESCs compared to the karyotypically variant hESC line BG01V. Pairwise comparisons of human feeders with hESCs resulted in a correlation coefficient of 0.66, which was less than the correlation coefficient of 0.71–0.74 observed between hESCs and their corresponding EBs. The large difference between human feeders and hESCs suggested that it would be possible to identify markers that were robust and reliable in distinguishing the two populations, and these markers would be sufficiently sensitive in detecting contamination of feeders. We examined the data to develop a list of genes that had high levels of expression in human feeder cells maintained in hESC medium but whose expression was low or absent in either ESCs or EBs. The absence of expression in EBs was used as a control for spontaneous differentiation of ESC colonies (including mesodermal differentiation) which may occur and the markers selected should be able to distinguish between these two events. A complete list of genes expressed at least ten-fold higher in human feeders is provided in Figure 3. Quantitative RT-PCR (qPCR) was used to verify the fold change of the expression of 4 genes, including THBS1, MMP3, TNFRSF11B and KRTHA4 (Figure 3C). Further confirmation can also be done using immunocytochemistry, as antibodies against these genes are commercially available. Human fibroblast feeder cells can be distinguished from hESCs and EBs. Bead array identified lists of genes that were uniquely expressed in human fibroblast feeders as compared to hESCs (A) and hEBs (B). The four genes whose expression was confirmed by qPCR (C) were in bold. In the graph (C), gene expression of each gene in feeder cells was designated as 1 fold and the bars represented fold decrease for each gene. Thus this comparison allowed us to distinguish between hESCs and human feeders and identify candidate markers that could detect feeder cell contamination should human feeders be used in the propagation of hESCs. Illumina bead array analysis confirmed that hESCs could be readily distinguished from EBs by global analysis. This raised the possibility that specific subsets of markers could be identified. We and others have used MPSS and EST scan and generated array data to make lists of hESC-specific genes . As discussed above, most hESC markers identified by MPSS have been detected in the present bead array analysis (Table 3), confirming the utility of these previously identified markers for use in assessing undifferentiated status of hESCs. In addition, we have generated a list of genes differentially expressed at higher level in EBs than in hESCs, a subset of which is shown in Table 5. These markers were common to all EB samples tested and included genes known to be expressed in ectoderm, endoderm and mesoderm. The entire set of differentially expressed genes is provided in Additional file 3. Thus, the bead array format, which allows multiple pairwise comparisons, can be used to identify genes that are expressed by all differentiating EB samples in the present study. Our data suggested that a core set of limited markers might be sufficient to monitor the process of differentiation. By suitable selection of different germ cell layer specific markers one may also assess the overall quality of differentiation toward germ cells. Genes which are differentially expressed at higher levels in EBs than in hESCs Our cluster analysis indicated that BG01, BG02 and BG03 cell lines were overall more similar to each other than to other lines (Figure 1 and 2), but nevertheless showed additional differences than technical or biological repeats of the same sample. This raised the possibility that this microarray strategy may be sufficiently sensitive to identify relatively cell type specific candidate genes that could be used to distinguish one hESC population from another or to identify differences that were due to varied isolation and growth conditions. As a test we looked for differences between BG01, BG02 and BG03, which were grown in the same laboratory under the same conditions. Lists of candidate genes are shown in Figure 4A, C and 4E and the comparison of these three lines are shown in scatter plots in Figure 4B, D and 4F. BG lines show small but distinct differences as assessed by bead array. These three hESC lines share high similarities as shown by the scatterplots of BG01 vs BG02 (B), BG01 vs BG03 (D) and BG02 vs BG03 (F). Comparisons of all three lines were made and lists of selected genes that were specifically expressed in BG01 (A), BG02 (C) and BG03 (F) are shown. Correlation coefficients (R) were generated using all genes with expression level >0 (black and blue dots), or all genes with detection confidence >0.99 (blue dots). Genes outside the two thin red lines were detected at >2.5- fold difference. We reasoned as well that such a global comparison should allow us to distinguish between male and female lines if genes present on the Y chromosome were expressed at high levels in the undifferentiated state and were detected by the bead array. Several such candidate genes were identified. The most robust were RPS4Y, RPS4Y2, and EIF1AY (Figure 5). To confirm that these were useful markers, we designed RT-PCR primers and tested their expression in a male (BG01) and a female (BG03) line (Figure 5B). We noted that several of these continued to be expressed at high levels as ESCs differentiated to form EBs and upon further differentiation (data not shown), suggesting that these markers might be used in adult stem cell and germ cell populations as well. Male and female hESC lines can be distinguished by genes identified by bead array. Five potential genes RPS4Y, RPS4Y2, EIF1AY, VCY, and AMELY are located in the Y chromosome. By comparing the expression level of these genes in all hESC lines, we have found that 3 out of 5 were specifically expressed in male hESC lines I6, BG01 and BG02 (A) and this was verified by RT-PCR in male line BG01 and female line BG03 (B). G3PDH was used as an internal control. *: represents the gene expression level is detected at <0.99 confidence. In summary, our data suggest that the bead array format is sufficiently sensitive and global that it can distinguish one cell line from another even if those two cell lines are grown in the same laboratory under virtually identical conditions. Bead array can also be used to distinguish between male and female lines. Our previous results have suggested that EC lines share many of the properties of hESCs and can be used as a useful model for initial testing of biological questions . More recently we have identified BG01V as a karyotypically abnormal variant that behaves much like its normal counterpart BG01, but is not subject to the same constraints of use as karyotypically normal hESCs . Given the sensitivity of the bead array analysis, we tested its ability to detect the overall similarities and differences between NTera2 and a pooled ESC sample or between the karyotypically abnormal BG01V and its normal parent line (Figure 6). Diploid pluripotent EC cell line NTera2 and karyotypically abnormal hESC line BG01V can be distinguished from normal hESCs using Illumina array. Comparison of NTera2 and pooled hESC sample resulted a correlation coefficient of 0.8997. Two lists of genes, which were specifically expressed in NTera2 (C) or in hESCs (E) were identified. Likewise, while sharing similarities with BG01 (B, correlation coefficient= 0.9043), BG01V was different from BG01 in expression for many genes, particularly genes from the TGFβ pathway (D, F). Black dots represent genes that were detected at >0 expression level, blue dots represent genes that were detected both at > 0 expression level and at >0.99 confidence. Genes plotted outside the two thin red lines were detected at >2.5- fold difference. Our results showed that, while NTera2 shared a high similarity with hESCs , it did have important differences with hESC lines. Examining these differences (summarized in Figure 6C and 6E), we noted that some reflected the origin of the tumor cells from which this line was derived . Several germ cell markers such as GAGE2, GAGE7 and GAGE8 were highly expressed in NTera2 but were absent (or present at low levels) in any of the hESC lines examined (See Figure 6C and Additional file 1. Note that the GAGE genes are highly similar in sequence, making it difficult to distinguish one family member from another through hybridization; thus, while all of these GAGE gene probes gave positive signal, it is difficult to say if the signal came from the specific gene itself or from cross-hybridization from one of the other family members). None of these were present in BG01V, indicating that the karyotypically abnormal variant is not the equivalent of a teratocarcinoma line such as NTera2. In addition to the expression of germ cell markers, we noticed a significant difference in the expression of genes in the TGFβ pathway, such as GDF3 (Figure 6C), TGFBI, CDKN1A, IGFBP7, IGFBP3, NODAL, CER1 and BMP2 (Figure 6E). This is consistent with the postulated role of this pathway in germ cell differentiation and suggests that TGFβ pathway cannot be reliably tested using NTera2 as a model for hESC. The BG01V showed clear differences from its normal counterpart and some major changes are summarized in Figure 6D and 6F. Early markers of differentiation appeared to be present at higher levels in BG01V as compared to any of the hESC lines examined, although hESC specific genes continued to be expressed at high levels (see Additional file 4). In particular, the Wnt pathway and the TGFβ signaling pathway (Figure 6D), both of which involved in the early process of differentiation , appeared to be activated (Additional file 4), suggesting that the role of growth factors and signaling in these early events cannot be readily studied in this cell line. In summary, the analysis highlighted the utility of the potential reference standards NTera2 and BG01V, demonstrated their general similarity and provided detail on potential caveats to their application. We have utilized a small fraction of the data to demonstrate the overall utility of this approach and its sensitivity in identifying small differences in cell populations. An additional potential application of such an analysis is the ability to examine the general state of a particular signaling pathway and determine whether it is active. By comparing across many samples, a procedure previously expensive and difficult in terms of the RNA and replicate requirement, one can rapidly identify key regulatory pathways. To test whether we could use such multiple pairwise comparisons to elucidate the major regulatory pathways that may be required for hESC self-renewal, we examined several metabolic pathways. The results of the analysis of the insulin/insulin-like growth factor (IGF) signaling pathway are shown in Figure 7. Using the same 4 groups of samples as in Figure 1, we conducted PAM (Prediction Analysis of Microarray) , in search for biomarkers used in diagnostic identification of these four groups, ES, EB, NS, and FB. In PAM, a list of significant IGF pathway genes whose expression characterizes each diagnostic class was obtained. The average gene expression level in each class was divided by the within-class standard deviation. The nearest centroid classification computed took the gene expression profile from a new sample and compared it to each of these class centroids. For cross-validation of prediction results, multiple classification processes were performed on two data sets randomly constructed each time from the entire gene expression dataset. The first dataset, consisting of 70% of the total data, was used as the training dataset, and the other dataset, containing the remaining 30% of data, was used for the data prediction and verification process. The final biomarkers were determined in such a way that the misclassification error rate was minimal. The resulting graph (Figure 7) showed the shrunken class centroids for genes that had at least one nonzero difference in each diagnostic class. The genes with nonzero components in each class were almost mutually exclusive and represented candidate biomarkers for the diagnosis of each class. All data analyses were performed using the bioconductor package . Identification of diagnostic markers by PAM. The shrunken class centroids for genes which have at least one nonzero difference are shown. The genes with nonzero components in each class were almost mutually exclusive and were the candidate molecular markers for the diagnosis of the four groups of cell populations, including, (from left to right) hESC derived mesenchyme and human fibroblast feeder cells ("FB", n = 5), undifferentiated hESCs ("ES", n = 11), hESC derived neural cells ("NS", n = 3), and differentiated ES cells and EB, ("EB", n = 6). The identified biomarkers can be used to distinguish the four groups of cell populations. Undifferentiated hESCs have been analyzed by EST scan, MPSS, SAGE and microarray . The goal of these experiments including our own is to develop a low cost reliable method to assess multiple samples to generate a global database of markers and to provide a method of identifying core measures of similarities and differences across multiple laboratories. We and others have proposed three alternative methods of assessment: Quantitative RT-PCR , focused arrays or a large scale array with bioinformatics tools being utilized to focus on appropriate subsets of genes . Each of these methods has its advantages and disadvantages. The present results suggest that the global Illumina bead array retains the advantages of low cost per sample associated with focused arrays yet still has the strength of the global attributes of MPSS or EST scan while requiring much less RNA and turnaround time. To test this array format we examined samples from a variety of laboratories in a blinded fashion to determine whether the array was sufficiently sensitive and rapid for routine assessment. Duplicates using 100 ng of RNA were run and results obtained forty-eight hours later. The resolution was sufficient that ESC samples could be distinguished from one another and a variant karyotypically abnormal subclone could be distinguished from the parent population (correlation coefficient = 0.9043). Aliquots of the pooled ES and pooled EB samples, which we had prepared for MPSS, were included in this run to compare these two methods directly. The current analysis confirms that comparison across platforms is difficult and that only positive results can be treated with any reliability. The absence of expression cannot be readily interpreted. In particular, genes expressed at low levels (greater than 70% of all genes detected) should not be assessed in cross platform comparisons. The limited concordance at low levels raises a question as to how many genes are actually expressed by any one cell line and whether the cutoff of 3 tpm used for MPSS or 100 intensity units for bead arrays is a reasonable cutoff. We used 100 units for our analysis and we would suggest that readers exercise similar caution. Nevertheless even at this higher cutoff the arrays were remarkably sensitive and allowed us to readily distinguish between samples including cells grown in the same laboratory. The basis of the sensitivity could be attributed to a limited set of genes and those genes could be identified for future use. For example BG01V, while much more similar to BG01 than to any other cell type, could still be distinguished from a biological replicate of BG01 by the expression of a particular subset of differentiation markers (Figure 6). EC cells such as NTera2 could be distinguished from hESCs by the expression of germ cell markers and the presence of a partially inactivated TGFβ (BMP) signaling pathway (Figure 6). Distinguishing ESCs from EBs was relatively straightforward. We have confirmed the utility of previously identified markers for use in this platform as well as identified an additional set of markers that can serve as biomarkers to distinguish between the hESC and EB states. A subset of these markers have been used to develop a qPCR assay that shows such a high sensitivity that changes in cell behavior can be detected after as little as twenty-four hours and the development of EBs can be reliably staged . During the identification of ES and EB specific markers, we have noticed that some known hESC markers, such as Nanog, was not detected in all populations of hESCs that were included in this analysis. Several ESC-specific gene, including Lin41, Sox2 and FoxD3, were not detected in the array either (Table 3). We believe that the problem with Lin 41, Sox2 and FoxD3 is a technical one as we were able to confirm expression using alternate methods. We are in progress of redesigning appropriate probes for these genes. In the case of the gene Nanog, there are several pseudo genes in the genome for Nanog and it has been a major technical challenge designing primers or probes that are specific and sensitive. We believe that a partial explanation for the variability in Nanog expression is due to the lack of sensitivity to this gene. However, immunocytochemistry while not strictly quantitative shows similar variability when used to assess Nanog expression in different cell lines . This large comparison between samples allowed us to identify markers that distinguish human feeder cells from hESC. While we have listed 19 potential markers (Figure 3) and identified several hundred potential markers as shown in Additional file 5, we suggest that as few as 3–4 genes may be sufficient. Previously we found that as few as four were satisfactory to distinguish between hESCs and hEBs, which are two much more closely related samples . In this study we have confirmed by qPCR the differential expression of four genes, THBS1, MMP3, TNFRSF11B and KRTH4, to separate human fibroblast feeders and hESCs (Figure 3). Several markers such as MMP3 and TNFRSF11B have commercially available antibodies (R&D systems) that may be used to further confirm contamination of feeder cells by immunocytochemistry. Efforts to identify other useful antibodies based on these results continue . While we have focused on the immediate utility of the Illumina array platform, it is important to remember that this array provides a global snapshot of cell state and the data obtained can be readily compared in order to determine key signaling pathways. The ability to compare multiple samples in one run enhances data selectivity and reliability. To make such analysis more readily available, we utilized several software tools including the software package available through Illumina. The BeadStudio software provided with the BeadLab and BeadStudio genetic analysis systems for use with the bead array datasets provides a useful set of analytical and presentation tools that allow straightforward comparisons, which are sufficient for average users. For detailed analysis we recommend using more specific commercial tools or software packages developed by NCBI. In summary, the Illumina bead array has several key strengths including high throughput, low cost and high sensitivity. By using this array, we can readily detect contaminating feeders and spontaneous differentiation, differentiate male and female lines and distinguish between one undifferentiated population and another. Such a global analysis allows us to assess context dependent signaling and identify biomarkers of particular states of cells. Our future efforts will focus on data mining and developing better cross platform comparison tools and generating focused high throughput arrays for quality control in clinical and research settings. The hESC lines H1, H7 and H9 (WiCell, Madison, WI) were cultured on feeder layers derived from mitotically inactivated HS27 human fibroblast cells (HS27, ATCC), or mouse embryonic fibroblsts or under feeder-free conditions on Matrigel (BD, Franklin Lakes, NJ) coated plates for at least 10 passages. Culture medium for all cultures was composed of DMEM/F12-Glutamax 1:1, 20% Knockout Serum Replacement, 2 mM nonessential amino acids, 100 μM beta-mercaptoethanol, 50 μg/ml Pen-Strep (all from Invitrogen, Carlsbad, CA), and 4 ng/ml human recombinant basic fibroblast growth factor (bFGF/FGF2; PeproTech Inc., Rocky Hill, NJ.) Feeder-free cultures were prepared for gene expression analysis by manually harvesting individual colonies with uniform typical undifferentiated ESC morphology. BG01 (46, XY), BG02 (46, XY), BG03 (46, XX), I6 (46, XY) and BG01V (BG01 karyotypic variant: 49, XXY, +12, +17): Cells were maintained for 3 (BG01V), 7 (BG02), 8 (BG01), or 21 (BG03) passages under feeder-free condition on fibronectin-coated plates in medium that had been conditioned by mouse embryonic fibroblasts for 24 hours. Culture medium was DMEM/F12, 1:1 supplemented with 20% Knockout Serum Replacement, 2 mM non-essential amino acids, 2 mM L-glutamine, 50 μg/ml Pen-Strep, 100 μM beta-mercaptoethanol, and 4 ng/ml of bFGF. Different hESC lines were grown in slightly different culture conditions as described above. H lines were grown on Matrigel coated dishes, while BG lines on fibronectin treated dishes. These coating substrata supported the growth of hESCs similarly, as evaluated by colony morphology, immunocytochemistry and proliferation rate (data not shown). Embryoid bodies (EBs) were prepared from BG lines as described in . Cells were aggregated and cultured on non-adherent substrata for fourteen days. NTera2 cells were purchased from ATCC and cultured in parallel with hESCssamples using protocols described previously . HS27 embryonic human newborn foreskin cells (ATCC CRL-1634) were grown in DMEM with 10%FBS. All samples included in this study can be found in Additional file 6. RNA was isolated from cultured cells using the Qiagen RNEasy kit (Qiagen, Inc, Valencia, CA). Sample amplification was performed using 100 ng of total RNA as input material by the method of Van Gelder et al . Amplified RNA synthesized from limited quantities of heterogenous cDNA was performed using the Illumina RNA Amplification kit (Ambion, Inc., Austin, TX) following the Manufacturer instructions. Labeling was achieved by use of the incorporation of biotin-16-UTP (Perkin Elmer Life and Analytical Sciences, Boston, MA) present at a ratio of 1:1 with unlabeled UTP. Labeled, amplified material (700 ng per array) was hybridized to a pilot version of the Illumina HumanRef-8 BeadChip according to the Manufacturer's instructions (Illumina, Inc., San Diego, CA). Amersham fluorolink streptavidin-Cy3 (GE Healthcare Bio-Sciences, Little Chalfont, UK) following the BeadChip manual. Arrays were scanned with an Illumina Bead array Reader confocal scanner according to the Manufacturer's instructions. Array data processing and analysis was performed using Illumina BeadStudio software. Differentially expressed genes between ES and EB were identified by ANOVA at p value 0.05 using bioconductor . Unsupervised hierarchical clustering analysis and principal component analysis (PCA) were conducted using software Pcluster and TreeView . PAM (prediction analysis of microarray) was employed for the identification of diagnostic markers from insulin pathway genes by using the software package bioconductor . PAM is a class prediction method for expression data mining. It can provide a list of significant genes whose expression characterizes each diagnostic class. The average gene expression level in multiple classes, such as ES, EB, NS, and FB, was divided by the within-class standard deviation for that gene. The nearest centroid classification computed by PAM takes the protein expression profile from a new sample, and compares it to each of these class centroids . Total RNA was isolated with TRIzol (Invitrogen. cDNA was synthesized using 2.5 μg total RNA in a 20-μl reaction with Superscript II (Invitrogen) and oligo (dT)12–18 (Promega; Madison, WI). One microliter RNase H (Invitrogen) was added to each tube and incubated for 20 minutes at 37°C before proceeding to the RT-PCR analysis. The PCR primers are: RPS4Y-forward: 5' AGATTCTCTTCCGTCGCAG 3', RPS4Y-reverse, 5' CTCCACCAATCACCATACAC 3'; EIFAY-forward, 5' CTGCTGCATCTTAGTTCAGTC 3'; EIFAY-reverse 5' CTTCCAATCGTCCATTTCCC 3'. Quantitative real time PCR gene specific primer pairs and probes were purchased from Applied Biosystems (Foster City, CA) for the following genes: MMP3 (Hs00233962_m1), TFRSF11B (Hs00171068_m1), THBS1 (Hs00170236_m1), KRTHA4 (Hs00606019_gH), and for internal control β-actin (ACTB, Hs99999903_m1). YL contributed significantly in array data analysis and validation and drafted the manuscript. SS provided hESC samples and validated array results. XZ provided a large number of the hESC samples for this report and participated in manuscript drafting. HL performed bioinformatics study under the supervision of MZ who also participated in all discussions of the manuscript. RG and FJM isolated total RNA from the human ES cell line WA09 (H9) cell line grown on feeder (human HS27 fibroblast) and feeder-free conditions for > 20 passages at the Burnham Institute for Medical Research. These RNA samples were analyzed by Illuminas Bead Array under the supervision of JFL. CMS provided NTera2 sample, participated in analysis of this EC line and helped draft the manuscript. HX performed validation of the array results. SCB, EC, DLB, TKM and SO participated in the array design and analysis. MPM monitored the whole project. MSR conceived of the study, designed the whole project, coordinated the collaboration, supervised YL, SS, HX and CMZ, and helped draft the manuscript.
PMC12101583
Inhibitory effect of Endostar on HIF-1 with upregulation of MHC-I in lung cancer cells
Endostar is a human recombinant endostatin which is an attractive anti-angiogenesis protein. Because inefficient antigen presenting MHC class I expression (which can be downregulated by HIF-1) is an important strategy for cancer immune evasion, besides its anti-angiogenesis effect, it remains unclear whether Endostar has an inhibitory effect on HIF-1 expression by upregulating MHC class I expression in cancer cells to facilitate immunotherapies, including PD-1/PD-L1 inhibitors. In this study, A549 and NCI-H1299 lung cancer cells were treated with Endostar (6.25 μg/ml, 12.5 μg/ml, and 25 μg/ml, respectively). HIF-1 expression was detected by Immunocytochemistry and Western blot. Proteins of the MHC class I α-heavy chain and β2 m light chain, STAT3 and pSTAT3 were detected by Western blot. The mRNAs of MHC class I α-heavy chain and β2 m light chain were detected by RT-qPCR. It was shown that decreased expression of HIF-1 and promotion of β2-microglobulin were observed after Endostar treatment. In addition, elevated levels of MHC class I α-heavy chain mRNA and protein, as well as downregulation of STAT3 and pSTAT3, were also observed following Endostar treatment. Endostar inhibited HIF-1 expression in A549 and NCI-H1299 lung cancer cells, upregulated expression of MHC class I α-heavy chain and β2 m light chain, with the upregulation of STAT3 and pSTAT3, suggesting involvement of STAT3 pathway. It is important because only in combination with MHC class I on target cells can tumor antigenic peptides be recognized by CD8+ CTLs which destroy target cells. However, MHC class I is frequently deficient in cancer cells.For the last several decades, lung cancer has long been the most common cancer worldwide and the leading cause of death from cancer with poor survival and high fatality rate of this disease. The individual risk for development of lung cancer arises from several factors, including smoking and age, radon exposure, asbestos, and other harmful substance exposure, socioeconomic deprivation, earlier diagnosis of a malignant tumor, earlier diagnosis of respiratory conditions, such as chronic obstructive pulmonary disease (COPD), family history, as well as particular rare hereditary disorders, such as Li Fraumeni syndrome and the recently described non-syndromic association with germline EGFR mutation. Besides, inefficient immunosurveillance, which was caused by immune evasion strategy used by cancer including lung cancer, plays an important role. Cancer cells can be eliminated by the host immune system because, even though they are self origined, the cancer cells differ from their normal counterparts in terms of their biological behaviors, antigenic characteristics, and biochemical characteristics. Due to an inefficient DNA damage repair system, tumor-specific neoantigens arise and are expressed in cancer cells. The neoantigens bind to the human leukocyte antigen (HLA) system class I, which are the major histocompatibility class I (MHC-I) molecules in humans. The neoantigens binding to HLA-I are presented for CTLs. CTLs eliminate cancer cells through two main mechanisms: direct perforin-dependent destruction and secretion of inflammatory cytokines, such as tumor necrosis factor (TNF) and interferon (IFN) alpha (INF-α), which increase tumor immune sensitivity. Despite the host immune system’s capability）to recognize and destroy cancer cells, the host immune system frequently fails to control cancer growth because cancer cells have acquired the capability to evade recognition and destruction by immune system. Hypoxia is beneficial for malignant tumors to get rid of immune surveillance. In solid tumors, because of uncontrolled growth of cancer cells and disorganized neoangiogenesis, cellular oxygen availability is reduced, leading to intratumoral hypoxic conditions. In many types of tumor cells, hypoxia-inducible factor 1 (HIF-1) is highly expressed. HIF-1 signaling pathways play an important role in metabolic adaptation to hypoxia stress. HIF-1 signaling also induces immune checkpoint molecules and immunosuppressive factors to express, which suppresses innate and adaptive immune systems in order to escape from immune attack. MHC class I molecules, which are necessary for antigen-presentation, are also down-regulated by HIF-1, thus limiting T cells recognizing tumor cells. Human cancers of different histological origins have shown altered or complete loss of MHC class I, which subsequently affects the final outcome of immunotherapy. Thus, there is a clear need to improve tumor immunogenicity. The high expression of HIF-1 in tumor cells and its role in MHC-I down-regulation make it an important target in cancer immunotherapy. Exploring drugs that inhibit HIF-1 and upregulate MHC-I is an important objective. Endostar may be an important candidate of such drugs. Endostar is a human recombinant endostatin which is an attractive anti-angiogenesis protein. It has been approved by the State Food and Drug Administration of China (CFDA) because of its clinical application on non-small cell lung cancer (NSCLC). Nevertheless, besides its anti-angiogenesis, it remains unclear whether Endostar also inhibits HIF-1 expression by upregulating MHC class I expression in cancer cells, which is helpful for immunotherapies, including PD-1/PD-L1 inhibitor, to induce effective responses of cell-mediated immunity against cancers such as lung cancer. This study was to demonstrate this effect of Endostar using A549 and NCI-H1299 lung cancer cells. Human A549 and NCI-H1299 lung cancer cells were maintained in standard cell culture conditions. The cells whose density were 9 × 10/well were seeded in 6-well culture plates for 24 hours culture to settle. The temperature was 37 °C. The atmosphere was humidified and contained 5% CO2 in F-12K (A549 cells) or RPMI-1640 (NCI-H1299 cells) medium supplemented with 10% FBS. Then, the cells were precultured for 24 h in the medium containing 1% FBS, followed by the addition of Endostar (a Modified Recombinant Human Endostatin expressed and purified in E. coli, which was purchased from Simcere Pharmaceutical Research Co., Ltd., Nanjing, China) into the wells to the indicated final concentration (6.25 μg/ml, 12.5 μg/ml, and 25 μg/ml with 0 μg/ml as control) for a further 24 h incubation. Cells were collected for analysis. All experiments were individually carried out three times. This study was approved by the Ethics Committee of the Affiliated Hospital of Chengde Medical University in biomedicine (CYFYLL2021091). To overexpress HIF-1, the cells were transiently transfected with Human HIF1A ORF mammalian expression plasmid (Sino Biological Inc., Beijing, China) using Lipofectamine™ LTX Reagent with PLUS™ Reagent (Invitrogen, CA, USA), according to the manufacturer’s instructions. Forty-eight hours after transfection, the cells were harvested for assays. The Endostar treated and untreated cells grown on glass slides were washed and fixed using cold methanol for 5 min. Next, in order to inhibit the endogenous peroxidase activity, cells were treated with 3% hydrogen peroxide in methanol. Then, in order to avoid nonspecific binding, cells were blocked with 10% normal serum. After overnight incubation at 4 °C with goat polyclonal primary antibody against HIF-1 (Affinity biosciences, Suzhou, China) (at a 1:100 dilution), the cells were hybridized by corresponding secondary antibody for 1 h, and finalized with a diaminobenzidine solution to detect the target antigen. The nucleus of cells was counterstained by hematoxylin before mounting. A light microscope was used to examine the slides. After extraction of the total cellular proteins from the Endostar treated and untreated cells grown in 6-well culture plates by lysis with radioimmunoprecipitation assay (RIPA) buffer (Beijing Solarbio Science & Technology Co., Ltd., Beijing, China) and quantification with BCA Protein Assay Kit (Beijing Solarbio Science & Technology Co., Ltd., Beijing, China), 30 μg of the proteins, diluted in loading buffer and denaturized for 5 minutes at the temperature of 100 °C following keeping on ice for 10 minutes, were subjected to 10% (for HIF1, MHC-1, STAT3, and pSTAT3) or 12% (for β2 m) sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). The proteins were then electrotransferred onto polyvinylidene fluoride (PVDF) membrane, blocked for 1 hour with 5% skimmed milk at room temperature, before overnight probe at the temperature of 4 °C using relevant primary antibodies, polyclonal rabbit anti-HIF-1 antibody (Affinity biosciences, Suzhou, China), recombinat monoconal rabbit anti-β2 m antibody (HuaBio Inc., Cambridge, USA), recombinat monoconal rabbit anti-HLA-ABC antibody (HuaBio Inc., Cambridge, USA), recombinat monoconal rabbit anti-STAT3 and anti-pSTAT3 antibody (HuaBio Inc., Cambridge, USA), and mouse monoclonal anti-β-actin antibody (Proteintech, Rosemont, Illinois, USA) following 2-h incubation with horseradish peroxidase-conjugated secondary antibodies. The protein bands were visualized using enhanced chemiluminescence (ELC) detection system (Amersham, Arlington Heights, IL, USA). The expression of β-actin serves as an endogenous control. The acquired images were quantified with normalizative to β-actin using ImageJ software (Rasband, W.S., ImageJ, U.S. National Institutes of Health, Bethesda, Maryland, USA). Total RNAs in Endostar treated and untreated cells grown in 6-well culture were extracted by using Superbrilliant® 6 min High-quality RNA Extraction Kit (Zhongshi Gene Technology, Tianjin, China) under Rnase free condition according to the protocol of manufacturer. Fast Quant RTkit (Tiangen Biotech Co., Ltd., Beijing, China) was utilized to synthesize cDNAs. Quantitative real-time PCR (qPCR) was conducted on Roche Cobas z 480 Real-Time PCR Detection System (Roche, Basel, Switzerland) using SuperReal PreMix Plus kit (Tiangen Biotech Co., Ltd., Beijing, China), and the responsible primers, β2 m (Invitrogen, Carlsbad, USA), HLA-ABC α heavy chain (Invitrogen, Carlsbad, USA), and β-actin (Thermo Fisher Scientific Inc., Waltham, Massachusetts, USA), were used. The expression of β-actin mRNA as a control was used for normalization of the mRNAs expression with 2 method to calculate the fold changes of the mRNAs. The data were described as mean ± standard deviation (SD) and the statistical significance was determined by one-way ANOVA before LSD test for multiple comparisons and independent-sample t-test for comparison between the two groups. The level of p < .05 was used to confirm that the differences were statistically significant. It was shown with western blot assays that, in A549 lung cancer cells, treatment with Endostar for 24 h decreased the levels of HIF-1 protein in the groups treated with 25 μg/ml Endostar with statistical significance compared with the control group (p = .015), 6.25 μg/ml group (p = .014) and 12.5 μg/ml group (p = .005), and in NCI-H1299 lung cancer cells, treatment with Endostar decreased the levels of HIF-1 protein in the groups treated with 25 μg/ml Endostar with statistical significance compared with the control group (p = .004) and 12.5 μg/ml group (p = .047) (Figure 1 in detail), which was similar to immunocytochemistry showing the reduction of HIF-1 protein in Endostar treated A549 and NCI-H1299 lung cancer cells compared with the control cells (Figure 2). Figure 1.Inhibition by Endostar on HIF-1 in A549 and NCI-H1299 lung cancer cells measured by Western blot assay. After treatment with Endostar for 24 h, the HIF-1 in A549 and NCI-H1299 lung cancer cells was measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; Hashtags indicate p < .05, significantly different compared with the group treated with 6.25 μg/ml Endostar; ampersands indicate p < .05, significantly different compared with the group treated with 12.5 μg/ml Endostar; one-way analysis of variance followed by LSD test.A bar graph (labeled with “HIF-1 in A549 cells”) with four bars, the means of the arbitrary units of 1.03, 1.03, 1.06, and 0.88 with asterisk, hashtag, and ampersand being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the HIF-1 protein expression levels of three independent experiments after treatment of A549 lung cancer cells with Endostar for 24h; a bar graph (labeled with “HIF-1 in NCI-H1299 cells”) with four bars, the means of the arbitrary units of 1.34, 1.00, 1.05, and 0.64 with asterisk and ampersand being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the HIF-1 protein expression levels of three independent experiments after treatment of NCI-H1299 lung cancer cells with Endostar for 24 h. Corresponding protein bands with Western blot were representatively shown under each bar graph. Figure 2.Inhibition by Endostar on HIF-1 in A549 and NCI-H1299 lung cancer cells measured by immunocytochemistry. After treatment with Endostar for 24 h, the HIF-1 in A549 and NCI-H1299 lung cancer cells was measured by immunocytochemistry. Photographs were taken under microscope (400× magnification).A picture (entitled A549 Cells) with four photographs of A549 lung cancer cells taken under microscope after immunocytochemistry, the Endostar concentration of control, 6.25µg/ml, 12.5µg/ml, and 25µg/ml being under each photograph, respectively. The intensity of tan color in the cytoplasm, which is lighter and lighter with the increase of the Endostar concentration, displays the levels of HIF-1 after treatment with Endostar for 24 h; a picture (entitled NCI-H1299 Cells) with four photographs of NCI-H1299 lung cancer cells taken under microscope after immunocytochemistry, the Endostar concentration of control, 6.25µg/ml, 12.5µg/ml, and 25µg/ml being under each photograph, respectively. The intensity of tan color in the cytoplasm, which is lighter and lighter with the increase of the Endostar concentration, displays the levels of HIF-1 after treatment with Endostar for 24 h. Inhibition by Endostar on HIF-1 in A549 and NCI-H1299 lung cancer cells measured by Western blot assay. After treatment with Endostar for 24 h, the HIF-1 in A549 and NCI-H1299 lung cancer cells was measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; Hashtags indicate p < .05, significantly different compared with the group treated with 6.25 μg/ml Endostar; ampersands indicate p < .05, significantly different compared with the group treated with 12.5 μg/ml Endostar; one-way analysis of variance followed by LSD test. Inhibition by Endostar on HIF-1 in A549 and NCI-H1299 lung cancer cells measured by immunocytochemistry. After treatment with Endostar for 24 h, the HIF-1 in A549 and NCI-H1299 lung cancer cells was measured by immunocytochemistry. Photographs were taken under microscope (400× magnification). Beta 2-microglobulin (β2 m) is the invariable light chain of the MHC class I heterodimer molecules and associated with MHC class I down-regulation which is possibly caused by high level of HIF-1. It was shown in this study that with the reduction of the HIF-1 level, unlike the β2 m mRNA levels without statistically significant differences among the Endostar treated and untreated groups (p = .098) according to RT-qPCR assay (Figure 3 in detail), the higher relative levels of β2 m protein were found in the A549 cells in the group which were treated with 25 μg/ml Endostar with statistical significance compared with the control group, 6.25 μg/ml group, and 12.5 μg/ml group (p = .017, p = .003, p = .026, respectively), indicating that the effect of Endostar on the β2 m expression in A549 cells may be mRNA independent, and in the NCI-H1299 cells in the groups which were treated with 6.25 μg/ml, 12.5 μg/ml, and 25 μg/ml (p = .000, p = .000, p = .000, respectively) Endostar, with statistical significance compared with the control group (Figure 4 in detail). Figure 3.Effects of Endostar on β2-microglobulin mRNA in A549 lung cancer cells measured by RT-qPCR assay. After treatment with Endostar for 24 h, the β2-microglobulin mRNA in A549 lung cancer cells was measured by RT-qPCR assay. Error bars indicate the SD. No significant differences among the groups were found (p > .05); one-way analysis of variance.A bar graph (labeled with “β2m mRNA in A549 cells”) with four bars, the means of 2−ΔΔCt values of 1.00, 1.01, 0.92, and 0.98 being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the β2-microglobulin mRNA levels of three independent experiments after treatment of A549 lung cancer cells with Endostar for 24 h. Figure 4.Upregulation by Endostar on β2-microglobulin protein in A549 and NCI-H1299 lung cancer cells measured by Western blot assay. After treatment with Endostar for 24 h, the β2-microglobulin protein in A549 and NCI-H1299 lung cancer cells was measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; Hashtags indicate p < .05, significantly different compared with the group treated with 6.25 μg/ml Endostar; ampersands indicate p < .05, significantly different compared with the group treated with 12.5 μg/ml Endostar; one-way analysis of variance followed by LSD test.A bar graph (labeled with “β2m in A549 cells”) with four bars, the means of the arbitrary units of 0.96, 0.86, 0.98, and 1.19 with asterisk, hashtag, and ampersand being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the β2-microglobulin protein expression levels of three independent experiments after treatment of A549 lung cancer cells with Endostar for 24 h; a bar graph (labeled with “β2m in NCI-H1299 cells”) with four bars, the means of the arbitrary units of 0.68, 1.08 with asterisk, 1.05 with asterisk, and 1.18 with asterisk being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the β2-microglobulin protein expression levels of three independent experiments after treatment of NCI-H1299 lung cancer cells with Endostar for 24 h. Corresponding protein bands with Western blot were representatively shown under each bar graph. Effects of Endostar on β2-microglobulin mRNA in A549 lung cancer cells measured by RT-qPCR assay. After treatment with Endostar for 24 h, the β2-microglobulin mRNA in A549 lung cancer cells was measured by RT-qPCR assay. Error bars indicate the SD. No significant differences among the groups were found (p > .05); one-way analysis of variance. Upregulation by Endostar on β2-microglobulin protein in A549 and NCI-H1299 lung cancer cells measured by Western blot assay. After treatment with Endostar for 24 h, the β2-microglobulin protein in A549 and NCI-H1299 lung cancer cells was measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; Hashtags indicate p < .05, significantly different compared with the group treated with 6.25 μg/ml Endostar; ampersands indicate p < .05, significantly different compared with the group treated with 12.5 μg/ml Endostar; one-way analysis of variance followed by LSD test. The loss of MHC class I expression was probably caused by high level of HIF-1. After a treatment of A549 lung cancer cells with Endostar for 24 hours, the relative levels of MHC-I α heavy chain mRNA demonstrated a slight enhancement without statistical significance (p = .840) in the groups treated with 6.25 μg/ml, 12.5 μg/ml, and 25 μg/ml Endostar compared with the control group (Figure 5 in detail). The relative levels of MHC-I protein α heavy chain in the A549 cells in the groups treated with 6.25 μg/ml, 12.5 μg/ml, and 25 μg/ml Endostar were statistically enhanced (p = .048, p = .007, p = .002, respectively) compared with the control group, and in the NCI-H1299 cells in the groups treated with 6.25 μg/ml, 12.5 μg/ml and 25 μg/ml Endostar were statistically enhanced as well (p = .008, p = .007, p = .005, respectively) compared with the control group (Figure 6 in detail). Figure 5.Effects of Endostar on the mRNA of MHC class I α-heavy chain in A549 lung cancer cells measured by RT-qPCR assay. After treatment with Endostar for 24 h, the mRNA of MHC class I α-heavy chain in A549 lung cancer cells was measured by RT-qPCR assay. Error bars indicate the SD. Slight enhancement without statistical significance were found among the groups with Endostar treatment (p > .05); one-way analysis of variance.A bar graph with four bars, the means of 2−ΔΔCt values of 1.00, 1.03, 1.03, and 1.05 being over each bar, which represents the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the MHC class I α-heavy chain mRNA levels of three independent experiments after treatment of A549 lung cancer cells with Endostar for 24 h. Figure 6.Upregulation by Endostar on the protein of MHC class I α-heavy chain in A549 and NCI-H1299 lung cancer cells measured by Western blot assay. After treatment with Endostar for 24 h, the protein of MHC class I α-heavy chain in A549 and NCI-H1299 lung cancer cells was measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; one-way analysis of variance followed by LSD test.A bar graph (labeled with “HLA-I in A549 cells”) with four bars, the means of the arbitrary units of 0.86, 0.99 with asterisk, 1.05 with asterisk, and 1.10 with asterisk being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the MHC class I α-heavy chain protein expression levels of three independent experiments after treatment of A549 lung cancer cells with Endostar for 24 h; a bar graph (labeled with “HLA-I in NCI-H1299 cells”) with four bars, the means of the arbitrary units of 0.71, 1.07 with asterisk, 1.08 with asterisk, and 1.10 with asterisk being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the MHC class I α-heavy chain protein expression levels of three independent experiments after treatment of NCI-H1299 lung cancer cells with Endostar for 24 h. Corresponding protein bands by Western blot were representatively shown under each bar graphs. Effects of Endostar on the mRNA of MHC class I α-heavy chain in A549 lung cancer cells measured by RT-qPCR assay. After treatment with Endostar for 24 h, the mRNA of MHC class I α-heavy chain in A549 lung cancer cells was measured by RT-qPCR assay. Error bars indicate the SD. Slight enhancement without statistical significance were found among the groups with Endostar treatment (p > .05); one-way analysis of variance. Upregulation by Endostar on the protein of MHC class I α-heavy chain in A549 and NCI-H1299 lung cancer cells measured by Western blot assay. After treatment with Endostar for 24 h, the protein of MHC class I α-heavy chain in A549 and NCI-H1299 lung cancer cells was measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; one-way analysis of variance followed by LSD test. HIF-1 gene was transfected into A549 and NCI-H1299 cells to overexpress HIF-1. It was shown that the relative levels of β2 m (β light chain) and α heavy chain of MHC-I protein were statistically decreased in HIF-1-over-expressing A549 cells and NCI-H1299 cells compared with the NC control A549 cells and NCI-H1299 cells (p = .019, p = .041, p = .000, and p = .048), respectively (Figure 7 in detail). Figure 7.Downregulation by HIF-1 over-expression on β2-microglobulin and MHC class I α-heavy chain in A549 and NCI-H1299 lung cancer cells measured by Western blot assay. After transfection with over-expressing HIF-1 gen into the cells for 48 h, the proteins of β2-microglobulin and MHC class I α-heavy chain in A549 and NCI-H1299 lung cancer cells were measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; independent-sample t test.A bar graph with four bars (the first two bars were labeled with “β2m in A549 cells” and the last two bars were labeled with “β2m in NCI-H1299 cells”), the means of the arbitrary units of 1.12, 0.9 with asterisk, 1.3, and 0.74 with asterisk, being over each bar, which represent the NC transfection group and HIF-1 transfection group of A549 cells, and the NC transfection group and HIF-1 transfection group of NCI-H1299 cells, respectively, displays the β2-microglobulin protein expression levels of three independent experiments after transfection with over-expressing HIF-1 gen or NC gen into the cells for 48 h; a bar graph (the first two bars were labeled with “HLA-I in A549 cells” and the last two bars were labeled with “HLA-I in NCI-H1299 cells”) with four bars, the means of the arbitrary units of 1.11, 0.90 with asterisk, 1.06, and 0.91 with asterisk being over each bar, which represent the NC transfection group and HIF-1 transfection group of A549 cells, and the NC transfection group and HIF-1 transfection group of NCI-H1299 cells, respectively, displays the HLA-I α-heavy chain expression levels of three independent experiments after transfection with over-expressing HIF-1 gen or NC gen into the cells for 48 h. Corresponding protein bands by Western blot were representatively shown under each bar graph. Downregulation by HIF-1 over-expression on β2-microglobulin and MHC class I α-heavy chain in A549 and NCI-H1299 lung cancer cells measured by Western blot assay. After transfection with over-expressing HIF-1 gen into the cells for 48 h, the proteins of β2-microglobulin and MHC class I α-heavy chain in A549 and NCI-H1299 lung cancer cells were measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; independent-sample t test. MHC class I may be regulated by JAK2/STAT3 signaling pathway. It was shown that the relative levels of STAT3 in A549 cells in the groups treated with 6.25 μg/ml, 12.5 μg/ml, and 25 μg/ml Endostar were statistically decreased (p = .004, p = .008, p = .010, respectively) compared with the control group (Figure 8 in detail). STAT3 is activated by phosphorylation. After a treatment of A549 lung cancer cells with Endostar for 24 hours, the relative levels of pSTAT3 were statistically decreased with statistical significance in the groups treated with 12.5 μg/ml (p = .037) and 25 μg/ml (p = .003) Endostar compared with the control group (Figure 8 in detail). Figure 8.The downregulation of STAT3 and pSTAT3 proteins in A549 lung cancer cells by Endostar measured by Western blot assay.After treatment with endostar for 24 h, the protein of STAT3 and pSTAT3 in A549 lung cancer cells was measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; Hashtags indicate p < .05, significantly different compared with the group treated with 6.25 μg/ml Endostar; one-way analysis of variance followed by LSD test.A bar graph (labeled with “STAT3 in A549 cells”) with four bars, the means of the arbitrary units of 1.21, 0.92 with asterisk, 0.96 with asterisk, and 0.96 with asterisk being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the STAT3 protein expression levels of three independent experiments after treatment of A549 lung cancer cells with Endostar for 24 h; a bar graph (labeled with “pSTAT3 in A549 cells”) with four bars, the means of the arbitrary units of 1.28, 1.09, 0.97 with asterisk, and 0.74 with asterisk and hashtags being over each bar, which represent the control group, the 6.25µg/ml Endostar group, the 12.5µg/ml Endostar group, and the 25µg/ml Endostar group, respectively, displays the pSTAT3 protein expression levels of three independent experiments after treatment of A549 lung cancer cells with Endostar for 24 h. Corresponding protein bands by Western blot were representatively shown under each bar graph. The downregulation of STAT3 and pSTAT3 proteins in A549 lung cancer cells by Endostar measured by Western blot assay.After treatment with endostar for 24 h, the protein of STAT3 and pSTAT3 in A549 lung cancer cells was measured by Western blot assay. Error bars indicate the SD. Asterisks indicate p < .05, significantly different compared with the control; Hashtags indicate p < .05, significantly different compared with the group treated with 6.25 μg/ml Endostar; one-way analysis of variance followed by LSD test. The host immune system exerts immunosurveillance to control cancer development. By inducing the expression of immunosuppressive factors, however, tumor HIF-1 signaling subdues both the innate and adaptive immune responses, thereby shielding the tumor from immune attacks. In addition, insufficient expression of MHC class I on cancer cells is one of the important strategies employed by cancer to evade the immune system. Therefore, inhibition of HIF-1 and improvement of MHC class I expression on cancer cells is important to induce effective responses of cell-mediated immunity against cancer such as lung cancer for immunotherapies, including PD-1/PD-L1 inhibitor. The effect of Endostar in inhibiting HIF-1 and promoting MHC class I expression on cancer cells, such as lung cancer cells, may counteract cancer immune evasion and thereby benefit cancer immunotherapy. However, this remains unclear. This study was designed to determine the effect of Endostar in inhibiting HIF-1 and promoting MHC class I expression on lung cancer cells. Endostar was administrated to A549 and NCI-H1299 lung cancer cells and the protein of HIF-1 and MHC class I and their mRNAs was detected by western blot and RT-qPCR. The lower HIF-1 and higher MHC class I were found in Endostar treated A549 and NCI-H1299 cells. These findings demonstrated inhibtion on HIF-1 and promotion on MHC class I by Endostar, suggesting the potential of Endostar to benefit lung cancer immunotherapy. Endostar, an N-terminal modified recombinant human endostatin, is a human recombinant endostatin, an attractive anti-angiogenesis protein. Therefore, Endostar has anti-angiogenesis effects. Endostar was developed as a specific drug permitted by the State Food and Drug Administration in China in 2005 for its use in NSCLC therapy. Endostatin is a natural protein, first isolated and extracted from mouse tumors by Judah Folkman, with a wide antitumor spectrum and strong antiangiogenic capacity. Angiogenesis is a physiological process of forming new blood vessels from existing blood vessels and circulating endothelial precursors. It is essential to the occurrence and development of tumors, which grow rapidly and metastasize eventually. Physiologically, angiogenesis is essential to physiological processes like embryogenesis, tissue growth, and regeneration. Oncologically, it is also important for cancer cells that grow rapidly and metastasize eventually because angiogenesis supplies oxygen and nutrients which are deficient in cancer cells for their rapid growth and eventual metastasis. Therefore, anti-angiogenesis is one of the most important cancer therapies. Endostar as a recombinant human endostatin with nine added amino acids (MGGSHHHHH) to maintain stability and a long half-life has effective antiangiogenic effect. Besides, it was shown in this study that Endostar inhibited HIF-1 with upregulation of MHC class I expression in A549 and NCI-H1299 lung cancer cells, benefiting cancer cell killing by effectory lymphocytes. HIF-1, a heterodimer consisting of a constitutive β-subunit and an oxygen-sensitive α-subunit, is a main transcriptional regulator responsible for metabolic adaptation to alterations in the oxygen environment. It involves in many physiological and pathological processes in the body and is closely associated with the pathogenesis of many diseases. In solid tumors, uncontrolled proliferation of cancer cells vs disorganized growth of blood vessels results in limited supply of nutrients and oxygen resulting in low oxygen tension; therefore, regions with hypoxic microenvironments are created. In these regions, the highly overexpressed HIF-1 is important to drive tumor growing, invasion, and metastasis in different human cancers. The association of a poor survival with the high expression of HIF-1α has been indicated by survival analysis in patients with lung cancer, and different SNPs in HIF-1α may have different effects on overall cancer risk in an ethnicity- and type-specific manner. Reduction of HIF-1α by Simvastatin enhanced Anti-tumor Effects of Bevacizumab in A549 cells. Targeting ATM/HIF-1α signaling by solanidine induced anti-angiogenesis and anti-cancer effect in lung cancer. Besides its involvement in various aspects of tumor development, such as tumor growth, invasion, metastasis, and angiogenesis, HIF-1 is also involved in tumor immune evasion which facilitates cancer cells to proliferate and metastasize, and contributes to failure in immunotherapy. Inhibition of HIF-1 expression in cancer cells contributes to cancer control and induction of effective anti-cancer immunity in cancer immunotherapy. Endostar has the effect to down-regulate HIF-1. It was shown in this study that 25 μg/ml Endostar inhibited HIF-1 expression in A549 and NCI-H1299 lung cancer cells. HIF-1 is induced by hypoxia in cancer cells, and in normoxia, it is not usually observed or only basal expression can be observed. In this study, experiments showed a high expression of HIF-1 in control and untreated cells. However, this did not necessarily mean that the expression of HIF-1 in control and untreated cells was really high because it was detected with the expression in Endostar treated A549 cells and NCI-H1299 cells as the backgroud. The expression detected by Western blot and Immunocytochemistry is relative quantification. The assays had been optimized for enough sensitivity to detect the reduced expression in Endostar treated A549 cells and NCI-H1299 cells. An effective adaptive response can be achieved through a multi-step antigen processing and pathway, namely the cellular antigen processing machinery (APM). Antigens must be processed into antigenic peptides by APM. These peptides are loaded onto an MHC class I molecule. MHC class I molecules are glycoproteins of heterodimers with a polymorphic heavy chain (α-chain) and an invariable β2 microglobulin (β2 m) light chain (β-chain). The α-chain is encoded by the Human Leukocyte Antigen-HLA A, B, and C genes in humans. There is a groove in the MHC class I molecules to preferentially bind 8-11mer peptides. The antigenic epitope binds to the exposed surface of this groove as a part of the MHC class I complex. It is recognized by the TCR on the T cells. This study showed that the expression of MHC class I α-heavy chain and β2 m light chain in A549 and NCI-H1299 lung cancer cells was improved by Endostar, which benefited cancer immunotherapy against the lung cancer cells. In the context of HIF-1 down-regulation by Endostar, which was shown in this study, to demonstrate the role of HIF-1 on the MHC class I α-heavy chain and β2 m light chain in A549 and NCI-H1299 lung cancer cells, the over-expressing HIF-1 gene was transfected into A549 and NCI-H1299 cells resulting in decrease of relative levels of MHC class I α-heavy chain and β2 m light chain. This results were in line with the decrease of HIF-1 by Endostar treatment accompanied with the enhancement of MHC class I α-heavy chain and β2 m light chain. In addition to the important role in metabolic adaptation to hypoxia stress caused by deficient supply of nutrients and oxygen because of uncontrolled growth of cancer cells and disorganized neoangiogenesis, the signaling pathways of HIF-1, a heterodimer highly expressed in a variety of tumor cells, suppress innate and adaptive immune systems to escape immune attack. That HIF-1 downregulates the antigen presenting MHC class I molecules is an important strategy for cancer to evade immune attack, because only in combination with MHC class I on the target cells, can tumor antigenic peptides be recognized by CD8+ CTL with the subsequent destruction of the target cells. Theoretically, inhibition of HIF-1 may upregulate the expression of MHC class I on cancer cells. This study demonstrated that 25 μg/ml Endostar inhibited expression of HIF-1 with the upregulation of MHC class I α-heavy chain and β2 m light chain in A549 and NCI-H1299 lung cancer cells, suggesting the potential for Endostar to facilitate cancer immunotherapy. Experimental evidence, however, is required and further research with experimental evidence remains to be performed in the future to confirm the possible suppression of immune evasion tactic with endostar treatment. Other limitations such as lack of an in vivo tumor model and further validation using flow cytometry remain to be supplementarily performed in the future. The mechanism of MHC class I regulation involves the innate immune molecule NLR family CARD domain containing 5 (NLRC5), which is a member of recently discovered NLRs-like receptor family of the highly conserved one. NLRC5 is an MHC class I gene transactivation factor which induces the MHC class I gene transcription and subsequently activates antigen presentation process. NLRC5 is transcriptionally regulated in JAK2/STAT3 signaling-dependent pathway. This study demonstrated that some concentrations of Endostar decreased the relative levels of STAT3 and pSTAT3 in A549 lung cancer cells with statistical significance, which is in line with the role of JAK2/STAT3 pathway in regulation of MHC class I via NLRC5, suggesting the underlying mechanism of MHC class I upregulation involving JAK2/STAT3 signaling pathway. Endostar (25 μg/ml) inhibited the expression of HIF-1 with the upregulation of MHC class I α-heavy chain and β2 m light chain in A549 and NCI-H1299 lung cancer cells, which showed the potential for Endostar to facilitate cancer immunotherapy. It is needed for future studies that are warranted to confirm the role of Endostar treatment.
PMC12612255
YTHDF2 regulates self non-coding RNA metabolism to control inflammation and tumorigenesis
The role of mA RNA methylation of self non-coding RNA remains poorly understood. Here we show that mA-methylated self U6 snRNA is recognized by YTHDF2 to reduce its stability and prevent its binding to Toll-like receptor 3 (TLR3), leading to decreased inflammatory responses in human and mouse cells and mouse models. At the molecular level, endosomal U6 snRNA binds to the LRR21 domain in TLR3, independent of mA methylation, to activate inflammatory gene expression, a mechanism that is distinct from that of the best known synthetic TLR3 agonist poly I:C. Both U6 snRNA and YTHDF2 are localized to endosomes via the transmembrane protein SIDT2, where YTHDF2 functions to prevent the U6-TLR3 interaction. We further show that UVB exposure inhibits YTHDF2 by inducing its dephosphorylation and autophagic protein degradation in human keratinocytes and mouse skin. Skin-specific deletion of Ythdf2 in mice enhanced the UVB-induced skin inflammatory response and promoted tumor initiation. Taken together, our findings demonstrate that YTHDF2 plays a crucial role in controlling inflammation by inhibiting mA U6-mediated TLR3 activation, suggesting that YTHDF2 and mA U6 are potential therapeutic targets for preventing and treating inflammation and tumorigenesis.N-methyladenosine (mA) RNA methylation is the most prevalent internal modification that occurs in messenger RNA (mRNA) and long non-coding RNA (lncRNA) of most eukaryotes. mA mRNA methylation regulates several aspects of RNA metabolism, including RNA decay, nuclear processing, translation, transcription, and RNA-protein interactions. At the molecular level, mA RNA modification is installed by writer complexes composed of factors including METTL3, METTL14, WTAP, and KIAA1429, or METTL16, and is removed by erasers FTO or ALKBH5. mA is recognized by mA-binding proteins including YTHDF1-3, YTHDC1, YTHDC1-2, and IGF2BP1-3, also known as mA readers, to regulate RNA fate. Among the mA writers, METTL16 is monomeric, and is distinct from METTL3/METTL14, which is an obligate heterodimer. The METTL16 ortholog mett-10 in C. elegans has been shown to deposit mA on SAM synthase to inhibit its proper splicing. Recent studies have demonstrated critical roles for METTL16 in the pathogenesis of leukemia and liver cancer in both mA-dependent and -independent mechanisms. However, the function of METTL16 remains incompletely understood. One mA-modified non-coding RNA is the small nuclear RNA (snRNA) U6. U6 snRNA is a non-coding RNA best known for its role in splicing. U6 interacts with three snRNAs, pre-mRNA substrates, and more than 25 protein partners to form the catalytic core of the spliceosome during splicing. Although commonly used as an internal standard to quantify the level of miRNA, U6 snRNA was recently recognized as a highly variably expressed gene in various human tissues including carcinoma tissues. Notably, U6 snRNA levels are significantly higher in human carcinoma tissue than in the corresponding normal tissue. Newly synthesized U6 appears transiently in the cytoplasm and undergoes maturation where it is accompanied by U6-associated proteins known as small nuclear ribonucleoprotein complexes, snRNPs, before returning to the nucleus. Among the U6 snRNPs, loss of LSM6 or LSM7 induces a cytosolic accumulation of U6 snRNA. Recently, U6 has been shown to be mA-modified by METTL16 at A43. In Schizosaccharomyces pombe, loss of the Mettl16 ortholog Mtl16 alters global splicing. In contrast, in mouse embryos, Mettl16 deletion had little effect on global splicing. These contrasting functions suggest species-specific roles for METTL16 in splicing. Altogether, the functional role of mA methylation on U6 snRNA in mammals as well as its reader remains unknown. Emerging evidence has demonstrated that inflammation, originally recognized for its pivotal role in pathogen defense, also plays critical roles in a number of diseases including cancer and autoimmune diseases, both of which can be induced or triggered by environmental factors such as UV radiation. However, the molecular mechanisms that regulate inflammation remain incompletely understood. Here we show that YTHDF2 recognizes mA-modified U6 snRNA to regulate U6 stability and binds to TLR3 in the context of inflammation and tumorigenesis, highlighting the crucial role of YTHDF2 and U6 mA methylation in controlling inflammation. Recently, we have shown that the mA reader YTHDF1 regulates the repair of genome damage caused by UVB stress. However, the role of other mA readers, including YTHDF2, in stress responses remains poorly understood. To determine the functional role of YTHDF2, we performed RNA-seq in HaCaT cells, non-tumorigenic human keratinocytes, to identify pathways affected by YTHDF2 loss. Our analysis showed that YTHDF2 knockdown upregulates genes in several pathways, including the TNF and IL-17 signaling pathway, signaling by interleukins, TLR3 cascade, and antiviral pathways (Fig. 1A, Supplementary Fig. S1A), suggesting that YTHDF2 may act as a regulator of inflammation. Next, to determine whether YTHDF2 has a role in the UVB stress response, we performed mass spectrometric analysis to identify pathways associated with YTHDF2-interacting proteins in HaCaT cells (Supplementary Table S1). We compared these pathways with pathways of genes upregulated by UVB irradiation from our previous work in HaCaT cells (GSE145924). Analyzing both data sets, we found that genes up-regulated by UVB share several pathways with YTHDF2-interacting proteins, including metabolism of RNA and ribonucleoprotein complex biogenesis (Fig. 1B, Supplementary Fig. S1B), suggesting that YTHDF2 may play an important role in the UVB-induced stress response.Fig. 1YTHDF2 controls inflammatory gene expression.A Pathway enrichment of genes upregulated upon YTHDF2 knockdown in HaCaT cells (RNA-seq, q < 0.05, DESeq2). B Overlap between the top 20 UVB-induced pathways (RNA-seq) and the top 10 pathways of YTHDF2-interacting proteins (mass spectrometry, 1 h post-sham or -UVB). C Representative H&E staining of skin sections 24 h after sham or UVB exposure in WT (YTHDF2) and DF2 cKO (K14Cre; YTHDF2) mice. Scale bar, 200 μm. n = 3 mice per group. D Quantification of epidermal thickness (n = 3). E CD45 cell counts in skin of sham- and UVB-irradiated WT and DF2 cKO mice (n = 5). F Box plots of YTHDF2 expression across systemic lupus erythematosus (SLE) datasets. Centre line, median; box, interquartile range; whiskers, minimum to maximum. Statistical analyses were conducted using a two-tailed unpaired Student’s t-test, with P values indicated (ns, P > 0.05). G Immunoblot of COX-2 and YTHDF2 in HaCaT cells with or without YTHDF2 knockdown. H–K qPCR analysis of TNF-α, IL-6, COX-2 and IL-1β mRNAs in HaCaT cells transduced with shNC (short hairpin RNA negative control), shDF2-1 or shDF2-2 (short hairpin RNAs targeting YTHDF2), with or without UVB. L. Immunoblot analysis of YTHDF2 and COX-2 in HaCaT cells with or without YTHDF2 knockdown, with or without UVB. M–Q qPCR analysis of YTHDF2, TNF-α, IL-6, COX-2 and IL-1β in NHEK cells transfected with siNC (small interfering RNA negative control) or siDF2 (siRNA targeting YTHDF2), with or without UVB. R Immunoblot of YTHDF2 in HaCaT cells expressing EV (empty vector) or OE DF2 (YTHDF2 overexpression), with or without UVB. S–U. qPCR analysis of IL-1β, IL 6 and TNF-α in cells as in R. V Immunoblot of COX-2 in cells as in R. β-actin served as internal control for qPCR (H–K, M–Q, S–U). Statistical analyses were conducted using a two-tailed unpaired Student’s t-test, with P values indicated (ns, P > 0.05). Data are shown as mean ± SE (n = 5 for E; n = 4 for H–K, M–Q, S–U) or mean ± SD (n = 3 for D). All experiments were conducted using biologically independent samples. A Pathway enrichment of genes upregulated upon YTHDF2 knockdown in HaCaT cells (RNA-seq, q < 0.05, DESeq2). B Overlap between the top 20 UVB-induced pathways (RNA-seq) and the top 10 pathways of YTHDF2-interacting proteins (mass spectrometry, 1 h post-sham or -UVB). C Representative H&E staining of skin sections 24 h after sham or UVB exposure in WT (YTHDF2) and DF2 cKO (K14Cre; YTHDF2) mice. Scale bar, 200 μm. n = 3 mice per group. D Quantification of epidermal thickness (n = 3). E CD45 cell counts in skin of sham- and UVB-irradiated WT and DF2 cKO mice (n = 5). F Box plots of YTHDF2 expression across systemic lupus erythematosus (SLE) datasets. Centre line, median; box, interquartile range; whiskers, minimum to maximum. Statistical analyses were conducted using a two-tailed unpaired Student’s t-test, with P values indicated (ns, P > 0.05). G Immunoblot of COX-2 and YTHDF2 in HaCaT cells with or without YTHDF2 knockdown. H–K qPCR analysis of TNF-α, IL-6, COX-2 and IL-1β mRNAs in HaCaT cells transduced with shNC (short hairpin RNA negative control), shDF2-1 or shDF2-2 (short hairpin RNAs targeting YTHDF2), with or without UVB. L. Immunoblot analysis of YTHDF2 and COX-2 in HaCaT cells with or without YTHDF2 knockdown, with or without UVB. M–Q qPCR analysis of YTHDF2, TNF-α, IL-6, COX-2 and IL-1β in NHEK cells transfected with siNC (small interfering RNA negative control) or siDF2 (siRNA targeting YTHDF2), with or without UVB. R Immunoblot of YTHDF2 in HaCaT cells expressing EV (empty vector) or OE DF2 (YTHDF2 overexpression), with or without UVB. S–U. qPCR analysis of IL-1β, IL 6 and TNF-α in cells as in R. V Immunoblot of COX-2 in cells as in R. β-actin served as internal control for qPCR (H–K, M–Q, S–U). Statistical analyses were conducted using a two-tailed unpaired Student’s t-test, with P values indicated (ns, P > 0.05). Data are shown as mean ± SE (n = 5 for E; n = 4 for H–K, M–Q, S–U) or mean ± SD (n = 3 for D). All experiments were conducted using biologically independent samples. UVB damage induces inflammation, which clinically presents as sunburn. To determine whether YTHDF2 regulates UVB-induced inflammation in vivo, we generated a mouse model with skin-specific conditional knockout of Ythdf2 (DF2 cKO) and assessed differences in UVB-induced histological alteration between wild-type (WT) and DF2 cKO mice. We noted that skin-specific YTHDF2 deletion increased epidermal thickness upon sham or UVB irradiation (Fig. 1C, D), suggesting that YTHDF2 loss enhances UVB-induced inflammation. Furthermore, flow cytometric analysis showed that skin-specific Ythdf2 deletion specifically increased the number of CD45 cells and TCRγδ T cells following UVB irradiation in the skin (Fig. 1E, Supplementary Fig. S2A, B); however, it did not significantly alter the number of other immune cells analyzed (Supplementary Fig. S2B–D). Next, we analyzed the effect of skin-specific Ythdf2 deletion on the systemic immune system in the spleen, blood, and lymph node (LN). Skin-specific Ythdf2 deletion did not significantly alter spleen weight or immune cell counts in the spleen or LN (Supplementary Fig. S3A–C). It is worthwhile to note that skin-specific Ythdf2 deletion increased the number of CD11b myeloid cells in both the blood and spleen, PMN-MDSCs and Ly6C high cells in the blood, and M-MDSC cells in the spleen, while it decreased Ly6C low cells in the blood (Supplementary Fig. S4A–C). These findings indicate that skin Ythdf2 deletion augments skin inflammation and likely alters the systemic inflammatory response following UVB irradiation. Given the role of YTHDF2 in modulating skin and systemic inflammation, we hypothesized that its expression may be altered in autoimmune disorders, where inflammation plays a central role in disease progression. Indeed, YTHDF2 expression was decreased in systemic lupus erythematosus (SLE) and type I diabetes as compared with healthy controls (Fig. 1F, Supplementary Fig. S5A). In contrast, we did not observe consistent changes in other autoimmune diseases (Supplementary Fig. S5B–D). These findings implicate that YTHDF2 may play an important, yet selective, role in autoimmune diseases, likely associated with its role in inflammation. Further investigations are warranted to define the role of YTHDF2 in the pathogenesis of SLE and type I diabetes. Prompted by the observed role of YTHDF2 in skin inflammation in vivo, we hypothesized that keratinocyte-intrinsic YTHDF2 controls inflammatory gene expression and suppresses UVB-induced inflammation. Indeed, we found that knockdown of YTHDF2 increased the expression of inflammatory mediators such as TNF-α, IL-6, COX-2/PTGS2, and IL-1β in both non-tumorigenic human HaCaT keratinocytes and normal human epidermal keratinocytes (NHEK) following UVB irradiation; it also increased the baseline expression of multiple inflammatory genes in A431 skin cancer cells (Fig. 1G–Q and Supplementary Fig. S6A–C). Furthermore, we found that skin-specific Ythdf2 deletion increased COX-2 expression upon sham or UVB irradiation (Supplementary Fig. S6D), supporting our findings in human cells that YTHDF2 loss enhances UVB-induced inflammatory gene expression. Conversely, forced overexpression of YTHDF2 inhibited UVB-induced expression of these inflammatory mediators (Fig. 1R–V). As an mRNA mA reader, YTHDF2 destabilizes mA-containing mRNAs. Intriguingly, although YTHDF2 knockdown increased mRNA levels of pro-inflammatory genes, it had no effect on the mRNA stability of IL-8, TGFβ, VEGF, COX-2, GM-CSF, IL-6, IL-1α, or IL-1β, while it decreased the mRNA stability of TNF-α and IL-16 in HaCaT cells (Supplementary Fig. S6E). These findings suggest that YTHDF2 loss may not directly inhibit mRNA decay of pro-inflammatory genes, but rather may indirectly enhance pro-inflammatory gene expression by activating upstream pathways that promote the expression of these genes. To determine the molecular mechanism by which YTHDF2 regulates pro-inflammatory gene expression, we analyzed mRNA mA modification levels in polyadenylated RNAs using ultra-high-performance liquid chromatography-tandem mass spectrometry (UHPLC-MS/MS). Indeed, loss of YTHDF2 increased mRNA mA enrichment in HaCaT cells (Fig. 2A). Next, we performed mA-seq to identify potential transcriptome-wide mA-modified gene targets for YTHDF2. YTHDF2 knockdown had little effect on mA enrichment in the 5′UTR, 3′UTR, CDS, or the total peak distribution in HaCaT cells (Fig. 2B, C). Sequence motif analysis of mA peaks showed enrichment for the previously identified mA target site motif (GGACU) (Fig. 2D). Overall, the levels of over 3800 genes were changed in cells with knockdown of YTHDF2 (Supplementary Fig. 7A). The downstream targets of YTHDF2 were selected following 3 criteria in YTHDF2 knockdown cells as compared with control cells: (1) increased RNA level, (2) increased mA enrichment, and (3) related to inflammation or tumorigenesis. From this analysis, we identified the top 30 upregulated genes that are potential mA targets of YTHDF2 (Supplementary Fig. S7B), consistent with YTHDF2’s known role in promoting decay of mA-modified mRNA. Finally, we selected and verified four genes – FOS, JUN, SOX4, and SOX9 – as potential YTHDF2 targets, based on increased mRNA levels, stability, and increased mA enrichment upon YTHDF2 knockdown, and as well as YTHDF2 binding to the mRNAs in HaCaT cells (Supplementary Fig. S7C–F, S8A, B). These findings indicate that FOS, JUN, SOX4, and SOX9 are downstream mA targets for YTHDF2. Further investigation is needed to elucidate the functional importance of these genes in YTHDF2’s function. Interestingly, YTHDF2 knockdown had little effect on the mA enrichment across the transcripts for TNF-α and IL-1β, while it increased the mA enrichment across the transcripts for IL-6 and COX-2 in HaCaT cells (Supplementary Fig. S9A–D). Since YTHDF2 knockdown had no effect on the mRNA stability of COX-2 or IL-6 (Supplementary Fig. S6E), future investigation is warranted to elucidate the role of mA methylation of COX-2 and IL-6 in their gene expression, including but not limited to translation.Fig. 2YTHDF2 binds to mA-methylated self U6 snRNA to induce U6 decay.A UHPLC–MS/MS for mA enrichment in mRNA in HaCaT cells with or without YTHDF2 knockdown (n = 2). No statistical analysis was performed. B Distribution of mA peaks across 5′UTR, CDS and 3′UTR. C Proportion of mA peaks in the indicated regions across all mRNAs. D Consensus mA motif identified by HOMER. E Pathway analysis of UVB-induced upregulated genes in HaCaT cells (q < 0.05, DESeq2). F Identification of U5-, U4- and U6-associated RNPs among YTHDF2-interacting proteins by mass spectrometry in HaCaT cells. G–I qPCR of U6 snRNA in HaCaT cells with or without YTHDF2 knockdown, with or without YTHDF2 overexpression, and with or without UVB irradiation. J qPCR of U6 snRNA in mouse skin with or without YTHDF2 deletion after sham or chronic UVB for 23 weeks (n = 3 mice). K mA-IP qPCR for enrichment of U6 snRNA in HaCaT cells with or without YTHDF2 knockdown. L RIP of YTHDF2–U6 snRNA interaction in HaCaT cells. M mA-IP qPCR of U6 snRNA in HaCaT cells with or without UVB (30 min). N RIP of YTHDF2–U6 snRNA interaction in HaCaT cells with or without UVB (30 min). O qPCR of U6 snRNA stability in HaCaT cells with or without YTHDF2 knockdown, treated with actinomycin D (ActD), with or without UVB (4 h time course). P qPCR of U6 snRNA in HaCaT cells with METTL16 knockdown (shNC, shM16-1, shM16-2). Q mA-IP qPCR of U6 snRNA in HaCaT cells with or without METTL16 knockdown, with or without UVB (30 min). R RIP of YTHDF2–U6 snRNA interaction in HaCaT cells with or without METTL16 knockdown, with or without UVB (30 min). 18S rRNA served as internal control (G–J, O–P). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are presented as mean ± SE (n = 4 for G–I, O–P) or mean ± SD (n = 2 for A; n = 3 for J–N, Q–R). All experiments were conducted using biologically independent samples. A UHPLC–MS/MS for mA enrichment in mRNA in HaCaT cells with or without YTHDF2 knockdown (n = 2). No statistical analysis was performed. B Distribution of mA peaks across 5′UTR, CDS and 3′UTR. C Proportion of mA peaks in the indicated regions across all mRNAs. D Consensus mA motif identified by HOMER. E Pathway analysis of UVB-induced upregulated genes in HaCaT cells (q < 0.05, DESeq2). F Identification of U5-, U4- and U6-associated RNPs among YTHDF2-interacting proteins by mass spectrometry in HaCaT cells. G–I qPCR of U6 snRNA in HaCaT cells with or without YTHDF2 knockdown, with or without YTHDF2 overexpression, and with or without UVB irradiation. J qPCR of U6 snRNA in mouse skin with or without YTHDF2 deletion after sham or chronic UVB for 23 weeks (n = 3 mice). K mA-IP qPCR for enrichment of U6 snRNA in HaCaT cells with or without YTHDF2 knockdown. L RIP of YTHDF2–U6 snRNA interaction in HaCaT cells. M mA-IP qPCR of U6 snRNA in HaCaT cells with or without UVB (30 min). N RIP of YTHDF2–U6 snRNA interaction in HaCaT cells with or without UVB (30 min). O qPCR of U6 snRNA stability in HaCaT cells with or without YTHDF2 knockdown, treated with actinomycin D (ActD), with or without UVB (4 h time course). P qPCR of U6 snRNA in HaCaT cells with METTL16 knockdown (shNC, shM16-1, shM16-2). Q mA-IP qPCR of U6 snRNA in HaCaT cells with or without METTL16 knockdown, with or without UVB (30 min). R RIP of YTHDF2–U6 snRNA interaction in HaCaT cells with or without METTL16 knockdown, with or without UVB (30 min). 18S rRNA served as internal control (G–J, O–P). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are presented as mean ± SE (n = 4 for G–I, O–P) or mean ± SD (n = 2 for A; n = 3 for J–N, Q–R). All experiments were conducted using biologically independent samples. As we did not observe canonical inflammatory genes in the candidate target list for YTHDF2 by mA-seq, we elected to re-examine the pathways for UVB-induced genes from our previous RNA-seq data (GSE145924). Indeed, UVB irradiation induced the expression of genes related to (non-coding RNA) ncRNA metabolic processes and ncRNA processing in HaCaT cells (Fig. 2E), suggesting a role for ncRNA in YTHDF2 regulation of the UVB stress response. In addition, our mass spectrometric analysis of YTHDF2-interacting proteins showed that YTHDF2 binds to several U6 snRNA-associated small nuclear ribonucleoproteins (U6 snRNPs) such as U6 snRNA-associated Sm-like protein LSM4, LSM6, U4/U6 small nuclear ribonucleoprotein Prp4 (PRPF4), U4/U6 small nuclear ribonucleoprotein Prp3 (PRPF3), and U4/U6.U5 tri-snRNP-associated protein 1/2 (SNUT1/2), as well as several U5-associated ribonucleoproteins in HaCaT cells (Fig. 2F, Supplementary Fig. S10A). These findings led us to hypothesize that YTHDF2 acts as a reader for mA-methylated self U6 snRNA. Indeed, we found that YTHDF2 knockdown increased U6 snRNA levels in HaCaT cells (Fig. 2G). Similarly, UVB irradiation also increased U6 levels in HaCaT cells by qPCR and Fluorescence In-Situ Hybridization (FISH) analysis (Fig. 2H-I, Supplementary Fig. S10B). Both baseline and UVB-induced U6 levels were decreased by YTHDF2 overexpression, while they were further increased by YTHDF2 knockdown in HaCaT cells (Fig. 2H-I). Skin-specific Ythdf2 deletion in mice increased U6 levels in the skin, which were further increased by UVB irradiation (Fig. 2J). Moreover, YTHDF2 knockdown increased mA enrichment in U6 snRNA in HaCaT cells and primary mouse keratinocytes (Fig. 2K, Supplementary Fig. S10C). Furthermore, using RNA immunoprecipitation (RIP) analysis, we found that YTHDF2 binds to U6 snRNA in HaCaT cells (Fig. 2L). At 30 min in HaCaT cells, UVB irradiation increased mA enrichment in U6 snRNA (Fig. 2M) and enhanced the interaction of U6 with YTHDF2 (Fig. 2N). We then hypothesized that YTHDF2 regulates U6 stability. Indeed, both YTHDF2 knockdown and UVB stress increased U6 snRNA stability in HaCaT cells (Fig. 2O). Notably, the combination of UVB irradiation and YTHDF2 knockdown had a similar effect on U6 stability as either UVB irradiation or YTHDF2 knockdown alone (Fig. 2O), suggesting that UVB irradiation increases U6 stability through inhibiting YTHDF2. Notably, the half-life of U6 in control cells is approximately 4 h, substantially shorter than the ~24 h half-life reported previously. It is possible that actinomycin D (ActD) is not specific for RNA Pol III, as it inhibits all three eukaryotic polymerases (I, II, and III) by binding to DNA. Therefore, the shorter half-life observed in our data suggests that the effect of ActD on other RNA polymerases might influence U6 levels and stability, as other RNA polymerases may modulate upstream regulators involved in U6 transcription, maturation, and/or stability. To determine whether binding of YTHDF2 to U6 snRNA is dependent on mA methylation, we assessed the effect of knocking down the known mA methyltransferase METTL16 on U6 snRNA levels. METTL16 knockdown increased U6 levels in HaCaT cells (Fig. 2P, Supplementary Fig. S10D), while it decreased both baseline and UVB-induced mA enrichment of U6 in HaCaT cells (Fig. 2Q). Furthermore, METTL16 knockdown reduced YTHDF2 binding to U6 under baseline condition and UVB stress in HaCaT cells (Fig. 2R). These results demonstrate that YTHDF2 binds to mA-modified U6 snRNA to mediate U6 decay. To determine the role of U6 in YTHDF2’s function, we assessed whether U6 knockdown rescues the effect of YTHDF2 knockdown on inflammatory gene expression. Indeed, U6 (RNU6-1) knockdown drastically inhibited the effect of YTHDF2 knockdown on the expression of cytokine genes in HaCaT cells (Fig. 3A-C and Supplementary Fig. S11A). Prior literature has shown that UVB exposure directly damages the snRNA U1 to promote inflammation. To determine whether the effect of U6 on cytokine expression occurs through direct UVB damage to U6 snRNA, we assessed the effect of UVB-irradiated synthetic U6 snRNA. tRNA was used as a negative control RNA. However, we found that in vitro UVB-irradiated U6 snRNA had no effect on IL-6 expression when transfected into HaCaT cells (Fig. 3D, E). Next, we assessed whether the effect of YTHDF2 inhibition is mediated by the direct damage on U6 snRNA by UVB irradiation. We examined the difference in inflammatory response in HeLa cells, a cell line with high transfection efficiency, with or without YTHDF2 inhibition transfected with sham- or UVB-irradiated synthetic non-mA-modified (U6)- and mA-modified U6 (mA U6) in vitro. Consistently, in vitro UVB irradiation of U6 or mA U6 molecules had no effect on IL-6 expression compared to sham (Fig. 3F). In WT HeLa cells, while both U6 and mA U6 induced IL-6 expression, an inflammatory gene robustly upregulated by UVB irradiation and/or YTHDF2 inhibition (Figs. 2I and 3B), U6 lead to a higher induction than mA U6 (Fig. 3F). We also observed that YTHDF2 deletion in HeLa cells increased IL-6 expression and abolished the difference between U6 and mA U6 (Fig. 3F).Fig. 3YTHDF2 controls inflammatory gene expression through U6 mA methylation.A–C qPCR of TNF-α, IL-6 and IL-1β mRNAs in HaCaT cells with or without YTHDF2 or U6 knockdown, with or without UVB irradiation. siU6, small interfering RNA targeting U6 snRNA. qPCR of U6 snRNA (D) and IL-6 mRNA (E) in HaCaT cells 30 h after transfection with sham- or UVB-irradiated purified tRNA or synthetic U6 snRNA. F qPCR of IL-6 mRNA in wild-type (WT) or YTHDF2 knockout (KO) HeLa cells 30 h after transfection with sham- or UVB-irradiated purified tRNA, synthetic U6 snRNA, or synthetic mA-modified U6. G, H qPCR of TNF-α and IL-6 mRNAs in A431 cells transduced with shNC (short hairpin RNA negative control), shMETTL16-1 or shMETTL16-2. I–L qPCR of TNF-α, IL-6, METTL16 and U6 snRNA in A431 cells with or without siMETTL16 or siU6. M Immunoblot of COX-2 in HaCaT cells with or without METTL16 knockdown, with or without UVB irradiation. N–Q qPCR of TNF-α, COX-2, IL-1β and IL-6 mRNAs in HaCaT cells with or without YTHDF2 or METTL16 knockdown, with or without UVB irradiation. R–W qPCR of TNF-α, IL-1β and IL-6 mRNAs in WT and YTHDF2 knockout (DF2 KO) HeLa cells with or without METTL16 knockdown, with or without UVB irradiation. siM16, small interfering RNA targeting METTL16. Housekeeping genes used were HPRT1 (A–C,N–W), 18S rRNA (D, K) and GAPDH (E–J, L). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are presented as mean ± SE (n = 4 for A–C, N–W) or mean ± SD (n = 3 for D–L). All experiments were conducted using biologically independent samples. A–C qPCR of TNF-α, IL-6 and IL-1β mRNAs in HaCaT cells with or without YTHDF2 or U6 knockdown, with or without UVB irradiation. siU6, small interfering RNA targeting U6 snRNA. qPCR of U6 snRNA (D) and IL-6 mRNA (E) in HaCaT cells 30 h after transfection with sham- or UVB-irradiated purified tRNA or synthetic U6 snRNA. F qPCR of IL-6 mRNA in wild-type (WT) or YTHDF2 knockout (KO) HeLa cells 30 h after transfection with sham- or UVB-irradiated purified tRNA, synthetic U6 snRNA, or synthetic mA-modified U6. G, H qPCR of TNF-α and IL-6 mRNAs in A431 cells transduced with shNC (short hairpin RNA negative control), shMETTL16-1 or shMETTL16-2. I–L qPCR of TNF-α, IL-6, METTL16 and U6 snRNA in A431 cells with or without siMETTL16 or siU6. M Immunoblot of COX-2 in HaCaT cells with or without METTL16 knockdown, with or without UVB irradiation. N–Q qPCR of TNF-α, COX-2, IL-1β and IL-6 mRNAs in HaCaT cells with or without YTHDF2 or METTL16 knockdown, with or without UVB irradiation. R–W qPCR of TNF-α, IL-1β and IL-6 mRNAs in WT and YTHDF2 knockout (DF2 KO) HeLa cells with or without METTL16 knockdown, with or without UVB irradiation. siM16, small interfering RNA targeting METTL16. Housekeeping genes used were HPRT1 (A–C,N–W), 18S rRNA (D, K) and GAPDH (E–J, L). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are presented as mean ± SE (n = 4 for A–C, N–W) or mean ± SD (n = 3 for D–L). All experiments were conducted using biologically independent samples. To explore the role of mA methylation of U6 in YTHDF2’s function in skin cancer, we next assessed the consequence of METTL16 inhibition on the expression of cytokines in the human A431 skin carcinoma cells. METTL16 knockdown enhanced the expression of IL-6 and TNF-α, two inflammatory genes robustly upregulated by UVB irradiation and/or YTHDF2 inhibition (Figs. 2I and 3B), under baseline conditions in A431 cells (Fig. 3G–H, Supplementary Fig. S11B). The effect of METTL16 knockdown was reduced by U6 knockdown in A431 cells (Fig. 3I-L), indicating that METTL16 suppresses cytokine expression through regulating U6. It is noteworthy that as compared with HaCaT and HeLa cells, A431 cells showed reduced basal METTL16 expression, which may render A431 cells sensitive to METTL16 knockdown-induced cytokine expression (Supplementary Fig. S11C). METTL16 knockdown increased UVB-induced COX-2 expression in HaCaT cells (Fig. 3M). However, dual knockdown of both METTL16 and YTHDF2 mimicked the effect of YTHDF2 knockdown alone in HaCaT cells (Fig. 3N–Q). Similarly, YTHDF2 knockout abolished the effect of METTL16 knockdown in HeLa cells (Fig. 3R–W, Supplementary Fig. S11D, E). Finally, to determine whether METTL16 inhibition affects dsRNA levels, as previously observed following METTL3 inhibition, and to assess whether U6 snRNA contributes significantly to the dsRNA pool, we compared dsRNA levels in A431 cells under control, METTL16 knockdown, and/or U6 knockdown conditions using immunofluorescence analysis with the anti-J2 antibody. However, METTL16 knockdown, U6 knockdown, or the combined knockdown had no effect on dsRNA levels in A431 cells (Fig. S11F-S11G), suggesting that METTL16 regulates inflammatory response via a distinct mechanism from METTL3 and that the effect of U6 depletion on inflammatory gene expression is not mediated by secondary dsRNA fragment accumulation. Together, these results demonstrate that YTHDF2 suppresses inflammatory gene expression, at least in part, by controlling the metabolism of mA methylated U6. Our RNA-seq data analysis showed that genes in the TLR3 cascade are upregulated by YTHDF2 depletion (Fig. 1A), suggesting that TLR3 signaling may contribute to the inflammatory response induced by YTHDF2 loss. TLR3 is a pattern recognition receptor best known for detecting dsRNA and structured RNAs resembling dsRNA to activate immune responses. Indeed, a small-scale siRNA screening for TLR3, TLR7, TLR8, MDA5, and RIG-I, which are all known sensors for RNA, showed that knockdown of TLR3 at least partially reverses the effect of YTHDF2 knockdown in HaCaT cells (Fig. 4A, B). In addition, TLR7/8 knockdown also partially reversed the effect of YTHDF2 knockdown in HaCaT cells (Fig. 4A, B), suggesting an important role of these receptors in YTHDF2’s function and warrants further investigation. Moreover, inhibition of TLR3 at least partially reversed the effect of either YTHDF2 knockdown or METTL16 knockdown both under UVB stress in HaCaT cells and under baseline condition in A431 cells (Fig. 4C–G and Supplementary Fig. S12A–J). Together, these data demonstrate that the effect of YTHDF2 on inflammation is mediated, at least in part, through TLR3.Fig. 4YTHDF2 interacts with mA U6 and thus inhibits mA U6 binding to TLR3.A, B qPCR of TNF-α and COX-2 mRNAs in shNC and shDF2 HaCaT cells transfected with siNC, siTLR3, siTLR7, siTLR8, siMDA5 or siRIG-I (siRNAs targeting TLR3, TLR7, TLR8, MDA5 and RIG-I, respectively). C, D qPCR of TNF-α and IL-6 mRNAs in shNC and shDF2 HaCaT cells transfected with siNC or siTLR3, 6 h after sham or UVB irradiation. E-G. qPCR of IL-8, IL-6 and IL-1β mRNAs in shNC and shDF2 A431 cells. H–J qPCR of IL-6, U6 snRNA and TLR3 in A431 cells with or without U6 and/or TLR3 knockdown. K Pull-down showing interaction between biotin-labelled U6 or mA-U6 and recombinant TLR3. L qPCR of U6 snRNA in mA-IP or flow-through fractions in A431 cells, showing the proportion of mA-modified U6. M Pull-down of biotin-labelled tRNA, U6 or truncated U6 oligos (1–26, 36–60, 54–91, 63–85) with TLR3 in lysates of HeLa cells overexpressing TLR3–FLAG. N Pull-down of biotin-labelled tRNA, U6 or mA-U6 with recombinant YTHDF2 and TLR3. O. Pull-down of biotin-labelled tRNA, U6 or mA-U6 with YTHDF2 or TLR3 in lysates from WT and DF2 KO HeLa cells overexpressing TLR3–FLAG. P RIP of U6 interaction with TLR3 (WT and mutants) in DF2 KO HeLa cells. Q–R qPCR of IL-6 mRNA in DF2 KO HeLa cells with TLR3 knockdown and overexpression of WT or mutant TLR3, following treatment with synthetic U6 or poly I:C (polyinosinic–polycytidylic acid, synthetic dsRNA analog). S Pull-down of biotin-labelled poly I:C with WT or mutant TLR3 in DF2 KO HeLa cells with TLR3 knockdown and overexpression of WT or mutant TLR3–FLAG. Housekeeping genes: HPRT1 (A–G, Q–R), GAPDH (H, J) and 18S rRNA (I). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are shown as mean ± SE (n = 4 for A–G) or mean ± SD (n = 3 for H–J, L, P–R). All experiments used biologically independent samples. A, B qPCR of TNF-α and COX-2 mRNAs in shNC and shDF2 HaCaT cells transfected with siNC, siTLR3, siTLR7, siTLR8, siMDA5 or siRIG-I (siRNAs targeting TLR3, TLR7, TLR8, MDA5 and RIG-I, respectively). C, D qPCR of TNF-α and IL-6 mRNAs in shNC and shDF2 HaCaT cells transfected with siNC or siTLR3, 6 h after sham or UVB irradiation. E-G. qPCR of IL-8, IL-6 and IL-1β mRNAs in shNC and shDF2 A431 cells. H–J qPCR of IL-6, U6 snRNA and TLR3 in A431 cells with or without U6 and/or TLR3 knockdown. K Pull-down showing interaction between biotin-labelled U6 or mA-U6 and recombinant TLR3. L qPCR of U6 snRNA in mA-IP or flow-through fractions in A431 cells, showing the proportion of mA-modified U6. M Pull-down of biotin-labelled tRNA, U6 or truncated U6 oligos (1–26, 36–60, 54–91, 63–85) with TLR3 in lysates of HeLa cells overexpressing TLR3–FLAG. N Pull-down of biotin-labelled tRNA, U6 or mA-U6 with recombinant YTHDF2 and TLR3. O. Pull-down of biotin-labelled tRNA, U6 or mA-U6 with YTHDF2 or TLR3 in lysates from WT and DF2 KO HeLa cells overexpressing TLR3–FLAG. P RIP of U6 interaction with TLR3 (WT and mutants) in DF2 KO HeLa cells. Q–R qPCR of IL-6 mRNA in DF2 KO HeLa cells with TLR3 knockdown and overexpression of WT or mutant TLR3, following treatment with synthetic U6 or poly I:C (polyinosinic–polycytidylic acid, synthetic dsRNA analog). S Pull-down of biotin-labelled poly I:C with WT or mutant TLR3 in DF2 KO HeLa cells with TLR3 knockdown and overexpression of WT or mutant TLR3–FLAG. Housekeeping genes: HPRT1 (A–G, Q–R), GAPDH (H, J) and 18S rRNA (I). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are shown as mean ± SE (n = 4 for A–G) or mean ± SD (n = 3 for H–J, L, P–R). All experiments used biologically independent samples. To further determine whether the effect of YTHDF2 inhibition on the inflammatory response is mediated through U6/TLR3, we assessed the effect of U6 knockdown, TLR3 knockdown, and their combined depletion. We found that either U6 knockdown or TLR3 knockdown decreased IL-6 expression, and that combined depletion of TLR3 and U6 had a similar effect to single knockdown of either U6 (RNU6-1) or TLR3 in A431 cells (Fig. 4H–J), suggesting that the effect of U6 on inflammation is mediated through TLR3 signaling. In addition, U6 knockdown had no effect on the dsRNA level in A431 cells (Fig. S11F), suggesting that U6 depletion does not affect secondary dsRNA fragment accumulation in regulating inflammatory response. Given the well-established role of U6 snRNA in regulation of splicing, we aimed to assess whether splicing inhibition affects the inflammatory response driven by YTHDF2 loss through U6 upregulation. We found that the splicing inhibitor Pladienolide B (PlaB) decreased the expression of the proinflammatory cytokines in both control and YTHDF2-knockdown cells (Supplementary Fig. S12K), mimicking the effect of U6 knockdown (Fig. 3). In parallel, PlaB also decreased the U6 snRNA level, which correlates with the decrease in cytokine expression caused by PlaB (Supplementary Fig. S12K). These findings raise the possibility that splicing inhibition by PlaB reduces global transcription or alters RNA metabolism, leading to decreased expression of inflammatory genes and U6 snRNA. To determine whether UVB-induced DNA damage also contributes to the induction of inflammation, we assessed the effect of TLR9 knockdown or cGAS inhibition. We found that either TLR9 knockdown or inhibition of cGAS had no effect on UVB-induced expression of inflammatory genes (Fig. S12L–S12M). These data suggest that UVB-induced inflammation is not mediated by UVB-induced DNA damage. To determine the specific mechanism by which YTHDF2 regulates U6 and TLR3 signaling, we performed several in vitro binding and pull-down assays. Consistent with observations in mRNA, we found that recombinant YTHDF2 preferentially binds to mA U6 more than U6 in vitro (Supplementary Fig. S13A). In contrast, recombinant TLR3 bound to both U6 and mA U6 with a similar affinity in vitro (Fig. 4K). An mA immunoprecipitation assay showed that 16.5% U6 snRNA is mA modified in A431 cells (Fig. 4L), similar to previous findings in mouse embryonic fibroblasts. Additionally, pull-down assays showed that the U6 36-60 sequence, where the A43 mA motif is located, preferentially binds with TLR3 in HeLa cells (Fig. 4M, Supplementary Fig. S13B, C). When recombinant TLR3 was incubated with U6 or mA U6 in the presence of recombinant YTHDF2, we found that U6 exhibited preferred binding with TLR3 (Fig. 4N). In comparison, mA U6 preferentially bound with YTHDF2, but not TLR3 (Fig. 4N). In wild-type (WT) HeLa cells, pull-down assays showed that U6 binds with TLR3 but not YTHDF2, while mA U6 binds with YTHDF2, not TLR3 (Fig. 4O), confirming the findings from in vitro binding assays (Fig. 4N). However, in YTHDF2 knockout cells, both U6 and mA U6 bound with TLR3 with mA U6 showing more prominent binding with TLR3 than U6 (Fig. 4O), suggesting that TLR3 prefers to bind with mA U6 in vivo, distinct from the findings from in vitro binding assays (Fig. 4K), possibly due to structural differences between endosomal TLR3 in vivo and recombinant TLR3 in vitro. Together, these findings demonstrate that YTHDF2 specifically inhibits TLR3 binding to mA U6. TLR3 recognizes its ligand, dsRNA, through its N-terminal ectodomain, which contains 23 leucine-rich repeats (LRRs). Mutational analysis has shown that the LRR20 motif of TLR3 binds to poly I:C, a synthetic analog of dsRNA and known TLR3 ligand. To determine the specific LRR motif of TLR3 that binds to U6, we generated various LRR deletion mutants of motifs (ΔLRR16-19, ΔLRR20, and ΔLRR21) that are in the C-terminus of TLR3 around the poly I:C-binding LRR20 motif to determine which mutants show loss of binding to U6 and activation by U6. A RIP assay showed that deletion of LRR21 leads to loss of binding to U6, while deletion of LRR20 had no effect in HeLa cells with YTHDF2 knockout and TLR3 knockdown (Fig. 4P). Intriguingly, deletion of LRR16-19 enhanced binding of TLR3 to U6 (Fig. 4P), suggesting that these motifs suppress the TLR3-U6 interaction, warranting future investigation to determine the underlying mechanism for such suppression. Further analysis showed that deletion of the LRR21 motif, but not the LRR20 motif, resulted in loss of induction of IL-6 by U6 in HeLa cells with YTHDF2 knockout and TLR3 knockdown (Fig. 4Q). In comparison, deletion of the LRR20 motif, but not the LRR21 or LRR16-19 motifs, led to loss of induction of IL-6 expression by Poly I:C in HeLa cells with YTHDF2 knockout and TLR3 knockdown (Fig. 4R), consistent with previous studies. Furthermore, a pull-down assay confirmed that LRR20 is required for binding to poly I:C in HeLa cells with YTHDF2 knockout and TLR3 knockdown (Fig. 4S). Together, these results demonstrate that the mA motif sequence of U6 snRNA interacts with and activates TLR3 via the LRR21 motif. TLR3 is an RNA sensor localized in endosomes. To determine the mechanism by which YTHDF2 regulates TLR3 activation, we assessed whether U6 and YTHDF2 also localize in the cytoplasm, particularly in endosomes, and whether YTHDF2, U6, and TLR3 interact. First, we found that both YTHDF2 knockdown and UVB irradiation increased the cytoplasmic U6 level in HeLa, HaCaT, and A431 cells (Fig. 5A and Supplementary Fig. S14A–E). Second, UVB irradiation increased both total level and endosomal proportion of U6 snRNA in HeLa cells (Fig. 5B). Third, we performed RNA fluorescence in situ hybridization (FISH) of U6 in combination with immunofluorescence analysis of the endosome marker Rab7 to confirm U6 localization in endosomes. While U6 snRNA was mainly detected in the nucleus, we observed that UVB increases U6 snRNA levels in the cytoplasm and U6 colocalization with the late endosome marker Rab7 in HaCaT cells, HeLa cells, and mouse skin (Fig. 5C, Supplementary Fig. S14F–S14G), supporting U6 localization in endosomes. YTHDF2 inhibition in HaCaT cells, mouse skin, or A431 cells increased U6 snRNA levels in the nucleus and cytoplasm and increased U6 colocalization with Rab7 (Fig. 5C, Supplementary Fig. S14F–S14I), while U6 knockdown decreased them in A431 and HaCaT cells (Supplementary Fig. S14H–S14I). These orthogonal validations further suggest a role for endosomal U6 snRNA in the inflammatory response. Next, we demonstrated YTHDF2 colocalization with endosomes using multiple approaches, including endosome fractionation in A431 cells and Endo-IP in 293 T cells, which were selected for their high plasmid transfection and protein expression (Fig. 5D, E). Endo-IP enabled rapid isolation of early/sorting endosomes through affinity capture of the early endosome-associated protein EEA1. Further, confocal imaging analysis in HeLa cells confirmed that YTHDF2 localization in endosomes, as shown by the colocalization of YTHDF2 with the late endosome marker Rab7 (Fig. 5F).Fig. 5Both U6 and YTHDF2 are localized in endosomes.A qPCR of nuclear (nuc) and cytoplasmic (cyto) fractions in HeLa cells with or without UVB irradiation (30 mJ/cm², 6 h). B. qPCR in HeLa cells with or without UVB irradiation (30 mJ/cm², 6 h). C Confocal images of HaCaT cells with or without YTHDF2 knockdown, 6 h after sham or UVB irradiation (20 mJ/cm²). D Immunoblot in A431 cells. E Endo-IP of endosomal fractions in 293 T cells. F Confocal images of HeLa cells; arrows indicate colocalization of YTHDF2 and Rab7. G qPCR in endosomes of HeLa cells with or without Dynasore. H Immunoblot in A431 cells with or without Dynasore. I qPCR of mRNA stability in A431 cells treated with or without Dynasore. J qPCR in endosomes in A431 cells with or without SIDT2 knockdown. K Immunoblot of WCL and endosomes in A431 cells with or without SIDT2 knockdown. L qPCR of mRNA stability in A431 cells with or without SIDT2 knockdown. M Immunoblot of WCL and endosomes in A431 cells with or without U6 knockdown. N qPCR in endosomes of HeLa cells with or without METTL16 knockdown. O qPCR in endosomes of A431 cells with or without YTHDF2 knockdown. P Schematic of human YTHDF2 and truncations (N- or C-terminal). Q Immunoblot of WCL and endosomes in HeLa cells transfected with EV, DF2 (WT), DF2 (N) or DF2 (C). Unmarked bands >25 kDa are nonspecific. qPCR of mRNAs in HeLa cells transfected with tRNA or U6 and treated with vehicle or Dynasore (R), and in A431 cells with or without YTHDF2 knockdown treated with vehicle or Dynasore (S). Housekeeping genes: 18S rRNA (B, I, L, N, O) and GAPDH (R–S). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are shown as mean ± SD (n = 3 for A–B, G, I, J, L, N, O, R; n = 6 for S). All experiments used biologically independent samples. A qPCR of nuclear (nuc) and cytoplasmic (cyto) fractions in HeLa cells with or without UVB irradiation (30 mJ/cm², 6 h). B. qPCR in HeLa cells with or without UVB irradiation (30 mJ/cm², 6 h). C Confocal images of HaCaT cells with or without YTHDF2 knockdown, 6 h after sham or UVB irradiation (20 mJ/cm²). D Immunoblot in A431 cells. E Endo-IP of endosomal fractions in 293 T cells. F Confocal images of HeLa cells; arrows indicate colocalization of YTHDF2 and Rab7. G qPCR in endosomes of HeLa cells with or without Dynasore. H Immunoblot in A431 cells with or without Dynasore. I qPCR of mRNA stability in A431 cells treated with or without Dynasore. J qPCR in endosomes in A431 cells with or without SIDT2 knockdown. K Immunoblot of WCL and endosomes in A431 cells with or without SIDT2 knockdown. L qPCR of mRNA stability in A431 cells with or without SIDT2 knockdown. M Immunoblot of WCL and endosomes in A431 cells with or without U6 knockdown. N qPCR in endosomes of HeLa cells with or without METTL16 knockdown. O qPCR in endosomes of A431 cells with or without YTHDF2 knockdown. P Schematic of human YTHDF2 and truncations (N- or C-terminal). Q Immunoblot of WCL and endosomes in HeLa cells transfected with EV, DF2 (WT), DF2 (N) or DF2 (C). Unmarked bands >25 kDa are nonspecific. qPCR of mRNAs in HeLa cells transfected with tRNA or U6 and treated with vehicle or Dynasore (R), and in A431 cells with or without YTHDF2 knockdown treated with vehicle or Dynasore (S). Housekeeping genes: 18S rRNA (B, I, L, N, O) and GAPDH (R–S). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are shown as mean ± SD (n = 3 for A–B, G, I, J, L, N, O, R; n = 6 for S). All experiments used biologically independent samples. To determine the mechanism by which U6 and YTHDF2 translocate to endosomes, we assessed the role of dynamin, a key factor in the endocytosis pathway that mediates biomolecule trafficking. Indeed, the dynamin inhibitor Dynasore reduced the endosomal proportion of U6 and the endosomal YTHDF2 level in HeLa cells and reduced U6 stability in A431 cells (Fig. 5G–I, supplementary Fig. S14J), indicating a critical role of endocytosis in the localization of U6 and YTHDF2 to endosomes. To further understand the molecular mechanism by which U6 and YTHDF2 are trafficked to endosomes, we assessed the role of SIDT2, a RNA transporter localized in late endosomes and lysosomes. Indeed, knockdown of SIDT2 reduced the total and endosomal proportion of U6, endosomal YTHDF2 levels, and U6 stability in A431 cells (Fig. 5J–L and supplementary Fig. S14K–S14L). These results mimic the effect of Dynasore and indicate that SIDT2 is required for the localization of U6 and YTHDF2 to endosomes. To examine whether YTHDF2 or U6 enters endosomes as a complex or individually, we assessed the effect of knockdown of U6, YTHDF2, or METTL16 on endosomal localization. Knockdown of U6 reduced the YTHDF2 levels in endosomes in A431 cells (Fig. 5M and supplementary Fig. S14M), indicating that U6 is required for YTHDF2 localization to endosomes. We found that knockdown of either YTHDF2 or METTL16 increased both endosomal U6 levels and total U6 levels in HeLa cells (Fig. 5N, O), indicating that neither YTHDF2 nor mA methylation is required for U6 entry into endosomes. In addition, we found that the C-terminus of YTHDF2 is required for its localization to endosomes, as wild-type (DF2 (WT)) full-length and C-terminal YTHDF2 (DF2 (C)), but not N-terminus YTHDF2 (DF2 (N)), were detected in endosomes (Fig. 5P, Q). Together, our data demonstrate that SIDT2 transports either the U6-YTHDF2 complex altogether or U6 alone into endosomes, while YTHDF2 entry into endosomes requires U6 snRNA. Next, to characterize whether YTHDF2 or U6 snRNA is trafficked intracellularly or extracellularly to endosomes, we assessed whether cells uptake either YTHDF2 or U6 snRNA secreted by other cells. However, when cultured with conditioned medium from control cells, no YTHDF2 or U6 snRNA was detected in endosomes of HeLa cells with YTHDF2 deletion or A431 cells with U6 snRNA knockdown, respectively (Fig. S14N–S14O), suggesting that intracellular trafficking pathways deliver YTHDF2 and U6 snRNA into endosomes. Lastly, to determine whether U6 snRNA entry into endosomes is required for the increased expression of inflammatory genes, we assessed the effect of Dynasore, an inhibitor of U6 endosomal entry (Fig. 5G), on cytokine expression. We found that Dynasore reduced expression of inflammatory genes induced by U6 in HeLa cells, or by YTHDF2 knockdown in A431 cells at baseline condition (Fig. 5R, S) and in HaCaT and HeLa cells treated with UVB irradiation (Fig. S14P, S14Q). In control cells with YTHDF2 expression, Dynasore also slightly reduced inflammatory gene expression in A431 cells and IL-6 expression in UVB-irradiated HaCaT cells, but not in UVB-irradiated HeLa cells (Fig. 5S and Fig. S14P, S14Q), suggesting a cell-type-dependent, YTHDF2-independent effect of Dynasore. Such effect of Dynasore may be due to the reduction of endosomal localization of non-mA-modified U6, which can be YTHDF2-independent, as non-mA-modified U6 is bound by TLR3 but not by YTHDF2 (Fig. 4K, O). Future investigation is needed to elucidate the specific molecular mechanism for regulating U6 snRNA metabolism and function. Taken together, these findings demonstrate that the endocytosis pathway mediates the localization of U6 and YTHDF2 to endosomes, thus regulating TLR3 activation and inflammation. Since both YTHDF2 loss and UVB treatment similarly increase U6 stability (Fig. 2O), we next asked whether UVB inhibits YTHDF2 by modulating YTHDF2 post-translational modifications. Mass spectrometric analysis showed that at 1 h, UVB induced dephosphorylation of YTHDF2 at serine 39 (S39) in HaCaT cells (Fig. 6A, Supplementary Fig. S15A). To confirm this observation, we generated an antibody specific for YTHDF2 phosphorylation at S39 and found that UVB decreases YTHDF2 S39 phosphorylation post-irradiation in HaCaT cells (Fig. 6B). We next re-examined YTHDF2-interacting proteins and found that UVB inhibits the binding of YTHDF2 with several kinases and CNOT1, the subunit for the CCR4–NOT deadenylase complex that mediates mA-methylated RNA decay, but promotes YTHDF2 binding to MYPT1 (myosin phosphatase target subunit 1) (Supplementary Fig. S15B–D), a regulatory subunit for myosin phosphatase that is a member of protein phosphatase type 1. We hypothesized that UVB might induce YTHDF2 dephosphorylation by the MYPT1-phosphatase complex to inhibit the YTHDF2-CNOT1 interaction and thus U6 decay. Coimmunoprecipitation (Co-IP) analysis demonstrated that UVB induces the YTHDF2-MYPT1 interaction in HeLa cells (Supplementary Fig. S15E). Further co-IP analysis demonstrated that the phosphomimetic S39D YTHDF2 mutant shows enhanced binding to MYPT1 as compared with WT YTHDF2 and the non-phosphorylatable S39A mutant in HeLa cells with YTHDF2 knockout (Fig. 6C), suggesting a potential role of the MYPT1–phosphatase complex in dephosphorylating YTHDF2 at S39. In addition, as compared with WT YTHDF2, we found that the phosphomimetic S39D YTHDF2 mutant shows increased binding to CNOT1, while the S39A mutant shows decreased binding in HeLa cells with YTHDF2 knockout (Fig. 6D), indicating that S39 phosphorylation of YTHDF2 is required for the YTHDF2-CNOT1 interaction. Consistently, WT YTHDF2, but not the S39A mutant, decreased U6 snRNA level and U6-induced-TNF-α expression, while both WT and S39A mutant YTHDF2 showed similar binding affinity to U6 in HeLa cells with YTHDF2 knockout (Fig. 6E, Supplementary Fig. S15F–I).Fig. 6YTHDF2 is inhibited by UVB irradiation.A Loss of YTHDF2 phosphorylation at serine 39 (S39) in HaCaT cells after UVB irradiation identified by mass spectrometry. B Immunoblot of p-YTHDF2 (S39) and YTHDF2 in HaCaT cells with or without UVB irradiation (1 h). C Co-IP of MYPT1 with FLAG-tagged YTHDF2 (EV, WT, S39D or S39A) in DF2 KO HeLa cells. D Co-IP of CNOT1 with FLAG-tagged YTHDF2 (EV, WT, S39D or S39A) in DF2 KO HeLa cells. E qPCR of U6 snRNA stability in DF2 KO HeLa cells transfected with DF2 WT or DF2 S39A. F RIP of U6 with CNOT1 in HeLa cells. qPCR of U6 snRNA stability in HeLa (G) and A431 (H) cells with or without CNOT1 knockdown. RIP of U6 with CNOT1 in HeLa cells with or without YTHDF2 knockout (I) or METTL16 knockdown (J). K Immunoblot of FLAG-tagged YTHDF2 in WCL and endosomes from HeLa cells transfected with EV, DF2 WT or DF2 S39A. L Immunoblot analysis of YTHDF2, p62, and LC3-I/II in HaCaT after sham (C) or UVB irradiation. M Unique spectrum counts for autophagy receptors including p62 f from mass spectrometry of YTHDF2-interacting proteins as compared with IgG. N Immunoblot of YTHDF2, p62, and LC3-I/II in HaCaT cells expressing shNC or shp62, 6 h after sham or UVB irradiation. O Co-IP of FLAG-tagged YTHDF2 with HA-tagged p62 (WT or ΔUBA) in HeLa cells. P Co-IP of FLAG-tagged YTHDF2 (WT, N-terminus or C-terminus) with p62 in HeLa cells. Housekeeping gene: 18S rRNA (E, G, H). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are shown as mean ± SD (n = 3 for E–J). All experiments were conducted using biologically independent samples. A Loss of YTHDF2 phosphorylation at serine 39 (S39) in HaCaT cells after UVB irradiation identified by mass spectrometry. B Immunoblot of p-YTHDF2 (S39) and YTHDF2 in HaCaT cells with or without UVB irradiation (1 h). C Co-IP of MYPT1 with FLAG-tagged YTHDF2 (EV, WT, S39D or S39A) in DF2 KO HeLa cells. D Co-IP of CNOT1 with FLAG-tagged YTHDF2 (EV, WT, S39D or S39A) in DF2 KO HeLa cells. E qPCR of U6 snRNA stability in DF2 KO HeLa cells transfected with DF2 WT or DF2 S39A. F RIP of U6 with CNOT1 in HeLa cells. qPCR of U6 snRNA stability in HeLa (G) and A431 (H) cells with or without CNOT1 knockdown. RIP of U6 with CNOT1 in HeLa cells with or without YTHDF2 knockout (I) or METTL16 knockdown (J). K Immunoblot of FLAG-tagged YTHDF2 in WCL and endosomes from HeLa cells transfected with EV, DF2 WT or DF2 S39A. L Immunoblot analysis of YTHDF2, p62, and LC3-I/II in HaCaT after sham (C) or UVB irradiation. M Unique spectrum counts for autophagy receptors including p62 f from mass spectrometry of YTHDF2-interacting proteins as compared with IgG. N Immunoblot of YTHDF2, p62, and LC3-I/II in HaCaT cells expressing shNC or shp62, 6 h after sham or UVB irradiation. O Co-IP of FLAG-tagged YTHDF2 with HA-tagged p62 (WT or ΔUBA) in HeLa cells. P Co-IP of FLAG-tagged YTHDF2 (WT, N-terminus or C-terminus) with p62 in HeLa cells. Housekeeping gene: 18S rRNA (E, G, H). Statistical analyses were performed using a two-tailed unpaired Student’s t-test. Data are shown as mean ± SD (n = 3 for E–J). All experiments were conducted using biologically independent samples. Next, we assessed the role of CNOT1 in U6 decay. Indeed, we found that CNOT1 binds to U6 in HeLa cells (Fig. 6F) and CNOT1 knockdown inhibits U6 decay in both HeLa and A431 cells (Fig. 6G, H, and Supplementary Fig. S15J, K). To elucidate the mechanism of YTHDF2-mediated U6 decay, we examined the role of YTHDF2 and METTL16-mediated mA methylation in CNOT1 binding to U6 snRNA. We found that either YTHDF2 deletion or METTL16 knockdown drastically reduces CNOT1 binding to U6 in HeLa cells (Fig. 6I, J and Supplementary Fig. S15L, M), supporting a critical role of YTHDF2 and U6 mA methylation in CNOT1 binding to U6 snRNA. To determine whether UVB irradiation affects CNOT1 expression, we assessed the effect of UVB irradiation on CNOT1 protein levels. We found that UVB irradiation had no effect on CNOT1 protein expression in either HeLa or HaCaT cells (Supplementary Fig. S15N). These results demonstrate that CNOT1 binds to U6 snRNA and mediates U6 decay through interacting with YTHDF2 in an mA-methylation dependent manner. Furthermore, the S39A mutant reduced endosomal localization in HeLa cells (Fig. 6K). These findings demonstrate that CNOT1 mediates U6 decay and dephosphorylation of YTHDF2 inhibits U6 decay and endosomal localization. In addition to inhibiting YTHDF2 phosphorylation, we observed that UVB down-regulates YTHDF2 protein in multiple cell lines and primary keratinocytes including HaCaT, CHL-1 (melanoma cells), HeLa, and NHEK cells (Fig. 6L, and Supplementary Fig. S16A–C). This occurred in parallel with autophagy induction, as shown by the downregulation of the selective autophagy receptor p62 and the induction of LC3-II, despite the increase in the YTHDF2 mRNA level in HaCaT cells (Fig. 6L, and Supplementary Fig. S16A-D). UVB irradiation induces an acute inflammatory response as observed by increased expression of inflammatory genes (Fig. 1). To determine whether the inflammatory response correlates with YTHDF2 protein abundance, we assessed YTHDF2 protein levels at different time points. Indeed, as compared with sham-irradiation, YTHDF2 protein level is decreased at 6 h post-UVB irradiation, while it was recovered at 24 h post-UVB irradiation in HaCaT cells (Supplementary Fig. S16D), likely due to increased YTHDF2 mRNA levels (Supplementary Fig. S16D). The temporal regulation of YTHDF2 protein expression by UVB irradiation negatively correlated with the increases in U6 snRNA levels (Fig. 2I) as well as inflammatory gene expression at 6 h followed by a decrease at 24 h post-UVB irradiation, further supporting an inhibitory role of YTHDF2 in UVB-induced inflammation. Importantly, UVB irradiation had little effect on other mA readers in HaCaT cells (Supplementary Fig. S16E). Furthermore, UVB irradiation also down-regulated YTHDF2 protein in mouse skin (Supplementary Fig. S16F, G). Next, to determine whether YTHDF2 down-regulation induced by UVB radiation is mediated through autophagy, a cellular self-eating process that degrades excessive or damaged proteins and organelles, we sought to establish whether autophagy inhibition reversed the effect of UVB irradiation on YTHDF2 and U6 levels. Using HaCaT cells with or without knockdown of the essential autophagy gene ATG5 or ATG7, we found that inhibition of either ATG5 or ATG7 prevented UVB-induced down-regulation of YTHDF2 and U6 up-regulation (Supplementary Fig. S16H–J). To identify the autophagy receptor responsible for UVB-induced autophagy of YTHDF2, we re-examined the list of YTHDF2-interacting proteins from our mass spectrometry analysis. We observed that YTHDF2 preferentially binds with the autophagy receptor p62 (also known as SQSTM1) in HaCaT cells (Fig. 6M, and Supplementary Fig. S16K). Indeed, knockdown of p62 prevented UVB-induced YTHDF2 down-regulation and U6 up-regulation in HaCaT cells (Fig. 6N and Supplementary Fig. 16L). Furthermore, in mouse skin, skin-specific conditional knockout (cKO) of Atg5, Atg7, or Sqstm1(p62) prevented UVB-induced YTHDF2 down-regulation (Supplementary Fig. S16M, N). Co-IP showed that p62 binds to YTHDF2 in HeLa and HaCaT cells. Additionally, the p62-YTHDF2 interaction requires p62’s ubiquitin-associated (UBA) domain in HeLa cells (Fig. 6O and Supplementary Fig. S16O), which is known to interact with ubiquitinated proteins. As both p62 and YTHDF2 can bind to RNA, we assessed whether the p62-YTHDF2 interaction is dependent on RNA. We found that removal of RNA had no effect on the interaction between p62 and YTHDF2 in HaCaT cells (Supplementary Fig. S16P). In addition, we found that the C-terminus, but not the N-terminus, of YTHDF2 is required for YTHDF2 binding to p62 (Fig. 6P). These results demonstrate that YTHDF2 is degraded via p62-dependent selective autophagy in response to UVB stress. As inflammation can promote tumorigenesis and the role of YTHDF2 in established cancers is cancer-type-dependent, we next investigated the function of YTHDF2 in skin tumorigenesis. Indeed, our RNA-seq data analysis showed that genes in the Pathways in cancer are upregulated by YTHDF2 knockdown in HaCaT cells (Fig. 1A), suggesting a role for YTHDF2 in skin cancer. In addition, we found that knockdown of YTHDF2 in A431 cells increases cell proliferation and migration in vitro, whereas forced overexpression of YTHDF2 in A431 cells decreases cell proliferation and migration (Fig. 7A–C, and Supplementary Fig. S17A). A431 xenografts in nude mice revealed that knockdown of YTHDF2 increased tumor growth, whereas forced overexpression of YTHDF2 decreased tumor growth (Fig. 7D, E, and Supplementary Fig. S17B). Furthermore, skin-specific Ythdf2 deletion accelerated skin tumorigenesis induced by chronic UVB irradiation in both male and female mice, demonstrating an inhibitory role of YTHDF2 in tumor initiation (Fig. 7F–I). Histological analysis showed that UVB irradiation induces skin tumor formation in both WT and DF2 cKO mice, while squamous skin carcinoma was detected in the DF2 cKO mice (Supplementary Fig. S17C). This was supported by immunofluorescence analysis that indicated expression of the basal keratinocyte marker KRT14 (K14) in both basal layer of mouse skin and mouse tumors, while the differentiation marker KRT10 (K10) was only detected in mouse skin, hair follicles, tumor from WT mice, but not in the tumor from DF2 cKO mice (Supplementary Fig. S17C).Fig. 7YTHDF2 suppresses skin tumorigenesis in mice through inhibiting TLR3.A Immunoblot confirming YTHDF2 knockdown and overexpression in A431 cells. B Cell proliferation of cells as in A.C Cell migration of cells as in A. D Tumor growth of A431 xenografts in nude mice. E Tumor weight for A431 xenografts. F-I Number of tumor per mouse (F and G) and percentage of tumor-free mice (H and I) in WT and DF2 cKO mice following chronic UVB irradiation. J Immunoblot of YTHDF2 and COX-2 in shNC and shDF2 A431 cells with or without U6 knockdown. K Cell proliferation of shNC and shDF2 A431 cells with or without U6 knockdown. Cell migration (L) and proliferation (M) in A431 cells with or without YTHDF2 knockdown, treated with or without Celecoxib (20 μM, COX-2 inhibitor, 24 h). Tumor volume (N) and tumor weight (O) in nude mice injected with shNC or shDF2 A431 cells with or without TLR3 knockdown. P Schematic of the role of YTHDF2 in regulating U6 snRNA decay and interaction with TLR3 to control UVB-induced inflammation and tumorigenesis. f Created in BioRender. Verghese, M. (https://BioRender.com/k4uvdti). Statistical analyses were conducted using a two-tailed unpaired Student’s t-test (B–F, H, K–O) and log-rank test (G, I). Data are shown as mean ± SD (n = 3 for B–E, K, L–O; n = 9 for each group in F, G; n = 8 [WT] and n = 9 [DF2 cKO] in H, I). All experiments used biologically independent samples. A Immunoblot confirming YTHDF2 knockdown and overexpression in A431 cells. B Cell proliferation of cells as in A.C Cell migration of cells as in A. D Tumor growth of A431 xenografts in nude mice. E Tumor weight for A431 xenografts. F-I Number of tumor per mouse (F and G) and percentage of tumor-free mice (H and I) in WT and DF2 cKO mice following chronic UVB irradiation. J Immunoblot of YTHDF2 and COX-2 in shNC and shDF2 A431 cells with or without U6 knockdown. K Cell proliferation of shNC and shDF2 A431 cells with or without U6 knockdown. Cell migration (L) and proliferation (M) in A431 cells with or without YTHDF2 knockdown, treated with or without Celecoxib (20 μM, COX-2 inhibitor, 24 h). Tumor volume (N) and tumor weight (O) in nude mice injected with shNC or shDF2 A431 cells with or without TLR3 knockdown. P Schematic of the role of YTHDF2 in regulating U6 snRNA decay and interaction with TLR3 to control UVB-induced inflammation and tumorigenesis. f Created in BioRender. Verghese, M. (https://BioRender.com/k4uvdti). Statistical analyses were conducted using a two-tailed unpaired Student’s t-test (B–F, H, K–O) and log-rank test (G, I). Data are shown as mean ± SD (n = 3 for B–E, K, L–O; n = 9 for each group in F, G; n = 8 [WT] and n = 9 [DF2 cKO] in H, I). All experiments used biologically independent samples. Next, we assessed whether YTHDF2 protein level is altered in human skin cancer. Immunofluorescence analysis showed that YTHDF2 was lower in human skin squamous cell carcinoma samples (both stage 1 and stage 2) than in normal human epidermis (Supplementary Fig. S17D, E). Furthermore, YTHDF2 was lower in stage 2 SCC than stage 1 SCC (Supplementary Fig. S17D, E), suggesting an active role for YTHDF2 in skin tumorigenesis as well as tumor progression. To determine the role of U6 in YTHDF2’s functional role, we assessed whether U6 inhibition rescues the effect of YTHDF2 knockdown on cell proliferation. U6 knockdown drastically inhibited the effect of YTHDF2 knockdown on COX-2 expression and cell proliferation in A431 cells (Fig. 7J, K and Supplementary Fig. S17F). Next, we assessed the importance of inflammatory genes in YTHDF2 function. Among these inflammatory mediators regulated by YTHDF2 (Fig. 1), COX-2 is a UVB-inducible enzyme that promotes inflammation and tumorigenesis. Indeed, the COX-2 inhibitor Celecoxib drastically suppressed the proliferation and migration of YTHDF2 knockdown and control cells; it also diminished the effect of YTHDF2 knockdown in A431 cells (Fig. 7L, M, Supplementary Fig. S17G). These findings indicate that YTHDF2 loss promotes tumor cell proliferation by increasing U6/COX-2. In addition, inhibition of TLR3 also reversed the tumor-promoting effect of YTHDF2 knockdown in mice from A431 cells (Fig. 7N, O and Supplementary Fig. S17H, I). These findings demonstrate that YTHDF2 suppresses tumorigenesis through inhibiting the TLR3 pathway. Despite emerging investigations into the critical roles of mA mRNA methylation in inflammation and cancer, the regulatory and functional role of mA methylation of self non-coding RNA in inflammation and tumorigenesis remains largely unknown. Here, we show that YTHDF2 recognizes mA methylation of non-coding U6 snRNA to control TLR3 activation in inflammation and tumorigenesis (Fig. 7P). mA-methylated U6 snRNA is recognized by YTHDF2, leading to U6 decay. In addition, both YTHDF2 and U6 localize to endosomes via SIDT2 and the endocytosis pathway. YTHDF2 binds to mA U6 to compete with TLR3 binding to mA U6, thus inhibiting the expression of inflammatory modulators, cell proliferation, and tumor growth. In addition, UVB irradiation inhibits YTHDF2 phosphorylation at S39, which is critical for YTHDF2 interaction with CNOT1 and localization in endosomes. YTHDF2 is also downregulated by UVB irradiation at later time points through p62-mediated selective autophagy. YTHDF2 is reduced in human skin cancer, SLE, and type I diabetes, as compared with normal controls. Skin-specific YTHDF2 deletion sensitizes mice to UVB-induced inflammation and tumorigenesis. Our findings demonstrated a critical role of YTHDF2 in regulating self non-coding RNA in inflammation and tumorigenesis (Fig. 7P). YTHDF2 was initially shown to be an mRNA-binding protein that recognizes mA methylation in mRNA and promotes mRNA decay by directly interacting with CNOT1 and recruiting the CCR4–NOT deadenylase complex. Here, we show that YTHDF2 binds to mA-modified self U6 snRNA to mediate its decay. As a key component of the spliceosome, U6 binds to and interacts with many protein factors. Upon mA methylation by METTL16, U6 structure may be altered such that a sequence on or around the mA site (A43) becomes available to preferentially bind to YTHDF2. Previously it has been shown that mA RNA methylation regulates the RNA-structure-dependent accessibility of RNA binding motifs to affect RNA–protein interactions for mRNAs. Thus, it is possible that mA methylation of U6 induces U6 structural remodeling to permit YTHDF2 binding. In addition, the interaction between YTHDF2 and U6 snRNP proteins may also facilitate YTHDF2 access to U6 snRNA. Moreover, we also show that CNOT1, which is critical for the CCR4-NOT mRNA deadenylase activity in the decay of mRNAs, including mA-methylated mRNAs, binds to U6 snRNA through interacting with YTHDF2 in an mA-methylation dependent manner, as YTHDF2 deletion or METTL16 knockdown inhibited the CNOT1-U6 snRNA interaction. Previously, the 3’ tail of a small fraction of U6 snRNA has been shown to be adenylated. Although U6 is not known to be polyadenylated, recent studies have detected mA methylated U6 snRNA in poly(A) RNA in worms. While this may represent remnants left after poly(A) enrichment from total RNA, it is also possible that a fraction of U6 snRNA is polyadenylated in worms and other organisms, which could regulate U6 decay upon mA methylation similar to poly(A) mRNAs via the YTHDF2-CNOT1 interaction. It is possible that adenylation and/or polyadenylation cooperates with mA methylation of U6 snRNA to regulate U6 snRNA turnover. Future investigation is warranted to elucidate the detailed mechanism by which YTHDF2 and CNOT1 regulate U6 snRNA decay. Nevertheless, our findings establish self U6 snRNA as a previously unrecognized non-coding RNA target of YTHDF2. Intriguingly, previous studies have also shown that loss of mett-10 in C. elegans increases U6 snRNA levels, while there is no mA reader protein like YTHDF2 in worms. There are three possibilities: (1) mA methylation of U6 may affect other U6 snRNA modifications that in turn regulate U6 snRNA stability, (2) mA methylation of other mett-10 target genes may regulate U6 snRNA transcription, maturation, and/or stability, or (3) worms may express other proteins that recognize mA-methylated U6 snRNA to mediate U6 snRNA decay. Future investigations are required to test these possibilities. In contrast, a previous report by Warda and colleagues did not detect a change to U6 snRNA levels upon METTL16 knockdown in HEK293 cells. These findings, together with our own, suggest that METTL16’s regulation of U6 snRNA levels may be context-dependent, warranting further investigation to elucidate the molecular basis for these differing effects across cell types and organisms. Previous reports have suggested that RNA modifications, including mA methylation, of in vitro-transcribed synthetic RNAs, inhibit activation of TLR3 and RIG-I. Accordingly, mA methylation has been shown to play critical roles in the immune recognition of circular RNAs and non-coding RNAs generated by back splicing, inhibiting their binding to RIG-I and subsequent RIG-I activation. U6 is best known for its role in splicing and functions in the nucleus. However, emerging evidence has shown that U6 can be localized in the cytoplasm and the nucleus. Recently, several snRNAs including U6 snRNA have been shown to be glycosylated and localized at the cell surface, suggesting emerging biological roles of cytoplasm and surface U6 snRNA in RNA biology. In addition, free U6 snRNA has a double-stranded 5’-short stem, 3’-telestem, and internal stem-loop (ISL) regions (Supplementary Fig. S13B), which may permit U6 binding to TLR3 as a self RNA pathogen-associated molecular pattern (PAMP). Our findings demonstrate that endogenous U6 snRNA is localized in endosomes and binds to TLR3 via the LRR21 motif, but not LRR20, the known motif that the synthetic ligand poly I:C binds, suggesting that TLR3 utilizes distinct LRR motifs for different RNA agonists. Our data also demonstrates that TLR3 prefers to bind to mA U6 than U6 when YTHDF2 is absent (Fig. 4O), suggesting that the mA motif and/or methylation may be critical for TLR3 binding. Intriguingly, we found that TLR3 preferentially binds to the U6 sequence containing the mA methylation site A43, which is also bound by YTHDF2 (Fig. 4M). Thus, YTHDF2 may compete with TLR3 in binding to mA U6, thereby inhibiting TLR3 activation. Distinct from the role of U1 snRNA in UVB-induced inflammation, in vitro irradiation of U6 with UVB had no effect on cytokine expression, further supporting the critical role of the U6 mA methylation sequence, but not UVB-induced de novo damage lesions of U6, in TLR3 activation. Under baseline conditions, METTL16 knockdown increases both total and endosomal U6 abundance and cytokine expression in METTL16-low A431 cells (Fig. 4N), suggesting that non-mA-modified U6 snRNA, which is not bound by YTHDF2, can also enter endosomes and be bound by TLR3 in a YTHDF2-independent manner. Intriguingly we show that both YTHDF2 and U6 are localized in endosomes, permitting YTHDF2 binding to U6 in endosomes to regulate TLR3 activation. YTHDF2 is known to localize in P-bodies to mediate the decay of mA-methylated mRNAs, while U6 is mainly localized in the nucleus as well as cytoplasm. Transporting RNA from the nucleus to endosomes and other membrane locations requires RNAs to cross a membrane at least once. The mammalian Sidt1 and Sidt2 genes, two homologs of the C. elegans sid-1 gene, have been shown to mediate the trafficking of RNAs across cellular or intracellular vesicle membranes such as lysosomes. Recently SIDT2 has been shown to be localized in endosomes. In addition, RNA transport by the C. elegans Sid-1 protein is shown to be specific for dsRNA. Indeed, previous studies have shown that free U6 snRNA contains double-stranded regions (Supplementary Fig. S13B), which may permit U6 to interact with SIDT2 and thus be transported to endosomes. Taken together, we show that both U6 snRNA and YTHDF2 are localized to endosomes to control TLR3 activation. Specifically, our findings demonstrate that both U6 snRNA and YTHDF2 enter endosomes through SIDT2- and dynamin-dependent intracellular trafficking pathway, while YTHDF2 is transported into endosome by the RNA transporter SIDT2 through binding to mA-methylated U6 snRNA. This model is supported by the following key findings (Fig. 5, supplementary Fig. S14): (1) SIDT2 knockdown or dynamin inhibition reduces entry of both U6 snRNA and YTHDF2 to endosomes, (2) to our best knowledge, YTHDF2 does not have an N-terminal peptide signal sequence that would enable YTHDF2 access to the endosomal lumen, (3) U6 knockdown reduces YTHDF2 levels in endosomes, (4) YTHDF2-N, a mutant lacking the m6A-RNA binding domain, showed reduced endosomal level, (5) knockdown of YTHDF2 or METTL16 increased U6 snRNA levels in endosomes, indicating that neither YTHDF2 nor mA methylation of U6 snRNA is required for U6 snRNA entry into endosomes, and (6) extracellular uptake for either U6 snRNA or YTHDF2 was not observed for endosome transportation. Taken together, these findings support a model that U6 snRNA is transported into endosomes at least in part by SIDT2 in a dynamin-dependent, mA methylation-independent, and YTHDF2-independent manner, while YTHDF2 entry into endosomes is dependent on U6 snRNA. UVB radiation causes damage to biomolecules including DNA, resulting in genomic instability. Besides genomic instability, UVB irradiation also induces inflammation, which manifests as sunburn, a critical enabling hallmark for tumorigenesis and aging. Our findings demonstrate that YTHDF2 inhibition enhances UVB-induced inflammation, leading to epidermal hyperplasia in mice. UV damage triggers a number of pathways such as the induction of COX-2 and proinflammatory cytokines in epidermal keratinocytes. Previously, we and others have demonstrated that COX-2 in keratinocytes acts as a pro-inflammatory and tumor-promoting factor in skin tumorigenesis and is upregulated in human skin cancer. UV-induced COX-2 upregulation can initiate the synthesis of the principle inflammation mediator PGE2 and thus increase keratinocyte proliferation and hyperplasia in the epidermis. Other cytokines such as TNF-α, IL-6, and IL-1β also play critical roles by contributing to the induction of COX-2 and other genes involved in epidermal inflammation and hyperplasia. In this study, we found that YTHDF2 inhibition increases UV-induced COX-2 expression in primary keratinocytes and cultured human keratinocytes and epidermal hyperplasia, and that the COX-2 inhibitor decreased cell proliferation in cells with YTHDF2 knockdown, supporting a critical role of COX-2 in increased cell proliferation caused by YTHDF2 inhibition. Our study also suggests that UVB might reduce YTHDF2 phosphorylation at S39 by promoting the interaction between YTHDF2 and MYPT1, which could be involved in the autophagic degradation of YTHDF2. Previous studies have shown that CHK1, a kinase activated by DNA damage and UV radiation, binds and phosphorylates MYPT1, leading to the recruitment of protein phosphatase 1β (PP1cβ) to dephosphorylate its substrate. It remains to be investigated whether UV radiation-induced CHK1 activation mediates MYPT1 phosphorylation, thus leading to increased MYPT1 binding to YTHDF2. In addition, YTHDF2 S39/T381 phosphorylation by EGFR/SRC/ERK signaling has previously been shown to promote YTHDF2 protein stability. It is possible that UVB-induced YTHDF2 dephosphorylation mediates YTHDF2 down-regulation by autophagy. In addition, we show that YTHDF2 S39 phosphorylation is critical for its interaction with CNOT1, U6 decay, and its endosomal trafficking, suggesting that YTHDF2’s function in promoting RNA decay and endosomal localization is regulated by its phosphorylation at S39. Future investigation is needed to determine whether phosphorylation at other sites regulates YTHDF2 activity. In addition to stabilizing U6 by inhibiting YTHDF2, UVB further increases U6 snRNA levels in YTHDF2-depleted cells (Fig. 2I-J), suggesting that UVB also induces U6 snRNA transcription, leading to TLR3 activation independent of YTHDF2. These YTHDF2-dependent and -independent mechanisms may serve as a multilayer defense mechanism to ensure the induction of a robust inflammatory and innate immune response in the skin under UVB stress to prevent microbe infection and maintain barrier function. Future investigation is warranted to fully map the molecular machinery that regulates U6 mA methylation, decay, and trafficking under baseline conditions, UVB stress, and other stresses. Taken together, our studies establish a critical role for YTHDF2 in recognizing mA methylation of U6 snRNA, thus increasing U6 decay and competing with mA U6 binding to TLR3 to suppress TLR3 activation in inflammation and tumorigenesis under baseline and UVB stress conditions. These findings may also open promising opportunities for targeting the YTHDF2/mA U6 axis in the intervention and treatment of epithelial cancers and autoimmune diseases, such as SLE. All research complied with relevant ethical regulations. Human tissue samples were obtained from US Biomax under protocols approved by Institutional Review Boards, with informed consent obtained from the donors, as documented by the vendor. All animal experiments were conducted in accordance with institutional guidelines and were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Chicago. Human tissue microarrays were obtained from US Biomax (Derwood, MD). According to the vendor’s documentation, all human tissues were collected under protocols approved by Institutional Review Boards, with informed consent obtained from the donors. The samples were provided in a fully de-identified form, ensuring that no personally identifiable information was accessible to the investigators. All research involving human-derived samples was conducted in accordance with the ethical principles set forth in the Declaration of Helsinki. Primary normal human epidermal keratinocytes (NHEK) were purchased from Invitrogen (Invitrogen, C-001-5 C) and Lonza (00192907). NHEK cells from Invitrogen were maintained in NHEK complete medium (Invitrogen, S-001-5) according to the manufacturer’s instructions, and NHEK cells from Lonza maintained in KGM Gold keratinocyte growth basal medium (Lonza, #00192151) and KGM Gold keratinocyte growth medium supplements and growth factors (Lonza, # 00192152). HaCaT cells (human keratinocyte, kindly provided by Dr. Fusenig), A431 cells (human squamous carcinoma cells, ATCC, CRL-1555), HEK-293T cells (human embryonic kidney cells, ATCC, CRL-3216), HeLa cells (WT and YTHDF2 knockout [KO] cells, kindly provided by Dr. Chuan He), and CHL-1 cells (melanoma cells, provided by the Comprehensive Cancer Center Core Facilities at the University of Chicago) were maintained in Dulbecco’s modified Eagle’s medium (Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum (Gibco), 100U/ml penicillin, and 100 µg/ml streptomycin (Invitrogen, Carlsbad, CA). Cells were treated with the following agents: Dynasore (80 μM, 16 h. Sigma, D7693), human cGAS inhibitor G140 (5 µM, 24 h. InvivoGen), Pladienolide B (500 nM, 3 h. MedChemExpress, HY-16399), followed by sham or UVB irradiation (20 mJ/cm, 6 h, unless indicated otherwise), or actinomycin D (2 μM. Sigma, SBR00013) post-sham or UVB irradiation. Cells were irradiated with sham or UVB irradiation (20 mJ/cm for HaCaT cells and 30 mJ/cm for other cell lines that exbibit decreased sensitivity to UVB stress) and then collected for analysis at 6 h or 24 h for the delayed UVB stress response such as inflammatory gene expression and YTHDF2 protein down-regulation, or 1 h for the early or direct UVB stress response such as YTHDF2-interacting proteins and YTHDF2 phosphorylation, unless indicated otherwise. All animal procedures were approved by the University of Chicago institutional animal care and use committee (IACUC). Athymic nude mice were obtained from Harlan Sprague-Dawley (now Envigo). For xenograft experiments, one million cells were injected subcutaneously into the right flanks of 6-week-old female nude mice. Tumor growth was monitored and measured weekly by a caliper, and tumor volume was calculated using the formula, Tumor volume (mm) = d X D/2, where d and D are the shortest and the longest diameters, respectively. Mice with wild type (WT, Ythdf2) & conditional skin-specific YTHDF2 homozygous knockout (DF2 cKO, K14Cre;Ythdf2), WT & ATG5 cKO (K14Cre;Atg5), WT & ATG7 cKO (K14Cre;Atg7), and WT & p62 cKO (K14Cre;p62) were generated and maintained on an SKH-1 hairless background, Both male and female mice (6–8 weeks old) were used for UVB irradiation experiments. The initial dose of UVB was 80 mJ/cm for the first week, followed by a weekly 10% increase until it reached 100 mJ/cm. WT and cKO were irradiated with UVB every other day 3 times a week, and tumor formation was recorded. For short UVB treatment, mice were irradiated with UVB irradiation (100 mJ/cm) 3 times every other day and skin samples were collected 24 h after the final UVB irradiation. All mice were housed in the University of Chicago Animal Resources Center (ARC) under specific pathogen-free conditions. ARC follows standard operating conditions, including a 12-h light/12-h dark cycle, ambient temperature of 21–23 °C, and relative humidity of 40–60%, with food and water provided ad libitum. Primary keratinocytes were isolated from neonatal mice as described previously in ref. . Briefly, WT (Ythdf2) and DF2 cHet (K14Cre; Ythdf2) neonatal mice were euthanized and rinsed briefly with sterile PBS containing penicillin-streptomycin, followed by a quick rinse with 70% ethanol and then submerging in fresh 70% ethanol for 10 minutes to allow thorough surface sterilization. Mice were then rinsed again with sterile PBS (with penicillin-streptomycin) and kept in fresh PBS on ice. Skin was collected on a 60 mm tissue culture dish containing 5 mL of cold sterile Ca-free 0.25% trypsin solution without EDTA (Invitrogen, Cat#15090-046, 10x stock, use PBS to dilute), with dermis facing down, and incubated at 4 °C for 15 to 24 h. Skin was then transferred to a dry, sterile tissue culture plate and epidermis was separated from dermis using a sterile pasteur pipette. Epidermis is minced and suspended in 6 ml of growth medium (10% FBS/DMEM, HaCaT medium) in a 15 mL tube, following by pipetting up and down to release keratinocytes from the epidermis. The suspension was then filtered through a sterile, 70 μm nylon filter (Thermo fisher, #22363548) into a fresh 50 mL tube to remove cornified sheets. The samples were then rinsed with 5 mL fresh growth medium, filtered through the same filter into the same 50 mL tube, and centrifuged at 160 g for 5 min at room temperature. Cells were resuspended in KGM® Gold Keratinocyte Growth Medium (Lonza, 00192060), counted, and seeded at 0.5–1 × 10 cells in 35 mm dishes. 48–72 h after plating, medium was removed and replaced with fresh culture medium for UVB irradiation experiments. WT and DF2 cHet cells were irradiated with UVB (30 mJ/cm) and then samples were collected at 6 h post-sham or -UVB irradiation for mA-IP qPCR analysis. pLKO.1 plasmids of shNC, shYTHDF2, shMETTL16, shATG5, shATG7, shp62, and shTLR3_GFP were obtained from Sigma. pLenti plasmids of overexpression of YTHDF2 were purchased from Applied Biological Materials Inc. (Abm). Lentivirus was produced by co-transfection into HEK-293T (human embryonic kidney) cells with lentiviral constructs together with the pCMVdelta8.2 packaging plasmid and pVSV-G envelope plasmid using GenJet™ Plus DNA In Vitro Transfection Reagent (Signagen, Ijamsville, MD). Virus-containing supernatants were collected 24–48 h after transfection and used to infect recipients. Target cells were infected in the presence of Polybrene (8 μg/ml) (Sigma-Aldrich, St. Louis, MO) and selected with puromycin (Santa Cruz Biotechnology, Santa Cruz, CA) at 1 μg/ml for 6 days. YTHDF2 (WT, N, and C) plasmids were kindly provided by Dr. Chuan He. The pCMV3-C-FLAG-TLR3 construct was obtained from Sino Biological (Catalog# HG10190-CF). Primers for this study were designed individually based on the specific deletions. The YTHDF2 mutants (S39A and S39D) and TLR3 deletion mutants were generated using a QuikChange Site-Directed Mutagenesis kit (Catalog# 200518), according to the manufacturer’s instructions. Sequences of the primers for cloning of TLR3 or YTHDF2 mutants are shown in Supplementary Table S2. Newly generated plasmids are available upon request from the corresponding authors. Cells were transfected with siRNA targeting negative control (siNC), an individual gene, or the combination of SIDT2, CNOT1, TLR3, TLR7, TLR8, MDA5, RIG-1, YTHDF2, U6, METTL16, TLR9, or MYPT1 (Dharmacon, Lafayette, CO). siU6, U6 snRNA, and mA-modified U6 snRNA (mA43) were purchased from Integrated DNA Technologies (IDT, Coralville, IA). tRNA (Roche, TRNABRE-RO SKU10109517001) was used as the control for structured RNA. RNAs were transfected using PepMute™ siRNA Transfection Reagent (Signagen, Ijamsville, MD), according to the manufacturer’s instructions. Quantitative real-time PCR assays (qPCR) were performed using a CFX Connect real-time system (Bio-Rad, Hercules, CA) with Bio-Rad iQ SYBR Green Supermix (Bio-Rad, Hercules, CA). cDNAs were prepared by using iScript Reverse Transcription Supermix for RT-qPCR (Bio-Rad, Hercules, CA) for mRNAs, or miRNA All-In-One cDNA Synthesis Kit (Applied Biological Materials Inc., BC, Canada) for non-coding RNAs. The threshold cycle number (CQ) for each sample was determined in triplicate or quadruplicate. The CQ for values for YTHDF2, TNF-α, IL-6, COX-2, IL-1β, GM-CSF, IL-8, MMP-9, VEGF, IL-16, IL-1α, FOS, JUN, SOX4, SOX9, CD74, U6 snRNA, ATG5, ATG7, p62, SIDT2, CNOT1, METTL16, and TLR3 were normalized against GAPDH, ACTB, 18S rRNA, or HPRT1, which are specifically indicated in figure legends. Sequences of the primers for qPCR are shown in Supplementary Table S3. RNA stability was assessed using the transcriptional inhibitor actinomycin D (2 μM). Cells were harvested at different time points after treatment with actinomycin D. Total RNA was isolated using an RNeasy plus mini kit (QIAGEN, Hilden, Germany) or TRIzol, following the manufacturer’s protocol. The HPRT1 housekeeping gene was used as a loading control, as HPRT1 mRNA does not contain mA modifications, is not bound by YTHDsF2, and is rarely affected by actinomycin D treatment. Immunofluorescence analysis was performed as described previously in refs. . For immunofluorescence analysis of cells, cells were fixed with 4% paraformaldehyde/PBS for 30 min and permeabilized with 0.5% Triton X-100 (Sigma-Aldrich, T8787)/PBS for 20 min. Cells were then washed with PBS with 0.05% Triton X-100. The samples were then incubated with a blocking solution of 3% albumin from chicken egg white (Sigma-Aldrich, A5503) in PBS for 1 h. For immunofluorescence analysis of tissues, formalin-fixed, paraffin-embedded tissue sections were pre-treated by antigen retrieval and incubated with a blocking solution of 3% albumin from chicken egg white (Sigma-Aldrich, A5503) in PBS for 1 h. After removal of the blocking solution, samples were incubated with primary antibodies at 4 °C overnight, and then the samples were washed three times with 0.1% Triton X-100/PBS for 10 min at room temperature (RT). Samples were then incubated with fluorochrome-conjugated secondary antibodies for 1 h at RT and washed three times with 0.1% Triton X-100/PBS for 10 min at room temperature. Cells were then fixed in Prolong Gold Antifade with DAPI (Invitrogen, P36931), Fluoromount™ Aqueous Mounting Medium (Sigma-Aldrich, F4680), or Anti-Fade Fluorescence Mounting Medium (Abcam, ab104135), and were then observed under a fluorescence microscope (Olympus IX71, Olympus Life Science, Japan). For confocal microscopy, cells were imaged with the SoRa Marianas Spinning Disk Confocal microscope (Intelligent Imaging Innovations, 3i). Hematoxylin and eosin (HE) staining of tissues and immunohistochemical analysis of YTHDF2 (Proteintech, #247441-1-AP) were performed by the Human Tissue Resource Center (HTRC) core facility at the University of Chicago. Epidermal thickness was measured on HE stained sections, as described previously. Two investigators independently scored the staining intensity blindly, as 3 (strong), 2 (medium), 1 (weak), and 0 (negative). ImageJ (NIH) was also used for analyzing cells and tissue. Information for the antibodies for the IF staining are shown in Supplementary Table S4. Cells were fixed with 4% paraformaldehyde (PFA) for 30 minutes at room temperature (RT). Permeabilization was performed using 0.5% Triton X-100 in PBS supplemented with RNase inhibitor (40 U/mL) for 30 min at 4 °C. Samples were then washed with Buffer A (20% Buffer A (Cat:SMF-WA1-60, BioSearch)+ 10% formamide) for 5 min at RT. Hybridization was carried out overnight at 37 °C using a 150 nM DIG-labeled U6 probe (U6 probe: 5’-CACGAATTTGCGTGTCATCCTT-3’, IDT) in hybridization solution (Cat:SMF-HB1-10, BioSearch) containing 10% formamide and the RNase inhibitor (40 U/mL). After hybridization, samples were thoroughly washed with PBS to prepare for immunofluorescence staining. Blocking was performed using goat serum for 30 minutes at RT. The anti-Rab7 (ab137029) primary antibody was diluted in blocking buffer and incubated with the samples for 2 h at RT. After washing with PBS, DIG was detected using the anti-DIG Alexa Fluor 488 antibody (1:50), and Rab7 was visualized using an Alexa Fluor 594-conjugated secondary antibody by incubation for 1 h at RT. Finally, samples were washed with PBS, counterstained with DAPI, and then mounted for imaging. For mouse tissue, formalin-fixed, paraffin-embedded (FFPE) tissue sections were first subjected to antigen retrieval. After washing, U6 probe hybridization was performed following the same protocol as for cells. After hybridization, tissue sections were washed with 0.1% Triton X-100 in PBS. Blocking was then performed using 3% chicken egg white albumin (Sigma-Aldrich, A5503) in PBS for 1 h at RT. Primary antibodies were applied, followed by incubation with the corresponding Alexa Fluor-conjugated secondary antibodies. Tissue samples were then fixed in Prolong Gold Antifade with DAPI (Invitrogen, P36931). Cells and tissue samples were then imaged with the SoRa Marianas Spinning Disk Confocal microscope (Intelligent Imaging Innovations, 3i) and a fluorescence microscope (Olympus IX71, Olympus Life Science, Japan), respectively. Cells were washed twice with ice-cold PBS and lysed with Cell Lysis Buffer (Cell Signaling, #9803) or RIPA buffer (Cell Signaling, #9806) containing inhibitors for proteases and phosphatases. Cell lysates were resolved by SDS-PAGE and transferred onto nitrocellulose membranes followed by blotting. The information for the antibodies for immunoblotting is shown in Supplementary Table S5. Single cell suspensions were prepared from spleen and skin tissues. Live/dead labeling was performed before the cell surface staining, using a Zombie NIR™ Fixable Viability Kit (Biolegend; catalog number 423106) diluted 1:1,000 in PBS for 10 min at room temperature in dark. Cells were labeled with specific fluorescence-conjugated antibodies to CD45, CD3, CD4, CD8, MHCII, CD11b, CD103, Ly6C, and/or Ly6G for 20 min on ice. Intracellular cytokine staining was performed using an eBioscience Intracellular Fixation & Permeabilization Buffer Set (catalog number 88-8824-00) according to the manufacturer’s instructions. Antibodies against FOXP3 (clone; NRRF-30) were used for regulatory T cell staining. Flow cytometric analysis was performed on a Fortessa 4–15 (BD Biosciences) and Attune NxT (Thermo Fisher Scientific) with Flowjo V10.6.1 used for analysis. The information for the antibodies is shown in Supplementary Table S6. The gating strategy is shown in Supplementary Fig. S18. 100 μg to 150 μg total RNA was extracted from cells using TRIzol following the manufacturer’s protocol. mRNA was purified using a Dynabeads mRNA DIRECT Kit following the manufacturer’s protocols. 1 μg mRNA was sonicated to ∼100 nt and mA containing mRNA fragments was enriched with an EpiMark N6-Methyladenosine Enrichment Kit following the manufacturer’s protocols. Finally, RNA was isolated using RNA Clean and Concentrator (Zymo Research) and subjected to library preparation with a TruSeq stranded mRNA sample preparation kit (Illumina). The mA enrichment in mRNA and U6 snRNA was analyzed using ultra–high-performance liquid chromatography-tandem mass spectrometry (UHPLC-MS/MS). Total RNA was extracted from cells using TRIzol following the manufacturer’s protocol. U6 snRNA was isolated as described previously. The mA in polyadenylated RNA was quantified using an Agilent 6460 LC-MS/MS spectrometer as described previously General pre-processing of reads: all samples were sequenced by Illumina Hiseq4000 with single end 80 bp read length. The adapters were removed by using cutadapt for mA-seq. RefSeq Gene structure annotations were downloaded from the UCSC Table Browser. Reads were aligned to the reference genome (hg38) using HISAT2. DESeq2 was used for differential gene expression analysis. Lowly expressed genes were pre-filtered out using reads count threshold of 10. An adjusted p value (DESeq2 default) threshold of 0.05 was used for all differential expression analysis. Aligned reads were extended to 150 bp (average fragment size) and converted from genome-based coordinates to isoform-based coordinates. The method used for peak calling was adapted from published work with modifications. To call mA peaks, the longest isoform of each gene was scanned using a 100 bp sliding window with a 10 bp step. To reduce bias from potential inaccurate gene structure annotation and the arbitrary usage of the longest isoform, windows with read counts of less than 1/20 of the top window in both mA-IP and the input sample were excluded. For each gene, the read counts in each window were normalized to the median count of all windows of that gene. A Fisher exact test was used to identify the differential windows between IP and input samples. The window was called as positive if the FDR < 0.01 and log2(Enrichment Score) >= 1. Overlapping positive windows were merged. The following four numbers were calculated to obtain the enrichment score of each peak (or window): read counts of the IP samples in the current peak/window (a), median read counts of the IP sample in all 100 bp windows on the current mRNA (b), read counts of the input sample in the current peak/window (c), and median read counts of the input sample in all 100 bp windows on the current mRNA (d). The enrichment score of each window was calculated as (a×d)/(b×c). The data have been deposited in the GEO repository with accession number GSE145925. Real-time quantitative PCR (qPCR) was performed to assess the relative mA abundance of the selected U6 snRNA and mRNA in mA antibody IP samples and input samples, as described previously. Briefly, total RNA was isolated with an RNeasy mini kit or TRIzol. While 500 ng RNA was saved as an input sample, the rest of the RNA was used for mA IP. 100 μg RNA was diluted into 500 μL IP buffer (150 mM NaCl, 0.1% NP-40, 10 mM Tris, pH 7.4, 100 U RNase inhibitor) and incubated with the mA antibody (Synaptic Systems, Goettingen, Germany). The mixture was rotated at 4 °C for 2 hours, then Dynabeads® Protein A (Thermo Fisher Scientific, Waltham, MA) coated with BSA was added into the solution and rotated for an additional 2 hours at 4 °C. After washing with the IP buffer with RNase inhibitors four times, the mA IP portion was eluted with elution buffer (5 mM Tris-HCL pH 7.5, 1 mM EDTA pH 8.0, 0.05% SDS, and 4.2 μl Proteinase K (20 mg/ml). The final eluted RNA was concentrated with an RNA Clean & Concentrator-5 kit (Zymo Research, Irvine, CA). The same amount of the concentrated IP RNA or input RNA from each sample was used for the cDNA library. The RNA expression was determined by the number of amplification cycles (Cq). The relative mA levels in genes were calculated by normalizing the mA levels (mA IP) against the expression of each gene (Input). Endo-IP was performed as described previously. Cells were harvested on ice by scraping in 2 ml DPBS and pelleted by centrifugation at 1000 x g for 2 min at 4 °C. Supernatant was discarded, and cell pellet was washed once with 1 ml of KPBS buffer (25 mM KCl, 100 mM potassium phosphate, pH 7.2), followed by centrifugation at 1000xg for 2 min at 4 °C. Cells were then resuspended in 500 µL of KPBS supplemented with the protease inhibitors (Thermo Fisher, 78442) and lysed with 30 strokes with a 2 ml Dounce homogenizer on ice. Lysed cells were then subject to centrifugation at 1000xg for 5 min at 4 °C, and the post-nuclear supernatants (PNS) were transferred to new tubes on ice. The α-FLAG M2 magnetic beads were washed with 1 mL KPBS buffer with the protease inhibitors and resuspend in the same buffer. Resuspended bead slurry was added to each PNS and incubated at 4 °C for 1 h with gentle rotation. Beads were eluted by addition of 120 µL 0.5% NP-40 in KBPS with the protease inhibitors and incubated for 30 min at 4 °C with gentle rotation, followed by immunoblot analysis. The interaction between protein and RNA was assessed using the Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore Co., Burlington, MA), according to the manufacturer’s instructions. Briefly, specific antibodies attached to magnetic beads were incubated with cell lysates. Then, the protein-RNA complexes were isolated, and the RNAs were extracted by the phenol-chloroform RNA extraction method. The relative interaction between the target proteins and RNAs was analyzed by qPCR with normalization against the corresponding input. Endosomes were isolated using the Minute™ Endosome Isolation and Cell Fractionation Kit (Invent biotech, Plymouth, MN), according to the manufacturer’s instructions. Briefly, the final isolated pellets were lysed with 1X Cell Lysis Buffer (Cat #9803, Cell Signaling, Danvers, MA) containing Invitrogen™ SUPERase•In™ RNase Inhibitor (Invitrogen, Carlsbad, CA). For RNA isolation from endosomes, TRizol or an RNA Clean & Concentrator-5 kit (Zymo Research, Irvine, CA) was used according to the manufacturer’s instructions. The relative levels of target RNAs in endosomes were normalized against HPRT1 mRNA, 18S rRNA, or the total U6 snRNA in the input samples. The nuclear, cytoplasmic, or total RNAs were isolated using the RNA Subcellular Isolation Kit (Active Motif, Carlsbad, CA) according to the manufacturer’s instructions. The relative RNA levels were measured by qPCR analysis. The nuclear and cytoplasmic proteins were isolated using the NE-PER™ Nuclear and Cytoplasmic Extraction Reagents kit (Thermo Scientific, 78833) according to the manufacturer’s instructions, followed by immunoblotting analysis. tRNA (Sigma-Aldrich, St. Louis, MO), U6 (IDT, Coralville, IA), mA43 U6 (IDT, Coralville, IA), or Poly I:C (Sigma-Aldrich, St. Louis, MO) were labeled with an RNA 3’ End Biotinylation Kit (Thermo Fisher Scientific, Waltham, MA). The quality and quantity of the biotin-labeled RNAs were confirmed using the Chemiluminescent Nucleic Acid Detection Module Kit (Thermo Fisher Scientific, Waltham, MA) according to the manufacturer’s instructions. Streptavidin-coated magnetic beads, Dynabeads™ MyOne™ Streptavidin C1 (Thermo Fisher Scientific, Waltham, MA), were incubated with the prepared IP samples, including biotin-labeled target RNAs with or without recombinant proteins such as human YTHDF2 (Active Motif, Carlsbad, CA) or human TLR3 (Abcam, Waltham, MA). 1X Cell Lysis Buffer (Cat# 9803, Cell Signaling, Danvers, MA) was used as an IP reaction buffer for cell lysates or recombinant proteins. The relative binding affinity between biotin-labeled RNAs and target proteins was measured by immunoblotting. All experiments were conducted using established cell lines and immunoprecipitation assays; ethical approval was not applicable. To characterize proteins associated with YTHDF2, immunoprecipitates were analyzed by mass spectrometry. A total of three samples were submitted: one IgG control and two YTHDF2 immunoprecipitates (treated and non-treated). Sample preparation, LC–MS/MS acquisition, and data processing were performed by MS Bioworks (MS Bioworks LLC, Ann Arbor, MI). For sample preparation, half of each immunoprecipitate was subjected to SDS-PAGE using a 10% Bis-Tris NuPAGE gel (Invitrogen) with the MES buffer system. Gels were run to ~2 cm, and the corresponding region was excised into ten equal slices. Gel slices were processed for in-gel digestion on a ProGest robot (DigiLab) using the following protocol: washing with 25 mM ammonium bicarbonate and acetonitrile, reduction with 10 mM dithiothreitol at 60 °C, alkylation with 50 mM iodoacetamide at room temperature, digestion with sequencing-grade trypsin (Promega) at 37 °C for 4 h, and quenching with formic acid. Supernatants were analyzed directly without further cleanup. Half of each digested sample was analyzed by nanoLC-MS/MS using a Waters M-Class HPLC system coupled to a Thermo Fisher Orbitrap Fusion Lumos mass spectrometer. Peptides were first loaded onto a trapping column and then separated on a 75 μm analytical column packed with Luna C18 resin (Phenomenex) at 350 nL/min. Elution was performed with a linear gradient of acetonitrile in 0.1% formic acid. The mass spectrometer was operated in data-dependent acquisition mode, with full MS scans acquired at 60,000 FWHM and MS/MS scans at 15,000 FWHM, using a 3 s cycle time. Data analysis was performed using Mascot (Matrix Science) against the SwissProt Human database (concatenated forward and reverse sequences plus common contaminants). Search parameters were as follows: enzyme specificity set to trypsin/P with up to two missed cleavages; carbamidomethylation of cysteine as a fixed modification; oxidation of methionine, N-terminal acetylation, pyro-glutamate formation (N-terminal glutamine), deamidation (N,Q), and phosphorylation (S,T,Y) as variable modifications; monoisotopic mass values; precursor ion tolerance of 10 ppm; and fragment ion tolerance of 0.02 Da. Mascot DAT files were parsed into Scaffold (Proteome Software) for validation and filtering. Data were filtered at 1% peptide and protein false discovery rate (FDR), requiring at least two unique peptides per protein The anti-YTHDF2 pS39 antibody was generated by Abclonal Technology. One-and-a-half-year-old New Zealand rabbits (2.5 kg), housed under SPF conditions, were subcutaneously injected with 700 μg of a modified antigen peptide (EPYL(S-p)PQAR-C-KLH) emulsified with Complete Freund’s Adjuvant (CFA) for the primary immunization, followed by five booster injections of 350 μg of the same peptide emulsified with Incomplete Freund’s Adjuvant (IFA) at 1-, 2-, and 3-week intervals. Terminal bleeds were collected after the final immunization. Polyclonal antibodies were purified from the terminal bleeds by antigen affinity chromatography using a column conjugated with the modified peptide (EPYL(S-p)PQAR-C), followed by depletion using a column conjugated with the non-modified peptide (EPYLSPQAR-C). Gene ontology (GO) enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using Metascape (https://metascape.org/), a web-based portal designed to provide a comprehensive gene list annotation and analysis resource, as described previously, using the default background gene list (all genes) provided by Metascape for initial pathway screening. Quantitative Venn diagrams were generated using the following web tool: https://www.deepvenn.com/. Genes with an adjusted p value less than 0.05 by DESeq2 were used for pathway analysis and Venn diagram analysis. Migration assays were performed as described previously. Briefly, 5 × 10 cells were suspended in 150 μl of serum-free medium and seeded onto 8-mm Pore Transwell Inserts (Corning, Corning, NY) for the migration assay. The lower chamber was filled with 900 μl of complete medium. Cells on the Transwell Inserts were then fixed with 4% paraformaldehyde/PBS for 30 min. Subsequently, fixed cells were stained with hematoxylin solution (Sigma-Aldrich, St. Louis, MO) for 1 h. After wiping off cells on the upper side of the filter on the Transwell Inserts using cotton swabs, microphotograms of the cells migrated onto the lower side of the filter were taken using light microscopy. Cells migrated onto the lower side of the filter were manually counted from the microphotograms. Mean cell numbers were quantified from about ten randomly selected squares (500 μm X 500 μm) per Transwell insert. Cell proliferation was assessed using the Cell Counting Kit-8 (CCK-8; Sigma-Aldrich, St. Louis, MO) according to the manufacturer’s instructions, with minor modifications. Briefly, cells were seeded in 24-well plates and incubated with CCK-8 solution for 1 h. After incubation, the culture medium containing the CCK-8 reagent was transferred to 96-well plates, and absorbance was measured at 450 nm using a microplate reader. The use of 24-well plates, instead of the standard 96-well format recommended in the kit manual, helped minimize several technical issues encountered during long-term culture, including high cell density effects, difficulties in handling very small volumes, and uneven medium evaporation at the plate edges. In addition, the larger well format reduced the risk of cell detachment during medium changes, which can easily occur in smaller wells due to suction effects. RNA Seq or microarray data used to assess the role of YTHDF2 in autoimmune diseases were originally found using the ADEx: Autoimmune Diseases Explorer database (https://adex.genyo.es). Data files were accessed and downloaded using the Gene Expression Omnibus (GEO). Data was imported into Prism for statistical analyses. Datasets analyzed were GSE110169_SLE (microarray), GSE50772 (microarray), GSE72509 (RNA-Seq), GSE11907_T1D (microarray), GSE110169 RA (microarray), GSE56649 (microarray), GSE45291 (microarray), GSE89408 (RNA-seq), GSE104174 (RNA-Seq), GSE124073 (RNA-Seq), and GSE51092 (microarray). Statistical analyses were performed using Prism v10 (GraphPad Software). Data are presented as the mean of at least three independent experiments and were analyzed using a two-tailed unpaired Student’s t-test or a Mann–Whitney U test, as indicated. Error bars indicate the SDs or SEs of the means as specified. P < 0.05 was considered statistically significant. All experiments were independently repeated at least three times with similar results. Representative images (e.g., micrographs, blots) are shown from experiments that were repeated independently with comparable outcomes. Sample sizes (n) refer to biologically independent replicates (e.g., independent mice, cell cultures, or tissue samples) unless otherwise indicated. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
PMC12116388
Cell-intrinsic metabolic phenotypes identified in patients with glioblastoma, using mass spectrometry imaging of C-labelled glucose metabolism
Transcriptomic studies have attempted to classify glioblastoma (GB) into subtypes that predict survival and have different therapeutic vulnerabilities. Here we identified three metabolic subtypes: glycolytic, oxidative and a mix of glycolytic and oxidative, using mass spectrometry imaging of rapidly excised tumour sections from two patients with GB who were infused with [U-C]glucose and from spatial transcriptomic analysis of contiguous sections. The phenotypes are not correlated with microenvironmental features, including proliferation rate, immune cell infiltration and vascularization, are retained when patient-derived cells are grown in vitro or as orthotopically implanted xenografts and are robust to changes in oxygen concentration, demonstrating their cell-intrinsic nature. The spatial extent of the regions occupied by cells displaying these distinct metabolic phenotypes is large enough to be detected using clinically applicable metabolic imaging techniques. A limitation of the study is that it is based on only two patient tumours, albeit on multiple sections, and therefore represents a proof-of-concept study.GB is the most common primary adult brain cancer. Transcriptomic analyses have attempted to classify GB into subtypes that could predict treatment response, and a recent study that used a pathway-based classification defined metabolism-associated subtypes with distinct therapeutic vulnerabilities. These included a mitochondrial subtype, which is associated with a more favourable clinical outcome and is sensitive to inhibitors of oxidative phosphorylation, and a glycolytic, plurimetabolic subtype that is resistant to multiple treatment types. An important question is the extent to which the metabolism displayed by tumour cells in vivo is cell-intrinsic and how much is defined by the tumour microenvironment (TME). We addressed this question by using mass spectrometry imaging (MSI) of rapidly excised tumour sections from patients with GB who were infused with [U-C]glucose immediately before surgery to image tumour cell metabolic activity in vivo and from a spatial transcriptomic analysis of adjacent sections. We infused three patients, two with GB and a third with an adenocarcinoma metastasis, with [U-C]glucose and performed MSI on rapidly excised tumour tissue that was dissected during tumour debulking surgery (Fig. 1a). We sampled 16 regions from two patients with GB and seven regions from the patient with a metastasis, the latter containing samples from normal-appearing cortex (Fig. 1b). There were no significant differences in lactate and glutamate C labelling between the tumour mass and tumour margin (Extended Data Fig. 1a).Fig. 1Intra-operative freezing rapidly arrests tissue metabolism and allows visualisation of metabolic activity.a, Patients were infused intra-operatively with [U-C]glucose and underwent multi-regional tumour sampling followed by rapid freezing in liquid nitrogen (<5 s). Tumour sections (10 μm) were analysed using DESI-MSI and MALDI-MSI, and contiguous sections were analysed by IMC. Blood samples were collected before, during and after infusion for plasma liquid chromatography–MS (LC–MS) analysis. Created in BioRender.com. b, Coronal MR images from the three patients who were infused, showing the locations of the sampled regions (blue stars). The number of regions sampled is shown for each patient; each region had three to six pieces of snap-frozen tissue. c, Plasma glucose (Glc) fractional labelling in the three infused patients. Tumour sampling commenced at 90 min. d, Glucose and metabolite fractional labelling in the tumour tissue: Glc ([U-C]glucose/([U-C]glucose + [U-C]glucose) (GB1 vs GB2, P = 0.0001; GB2 vs Metastasis, P = 0.004); lactate (Lac) ([U-C]lactate/([U-C]lactate + [U-C]lactate) (GB1 vs GB2, P < 0.0001; GB2 vs Metastasis, P = 0.0002; GB1 vs Metastisis, P < 0.0001); and glutamate (Glu) ([C2]glutamate/([C2]glutamate + [U-C]glutamate) (GB1 vs GB2, P = 0.0069; GB1 vs Metastasis, P = 0.0008). e, ATP/ADP and PCr signal intensity ratios in each tumour section (PCr/ATP GB1 vs Metastasis, P = 0.0001; GB2 vs Metastasis, P = 0.0033). f, Representative tumour sections from each patient showing PCr signal intensities and ATP/ADP signal intensity ratios. g, Representative sections from the metastasis and from normal-appearing brain (data reproduced in three additional sections of normal cortex and nine of the metastasis); (i) Pan-cytokeratin (CK)-positive (green) adenocarcinoma cells formed tubules around a central Collagen I-positive vessel (red). The tubules are surrounded by vimentin-positive (pink) cortex. (ii) Normal-appearing cortical tissue was obtained from the margin of the metastasis and showed vimentin-positive cells and an absence of adenocarcinoma cells. h, Relative abundance of [U-C]lactate (GB1 vs GB2, P < 0.0001; GB1 vs Metastasis, P = 0.0016) and [C2]glutamate (GB1 vs Metastasis, P = 0.0394) in tumour sections from GB1 (n = 22), GB2 (n = 12), the adenocarcinoma metastasis (n = 9) and normal-appearing cortex (n = 3). Data are means; error bars, s.d.; every dot is a tissue section. Asterisks refer to P values obtained from a one-way ANOVA followed by Tukey’s multiple comparisons test (*P < 0.05, **P < 0.005, ***P < 0.0005, ****P < 0.00005).Source data a, Patients were infused intra-operatively with [U-C]glucose and underwent multi-regional tumour sampling followed by rapid freezing in liquid nitrogen (<5 s). Tumour sections (10 μm) were analysed using DESI-MSI and MALDI-MSI, and contiguous sections were analysed by IMC. Blood samples were collected before, during and after infusion for plasma liquid chromatography–MS (LC–MS) analysis. Created in BioRender.com. b, Coronal MR images from the three patients who were infused, showing the locations of the sampled regions (blue stars). The number of regions sampled is shown for each patient; each region had three to six pieces of snap-frozen tissue. c, Plasma glucose (Glc) fractional labelling in the three infused patients. Tumour sampling commenced at 90 min. d, Glucose and metabolite fractional labelling in the tumour tissue: Glc ([U-C]glucose/([U-C]glucose + [U-C]glucose) (GB1 vs GB2, P = 0.0001; GB2 vs Metastasis, P = 0.004); lactate (Lac) ([U-C]lactate/([U-C]lactate + [U-C]lactate) (GB1 vs GB2, P < 0.0001; GB2 vs Metastasis, P = 0.0002; GB1 vs Metastisis, P < 0.0001); and glutamate (Glu) ([C2]glutamate/([C2]glutamate + [U-C]glutamate) (GB1 vs GB2, P = 0.0069; GB1 vs Metastasis, P = 0.0008). e, ATP/ADP and PCr signal intensity ratios in each tumour section (PCr/ATP GB1 vs Metastasis, P = 0.0001; GB2 vs Metastasis, P = 0.0033). f, Representative tumour sections from each patient showing PCr signal intensities and ATP/ADP signal intensity ratios. g, Representative sections from the metastasis and from normal-appearing brain (data reproduced in three additional sections of normal cortex and nine of the metastasis); (i) Pan-cytokeratin (CK)-positive (green) adenocarcinoma cells formed tubules around a central Collagen I-positive vessel (red). The tubules are surrounded by vimentin-positive (pink) cortex. (ii) Normal-appearing cortical tissue was obtained from the margin of the metastasis and showed vimentin-positive cells and an absence of adenocarcinoma cells. h, Relative abundance of [U-C]lactate (GB1 vs GB2, P < 0.0001; GB1 vs Metastasis, P = 0.0016) and [C2]glutamate (GB1 vs Metastasis, P = 0.0394) in tumour sections from GB1 (n = 22), GB2 (n = 12), the adenocarcinoma metastasis (n = 9) and normal-appearing cortex (n = 3). Data are means; error bars, s.d.; every dot is a tissue section. Asterisks refer to P values obtained from a one-way ANOVA followed by Tukey’s multiple comparisons test (*P < 0.05, **P < 0.005, ***P < 0.0005, ****P < 0.00005). Source data Tissue sampling began 90 min after the start of infusion, during which time the plasma glucose fractional enrichment reached a steady state (Fig. 1c). Fractional labelling of lactate, an end product of the glycolytic pathway, and of [C2]glutamate, which is labelled via α-ketoglutarate in the tricarboxylic acid (TCA) cycle (Extended Data Fig. 1b) in GB1 and GB2, were less than that of tumour glucose (Fig. 1d), suggesting that neither had reached isotopic steady state and were therefore a measure of glycolytic and TCA cycle activity, respectively. There was a direct correlation between [C2]glutamate signal intensities and the intensities of the TCA cycle intermediates [C2]malate, [C2]fumarate and [C2]succinate (Extended Data Fig. 1c). The signal from ceramide-1-phosphate (752.596 m/z) confirmed minimal variation in instrument performance during acquisition and uniformity of tissue preparation (Extended Data Fig. 1d). Mass errors on the metabolite ions and the mass isotopologue distributions of all detectable labelled metabolites are given in Supplementary Tables 1 and 2, respectively. The sampling technique rapidly arrested tissue metabolism, as indicated by the ATP/ADP and phosphocreatine PCr/ATP ratios (Fig. 1e,f), which decline rapidly in hypoxic or ischaemic brain tissue. Preservation of the ATP, ADP and PCr concentrations was also demonstrated using P nuclear magnetic resonance (NMR) measurements (Extended Data Fig. 2a,b). The metabolite C labelling observed was assumed, therefore, to be similar to that present in vivo. The ATP and PCr signal intensities were similar in the two GB tumours but significantly higher than those in the metastasis (Extended Data Fig. 2c). The concentrations of plasma amino acids and lactate did not change significantly during the infusion protocol. The plasma concentrations of C-labelled lactate and glutamine, although less than 10% of the unlabelled concentrations, could have contributed to some of the labelled lactate and, via glutamine, the labelled glutamate observed in the tumour sections (Extended Data Fig. 2d and Supplementary Table 3). The metastatic adenocarcinoma occupied small, discrete areas surrounded by gliotic and normal-appearing brain parenchyma (Fig. 1g) and therefore provided apparently normal brain tissue for comparative analysis. We segmented the mass spectrometry (MS) images using seven C-labelled metabolites from glycolysis ([U-C]pyruvate, [U-C]lactate) and the TCA cycle ([C2]fumarate, [C2]succinate, [C2]malate, [C2]glutamate and [C2]glutamine). This separated the tumour, gliotic and normal-appearing brain parenchyma and agreed with tissue classification performed by a histopathologist (Extended Data Fig. 2e). Next, we assessed glycolytic activity and TCA cycle activity in the tumours and in normal-appearing brain. [U-C]lactate signals were significantly lower in GB1 than in GB2 and the metastasis, whereas the [C2]glutamate signals in GB1 were significantly higher than in the metastasis (Fig. 1h), reflecting higher glycolytic and lower TCA cycle activities in the metastasis. Glutamate labelling in the GB tumours was comparable to that in normal brain. We segmented the GB MS images using the same seven C-labelled metabolites from glycolysis and the TCA cycle. We reasoned that the activity of these two pathways could result in four cellular states; however, the MSI spectra did not contain a high glycolytic, high TCA cycle phenotype, and images were better segmented assuming only three metabolic states (Fig. 2a and Extended Data Fig. 3a): state 3 (high glycolysis, low TCA cycle), state 2 (low glycolysis, high TCA cycle) and state 1 (low activity in both pathways). Segmentation assuming three or four metabolic states showed that these states occupied distinct and spatially extensive regions (Fig. 2b and Extended Data Fig. 3b–e). As observed previously from transcriptomic data, each tumour had a predominant metabolic phenotype, with GB1 containing more of state 2 and GB2 containing more of state 3 (Extended Data Fig. 3f). Nevertheless, regions occupied by one of these three metabolic states co-existed in both tumours, sometimes within a single tumour section. Spatial RNA sequencing of adjacent tumour sections and segmentation using Hallmark oxidative and glycolytic gene sets also found three metabolic states: glycolytic, oxidative and mixed (Extended Data Fig. 4a), which showed a strong correlation with the MSI data (Fig. 2c) and an overall concordance of 65% (Extended Data Fig. 4b).Fig. 2Identification of metabolic phenotypes in GB tumour sections.a, Heatmap showing the average intensities of the seven metabolites used for segmentation of the three metabolic regions. b, Representative example of three tumour sections segmented into three clusters. The region containing state 3 cells (blue) showed high [U-C]lactate and [U-C]pyruvate labelling and was considered to have a glycolytic phenotype. The region containing state 2 cells (yellow) showed high labelling of [C2]glutamate and was considered to have an oxidative phenotype. The region containing state 1 cells (red) showed low labelling of glycolytic and TCA cycle metabolites. c, Metabolic segmentation based on spatial RNA sequencing of sections contiguous with those shown in b. Blue spots correspond to a glycolytic phenotype, and yellow spots correspond to an oxidative phenotype based on Hallmark gene sets. Red spots correspond to cells with low glycolytic and oxidative gene signatures. d, ATP/ADP and PCr/ATP signal intensity ratios (mean; error bars, s.d.) in normal-appearing brain (n = 4) and in the three metabolically defined regions (region 1, n = 22; region 2, n = 14; region 3, n = 17); every dot is a unique region. No statistical significance was identified using one-way ANOVA with Tukey’s multiple comparisons test. e, Redox status (mean; error bars, s.d.) in normal-appearing brain tissue and GB regions quantified using AsA:DHA, GSH:GSSG and [U-C]lactate/pyruvate ratios (normal brain, n = 4; region 1, n = 22; region 2, n = 14; region 3, n = 17). No statistical significance was identified using one-way ANOVA with Tukey’s multiple comparisons test. f, Quantification of Ki67 cells, immune cells and blood vessels (mean; error bars, s.d.) in the three metabolically defined regions based on immunohistochemical analysis (region 1, n = 17; region 2, n = 4; region 3, n = 8). No statistical significance was identified using one-way ANOVA with Tukey’s multiple comparisons test. g, Spatial co-assignment of deconvolved tumour/TME signals (first row) and MSI labels (second row). h, Violin plots showing tumour signal distribution by metabolic phenotype. Box plots display the median (50th percentile) as the central line, with boxes spanning the 25th and 75th percentiles. Whiskers extend to the minimum and maximum values within 1.5× the interquartile range; Glyco, n = 5,440; Low, n = 8,038; Oxphos, n = 4,508. i, Spatial uniform manifold approximation and projection (UMAP) plot of tumour-enriched spots from all metabolic phenotypes. j, Quantification of TME-deconvolved populations (x axis, population; y axis, normalized weight per spot for each phenotype). Box plot median and range as in h. k, Spatial maps showing that areas enriched for tumour cells display all three metabolic states. Highlighted areas show that these metabolic phenotypes arise from cells displaying strong tumour signals rather than from cells of the TME.Source data a, Heatmap showing the average intensities of the seven metabolites used for segmentation of the three metabolic regions. b, Representative example of three tumour sections segmented into three clusters. The region containing state 3 cells (blue) showed high [U-C]lactate and [U-C]pyruvate labelling and was considered to have a glycolytic phenotype. The region containing state 2 cells (yellow) showed high labelling of [C2]glutamate and was considered to have an oxidative phenotype. The region containing state 1 cells (red) showed low labelling of glycolytic and TCA cycle metabolites. c, Metabolic segmentation based on spatial RNA sequencing of sections contiguous with those shown in b. Blue spots correspond to a glycolytic phenotype, and yellow spots correspond to an oxidative phenotype based on Hallmark gene sets. Red spots correspond to cells with low glycolytic and oxidative gene signatures. d, ATP/ADP and PCr/ATP signal intensity ratios (mean; error bars, s.d.) in normal-appearing brain (n = 4) and in the three metabolically defined regions (region 1, n = 22; region 2, n = 14; region 3, n = 17); every dot is a unique region. No statistical significance was identified using one-way ANOVA with Tukey’s multiple comparisons test. e, Redox status (mean; error bars, s.d.) in normal-appearing brain tissue and GB regions quantified using AsA:DHA, GSH:GSSG and [U-C]lactate/pyruvate ratios (normal brain, n = 4; region 1, n = 22; region 2, n = 14; region 3, n = 17). No statistical significance was identified using one-way ANOVA with Tukey’s multiple comparisons test. f, Quantification of Ki67 cells, immune cells and blood vessels (mean; error bars, s.d.) in the three metabolically defined regions based on immunohistochemical analysis (region 1, n = 17; region 2, n = 4; region 3, n = 8). No statistical significance was identified using one-way ANOVA with Tukey’s multiple comparisons test. g, Spatial co-assignment of deconvolved tumour/TME signals (first row) and MSI labels (second row). h, Violin plots showing tumour signal distribution by metabolic phenotype. Box plots display the median (50th percentile) as the central line, with boxes spanning the 25th and 75th percentiles. Whiskers extend to the minimum and maximum values within 1.5× the interquartile range; Glyco, n = 5,440; Low, n = 8,038; Oxphos, n = 4,508. i, Spatial uniform manifold approximation and projection (UMAP) plot of tumour-enriched spots from all metabolic phenotypes. j, Quantification of TME-deconvolved populations (x axis, population; y axis, normalized weight per spot for each phenotype). Box plot median and range as in h. k, Spatial maps showing that areas enriched for tumour cells display all three metabolic states. Highlighted areas show that these metabolic phenotypes arise from cells displaying strong tumour signals rather than from cells of the TME. Source data To exclude perfusion and hypoxia as explanations for the observed metabolic heterogeneity, we assessed cellular energy status from measurements of the ATP, ADP and PCr concentrations. All tumour regions had ATP/ADP and PCr/ATP ratios comparable to those in normal brain (Fig. 2d). ADP was more abundant in the more oxidative tumour regions (Extended Data Fig. 4c), consistent with a high intramitochondrial ADP concentration. The three metabolic states and normal brain showed similar redox status, as assessed from measurements of the ascorbic acid to dehydroascorbic acid (AsA:DHA) and reduced to oxidized glutathione (GSH:GSSG) ratios, reflecting the NADPH/NADP ratio, and the [U-C]lactate/[U-C]pyruvate ratios, reflecting the NADH/NAD ratio (Fig. 2e). Next, we analysed contiguous sections by imaging mass cytometry (IMC) for the presence of immune cells, blood vessels and proliferating cells. We defined five immune phenotypes: CD3CD45CD4 (helper T cells); CD3CD45CD8 (cytotoxic T cells); CD3CD45CD8GZMB (activated T cells); CD45GZMB (natural killer or neutrophils); and CD68 (macrophages or microglia). Vascular phenotypes included large vessels (ASMACollagenIpanCKCD31) and small vessels or capillaries (CD31CollagenI). Regions occupied by the different metabolic states showed similar numbers of Ki67 cells, immune cell phenotypes and blood vessels (Fig. 2f and Extended Data Fig. 5a,b), with the only significant difference being slightly fewer CD68 cells in the oxidative regions. However, the proportion of immune cells was small, representing less than 10% of the total cell population. Cell density could also not account for the observed metabolic heterogeneity (Extended Data Fig. 5c). Regions of necrosis, pseudopalisading necrosis and microvascular proliferation showed a lack of correlation with metabolically distinct areas (Extended Data Fig. 5d). Tumour areas identified by a strong malignant signal in the spatial transcriptomics data showed the three metabolic states in spatially coherent areas on co-registered MS images (Fig. 2g–i), whereas regions identified as containing immune cells were predominantly glycolytic and those containing neurons were predominantly oxidative (Fig. 2j). Regions that showed a strong malignant signal and that were largely devoid of TME signals displayed all three metabolic states in the MS images (Fig. 2k). There were no differences in carbonic anhydrase IX (CAIX) transcript levels between the three metabolic states (Extended Data Fig. 6a), suggesting similar levels of hypoxia, or in immune, vascular, neuron, astrocyte and oligodendrocyte cell populations (Extended Data Fig. 6b). The bioenergetic status and microenvironment of the three metabolic states indicate that differences in metabolic activity are unlikely to have been influenced by differences in tissue perfusion, the presence of necrosis, differences in cell proliferation or the presence of immune cell infiltrates, but rather represent tumour-cell-intrinsic metabolic phenotypes. Despite there being no significant correlation between metabolic phenotype and blood vessel density, we nevertheless investigated a possible relationship between metabolic phenotype and proximity to the vasculature. We selected large vessels (Collagen I and αSMA) (Extended Data Fig. 7a), co-registered these with contiguous MSI sections and measured C-labelled glycolytic and TCA cycle metabolites at 65 μm intervals from the blood vessel lumen. Regardless of distance, there were no differences in lactate and glutamate labelling (Extended Data Fig. 7b), the number of proliferating cells, as indicated by Ki67 staining (Extended Data Fig. 7c), the NADPH/NADP ratio (AsA:DHA ratio) or the NADH/NAD ratio ([U-C]lactate/[U-C]pyruvate ratio) (Extended Data Fig. 7d,e). However, there was a decrease in the number of immune cells with increasing distance from the vessel wall (Extended Data Fig. 7f,g). To confirm that the metabolic phenotypes are tumour-cell-intrinsic and not a consequence of differences in the TME, we derived 30 primary cell lines from 26 patients with GB (two patients had two cell lines derived from multi-regional tumour sampling). These were grown as neurospheres with [U-C]glucose before snap-freezing and sectioning for MSI analysis (Fig. 3a–c). Three-dimensional culture has been shown to more closely approximate the behaviour of the primary tumour. The MS images were segmented using the same k-means clustering as for the human data (Fig. 3d). The neurospheres derived from each cell line showed distinct metabolic states (Fig. 3a). Neurosphere diameter was similar for all three metabolic phenotypes, which showed similar AMP, ADP, ATP and PCr signal intensities and ATP/ADP ratios (Fig. 3e). Although the segmentation allowed us to group the spheres and the cells from which they were derived into distinct metabolic phenotypes, these nevertheless represent a continuum of glycolytic and TCA cycle activities (Extended Data Fig. 8a). RNA sequencing showed that spheres with a glycolytic phenotype had an upregulation of glycolytic genes and those regulated by hypoxia (Extended Data Fig. 8b). However, there was no difference in the oxidative or TCA cycle gene expression profiles.Fig. 3Primary neurospheres recapitulate metabolic phenotypes in patients with GB.a, k-means clustering map of 30 cell lines grown as neurospheres in Matrigel domes (outlined in grey) using the seven C-labelled metabolites from glycolysis and the TCA cycle. Red corresponds to metabolic state 1 (low glycolytic, low TCA cycle activity), yellow corresponds to metabolic state 2 (high TCA cycle activity) and blue represents metabolic state 3 (high glycolytic activity). b, Representative H&E-stained sections of spheres displaying one of the three metabolic states. Red, yellow and blue outlines correspond to the sphere domes in a. c, Signal intensity maps for [U-C]lactate, [C2]glutamate and the ATP/ADP ratio in spheres shown in b. d, Heatmap showing the average signal intensities of the seven metabolites used for segmentation of the three metabolic states. e, Top: sphere diameter for each metabolic phenotype. Each point represents a single sphere. State 3 (glycolytic phenotype) had a higher average sphere diameter than state 1 (low glycolytic, low TCA cycle activity) (P = 0.027). Bottom: ATP, ADP, AMP and PCr signal intensities in neurospheres with different metabolic phenotypes. f, k-means segmentation of MS images of spheres derived from multi-regional tumour sampling (GTP2 and AT21). These results are part of the k-means analysis performed on all 30 neurosphere lines. g, Signal intensities of [U-C]lactate and [C2]glutamate in the indicated neurospheres (****P < 0.0001). The colour of the bar corresponds to the metabolic phenotype. Data are means; error bars, s.d.; each dot represents a single neurosphere. Asterisks refer to P values obtained from one-way ANOVA followed by Tukey’s multiple comparisons test or unpaired t-test (*P < 0.05, ***P < 0.0005, ****P < 0.00005).Source data a, k-means clustering map of 30 cell lines grown as neurospheres in Matrigel domes (outlined in grey) using the seven C-labelled metabolites from glycolysis and the TCA cycle. Red corresponds to metabolic state 1 (low glycolytic, low TCA cycle activity), yellow corresponds to metabolic state 2 (high TCA cycle activity) and blue represents metabolic state 3 (high glycolytic activity). b, Representative H&E-stained sections of spheres displaying one of the three metabolic states. Red, yellow and blue outlines correspond to the sphere domes in a. c, Signal intensity maps for [U-C]lactate, [C2]glutamate and the ATP/ADP ratio in spheres shown in b. d, Heatmap showing the average signal intensities of the seven metabolites used for segmentation of the three metabolic states. e, Top: sphere diameter for each metabolic phenotype. Each point represents a single sphere. State 3 (glycolytic phenotype) had a higher average sphere diameter than state 1 (low glycolytic, low TCA cycle activity) (P = 0.027). Bottom: ATP, ADP, AMP and PCr signal intensities in neurospheres with different metabolic phenotypes. f, k-means segmentation of MS images of spheres derived from multi-regional tumour sampling (GTP2 and AT21). These results are part of the k-means analysis performed on all 30 neurosphere lines. g, Signal intensities of [U-C]lactate and [C2]glutamate in the indicated neurospheres (****P < 0.0001). The colour of the bar corresponds to the metabolic phenotype. Data are means; error bars, s.d.; each dot represents a single neurosphere. Asterisks refer to P values obtained from one-way ANOVA followed by Tukey’s multiple comparisons test or unpaired t-test (*P < 0.05, ***P < 0.0005, ****P < 0.00005). Source data Next, we looked at neurospheres derived from multi-regional sampling of the same tumour to determine whether the neurospheres captured the regional metabolic heterogeneity observed in the tumour samples. GTP2 Med (medial tumour) and GTP2 Lat (lateral tumour) formed similar-sized neurospheres but showed different metabolic phenotypes, with GTP2 Med being more glycolytic and clustering into state 3 and GTP2 Lat clustering into state 1. Similarly, AT21 Ant (anterior) and AT21 Post (posterior), from another patient, displayed an oxidative and mixed metabolic phenotype, respectively (Fig. 3f,g). Therefore, metabolic heterogeneity observed in vivo was retained following tissue dissociation and growth outside of the native TME. To test the robustness of the tumour-cell-intrinsic metabolic phenotype, we grew a subset of neurospheres, representative of the highly glycolytic to the more oxidative phenotypes, under normoxic and hypoxic conditions (0.5% O2) and compared their transcriptomes. Despite prolonged exposure to hypoxia (160 h), there was minimal change in the transcriptomes (Extended Data Fig. 8c). Gene set enrichment analysis of the PC1 loadings showed E2F and MYC targets and mTORC1 pathway genes in the top ten under both normoxic and hypoxic conditions, with hypoxia and glycolysis genes under normoxic conditions and oxidative phosphorylation, and epithelial mesenchymal transition genes under hypoxic conditions (Supplementary Table 4). To further test the cell-intrinsic nature of the metabolic phenotypes, we implanted A11, AT8, S2 and AT5 into the brains of athymic rats and found that the cells retained the metabolic phenotypes that were observed when they were grown as neurospheres. AT8 and S2 formed more glycolytic xenografts, as demonstrated by higher [U-C]lactate labelling, whereas AT5 xenografts were more oxidative with higher abundance of [C2]glutamate (Fig. 4a,b). There were no differences in cell proliferation (Ki67 staining), cell death (CC3 staining) or vascularization (CD3 staining) between these three models, which paralleled the finding in the human data that proliferation rate and vascularization were not responsible for the differences in the observed metabolic phenotypes (Extended Data Fig. 8d,e). We have shown previously that S2 cells are more sensitive than A11 cells to irradiation (referred to as GB1 and GB4, respectively, in the earlier publication) and that S2 xenografts are more sensitive than A11 xenografts to treatment with temozolomide plus irradiation. Although when evaluated alongside all the other neurospheres, A11 and S2 appeared similarly glycolytic (Extended Data Fig. 8a), S2 xenografts showed greater TCA cycle activity than A11 xenografts (Fig. 4c,d). We have shown previously that A11 cells show higher glycolytic activity and S2 cells higher oxygen consumption and lower glycolytic activity, and demonstrated this in the corresponding xenografts using deuterium magnetic resonance spectroscopy and spectroscopic imaging measurements of deuterium-labelled glucose metabolism in vivo. Treatment of cells with AZD2014 (mTOR1 and mTOR2 inhibitor), imatinib (PDGFR; tyrosine kinase inhibitor) and gefitinib (EGFR; tyrosine kinase inhibitor) showed that cell viability was correlated with metabolic phenotype, where oxidative cells were more drug resistant (Extended Data Fig. 8f–h).Fig. 4Neurospheres retain their metabolic signatures as orthotopically implanted xenografts, and these signatures correlate with drug response.a, Representative MSI sections from rat brains implanted with AT8, S2 and AT5 (n = 3 independent tumours per model). Relative signal intensities for [U-C]lactate (top) and [C2]glutamate (bottom); H&E-staining of the corresponding sections. b, The relative signal intensities of [U-C]lactate (****P < 0.0001) and [C2]glutamate (AT8 vs AT5, P = 0.0012; S2 vs AT5, P = 0.0406; AT8 vs AT5, P = 0.0012) in neurospheres (AT8, n = 4; S2, n = 5; AT5, n = 4) and the respective xenografts (AT8, n = 3; S2, n = 3; AT5, n = 3) expressed as mean values; error bars, s.d. ([C2]glutamate AT8 vs AT5, P = 0.0015; S2 vs AT5, P = 0.0049). *P < 0.05; **P < 0.005. c, Relative abundance (mean values; error bars, s.d.) of labelled malate and fumarate in sections of A11 and S2 xenografts. d, Top: labelled malate and fumarate signal intensities in MS images of the brains of rats implanted with A11 (n = 12) and S2 (n = 12) xenografts. Bottom: H&E-staining of the corresponding sections. Asterisks refer to P values obtained from one-way ANOVA followed by Tukey’s multiple comparisons test or unpaired t-test (****P < 0.0001).Source data a, Representative MSI sections from rat brains implanted with AT8, S2 and AT5 (n = 3 independent tumours per model). Relative signal intensities for [U-C]lactate (top) and [C2]glutamate (bottom); H&E-staining of the corresponding sections. b, The relative signal intensities of [U-C]lactate (****P < 0.0001) and [C2]glutamate (AT8 vs AT5, P = 0.0012; S2 vs AT5, P = 0.0406; AT8 vs AT5, P = 0.0012) in neurospheres (AT8, n = 4; S2, n = 5; AT5, n = 4) and the respective xenografts (AT8, n = 3; S2, n = 3; AT5, n = 3) expressed as mean values; error bars, s.d. ([C2]glutamate AT8 vs AT5, P = 0.0015; S2 vs AT5, P = 0.0049). *P < 0.05; **P < 0.005. c, Relative abundance (mean values; error bars, s.d.) of labelled malate and fumarate in sections of A11 and S2 xenografts. d, Top: labelled malate and fumarate signal intensities in MS images of the brains of rats implanted with A11 (n = 12) and S2 (n = 12) xenografts. Bottom: H&E-staining of the corresponding sections. Asterisks refer to P values obtained from one-way ANOVA followed by Tukey’s multiple comparisons test or unpaired t-test (****P < 0.0001). Source data The concentrations of unlabelled serine, threonine, glutamine and glutamate were significantly higher in oxidative regions in the patient tumour sections (ANOVA, Tukey’s P < 0.05), and in the neurospheres, the concentrations of unlabelled leucine/isoleucine, glutamine, glutamate, histidine and phenylalanine were significantly higher (ANOVA, Tukey’s P < 0.005) (Extended Data Fig. 9). The oxidative phenotype in GB has been associated with increased fatty acid oxidation, and we observed that the concentrations of unlabelled fatty acids were higher in the more glycolytic regions in both the human and neurosphere data, as were intermediates in the pentose phosphate pathway. The extent to which tumour metabolism is driven by cell-intrinsic mechanisms or microenvironmental pressures is an open question in tumour biology. To address this question, we measured the metabolic activity of GB within its native microenvironment using MS imaging of isotope labelling in rapidly quenched tissue from patients with GB. A similar approach has been used to map heterogeneity in fatty acid synthesis in gliomas implanted orthotopically in mice, to map metabolic activities in kidney and brain in mice and to image glycolytic activity in a lung metastasis model. Using a targeted approach, we identified distinct glycolytic and oxidative metabolic phenotypes. Although recent reports have identified metabolic phenotypes from a transcriptomic analysis, we describe here the classification of metabolic phenotypes based on measurements of metabolic activity in patient tumours in vivo at relatively high spatial resolution (65 μm). The metabolic phenotypes occupied distinct territories that did not show significant differences in the cellular composition of their microenvironments, including immune cell infiltration, proliferative index and vascularization. The presence of these metabolic phenotypes in distinct territories and their cellular composition was confirmed by spatial transcriptomics. Importantly, we have shown that these tumour-cell-intrinsic metabolic phenotypes can be independent of the TME. Previous studies have also shown that tumour subtype-specific protein and gene expression profiles can be independent of the tumour niche. We observed no change in lactate and glutamate labelling with distance from the blood vessels. This appears to be inconsistent with a study in an orthotopically implanted glioma cell line model (U87MG) in immunocompromised mice that showed high mitochondrial activity in cells adjacent to vessels and increased expression of hypoxia-related genes at increasing distance from the vessels. However, there were no significant increases in lactate concentration or differences in α-ketoglutarate concentrations in the cells with distance from the vessels in this study, although these concentrations may have been affected by the flow cytometric method used to sort the cells. An MSI study on an orthotopically implanted syngeneic murine model of isocitrate dehydrogenase 1 mutant GB in animals fed for 48 h with [U-C]glucose showed a remarkable degree of metabolic homogeneity in the tumours, in contrast to what was observed here in tumours from patients with GB, emphasising the significance of the observed metabolic phenotypes. The cell-intrinsic nature of the metabolic phenotypes was confirmed using neurospheres grown in vitro, which reproduced the metabolic phenotypes observed in the patients and were preserved following their orthotopic implantation in rats. Exposing the neurospheres to chronic hypoxia did not lead to significant changes in the neurosphere transcriptomes, demonstrating the robustness of these phenotypes and further underlining their independence from the TME. This is in contrast to a similar study in patients with lung cancer who were infused with [U-C]glucose, which concluded that given the observed differences in glucose oxidation in the TCA cycle in well-perfused versus poorly perfused regions, tumour perfusion in this case overrides tumour-cell-intrinsic metabolic phenotypes. A recent study in mouse models of leukaemia, pancreatic, lung and colon cancer showed that these tumours suppress TCA cycle activity relative to normal tissue. By contrast, the glutamate labelling observed here in the GB tumours was similar to that in normal-appearing brain. Similar observations of substantial glucose oxidation in the TCA cycle have been made previously in human lung tumours and in non-imaging studies of patients with GB who were infused with [U-C]glucose. Metabolic phenotypes with distinct therapeutic vulnerabilities have been identified in several cancers, including GB, in which GB cells with an oxidative phenotype were shown to be more sensitive to inhibitors of mitochondrial complex I and to radiation treatment. We have shown previously that S2 cells are more sensitive to irradiation in vitro and in vivo and have shown here, and previously, that they have a more oxidative phenotype with higher TCA cycle activity than A11 cells, which show a more glycolytic phenotype. Previous RNA sequencing studies have shown that A11 has a mesenchymal phenotype, whereas S2 cells have a neural progenitor cell-like phenotype, consistent with a previously identified association between the mesenchymal and glycolytic phenotypes. Cells displaying these distinct metabolic phenotypes occupy territories that are sufficiently large to be imaged clinically using techniques such as hyperpolarized C magnetic resonance imaging (MRI) and deuterium metabolic imaging. These metabolic phenotypes show a correlation with treatment responsiveness, suggesting that a personalized therapy approach may be possible in which metabolic imaging and phenotyping could be used to guide subsequent treatment selection. We have demonstrated here high-resolution imaging of isotope labelling of cellular metabolites in a human tumour in vivo. In conjunction with studies on patient-derived neurospheres and orthotopically implanted xenografts, we have demonstrated the presence of different metabolic phenotypes within GB that are tumour-cell-intrinsic and largely independent of the TME. Three male patients from Addenbrooke’s Hospital, Cambridge, were infused with [U-C]glucose. The selection criteria included first clinical presentation, MRI consistent with GB and no significant co-morbidities. Following induction of anaesthesia, a pyrogen-free 5% solution of [U-C]glucose in sterile saline (Merck) was administered as a bolus of 8 g over 10 min followed by 8 g h continuous infusion, as described previously for patients with GB and several other tumour types. Arterial blood was collected by a peripheral arterial line before bolus administration and then every 15 min following the start of infusion and at 15 min after the end of the infusion. Tumour sampling was guided by intra-operative Stealth navigation and assessment of 5-ALA fluorescence. Tumours were sampled between 90 and 150 min. A pituitary ronguer was used to transfer tumour samples directly into liquid nitrogen. The freezing time was <5 s between tissue devascularization and immersion in liquid nitrogen. The study was approved by the Central Cambridge Research Ethics Committee and was compliant with the Health Research Authority. The study adhered to the principles of the Declaration of Helsinki and the Guidelines for Good Clinical Practice. Participants did not receive financial compensation and gave informed consent. Frozen tumour samples were embedded in a hydroxypropyl methylcellulose/polyvinylpyrrolidone hydrogel, and 10 µm-thick cryo-sections were obtained. Sections were thaw-mounted onto Superfrost microscope slides for desorption electrospray ionization (DESI) and IMC experiments (Thermo Scientific), while sections prepared for MALDI-MSI were thaw-mounted onto conductive ITO-coated slides (Bruker Daltonik). The sections were dried immediately and sealed in vacuum pouches for storage at −80 °C. Human tumour samples were treated with UV-C light before MSI analysis to minimize aerosolization of potential pathogens. Decontamination was performed in a sensor-controlled UV chamber (Opsytec Dr. Gröbel) at 250 mJ cm. DESI-MSI analysis was performed on a Q-Exactive mass spectrometer (Thermo Scientific) equipped with an automated 2D-DESI ion source (Prosolia). The spectrometer was used with a home-made Swagelok DESI sprayer and a mixture of 95% methanol, 5% water delivered at a flow rate of 1.5 µl min and nebulized with nitrogen at a backpressure of 6 bar. Human tumour samples were analysed in the mass range from 70 to 280 m/z with a mass resolution of 140,000 at 200 m/z. The injection time was set at 500 ms, and data were acquired with a pixel size of 65 µm. Following analysis, sections were stained with H&E and co-registered with the DESI-MS images. The resulting .raw files were converted into .mzML files using ProteoWizard msConvert (v.3.0.4043) with the built-in peak picking algorithm and subsequently compiled to an .imzML file (imzML converter, v.1.3). All subsequent data processing was performed in SCiLS Lab (v.2022b, Bruker Daltonik). MALDI-MSI analysis was performed using a RapifleX Tissuetyper instrument (Bruker Daltonik) operated in negative ion detection mode. 9-Aminoacridine, prepared in an 80:20 methanol-to-water ratio, was used as a matrix and spray-deposited using an automated spray system (M3-Sprayer, HTX technologies). Mass spectra were acquired from 180 to 1,000 m/z with a pixel size of 40 µm. A total of 350 laser shots were summed up per pixel. For all experiments, the laser was operated with a repetition rate of 10 kHz. Raw data were directly uploaded and processed with SCiLS Lab (v.2022b) software. DESI and MALDI data and images were normalized to the total ion current to compensate for signal variation during the course of the experiments, and acquisition parameters and data processing were identical for human tissue, neurospheres embedded in Matrigel and xenograft sections. Snap-frozen tumour samples were homogenized with 5 µl mg of 2 M perchloric acid. The extract was centrifuged at 13,000g for 15 min, and the pH of the supernatant was adjusted to 7.0 using 2 M KOH. Extracts were lyophilized and dissolved in 550 µl deuterium oxide containing methylenediphosphonic acid at 100 nmol g tissue, which was added as a chemical shift and intensity standard. Proton-decoupled P NMR spectra were acquired with more than 8,000 repetitions into 32,768 data points with an 11 µs 90° pulse, a repetition time of 7.2 s and a spectral width of 57 ppm (14,006 Hz). Arterial blood collected during intra-operative infusion was centrifuged at 2,000g and 4 °C for 20 min to collect the plasma, which was snap-frozen and stored at −80 °C. Samples were thawed on wet ice and aliquots diluted 50-fold with cold methanol:acetonitrile:water (50:30:20) in chilled tubes, vortexed for 10 min and centrifuged at 21,100g for 10 min at 4 °C, and then the supernatants stored at −80 °C. On the day of analysis, supernatants were centrifuged and aliquots transferred to a 96-well plate for analysis by hydrophilic interaction liquid chromatography (HILIC) high-resolution mass spectrometry (HRMS). The HILIC–HRMS system consisted of a Shimadzu Nexera X2 UHPLC and Sciex 6600 Triple TOF mass spectrometer, using a SeQuant ZIC-pHILIC 5 µm 150 × 2.1 mm column (with a ZIC-pHILIC 20 × 2.1 mm guard column) at 45 °C. The liquid chromatography gradient started at 80% acetonitrile and 20% 20 mM ammonium carbonate (pH 9.4), changing to 20% acetonitrile over 17 min at 200 µl min, with a further 11.5 min of column re-equilibration. Each sample was injected twice for analysis in both positive and negative electrospray ionization with a full-scan m/z range of 75–1,000. Data were acquired using Sciex Analyst TF and processed using Sciex MultiQuant software. Extracted ion chromatograms were generated from the theoretical m/z ± 20 ppm. Peak integration was reviewed manually, and the peak area of each metabolite and isotope was exported. Primary human cell lines were derived at Addenbrooke’s Hospital, Cambridge, UK, as described previously. Tissue collection was approved by a Regional Ethics Committee (REC18/EE/0283) and was compliant with the UK Human Tissue Act 2004. Resected tissue samples were washed with Hanks’ balanced salt solution and minced using sterile razor blades, followed by digestion with Accutase (Sigma-Aldrich) for 60 min at 37 °C. Single cells were isolated by filtration through a 40 µM filter (Falcon). Cells were centrifuged at 350g, 21 °C, and the pellet was incubated with 2–3 ml of Red Blood Cell Lysis Buffer (Sigma-Aldrich) for 5 min at room temperature (21 °C). Following centrifugation, cells were seeded at a density of 15,000 cells per cm in serum-free Neurobasal A medium (Gibco) supplemented with B27, N2, EGF and FGF growth factors (Sigma-Aldrich). Cells were allowed to form aggregates, and the medium was changed 3 days post derivation. For neurosphere formation, cells were seeded in Ultra Low Attachment 96-well plates (Corning) at a density of 10,000 cells per well. Sphere diameter was monitored using an IncuCyte microscope (Sartorius). Spheres were embedded in 150 μl Matrigel (Corning) domes in 24-well plates (Corning). The domes were covered with 2 ml of fresh Neurobasal medium and grown for 24 h. The domes were then washed three times with PBS before 2 ml of fresh glucose-free Neurobasal medium supplemented with 25 mM [U-C]glucose was added. Following incubation for 24 h, the medium was removed and the domes were washed three times with ice-cold PBS and then snap-frozen in liquid nitrogen. The domes were sectioned to a thickness of 10 µm and analysed using DESI-MSI and MALDI-MSI, as described above. Black, clear-bottomed 96-well plates (Corning) pre-coated with ECM (Merck) were seeded at 5,000 cells per well and incubated at 37 °C in 200 μl of Neurobasal medium supplemented with growth factors, as described above. The medium was then replaced with either fresh medium or medium containing the following drug concentrations: AZD2014, 1 μM; gefitinib, 1 μM; imatinib, 10 μM. The cells were incubated for 72 h before re-imaging using the IncuCyte microscope, and cell numbers were determined using the Sartorius cell confluency module. At the end of the incubation, the medium was aspirated, and 100 μl of PBS containing 3 μM propidium iodide was added to each well to measure necrotic cell death. The plates were imaged using the IncuCyte 10× cell-by-cell module using the red channel. The experiment was repeated three times with six technical replicates for each drug treatment. Neurospheres in Matrigel domes were grown at 37 °C, with one plate incubated in atmospheric O2 and the other grown in 0.5% O2, 0.5 Pa (Avatar, XCellbio). Medium was refreshed every 48 h, and on day 10 it was removed, the plates placed on ice and the wells washed with 1 ml of ice-cold PBS, followed by the addition of 1 ml Corning Cell Recovery Solution. The plates were incubated at 4 °C for 1 h and then washed four times with ice-cold PBS. RNA was isolated from cell pellets using a QIAshredder spin column (Qiagen) and AllPrep DNA/RNA Mini Kit (Qiagen), quantified using a Qubit fluorometer (Thermo Fisher Scientific) and quality tested using an Agilent 4200 TapeStation. For library preparation, the Illumina TruSeq Stranded mRNA Kit was used, and single-read sequencing was performed on a HiSeq 4000 machine (Illumina). Quality control of raw sequence data was carried out using FastQC (v.0.11.8). Some reads were trimmed to remove adaptor content using Trimmomatic (v.0.39). Reads were aligned to GRCh38 Ensembl release 102 using STAR (v.2.7.7a) and alignment quality control was carried out using Picard tools (v.2.25.1). Quantification was carried out using Salmon (v.1.6.0) against a reference transcriptome for the same genome release. Differential gene expression analysis was carried out in R (v.4.2.2) using the DESeq2 package (v.1.38.3) with default parameters. Multiple testing correction of P values was carried out using the Benjamini–Hochberg method. Genes were determined to be differentially expressed at an adjusted P value of 0.05. Gene set enrichment analysis was carried out using clusterProfiler (v.4.6.0). Experiments were performed under the authority of a Home Office project licence (PP5634271) and approved by an Animal Welfare and Ethical Review Body at the Cancer Research UK (CRUK) Cambridge Institute, University of Cambridge. Athymic, female nude rats that were at least 9 weeks old were implanted orthotopically with the primary GB lines at passage 10. Animals were anaesthetized using 2% isoflurane (Isoflo, Abbott Laboratories) in O2/air (25/75%, vol/vol, 2 l min) with 5 mg kg Carpofen (Zoetis) and 0.3 mg ml buprenorphine hydrochloride (Alstoe) subcutaneous analgesia. Body temperature was maintained using a heating pad. A stereotactic surgical frame (Kopf) was used to secure the animal’s head. A midline incision was made followed by a 1 mm burr hole anterior and to the right of the bregma. A total of 1 × 10 cells were injected in 5 µl of Neurobasal medium at a depth of 4 mm. The burr hole was closed with bone wax (Ethicon) and skin with 6/0 vicryl (Ethicon). Tumour growth was monitored using T2-weighted MRI. Specifically, animals were anaesthetized using 2% isoflurane (Isoflo, Abbott Laboratories) in O2/air (25/75%, vol/vol, 2 l min) and placed supine inside a 7T magnet (Agilent), and T2-weighted MRI was used to monitor tumour growth. A 72 mm H volume coil was placed around the animal’s head, and breathing rate and temperature were monitored with a small animal instruments monitoring system (SAII). Axial H T2-weighted images were acquired using a fast spin-echo sequence with an echo time of 50 ms, pulse repetition time of 1,500 ms, flip angle of 60–90° and a slice thickness of 2.0 mm, field-of-view of 40 mm × 40 mm and 128 × 128 or 256 × 256 data points. Three animals per cell line were administered with [U-C]glucose as a bolus at 0.4 mg g, followed by continuous infusion of 0.012 mg g min at 300 µl h for 120 min (ref. ). The brains were snap-frozen in liquid nitrogen, cryo-sectioned at a thickness of 10 µm and analysed with DESI-MSI and MALDI-MSI, as described above. The 10× Genomics Visium platform was used and analysed with the Space Ranger pipeline. Downstream analyses were conducted in R using the Seurat package. Samples were processed individually using the SCTransform() function. Spots were filtered based on standard quality control thresholds (for example, mitochondrial gene percentage of >20%, nCount_Spatial of <1,000 and nFeature_Spatial of <1,000). Data were re-corrected across samples using PrepSCTFindMarkers() for joint numerical analyses. Gene set scoring was performed using the Hallmark gene sets and the UCELL package, excluding mitochondrial genes from the oxidative phosphorylation score calculations. To annotate spatial spots, genes from Hallmark gene sets of interest were combined and subjected to k-means clustering (k = 3). To annotate TME spots, we performed two deconvolution steps with robust cell type decomposition, using previously published reference cell annotations. First, a balanced normal reference was sampled from the non-neoplastic cells, combined with the neoplastic cells and used as input for robust cell type decomposition. The assignment was further confirmed by histological evaluation. To label TME niches, we further deconvolved TME signals into oligodendrocytes, astrocytes, vascular cells, immune cells (macrophages) and neurons. Given that each spatial spot contained multiple cells, we normalized the deconvolution weights to a maximum of one, providing a relative abundance. To mitigate over-assignment to sorted TME cell types, we included unsorted cell types (neoplastic + oligodendrocyte precursor cell with high transcriptomic similarity) in this deconvolution step as a ‘block’ effect. Joint feature plots (tumour and TME) were generated using a custom modification of the SpatialFeaturePlotBlend() source function. The original repository is available at https://github.com/george-hall-ucl/SpatialFeaturePlotBlend. For the TME population-level quantification, for each spot, the weights of all deconvolved populations were normalized to sum to one, thereby representing the relative contribution of each cell population within that specific spot. Subsequently, these normalized weights were plotted across the three metabolic states. Total ion count-normalized MSI data were extracted using the SCiLS Lab API (v.2022b; Bruker Daltonik), and metabolic labels were spatially smoothed to provide coherent spatial regions. Immediately neighbouring pixels (≤8) were identified for each MSI pixel, and across five iterations, for each pixel that had at least two neighbours and for which the majority of neighbours were assigned a different label, the label was replaced with the majority label. An image was created using spatial coordinates, in which pixels were coloured using the first three principal components as RGB channels. Landmarks were identified between this MSI dimensionality reduction image and the H&E image of the corresponding tissue section, and subsequently between the H&E image and the corresponding H&E image of the contiguous tissue section used for spatial transcriptomic analysis. These landmarks were then used to map the MSI coordinates onto the spatial transcriptomic coordinate space using an affine transform. MSI labels were finally transferred to spots on the spatial transcriptomic image using the k-nearest neighbours algorithm (k = 3). Regions were defined using k-means clustering and fitted using the Kmeans function in the amap R package and performed independently on the neurosphere, human and metastases datasets. The values for each of the seven C-labelled metabolites used for metabolic clustering were standardised into z-scores (by subtracting the metabolite’s mean and dividing by the metabolite’s standard deviation). Each pixel was then assigned to three or four groups, using k-means clustering with Manhattan distances. Antibodies used for immunohistochemistry are described in Supplementary Table 5. Antibodies were tagged using the Fluidigm Maxpar Antibody Labelling Kit. Slides were fixed with 4% paraformaldehyde in PBS for 10 min, washed three times in PBS, permeabilized using a 1:1,000 dilution of Triton X-100 in casein solution, washed another three times in PBS and then blocked for 30 min with casein solution (Thermo Fisher). Antibodies were diluted in casein solution, and the slides were incubated overnight at 4 °C. The slide was then washed three times in PBS, and nuclei were stained with DNA intercalator-iridium (Fluidigm) at a dilution of 1:400 in PBS for 30 min. The slide was washed three times in PBS, 30 s in deionized water and then dried at room temperature. A region for IMC analysis was selected using consecutive H&E-stained sections and the DESI-MSI data. IMC analysis was performed using a Hyperion Instrument (Fluidigm Corporation) with an ablation energy of 4 db and an ablation frequency of 200 Hz. IMC images were produced using MCD viewer (v.1.0; Fluidigm), and analysis was performed using HALO (Indica Labs). Tumour-bearing rat brains were snap-frozen in liquid nitrogen and sectioned at 6 μm thickness for immunohistochemistry analysis using Leica’s Polymer Refine Kit (antibodies listed in Supplementary Table 6). Images were analysed using Aperio image-viewing software and HALO (v.3.6.4134.137). HALO (v.3.6.4134.137) and HighPlex FL (v.4.1.3) modules were used for automated image analysis. Optical densities for weakly, moderately and strongly stained cells used for the automated quantitative analysis of scanned sections were as follows: Ki67 – (nuclear) 7, 40.7522, 54.385, p53 – (nuclear) 1.8, 3.5929, 5.1327, CC3 – (cytoplasm) 1.9427, 4.646, 6.1947, Vimentin – (cytoplasm) 13.3343, 25.7522, 41.2035, CD3 – (nuclear) 2.2832, 32.6875, 63, GZMB – (nuclear) 0.6956, 0.8673, 1.6106, CD4 – (cytoplasm) 3.9823, 6.7699, 10.354, CD8A – (nuclear) 1.6327, 18.7679, 26, CD68 – (cytoplasm) 10.4106, 45.9027, 78.6903, CD45 – (cytoplasm) 2, 5.3363, 7.115, ASMA – (cytoplasm) 2.8, 12.1327, 18.6239, CD31 – (cytoplasm) 2.274, 4.5841, 6.876, panCK – (cytoplasm) 1.836, 2.8673, 3.9823, Collagen I – (cytoplasm) 28.293, 104.7301, 179.58. Five cellular phenotypes were identified: CD4 helper T cells (CD45CD4CD3); cytotoxic T cells (CD3CD45CD8A); activated cytotoxic T cells (CD3CD45CD8AGZMB); natural killer cells and neutrophils (CD45GZMB); and macrophages and microglia (CD68). A random forest classifier was used to distinguish vessels and non-vessels within the images. Large vessels were positive for Collagen I and CD31. Annotations were created manually on several images, and these were used to train the classifier. The five cellular phenotypes were plotted spatially, three 60 µm bands travelling out from the vessels were defined and the total number of cells displaying these cellular phenotypes in each band was determined. Sample size for human tumour collection was determined intra-operatively based on patient and tumour factors (for example, proximity to eloquent brain). A minimum of six samples were collected for each tumour region. For xenograft and neurosphere studies, a minimum of three independent biological replicates were used (three different rats; three different cell line passages with a minimum of six technical replicates). No randomization or blinding was used, and all data were included in the analysis. Statistical analyses were performed using GraphPad Prism (v.10.0.3), R (v.4.3.0) and SCiLS. Bioinformatics analyses were performed in R. Statistical tests were two-sided unless stated otherwise. Student’s t-tests and one-way ANOVA with Tukey’s multiplicity correction were used to test the equality of means between two or more groups, respectively. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
PMC12081701
PRMT3 reverses HIV-1 latency by increasing chromatin accessibility to form a TEAD4-P-TEFb-containing transcriptional hub
Latent HIV-1 presents a formidable challenge for viral eradication. HIV-1 transcription and latency reversal require interactions between the viral promoter and host proteins. Here, we perform the dCas9-targeted locus-specific protein analysis and discover the interaction of human arginine methyltransferase 3 (PRMT3) with the HIV-1 promoter. This interaction reverses latency in cell line models and primary cells from latently infected persons by increasing the levels of H4R3Me2a and transcription factor P-TEFb at the viral promoter. PRMT3 is found to promote chromatin accessibility and transcription of HIV-1 and a small subset of host genes in regions harboring the classical recognition motif for another transcription factor TEAD4. This motif attracts TEAD4 and PRMT3 to the viral promoter to synergistically activate transcription. Physical interactions among PRMT3, P-TEFb, and TEAD4 exist, which may help form a transcriptional hub at the viral promoter. Our study reveals the potential of targeting these hub proteins to eradicate latent HIV-1.A major challenge to the current efforts to cure HIV-1 infection is the persistence of latent viruses. Addressing this challenge hinges on unraveling the intricate interplay between the host and viral factors that are key for HIV-1 latency. Current antiretroviral agents effectively prevent actively transcribing viruses from replicating, but have no activity against the latent viral reservoirs. One of the underlying obstacles is to identify the key chromatin-associated host protein(s) that interact with the 5’ long terminal repeats (LTR) containing the viral gene promoter and control viral transcription during latency reactivation. Such interactions are pivotal in determining the fate of a virus -- latency versus reactivation -- a balance that must be understood in order for us to achieve the elusive goal of a functional cure for HIV. Protein arginine methyltransferase 3 (PRMT3), a member of the protein arginine methyltransferase family, functions in epigenetic regulation of gene expression through histone modification and is implicated in a wide array of normal and diseases processes ranging from carcinogenesis, hepatic steatosis, to antiviral innate immunity. Given its diverse physiological and pathological functions, significant attention has been directed toward the exploration and development of selective allosteric inhibitors of PRMT3. Despite its well-established functions in many processes of a host cell, the role of PRMT3 in regulating HIV-1 transcriptional reactivation and latency reversal remains largely unclear. In the current work, we applied a nuclease-deactivated Cas9 (dCas9)-based DNA-protein interaction screen to investigate host proteins that potentially interact with the HIV-1 5’ LTR. This viral DNA interactome screen has identified PRMT3 as a top candidate. Subsequent functional characterization demonstrates that PRMT3 interacts with the LTR to regulate Tat-dependent reactivation of HIV-1 transcription by regulating the H4R3Me2a level and increasing chromatin accessibility at the viral LTR. This increased accessibility creates a transcriptional hub wherein PRMT3 forms a complex with transcription factors TEAD4 and P-TEFb. The interaction between PRMT3 and TEAD4, and their co-regulation of HIV-1 transcription through binding to the TEAD4-recognition motifs within the LTR, reveal a regulatory pathway whereby host protein and viral DNA interact to control HIV-1 gene expression. Thus, in addition to depicting a previously unrecognized role of PRMT3 as a key host factor that controls viral transcription and latency reversal, our findings establish a basis for the future development of targeted therapeutics for combating HIV-1 infection and latency. Seeking host factors that specifically associate with the HIV-1 LTR, we conducted an LTR-interacted proteome analysis based on a refined version of a previously described nuclease-deficient Cas9 (dCas9)-targeted chromatin-based purification strategy (CLASP) (Cas9 locus-associated proteome) (Fig. 1a). The screen used NH1 cells, which is a modified HeLa cell line that harbors an integrated HIV-1 LTR-driven luciferase reporter gene in the host genome (Fig. 1b). To get the most effective sgRNAs that target LTR, we used two methods for screening. We individually inserted the sgRNAs into the Cas9 plasmid, which was then co-transfected with a Tat-expression plasmid into NH1 cells that harbors the LTR-driven luciferase reporter gene. If a sgRNA-Cas9 combination can target and disrupt the LTR sequence, a reduction in luciferase activity is observed. We designed seven sgRNAs targeting different regions of the HIV-1 LTR (Fig. 1b). An sgRNA targeting the yeast Gal4 gene was used as the negative control. Among the designed sgLTRs, the top three sgLTRs (# 5, 6, 7 in pink, Fig. 1b) exhibiting the capacity to attenuate the LTR driven-luciferase expression activated by HIV-1 Tat (Fig. 1c). Then, we performed anti-Flag ChIP-qPCR in cells transfected with the sgLTRs-dCas9-3 × Flag plasmids to further verify the abilities of the sgLTRs to target dCas9-3 × Flag to the LTR, and found that sgLTR-1 and -2 failed to recruit dCas9 to LTR efficiently (Supplementary Fig. 1a, b), sgLTR-5 and -6 displayed the most significant effect, whereas sgLTR-3, -4 and -7 all displayed a similar and partial effect (Supplementary Fig. 1c, d). Thus, sgLTR-5, -6, -7 were selected for subsequent ChIP-qPCR analysis. By utilizing the combination of the three sgRNAs, we found that in comparison to sgGal4, these sgRNAs can guide dCas9 to the LTR as revealed by ChIP-qPCR analysis (Fig. 1d). Next, we performed in vitro transcription reactions to generate these sgRNAs, which were then used for dCas9/sgRNAs complex formation. This complex was subsequently incubated with sheared chromatin that contained the LTR-bound proteins for dCas9-3 × Flag immunoprecipitation, following with mass spectrometry analysis.Fig. 1dCas9-targeted HIV-1 LTR-interactome analysis identified PRMT3 as a LTR-binding factor.a Schematic of dCas9-targeted proteome analysis workflow. (1) sgRNAs targeting the LTR were complex formation with dCas9-3 × Flag; (2) Crosslinking of proteins bound to the LTR; (3) dCas9-3 × Flag-sgRNA complex was incubated with sheared fragments from NH1 cells that stably expressed LTR-luciferase reporter; (4) Immunoprecipitation was used to capture LTR bound proteins; (5) LTR-interacted proteins were identified by mass spectrometry. b Schematic of the position of 7 sgRNAs. c Luciferase activity was measured in NH1 cells transfected with a plasmid encoding each sgRNA and control sgRNA (sgGal4), which expresses the Cas9 in the presence of Tat (p = 0.0016, p = 0.0024, p = 0.0042). d ChIP-qPCR analyses were conducted to assess the occupancy of 3 × Flag-dCas9 at the Promoter (Left) or Nascent region (Right) of LTR when cells were transfected with control plasmid, a mixture of plasmids containing sgLTRs-5, 6, 7 or the sgGal4 plasmid (Promoter: p = 0.0009, p = 0.0011, Nascent: p = 0.000004, p = 0.000003). e Proteins showing over 1.5-fold enrichment relative to the control sample identified from 3 × Flag-dCas9 immunoprecipitation following by mass spectrometry analysis were listed. f ChIP-qPCR analyses were conducted to assess the occupancy of PRMT3 at LTR-luciferase region (p = 0.0001, p = 0.0012, p = 0.0005, p = 0.0008). The schematic display of the four primers was shown on the top of the panel. g ChIP-qPCR analyses were conducted to assess the occupancy of PRMT3 at LTR-luciferase regions in the absence or presence of Tat (p = 0.0022, p = 0.0012, p = 0.0045, p = 0.0017). h CUT&Tag sequencing was performed in primary CD4 T cells isolated from virologically suppressed HIV-1-infected patient cells that were treated with PMA, by using IgG or PRMT3 antibody. The alignment results of sequencing with the HIV-1 genome in three patients were displayed. Error bars = mean +/− SD of three biological replicates. *p < 0.05, **p < 0.01, ***p < 0.001, two-tailed t test. Source data are provided as a Source Data file. a Schematic of dCas9-targeted proteome analysis workflow. (1) sgRNAs targeting the LTR were complex formation with dCas9-3 × Flag; (2) Crosslinking of proteins bound to the LTR; (3) dCas9-3 × Flag-sgRNA complex was incubated with sheared fragments from NH1 cells that stably expressed LTR-luciferase reporter; (4) Immunoprecipitation was used to capture LTR bound proteins; (5) LTR-interacted proteins were identified by mass spectrometry. b Schematic of the position of 7 sgRNAs. c Luciferase activity was measured in NH1 cells transfected with a plasmid encoding each sgRNA and control sgRNA (sgGal4), which expresses the Cas9 in the presence of Tat (p = 0.0016, p = 0.0024, p = 0.0042). d ChIP-qPCR analyses were conducted to assess the occupancy of 3 × Flag-dCas9 at the Promoter (Left) or Nascent region (Right) of LTR when cells were transfected with control plasmid, a mixture of plasmids containing sgLTRs-5, 6, 7 or the sgGal4 plasmid (Promoter: p = 0.0009, p = 0.0011, Nascent: p = 0.000004, p = 0.000003). e Proteins showing over 1.5-fold enrichment relative to the control sample identified from 3 × Flag-dCas9 immunoprecipitation following by mass spectrometry analysis were listed. f ChIP-qPCR analyses were conducted to assess the occupancy of PRMT3 at LTR-luciferase region (p = 0.0001, p = 0.0012, p = 0.0005, p = 0.0008). The schematic display of the four primers was shown on the top of the panel. g ChIP-qPCR analyses were conducted to assess the occupancy of PRMT3 at LTR-luciferase regions in the absence or presence of Tat (p = 0.0022, p = 0.0012, p = 0.0045, p = 0.0017). h CUT&Tag sequencing was performed in primary CD4 T cells isolated from virologically suppressed HIV-1-infected patient cells that were treated with PMA, by using IgG or PRMT3 antibody. The alignment results of sequencing with the HIV-1 genome in three patients were displayed. Error bars = mean +/− SD of three biological replicates. *p < 0.05, **p < 0.01, ***p < 0.001, two-tailed t test. Source data are provided as a Source Data file. Mass spectrometry detected a total of 28 proteins with at least 1.5-fold enrichment relative to the non-targeting negative control (Fig. 1e and Supplementary Data 1). PRMT3, which has been shown to function in transcriptional regulation and antiviral innate immunity, was among the hits with the highest fold enrichment of sgLTR/sgGal4, and was therefore selected for subsequent investigation. We performed ChIP-qPCR analysis to examine the binding of PRMT3 to the LTR. PRMT3 was significantly enriched at LTR-luciferase reporter region (Fig. 1f), which can be further increased in the presence of Tat (Fig. 1g). The binding of PRMT3 at the LTR was also verified in Jurkat 2D10 cells (Supplementary Fig. 2). Notably, we performed Cleavage Under Targets & Tagmentation (CUT&Tag) sequencing by using IgG or PRMT3 antibody in CD4 T cells isolated from HIV-1 infected patient cells, and demonstrated that while IgG displayed undetectable signal at the LTR, PRMT3 showed significant binding signal around LTR (Fig. 1h). Thus, these results demonstrated that PRMT3 is an HIV-1 LTR binding factor, which is further enhanced in the presence of Tat. To examine the potential impact of the HIV-1 LTR-associated PRMT3 on Tat-dependent HIV-1 transcription, we induced PRMT3 knockdown (KD) by shRNAs in NH1 cells. PRMT3 KD did not affect basal HIV-1 transcription but did result in a significant decrease in Tat-dependent viral transcription (Fig. 2a). In contrast, PRMT3 overexpression significantly elevated Tat-dependent but not the Tat-independent HIV-1 transcription (Fig. 2b). The effect of PRMT3 on HIV-1 transcriptional activation was further measured in PRMT3 knockout (KO) cells (Supplementary Fig. 3a, b), which showed no significant defects in cell growth in comparison to the WT cells (Supplementary Fig. 3c). Similar to the KD result above, PRMT3 KO significantly decreased Tat-activated HIV-1 transcription (from 127.5-fold to 25.75-fold), and the plasmid-based re-expression of PRMT3 in the KO cells markedly restored Tat-dependent HIV-1 transcription (from 25.75-fold to 84.12-fold) (Fig. 2c). Together, these results demonstrate that PRMT3 promotes Tat-dependent HIV-1 transcription.Fig. 2PRMT3 enhances Tat-dependent HIV-1 transcription and latency reversal.a–c Luciferase activities were measured in NH1 cells transfected with shRNA targeting PRMT3 (a) (p = 0.0021, p = 0.0024), Flag-PRMT3 (b) (p = 0.0098), or in NH1 (WT) and PRMT3 knockout (KO) cells cotransfected with control vector or Tat together with or without Flag-PRMT3 (p = 0.0012, p = 0.0006, p = 0.0015, p = 0.0053). d–h Luciferase activities were measured in WT and KO cells treated with DMSO, JQ1 (d) (p = 0.0002, p = 0.0036), PMA (e) (p = 0.0009, p = 0.0036), or NH1 cells added with Tat (f) (p = 0.0131, p = 0.0050, p = 0.0009), treated with JQ1 (5 μM) (g) (p = 0.0013, p = 0.0004, p = 0.00001) or PMA (200 nM) (h) (p = 0.0022, p = 0.0012, p = 0.0008), together with or without SGC707. i, j The relative HIV-1 RNA copies (i) (p = 0.0009, p = 0.0015, p = 0.00003) and p24 expression (j) (p = 0.0012, p = 0.0006, p = 0.0006) in HIV-1 CRF01-AE infected MT4 cells treated with SGC707 was measured. k, l. The HIV-1 RNA copies were measured in HIV-1 infected WT or KO TZM-bl cells (k) (p = 0.0122). The PRMT3 level (right in panel k) and relative mRNA level of gag was detected (l) (p = 0.0012). m, n Representative flow cytometry analysis of 2D10 cells treated with JQ1 (m) or PMA (n) with or without SGC707. o, p The percentage of GFP cells indicates the HIV-1 transcription activation (o: p = 0.0184, p = 0.0022, p = 0.0013, p: p = 0.0003, p = 0.0022, p = 0.0006). q Schematic of HIV-1 detection in patients’ primary cells. r, s HIV-1 RNA copies were detected in CD4 T cells treated with DMSO, JQ1, JQ1 + SGC707 (r) (p = 0.0277), or PMA, PMA + SGC707 (s) (p = 0.0277). Error bars = mean +/− SD of three biological replicates except (r and s) (n = 6). *p < 0.05, **p < 0.01, ***p < 0.001, two-tailed t test. Wilcoxon’s matched-pairs signed-rank test was performed in r and s. Source data are provided as a Source Data file. a–c Luciferase activities were measured in NH1 cells transfected with shRNA targeting PRMT3 (a) (p = 0.0021, p = 0.0024), Flag-PRMT3 (b) (p = 0.0098), or in NH1 (WT) and PRMT3 knockout (KO) cells cotransfected with control vector or Tat together with or without Flag-PRMT3 (p = 0.0012, p = 0.0006, p = 0.0015, p = 0.0053). d–h Luciferase activities were measured in WT and KO cells treated with DMSO, JQ1 (d) (p = 0.0002, p = 0.0036), PMA (e) (p = 0.0009, p = 0.0036), or NH1 cells added with Tat (f) (p = 0.0131, p = 0.0050, p = 0.0009), treated with JQ1 (5 μM) (g) (p = 0.0013, p = 0.0004, p = 0.00001) or PMA (200 nM) (h) (p = 0.0022, p = 0.0012, p = 0.0008), together with or without SGC707. i, j The relative HIV-1 RNA copies (i) (p = 0.0009, p = 0.0015, p = 0.00003) and p24 expression (j) (p = 0.0012, p = 0.0006, p = 0.0006) in HIV-1 CRF01-AE infected MT4 cells treated with SGC707 was measured. k, l. The HIV-1 RNA copies were measured in HIV-1 infected WT or KO TZM-bl cells (k) (p = 0.0122). The PRMT3 level (right in panel k) and relative mRNA level of gag was detected (l) (p = 0.0012). m, n Representative flow cytometry analysis of 2D10 cells treated with JQ1 (m) or PMA (n) with or without SGC707. o, p The percentage of GFP cells indicates the HIV-1 transcription activation (o: p = 0.0184, p = 0.0022, p = 0.0013, p: p = 0.0003, p = 0.0022, p = 0.0006). q Schematic of HIV-1 detection in patients’ primary cells. r, s HIV-1 RNA copies were detected in CD4 T cells treated with DMSO, JQ1, JQ1 + SGC707 (r) (p = 0.0277), or PMA, PMA + SGC707 (s) (p = 0.0277). Error bars = mean +/− SD of three biological replicates except (r and s) (n = 6). *p < 0.05, **p < 0.01, ***p < 0.001, two-tailed t test. Wilcoxon’s matched-pairs signed-rank test was performed in r and s. Source data are provided as a Source Data file. Since JQ1 and phorbol myristate acetate (PMA) are known latency reversal reagents that promote Tat-mediated recruitment of P-TEFb to the HIV-1 LTR to activate transcription, we also monitored PRMT3’s effects on HIV-1 transcription upon exposure to JQ1 and PMA. We observed that PRMT3 KO significantly decreased the extent of HIV-1 transcription induced by JQ1 (from 10.35-fold to 4.14-fold) and by PMA (from 20.28-fold to 12.67-fold) (Fig. 2d, e), supporting the view that PRMT3 regulates HIV-1 transcription in a Tat/P-TEFb dependent manner. Furthermore, we also tested the effect of PRMT3 on HIV-1 transcription using a tetracycline inducible shRNA expression system. We found that after doxycycline-induced PRMT3 knockdown using shRNAs targeting two different regions of PRMT3, a significant decrease in Tat-, JQ1-, or PMA-induced HIV-1 transactivation was observed (Supplementary Fig. 4a–c). To investigate whether PRMT3’s methyltransferase activity is required for the observed stimulatory effects, we treated NH1 cells with a PRMT3 inhibitor (SGC707). The treatment significantly impaired Tat-, JQ1-, and PMA-activated HIV-1 transcription, in a dose-dependent manner (Fig. 2f–h), which showed no obvious inhibitory effect in PRMT3 KO cells (Supplementary Fig. 5a, b). In addition, the catalytic inactive form PRMT3 E338Q was used for further validation of the impact of methyltransferase activity of PRMT3 for HIV-1 transcriptional activation. We found that this E338Q mutant PRMT3 showed significantly decreased ability to support Tat-activation of HIV-1 transcription in comparison to the WT PRMT3 (Supplementary Fig. 5c). These findings establish that PRMT3’s arginine methyltransferase activity contributes to the observed induction of HIV-1 transcription. To demonstrate that the effect of PRMT3 on HIV-1 transcription contributes to HIV-1 replication, we measured HIV-1 replication in MT4 cells, TZM-bl cells, and in the primary CD4 T cells isolated from virologically suppressed HIV-1-infected patients. MT4 cells were infected with HIV-1 CRF01-AE (GX002: accession: GU564222), which is the predominantly transmitted HIV-1 strain in China, and treated with DMSO or SGC707. We found that inhibition of PRMT3 by SGC707 significantly decreased the replication of HIV-1 in MT4 cells by measuring the pol and p24 expression levels in the supernatant (Fig. 2i, j). We also constructed PRMT3 gene knockout in the TZM-bl cell line, a Hela based cell line expressing CD4, CCR5, and CXCR4. WT and PRMT3 KO TZM-bl cells were infected with HIV-1 CRF01-AE. We showed that knockout of PRMT3 resulted in a decrease in HIV-1 replication by measuring the cellular HIV-1 RNA copies and gag expression (Fig. 2k, l). Finally, we measured the effects of PRMT3 knockdown on HIV-1 replication in MT4 cells or primary CD4 T cells isolated from HIV-1 infected patients by targeting PRMT3 using two different PRMT3 siRNAs, and found that knockdown of PRMT3 significantly decrease HIV-1 replication (Supplementary Fig. 6a–e). The impact of PRMT3 on HIV-1 replication was also verified by performing SGC707 treatment, PRMT3 knockdown, or knockout in cells that were infected with another strain, NL4-3 (Supplementary Fig. 7a–e), suggesting that the regulation by PRMT3 on HIV-1 replication is conserved across different HIV-1 subtypes. Together, these data demonstrated that PRMT3 regulates HIV-1 transcription and replication. Given that HIV-1 transcriptional activation is essential for HIV-1 latency reversal, we anticipated that the inhibition of PRMT3 activity would repress HIV-1 latency reversal. Indeed, experiments involving FACS analysis of a well-known HIV-1 latency model, Jurkat 2D10 cells containing an integrated HIV-1-GFP fusion reporter, showed that the SGC707-treated group had significantly impeded JQ1- and PMA-mediated latency reversal compared to the vehicle (DMSO) group (Fig. 2m, n and Supplementary Fig. 8a, b). The percentage of GFP cells indicates the activation effect of HIV-1 transcription (Fig. 2o, p). We then asked whether the inhibition of PRMT3 activity can impair latency reversal of the HIV-1 reservoir in latently infected patients under antiviral treatment. We isolated primary CD4 T cells from 6 HIV-1-infected individuals undergoing suppressive antiretroviral therapy (ART) and then monitored the reactivation of latent HIV-1 transcription (assessed as HIV-1 RNA copies) in cells treated with JQ1 or PMA in the presence or absence of SGC707 (Fig. 2q). The data show that SGC707 significantly inhibited the JQ1- and PMA-induced HIV-1 latency reversal (Fig. 2r, s). Taken together, the results demonstrate that PRMT3 interacts with the HIV-1 LTR and that its arginine methyltransferase activity promotes Tat-dependent HIV-1 transcription and latency reversal. A previous study reported that PRMT3 activates host gene transcription by enhancing the levels of histone H4 arginine 3 asymmetric di-methylation (H4R3Me2a) at promoter regions. In light of these findings and our results showing that the arginine methyltransferase activity of PRMT3 is required for HIV-1 transactivation, we examined whether PRMT3 promotes Tat-activation of HIV-1 transcription by increasing the H4R3Me2a level at the LTR. We examined the global H4R3Me2a levels in wild-type (WT) NH1 and PRMT3 KO cells. Immunoblotting showed a notable reduction in the H4R3Me2a level in the KO cells compared to WT control cells (Fig. 3a). Similarly, reduced H4R3Me2a levels were observed in various cell lines (Jurkat, HeLa, and HEK293T) treated with SGC707 (Fig. 3b).Fig. 3PRMT3 increases H4R3Me2a levels and chromatin accessibility at HIV-1 LTR to promote Tat-dependent recruitment of P-TEFb to the LTR.a, b H4R3Me2a levels were examined in WT and PRMT3 KO cells (a), or in indicated cells treated with DMSO or SGC707 (b). Densitometric analyses of immunoblots are shown. c, d H4R3Me2a occupancy at LTR-luciferase regions in WT and KO cells (c) (p = 0.0050, p = 0.0133, p = 0.0102, p = 0.0047), and cells treated with DMSO or SGC707 (d) (p = 0.0064, p = 0.0009, p = 0.0256, p = 0.1042) were measured. e, f ATAC-Seq was performed in WT and KO cells transfected with Tat using three biological replicates (e). Normalize depth (Reads per million) was calculated using LTR counts divided by the number of sequences to HIV-1 × 1,000,000. The center line, upper, and lower edges of the boxplot represented the mean, first, and third quartile (f). g–i WT and KO cells were cotransfected with Tat and Flag-CycT1 (g) (p = 0.0029, p = 0.0028, p = 0.0109, p = 0.0019) or Flag-CDK9 (h) (p = 0.0050, p = 0.0042, p = 0.0135, p = 0.0183). F1C2 cells were treated with DMSO or SGC707 (i) (p = 0.0007, p = 0.0035, p = 0.0014, p = 0.00009). The occupancy of Flag-CycT1 or Flag-CDK9 at LTR-luciferase regions was measured. j–l Luciferase activities were measured for Tat transfected (j) (p = 0.0017, p = 0.0158), JQ1 (k) (p = 0.0005, p = 0.0065), or PMA (l) (p = 0.000003, p = 0.0011) treated WT and KO cells when CycT1 overexpression. m–q Luciferase activities were measured for Tat transfected (m) (p = 0.0005, p = 0.0041), JQ1 (n) (p = 0.0033, p = 0.0137), or PMA (o) (p = 0.0042, p = 0.0019) treated cells overexpression CycT1 when treated with SGC707, in WT, CDK9 KD cells (p = 0.0147, p = 0.0099, p = 0.0270) (p), or in NH1 cells transfected with indicated plasmid together with Tat (q) (p = 0.0008, p = 0.0025, p = 0.0003). Error bars = mean +/− SD of three biological replicates. *p < 0.05, **p < 0.01, ***p < 0.001, n.s. denotes no significance, two-tailed t test except (f) was assessed by Wilcoxon’s test. Source data are provided as a Source Data file. a, b H4R3Me2a levels were examined in WT and PRMT3 KO cells (a), or in indicated cells treated with DMSO or SGC707 (b). Densitometric analyses of immunoblots are shown. c, d H4R3Me2a occupancy at LTR-luciferase regions in WT and KO cells (c) (p = 0.0050, p = 0.0133, p = 0.0102, p = 0.0047), and cells treated with DMSO or SGC707 (d) (p = 0.0064, p = 0.0009, p = 0.0256, p = 0.1042) were measured. e, f ATAC-Seq was performed in WT and KO cells transfected with Tat using three biological replicates (e). Normalize depth (Reads per million) was calculated using LTR counts divided by the number of sequences to HIV-1 × 1,000,000. The center line, upper, and lower edges of the boxplot represented the mean, first, and third quartile (f). g–i WT and KO cells were cotransfected with Tat and Flag-CycT1 (g) (p = 0.0029, p = 0.0028, p = 0.0109, p = 0.0019) or Flag-CDK9 (h) (p = 0.0050, p = 0.0042, p = 0.0135, p = 0.0183). F1C2 cells were treated with DMSO or SGC707 (i) (p = 0.0007, p = 0.0035, p = 0.0014, p = 0.00009). The occupancy of Flag-CycT1 or Flag-CDK9 at LTR-luciferase regions was measured. j–l Luciferase activities were measured for Tat transfected (j) (p = 0.0017, p = 0.0158), JQ1 (k) (p = 0.0005, p = 0.0065), or PMA (l) (p = 0.000003, p = 0.0011) treated WT and KO cells when CycT1 overexpression. m–q Luciferase activities were measured for Tat transfected (m) (p = 0.0005, p = 0.0041), JQ1 (n) (p = 0.0033, p = 0.0137), or PMA (o) (p = 0.0042, p = 0.0019) treated cells overexpression CycT1 when treated with SGC707, in WT, CDK9 KD cells (p = 0.0147, p = 0.0099, p = 0.0270) (p), or in NH1 cells transfected with indicated plasmid together with Tat (q) (p = 0.0008, p = 0.0025, p = 0.0003). Error bars = mean +/− SD of three biological replicates. *p < 0.05, **p < 0.01, ***p < 0.001, n.s. denotes no significance, two-tailed t test except (f) was assessed by Wilcoxon’s test. Source data are provided as a Source Data file. Next, we investigated the impact of PRMT3 on H4R3Me2a levels specifically at the HIV-1 LTR in WT NH1 and PRMT3 KO cells in the presence of Tat, as well as in NH1 cells expressing Tat and treated with DMSO or SGC707. In both the KO cells and cells upon PRMT3 inhibition, the H4R3Me2a levels were significantly decreased at the LTR (Fig. 3c, d). Having found that PRMT3 deposits H4R3Me2a marks at the LTR, we also performed ATAC-seq in WT and PRMT3 KO NH1 cells that were transfected with a plasmid expressing Tat to examine the effect of PRMT3 on chromatin accessibility at the LTR. Our data showed that the accessibility at the LTR was significantly decreased in PRMT3 KO NH1 cells in comparison to WT cells (Fig. 3e, f). We found that the level of Pol II and Pol II pSer2 decreased at the LTR-luciferase reporter region after PRMT3 KO (Supplementary Fig. 9a, b), which is consistent with the effect of PRMT3 on chromatin accessibility. Together, these data indicate that PRMT3 promotes Tat-activation of HIV-1 transcription by depositing H4R3Me2a marks and enhancing local chromatin accessibility at the LTR. Given the established mechanism of P-TEFb recruitment to the LTR by Tat for HIV-1 transactivation, we asked whether PRMT3’s deposition of H4R3Me2a and change of chromatin accessibility affect the recruitment of P-TEFb. We determined P-TEFb’s occupancy at the LTR in WT and PRMT3 KO cells or in DMSO and SGC707 treated F1C2 cells that stably expressed a Flag-tagged CDK9 by conducting anti-Flag-CycT1 or anti-Flag-CDK9 ChIP-qPCR, and discovered a significant reduction in P-TEFb binding at the LTR in PRMT3 KO cells or in SGC707 treated cells (Fig. 3g–i), reinforcing the notion that PRMT3’s regulatory effect on HIV-1 transcription involves modulating chromatin accessibility and consequently P-TEFb’s recruitment at the LTR. Next, we investigated whether the impact of PRMT3 on P-TEFb recruitment affected the P-TEFb-mediated Tat-dependent HIV-1 transcription by measuring the luciferase level in HeLa cells that were transfected with an empty control vector or plasmid expressing CycT1 under different conditions. The data showed that a significantly impaired Tat-, JQ1-, or PMA-mediated activation of HIV-1 transcription upon PRMT3 deletion (Fig. 3j–l) or inhibition (Fig. 3m–o) was observed especially when CycT1 was overexpressed. Furthermore, we also determined the dependence on another subunit of P-TEFb, CDK9, for PRMT3-mediated activation of HIV-1 transcription. CDK9 knockdown significantly disrupted PRMT3’s stimulatory effect on Tat transactivation (Fig. 3p), further highlighting the requirement of both PRMT3 and P-TEFb for HIV-1 transcriptional activation. To further confirm this point, we co-transfected CycT1 and PRMT3 into NH1 cells to test the effect of co-activation of HIV-1 transcription by PRMT3 and P-TEFb, and observed that the co-expression of CycT1 and PRMT3 resulted in higher transcriptional activation of HIV-1 than the expression of either alone (Fig. 3q). Together, these data suggested that PRMT3 promotes HIV-1 transcription by increasing H4R3Me2a levels and chromatin accessibility at the LTR for enhancement of Tat-dependent recruitment of P-TEFb to the LTR. In addition to PRMT3’s stimulatory effect on chromatin accessibility and transcription from HIV-1, we also examined its potential regulation of specific host genes. By performing ATAC-seq in WT and PRMT3 KO NH1 cells, we detected 9392 down-regulated and 6473 up-regulated regions, including promoter, exon, intron, UTR, and distal intergenic regions in the KO cells (Fig. 4a, Supplementary Fig. 10a–c and Supplementary Data 2). However, PRMT3 does not appear to cause a widespread, global change in chromatin accessibility across the whole genome as evidenced in the analysis from ATAC-seq of WT and PRMT3 KO NH1 cells (Fig. 4b).Fig. 4PRMT3 affects chromatin accessibility and transcription of a subset of host genes.a The Volcano plot displays the differential loci in WT and PRMT3 KO cells. The x-axis value indicates the log2fold change of average density of peaks in WT and KO cells. The y-axis value indicates –log10P. The dotted line represents p = 0.05. b A heatmap showing the distribution of accessible regions upstream of the TSS and downstream of the TES identified by ATAC-seq using three biological replicates of WT and KO cells. c The Volcano plot showing differentially expressed genes, down-regulated (blue) and up-regulated (red), in WT and KO cells using RNA-seq. d The Volcano map was used to display the genes with differential transcription level and chromatin accessibility in WT and KO cells from the combining analysis of ATAC-Seq and RNA-Seq results. e The plot (Up) displayed the normalized ATAC-Seq signal of three sets of genes, and the heatmaps (Down) showed the ATAC-seq signal for each gene that displayed in a row. f The related signaling pathway of PRMT3-regualted genes showed by KEGG analysis. g–i Representative tracks of ATAC-seq (WT: blue; KO: pink) and RNA-Seq (WT: blue; KO: pink) signals showing decreased ATAC-Seq signals along with decreased RNA-seq signals within PRDM1, SLITRK5, and TGFB2 regions in KO cells compared to those in WT cells. The boxed area indicates the representative decreased areas. The schematics for the three genes were displayed at the bottom of the figure (green). j–l RT-q-PCR analysis showed the expression of three representative genes in WT and KO cells (j: p = 0.0054, k: p = 0.0005, l: p = 0.0002). GAPDH was used for normalization in the RT-qPCR. m–o ChIP-qPCR analysis detected PRMT3's association with three regions of the promoters (P1-P3) of selected genes(m: p = 0.0032, p = 0.0016, p = 0.0038, n: p = 0.0047, p = 0.0043, p = 0.0037, o: p = 0.0019, p = 0.0004, p = 0.0061). Error bars = mean +/− SD of three biological replicates. *p < 0.05, **p < 0.01, ***p < 0.001, two-tailed t test. Source data are provided as a Source Data file. a The Volcano plot displays the differential loci in WT and PRMT3 KO cells. The x-axis value indicates the log2fold change of average density of peaks in WT and KO cells. The y-axis value indicates –log10P. The dotted line represents p = 0.05. b A heatmap showing the distribution of accessible regions upstream of the TSS and downstream of the TES identified by ATAC-seq using three biological replicates of WT and KO cells. c The Volcano plot showing differentially expressed genes, down-regulated (blue) and up-regulated (red), in WT and KO cells using RNA-seq. d The Volcano map was used to display the genes with differential transcription level and chromatin accessibility in WT and KO cells from the combining analysis of ATAC-Seq and RNA-Seq results. e The plot (Up) displayed the normalized ATAC-Seq signal of three sets of genes, and the heatmaps (Down) showed the ATAC-seq signal for each gene that displayed in a row. f The related signaling pathway of PRMT3-regualted genes showed by KEGG analysis. g–i Representative tracks of ATAC-seq (WT: blue; KO: pink) and RNA-Seq (WT: blue; KO: pink) signals showing decreased ATAC-Seq signals along with decreased RNA-seq signals within PRDM1, SLITRK5, and TGFB2 regions in KO cells compared to those in WT cells. The boxed area indicates the representative decreased areas. The schematics for the three genes were displayed at the bottom of the figure (green). j–l RT-q-PCR analysis showed the expression of three representative genes in WT and KO cells (j: p = 0.0054, k: p = 0.0005, l: p = 0.0002). GAPDH was used for normalization in the RT-qPCR. m–o ChIP-qPCR analysis detected PRMT3's association with three regions of the promoters (P1-P3) of selected genes(m: p = 0.0032, p = 0.0016, p = 0.0038, n: p = 0.0047, p = 0.0043, p = 0.0037, o: p = 0.0019, p = 0.0004, p = 0.0061). Error bars = mean +/− SD of three biological replicates. *p < 0.05, **p < 0.01, ***p < 0.001, two-tailed t test. Source data are provided as a Source Data file. To examine the effects of PRMT3-driven chromatin modifications on host gene expression, RNA-Seq analysis was conducted in WT and PRMT3 KO cells, and the results showed that PRMT3 KO resulted in altered expression of 256 genes (119 down-regulated, 137 up-regulated) (Fig. 4c, Supplementary Fig. 10d–f and Supplementary Data 3). Further integrated analysis combining both ATAC-Seq and RNA-Seq showed that only a limited subset of host genes is regulated by PRMT3 through altering chromatin accessibility (Fig. 4d). The signal of chromatin accessibility in WT cells is significantly higher than that in KO cells, which correlates with the gene expression level (Fig. 4e, Up). The signal for each gene was displayed in a row (Fig. 4e, Down). KEGG analysis showed that these PRMT3-regulated genes are predominantly involved in several signaling pathways (Fig. 4f). A subsequent integrated analysis of the PRMT3 KO cells revealed significant reductions in both chromatin accessibility and expression levels of a few genes, including PRDM1, SLITRK5, and TGFB2, pinpointing specific loci where PRMT3’s absence markedly reduced transcriptional activity (Fig. 4g–i). A parallel analysis by qRT-PCR validated that PRMT3 KO resulted in significantly decreased mRNA production from these genes (Fig. 4j–l). In light of our earlier demonstration that PRMT3 binds to the HIV-1 LTR, we hypothesized that PRMT3 likely associates with the promoter regions of the three genes, PRDM1, SLITRK5, and TGFB2, to deposit H4R3Me2a to activate their transcription. To test this hypothesis, we performed ChIP-qPCR to target multiple promoter regions and detected a significant enrichment of the PRMT3 signal at the tested regions of all three examined loci (Fig. 4m–o). Thus, in addition to showing that PRMT3 binds to the HIV-1 LTR and catalyzes deposition of H4R3Me2a marks that promote Tat-dependent HIV-1 transcription, these results support the notion that PRMT3 also epigenetically regulates the transcription of a subset of host genes that play key roles in certain signal transduction pathways. Next, ATAC-seq footprinting analyses were performed in WT and PRMT3 KO cells to investigate the reason behind the apparently quite narrow specificity of the PRMT3-targeted loci. The analyses using Discriminative Regular Expression Motif Elicitation (DREME) enrichment analyses reveal that PRMT3 impacts accessibility at a specific motif containing a core GGAAT sequence located in a somewhat less conserved but still preferred context containing additional nucleotides (Fig. 5a), a sequence that is also present within the HIV-1 LTR at two distinct locations (Fig. 5b).Fig. 5PRMT3 co-localizes with TEAD4 at LTR for their synergistic activation of HIV-1 transcription through a motif containing the GGAAT core sequence.a DREME enrichment analyses of ATAC-seq in WT and PRMT3 KO cells. b GGAAT-motif location in LTR mutants. c PRMT3’s occupancy at LTR-luciferase regions in cells transfected with WT or Δ9 + 7 bp LTR-luciferase reporter (p = 0.0108, p = 0.0020, p = 0.0100, p = 0.0013) was measured. d Luciferase activities were measured in cells cotransfected with WT or mutant LTR-luciferase reporter together with control or Flag-PRMT3 plasmid with or without Tat (p = 0.0056, p = 0.0158, p = 0.0091). e The reported TEAD4 binding motif. f Venn diagram of PRMT3- against TEAD4-associated genes. g Myc-TEAD4’s occupancy at LTR-luciferase regions (p = 0.0029, p = 0.0425; p = 0.0026, p = 0.0029; p = 0.0006, p = 0.0184; p = 0.0001, p = 0.0015) in cells with or without Tat was measured. h–j Luciferase activities were measured in NH1 cells transfected with shRNAs targeting TEAD4 (h) (p = 0.0020, p = 0.0009), the plasmid expressing Myc-TEAD4 (i) (p = 0.0134), or in HeLa cells cotransfected with WT or Δ9 + 7 bp LTR construct together with Myc-TEAD4 and Tat (p = 0.0017, p = 0.0291) (j). k–m Myc-TEAD4’s occupancy at LTR-luciferase regions in cells transfected with WT or Δ9 + 7 bp LTR (k) (p = 0.0110, p = 0.0053, p = 0.0262, p = 0.0032), or in WT and KO cells (l) (p = 0.0090, p = 0.0067, p = 0.00006, p = 0.0007), or in cells treated with DMSO or SGC707 (m) (p = 0.0016, p = 0.0034, p = 0.0019, p = 0.0005) was measured. n Luciferase activities were measured in cells cotransfected with Myc-TEAD4 and Flag-PRMT3 (p = 0.0046, p = 0.0059) versus cells transfected with either alone (p = 0.0006, p = 0.0037). o, p PRMT3’s occupancy at LTR-luciferase regions with or without TEAD4 overexpression (o) (p = 0.0007, p = 0.0104, p = 0.0061, p = 0.0122), or knockdown (p) (p = 0.00003, p = 0.0016, p = 0.0016, p = 0.0013) was measured. Error bars = mean +/− SD of three biological replicates. *p < 0.05, **p < 0.01, ***p < 0.001, two-tailed t test. Source data are provided as a Source Data file. a DREME enrichment analyses of ATAC-seq in WT and PRMT3 KO cells. b GGAAT-motif location in LTR mutants. c PRMT3’s occupancy at LTR-luciferase regions in cells transfected with WT or Δ9 + 7 bp LTR-luciferase reporter (p = 0.0108, p = 0.0020, p = 0.0100, p = 0.0013) was measured. d Luciferase activities were measured in cells cotransfected with WT or mutant LTR-luciferase reporter together with control or Flag-PRMT3 plasmid with or without Tat (p = 0.0056, p = 0.0158, p = 0.0091). e The reported TEAD4 binding motif. f Venn diagram of PRMT3- against TEAD4-associated genes. g Myc-TEAD4’s occupancy at LTR-luciferase regions (p = 0.0029, p = 0.0425; p = 0.0026, p = 0.0029; p = 0.0006, p = 0.0184; p = 0.0001, p = 0.0015) in cells with or without Tat was measured. h–j Luciferase activities were measured in NH1 cells transfected with shRNAs targeting TEAD4 (h) (p = 0.0020, p = 0.0009), the plasmid expressing Myc-TEAD4 (i) (p = 0.0134), or in HeLa cells cotransfected with WT or Δ9 + 7 bp LTR construct together with Myc-TEAD4 and Tat (p = 0.0017, p = 0.0291) (j). k–m Myc-TEAD4’s occupancy at LTR-luciferase regions in cells transfected with WT or Δ9 + 7 bp LTR (k) (p = 0.0110, p = 0.0053, p = 0.0262, p = 0.0032), or in WT and KO cells (l) (p = 0.0090, p = 0.0067, p = 0.00006, p = 0.0007), or in cells treated with DMSO or SGC707 (m) (p = 0.0016, p = 0.0034, p = 0.0019, p = 0.0005) was measured. n Luciferase activities were measured in cells cotransfected with Myc-TEAD4 and Flag-PRMT3 (p = 0.0046, p = 0.0059) versus cells transfected with either alone (p = 0.0006, p = 0.0037). o, p PRMT3’s occupancy at LTR-luciferase regions with or without TEAD4 overexpression (o) (p = 0.0007, p = 0.0104, p = 0.0061, p = 0.0122), or knockdown (p) (p = 0.00003, p = 0.0016, p = 0.0016, p = 0.0013) was measured. Error bars = mean +/− SD of three biological replicates. *p < 0.05, **p < 0.01, ***p < 0.001, two-tailed t test. Source data are provided as a Source Data file. To examine the dependence of PRMT3 on this motif for LTR-binding, we constructed two LTR mutants harboring a single Δ13 bp (− 224 ~ − 236bp) or double Δ9 + 7 bp (− 226 ~ − 234bp, + 84 ~ + 90 bp) deletions of the core GGAAT sequence plus a few surrounding nucleotides (Fig. 5b). The WT or LTR mutants were fused to a luciferase reporter gene and co-transfected with Tat into HeLa cells. ChIP-qPCR analysis of PRMT3 bound to the WT, Δ13 bp, or Δ9 + 7 bp LTR-luciferase reporter region was performed. We found that the association of PRMT3 with the Δ9 + 7 bp LTR-luciferase reporter region was significantly decreased (Fig. 5c), while the association of PRMT3 with the Δ13 bp LTR-luciferase reporter showed no significant change in comparison with the WT (Supplementary Fig. 11), suggesting that PRMT3’s activation of HIV-1 transcription is dependent on its association with the motif within LTR. To investigate whether Tat-activated HIV-1 transcription by PRMT3 is dependent on this motif, we cotransfected the WT or mutant LTR constructs with Flag-tagged PRMT3 with or without a Tat-expressing plasmid into HeLa cells. Measurement of luciferase signals showed that Flag-PRMT3 significantly activated Tat-dependent transcription of the WT LTR. In contrast, it failed to activate transcription from the double-deletion mutant Δ9 + 7 bp LTR, while the Δ13 bp single deletion resulted in a milder decrease in luciferase signal (Fig. 5d). This result suggests that the activation of HIV-1 transcription by PRMT3 depended on the motif containing the core sequence GGAAT. Interestingly, this GGAAT sequence is the same core sequence in the known binding motif for sequence-specific transcription factor TEAD4 (Fig. 5e). Venn diagram analysis of differential expression genes (DEGs) identified in PRMT3 KO cells in the current study compared with those detected in TEAD4 KD cells by Biswarup and colleagues showed that 30 genes are jointly regulated by both PRMT3 and TEAD4 (Fig. 5f), indicating a possible cooperative transcriptional regulation mechanism facilitated by their interaction at the motif. To investigate TEAD4’s potential contribution to Tat-dependent HIV-1 transcriptional activation, we performed ChIP-qPCR analysis and found that TEAD4 indeed physically associates with the LTR, which is further increased in the presence of Tat (Fig. 5g). Next, we knocked down TEAD4 expression in NH1 cells and found that the KD significantly decreased Tat-dependent but not basal HIV-1 transcription (Fig. 5h). Conversely, NH1 cells transfected with a plasmid overexpressing TEAD4 showed significantly increased Tat-activated HIV-1 transcription (Fig. 5i). Given this positive effect on Tat-transactivation displayed by TEAD4, we examined whether it is dependent on the GGAAT motif in the LTR. HIV-1 transcriptional activation was measured after co-transfecting HeLa cells with plasmids expressing TEAD4 and the luciferase reporter gene under the control of WT LTR or the mutant LTR lacking both GGAAT motifs. In the presence of Tat, TEAD4 significantly activated Tat-dependent HIV-1 transcription from WT but not the mutant LTR (Fig. 5j). Further ChIP-qPCR analysis of TEAD4’s binding to the LTR-luciferase reporter region showed significantly lower TEAD4 binding to the mutant LTR-luciferase reporter region compared to the WT (Fig. 5k). These results demonstrate that TEAD4 activates Tat-dependent HIV-1 transcription by associating with the LTR through the binding motif containing the core GGAAT sequence. Building upon our aforementioned data illustrating PRMT3’s regulation of HIV-1 transcription through modulating chromatin accessibility, we compared TEAD4 enrichment at the LTR between WT and PRMT3 KO cells to determine whether PRMT3 influences TEAD4’s recruitment to the LTR. Our ChIP-qPCR analyses showed that in PRMT3 KO cells, there was a significantly decreased TEAD4 occupancy at the LTR-luciferase reporter region (Fig. 5l). Notably, a significant decrease was also observed upon exposure to the PRMT3 inhibitor SGC707 (Fig. 5m). In addition, the binding of TEAD4 in DMSO or SGC707 treated Jurkat 2D10 cells was detected and results showed that TEAD4’s association with LTR significantly decreased in SGC707 treated cells (Supplementary Fig. 12). These data suggested that PRMT3 promotes the TEAD4 binding at the LTR. In light of the above results, we next assessed the combinatorial effects of PRMT3 and TEAD4 on Tat-dependent HIV-1 transcription by measuring luciferase signals in NH1 cells co-transfected with plasmids expressing both TEAD4 and PRMT3 or either alone. The data showed that the co-expression of both proteins resulted in a higher level of activation compared to either alone (Fig. 5n), indicating that the two proteins cooperate to increase HIV-1 transcription. As a potential explanation of this cooperativity, we found that the TEAD4 overexpression led to a significantly enhanced association of PRMT3 at the LTR (Fig. 5o), while TEAD4 knockdown resulted in a significant decrease in PRMT3’s binding at the LTR-luciferase reporter region (Fig. 5p), suggesting that TEAD4 promotes the PRMT3 binding. In addition to the HIV-1 LTR, we asked whether this PRMT3-TEAD4 mutual promotion and co-localization also occur on host genes. To address this, we conducted ChIP-qPCR to examine the presence of TEAD4 at the loci, including PRDM1, SLITRK5, and TGFB2, where we previously detected the PRMT3 occupancy. Indeed, TEAD4 binding was detected at the promoters of all three examined host genes (Supplementary Fig. 13a−c), which was significantly decreased in PRMT3 KO cells or cells exposure to PRMT3 inhibitor treatment (Supplementary Fig. 14a−f), expanding the relevance of our findings beyond HIV-1 to potentially include cooperative PRMT3/TEAD4 transcriptional regulation of host genes. Together, these findings collectively reveal a cooperative activation of HIV-1 transcription, as well as a small subset of host genes, by PRMT3 and TEAD4, which apparently results from the mutual promotion of bindings to a common motif. In light of the above results showing co-localization of PRMT3 and TEAD4 at the LTR, resulting in their cooperative activation of HIV-1, we examined whether PRMT3 physically interacts with TEAD4. Indeed, co-IP experiments revealed a strong interaction between the two (Fig. 6a, b). The interaction between PRMT3 and TEAD4 was also demonstrated in the GST pull-down assay (Fig. 6c). Furthermore, consistent with the earlier demonstration that Tat promotes TEAD4’s binding at the LTR-luciferase reporter region (Fig. 5g), the co-IP experiments showed that Tat also interacts with TEAD4 (Fig. 6d, e), again supporting a Tat-dependent mechanism for HIV-1 transactivation by TEAD4.Fig. 6PRMT3 interacts with TEAD4 and P-TEFb to form a transcriptional hub that co-localizes in the nucleus.a, b HEK293T cells were cotransfected with empty vector or Flag-PRMT3 with Myc-TEAD4 (a), or Myc-TEAD4 with Flag-PRMT3 (b) for 40 h. Flag-PRMT3 (a) or Myc-TEAD4 (b) was immunoprecipitated using anti-Flag antibody or anti-Myc antibody, following with western blot analysis for the indicated protein. c GST pull-down assay was performed using GST-PRMT3 and His-TEAD4 that were purified from E. Coli. The products from the pull-down assay were subjected to western blot analysis for the detection of the indicated proteins. d, e HEK293T cells that were cotransfected with empty vector or plasmid expressing Myc-TEAD4 together with Flag-Tat (d), or plasmid expressing Flag-Tat together with Myc-TEAD4 (e) for 40 hours. Myc-TEAD4 (d) or Flag-Tat (e) was immunoprecipitated using anti-Myc or anti-Flag IP, following with western blot analysis for the indicated protein. f, g Whole cell lysates of HEK293T cells that were transfected with empty vector or plasmid expressing Flag-PRMT3 (f), or plasmid expressing Flag-CycT1 (g) for 40 h were used for anti-Flag IP, following with western blot analysis for the indicated protein. h, i Whole cell lysates of HEK293T cells that were transfected with empty vector or plasmid expressing Myc-TEAD4 (h), or plasmid expressing Flag-CycT1 together with Myc-TEAD4 (i) for 40 h were used for anti-Myc (h) or anti-Flag (i) IP following with western blot analysis for the indicated protein. j–l Whole cell lysates of Hela cells was used for chromatin pellets collection with lysis buffer and immunoprecipitated using anti-IgG, anti-PRMT3, anti-TEAD4, or anti-CDK9 antibody, following with western blot analysis for the indicated protein. m HEK293T cells were cotransfected with plasmids expressing HA-PRMT3 together with EGFP-CycT1 (green), Myc-TEAD4, or EGFP-CycT1 + Myc-TEAD4 for 40 h. Protein localization was detected by immunofluorescence using anti-HA antibodies (yellow) or anti-Myc antibodies (red), and 40,6-diamidino-2-phenyl-indole (DAPI) (nuclei, blue) for confocal microscopy analysis. Scale bars, 10 μm. All western blots are representative of three independent experiments. a, b HEK293T cells were cotransfected with empty vector or Flag-PRMT3 with Myc-TEAD4 (a), or Myc-TEAD4 with Flag-PRMT3 (b) for 40 h. Flag-PRMT3 (a) or Myc-TEAD4 (b) was immunoprecipitated using anti-Flag antibody or anti-Myc antibody, following with western blot analysis for the indicated protein. c GST pull-down assay was performed using GST-PRMT3 and His-TEAD4 that were purified from E. Coli. The products from the pull-down assay were subjected to western blot analysis for the detection of the indicated proteins. d, e HEK293T cells that were cotransfected with empty vector or plasmid expressing Myc-TEAD4 together with Flag-Tat (d), or plasmid expressing Flag-Tat together with Myc-TEAD4 (e) for 40 hours. Myc-TEAD4 (d) or Flag-Tat (e) was immunoprecipitated using anti-Myc or anti-Flag IP, following with western blot analysis for the indicated protein. f, g Whole cell lysates of HEK293T cells that were transfected with empty vector or plasmid expressing Flag-PRMT3 (f), or plasmid expressing Flag-CycT1 (g) for 40 h were used for anti-Flag IP, following with western blot analysis for the indicated protein. h, i Whole cell lysates of HEK293T cells that were transfected with empty vector or plasmid expressing Myc-TEAD4 (h), or plasmid expressing Flag-CycT1 together with Myc-TEAD4 (i) for 40 h were used for anti-Myc (h) or anti-Flag (i) IP following with western blot analysis for the indicated protein. j–l Whole cell lysates of Hela cells was used for chromatin pellets collection with lysis buffer and immunoprecipitated using anti-IgG, anti-PRMT3, anti-TEAD4, or anti-CDK9 antibody, following with western blot analysis for the indicated protein. m HEK293T cells were cotransfected with plasmids expressing HA-PRMT3 together with EGFP-CycT1 (green), Myc-TEAD4, or EGFP-CycT1 + Myc-TEAD4 for 40 h. Protein localization was detected by immunofluorescence using anti-HA antibodies (yellow) or anti-Myc antibodies (red), and 40,6-diamidino-2-phenyl-indole (DAPI) (nuclei, blue) for confocal microscopy analysis. Scale bars, 10 μm. All western blots are representative of three independent experiments. As both PRMT3 and TEAD4 regulate HIV-1 transcription in a Tat-dependent manner, and HIV-1 transcription is known to be highly dependent on the Tat/P-TEFb complex, we also examined whether PRMT3 and TEAD4 both interact with P-TEFb. Co-IP experiments revealed clear signal of both CycT1 and CDK9 in the pull-down products upon immunoprecipitation of Flag-PRMT3 (Fig. 6f). Consistent with this result, PRMT3 was present in the immunoprecipitation products of Flag-CycT1 (Fig. 6g). Furthermore, we performed reciprocal co-IPs using Myc-TEAD4 or Flag-CycT1 as the bait to test whether TEAD4 and P-TEFb can also interact with each other. The data in Fig. 6h and i indicate that in addition to pulling down PRMT3, both TEAD4 and CycT1 were able to co-precipitate each other. Furthermore, the interaction among PRMT3, TEAD4, and P-TEFb was also demonstrated by performing endogenous IP (Fig. 6j–l). The above co-IP results have revealed physical interactions that exist among PRMT3, TEAD4, and P-TEFb. To validate this conclusion, we performed immunofluorescence staining and found that PRMT3, TEAD4, and CycT1 demonstrated pairwise partial co-localization in the nucleus (Fig. 6m). Collectively, these biochemical and cell biology results support a scenario wherein PRMT3 modifies chromatin to form a nuclear transcriptional hub that also contains TEAD4 and P-TEFb for transcriptional activation and latency reversal of HIV-1 (Fig. 7).Fig. 7The hypothesis model of PRMT3’s role in the regulation of HIV-1 transcription and latency reversal.PRMT3 increases chromatin accessibility by elevating H4R3Me2a level to form a transcriptional hub containing TEAD4, P-TEFb, and Tat at the HIV-1 promoter for viral transcription and latency reversal. PRMT3 increases chromatin accessibility by elevating H4R3Me2a level to form a transcriptional hub containing TEAD4, P-TEFb, and Tat at the HIV-1 promoter for viral transcription and latency reversal. The elusive nature of latent HIV-1 reservoirs presents a formidable challenge to eradicating the virus from infected individuals. Despite the various strategies employing either the latency reversal agents (LRAs) or latency promoting agents (LPAs), which are aimed at either purging or deeply silencing the latent viral reservoirs, these agents have yet to be made into effective drugs for curing HIV/AIDS and targets are still needed for the development of therapies. Furthermore, the premise of using the current and future versions of LRAs or LPAs for treating HIV-1 hinges on the potential to epigenetically modulate the HIV-1 promoter activity, aiming for a sustained activation or suppression of viral transcription, respectively. The significance of targeting epigenetic control of viral transcription was highlighted recently in a human clinical trial, where the combination of panobinostat, a potent pan-histone deacetylase inhibitor, with interferon-α2a resulted in an enhanced vulnerability of latent HIV-1 reservoir cells, underscoring the pivotal role of epigenetic manipulation in combating HIV-1 persistence. In the current study, by performing the dCas9-targeted LTR interactome screening, we have identified PRMT3 as a positive regulator to promote HIV-1 latency reversal. Mechanistically, PRMT3 activates HIV-1 transcription by interaction with TEAD4 and the two proteins co-localize at the LTR by using specific TEAD4-binding motifs, thereby regulating chromatin accessibility and facilitating P-TEFb/Tat recruitment. Thus, our findings propose a promising epigenetic strategy to combat latent HIV-1 infection, underscoring PRMT3 and its partners as promising therapeutic targets for anti-HIV-1 drug development. Our study demonstrates TEAD4 as a positive regulator of HIV-1 transcription, acting alongside PRMT3 through a specific interaction with the LTR’s GGAAT motif. This discovery expands our current understanding of TEAD4’s regulatory roles in host cells and suggests its potential regulation of HIV-1 viral expression. The specificity of TEAD4’s interaction with the motif containing the core GGAAT sequence within the HIV-1 LTR and its synergy with PRMT3 in promoting Tat-dependent HIV-1 transcription and a specific subset of host gene transcription is particularly intriguing given TEAD4’s established roles in cell survival, proliferation, tissue regeneration, stem cell maintenance, embryonic trophoblast and organ development and tumorigenesis. Specifically, we selected three host genes, PRDM1, SLITRK5, and TGFB2, for further validation of our proposed PRMT3-TEAD4 regulation model. Our data demonstrate that in addition to the HIV-1 LTR, PRMT3 and TEAD4 indeed associated with these cellular gene promoters, where the binding of TEAD4 significantly decreases after the depletion of PRMT3, suggesting a common regulatory mechanism involving the PRMT3-TEAD4 complex that exists at the HIV-1 LTR, as well as the promoters of selected host genes. At the HIV-1 LTR, the interaction of the PRMT3-TEAD4 complex with Tat and the P-TEFb provides further regulation of HIV-1 transcription. On cellular gene promoters such as the three genes mentioned above, it is conceivable that transcription can also be modulated through controlling the function and/or binding of the PRMT3-TEAD4 complex to the GGAAT motif in response to changes in physiological or pathological conditions. This co-regulation of viral and host gene transcription by PRMT3-TEAD4 underscores a finely tuned regulatory mechanism that integrates the cellular and viral transcriptional control. Furthermore, TEAD4 has also been demonstrated to interact with P-TEFb, which is the core transcriptional component of the Super Elongation Complex (SEC) essential for Tat-activated HIV-1 transcription. Given the fact that TEAD4 is a DNA sequence-specific transcription factor, it remains to be investigated whether it plays a key role in recruiting SEC to both host and HIV-1 gene promoters that contain the TEAD4-binding motif. While previous studies have indicated PRMT3’s involvement as a key factor in mediating host responses to viral infection, its specific role in HIV-1 infection and latency remains largely unexplored. For instance, in zebrafish, PRMT3 has been shown to negatively regulate antiviral responses. Similarly, recent studies utilizing the PRMT3 inhibitor, SGC707 treatment, or PRMT3 knockout mice have revealed PRMT3’s facilitation of HSV-1 infection. In conjunction with our current discovery demonstrating PRMT3’s promotion of HIV-1 transcription to reverse latency, these findings suggest that PRMT3 may serve as a potential target for broad-spectrum antiviral drugs. However, future comprehensive research is imperative to fully comprehend the extent of PRMT3’s involvement in mediating viral-host interactions. It is noteworthy that besides PRMT3, other PRMTs have also been implicated in regulating anti-HIV activity. For instance, PRMT6 has been shown to inhibit HIV-1 replication in vitro by directly methylating several HIV-1 proteins, including Tat, Rev, and nucleocapsid protein, thereby interfering with their functions. Similarly, PRMT2 was recently discovered to suppress HIV-1 transcription by methylating Tat and promoting the phase separation of P-TEFb and Tat. Thus, in addition to PRMT3, other members of the PRMT family should be explored as potential targets for developing effective antiviral drugs. In addition to its effects on viral gene expression, PRMT3’s regulation of host gene transcription has been implicated in various cellular processes, including tumorigenesis, oxaliplatin resistance in liver cancer, retinoic acid signaling, and hepatic lipogenesis. However, the precise mechanism(s) underlying these transcriptional effects of PRMT3 remain unclear. In this study, the comparison of ATAC-Seq and RNA-Seq data between WT and PRMT3 KO cells reveals that PRMT3 selectively regulates chromatin accessibility and transcription of a small subgroup of human genes, including some well-studied genes such as TGFB2, which is relevant to the potential pathogenic effects of PRMT3 in tumorigenesis. The selectivity of PRMT3’s action is apparently achieved through forming a complex with TEAD4, which then co-localizes at specific gene promoter regions via binding to the TEAD4-recognition motif. Together, these findings have revealed the mechanistic basis for PRMT3’s control of the transcription of HIV-1 and a selected group of host genes. Although the PRMT3-created transcription hub containing TEAD4 and P-TEFb is demonstrated in the current study as important for HIV-1 to escape latency, it is presently unknown whether it also plays a role during the establishment of latency. It is conceivable that the loss of expression/function of any component of the hub could be responsible in this latter process. Further studies are thus necessary to test this hypothesis and explore the possibility of targeting the PRMT3 transcription hub to cure HIV/AIDS. The HIV-positive patient’s sample collection was approved by the Fifth Medical Center of the Chinese PLA General Hospital (KY-2023-10-67-1). The committee approved the experiments and confirmed that all experiments conform to the relevant regulatory standards. The informed consent was obtained from all the patients for the publication of data in this study. NH1, PRMT3 KO, HeLa, HEK293T, F1C2, and CDK9 KD cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM; BI) supplemented with 10% fetal bovine serum (FBS) (Corning), 1% penicillin-streptomycin at 37 °C with 5% CO2. Jurkat 2D10 and MT4 cells were cultured in RPMI 1640 with 10% FBS and 1% penicillin-streptomycin at 37 °C with 5% CO2. For HeLa cell and HEK293T cell expression, DNA was transfected with polyethylenimine (PEI) (Polysciences). All transfections were performed according to the manufacturer’s protocol. The Luciferase Reporter Assay System was purchased from Promega. Primers, plasmids, and antibodies are listed in Supplementary Data 4, 5. sgRNAs synthesis were transcribed by using HiScribe™ Quick T7 High Yield RNA Synthesis Kit (NEB) in vitro following the instructions. The plasmid PCT310 containing dCas9-3 × Flag was obtained from Robert Tjian Lab as a gift. GST-tagged dCas9-3 × Flag was expressed and purified by using the E. coil BL21 expression system slightly modified from the previous study. The product was eluted from Ni-NTA resin by using elution buffer [500 mM NaCl, 20 mM Tris (pH7.5), 5% glycerol, 200 mM imidazole, 20 mM 2-mercaptoethanol] and analyzed by running the SDS-PAGE and followed by Coomassie Blue Staining. Protein was aliquoted and stored at − 80 °C. Then, the complex formation was performed by incubating 2.5 μg sgRNAs with 10 μg dCas9-3 × Flag in 200 μl reaction volume at a molar ratio of 3:1 at room temperature for 1 h. One million NH1 cells with an integrated with HIV-1 5’ LTR-driven luciferase reporter construct, which contains 424 bp of LTR. Cells were fixed with 1% formaldehyde for 15 min at room temperature and 0.125 M Glycine for 5 min to quench unreacted formaldehyde at room temperature. Use 20 ml of cold PBS to wash cells twice. Scrape cells with 2 ml cold PBS containing protease inhibitor and dithiothreitol (DTT). Spin at 700 × g at 4 °C for 3 min to pellet cells. Cells were lysed with lysis buffer [1% SDS, 10 mM EDTA, 50 mM Tris, pH 8.1] and incubated on ice for 20 min. Cells were sheared with a Covaris sonicator M220 until genomic DNA was visualized to be at 200–1000 bp. The DNA was cleared with a high-speed spin (15,000 g for 10 min) at 4 °C. The DNA was diluted by ChIP Dilution Buffer [0.01% SDS, 1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris-HCl, pH 8.1, 167 mM NaCl]. dCas9-3 × Flag/sgRNAs complex was added to the sheared DNA and incubated at 4 °C overnight. The anti-Flag M2 agarose resin (Sigma) was added, followed by a 3 h incubation at 4 °C. The resin was spun down at 1500 × g at 4 °C for 1 min and washed with high salt wash buffer [0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, pH 8.1, 300 mM NaCl] for once, low salt wash buffer [0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, pH 8.1, 150 mM NaCl] for three times, and TE buffer [10 mM Tris-HCl, 1 mM EDTA, pH 8.0] for twice, with each wash for 5 min rotated at 4 °C. The products concentrated by beads were used for subsequent experiments. The sample of dCas9-3 × Flag-sgLTRs and dCas9-3 × Flag-sgGal4 (control) immnuoprecipitated proteins were prepared and analyzed by MS. In general, the eluted LTR binding proteins were reduced in 20 mM DTT at 95 °C for 5 min, and subsequently alkylated in 50 mM iodoacetamide for 30 min in the dark at room temperature. After alkylation, the samples were transferred to a 10 kD centrifugal spin filter (Millipore) and sequentially washed with 200 μl of 8 M urea for three times and 200 μl of 50 mM ammonium bicarbonate for two times by centrifugation at 14,000 × g. Next, tryptic digestion was performed by adding trypsinat 1:50 (enzyme/substrate, m/m) in 200 μl of 50 mM ammonium bicarbonate at 37 °C for 16 hours. Peptides were recovered by transferring the filter to a collection tube and spinning at 14,000 × g. To increase the yield of peptides, the filter was washed twice with 100 μl of 50 mM ammonium bicarbonate. Peptides were desalted by StageTips. MS experiments were performed on a nanoscale EASY-nLC 1200 UHPLC system (Thermo Fisher Scientific) connected to an Orbitrap Fusion Lumos equipped with a nanoelectrospray source (Thermo Fisher Scientific). Mobile phase A contained 0.1% formic acid (v/v) in water; mobile phase B contained 0.1% formic acid in 80% acetonitrile (ACN). The peptides were dissolved in 0.1% formic acid (FA) with 2% acetonitrile and separated on an RP-HPLC analytical column (75 μm × 25 cm) packed with 2 μm C18 beads (Thermo Fisher Scientific) using a linear gradient ranging from 5% to 22% ACN in 90 min and followed by a linear increase to 35% B in 20 min at a flow rate of 300 nl/min. The Orbitrap Fusion Lumos acquired data in a data-dependent manner, alternating between full-scan MS and MS2 scans. The spray voltage was set at 2.2 kV, and the temperature of the ion transfer capillary was 300 °C. The MS spectra (350–1500 m/z) were collected with 120,000 resolutions, AGC of 4 × 10, and 50 ms maximal injection time. Selected ions were sequentially fragmented by HCD with 30% normalized collision energy, specified isolated windows 1.6 m/z, and 15,000 resolutions. AGC of 5 × 10 and 40 ms maximal injection time were used. Dynamic exclusion was set to 40 s. Unassigned ions or those with a charge of 2 + and >7 + were rejected for MS/MS. Raw data was processed using Proteome Discoverer (PD, version 2.4), and MS/MS spectra were searched against the reviewed SwissProt human proteome database. Only peptides with at least six amino acids in length were considered. The peptide and protein identifications were filtered by PD to control the false discovery rate (FDR) <1%. At least one unique peptide was required for protein identification. sgRNA sequence was ligate into plasmid pSpCas9 (BB)−2A-Puro (PX459), which was purchased from Addgene (#48139). Cells were transfected with the plasmid expressing sgPRMT3 by using Lipofectamine 3000 (Invitrogen) according to the manufacturer’s protocol. After 48 h of transfection, cells that did not contain the plasmid were killed with puromycin for 2 days, then changed into DMEM with no antibiotics and plated into 96-well plate for single clone selection. After verification with western blot, the genome was extracted and sent for sequencing. HEK293T cells were transfected with 15 μg of total plasmid DNA encoding Flag-tagged or Myc-tagged bait protein by using PEI. After 48 h, cells were collected, lysed by resuspended with 1 ml whole cell lysis buffer [20 mM HEPES (pH 7.9), 150 mM NaCl, 1.5 mM MgCl2, 5% Glycerol, 0.5 mM EDTA, 0.1% NP40] and incubated for 30 min at 4 °C by gentle rotating. The supernatant was collected by centrifuging the cells at 19,000 × g for 10 min for twice. 20 μl M2 agarose resin (Sigma) or anti-Myc agarose beads (Lablead) were added to each reaction volume and incubated overnight at 4 °C. The resin was collected by centrifuging at 1500 × g at 4 °C for 1 min and washed with high salt wash buffer [20 mM HEPES (pH 7.9), 300 mM KCl, 10% Glycerol, 0.2 mM EDTA, 0.2% NP40] for once, low salt wash buffer [20 mM HEPES (pH 7.9), 100 mM KCl, 10% Glycerol, 0.2 mM EDTA, 0.2% NP40] for three times Then, IP products were boiled with 1× SDS loading buffer at 100 °C for 10 min. Samples were loaded on an SDS-PAGE gel and transferred to membranes for WB analysis. The endogenous IP was performed as previously described. Cells were subjected to chromatin fractionation prior to immunoprecipitation. Cells were lysed in EBC-150 buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 0.5% NP-40, 2 mM MgCl2, supplemented with protease and phosphatase inhibitor cocktails) for 20 min at 4 °C, followed by centrifugation to remove cytoplasmic proteins. Subsequently, the chromatin fraction was solubilized in EBC-150 buffer with 500 U/ml Benzonase and antibody recognizing PRMT3, TEAD4, or CDK9 for 1 h at 4 °C under rotation. Next, the NaCl concentration of the lysis buffer was increased to 300 mM, and lysates were incubated for another 30 min at 4 °C. The lysates were cleared from insoluble chromatin and were subjected to immunoprecipitation with protein A agarose beads (Millipore) for 1.5 h at 4 °C. The beads were then washed 4–6 times with EBC-300 buffer (50 mM Tris, pH 7.5, 300 mM NaCl, 0.5% NP-40, 1 mM EDTA) and boiled in sample buffer. Bound proteins were resolved by SDS-PAGE and immunoblotted with the indicated antibodies. NH1, a HeLa-based cell line, which contains an integrated HIV-1 LTR fused with a luciferase reporter gene in the genome was used for evaluation of HIV-1 transcription. Cells were seeded into 12-well plates at a density of 1 × 10 per well and cultured for 24 h. Then, control plasmid, plasmids expressing target protein, shRNA were transfected in to NH1 cells or HeLa cells together with HIV-1 LTR. For drug treatment, the WT NH1 cells or PRMT3 KO cells were treated with JQ1 (5 μM) or PMA (200 nM) or with these reagents in combination with SGC707 (0.25 mM) for 22 h. Each sample was performed in biological triplicates. 48 h later, cells were collected and lysed with 5× lysis buffer in the kit (Promega). Cells were put into − 80 °C for at least 2 h and vortex for 15 s. The supernatant was tested for luciferase activity. Cell growth was measured using CellTiter-Glo® Luminescent Cell Viability Assay following the instruction as described in the kit (Promega). Prepare WT NH1 and PRMT3 KO cells with medium on a 96-well plate with opaque walls for 100 μl/well. Equilibrate the plate and its contents to room temperature, which will take about 30 min. Each well was supplemented with 100 μl CellTiter-Glo reagents. Mix the contents on an orbital shaker for 2 min to induce cell lysis. Incubate the plate at room temperature for 10 min to stabilize the luminescent signal values, and detect the luminescence signal. Peripheral blood mononuclear cells (PBMCs) were isolated from whole blood using a density gradient centrifugation over the Ficoll medium (TBDScience). CD4 T cells were isolated from PBMCs using the EasySep negative-selection Human CD4 T Cell Enrichment Kit (STEMCELL Technologies) according to the manufacturer’s instructions after stabled for 3 hours in RPMI 1640 medium (Gibco) containing 10% FBS, 1% penicillin-streptomycin, 2 mM L-glutamine, IL-2 (10 ng/ml), and phytohemag-glutinin (1 μg/ml). CD4 T cells were treated with JQ1 (5 μM) or PMA (200 nM) or with these reagents in combination with SGC707 (0.25 mM) for 22 h. Detailed information on the patients is shown in Supplementary Data 6. Nucleofection in MT4 cells and human primary CD4 T cells were performed as previously described. MT4 and primary CD4 T cells were counted at 2 × 10/well and suspended with 100 µl transfection buffer (82 μl of Nucleofector® Solution + 18 μl of supplement, Lonza Nucleofector SF Kit for MT4 cell, Lonza P3 Primary Cell Kit for primary CD4 T cells) containing 300 nM of siRNA. The mixture of cells and siRNA were resuspended and transferred to the Nucleocuvette™ Vessels. Cells were subjected to nucleofection using 4D-Nucleofector® X Unit-Lonza Electrotransfer Instrument with program CA137 for MT4 cells and program EH100 for primary CD4 T cell. After nucleofection, 200 μl Opti-MEM was added and incubated for 10 minutes at 37 °C. The samples were gently resuspended and transferred to 12-well plates, and incubated with culture medium for 48 h. The virus suspension was prepared with fresh 1640 cell culture medium. Cells were infected with HIV-1 at a multiplicity of infection (MOI) of 0.005, gently mixed. The virus-cell suspension was transferred to a 24-well plate culture plate, and cultured at 37 °C, 5% CO2. WT and PRMT3 KO TZM-bl cells were plated at a density of 5 × 10 cells/ ml the day before viral infection. Cells were infected with HIV-1 CRF01-AE or NL-43 at an MOI of 0.005. For the SGC707 treatment experiment, cells were infected with HIV-1 at an MOI of 0.05. The virus-cell suspension was incubated for 48 h or 72 h before collection. The viral supernatant or cells were collected for measurement of HIV-1 RNA copies, pol, gag expression, or p24 level. Total cellular DNA was extracted from patient CD4 T cells using the Qiagen QIA Symphony DNA Mini kit (Qiagen). Magen HiPure Total RNA Plus Mini Kit, respectively, according to the manufacturer’s protocols. The SUPBIO total HIV-1 DNA Quantitative PCR Kit was used for the simultaneous quantitation of total HIV-1 DNA and cell numbers. The SUPBIO HIV-1 usRNA Quantitative PCR Kit was used for the quantification of cell-associated HIV-1 RNA. The measurement of relative HIV-1 RNA copies in cells was also performed by quantitation of the pol expression and calculation of the viral copies using the standard curve, which was normalized to the expression of GAPDH. The p24 expression in the viral supernatant was detected using the lentivirus Titer p24 Assay Kit (EasyQuarter). In general, the HIV p24 standard curve was prepared. 25 µl of lysis buffer was added to each well, followed by the addition of 75 µl sample or standard to the appropriate wells. 75 µl enzyme conjugate was added to the reaction wells and vibrated for 60 seconds, which was then placed at 37 °C for 50 min. Wells were washed with 300 µl wash buffer for 5 times. Then, 100 µl of TMB development solution was added to each well and incubated for 10 min at 37 °C. 50 µl of stop solution was added to each well and vibrated for 60 s. The OD 450 nm was measured. The ChIP experiment was performed as previously described with slightly modified. DNA was sheared as described above. Antibodies of control IgG (Cell Signaling Technology) or PRMT3 antibody (Novus) were added to the clean sheared DNA at 4 °C overnight. Dyna beads (Invitrogen) were added the after day and rotated at 4 °C for 3 h. The immune complex was washed by using high salt wash buffer for once, low salt wash buffer for three times, and TE buffer for two times, with each wash for 5 min rotated at 4 °C, and eluted with 200 μl elution buffer for 1 h. The eluted product was isolated from the resin and added 8 μl 5 M NaCl for reverse crosslinking at 65 °C for 5 h. Add 1 μl of RNase A and incubate for 30 min at 37 °C. Then, add 4 μl 0.5 M EDTA, 8 μl 1 M Tris-HCl pH 6.5, and 1 μl Proteinase K and incubate at 50 °C for 4 h. The DNA product was purified by using the Cycle Pure Kit (Omega). The amount of ChIP DNA product at different LTR regions was detected by using q-PCR. CUT&Tag was performed following the protocol provided in the Hyperactive Universal CUT&Tag Assay Kit for Illumina Pro (TD904-01, Vazyme). NH1 cells were harvested and counted, and aliquots of 50,000 cells were centrifuged at 600 × g for 5 min at room temperature. Cells were washed twice in 1.5 ml Wash Buffer with gentle pipetting. A total of 100 µl of cell (nuclei) suspension was transferred to an 8-strip tube containing activated Con A-coated magnetic beads, mixed by inversion, and incubated at room temperature for 10 min, with occasional mixing (2–3 times by inversion). After a brief centrifugation (<100 × g), the reaction mix was collected, and the tube was placed on a magnetic separation rack. Once the solution cleared (~2 min), the supernatant was discarded. Each sample was resuspended in 50 µl of precooled Antibody Buffer, and 1 µg of IgG or PRMT3 antibody was added. The mixture was briefly centrifuged to collect the contents at the bottom of the tube and incubated overnight at 4 °C. The primary antibody was removed using the magnet stand, and the cells were incubated at room temperature with a 1:100 dilution of the secondary antibody in 50 µl of Dig-Wash buffer for 30 minutes with rotation. Unbound antibodies were removed by washing the cells three times for 2 minutes in 0.2 ml Dig-Wash buffer using the magnet stand. For each sample, 2 µl of pA/G-Tnp Pro was mixed with 98 µl of Dig-300 Buffer, yielding a final concentration of 0.04 µM. After removing the liquid on the magnet stand, 100 µl of the reaction mix was added to the cells and incubated at room temperature for 1 hour with gentle rotation. Unbound pA-Tn5 protein was removed by washing the cells three times for 5 min in 0.2 ml Dig-300 Buffer. Fragmentation was performed by adding 50 µl of TruePrep Tagment Buffer L to the sample and incubating at 37 °C for 1h. The sample was briefly centrifuged, and 2 µl of 10% SDS along with an appropriate amount of DNA Spike-in (adjusted based on target protein abundance and cell input) was added, followed by incubation at 55 °C for 10 min, with mixing 2–3 times by inversion. DNA was purified using DNA extraction beads, and the purified DNA was sent for sequencing. Raw data of fastq format were processed through in-house perl scripts. Clean data were obtained by removing reads containing adapter, reads containing ploy-N and low quality reads from the raw data. Clean reads were mapped to the reference genome using Bowtie2 software. Macs2 was used to call peaks with qvalue <0.05. Peaks were annotated by ChIPseeker package. Differential peak analysis was performed using the DiffBind package (version 3.6.2) in R (version 4.1.2). The DiffBind workflow was initiated by importing the peak files generated by MACS2. The input data were normalized using the “read count” method, and the read counts were then modeled using an edgeR-based negative binomial generalized linear model (GLM) to account for library size differences and overdispersion. Differential peaks were identified using the GLM likelihood ratio test with a false discovery rate (FDR) correction for multiple testing. Peaks with an adjusted p-value <0.05 were considered differentially enriched. All CUT&Tag-seq data analysis was conducted by Beijing SeqWisdom Biotechnology Co., Ltd. The Jurkat-based latency model 2D10 (previously generated by Dr. Jonathan Karn’s laboratory) was used to evaluate the effect on HIV-1 transcription in T cells by PRMT3 inhibitor treatment. 1 × 10 cells were cultured in a 24-well plate and treated with the drug 18 h before cells were collected for FACS analysis. Cells were treated with DMSO or PRMT3 inhibitor SGC707 together with JQ1 or PMA for HIV-1 transcriptional activation. After 18 h of treatment, cells were collected and then resuspended in cold PBS twice. Quantification of the GFP cells was performed using a BD FACSCalibur cytometer. Gating strategies were described in the Supplementary Fig. 8. Data were analyzed with the Flowjo software and plotted as bar graphs with error bars representing standard deviations. The GST-pull down assays were performed as previously described. The bacterial expression and purification of GST-tagged PRMT3 and His-tagged TEAD4 were performed as described previously. GSTSep Glutathione MagBeads (Yeasen) were washed twice with the ice-cold interaction buffer (20 mM HEPES, pH: 7.5, 150 mM KCl, 0.2 mM DTT, 1 mM EDTA, and 10% glycerol) and were blocked at 4 °C for 1 h with a blocking buffer (interaction buffer containing 5% BSA). After the blocking step, 40 μl beads were incubated with the equal volume of GST or GST-tagged PRMT3, respectively, at 4 °C overnight with rotation in the roller. Magnetic beads were collected with magnetic holders and washed with high salt wash buffer [20 mM HEPES (pH 7.9), 300 mM KCl, 10% Glycerol, 0.2 mM EDTA, 0.2% NP40] for once, low salt wash buffer [20 mM HEPES (pH 7.9), 100 mM KCl, 10% Glycerol, 0.2 mM EDTA, 0.2% NP40] for three times. After the washing step, the beads were incubated with His-tagged TEAD4 protein at 4 °C for 3 h with rotation in the roller. Beads were washed with high salt wash buffer for once, low salt wash buffer for three times and boiled with 1× SDS loading buffer at 100 °C for 10 min for western blot detection. Library preparation and sequencing were performed as previously reported. For each library preparation, 5 × 10 WT NH1 cells, PRMT3 KO cells, or cells transfected with a plasmid expressing Tat were used. Cells were washed twice with 500 µl of ice-cold PBS and then lysed in 1× lysis buffer to isolate the nuclei. The TruePrep DNA Library Prep Kit V2 for Illumina and TruePrep Index Kit V2 for Illumina (Vazyme Biotech) were used for library construction. Libraries were purified and selected using AMPure XP beads. The mass concentration of the libraries was determined using a Qubit 3.0 Fluorometer, while the molar concentration was measured with the StepOnePlus™ Real-time PCR system. The lengths of inserted fragments were assessed using the Agilent HS2100 analyzer. The qualified libraries were then sequenced on the Illumina HiSeq X Ten platform in paired-end 150 bp format by Annoroad Gene Technology Co., Ltd. Raw data were stored in FASTQ format, which included the base sequence along with corresponding quality information. Quality control was performed on the raw FASTQ data using Trimmomatic (v0.36). In this step, clean data (clean reads) were generated by removing reads containing adapters, poly-N sequences, and low-quality reads from the raw data. In addition, the Q20, Q30, and GC content of the clean data were calculated. All downstream analyses were based on the high-quality clean data. The clean reads were mapped to the reference human genome using Bowtie2, and the mapping results were visualized using the Integrative Genomics Viewer. Reads with mapping quality below 30, or PCR duplicates, were removed. The remaining high-quality mapped reads were used for further peak calling. Tn5 transposase generates a 9 bp sticky end when cleaving DNA. To improve the accuracy of transposase binding site representation, the reads were adjusted: + 4 bp was added to the positive strand and -5 bp was added to the negative strand. Peak calling was performed using the nomodel option with the --shift -75 parameter (shifting the 5’ end of each read 75 bp towards the left) and the --extsize 150 parameter (extending the moved reads 150 bp to the right to form the final “virtual fragment”). Motif analysis was performed using HOMER’s findMotifsGenome.pl tool. The input files included the peak file and the genome FASTA file. The DNA sequences corresponding to the peaks were extracted and compared with the motif database to identify enriched motifs. Peaks were annotated using the ChIPseeker package. Differential peak analysis was conducted using MAnorm, which identifies differential peaks based on peak density. The MAnorm workflow was initiated by importing the peak files generated by MACS2. Peaks with an adjusted p-value <0.05 were considered differentially enriched. All data analysis was conducted by Beijing SeqWisdom Biotechnology Co., Ltd. Total RNA was extracted from 1 × 10 cells (in triplicate) using TRIzol reagents, and genomic DNA was removed with a DNA-free Kit (Thermo Fisher Scientific). RNA library preparation was performed using the VAHTS Universal V6 RNA-seq Library Prep Kit for Illumina, following the manufacturer’s instructions. Quality control and sequencing were conducted by Annoroad Gene Technology Co., Ltd. RNA-seq data processing was performed as previously described. Low-quality reads and adapter sequences were removed using fqtools. To ensure high-quality data for analysis, a custom Perl script was used to filter the raw data. Reads containing more than 5 base pairs (bp) of adapter sequences were considered contaminated, and if either paired-end (PE) read was contaminated, both reads were discarded. Reads were classified as low quality if more than 15% of their bases had a Phred quality score of ≤19, and if either PE read met this criterion, both reads were removed. Reads with more than 5% ambiguous (N) bases were also discarded, and if one PE read contained excessive N bases, both reads were removed. The reference genome and annotation files were obtained from the ENSEMBL database (http://www.ensembl.org/index.html). Genome indexing was performed using Bowtie2 v2.2.3, and clean reads were aligned to the reference genome using HISAT2 v2.1.0. HISAT2, a successor to TopHat2, employs a modified Burrows-Wheeler Transform (BWT) algorithm to improve alignment speed and computational efficiency. Mapping results were visualized using the Integrative Genomics Viewer (IGV) through heatmaps, histograms, scatter plots, or other graphical representations. Differentially expressed genes (DEGs) were identified using the R/Bioconductor package DESeq2, with a Benjamini-Hochberg adjusted p-value <0.05 and a log2 fold-change >1 as the cutoff. Enrichment analysis was performed using ClusterProfiler, with significance set at adjusted p-value <0.05. All RNA-seq data analysis was conducted by Beijing SeqWisdom Biotechnology Co., Ltd. Cells were seeded on glass coverslips in 24-well plate cotransfected HA-PRMT3 with EGFP-CycT1, Myc-TEAD4, or EGFP-CycT1 and Myc-TEAD4, cells were washed 3 times with PBS for 3 min each time, fixed with 4% paraformaldehyde in PBS for 15 min, permeabilized with 0.2% Triton X-100 in PBS for 30 min at room temperature and blocked with 3% BSA in PBST (0.2% Triton X-100 in PBS) for 30 min at room temperature, then, incubated with primary antibodies overnight at 4 °C. After washes, cells were incubated with Alexa-Fluor-488- or Alexa-Fluor-555-conjugated secondary antibodies (Bioss) for 1 hour at room temperature, and incubated with Hoechst 33342 (Bioss) for 5 min. The coverslips were mounted on glass slides in Mounting Medium and sealed. Immunofluorescence was detected using a Leica confocal microscope. Colocalization of different channels was performed using LAX (Leica). Statistical significance was determined using Origin software 2024b. Sample size, p-values and error bars are described in the text or figure legends. Error bars = mean +/− SD of three biological replicates. A two-tailed Student’s t test was conducted using SPSS. p < 0.05 (*), p < 0.01 (**) and p < 0.001 (***) are considered statistically significant. No statistical method was used to predetermine sample size, and no data were excluded from the analyses. All experiments were repeated at least three times, with similar results obtained. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
PMC1802744
"NeuroStem Chip": a novel highly specialized tool to study neural differentiation pathways in human stem cells
Human stem cells are viewed as a possible source of neurons for a cell-based therapy of neurodegenerative disorders, such as Parkinson's disease. Several protocols that generate different types of neurons from human stem cells (hSCs) have been developed. Nevertheless, the cellular mechanisms that underlie the development of neurons in vitro as they are subjected to the specific differentiation protocols are often poorly understood. We have designed a focused DNA (oligonucleotide-based) large-scale microarray platform (named "NeuroStem Chip") and used it to study gene expression patterns in hSCs as they differentiate into neurons. We have selected genes that are relevant to cells (i) being stem cells, (ii) becoming neurons, and (iii) being neurons. The NeuroStem Chip has over 1,300 pre-selected gene targets and multiple controls spotted in quadruplicates (~46,000 spots total). In this study, we present the NeuroStem Chip in detail and describe the special advantages it offers to the fields of experimental neurology and stem cell biology. To illustrate the utility of NeuroStem Chip platform, we have characterized an undifferentiated population of pluripotent human embryonic stem cells (hESCs, cell line SA02). In addition, we have performed a comparative gene expression analysis of those cells versus a heterogeneous population of hESC-derived cells committed towards neuronal/dopaminergic differentiation pathway by co-culturing with PA6 stromal cells for 16 days and containing a few tyrosine hydroxylase-positive dopaminergic neurons. We characterized the gene expression profiles of undifferentiated and dopaminergic lineage-committed hESC-derived cells using a highly focused custom microarray platform (NeuroStem Chip) that can become an important research tool in human stem cell biology. We propose that the areas of application for NeuroStem microarray platform could be the following: (i) characterization of the expression of established, pre-selected gene targets in hSC lines, including newly derived ones, (ii) longitudinal quality control for maintained hSC populations, (iii) following gene expression changes during differentiation under defined cell culture conditions, and (iv) confirming the success of differentiation into specific neuronal subtypes.Modern DNA microarrays permit a comprehensive analysis of quantitative and qualitative changes in RNA transcript abundance, outlining the cross-sections of gene expression and alterations of these in response to genetic or environmental stimuli. Genome-scale microarrays (cDNA- or oligonucleotide-based) are most valuable when screening populations of cells for the novel genes reflecting potential diagnostic and prognostic markers or for an identification of novel therapeutic targets. On the other hand, custom microarray platforms that focus on specific pre-selected subset of genes relevant to a particular field of investigation can be less costly and more suitable for detection of smaller gene expression changes. Microarray technology has added important information on both normal development and pathological changes in neurons. This is well illustrated by multiple studies on substantia nigra dopaminergic neurons, which degenerate in Parkinson's disease (PD) [1-5]. The shortcomings of pharmacological therapies in PD have stimulated a search for alternative treatment strategies. In successful cases, transplants of human embryonic mesencephalic dopaminergic neurons can both restore dopaminergic neurotransmission and provide some symptomatic relief [6-8]. A wider application of neural transplantation in PD is, however, currently not feasible due to the unpredictable and variable outcome, the risks of unwanted side-effects (dyskinesias) and ethical and practical problems associated with using donor cells obtained from aborted embryos and fetuses . Human embryonic stem cells (hESCs) are considered a promising future source of cells for cell replacement therapy in PD and other neurological conditions . They could constitute a virtually infinite source of self-renewing cells that can be persuaded to differentiate into specific types of neural cells, including dopaminergic neurons [14-16]. The molecular mechanisms that govern development of cultured hESCs into specific types of neural cells are not fully understood. To promote our understanding of such mechanisms, it would be valuable to have tools that readily and reproducibly can help to characterize the cells as they differentiate from pluripotent stem cells into post-mitotic neurons. This important issue was addressed in earlier studies by Luo et al. and Yang et al., who designed small-to-moderate scale custom microarray platforms (281 and 755 gene targets, respectively) . In addition SuperArray Bioscience Corporation (Frederick, MD, USA) have manufactured a range of small-scale arrays (263 gene targets for human array; ). We sought to create an improved and updated microarray platform for hESC/neuronal differentiation-oriented gene expression studies. Therefore, we generated a specialized large-scale DNA microarray platform (the "NeuroStem Chip") that has over 1,300 pre-selected gene targets and multiple controls spotted in quadruplicates (~46,000 spots total). Here we introduce the platform and the advantages it can offers to neuroscientists and stem cell biologists: particularly, in the niche of gene expression-oriented characterization of the samples using an assay of pre-selected, already established gene targets. In the current study, we use the NeuroStem Chip to characterize an undifferentiated population of pluripotent hESCs (cell line SA02, Cellartis AB, Göteborg, Sweden) and compare the gene expression in those cells with that of a hESC-derived cell population rich in neurons, including tyrosine hydroxylase-positive dopaminergic neurons. Stem cells have unique biological characteristics, but only a limited number of genes are currently recognized as established stem cell markers. Examples include POU domain, class 5, transcription factor 1 (Oct3/4), signal transducer and activator of transcription 3 (Stat3), teratocarcinoma-derived growth factor (Tdgf1), Enk-pending (Nanog), undifferentiated embryonic cell transcription factor 1 (Utf1) and DNA methyltransferase 3B (Dnmt3b) . At the same time, hundreds of genes are suggested as candidate markers for "stemness", but their coupling to the undifferentiated stem cell state is not yet fully verified . The concept of "stemness" (term introduced in 1986 by Grossman & Levine) is defined as "core stem cell properties that underlie self-renewal and the ability to generate differentiated progeny" . Considering the complexity of the processes involved, stemness can hardly be ensured by co-operation of just a few genes. Nevertheless, three stemness genes (namely, Oct3/4, Stat3 and Nanog) are considered "master"-genes that control the self-renewing process . Various types of stem cells, such as hematopoietic, mesenchymal and neural (HSCs, MSCs and NSCs, respectively), embryonic germ and embryonic carcinoma cells (EGCs and ECCs, respectively) are all characterized by variations in gene expression profiles, and only a few gene markers are associated with all these cell types . We have aimed to embrace the most comprehensive set of those genes into a solitary array, the NeuroStem Chip. Thereby, it is possible to employ it to monitor the relative expression levels of numerous known and candidate stemness genes in a single experiment. Similar to the genetic bases underlying stemness, cell differentiation is associated with altered expression levels of certain recognized or candidate genes . We therefore incorporated gene markers of development and differentiation in general, and that of neuronal and dopaminergic differentiation in particular, into the NeuroStem Chip. Examples include markers for the processes of neuronal maturation, axonal branching, neural/neuronal survival, etc. Finally, we ensured that known markers for specific types of neurons, allowing identification of individual cell types, were present on the chip. We paid special attention to genes associated with the differentiation and maturation of dopaminergic neurons. In many published studies, the expression of only a single (tyrosine hydroxylase, TH) or 2–3 markers for dopaminergic neurons (e.g. amino acid decarboxylase (AADC), dopamine transporter (DAT), vesicular monoamine transporter 2 (VMAT2)) have been used to indicate dopaminergic identity of neurons. In contrast, the NeuroStem Chip includes oligonucleotide probes for 88 genes related to dopaminergic neurons, thus being more comprehensive in this sense, compared to other existing microarray platforms, including focused ones . Those entries encompass recognized and candidate markers for dopaminergic neurons (mature and early) and progenitors, as well as markers for the maturation and differentiation of the latter (Table 1). Table 2 represents conditional functional breakdown of genes targeted by the NeuroStem microarray platform. A number of important gene groups that are included in the chip are not mentioned in Table 2. Among these, entries related to Dickkopf gene family, galanin-, melatonin-, vasoactive intestinal peptide (VIP)-, cAMP response element-binding protein (CREB)- and B cell leukemia 2 (Bcl2) oncogene-related are present. Many of them play potentially important, yet undefined, roles in the biology of stem cells. Additionally, we included some genes implicated in disease mechanisms of neurodegenerative disorders (most importantly, Parkinson's disease and Alzheimer's disease) in the chip. Furthermore, we incorporated a number of markers for distinct differentiation pathways (e.g. hematopoietic and pancreatic) and cell types (e.g. cancer subtypes and a range of normal cell types) to serve as essential controls. Taken together, we believe that in its present form NeuroStem Chip represents currently most comprehensive gene expression platform for studies on stem cells, neural/neuronal differentiation, human neurodegeneration and neuronal survival, both in vivo and in vitro. The complete layout of NeuroStem Chip will be disclosed to the academic community, upon request. The microarray format we selected relies on long oligonucleotide molecules (69–71 nucleotides) printed over a solid surface. We spotted the synthesized oligonucleotides (Operon Biotechnologies) with a constant concentration across the slides, and evaluated the quality and consistency of spotting in a series of control experiments. We then illustrated the utility and technical reliability of the NeuroStem Chip by characterizing the gene expression profile of commonly utilized hESC line SA02 (Sahlgrenska 2; ), including (i) undifferentiated cells and (ii) cells committed towards neuronal/dopaminergic differentiation pathway. For the first of these, we used total RNA sample purified from hESC colonies that exhibited morphology consistent with cell proliferation and the absence of spontaneous differentiation (Figure 1A). We also evaluated the expression of the cell cycle marker Ki67 and the pluripotency marker OCT3/4 in the sample by immunocytochemistry (Figure 1B–E). Co-culturing of ESCs with murine stromal cells (including PA6 cell line) rapidly generates dopaminergic neurons from ESCs by an unexplained mechanism termed stromal cell-derived inducing activity (SDIA; ). We therefore committed hESCs toward the neuronal/dopaminergic differentiation pathway by co-culturing with PA6 cells for 16 days, resulting in appearance of cells positive for early and late neuronal markers, including nestin, β-III-tubulin, and TH, the established marker of dopaminergic neurons (Figure 2). To verify the expression of some key stem cell- and neural phenotype-associated genes we performed RT-PCR comparing RNA samples from the undifferentiated hESCs with hESCs of the same line differentiated toward neuronal/dopaminergic pathway, as described above. The expression profile outlined by RT-PCR confirmed the identity of the sample used (Figure 3). After performing RNA integrity tests, we incorporated fluorescent labels to the amplified RNA samples from hESCs (Cyanine 3-CTP (Cy3) and Cyanine 5-CTP (Cy5)), hESC-derived cells containing TH-positive neurons (Cy3 and Cy5) and human universal reference RNA (Cy5), and hybridized aliquots with NeuroStem microarray slides using the following conditions: hESC vs. reference, Cy3 : Cy5 = (i) 20:10 pmol, and (ii) 10:5 pmol; and hESC vs. hESC-derived cells, Cy3 : Cy5 = (iii) 30:20 pmol, respectively. Universal reference RNA has been previously established as a standard reference material for microarray experiments, proving an ability to effectively hybridize to a large fraction of microarray spots . We performed two-color hybridizations (e.g. for the experiment vs. reference) following an established protocol , and included dye-flip technical replicates in the analysis (Figure 4). Using the online software program BASE we sequentially filtered the data by background subtraction, negative flagging, negative intensities and for inconsistent data amongst replicates . Figure 5A shows a comparison of the spot intensities prior to normalization (M versus A plot), with the Log2 of the expression ratio between Cy3/Cy5 being plotted as a function of the log10 of the mean of the total expression intensities for Cy3 and Cy5 channels. The deviation of the line from zero revealed a need for normalization, so prior to data analyses we normalized signals using a locally weighted scatterplot-smoothing regression (LOWESS) algorithm (Figure 5A–B; fitted line) implemented in BASE. Since the reproducibility of two-color microarray gene expression data is critically important, we calculated Pearson correlation coefficients of the reporters present in the filtered database comparing the average expression ratios (7005 for hESCs vs. universal reference; 6947 for undifferentiated vs. neuronal/dopaminergic lineage-committed hESCs). Results obtained revealed that data were consistent across technical replicates (dye-swap and amount of loaded material), showing general high reproducibility: e.g., correlation coefficients were greater than 0.96 for technical replicates and 0.78 for dye-swapping samples in hESCs vs. universal reference hybridizations (Table 3). To detect genes with high expression levels in hESC samples, we filtered data for intensity values >100 in the hESC sample and performed clustering analysis using the TIGR MultiExperiment Viewer (MEV; ). To visualize variations of spot/reporter per technical replicate, hierarchical clustering was performed by K-means classifier based on the linear-correlation-based distance (Pearson, centred) method. The optimal number of clusters was determined empirically to produce the most balanced ratio of entries to cluster of highly expressed genes. A cluster of 101 genes up-regulated in the hESC sample [see Additional file 1], was plotted in a centroid graph (Figure 5C); the variation across technical replicates was low. We merged technical replicates to generate a list of the most up-regulated genes expressed in the hESC sample compared to the universal reference RNA (Table 4). Standard error of the mean expressed as percentage was calculated for the 4 technical replicates, and was 6.7% for the top 25 genes up-regulated in hESC samples, compared to universal reference RNA. We performed the analysis of microarray data, as described in the Methods, and spot error values were generally in the lower range, indicating high stringency of the signals and low variance. As seen in Table 4 and Table 5, the NeuroStem Chip identified numerous genes associated with stem cells. In particular, homeo box expressed in ES cells 1 (Hesx1) gene was identified as the most up-regulated in the ES cell preparation, compared to universal reference RNA. Highly expressed in pluripotent ESCs, Hesx1 expression is down-regulated upon embryonic stem cell differentiation , as also clearly seen in differentiation experiment of our own (Table 4). Similarly, Gremlin 1 homolog, cysteine knot superfamily gene (Grem1, also known as Cktsf1b1 and Dand2) is a recognized factor of cell-fate determination of ESCs . Many more genes highly up-regulated in the hESC sample in comparison with universal reference RNA are associated with stem cells: further examples include Gap junction protein α1 (Gja1) and Zic family member 3 heterotaxy 1 (Zic3) (Table 4) . The expression of fibroblast growth factor receptor 2 (Fgfr2) is of particular interest. Basic fibroblast growth factor (FGF2, bFGF) supports hESC proliferation and their ability to maintain undifferentiated phenotype when cultured in vitro . Moreover, in some hESC lines a very high concentration of FGF2 could substitute for the need of feeder cells . At the same time, genes listed in Table 4 represent the most highly up-regulated entries in a relatively limited group of genes (Figure 5C). Many other genes involved in maintenance of ESC phenotype (i.e. established or candidate markers of stem cells) have lower levels of expression (Table 5). Examples include undifferentiated embryonic cell transcription factor 1 (Utf1), DNA methyltransferase 3B (Dnmt3b), developmental pluripotency associated 4 (Dppa4, a newly established pluripotency marker ) and numerous candidate markers of "stemness": e.g. genes for KIAA1573 protein, forkhead box O1A (Foxo1a), high-mobility group box 1 (Hmgb1), C-terminal binding protein 2 (Ctbp2) and left-right determination factor 1 (Lefty1), as well as others. For numerous established or candidate markers of stem cells the expression levels were not considerably higher (Log2 ratio < 1) in the hESC sample compared to the universal reference RNA. For example, the expression of Nanog, DNA (cytosine-5-)-methyltransferase 3α (Dnmt3a), MutS homolog 2, colon cancer, nonpolyposis type 1 (E. coli) (Msh2), Thy-1 cell surface antigen (Thy1), high-mobility group box 2 (Hmgb2), transcription factor 3 (Tcf3), Nanos homolog 1 (Nanos1), MyoD family inhibitor (Mdfi), Calumenin (Calu) and soluble thymidine kinase 1 (Tk1) was detected in hES SA02 cells with Log2 ratio value < 1. Expression levels of those genes range from being inconsiderably higher to nearly equal to that in universal reference RNA sample. We believe that those findings could be explained by cellular composition of human universal reference RNA sample , which includes pooled RNA samples from proliferating cells (e.g., skin and testis cell lines). Thus, the relative difference between gene expression of certain markers of stem cells in undifferentiated hESCs and universal reference RNA is naturally decreased. Taken together, the gene expression signature of hES SA02 cell line profiled by NeuroStem Chip is indeed characteristic for pluripotent stem cells, providing proof-of-concept. Notably, comparison of expression profiles of undifferentiated hESCs and hESC-derived cells committed toward dopaminergic differentiation pathway by co-culturing with SDIA for 16 days have revealed that many of the stem cell marker genes mentioned above were down-regulated in differentiation (Table 5). Expectedly, Hesx1, Grem1, Dnmt3b, Utf1 and Nanog could be listed among these. At the same time, numerous other genes, including Pitx2, Dlk1 and Msx1 were up-regulated in the latter sample ([see Additional file 2], Figure 3). Table 1 lists 24 dopaminergic system-related entries (e.g., Ptx3, Th, Lhx1) with gene expression up-regulated by Day 16 of hESC differentiation protocol; few more genes have demonstrated less prominent up-regulation (Log2 ratio values in the range of 0.7/0.97–1.0). The gene expression profiles generated are therefore consistent with the results of earlier studies utilizing hSC-derived samples with similar characteristics . Diversity of NeuroStem Chip entries responsive to hESC commitment toward neuronal/dopaminergic differentiation pathway clearly illustrates the complexity of that pathway. The cell population obtained after 16 day exposure to SDIA is highly heterogeneous. Only around 0.2% of the cells are TH-positive cells (Figure 2). This heterogeneity, with an apparent presence of residual pluripotent cells explains the presence of stem cell marker genes, including homeobox transcription factor Nanog, as revealed by RT-PCR data (Figure 3). It would be therefore impossible to apply the platform to identify novel genes associated with the process of differentiation; for that application, the genome-scale microarray platforms (e.g., Affymetrix) are clearly superior. Nevertheless, being based upon a moderate assay of pre-selected specific gene targets, the comparative analysis of microarray data derived from undifferentiated and dopaminergic differentiate pathway-committed hESCs provides a valuable cross-cut of complex relationship between factors driving or indicative to neuronal/dopaminergic differentiation [see Additional file 2]. RT-PCR analyses have validated the overall reliability of NeuroStem microarray platform: all of the entries detected in the hybridization experiments have demonstrated similar trends when analyzed by RT-PCR means (Figure 3). Those entries include Sox2, En1 and Nanog (ratio of differentiated/undifferentiated hESC sample normalized spot intensity < 0.75, down-regulated), Gadph, Aldh1a1, Sdha, Tubb and Nestin (ratio .1.0, unchanged), Actb, Th, Msx1 and Pitx2 (ratio >1.25, up-regulated). Some of the housekeeping genes (Gapdh, Sdha, Tubb, Actb) have somewhat different expression in undifferentiated vs. differentiated cells, consistent with previous reports on certain established housekeeping genes (including Gapdh) being variable in human samples . Importantly, all the observed gene expression trends were similar in both microarray and RT-PCR. Our experiment therefore confirms that the NeuroStem Chip microarray platform can still identify gene expression changes related to early stages of differentiation of hESC into dopaminergic neurons. Recent technological advances have led to DNA microarrays which contain over hundred thousand of spots of DNA material, reaching a truly genomic scale. Highly specialized DNA microarrays of smaller scale (e.g. the NeuroStem Chip) still have an important role in the directed studies in particular fields. Since they are significantly less expensive, compared to many recognized large-scale platforms (e.g. Affymetrix Human Genome platforms), they have a clear advantage in routine work involving samples from, e.g., multiple cell culture conditions. While there is a risk that one will miss out on changes in genes previously not believed to be relevant to neural differentiation, the restricted number of genes in the NeuroStem Chip also simplifies analysis and adds power. NeuroStem Chip is comparable to other stem cell-related focused microarray platforms in regards to manufacturing costs and technical simplicity of the recommended hybridization protocols. At the same time, it currently implies an advantage in both the scale and the spectrum of pre-selected, specific gene targets assayed. Some suggested areas of application for NeuroStem microarray platform could be the following: (i) characterization of the expression of established, pre-selected gene targets in human stem cell (hSC) lines, including newly derived ones, (ii) longitudinal quality control for maintained hSC populations, (iii) following gene expression changes during differentiation under defined cell culture conditions, and (iv) confirming the success of differentiation into specific neuronal subtypes. In addition, the NeuroStem Chip can be used to characterize gene changes in intracerebral grafts of human cells, even when they are transplanted into experimental animals. We specifically wish to stress that we are about to make the NeuroStem Chip available at a non-profit cost to the research community. We believe it has the potential to become an important screening tool in the expanding field of hSC studies in application to neurological/neurodegenerative disorders. Undifferentiated hESCs of SA02 (Sahlgrenska 2) line (Cellartis AB, Göteborg, Sweden; see NIH Human Embryonic Stem Cell Registry at ) were maintained over a monolayer of human "feeder cells" (hFCs; human foreskin fibroblasts, ATCC; cell line CCD-1112Sk). Feeder cells were grown in hFC medium (Iscove's modified Dulbecco's medium (IMDM) supplemented with 10% heat-inactivated FCS (Stem Cell Technologies, USA) and 0.5% Penicillin/Streptomycin mix) for 11 passages. One day prior to hESC plating, hFC medium was washed away from the hFCs, the latter were resuspended in a hESC proliferation medium (VitroHES media (Vitrolife AB, Sweden) supplemented with 4 ng/ml human recombinant basic FGF (hrbFGF, Biosource International, USA) and plated in a central ring of gelatinized in vitro fertilization (IVF) dishes with a cell density of 120,000 cells/dish. The outer rings of the IVF dishes were filled with Dulbecco's modified Eagle medium (DMEM) supplemented with 0.5% Penicillin/Streptomycin mix. One half of the culture medium was replaced every other day. The cells were maintained at 37°C, 5% CO2, 95% humidity settings. Every 6 days, fragments of the hESC colonies (around 10–14 colonies per dish, measuring around 0.015 × 0.015 mm) that had an unaltered morphology (indicating lack of spontaneous differentiation) were mechanically cut from dishes using stem cell knives/transfer pipettes (SweMed Lab International AB, Sweden) and then plated on fresh hFCs. Co-culturing with the PA6 stromal cell line (MC3T3-G2/Pa6, from RIKEN Cell Bank Japan (RCB 1127), derived from newborn mouse calvaria rapidly generates high numbers of DA neurons from mouse and monkey ESCs by an unknown mechanism named stromal-derived inducing activity (SDIA; ). For differentiation experiments, PA6 cells were plated on gelatine-coated T25 flasks at 16 × 10cells/cm(400,000 cells/flask) density 2 days prior to introducing hESCs into the co-culture and cultured at PA6 culturing media (containing minimum essential medium alpha (α-MEM) supplemented with 10% FCS and 0.5% Penicillin/Streptomycin). Alternatively, PA6 cells were plated over Type I collagen-coated glass cover-slips placed in wells of 4-well-plates (50,000 cells/well, for immunocytochemical (ICC) analysis). Three hours prior to initiation of co-culture, PA6 cells were rinsed 3 times with PBS and media was replaced with co-culture media (Glasgow's modified Eagle's media (G-MEM) supplemented with 8% knock-out serum replacement (KSR), 2 mM glutamine, 0.1 mM non-essential amino-acids (NEAA), 1 mM pyruvate, 0.1 mM β-mercaptoethanol (βME) and 4 ng/μl bFGF). Fragments of hESC colonies (80–90 per flask; 4–5 per well of 4-well-plate) presenting undifferentiated morphology were manually passaged onto the confluent PA6 monolayer and cell co-cultures were maintained at 37°C, 5% CO2, 95% humidity settings. One half of the co-culture medium was replaced every other day for first 10 days, and daily onwards. IVF dishes with hESCs grown atop hFCs and 4-well plate dishes with hESCs growing atop PA6 cells were fixed with 4% paraformaldehyde (PFA) for 15 minutes at the day of passage/harvest (Day 6 of hESC/hFC co-culturing) and Day 16 of co-culturing with PA6 cells, respectively. Cells were pre-incubated with blocking solution containing PBS, 0.5% Triton X-100 and 5% of donkey serum. They were then incubated with primary antibodies in blocking solution overnight at room temperature. After three washes with PBS, cells were incubated with the donkey anti-rabbit IgG conjugated with FITC or anti-mouse Cy3 (1:200, Jackson ImmunoResearch Laboratories). Cells were then washed once with PBS, incubated with 1:1000 DAPI in PBS for 10 minutes, followed by another wash with PBS. Coverslips were mounted onto glass slides with PVA mounting medium containing anti-fading reagent DABCO. The following primary antibodies were used: mouse anti-Oct3/4 (1:500, Santa Cruz Biotechnology Inc.); rabbit anti-Ki67 (1:200, Novocastra Ltd.); rabbit anti-TH (1:500, Chemicon). Immunostained cell cultures were visualized using a Zeiss fluorescent microscope attached to a Nikon digital camera. Using RT-PCR, all RNA samples used in this study were tested negative for the presence of gDNA (data not shown). The intron-spanning primers for RT-PCR were selected from published works or designed using Oligo 4.0 software (Molecular Biology Insight) or Clone Manager Suite 7.1 (Sci Ed Software, NC, USA) and ordered from TAG Copenhagen A/S, Denmark, as the following: Sox2, SRY-box 2: 5'-TAC CTC TTC CTC CC CTC CA-3', 5'-ACT CTC CTC TTT TGC ACC CC-3'; En1, Engrailed 1: 5'-AAG GGA CGA AAC TGC GAA CTC C-3', 5'-GAC ACG AAA GGA AAC ACA CAC TCT CG-3' ; Nanog: 5'-TGC TTA TTC AGG ACA GCC T-3', 5'-TCT GGT CTT CTG TTT CTT GAC T-3' ; Gapdh, glyceraldehydes-3-phosphate dehydrogenase: 5'-GGA AGG TGA AGG TCG GAG TCA A-3', 5'-GAT CTC GCT CCT GGA AGA TGG T-3'; Aldh1A1, aldehyde dehydrogenase 1 family, member A1: 5'-GGG CAG CCA TTT CTT CTC AC-3', 5'-CTT CTT AGC CCG CTC AAC AC-3' ; Sdha, succinate dehydrogenase: 5'-TGG GAA CAA GAG GGC ATC TG-3', 5'-CCA CCA CTG CAT CAA ATT CAT G-3' ; Tubb, β-tubulin: 5'-CTC ACA AGT ACG TGC CTC GAG-3', 5'-GCA CGA CGC TGA AGG TGT TCA-3'; Nestin: 5'-AGA GGG GAA TTC CTG CT GAG-3', 5'-CTG AGG ACC AGG ACT CTC TA-3' ; Actb, β-actin: 5'-CAT CGA GCA CGG CAT CGT CA-3', 5'-TAG CAC AGC CTG GAT AGC AAC-3' ; Th, Tyrosine hydroxylase: 5'-CGA GCT GTG AAG GTG TTT G-3', 5'-TTG GTG ACC AGG TGA TGA C-3'; Msx1, homolog of Drosophila muscle segment homolog 1: 5'-CTC AAG CTG CCA GAA GAT GC-3', 5'-TCC AGC TCT GCC TCT TGT AG-3'; Pitx2, paired-like homeodomain transcription factor 2: 5'-ACC TTA CGG AAG CCC GAG TC-3', 5'-TGG ATA GGG AGG CGG ATG TA-3' . cDNA was synthesized from 1 mg of total RNA using SuperScript II (Invitrogen), and RT-PCR amplifications were performed using the MiniOpticon system (Bio-Rad) with REDTaq Polymerase (Sigma-Aldrich) essentially as described by the manufacturer. Following initial denaturation for 5 min at 95°C, DNA amplifications were performed for 35 (En1, Nanog, Aldh1a1), 33 (Sox2, Nestin, Th, Msx1), 32 (Tubb, Pitx2), 27 (Sdha), 25 (Actb) or 22 (Gapdh) cycles of 1 min at 95°C, 1 min at 55°C (En1, Pitx2), 57°C (Sox2, Nanog, Sdha, Nestin), 58°C (Tubb), 58.5°C (Aldh1a1, Th) or 59°C (Gapdh, Actb, Msx1), and 1 min at 72°C. The final extension was 5 min at 72°C. Twenty μl volumes of RT-PCR products were analyzed by electrophoresis at 1% agarose gels and visualized by ethidium bromide staining. RNA purification and fluorescent dye incorporationFor RNA purification of undifferentiated hESCs, the latter were mechanically separated from hFCs, collected in a 500 μl volume of VitroHES media, rinsed in PBS buffer and spun down at 300 rcf for 5 min. hESC-derived cells grown atop PA6 cells were harvested using a papain dissociation kit (Worthington Biochemical Corporation), rinsed in PBS buffer and spun down as described above. The resulting cell pellets were resuspended in RLT buffer (Qiagen, USA), passed through the shredder column (Qiagen) and stored at -80°C until the RNA sample was purified following the RNeasy Micro Kit (Qiagen) protocol (without carrier RNA); with DNase I (Quiagen) treatment incorporated to the latter. RNA integrity was tested using both ND-1000 specrophotometer (NanoDrop, USA) and RNA Nano LabChip/2100 Bioanalyzer system (Agilent Technologies, USA). Fluorescent label (24 nmol of the Cyanine 3-CTP (Cy3); PerkinElmer, USA) was incorporated to 350–500 ng of total RNA amplified using Low RNA Input Fluorescent Linear Amplification Kit (Agilent Technologies), generally following the kit manufacturer's protocol. Similarly, 24 nmols of the Cyanine 5-CTP (Cy5; PerkinElmer) fluorescent label were incorporated to 400 ng sample of Human Universal Reference RNA (Stratagene, USA); in addition, dye-swap replicate amplification were performed. Amplified fluorescent cRNA samples were purified using RNeasy mini-columns (Quiagen), and fluorescence of the eluted products was measured using ND-1000 specrophotometer (NanoDrop). Long oligonucleotide probes (69–71 nucleotides) matching gene targets of interest were selected from Operon V2 and V3 human AROS sets (Operon Biotechnologies Inc., USA). Arrays were produced by the SweGene DNA Microarray Resource Centre, Department of Oncology at Lund University (Sweden) using a MicroGrid II 600R arrayer fitted with MicroSpot 10 K pins (Harvard BioRobotics, USA). Printing was performed in a temperature- (18–20°C) and humidity- (44–49% RH) controlled area on Corning UltraGAPS aminosilane slides (Corning Inc., USA) with 140 μm spot-to-spot centerdistance and 90–110 μm average spot size. Following printing, arrays were dried for 48 hours andstored in a dessicator until used. Microarray slides were UV-cross-linked (800 mJ/cm), pre-hybridizedwith fluorescently labeled samples using the Pronto! Universal Microarray Hybridization Kit (Corning) and subsequently hybridized with test (Cy3-labeled)/reference (Cy5-labeled) RNA samples (or in reverse dye-labeling order) at 42°C for 17 h using a MAUI hybridization station (BioMicro Systems Inc., USA) and the Pronto! Universal Microarray Hybridization Kit, generally following manufacturer's instructions, with several minor adaptations . Immediately following the washing steps, the fluorescence intensities were measured using a confocal laser scanner (G2505B, Agilent Technologies). After image formatting by Tiff Image Channel Splitter Utility (Agilent Technologies) and grid annotation, a complete set of spots was visually inspected for each slide. Using GenePix Pro (Molecular Devices Corp. USA) flags for artifactual spots were annotated for each spot. Median pixel intensity minus the median local background for both dyes was used to obtain a test over reference intensity ratio. Data normalization was performed per array subgrid using LOWESS curve fitting with a smoothing factor of 0.33 . All normalizations, filtering, merging of technical replicates and analyses were performed in the BioArray Software Environment database . To visualize sample-dependent variation of spot intensities, data was uploaded to the TIGR MultiExperiment Viewer (MEV; ). Overall design of the project was performed by joint effort of all coauthors. S.V.A. developed the NeuroStem Chip design, participated in hESC growth and differentiation, and performed RNA sample purifications, fluorescent sample preparations, microarray hybridizations, microarray data formatting and RT-PCR experiments. N.S.C. performed all computer analysis of microarray data. A.S.C. participated in hESC growth and differentiation and performed extensive characterization of hESCs on all stages of differentiation protocol. All authors have contributed to the writing and approved the final manuscript.
PMC11116453
Single-cell and spatial transcriptomics analysis of non-small cell lung cancer
Lung cancer is the second most frequently diagnosed cancer and the leading cause of cancer-related mortality worldwide. Tumour ecosystems feature diverse immune cell types. Myeloid cells, in particular, are prevalent and have a well-established role in promoting the disease. In our study, we profile approximately 900,000 cells from 25 treatment-naive patients with adenocarcinoma and squamous-cell carcinoma by single-cell and spatial transcriptomics. We note an inverse relationship between anti-inflammatory macrophages and NK cells/T cells, and with reduced NK cell cytotoxicity within the tumour. While we observe a similar cell type composition in both adenocarcinoma and squamous-cell carcinoma, we detect significant differences in the co-expression of various immune checkpoint inhibitors. Moreover, we reveal evidence of a transcriptional “reprogramming” of macrophages in tumours, shifting them towards cholesterol export and adopting a foetal-like transcriptional signature which promotes iron efflux. Our multi-omic resource offers a high-resolution molecular map of tumour-associated macrophages, enhancing our understanding of their role within the tumour microenvironment. Subject terms: Non-small-cell lung cancer, Non-small-cell lung cancer, Tumour immunologyLung cancer is the second most commonly diagnosed cancer and the first cause of cancer death worldwide, with a 5-year survival of ~6% in patients with the most advanced stages. Non-small-cell lung cancer (NSCLC) is the most common type of lung cancer (~85% of total cases), followed by small-cell lung cancer (15% of total cases). Lung cancer is a complex disease in which the tumour microenvironment plays a critical role and macrophages (Mɸ) are intimately involved in the progression of the disease. In particular, tumour-associated Mɸ (TAMs) can exhibit a dual role, contributing to tumour promotion by suppressing the immune response, facilitating angiogenesis, and aiding in tissue remodelling, but also tumour suppression by promoting inflammation and engaging in cytotoxic activity against cancer cells. The intricate interplay between lung cancer and Mɸ highlights the importance of understanding their dynamic relationship in order to develop more effective therapeutic strategies. Within NSCLC, adenocarcinoma (LUAD) is the most common histological subtype, followed by squamous-cell carcinoma (LUSC). Lobectomy (i.e., the anatomical resection of a lung lobe) is currently the gold standard for the treatment of early stages of NSCLC (stage I/II), while patients with unresectable stage III or metastatic stage IV NSCLC are treated with a combination of chemotherapy and neoadjuvant targeting vascular endothelial growth factor (VEGF) or immune checkpoint inhibitors (ICIs) like PD1, PDL1 and CTLA4. Advancements made in the last decade in uncovering predictive biomarkers have paved the way for novel therapeutic prospects in the fields of targeted therapy and immunotherapy on the basis of tumour histology and PDL1 expression. A number of studies have employed single-cell technologies to explore transcriptional changes in NSCLC. They have extensively examined the lung tumour microenvironment revealing diverse T-cell functions linked to patient prognosis, relevance of diversity of B cells in NSCLC for anti-tumour therapy, multiple states of tumour-infiltrating myeloid cells, proposing them as a new target in immunotherapy, as well as the association of tissue-resident neutrophils with anti-PDL1 therapy failure. They further unveiled tumour heterogeneity and cellular changes in advanced and metastatic tumours as well as tumour therapy-induced transition of cancer cells to a primitive cell state. In many of these studies, a limited number of cells was analysed per patient, and often there was no systematic collection of patient-matched non-tumour tissue, thus restricting dissection of the biological heterogeneity within tumour and adjacent non-tumour tissue. Additionally, with some exceptions, LUAD and LUSC were considered as a single entity thus hindering the investigation of specific hallmarks of the two cancer types which are radically distinct both at the molecular and pathological level. While single-cell RNA-seq (scRNA-seq) can identify cell types and their states at high resolution within tissues, it lacks the capability to pinpoint their spatial distribution or capture the local cell–cell interactions as well as ligands and receptors that mediate these interactions. Therefore, impeding our ability to fully explore the tumour microenvironment (TME) and the complexity of cell–cell interactions therein. To overcome above mentioned limitations, we combined scRNA-seq data from nearly 900,000 cells from 25 treatment-naive patients with LUAD or LUSC and spatial transcriptomics from eight patients to investigate the differences in cellular organisation in tumour and adjacent non-tumour tissue. We further examined Mɸ populations and molecular changes they undergo in the tumour environment, some of which resemble those observed in Mɸ during human foetal development. To determine the heterogeneity of immune and non-immune cellular states and their spatial landscape in LUAD and LUSC, we collected lung tissue resections from 25 treatment-naive patients with either LUAD (n = 13), LUSC (n = 8) or undetermined lung cancer (LC, n = 4), and two healthy deceased donors (Fig. 1A, B and Supplementary Data 1). We collected both tumour and matched normal non-tumorigenic tissue (i.e., background), isolated CD45+ immune cells (Supplementary Fig. 1A) as well as tumour and other non-immune populations (using CD235a column to deplete erythroid cells), and performed scRNA-seq. In addition, tumour and background tissue sections from eight patients (of the aforementioned 25) were processed for spatial transcriptomics using the 10x Genomics Visium platform (n = 36 sections in total) (Fig. 1A and Supplementary Data 1). A Study overview. Single-cell suspensions of resected tumour tissue, adjacent non-involved tissue (background) and healthy lung from deceased donors were enriched for CD45+ or CD235− and subjected to scRNA-seq. Cryosections of fresh, flash-frozen tumour, background and healthy tissues were used for 10x Visium spatial transcriptomics. B Cohort overview. Symbols represent individual patients and performed analyses. C UMAP projection of tumour and combined background+healthy datasets. D Dotplot of representative genes used for broad cell-type annotations in tumour samples. E Contour plot showing the co-expression of myeloid (LYZ, CD68, MRC1) and epithelial (EPCAM) genes in AT2 cells (44,399 cells), CAMLs (2520 cells) and AIMɸ (16,120 cells). Normalised, scaled and log-transformed gene expression. F Boxplot showing normalised, scaled and log-transformed gene expression of myeloid (LYZ, APOE, CD68, MRC1) and epithelial (EPCAM, KRT8, KRT19) genes in AT2 cells, CAMLs and AIMɸ. Boxes: quartiles. Whiskers: 1.5× interquartile range. G Relative proportion of non-immune cell subsets in tumour and background, calculated within the CD235− enrichment. Arrows indicate increase (↑) or decrease (↓) in tumour versus background. Pairwise comparisons by two-sided Wilcoxon rank test and Bonferroni correction for multiple comparisons. **P < 0.01. Arrows without asterisks indicate that the cell type was found only in tumour or background. H Relative proportion of broad immune cells in tumour and background, calculated within all immune cells identified in the CD235- enrichment. Arrows indicate an increase (↑) or decrease (↓) in tumour versus background. Pairwise comparisons by two-sided Wilcoxon rank test and Bonferroni correction for multiple comparisons. *P < 0.05, **P < 0.01, ***P < 0.001. Arrows without asterisks indicate that the cell type was found only in tumour or background. I Relative proportion of NK, DC, B, T and macrophage subsets within the broad annotations in tumour and background, calculated within the CD235- enrichment. Arrows indicate increase (↑) or decrease (↓) in tumour versus background. Pairwise comparisons by two-sided Wilcoxon rank test and Bonferroni correction for multiple comparisons. ***P < 0.001. Arrows without asterisks indicate that the cell type was found only in tumour or background. Following quality control (QC) on the scRNA-seq dataset, we identified 895,806 high-quality cells in total, of which 503,549 were from tumour and 392,257 from combined background and healthy tissue (from here on referred to as B/H). After performing normalisation and log1p transformation, highly-variable gene selection, dimensionality reduction, batch correction, and Leiden clustering, cells originating from tumour and B/H were separately annotated into distinct broad cell types and visualised via Uniform Manifold Approximation and Projection (UMAP) (Fig. 1C, Supplementary Fig. 1B, C, and “Methods”). We identified clusters of myeloid cells with transcriptional signatures of monocytes, macrophages, dendritic cells (DCs), as well as mast cells, natural killer (NK) cells, T cells, B cells and non-immune cells (Fig. 1C, D). We did not detect neutrophilic granulocytes, most probably due to their sensitivity to degradation after collection and in particular to the freezing-thawing cycle. Finally, we identified a cluster characterised by the co-expression of myeloid (LYZ, CD68, CD14, MRC1) and epithelial genes (KRT19, EPCAM) (Fig. 1D–F). These cells were found within the tumour and exhibited similarities to previously described cancer-associated macrophage-like cells (CAMLs). CAMLs represent a distinct population of large myeloid cells with concomitant epithelial tumour protein expression. These unique cells have been observed in blood samples of patients with various malignancies, including NSCLC. The abundance of CAMLs exhibits a direct correlation with response to therapeutic interventions, highlighting their functional significance. Even after further subclustering, CAMLs maintained their distinct dual myeloid-epithelial signature (Supplementary Fig. 1D). It is noteworthy that doublet detection software Scrublet assigned a low doublet score to CAMLs, suggesting their expression profile is unlikely to be explained as a combined signature arising from the coincidental sequencing of a tumour cell and a macrophage (Supplementary Fig. 1E). All clusters included cells from multiple patients, with the cluster size ranging from 2520 to 124,459 cells (Supplementary Fig. 1F, G). Furthermore, we conducted reference-query mapping using scArches to confirm the consistency of our annotations in the tumour and B/H dataset (Supplementary Fig. 2A–C and Supplementary Notes). The composition of the immune and non-immune compartment was markedly different between the tumour and background. In the tumour, we detected fibroblasts and a decrease in the fraction of lymphatic endothelial cells (LECs) (Padj = 0.0025, Fig. 1G and Supplementary Data 2). Furthermore, the population of epithelial cells showed higher diversity, with the presence of alveolar type II (AT2), atypical epithelial cells which downregulated epithelial markers (KRT19, EPCAM, CDH1), transitioning epithelial cells which upregulated myeloid markers (LYZ), and cycling epithelial cells in tumour tissues (Fig. 1G, Supplementary Notes, and Supplementary Fig. 2D, E). These differences are in agreement with the fact that in tumour specimens, epithelial cells are likely to be a mixture of mutant tumour and non-mutant normal cells, and suggest that neoplastic transformation leads to further diversity of cell states. We did not detect alveolar type I (AT1) or basal cells, possibly due to their loss during dissociations, as previously reported by others. As previously reported, the proportion of monocytes and immature myeloid cells was significantly reduced in tumour samples compared to background (Padj = 0.022 and Padj = 0.00001, respectively), while DCs and B cells were overall expanded (Padj = 0.0023 and Padj = 0.0044, respectively; Fig. 1H and Supplementary Data 3). To get further insight into the cellular composition of tumour versus background tissue, we subclustered each of the broad clusters and identified 46 cell types/states (Supplementary Fig. 2D, E, Supplementary Data 4 and 5, Supplementary Fig. 3, and Supplementary Notes). In the tumour, we found that a significantly higher proportion of NK cells had a lower cytotoxicity phenotype (Supplementary Notes), and that the significant majority of DCs were derived from monocytes (i.e., mo-DC2), (Supplementary Notes) compared to background (Padj = 0.00002 and Padj = 0.00002, respectively, Fig. 1I and Supplementary Data 6). This is consistent with the monocytic origin of mo-DC2s under inflammatory conditions. Similarly, we found an expansion of B cells expressing LYZ and TNF, and depletion of NKB cells (Fig. 1I and Supplementary Notes). Among T cells, tumour samples showed an accumulation of regulatory T cells (Tregs), known to hinder the immune surveillance of tumours (Fig. 1I). Conversely, there was a reduction of exhausted cytotoxic T cells (Padj = 0.00002) in the tumour and absence of T cells, which have been associated with survival in NSCLC (Fig. 1I and Supplementary Data 6). T cells are capable of recognising and lysing diverse ranges of cancer cells, and thus have been suggested for a role in pan-cancer immunotherapy. Finally, we saw an increase in heterogeneity and proportion of anti-inflammatory Mɸ (AIMɸ), with a subset of cycling anti-inflammatory Mɸ, STAB1 + Mɸ (Fig. 1I) and CAMLs (Fig. 1H) being abundantly present in tumour tissue. Interestingly, we found a strong negative correlation between the frequency of STAB1 + Mɸ/AIMɸ and T/NK cells across patients, highlighting the key role of Mɸ in restraining the infiltration of cytotoxic cells in the lung tumour tissue (Fig. 2A). This is in line with a recent work describing that monocyte-derived Mɸ in human NSCLC acquire an immunosuppressive phenotype and restrain the infiltration of NK cells. A Heatmap showing the Pearson correlation between the relative cell-type abundance for each immune cell type (calculated within the CD235− enrichment). Colour indicates the Pearson correlation value, asterisks indicate the level of significance of the two-sided association test computed on Pearson’s product-moment correlation coefficients (*P < 0.05, **P < 0.01, ***P < 0.001). B Heatmap showing the number of LR interactions between all cell types summarised by broad cell annotations in LUAD (left) and LUSC (right). Rows were hierarchically clustered using the complete linkage method on euclidean distances. C Sankey diagram showing the tumour-specific interactions in LUAD and LUSC for selected ICIs detected by cellphoneDB. Line colour identifies whether the LR interaction between each cell type was found in LUAD only (orange), in LUSC only (green) or in both tumour types (blue). D Dotplot for the ICI genes and cell types highlighted in (C), split by tumour type. The size of each dot represents the percentage of cells in the cluster expressing the gene, while the colour represents the mean normalised scaled log-transformed expression of each gene in each group. E Sankey diagram showing the tumour-specific interactions in LUAD and LUSC for VEGFA/B interactors detected by cellphoneDB. Line colour identifies whether the LR interaction between each cell type was found in LUAD only (orange), in LUSC only (green) or in both tumours (blue). F Sankey diagram showing the tumour-specific interactions in LUAD and LUSC for EGFR interactors detected by cellphoneDB. Line colour identifies whether the LR interaction between each cell type was found in LUAD only (orange), in LUSC only (green) or in both tumours (blue). LUAD and LUSC have very different prognoses and are often considered as different clinical entities. To examine if differences in clinical features stem from distinct cellular composition, we compared the frequency of immune and non-immune cell subsets within CD235- samples from LUAD versus LUSC patients. We observed minor differences in cell frequency that did not reach statistical significance after P value correction (Supplementary Fig. 4A and Supplementary Data 7 and 8). Furthermore, there was no clear association between the frequency of immune and non-immune cells observed in patients and the cancer subtype, cancer stage or sex (Supplementary Fig. 4B, C), suggesting that the TME composition is rather similar in LUAD and LUSC. While LUAD and LUSC shared similar cellular compositions, the observed clinical distinctions may arise from varying intercellular interactions. Therefore, we examined whether different cell–cell interaction networks were employed within the TME in LUAD versus LUSC. To this end, we identified a putative list of cell–cell interactions exclusively observed in each tumour type environment by inferring statistically significant ligand–receptor pairs (L–Rs) that were not detected in background or healthy and their corresponding cell types, using CellPhoneDB. Although the two tumour subtypes showed a similar interaction network that mostly involved interactions between non-immune cells, AIMɸ and T cells (Fig. 2B), there were also some notable differences. First, we identified overall a higher number of L–Rs in the LUAD dataset (Supplementary Fig. 4D and Supplementary Data 9–12), which was not driven by a difference in the number of cells in the LUAD (n = 105,749 cells) vs LUSC (n = 230,066 cells) dataset. Secondly, several pairs of immune checkpoint inhibitors (ICI) and their respective inhibitory molecules were differentially co-expressed in LUAD versus LUSC (Fig. 2C, D). For example, LGALS9-HAVCR2 (TIM3), NECTIN2-CD226 (DNAM1) and NECTIN2/NECTIN3-TIGIT were frequently identified in LUAD, and the putative ICI CD96-NECTIN1 was found preferentially in LUSC (Fig. 2C, D). In contrast, CD80/CD86-CTLA4 and HLAF-LILRB1/2 were found in both tumour subtypes (Fig. 2C, D). LILRBs (leucocyte Ig-like receptors) are emerging as potential targets for next-generation immunotherapeutics as their blocking can potentiate immune responses. The most commonly used immunotherapies for lung cancer block the interaction between PD1 and PDL1, and recent clinical trials suggested that anti-CTLA4 and anti-PD1 combination therapy improved the survival of patients independent of tumour PD1 expression. Within our dataset, we did not observe PD1-PDL1 interactions in either of the tumour subtypes (Fig. 2C, D). Our initial analysis suggests that other ICIs (such as CTLA4, TIGIT, LILRB1/2 and TIM3) might be promising targets in the treatment of NSCLC. Of the significant L–Rs detected in both LUAD and LUSC we noted several pairs involved in angiogenic signalling in different populations of myeloid cells such as VEGFA/B-FLT1, VEGFA-KDR and VEGFA-NRP1/2. Although VEGFA and VEGFB were found to be expressed in both LUAD and LUSC, their receptors were more frequently found in LUAD, especially in fibroblasts (Fig. 2E and Supplementary Fig. 4E). Similarly, we observed significant expression of EGFR ligands signalling in AT2 and cycling epithelial cells, such as EGFR-EREG, EGFR-AREG, EGFR-HBEGF and EGFR-MIF, although MIF expression was found more frequently in cells from LUSC (Fig. 2F and Supplementary Fig. 4F). Finally, we observed key co-stimulatory signals required to support lymphoid cell activation, such as CD40-CD40LG, CD2-CD58, CD28-CD86, CCL21-CCR7, and TNFRSF13B/C-TNFSF13B (TACI/BAFFR-BAFF) (Supplementary Fig. 4G), which are often associated with the presence of ectopic lymphoid organs mainly consisting of B cells, T cells, and DCs i.e., tertiary lymphoid structures (TLS). TLS are usually correlated with the longer relapse-free survival in NSCLC. The significant L–Rs and their interacting cell types were calculated based on the co-expression of genes in different cell-type clusters from the scRNA-seq dataset using CellPhoneDB. However, in order to discern biologically significant interactions, it is essential to ascertain whether the cell types identified as interacting are indeed physically co-located. To achieve this, we considered how the scRNA-seq-identified cell types are spatially arranged on tissue sections. We applied an integrative approach which combines the scRNA-seq of the tumour and background samples with the spatial transcriptomic (STx) profile of the fresh frozen tumour and background tissue sections. We performed 10× Visium on two consecutive, 10-µm sections, from eight patients, seven of which matched the samples used for the scRNA-seq. We analysed 36 sections in total (ntumour = 20, nbackground = 16) with an average UMI count of 6894/spot in tumour and 3350/spot in the background. Next, we used cell2location and cell-type specific expression profiles from our scRNA-seq dataset to deconvolute cell-type abundances on the tissue (Fig. 3A, see “Methods”). A Spatial images depicting the cell abundance estimated by cell2location for AT2 cells, AIMɸ and Tregs on a representative tumour section. B Relative proportion of immune (left) and non-immune (right) cell types calculated on the cell abundance estimations by cell2location in tumour and background sections. Immune cells were grouped according to their broad annotations. Arrows indicate an increase (↑) or a decrease (↓) in the tumour, compared to the background. Pairwise comparisons were performed with a two-sided Wilcoxon rank test and Bonferroni correction for multiple comparisons. *P < 0.05, **P < 0.01, ***P < 0.001. Arrows without asterisks indicate that the cell type was found only in the tumour or background. Please refer to Supplementary Data 13 and 14 for the exact P values. C Heatmap of spatial LR colocalization. LR gene pair co-expression was estimated in each spot for all sections, and the frequency of colocalising vs. non-colocalising spots in the tumour and background was compared using a χ test followed by Bonferroni multiple comparison correction. Dark-grey tiles indicate that the frequency of colocalising gene pairs was significantly different in tumour and background sections. Green column annotations indicate the LR pairs which were significant in at least four out of eight patients. Row annotations indicate tumour type. D Boxplot showing the frequency of colocalising LR pairs significantly different in tumour vs background in each section analysed. N = 8 patients. Boxes are plotted with default settings in the Python Seaborn package, i.e., boxes show quartiles with whisker length being 1.5 times the interquartile range. Source data is provided as a Source Data file. E Spatial images depicting the location of spots in which the LR pair was found co-expressed in tumour (top) and background (bottom), for NRP1-VEGFA, NECTIN2-TIGIT, PD1-PDL1, CD96-NECTIN1 and HAVCR2-LGALS9. Representative sections from one patient. Once the cell types were resolved on the tissue sections, we examined the frequency of different cell types across all sections from tumour and background tissue. The cell-type abundance in tumour and background were computed by summing up the posterior 5% quantile (q05) value of estimated cell abundance by cell2location, across spots that passed QC (“Methods”). Our analysis confirmed that the differences in the frequency of cell types across all sections in tumour versus background was in line with the results obtained in the scRNA-seq data (Fig. 3B). For example, in tumours we found an increase in the proportion of B cells (Padj = 0.0372) and cycling AT2 cells (Padj = 0.0147) compared to the background tissue, and a decrease in the proportion of immature cells (Padj = 0.0012), NK cells (Padj = 0.0012), and LECs (Padj = 0.00077, Supplementary Data 13 and 14). However, the proportions of other cell types estimated from the scRNA-seq data or the STx data within the tumour or background showed some discrepancies (Supplementary Fig. 4H, I). This was particularly evident within the non-immune populations, where STx estimated higher proportions of LECs, activated adventitial fibroblasts and cycling subsets, compared to scRNA-seq. Disparities in cell proportions between different methodologies were previously shown by others, underscoring the potential influence of distinct sampling biases inherent to scRNA-seq and STx techniques like Visium. In the case of scRNA-seq, variations in cell digestion sensitivity can lead to differential representation of cell types. Meanwhile, with Visium, discrepancies might arise from variations in the location of tumour resections as well as differences in sample sizes compared to scRNA-seq studies. Nevertheless, the overall concordance in the results obtained by scRNA-seq and Visium suggests that our spatial “map” of different cell types faithfully represents their distribution in the tissue. Next, we examined the spatial co-localisation of the L–Rs identified by cellphoneDB. The L–Rs were considered to co-localise if both genes were expressed in the same spot and above median value for the given genes across the section spots. We then compared the frequency of spots in which L–R genes were colocalising versus non-colocalising in the matched tumour versus background sections, using a χ test (“Methods”). Due to the low number of tissue blocks collected from LUSC and LUAD patients (NLUSC = 3, NLUAD = 5), the statistical power was not sufficient to perform a comparative analysis between spatial localisation of LUAD/LUSC-specific L–Rs. Nevertheless, we confirmed that several of the aforementioned tumour-specific L–Rs colocalized significantly more in tumour than in background sections, including NRP1-VEGFA and the ICIs NECTIN2-TIGIT, LGALS9-HAVCR2, and CD96-NECTIN1 (Fig. 3C–E and Supplementary Data 15). Consistent with the cellphoneDB results, we found no significant colocalization of PD1-PDL1 in the tumour sections. Tumour samples obtained from surgical resection contain both malignant and residual normal epithelial cells. A significant challenge in scRNA-seq of human tumours lies in the differentiation of cancer cells from non-malignant counterparts. Therefore, we applied Copynumber Karyotyping of Tumors (CopyKAT) to discern genome-wide aneuploidy within individual cells. The principle driving the computation of DNA copy number events from scRNA-seq data is rooted in the notion that the expression levels of neighbouring genes can provide valuable information to infer genomic copy numbers within that specific genomic segment. Since aneuploidy is common in human cancers, cells with genome-wide CNAs are considered as tumour cells. Analysis using CopyKAT revealed extensive, patient-specific CNAs in tumour tissue (Fig. 4A and Supplementary Fig. 5A) but not in the background. Within individual tumour samples, the CNAs were detected in AT2 and cycling AT2 cells, and in some patients these genetic alterations were shared between AT2/cycling AT2 cells and atypical epithelial cells, suggesting a close lineage relationship between different epithelial subpopulations (Fig. 4A and Supplementary Fig. 5A). We confirmed this finding by inferring the trajectory of non-blood cell populations in tumour using Partition-Based Graph Abstraction (PAGA). PAGA showed differentiation continuity between AT2 cells, cycling AT2/epithelial cells, and atypical epithelial cells on one side and ciliated epithelial cells and transitioning epithelial cells on the other (Fig. 4B). Furthermore, blinded histological evaluation confirmed the overlap between pathologist-defined tumour sites and AT2 and cycling AT2 cells predicted by cell2location, suggesting their tumour cells status (Fig. 4C). Less overlap was observed for atypical epithelial cells (Fig. 4C). The differential expression analysis (DEA) of AT2 cells from tumours compared to background showed upregulation of genes involved in hypoxia, TP53 pathways, and metabolic rewiring in tumours. AT2 cells in tumour-upregulated genes involved both in glycolysis and oxidative phosphorylation (Fig. 4D and Supplementary Data 16). While the importance of glycolysis in tumour cells is well-established, it was recently reported that human NSCLC use glucose and lactate to fuel the tricarboxylic acid (TCA) cycle. In addition, the tumour AT2 cells were noted to express more LYPD3 compared to background AT2 cells (log2FC = 2.04, Padj = 0.039, Supplementary Data 16), an adhesion protein which has previously been connected to poor prognosis in NSCLC and is currently being targeted in preclinical and clinical studies. A CNA analysis. The plot shows chromosomal gains (red lines) and losses (blue lines) estimated by CopyKat in each chromosome arm for different cell types and patients in the tumour dataset. All immune cell types were grouped together for plotting purposes. B PAGA graph overlaid on the diffusion maps (force-directed layout—FLE embedding) computed for non-immune cell types in tumour. C First three panels—Representative blind annotations from a qualified pathologist, indicating the areas of tumour infiltration (left), binning of the tumour area on the Visium spots (centre) and the spots that passed QC (right). The last three panels—cell2location estimation for AT2 cells (left), Cycling AT2 cells (centre) and Atypical epithelial cells (right) on the same sections, overlaid with the pathologist’s annotation for the tumour infiltration (green contour). D Overrepresentation analysis on gene ontology—biological processes (GO:BP) and REACTOME database by clusterProfiler R package, using DEGs upregulated by AT2 cells in tumour vs background. Source data is provided as a Source Data file. E Detailed overview of CNAs in AT2 and CAMLs from the tumour of one representative patient. Bars indicate the frequency of cells harbouring chromosomal gains (red bar) or losses (blue bars) in specific chromosomal regions. F Scatterplot of the KL divergence for losses (x axis) and gains (y axis) between each cell type in the tumour dataset calculated using their gain and loss distribution. All immune cell types were grouped together for plotting purposes. G Spatial images depicting the cell abundance estimated by cell2location for AT2 cells and CAMLs on three representative tumour sections. H Hierarchical clustering of the correlation distance calculated on cell-type composition (as estimated by cell2location) across spots that passed QC in all tumour sections. I Non-negative matrix factorisation built on the q05 estimation of cell-type abundance across spots that passed QC (as estimated by cell2location) in all tumour sections. Interestingly, the population of CAMLs also showed substantial CNAs that were similar to those of AT2 cells and cycling AT2 cells from the same patient (Fig. 4A, E and Supplementary Fig. 5A, B). To measure the difference of the distribution of genomic gain and loss between cell types in a statistically robust manner, we calculated the Kullback–Leibler (KL) divergence (Fig. 4F and Supplementary Fig. 5C). CAMLs had KL divergence values comparable to CNA-harbouring tumour cells, thus confirming the similarity of their CNA profiles (Fig. 4F and Supplementary Fig. 5C). As CAMLs co-expressed a wide array of myeloid genes as well as typical epithelial genes (Fig. 1D–F and Supplementary Fig. 1D), had a low doublet score and shared the same CNA signature as tumour cells, we hypothesised that these cells might represent a subset of Mɸ tightly attached to a cancer cell. It is possible that these Mɸ were undergoing phagocytosis or fusion. CAMLs have been previously isolated from peripheral blood of cancer patients and described to facilitate circulating tumour cells seeding of distant metastases. Our analysis suggested that CAMLs can also be isolated from tumour tissue. To validate that CAMLs are in physical proximity to tumour cells in situ we examined our STx sections. We calculated across all sections (8 patients, nsections = 20) the Pearson correlation between the relative abundance of the cell types that reside in the same spot and are therefore co-localised. Our analysis showed that CAMLs indeed co-localised with AT2 cells (Fig. 4G, H). We confirmed this finding using non-negative matrix factorisation (NMF) on the absolute cell-type abundances estimated by cell2location that defined factors of co-occurring cell states (Fig. 4I). To determine the specific Mɸ population from which CAMLs likely originate, we employed PAGA to elucidate the differentiation path of the myeloid cell population in our tumour dataset (Supplementary Fig. 5D). The analysis revealed continuity of the differentiation transitions between diverse populations of myeloid cells. Within the PAGA trajectory, alveolar Mɸ (AMɸ) and AIMɸ showed high PAGA connectivity indicating their high transcriptional similarity. Both AIMɸ and AMɸ showed the strongest connectivity on the PAGA trajectory with STAB1 + Mɸ which, in turn, were linked with CAMLs. In line with trajectory analysis, CAMLs co-expressed many of the genes specific to STAB1 + Mɸ (Supplementary Fig. 2A), supporting the hypothesis that CAMLs are likely derived from STAB1 + Mɸ following their close interaction with tumour cells. Finally, DEA analysis between CAMLs from LUSC versus LUAD patients, showed upregulation of KRT17, KRT5 and KRT6A in LUSC samples (Supplementary Data 17). These KRT genes were previously identified as markers of LUSC in multiple studies, which supports hypothesis that CAMLs arise from the interaction between Mɸ and tumour cell. Mɸ, traditionally categorised into distinct M1 (classically activated) and M2 (alternatively activated) phenotypes, are now understood to exist along a dynamic spectrum of functional states. This concept of Mɸ plasticity underscores their ability to seamlessly transition between pro-inflammatory and anti-inflammatory roles in response to intricate cues from their microenvironment (Supplementary Fig. 5D). To better understand the transcriptional changes that different Mɸ populations undergo in the TME, we performed DEA. In tumours, both AMɸ and AIMɸ upregulated genes involved in cholesterol and lipid transport and metabolism (such as ABCA1, APOC1, APOE, FABP3 and FABP5) compared to the background tissue (Fig. 5A, B and Supplementary Data 18 and 19). Cholesterol plays a vital role in tumour growth due to the high demand of newly synthesised cellular membranes during cancer cell proliferation. Hypoxia-related genes were upregulated in AT2 cells in tumour compared to the background (Fig. 4D), which can promote cholesterol auxotrophy in tumour cells by suppressing cholesterol synthesis, thereby forcing them to rely on exogenous cholesterol uptake. In our dataset, we detected higher expression of the cholesterol exporter ABCA1 and no expression of low-density lipoprotein receptor (LDLR) in AMɸ and AIMɸ, the latter gene being responsible for the uptake of cholesterol-carrying lipoprotein particles into cells, suggesting preferential export of cholesterol from TAMs to the TME (Fig. 5A). Interestingly, we also noted a high expression of TREM2 in both AMɸ and AIMɸ (Fig. 5A), which plays a prominent role in efflux of cholesterol in microglia. To validate the increased levels of cholesterol in the TME, we stained matched tumour and background tissue sections with BODIPY™ 493/503, a stain targeting cholesterol and other neutral lipids. We found a significant increase in the BODIPY signal in the tumour sections, compared to background tissue (Fig. 5C, D), confirming an increased availability of neutral lipids in the tumour, possibly as a result of an increased export by TAMs. A Volcano plot of DEGs (red) for AIMɸ in tumour vs background, extracted using the py_DESeq2 package. B Overrepresentation analysis on gene ontology—biological processes database by clusterProfiler R package, using the DEGs upregulated by Alveolar Mɸ and AIMɸ in tumour vs background. Source data is provided as a Source Data file. C IHC for CD68 and neutral lipids (BODIPY 493/503) on tumour and background tissue sections. Maximum intensity projection of Z-stacks. Scale bar 50 µm. D Area covered by the BODIPY signal in tumour and background section. The difference in BODIPY area coverage was determined with a paired, two-sided t test, matching tumour and background sections from the same patients. N = 5 patients. Source data is provided as a Source Data file. E IHC for CD68 and STAB1 on tumour (left) and background (right) tissue sections. Maximum intensity projection of Z-stacks. Inlets show a detailed magnification on a single cell. Scale bar 20 µm. F Quantification of STAB1+ cells within the CD68+ macrophage population. The fraction of the STAB1 + CD68+ area is shown as a percentage of the total CD68+ area. Data are presented as mean value and standard deviation (n = 3 biological replicates). Source data is provided as a Source Data file. G Staining for CD68, STAB1 and PanCK on tumour tissue sections. Maximum intensity projection of Z-stacks. Inlets show a detailed magnification on a single cell. Scale bar 20 µm. H Quantification of STAB1 + CD68+ cells within the CD68+ macrophage population in NSCLC. Data are presented as mean value and individual data points (n = 2 biological replicates). Source data is provided as a Source Data file. I Dotplot showing the expression of the “STAB1 signature genes” across all macrophage subsets and CAMLs in tumour. J Volcano plot of DEGs identified by py_DESeq2 (red) for Alveolar Mɸ vs STAB1 Mɸ in tumour. K Overrepresentation analysis on gene ontology— biological processes database by clusterProfiler R package, using the DEGs from Alveolar Mɸ vs STAB1 Mɸ (top) and AIMɸ vs STAB1 Mɸ (bottom) in tumour (left—upregulated by STAB1 Mɸ; right—upregulated by Alveolar Mɸ or AIMɸ). Source data is provided as a Source Data file. STAB1 + Mɸ were identified in the tumour resections (Fig. 5E–H, Supplementary Fig. 2 and Supplementary Notes), so we used DEA to identify a set of genes that were specific for STAB1 + Mɸ compared to tumour AIMɸ or AMɸ. We identified 20 genes, from here on referred to as “STAB1 signature genes” (Fig. 5I). Interestingly, STAB1 + Mɸ uniquely expressed SLC40A1, which encodes for the ferroportin, the only known protein that exports ferrous iron from the cytoplasm across the plasma membrane and is key for the iron-releasing activity of macrophages (Fig. 5I, J and Supplementary Data 20 and 21). Ferroportin-mediated release of free iron by M2 Mɸ was reported to promote the proliferation of renal carcinoma cells in vitro, possibly by supporting the high iron requirement due to increased DNA synthesis. Furthermore, compared to AMɸ, STAB1 + Mɸ expressed lower levels of ferritin heavy chain 1 (FTH1) and ferritin light chain (FTL) encoding for the iron storer ferritin (Fig. 5J and Supplementary Data 20). Consistent with the hypothesis of their sustained export of free iron to the extracellular milieu, STAB1 + Mɸ downregulated genes involved in iron sequestration (Fig. 5K). Taken together, our analysis suggests that macrophages undergo “reprogramming” within the TME and adopt a transcriptional signature that facilitates cholesterol efflux and iron export, thus supporting tumour progression. Embryonic development shares many characteristics with tumour tissue, including rapid cell division, cellular flexibility, and a highly vascular microenvironment. It has been recently reported that during tumorigenesis, Mɸ can undergo oncofoetal reprogramming and acquire a foetal-like transcriptional identity that supports tumour growth and metastasis. Considering that some of the STAB1 signature genes are typically expressed by foetal Mɸ (such as STAB1, FOLR2, SLC40A1, MERTK, GPR34 and F13A1), we wanted to explore if further transcriptional commonalities exist between tumour-originating STAB1 + Mɸ and Mɸ isolated from human foetal lung. To this end, we combined tumour- and background-originating myeloid cells from our dataset (n = 347,364 cells) with myeloid and progenitor cells from a publicly available foetal lung scRNA-seq dataset (n = 6,947 cells) using Harmony. Next, we performed Leiden clustering on the neighbourhood graph and examined how cell types are distributed within the clusters (Supplementary Fig. 6A, B). To examine similarity in their gene expression profile, we applied hierarchical clustering and built a dendrogram by estimating the correlation distance between cell types on the harmonised PC embedding space, under the complete linkage criterion of hierarchical clustering (Fig. 6A). A Hierarchical clustering of the correlation distance calculated on each cell in the harmonised (tumour myeloid + background myeloid + foetal lung myeloid) PC space. B Violin plot showing the expression level of the “STAB1 gene signature” across myeloid cell and progenitor populations identified in a publicly available human foetal lung atlas. C Dotplot of the expression of each gene in the “STAB1 gene signature” in selected foetal lung macrophage populations. The size of each dot represents the percentage of cells in the cluster expressing the gene, while the colour represents the mean expression of each gene in each cluster. D Violin plot showing the expression level of the “STAB1 gene signature” across the clusters identified in the publicly available MoMac-VERSE dataset. E Dotplot of the expression of each gene in the “STAB1 gene signature” in selected macrophage populations from the MoMac-VERSE. The size of each dot represents the percentage of cells in the cluster expressing the gene, while the colour represents the mean expression of each gene in each cluster. F Violin plot showing the expression level of the “AMɸ gene signature” across myeloid cell and progenitor populations identified in the publicly available “MoMac-VERSE” dataset. G Violin plot showing the expression level of the “AMɸ gene signature” across myeloid cell and progenitor populations identified in a publicly available human foetal lung atlas. H Dotplot of the expression of each gene in the “AMɸ gene signature” in selected macrophages populations identified in the “MoMac-VERSE” dataset. The size of each dot represents the percentage of cells in the cluster expressing the gene, while the colour represents the mean expression of each gene in each cluster. I Dotplot of the expression of each gene in the “AMɸ gene signature” in selected foetal lung macrophage populations. The size of each dot represents the percentage of cells in the cluster expressing the gene, while the colour represents the mean expression of each gene in each cluster. We observed that tumour cDC2 exhibited the strongest correlation with background cDC2, whereas tumour mo-DC2 displayed the highest correlation with foetal DC2 and, in a broader context, with background mo-DC2. The population of pDC from tumour, background and foetal lung were closely correlated. Similarly, tumour monocytes were correlated with foetal classical monocytes and background monocytes. In contrast, macrophage populations in tumour, and in particular STAB1 + Mɸ, were correlated with foetal macrophages. STAB1 + Mɸ clustered predominantly with foetal SPP1 + Mɸ (Fig. 6A), which accounted for over 80% of all foetal lung macrophages reported in ref. . Consistent with this finding, SPP1 + Mɸ had a high expression of the “STAB1 signature genes” compared to other haematopoietic populations (Fig. 6B, C). Our analysis substantiates the idea that monocytes within the tumour environment, as they undergo differentiation into anti-inflammatory macrophages, acquire a transcriptional signature akin to that of foetal macrophages. This distinctive transcriptional signature was not observed in the macrophages from surrounding normal tissue. To further examine the prevalence of STAB1 + Mɸ in other pathologies, including other cancers, we examined the expression of “STAB1 signature genes” across a diverse group of myeloid cells using a published atlas of human monocytes and Mɸ collected from 12 different healthy and pathologic tissues (n = 140,327 cells), called MoMac-VERSE. The cluster of “HES1+ macrophages” identified in MoMac-VERSE showed the highest expression of the “STAB1 signature genes” (Fig. 6D, E). Similar to STAB1 + Mɸ, HES1+ macrophages accumulated in tumours of lung cancer patients but also liver cancer patients and were suggested to represent a cluster of “long-term resident-like” Mɸ with foetal-like transcriptional signature. In contrast, “C1Q” Mɸ from MoMac-VERSE, which have been described as lung alveolar Mɸ, had a high expression of genes unique to our tumour alveolar AMɸ (from here on referred as “AMɸ signature genes”, Fig. 6F, H). In the context of foetal lung, a rare population of APOE + Mɸ, which accounted for less than 1% of all foetal lung macrophages reported in ref. , had a high AMɸ signature genes score (Supplementary Notes and Fig. 6G, I, see “Methods”). Taken together, our analysis suggests that tumour macrophages, especially STAB1 + Mɸ, exhibited a transcriptional signature reminiscent of Mɸ during foetal lung development, suggesting that they have undergone oncofoetal reprogramming within the NSCLC tumour environment. Our study represents a large single-cell multiomics analysis of samples collected from treatment-naive patients with NSCLC. We integrated scRNA-seq data from nearly 900,000 cells from tumour resections and adjacent non-malignant tissue from 25 treatment- naive patients with spatial transcriptomics to build an atlas of immune and non-immune compartments in lung cancer. LUAD and LUSC, the two most common NSCLC subtypes, exhibit markedly different prognostic outcomes and have shown potential for subtype-specific therapies. Despite similar cell-type composition, we observed significant differences in the co-expression of several ICIs and inhibitory molecules between LUAD and LUSC, highlighting therapeutic opportunities. LUAD samples frequently expressed TIGIT and TIM3 (HAVCR2), while in LUSC we found the putative ICI CD96-NECTIN1. While different advanced clinical trials targeting TIGIT, including in patients affected by NSCLC, are ongoing, progress on TIM3 and CD96 is more limited. A first-in-human phase-I study evaluating the anti-CD96 monoclonal antibody GSK6097608 as monotherapy alone or in combination with anti-PD1 (dostarlimab) started recruiting patients only recently. Taken together, our data suggest that LUAD and LUSC patients might benefit from specific immunotherapy targeting ICIs as TIM3, TIGIT and CD96. The TME plays a crucial role in modulating the population and behaviour of Mɸ. We found that, compared to the adjacent non-tumour tissue, tumour resections harboured a lower proportion of monocytes but a higher proportion of monocyte-derived cells, such as mo-DC2s and anti-inflammatory Mɸ, suggestive of an enhanced monocyte differentiation in the TME. The prevalence of anti-inflammatory Mɸ, including STAB1 + Mɸ, exhibited an inverse relationship with the abundance of natural killer (NK) cells and T cells in the tumour environment; and the NK cells within the tumour exhibited reduced cytotoxic activity. Our results are in line with the recent findings that the removal of tumour cell debris by lung Mɸ leads to their conversion into an immunosuppressive phenotype, consequently hindering the infiltration of NK cells into the TME. Mɸ with elevated levels of tumoural debris were reported to upregulate genes involved in cholesterol trafficking and lipid metabolism, a characteristic shared with anti-inflammatory Mɸ in our dataset. As a result, they downregulated co-stimulatory molecules, cytokines and chemokines essential for the recruitment of CD8 + T cells, therefore becoming more immunosuppressive. Among the Mɸ population within tumours, we also identified STAB1 + Mɸ that exhibited the highest level of immunosuppression markers. These STAB1 + Mɸ displayed a gene expression pattern akin to that of foetal lung Mɸ and demonstrated a modified iron metabolism, marked by the increased expression of genes associated with iron release in the TME. Therefore, we hypothesise that STAB1 + Mɸ might play a crucial role in supporting tumour progression by sustaining the increased iron requirement of highly-cycling tumour cells. In a subcutaneous LLC1 Lewis lung adenocarcinoma model, mice lacking Stab1 expression in Mɸ, tumour growth was diminished. This outcome was attributed to a shift towards a pro-inflammatory phenotype in TAM and a robust infiltration of CD8 + T cells within the TME. STAB1 + Mɸ displayed a transcriptional resemblance to CAMLs, which concurrently expressed genes associated with both Mɸ and epithelial cells, and exhibited copy number alterations (CNAs) similar to those found in tumour cells. STAB1+ plays a pivotal role in facilitating the adhesion and engulfment of apoptotic cells by engaging in a specific interaction with phosphatidylserine, supporting the hypothesis of a strong interaction of a Mɸ with a tumour cell in CAMLs. In previous studies, CAMLs were identified by immunofluorescence in the peripheral blood of individuals affected by various solid tumours and were proposed to facilitate the dissemination and establishment of circulating tumour cells in distant metastatic sites. Here, we report their presence in multiple tumour resections, based on a combination of a compound gene expression signature, tumour-specific copy number alterations and physical proximity to tumour cells, as evident from Visium sections. Taken together, our comprehensive dataset allowed identifying a multitude of molecular changes in the Mɸ population of the lung tumour microenvironment, which will help pave the way for the development of therapeutic strategies against NSCLC. Tissue used in the research study was obtained from the Papworth Hospital Research Tissue Bank. Written consent was obtained for all tissue samples using Papworth Hospital Research Tissue Bank’s ethical approval (East of England— Cambridge East Research Ethics Committee). Human tumour and adjacent background tissues, collected from the edges of the lungs, were obtained from 25 patients following tumour resection. Human healthy lung samples were obtained from two healthy deceased donors. Both healthy samples were evaluated by an expert pathologist to exclude the presence of malignancies. The human material was provided by the Royal Papworth Tissue Bank (T02229), in accordance with the HMDMC Human Tissue Act Sample Custodian Form Version 7.0 (UK NRES REC approval reference number(s): 08/H0304/56 + 5; HMDMC 16 | 094). NSCLC FFPE tumour blocks (n = 2) used for validation of STAB1+ macrophages with Akoya were obtained from 2 different donors and purchased from BioIVT (ex-Asterand Bioscience). Informed Consent Form (ICF) and Institutional Review Board Approval Letter (IRBA) were obtained for all tissue samples. Sex was assigned (15 male and 12 female patients/donors). Sex-based analyses were not performed due to the limited sample size. Gender was not determined. Tissues were kept in cold complete RPMI medium (RPMI [Invitrogen] supplemented with 10% FBS [Sigma Millipore, catalogue number: F9665], 2 mM L-Glutamine [Life Technologies, catalogue number: 25030-024] and 100 U/ml Penicillin-Streptomycin [Thermofisher, catalogue number: 15140122]) until dissociation, which was performed on the same day of collection. Single-cell suspensions were generated as follows: tissues were placed into a petri dish and cut into small pieces of 2–4 mm and transferred into a 1.5-ml tube containing the digestion mix (complete RPMI media supplemented with 1 mg ml collagenase IV and 0.1 mg ml DNase I) and minced using surgical scissors. Minced tissues were incubated for 45 min at 37 °C and vortexed every 15 min. Digested tissues were passed through a 100-μm strainer into a falcon tube prefilled with cold PBS. Cells were then centrifuged for 5 min at 300 × g, 4 °C and the pellet was resuspended into 1× RBC lysis buffer (eBioscience) for 2 min at room temperature, after which 20 ml of cold PBS were added to stop the lysis reaction. Cells were cryopreserved in 5% DMSO in KnockOut Serum Replacement (KOSR; Gibco, catalogue number: 10828010) until further use. On the day of FACS sorting, cells were rapidly thawed at 37 °C and transferred to complete RPMI media. Live-cell enrichment was performed using MACS Dead Cell Removal Kit (Miltenyi Biotec) following the manufacturer’s instructions. Red blood cells were further depleted by negative selection using CD235a Microbeads (Miltenyi Biotec) and MACS LS columns (Miltenyi Biotec), following the manufacturer’s instructions. For FACS sorting, cells were stained with Zombie Aqua to exclude dead cells and the cocktail of antibodies for 30 min at 4 °C. Cells were centrifuged for 5 min at 300 × g, 4 °C, resuspended in 500 μl of 5% FBS in PBS and subsequently filtered into polypropylene FACS tubes. Immune cells were sorted as live, CD45 + ; MDSC were sorted as live, CD45 + , Lineage- (Lin: CD3, CD56, CD19), CD33 + , HLA-DR-/low (Supplementary Data 22 and Supplementary Fig. 1A). Cells were sorted into a 1.5-ml tube, counted and submitted for 10x scRNA-seq library preparation. Each cell suspension was submitted for 3’ single-cell RNA sequencing using Single Cell G Chip Kit, chemistry v3.1 (10x Genomics Pleasanton, CA, USA), following the manufacturer’s instructions. Libraries were sequenced on an Illumina NovaSeq 6000, and mapped to the GRCh38 human reference genome using the CellRanger toolkit (version 3.1.0). Integrating numerous samples, notably from diverse cancer subtypes and adjacent normal tissues, is challenging due to variations in gene programmes between samples. Consequently, these differences often hinder a coherent biological alignment when attempting simultaneous embedding. Most current integration techniques, primarily focused on batch correction, operate under the assumption of shared cell states across samples. However, while they aim to mitigate technical disparities, they might inadvertently erase genuine biological distinctions. Therefore, we applied the QC filtration and doublet removal on the merged dataset (Tumour + B/H) but we split the datasets between tumour and B/H for HVG selection, PCA, batch correction (using Harmony), clustering and annotations. Starting from the unnormalised, uncorrected gene expression matrices produced (per sample) by the CellRanger protocol, we performed careful downstream analysis of the scRNA-seq data. For each CellRanger output (corresponding to a specific technical and biological replicate of the separate tumour, background and healthy data) we identified low-quality cells or empty droplets by applying the barcodeRanks and emptyDrops functions using the R package DropletUtils. Following per-sample droplets removal, the complete set of cell expression matrices was merged (we merged tumour, background, and healthy samples), and quality control (QC) was applied to the resultant merged matrix. The remaining analysis is implemented using standard approaches in the Scanpy framework. The QC is based on three parameters: the total UMI count (lower-upper threshold [400, 100,000]), the number of detected genes (lower-upper threshold [180, 6000]), and the proportion of mitochondrial gene count per cell (20% fraction upper bound). We applied Scrublet to remove potential doublets with 0.06 as the expected doublet rate and then filtered the results using the parameter values (2 for minimum read count of cell, 3 for minimum detected cell of gene, 85 for minimum gene variability percentage, and 30 for the number of principal components used to embed the transcriptomes prior to k-nearest-neighbour graph construction). The resulting merged and filtered expression matrix is then normalised using the scaling factor 10,000, followed by log1p transformation. For dimensionality reduction, we first selected sets of highly-variable genes (HVGs) from the initial gene set of 25,718. Starting from the HVG selection, the merged matrix was split into two separate matrices: tumour, and combined background/healthy which we refer to as B/H. After HVG selection, 1604 genes were selected from the tumour matrix and 1486 from B/H. From these separate HVG sets, we applied dimensionality reduction using Principal Component Analysis (PCA). Next, we performed PCA separately for tumour and for B/H and retained the top 15 components, according to the Scree plot elbow rule. The resulting matrix is then batch corrected to account for additional technical variations arising between samples which are non-biological in origin. We apply batch correction by using harmonypy (a Python version of the original harmonyR package), based on recommended benchmarking against other procedures. Following between-sample batch correction, we computed a neighbourhood graph and applied Leiden clustering (with Leiden resolution being 1) to the 15-dimensional harmonised PCA space. For visualisation purposes, we used Uniform Manifold Approximation and Projection (UMAP) manifold embedding to capture the global features of the 15-dimensional clustered manifold and represent the global structure in two and three dimensions. We identified top 100 representative genes for each cluster by performing the Wilcoxon signed-rank test with the Bonferroni correction, followed by a filtering to obtain genes overexpressed in the target group (minimum log fold change as 0) and expressed in at least 30% of cells within the group. We did not control the fraction of gene expression of other clusters, by setting the maximum threshold as 100%. We then annotated each cell cluster according to the the expression profile of these marker genes and the expression of other canonical genes significant for different lung cell types based on the literature (see extended results). The annotation procedure was done iteratively. With this approach we generated two separate annotated UMAPs, together with associated marker genes, for the tumour and B/H datasets. To compare cell-type abundances, we calculated the proportion of each cell type within each patient and broad cell annotation in the unenriched (CD235-) samples. We contrasted cell-type proportions between groups (tumour vs. background or LUAD vs. LUSC) using a Wilcoxon rank-sum test. Finally, we corrected for multiple testing using a two-sided Bonferroni correction independently for each group analysed. The association between the relative cell-type abundance for each immune cell type was evaluated on the Pearson’s product-moment correlation coefficients. To test consistency in cell-type annotation performed separately in tumour and B/H, we performed reference-query mapping from tumour to B/H using scArches. For the 828,191 immune cells (464,952 in tumour and 363,239 in B/H) identified through our separate annotations, we selected a common set of 10,000 HVGs. We first built an scVI model and trained it on the tumour dataset using broad cell types for reference, and applied scHPL method (provided in the scArches package, parameters set to use KNN classifier, 100 neighbours and with PCA dimensionality reduction) to obtain the hierarchy for the tumour cell types. We then applied the B/H dataset to the pretrained reference model for a query, and predicted B/H broad cell types based on tumour hierarchy (probability threshold set as 0.2). Finally, we compared the predicted cell types with our separate annotations in B/H using a heatmap to visualise the confusion matrix. We initially identified a putative long list of cell–cell interactions differentially observed in the tumour environment by inferring statistically significant ligand–receptor pairs, and their corresponding cell types, using CellPhoneDB. We treated the tumour (LUAD or LUSC), background, and healthy scRNA-seq profiles as independent datasets and ran CellPhoneDB separately. To reduce the impact of randomness in the way CellPhoneDB samples from input datasets, we required that any ligand–receptor pair of interest from the CellPhoneDB database be expressed in at least 30% of cells in a particular cell-type cluster of interest. The final ligand–receptor lists were further filtered by requiring that the mean log(1 + expression) of the ligand–receptor pair be greater than 1.0, and the Bonferroni-adjusted P value be less than 0.01. From these filtered long lists, ligand–receptor pairs and corresponding cell types relevant to the tumour data are identified. When evaluating the ligand–receptor lists calculated with CellPhoneDB, we did not run on the complete datasets due to the difficulty in scaling up the CellPhoneDB statistical permutation tests to scRNA-seq with more than 10 cells. Instead, we separately stratified the tumour, healthy and background datasets such that the proportion of cell types, patients, and samples in the reduced 50% of the data recapitulated the proportions in the full dataset. Differentiation expression analysis (DEA) was performed for AT2 cells, anti-inflammatory macrophages and alveolar macrophages using a pseudo-bulk approach to compare tumour versus background. Pseudobulks were built for each patient by summing raw gene counts across all cells in each cell type investigated. The patients 1 and 4 were not included in the analysis as their cancer subtype and stage were not known at the time of analysis. Since there were differences in the cell count between datasets we downsampled the biggest cluster to the size of the smaller. The downsampling routine was repeated 100 times, such that 100 new datasets were created that match the smaller dataset. DEA was performed using sample-level pseudobulks and a Pythonic version of the DESeq2 pipeline (py_DESeq2), including the patient information as co-variate. The median adjusted p value by Benjamini–Hochberg procedure and median log2FC for each differentially expressed gene (DEG) was calculated across 100 iterations. We verified the robustness of this choice of 100 iterations by visualising the variability of the median p value across iterations, in order to assess its stability (Supplementary Fig. 6C). DEGs were filtered with median(padj)≤0.05 and |median(logFC)|≥1. Prior to performing overrepresentation analysis, the genes that were commonly upregulated in more than 50% of the contrasts were removed (DNAJB1, HSPA1A, HSPA1B, HSPB1, HSPE1, IGHA1, IGKC, IGLC2). DEGs were used to perform gene ontology (GO) overrepresentation using the clusterProfiler package. To define STAB1 + Mɸ and AMɸ gene signatures, we compared DEA results and intersected the genes significantly upregulated by STAB1 + Mɸ (or AMɸ) compared to the other Mɸ populations in tumour. To analyse myeloid cell trajectory in tumour dataset, we recomputed a neighbourhood graph from the same 15-dimensional harmonised PCA space as above, but only within myeloid cell populations. We next applied PAGA within the Scanpy package to the neighbourhood graph. In parallel, we computed the diffusion map and its force-directed layout for visualisation using the Pegasus package. We finally overlaid the PAGA network with the diffusion map using the scVelo package. We repeated the same analysis workflow but on non-immune cells in the tumour dataset. We applied the CopyKAT package to the single-cell RNA-seq data to obtain copy number calls. The Copykat pipeline was extended to obtain confident copy number calls per cell, per chromosome arm, beyond the hierarchical clustering the standard pipeline produces. Per cell copy number calls were obtained as follows: first, the regular CopyKAT (v1.0.5) pipeline was run on the unmodified UMI counts of a particular patient/environment (i.e., tumour or background) combination with default parameters, except for norm.cell.names. The norm.cell.names parameter allows for specifying which cells are used as confident diploid normals during expression normalisation. CopyKAT was set to use all cells labelled as cDC2 dendritic cells, as they are available in great numbers across all patients and an initial inspection of their expression profiles revealed no systematic copy number alterations. After CopyKAT has completed, a calling step was applied that is aimed to call whole chromosome arm alterations in individual cells. We reasoned that, on a chromosome arm basis, the distribution of binned-and-normalised expression from CopyKAT should be significantly different (higher or lower) than the distribution of the same bins in all confidently diploid cells. For each chromosome arm, we model the distribution of all data bins from the confidently diploid cells as a normal distribution. Each bin on that same chromosome arm from a candidate aneuploid cell is then tested against that distribution. Finally, when more than 50% of bins across that chromosome arm are significant, the arm is marked as altered in that cell. The above-described procedure yields a conservative true/false call per cell, per chromosome arm without directly distinguishing between gains and losses. To obtain a profile with gains and losses as is shown in Fig. 4A, we discretise the values for each bin in each cell: If the arm is altered and the expression value of the bin is negative: −1, if the arm is altered and the expression value is positive: +1, if the arm is unaltered: 0. The discretized values are then finally summed per bin across all cells of a particular cell type and divided by the number of cells of that cell type to obtain the fraction of cells with an alteration as shown in Fig. 4A. Tissues were frozen in dry-ice-cooled isopentane and stored in air-tight tissue cryovials at −80 °C. The tissues were embedded in an optimal cutting temperature compound (OCT) and cryosectioned in a pre-cooled cryostat at 10 μm thickness on SuperFrost slides. On the day of the experiment, slides were thawed at room temperature for less than 5 min, then immersed in a fixation solution (4% PFA in PBS) for 20 min. After three washes with PBS, each section was permeabilized with freshly prepared 0.2% Triton-X100 (Sigma Aldrich) for 10 min at room temperatures, followed by three washes in PBS. Unspecific binding was blocked by incubating the sections in PBS + 2.5% BSA for 1 h at room temperature. Following two washes in PBS, sections were incubated with recombinant rabbit anti-CD68 (Abcam ab213363, 1:50) and mouse anti-STAB1 (Santa Cruz Biotechnology sc-293254, 10 µg/ml) in PBS + 0.5% BSA overnight at 4 °C. Primary antibodies were removed and sections washed three times with PBS, then incubated with the appropriate secondary antibodies (goat anti-rabbit AlexaFluor 594 and goat anti-mouse AlexaFluor 488 Abcam) 1:500 in PBS + 0.5% BSA for 2 h at room temperature, protected from light. Two confocal immunohistochemistry z-stacks each for tumour and background tissue from three patients were analysed. Using Fiji (ImageJ) software, the STAB1+ and CD68+ areas were segmented by automatic thresholding and quantified in each image of the z-stack. To assess the levels of cholesterol and neutral lipids we further stained tumour and background tissue sections with BODIPY™ 493/503 (Invitrogen). After three washes in PBS, sections were incubated with a 10 µg/ml solution of BODIPY™ 493/503 in PBS (1:100 from a stock 1 mg/ml solution in DMSO) for 15 min at room temperature. Following four washes in PBS, sections were incubated for 90 s with TrueVIEW (Vector Laboratories), washed by immersing in PBS for 5 min, then tap-dried and mounted in VECTASHIELD Vibrance™ Antifade. Sections were imaged using a Zeiss LSM 710 confocal microscope at ×20 (Plan-Apochromat ×20/0.8 M27) and ×63 (Plan-Apochromat ×63/1.40 Oil DIC M27) magnification. Tile scans were set to cover an area of 3541 × 3542 microns for all sections. ImageJ was used to remove background BODIPY signals and calculate the area covered by the thresholded BODIPY on the stitched images. To compare the area covered by BODIPY in tumour and background, we used a paired t test at a patient level, after confirming the normal distribution of the data using a Shapiro–Wilk test. To investigate the oncofetal reprogramming of myeloid cells in NSCLC, we took advantage of a published scRNA-seq dataset of foetal lung myeloid cells and the published “MoMac-VERSE”. The expression of the “STAB1 signature genes” and of the “AMɸ signature genes” across lung foetal myeloid cells was determined using the AddModuleScore function in Seurat v4.3. To combine foetal lung and adult lung tumour-infiltrating myeloid cells, we isolated the myeloid cells from our tumour and background datasets and integrated those with the aforementioned foetal lung myeloid dataset using the Pegasus package, following the following workflow: (i) remove rarely expressed genes (less than 10 cells), normalisation and log1p transformation, (ii) robust and highly-variable gene selection, (iii) PCA with optimal PC number determined by random matrix theory (resulting in 75 PCs), (iv) batch effect correction using Harmony, and (v) Leiden clustering on neighbourhood graph. The dendrogram was built by estimating the correlation distance between cell types on the harmonised PC embedding space, under complete linkage criterion of hierarchical clustering. The UMAP was computed to obtain a 2D summary of the harmonised PC space. Tissues were frozen in dry-ice-cooled isopentane and stored in air-tight tissue cryovials at −80 °C. Prior to undertaking any spatial transcriptomics protocol, the tissues were embedded in OCT compound and tested for RNA quality with an Agilent BioAnalyser. Tissues with RNA integrity (RIN) values > 7 were cryosectioned in a pre-cooled cryostat at 10 μm thickness. Two consecutive sections were cryosectioned at 10 μm thickness in a pre-cooled cryostat and transferred to the four 6.5 mm × 6.5 mm capture areas of the gene expression slide. Slides were fixed in methanol for 30 min prior to staining with H&E and then imaged using the Nanozoomer slide scanner. The tissues underwent permeabilization for 24 min. Reverse transcription and second strand synthesis was performed on the slide with cDNA quantification using qRT-PCR using KAPA SYBR FAST-qPCR kit (KAPA Biosystems) and analysed on the QuantStudio (ThermoFisher). Following library construction, these were quantified and pooled at 2.25 nM concentration. Pooled libraries from each slide were sequenced on NovaSeq SP (Illumina) using 150 base pair paired-end dual-indexed set-up to obtain a sequencing depth of ~50,000 reads as per 10x Genomics recommendations. The sequencing libraries were then processed by SpaceRanger (version 1.1.0) on the reference GRCh38 human reference genome to estimate gene expression on spots. We used cell2location to deconvolute the cellular composition of each capture area (spot). As our scRNA-seq cells were annotated independently for tumour and the combined B/H datasets, we applied the deconvolution model separately as well, using tumour annotation to infer spatial cell composition of tumour sections, and background annotations for background datasets. Only spots with total UMI counts above 800 were used in downstream analysis. The cell-type abundance in tumour and background sections were computed by summing up the q05 cell abundance, as estimated by cell2location, across spots that passed QC. Cell-type composition was computed by normalising each cell type’s abundance with the total abundance of all cell types. We compared cell-type composition between tumour and background with Wilcoxon signed-rank test, followed by Bonferroni correction. On tumour sections, we estimated the correlation distance on cell-type composition across valid spots, applied hierarchical clustering with complete linkage, and visualised the results as a dendrogram. In addition, we applied non-negative matrix factorisation analysis to the q05 estimation of cell-type abundance with eight factors. To study the expression of ligand–receptor pairs on the 10X Visium, we first binarised the expression of each gene in the LR pairs in the spots that passed QC. We considered a gene being expressed in a spot if its cell2location estimated abundance were higher than the median counts for that gene in the corresponding section. We counted spots where both genes in each LR pair were either co-expressed or not, in tumour and background sections from the same patient, and subsequently, applied the χ test on the contingency table. To correct for multiple comparisons, we adjusted the P value using a conservative Bonferroni correction for all the LRs enriched in tumours in the cellphoneDB analysis (309 * 8 patients). LRs were considered significantly enriched in tumour if the Bonferroni-adjusted P value was lower than 0.05 in at least four patients. 5 μm thick sections were generated from NSCLC FFPE tumour blocks. An antibody cocktail was prepared with optimal dilutions of each of the following conjugated antibodies: anti-human Stabilin-1 antibody (clone #840449, catalogue #MAB3825, R&D systems) was conjugated to a custom oligo barcode according to instructions in Akoya Biosciences’ antibody conjugation kit (Conjugation kit, #7000009; Akoya) while human CD68 (clone #KP1, catalogue #4550113, Akoya) and human PanCK (clone AE-1/AE-3, catalogue #4150020, Akoya) were obtained directly pre-conjugated to oligo barcodes from Akoya Biosciences. Complementary oligo-conjugated fluorophore reporters were obtained from Akoya Biosciences. Tissue multiplexed immunofluorescence staining and image acquisition were performed according to Akoya Phenocycler-Fusion user guide (PD-000011 Rev. A., Akoya). OME-TIFF files were generated and processed for image analysis. Analysis of the multiplexed immunofluorescence images (generated from Akoya Phenocycler-Fusion platform) was performed using Visiopharm (version 2023.09.3.15043 × 64) on the entire tissue area. Briefly, cell segmentation (including both nuclear and cytoplasmic segmentation) was first performed using Visiopharm’s “Cell Detection, AI (Fluorescence)” (version 2023.09.3.15043 × 64) with its default parameters. After cell segmentation, Visiopharm’s “Phenoplex Guided Workflow” was used. DAPI (nucleus), CD68 (cell body) and STAB1 (cell body) variables were selected and manually thresholded to define positive and negative cells for each marker and generate a co-occurrence matrix. Macrophages were defined as [DAPI + , CD68 + ] while STAB1+ macrophages were defined as [DAPI + , CD68 + , STAB1 + ]. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. The authors are greatly thankful to the Papworth Hospital Research Tissue Bank for providing samples with data, and in particular to D. Rassl. The authors would like to thank L. Campos for the annotation of tumour histologies; A.M. Ranzoni, B. Myers and E. Panada for sample collection and processing; M. Nelson for computational support with initial clustering of scRNA-Seq and application of cell2location; Alessandro Di Tullio, GSK for insightful discussions; Cancer Research UK Cambridge Institute (CRUK CI) (Grant # CTRQQR-2021\100012) Genomics Core Facility for library preparation and sequencing services; Wellcome Sanger Institute (WSI) DNA pipelines for their contribution to sequencing the data; S. Leonard from New Pipeline Group (NPG) for pre-processing of sequencing data; the Cambridge NIHR BRC Cell Phenotyping Hub for support with cell sorting. We thank R. Möller, P. Rainer, and U. Tiemann for critically reading the manuscript. This study was conceived and funded by Open Targets (OTAR2060, A.C.); Core support grants from the Wellcome Trust and Wellcome Sanger Institute and both Wellcome and the MRC to the Wellcome Trust-Medical Research Council Cambridge Stem Cell Institute (203151/Z/16/Z, A.C.); European Research Council (CONTEXT 101043559, A.C.); Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them. Nature Communications thanks Charles Powell and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. The scRNA-seq and Visium datasets generated in this study are publicly available at BioStudies (https://www.ebi.ac.uk/biostudies/) with accession numbers E-MTAB-13526 and E-MTAB-13530, respectively. The remaining data are available within the Article, Supplementary Information or Source Data file Source data are provided with this paper. The scripts used for all the analyses and to produce all the figures in the manuscript are available at https://gitlab.com/cvejic-group/lung and https://github.com/sdentro/copykat_pipeline. The authors declare no competing interests. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Marco De Zuani, Haoliang Xue. The scRNA-seq and Visium datasets generated in this study are publicly available at BioStudies (https://www.ebi.ac.uk/biostudies/) with accession numbers E-MTAB-13526 and E-MTAB-13530, respectively. The remaining data are available within the Article, Supplementary Information or Source Data file Source data are provided with this paper. The scripts used for all the analyses and to produce all the figures in the manuscript are available at https://gitlab.com/cvejic-group/lung and https://github.com/sdentro/copykat_pipeline.
PMC10968586
Use of Oleuropein and Hydroxytyrosol for Cancer Prevention and Treatment: Considerations about How Bioavailability and Metabolism Impact Their Adoption in Clinical Routine
The fact that the Mediterranean diet could represent a source of natural compounds with cancer-preventive and therapeutic activity has been the object of great interest, especially with regard to the mechanisms of action of polyphenols found in olive oil and olive leaves. Secoiridoid oleuropein (OLE) and its derivative hydroxytyrosol (3,4-dihydroxyphenylethanol, HT) have demonstrated anti-proliferative properties against a variety of tumors and hematological malignancies both in vivo and in vitro, with measurable effects on cellular redox status, metabolism, and transcriptional activity. With this review, we aim to summarize the most up-to-date information on the potential use of OLE and HT for cancer treatment, making important considerations about OLE and HT bioavailability, OLE- and HT-mediated effects on drug metabolism, and OLE and HT dual activity as both pro- and antioxidants, likely hampering their use in clinical routine. Also, we focus on the details available on the effects of nutritionally relevant concentrations of OLE and HT on cell viability, redox homeostasis, and inflammation in order to evaluate if both compounds could be considered cancer-preventive agents or new potential chemotherapy drugs whenever their only source is represented by diet. Keywords: oleuropein, hydroxytyrosol, cancer, nutritionCancer insurgence and progression are complex processes, depending on the combination of unmodifiable genetic and modifiable environmental/lifestyle-related factors. With this premise, it sounds perfectly understandable that scientific evidence has corroborated the role of a healthy diet and dietary intervention as potentially beneficial approaches contributing to cancer prevention . Epidemiological and experimental evidence has confirmed that the so-called Mediterranean diet is a source of molecules that may mitigate cancer risk factors like chronic inflammation and redox imbalance, thus participating in the prevention of carcinogenesis in terms of loss of cell cycle regulation and proper immune modulation, as well as in the inhibition of angiogenesis and metastasis. Moreover, some of these natural compounds may have a cytotoxic effect, making them interesting alternatives to or candidates for integration into conventional therapeutic approaches . Among Mediterranean diet phenols, secoiridoid oleuropein (OLE) is the most abundant phenolic compound in Olea europaea L. tree leaves (OLE content up to 14–19% in olive leaves), followed by its degradation derivative hydroxytyrosol (3,4-dihydroxyphenylethanol, HT, 2.28 mg/g of olive leaf extract) (Figure 1) . Molecular structure of OLE and HT. As it can be appreciated from the image, OLE structure is characterized by the presence of an ester bond between elenolic acid and HT (thus, HT may be released as an OLE decomposition product). Molecular structures were designed with MolView Copyright© 2014–2023 Herman Bergwerf available at https://molview.org/ (accessed on 13 February 2024). OLE and HT are also found in the fruit of Olea europaea L. and in olive oil; thus, they are easily ingested as part of a routine diet, but they can also be obtained from other sources, e.g., olive mill wastewater . Both compounds have attracted attention for their accessibility, safe profile, powerful antioxidant and scavenging activity against reactive oxygen species (ROS), and controversial anti-inflammatory action. For more than two decades, OLE and HT (together or alone) have been the focus of intense research efforts in the context of infectious diseases and prevention/management of chronic non-communicable diseases, including cancer, with encouraging results from in vitro and in vivo models . On this basis, it would be difficult to understand the reasons behind the lack of systematic testing of OLE and HT as supplements to prevent the insurgence of cancer or support the management of hematological malignancies and solid tumors. This becomes clearer considering that, despite promising proof in the field, experimental evidence about OLE and HT bioavailability in humans and animals clearly demonstrates that OLE and HT act as cancer-preventive agents and cytotoxic drugs mainly at concentrations far from plasma levels reachable through nutrition, an aspect often interpreted as marginal that we discuss in detail in this review. Also, as explained in the following paragraphs, the complexity and diversity of molecular mechanisms resulting in net OLE and HT action has led to questions regarding the possibility that these compounds might even facilitate the expansion of neoplastic clones at nutritionally relevant concentrations. In this review, we discuss the available data on the use of OLE and HT as anti-cancer drugs and the feasibility of their application in the context of clinical routine now or in the near future. OLE and HT activity against solid tumor insurgence and development has been challenged in a large number of experimental models, both in vivo and in vitro. No single shared molecular mechanism and/or triggered cellular response seems implicated in OLE and HT cytotoxicity, resulting in an articulated frame that imposes a separate dissertation for every type of studied solid tumor. To facilitate critical interpretation, in each subsection, OLE and HT assayed doses are indicated, with the half-maximal inhibitory concentration (IC50) and the half-maximal effective concentration (EC50) reported as exact values, mean ± standard deviation (S.D.), or mean ± standard error of the mean (S.E.M.) whenever provided by the authors. A list of experimental models used to study OLE and HT cytotoxicity in cancer cells is reported in Table 1. List of experimental models used to study OLE- and HT-dependent cytotoxicity in cancer and hematological malignancies. OLE, oleuropein; HT, hydroxytyrosol; Ref., reference(s); [C], cytotoxic concentrations reported in the cited literature; HFD, high-fat diet; 0.02% HFD, HFD containing 0.02% OLE; 0.04% HFD, HFD containing 0.04% OLE; AOM, azoxymethane; DSS, dextran sulfate sodium; *, non-cytotoxic but cell viability-affecting effects; §, olive pulp extract with a 70% content of phenols, of which 70% is represented by HT. Malignant melanoma is a malignancy arising from the transformation of melanocytes, with increasing incidence worldwide. On the basis of the tissue where the primary lesion appears, four major subtypes can be distinguished: cutaneous melanoma (non-glabrous skin), acral melanoma (glabrous skin), mucosal melanoma (melanocytes in the mucosal tissues), and uveal melanoma (uveal tract of the eye). Among these major subtypes, it is possible to distinguish some particular variants: amelanotic/hypomelanotic melanoma, a subtype of cutaneous melanoma with low or absent melanin; desmoplastic melanoma, a spindle cell tumor exhibiting signs of dense scar-like fibrosis; spitzoid melanoma, sharing histopathological characteristics with Spitz nevi; acral lentiginous melanoma, with a lentiginous growth pattern . OLE seems to be effective in the prevention of skin carcinogenesis in vivo. Orally administered 10 mg/kg and 25 mg/kg OLE reduced skin carcinogenesis (expressed in terms of number of tumors per mouse) in UVB-irradiated albino hairless HOS: HR-1 mice . For 25 mg/kg OLE, this effect was associated with a persistent reduction in (I) the total volume of tumors per mouse, (II) the expression levels of invasion enzymes matrix metalloproteinase 2 (MMP2), pro-MMP9, and MMP9, (III) tissue angiogenesis marker vascular endothelial growth factor (VEGF) and cyclooxygenase-2 (COX-2) levels, and (IV) the percentage of skin Ki-67+ cells and platelet endothelial cell adhesion molecule-1 (PECAM-1 or CD31)+ areas . OLE also seems effective in the prevention of melanoma growth and metastasis. In an in vivo B16F10 (mouse melanoma cell line) allograft model of high-fat diet (HFD)-induced melanoma progression in C57BL/6N mice, HFD containing 0.02% and 0.04% OLE reduced HFD-driven tumor growth and lymph node metastasis, with a mechanism involving (I) inhibition of cell proliferation, as indicated by the reduction in the percentage of cells positive for proliferation markers Ki67, Cyclin D1, and cyclin-dependent kinase 4 (CDK4) cells, (II) suppression of angiogenesis (reduction in CD31, VE-cadherin expression, VEGF-A, VEGF-C, VEGF-D, VEGF receptor 2 -VEGFR2- and VEGFR3), and (III) inhibition of lymphangiogenesis, as proved by staining for lymphatic vessel endothelial hyaluronan receptor (LYVE-1). According to further in vitro experimental assays, OLE anti-angiogenetic and lymphangiogenetic action relies on the inhibition of lipid and M2-macrophage accumulation . In vitro results for OLE are conflicting. Incubation of human amelanotic melanoma cell line C32 with 100 μM, 400 μM, and 1000 μM OLE for 72 h promoted cell viability . On the contrary, incubation of human melanoma cell lines A375, WM266-4, and M21 with 250 μM, 500 μM, and 800 μM OLE for 72 h produced a dose-dependent decrease in cell viability. As deepened in A375 cells, an increase in OLE effectiveness with time may be detected. OLE induced inhibition of cell viability after 24 h at a concentration of 800 μM, whereas after 48 and 72 h, a decrease in the percentage of viable cells became significant with a concentration as low as 250 μM . After 48 h treatment, 500 μM OLE increased the rate of apoptosis in A375 cells. In addition, 24 h incubation with 250 μM OLE was sufficient to reduce invasiveness of A375 cells, while 48 h incubation with the same OLE concentration reduced the phosphorylation of pro-survival kinase Akt . As regards in vitro evidence for HT, the effects on cell growth seem to be cell line-dependent. Treatment of C32 cells with 100 μM HT for 72 h increased cell viability, which was instead reduced by incubation with 400 μM and 1000 μM HT . Incubation of human melanoma cell line A375 with 100 μM and 200 μM HT significantly diminished cell viability after 48 h. On the contrary, 48 h treatment of melanoma cell line MNT1 with 100 μM and 200 μM HT had no significant effect on cell viability. The authors attributed this dissimilarity between the mentioned cell lines to differences in the active metabolic pathways. Expression analysis revealed a significant transcriptional upregulation of lactate dehydrogenase B (LDHB) and LDHC (accounting for lactate conversion into pyruvate) and glutamine synthetase (GLUL) in MNT1 cells in comparison with the A375 cell line, while sodium-coupled neutral amino acid transporter 1 (SNAT1) and SNAT2 (involved in glutamine transport within the cell), monocarboxylate transporter 4 (MCT4, accounting for lactate export), glycolytic enzyme glucose-6-phosphate dehydrogenase (G6PD), and excitatory amino acid transporter 3 (EEAT3) were downregulated in MNT1 cells vs. A375 . In a further report, treatment of human melanoma cell lines A375, HT-144, and M74 with 50–250 μM HT produced a dose- and time-dependent decrease in cell viability after 24, 48, and 72 h . A detailed analysis on A375 cells (treated with 250 μM, 375 μM, and 500 μM HT) and HT-144 cells (incubated with 250 μM, 350 μM, and 450 μM HT) performed for 24 and 48 h revealed an increase in the rate of apoptosis in both cell lines, with a dose- and time-dependent increase in tumor suppressor p53 and reduction in growth-promoting kinase Akt protein levels . The activation of apoptosis pathway was further confirmed by an increase in apoptosis markers pro-activated and cleaved (activated) forms of caspase-3, a dose- and time-dependent increase in cleavage (activation) of poly ADP-ribose polymerase 1 (PARP-1), and a dose-dependent increase in the phosphorylation of histone H2AX (γH2AX) . HT-mediated induction of apoptosis was related to ROS accumulation in both cell lines at the indicated HT concentrations, which was detected after 24 and 48 h . Thyroid cancer is a category of neoplastic lesions with a highly variable degree of aggressiveness, arising from parafollicular C cells (resulting in medullary thyroid cancers) and follicular thyroid cells (producing follicular thyroid cancer, papillary thyroid cancer, poorly differentiated thyroid cancer, Hürthle cell cancers, and anaplastic thyroid cancer) . In vitro, treatment of human papillary thyroid carcinoma cell line TPC-1 and poorly differentiated thyroid gland carcinoma cell line BCPAP with 50–100 μM OLE for 48 h produced a significant reduction in cell viability attributable to S phase and G2/M phase cell cycle block, respectively. In both cell lines, 50–100 μM OLE exerted an antioxidant activity against hydrogen peroxide (H2O2)-induced perturbation of ROS homeostasis. Also, 100 μM OLE caused a short-lasting (30 to 60 min) reduction in phosphorylated forms pro-survival kinases ERK (phospho-ERK) and Akt (phospho-Akt) in TPC-1 and BCPAP cells . Incubation of papillary thyroid cancer cell lines TPC-1 and FB-2 with 324–973 μM HT decreased cell viability after 24 and 48 h in a dose-dependent manner. A stronger action was exerted on follicular thyroid cancer cell line WRO, whose cell viability was reduced even at lower doses of HT after 24 h (162 μM) and 48 h (65 μM) treatment . After 24 h incubation, 324 μM HT elicited an increase in the percentage of apoptotic and necrotic cells in all the three mentioned cell lines (with a concomitant downregulation of pro-proliferative cyclin D1 and upregulation of tumor suppressor p21 at both mRNA and protein levels), increased protein level of tumor suppressor p53, and activated the intrinsic pathway of apoptosis, as corroborated by the increase in cleaved PARP and cleaved caspase-3 levels, Bcl-2-associated agonist of cell death (Bad) and caspase-9 protein levels, and the release of mitochondrial cytochrome c . Currently, lung cancer represents the most commonly diagnosed cancer and the main cause of cancer-related deaths worldwide, including small cell carcinoma and more common non-small cell carcinoma . In adenocarcinomic human alveolar epithelial cells A549 (a model for non-small cell lung cancer), 50 μM and 150 μM OLE-induced apoptosis after 24 h incubation was mediated by the decrement in Bcl-2 and Bcl-XL anti-apoptotic proteins flanked by the increase in (I) mitochondrial-located pro-apoptotic protein Bax, (II) cytochrome c release from the mitochondria, (III) activation of apoptosome component apoptotic protease activating factor-1 (Apaf-1), (IV) activation of caspase-3, and (V) mitochondrial methylglyoxal detoxicating enzyme Glo2 (mGlo2), which physically interacted with Bax . Consistent with these data, incubation of non-small cell lung cancer cell line H1299 with 50–200 μM OLE for 24 h elicited a dose-dependent G2/M phase cell cycle block and apoptosis, with effects on Bcl-2, Bax, cytochrome c, and caspase-3 that were similar to those documented for A549 cells . However, the underlying molecular mechanism ruling OLE activity differed between the two cell lines. The effects elicited by 150 μM OLE on mGlo2 levels in A549 cells were strictly dependent on OLE-induced increase in superoxide dismutase 2 (SOD2) detoxicating action against superoxide (O2), and on the inhibition of the Akt signalling pathway. This was not surprising, since in A549 cells, O2 supports Akt activations, promoting cell survival . In H1299 cells, the observed apoptosis was instead determined by OLE-induced phosphorylation of p38 mitogen-activated protein kinase (MAPK), accompanied by an increased rate of phosphorylation of activating transcription factor-2 (ATF-2), involved in cell cycle regulation, as documented in both tumorigenesis and cell death, and the upregulation of genes ruling cell metabolism and apoptosis . In A549 cell line, HT showed an increase in effectiveness of its anti-proliferative activity with respect to time, with the IC50 values changing from 230.60 μM to 149.36 μM in 72 h . This piece of data was confirmed in another report. In fact, for A549 cells, mean IC50 ± S.E.M. = 147.0 ± 16.5 μM was reported after 48 h incubation . Malignant pleural mesothelioma arises from mesothelial cells and is characterized by a pronounced aggressiveness and poor prognosis. On the basis of histological features, mesothelioma is classified as epithelioid, sarcomatoid, or biphasic, with epithelioid type offering the best median survival . Pleural epithelioid mesothelioma REN cell line was utilized to demonstrate that OLE exhibited a cytotoxic activity (IC50 = 25 μg/mL, ≈46 μM), and that both OLE and HT (10–100 μM) mobilized extracellular Ca in a dose-dependent manner . Breast cancer is the most frequent malignant tumor in women worldwide, with a constantly rising incidence . Treatment of breast cancer is based on the molecular subtype, a classification that in the first instance takes into account the immunohistochemically assayed expression of hormone receptors estrogen receptor (ER) and progesterone receptor (PR), and gene amplification or overexpression of human epidermal growth factor receptor 2 (HER2). Triple negative breast cancer identifies a category of tumors lacking the three mentioned receptors . In an in vivo model of tumor xenograft (triple negative MDA-MB-231 cell line) in BALB/c OlaHsd-foxn1 mice, animals receiving 50 mg/kg OLE for 4 weeks showed a decrease in tumor size, together with a reduction in actors involved in cell growth/proliferation: transcription factor NF-κB and cyclin D1 . Instead, levels of tumor suppressor p21 increased after OLE injection. The effects elicited by OLE were accompanied by the induction of apoptosis, as demonstrated by caspase-3 activation, increase in Bax levels, and reduction in Bcl-2 protein expression . In vitro, OLE action appears to be independent of HER2 gene amplification/overexpression and hormone receptor status. Using MCF-7 cell line, which is devoid of HER-2 overexpression , 200 μg/mL (≈370 μM) OLE for 48 h showed a specific cytotoxic effect on MCF-7 cell line, leaving the non-cancerous cell line MCF-10A unharmed. The observed effect on cell viability was accompanied by the upregulation of the expression of Prdx1-Prdx6, encoding for antioxidant and chaperone proteins peredoxins . The induction of apoptosis triggered by the same concentration of OLE in MCF-7 cells was confirmed in another study after 12 h incubation . Treatment of MCF-7 cells with 200 μM and 400 μM OLE for 24 h produced a notable decrease in cell viability. Specifically, at a concentration of 400 μM, OLE increased MCF-7 cell death by apoptosis induction . OLE effects might depend on the upregulation of p53 and Bax, and downregulation of Bcl-2, as demonstrated by incubating MCF-7 cells with 200 μM OLE for 48 h . Other studies using a very large concentration of OLE (600 μg/mL, ≈1100 μM) shed some light on a different mechanism ruling OLE-dependent induction of apoptosis. In fact, treatment of MCF-7 cells with the mentioned OLE concentration for 48 and 72 h reduced histone deacetylase 2 (HDAC2) and HDAC3 gene transcription in a time-dependent manner , and downregulated oncomiRs miR-21 and miR-155 . However, OLE-mediated effects on ER-positive cell viability may also be appreciated at lower doses. In fact, 30 μM and 50 μM OLE were able to reduce MCF-7 cell viability after 48 h, with no significant increase in the apoptotic rate . Moreover, 150 μM OLE reduced cell viability of ER-positive MCF-7 and T47D cells after 24 h . Further experiments on MCF-7 cell line demonstrated a dose-dependent OLE-mediated antiproliferative effect on 17β-estradiol (E2)-induced cell growth when OLE concentration was used in the range of 10–75 μM. Instead, concentrations ≥ 100 μM were found to be toxic . As regards a possible anti-estrogenic action, 10 μM OLE had no effect on estrogen receptor α (ERα) basal activation, and 10–75 μM OLE had irrelevant activity on E2-induced ERα activation and E2-modulated ERα expression, but reduced E2-induced ERK1/2 phosphorylation . In ER-positive cells, invasiveness may be suppressed by OLE-mediated induction of autophagy. In fact, treatment of MCF-7 and T47D cell lines with 100 μM OLE reversed hepatocyte growth factor (HGF)- and 3-methyladenine (3-MA, an autophagy inhibitor)-induced migration, upregulating LC3II/LC3I and Beclin1, while downregulating p62 . OLE was also able to reduce proliferation in triple negative breast cancer cell lines MDA-MB-231 and MDA-MB-468 after 48 h incubation (IC50 = 500 μM), with RNA sequencing revealing alterations of the expression profile of genes involved in cell death, apoptosis, and response to stress . Other reports documented that a dose as low as 50 μM was able to reduce MDA-MB-231 cell viability after 72 h incubation and IC50 = 36.2 μM was estimated for 72 h treatment . A further attempt to deepen the mechanism of action in triple negative breast cancer showed that 12.5–100 μM OLE affected MDA-MB-231 cell viability in a dose- and time-dependent manner, reducing cellular migration and invasion capabilities, especially at doses ≥ 25 μM. OLE was able to induce dose-dependent apoptosis after 72 h incubation. A deeper analysis, performed using a dose of 100 μM, revealed caspase-3/7 activation after 48 and 72 h, and a reduction in NF-κB phospho-p65 nuclear localization after 12 h . In a breast cancer cell line identified by the authors only by the letters “MDA”, 200 μg/mL OLE was able to interfere with the metastatic process, causing a time-dependent increase in mRNA levels of MMP inhibitors TIMP metallopeptidase inhibitor 1 (TIMP1) and TIMP3, while TIMP4 showed no further increase after 48 h incubation. Simultaneously, the same concentration of OLE triggered a time-dependent decline in MMP2 and MMP9 mRNA levels . As regards HT activity in vivo, in a model of dimethylbenz[α]anthracene-induced mammary tumors in Sprague–Dawley rats treated with HT (0.5 mg/kg, 5 days/week for 6 weeks) reduced tumor growth, modulating the expression of genes involved in apoptosis and cell proliferation/survival . The relation between HT cytotoxic action in vitro and HER2 overexpression/hormone receptor status is less clear. Incubation of MCF-7 cell line with 50 μg/mL HT (≈324 μM) for 12 h was sufficient to trigger apoptosis , despite the fact that this result was not uniformly reproduced. In other experimental settings involving MCF-7 cells, the anti-proliferative activity of HT became evident at higher as well as even lower concentrations. HT used in the range of 5–400 μM for 16 h had no effect on MCF-7 cell proliferation; it became inhibited only at 600 μM . Another report documented a dose- and time-dependent effect of HT on MCF-7 cell viability, with 250 μM decreasing the percentage of viable cells after 72 h treatment, and 400 μM HT reducing cell viability as a consequence of 48 and 72 h incubation . However, at sub-lethal concentration (200 μM), HT had an impact on both (I) oxidative stress response, augmenting mRNA and protein levels of transcription factor nuclear respiratory factor 2 (Nrf2) and upregulating the transcription of its targets glutathione S-transferase alpha 2 (GSTA2) and heme oxigenase-1 (HO-1), and (II) energy homeostasis, reducing mRNA level of mitochondria biogenesis regulator PPARγ coactivator-1α (PGC-1α), while increasing its protein level and downregulating expression (in terms of mRNA) of mitochondrial function regulators estrogen-related receptor α (ERRα) and deacetylase sirtuin 3 (SIRT3) . Other research records showed that a reduction in ER-positive cell viability may be obtained even at lower HT doses, as demonstrated by 72 h treatment with 50 μM and 100 μM HT, which significantly diminished the percentage of viable MCF-7 cells , and by 100 μM and 150 μM HT-induced reduction in MCF-7 and T47D cell viability after 24 h . Incubating MCF-7 cells with 10–75 μM HT showed an inhibitory effect on E2-induced cell growth, with HT becoming toxic at concentrations ≥ 100 μM. ERα basal activation was induced by 10 μM HT, but (similarly to OLE) when HT was used at concentrations of 10–75 μM, it exhibited no effect on E2-induced ERα activation and E2-modulated ERα expression, while reducing levels of phospho-ERK1/2 (pERK1/2) . As regards invasiveness of ER-positive cells, experimental evidence obtained in MCF-7 and T47D cells confirmed that 50 μM HT elicited effects similar to those recorded for OLE on HGF- and 3-MA-induced cell migration . In human triple negative breast cancer cells MDA-MB-231, HT seemed to lose effectiveness with respect to time, with IC50 values changing from 107.17 μM to 183.65 μM in 72 h . For the same cell line, higher IC50 values (230 μM) were reported for 72 h treatment . On the contrary, another study demonstrated that 100 μM HT promoted MDA-MB-231 cell viability during the first 24 h, becoming ineffective after 48 and 72 h, whereas 250 μM HT became able to reduce cell viability after 72 h treatment, and 400 μM HT significantly diminished cell viability at all the three assayed time points (24, 48, and 72 h) . This piece of data is openly conflicting with reports documenting a loss of MDA-MB-231 viability after 72 h incubation at all tested HT doses (10–100 μM) . The reduction in triple negative breast cell viability may depend on HT acting as a copper chelator, thus perturbating copper homeostasis, as demonstrated in MDA-MB-231 cells by 100 μM HT-mediated increase in the copper chaperone for superoxide dismutase (CCS) after 48 h, and reduction in the subunit II of the complex IV of the mitochondrial respiratory chain cytochrome c oxidase (CcO) after 72 h. Also, HT altered epithelial and mesenchymal markers, and reduced MDA-MB-231 aggressiveness and migration by diminishing copper-dependent Akt phosphorylation. Similar conclusions were drawn for MDA-MB-468 cells . Other triple negative breast cancer cell lines were identified as particularly resistant towards HT-mediated cytotoxicity, e.g., SUM159 (IC50 = 300 μM) . In naturally HER2-overexpressing human breast cancer cells SKBR3, HT concentrations up to 100 μM failed to elicit a cytotoxic response after 5 days, and only weakly interfered with cell proliferation, but reduced HER2 protein expression after 48 h. However, in engineered HER2-overexpressing MCF-7 cells, 100 μM HT efficaciously triggered apoptosis, diminished cell proliferation, and downregulated HER2 . Other authors suggest that in vitro cytotoxic action of OLE and HT may depend on cell density in culture, hypoxia, and ROS homeostasis. Han et al. observed that incubation with 200 μg/mL OLE and 50 μg/mL HT for 12 h exhibited the most efficient inhibition of cell growth when MCF-7 seeded cell number did not exceed 2 × 10 cell/well in a 96-well plate . Hypoxia increased HT cytotoxicity in MCF-7 cell line, an effect that became evident at 400 μM (vs. 600 μM in normoxic conditions). In hypoxic conditions, 200 μM HT was also able to exert the same actions on oxidative stress response and energy homeostasis as those reported above for normoxic MCF-7 cells . Moreover, in hypoxic MCF-7 cells, 75–200 μM HT reduced the amount of PARP-1, a DNA-binding protein involved in oxidative stress response, but inhibition of PARP-1 activity was achieved only with 200 μM HT . The same concentration of HT accounted for the reduction in the phosphorylated form of kinase mammalian target of rapamycin (mTOR), which in turn produced a reduction in hypoxia inducible factor-1α (HIF-1α), one of the two subunits of the heterodimeric transcription factor HIF-1, ruling the cellular response to hypoxia . More importantly, 200 μM HT upregulated the transcription of angiogenic factors adrenomedullin (AM) and VEGF with an HIF-1α-independent mechanism . Treatment of MDA-MB-231 cells with 200 μg/mL olive leaf extract containing 87% OLE for 24 h induced S phase cell cycle arrest and apoptosis by a mechanism that was strongly dependent on OLE-mediated ROS accumulation and downregulation of protein expression of catalase (CAT) and SOD2 . Treatment of MCF-7 cells and another breast cancer cell line generically indicated by the authors simply as “MDA” with 25–100 μM HT for 72 h reduced cell viability in a dose-dependent fashion (mean IC50 ± S.D. = 52 ± 4 μM for of MDA and 58 ± 8 μM for MCF-7), mainly in culture conditions favoring H2O2 accumulation . HT may also have a role in changing the tumor microenvironment. Aged quiescent normal human fibroblasts were able to stimulate MDA-MB-231 proliferation by synthetizing and secreting large amounts of chemokine C-C motif ligand 5 (CCL5), which in turn activated pro-proliferative ERK1/2 and cyclin D1 signalling pattern. Treatment of 100 μM and 200 μM HT prevented CCL5 accumulation in aged quiescent normal human fibroblasts, limiting the proliferation of MDA-MB-231 cells. A similar growth inhibition was also obtained for MCF-7 cells . Hepatocellular carcinoma (HCC) or hepatoma is the most frequent primary liver tumor and the sixth most common cancer worldwide, with a very modest survival rate and a complex management . Results on OLE-mediated effects on hepatoma cells are conflicting. In vitro, 20–80 μM OLE reduced hepatoma Huh7 and HepG2 cell line viability in a dose-dependent manner in 24 h . As determined in HepG2 cells, OLE effects on cell viability are linked to apoptosis induction, mediated by an increase in Bax and cleaved (i.e., activated) caspase-9, caspase-8, caspase-3, and PARP-1 levels, and a decrease in phospho-Akt and Bcl-2 protein levels. After treating HepG2 cells with 50 μM OLE for 24 h, the measured reduction in cell viability seemed related to OLE-mediated accumulation of ROS . However, one report documented the absence of effects on cell viability after treating HepG2 cells with 10–10 M OLE for 24 h , and another study described no effect on HepG2 cell viability after 48 h incubation with 15–200 μM OLE . The dose-dependent cytotoxic effect of HT on HCC was demonstrated in human hepatoma HepG2 and Hep3B cell lines, with a time-dependent increase in effectiveness. In fact, 80–200 μM HT significantly affected cell metabolic activity after 48 h, whereas a concentration as low as 30 μM was sufficient to reduce the percentage of viable cells in both cell lines after 72 h of incubation . These effects might rely on inhibition of lipogenesis. In fact, in both cell lines’ fatty acid synthase (FASN) activity was consistently inhibited by 200 μM HT (with significant enzymatic inhibition starting with 30 μM and 80 μM HT for HepG2 and Hep3B, respectively), whereas farnesyl diphosphate synthase (FPPS) lipogenic enzyme activity was inhibited only in HepG2 cells . A further report confirmed the dose- and time-dependent inhibition of cell viability of human HCC cell lines HepG2, Hep3B, and Huh-7, and ascitic fluid-derived cell line SK-HEP-1 for HT doses up to 400 μM, with G2/M cell cycle arrest and inactivation of cell growth-promoting Akt and NF-κB pathways . Instead, the role of redox homeostasis in HT-dependent cytotoxicity is less clear in the context of HCC. Cell antioxidant capacities were promoted by 48 h treatment with 80–200 μM HT in HepG2 cells; on the contrary, the same effect was elicited by only 48 h treatment with 30 μM HT in Hep3B cells . However, HT triggers other molecular mechanisms that may influence HCC cell behavior besides affecting cell viability, at least in vitro. Treatment of human hepatoma HepG2 cells with 50–200 μM HT determined an increase in intracellular ionized calcium levels ([Ca]i) with the contribution of both Ca influxes and mobilization of endoplasmic reticulum depots. Anyway, Ca dynamics seemed to be unrelated to HT-dependent reduction in cell viability , but they might be implicated in changes in the cell secretome , although definitive proofs are still missing. Cholangiocarcinoma is the term used to identify rare heterogeneous cancers arising in intrahepatic and extrahepatic bile ducts . Instead, gallbladder cancer is the most common biliary tract cancer, characterized by a very unfavorable prognosis . Treatment of cholangiocarcinoma cell line KMBC and TFK-1, and gallbladder carcinoma GBC-SD cells with 25–200 μM HT for 24, 48, and 72 h reduced cell proliferation in a time- and dose-dependent manner. A time- (24–72 h) and dose (75–150 μM)-dependent reduction in ERK1/2 phosphorylation was detected in all the three mentioned cell lines. The 72 h incubation of KMBC, TFK-1, and GBC-SD cells with 75 μM and 150 μM HT caused G2/M phase cell cycle arrest, induced apoptosis, augmented the levels of cleaved PARP, Bax, cleaved caspase-3, and cleaved caspase-9, and reduced Bcl-2 levels. Also, treatment of TFK-1 cell xenograft tumor grown in nude BALB/c mice with peritoneal injection of 500 mg/kg/day HT for 3 weeks reduced tumor growth in vivo . Colorectal cancer, including tumors of the colon and/or rectum, is the third most common cancer and the second most common cancer-related cause of death worldwide. It is characterized by high rates of acquired multidrug resistance, leading to chemotherapy failure, relapse, and development of lethal disease . The classical Dukes’ classification distinguishes between “type A” and “type B” tumors, according to the absence or presence of tumor infiltration in extra-rectal tissues, respectively, and designates as “type C” those tumors with regional lymph node metastasis, and as “type C1” and “type C2” those malignancies exhibiting involvement of more distant lymph nodes . In an in vivo model of azoxymethane (AOM)/dextran sulfate sodium (DSS)-induced colorectal cancer in C57BL/6 mice, 50 and 100 mg/kg OLE reduced the incidence of colonic neoplasms, and levels of proliferation regulators NF-κB subunit p65, phosphorylated form of signal transducer and activator of transcription 3 (STAT3), and phospho-Akt, while increasing Bax protein expression. In addition, 100 mg/kg OLE reduced cell proliferation (as confirmed by reduced expression of proliferation marker Ki67) . Colorectal carcinoma cell lines may exhibit different sensitivity towards OLE-mediated effects on cell growth. In fact, treatment of colon carcinoma RKO cell line with 20–80 μM OLE diminished cell viability in a dose-dependent fashion . Similarly, 10–100 μM OLE reduced viability of Dukes’ type C colorectal adenocarcinoma SW620 cell line after 72 h incubation, becoming able to increase the rate of apoptosis at the highest concentration tested . On the contrary, 10–50 μM OLE had no effect on the viability of colon adenocarcinoma HT29 cell line, but induced apoptosis only at a concentration of 100 μM . Incubation with 400 μM and 800 μM OLE reduced HT-29 cell viability at all three tested time points (24, 48, and 72 h), inducing G0/G1 phase cell cycle arrest after 24 h treatment. However, only 800 μM OLE was able to promote apoptosis in HT-29 cells after 24 h incubation . Treatment with 400 μM and 800 μM OLE produced a reduction in protein expression of HIF-1α, which was detected after 2 h and persisted up to 48 h. Instead, p53 protein expression was increased only by 48 h incubation with 800 μM OLE . In vitro, the effects of low or nutritionally relevant HT doses are poorly explored, and HT is suspected to interfere with colorectal cancer insurgence acting as a methylation pattern modifying agent. In fact, incubation of human colorectal adenocarcinoma Caco2 cell line with 10 μM HT increased DNA methylation, leading to the repression of the crucial colorectal cancer promoter endothelin receptor type A (EDNRA) . Experimental evidence supports a pro-oxidant action of low doses of HT in colorectal cancer cells. Treatment of human colorectal carcinoma HCT116 cells and human Dukes’ type C colorectal adenocarcinoma SW620 cells with 5–20 μM HT for 24 h determined a raise in the apoptotic rates of both cell lines, ruled by a dose-dependent ROS accumulation detected after 4 h of incubation, which in turn was caused by HT direct inhibition of thioredoxin (Trx) reductase 1 (TrxR1), a key player in redox homeostasis . At higher concentrations, effects triggered by HT appear different and sometimes poorly reproducible. Treatment of Caco2 and HT-29 cell lines with 100 μM HT for 8 h led to G1 phase cell cycle arrest, whereas extending the treatment for 48 h caused activation of caspase-3. Also, incubation of both cell lines with 150 μM for 8 h produced an increase in apoptosis rate, but HT contributed to the increase in the percentage of necrotic cells only in HT-29 cells . After 24 and 48 h incubation, 6 ppm (≈39 μM) HT was not sufficient to decrease HT-29 cell viability, but after 72 h treatment, there was a significant cytotoxic effect, with IC50 = 12 ppm (≈78 μM). At this concentration, HT elicited an increase in the transcription of tumor suppressor genes encoding for p21 and p27, while decreasing the expression of CCND1 (encoding for cyclin D1) . After 24 h, 100 μM HT showed no effect on cell viability of HT-29 cell line. However, 72 h incubation with 100 μM HT efficaciously reduced cell viability of and triggered apoptosis in HT-29 cells . On the contrary, another report demonstrated a dose-dependent increase in the apoptotic rates of HT-29, HCT-116, and LoVo cells as well as Dukes’ type B colorectal carcinoma cell line SW480 for 24 and 48 h treatment with 100–400 μM HT . Further evidence supports HT-dependent reduction in cell viability of and induction of apoptosis in HT-29 cells also at higher concentrations (600–800 μM) , and one study documented IC50 = 750 μM for HT29 and HT29–19A cell lines . A more detailed analysis revealed that treatment of HT-29 cells with 400 μM HT for 24 h produced S and G2/M phase cell cycle arrest and induced apoptosis, flanked by apoptosis ruling events: loss of mitochondrial potential, decrease in anti-apoptotic Bcl-2 protein level, reduction in phosphorylation of Bad, increase in pro-apoptotic Bax and Bak, cytochrome c release, and activation of caspase-3 . A similar S phase cell cycle arrest was also observed for 800 μM HT, and induction of apoptosis also took place after 24 h incubation of HT-29 cells with 600 μM and 800 μM HT . A 16 h incubation with 400 μM HT triggered endoplasmic reticulum stress in HT-29 cells, with activation of unfolded protein response, as indicated by the increase in the spliced form of X-box binding protein 1 (XBP-1) mRNA, the upregulation of chaperone 78 kDa glucose-regulated protein (GRP78/Bip) at both mRNA and protein levels, the transient increase (after 2 h) in phosphorylation of PKR-like ER-associated kinase (PERK) and translation initiation factor-2 (eIF2a) (accounting for inhibition of protein synthesis), and time-dependent increase in CHOP protein levels (responsible for ROS production and Bcl-2 downregulation) and NADPH oxidase 4 (NOX4) . In the same cellular model, apoptosis seemed regulated by HT-mediated increase in the phosphorylated form of c-jun N-terminal kinase (JNK), which is a central player in establishing cell faith (apoptosis vs. proliferation) . Accordingly, the phosphorylated form of JNK target, c-jun, was also increased in HT-29 cells as a response to treatment with HT. Both JNK and c-jun phosphorylations seem to reach a peak after 30 min and 4 h of incubation. In addition, 400 μM HT induced activator protein 1 (AP-1) transcriptional activity (responsible for the expression of cell cycle regulator genes) , promoted a dynamic phosphorylation of ERK1/2 (with two peaks of phosphorylation—after 30 min and 2 h of incubation—both followed by a reduction in phospho-ERK1/2 levels), and abolished Akt phosphorylation. All HT-dependent effects producing apoptosis as a result seemed to be mediated by protein phosphatase 2A (PP2A) . Moreover, 400 μM HT reduced HIF-1α protein levels and augmented the protein levels of proliferation and inflammation modulator peroxisome proliferator-activated receptor γ (PPARγ) after 48 h incubation; 400 μM HT also elicited an increase in p53 levels starting after 16 h and lasting for 48 h . Similarly, 800 μM HT had analogous effects on HIF-1α and PPARγ protein expression starting after 24 h of incubation, but did not change p53 levels . Finally, treatment of HT-29 cells with 200 μM HT caused mobilization of endoplasmic reticulum Ca depots, but the contribution of this ionic species to HT-induced apoptosis/effects was not explained . In SW620 cells, HT reduced cell viability at all tested concentrations (10–100 μM), with a mechanism related to a decrease in FASN transcription and activity after 72 h exposure. A detailed analysis of the cell cycle revealed that 10–50 μM HT produced S phase cell cycle arrest, with apoptosis induction when HT was used at a concentration of 100 μM . Treatment of HCT116 cell line with 100–300 μM HT and LoVo cell line with 100–400 HT μM for 72 h led to a dose-dependent decrease in cell viability, with IC50 calculated as 92.83 μM and 140.8 μM for HCT116 and LoVo cells, respectively. Cell cycle analysis revealed that 72 h incubation of HCT116 and LoVo cells with 0.0154 mg/mL (≈100 μM) HT and 0.0231 mg/mL (≈150 μM) HT, respectively, caused G2/M cell cycle arrest and promoted apoptosis . As regards the role of ROS in high-dose HT-mediated effects, a report documented a dose-dependent reduction in cell viability of SW480 (mean IC50 ± S.D. = 82 ± 10) and HCT116 (mean IC50 ± S.D. = 55 ± 7) cells treated with 50–100 μM for 72 h, in conditions favoring the accumulation of extracellular H2O2 . Pancreatic cancer, or pancreatic ductal adenocarcinoma, arises from the malignant transformation of pancreatic ductal cells, and represents the third leading cause of cancer-related death worldwide, with extremely low 5-year and overall survival rates . Up to now, the study of OLE and HT effects in pancreatic cancer in vivo has been limited. In an in vivo orthotopic model of pancreatic cancer, obtained by injecting mouse pancreatic cancer cells Panc02 in C57BL/6 mice, proved that 200 mg/kg HT for 10 days suppressed tumor growth and proliferation. These results seem to depend on the modulation of the tumor microenvironment, since HT was able to reduce the accumulation of myeloid-derived suppressor cells in lymphoid organs, bone marrow, and tumor tissues . Results in vitro underline the diverse sensitivity that cell lines exhibit against OLE- and HT-mediated effects on cell growth. Neither OLE nor HT exhibited a cytotoxic effect on pancreatic ductal adenocarcinoma cell lines BxPC-3 and CFPAC-1 for doses up to 300 μM, but both were able to decrease cell viability of pancreatic ductal adenocarcinoma cell line MIA PaCa-2 (IC50 = 150.1 μM for OLE and 75.1 for HT) after 72 h incubation. Both 200 μM OLE and 100 μM HT produced G2 phase cell cycle arrest after 24 h treatment. The 48 h incubation with the same concentrations of OLE and HT led to caspase-3-mediated apoptosis. The observed effects may be dependent on OLE- and HT-mediated upregulation of AP-1 components c-jun and fos . Similarly, treatment of pancreatic ductal adenocarcinoma cell line PANC-1 with 10 μM, 32 μM, 100 μM, and 320 μM HT for 24, 48, and 72 h reduced cell viability in a dose- and time-dependent fashion; doses as low as 19 μM were able to induce apoptosis (as detected after 24 h), with caspase-9 and Bax upregulation, and MMP-2 and MMP-9 downregulation at mRNA levels after 72 h incubation . Treatment of mouse pancreatic cancer cell line Panc02 with 50 μM, 150 μM, and 200 μM HT for 48 h elicited a dose-dependent inhibition of cell proliferation and induction of apoptosis. Further experimental evidence revealed that 100 μM HT inhibited the expression of phospho-STAT3 and Cyclin D1 after 24 h . Cervical cancer is the fourth most common cancer among women worldwide, and the protagonist of an internationally shared effort towards prevention. Cervical cancer cases may be classified into two major histological subtypes: squamous cell carcinoma and adenocarcinoma . In human cervical adenocarcinoma HeLa cells, 50–100 μM OLE promoted the accumulation of S phase cells with only a modest increase in the percentage of sub-G1 cells, whereas 200 μM OLE was able to induce G2/M cell cycle arrest and apoptosis through the augmented expression of pro-apoptotic Bax protein, the reduction in anti-apoptotic Bcl-2, the release of mitochondrial cytochrome c in the cytosol, and caspase-9 mediated activation of caspase-3. These effects were connected to (I) an increase in the phosphorylated form of apoptosis regulator JNK, which in turn ruled an increase in the phosphorylated forms of cell cycle controller activator protein 1 (AP-1) components c-Jun and ATF-2, and (II) an increase in the levels of tumor suppressors p53 and p21 . Another report documented the cytotoxic effect of 100–1200 μg/mL (≈185–2220 μM) OLE in HeLa cells after 24 and 48 h treatments in a dose- and time-dependent manner, with 24 h IC50 = 600 μg/mL and 48 h IC50 = 300 μg/mL. A 50 μg/mL, 100 μg/mL, 200 μg/mL, 400 μg/mL, and 600 μg/mL OLE-dependent increase in the apoptotic cell rate was detected after 48 h incubation . Treatment with 300 μg/mL OLE downregulated invasion-promoting miR-181b-3p, miR-221-3p, radiosensitivity-promoting miR-16-5p, and anti-apoptotic Bcl-2 and Mcl1, while upregulating invasion-inhibiting miR-29a-3p, miR-34a-5p, miR-125-5p and apoptosis-promoter Fas, p53, TNF receptor superfamily member 10b (TNFRSF10B9, also known as TRAIL), and Bid . Ovarian cancer is a highly aggressive neoplasm whose complex management (due to molecular and cellular heterogeneity) together with its commonly delayed diagnosis have made it the fifth cause of death in women. The great majority of ovarian cancers are classified as epithelial belonging to the serous subtype . Treatment of high-grade ovarian serous adenocarcinoma cell line HEY with 200 μM and 400 μM OLE led to a reduction in cell viability, with accumulation of cells in G2/M phase and induction of apoptosis. A detailed analysis revealed that 400 μM OLE acted as a pro-oxidant agent, increasing oxygen radical generator labile iron pool (LIP) and ROS levels . Similarly, treatment of ovarian serous adenocarcinoma OVCAR-3 cell line with 200 μg/mL olive leaf extract containing 87% OLE for 24 h induced S/G2M phase cell cycle arrest and apoptosis. These effects were dependent on OLE-mediated ROS accumulation and reduction in CAT and SOD2 protein levels . Prostate cancer is the fifth leading cause of cancer-related deaths among men, whose therapeutic options are mainly defined on the basis of a combination of life expectancy, disease stage/risk classification/presence of metastasis, and cancer sensitivity towards androgens . Treatment of benign prostatic hyperplasia (BPH) epithelial cell line BPH-1 and prostate cancer cell lines LNCaP and DU145 with 100 μM and 500 μM OLE for 72 h determined a reduction in cell viability, probably as a consequence of the reduction in Akt phosphorylation levels, with cells showing signs of necrosis for the highest of the two tested OLE doses. Treatment with 100 μM and 500 μM OLE for 72 h also produced different effects on ROS homeostasis in the three cell lines. In BPH-1, a reduction in ROS was measured, while no effect was recorded for LNCaP cells, and a pro-oxidant activity with a marked ROS increase was exerted by OLE on DU145 cells. Measurement of non-protein thiol groups as a marker of oxidative stress defense revealed that OLE caused a dose-dependent increase in thiol groups in both BPH-1 and LNCaP cells, but only 100 μM OLE augmented thiol groups in DU145 cells; on the contrary, 500 μM OLE diminished thiol groups in this cell line. As regards oxidative damage protecting enzyme heme oxygenase-1 (HO-1), both 100 μM and 500 μM OLE increased HO-1 protein levels in BHP-1 and LNCaP cells, while in DU145 cells, OLE treatment decreased HO-1 enzyme expression. The levels of γ-glutamylcysteine synthetase (γ-GCS) (involved in reduced glutathione synthesis) were not affected by OLE treatment in the three studied cell lines . ROS homeostasis also seems to play a fundamental role in HT cytotoxic action on prostate cancer cells. Treatment of prostate cancer cell line PC3 with 80 μM HT caused mitochondrial dysfunction and triggered O2-dependent apoptosis . Similarly, 50–150 μM HT reduced cell viability of PC3 and LNCaP cells after 72 h treatment in a dose-dependent fashion (mean IC50 ± S.D. = 103 ± 7 μM and 146 ± 12 μM for PC3 and LNCaP cells, respectively), with a mechanism that seemed dependent on the ability of culture conditions to favor H2O2 accumulation (quantified in the medium after 24 h incubation with 100 μM HT as mean ± S.D. = 12.0 ± 2.9 and 8.8 ± 3.4 for PC3 and LNCaP cells, respectively) . HT action might overcome androgen sensitivity. Incubation of androgen-sensitive LNCaP and castration-resistant C4-2 cell lines with 50–400 μM HT for 48 and 72 h reduced cell viability in a dose-dependent manner, with the IC50 values for LNCaP = 190 μM and 86.9 μM after 48 and 72 h, respectively, and the IC50 values measured for C4–2 = 176 μM and 76.5 μM after 48 and 72 h, respectively. Cell cycle analysis after 48 h incubation with 100–300 μM demonstrated a dose-dependent cellular accumulation in the G1 phase and apoptosis induction, with (I) caspase-3/7 and PARP activation, (II) reduction in Bcl-2, Bcl-XL, cell cycle progression drivers cyclins D1 and E, and CDK2 and CDK4 protein levels, (III) increased Bax protein levels, (IV) reduced NF-κB p65 nuclear localization, (V) reduction in activation levels of cell growth regulators Akt and STAT3, and (VI) decreased androgen receptor protein levels and activity . The 48 h incubation of human prostate cancer cell lines LNCaP, 22Rv1, and PC-3 with 30–300 μM HT decreased cell viability in a dose-dependent fashion. As indicated by the authors, mean IC50 ± S.E.M. = 41.17 ± 2.79 μM for LNCaP cells, 9.32 ± 0.50 for 22Rv1 cell line, and 28.88 ± 2.25 for PC-3 cells. Treatment of PC-3 cells with 30 μM and 100 μM HT for 24 h reduced the migration rate in a dose-dependent manner. Treatment of 22Rv1 cells with 10 μM HT reduced prostatosphere number and size after 10 days, and phosphorylation of ERK1/2, cAMP response element-binding (CREB) protein, and JNK after 24 h . Osteosarcoma is a primary malignant bone tumor historically known for its extremely low survival rates even in the case of surgical resection/amputation, which have been improved by the introduction of neoadjuvant and adjuvant chemotherapy . Data about OLE cytotoxic effect in vitro are conflicting, and partially dependent on the used cell line and the duration of the treatment. In human 143B osteosarcoma cells, OLE showed a dose-dependent inhibition of proliferation when used at concentrations of 62.5 μM, 125 μM, and 250 μM for 24 h, and 1 μM–250 μM for 48 h. Also, 100 μM OLE induced autophagy after 48 h and exhibited an anti-migratory effect after 60 h incubation . In human MG-63 cells, 3–50 μg/mL (≈5–92.5 μM) OLE for 24 h reduced cell viability in a dose-dependent manner with IC50 = 22 μg/mL ± 3.6 . Another report documented no effect on MG-63 cell viability at OLE concentrations of 50 μM and 100 μM, while a dose-dependent cytotoxic effect was documented for 200 μM and 400 μM OLE after 24, 48, and 72 h incubation, with IC50 = 346 μM for 48 h incubation . Treatment of MG-63 cells with 20 μg/mL (≈37 μM) OLE for 48 h boosted the transcription of autophagy-related genes unc-51, like autophagy activating kinase 1 (ULK1), activating molecule in BECN1-regulated autophagy protein 1 (AMBRA1), and Bcl-2 interacting protein 3 like (BniP3L), as well as protein levels of ubiquitin binding protein p62, suggesting that OLE activity relies on autophagy induction . In Saos2 osteosarcoma cell line, 50 μM OLE caused no changes in cell viability after 24, 48, and 72 h incubation. Treatment with 100 μM OLE affected cell viability only after 48 h incubation, causing no detectable effects after 24 and 72 h incubation. Instead, 200 μM OLE started being effective as a cytotoxic agent after 48 h incubation, with effects confirmed after 72 h incubation. On the contrary, 400 μM showed a consistent cytotoxic action at all the assayed time points (24, 48, and 72 h) . Neuroblastomas are clinically heterogeneous embryonal neuroendocrine tumors, driving important research efforts towards the improvement of survival rates for patients with high-risk disease . Both OLE and HT exhibit a cytotoxic action against human neuroblastoma cell line SH-SY5Y. After 48 h treatment, IC50 for OLE was 350 μM. OLE induced cell cycle arrest and apoptosis through upregulation of CDK inhibitors p53, p21, p15, and p16, together with downregulation of cyclins D1, D2, and D3, and CDK4 and CDK6 . HT was also effective in the reduction in cell viability of SH-SY5Y cells after 72 h incubation, with mean EC50 ± S.D. = 114.02 ± 1.69 μM, inducing apoptosis . Glioma, a neoplastic lesion arising from glial cells, is the most frequent tumor of the central nervous system, and glioblastoma multiforme accounts for the large majority of all gliomas. Glioblastomas are extremely aggressive and difficult-to-treat cancers, with very modest survival expectancy . Treatment of human glioblastoma cell lines A-172 and U-251 with 200 μM and 400 μM OLE for 24 h reduced cell viability in a dose-dependent fashion, triggering apoptosis and suppressing cell migration and invasion abilities with a mechanism involving the increase in Bax, MMP2, MMP9, and phospho-Akt levels, the reduction in Bcl-2 protein levels, and caspase-9 and caspase-3 activation . In glioblastoma cell line T98G, 277.5 μM and 555 μM OLE for 24 h diminished cell viability in a dose-dependent manner. Treatment with 555 μM OLE strongly increased the expression of miRNAs inhibiting tumor growth: miR-181b, miR-137, and Let-7d. On the contrary, 277.5 μM OLE increased Let-7d levels only. Expression of miR-153 was not changed by OLE treatments . In this section, the most important results about less investigated malignancies are summarized. Head and neck squamous cell carcinoma is the sixth most common cancer and arises from mucosal epithelium in the oral cavity, pharynx, and larynx . Treatment of human cell lines Tu686 and CAL-27 with 50–200 μg/mL OLE (≈92.5–370 μM) significantly reduced cell viability, while no effect was elicited by concentrations ≤ 25 μg/mL (≈46 μM). Consistently, 25 μg/mL did not change the apoptotic rate in the mentioned cell lines, but inhibited epithelial-to-mesenchymal transition (EMT) induced by transforming growth factor-β1 TGF-β1 . The effects of OLE on EMT-related proteins were also verified in a in vivo model of 686LN-M2 cell line xenograft in BALB/c nude mice . Among gastric cancers, gastric adenocarcinoma is the most common subtype . Treatment of human gastric adenocarcinoma cell line CRL-1739 with 50–500 μM OLE for 24 h reduced cell viability in a dose-dependent fashion, with results becoming significant for values between 200 μM and 500 μM. The authors calculated that IC50 = 42 μM. As demonstrated by incubating the mentioned cell line with 200–500 μM OLE, 24 h exposure elicited the accumulation of ROS and apoptosis . Seminoma is the result of germ cell malignant transformation. It is generally associated with an excellent prognosis, and may involve testicles or extra-gonadal sites . The 48 h incubation of human seminoma cell lines SEM-1 and TCAM-2 with 15–200 μM OLE reduced cell viability in a dose-dependent manner, with IC50 = 140 μM and 50 μM for SEM-1 and TCAM-2, respectively. Treatment of the mentioned cell lines with OLE concentrations corresponding to their respective IC50 values produced apoptosis, the reduction in protein levels of cyclin D1 and nuclear localization of NF-κB, the increase in Bax and p21, and impaired cell migratory capacities through downregulation of TGF-β1 . Hematological malignancies arise from the loss of hematopoietic homeostasis, and may be configured as a large category including both myeloid and lymphatic neoplasms (leukemia, lymphoma, and multiple myeloma) whose incidence is pretty variable at the regional level . Only limited experimental evidence supporting OLE and HT preventive and anti-cancer properties is available in such a context (Table 1). Acute promyelocytic leukemia is a form of acute leukemia characterized by a chromosomal rearrangement involving PML::RARA fusion , making this malignancy sensitive to differentiating agents inducing PML-RARA fusion protein degradation by targeting the RARA or the PML part (all-trans-retinoic acid and arsenic-trioxide, respectively) . In vitro, treatment of promyelocytic leukemia cell line HL-60 with 50–100 μM HT reduced cell viability and triggered apoptosis, as indicated by the cleavage of PARP, the activation of caspase-3, and cytochrome c release . According to another report, 100 μM HT reduced DNA synthesis in HL-60 cells, with promotion of cell differentiation, cell accumulation in G0/G1 phase, and apoptosis of S phase cells . The detected cytotoxic effect of HT was caused by HT-mediated H2O2 accumulation in cell culture medium, overcoming HL-60 ability to clear ROS . The accumulation of H2O2 and the extent of apoptosis were inversely correlated with cell density . As hypothesized by the authors, in their experimental setting, the first step of HT-mediated H2O2 accumulation is the auto-oxidation of HT by O2, with production of the corresponding o-quinone and superoxide O2; the process is accelerated by SOD . These findings were confirmed in a study using an HT-rich natural extract of the olive pulp as a source of HT . Effects of HT on cell cycle and induction of apoptosis in replicating cells seemed to be related to an increase in cyclin D3 flanked by a decrease in CDK6 when HT concentration was 100 μM, whereas it was linked to an increase in p21 and p27 mRNA and protein levels when HT was used at a concentration of 75 μM. As suggested by the authors, these observations may be explained in the light of cyclin D3-dependent CDK6-mediated phosphorylation of pRB, which in turn causes the release of E2F transcription factor and G1/S transition as well as DNA synthesis. Intriguingly, both p21 and p27 are able to inhibit CDK6-mediated pRB phosphorylation and accelerate HL-60 differentiation . Chronic myelogenous leukemia (also known as chronic myeloid leukemia) is a myeloproliferative neoplastic disease characterized by the chromosomal translocation t(9;22)(q34;q11.2), resulting in the presence of the Philadelphia chromosome. The mentioned translocation produces the BCR- ABL1 fusion oncogene, encoding for a constitutively active tyrosine kinase, making leukemia cells sensitive to tyrosine kinase inhibition, with very good 5-year survival rates (90%) . Incubation of chronic myelogenous leukemia cell line K562 for 4 days with 200 μg/mL (≈370 μM) and 400 μg/mL (≈740 μM) OLE reduced cell density and viability. After 48 h treatment with 200 μg/mL OLE, a significant activation of caspase-1 was measured. Also, 200 μg/mL OLE caused a reduction in Prdx-1 protein level after 8 h of treatment, which was maintained after 24 h . Treatment of K562 cells with concentrations of HT up to 1000 μM produced a dose-dependent reduction in cell viability (EC50 = 147 μM) with concomitant increase in caspase 3/7 activity . Acute monocytic leukemia is the expression formerly utilized to indicate a neoplastic lesion arising from the loss of normal maturation along the monocytic lineage, and is now included in the larger category labeled as acute myeloid leukemia (that is the most frequent leukemia among adults) . THP-1 is probably the most known acute monocytic leukemia cell line. Among all HT doses tested (1–40 μM), only 20 μM affected viability of THP-1 cells after 72 h . Instead, in U937 cells (a human acute myeloid leukemia cell line of pro-monocytic origin) , 75 μM and 200 μM HT increased cell death and apoptosis . T-cell acute lymphoblastic leukemia (T-ALL) represents the consequence of the loss of proper regulation of T-cell development, resulting in the accumulation of immature progenitors. With conventional therapies, survival is lower among adult patients vs. pediatric subjects, mainly because of treatment-associated toxicity and higher relapse rates in adults . Treatment of T-ALL cell line CCRF-CEM with HT concentrations up to 1000 μM produced a dose-dependent reduction in cell viability (EC50 = 338 μM), flanked by the increase in caspase 3/7 activity . Despite being considered as a T-ALL cell line, Jurkat cells are immunophenotypically different from T-ALL cells; thus, results obtained in this cell model retain modest reliability . A study performed using an HT-rich natural extract of the olive pulp as a source of HT demonstrated a dose-dependent reduction in cell viability and induction of apoptosis, together with ROS accumulation . As the literature analysis in Section 2 and Section 3 corroborates, methodological standardization for the study of the effects of natural products may reveal to be challenging, mainly as a consequence of the combination of the multiplicity of molecular patterns triggered and effects elicited by these substances, together with the intrinsic variability of the biological systems (cancer cells and interaction with the surrounding tissue cells) of interest. The dissertation above seems to validate the use of OLE and HT as cytotoxic agents, with the chance of a further employment of both compounds as microenvironment modulators (OLE- and HT-dependent effects on cancer cell viability and behavior are summarized in Figure 2). However, before extrapolating any conclusion, the reported data should be interpreted and discussed in the light of factors determining the feasibility of such a purpose. In the next paragraphs, we discuss the role of OLE and HT absorption, bioavailability and toxicity to non-cancer cells, OLE and HT interaction with chemotherapy drugs, OLE- and HT-mediated repercussions on drug metabolism, and OLE and HT antioxidant activity in the context of chemotherapy-induced oxidative stress as determinants of the net OLE and HT activity on cancer insurgence and growth in humans. Factors influencing OLE and HT effects in human cancers are reported in Figure 2. Schematization of molecular in vitro and in vivo effects determining OLE- and HT-mediated suppression of cancer cell growth, and factors influencing OLE and HT cytotoxicity in humans. In the upper left part of the image, mediators whose expression is downregulated and/or activity is suppressed and/or levels are reduced by OLE and HT are indicated. In the upper right part of the figure, pathways induced and mediators upregulated by OLE and HT are reported. OLE and HT induce the activity and/or increase the levels of pro-apoptotic mediators, while inhibiting patterns involved in proliferation and survival. The role of some actors (like autophagy marker p62 and intracellular calcium levels) still needs to be elucidated. ROS function is still poorly understood, since oxidative stress seems involved in OLE- and HT-mediated reduction in cancer cell viability, despite OLE and HT being antioxidants. It is worth noting that these results are not expected to be automatically confirmed in cancer patients, since the role of modest OLE and HT bioavailability, OLE and HT interference with drug metabolism and chemotherapy effect, and OLE and HT toxicity to non-cancer cells have been barely investigated in the context of cancer prevention and cancer treatment. Molecular structures were designed with MolView Copyright© 2014–2023 Herman Bergwerf available at https://molview.org/ (accessed on 13 February 2024); the artwork was made with Microsoft Paint 3D version 6.2310.24037.0 (Microsoft Corporation, Redmond, WA, USA). Information about OLE and HT absorption through the digestive tract and bioavailability is limited, and often arises from a combination of experimental proofs obtained from humans and animal models. After ingestion, OLE remains mostly stable in the acid gastric environment, although non-enzymatic hydrolysis may account for an increased amount of HT reaching the small intestine. OLE is poorly absorbed through the small intestine (mainly via diffusion), as demonstrated in rats and humans, enters systemic circulation, undergoes sulphate and glucuronide conjugation and/or enzymatic conversion in HT, and is eliminated in urine mainly as aglycon and glucuronide derivatives . In the large intestine, OLE is metabolized by gut microbiota, producing HT . After ingestion, HT is more largely absorbed in the intestine by diffusion, is mainly metabolized into glucuronide and sulphate conjugates, and is excreted in urine mostly in its glucuronide conjugated form . For both OLE and HT (and their metabolites), the maximum excretion rate is reached in 4 h in humans . All these aspects may represent a potential source of difficulties if the objective is achieving and maintaining pharmacologically relevant concentrations of OLE and HT in plasma after ingestion. Whenever performing a critical evaluation of OLE and HT availability, data from animal models should be taken into account with caution, since it has been demonstrated that the rate of excretion of HT differs between humans and rats (being more rapid in humans) . Thus, only data related to OLE and HT plasma concentrations in humans are listed in this section. In a group of volunteers, the efficacy of OLE delivery was assayed for liquid and capsule preparations, each containing a lower (64 mg total olive phenols, with 51.1 mg OLE) or a higher (96 mg total, with 76.6 mg OLE) dose. The best performance was offered by higher dose liquid preparation, with a mean peak of plasma OLE ± S.D. = 3.55 ± 2.27 ng/mL and a mean time to peak ± S.D. = 20 ± 12 min, whereas the worst OLE peak plasma values were detected for the lower dose capsule preparation (mean peak ± S.D. = 0.52 ± 0.24 ng/mL, mean time to peak ± S.D. = 40 ± 27 min) . In humans, the assumption of 5 mg HT added to extra virgin olive oil produced a plasma peak of 3.79 ng/mL after 30 min, followed by a rapid decline in HT plasma concentration (minimum reached value < 2 ng/mL after 240 min) . This result matched another report documenting that the consumption of HT-enriched biscuits (containing 5.25 mg HT) determined the appearance of HT metabolites in the volunteers’ plasma between 30 min and 1 h . A similar pharmacokinetic effect was reported in other experimental settings using olive-derived watery supplements as a source of HT . The ingestion of 40 mL of high (366 mg/kg)-phenolic-compound-content olive oil led to a plasma HT concentration peak of ≈15 μM (mean time to peak ± S.D. = 0.91 ± 0.84 h, mean estimated elimination half-life ± S.D. = 3.00 ± 1.46 h), whereas the same amount of low- (2.7 mg/kg) and medium (164 mg/kg)-phenolic-compound-content olive oil produced an HT peak of ≈5 μM or less . The assumption of 25 mL of extra virgin olive oil led to a maximum plasma concentration = 4.4 ng/mL, (time to peak = 0.25 h) , and the ingestion of 25 mL of low-phenolic-content (10 mg/kg), moderate-phenolic-content (133 mg/kg), and high-phenolic-content (486 mg/kg) olive oil produce a plasma HT peak of ≈5 nM, 25 nM, and 50 nM, respectively . On the basis of these pieces of data, it becomes evident that cytotoxicity and anti-cancer effects of OLE and HT were recorded at concentrations largely exceeding those reachable with diet/olive oil consumption, and OLE and HT pharmacokinetics does not match the requested treatment duration to exert an anti-proliferative effect. Thus, it is difficult to imagine how OLE and HT may be used as cancer-preventive/treating agents if the route of administration is ingestion. Also, given that both phenols are extensively metabolized and rapidly excreted, the safety and efficacy of other routes of administration (e.g., intravenous) should be assessed in detail. However, even at high concentrations, OLE and HT seem to be selectively cytotoxic for cancer cells, with no or negligible/minimal effects on non-cancer cells, as demonstrated for embryonic rat cardiomyoblasts H9c2(2-1), human breast epithelial cell line MCF-10A , nonmalignant human bronchial epithelial BEAS-2B cell line , normal colonic cell line CCD-841CoN , human normal liver cell line (HL-7702) , human normal prostate epithelial cells PWLE2 , human bile duct cell line HIBEpiC , human fibroblasts WI-38 , normal skin fibroblast cell line WS1 , human GN61 gingival fibroblasts , human lymphocytes , human PBMCs , and normal human fibroblasts . Thus, OLE and HT are generally considered safe on the basis of experimental data, despite sporadic reports against the trend, as happened for HT, which proved to be toxic for human non-tumorigenic pancreas cells HPDE and human normal prostate RWPE-1 cells . The evaluation of OLE and HT safety profile in cancer patients is still pending. Some experimental proof has demonstrated that OLE and HT may potentiate the effect of both routinely employed and new potential anti-cancer drugs . In human melanoma cell line A375, the combination of 250 μM OLE with alkylating agent dacarbazine was more effective in reducing cell viability than dacarbazine alone . In female breast cancer patients undergoing neoadjuvant chemotherapy, orally administered 15 mg/kg HT determined a significant decrease in plasma levels of TIMP-1 during treatment with epirubicin and cyclophosphamide . In an in vivo model of tumor xenograft (triple negative MDA-MB-231 cell line) in BALB/c OlaHsd-foxn1 mice, peritoneally injected 50 mg/kg OLE for 4 weeks exhibited a synergistic effect with doxorubicin on inhibition of tumor growth and induction of apoptosis . Combination of paclitaxel with HT reduced MCF-7 and MDA-MB-231 cell viability in vitro, and tumor volume in breast cancer-bearing Sprague–Dawley rats (in vivo injected HT dose = 0.5 mg/kg/day) . In HT-29 and WiDr colorectal cancer cell lines, combination of 10 μM HT and monoclonal anti-epidermal growth factor receptor (EGFR) antibody cetuximab reduced cell growth, both in the presence and in the absence of epidermal growth factor (EGF) stimulation, inducing G1/S and G2/M phase cell cycle arrest . In human osteosarcoma MG-63 cell line, 20 μg/mL (≈37 μM) OLE had an additive effect on anthracycline Adriamycin-induced reduction in cell viability, with a mechanism that did not alter the G2/M phase blockade elicited by Adriamycin . Similarly, in 143B osteosarcoma cells, OLE showed a synergistic, antiproliferative, and anti-migratory effect in combination with estradiol metabolite 2-methoxyestradiol at all OLE concentrations tested (1–250 μM for proliferation, and 100 μM for wound healing assay) . On the contrary, HT did not modify doxorubicin-mediated growth inhibition of human osteosarcoma cells U-2 OS . In neuroblastoma cells T98G, 277.5 μM and 555 μM OLE showed a synergistic effect with alkylating agent temozolomide on cell viability, increasing the expression of miRNAs involved in tumor growth suppression, mainly Let-7d . OLE and HT may also magnify the efficacy of other types of cancer treatment. In nasopharyngeal cancer cell line HNE1 and HONE1, 200 μM OLE enhanced cell radiosensitivity in vitro and in vivo after injection in BALB/C nude mice, with a mechanism involving OLE-dependent removal of HIF-1α hypoxic repression exerted at miR-519d promoter region, upregulation of miR-519d, and miR-519d targeting of DNA damage-regulated protein 1 (PDRG1) . Despite these encouraging results, since both OLE and HT may act as transcriptional regulators and are extensively metabolized by the liver (see sections above), a careful analysis of the effects of both phenols on phase I and phase II enzyme kinetics and expression should be performed before considering the use of OLE and HT during chemotherapeutic treatments. Preliminary data obtained in human liver microsomes point towards OLE-mediated inhibition of CY3A and CYP1A2 activity . On the contrary, a study performed in 129/Sv WT and Ppara-null mice demonstrated that OLE (ingested with food, thus absorbed through the intestinal wall) stimulated the transcription of cytochrome P450 genes Cyp1a1, Cyp1a2, Cyp1b1, Cyp3a14, Cyp3a25, Cyp2c29, Cyp2c44, Cyp2d22, and Cyp2e1 in the liver, with a mechanism mediated by peroxisome proliferator-activated receptor α (PPARα) . Studies defining the effects of OLE and HT on the expression of and interaction with phase I and phase II enzymes are still missing, but they should be considered absolutely necessary in order to determine if OLE and HT are able to modify the metabolism of other drugs. Besides their possible synergic/additive actions, OLE and HT might also be seen as useful support agents during cancer treatment. A lot of experimental data in vivo and in vitro have definitively demonstrated the ROS scavenger ability of OLE and HT, which can also act on antioxidant cellular mechanisms restoring ROS homeostasis, including promotion of the increase in reduced glutathione levels (GSH), depletion of lipid peroxidation product malondialdehyde (MDA), intensification of the expression and/or activity of detoxicating enzymes SOD, CAT, glutathione-S-transferase (GST), and glutathione peroxidase (GSH-Px), and nuclear factor E2-related factor 2 (Nrf2) upregulation/transactivation, which in turn regulates the expression of fundamental enzymes protecting cells from oxidative damage, like HO-1 . Radical and non-radical ROS (including hydrogen peroxide H2O2, superoxide anion radical O2, and hydroxyl radical OH) may have a pro-tumorigenic effect; they are involved in cancer insurgence determining DNA damage, genomic instability, and interference with signalling and metabolic pathways, enhance cell proliferation through the activation of pro-survival pathways in cancer cells while promoting the reorganization of cellular antioxidant capacities and adaptation to hypoxic conditions, are involved in the development of anti-cancer therapy resistance (through the expansion of cell antioxidant capacities), push the activation of metastasis cellular programs via EMT, and boost immunosuppression and angiogenesis in the cancer microenvironment. However, if ROS concentration overcomes cancer cell defenses against oxidative stress and damage, or cancer cell-produced ROS balance is perturbated, ROS may account for fatal cell damage and trigger apoptosis . This may explain the fact that the mechanism of action of some common cancer treatments (e.g., radiation, inorganic compounds, tyrosine kinase inhibitors, monoclonal antibodies, protease inhibitors, pyrimidine analogues, alkylating agents, and anthracyclines) relies on oxidative stress and ROS-dependent apoptosis . OLE and HT cytotoxic actions themselves in part depend on ROS generation ; moreover, 200 μM OLE reduced HIF-1α mRNA and protein levels in HNE-1 and HONE-1 nasopharyngeal cancer cell lines , and pro-apoptotic OLE concentration corresponding to 100 μM was able to increase ROS production in MDA-MB-231 cell line, with the maximum peak obtained after 4 h incubation . ROS-mediated damage at least in part accounts for chemotherapy-dependent toxicity detected at the tissue level on non-cancer cells . However, OLE and HT have shown an important ability to mitigate the toxicity elicited by chemotherapeutic agents mainly through their largely demonstrated antioxidant and ROS scavenger activity. In fact, in vivo, 50 mg/kg, 100 mg/kg, and 200 mg/kg OLE showed a dose-dependent antioxidant activity, accounting for amelioration of cisplatin-induced pancreatic, liver, lung, and stomach damage in Spraque–Dawley rats . In the same in vivo model of cisplatin-induced oxidative stress, 50 mg/kg, 100 mg/kg, and 200 mg/kg OLE improved anemia, thrombocytopenia, and leukopenia . In cyclophosphamide- and epirubicin-induced toxicity in Sprague–Dawley rats, four cycles of 150 mg/kg/week OLE reduced lipid peroxidation and increased the activity of antioxidant enzymes in heart, kidney, and liver . Moreover, in a model of cyclophosphamide-induced immunosuppression in broilers, a solution containing 200 mg/L HT reduced duodenal MDA levels while increasing the activity of antioxidant enzymes . In vitro, 50 μM and 70 μM HT reduced doxorubicin-mediated toxicity in embryonic rat cardiomyoblasts H9c2(2-1) after 48 h incubation. Also, 24 and 48 h incubation with 50 μM HT reduced doxorubicin-dependent intracellular ROS accumulation, increasing SOD2 levels and protecting cells from doxorubicin-induced apoptosis . OLE- and HT-dependent redox homeostasis restoration might represent a potential issue with respect to OLE and HT use at nutritionally relevant concentrations as both anticancer drugs and detoxicating agents (especially during chemotherapy). Low concentrations of both OLE and HT (10 μM) protected HL-60 and PBMCs from H2O2-induced DNA damage, and lymphocytes from PMA-stimulated monocyte-mediated oxidative DNA damage . This piece of information sounds particularly alarming, since arsenic trioxide (As2O3) is one of the agents utilized to treat acute promyelocytic leukemia, especially in combinatory first-line therapy, and its mechanism of action relies on ROS increase . Incubation with 30 μM HT for 48 h was not sufficient to reduce viability of Hep3B cells, but significantly improved cellular antioxidant capacities . Similarly, 5–200 μM HT was ineffective as a cytotoxic reagent, but reduced the level of oxidative stress in MCF-7 cells after 16 h incubation . Erastin is a ferropoptosis inducer that is currently under evaluation as an anti-cancer treatment . In ovarian carcinoma HEY cells, 100 μM OLE was not able to affect cell viability, but acted as an antioxidant, decreasing endogenous and erastin-dependent LIP and ROS levels, preventing erastin-dependent ROS accumulation in mytochondria and increasing glutathione peroxidase 4 (GPX4) levels that were reduced by erastin treatment, finally counteracting erastin-induced cell death . It remains to be demonstrated if nutritionally relevant and/or non-cytotoxic concentrations of OLE and HT may offer a survival advantage to neoplastic cells. A similar scenario has been described for other antioxidants (e.g., N-acetyl-L-cysteine, alpha-tocopherol and carotenoids), which showed a pro-tumorigenic action by exerting beneficial effects on cancer cells . It will be fundamental to deepen these pieces of data in order to rule out the chance that OLE and HT may promote cancer growth and metastasis. Another risk factor for tumorigenesis is inflammation. ROS imbalance itself may drive inflammation, inducing the production of pro-inflammatory cytokines interleukin-1 (IL-1), IL-6, IL-8, and tumor necrosis factor-α (TNF-α) . Persisting (chronic) or poorly controlled inflammation may directly promote tumorigenesis due to the action of ROS and the recruitment of immune cells triggering the release of growth signals and the activation of tissue repair mechanisms. After cancer insurgence, the production of inflammatory mediators sustained by cancer cells and surrounding actors (tissue macrophages, cancer-associated fibroblast, infiltrating immune cells, and endothelial cells, among the others) contributes to the creation of an environment favoring immune evasion, cell survival, invasion, and angiogenesis. Chemotherapy may also account for an exacerbation of inflammation, whose biological meaning is poorly understood, being recognized as immune-activating as well as a contributor to the failure of the therapeutic regimen . OLE and HT exhibit an anti-inflammatory activity that has been demonstrated in multiple in vivo and in vitro models, although some experimental variables related to the timing of administration of inflammatory stimuli might have affected the reproducibility of the results, mainly in vitro, and nutritionally relevant concentrations of OLE and HT often showed no anti-inflammatory activity on human peripheral blood mononuclear cells . Proof of OLE and HT modulation of inflammation in cancer cells has been obtained mostly from colorectal cancer. In an in vivo model of AOM/DSS-induced colorectal cancer in C57BL/6 mice, 50 mg/kg and 100 mg/kg OLE reduced IL-6, TNF-α, and IFN-γ colon tissue levels, as well as COX-2 levels . Treatment of HCT116 and LoVo cells with 0.0154 mg/mL (≈100 μM) HT and 0.0231 mg/mL (≈150 μM) HT, respectively, for 72 h reduced phosphorylation of NF-κB p65, in turn leading to a reduction in pro-inflammatory cytokines TNF-α and IL-8 at both mRNA and protein levels. HT-dependent anti-inflammatory effect was also mediated by HT-elicited increase in protein levels of PPARγ. In addition, HT acted as a PPARγ agonist, promoting its transcriptional activity . In HT-29 cell line, 100–400 μM reduced TNF-α-induced NF-κB activation in a dose-dependent manner . In in vitro models, the frame appears completely different. The only HT concentrations that elicited a reduction in IL-6 levels were 80 μM for HepG2 and 30 μM for Hep3B. HT doses of 100 μM and 200 μM in HepG2 and 80 μM, 100 μM, and 200 μM in Hep3B cells produced an increase in IL-6 release . Treatment of K562 cells with 100 μM HT produced transcriptional effects including the downregulation of IL-10 receptor and the upregulation of inflammatory mediators IL-6 and IL-8 . OLE and HT may participate in control of cancer-associated and chemotherapy-dependent tissue inflammation by acting on non-tumor cells. As demonstrated in vivo, treatment of BALB/cN mice with 5, 10, and 20 mg/kg OLE suppressed signs of cisplatin-induced renal inflammation, including tissue modulation of TNF-α levels . Four cycles of 150 mg/kg/week OLE in Sprague–Dawley rats reduced serum TNF-α and IL-6 in an experimental model of epirubicin and cyclophosphamide toxicity . In a model of cyclophosphamide-induced immunosuppression in broilers, a solution containing 200 mg/L HT promoted the duodenal expression of anti-inflammatory cytokines IL-4 and IL-10 . It remains to be assessed if such a modulatory activity is able to contribute to the immunosuppressant environment surrounding cancer cells. From the analysis of the literature reported in this review, it becomes evident that the multifaceted nature of OLE and HT interaction with molecular mediators in cancer cells and non-cancer tissues determines the need for safe strategies to improve OLE and HT bioavailability and delivery, also offering a more stable and highly selective anti-proliferative activity throughout time. Different approaches have been evaluated to reach this objective through the development of OLE and HT derivatives or carrier systems. For example, peracetylated OLE had a stronger anti-proliferative and antioxidant activity than OLE in thyroid and breast cancer in vitro ; iron oxide nanoparticles coated with glucose and conjugated with OLE showed a notable cytotoxic action against colorectal cancer cell line SW480 ; chitosan nanoencapsulation of HT was shown to be unaffected by time-dependent dynamic changes in efficacy observed for free HT in lung and breast cancer cell lines ; hydroxytyrosyl dodecyl ether, an HT alkyl ether derivative, had a very strong cytotoxic activity at low concentrations (IC50 = 19.9 ± 4.6 μM) in lung cancer cells in vitro ; hydroxytyrosyl oleate, an HT ester, exerted the same effects shown by HT on SH-SY5Y neuroblastoma cells at lower concentrations ; HT-loaded poly lactide-co-glycolide-co-polyacrylic acidnanoparticles exhibited cytotoxic and transcriptional regulating effects at lower concentrations (6 ppm) than free HT in colorectal cancer in vitro . Anyway, it would be highly recommendable to address future OLE- and HT-specific research directions towards the exploration of the true effect of nutritionally relevant concentrations of these phenols (or OLE and HT concentrations that can realistically be maintained in plasma) on cancer cells, since even most recent evidence is not able to determine if the administration of OLE and HT in humans could be detrimental in terms of tumor growth and anti-cancer drug metabolism/effects. Up to now, conclusive studies assessing the influence of nutritionally relevant concentrations of OLE and HT on tumor insurgence and expansion, as well as proof of the absence of OLE- and HT-mediated alterations of chemotherapy metabolism, are still missing. Another important reason for concern is represented by OLE and HT context-dependent dual action as both antioxidant and pro-oxidant agents, since such a behavior may alter the efficacy of cancer treatment in unexplored ways. Also, OLE and HT modulatory activity on immune cells deserves further investigation. Currently, neither OLE nor HT could be safely considered as cancer-preventive agents or drugs for combinatory therapies without excluding the possibility that nutritionally relevant concentrations of these compounds might facilitate neoplastic cell expansion or even treatment escape. F.P. was supported by Fondazione Umberto Veronesi. The authors declare no conflicts of interest. This research received no external funding. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
PMC11641291
Gliotic Response and Reprogramming Potential of Human Müller Cell Line MIO-M1 Exposed to High Glucose and Glucose Fluctuations †
Retinal neurodegeneration (RN), an early marker of diabetic retinopathy (DR), is closely associated with Müller glia cells (MGs) in diabetic subjects. MGs play a pivotal role in maintaining retinal homeostasis, integrity, and metabolic support and respond to diabetic stress. In lower vertebrates, MGs have a strong regenerative response and can completely repair the retina after injuries. However, this ability diminishes as organisms become more complex. The aim of this study was to investigate the gliotic response and reprogramming potential of the human Müller cell line MIO-M1 cultured in normoglycemic (5 mM glucose, NG) and hyperglycemic (25 mM glucose, HG) conditions and then exposed to sustained high-glucose and glucose fluctuation (GF) treatments to mimic the human diabetic conditions. The results showed that NG MIO-M1 cells exhibited a dynamic activation to sustained high-glucose and GF treatments by increasing GFAP and Vimentin expression together, indicative of gliotic response. Increased expression of SHH and SOX2 were also observed, foreshadowing reprogramming potential. Conversely, HG MIO-M1 cells showed increased levels of the indexes reported above and adaptation/desensitization to sustained high-glucose and GF treatments. These findings indicate that MIO-M1 cells exhibit a differential response under various glucose treatments, which is dependent on the metabolic environment. The in vitro model used in this study, based on a well-established cell line, enables the exploration of how these responses occur in a controlled, reproducible system and the identification of strategies to promote neurogenesis over neurodegeneration. These findings contribute to the understanding of MGs responses under diabetic conditions, which may have implications for future therapeutic approaches to diabetes-associated retinal neurodegeneration.Diabetic retinopathy (DR) is the most common and sever microvascular complication of diabetes mellitus (DM), accounting for the majority of vision damage to the retina and blindness in adults . In recent years, the concept of DR as a microvascular disease has evolved, as it is now recognized as a more complex diabetic complication in which retinal neurodegeneration (RN) plays a significant role . In this contest, RN has recently been considered as an early marker of DR, since it seems to precede vascular damage . The neurovascular unit (NVU) in the retina refers to the functional coupling and interdependency of neurons (e.g., ganglion cells, amacrine cells, horizontal and bipolar cells), glia (e.g., Müller cells and astrocytes), immune cells (e.g., microglia and perivascular macrophages), and highly specialized vascular cells (e.g., endothelial cells and pericytes) . The impairment of the NVU represents a primary event in the pathogenesis of DR and is characterized by RN and early microvascular alterations. The hallmarks of diabetes-induced RN, which include neural-cell apoptosis, diminished retinal neuronal function, and reactive gliosis, have been observed to occur prior to the onset of overt microangiopathy in experimental models of DR, diabetic patients, and post-mortem human retinas . Müller Glia cells (MGs) are the main retinal glial cells, extending throughout the entire thickness from the inner to the outer limiting membranes, providing an anatomical link between the retinal neurons and the retinal blood vessels and thus responsible for the homeostatic and metabolic support of retinal nerve cells . MGs also play a pivotal role in maintaining neuronal health through the recycling of neurotransmitters and the prevention of neurotoxicity resulting from glutamate excess. Additionally, they regulate ionic balance by buffering K ions and supply lactate as an energy source for photoreceptors . In response to diabetic stress, MGs undergo morphological changes and exhibit an increased expression of the intermediate filament proteins, such as glial fibrillary acidic protein (GFAP), a key marker of reactive gliosis . Reactive gliosis plays a dual role in the context of retinal damage, with the potential to both protect against damage and contribute to its progression in a diabetic milieu. The formation of a glial scar, while stabilizing the tissue structure, may also impair neuronal function through the release of proinflammatory mediators . The ambivalent role of MGs gliosis in DR underscores the complexity of their involvement, thereby emphasizing the necessity for a comprehensive understanding of the regulatory mechanisms involved. The impact of fluctuating glycemic conditions on MGs responses introduces an additional layer of complexity. Fluctuations in blood glucose levels have been demonstrated to trigger diverse signaling pathways within MGs, resulting in an increase of the inner nuclear layer (INL) in subjects with diabetes , potentially leading to differential activation states or pathological outcomes . As a matter of fact, in recent years significant advancements have been made in understanding the pathophysiological roles of MGs in DR . Studies in lower vertebrates, including fish and amphibians, have demonstrated that, following reactive gliosis, the sonic hedgehog (SHH) signaling pathway is activated, thereby promoting the reprogramming of MGs into various retinal cell types and contributing to retinal regeneration . The capacity for retinal regeneration is notably limited in mammals, including humans. In response to injury, MGs typically undergo a process of reactive gliosis, which often results in scarring rather than regeneration . To achieve significant regenerative processes, additional manipulations of the SHH pathway may be required . Recent studies have identified the potential genes involved in reprogramming MGs into retinal neurons, such as Ascl1, Sox2, and Lin28, which can be modulated to stimulate functional neuronal regeneration . Despite the advancement of knowledge regarding the neurogenic process, however, the existing literature lacks insights into the influence of glucose metabolism on these processes. In particular, as far as we know, there is no evidence in the literature indicating the effect of glucose metabolism on retinal neurogenesis. Our previous in vitro studies, involving rat retinal Müller cell line rMC-1, have demonstrated that these cells are activated in response to both high glucose levels and glucose fluctuations (GFs), and that this activation is associated with cellular increased levels of GFAP and aquaporin-4 (AQPs) . These findings contributed to elucidating the dynamic responses of Müller cells under varied glycaemic conditions, suggesting the existence of a complex regulatory mechanism sensitive to metabolic fluctuations. However, to better understand the mechanisms underlying DR in humans, it is essential to investigate whether or not similar responses occur in human Müller cell lines, as cellular responses may differ between animal models and humans, particularly due to differences in complexity and disease progression. The present study aims to fill this gap by investigating the gliotic response of human Müller cell line MIO-M1, a well-established model in human retinal research, exposed to sustained high-glucose and glucose fluctuation treatments that mimic the different glycemic conditions that can occur in diabetic patients. Furthermore, the study aims to investigate the reprogramming potential of MIO-M1 cells under different glucose stress conditions, focusing on their potential for dedifferentiation and neurogenesis. These processes are of significant interest in understanding how glial cells respond to pathological conditions such as DR. This approach allowed one to explore not only the gliotic response but also the potential for reprogramming Müller cells into a more progenitor-like state, a key aspect for the development of future regenerative therapies. The results obtained might contribute to elucidating the dynamic responses of MGs under different glycemic conditions, suggesting the existence of a complex regulatory mechanism sensitive to glucose metabolic stress. WB analysis carried out on MIO-M1 cells cultured under HG condition showed a significant increase in GFAP levels compared to MIO-M1 cells cultured under NG condition (fold change of about 2 in HG vs. NG), indicating an enhanced gliotic response under hyperglycemic stress conditions (Figure 1A,B). Such significant increases of GFAP levels was also observed when NG MIO-M1 cells were exposed to sustained high-glucose (II) and GF (III) treatments compared to their basal glucose (I) (Figure 2A,B and Figure S1). Although a clear trend towards increased GFAP expression was also observed across GF treatments IV and V, these increases were not statistically significant (Figure 2A,B). In contrast, MIO-M1 cells cultured under HG condition showed a markedly different response, with no significant changes in GFAP levels when these cells were exposed to sustained high-glucose (II) and GF (III–V) treatments compared to their basal glucose treatment (I) (Figure 2A,B). The impact of the experimental culture conditions described above on GFAP expression in MIO-MI cells was also investigated by IF, and cellular bipolar or radial morphologies, typical of quiescent and activated Müller cells, respectively , were also recorded. Under NG condition, GFAP expression was observed in a limited number of MIO-MI cells, approximately 15%, which were GFAP positive, consistent with literature data . Typical of low active Müller cell status, GFAP staining in NG MIO-M1 cells was faint, while bipolar or radial morphologies were equally distributed (Figure 3A,B). In contrast, MIO-M1 cells maintained in the HG condition showed intense GFAP staining, with almost 80% displaying radial morphology (Figure 3A,B), indicating that these cells are in a more activated state. Similarly to the WB results, in NG MIO-M1cells, subjected to all different glucose treatments (II–V), a significant increase in the GFAP intensity and radial morphology were observed, compared to the basal glucose treatment (I) (NG: 47.85% condition I vs. 69,63% condition II; 77.80% condition III; 74.21% condition IV; 81.42% condition V) (Figure 4A,B). In contrast, in HG MIO-M1 cells subjected to any experimental glucose treatments, no significant change in GFAP positivity or intensity and morphology were detected compared to the basal glucose treatment (Figure 4A,C). IF for vimentin filaments, another well-established marker of Müller cell stress response, on MIO-M1 cells subjected to all different glucose treatments gave similar results to those described for GFAP and cell morphology changes. Briefly, NG MIO-M1 cells in the basal glucose treatment (I) exhibited intact cytoskeletal integrity, with well-organized and linear vimentin filaments. Following the exposure of NG MIO-M1 cells to sustained high-glucose (II) and GF (III–V) treatments, a significant increase in vimentin expression was observed, compared to the basal glucose treatment (I). This increase was associated with a more disorganized appearance of vimentin filaments, reflecting increased cellular stress and cytoskeletal reorganization (Figure 5). Such a result was also observed in HG MIO-M1 cells maintained in the basal glucose treatment (I) and when these cells were exposed to all different glucose treatments (II–V) (Figure 5). In order to investigate the activation of the neurogenic potential of MIO-M1 cells under the various glucose treatments investigated here, the expression and cellular localization of sonic hedgehog (SHH), a key signaling protein capable of regulating the differentiation process during retinal development, was assessed . WB analysis of MIO-M1 cells maintained under NG and HG conditions revealed a significant increase in SHH protein levels in the latter. IF observations confirmed such results and provided insights into the differential cytoplasmic localization of the protein between NG and HG MIO-M1 cells. In the majority of NG MIO-M1 cells (about 85%), SHH was widespread throughout the cytoplasm. In contrast, in approximately 70% of HG-MIO-M1 cells, although SHH maintained a widespread distribution, it was also predominantly localized in punctate structures (dot-like formations) within the cytoplasm and near the plasma membrane (Figure 6C,D). The WB results reported in Figure 7A,B show that SHH expression was significantly increased in NG MIO-M1 cells subjected to GF treatments (III–V), compared to their basal glucose treatment (I). Although a clear trend towards increased SHH expression was also observed in response to sustained high-glucose treatment (II), this increase was not statically significant (Figure 7A,B). Conversely, HG MIO-M1 cells subjected to treatments II and V showed no change in SHH expression compared to the basal glucose treatment, while a decrease of the SHH protein level in cells exposed to GF treatments III and IV was observed (Figure 7A,B). IF basically confirmed these findings at the protein level and revealed a distinct cytoplasm localization of the protein in NG MIO-M1 cells exposed to the different glucose treatments (II–V) compared to the basal glucose treatment (I). In these cells, the increased level of SHH expression observed under sustained high-glucose (II) and GF (III–V) treatments was accompanied by a significant change in the intracellular localization of SHH, showing a more punctate pattern, consistent with that of HG MIO-M1 cells (Figure 7C,D). In contrast, HG MIO-M1 cells exposed to all different glucose treatments showed no changes of intracellular SHH distribution compared to the basal glucose treatment (Figure 7C,D). These results underline the notion that Müller glia undergo adaptation or desensitization when cultured in a chronically hyperglycemic environment and subsequently subjected to further glucose-induced stress. IF observations and q-PCR analyses for SRY-Box Transcription Factor 2 (SOX2), a transcription factor that plays an important role in reprogramming Zebrafish Müller glia , showed expression changes of this transcription factor parallel to that of SHH. In short, while approximately 55% of NG MIO-M1 cells were positive for the protein, in line with Lawrence et al. (2007) , this increased to 80% in HG MIO-M1 cells (Figure 8A,B). q-PCR analysis confirmed a higher SOX2 transcript in these latter in comparison to the former (Figure 8C). Additionally, q-PCR analysis revealed a significant increase of SOX2 mRNA levels in NG MIO-M1 cells exposed to sustained high-glucose (II) and GF (III–V) treatments compared to the basal glucose treatment (I) (Figure 8D). In contrast, no significant changes in SOX2 mRNA levels were observed in HG MIO-M1 cells when the cells were exposed to the same glucose treatments (Figure 8D). Müller glial cells play fundamental roles in retinal tissue functions. Therefore, studies into their biology and functions may contribute to understanding the causes of retinal pathologies and to develop strategies to alleviate their outcomes. The aim of this study was to investigate the impact of high glucose and glucose fluctuations on critical cellular processes, such as gliosis and reprogramming in Müller cells. These processes are of pivotal importance in the onset and progression of retinal neurodegeneration and diabetic retinopathy. In the present study, we observed a significant increase in GFAP expression in NG MIO-M1 cells exposed to sustained high-glucose and GF treatments, indicating a glucose stress-induced gliotic response. This response was associated with other markers of reactive gliosis, such as the morphological changes associated with the transition from a bipolar to a radial morphology and the overexpression/reorganization of vimentin intermediate filament. These findings indicate that NG MIO-M1 cells are sensitive to glucose changes in their surrounding environment, potentially functioning as a protective or reactive mechanism in response to metabolic stress. The remarkable sensitivity to glucose stress and the prompt reactive gliosis observed in MIO-M1 cells were found to be similar to those observed in Müller cells of diabetic subjects . This finding confirms the suitability of MIO-M1 cells as an in vitro model to study Müller cell response. In contrast, HG MIO-M1 cells maintain a basal high level of both GFAP and Vimentin together to less organized intermediate filaments across the different glucose treatments, suggesting a limited and static response to sustained high-glucose and GF treatments. These findings highlight that MIO-M1 cells in chronic hyperglycemic conditions may adapt or develop enhanced tolerance over time due to constant and prolonged exposure to high glucose levels. The adaptation or desensitization of HG MIO-M1 cells to sustained high-glucose and GFs may indicate a saturation of the gliotic response, which is characteristic of prolonged hyperglycemic states, as observed in uncontrolled diabetes . This adaptation or desensitization may result in a reduction in the cellular capacity to respond to additional metabolic challenges or stress. To understand the clinical consequences of Müller cell gliosis it is essential to also consider the reprogramming process, which may include dedifferentiation of Müller cells and regeneration of various retinal neurons, contributing to retinal tissues repair under certain conditions. In species such as birds , zebrafish , and rodents , Müller glia contribute as the primary source of retinal regeneration. When the retina is damaged, Müller glia can dedifferentiate into Müller glia-derived progenitor cells (MGPCs), acquiring a progenitor-like phenotype and starting to proliferate, thus contributing to retinal repair . Notably, Müller glia in lower vertebrates exhibit a remarkable ability to regenerate retinal neurons, contrasting with the limited regenerative capacity seen in mammals, including humans , where Müller cells typically respond to injury with reactive gliosis, leading to scarring rather than regeneration . The SHH signaling pathway is essential for the proper development of all vertebrate retinas , and several studies have demonstrated the involvement of SHH signaling in the proliferation and differentiation of MGPCs, contributing to retinal regeneration in lower vertebrate . However, in mammals, the SHH pathway alone may not be sufficient to overcome the intrinsic limitations of Müller cells to regenerate neuronal cells. In our study, we observed an increase in SHH expression in NG MIO-M1cells exposed to sustained high-glucose and GF treatments. This suggests that these cells are engaging in pathways associated with cellular dedifferentiation and potential reprogramming towards a progenitor state. The increased SHH protein expression in NG cells was accompanied by significant changes in the intracellular localization of SHH, exhibiting a more spotted/punctate pattern in the cytoplasm and in close proximity to the plasma membrane. These findings suggest an enhanced state of cellular activity with increased synthesis and potential accumulation of SHH, aligning with the findings in lower vertebrates and rodents, where increased SHH signaling plays a role in Müller cell reprogramming and retinal regeneration . Furthermore, significantly higher levels of SHH were observed in HG MIO-M1 cells compared to NG MIO-M1 cells. Conversely, HG MIO-M1 cells exposed to GF treatments showed a decrease in SHH protein expression levels. A trend toward a decrease was also observed in response to sustained high-glucose. This finding is consistent with a previous study in which the authors observed a decrease in SHH signaling in reactive astrocytes of the cerebral cortex after acute, focal injury, particularly in cells proximal to the lesion site . This highlights that the negative regulation of SHH activity in astrocytes is context-dependent and varies with the degree of cellular damage. In our study, the decreased SHH levels observed in HG MIO-M1 cells exposed to GFs indicate that the severity of metabolic stress may influence SHH signaling pathways through a similar mechanism. These observations underscore the notion that glial cells, including Müller cells and astrocytes, exhibit differential SHH responses depending on the extent of glucose stress, highlighting the dynamic regulation of SHH signaling in glial cells. Although the SHH level has decreased, the cells that express it maintain a dot-like distribution of SHH inside the cells. Our findings suggest that, under NG condition, exposure to sustained high-glucose and GFs stimulates SHH expression, potentially promoting the dedifferentiation and reprogramming of Müller cells. However, in HG-adapted cells, additional metabolic stress from GF treatments leads to decreased SHH expression, possibly impairing regenerative capacity and enhancing gliotic responses. These observations indicate that the severity and fluctuation of glucose stress influence SHH signaling pathways, affecting the balance between neuroprotection and gliosis in the retina. A previous study demonstrated that high-glucose conditions have been associated with increased SOX2 levels in human Müller glial cells, which may support cell survival and regeneration under stress conditions . In our experiments, SOX2 was significantly upregulated in HG MIO-M1 cells, compared to NG MIO-M1 cells. Furthermore, NG MIO-M1 cells have the capacity to modulate this gene when exposed to sustained high-glucose and GF treatments, exhibiting a consistent upregulation of SOX2. These findings are consistent with the observed upregulation of SHH in NG MIO-M1 cells exposed to different glucose treatments. Unlike NG MIO-M1cells, HG MIO-M1 cells exposed to the same glucose treatments do not show variations in SOX2 expression. This different behavior of HG MIO-M1 cells, with regards to SOX2 and SHH expression, may be due to the different roles and regulatory mechanisms in the glial cells of these two genes. SOX2 is a transcription factor crucial for maintaining stemness and promoting cell progenitor proliferation, and its upregulation could be a response to cellular stress in order to maintain or enhance regenerative capacity. In contrast, SHH signaling, which is involved in cell differentiation and tissue patterning, may be more sensitive to metabolic perturbations with a more dynamic regulation, which can lead to its downregulation under glucose fluctuations treatments. However, despite the observed upregulation of SHH and SOX2 in NG MIO-M1 cells, the expression of rhodopsin, a marker of mature neuronal cells, was not detected. This indicates that, although the cells possess the potential to differentiate into neuronal cells, the conditions employed in our protocol were not sufficient to fully induce this differentiation. The lack of mature neuronal marker expression is likely due to the timing and duration of the experimental conditions. It may therefore be worthwhile to optimize the protocol in order to more accurately reflect the diabetic conditions in humans, with a view to promoting full neuronal differentiation. Given the role of SHH in Müller cells reprogramming and the observed upregulation under sustained high-glucose and GF treatments, future experiments could investigate the effect of exogenous SHH treatment on human Müller cells. This approach would provide insights into the therapeutic potential of modulating SHH signaling to mitigate the adverse effects of reactive gliosis and enhance regenerative processes in the diabetic retina. Although the current study provides important insights into the gliotic and reprogramming potential of Müller cells under different glucose treatments, some limitations should be considered. The study employed the MIO-M1 cell line, which, although well established in human retinal research, may not fully capture the physiological complexity of Müller cells found in human retina. Future research could benefit from using primary Müller cells to better reflect the in vivo context of human diabetic retinopathy. In addition, the effects of other stressors, such as oxidative stress or hypoxia, which also play a role in diabetic retinal damage, were not included in our model. Inclusion of these factors could provide a more comprehensive understanding of the cellular mechanisms involved. Despite these limitations, the current study contributes important insights into the gliotic and reprogramming potential of Müller cells under varying glucose conditions. The Müller cell line (MIO-M1; YB-H3309, Ybio, Shanghai, China) was maintained under controlled conditions at 37 °C and cultured in Dulbecco’s Modified Eagle Medium (DMEM) (11966-025, Gibco, Grand Isle, VT, USA). The culture medium was supplemented with 10% fetal bovine serum (FBS; Gibco), 5 mg/mL streptomycin, 5 U/mL penicillin (Gibco), and 5 mM (1 g/L) glucose or 25mM (4.5 g/L) glucose. The experimental conditions, referred to as normoglycemic (NG) at 5 mM glucose and hyperglycemic (HG) at 25 mM glucose, represent in humans the normal physiological levels of glucose (100 mg/dL) and the hyperglycemic conditions, respectively . In order to evaluate the influence of sustained high-glucose and glucose fluctuations (GFs) on cellular function, MIO-M1 cultured in NG and HG conditions were plated at a density of 10,000 cells/cm and exposed to different glucose treatments over 96 h, with media changes every 24 h. Treatments were as follow: Treatment I = constant basal glucose medium (5 mM for NG cells and 25 mM for HG cells); Treatment II = sustained high-glucose medium (25 mM for NG cells and 45 mM for HG cells); Treatment III = alternating basal (5 mM for NG cells and 25 mM for HG cells) and high-glucose medium (25 mM for NG cells and 45 mM for HG cells) every 24 h; Treatment IV = basal glucose medium (5 mM for NG cells and 25 mM for HG cells) for 72 h followed by high-glucose medium (25 mM for NG cells and 45 mM for HG cells) for the last 24 h; Treatment V = alternating low- (3 mM for NG cells and 5 mM for HG cells) and high-glucose (25 mM for NG cells and 45 mM for HG cells) medium every 24 h. A detailed scheme of the cell culture conditions and treatments is provided in Supplementary Figure S1. MIO-M1 cell lysates were prepared with RIPA buffer (25 mM Tris–HCl pH 7.5, 150 mM NaCl, 1% NP-40, 0.1% sodium dodecyl sulphate (SDS), 1% sodium deoxycholate, 10 mM sodium fluoride (NaF), 1 M phenylmethylsulphonyl fluoride (PMSF), 1 M sodium vanadate (NaVO3), containing EDTA-free protease inhibitor cocktail (Sigma-Aldrich, St. Louis, MO, USA) and PhosSTOP™ (Roche, Penzberg, Germany). The samples were incubated for 30 min at 4 °C, centrifuged at 13,000 rpm for 10 min at 4 °C, and then sonicated once (5 s, 10% power) . The protein concentration was determined using the Bradford assay. Protein samples were separated by electrophoresis on 12% (v/v) sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS–PAGE) gels and transferred to a polyvinylidene difluoride (PVDF) Transfer Membrane Hybond™ (Amersham Biosciences, Amersham, UK). Membranes were then blocked with 5% non-fat dry milk in phosphate buffered saline (PBS) containing 0.05% (v/v) Tween 20 (PBS-T) for 1 h at room temperature, followed by overnight incubation at 4 °C with the primary antibodies. The following primary antibodies were used: rabbit anti-GFAP (ab7260, Abcam, Cambridge, MA, USA, 1:5000); goat anti-SHH (ab240438, Abcam 1:250); mouse anti-HSP90 (ab79849, Abcam, 1:1000); rabbit anti-α-tubulin (SC-9104, La Santa Cruz Biotechnology, Dallas, TX, USA, 1:500) diluted with 5% BSA in PBS-T. Secondary HPR-conjugated antibodies (Amersham Biosciences, Uppsala, Sweden) diluted 1:5000 in 1% (w/v) non-fat dry milk in PBS-T were used to incubate the membrane for 1 h at room temperature. Immunoreactive bands were detected by Amersham™ ECL™ Prime (Amersham Biosciences), according to the manufacturer’s protocol. Chemical luminescent signals were detected using ImageQuant LAS 4000 mini (GE Healthcare, Chicago, IL, USA). Densitometric analysis of the bands was performed using ImageJ software v1.53k. For normalization of protein expression, membranes were stripped using the Re-Blot Plus Mild Solution (10×) (Merck KGaA, Darmstadt, Germany), following the manufacturers’ protocol, and re-probed with rabbit anti-β-tubulin, mouse anti-HSP90. MIO-M1 cells grown on a 96-well plate were treated as indicated above. After washing with PBS, they were fixed with 4% paraformaldehyde (PFA) in PBS for 15 min and then permeabilized with 0.2% Triton X-100 in PBS for 10 min. Cells were incubated with blocking buffer (3% BSA and 0.05% Tween-20 in PBS) for 1 h at room temperature, then incubated overnight at 4 °C with the following antibodies: rabbit anti-GFAP antibody (ab7260, Abcam, 1:500); goat anti-SHH antibody (ab240438, Abcam; 1:50); mouse anti-SOX2 antibody (sc-365823, La Santa Cruz Biotechnology, 1:500); rabbit anti-Vimentin antibody (ab45939, Abcam; 1:200). Subsequently, cells were incubated with the following secondary antibodies: anti-rabbit conjugated with Alexa-Fluor-488 or Alexa-Fluor-568, anti-mouse conjugated with Alexa-Fluor-568 and with an anti-goat conjugated with Alexa-Fluor-568 (Invitrogen, Carlsbad, CA, USA, 1:500) diluted in PBS-T for 1 h at room temperature. Following the staining of the nuclei with Hoechst 33242 dye (0.5 µg/mL, Invitrogen, Carlsbad, CA, USA) diluted in PBS, the cells were examined using a Leica DM6000 B (Leica Microsystems, Wetzlar, Germany). Images were captured using the LAS AF acquisition software (2.6.0.7266, Leica Microsystems). For the quantification of MIO-M1 cells, 20 fields at 20× magnification were examined for GFAP, SOX2, and SHH in all glucose conditions. Immunoreactive cells were counted in 3 independent experiments using the ImageJ program v1.53k; data were obtained by counting at least 500 cells for each group in at least three independent experiments. RNA was extracted from mouse tissues using Trizol Protocol (TRIzol™ Reagent, Invitrogen, ThermoFisher Scientific, Catalog number: 15596026, Waltham, MA, USA), according to the manufacturer’s recommendations. An amount of 1 μg of RNA was reverse transcribed using random primers and the QuantiTect Reverse Transcription Kit (Qiagen, Hilden, Germany), following the manufacturer’s specifications. Gene expression was measured using iTaq Universal SYBR Green Supermix (Biorad Laboratories, Hercules, CA, USA). Real-time PCR was performed in the LightCycler 96 Real-Time PCR System (Roche Diagnostics GmbH, Mannheim, Germany). We used the following primers for the SOX-2 gene: SOX-2-F: GCTACAGCATGATGCAGGACCA; SOX-2-R: TCTGCGAGCTGGTCATGGAGTT; RHODOPSIN-F: AGCTCGTCTTCACCGTCAAGGA; RHODOPSIN-R: CCAGCAGATCAGGAAAGCGATG. The GAPDH gene was used as a housekeeping gene for all experimental samples: GAPDH-F: TCGGAGTCAACGGATTTGGT; GAPDH-R: GAATTTGCCATGGGTGGAAT. Gene expression levels were calculated using the 2 method. All the results are expressed as the mean ± SEM (standard error of the mean) of at least three independent experiments. Statistically significant differences were assessed using Prism 6.05 (GraphPad PRISM Software, Inc., La Jolla, CA, USA) with Student’s t-test for statistical comparison between groups. Differences between means were considered statistically significant when p-values were at least <0.05. The results showed that MIO-M1 Müller cells exhibit distinct responses to different glucose treatments, which are strongly dependent on their metabolic environment. This differential response could have implications for the onset and progression of diabetic retinopathy. The increased levels of activation and dedifferentiation markers observed in NG MIO-M1 cells in response to different glucose treatments suggest a protective response that may occur in early or well-controlled diabetes. In contrast, the lack of response observed in HG MIO-M1 cells exposed to different glucose treatments may reflect an exhausted gliotic and reprogramming capacity, which could contribute to the development and progression of diabetic complications such as retinal neurodegeneration and diabetic retinopathy observed in uncontrolled diabetic patients exposed to prolonged metabolic glucose stress. Although our results provide valuable insights, further in-depth studies are needed to elucidate the mechanisms involved and to explore the potential therapeutic implications for mitigating retinal neurodegeneration associated with diabetes. Extending these studies to primary Müller cells in future research would provide a more physiologically relevant system, helping to validate and refine our findings. This approach would provide a deeper understanding of the cellular response to sustained high-glucose and glucose fluctuations and could improve the development of targeted therapeutic strategies for retinal pathologies in human patients.
PMC12133578
Immune–epithelial–stromal networks define the cellular ecosystem of the small intestine in celiac disease
The immune–epithelial–stromal interactions underpinning intestinal damage in celiac disease (CD) are incompletely understood. To address this, we performed single-cell transcriptomics (RNA sequencing; 86,442 immune, parenchymal and epithelial cells; 35 participants) and spatial transcriptomics (20 participants) on CD intestinal biopsy samples. Here we show that in CD, epithelial populations shifted toward a progenitor state, with interferon-driven transcriptional responses, and perturbation of secretory and enteroendocrine populations. Mucosal T cells showed numeric and functional changes in regulatory and follicular helper-like CD4 T cells, intraepithelial lymphocytes, CD8 and γδ T cell subsets, with skewed T cell antigen receptor repertoires. Mucosal changes remained detectable despite treatment, representing a persistent immune–epithelial ‘scar’. Spatial transcriptomics defined transcriptional niches beyond those captured in conventional histological scores, including CD-specific lymphoid aggregates containing T cell–B cell interactions. Receptor–ligand spatial analyses integrated with disease susceptibility gene expression defined networks of altered chemokine and morphogen signaling, and provide potential therapeutic targets for CD prevention and treatment. Subject terms: Translational immunology, Coeliac diseaseCeliac disease (CD) is a common gastrointestinal disorder affecting 1–2% of European and North American populations, in which small intestinal inflammation and damage are driven by aberrant adaptive immune responses to gluten. The only treatment is a lifelong gluten-free diet (GFD). There is an unmet therapeutic need for those living with CD, including refractory CD, where ongoing tissue damage occurs despite a GFD. A strong genetic component drives CD, dominated by HLA-DQ2 and HLA-DQ8 (ref. ), with association studies identifying over 40 non-HLA genomic loci, implicating over 100 candidate genes and a role for immunoregulatory mechanisms. Murine models implicate viral infection as a trigger of loss of tolerance driving CD pathogenesis, a hypothesis supported by epidemiological studies. CD pathophysiology is multifactorial with several cell types implicated. Dietary gluten is deamidated by tissue transglutaminase 2, and deamidated gluten peptides presented via HLA-DQ2/HLA-DQ8 to CD4 T cells. Gluten-specific CD4 T cells possess a distinct type 1 helper T (TH1)/follicular helper T (TFH) cell phenotype, emphasizing the importance of T cell–B cell interactions. Tissue plasma and B cells may present gluten peptides via HLA-DQ. Subsequent stimulation of disease-specific plasma cells drives anti-tissue transglutaminase and anti-deamidated gliadin peptide antibody production. Gluten-specific T cells are necessary but not sufficient to generate mucosal damage. The mechanisms by which this response leads to tissue architectural change are incompletely understood. Intraepithelial lymphocytes (IELs), mainly CD8 T IELs, are highly enriched in CD, likely driven by epithelial and myeloid-derived interleukin (IL)-15, in combination with CD4 T cell-derived IL-2, IL-21 and interferon gamma (IFNγ). IELs may be directly involved in EC killing in a T cell antigen receptor (TCR)-independent manner, via NKG2C and NKG2D and their epithelial ligands MICA and HLA-E. However, the transcriptional state and involvement of TCR signaling in these CD8 T cell populations remains unclear. While novel treatments are under development, recent therapeutic trials targeting gluten degradation, gluten-specific CD4 T cell tolerance and IL-15 have been unsuccessful. However, therapies including tissue transglutaminase inhibitors and inducers of immune tolerance have shown promise. Single-cell transcriptomics have redefined cellular landscapes in the gastrointestinal tract, offering insights into CD immunopathology. Recent studies have sought to understand the cellular basis of CD using mass cytometry, including studies of refractory CD, gluten-specific T cells, and mucosal and circulating T cells. Single-cell RNA sequencing (scRNA-seq) has been used to study mucosal immune cells, T cells, circulating immune cells and mucosal plasma cells. Here, we combined single-cell and spatial transcriptomics to define the network of intestinal immune, epithelial and parenchymal cell populations in adults and children with CD. Our description of spatially localized immune–parenchymal interactions driving inflammation and remodeling of the mucosa, and with specific disease-associated T cell subsets occupying distinct mucosal niches, will facilitate identification of therapeutic targets. We generated scRNA-seq profiles of duodenal epithelial, immune and parenchymal populations from 35 participants: 21 with CD (16 children, 5 adults) and 14 controls (5 children, 9 adults; Fig. 1 and Supplementary Table 1). We used complementary single-cell techniques for adult and pediatric datasets, with 86,442 cells sequenced. In adults (datasets 1 and 3), we performed scRNA-seq (10x Genomics) on epithelial, immune (Supplementary Fig. 1a,b), stromal and endothelial cells. In children (dataset 2), we performed targeted scRNA-seq (BD Rhapsody; 504 targeted gene primer pairs) and surface protein expression (79 oligonucleotide-conjugated antibodies) on intestinal immune cells (Supplementary Fig. 1c,d and Supplementary Tables 2 and 3). Schematic of scRNA-seq, RNA-seq, TCR-seq, spatial transcriptomics, and flow cytometry experiments and datasets. Dataset 1: ECs and total mucosal CD45 cells were isolated from intestinal biopsy samples before scRNA-seq library preparation using the 10x Genomics platform. Dataset 2: total mucosal CD45 cells were isolated from intestinal biopsy samples before combined targeted scRNA-seq and multiplex surface antibody characterization using the BD Rhapsody platform. Dataset 3: scRNA-seq (10x Genomics) was performed on intestinal stromal and endothelial cells. Datasets 4 and 5: OCT-embedded frozen duodenal biopsy samples were sectioned and used for spatial transcriptomics (10x Visium). Datasets 6 and 7: mucosal CD8 T cells were isolated before bulk RNA-seq and TCR-seq. Dataset 8: mucosal CD8 and γδ T cells were isolated before scRNA-seq library preparation using the 10x Genomics platform. Dataset 9: flow cytometry of circulating CD8 T cells. Study participant numbers and disease characteristics, as well as cell numbers after the quality-control pipeline, are indicated. ILC, innate lymphoid cell; HC, healthy controls; ACD, active celiac disease; TCD, treated celiac disease. We analyzed EPCAM epithelial populations from dataset 1. Nine transcriptionally distinct epithelial cell (EC) clusters were identified, representing progenitor, secretory and absorptive lineages along the developmental progression of the crypt–villus axis (Fig. 2a,b, Extended Data Fig. 1a and Supplementary Table 4). BEST4 enterocytes (BEST4CA7CPA2), first identified in the colon, were seen, expressing CFTR and showing chloride channel activity (Fig. 2b and Extended Data Fig. 1a). Goblet cells (ITLN1MUC2SPINK4) and tuft cells (PLCG2TRPM5IRAG2) were also identified. a, UMAP plot of small intestinal epithelial EPCAM cells in HCs (n = 3) and in participants with CD (n = 5). b, Bubble plot showing the expression of selected genes defining specific cluster identities. Scaled gene expression indicated by color; proportion of cells expressing the gene indicated by bubble size. c, Local neighborhood enrichment of EPCAM cells in ACD versus HCs. Color indicates enrichment (log fold change (FC)) of cells in ACD versus HCs in that UMAP neighborhood; size of dot indicates false discovery rate (FDR)-adjusted −log10 values. d, TA cells (left) and early enterocytes (right) in HCs and CD, as a proportion of total EPCAM cells. e, Pseudotime trajectory of gene expression of EPCAM ECs, colored by pseudotime axis (left), cluster identity (middle) and lineage (right). Arrows indicate putative direction of cell differentiation. f, Density of cells along pseudotime trajectory axis split by disease state: ACD (red), TCD (blue) and HCs (gray). g, Smoothed heat map showing expression of selected genes related to intestinal absorption along pseudotime trajectories relating to secretory (toward left) and absorptive (toward right) lineage. h, Volcano plot displaying differentially expressed gene transcripts between HCs and ACD in total ECs. d, Unpaired two-tailed t-test. Data are presented as mean values ± s.e.m. (a) Bubble plot showing the expression of selected GO term gene sets defining specific cluster identities and functions. Gene set expression indicated by colour. (b) Local neighbourhood enrichment of EPCAM cells in treated CD vs healthy controls (HC). Colour indicates enrichment (log fold change) of cells in active CD vs HC in that UMAP neighbourhood, size of dot indicates –log10FDR. (c) Transit amplifying cells in HC (n = 3), active CD (n = 3), and treated CD n = 2), as a proportion of total EPCAM cells (mean ± SEM). (d) Proportion of EPCAM cells in predicted cell cycle states (G1, G2M, S), stratified by disease state. (e) UMAP overlay of predicted cell cycle states (G1, G2M, S), stratified by disease state. (f) Violin plot of CCL25 expression by epithelial cell type. (g) Expression of GO term gene sets associated with metabolic and absorptive function in mature enterocytes, stratified by disease state. (h) UMAP overlay of expression of selected GO term gene sets associated with metabolic and absorptive function, stratified by disease state. (i) Violin plots of expression of genes associated with absorptive function in mature enterocytes, stratified by disease state. (j–l) Volcano plots displaying differentially expressed gene transcripts between HC and active CD in the three cell lineages: epithelial progenitors (i), absorptive cells (j), and secretory cells (k). (m, n) Venn diagrams showing shared differentially expressed GO term gene sets between progenitor, absorptive, and secretory lineages, either upregulated (l) or downregulated (m) in active CD compared to HC. A LYZ Paneth cell-like population (MMP7REG1ASOD3PLA2G2A) was also identified (Fig. 2a,b), although defensin gene expression was not detected. This population expressed PGC, mucins including MUC5AC, MUC1 and MUC6 and AQP5, suggesting it also contained Brunner’s gland cells or ectopic gastric pyloric gland cells. This cell type was enriched in active celiac disease (ACD; Fig. 2c,d), perhaps in response to IFNγ. Thus, this population could represent inflammation-driven gastric cell metaplasia. Transit-amplifying (TA) cells were increased in CD, along with enrichment of uniform manifold approximation and projection (UMAP) areas corresponding to EC progenitors (stem cells, TA cells and early enterocytes; Fig. 2c,d). This persisted in treated celiac disease (TCD; Extended Data Fig. 1b,c). In parallel, more actively cycling ECs were observed in ACD and TCD (Extended Data Fig. 1d,e). Pseudotime analyses identified epithelial developmental trajectories, from undifferentiated progenitor states toward absorptive and secretory lineages (Fig. 2e). In CD, ECs were shifted to earlier pseudotime states, with loss of mature ECs (Fig. 2f). CCL25, encoding the ligand for CCR9 (implicated in CD pathogenesis), was expressed predominantly by progenitor cells (Fig. 2b and Extended Data Fig. 1f). We examined putative EC functions through functional gene-set analysis (Extended Data Fig. 1a), identifying functions of secretory Paneth-like/Brunner’s gland cells (secreted protein and vesicle pathways), BEST4 enterocytes (chloride/anion channel activity), tuft cells (taste perception) and enteroendocrine cells (EECs; peptide hormone processing/secretion). Mature enterocytes expressed key metabolic and macronutrient catabolic pathways, and active transport and absorption mechanisms. Early ECs and TA cells did not express these pathways. Absorptive function genes were limited to cell states at the end of absorptive epithelium pseudotime trajectories, consistent with EC development along the crypt–villus axis (Fig. 2g). Notably, gene sets related to lipid, carbohydrate, cholesterol, vitamin and iron processing and absorption were all downregulated in mature enterocytes in ACD (Extended Data Fig. 1g–i). These transcriptional changes normalized in TCD, although some pathways, including fructose metabolism and lipid catabolism, remained reduced (Extended Data Fig. 1h). Overall, absorptive capacity is reduced in ACD not simply by reduction in villus surface area, but through a relative increase of EC progenitors lacking absorptive machinery, and pathway downregulation in mature enterocytes. ECs in ACD upregulated multiple antigen-presentation molecules, including classical HLA class I and class II genes (except HLA-DQ) and nonclassical genes including HLA-E and HLA-F (Fig. 2h). Interferon-stimulated genes (types I and II) dominated the epithelial response, including STAT1 (Fig. 2h and Supplementary Table 5). The major disease-associated responses were observed in all EC lineages (Extended Data Fig. 1j–l), including antigen-presentation pathways, type I/II interferon responses, lymphocyte-mediated immunity and cytotoxicity and cell adhesion regulation (Extended Data Fig. 1m,n). Some transcriptional changes were cell-type specific. IL32 was highly expressed in ACD by mature enterocytes (Extended Data Fig. 1k), perhaps regulated by interferons. The reduction of fatty acid catabolism/transport (APOA1, FABP2), metal ion transport (iron: FTH1, FTL; zinc: SLC39A4) and carbohydrate metabolism (ALDOB, PCK1) was restricted to absorptive lineages, mainly mature enterocytes (Extended Data Fig. 1k,n). Progenitor cells upregulated genes associated with cell division and differentiation, and downregulated those associated with tissue repair and homeostasis (Extended Data Fig. 1m,n). Secretory lineages showed increased expression of gut hormone genes, LYZ, and chemokines (CXCL17, CXCL2; Extended Data Fig. 1l). The duodenum, where CD inflammation predominates, has sensory and neurohormonal functions. We extended EEC clustering, revealing multiple transcriptional states, including NEUROG3 progenitors and EEC subtypes, which showed similar CD-related transcriptional changes to other ECs (Extended Data Fig. 2). EEC proportions altered in CD, with increases in NEUROG3 progenitor cells and somatostatin-producing D cells (Extended Data Fig. 2i–k). (a) UMAP plot of intestinal enteroendocrine cells (EECs) in healthy controls (HC) and celiac disease (CD) (n = 8). (b) UMAP plots overlaid with expression of genes identifying enteroendocrine cell types. (c) Bubble plot showing the expression of selected genes defining specific cluster identities. Gene expression indicated by colour, proportion of cells expressing the gene indicated by bubble size. (d–g) UMAP overlay of expression of selected GO term gene sets associated with peptide hormone secretion and enteroendocrine differentiation, stratified by disease state. (h) Volcano plot displaying differentially expressed gene transcripts between HC and active CD in enteroendocrine cells. (i) Local neighbourhood enrichment of enteroendocrine cells in treated CD vs HC. Colour indicates enrichment (log fold change) of cells in active CD vs HC in that UMAP neighbourhood, size of dot indicates –log10FDR. (j) Volcano plot of enteroendocrine subset proportion (log2 fold change) between CD and HC. (k) NEUROG3+ progenitors and D cells in HC (n = 3) and CD (n = 5), as a proportion of total enteroendocrine cells (mean ± SEM). Unpaired two-tailed t-test. In adults (dataset 1), CD4 T cells formed subsets dominated by TH1-polarized and IL-17-producing helper T (TH17)-polarized effectors, as well as small naive and FOXP3 regulatory populations (Fig. 3a–c and Supplementary Table 6). There was a cluster of TFH-like CD4 T cells expressing PDCD1, BTLA, CD28, ICOS and intermediate CXCR5. Dataset 2 (pediatric) contained analogous subsets (Extended Data Fig. 3a), including CD31CR2 recent thymic emigrants, a CCR7 TFH-like subset and the TFH-like subset expressing PD1, ICOS, CTLA4, BTLA and CD161 at the protein level (Fig. 3d,e). a–c, Intestinal CD4 T cells in health and CD in dataset 1 (adult—10x Genomics). a, UMAP plot of intestinal CD4 T cells in health and CD (n = 8). b, Bubble plot showing the expression of selected genes defining specific cluster identities. Scaled gene expression indicated by color; proportion of cells expressing the gene indicated by bubble size. c, CD4 T cell UMAP plots overlaid with expression of TNFSF8, PDCD1, TOX2, CXCR3, CXCL13, CD200, CXCR5 and TRBV7-2. Intestinal CD4 T cells in health and CD in dataset 2 (pediatric—BD Rhapsody; d–f). d, UMAP plot of intestinal CD4 T cells in health and CD (n = 15). e, Bubble plot showing the expression of selected genes and proteins defining specific cluster identities. Scaled gene/protein expression indicated by color; proportion of cells expressing the gene/protein indicated by bubble size. f, Local neighborhood enrichment of CD4 cells in ACD versus HCs (dataset 1). Color indicates enrichment (log fold change) of cells in ACD versus HCs in that UMAP neighborhood; size of dot indicates −log10FDR. g, Scatterplot of mean proportion (± s.e.) of CD4 T cell clusters in HCs (n = 3) versus ACD (n = 5) in dataset 1. Clusters above the line of unity are enriched in ACD. h,i, Treg (h) and TFH (i) CD4 T cell populations in HCs and CD, as a proportion of total CD4 T cells in dataset 1 (HCs n = 3, ACD n = 5) and dataset 2 (HCs n = 5, ACD n = 10). j, UMAP plot of CD4 T cells in dataset 2, overlaid with IL21 and IFNG expression. k, UMAP plot of CD4 T cells in dataset 1, overlaid with CXCL13, IL21, IFNG and TNFSF8 expression. h,i, Two-sided Mann–Whitney test. Data are presented as mean values ± s.e.m. Ab, antibody; Tc17, IL17CD8 T cells; DP, CD4CD8 double positive cells. (a) Sankey plot showing predicted analogous CD4 T-cell clusters between Dataset 1 and Dataset 2 datasets. (b) Bubble plot showing the expression of selected genes associated with CD4 T-cells clusters, including genes previously associated with gluten-specific CD4 T-cells (Christophersen et al, 2019). Gene expression indicated by colour, proportion of cells expressing the gene indicated by bubble size. (c) Bubble plot of expression of chemokine, cytokine, and TNF family member genes by CD4 T-cell clusters in Dataset 2 dataset. Gene expression indicated by colour, proportion of cells expressing the gene indicated by bubble size. (d) Bubble plot showing expression of canonical Tfh and Th17 gene signatures in CD4 + T cell clusters. (e) Heatmap of transcription factor (TF) regulon expression by CD4 T-cell clusters in Dataset 1. (f) Transcription factor gene expression by CD4 T-cell clusters in Dataset 2. (g) UMAP plots of CD4 T-cells in Dataset 1, overlaid with expression of selected TF regulons. This TFH-like population in adults and children showed similar phenotypic profiles to those of gut-resident gluten-specific CD4 T cells in CD (Extended Data Fig. 3b), and expressed TOX2, CD200, IL21 and CXCL13. The cluster showed enrichment of TRBV7-2, a V-gene enriched in gluten-specific CD4 T cell HLA-DQ2.5 TCR repertoires. Treg and TFH-like CD4 T cells were increased in ACD in adults and children (Fig. 3f–i). T cell populations showed distinct cytokine and chemokine expression patterns (Extended Data Fig. 3c). The CD-associated TFH-like population, showed high CXCL13 and IL21 expression, with IFNG and IL21 coexpression (Fig. 3j,k), similarly to gluten-specific T cells. TFH-like cells expressed TNFSF8, CCL1, CCL22 and CXCL10, as well as IL17F (Extended Data Fig. 3c). IL17F expression was not seen in the IL17ARORCIL23RCCR6 TH17 population, nor did the TH17 cluster show TRBV7-2 enrichment (Extended Data Fig. 3b,d). Oral gluten challenge in CD drives rapid circulating cytokine responses, including IL-2, CXCL8, CXCL10 and IL-6 (ref. ). CXCL8 expression was highest in CCR7 TFH CD4 T cells, CXCL10 was detected in TFH-like CD4 T cells, while IL6 was detected in Treg cells (Extended Data Fig. 3c). IL2 expression was low within the CD4 compartment, as expected without gluten challenge. We examined transcription factor (TF), and regulon expression within CD4 subsets, with canonical TFs and regulons of TH17 and Treg cell function expressed as expected (Extended Data Fig. 3e–g). IKZF1 and its regulon were upregulated in TFH-like cells, with intermediate expression of RUNX1, BATF and IRF3. We examined B cell lineages in dataset 2 (pediatric; Extended Data Fig. 4a,b). Both IgA and IgM plasma cells were increased in CD (Extended Data Fig. 4c–f). A population of CXCR5 B cells (MS4A1CD19CD20) were present, with a shift toward the CD27 memory B cell phenotype in CD. B cell and plasma cell clusters were examined in the Dataset 2 (pediatric) dataset (BD Rhapsody) in healthy controls (HC) (n = 5) and CD (n = 10). (a) UMAP plot of intestinal B and plasma cell clusters. (b) Bubble plot showing the expression of selected genes and surface proteins defining specific cluster identities. Gene/protein expression indicated by colour, proportion of cells expressing the gene indicated by bubble size. (c) Intestinal B cell and plasma cell frequencies in HC and CD, as a proportion of total CD45 cells. (d) Intestinal B cell and plasma cell cluster frequencies in HC and CD, as a proportion of total CD45 cells. (e) Proportion of intestinal CD27- and CD27 + B cells in HC and CD, as a proportion of total CD45 cells. (f) Proportion of intestinal IgA and IgM plasma cells in HC and CD, as a proportion of total CD45 cells. (g) Bubble plot showing the expression of genes and proteins related to MHC class II expression. (h) UMAP plots overlaid with the expression of genes CD74 and HLA-DQB1, and the surface protein expression of HLA-DQA1/A2. (c-f) Data presented as mean ± SEM. (f) Unpaired two-tailed t-test. Gene signatures of age-related B cells (an inflammation-associated population in autoimmune disease), including ITGAM, ITGAX, CD86 and BATF, were expressed most highly in CD27 B cell populations, while a key age-related B cell TF, TBX21, was highly expressed in cycling B cells (Extended Data Fig. 4b). HLA class II gene and protein expression, specifically HLA-DQ, was highest in CD27 and cycling B cells (Extended Data Fig. 4g,h). Intestinal myeloid cell populations are impacted by CD and may be involved in antigen presentation and oral tolerance. Myeloid cells (dataset 2) formed 11 transcriptionally distinct clusters, including macrophages, conventional dendritic cells and plasmacytoid dendritic cells (Supplementary Fig. 2a–c). HLA-DQ expression was highest on macrophage populations, particularly CD163 cells. In contrast to prior studies, CD163 macrophages were reduced in ACD, with expansion of a conventional dendritic cell 2 population, which showed increased IL-1B expression (Supplementary Fig. 2d,e). Intestinal CD8 T cells showed considerable heterogeneity in transcriptional states, with multiple tissue-resident memory CD8 T (TRM) cells, including an ITGAEIL7R population, a CCL4CD69ITGAE population and two subsets of ITGAE TRM cells (Fig. 4, Extended Data Fig. 5a and Supplementary Table 7). These aligned with gene signatures defining subsets of bona fide human TRM cells. FGFBP2 effectors aligned with previously described ITGB2ITGAE TRM cells, while TRM(1), TRM(2) and cycling subsets aligned with CD103 TRM cells (Extended Data Fig. 5b). CCL4 and IL7R populations likely represent intermediate states in TRM cell development. Small natural IEL and cycling MKI67 populations were seen (Extended Data Fig. 5b,c). Analogous CD8 T cell subsets were seen in dataset 2 (Fig. 4d,e and Extended Data Fig. 5a), with additional resolution for tissue-resident γδ T cells, and innate-like T cells (mucosal-associated invariant T cells and Vδ2Vγ9 cells). a–c, Dataset 1 intestinal CD8 T cells in health and CD (adult—10x Genomics). a, UMAP plot of intestinal CD8 T cells in health and CD (n = 8). b, Bubble plot showing the expression of selected genes defining specific cluster identities. Scaled gene expression indicated by color; proportion of cells expressing the gene indicated by bubble size. c, UMAP plots overlaid with expression of IL7R, GZMK, ITGAE, CXCR6, GZMA, LAYN, ENTPD1, TNFRSF9, TIGIT and HLA-DRB1. Dataset 2 intestinal CD8 T cells in health and CD (pediatric—BD Rhapsody; d–f). d, UMAP plot of intestinal CD8 T cells in health and CD (n = 15). e, Bubble plot showing the expression of selected genes and proteins defining specific cluster identities. Gene/protein expression indicated by color; proportion of cells expressing the gene/protein indicated by bubble size. f, Local neighborhood enrichment of CD8 cells in ACD versus HCs (dataset 1). Color indicates enrichment (log fold change) of cells in ACD versus HCs in that UMAP neighborhood; size of dot indicates −log10FDR. g, Scatterplot of mean proportion (± s.e.) of CD8 T cell clusters in HCs (n = 3) versus ACD (n = 5). Clusters above the line of unity are enriched in ACD. h,i, TRM(2) (h) and cycling (i) CD8 T cell phenotype populations in HCs and CD, as a proportion of total CD8 T cells in dataset 1 (HCs n = 3, ACD n = 5) and dataset 2 (HCs n = 5, ACD n = 10). h,i, Two-sided Mann–Whitney test. Data are presented as the mean values ± s.e.m. nIEL, natural intraepithelial lymphocyte. (a) Sankey plot showing predicted analogous CD8 T-cell clusters between Dataset 1 and Dataset 2 datasets. (b) Bubble plot showing the expression of selected genes associated with CD8 TRM cell subsets. Gene expression indicated by colour, proportion of cells expressing the gene indicated by bubble size. (c) UMAP plot of CD8 T-cells in Dataset 1, overlaid with MKI67, KLRC1, and KLRC2 expression. (d) Bubble plot showing the expression of KIRs and NKRs associated with previously described CD8 T-cell subsets in CD. (e) Natural IEL populations in HC (n = 3) and CD (n = 5), as a proportion of total CD8 T-cells in Dataset 1 (above) and Dataset 2 (below) (mean ± SEM). (f) Stacked plot of predicted CD8 T-cell cluster origin of cycling CD8 T-cells, stratified by disease state. (g) Volcano plot displaying differentially expressed gene transcripts between active and treated CD in TRM(2) CD8 cells. (h) Violin plot of IFNG expression in CD8+ clusters in Dataset 2, split by disease state. (i) Violin plot of IFNG expression in TRM(2) and cycling CD8+ clusters in Dataset 1, split by disease state. (j) Heatmap of transcription factor (TF) regulon expression by CD8 T-cell clusters in Dataset 1. (k) Transcription factor gene expression by CD8 T-cell clusters in Dataset 2. (l) UMAP plots of CD8 T-cells in Dataset 1, overlaid with expression of selected TF regulons. (e) Unpaired two-tailed t-test. We analyzed subsets relevant to CD, including natural killer (NK)-receptor expressing IELs and killer-cell immunoglobulin-like receptor (KIR)-positive CD8 T cells. KLRC1 (NKG2A) was expressed by CCL4 cells, while KLRC2 (NKG2C) was expressed by resident IL7R, TRM(1) and TRM(2) subsets (Extended Data Fig. 5c,d). Inhibitory KIR molecule expression was confined to a small FGFBP2 effector population. TRM(2) and cycling populations were enriched in ACD, but not TRM(1) cells (Fig. 4f,g). TRM(2) cells were rare in health, but increased to form 20–40% of CD8 T cells in ACD, which persisted in TCD (Fig. 4h). Natural IELs were reduced in ACD (Extended Data Fig. 5e). Cycling CD8 T cells increased to form 2–4% of cells in ACD (Fig. 4i). Most cycling cells showed a TRM(2) phenotype (Extended Data Fig. 5f). As TRM(2) CD8 T cells were increased in proportion and proliferating in ACD, we profiled them in depth (Fig. 5a–e and Extended Data Fig. 6). TRM(2) CD8 T cells showed a CD103 tissue-resident phenotype, high GZMA and absent GZMK expression, along with high expression of CXCR6, activation markers (HLA-DR) and genes expressing co-stimulatory and co-inhibitory molecules (TIGIT, TNFRSF9 (4-1BB), ENTPD1 (CD39) and LAYN (Fig. 4b,c). Comparison of TRM(2) cells in ACD versus TCD showed increased expression of activation markers and increased effector function with IFNG, GZMB and IL32 expression (Extended Data Fig. 5g). a, Pseudotime trajectory of gene expression of tissue-resident CD8 T cell clusters (dataset 1—adult), colored by pseudotime axis (left) and cell cluster (right). Arrows indicate direction of differentiation. b, Pseudotime trajectory, split by disease state, and colored by differentiation branch. The proportion of CD8 TRM cells differentiating down branches 1 and 2 in each disease state is indicated. c, Bubble plot of expression of chemokine, cytokine and TNF family member genes by CD8 T cell clusters in dataset 2 (pediatric). Scaled gene expression indicated by color; proportion of cells expressing the gene indicated by bubble size. d, UMAP plots of CD8 T cells in dataset 2 (pediatric), overlaid with IFNG, CCL20 and FASLG expression. e, TCR clonal overlap (Morisita–Horn) between CD8 T cell clusters in dataset 1. f, Volcano plot of TRBV segment usage within the TCR repertoire of TRM(2) cells between HCs and CD. Black, high-frequency TRBV segments used by >1% of total clones; gray, low-frequency TRBV segments used by <1% of total clones. g, Volcano plot of TRBV segment gene expression (left) and normalized expression of TRBV28 (right) in bulk RNA-seq data from sorted intraepithelial CD8 T cells (dataset 3; HCs n = 3, ACD n = 4, TCD n = 3, potential CD n = 2). h, Volcano plot of TRBV segment usage (left), and proportion of unique CDR3β clonotypes (right above) and proportion of top 100 most common clonotypes (right, below) using the TRBV28 V segment in bulk TCR-seq of CD8 mucosal T cells in HCs and CD (dataset 4; HCs n = 8, ACD n = 7, TCD n = 5). f, Negative binomial model without multiple comparisons. g, Negative binomial model with Benjamini–Hochberg multiple testing. h, One-way analysis of variance with Holm–Sidak’s multiple-comparisons test. (a, b) Single-cell TCR-seq of CD8 T-cells in Dataset 1. (a) Upset plot of T-cell clonotype overlap between CD8 T-cell clusters. (b) TRBV segment usage in TCR clonotypes in TRM(2) CD8 T-cells. (c–e) TCR-seq of sorted mucosal CD8 T-cells (n = 20; Dataset 4). (c) TRBV gene usage by unique CDR3 clonotypes in intestinal CD8 T-cell TCR repertoires in health and celiac disease. (d) TRBJ gene usage by TRBV28 (red) and non-TRBV28 (black) clonotypes in intestinal CD8 T-cell TCR repertoires in CD (n = 12) (mean ± SEM). (e) Amino acid usage within CDR3 sequences of TRBV28 (red) and non-TRBV28 (black) clonotypes in intestinal CD8 T-cell TCR repertoires in CD (n = 12) (median, IQR, range). (f) Proportion of unique CDR3β clonotypes using the TRBV28 V-gene in health (n = 5), active ulcerative colitis (n = 5), and irAE colitis (n = 11) (mean ± SEM). Data from Sasson and colleagues. (g) Proportion of unique paired CD8 clonotypes using the TRBV28 V-gene in health (n = 3) and active ulcerative colitis (n = 3) (mean ± SEM). Data from Corridoni and colleagues. (h) Proportion of unique paired T cell clonotypes using the TRBV28 V-gene in uninflamed (n = 11) and inflamed (n = 11) biopsies in ileal Crohn’s disease (mean ± SEM). Data from Martin and colleagues. (c, d, f) Two-way ANOVA with Holm-Sidak’s multiple comparisons test. (e) Multiple two-tailed t-tests with Benjamini, Krieger, and Yetutieli two-stage step up control of multiple comparisons with FDR < 0.01. (g,h) Mann-Whitney two-tailed test. We examined pseudotime trajectories of tissue-resident clusters with TCR repertoire clonal sharing (TRM(1), TRM(2), IL7R and CCL4 effectors) to infer putative differentiation pathways (Fig. 5a,b,e). The pseudotime trajectory showed two branches, formed predominantly of TRM(1) cells in branch 1 and TRM(2) cells in branch 2, developing from IL7R and CCL4 populations (Fig. 5a). While branch 1 cells were seen in both controls and CD, strikingly, branch 2 was almost totally restricted to ACD and TCD (Fig. 5b). We examined cytokine, chemokine and TF expression by CD8 T cell subsets. The predominant CD8 sources of IFNG were CD8 TRM(2) and cycling clusters (Fig. 5c,d and Extended Data Fig. 5g–i). These populations also expressed the chemokine CCL5, CD70 and FASLG. Natural IELs (reduced in ACD), produced CCL2, CXCL2, CXCL3, IL12, IL18 and type I interferon. TRM(1) and TRM(2) CD8 subsets showed distinct TF and regulon profiles; TRM(2) cells were associated with the TF regulons BACH1, CEBPZ, CREM, IRF4 and NR3C1 and TF expression of RORA, PRDM and FOXO1 (Extended Data Fig. 5j–l). CD8 T cell-induced epithelial damage is thought to be mediated via TCR-independent mechanisms. We hypothesized that CD8 T cell TCR repertoires would be similar in health and disease. Single-cell TCR sequences were examined, which showed expected clonal overlap between tissue-resident populations (Fig. 5e and Extended Data Fig. 6a). Cluster TRBV gene usage was examined between health and CD. Several high-frequency TRBV segments (>1% total clones) were overrepresented in CD (Fig. 5f and Extended Data Fig. 6b). However, statistical power was limited due to low clonotype numbers. Consequently, we sorted intraepithelial CD8 T cells from 12 adults with and without CD (dataset 3) and performed bulk RNA-seq. This showed significant enrichment of one TRBV segment, TRBV28, enriched in ACD and TCD, but not controls (Fig. 5g). TRBV28 was the high-frequency V segment with the highest fold change for enrichment in CD within the TRM(2) population (Fig. 5f). We validated this by performing bulk TCR repertoire sequencing on 1,068,814 mucosal CD8 T cells from 20 donors with and without CD (dataset 4). Again, TRBV28 was highly upregulated in CD, forming 10% of unique CDR3 sequences in ACD and TCD, versus 2% in controls (Fig. 5h). TRBV28 was also enriched within the top 100 most expanded clonotypes. No association with TRAV usage was seen. Clonotypes containing TRBV28 in CD paired with multiple TRBJ segments, and showed altered CDR3 amino acid usage, with enrichment of germline-encoded and non-germline-encoded leucine residues (Extended Data Fig. 6c–e). We examined bulk TCR repertoires of intestinal CD8 T cells of colonic and small intestinal biopsy samples from three separate studies examining non-CD inflammatory gastrointestinal conditions. There was no signal for enrichment of TRBV28 gene usage in these disease settings (Extended Data Fig. 6f–h). We hypothesized that differences in mucosal CD8 TCR repertoire/phenotype may be mirrored within gut-homing CD8 T cells in the circulation, as seen following gluten challenge. We examined TRBV28 usage by circulating CD8 T cells using flow cytometry (dataset 9). Using TCR sequencing (TCR-seq), we validated the specificity of the TRBV28-specific antibody clone (JOVI.3; Extended Data Fig. 7a–c). As expected, there was no difference in the fraction of TRBV28 cells in total peripheral CD3 or CD8 T cell compartments in participants with and without CD. However, within CD8 T cell populations expressing gut-specific chemokines (CCR9) or integrins (CD103/β-integrin), the fraction of TRBV28 cells was increased in ACD and TCD (Extended Data Fig. 7d–h). (a) TRBV gene usage (proportion of unique CDR3 clonotypes) of sorted CD3 JOVI.3 T-cells, determined by bulk TCR-seq. (b) TRBJ gene usage of TRBV28 clonotypes from sorted CD3 JOVI.3 T-cells. (c) TRAV gene usage (proportion of unique CDR3 clonotypes) of sorted CD3 JOVI.3 T-cells. Flow cytometry examining TRBV28 surface protein expression on circulating CD8 + T-cells in healthy controls (n = 8), active CD (n = 13), and treated CD (n = 7). (d) Representative flow cytometry gating strategy. (e) TRBV28 expression by total CD3 T-cells (left) and total CD8 T-cells (right) in peripheral blood in health and CD. (f) Representative plots of staining for CCR9, β7-integrin, and CD103. (g, h) TRBV28 expression by CCR9 (left), β7-integrin (middle), and CD103 (right) CD8 T-cells in peripheral blood, categorised by health and CD (g), and by health, active and treated CD (h). (e,g) Mann-Whitney two-tailed test. (e,g,h) Mean ± SEM shown. Intraepithelial duodenal γδ T cells are increased in CD, although their role is unclear. We analyzed a further dataset of 5,552 sorted intestinal CD8 αβ and γδ T cells (dataset 8; Extended Data Fig. 8). Clustering of cell transcriptional states recapitulated the key populations described above (Extended Data Fig. 8a,b). As previously, the TRM(2) population (in this case split into IFNG and IKZF2 subpopulations) was increased in ACD, along with cycling T cells (Extended Data Fig. 8c). Sorted intestinal γδ+ and αβ + CD8 T-cells in CD. (a) UMAP plot of intestinal CD4 T-cells in health and CD (n = 4). (b) Bubble plot showing the expression of selected genes defining specific cluster identities. Scaled gene expression indicated by colour, proportion of cells expressing the gene indicated by bubble size. (c) Proportion of transcriptional clusters in HC and CD. (d) UMAP overlay and (e) stacked plots showing the proportion of γδ + T cells in each transcriptional cluster. (f) UMAP plot overlaid with the expression of genes of interest. (g) UMAP plot overlaid with the clonal expansion of TCR clones. Clonal expansion expressed as quintiles, with Q1 showing the most expanded clones. (h) Frequency of quintiles of clonal expansion in TCR sequences expressing TRBV28. (i) UMAP plot overlaid with TRBV28-expressing CD8 + TCR clonotypes, split by quintiles of clonal expansion. γδ T cells showed overlapping transcriptional profiles with mucosal CD8 αβ T cells, albeit with enrichment within specific clusters (Extended Data Fig. 8d,e). γδ T cells were most enriched within a natural IEL phenotype cluster and the GZMK/FGFBP2 effector populations, and were also present in the CCL4 effector and IKZF2 TRM(2) population. γδ T cells were uncommon within IFNG TRM(2) and cycling clusters. TRDV1 and TRDV3 expression was higher in the CCL4, IKZF2 TRM(2) and natural IEL populations, with TRDV3 in particular enriched in the natural IEL cluster (Extended Data Fig. 8f). We analyzed the TCR repertoire of CD8 T cells in this dataset. The TRM and IL7R clusters showed greatest clonal expansion (Extended Data Fig. 8g). In all participants with CD, TRBV28-containing clonotypes were more clonally expanded than their non-TRBV28 counterparts. TRBV28 clonotypes were enriched in the top quintile of expanded clones, which were almost exclusively found within the TRM(2) and cycling clusters (Extended Data Fig. 8h,i). We validated these findings through bulk RNA-seq of sorted intestinal αβ CD8 and γδ T cells from participants with and without CD (dataset 6). Gene-set enrichment analysis of CD8 T cell gene expression in ACD showed upregulation of TCR activation gene sets, and enrichment of cluster marker gene sets from TRM(2) and cycling populations (Supplementary Fig. 3a–c), with upregulation of CXCR6, ENTPD1 and MKI67 (Supplementary Fig. 3d). CD8 T cells showed upregulation of IFNG and IL26 (Supplementary Fig. 3e). There was a shift from KLRC1 (NKG2A) to KLRC2 (NKG2C) expression, but KLRK1 (NKG2D) expression was not increased. Inhibitory KIRs were upregulated in this dataset, consistent with recent findings. Gene expression between health and CD was different in γδ and αβ CD8 T cells (Supplementary Fig. 3f,g). IFNG and MKI67 expression were not increased to the same extent in γδ T cells, nor were TRM(2) IFNG cluster markers like ENTPD1. There were also differences in NK cell receptor changes, KIRs, PDCD1 and TYROBP, a natural IEL marker (Supplementary Fig. 3h). Bulk γδ TCR repertoire sequencing (dataset 7) revealed a skewed TRGV repertoire, with reduced TRGV4 and increased TRGV3 use in ACD (Supplementary Fig. 4a,b), which persisted after treatment, as previously described. Most TRD CDR3 sequences were private; however, increased sequence sharing was noted between ACD repertoires (Supplementary Fig. 4c), with longer shared CDR3 sequences in ACD (Supplementary Fig. 4d). Previously reported CD-associated TRDV CDR3 motifs, were increased in ACD; however, we were unable to replicate the previously described association between the TRDV H-J1 motif and CD (Supplementary Fig. 4e–i). Samples from adults with TCD (GFD with good symptomatic, serological and histological response) were included in scRNA-seq, bulk RNA-seq and TCR-seq experiments (Supplementary Table 1). We hypothesized that cell-type and transcriptional changes would normalize with treatment. However, many biological changes persisted. Specifically, EC changes including increased TA cell proportions and cycling cells (Extended Data Fig. 1b–e) and the shift toward progenitor states (Fig. 2f) persisted despite treatment. EEC changes also persisted. However, absorptive function gene expression within mature enterocytes had predominantly normalized, aside from the ongoing reduction in fructose metabolism and lipid catabolism (Extended Data Fig. 1g). While TFH-like CD4 T cells and Treg cells returned toward control levels on treatment, the CD8 compartment remained perturbed, with reduced natural IELs and increases in TRM(2) CD8 T cells. However, the TRM(2) population showed reduced activation in TCD (lower IFNG, IL32 and pro-inflammatory markers). Intestinal CD8 TCR repertoire changes remained (specifically TRBV28 enrichment), as did increased circulating TRBV28 gut-homing CD8 T cells. We next analyzed duodenal stromal and endothelial populations in CD (dataset 3; Supplementary Fig. 5a–d). Annotation of stromal populations based on previous descriptions showed S1, S2 and S3 fibroblasts, as well as myofibroblasts, with S1 fibroblasts most common in the duodenum. The pro-inflammatory S4 phenotype seen in colonic inflammatory bowel disease was not seen. Differential gene expression and gene-set enrichment analysis showed upregulation of interferon-induced genes including STAT1, the major histocompatibility complex class II invariant chain (CD74), and SLIT2, encoding a secreted protein involved in intestinal homeostasis (Supplementary Fig. 5e,f). Analysis of endothelial cells revealed arterial, capillary, venous and lymphatic populations, with upregulation of interferon-stimulated genes in CD (Supplementary Fig. 5g–i). We next performed spatial transcriptomics on duodenal biopsy samples (dataset 4; Fig. 6). Spatial transcriptomics showed 13 transcriptionally distinct regions within the mucosa, representing compartments of the crypt–villus axis (stem cell niche, lower-crypt and mid-crypt regions and villus zones), stromal cell-rich regions, several lamina propria regions with immune cell infiltrates dominated by plasma cell signatures and lymphoid aggregates (LAs; Fig. 6a,b). In health, the epithelial villus compartments dominated; these were reduced in ACD (Fig. 6c–e). These villus regions expressed absorptive function genes, predominantly in the most mature villus compartment (Fig. 6b and Supplementary Fig. 6b). In contrast, immune-rich regions and LAs were greatly expanded in ACD (Fig. 6d,e). These regions were themselves spatially organized, with LAs closely associated with lower-crypt and immune-rich regions, and telocyte-rich regions with villus structures (Fig. 6g). a, UMAP overlay of all spatial transcriptomics tissue-covered spots with transcriptome-driven clustering analysis, colored by region. b, Bubble plot showing the expression of selected genes defining spatial regions. Scaled gene expression indicated by color; proportion of cells expressing the gene indicated by bubble size. c, Visualization of transcriptionally distinct spatial regions overlaid on representative HC tissue section. d, Proportion of intestinal mucosa formed in different regions in HCs (above) and ACD (below). Immune-rich and LA regions are highlighted. e, Local neighborhood enrichment of intestinal mucosal regions in ACD versus HCs. Color indicates enrichment (log fold change) of cells in ACD versus HCs in that UMAP neighborhood. f, Volcano plot of differential gene expression between HCs and ACD within villus tip spatial regions. g, The spatial relationships between different regions in ACD can be visualized using a network plot. Regions that are more likely to be adjacent to another region are connected by arrows colored by the percentage of adjacent spots. Region size is indicated by size and color of the region circle. h, Integrating scRNA-seq reference data localizes single-cell transcriptomes to spatial regions. These data are used to generate network plots visualizing colocalization of cell types together in ACD. Cell-type nodes close together and linked by connecting lines are more often located in the same spots. In ACD, mature enterocytes colocalize with TRM(2) CD8 T cells (lower red box), while TFH-like CD4 T cells localize with B cells, Treg cells and plasma cells (upper red box). These immune-rich regions and LAs showed gene expression patterns associated with B cells (CD19, MS4A1 (CD20)), plasma cells (IGHM, IGHA1, TXNDC5) and T cells (CD3D, CCR7, CXCL13). Signals of cellular proliferation (MKI67, REG1A) were highly localized in specific clusters. In health, cellular proliferation was limited to stem-cell and lower-crypt regions, while in ACD, proliferative markers were more dispersed, including in immune-rich and LA regions (Supplementary Fig. 6c,d). In ACD, villus top regions showed increases in interferon-stimulated genes, markers of proliferation, and IL32 (Fig. 6f), analogous to epithelial scRNA-seq results. TCR genes, tissue-residency markers (ITGAE, CXCR6, KLRB1), and cytotoxic CD8 markers (GZMA, KLRD1, KLRK1) were increased in ACD, suggesting tissue-resident cytotoxic CD8 T cell enrichment in villi. We integrated spatial transcriptomics with scRNA-seq data to predict cell-type locations in mucosa (Supplementary Figs. 6e–j and 7). In ACD, TA cell signatures expanded from crypt bases to most villus regions, while mature EC signatures were restricted to superficial epithelial layers. In ACD, LAs showed highly localized enrichment of TFH-like CD4 T cell (CXCR5, CXCL13) and B cell (CD19, MS4A1; Fig. 6b and Supplementary Fig. 7a) signatures. Plasma cell signatures were expanded in neighboring immune-rich regions. In ACD, CD8 TRM(2) cell signatures were highly enriched in villus tip regions, colocalized with mature enterocytes (Fig. 6h and Supplementary Figs. 6h–j and 7b). To further study LAs in CD, we performed further spatial transcriptomics experiments on duodenal biopsy samples in participants with and without CD (dataset 5; Fig. 7 and Extended Data Fig. 9a–c). Analysis of spatial regions recapitulated our description of key transcriptional regions in the duodenal mucosa (Fig. 7a,b and Extended Data Fig. 9c), with enrichment of proliferating areas at crypt bases, MUC5AC and PGC epithelium, immune-rich areas containing plasma cells, and LAs (Fig. 7c–f). These LAs were enriched in both ACD and TCD. a, UMAP overlay of all spatial transcriptomics tissue-covered spots with transcriptome-driven clustering analysis, colored by region. b, Bubble plot showing the expression of selected genes defining spatial regions. Scaled gene expression indicated by color; proportion of cells expressing the gene indicated by bubble size. c,d, Local neighborhood enrichment of intestinal mucosal regions in ACD versus HCs (c) and TCD versus HCs (d). Color indicates enrichment (log fold change) of cells in CD versus HCs in that UMAP neighborhood. e, Proportion of intestinal mucosa formed in different regions in HCs, ACD and TCD. f, Proportion of immune-rich and LA regions in HCs, ACD and TCD. g,h, Detailed examination of a representative LA in ACD (seen in 5/10 CD sections). g, Hematoxylin and eosin (H&E)-stained section of duodenal biopsy with LA circled. h, Spatial regions overlaid onto the section show the LA near the lower-crypt/stem-cell niche region, and near the muscularis mucosa. i, Predicted cell-type locations in regions overlaid onto the section. j–l, Bubble plots of gene expression within LAs and other regions, paired with gene expression overlaid onto an ACD section with LA, including TFH/Treg cell gene signatures (j), B/plasma cell gene signatures (k), and chemokines and associated receptors (l). m, Stromal cell gene expression overlaid onto a representative ACD section with LA. (a–c) Dataset 5 spatial transcriptomics descriptive data and integration with Dataset 4. (a) UMAP overlaid with disease state. (b) UMAP overlaid with study subject ID. (c) UMAP overlaid with Dataset 4 region labels. (d, e) Cell type predictions and spatially restricted gene expression within lymphoid aggregates within the duodenal mucosa in CD. Lymphoid aggregate regions are highly localised in intestinal biopsies from CD subjects (b). Cell type predictions (c) indicate Tfh-like CD4 T-cells, as well as B cells, are present in these regions, with plasma cells and Tregs surrounding these regions. This is supported by the expression of MS4A1 (CD20), CXCR4, CXCR5 and CXCL13. These lamina propria LAs were located adjacent to stem cell niches and muscularis mucosa (Fig. 7g–i), with enrichment of TFH-like CD4 T cell, Treg cell and B cell gene signatures. Plasma cell signatures were more widely dispersed in immune-rich regions (Fig. 7i–k and Extended Data Fig. 9d,e). Genes for chemokines and receptors, including CXCR5, CCR7, CCL19, CCL21, CXCL13 and CXCL14 were enriched specifically within LAs (Fig. 7l), as were genes associated with S3 stromal cells, and pro-inflammatory stroma seen in inflammatory bowel disease (Fig. 7m). We examined the expression of chemokines, cytokines and tumor necrosis factor (TNF) superfamily members, as well as receptor–ligand coexpression, within regions in the spatial transcriptomics dataset (Fig. 8a,b and Supplementary Fig. 8). Signaling pathway expression was region dependent, indicating highly localized mucosal signaling circuits (Fig. 8a). Within CD-specific LAs, chemokine signaling circuits involving CXCR5–CXCL13, CCR7–CCL19, CXCR4 and integrins ITGB2 and ITGAM were upregulated (Fig. 8a and Supplementary Fig. 8a,b). TNF superfamily receptor–ligand pathways were upregulated in LAs, including TNF, lymphotoxins A and B, CD40, TNFRSF8 (CD30), TNFRSF6B, TNFRSF18 (GITR) and TNFRSF4 (OX40) (Supplementary Fig. 8c,d). There was also evidence of IL2 and IL21 signaling, as well as possible involvement of IL23A and IL26 pathways (Supplementary Fig. 9e,f). LAs found in CD were enriched with signaling pathways involving CXCR4, CXCR5 and CXCL13 (Figs. 7l and 8a,b). Single-cell examination of CXCL13, CCR7 and ITGB2 signaling interactions in dataset 1 implicated TFH-like CD4 T cell, B cell and myeloid cell interactions as drivers of these signaling pathways in LAs (Fig. 8b). a, Bubble plot of region-specific receptor–ligand expression within the duodenal mucosa. Scaled receptor–ligand (RL) expression indicated by color; proportion of regions expressing the receptor–ligand genes indicated by bubble size. b, Circos plots of selected receptor–ligand pair expression between cell types in CD (dataset 1). c, A proposed schematic for the spatially resolved cellular ecosystems within the duodenal mucosa in CD. LTo, lymphoid tissue organizer. Figure created with BioRender.com. Villus tip regions, demonstrated region-specific activation of T cell and immune-related pathways, including increased expression of IL15, IL18, IFNG and IL32, and interactions including CXCR6–CXCL16, CCR9–CCL25, DPP4–ADA and HLA-E interactions with KLRC2 and KLRD1 (Fig. 8a and Supplementary Fig. 8). There was also high expression of TNF superfamily members associated with apoptotic pathways, including FASLG, TNFSF10 (TRAIL), TNFSF11 (RANKL), TNFSF12 (TWEAK) and TNFSF13 (APRIL) interactions. Examination of single-cell signaling pathways implicated CD8 T cells in chemokine and type II interferon signaling, including CCR9–CCL25 axis interactions between progenitor ECs and cycling CD8 T cells (Fig. 8b). Wnt signaling pathways were enriched in telocyte-rich areas and neighboring villus structures (Fig. 8a). Such morphogen gradients may shape villus structure and morphology, perhaps driven by subepithelial telocytes. To understand how genetic susceptibility can drive CD inflammatory responses, we examined putative genome-wide association study (GWAS) candidate gene expression in spatial regions. Villus, telocyte-rich and LA regions showed enriched expression of multiple GWAS candidates (Extended Data Fig. 10a). Expression of MMP9, CTLA4, ICOS, ITGA4, GPR183, IL21 and IL21R was enriched within CD-specific LA regions (Extended Data Figs. 9e and 10a), while IL2RA, CCR1, XCR1, TNFSF11 (RANKL) and TNFRSF9 (CD137, 4-1BB) were increased in telocyte-rich regions. (a) Bubble plot of CD-associated GWAS candidate gene expression within the duodenal mucosa. Scaled gene expression indicated by colour, proportion of regions expressing the genes indicated by bubble size. (Genes identified from van der Graff et al. 2021, with at least one line of supporting evidence for direct gene association with CD). (b) Barplot of significance (–log10(FDR)) of CD-specific GWAS signal enrichment in scRNA-seq cell clusters in immune and epithelial subpopulations in cells from healthy and UC donors (n = 8). SNPsea empirical distribution P value with multiple testing correction (Benjamini–Hochberg). (c) Heatmap of CD-associated GWAS candidate gene scaled expression within immune and epithelial cell populations identified in Dataset 1 scRNA-seq dataset. (d) Violin plots of selected GWAS candidate gene expression in TRM(2) CD8 T-cells in health and CD. (e) Heatmap of CD-associated GWAS candidate gene scaled expression within immune cell population identified in Dataset 2 scRNA-seq dataset. scRNA-seq data were examined for cell-type-specific expression of CD genetic susceptibility loci. GWAS candidate genes were most prominent in T cell subsets (Extended Data Fig. 10b), with the highest signal enrichment in cycling CD8 T cells. Specific putative GWAS candidate genes drove these associations (Extended Data Fig. 10c), with TFH-like CD4 T cells expressing the IL21, PTPN2, ITGA4, CD28 and ICOS and TRM(2) CD8 T cells expressing CXCR6, TNFRSF9 and TNFRSF14, and showing CD-related changes in STAT expression (Extended Data Fig. 10d). Cell-type-specific gene expression patterns were recapitulated in dataset 2 (Extended Data Fig. 10e). This multi-omics study provides an integrated single-cell transcriptomic and proteomic assessment of intestinal immune, epithelial and parenchymal cell populations in adult and pediatric CD, contextualized through integration with spatial transcriptomics analysis. Our results show that perturbations of immune and epithelial cell states are spatially localized within distinct mucosal niches. Disease-associated cell types, including gluten-specific TFH-like CD4 T cells and CD8 TRM cells, occupy distinct LA and villus niches, respectively, with cell–cell interactions best understood through spatial colocalization (Fig. 8c). The application of receptor–ligand analyses implicates broader cytokine and chemokine perturbations in CD than those described previously, including IL-32, CXCL13, CCL19, CXCL16, CXCL8 and CCL25. Our understanding of human duodenal lymphoid structures is incomplete. Isolated gut lymphoid structures may act as immune-inductive sites. In CD, it remains unclear where the immune response to gluten is primed, or where subsequent antigen presentation occurs. Both myeloid and B cell lineages have been posited as relevant antigen-presenting cells (APCs) in CD. The discovery of highly localized LAs where B cells and (likely gluten-specific) TFH CD4 T cells are co-located implicates these sites in gluten peptide antigen presentation. Several aspects of the mucosal epithelial and immune response remained perturbed despite a GFD. These findings help explain the observation that subtle abnormalities in duodenal biopsy samples remain despite treatment, with reduced villus height/crypt depth ratios on morphometric analysis. Whether this represents subclinical inflammation from ongoing low-level antigen exposure, slow mucosal healing or a long-term (perhaps epigenetic) response to prior inflammation is unclear. Participants often report ongoing symptoms despite a GFD, and this epithelial and immunological ‘scar’ from prior inflammation could underpin this, representing a therapeutic target. Intestinal tissue-resident CD8 T cell perturbations, including increases in TRM(2) populations, persisted despite GFD treatment. Therefore, this CD8 TRM(2) state may represent a distinct T cell fate, rather than an activated phenotype alone. TRM cells are long-lived memory populations that persist as immunological sentinels in barrier tissues, and this result is consistent with prior work showing a permanent reshaping of γδT cell-resident populations following CD-driven inflammation. CD8 T cell populations exhibited changes in TCR repertoire, a finding validated in multiple datasets. These TCR repertoire changes, along with upregulation of TCR signaling gene sets, may indicate that TCR-dependent activation is relevant in CD, involving a separate mechanism to previously described NKG2C/NKG2D pathways, and invoke the possibility of TCR-targeted disease therapies. IL-15 and NK cell receptor signaling could lead to a reduction in TCR activation threshold, which could enable recognition of low-affinity antigens, either self-antigens or those of microbial or dietary origin. The persistence of CD8 TRM(2) cells in TCD could represent ongoing antigen exposure. There is a series of enterocyte transcriptional states in the human small intestine, with absorptive cellular machinery generally limited to mature ECs, consistent with murine studies. The shift to progenitor states in CD may increase CCL25 expression, implicating the CCL25–CCR9 axis in disease. This shift toward progenitor states underpins CD-associated malabsorption, beyond reduction in intestinal surface area. In contrast to the term ‘villus atrophy’, we observe that the CD epithelium is hyperproliferative, and so the loss of villus structures requires additional explanation. Spatial transcriptomics data indicate specific regions responsible for WNT signaling, and we hypothesize that CD inflammation drives morphogen signaling shifts causing mucosal remodeling. These morphogen responses may not be CD specific, and may underpin histological similarities seen with mimics such as environmental enteropathy, monogenic enteropathies and olmesartan enteropathy. Integrating single-cell and spatial transcriptomics data, we have dissected the molecular and cellular basis of the histological changes in CD, including villus epithelial changes, crypt hyperplasia and intraepithelial lymphocytosis. This cellular, spatial transcriptomics description builds on the Marsh–Oberhuber histological description, with complex, highly localized, mucosal cell communities, including focal lymphoid organization where specific cell types, including gluten-specific TFH-like CD4 T cells, B cells and Treg cells, are co-located. Overall, our study of the mucosal cellular and spatial landscape in CD provides a detailed foundation from which to explore potential therapeutic targets, and highlights the need to explore the clinical implications of the prolonged epithelial–immune scar in TCD. Study participants with CD, and HCs, were identified via Oxford University Hospitals NHS Trust CD clinic and endoscopy service (Oxford, UK). Blood and intestinal biopsy samples were taken at endoscopy with informed consent under the Oxford Gastrointestinal Illnesses Biobank study (REC: 21/YH/0206). Study participant demographics and study inclusion/exclusion criteria are summarized in Supplementary Table 1. Participants were not compensated financially. Peripheral blood mononuclear cells were extracted from whole blood or leukocyte cones via density gradient centrifugation. Briefly, peripheral blood was diluted at a 1:1 ratio with Dulbecco’s phosphate buffered saline without calcium or magnesium (PBS) and layered over Lymphoprep (Axis-Shield) before centrifugation (973g for 30 min at 20 °C without brake). The mononuclear layer was retrieved and washed twice in PBS or R10 culture media (RPMI-1640 (Sigma-Aldrich), 10% FCS, 1% penicillin/streptomycin, 1% l-glutamine). Viable mononuclear cells stained with Trypan blue were counted manually by microscopy using a hemocytometer before downstream applications. If remnant red blood cells were present, they were lysed with ammonium–chloride–potassium solution for 2–3 min, then washed again in R10. Samples were cryopreserved in freezing medium (90% FCS (Sigma-Aldrich), 10% dimethylsulfoxide (Sigma-Aldrich)). When needed, samples were thawed rapidly in a water bath (37 °C), then washed twice in R10 before downstream use. Intestinal biopsy samples were collected at endoscopy from duodenum. Biopsy samples for intestinal lymphocyte extraction were immediately placed in sterile R10 medium (as above) or MACS Tissue Storage Solution (Miltenyi Biotec) on ice for transportation, before cryopreservation in CryoStor Cs10 (STEMCELL Technologies). This approach preserves immune cell viability and surface marker expression. When required, samples were rapidly thawed in a 37 °C water bath and washed in 20 ml R10 before tissue dissociation. For immune cell isolation from duodenal biopsy samples for scRNA-seq (10x Genomics, dataset 1), biopsy samples were incubated in R10 medium with 1 mg ml Collagenase D (Roche) and 100 mg ml DNase (Thermo Fisher Scientific) for 1 h in a shaking incubator at 37 °C. Biopsy samples were then dissociated by vigorous agitation using a GentleMACS Dissociator (Miltenyi Biotec), then strained through a 70-μm filter. The mononuclear cells were isolated on a discontinuous 70% and 35% Percoll gradient (GE Healthcare) by centrifugation at 700g for 20 min without brake. Mononuclear cells were collected from the interface and washed in R10. Cells were washed with R10 medium before antibody staining and downstream applications. The complete protocol with all steps is available at https://www.protocols.io/view/freezing-and-processing-intestinal-biopsies-for-th-dm6gp8745lzp/v1/ (Oxford HCA, 2019). The method for EC isolation from duodenal biopsy samples for scRNA-seq (10x Genomics, dataset 1) was adapted from ref. . Biopsy samples were washed in wash medium (HPGA, 1 mM EDTA, 1 mM dithiothreitol), then incubated in chelation medium (HPGA, 1 mM EDTA) at 37 °C for 40 min with agitation. The supernatant, which was removed and replaced every 10 min and contained epithelial crypts, was digested into a single-cell suspension by dissociation in a shaking incubator with TrypLE Express and DNase (50 µg ml) for 60 min at 37 °C. The epithelial single-cell suspension was washed with PBS and passed through a 30-µm filter. Cell counts and viability were confirmed with a manual hemocytometer before further processing. The complete protocol with all steps is available at https://www.protocols.io/view/isolation-of-cells-from-the-epithelial-layer-of-fr-e6nvw9e6dgmk/v1/ (Oxford HCA, 2019). For immune cell isolation from duodenal biopsy samples for scRNA-seq and proteomics (BD Rhapsody, dataset 2), samples were thawed, then diluted with warm X-VIVO (Lonza) + 1% AB serum (Sigma-Aldrich). Biopsy samples underwent enzymatic and mechanical digestion using 0.042 mg ml Liberase TL (Roche) and 1 mg ml DNAse I (Thermo Fisher). Samples were placed horizontally in a shaking incubator for 20 min at 37 °C and homogenized using a gentleMACS Dissociator (Miltenyi Biotech). Dissociated cells were passed through a 70-μm strainer. Lymphocytes were enriched by Percoll density centrifugation as above. Duodenal biopsy samples for flow cytometry were processed as for scRNA-seq of immune cells (see above). Cells were washed with R10 medium before antibody staining and downstream applications. Duodenal biopsy samples for fluorescence-activated cell sorting (FACS) of IELs for RNA-seq and TCR repertoire sequencing were placed in 10 ml HBSS with 1 mM EDTA and 1 mM dithiothreitol (both Sigma-Aldrich) and placed in a shaking incubator (200 rpm, 37 °C) for 15 min. IELs were strained through a 70-μm filter and washed two to three times with R10 before downstream applications. For spatial transcriptomics, single intestinal biopsy samples were embedded in OCT cryo-embedding matrix (Thermo Fisher Scientific) then frozen in isopentane (Sigma-Aldrich) suspended over liquid nitrogen or dry ice, and stored at −80 °C until use. For surface marker staining, cells were stained in 50 ml of FACS buffer (PBS + 1 mM EDTA + 0.05% BSA) for 30 min at 4 °C. Surface antibodies and clones used are listed in Supplementary Table 3. Antibodies were purchased from BioLegend, BD Biosciences, Miltenyi Biotec or Thermo Fisher Scientific. After staining, cells were stored at 4 °C protected from light until data acquisition. Flow cytometry data were acquired on a BD LSR II flow cytometer (BD Biosciences). FACS samples were surface stained as above, with Sytox Green (Thermo Fisher Scientific) used as a viability dye. FACS was performed on an Aria III (BD Biosciences; 70-mm nozzle). For sorting by FACS for scRNA-seq of intestinal immune populations, cells were stained with EpCam-PE, CD27-BV421 and CD45-APC-Cy7. Live CD45 or CD27 cells were sorted to include all mucosal immune cell populations, including long-lived CD27 plasma cells, which can downregulate surface CD45 expression. For sorting by FACS for scRNA-seq of intestinal epithelial populations, cells were stained with EpCAM-PE and CD45-AF700, with live EpCAM cells sorted. For sorting by FACS for bulk RNA-seq or TCR-seq of CD8 IEL populations, cells were stained with CD45-BV785, CD3-BV711, αβTCR-APC, γδTCR-PE, CD4-BV650 and CD8a-AF700, with live CD45CD3αβTCRCD8CD4 cells sorted. Libraries were generated using 10x Genomics Chromium Single Cell V(D)J Reagents Kits (v1 Chemistry) per the manufacturer’s instructions. Sorted cells suspended in PBS (plus 0.04% BSA) at a concentration of 1,000 cells per microliter were loaded into one lane of a Chromium controller. Library quality and quantity were assessed using a TapeStation (Agilent) and Qubit Fluorometer (Thermo Fisher Scientific). Libraries were sequenced on an Illumina HiSeq 4000 following the manufacturer’s instructions. Library generation and sequencing were performed at the Sanger Institute, Cambridge. Sorted CD45 cells were stained with a cocktail of 79 oligonucleotide-conjugated AbSeq antibodies (BD Biosciences, for 45 min at 4 °C. Cells were then washed to remove residual unbound AbSeq antibodies and loaded onto three BD Rhapsody cartridges (BD Biosciences) for single-cell capture. AbSeq antibodies used in this study are listed in Supplementary Table 2. Single-cell capture and cDNA library preparation were performed using the BD Rhapsody Express Single-Cell Analysis System (BD Biosciences), according to the manufacturer’s instructions. Briefly, cDNA was amplified—ten cycles for resting cells and nine cycles for in vitro-stimulated cells—using the Human Immune Response Primer Panel (BD Biosciences; Supplementary Table 6), containing 399 primer pairs and a supplementary panel of 105 primer pairs (BD Biosciences; Supplementary Table 3). The resulting PCR1 products were purified using AMPure XP magnetic beads (Beckman Coulter), and the respective mRNA and AbSeq/Sample Tag products were separated based on size selection, using different bead ratios (0.7× and 1.2×, respectively). The purified mRNA and Sample Tag PCR1 products were further amplified (ten cycles), and the resulting PCR2 products purified by size selection (1× and 1.2× for the mRNA and Sample Tag libraries, respectively). The concentration, size and integrity of the resulting PCR products were assessed using both Qubit (High-Sensitivity dsDNA Kit; Thermo Fisher) and the Agilent 4200 TapeStation system (High Sensitivity D1000 ScreenTape; Agilent). The final products were normalized to 2.5 ng μl (mRNA), 0.5 ng μl (Sample Tag) and 0.275 ng μl (AbSeq) and underwent a final round of amplification (six cycles for mRNA and eight cycles for Sample Tag and AbSeq) using indexes for Illumina sequencing to prepare the final libraries. Final libraries were quantified using Qubit and Agilent TapeStation and pooled (~60%/38%/2% mRNA/AbSeq/Sample Tag ratios, respectively) to achieve a final concentration of 5 nM. Final pooled libraries were spiked with 10% PhiX control DNA to increase sequence complexity and sequenced (75 base pairs (bp), paired-end) on a HiSeq 4000 sequencer (Illumina). Cryopreserved, OCT-embedded duodenal biopsy samples were stored at −80 °C until use. Before performing the full protocol, a tissue permeabilization optimization was performed (10x Genomics, Visium Spatial Tissue Optimization), which identified 11 min as the optimum permeabilization time. Samples were processed for spatial transcriptomics per the manufacturer’s instructions (10x Genomics, Visium Spatial), with 2 × 10-μm sections cut on a pre-cooled cryostat for each sample onto two 6.5 × 6.5-mm capture areas, each with approximately 5,000 oligonucleotide-barcoded 55-μm-diameter spots. Slides were fixed, H&E stained and imaged on a Leica DMI8 Widefield microscope at a magnification of ×40. Tissue was permeabilized per instructions for 11 min, followed by reverse transcription and second-strand synthesis performed on the slide. cDNA quantification was performed using qPCR using KAPA SYBR FAST-qPCR kit (KAPA Biosystems) on a CFX96 Thermal Cycler instrument (Bio-Rad). Following library construction per instructions, the spatial transcriptomics libraries were quantified and pooled at a concentration of 4 nM with a sample ratio corresponding to the approximate surface area of tissue coverage obtained from the H&E imaging. Pooled libraries were sequenced on a NextSeq (Illumina) using a 150-bp paired-end dual-indexed setup (High output, v2.5, Illumina) loaded at a concentration of 1.8 pM, and sequenced to a manufacturer-recommended depth of a minimum of 50,000 reads per tissue-covered spot. Cryopreserved, OCT-embedded duodenal biopsy samples were stored at −80 °C until use. Spatial transcriptomics was performed using the Visium CytAssist (10x Genomics) workflow for fresh-frozen tissue according to the manufacturer’s instructions. Ten-micron sections were cut on a cryostat pre-cooled to −20 °C and placed on 11 × 11-mm areas (four sections per area) on SuperFrost Plus slides (Thermo Fisher). Sections were fixed, H&E stained and imaged on an Axioscan Z1 slide scanner (Zeiss) at a magnification of ×20. Sections were de-stained and subjected to on-slide probe hybridization and ligation followed by probe transfer onto Visium CytAssist Spatial Gene Expression slides, each containing an 11 × 11-mm capture area covered by approximately 14,000 55-µm-diameter oligonucleotide-barcoded spots. Probes were extended and cDNA quantified by qPCR using a KAPA SYBR FAST-qPCR kit (KAPA Biosystems) on a CFX96 Thermal Cycler instrument (Bio-Rad), followed by off-slide library construction per the instructions. Libraries were quantified and pooled at a concentration of 2 nM with a sample ratio corresponding to the approximate surface area of tissue coverage obtained from the H&E imaging. Pooled libraries were sequenced on a NextSeq 500 instrument (Illumina) using a 150-bp paired-end dual-indexed setup (High output, v2.5, Illumina) at a manufacturer-recommended depth of a minimum of 50,000 reads per tissue-covered spot. A TRIzol nucleic acid extraction method was used to extract RNA from low numbers of sorted lymphocytes, as previously described, except that phase-lock gel tubes were replaced with standard 1.5 ml microcentrifuge tubes. Briefly, after sorting, cells were centrifuged (500g, 5 min), resuspended in 1 ml TRIzol, then frozen at −80 °C until RNA extraction. For RNA extraction, samples were thawed, mixed with 200 μl chloroform and centrifuged (14,000g, 5 min). A total of 500 μl of the aqueous phase was taken and RNA was extracted using the Agencourt RNAdvance Tissue Isolation kit. RNA concentration and purity were assessed on a 2100 Bioanalyzer instrument (Agilent). Bulk RNA-seq was performed using the Smart-seq2 protocol at the Oxford Genomics Centre (University of Oxford). Around 10 ng RNA was used as a template from each sample for library generation. Barcoded samples were pooled, and External RNA Controls Consortium RNA (1:100,000 dilution) was added before 75-bp paired-end sequencing on an Illumina HiSeq 4000 instrument. Bulk TCR repertoire sequencing was performed using the amplicon-rescued multiplex-PCR method (iRepertoire). This method performs an initial first-round RT–PCR with TCR V and C gene-specific primers for the relevant TCR chain, followed by further amplification steps with universal primers for the exponential phase of amplification. This method is designed to provide quantitative, deep sequencing of the TCR repertoire, with minimal bias. Library generation was performed following the manufacturer’s instructions, except for using 96-well plates. The quality, size distribution, concentration and presence of contaminating primer dimers of the final product was assessed using agarose gel electrophoresis, a spectral photometer (Nanodrop, Thermo Fisher Scientific), and the Bioanalyser DNA 1000 assay using a 2100 Bioanalyzer instrument (Agilent). Libraries were quantified using the KAPA Library Quantification Kit (Roche) on a CFX96 Thermal Cycler instrument (Bio-Rad) before equimolar pooling. A PhiX library spike-in was added (10%) due to the low diversity of the TCR library, before 300-bp paired-end sequencing on an Illumina MiSeq instrument at the Oxford Genomics Centre. This was an observational, descriptive study. The experiments were not randomized. No statistical method was used to predetermine sample size, but our sample sizes are similar to those reported in previous publications. No data were excluded from analyses. The investigators were not blinded to allocation during experiments and outcome assessment. Data collection and analysis were not performed blind to the conditions of the experiments. Data distributions of transcriptomics datasets were tested to ensure they met the assumptions of statistical tests. For other datasets, data distribution was assumed to be normal, but this was not formally tested. All statistical analyses and graphs, except transcriptional data, were performed using Prism Software v9 and v10 (GraphPad). Specific statistical tests are described in relevant figure legends. All data are presented as the mean ± s.e.m. unless stated otherwise. Flow cytometry data were analyzed using FlowJo v9.9.5 and v10.6.1. Raw sequencing read data were subjected to quality control and aligned to the human reference hg38 genome using STAR aligner. The DESeq2 R package was used for downstream differential expression analysis. Bulk TCR repertoire analysis was performed using the iRepertoire analysis pipeline. Raw read data were processed with either the Cell Ranger pipeline or the BD Genomics pipeline. Downstream analyses were carried out in R using the Seurat pipeline. Raw read data were processed using the Space Ranger pipeline. Downstream analyses were carried out in R using the Seurat pipeline. Raw sequencing read data were processed with the Cell Ranger VDJ pipeline. Downstream analyses were carried out in R. Additional details for computational data analysis are provided in Supplementary Methods. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41590-025-02146-2. This work was supported by a Human Cell Atlas grant from Wellcome (WSSS 211276/Z/18/Z and 218597/Z/19/Z), Wellcome (222426/Z/21/Z), the Oxford-BMS Translational Fellowship Scheme, Beyond Celiac, Coeliac UK, Academy of Medical Sciences, the JDRF/Wellcome Diabetes and Inflammation Laboratory Strategic Award, and by the NIHR Biomedical Research Centres, University of Oxford. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, Wellcome, JDRF/Breakthrough T1D or the Department of Health. D. Aschenbrenner is an employee and shareholder of Novartis Pharma AG; D.R. is an employee of Janssen Immunology, Translational Sciences & Medicine; this article reflects the author’s personal opinions and not those of their employers. This work was supported by Johnson and Johnson Innovative Medicine via the Cartography Consortium through contribution of the stromal dataset. We thank the clinicians, biobankers and endoscopists in the Department of Gastroenterology, Oxford University Hospitals, especially V. Cheung, H. Ferry for expertise in cell sorting, D. Fawkner-Corbett for advice on EC isolation approaches, L. Wicker and L. Ciacchi for their comments, the MRC WIMM Wolfson Imaging Centre and Sequencing Facility, and the study participants. Nature Immunology thanks Bana Jabri and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: L. A. Dempsey, in collaboration with the Nature Immunology team. Raw and processed data are available on Zenodo (10.5281/zenodo.15069144 (ref. )) and Gene Expression Omnibus (GSE252545). M.E.B.F. consults for Takeda Pharmaceuticals. N.M.P. has consulted for Infinitopes. D.R. is an employee of Janssen Immunology. D. Aschenbrenner is an employee of Novartis Institute of BioMedical Research. S.A.T. is a scientific advisory board member of ForeSite Labs, OMass Therapeutics, Qiagen and Xaira Therapeutics, a cofounder and equity holder of TransitionBio and Ensocell Therapeutics, a non-executive director of 10x Genomics, and a part-time employee of GlaxoSmithKline. J.A.T. consults for GSK, Vesalius, Avammune, Azenta, Dasman Diabetes Institute and Immunocore. P.K. has consulted for AZ, Infinitopes and UCB, and received research funding from Immunocore, all on unrelated projects. The other authors declare no competing interests. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Michael E. B. FitzPatrick, Agne Antanaviciute. These authors jointly supervised this work: Holm H. Uhlig, John A. Todd, Paul Klenerman. is available for this paper at 10.1038/s41590-025-02146-2. Raw and processed data are available on Zenodo (10.5281/zenodo.15069144 (ref. )) and Gene Expression Omnibus (GSE252545).
PMC12408821
Mapping the arterial vascular network in an intact human kidney using hierarchical phase-contrast tomography
The architecture of kidney vasculature is essential the organ's specialised functions, yet is challenging to structurally map in an intact human organ. Here, we combined hierarchical phase-contrast tomography (HiP-CT) with topology network analysis to enable quantitative assessment of the intact human kidney vasculature, from the renal artery to interlobular arteries. Comparison with kidney vascular maps described for rodents revealed similar topologies to human, but human kidney vasculature possessed a significantly sharper decrease in radius from hilum to cortex, deviating from theoretically optimal flow resistance for smaller vessels. Structural differences in kidney hilar, medullary and cortical vasculature reflected unique functional adaptations of each zone. This work represents the first time the arterial vasculature of an intact human kidney has been mapped beyond segmental arteries, potentiating novel computational models of kidney vascular flow in humans. Our analyses have implications for understanding how blood vessel structure collectively scales to facilitate specialised functions in human organs. Subject terms: Engineering, Kidney diseasesThe vasculature of the kidney is highly specialised and serves multiple functions, including the delivery of oxygen and nutrients to the organ’s parenchyma, whilst also facilitating plasma ultrafiltration and solute reabsorption. Despite only comprising approximately 1% of body weight, the kidney receives up to 20% of cardiac output. Blood enters the kidney through the renal artery, which branches from the abdominal aorta and enters the kidney hilum. Once within the kidney, the renal artery divides hierarchically, first into segmental or renal feeding arteries which pass through the kidney pelvis, then branching into interlobar arteries which pass through columns between the pyramids of the kidney medulla. At the distal end of the kidney columns, interlobar arteries branch into arcuate arteries that arch around the outer surface of the kidney pyramids. From these, the interlobular vessels branch and penetrate the surrounding kidney cortex, before finally terminating at afferent arterioles. This complex network perfuses specialised capillary networks, including glomerular capillaries across which plasma ultrafiltration occurs, efferent arterioles and peritubular capillaries or vasa recta, which facilitate dynamic solute exchange in the cortex and medulla, respectively. Thereafter, venous return follows the arterial supply out of the organ. Structural and molecular changes to the kidney vasculature are a common feature of kidney pathologies, including multiple aetiologies of chronic kidney disease (CKD) and transplant rejection in both animal models and patients. Therefore, studying kidney vascular patterning has implications for understanding the basis of kidney function in health and disease, and aids surgical planning for tumour resection, nephrectomy and transplantation. Vascular geometries also have a central role to play in computational models that underpin the creation of digital twins, such as through the generation of synthetic data, and blood flow modelling, which are playing an increasing role in biomedical research. Vascular imaging of the kidney has advanced following technological innovations in micro-computed tomography (μCT), magnetic resonance imaging (MRI) and μMRI, ultrasound, lightsheet microscopy and photoacoustic imaging. These techniques have been used to generate quantitative analyses of vascular network geometry in intact kidneys of model organisms, particularly rodents, in which kidney diameter reaches up to 12 mm. Comparatively human kidneys, with a diameter of approximately 5 cm are far more challenging to image at high resolution whilst still intact. Corrosion casting of human kidneys has highlighted vascular heterogeneity and generated intricate 3D casts (down to 100 µm) but provides limited quantitative or accessible digitised geometries of the vascular network. Optical clearing and lightsheet microscopy have been used to quantify portions of the human kidney vascular network. However, as far as the authors are aware, there is no published dataset capturing the intact vascular network of the human kidney beyond approximately six vessel divisions without physical sectioning or subsampling of the tissue. MRI has been used to quantify larger vessels both in vivo and post mortem, but for large volumes of interest (VOI), lacks the resolution capable of imaging small vessels and arterioles. µMRI can be used to image down to ~50 µm/voxel, but is limited to smaller biological samples such as rodent kidneys. CT and µCT have been used extensively to image and analyse rodent renal vasculature, and have also been applied to study the vasculature of ex vivo human lung and fetal kidney. However, no detailed segmentation and quantitative analysis of vascular networks in the human kidney have been performed down to the level of arterioles, because of a lack of available imaging data. Due to this limitation, analysis of human kidney vascular networks is often focused on the first three, large branches of the arterial tree, or limited to subregions within the network. Where multiscale modelling has been performed, parameters from rodent kidneys are assumed to be representative of human kidney vascular networks. However, semi-quantitative studies of human kidney vascular casts have shown large anatomical variation in segmental artery patterning, whilst smaller vessels such as the arcuate arteries, interlobular arteries and afferent or efferent arterioles have not been assessed quantitatively at the organ scale. One imaging modality that could address the challenge of imaging intact organ vascular networks is synchrotron phase-contrast tomography. Hierarchical phase-contrast tomography (HiP-CT) is a technique which leverages the European Synchrotron Radiation Facility’s (ESRF) Extremely Brilliant Source (EBS), a high-energy fourth-generation synchrotron source, to image intact human organs. By utilising the high spatial coherence of the ESRF-EBS and the long beamlines available at ESRF, the development of HiP-CT has allowed the scaling of synchrotron phase-contrast tomography to sample sizes up to and including intact human organs. Datasets created with HiP-CT are hierarchically nested three-dimensional (3D) volumes at multiple resolutions, with exceptional soft tissue contrast spanning from small VOI to the whole intact organ (Fig. 1A). As an example of HiP-CT’s potential, we have previously profiled human glomerular morphology and number across cubic centimetres of intact human kidney. However, the soft tissue contrast achievable with HiP-CT, coupled with its high spatial resolution, potentiates the visualisation and quantification of vascular networks across whole human organs, and could address the limitations of current imaging technologies used to map kidney vascular architecture. A Overview of the hierarchical image volumes that can be acquired with Hierarchical Phase-Contrast Tomography (HiP-CT). Brown, cyan and yellow volumes show the whole organ acquired at 25 µm per voxel, with sub-volumes (Aii) acquired at 6 and 2.6 µm per voxel, respectively, in the intact human kidney. Bi–iv Vascular segmentation performed across the three resolutions of HiP-CT data, enabling the intact arterial network to be visualised and segmented. Red arrows in (Biv) indicate segmented glomeruli. Ci Diagram of the anatomical organisation of the human kidney arterial network, with insert in (Cii) showing the smaller arterioles and capillaries. D The vasculature of the HiP-CT imaged kidney was partitioned into four territories, with each territory denoted by a different colour. Here, we demonstrate how the arterial network of an intact human kidney can be extracted and quantified across multiple length scales using HiP-CT without use of a vascular contrast agent. Our pipeline utilises the benefits of HiP-CT, such as validation of segmentation using multiscale data, whilst also providing solutions for the technical challenges associated with HiP-CT, for example, the collapse of large vessels. Within the human kidney, we delineated the extent and morphology of the vasculature, from the renal artery down to the interlobular arteries. In doing so, we were able to quantify heterogeneity in vascular architecture within the context of ordering schemes describing morphological network branching. We also demonstrate how the multiscale nature of HiP-CT allows estimation of the vascular network between the interlobular arteries and afferent arterioles in smaller VOIs, which we describe as 'local' scans,within the still-intact kidney. We perform a quantitative comparison between our human and previously published rodent kidney vascular networks, the latter of which has been used as inputs for biophysical modelling of kidney vascular blood flow. We further demonstrate how the label-free nature and exceptional soft tissue contrast of HiP-CT allow vascular heterogeneity to be quantified in the context of other anatomical features, such as the different compartments of the kidney. Such spatial variations highlight the link between regional structure and function, reinforcing the importance of quantitative analyses for understanding and modelling regional microenvironments within the human kidney. Using HiP-CT in a hierarchical fashion, we imaged the whole intact kidney obtained from a 63-year-old male organ donor. We initially performed an overview scan of the entire kidney at 25 μm per voxel, followed by selecting and imaging representative VOIs at 6.5 μm per voxel and 2.6 μm per voxel (Fig. 1A). As these image volumes are inherently aligned, expert annotation using renal anatomical landmarks (Fig. 1B) was applied to the image volumes taken at each resolution to produce a multiscale segmented model of the kidney’s arterial network (Supplementary Movie 1). From the segmented data, we were able to identify examples of and interconnect all known anatomical subdivisions of the kidney arterial system (Fig. 1C). The segmental pattern of anterior, posterior, superior and inferior territories supplying the kidney parenchyma was clearly delineated. Each vascular territory (Fig. 1D and Supplementary Movie 2) had a corresponding kidney arterial branch originating from the hilum, which bifurcated before hierarchical branching towards the cortical parenchyma. We next sought to quantitate the arterial network in a reliable and reproducible manner. As we have previously shown that quantitative features of vascular networks can vary by the image processing pipeline used, we developed our own bespoke image processing pipeline (Fig. 2), involving reduction of the initial HiP-CT image to a skeleton, or spatial graph representation, of the arterial network. The graph representation comprises a set of ‘nodes’; defined as 3D locations where vessels meet or end, and ‘segments’; defined as the connections between these nodes (see Supplementary Fig. 6A and Fig. 4A). Our pipeline comprises 8 steps, which are fully detailed in our Supplementary Note 2, and enables the generation of a spatial graph from segmented HiP-CT data, with quantification of error in segmentation (from multiscale comparison) and skeletonization (through application of the skeletonization metric). Step 1: Segmentation is performed with quantitative validation using a higher resolution VOI. Step 2: Skeletonization is optimised by comparison of skeletonization algorithms and the skeleton super-metric. The super-metric is a projection of the distance vector between the reconstructed skeleton and the segmented image onto a weighted space. It contains 5 contributing terms: network volume (Vol.), connected components (CC), Euler Number, Centerline sensitivity (cl sens.) and Bifurcation DICE (BB DICE). Step 3: An initial truncated Strahler order (tSO) calculation is made on the skeletonised network. Step 4: Using the tSO from Step 3 the network can be split into larger calibre (tSO ) and smaller calibre vessels (tSO ). The larger calibre vessel can then be smoothed as shown in insets, orange arrows show the points where smoothing has noticeably acted on regions of larger vessels. Step 5: tSO vs mean radius is plotted for every segment (blue circles); potential collapsed vessels (red crosses) are flagged for all larger vessels and identified in smaller calibre segments by those as having a radius below the 90% percentile for their tSO. Step 6: The segments identified as outliers are presented to an annotator in an interact pop-up window, which allows the annotator to visualise the segment and manually confirm if it should be corrected for collapse. Step 7: For vessels which are confirmed as collapsed, planes which are normal to the centreline of the vessel (indicated by orange arrows) are created at every point along the centreline, these are presented in pop-up windows to the annotator, as are the 2D image for each orthogonal plane (lower panels). From these 2D planes, the collapsed vessel is identified (red cross) and the perimeter (yellow dashed line) is extracted. The perimeter is used to calculate an equivalent radius and assigned as the new radius of the segment. Step 8: The new radii at each point are plotted, and outliers are removed to reduce the effect of any remaining tortuosity in the centerlines. At this stage, the annotator can also manually remove planes that are visibly affected by residual tortuosity. A Schematic diagram of how the metrics in (B–E) are calculated. B The length:diameter ratio. C The branching angle between the child and parent segments. D The tortuosity of segments, E their radius, and F the inter-vessel distance as measured between the midpoint of each segment. Our pipeline first (Fig. 2, Step 1) assesses validation of the segmentation. By aligning segmentations from scans taken at 13 µm per voxel, with VOIs captured at 50 µm per voxel, the higher resolution scans served as ‘ground truth’ for the lower resolution scanning. We used the cl-DICE metric to quantify the overlapping vessel portions finding that 97% of vessels with a vessel lumen radius greater than 50 µm are detectable at 50 µm per voxel. Our next step (Fig. 2, Step 2) comprises the optimisation of the skeletonization algorithm. We applied three different skeletonization algorithms, and utilised the recently developed skeleton super-metric to determine the most suitable algorithm and its parameter optimisation. We found that the Centerline Tree algorithm (Amira-Avizo v2021.1) was the best candidate algorithm, as indicated by its lower super-metric value in comparison to other skeletonization algorithms (Fig. 2, Step 2). Thereafter, several steps were implemented to correct the skeleton for HiP-CT specific challenges (Fig. 2, Steps 3–8), namely the multiscale nature of the vasculature, and the presence of collapsed vessels as a consequence of the ex vivo, label-free HiP-CT protocol. The challenge of multiscale vascular trees was corrected using a truncated Strahler ordering system, which partitions the network into larger or smaller calibre vessels. Smoothing was then applied to all large calibre vessels to reduce tortuosity in the vessel centreline, an artefact which occurs due to the sensitivity of skeletonization algorithms to noise along the vessel surface (Fig. 2, Steps 3 and 4). Following this multiscale smoothing approach, Fig. 2 Steps 5 and 6 involved the identification and manual verification of collapsed vessels. Initially, all large-calibre vessels were flagged as potentially collapsed. Additionally, smaller calibre vessels that were potentially collapsed were identified based on their categorisation below the 10th percentile for radius in their truncated Strahler order (Fig. 2, Step 5). Once identified, collapsed vessels were subject to a bounding box, automatically extracted and thereafter manually determined whether correction of the radius was required to account for collapse (Fig. 2, Step 6). For vessels requiring correction, cross-sectional planes along the vessel centreline were extracted, and the radius was calculated based on the cross-sectional perimeter (Fig. 2, Step 6). Finally, the identification of outlier radii in these cross-sectional planes was performed, using a 95th and 5th percentile windowing for radius along the vessel length. Additionally, an option was applied to manually flag planes that appeared compromised by residual tortuosity in the vessel centreline (Fig. 2, Step 8). The result of this novel pipeline, when applied to our HiP-CT data of human kidney, was the generation the first open-source spatial graph of the intact human kidney arterial vasculature, ranging from renal artery to interlobular arteries. We were able to identify 97% of vessels >50 µm radius across the whole intact human kidney. The final network consisted of 10,193 nodes, 376,603 points and 10,190 vessels. The total network volume was 1.68 × 10 µm, with a length of 2.3 × 10 µm. This spatial graph, which is provided in our Supplementary Information, captures the morphological features and connectivity of the human kidney arterial vasculature, which was then used for downstream analyses as described below. Having created a reproducible spatial graph of the human kidney arterial vasculature, we then performed topological generation and truncated Strahler ordering analyses. This resulted in nine truncated Strahler orders (Fig. 3A) and twenty-five topological generations (Fig. 3B). As the main artery supplying the kidney was cut during autopsy, we inferred that 10 truncated Strahler orders, representing 26 topological generations, were imaged over the intact human kidney with HiP-CT. Rendering of the human kidney vascular network, with vessels coloured according to A Strahler order and B topological generation. Ci Plot showing the number of vessels per truncated Strahler order, with fit for the log plot to calculate branching ratio. Cii Truncated Strahler order against cumulative vascular volume fraction. D One of the VOIs with all glomeruli segmented. E A region within the VOI in (D) segmented at 2.6 µm per voxel, showing connection down to afferent glomeruli. Inset shows the six glomeruli that were connected back to the whole network. The five red arrows indicate glomeruli arising from non-terminal arteries, while the black arrowhead indicates a glomerulus arising from a terminal artery. Strahler ordering, and other approaches to classify vascular networks, have potential caveats (See Supplementary Note 4), for example, the Strahler (or truncated Strahler) order of any individual vessel depends upon the downstream network, and thus on the identification of the network endpoint. Ideally, this endpoint would correspond to the afferent arteriole entering the renal glomerulus. However, at 50 µm per voxel resolution, we were unable to detect afferent arterioles, and thus, the smallest vessels, defined as truncated Strahler order 1 vessels, were the interlobular arteries. Diameter-based statistical approaches to estimation of the Strahler order of the kidney vascular network’s terminal ends were not appropriate to correct for this due to the ex vivo and non-perfused nature of HiP-CT, as well as the connectivity of glomeruli relative to terminal vessel ends. Thus, we applied truncated Strahler ordering to our spatial graph and report how morphological features of vessels in the network vary with truncated Strahler order. We also mapped our truncated Strahler orders to known anatomical subdivisions of the arterial tree to give anatomical context. This resulted in the following classification: truncated Strahler orders 7–9 (n = 25 segments; mean radius = 929 ± 477 µm) mapped to the branches of the kidney artery entering the kidney hilum. Orders 5–6 comprised interlobar arteries (n = 219 segments; mean radius = 417 ± 247 µm), and orders 2–4 arcuate arteries (n = 4841 segments; mean radius = 78 ± 45 µm). Finally, interlobular arteries fell within orders 1–3 (n = 9430 segments; mean radius = 55 ± 23 µm). We further plotted the cumulative volume of the kidney vascular network (Fig. 3Cii), finding that over 20% of the volume of the network lies within Strahler orders 1–4, corresponding to segments from interlobular arteries and arcuate arteries. We found 5105 truncated Strahler order 1 segment and identified a logarithmic relationship between truncated Strahler order and segment number (Fig. 3Ci). Using this relationship, we determined the branching ratio within this subsection of the vascular tree to be 2.92, a value which is similar to that of the human pulmonary arterial tree (3.0) and the rat kidney vasculature (2.85). To provide further context to the truncated Strahler order and to investigate the small-calibre vessels within the human kidney, we leveraged the hierarchical capability of HiP-CT. Using high-resolution VOIs, we segmented and counted all glomeruli within each of the 3 high-resolution VOIs of the HiP-CT data (Fig. 3D). We extrapolated from these VOIs to the total of ~1.2 million glomeruli in the intact kidney, which aligned well with estimates for adult males within a similar age range. Given the 5105 truncated Strahler order 1 segments, the branching ratio of 2.921 and the total number of glomeruli, we estimated that there are a further 4–5 truncated Strahler orders between the end of our whole organ network and the afferent arterioles. To further evaluate this estimate, we assessed on high-resolution VOI, connecting six afferent arterioles of individual glomeruli back to the main vessel tree (Fig. 3E, Supplementary Movie 4 and Supplementary Fig. 8). Of the 6 glomeruli, 5 originated from non-terminal arteries and one from a terminal artery (Fig. 3E, red arrows and black arrows respectively). This supports recent findings, in the rat kidney, which similarly demonstrated the existence of non-terminal branch arterioles, with potential contributions to the synchronicity of blood flow within the kidney. The existence of non-terminal glomeruli also prevents the application of the statistical methods which have previously been used to estimate the true Strahler order of a truncated network. Given the presence of non-terminal glomeruli, statistical estimation of true Strahler order would necessitate connecting a larger number (~1000) of glomeruli back to the main tree to generate an accurate statistical representation of the proportion of terminal to non-terminal glomeruli. Such an estimation cannot be made with this dataset as, even with our highest resolution scans, the small vessels connecting the glomeruli to the main vascular tree could not be annotated reliably for a large number of cases. However, our estimation of total glomeruli number provides a data-driven estimate for the number of missing orders, and thus gives the context needed to support our use of the truncated Strahler order for our ongoing analysis. Vascular network geometric properties, including vessel diameters, lengths and branching angles, are key metrics for quantitative and objective comparison of vascular networks in health or disease. Thus, we extracted and reported the metrics for the human kidney vasculature. Data were grouped according to truncated Strahler order (Fig. 4, Table 1) to enable quantitative comparison to rat and other human organ data. The raw data for each segment, which may serve as inputs for computational models, have been provided as Supplementary Information. Human kidney vascular branching metrics by truncated Strahler generation (means with standard deviation are shown) Further quantitative analysis of the human kidney vascular network revealed that, as truncated Strahler order increased, there was a reduction in the ratio of vessel length:diameter (Fig. 4B). In contrast, the mean radius (Fig. 4E) and inter-vessel distance increased (Fig. 4F). Tortuosity did not vary significantly with truncated Strahler order (Fig. 4D); with most segments possessing tortuosity close to 1, thus implying limited deviation from a straight path. These findings are consistent with anticipated trends for a healthy tissue, wherein a vascular network is assumed to be a fractal structure, with branching pattern driven by optimised delivery of blood to the whole organ. Interestingly, within truncated Strahler orders 8–6, the mean branching angle was approximately 150°, decreasing to 130° for truncated Strahler orders 3–1 (Fig. 4C). Importantly, the latter value of 130° is the predicted optimal theoretical branching angle for volume-constrained vascular growth. Simulation of kidney haemodynamics has previously been performed using μCT data from the rat kidney. To facilitate comparison between existing rat data and our human HiP-CT results, we aligned our network based on the Strahler order allocated to the segmental arteries, thus aligning Strahler order 9 in the previously published rat dataset to truncated Strahler order 8 of our human data. We then related normalised vessel metrics from each species, matching anatomically defined vessel types. The expected increase in vessel radius with order followed a similar trend between human and rat kidney (Fig. 5A). However human kidney vessel radii increased to a greater extent across Strahler orders than in the rat kidney, evaluated based on a fit of log(radius) to Strahler order (Fig. 5B), (p < 0.0001 Sum-of-F test F (DFn, DFd) = 700.6 (2, 12)). To provide additional insights into this difference observed between human and rat kidneys, we extracted radial scaling exponents of the human vascular network. The radial scaling exponent provides insight into how the network has structurally developed with respect to its functions such as the efficiency of blood flow and nutrient delivery to meet metabolic demands and minimise flow resistance. Exponents of 0.33 and 0.5 each have theoretical bases in different models, (i) Murray’s law (expected exponent of 0.33 for all the whole network, derived from considering the energy balance between energy of flow and viscous drag), (ii) the West-Brown-Enquist (WBE) model which predicts 0.5 in larger vessels and 0.33 in smaller vessels, resulting from balancing the energy for metabolic distribution of blood across a fractal-like network. However, deviations from these exponent values and other variants of vascular scaling models have been widely reported. A Normalised radius against Strahler order for our data and for previously published rat kidney vascular data derived from Nordsletten et al.. B The data are plotted for log(Radius), showing a similar pattern but with a significant statistical difference is found between the best fit for the two datasets. Ci Plot of Parent vessel cubed against Sum of cubed child vessels, Murray’s Law is shown in orange hatched line, points from each Truncated Strahler order are differentiated for clarity. Cii Plot of the log of the number of terminal downstream network ends against the radius for all segments in the network. The purple line shows the fit from the Standard Major axis regression, with the intercept a which is the radial scaling exponent and the 95% confidence intervals shown on the plot. Previously in the rat kidney, Nordsletten et al. demonstrated a deviation from Murray’s law by ~1% for the rat kidney. Figure 5Ci shows the cubed parent radius plotted against the cube sum of the child radii for our dataset, the theoretical Murray’s law is overlaid (orange hatched) to allow qualitative comparison to previous literature. To quantitatively evaluate whether our data are better represented by Murray’s law (expected radial scaling exponent 0.33) the WBE model (radial scaling exponent 0.33 in small and 0.5 in large vessels) or another model, we extracted the number of downstream terminal ends of the network for each vessel segment and the radius of that segment. Through a log-log plot of the data (Fig. 5Cii), the theoretical value of the exponent a was found to be 0.55. This value is higher than Murray’s law and closer to the WEB model and the values found in ref. for the human pulmonary artery system. We then sought to compare heterogeneity in morphology of the human kidney vasculature according to anatomical regions within the human kidney, which may reflect specialised vascular functions. For example, the medulla of the kidney is predominantly vascularised by vasa recta; specialised capillaries which possess low oxygen tension. This configuration leads to physiological hypoxia that is inherent to the medulla’s urinary concentration mechanisms. Further reflecting the importance of vascular morphology is the longstanding hypothesis, supported by blood oxygenation level-dependent MRI studies, that vascular rarefaction in CKD results in hypoxia within the kidney cortex. In turn, this stimulates neighbouring cells into a pro-fibrotic phenotype, manifesting in replacement of normal kidney tissue by fibrosis and heralding loss of organ function. Thus, regional heterogeneity of vascular morphology is fundamental for sustaining local microenvironmental features, such as hypoxia, that influence specialised organ functions. However, regional heterogeneity in vascular structure has not been quantitatively explored in the human kidney. Leveraging the contrast-free approach of HiP-CT, we were able to segment the kidney into known anatomical compartments, including hilum, medulla, intramedullary kidney columns and cortex (Fig. 6Ai). We compartmentalised the vascular network according to these anatomical compartments (Fig. 6Aii). The total tissue volume of each compartment, in addition to the number of vessels, length, radius and volume of segmented vessels within each compartment, were quantified (Table 2). Most of the tissue volume of the human kidney was occupied by the cortex (63.7%) as compared with the medulla (23.5%), hilum (8.7%) or intermedullary pillars (4.1%). The number of segments of the vascular network within each compartment followed this trend. We then quantified (Fig. 6Aiii) and mapped (Fig. 6Bi–Biii) the inter-vessel distance, compartmentalised by hilum, medulla, cortex, and intermedullary pillars. Mean inter-vessel distances were calculated for each compartment, assessing the distribution of inter-vessel distance from the renal artery down to interlobular arteries (Table 2). The medulla had the highest inter-vessel distance. Whilst the cortex had a comparatively smaller inter-vessel distance than medulla and hilum, a large standard deviation for this value was noted within the cortex. This is illustrated by the heatmap in Fig. 6Bi, Bii, which identified small areas with inter-vessel distance >4.5 mm localised towards the kidney capsule. Ai 3D surface masks of the kidney cortex (green), medulla (yellow) and hilum (pink), inter-medullar pillars (dark blue). Aii 3D reconstruction vasculature colour according to anatomical compartment within the human kidney cortex (green), medulla (yellow) and hilum (pink), inter-medullar pillars (dark blue). Aiii Inter-vessel distances are plotted against the total number of vessel voxels for each kidney compartment. Bi Visual heatmap of inter-vessel distance for the entire human kidney, where pink represents the largest inter-vessel distance (>4.5 mm) and white (0 mm) the smallest. Bii A digital zoomed region within cortex and medulla. Biii The 2D slice of the associated HiP-CT raw image with the compartments overlaid. Human kidney vascular branching metrics compartmentalised by spatial zone with the organ *Segments that crossed over two regions were excluded. Owing to the limited volume of tissue that can be imaged at high resolution using ex vivo 3D imaging modalities, such as μCT and lightsheet microscopy, and insufficient resolution of technologies routinely used in clinical practice, such as CT and MRI; it had previously been impractical to capture the vascular network of the intact adult human kidney beyond the very largest arteries. Here, using a synchrotron-phase contrast tomography technique, termed HiP-CT, we were able to image, segment and quantify the human kidney arterial network within an intact human kidney from renal artery to down to the level of interlobular arteries, without the need for exogenous contrast agents. With HiP-CT, we show that vessels which have not been imaged in the intact human kidney previously, namely the interlobar to interlobular arteries, occupy approximately 20% of the arterial vascular volume of the organ. By imaging VOIs in the intact kidney at higher resolution, and aligning this with our lower resolution scans, we further demonstrate that, akin to rat, and varying from the traditional hierarchy of the kidney vasculature observed in nephrology and anatomical textbooks, glomeruli in humans can originate from non-terminal arterioles. In further comparisons with the rat kidney vasculature, we found that although similar trends in vascular radius were seen, there was a significant difference in the change in radius with vessel order between species. This may be explained by the larger radii range in the human kidney between renal artery and afferent arterioles, relative to the volume of the human kidney; but could also be dependent on the difference in approach to calculation of the Strahler order for each study. We also found that the exponent for radial scaling is closer to the WBE model (0.5) than Murray’s law value of 0.33. This is broadly in alignment with previous work, where exponents of 0.47–0.58 were found for trees with vascular diameters ≥200 µm and 70–20≥ µm, respectively. Wide variation between theoretical exponents and those derived from real imaging data is widely accepted and often attributed to the complexity of real vessels including factors such as mechanical strain, the elastic nature of arteries during pulsatile flow and turbulent flow patterns. Specific to the kidney, while Murray’s law or the WBE model assume idealised flow-optimised network, or ideal fractal scaling, kidneys likely exhibit non-optimal but functional scaling due to their high-resistance, low-compliance vascular network which is needed to support hemodynamic fluctuations due to changes in glomerular filtration and autoregulation. However, it should also be noted that due to the ex vivo nature of HiP-CT, and the consequent lack of vascular tone, extracting such radial scaling laws from these data may have additional sources of error compared to in vivo imaging techniques. Deviations from theoretical laws support the idea that vascular systems adapt based on tissue-specific demands rather than universal optimisation principles. This idea can be further supported by examining regional heterogeneity of vascular morphology in different anatomical zones of the kidney. The segmentation of hilar, medullary, intramedullary and cortical zones of HiP-CT images from the same kidney support this hypothesis. For example, the increased inter-vessel distance observed within the medulla, as compared to the cortex, is pertinent. The medulla experiences physiological hypoxia, and increased inter-vessel distance, paired with the oxygen diffusion limit, provides a potential anatomical rational for this phenomenon, in addition to the unique solute and gas exchange mechanisms that take part in this region of the kidney. The data provided in this study, and resultant insights into how morphology of the kidney vasculature varies by different renal compartments, could shed light on the mechanisms underpinning the unique cellular and molecular adaptations of specialised endothelia across the kidney vascular network. Our pipeline and the HiP-CT data provide a framework to potentially study how the vasculature within each anatomical compartments is differentially affected by kidney disease, with potential for understanding the basis of vascular rarefaction and pathological hypoxia. Further studies with higher resolution HiP-CT or with microfill of the human kidney could potentially preserve vessel radius more accurately and resolve capillaries allowing extension of the work. Such information is important to acquire in human samples, as it could potentially influence simulations of haemodynamics, oxygenation or drug delivery; and generation of synthetic vessel trees for in silico experiments. The human kidney vasculature is exquisitely specialised to meet the physiological demands of the kidney. Underpinning this specialisation is the cellular and molecular heterogeneity of endothelial beds within the renal vasculature, of which we are gaining an increasing understanding due to the advent of improved techniques such as single-cell and spatially-resolved transcriptomics. The rapid and recent advances in our understanding of cellular and molecular heterogeneity of the kidney vasculature has not been matched by structural insights, likely because of limitations in imaging technologies. We have overcome many of these limitations using HiP-CT, where the exceptional contrast, coupled with appreciable spatial resolution at scale, allows us to capture and segment the 3D vascular architecture of an intact human kidney. Furthermore, within high-resolution VOIs, HiP-CT allows glomeruli and afferent arterioles to be segmented and, in selected cases, be connected back to the vascular tree of the intact whole organ. Robust and reproducible analysis of vascular networks relies on the careful application of a multi-stage image processing pipeline, which we have outlined in this paper. We have developed an approach which utilises multiple annotators and comparison to higher resolution scans to validate segmentation accuracy as a crucial first step. Following segmentation, we have developed a skeletonisation approach, which can be scaled to large datasets, and also provides corrections for radius estimation when portions of the vasculature have collapsed. Finally, we applied a truncated Strahler ordering to the vessel spatial graph, providing a meaningful ordering system with respect to known anatomical vessel descriptions, as well as facilitating quantification of individual vessels within the vascular hierarchy. By developing and applying this pipeline, we have produced quantitative vascular branching metrics from an intact human organ for the first time. These metrics exceed other studies on cadaveric human kidney cast and dye injections, which report arterial branches corresponding to truncated Strahler orders 7–9. We provide quantitative comparison between the human kidney vasculature and that of the rat, the latter of which has been key for inputs to generate biophysical models of kidney haemodynamics. The quantitative analysis pipeline performed in this paper serves multiple purposes. First, it allows the whole kidney vasculature dataset to be represented in a single spatial graph, comprising only kilobytes of data. This spatial graph, which is provided as Supplementary Data, is readily quantifiable. Whereas prior simulations of kidney haemodynamics and perfusion have relied on seminal μCT studies performed in rat, we provide, for the first time, a map of the kidney arterial network from renal artery to interlobular arteries. We demonstrated our segmentation approach to be accurate, with 97% of vessels of >50 µm radius captured across the intact human kidney. These data thus provide vital inputs for biophysical modelling of kidney physiology. The data also serves as a reference to study kidney diseases, in which vascular rarefaction is a pathophysiological hallmark. The pipeline described could be used to generate vascular maps from multiple kidneys, or other human organs, potentiating spatial ‘atlases’ of human organ vasculature across healthy and pathological contexts. Beyond these, our openly available dataset has immediate practical applications, such as providing inputs for bioprinting and tissue engineering of artificial kidneys or planning surgical resection of kidney tumours whilst preserving kidney function. These datasets can also be used as a tool for medical education and training, as well as for the creation and advancement of surgical methods. There are several limitations of this work. Firstly, the low throughput of HiP-CT vascular segmentation warrants discussion. Here, we present the complete analysis from a single kidney as a framework for future studies of kidneys in health and disease or other intact human organs. The accuracy of the segmentation, however, lays a foundation for tools such as machine learning methods for automated segmentation of blood vasculature from imaging data. HiP-CT imaging still cannot resolve afferent arteriole or capillary resolution across the whole organ, meaning that the contributions of peritubular capillaries or vasa recta are not incorporated. This also creates challenges for applying ordering schemes such as the Strahler order, where the true 0th order is the capillary bed. Previous approaches to estimating the distance of a terminal end in a truncated network from the capillary bed have relied on utilising diameter measurements of vessels to iteratively update the Strahler order of terminal ends. This facilitates the creation of a connectivity matrix to estimate the downstream network. However, diameter estimation is less accurate for HiP-CT, where vascular collapse makes radius estimates less consistent than, for example, when using microfill techniques. Using the high-resolution VOIs, we also demonstrated that glomeruli frequently emanate from non-terminal arterioles. Without connecting on the order of 1000s of glomeruli back to the main vascular tree as performed for small portions of rat kidney, a connectivity matrix cannot be developed. However, the utility of any vascular classification scheme relies upon the ability to distinguish morphologically distinct vessel types, and to show logarithmic relations between morphology and classification orders. Our truncated Strahler approach creates vessel orders which are able to separate morphologically distinct vessels (Supplementary Note 4), as well as demonstrating logarithmic relationships for radius and vessel number. We also found that truncated Strahler ordering also aligns well with the anatomically defined vessel classifications, as was the case for rat kidney. The future of HiP-CT and mapping the kidney vasculature is promising. The upcoming ESRF beamline (BM18) enables longer propagation distances than shown here, dramatically increasing the contrast sensitivity for the lower resolution scans. Further developments in scanning and data handling have already extended the capabilities of HiP-CT to create whole kidney overview datasets, with voxel sizes down to 9 µm/voxel, and to submicron voxel sizes in VOIs. Thus, future studies can leverage the greater detail available on low-resolution scans of the whole kidney, providing the potential to further assess phenomena such as the emergence of glomeruli from non-terminal arterioles, or potentially map entire organs down to the capillary level. As these developments unfold, we have created an open-access data portal (https://human-organ-atlas.esrf.eu/), enabling download and use of HiP-CT data by biomedical researchers across the world. In summary, we have achieved quantitative mapping of the arterial network of an intact human kidney, from renal artery to interlobular arteries, for the first time. This vital step progresses our understanding of how physical properties of the kidney vasculature relate to cellular and molecular heterogeneity, whilst generating key inputs for future biophysical modelling of human kidney vascular physiology. Ultimately, we envisage that mapping of microstructural detail will become routine at the scale of the whole kidney, providing a means to link cellular events with organ physiology and pathology. An intact human kidney was obtained from a 63-year-old male (cause of death: pancreatic cancer), who consented to body donation to the Laboratoire d’Anatomie des Alpes Françaises before death. Transport and imaging protocols were approved by the French Health Ministry. Post mortem examination was conducted according to Quality Appraisal for Cadaveric Studies scale recommendations. The body was embalmed by injecting 4500 mL of 1.15% formalin in lanolin, followed by 1.44% formalin, into the right carotid artery, before storage at 3.6 °C. During evisceration of the right kidney, vessels were exposed, and the surrounding fat and connective tissue were removed. The kidney was post-fixed in 4% neutral-buffered formaldehyde at room temperature for one week. The kidney was then dehydrated through an ethanol gradient over 9 days to a final equilibrium of 70%. Each solution was four-fold greater than the volume of the organ, and, during dehydration, the solution was degassed using a diaphragm vacuum pump (Vacuubrand, MV2, 1.9m/h) to remove excess dissolved gas. The dehydrated kidney was transferred to a polyethylene terephthalate jar where it was physically stabilised using a crushed agar-agar ethanol mixture, and then imaged. Imaging was performed on the BM05 beamline at the ESRF following the HiP-CT protocol. Initially, the whole kidney was imaged at 25 µm per voxel (isotropic edge length). VOIs within the same kidney were also imaged at 6.5 and 2.6 µm per voxel. Tomographic reconstruction was performed using the PyHST2 software and following the steps detailed in previous studies. Briefly, a filtered back-projection algorithm, with single-distance phase retrieval, coupled to an unsharp mask filter, was applied to the collected radiographs. Reconstruction and scanning parameters are provided in Supplementary Note 1, Supplementary Tables 1 and 2. The reconstructed volumes were binned (averaged) to 50, 13, and 5.2 µm per voxel, respectively, to increase the signal-to-noise ratio, reduce inter-annotator variability and reduce computational load for subsequent image segmentation and quantification (see Supplementary Fig. 1). All reconstructed image volumes and metadata can be accessed at human-organ-atlas.esrf.eu. A table for direct DOI links for each dataset is provided in Supplementary Table 2. Prior to manual segmentation, images were filtered to enhance blood vessel contrast using Amira-Avizo (v2021.1) software. A 3D median filter (iterations = 2 and 26 neighbourhood analysis) was used to reduce image noise. Image normalisation was performed using background detection correction (default parameter settings). A manual segmentation of the arterial networks was performed in Amira-Avizo using a combination of methodologies. First, a 3D region growing tool was used, where the user selects an initial voxel within a vessel lumen along with set intensity and contrast thresholds. Any voxel within the connected neighbourhood of the initially selected voxel with an intensity and contrast within thresholds are added to the region. Multiple seeds points and thresholds, as well as manual limits on the region, are used by the annotator to ensure the lumens of all vessels are accurately identified. In areas of collapsed or blood-filled vessels, annotators manually paint lumen voxels utilising three orthogonal views to ensure connection of the vascular network. An annotator continues this process in an iterative fashion by selecting seed points, altering the thresholds and manually correction, resulting in expansion of an interconnected vascular network (Method shown in Supplementary Movie 3). Once the first annotator has filled the interior of all vessels, data are passed to a second annotator, who repeats the process, but starting in reverse slice order. A third annotator serves as a proofreader by quantitatively reviewing the labels. The proofreader is presented with 5–9 randomised 2D slices of the data within any one of three orthogonal planes. They then count the number of vessels cross-sections present in the slice, recording the true positive and false negative number of vessel cross-sections that have been segmented. The proofreader returns the data to the initial two annotators, highlighting areas where vessels are not identified. This three-annotator process repeats iteratively until the proofreader does not find any false negatives. This method was applied to segment the kidney arterial network from the intact human kidney from the imaging data at 50 µm per voxel, and portions of the same network in the 13 and 5.2 µm per voxel datasets, approximately 250–300 h were needed to segment the kidney in this way. A second approach to independently and quantitatively validate the segmentation of the lowest resolution data was performed using segmented VOIs of the higher resolution, 13 µm per voxel dataset. Here, the 13 µm per voxel VOIs were rigidly registered to the whole organ volume using the affine registration toolkit (Amira-Avizo) (See Supplementary Note 1.1, Supplementary Fig. 2 and Supplementary Tables 3 and 4). Overlapping portions of the 13 µm voxel segmentations and 50 µm per voxel datasets were extracted, and the 50 µm per voxel datasets were upsampled to the resolution of the 13 µm voxel dataset. An overlap measure, termed topological precision and recall score, following Paetzold et al., was applied (see Supplementary Note 2.1 and Supplementary Fig. 3). To quantify branching metrics of the human kidney vasculature, the segmented 3D vascular network at 50 µm per voxel was skeletonised using the centreline tree algorithm in Amira-Avizo v2021.1. The choice of skeletonisation algorithm and the parameterising of the algorithm were optimised by utilising the super-metric approach, outlined by Walsh and Berg et al. (tube parameters: slope = 4 and zeroval = 10, see Supplementary Note 2.2 and Supplementary Fig. 4 for parameter optimisation results). The resulting spatial graph describes the vessel network in terms of ‘nodes’, ‘points’, ‘segments’, and ‘subsegments’. A segment is defined as being between a start and end node, corresponding to either a branching point leading into another segment branch, or a terminal end where no further branches were detectable. Between the start and terminal node of each segment lie subsegments with ‘points’, marking the start and end of each subsegment. Each subsegment has an associated radius and length (Supplementary Fig. 6A). A multiscale smoothing approach was applied to the larger vessels (those of Truncated Strahler order greater than 5). Iterative weighted smoothing was performed, where the smoothed location of any point is given by iteratively calculating weighted average of the current and two neighbour points. Parameter values were found empirically as 0.8, 0.1 and 15 for the neighbour points, current point and iterations, respectively. This reduced the tortuosity in the larger vessels (Fig. 2, Step 4), which occurs artefactually due to noise on the surface of the segmented large vessels, and which if uncorrected, impacts severely on the correction for collapsed radius vessels. Correction for collapsed vessels was delineated into two distinct cases. One case is a scenario in which there is a small collapsed portion in an otherwise patent vessel (Supplementary Fig. 5Bi, Bii). The second case applied when the majority of the vessel is collapsed (Supplementary Fig. 5Cii). The reduction in radius in the skeletonised form of the networks can be seen in Supplementary Fig. 5Bii, Cii. Correction for short subsegment collapsed vessels was performed by plotting radius along each segment. Subsegments where the radius was below the 5th percentile for that segment were replaced with the nearest neighbour. For larger collapsed vessels, the process is fully described in the “Results” section. Topological/morphological metrics of the network were calculated from the spatial graph as follows, with code provided at https://github.com/HiPCTProject/Skeleton_analysis: Branching angle is calculated as either: (a) the angle between the two child segments from a common parent segment, or (b) the angle between a child segment and its parent segment. In both cases, the vector for the segment of parent and child were calculated between the start node and end node, irrespective of vascular tortuosity. Tortuosity is defined as the Euclidean distance between start and end node of a segment, divided by the sum of all subsegment lengths. Radius is calculated per segment as the mean of all subsegment radii. In cases where larger vessels had fully collapsed (See “Results” for details), the radius was defined as the equivalent radius for a circular vessel with the same length perimeter as the vessel cross-section in the binary image. Length is defined as the sum of all subsegment lengths. Inter-vessel distance is calculated by two approaches to facilitate different analyses. First, using the segmentation binary image, the distance of every non-vessel voxel from its nearest vessel voxel was calculated via a 3D distance transform (ImageJ) applied to the binary vessel segmentation. Second, using the skeleton form, the Euclidean distance between the midpoint of every segment to its nearest-neighbouring segment midpoint was calculated. In addition to the above metrics, we also assessed vessel generation, or order, using two methods. First, we used a variation of the centripetal system, known as the Strahler ordering system, wherein the most distal, smallest segments are assigned as the first order. If two segments with the same order intersect, the resulting segment has one Strahler order greater. Alternatively, if two segments with different orders intersect, the higher order of the two is given to the resulting segment (Supplementary Fig. 6B). We used a variant of the Strahler order approach, which we term the truncated Strahler approach. Our 50 µm per voxel dataset does not provide sufficient resolution to image or segment down to afferent arterioles. Thus, the network created from this 50 µm per voxel dataset is truncated at the interlobular arteries. We assigned the terminal ends of our network as the first Strahler order, as opposed to applying statistical estimates to determine the Strahler order of these terminal ends based on diameter relative to the afferent arterioles, as performed previously. Detailed discussion of this approach and alternative ordering approaches are discussed in the Supplementary Note 3. Second, we took a centrifugal, or ‘topological’ approach, starting with the most proximal artery as generation one. At each branching node the generation is increased, an approached which has been previously utilised (Supplementary Fig. 6C). From the ordering analyses, we assessed the branching ratio () defined as the anti-log of the reciprocal for the linear fit to the plot of truncated Stahler order (O), against the logarithm of the number of segments (N) in each order: The radius of the arterial network in the human kidney obtained from this study was compared to those of the rat kidney taken, which was scanned with 20 and 4 µm per voxel using a microfilling approach. The radial scaling exponent for vascular networks refers to the relationship between the radii of parent and daughter vessels at a bifurcation. It is formulated as: Where is the scaling exponent. Murray’s Law describes an optimisation principle that minimises energy costs in blood flow by balancing viscous dissipation and metabolic maintenance, predicting a cubic relation, leading to a scaling exponent a = 0.33. Murray’s law is derived from assuming minimal work in maintaining blood transport, leading to vessel radii scaling with the cube root of flow rate. Murray’s law also assumes, constant uniform metabolic demand of the tissue, laminar flow and that blood is a Newtonian fluid. In contrast, the West, Brown, and Enquist (WBE) model describes vascular networks as fractal-like structures that optimise metabolic energy distribution across the vascular network in its entirety. It predicts a 0.5 scaling exponent for large vessels and 0.33 for small vessels, accounting for hierarchical branching, where the emphasis is on efficiently delivering nutrients and waste exchange throughout the system. Both models provide insights into vascular architecture but differ in scope, with real vascular networks often deviating due to biological variability and tissue-specific adaptations. Calculation of the radial exponent is done in this work following the regression-based method outlined in refs. . For each vessel in the network, the number of downstream endpoints of the network is counted. The radius and number of downstream tips are related by Eq 4. from ref. : Where = radius of a vessel in the network, = the number of downstream endpoints from that vessel, =the radial scaling exponent. Plotting the log-log relation of these two variables allows a to be estimated by regression analysis. Segmentation of the compartments within the human kidney, including cortex, medulla, intermedullary pillars and hilum, was performed in Dragonfly (version: 2021.3) using a 2D convolutional neural network (CNN). The final hyperparameters of the CNN are given in Supplementary Table 4. Correction of the CNN output was manually performed in by an expert in Amira-Avizo v2021.1 to provide the final compartment delineations. These compartments were used to group and then analyse vascular network parameters. 40 3D patches (512 × 512 × 512) of the highest resolution data, captured at 2.6–5.6 µm per voxel, were extracted from multiple human kidneys scanned by HiP-CT, and the glomeruli were manually segmented. The widely utilised network nnU-net was trained using 35:5 cubes for a train:test split and a 70:30 training validation split. Training using 5-fold cross validation achieved a final DICE score of 0.928, 0.860, 0.906 for training, validation and test data, respectively. See Supplementary Note 3 for training results and nnU-net configuration. The plan files detailing all parameters for the training nnU-net are provided in Supplementary data. This trained network was used to perform inference of two VOIs of high-resolution data from the human kidney in this study, and count the number of glomeruli in each. Utilising the kidney anatomical compartment segmentation from above, the volume of cortical tissue within these high-resolution VOIs was calculated. For each VOI, the number of glomeruli and the volume of cortex in each VOI were used to estimate the total number of glomeruli in the entire kidney. Estimates of total glomerular number extrapolated to the entire kidney, from each VOI, were: 1.28 × 10 and 1.12 × 10 for VOI 3.1 and VOI 2.1, respectively. Statistical comparisons of vascular network morphology between human and rat kidney were performed in GraphPad Prism (version: 10.1.2). For all statistical tests, a p-value of less than 0.05 was considered statistically significant. In both the rat and human datasets, the segmental/feeding renal arteries were identified to be at Strahler orders 8 and truncated Strahler order 9, respectively. Radius against Strahler order were normalised to the 9th truncated Strahler order of the human data. Log of radius against truncated Strahler generation for the human kidney; and radius against Strahler Order of the rat kidney, were plotted facilitating a linear least squares regression analysis. A sum of squares F test was performed with the null hypothesis that a single set of global parameters for slope and intercept would fit vessel radius or vessel length for both the rat and human cases. For calculating the fit of the radial scaling exponent (a), we followed the approach of ref. applying a Standard major axis regression to account for measurement error in both variables. This was performed in Matlab 2023a using the gmregress.m function with an alpha significance set to 0.05. This project has been made possible in part by grants DAF2020-225394 and 2022-316777 (10.37921/331542rbsqvn) from the Chan Zuckerberg Initiative DAF, an advised fund of the Silicon Valley Community Foundation, and grant CZIF2021-006424 from the Chan Zuckerberg Initiative Foundation, the MRC (MR/R025673/1), the RAEng (CiET1819-10), and ESRF beamtimes (md1252 and md1290). D.A.L.’s laboratory is supported by a Wellcome Trust Investigator Award (220895/Z/20/Z) and the NIHR Biomedical Research Centre at Great Ormond Street Hospital for Children NHS Foundation Trust and University College London. D.J.J. was supported by a Rosetrees Trust PhD Plus Award (PhD2020\100012), a Foulkes Foundation Postdoctoral Fellowship, a Wellcome Trust Accelerator Award (314710/Z/24/Z) and the Specialised Foundation Programme in the East of England Foundation Schools. The imaging data that form the basis of the study findings are freely available at the ESRF data repository (https://human-organ-atlas.esrf.eu) with links provided in Supplementary Note Supplementary Table 2. The spatial graph data of the kidney arterial network, along with the computed morphological parameters, can be accessed via links provided in Supplementary Note 7. Codes for skeleton correction and topological analysis are available at https://github.com/HiPCTProject/Skeleton_analysis. Claire Walsh is an Associate Editor for NPJ Imaging, but was not involved in the editorial review of, or the decision to publish this article. All other authors declare no competing interests. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The imaging data that form the basis of the study findings are freely available at the ESRF data repository (https://human-organ-atlas.esrf.eu) with links provided in Supplementary Note Supplementary Table 2. The spatial graph data of the kidney arterial network, along with the computed morphological parameters, can be accessed via links provided in Supplementary Note 7. Codes for skeleton correction and topological analysis are available at https://github.com/HiPCTProject/Skeleton_analysis.
PMC6892818
CD40 induces renal cell carcinoma-specific differential regulation of TRAF proteins, ASK1 activation and JNK/p38-mediated, ROS-dependent mitochondrial apoptosis
A unique feature of CD40 among the TNF receptor (TNFR) superfamily is its exquisitely contextual effects, as originally demonstrated in normal and malignant B-lymphocytes. We studied renal cell carcinoma (RCC) in comparison to normal (human renal proximal tubule) cells, as a model to better understand the role of CD40 in epithelial cells. CD40 ligation by membrane-presented CD40 ligand (mCD40L), but not soluble CD40 agonist, induced extensive apoptosis in RCC cells; by contrast, normal cells were totally refractory to mCD40L. These findings underline the importance of CD40 ‘signal-quality’ on cell fate and explain the lack of pro-apoptotic effects in RCC cells previously, while confirming the tumour specificity of CD40 in epithelial cells. mCD40L differentially regulated TRAF expression, causing sustained TRAF2/TRAF3 induction in RCC cells, yet downregulation of TRAF2 and no TRAF3 induction in normal cells, observations strikingly reminiscent of TRAF modulation in B-lymphocytes. mCD40L triggered reactive oxygen species (ROS) production, critical in apoptosis, and NADPH oxidase (Nox)-subunit p40phox phosphorylation, with Nox blockade abrogating apoptosis thus implying Nox-dependent initial ROS release. mCD40L mediated downregulation of Thioredoxin-1 (Trx-1), ASK1 phosphorylation, and JNK and p38 activation. Although both JNK/p38 were essential in apoptosis, p38 activation was JNK-dependent, which is the first report of such temporally defined JNK-p38 interplay during an apoptotic programme. CD40-killing entrained Bak/Bax induction, controlled by JNK/p38, and caspase-9-dependent mitochondrial apoptosis, accompanied by pro-inflammatory cytokine secretion, the repertoire of which also depended on CD40 signal quality. Previous reports suggested that, despite the ability of soluble CD40 agonist to reduce RCC tumour size in vivo via immunocyte activation, RCC could be targeted more effectively by combining CD40-mediated immune activation with direct tumour CD40 signalling. Since mCD40L represents a potent tumour cell-specific killing signal, our work not only offers insights into CD40’s biology in normal and malignant epithelial cells, but also provides an avenue for a ‘double-hit’ approach for inflammatory, tumour cell-specific CD40-based therapy.CD40 is a member of the tumour necrosis factor (TNF) receptor (TNFR) superfamily and interaction with its ligand, CD40L (CD154), plays a crucial role in immune responses. Yet, apart from hematopoietic cells, such as B cells and antigen-presenting cells, CD40 is also expressed by epithelial cells of various origins, including bladder, liver and ovarian carcinoma cells, normal epithelial cells as well as endothelial cells. There is accumulating evidence that CD40–CD40L signalling may strongly influence non-lymphoid cell fate in addition to inducing cytokine secretion. CD40 lacks kinase activity and related intracellular signalling motifs, thus utilises adapter molecules TNFR-associated factors (TRAFs) for signal transduction, with TRAF2 and TRAF3 being the main TRAF proteins that play significant and often opposing roles in CD40 signalling. After CD40L-mediated activation, CD40 translocates to lipid rafts, where it associates with various TRAFs to activate downstream mitogen-activated protein kinases (MAPKs). Recent studies provided evidence that the consequences of CD40 ligation may differ in normal and malignant cells, thus its effects may be highly context-specific. Moreover, the ‘quality’ of the CD40 signal may determine whether CD40L–CD40 interactions are pro-apoptotic: extensive receptor cross-linking by membrane-presented CD40L (mCD40L) causes extensive apoptosis, while soluble agonists (e.g. agonistic antibodies) cause little apoptosis. However, the mechanisms that define these properties of CD40 have only recently started to become investigated. CD40 expression has been demonstrated in normal renal cells and their malignant counterparts (renal cell carcinoma, RCC). CD40 is expressed by human RCC lines in vitro and its stimulation by soluble agonist triggered secretion of pro-inflammatory cytokines, yet no direct pro-apoptotic effects have been reported. Others reported that in primary RCC cultures stimulation of CD40 by soluble agonists caused proliferation and enhanced cell motility. By contrast, in normal proximal tubule cells cultured in vitro CD40 ligation induced both pro-inflammatory cytokine secretion and anti-inflammatory signals. Therefore, the consequences of CD40 ligation in both normal and RCC cells remain partially controversial and essentially unexplored. Here, we provide the first detailed investigation on the effect of CD40 ligation in human RCC cells as well as their normal counterparts. We demonstrate that mCD40L induced extensive apoptosis accompanied by pro-inflammatory cytokine secretion in malignant cells, while soluble CD40 agonist is weakly apoptotic. Importantly, normal cells were completely refractory to CD40-mediated killing. CD40 ligation differentially regulates TRAF2 and TRAF3 expression in normal vs. malignant cells and apoptosis induction involves a signalling pathway that entrains ASK1 activation and reactive oxygen species (ROS)-mediated death, with JNK and p38 being sequentially involved in the induction of the intrinsic Bak/Bax-associated mitochondrial apoptotic pathway. To investigate CD40 expression in RCC cells, we used the well-characterised cell lines ACHN, 786-O and A-704 (ref. ). Expression was also examined in primary human renal proximal tubule cells, HRPT. By immunoblotting, all RCC lines were CD40 positive, with the greatest expression observed in A-704 and ACHN, while lower expression was observed in 786-O and in normal (HRPT) cells (Fig. 1a). Expression was compared to HCT116 (CD40 positive) and SW480 (CD40 negative) colorectal carcinoma (CRC) lines.Fig. 1CD40 expression by normal (HRPT) and malignant (RCC) cells and its regulation by pro-inflammatory cytokines.a Detection of CD40 expression in RCC lines ACHN, 786-O and A-704 and in normal HRPT cells by immunoblotting. Lysates from HCT116 and SW480 cell cultures served as CD40+ve and CD40-ve controls, respectively. Correct loading (20 µg protein/lane) was confirmed by β-actin detection. b Surface expression of CD40 by HRPT cells and RCC lines was examined by flow cytometry using CD40-PE (red histograms) in comparison to control PE-conjugated (black histograms) antibody. HCT116 and EJ cell lines served as CD40+ve controls. c RCC cells were treated with TNF-α (green histograms) or IFN-γ (purple histograms) for 48 h and CD40 expression was assessed alongside untreated cells by flow cytometry. Overlay histograms from a representative experiment are shown on the left and a summary of two independent experiments is shown on the right. Bars show mean median fluorescence intensity (MFI) of three technical replicates ± SEM. a Detection of CD40 expression in RCC lines ACHN, 786-O and A-704 and in normal HRPT cells by immunoblotting. Lysates from HCT116 and SW480 cell cultures served as CD40+ve and CD40-ve controls, respectively. Correct loading (20 µg protein/lane) was confirmed by β-actin detection. b Surface expression of CD40 by HRPT cells and RCC lines was examined by flow cytometry using CD40-PE (red histograms) in comparison to control PE-conjugated (black histograms) antibody. HCT116 and EJ cell lines served as CD40+ve controls. c RCC cells were treated with TNF-α (green histograms) or IFN-γ (purple histograms) for 48 h and CD40 expression was assessed alongside untreated cells by flow cytometry. Overlay histograms from a representative experiment are shown on the left and a summary of two independent experiments is shown on the right. Bars show mean median fluorescence intensity (MFI) of three technical replicates ± SEM. Flow cytometry corroborated cell-surface CD40 expression (Fig. 1b), in comparison to HCT116 and urothelial cell carcinoma (UCC) line EJ. Pro-inflammatory cytokines can up-regulate CD40 expression in epithelial cells such as urothelial (EJ) and colorectal (HCT116). Treatment with TNF-α and particularly IFN-γ increased expression, with the greatest induction observed in 786-O (Fig. 1c). Hence, both normal (HRPT) and malignant (RCC) cells express CD40 and expression can be increased by TNF-α and IFN-γ treatment. We initially activated CD40 in the panel of RCC lines using soluble agonist, in particular cross-linked agonistic anti-CD40 G28-5 mAb. Treatment caused limited RCC cell death at 48 h, as was observed in positive control EJ (Fig. 2a)—similar results were obtained at earlier/24 h or later/72 h time-points (not shown).Fig. 2Membrane-CD40L (mCD40L) but not soluble CD40 agonist induces extensive cell death (apoptosis) in RCC lines.a ACHN, 786-O, and A-704 cells were either untreated (Control) or treated with 10 µg/ml cross-linked agonistic anti-CD40 mAb (G28-5) for 48 h and death was detected using the CytoTox-Glo assay (see Methods). Results are representative of at least two experiments and are presented as background-corrected relative luminescence unit (RLU) readings. EJ cells were included as positive controls. Bars show mean RLU of 3–4 technical replicates ± SEM. b ACHN, 786-O, and A-704 cells were co-cultured with 3T3Neo (Control) or 3T3CD40L (mCD40L) effector cells for 48 h and death was detected using the CytoTox-Glo assay. Results are typical of tive independent experiments and are presented as background-corrected relative luminescence units (RLU). EJ cells served as positive controls for mCD40L-mediated apoptosis. Bars show mean RLU of 5–6 technical replicates ± SEM. c Raw data from the experiments shown in (a) and (b) are presented as Cell death fold increase in RLU relative to control, in order to compare the degree of apoptosis induction for mCD40L vs. G28-5 mAb treatment in all RCC lines (and control EJ cells). Bars show mean fold change ± SEM. d RCC cell lines were treated with mCD40L or G28-5 mAb alongside appropriate controls in the presence (+) or absence (–) of IFN-γ (10 units/ml) and cell death was detected using the CytoTox-Glo assay. Results are presented as Cell death fold increase in background-corrected RLU readings relative to control and are representative of two independent experiments. Bars show mean fold change of 5–6 technical replicates ± SEM. e RCC lines were treated with mCD40L for 24, 48 and 72 h and effector caspase-3/7 activity was assessed (see Methods). Results are presented as Caspase activity fold increase in background-corrected fluorescence relative to control and are representative of three experiments. Bars show mean fold caspase activity of 5–6 technical replicates ± SEM. f RCC lines were treated with mCD40L for 48 h alongside appropriate negative controls (and positive control, staurosporine-treated cells) and DNA fragmentation was assessed (see Methods). Results are presented as % DNA fragmentation relative to positive control (maximal DNA fragmentation) and are representative of at least two independent experiments. Bars show mean % DNA fragmentation of six technical replicates ± SEM. a ACHN, 786-O, and A-704 cells were either untreated (Control) or treated with 10 µg/ml cross-linked agonistic anti-CD40 mAb (G28-5) for 48 h and death was detected using the CytoTox-Glo assay (see Methods). Results are representative of at least two experiments and are presented as background-corrected relative luminescence unit (RLU) readings. EJ cells were included as positive controls. Bars show mean RLU of 3–4 technical replicates ± SEM. b ACHN, 786-O, and A-704 cells were co-cultured with 3T3Neo (Control) or 3T3CD40L (mCD40L) effector cells for 48 h and death was detected using the CytoTox-Glo assay. Results are typical of tive independent experiments and are presented as background-corrected relative luminescence units (RLU). EJ cells served as positive controls for mCD40L-mediated apoptosis. Bars show mean RLU of 5–6 technical replicates ± SEM. c Raw data from the experiments shown in (a) and (b) are presented as Cell death fold increase in RLU relative to control, in order to compare the degree of apoptosis induction for mCD40L vs. G28-5 mAb treatment in all RCC lines (and control EJ cells). Bars show mean fold change ± SEM. d RCC cell lines were treated with mCD40L or G28-5 mAb alongside appropriate controls in the presence (+) or absence (–) of IFN-γ (10 units/ml) and cell death was detected using the CytoTox-Glo assay. Results are presented as Cell death fold increase in background-corrected RLU readings relative to control and are representative of two independent experiments. Bars show mean fold change of 5–6 technical replicates ± SEM. e RCC lines were treated with mCD40L for 24, 48 and 72 h and effector caspase-3/7 activity was assessed (see Methods). Results are presented as Caspase activity fold increase in background-corrected fluorescence relative to control and are representative of three experiments. Bars show mean fold caspase activity of 5–6 technical replicates ± SEM. f RCC lines were treated with mCD40L for 48 h alongside appropriate negative controls (and positive control, staurosporine-treated cells) and DNA fragmentation was assessed (see Methods). Results are presented as % DNA fragmentation relative to positive control (maximal DNA fragmentation) and are representative of at least two independent experiments. Bars show mean % DNA fragmentation of six technical replicates ± SEM. To investigate if signal ‘quality’ (extent of receptor cross-linking) is critical in inducing RCC cell death, we treated the panel of RCC lines with membrane-CD40L (mCD40L). We utilised a culture system which involves co-culture of target (epithelial) cells with effectors (fibroblasts) expressing mCD40L, alongside non-ligand expressing fibroblasts (Control). We performed pre-titration experiments to (a) determine optimal target:effector ratios (0.6:1, 0.8:1 and 1:1) and (b) assess cell death at different time-points (24, 48, 72 h) post-receptor ligation (Supplementary Fig. 1). mCD40L caused marked cell death by 48 h (0.8:1 ratio) in all RCC lines, comparable to EJ (Fig. 2b), with similar observations at 72 h (Supplementary Fig. 1). Death induction by mCD40L was extensive and several-fold higher than the effect of G28-5 mAb which was minimal (Fig. 2c). In light of its ability to up-regulate CD40 (Fig. 1c), we examined whether IFN-γ could augment soluble agonist (or mCD40L)-mediated effects; IFN-γ could not sensitise RCC cells to G28-5 mAb, although it caused some enhancement in mCD40L-mediated death (Fig. 2d). With regard to the ‘nature’ of mCD40L-killing, in addition to its ability to cause plasma membrane integrity compromisation (Fig. 2a–d), mCD40L caused an increase in effector caspase-3/-7 activity, with activation occuring within 24 h post-ligation (Fig. 2e). In addition, mCD40L triggered extensive DNA fragmentation (Fig. 2f) in RCC lines by 48 h (fragmentation evident at 24 h—also Fig. 3b). These results show that CD40 induces extensive killing in RCC cells, which is dependent on the mode of receptor ligation but irrespective of the level of CD40-positivity, and mCD40L engages a death pathway with classical apoptotic features.Fig. 3mCD40L is a highly pro-apoptotic signal in malignant (RCC) but not in normal HRPT cells.a HRPT cells were co-cultured with 3T3Neo (Control) or 3T3CD40L (mCD40L) effector cells to assess the effect of mCD40L-mediated CD40 ligation. The co-cultures involved a defined number of effector cells (10,000/well) and three different densities of HRPT cells (6000, 8000 and 10,000/well). Cell death was detected at 24, 48 and 72 h using the CytoTox-Glo assay (see Methods). Results are presented as Cell death fold increase in background-corrected RLU readings (mCD40L relative to control) and are representative of three independent experiments. 786-O RCC cells served as positive controls for mCD40L-mediated apoptosis. Bars show mean fold change of 5–6 technical replicates ± SEM. b HRPT cells (seeded at 8000 cells/well) were either treated with 10 µg/ml cross-linked agonistic anti-CD40 mAb (G28-5) or mCD40L (by co-culture as above) for 24, 48 and 72 h alongside appropriate negative controls (and positive control, staurosporine-treated cells) and DNA fragmentation was assessed (see Methods). Results are presented as % DNA fragmentation relative to positive control (maximal DNA fragmentation) and are representative of two experiments. 786-O RCC cells served as positive control for mCD40L-mediated DNA fragmentation. Bars show mean % DNA fragmentation of three technical replicates ± SEM. a HRPT cells were co-cultured with 3T3Neo (Control) or 3T3CD40L (mCD40L) effector cells to assess the effect of mCD40L-mediated CD40 ligation. The co-cultures involved a defined number of effector cells (10,000/well) and three different densities of HRPT cells (6000, 8000 and 10,000/well). Cell death was detected at 24, 48 and 72 h using the CytoTox-Glo assay (see Methods). Results are presented as Cell death fold increase in background-corrected RLU readings (mCD40L relative to control) and are representative of three independent experiments. 786-O RCC cells served as positive controls for mCD40L-mediated apoptosis. Bars show mean fold change of 5–6 technical replicates ± SEM. b HRPT cells (seeded at 8000 cells/well) were either treated with 10 µg/ml cross-linked agonistic anti-CD40 mAb (G28-5) or mCD40L (by co-culture as above) for 24, 48 and 72 h alongside appropriate negative controls (and positive control, staurosporine-treated cells) and DNA fragmentation was assessed (see Methods). Results are presented as % DNA fragmentation relative to positive control (maximal DNA fragmentation) and are representative of two experiments. 786-O RCC cells served as positive control for mCD40L-mediated DNA fragmentation. Bars show mean % DNA fragmentation of three technical replicates ± SEM. Having demonstrated extensive CD40-induced apoptosis in RCC cells, we tested whether CD40 may cause cytotoxicity to their normal counterparts. When we treated HRPT cells with mCD40L (as above), no apoptosis was detectable, although mCD40L induced apoptosis in 786-O cells (Fig. 3a). Concordantly, neither soluble agonist G28-5 mAb nor (and more importantly) mCD40L induced DNA fragmentation in HRPT cells, in comparison to mCD40L-treated 786-O cells (Fig. 3b). Therefore, although mCD40L induces extensive apoptosis in malignant (RCC) cells, normal cells remain totally refractory to mCD40L. Using a membrane-based array approach we determined global cytokine/chemokine secretion induced by mCD40L in 786-O RCC cells compared to control EJ (Supplementary Fig. 2). The cytokines/chemokines mostly induced in both 786-O and EJ cells were interleukin-8 (IL-8), IL-8-related chemokine GRO-α, IL-6, MCP-1/CCL2 and granulocyte-macrophage colony-stimulating factor (GM-CSF). We then quantified secretion of IL-8, IL-6 and GM-CSF in RCC and in normal (HRPT) cells by enzyme-linked immunosorbent assay (ELISA). Despite basal levels of IL-8, mCD40L rapidly augmented IL-8 secretion in all RCC lines (3–6 h) (Fig. 4a). IL-8 induction was striking in HRPT cells, where mCD40L continued to induce secretion after 12 h (Fig. 4a). mCD40L caused gradual induction of IL-6 in all RCC lines, although this was less-pronounced in normal cells (Fig. 4b). Interestingly, although both RCC and normal cells secreted low basal levels of the cytokine, mCD40L rapidly and markedly induced GM-CSF secretion in RCC cells and, albeit to a lesser extent, in HRPT cells (Fig. 4c).Fig. 4Induction of pro-inflammatory cytokine secretion in human RCC lines vs. normal HRPT cells is dependent on the mode of CD40 ligation.a–c RCC cell lines ACHN, 786-O and A-704, the positive control EJ and normal HRPT cells were treated with mCD40L. IL-8 (a), IL-6 (b) and GM-CSF (c) secretion in comparison to controls was detected at the indicated time-points post CD40 ligation (3, 6, 12 and 24 h) by ELISA assays (see Methods). Bars represent mean cytokine concentration (pg/ml) ± SEM for three technical replicates and are representative of two independent experiments. d, e RCC cell line 786-O (d) and normal HRPT cells (e) were treated with ‘mCD40L’ alongside negative controls (‘Control’), and CD40 was also activated with agonistic anti-CD40 mAb (‘G28-5′), and secretion of GM-CSF was measured by ELISA as above. Untreated RCC cells (‘786-O’ in d) and normal cells (‘HRPT’ in e), respectively, were also included as background controls. Bars show mean GM-CSF concentration (pg/ml) ± SEM for two technical replicates and are typical of two independent experiments. a–c RCC cell lines ACHN, 786-O and A-704, the positive control EJ and normal HRPT cells were treated with mCD40L. IL-8 (a), IL-6 (b) and GM-CSF (c) secretion in comparison to controls was detected at the indicated time-points post CD40 ligation (3, 6, 12 and 24 h) by ELISA assays (see Methods). Bars represent mean cytokine concentration (pg/ml) ± SEM for three technical replicates and are representative of two independent experiments. d, e RCC cell line 786-O (d) and normal HRPT cells (e) were treated with ‘mCD40L’ alongside negative controls (‘Control’), and CD40 was also activated with agonistic anti-CD40 mAb (‘G28-5′), and secretion of GM-CSF was measured by ELISA as above. Untreated RCC cells (‘786-O’ in d) and normal cells (‘HRPT’ in e), respectively, were also included as background controls. Bars show mean GM-CSF concentration (pg/ml) ± SEM for two technical replicates and are typical of two independent experiments. Although IL-8 secretion is induced by soluble CD40 agonist, GM-CSF secretion may only be triggered by mCD40L. We therefore compared GM-CSF secretion in malignant (786-O) and normal (HRPT) cells following CD40 ligation by G28-5 mAb and mCD40L. mCD40L caused rapid GM-CSF induction in 786-O cells that peaked at 3 h post-ligation, whereas soluble agonist caused little induction of GM-CSF even after 12 h (Fig. 4d). Concordantly, when we compared soluble vs. mCD40L in normal HRPT cells, we found that mCD40L caused more marked GM-CSF secretion than did soluble agonist (Fig. 4e). Therefore, mCD40L induces rapid and sustained pro-inflammatory cytokine secretion, compared to soluble agonist, both in RCC and normal cells, despite its differential effects in cell fate. To investigate mCD40L-triggered signalling in renal cells, we initially focused on receptor-proximal events and assessed TRAF protein regulation. mCD40L caused rapid upregulation of TRAF1 in all RCC lines, in comparison to low basal expression. Induction of TRAF1 was also evident in HRPT cells, although this response was slower (observed after 6 h) and less-pronounced when compared to RCC cells (Fig. 5a).Fig. 5Differential regulation of TRAF proteins by mCD40L in normal (HRPT) and malignant (RCC) cells.RCC lines ACHN, 786-O and A-704, and normal HRPT cells were treated with mCD40L for the indicated time periods (1.5, 3 and 6 h) and expression of TRAF1 (a), TRAF2 (b) and TRAF3 (c) was detected by immunoblotting (40 µg protein/lane) in controls (‘C’) vs. mCD40L-treated cells (‘mL’). Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As positive controls for TRAF1 (a), TRAF2 (b) and TRAF3 (c) protein expression induction, in RCC cell immunoblotting experiments (top panels) lysates from HCT116 cells that were treated with control (‘C’) or treated with mCD40L (‘mL’) for 6 h were included, while for HRPT cell experiments (bottom panels) lysates from ACHN cells untreated or treated with mCD40L for 6 h were used. Lysate from effector (3T3CD40L) cells alone (20 µg protein/lane) served as negative control (NC) and confirmed the human-protein specificity of the anti-TRAF1 (a), anti-TRAF2 (b) and anti-TRAF3 (c) antibodies. RCC lines ACHN, 786-O and A-704, and normal HRPT cells were treated with mCD40L for the indicated time periods (1.5, 3 and 6 h) and expression of TRAF1 (a), TRAF2 (b) and TRAF3 (c) was detected by immunoblotting (40 µg protein/lane) in controls (‘C’) vs. mCD40L-treated cells (‘mL’). Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As positive controls for TRAF1 (a), TRAF2 (b) and TRAF3 (c) protein expression induction, in RCC cell immunoblotting experiments (top panels) lysates from HCT116 cells that were treated with control (‘C’) or treated with mCD40L (‘mL’) for 6 h were included, while for HRPT cell experiments (bottom panels) lysates from ACHN cells untreated or treated with mCD40L for 6 h were used. Lysate from effector (3T3CD40L) cells alone (20 µg protein/lane) served as negative control (NC) and confirmed the human-protein specificity of the anti-TRAF1 (a), anti-TRAF2 (b) and anti-TRAF3 (c) antibodies. We then investigated expression of the most ‘central’ TRAFs in CD40 signalling, TRAF2 and TRAF3. We observed rapid and marked TRAF2 induction in all RCC lines; yet, interestingly, in HRPT cells, which expressed basal TRAF2 protein levels, mCD40L caused TRAF2 downregulation (Fig. 5b). Equally strikingly-differential was the effect of mCD40L on TRAF3 expression, as in malignant cells mCD40L caused rapid, marked and sustained TRAF3 upregulation, whereas no induction of TRAF3 expression was detectable in normal (HRPT) cells (Fig. 5c). Therefore, mCD40L regulates early signalling events in renal cells and, more importantly, TRAF2 and TRAF3 regulation is fundamentally different between normal and malignant cells. Pharmacological blockade of MEK/ERK and NF-κB had no effect on mCD40L-mediated death. By contrast, AP-1 inhibition attenuated cell death, while JNK blockade fully abrogated mCD40L-mediated killing (Fig. 6). As these findings functionally implicated the JNK/AP-1 pathway in apoptosis, we examined the expression of key components of this signalling axis. mCD40L induced phosphorylation of both MKK4 and MKK7, as well as triggering sustained phosphorylation of downstream target JNK (Fig. 6a).Fig. 6Regulation of intracellular signalling pathways and their functional role in mCD40L-mediated tumour cell apoptosis.a ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (1.5, 3 and 6 h) and expression of phosphorylated-MAPKs MKK4 (p-MKK4), MKK7 (p-MKK7), JNK (p-JNK) and p38 (p-p38) was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As positive controls for p-MKK4/7, p-JNK and p-p38 protein expression induction, lysates from HCT116 cells that were treated with control (‘C’) or treated with mCD40L (‘mL’) for 6 h were included. Lysate from effector (3T3CD40L) cell monocultures served as negative control (NC) and confirmed the human-protein specificity of the antibodies. b–f ACHN, 786-O and A-704 cells were treated with mCD40L in the absence (vehicle control—denoted ‘Control’) or presence of the indicated concentration (12.5, 25 and 50 μM) of JNK inhibitor SP600125 (b), p38 inhibitor SB202190 (c), AP-1 inhibitor NDGA (d), MEK/ERK inhibitor U0126 (e) and NF-κB inhibitor (PDTC). Cell death was detected 48 h later using the CytoTox-Glo assay (see Methods). Results are presented as Cell death fold increase in background-corrected RLU readings relative to control (mCD40L treatment vs. controls) and are representative of three independent experiments. Bars show mean fold change of 5–6 technical replicates ± SEM. g ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (3 and 6 h) in the presence of 25 μM JNK inhibitor SP600125 or p38 inhibitor SB202190 and expression of phosphorylated-MAPKs JNK (p-JNK) and p38 (p-p38) was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). ACHN, 786-O and A-704 cells treated with mCD40L for 6 h in the absence of inhibitor (vehicle controls) were also included (denoted as positive control, ‘PC') for each experiment. Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). a ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (1.5, 3 and 6 h) and expression of phosphorylated-MAPKs MKK4 (p-MKK4), MKK7 (p-MKK7), JNK (p-JNK) and p38 (p-p38) was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As positive controls for p-MKK4/7, p-JNK and p-p38 protein expression induction, lysates from HCT116 cells that were treated with control (‘C’) or treated with mCD40L (‘mL’) for 6 h were included. Lysate from effector (3T3CD40L) cell monocultures served as negative control (NC) and confirmed the human-protein specificity of the antibodies. b–f ACHN, 786-O and A-704 cells were treated with mCD40L in the absence (vehicle control—denoted ‘Control’) or presence of the indicated concentration (12.5, 25 and 50 μM) of JNK inhibitor SP600125 (b), p38 inhibitor SB202190 (c), AP-1 inhibitor NDGA (d), MEK/ERK inhibitor U0126 (e) and NF-κB inhibitor (PDTC). Cell death was detected 48 h later using the CytoTox-Glo assay (see Methods). Results are presented as Cell death fold increase in background-corrected RLU readings relative to control (mCD40L treatment vs. controls) and are representative of three independent experiments. Bars show mean fold change of 5–6 technical replicates ± SEM. g ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (3 and 6 h) in the presence of 25 μM JNK inhibitor SP600125 or p38 inhibitor SB202190 and expression of phosphorylated-MAPKs JNK (p-JNK) and p38 (p-p38) was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). ACHN, 786-O and A-704 cells treated with mCD40L for 6 h in the absence of inhibitor (vehicle controls) were also included (denoted as positive control, ‘PC') for each experiment. Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). p38 has a well-documented role in stress-related responses in the kidney and has been implicated in renal cell apoptosis triggered by TNF-α, yet no involvement in CD40-mediated apoptosis has been reported. p38 inhibitor SB202190 attenuated mCD40L-apoptosis (Fig. 6c), as effectively as did the JNK inhibitor SP600125 (Fig. 6b). Concordantly, mCD40L triggered p38 phosphorylation and this response was sustained (Fig. 6a). Interestingly, the induction of p-p38 lagged behind the activation of p-JNK, as JNK phosphorylation plateaued by 1.5 h whereas maximal p-p38 expression was evident at later time-points (3 h in A-704 or 6 h in ACHN cells) (Fig. 6a). To discover whether JNK and p38 operated independently (but the timing of activation differed) or whether JNK caused downstream p38 activation, we treated RCC cells with mCD40L in the presence of JNK or p38 inhibitor and p-JNK and p-p38 expression was assessed. p38 inhibitor SB202190 attenuated p-p38 induction (Fig. 6g) but had no effect on the induction of p-JNK expression. However, the JNK inhibitor SB600125 completely-suppressed both JNK and p38 phosphorylation (Fig. 6g). Therefore, mCD40L-mediated apoptosis involves both the JNK/AP-1 and p38 pathways, but p38 activation appears to be JNK-mediated. Having demonstrated mCD40L-mediated effector caspase (-3/7) activation (Fig. 2e), biochemical inhibition of initiator caspase-8 and -10 showed little effect on mCD40L-killing, even though the caspase-8 biochemical inhibitor effectively blocked TRAIL-mediated cell death (not shown). By contrast, caspase-9 blockade markedly reduced death and pan-caspase inhibitor nearly completely blocked apoptosis (Fig. 7a). This suggested that mCD40L-mediated apoptosis is caspase-dependent and apoptosis may entrain the ‘intrinsic’/mitochondrial pathway. We used whole-cell lysates and assessed whether mCD40L induced Bak and Bax, the key regulators of mitochondrial outer membrane permeabilisation (MOMP), cytochrome c release and caspase-9 activation. We could detect basal Bak and Bax expression in all RCC lines but mCD40L triggered marked induction of Bak and particularly Bax expression 6 h post-ligation (Fig. 7b) (no induction observed <3 h—not shown). Bax levels plateaued more rapidly, whereas Bak induction was gradual until expression peaked 24 h post-treatment. Interestingly, blockade of the JNK/AP-1 and p38 pathways fully abrogated induction of both Bax and Bak (Fig. 7c). Therefore, mCD40L-mediated death in RCC cells is caspase-dependent and involves JNK/p38-mediated induction of the mitochondrial apoptotic pathway.Fig. 7Role of caspase activation and induction of the mitochondrial (intrinsic) pathway during mCD40L-mediated tumour cell apoptosis.a ACHN, 786-O and A-704 cells were treated with mCD40L in the absence (vehicle control—denoted ‘Control’) or presence of 100 μM of inhibitor of caspase-10 (z-AEVD-FMK), caspase-8 (z-IETD-FMK), caspase-9 (z-LEHD-FMK) or pan-caspase inhibitor (z-VAD-FMK). Cell death was detected 48 h later using the CytoTox-Glo assay (see Methods). Results are presented as Cell death fold increase in background-corrected RLU readings relative to control (mCD40L treatment vs. controls) and are representative of three independent experiments. Bars show mean fold change of 4–6 technical replicates ± SEM. b ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (6, 12 and 24 h) and expression of Bak and Bax was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As positive controls for Bak and Bax protein expression induction, lysates from HCT116 cells that were treated with control (‘C’) or treated with mCD40L (‘mL’) for 24 h were included. Lysate from effector (3T3CD40L) cells alone served as negative control (NC) and confirmed the human-protein specificity of the antibodies. c ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (12 and 24 h) in the presence of 25 μM JNK inhibitor SP600125 or p38 inhibitor SB202190 and expression of Bak and Bax was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). ACHN, 786-O and A-704 cells treated with mCD40L for 24 h in the absence of inhibitor (vehicle controls) were also included (denoted as positive control, ‘PC') for each experiment. Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). a ACHN, 786-O and A-704 cells were treated with mCD40L in the absence (vehicle control—denoted ‘Control’) or presence of 100 μM of inhibitor of caspase-10 (z-AEVD-FMK), caspase-8 (z-IETD-FMK), caspase-9 (z-LEHD-FMK) or pan-caspase inhibitor (z-VAD-FMK). Cell death was detected 48 h later using the CytoTox-Glo assay (see Methods). Results are presented as Cell death fold increase in background-corrected RLU readings relative to control (mCD40L treatment vs. controls) and are representative of three independent experiments. Bars show mean fold change of 4–6 technical replicates ± SEM. b ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (6, 12 and 24 h) and expression of Bak and Bax was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As positive controls for Bak and Bax protein expression induction, lysates from HCT116 cells that were treated with control (‘C’) or treated with mCD40L (‘mL’) for 24 h were included. Lysate from effector (3T3CD40L) cells alone served as negative control (NC) and confirmed the human-protein specificity of the antibodies. c ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (12 and 24 h) in the presence of 25 μM JNK inhibitor SP600125 or p38 inhibitor SB202190 and expression of Bak and Bax was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). ACHN, 786-O and A-704 cells treated with mCD40L for 24 h in the absence of inhibitor (vehicle controls) were also included (denoted as positive control, ‘PC') for each experiment. Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As activation of JNK by TNFRs can be ROS-dependent, we detected ROS production in RCC cells. mCD40L caused rapid ROS release (30 min) and levels peaked at 1 h (Fig. 8a); thereafter, ROS levels remained high (Supplementary Fig. 3). By contrast, non-apoptotic G28-5 mAb induced modest changes in ROS (Fig. 8b). Induction of ROS was critical in apoptosis, as the ROS scavenger N-acetyl l-cysteine (NAC) markedly attenuated mCD40L-mediated death (Fig. 8c).Fig. 8ROS generation by mCD40L, but not soluble CD40 agonist, and activation of ROS-dependent signalling for apoptosis induction.a ACHN, 786-O and A-704 cells were co-cultured with 3T3Neo or 3T3CD40L effector cells for 1 h and intracellular ROS was detected by treatment with 1 μM H2DCFDA (see Methods). Background-corrected relative fluorescence unit (RFU) readings obtained were used to present the data as ROS induction fold increase, which is fold change in RFU detected for H2DCFDA-treated 3T3Neo/RCC cell (‘Control’) and 3T3CD40L/RCC cell (‘mCD40L’) co-cultures vs. untreated co-cultures. Results are representative of three independent experiments. Bars show mean fold change of 5–6 technical replicates ± SEM. b To compare the differences in the extent of ROS generation for mCD40L vs. soluble agonist, ACHN cells were treated with ‘mCD40L’ (alongside negative controls) or agonistic anti-CD40 mAb (‘G28-5′) for the indicated time periods (30 min, 1 h, 2 h and 3 h) and ROS levels were detected as in a. For mCD40L treatments, data are presented as background-corrected RFU readings of mCD40L relative to controls; for G28-5 treatments, data are presented as RFU readings relative to untreated cells. Results (ROS induction fold increase) are representative of two independent experiments. Bars show mean fold change of 4–5 technical replicates ± SEM. c, d ACHN, 786-O and A-704 cells were treated with mCD40L in the absence (vehicle control—denoted ‘Control’) or presence of the indicated concentration of inhibitors NAC (c) and DPI (d). Cell death was detected 48 h later using the CytoTox-Glo assay (see Methods). Results are presented as Cell death fold increase in background-corrected RLU readings relative to control (mCD40L treatment vs. controls) and are representative of three independent experiments. Bars show mean fold change of 4–5 technical replicates ± SEM. e ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (1.5, 3 and 6 h) and expression of phospho-p40phox, phospho-ASK1 and Trx-1 was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As positive controls, lysates from HCT116 cells that were treated with control (‘C’) or treated with mCD40L (‘mL’) for 6 h were included. Lysate from effector (3T3CD40L) cell monocultures served as negative control (NC) and confirmed the human-protein specificity of the antibodies. a ACHN, 786-O and A-704 cells were co-cultured with 3T3Neo or 3T3CD40L effector cells for 1 h and intracellular ROS was detected by treatment with 1 μM H2DCFDA (see Methods). Background-corrected relative fluorescence unit (RFU) readings obtained were used to present the data as ROS induction fold increase, which is fold change in RFU detected for H2DCFDA-treated 3T3Neo/RCC cell (‘Control’) and 3T3CD40L/RCC cell (‘mCD40L’) co-cultures vs. untreated co-cultures. Results are representative of three independent experiments. Bars show mean fold change of 5–6 technical replicates ± SEM. b To compare the differences in the extent of ROS generation for mCD40L vs. soluble agonist, ACHN cells were treated with ‘mCD40L’ (alongside negative controls) or agonistic anti-CD40 mAb (‘G28-5′) for the indicated time periods (30 min, 1 h, 2 h and 3 h) and ROS levels were detected as in a. For mCD40L treatments, data are presented as background-corrected RFU readings of mCD40L relative to controls; for G28-5 treatments, data are presented as RFU readings relative to untreated cells. Results (ROS induction fold increase) are representative of two independent experiments. Bars show mean fold change of 4–5 technical replicates ± SEM. c, d ACHN, 786-O and A-704 cells were treated with mCD40L in the absence (vehicle control—denoted ‘Control’) or presence of the indicated concentration of inhibitors NAC (c) and DPI (d). Cell death was detected 48 h later using the CytoTox-Glo assay (see Methods). Results are presented as Cell death fold increase in background-corrected RLU readings relative to control (mCD40L treatment vs. controls) and are representative of three independent experiments. Bars show mean fold change of 4–5 technical replicates ± SEM. e ACHN, 786-O and A-704 cells were treated with mCD40L for the indicated time periods (1.5, 3 and 6 h) and expression of phospho-p40phox, phospho-ASK1 and Trx-1 was detected in controls (‘C’) vs. mCD40L-treated cells (‘mL’) by immunoblotting (40 µg protein/lane). Equal loading for human epithelial cell lysate was confirmed by CK18 detection (see Methods). As positive controls, lysates from HCT116 cells that were treated with control (‘C’) or treated with mCD40L (‘mL’) for 6 h were included. Lysate from effector (3T3CD40L) cell monocultures served as negative control (NC) and confirmed the human-protein specificity of the antibodies. Apoptosis signal-regulating kinase 1 (ASK1) MAP3K acts upstream of p38/JNK and its activation is ROS-mediated. mCD40L triggered phosphorylation of ASK1 in all RCC lines (Fig. 8e, right panels), the timing of which was in line with ROS induction (Fig. 8b). Notably, ROS induction occurred far sooner than apoptosis induction (Fig. 2); apart from the mitochondria, the other source of ROS for cell signalling functions is the NADPH oxidase (Nox) complex. mCD40L triggered phosphorylation of the regulatory Nox-2 subunit p40phox (Fig. 8e) and Nox inhibition using diphenyleneiodonium (DPI) suppressed mCD40L-mediated death (Fig. 8d). Because ASK1 activation occurs following ‘release’ from its inhibitor Thioredoxin (Trx), facilitated by ROS-mediated Trx oxidation, we explored the effect of mCD40L on Trx-1. Notably, in untreated RCC cells we observed high basal Trx-1 expression, with expression being most pronounced in the most rapidly proliferating 786-O cells (Fig. 8e). Intriguingly, mCD40L rapidly attenuated Trx-1 protein expression either fully (ACHN and A-704 cells) or partially (786-O) (Fig. 8e). Therefore, mCD40L-mediated RCC death is ROS-dependent, pro-apoptotic signalling entrains activation of Nox as well as ASK1 activation, accompanied by Trx-1 downregulation. A unique feature of CD40 among TNFR members is its exquisitely contextual influence on cell fate; the consequences of CD40 ligation appear to be different in normal and malignant B-lymphocytes, yet recent evidence suggests that this may also apply to epithelial cells. In parallel, the ‘quality’ of the signal may determine whether CD40L–CD40 interactions are pro-apoptotic; mCD40L causes apoptosis, while soluble agonists are mainly growth-inhibitory. The molecular signalling pathways underpinning these two fundamental properties of CD40, i.e. the TRAF members involved, MAPK pathways invoked and death mechanisms employed, have only recently become the subject of detailed investigation. Our study provides the first systematic investigation of the effects of CD40 ligation in RCC cells as well as compared these effects to normal (HRPT) cells. Normal and malignant cells expressed CD40 which could be up-regulated by TNF-α/IFN-γ. We found that soluble CD40 agonist had little effect; however, mCD40L induced extensive killing, irrespective of the level of CD40-positivity. Cross-linked soluble agonist (G28-5 mAb) even in combination with IFN-γ could not trigger death in RCC cells. mCD40L-mediated death had apoptotic features, including plasma membrane compromisation, DNA fragmentation and caspase-3/-7 activation, and caspase activity was essential for apoptosis. Collectively, these findings highlighted the importance of the mode of CD40 ligation in determining functional outcome in carcinoma cells, and provided an explanation for lack of detectable pro-apoptotic effects in RCC cells in previous studies which employed soluble CD40 agonists. Due to its strong pro-apoptotic effect in malignant cells, it was essential to investigate the effect of mCD40L in normal (HRPT) cells. Previous studies reported that soluble CD40 agonists stimulated IL-8 and MCP-1 induction and MAPK signalling, but did not cause cell death in HRPT cells in vitro, despite ROS induction. We have now shown that despite being highly pro-apoptotic in RCC cells, mCD40L remains non-apoptotic in HRPT cells, thus normal cells are refractory to CD40-killing. This provides for the first time evidence for a tumour cell-specific pro-apoptotic effect in RCC, as well as adding support to previous observations that mCD40L is non-apoptotic in normal human uro-epithelial cells. CD40 signalling has been well-characterised in the context of B-cell function, where the role of the TRAFs in activating signalling pathways, including NF-кB, JNK/AP-1 and p38, has been elegantly studied by Bishop et al.. CD40-mediated TRAF modulation and subsequent MAPK signalling in normal and malignant epithelial cells is far less characterised. Pro-apoptotic CD40 ligation differentially regulated TRAF expression in renal cells; mCD40L induced TRAF1 expression; however, in normal cells TRAF1 induction was slower. Most dramatically different was the regulation of TRAF2/3; mCD40L caused their rapid and sustained induction in RCC cells, yet in HRPT cells we observed downregulation of TRAF2 and no TRAF3 induction. Our observations in normal cells are strikingly reminiscent of the regulation of TRAFs in normal B-lymphocytes, where TRAFs 2 and 3 are either downregulated or not activated and corroborate for the first time previous observations in normal uro-epithelial cells. As there is evidence that in UCC and CRC cells TRAF3 plays a critical role in mCD40L-mediated apoptosis by triggering JNK activation and apoptosis, it is tempting to speculate that TRAF3 might play an equally important role in RCC cell apoptosis. However, previous studies reported no clear patterns of TRAF2 regulation; intriguingly, in RCC cells mCD40L induced TRAF2 as effectively as TRAF3, and future studies should address the possible functional role/interplay of TRAFs 2/3 in CD40 signalling in RCC cells. MAPKKs such as MKK4/MKK7 can trigger death-inducing JNK activation and JNK/AP-1 induction underpins mCD40L-mediated death in UCC cells; however, p38 activation has never been directly implicated in CD40-mediated carcinoma cell death. We now show that mCD40L caused phosphorylation of MKK4/MKK7 followed by JNK and p38 phosphorylation, and activation of both JNK and p38 was essential in CD40-mediated apoptosis in RCC cells. Although it has been widely reported that JNK/p38 can be simultaneously activated but act independently for apoptosis induction, during mCD40L-mediated apoptosis in RCC cells, p38 activation was fully dependent on JNK activity, suggesting JNK as the direct p38 activator. To our knowledge, this is the first report for such a temporally defined interplay between these two key MAPKs during any known apoptotic programme. Bcl-2 family proteins Bak/Bax form oligomers to induce MOMP and subsequently apoptosis. Bak and particularly Bax were induced in RCC lines, induction was sustained during apoptosis and was JNK/p38-controlled. ROS are products of mitochondrial metabolism, but also influence cell proliferation/survival. Inability to control ROS by antioxidant pathways causes oxidative stress leading to cellular damage, therefore ROS generation represents ‘proliferation-at-a-cost’ for tumour cells, rendering them susceptible to a ‘lethal pro-apoptotic threshold’. Unlike soluble agonist, mCD40L triggered rapid ROS production in RCC lines which was sustained and was critical in cell death. Interestingly, ROS production occurred far sooner than apoptosis induction; another major ROS source is the Nox complex and studies in B cells have implicated the regulatory Nox-2 subunit p40phox in CD40 signalling. The demonstration that p40phox is activated by mCD40L and blockade of Nox abrogated apoptosis implies that the initial ‘wave’ of ROS may be the Nox-dependent. These findings suggest a role of Nox in CD40 signalling in RCC cells and support the recent observation of Nox-/ROS-dependent mCD40L-mediated apoptosis in UCC cells. ROS can trigger apoptosis via activation of JNK upstream of which is the ‘redox-sensor’ ASK1, normally inhibited by Trx-1. ROS elevation causes Trx-1 release, permitting ASK1 oligomerization and full-activation and MKK4/MKK7 phosphorylation to subsequently activate JNK and trigger apoptosis. mCD40L caused ASK1 phosphorylation in RCC cells linked with ROS induction, and in line with the MKK4/MKK7 phosphorylation observed, indicating ASK1 as the upstream inducer of JNK, which in turn mediates p38 activation. Thus, mCD40L triggers an ASK1–MKK4/7–JNK–p38 signalling pathway concomitantly with activation of the ROS-generating Nox complex and cell death is ROS-dependent. Interestingly, high basal intracellular Trx-1 expression was detected in RCC lines, supportive of the reported over-expression of the Trx antioxidant pathway in carcinomas. Notably, we detected substantial amounts of soluble Trx-1 protein in RCC cell culture supernatants (data not shown), supporting Trx-1 over-expression. Strikingly, mCD40L rapidly suppressed Trx-1 expression in RCC cells, which could facilitate ASK1 activation. As Trx-1 expression is under the control of the master-regulator Nrf2 (ref. ), future studies should determine if mCD40L-mediated signalling can modulate Nrf2 to control Trx-1 levels. Inflammatory responses triggered by cancer-cell death linked with induction of ‘immunogenic cell death' to prime/enhance T-cell responses against cancer-cell-derived tumour-associated antigens are regarded essential for cancer therapy. In CRC and UCC cells, soluble agonist and mCD40L caused IL-6 and IL-8 secretion; however, only mCD40L induced GM-CSF release. Soluble CD40 agonists can mediate cytokine secretion in RCC cells, particularly MCP-1 (refs. ), but mCD40L triggered more marked secretion of several cytokines, while GM-CSF secretion was specifically triggered by mCD40L in normal and RCC cells. Thus, in addition to defining the pro-apoptotic capacity of CD40 ligation, signal ‘quality’ also determines the repertoire/extent of pro-inflammatory cytokine secretion in renal cells. GM-CSF is a pleiotropic cytokine that enhances recruitment of neutrophils and macrophages towards tumours and assists in tumour cell lysis via macrophages and dendritic cells (DC); therefore, GM-CSF release concomitantly with mCD40L-mediated apoptosis may be of therapeutic value. A number of important studies by Wiltrout and colleagues provided evidence for the therapeutic potential of CD40 in RCC. They demonstrated that CD40 expression in RCC is linked with prolonged patient survival. Moreover, CD40 activation resulted in recruitment of monocytes and T cells into established RCC tumours in vivo where agonistic mAb increased the presence of DC and caused reduction in tumour size; these effects were immune-mediated, assisted by tumour-cell cytokine secretion and independent of tumour CD40-status. Importantly, however, it was suggested that RCC tumours could be targeted more effectively by combining CD40-mediated immune activation together with delivery of the CD40 signal to the tumour itself. In light of the ability of mCD40L, but not soluble CD40 agonist, to provide a potent tumour cell-specific killing signal, our work has not only offered insights into the underpinning biology of CD40’s effects in normal and malignant epithelial cells, but has also provided a novel avenue for an improved, ‘double-hit’ approach for inflammatory, tumour cell-specific CD40-based approach for cancer therapy. ACHN, 786-O and A-704 lines were from the ATCC, supplied via Sigma (Sigma, Dorset, UK) or LGC Standards (LGC, Middlesex, UK) and were adapted for culture in DR medium (1:1 v/v mixture of DMEM and RPMI), supplemented with 5% fetal bovine serum (FBS) (Sigma). EJ and HCT116 lines were cultured as previously. 3T3Neo and 3T3CD40L fibroblasts were maintained in DR/10% FBS and DR/10% FBS supplemented with 0.5 mg/ml G418 (Invivogen, supplied by Source BioScience LifeSciences), respectively, with omission of the antibiotic during co-culture experiments (below). Human renal proximal tubule epithelial cells (HRPTEpiC) (Caltag Medsystems, Bucks, UK), referred to as HRPT, were maintained in EpiCM supplemented with FBS, epithelial cell growth supplement and penicillin/streptomycin as recommended by the supplier. For activation of CD40 by soluble agonist, epithelial cells were treated with G28-5 mAb cross-linked with goat anti-mouse IgG antibody (Sigma). Unless otherwise stated, cells were seeded at 0.8 × 10 cells/well in 96-well plates or 5 × 10 cells/well in 24-well plates, and following overnight incubation they were treated with G28-5 mAb at 10 µg/ml cross-linked with 2.5 µg/ml goat anti-mouse Ig for 48 h. For CD40 activation by mCD40L, 3T3neo (control) and mCD40L-expressing 3T3CD40L (mCD40L) fibroblasts (effector cells) were growth-arrested by treatment with 10 µg/ml of Mitomycin C (MMC; Santa Cruz, supplied by Insight Biotechnology, Middlesex, UK) for 2 h in DR/5% medium and seeded either in 96-well plates at 1 × 10 cells/well for apoptosis detection assays, 10 cm culture dishes at 3 × 10 cells/dish for protein lysate preparation, or 24-well plates at 6 × 10 cells/well for culture-supernatant collection. Following overnight incubation, culture medium was removed and epithelial cells were added at 0.8 × 10 cells/well in 96-well plates, 3 × 10 in 10 cm dishes, or 5 × 10 cells/well in 24-well plates, respectively. In accordance with published guidelines regarding use and interpretation of assays for monitoring cell death, a minimum of two assays were routinely utilised for detection of cell death (apoptosis). This involved use of (a) CytoTox-Glo assay (Promega, Southampton, UK) for detection of compromisation of cell membrane integrity, (b) SensoLyte Homogenous AFC Capase-3/7 assay (Anaspec, Cambridge Bioscience, Cambs, UK) for effector caspase-3/7 activation, and (c) DNA Fragmentation ELISA (Roche Diagnostics, West Sussex, UK) for detection of fragmented DNA in culture supernatants. Full experimental details on the use of these assays for the co-culture system of effector/3T3 and target/epithelial cells have been described recently elsewhere. Whole-cell lysates were prepared from epithelial cells cultured alone or from co-cultures with 3T3Neo (controls) and 3T3CD40L (mCD40L-treated) cells, and were denoted as ‘C’ and ‘mL’, respectively, in the figures. Lysate from effector (3T3CD40L) cells alone served as a negative control (denoted as NC). Lysates were separated by 4–12% SDS-PAGE, and electroblotting performed using Immobilon-FL PVDF membrane (Thermo Fisher Scientific, Loughborough, UK) as detailed elsewhere. For epithelial cell-alone lysates, antibodies used were CD40 (sc-13128) (Insight Biotechnology) and β-actin (A5441) (Sigma). For co-culture lysates, antibodies used were TRAF1 (sc-7186), TRAF2 (sc-876) and TRAF3 (sc-949) (Insight Biotechnology); phospho-MKK4 (#4514), phospho-MKK7 (#4171), phospho-ASK1 (#3765), phospho-JNK (#9255), phospho-p38 (#9255), Thioredoxin-1 (#2285S) and phospho-p40phox (#4311) (Cell Signalling Technologies, supplied by New England Biolabs, Herts, UK); Bak (AF816) and Bax (2282-MC-100) (R&D Systems). Antibody binding was detected using goat anti-rabbit IRDye® 800 (Tebu-bio, Cambs, UK) or goat anti-mouse Ig Alexa Fluor 680 (Thermo Fisher Scientific). For lysates prepared from 3T3neo/3T3CD40L (effector) and epithelial (target) cell co-cultures, sample loading was corrected/adjusted for epithelial cells according to reactivity with human-specific anti-cytokeratin 18 (CK18) antibody (C8541) (Sigma) and not using antibodies detecting non-phosphorylated intracellular signalling mediators, as detailed elsewhere. Immunolabelling was visualised using an Odyssey Infra-Red imaging system (LiCor, Cambs, UK). Inhibitors for AP-1 (NDGA), NF-κB (PDTC), NADPH oxidase (Nox) (DPI) and the antioxidant N-acetyl l-cysteine (NAC) were from Sigma. Inhibitors for JNK (SP600125), p38 (SB202190) and MEK/ERK (U0126) were from Enzo. Inhibitors for caspase-8 (z-IETD-FMK), -9 (z-LEHD-FMK), -10 (z-AEVD-FMK) and pan-caspase inhibitor (z-VAD-FMK) were from R&D Systems (supplied by Bio-Techne, Abingdon, UK). NAC was reconstituted in DR/5% medium and its pH adjusted to 7.0 using 1 M NaOH solution, and was filter-sterilised before use. All other inhibitors were reconstituted in DMSO (Sigma). RCC cells were co-cultured with 3T3Neo or 3T3CD40L in the presence of inhibitors for 48 h and apoptosis determined as above. Cells treated with DMSO alone (vehicle controls) were included. Epithelial cells were either (a) cultured in 24-well plates at 5 × 10 cells/well and treated with 10 µg/ml of cross-linked G28-5 mAb or (b) co-cultured at 5 × 10 cells/well with 6 × 10 cells/well growth-arrested 3T3CD40L and 3T3Neo in 24-well plates. Culture supernatants were collected at specified time-points post-receptor ligation, and secretion of IL-6, IL-8 and GM-CSF was measured by cytokine-specific ELISA or membrane array-based detection (R&D Systems/Bio-Techne). For ELISA assays measurements were made spectrophotometrically and for membrane arrays by fluorescence measurements on an Odyssey Infra-Red imaging system (LiCor), as recommended by the manufacturer. For detection of CD40 expression, cells were cultured in flasks until approximately 80% confluent. Alternatively, for treatment with cytokines, cells were seeded in 24-well plates at 5 × 10 cells/well and, after overnight incubation, they were treated with fresh medium containing 10 units/ml TNF-α or IFN-γ (R&D Systems/Bio-Techne) for 48 h. Cells were then harvested, washed and re-suspended in FACS buffer. CD40 expression was detected using PE-conjugated mouse anti-human CD40 antibody and an isotype control IgG1-PE (BD Biosciences, Berks, UK). In all, 10,000 events were acquired on a Guava EasyCyte flow cytometer and results analysed using EasyCyte software (Millipore, Watford, UK). For detection of intracellular ROS levels, cells were treated with 1 μM H2DCFDA (Thermo Fisher Scientific) for 30 min, fluorescence measurements (relative fluorescence units, RFU) were taken spectrophotometrically and results expressed as fold increase in RFU, as described in detail previously. It is noted that when co-cultures were performed, effector (3T3Neo/3T3CD40L) cells were not growth-arrested using MMC to minimise background 3T3 cell-associated fluorescence. Mean values and standard error of the mean (SEM) were used as descriptive statistics. Two-tailed, paired or non-paired Student’s t-tests were used for evaluation of statistical significance. For graphical purposes in the figure captions: *p < 0.05, **p < 0.01 and ***p < 0.001, while NS denotes non-significance (p > 0.05).
PMC12256823
Bi-allelic variants in POPDC2 cause an autosomal recessive syndrome presenting with cardiac conduction defects and hypertrophic cardiomyopathy
POPDC2 encodes the Popeye domain-containing protein 2, which has an important role in cardiac pacemaking and conduction, due in part to its cyclic AMP (cAMP)-dependent binding and regulation of TREK-1 potassium channels. Loss of Popdc2 in mice results in sinus pauses and bradycardia, and morpholino-mediated knockdown of popdc2 in zebrafish results in atrioventricular (AV) block. We identified bi-allelic variants in POPDC2 in four families with a phenotypic spectrum consisting of sinus node dysfunction, AV conduction defects, and hypertrophic cardiomyopathy. Using homology modeling, we show that the identified variants are predicted to diminish the ability of POPDC2 to bind cAMP. In in vitro electrophysiological studies, we demonstrated that, in contrast with wild-type POPDC2, variants found in affected individuals failed to increase TREK-1 current density. While muscle biopsy of an affected individual did not show clear myopathic disease, it showed significantly reduced abundance of both POPDC1 and POPDC2, suggesting that stability and/or membrane trafficking of the POPDC1-POPDC2 complex is impaired by pathogenic variants in either protein. Single-cell RNA sequencing from human hearts demonstrated that co-expression of POPDC1 and POPDC2 was most prevalent in AV node, AV node pacemaker, and AV bundle cells. Using population-level genetic data of more than 1 million individuals, we show that none of the familial variants were associated with clinical outcomes in heterozygous state, suggesting that heterozygous family members are unlikely to develop clinical manifestations and therefore might not necessitate clinical follow-up. Our findings provide evidence for bi-allelic variants in POPDC2 causing a Mendelian autosomal recessive cardiac syndrome.The rhythmic contraction of the heart is orchestrated by the cardiac pacemaker and conduction system. Electrical activity in the heart arises in the sinus node, located in the right atrium near the entrance of the superior vena cava. The electrical impulse then spreads through the atria to the atrioventricular (AV) node and is subsequently propagated through the bundle of His and bundle branches to the Purkinje fibers from where it spreads throughout the ventricles. Cardiac conduction defects (CCDs; MIM: 115080) are primarily the consequence of age-related degeneration, structural heart disease, or post-operative complications. Presentation of CCDs in the young should raise suspicion of a genetic disorder. Rare variants in genes encoding cardiac ion channels (e.g., SCN5A, MIM: 600163; TRPM4, MIM: 606936; and HCN4, MIM: 605206), transcription factors (e.g., TBX5, MIM:601620; NKX2-5, MIM: 600584), constituents of the inner nuclear membrane (e.g., LMNA, MIM: 150330; EMD, MIM: 300384), gap junction proteins (e.g., GJC1, MIM: 608655), and others (e.g., GNB5, MIM: 604447; TNNI3K, MIM: 613932; PRKAG2, MIM: 602743) have been implicated in inherited CCD presenting in isolation or in presence of other cardiac or extracardiac features. However, many affected individuals with early-onset CCD remain genetically unexplained. Bi-allelic variants in POPDC1 (also known as BVES, MIM: 604577), encoding the Popeye domain-containing protein 1, are associated with muscular dystrophy and AV block (MIM: 604577). In mice, knockout of Popdc1 or Popdc2 resulted in stress-induced sinus pauses and sinus bradycardia. In zebrafish, morpholino knockdown of popdc1 or popdc2 resulted in second-degree AV block and bradycardia. The TWIK-related potassium channel 1 (TREK-1, encoded by KCNK2; MIM: 603219) is an established interacting protein of POPDC2 (MIM: 605823) and co-expression of POPDC2 and TREK-1 has been shown to result in a 2-fold higher TREK-1 current in comparison to expression of TREK-1 alone. Here, we provide evidence for bi-allelic loss-of-function (LOF) variants in POPDC2 as the cause of an autosomal recessive syndrome in four families, consisting of a phenotypic spectrum including sinus node disease and AV conduction defects with hypertrophic cardiomyopathy (HCM; MIM: 192600). Family A was referred to the Department of Human Genetics of the Amsterdam UMC (Amsterdam, the Netherlands) for genetic testing and counseling for CCD and HCM. To follow up on the findings from exome sequencing in this family, we studied 78 individuals that were diagnosed with a similar clinical presentation to family A (i.e., CCDs and HCM, cohorts 1–3). In addition, we studied 96 HCM individuals without CCDs (cohort 4). In all 174 individuals, genetic testing had ruled out causative variants in established arrhythmia and/or cardiomyopathy genes. Families C and D were identified via a genetic and phenotypic match through DECIPHER and GeneMatcher, respectively. The study protocol was approved by the Amsterdam University Medical Center Research Ethics Committee and the local Institutional Review Boards of contributing centers. Signed informed consent was obtained from the affected individuals or their parents. Details on case recruitment and DNA-sequencing methods for each family can be found in the supplemental information and Table S1. To ensure the privacy of the affected individuals and their families, (1) ages are presented as non-overlapping age ranges (i.e., 0–5, 6–10, and 11–15 years), (2) pedigrees were modified, (3) information related to ancestry/country or origin/nationality are not reported, and (4) clinical descriptions were minimized. The censoring undertaken for privacy reasons does not affect the ability to evaluate the presented data. Sex was not considered as a biological variable. Two homology models were generated using either SWISS-MODEL or AlphaFold2 Multimer. The SWISS-MODEL web server used a cyclic AMP (cAMP)-regulatory protein from Yersinia pestis (6DT4) as a template to generate the homology model for the Popeye domain of POPDC2. A dimer of full-length POPDC2 was generated using AlphaFold Multimer, and the intrinsically disordered C-terminal residues 275–364 were deleted to simplify figure presentation. A final homology model of POPDC2 was created by replacing residues 128–213 of the AlphaFold Multimer model with residues 128–213 of the SWISS-cAMP model after superimposing the individual Popeye domains. AlphaMissense was used to generate the predicted pathogenicity of single-amino-acid substitutions and deleted regions. AlphaMissense uses language modeling to understand amino-acid distributions based on sequence context, then it incorporates structural information using an AlphaFold-derived system to consider a protein’s three-dimensional form when assessing a variants’ impact. It also utilizes weak labels from population frequency data to refine predictions without human biases. Using these models, we evaluated the consequence of POPDC2 variants found in families A–D and five variants that occurred homozygously in individuals from the Genome Aggregation Database v2.1.1 (gnomAD), which are not expected to cause disease (Table S2) Single-nucleus RNA-sequencing (snRNA-seq) data and Visium Spatial gene expression data were obtained from a previously published study. Processed data of single-cell RNA-sequencing (scRNA-seq)/snRNA-seq and Visium data are available for browsing gene expression and download from the Heart Cell Atlas (https://www.heartcellatlas.org). Annotated, log-normalized count matrices for both modalities were downloaded and specifically analyzed for expression of POPDC1, -2, and -3 using Scanpy package for Python run in Jupyter Notebook. Original histological annotation of tissue sections was used. The cell-state annotation was adapted from the original study. All atrial cardiomyocytes were pooled in one category, and all ventricular cardiomyocytes were pooled together. From the conduction system cells, sinoatrial node pacemaker (SAN P) cells and Purkinje cells are shown separately; due to low cell numbers, atrioventricular node pacemaker (AVN P) and bundle cells are shown together. Sinus node and AV node/His scRNA-seq data from mice were obtained from a previously published study. Using these data, t-distributed stochastic neighbor embedding (t-SNE) maps with a perplexity of 50 were generated on the R2 environment Genomics Analysis and Visualization Platform (http://r2.amc.nl). The cells were subsequently clustered into different populations using the t-SNE DBSCAN tool. Sentinel gene expression was used to characterize the different clusters (e.g., sinus node cells, expressing higher levels of Tbx3, Isl1, and Hcn4). Thereafter, Popdc1-3 expression intensities were plotted on the t-SNE maps to identify their expression profiles across the present tissue clusters. hTREK-1a cloned in pIRES2-EGFP was obtained from Drs. Delphine Bichet and Florian Lesage (Université Nice Sophia Antipolis, France). Full-length POPDC2 cDNA sequences (NM_001308333-hg19; wild type [WT], c.516_527del: p.Gln172_Tyr176delinsHis and c.788G>A:p.Arg263His) were synthesized, cloned into pBluescript IISK+ (GeneCust, Boynes, France), and subsequently subcloned into pIRES-GFP (pCGI). Details on cell preparation and expression can be found in the supplemental information. INa and TREK-1 currents were measured with ruptured and amphotericin-perforated patch-clamp technique, respectively, using an Axopatch 200B amplifier (Molecular Devices Corporation, Sunnyvale, CA, USA). Voltage control, data acquisition, and analysis were accomplished using custom software. INa recordings were low-pass filtered with a cutoff frequency of 5 kHz and digitized at 20 kHz, while this was 2 and 4 kHz, respectively, for TREK-1 current measurements. Series resistance was compensated by ≥80%, and potentials were corrected for the calculated liquid junction potential. Cell membrane capacitance (Cm) was calculated by dividing the time constant of the decay of the capacitive transient after a −5-mV voltage step from −40 mV by the series resistance. Patch pipettes were pulled from borosilicate glass (Harvard Apparatus, UK) and had resistances of 2.5–3.5 MΩ after filling with the solutions as indicated below. Measurements from a minimum of nine cells from three independent transfections were acquired for each condition. TREK-1 currents were recorded at 36 ± 0.2°C. Cells were superfused with solution containing (in mM) NaCl 140, KCl 5.4, CaCl2 1.8, MgCl2 1, glucose 5.5, and HEPES 5 at pH 7.4 (NaOH). Pipette solution contained (in mM) K-gluc 125, KCl 20, NaCl 5, amphotericin-B 0.88, and HEPES 10 at pH 7.2 (KOH). TREK-1 currents were measured using 500-ms voltage-clamp steps to test potentials ranging from −100 to +50 mV from a holding potential of −80 mV. The TREK-1 current was measured at the end of the voltage-clamp step and current densities were calculated by dividing current amplitude by Cm. INa was measured at room temperature using a bath solution containing (in mM) NaCl 20, CsCl 120, CaCl2 1.8, MgCl2 1.0, glucose 5.5, and HEPES 5.0 at pH 7.4 (CsOH). Pipettes were filled with solution containing (in mM) NaF 10, CsCl 10, CsF 110, EGTA 11, CaCl2 1.0, MgCl2 1.0, Na2ATP 2.0, and 10 HEPES at pH 7.2 (CsOH). The INa density and voltage dependence of activation were determined by 50-ms depolarizing pulses to test potentials ranging from −80 to +40 mV from a holding potential of −120 mV. Voltage-dependent inactivation was obtained by measuring the peak currents during a 50-ms test step to −20 mV, which followed a 500-ms prepulse to membrane potentials between −140 and 0 mV to allow inactivation. The holding potential was −120 mV. All voltage-clamp steps were applied with a 5-s cycle length. Peak INa was defined as the difference between peak and steady-state current. Current density was calculated by dividing the measured currents by Cm. To determine the activation characteristics of INa, current-voltage curves were corrected for differences in driving force and normalized to maximum peak current (Imax). Steady-state activation and inactivation curves were fitted using the Boltzmann equation I/Imax = A/ to determine the membrane potential for half-maximal (in)activation (V1/2) and the slope factor k. Data are presented as mean ± standard error of the mean (SEM). Statistical analysis was carried out with SigmaStat 3.5 software. Normality and equal variance assumptions were tested with the Kolmogorov-Smirnov and the Levene median test, respectively. Groups were compared with one-way ANOVA. p < 0.05 defines statistical significance. The spontaneous electrical activity of a single human sinus nodal pacemaker cell was simulated using the comprehensive mathematical model developed by Fabbri et al. For simulations of a single human atrial cell, we used the model by Maleckar et al. The CellML code of both models, as available from the CellML Model Repository at https://www.cellml.org/, was edited and run in version 0.9.31.1409 of the Windows based Cellular Open Resource (COR) environment. TREK-1 currents are not included in both original models. To study whether the homozygous loss LOF variants in POPDC2 contribute to bradycardia via TREK-1 current changes, we fitted our experimental data of the TREK-1 current-voltage relationship (Figure 3A) and implemented the thus-obtained TREK-1 current in both models as a control over a range of TREK-1 current densities. Subsequently, the TREK-1 density was reduced to 59% of the TREK-1 + POPDC2-WT current according to the effects induced by the POPDC2 variants (Figure 3A) to assess the functional effects of the variants. All simulations were run for a period of 200 s, which appeared a sufficiently long time to reach steady-state behavior. The analyzed data are from the final 10 s of the 200-s period. The recent availability of population-level cohorts with both clinical as well as whole-genome sequencing and well-imputed array genotyping data now provide the opportunity for orthogonal validation of genetic findings through complementary population-level analysis. Samples were included from four large population biobanks (total n = 1,089,031) with genetic data, namely deCODE genetics in Iceland (n = 173,025), UK Biobank (n = 428,503), Copenhagen Hospital Biobank and the Danish Blood Donor study in Denmark (n = 487,356), and Intermountain in Utah, USA (n = 138,006). Disease status was obtained from electronic health records and ascertained using the following International Classification of Diseases 10th revision codes: atrioventricular block (I44.1 and I44.2), bradycardia (R00.1), cardiac arrest (I46), hypertrophic cardiomyopathy (I42.1 and I42.2), muscular dystrophy (G71.0), myocarditis (I40, I41 and I51.4), and sinus node dysfunction (I49.5). Pacemaker implantation was defined based on procedure codes (deCODE: Nomesco Classification of Surgical Procedures [NCSP] codes FPE/FPSE and FPF/FPSF. Copenhagen Hospital Biobank: NCSP codes FPE/FPSE and FPF/FPSF. UK Biobank: National Clinical Coding Standards OPCS code K60). Heart rate was available only in the UK Biobank and deCODE. In the UK Biobank, heart rate was obtained during blood-pressure measurement at assessment. Both measurements were taken twice, and multiple measurements for one individual were averaged. In deCODE, heart-rate measurements were sourced from electrocardiograms (ECGs) from Landspitali University Hospital in Iceland between 1998 and 2015. Mean values from sinus-rhythm ECGs were obtained for each individual, which were subsequently standardized and adjusted for age and sex. Details on each biobank and DNA genotyping and sequencing methods for each biobank can be found in the supplemental information. We defined different models to group together various types of variants:(1)LOF: only LOF variants according to Variant Effect Predictor (VEP).(2)LOFTEE: high-confidence LOF variants according to LOFTEE.(3)LOFCADD: LOF and missense (MIS) when predicted deleterious with CADD phred score ≥ 25.(4)LOF1MISID: LOF and MIS when predicted deleterious by at least one of the following prediction methods: MetaSVM, MetaLR, or CADD phred score ≥ 25. LOF: only LOF variants according to Variant Effect Predictor (VEP). LOFTEE: high-confidence LOF variants according to LOFTEE. LOFCADD: LOF and missense (MIS) when predicted deleterious with CADD phred score ≥ 25. LOF1MISID: LOF and MIS when predicted deleterious by at least one of the following prediction methods: MetaSVM, MetaLR, or CADD phred score ≥ 25. In all models, we used minor allele frequency (MAF) <2% to select variants for analyses. For case-control analyses, we used logistic regression under an additive model to test for association between carrying an LOF variant in POPDC2 (LOF, MIS according to each model A–D) and phenotypes, in which disease status was the dependent variable and genotype counts as the independent variable. For the analyses, we used software developed at deCODE genetics. For testing association with heart rate, measurements were inverse-normal transformed and analyzed using a linear mixed model implemented in BOLT-LMM. Meta-analysis was performed on the summary results from IS, UK, DK, and US when available, using a fixed-effects inverse-variance-weighted method. To uncover the genetic cause of sinus node disease, AV conduction defects and HCM in a child (age at presentation 11–15 years; family A), we performed whole-exome sequencing in the proband, both parents, and two of his unaffected siblings. His parents are first cousins (Figure 1, individual II-3 in the pedigree; Tables 1 and S3; Figure S1).Figure 1Bi-allelic variants in POPDC2 cause a recessive syndrome with sinus node disease and atrioventricular conduction defects with HCM(A) Pedigrees of families A–D. Closed symbols indicate affected individuals. Males are indicated by squares and females by circles. A double line indicates a consanguineous relationship. The arrows point to the probands. Affected individual 6 from family D (highlighted in the pedigree with a filled red box) was diagnosed with bradycardia resulting in an arrest and first-degree AV block during an episode of fulminant myocarditis.(B) Selection of ECG abnormalities: (1) affected individual II-3 from family A (second-degree AV block type Wenckebach and sinus pause [indicated by an arrowhead]; see Figure S1 for longer Holter registration), (2) affected individual II-4 from family B (second-degree AV block type Wenckebach), affected individual II-3 (non-sustained ventricular tachycardia), and affected individual II-1 (2:1 second-degree AV block) from family C. Arrows point to a non-conducted P wave. Upper right panel: cardiac MRI at the age of 11–15 years showing marked hypertrophy of the interventricular septum (23 mm, Z score: 16.43; height, 160 cm; weight, 49 kg) in the proband of family A.(C) POPDC2 protein domain structure and location of variants found in affected individuals. AV, atrioventricular; CTD, carboxy-terminal domain; ECD, extracellular domain; ND, genotype not determined; regions I/II/II, transmembrane region 1–3; WT, wild type.Table 1Clinical characteristics of affected individuals with bi-allelic POPDC2 variantsFamily AFamily BFamily CFamily DII-3II-4II-1II:2II:3II-1POPDC2 variantp.Gln172_Tyr176delinsHisp.Arg263Hisp.Arg263Cysp.Trp188Ter; p.Leu37Serfs20Effect on TREK1 currentLOFLOFnot testednot testedConsanguinityyesyesnonoSexMMMMMMAge at presentation (years)11–1521–2521–2521–2516–2011–15Symptoms at presentationpalpitationspalpitationsnonenonecardiac arrestchest painCardiac arrhythmia and ECG abnormalitiesCardiac arrestnononoarrest following bradycardiaarrest following bradycardiaMinimum heart rate (BPM)333033N/AN/A67Sinus node diseasesinus pauses (3 s)sinus bradycardianoSA-block sinus pauses (5 s)SA blocksinus arrestnoAV conduction disease2 AVB (type 1)PQ time of 200 mssecond AVBfirst and second AVB (type 1 and 2:1)first AVBfirst and second AVB (type 1)first AVBAtrial arrhythmiaatrial fibrillation and flutternoatrial fibrillation and flutternoepisodes of high atrial ratenoVentricular arrhythmiamonomorphic NSVT, PVCsnomonomorphic NSVTnomonomorphic NSVTVTStructural cardiac and extracardiac abnormalitiesHypertrophic cardiomyopathyyesyesnonononoCardiac MRI findings (mm)septal hypertrophy (23 mm)septal hypertrophy (16mm)N/AN/Apossible myocarditissignificant LV fibrosis and inflammationMyocarditisnonononopossible myocarditisclinically (not on cardiac histology)TreatmentPacemaker implantation (age, years)yes (15)noyes (23)yes (21)yes (17)yes (15), temporaryICD implantation (age, years)yes (15)nononoyes (33)noAppropriate ICD shocksnoN/AN/AN/AnoN/AAVB, atrioventricular block, BPM, beats per minute; ICD, implantable cardiac defibrillator; MRI, magnetic resonance imaging; N/A, not available, NSVT, non-sustained ventricular tachycardia; PVC, premature ventricular contractions; SA, sinoatrial; TTE, transthoracic echocardiogram; VT, ventricular tachycardia. Bi-allelic variants in POPDC2 cause a recessive syndrome with sinus node disease and atrioventricular conduction defects with HCM (A) Pedigrees of families A–D. Closed symbols indicate affected individuals. Males are indicated by squares and females by circles. A double line indicates a consanguineous relationship. The arrows point to the probands. Affected individual 6 from family D (highlighted in the pedigree with a filled red box) was diagnosed with bradycardia resulting in an arrest and first-degree AV block during an episode of fulminant myocarditis. (B) Selection of ECG abnormalities: (1) affected individual II-3 from family A (second-degree AV block type Wenckebach and sinus pause [indicated by an arrowhead]; see Figure S1 for longer Holter registration), (2) affected individual II-4 from family B (second-degree AV block type Wenckebach), affected individual II-3 (non-sustained ventricular tachycardia), and affected individual II-1 (2:1 second-degree AV block) from family C. Arrows point to a non-conducted P wave. Upper right panel: cardiac MRI at the age of 11–15 years showing marked hypertrophy of the interventricular septum (23 mm, Z score: 16.43; height, 160 cm; weight, 49 kg) in the proband of family A. (C) POPDC2 protein domain structure and location of variants found in affected individuals. AV, atrioventricular; CTD, carboxy-terminal domain; ECD, extracellular domain; ND, genotype not determined; regions I/II/II, transmembrane region 1–3; WT, wild type. Clinical characteristics of affected individuals with bi-allelic POPDC2 variants AVB, atrioventricular block, BPM, beats per minute; ICD, implantable cardiac defibrillator; MRI, magnetic resonance imaging; N/A, not available, NSVT, non-sustained ventricular tachycardia; PVC, premature ventricular contractions; SA, sinoatrial; TTE, transthoracic echocardiogram; VT, ventricular tachycardia. No likely causal variant was found in genes previously associated with Mendelian cardiomyopathies or arrhythmia syndromes (either recessive or dominant; Table S4). In line with the expected recessive inheritance pattern, we identified a rare segregating homozygous variant, NM_001308333.2:c.516_527del; NP_001295262.1:p.(Gln172_Tyr176delinsHis), in the Popeye domain-containing protein 2 (POPDC2), as the most likely variant consistent with a recessive mode of inheritance and parental consanguinity (see Table S5 for overview of the two segregating coding-region variants identified in the homozygous state). Of note, the only other homozygous segregating variant has been found 15 times in gnomAD and was therefore found unlikely to cause the disease in this individual. No de novo variants were found in the index affected individual. The cardiac arrhythmia phenotype in the affected individual is consistent with studies in mice that showed that loss of Popdc2 resulted in sinus pauses and bradycardia and with findings made in zebrafish where morpholino knockdown of popdc2 resulted in AV block. We therefore found the p.Gln172_Tyr176delinsHis variant as an excellent candidate for this cardiac disorder. We then searched for additional individuals carrying bi-allelic variants in POPDC2 by screening this gene in 78 individuals that presented with a similar phenotype to family A (i.e., CCD with HCM) and a search using GeneMatcher and DECIPHER. In total, we identified six affected individuals from four unrelated families of different ancestries (affected individuals 3–5 are siblings, Figures 1A and 1B; Tables 1 and S3) who harbor either homozygous or compound heterozygous rare POPDC2 variants. All six affected individuals were diagnosed with CCDs with or without HCM. In summary (see Tables 1 and S3; Figure 1B), (1) AV conduction disease was present in all six individuals, mainly consisting of first-degree AV block or second-degree AV block type 1 (Wenkebach), (2) sinus node disease presenting as sinus bradycardia and sinus pauses was present in four out of six individuals, (3) cardiac arrest accompanied by sinus bradycardia or asystole was seen in two out of six individuals, (4) atrial arrhythmia (i.e., atrial flutter or fibrillation) was detected in three out of six individuals, (5) non-sustained ventricular tachycardia occurred in four out of six individuals, and (6) HCM was diagnosed in two out of six individuals (probands in families A and B with no other genetic variant causing HCM). A pacemaker was implanted in five out of six affected individuals with the age at implantation ranging from 15 to 23 years. In two out of six affected individuals, an implantable cardioverter-defibrillator (ICD) was implanted to prevent lethal ventricular arrhythmia (ranging from 15 to 33 years), but no appropriate ICD shocks occurred. Unlike the recessive syndrome associated with the other POPDC genes (i.e., POPDC1 and POPDC3-related limb-girdle muscular dystrophy), none of the affected individuals showed signs of muscular dystrophy. Of note, individual 6 (II-1 in family D; Figure 1A) presented with acute myocarditis, for which he was admitted to the intensive care unit. Detailed phenotypic descriptions of all affected individuals can be found in Note S1. We noted an autosomal recessive mode of inheritance in families A and C wherein the identified variants in POPDC2 were inherited from each of the unaffected parents. While we expect a recessive inheritance, we could not assess the inheritance pattern in individuals 2 and 6 (from family B, II-4; and D, II-1; Figure 1A) as DNA of (one of) the parents was not available. In total, we report five POPDC2 variants (RefSeq transcript: NM_001308333.2; protein ID: NP_001295262.1), of which two are missense variants (c.788G>A [p.Arg263His], c.787C>T [p.Arg263Cys]), one in-frame insertion deletion (c.516_527del [p.Gln172_Tyr176delinsHis]) and two are expected to result in protein truncation (c.563G>A [p.Trp188Ter], c.110_113del [p.Leu37SerfsTer20], Table S2). The MAF of POPDC2 variants identified in affected individuals ranged from 1.4 × 10 to 1.3 × 10 in the gnomAD v4.0 (accessed March 2024; Table S2). Furthermore, none of the variants were found homozygous in 730,947 exomes and 76,215 genomes from gnomAD, suggesting that homozygosity for these variants is not well tolerated. In line with this, all variants are predicted damaging by multiple in silico prediction tools (Table S2). Intriguingly, in families B and C, the same residue is affected: p.Arg263His and p.Arg263Cys, respectively. Of note, a homozygous variant (c.787G>A [p.Arg261Gln]) in POPDC3, which affects the paralogous residue of p.Arg263 in POPDC2 (families B and C), has been associated with muscular dystrophy without cardiac arrhythmia (MIM: 604577). In family C, the three affected siblings homozygous for p.Arg263Cys inherited the variant from the heterozygous and clinically unaffected parents (II-1, II-2, and II-3 in Figure 1A). We performed a look-up of the p.Arg263Cys variant in recent whole-genome sequencing dataset consisting of young affected individuals (n = 226) receiving a pacemaker because of AV block, and none of the individuals was homozygous for p.Arg263Cys, nor were there any individuals with bi-allelic variants in POPDC2 (Note S2). In family D, we identified both p.Trp188Ter and p.Leu37Serfs20 variants that are expected to result in LOF due to premature truncation of the protein and/or nonsense-mediated decay (individual II-1 in Figure 1A). The cardiac arrest accompanied by sinus bradycardia and first-degree AV block he displayed fitted the conduction disease seen in the other POPDC2-affected individuals. However, although the cardiac biopsy was negative for myocarditis, a potential causal role of myocarditis cannot be ruled out, as the disease can be focal and patchy, potentially reducing sensitivity due to sampling error. To explore the structural and functional consequences of the homozygous and compound heterozygous POPDC2 variants, we generated a predicted structural model of POPDC2, as no experimentally determined structures are currently available. SWISS-MODEL and AlphaFold Multimer were used since these protein-structure-prediction programs could generate dimeric models of POPDC2, and dimerization has been shown to be critical for the activity of POPDC1. As a complementary approach, AlphaMissense was used to predict relative pathogenicity scores of the identified single-amino-acid substitutions and deleted regions. AlphaFold Multimer generated a high-confidence dimer (Figure S2A) for full-length POPDC2 with a dimeric transmembrane domain composed of six transmembrane α helices (three α helices from each subunit) and two cAMP-binding Popeye domains that also contained an extensive dimer interface (Figure 2A). In the AlphaFold model, Arg263 (altered in families B and C) was pointing into the cAMP-binding pocket of the Popeye domain from the adjacent subunit. SWISS-MODEL was the only program able to model a cAMP-bound state of the Popeye domain (Figure S2B). To generate a model of the cAMP-bound state of full-length POPDC2, we merged the AlphaFold Multimer and SWISS-MODEL predictions by swapping the Popeye domains (see section material and methods). This created a merged model of POPDC2 that included the transmembrane domain and the Popeye domains bound to cAMP (Figure 2A).Figure 2Functional characterization of the POPDC2 variants(A) Structural model of POPDC2 bound to cAMP generated using AlphaFold2 Multimer and SWISS-MODEL. Dimer subunits are shown in green and cyan; cAMP molecules, green and cyan sticks. N′ and C′ indicate the N and C termini. The positions of the transmembrane and Popeye domains are indicated by the labels. The intrinsically disordered C termini (residues 275–364) are shown as dotted lines.(B) Zoom-in of the predicted cAMP-binding pocket of POPDC2. Dimer subunits are shown in green and cyan; cAMP, green sticks; residues p.Gln172_Tyr176delinsHis, orange; p.Arg263, cyan sticks.(C) Structural models of the POPDC2 variants p.Leu37Serfs20 and p.Trp188Ter, which would both generate truncated proteins that would lack the ability to bind cAMP.(D) Homology model of POPDC2 by AlphaFold Multimer color coded by the average pathogenicity score for each residue as predicted by AlphaMissense.(E) Heatmap of predicted effects of amino-acid substitutions on POPDC2. AlphaMissense (AM) scores range from zero to one, with higher scores corresponding to increased pathogenicity.(F) Homology model of POPDC2 by AlphaFold Multimer color coded by the average pathogenicity score for each residue as predicted by AlphaMissense. The position of non-disease-associated variants found in general population are shown as blue spheres, indicating predicted AlphaMissense pathogenicity scores <0.1.(G) Typical examples of TREK-1 currents upon 500-ms voltage-clamp steps to membrane potentials ranging from −100 to +50 mV from a holding potential of −80 mV in absence or presence of wild-type (WT) and mutant POPDC2.(H) Average current-voltage relationships of TREK-1 currents in absence or presence of WT and mutant POPDC2.(I) TREK-1 current amplitude at +50 mV in absence or presence of WT and mutant POPDC2. p < 0.05 with one-way ANOVA. Error bars indicate the standard error of the mean (SEM). Functional characterization of the POPDC2 variants (A) Structural model of POPDC2 bound to cAMP generated using AlphaFold2 Multimer and SWISS-MODEL. Dimer subunits are shown in green and cyan; cAMP molecules, green and cyan sticks. N′ and C′ indicate the N and C termini. The positions of the transmembrane and Popeye domains are indicated by the labels. The intrinsically disordered C termini (residues 275–364) are shown as dotted lines. (B) Zoom-in of the predicted cAMP-binding pocket of POPDC2. Dimer subunits are shown in green and cyan; cAMP, green sticks; residues p.Gln172_Tyr176delinsHis, orange; p.Arg263, cyan sticks. (C) Structural models of the POPDC2 variants p.Leu37Serfs20 and p.Trp188Ter, which would both generate truncated proteins that would lack the ability to bind cAMP. (D) Homology model of POPDC2 by AlphaFold Multimer color coded by the average pathogenicity score for each residue as predicted by AlphaMissense. (E) Heatmap of predicted effects of amino-acid substitutions on POPDC2. AlphaMissense (AM) scores range from zero to one, with higher scores corresponding to increased pathogenicity. (F) Homology model of POPDC2 by AlphaFold Multimer color coded by the average pathogenicity score for each residue as predicted by AlphaMissense. The position of non-disease-associated variants found in general population are shown as blue spheres, indicating predicted AlphaMissense pathogenicity scores <0.1. (G) Typical examples of TREK-1 currents upon 500-ms voltage-clamp steps to membrane potentials ranging from −100 to +50 mV from a holding potential of −80 mV in absence or presence of wild-type (WT) and mutant POPDC2. (H) Average current-voltage relationships of TREK-1 currents in absence or presence of WT and mutant POPDC2. (I) TREK-1 current amplitude at +50 mV in absence or presence of WT and mutant POPDC2. p < 0.05 with one-way ANOVA. Error bars indicate the standard error of the mean (SEM). Based on the structural model, all variants (p.Gln172_Tyr176delinsHis, p.Arg263His, p.Arg263Cys, and p.Trp188Ter/p.Leu37Serfs20) are predicted to diminish the ability of POPDC2 to bind cAMP. The first variant (p.Gln172_Tyr176delinsHis, family A) replaces five residues with a single histidine residue. The deleted residues are predicted to form one beta strand that is part of a three-stranded beta sheet at the base of the cAMP-binding pocket (Figure 2B). Replacement of the five residues with a single histidine residue would significantly alter the structural integrity of the cAMP-binding Popeye domain and directly affect the ability of POPDC2 to bind cAMP. The second and third variants, p.Arg263His (found in family B) and p.Arg263Cys (found in family C), are also both predicted to interfere with proper cAMP binding, as the positively charged Arg residue is predicted to be near the negatively charged cyclic phosphate group of cAMP (Figure 2B). Change of the Arg to either His or Cys would eliminate a key interaction predicted to stabilize cAMP binding and thus reduce or eliminate the ability of POPDC2 to bind cAMP. In support of Arg263 being involved in a specific interaction to stabilize cAMP binding, alteration of Arg263 to any other residue is predicted to be pathogenic by AlphaMissense, with pathogenicity scores of 0.82 and 0.79 for the p.Arg263His and p.Arg263Cys variants, respectively. The compound heterozygous variants are also predicted to eliminate cAMP binding by POPDC2 (Figure 2C). p.Leu37SerfsTer20 is a truncated protein that completely lacks the cAMP-binding Popeye domain and would also generate an incomplete transmembrane domain. p.Trp188Ter would retain the dimeric transmembrane domain, but truncate the Popeye domain, thus leaving an incomplete domain incapable of binding to cAMP (Figure 2C). In addition, both variants could be subject to nonsense-mediated decay, which would result in no protein product being made. Notably, residues with high pathogenicity scores (>0.5) all cluster in the regions of the POPDC2 structure that are involved in either dimerization or cAMP binding (Figures 2D and 2E), which emphasizes the importance of cAMP binding and dimerization for POPDC2 function. Consistently, other variants within POPDC2, found homozygously in gnomAD, and without a clear disease association (p.Arg17Val, p.Val29Ile, p.Arg98Cys, p.Val270Ile, and p.Ala321Thr), are all positioned away from the cAMP-binding pocket and are predicted to be benign by AlphaMissense with low pathogenicity scores <0.1 (Figures 2E and 2F). Taken together, the disease-associated variants reported here are primarily located in regions of the POPDC2 structure that are critical for protein function. We then aimed to functionally characterize the POPDC2 variants. TREK-1 is a recognized interacting protein of POPDC2, and co-expression of POPDC2 and TREK-1 has been shown to increase TREK-1 current in comparison to expression of TREK-1 alone. Furthermore, cardiac-specific TREK-1-deficient mice display a sinus node phenotype characterized by bradycardia with frequent episodes of sinus pauses, partially resembling the phenotype in the affected individuals with bi-allelic POPDC2 variants presented here. We therefore hypothesized that the effect of the p.Gln172_Tyr176delinsHis (family A) and p.Arg263His (family B) variants is mediated through modulation of the TREK-1 current. We did not test the variants found in family C (p.Arg263Cys) and family D (p.Trp188Ter/p.Leu37Serfs20) as they affect the same residue as the variant in family B (p.Arg263His) or are expected to result in premature truncation of the protein and possibly nonsense-mediated decay, respectively. We co-transfected HEK293 cells with WT and mutant POPDC2 with TREK-1 containing plasmids. As expected, co-expression of WT POPDC2 with TREK-1 increased TREK-1 current density (Figures 2G–2I). However, when we co-expressed TREK-1 with a p.Gln172_Tyr176delinsHis or p.Arg263His POPDC2-containing plasmid, no increase in TREK-1 current density was observed, comparable to the co-expression of TREK-1 with an empty vector (Figures 2G–2I). We then tested the effect of the POPDC2 variants on sodium current (INa) as it was recently hypothesized that POPDC2 may interact with NaV1.5. However, expression of WT, p.Gln172_Tyr176delinsHis, or p.Arg263His POPDC2 in an HEK293 cell line stably expressing human NaV1.5 channels (encoded by SCN5A) showed an effect neither on INa density nor on INa gating properties by WT or mutant POPDC2 (Figure S3). Western blot analysis showed levels of expression of the mutant proteins that were comparable to the levels seen for WT POPDC2 (Figure S4). To evaluate whether the observed decrease in TREK-1 current density, associated with the POPDC2 variants, is responsible for the bradycardia observed in affected individuals, we conducted computer simulations using comprehensive mathematical models of both a human sinus nodal pacemaker cell and a human atrial cell. As shown in Figure 3A (blue line), the experimental data on the voltage dependence of the TREK-1 + POPDC2-WT current could be well fitted (r > 0.99) with the relationship ITREK-1 = −117.1 + 284.24 × exp(Vm/90.52), in which ITREK-1 and Vm denote TREK-1 current density (in pA/pF) and membrane potential (in mV), respectively. The mutant data could be well fitted by scaling down ITREK-1 to 59% of the TREK-1 + POPDC2-WT current over the entire volage range (Figure 3A, red line). We started our simulations with incorporating ITREK-1, which is not present in the original model cell, into the human sinus node model cell. Because data on the density of ITREK-1 in human sinus node cells are lacking, we set its density to 1.2 pA/pF (at +30 mV), as observed in mouse sinus node cells by Unudurthi et al. This, however, resulted in cessation of pacemaker activity in the human sinus node model cell. We then repeated our simulations with a 10 times lower density of ITREK-1. With this density, the cycle length of the simulated action potential amounted to 1,160 ms (Figure 3B, top panel, blue solid line), as compared to 813 ms in the original model (Figure 3B, top panel, gray dotted line). Cycle length was reduced by 16% from 1,160 to 973 ms upon reduction of ITREK-1 density to 59%, thus simulating the effect of the variants in POPDC2 (Figure 3B, top panel, red solid line; Figure 3C, vertical arrow). The decrease in cycle length was mainly due to an increase in diastolic depolarization rate (Figure 3B, top panel) as a result of the decrease in the small but effective ITREK-1 during this phase of the action potential (Figure 3B, bottom panel). Qualitatively similar results were obtained with other ITREK-1 densities. The observed effects on cycle length are summarized in Figure 3C.Figure 3Functional effects of the POPDC2 variants from in silico modeling(A) Fits to the experimental data on TREK-1 currents in absence or presence of wild-type (WT) and mutant POPDC2. The fit to the mutant data were obtained by scaling the fit to the WT data by a factor of 0.59.(B) Membrane potential (top) and associated TREK-1 current (bottom) of a single human sinus nodal pacemaker cell as simulated using the comprehensive mathematical model developed by Fabbri et al. ITREK-1 was introduced into the original model cell using the fits of (A). ITREK-1 magnitude was set to 0.12 pA/pF at a membrane potential of +30 mV(C) Cycle length of the simulated single human sinus nodal pacemaker cell as a function of ITREK-1 magnitude.(D) Membrane potential (top) and associated TREK-1 current (bottom) of a single human atrial cell as simulated using the comprehensive mathematical model developed by Maleckar et al. ITREK-1 was introduced into the original model cell using the fits of (A). ITREK-1 magnitude was set to 4.0 pA/pF at a membrane potential of +30 mV. Action potentials were elicited at a rate of 1 Hz with a 1-ms, 20% suprathreshold stimulus current.(E) Diastolic potential of the simulated single human atrial cell as a function of ITREK-1 magnitude.(F) Threshold stimulus current of the simulated single human atrial cell as a function of ITREK-1 magnitude. Functional effects of the POPDC2 variants from in silico modeling (A) Fits to the experimental data on TREK-1 currents in absence or presence of wild-type (WT) and mutant POPDC2. The fit to the mutant data were obtained by scaling the fit to the WT data by a factor of 0.59. (B) Membrane potential (top) and associated TREK-1 current (bottom) of a single human sinus nodal pacemaker cell as simulated using the comprehensive mathematical model developed by Fabbri et al. ITREK-1 was introduced into the original model cell using the fits of (A). ITREK-1 magnitude was set to 0.12 pA/pF at a membrane potential of +30 mV (C) Cycle length of the simulated single human sinus nodal pacemaker cell as a function of ITREK-1 magnitude. (D) Membrane potential (top) and associated TREK-1 current (bottom) of a single human atrial cell as simulated using the comprehensive mathematical model developed by Maleckar et al. ITREK-1 was introduced into the original model cell using the fits of (A). ITREK-1 magnitude was set to 4.0 pA/pF at a membrane potential of +30 mV. Action potentials were elicited at a rate of 1 Hz with a 1-ms, 20% suprathreshold stimulus current. (E) Diastolic potential of the simulated single human atrial cell as a function of ITREK-1 magnitude. (F) Threshold stimulus current of the simulated single human atrial cell as a function of ITREK-1 magnitude. One may argue that the POPDC2 variant-induced decrease in ITREK-1 affects pacemaker activity by changing the excitability of the atrial tissue surrounding the sinus node. This was assessed in simulations of a human atrial cardiomyocyte after incorporating ITREK-1, which was not represented in the original model cell. Figure 3D shows the results obtained with an ITREK-1 density of 4.0 pA/pF (at +30 mV). Simulating the effect of the variants in POPDC2 by lowering the density of ITREK-1 to 59% led to a 0.63-mV depolarization of diastolic potential, a 5.1% decrease in threshold current (Figures 3E and 3F, vertical arrows), and a 23-ms increase in action-potential duration (APD) at 90% repolarization. The relatively small effects on diastolic potential and threshold current were obtained with a probably overestimated ITREK-1 density, given the large effect of ITREK-1 on the APD of the original model (Figure 3D, top panel). Smaller, but qualitatively similar, effects on diastolic potential and threshold current were obtained with lower ITREK-1 densities (Figures 3E and 3F), associated with a less prominent effect of ITREK-1 on the APD of the original model. Overall, despite the observation of bradycardia in mice lacking TREK-1, findings from our simulation studies do not provide an explanation for the clinically observed sinus bradycardia in individuals with POPDC2 variants based on a decrease in ITREK-1 per se. Individual II-3 from family A (Figure 1A) underwent muscle biopsy of the m. vastus lateralis (Supplemental information). Histological analysis disclosed mild fiber size variability and a slight increase of connective tissue. Neither nuclear centralizations nor fiber splittings were observed. No necrotic or degenerated fibers were detected (Figures 4A and 4B). Non-specific myopathic features and increased connective tissue were previously observed in muscle samples from individuals with pathogenic POPDC1 variants, while affected individuals with POPDC3 variants displayed typical features of muscle dystrophies, although the severity of the histopathological findings differed among affected individuals.Figure 4Evaluation of muscle biopsy from the proband in family A(A and B) (A) Hematoxylin and eosin (H&E) stain and (B) modified Gomori trichrome (MGT) staining of affected individual and, in the inset, control muscle. Scale bar, 50 μm.(C) Immunofluorescent staining for caveolin-3 (red), POPDC1 (green), and merge for affected individual. The inset shows the corresponding immunofluorescence staining for control. Scale bar, 50 μm.(D) Immunofluorescent staining for caveolin-3 (red), POPDC2 (green), and merge for affected individual. The corresponding control staining is shown in the inset. Scale bar, 50 μm.(E–G) Ultrastructural findings. (E) Tubular aggregates in subsarcolemmal region. (F) Sarcolemma alteration (asterisks). (G) Increase in lipid droplets. Scale bar (E and F), 0.84 μm; (G), 3.33 μm. Evaluation of muscle biopsy from the proband in family A (A and B) (A) Hematoxylin and eosin (H&E) stain and (B) modified Gomori trichrome (MGT) staining of affected individual and, in the inset, control muscle. Scale bar, 50 μm. (C) Immunofluorescent staining for caveolin-3 (red), POPDC1 (green), and merge for affected individual. The inset shows the corresponding immunofluorescence staining for control. Scale bar, 50 μm. (D) Immunofluorescent staining for caveolin-3 (red), POPDC2 (green), and merge for affected individual. The corresponding control staining is shown in the inset. Scale bar, 50 μm. (E–G) Ultrastructural findings. (E) Tubular aggregates in subsarcolemmal region. (F) Sarcolemma alteration (asterisks). (G) Increase in lipid droplets. Scale bar (E and F), 0.84 μm; (G), 3.33 μm. In healthy skeletal muscle fibers, POPDC1 and POPDC2 were robustly localized in the sarcolemma. Conversely, in the affected individuals’ muscle, immunofluorescence staining showed a significant reduction of POPDC1 (Figure 4C) and POPDC2 (Figure 4D) levels. On the other hand, the immunofluorescent signal for caveolin-3, which also stains muscle membranes, showed normal distribution and intensity in the muscle sample of the affected individual II-3 from family A (Figure 1A) compared to control. The severe reduction of POPDC1 and POPDC2 levels was also documented by SDS-PAGE analysis of muscle protein lysates (Figure S5). These findings suggest that POPDC2 p.Gln172_Tyr176delinsHis affects the stability of POPDC2, hampering its membrane localization and leading to the secondary reduction of POPDC1. Indeed, the stability and/or membrane trafficking of the POPDC1-POPDC2 complex have been found to be impaired by genetic variants in each of the two proteins. Ultrastructural examination detected the presence of tubular aggregates in few muscle fibers (Figure 4E). Alterations in the structure of the sarcolemma, characterized by small microvilli-like projections, together with subsarcolemmal vacuoles were also observed. Basal lamina appeared unstructured and enlarged in some fibers (Figure 4F, asterisks). We also noted increased level of lipids, which in some cases were arranged to form rows of droplets (Figure 4G). Heterogeneous transmission electron microscopy findings were previously observed in muscle samples from affected individuals harboring variants in POPDC1 but not in those with POPDC3 pathogenic variants (MIM: 605824). To explore the anatomical distribution of POPDC1-3 expression (POPDC1’s previous official gene name is BVES), we analyzed a previously published Visium Spatial and single-nucleus transcriptomic dataset. For the Visium Spatial dataset, annotation of histological tissue sections of the AV node and the sinus node was used and expression of POPDC1-3 was explored across those structures (Figure 5A). POPDC2 showed higher expression than POPDC1/BVES and POPDC3. Cardiomyocyte-rich structures showed higher POPDC2 expression compared to cardiomyocyte-poor regions (e.g., cardiac skeleton or fat) (Figure 5B). We then interrogated the co-expression pattern of POPDC1/BVES and POPDC2, as (1) Swan et al. described the necessity of POPDC1-2 co-expression for their trafficking to the cell membrane, which is required for their proper function; and (2) variants in POPDC1/BVES-POPDC2 have also been found to affect the stability and/or membrane trafficking of the complex as corroborated here in the muscle biopsy of the affected individual II-3 from family A (Figure 4). The most prominent area where POPDC1/BVES and -2 expression was seen co-localizing in the same Visium spots was the AV node (Figure 5C). The sinus node expressed POPDC2, but expression of POPDC1 was sparse (Figure 5B).Figure 5Expression of POPDC1-3 in human hearts(A) Overview of single-nucleus and spatial-transcriptomics data analysis from a previously published human heart cell atlas..(B) Spatial-transcriptomics (Visium) analysis of POPDC1 (BVES), POPDC2, and POPDC3 expression across different anatomical regions and histological microstructures in adult human hearts. The anatomical sites sampled included AVN, SAN, left ventricle free wall, left ventricular apex, interventricular septum, left atrium, and right atrium.(C) Percentage of spatial spots where co-expression of both POPDC1/BVES and -2 was detected is shown for each histological feature.(D) POPDC family gene expression across cell types in adult human heart profiled by snRNA-seq (10× Genomics).(E) POPDC1/BVES, -2, and -3 gene expression in cardiomyocyte cell states in adult human hearts.(F) Percentage of cells that co-express POPDC1/BVES and POPDC2 in the same single cardiomyocyte.Figure created with BioRender. Error bars indicate the 95% confidence interval. Expression of POPDC1-3 in human hearts (A) Overview of single-nucleus and spatial-transcriptomics data analysis from a previously published human heart cell atlas.. (B) Spatial-transcriptomics (Visium) analysis of POPDC1 (BVES), POPDC2, and POPDC3 expression across different anatomical regions and histological microstructures in adult human hearts. The anatomical sites sampled included AVN, SAN, left ventricle free wall, left ventricular apex, interventricular septum, left atrium, and right atrium. (C) Percentage of spatial spots where co-expression of both POPDC1/BVES and -2 was detected is shown for each histological feature. (D) POPDC family gene expression across cell types in adult human heart profiled by snRNA-seq (10× Genomics). (E) POPDC1/BVES, -2, and -3 gene expression in cardiomyocyte cell states in adult human hearts. (F) Percentage of cells that co-express POPDC1/BVES and POPDC2 in the same single cardiomyocyte. Figure created with BioRender. Error bars indicate the 95% confidence interval. To validate these findings and to explore POPDC1 and -2 co-expression within the same cell, we analyzed snRNA-seq data from eight regions of adult human hearts, including the conduction system. POPDC2 expression was almost entirely restricted to atrial and ventricular cardiomyocytes compared to other cell types (Figure 5D). Other genes from the POPDC family (i.e., POPDC1 and POPDC3) also showed a preferential expression in cardiomyocytes, but their expression was less prevalent than POPDC2. The snRNA-seq object was then sub-set to cardiomyocytes only and expression of POPDC1-3 interrogated. POPDC2 was expressed in all cardiomyocyte cell states, while POPDC1 showed the highest expression in the pacemaker cells (P cells) of the AV node and the AV bundle cells (Figure 5E). The co-expression of POPDC1 and -2 within the same single-cell was most prevalent in AV node P cells and AV bundle cells, which concurs with the spatial transcriptomic data (Figure 5F). We then also assessed the cardiac expression of Popdc2 in mice utilizing previously published scRNA-seq data of the sinus node and AV node/His region, dissected from embryonic-day-16.5 mouse hearts. Consistent with the phenotype observed in the affected individual, Popdc2 was expressed in cardiomyocytes as well as sinus node and AV node/His-region cells (Figure S6), corroborating observations from previous studies in mice and our observations reported here in human hearts (Figures 5A–5F). We investigated whether heterozygous carriers of the POPDC2 variants found in families A–D (i.e., p.Gln172_Tyr176delinsHis, p.Arg263His, p.Arg263Cys, p.Trp188Ter, and p.Leu37Serfs20) showed (sub-)clinical manifestations of the recessive POPDC2 syndrome in four large population biobanks (total n = 1,089,031) with genetic data, namely deCODE genetics in Iceland (n = 173,025), UK Biobank (n = 428,503), Copenhagen Hospital Biobank and the Danish Blood Donor study in Denmark (n = 487,356), and Intermountain in Utah, USA (n = 138,006). We searched their association with (1) AV block, (2) bradycardia, (3) cardiac arrest, (4) HCM, (5) myocarditis, (6) pacemaker implantation, (7) sinus node dysfunction, and (8) heart rate. We detected 426 heterozygous carriers of the five familial POPDC2 variants (p.Gln172_Tyr176delinsHis, n = 62; p.Arg263His, n = 34; p.Arg263Cys, n = 156; p.Trp188Ter, n = 162; and p.Leu37Serfs20, n = 12) among 1,089,031 participants in the included biobanks but no homozygotes or compound heterozygotes. None of the variants showed statistical association with (sub-)clinical outcomes after Bonferroni correction, suggesting that heterozygous family members are unlikely to develop clinical manifestations and therefore might not necessitate clinical follow-up (Table S6). We then conducted genetic burden analyses, wherein rare variants in POPDC2 are aggregated, with the phenotypes above and muscular dystrophy as this is a clinical hallmark of recessive syndromes associated with the two other members of the POPDC family (i.e., POPDC1 and POPDC3). We found a significant association with bradycardia after Bonferroni adjustment (p < 1.4 × 10) for the following variant sets: LOFCADD (p = 3.3 × 10, odds ratio [OR] 1.6, LOF and missense when predicted deleterious with CADD phred score ≥25) and LOF1MISID (p = 2 × 10, OR 1.59, LOF and missense when predicted deleterious by at least one of the following prediction methods: MetaSVM, MetaLR, or CADD phred score ≥25; Table S7). Although no association with a clinically relevant phenotype was detected, an association with slow heart rate was seen in heterozygous carriers in the general population. Whether this has clinical relevance remains to be explored. We identified bi-allelic variants in POPDC2 in four families that presented with a phenotypic spectrum consisting of sinus node dysfunction, AV conduction defects, and HCM. Using homology modeling, we show that the identified POPDC2 variants are predicted to diminish the ability of POPDC2 to bind cAMP. In in vitro electrophysiological studies, we demonstrated that, while co-expression of WT POPDC2 with TREK-1 increased TREK-1 current density, POPDC2 harboring variants found in the affected individuals failed to increase TREK-1 current density. scRNA-seq from human hearts demonstrated that co-expression of POPDC1 and -2 was most prevalent in AV node, AV node pacemaker, and AV bundle cells. Sinus node cells expressed POPDC2 abundantly, but expression of POPDC1 was sparse. Together, these results concur with predisposition to AV node disease in humans with LOF variants in POPDC1 and POPDC2 and presence of sinus node disease in POPDC2 but not in POPDC1-related disease in human. Our findings provide evidence for bi-allelic variants in POPDC2 as the cause of a Mendelian autosomal recessive cardiac syndrome. Several observations support the causality of the identified POPDC2 variants (encoding p.Gln172_Tyr176delinsHis, p.Arg263His, p.Arg263Cys, p.Trp188Ter, and p.Leu37Serfs20). (1) The cardiac arrhythmia phenotype of the affected individuals harboring these variants in the homozygous or compound heterozygous state is similar to observations of sinus pauses and bradycardia that were made in mice lacking Popdc2 and to AV block observed in zebrafish after morpholino knockdown of popdc2. (2) The variants affect highly conserved residues and are predicted damaging by multiple in silico prediction tools. (3) The variants are extremely rare in gnomAD and were not found in homozygous state. (4) A change affecting the paralogous residue of Arg263 (families B and C) in POPDC3, p.Arg261Gln, has been associated with muscular dystrophy (MIM: 605824). The dimeric homology model of POPDC2 predicted all disease variants we identified in the homozygous or compound heterozygous state to critically affect cAMP binding. One variant leads to the substitution of five residues with a single histidine residue (p.Gln172_Tyr176delinsHis; Figure 2B), and another two (p.Trp188Ter/p.Leu37Serfs20; Figure 2C) are expected to produce a truncated protein, all of which are expected to critically affect the ability of POPDC2 to bind cAMP. In addition, both transcripts could be subject to nonsense-mediated decay, which would result in no protein product being made. While the p.Arg263His and p.Arg263Cys variants are not predicted to disrupt protein folding, they are expected to eliminate a key interaction predicted to stabilize cAMP binding. Based on these predictions, we hypothesize that cAMP binding is critical for POPDC2 function and, when affected, causes disease. Although heterozygous POPDC2 variants have been proposed to increase susceptibility for cardiac conduction disease in humans by Rinné et al., no Mendelian recessive disorder has been linked to this gene so far. In the study by Rinné et al., a heterozygous nonsense variant in POPDC2 (p.Trp188Ter; rs144241265, allele frequency of 1.70 × 10 among European individuals in gnomAD) was identified in the heterozygous state in a monozygotic twin pair presenting with AV block and in an unrelated family with a mother and son both diagnosed with first-degree AV block (i.e., prolongation of the PR-interval on the ECG). The twin pair inherited the variant from the unaffected mother, indicating that it does not fully explain the phenotype observed in the two siblings. In support of this, single-variant analysis that we conducted in 162 carriers of the p.Trp188Ter variant in more than 1 million individuals from the general population did not show an association with sinus node or AV node conduction disease in heterozygous state. Thus, while the findings from Rinné et al. are interesting and support our findings, they do not establish pathogenic genetic variation in POPDC2 as a (recessive or dominant) Mendelian cause of cardiac conduction disease in human. Based on our findings, we recommend the inclusion of POPDC2 in clinical genetic-testing panels for affected individuals presenting with unexplained sinus node dysfunction, AV conduction defects with or without HCM. Using population-level genetic data of more than 1 million individuals, we showed that none of the variants were associated with clinical outcomes in heterozygous state, suggesting that heterozygous family members are unlikely to develop clinical manifestations and therefore might not necessitate clinical follow-up. Bi-allelic variants in POPDC1 and POPDC3 have been associated with muscular dystrophy, with and without CCD, respectively, while the affected individuals reported here presented with isolated cardiac disease. Specifically, none of the affected individuals reported muscular weakness, atrophy, or cramps. Furthermore, muscle biopsy in the proband from family A did not show evident signs of muscle disease. Also, in contrast to POPDC1- and PODPC3-affected individuals, normal serum creatine kinase levels were found in the affected individuals reported here. While we did not detect a muscular phenotype in the affected individuals, currently aged 22–50 years, we cannot exclude subtle or age-dependent expression of muscular defects in POPDC2-related disease. Although all three POPDC proteins are expressed in both skeletal and cardiac muscle, differences in levels of expression between the POPDC proteins might, in part, determine phenotypic expression. Indeed, POPDC2 is predominantly expressed in cardiac tissue, whereas POPDC3, which presents with isolated muscular dystrophy, has a predominant expression in skeletal muscle. Western blot analysis and immunostaining of POPDC1 and POPDC2 in muscle biopsies obtained in the proband from family A showed significant reduction of the expression of both POPDC1 and POPDC2. These findings are in line with previous studies that suggested that stability and/or membrane trafficking of the POPDC1-POPDC2 complex is impaired by variants in each of the two proteins. Using spatial transcriptomics and scRNA-seq from human hearts, we showed that co-expression of POPDC1 and 2 was most prevalent in AV node, AV node pacemaker, and AV bundle cells. On the other hand, in the sinus node, POPDC2 was abundantly expressed, but expression of POPDC1 was sparse. While these results support the observed predisposition to AV node disease in affected individuals with POPDC2 LOF variants, proper statistical testing using pseudo-bulk counts was not possible due to a low number of donors with conduction-system data. Therefore, these results should be treated as hypothesis generating. Together, these results concur with predisposition to AV node disease in humans with LOF variants in POPDC1 and POPDC2 and presence of sinus node disease in POPDC2- but not in POPDC1-related disease in human. POPDC proteins are established interacting partners of the potassium channel TREK-1, which is known to underlie a background potassium current and is highly expressed in the sinus node. Co-expression of TREK-1 with any of the three POPDC proteins leads to an increase in TREK-1 current, a process modulated by the level of cAMP. Furthermore, cardiac-specific TREK-1-deficient mice display a sinus node phenotype characterized by bradycardia with frequent episodes of sinus pauses, partially resembling the phenotype in the affected individuals with POPDC2 variants presented here. In an effort to shed light on the electrophysiological mechanism by which the variants in POPDC2 lead to bradycardia, we therefore conducted co-expression studies of WT and mutant POPDC2 with TREK-1. In vitro, both POPDC2 variants tested failed to increase TREK-1 current and, by virtue of observations of bradycardia in TREK-1-deficient mice, these variants would be expected to be associated with bradycardia. Notwithstanding the clear observation of bradycardia in TREK-1-deficient mice, how loss of background potassium current, causing an increase in diastolic net inward current, leads to bradycardia and sinus pauses is unclear. Our in silico modeling studies showed that a 41% reduction of TREK-1 current, simulating the effect of the variants in POPDC2, leads to an increase in diastolic depolarization rate and spontaneous firing. These findings are in agreement with experiments using isolated sinus node cells of TREK-1-deficient mice and computer simulations using a rabbit sinus node cell model. In a murine cardiac muscle cell line (HL-1), a stop variant in Popdc2 associated with TREK-1 reduction, the maximum diastolic potential (MDP) was depolarized and the action potential upstroke velocity was reduced. In addition, a slower spontaneous firing rate was observed. In contrast, another study that examined loss of TREK-1 channel function in HL-1 cells showed an increase in spontaneous firing rate. Bradycardia may also be induced via changes in excitability of atrial cardiomyocytes surrounding the sinus node. In our simulated human atrial cell, reduction in TREK-1 current density slightly depolarized the MDP and increased the APD, consistent with findings in rat ventricular myocytes, but the excitability was hardly affected. Thus, while collectively these data support a role for TREK-1 in cardiac pacing and a causal effect of POPDC2 variants through modulation of TREK-1 current, the exact cellular electrophysiological mechanism remains unclear. Differences in the cellular models used cannot be excluded. Furthermore, the differences in effects of TREK-1 deficiency observed in vivo and in vitro in TREK-1-deficient mice suggest that other factors (such as altered sympathetic or parasympathetic stimulation in vivo) also contribute to the bradycardia observed at baseline in these mice. While an effect through modulation of sodium channel function could be postulated, no such effect was observed in our patch-clamp studies. Although cardiac arrhythmias in affected individuals with recessive POPDC2 variants are consistent with observations in mice and zebrafish, the role of POPDC2 in cardiac hypertrophy remains unexplained. One of the potential mechanisms underlying cardiac hypertrophy in affected individuals with recessive variants in POPDC2 is modulation of TREK-1. Since TREK-1 is activated by biomechanical stretch, the role of TREK-1 in cardiac responses to chronic pressure was recently studied using transverse aortic constriction. Notably, while no clear structural cardiac differences were seen at baseline, mice lacking TREK-1 exhibited an exaggerated pressure-overload-induced concentric hypertrophy with preserved systolic and diastolic cardiac function compared to WT mice. Affected individual 6 from family D (II-1 in Figure 1A) was diagnosed with bradycardia resulting in an arrest and first-degree AV block during an episode of fulminant myocarditis. We therefore cannot exclude a causal role of myocarditis in this case. However, (1) the conduction phenotype fits with the phenotypic characteristic of the other five affected individuals we report here; (2) the affected individual carried bi-allelic truncating variants in POPDC2 likely to result in complete LOF, which is a known mechanism for disease in animal models; (3) there was no individual among 125,748 exomes and 15,708 genomes in gnomAD that carried both variants found in compound heterozygosity (p.Trp188Ter and p.Leu37Serfs20) in the affected individual from family D; (4) gnomAD contains 70 predicted LOF POPDC2 variants and none of them occurs in the homozygous state; and (5) knockin mice of the p.Trp188Ter variant (found in compound heterozygosity in family D) displayed stress-induced sinus bradycardia and pauses. In aggregate, these data suggest a causal role for these variants in this individual. While at this stage speculative, some of the genetic cardiomyopathies (in particular in DSP) are associated with intermittent myocardial inflammatory episodes that appear clinically similar to myocarditis or sarcoidosis. Presentation with (recurrent) myocarditis-like episodes has been reported for arrhythmogenic cardiomyopathy (ACM). Among 560 probands and family members with ACM, Bariani et al. reported an episode resembling myocarditis (i.e., “hot phase”) in 23 cases (5%), particularly in pediatric affected individuals and carriers of desmoplakin (MIM: 125647) variants. Furthermore, in a population-based cohort of 336 consecutive affected individuals with acute myocarditis, a significant enrichment of pathogenic variants in genes associated with dilated or arrhythmogenic cardiomyopathy was found (8%) in comparison with controls (<1%, p = 0.0097). The question remains whether the myocarditis exposed the underlying POPDC2-related conduction disease or whether myocarditis is part of the POPDC2 phenotypic spectrum. While we find it unlikely, compound heterozygosity of the truncating POPDC2 variants might have been irrelevant in this case, and the phenotype could be fully the consequence of myocarditis. This study is limited by its relatively small sample size, with four families presenting bi-allelic POPDC2 variants, which might have not allowed us to provide the full clinical spectrum of disease associated with recessive POPDC2 variants. Finally, while population-level data suggest no clinical manifestations in heterozygous carriers, longer-term clinical follow-up would be needed to confirm this in diverse populations and the heterozygous family members in POPDC2 families. We here provide robust association of bi-allelic variants in POPDC2 with a Mendelian autosomal recessive cardiac syndrome consisting of sinus node dysfunction, AV conduction defects, and HCM. Future studies will help to illuminate the full clinical spectrum of the disease in individuals with bi-allelic variants as well as the clinical presentation of heterozygous carriers, thus elucidating the underlying mechanism of disease. The genetic data from families A–D supporting the current study have not been deposited in a public repository because of ethical restrictions but are available from the corresponding author on request. The PDB coordinates for the POPDC2 homology models are available in supplemental information. We thank the families for their participation and collaboration. N.L. is supported by the 10.13039/501100003246Dutch Research Council (ZonMW VENI and Off-road), The Auxilium & Caritas Tulips Fellowship, and the De Snoo van 't Hoogerhuijs Award. C.R.B., A.V.P., and N.L. acknowledge the support from the 10.13039/100002129Dutch Heart Foundation (CVON 2018-30 PREDICT2 and CVON2014-18 CONCOR-GENES to C.R.B.) and the 10.13039/501100003246Netherlands Organisation for Scientific Research (VICI fellowship, 016.150.610, to C.R.B.). This work was supported in part by the 10.13039/100000002NIH awards R35GM128666 (M.V.A.) and T32GM092714 (F.Z.B.), a Sloan Research Fellowship (M.V.A.), and an American Heart Association Fellowship 23PRE1019634 (L.W.). This study makes use of data generated by the DECIPHER community. A full list of centers that contributed to the generation of the data is available from http://decipher.sanger.ac.uk and via email from decipher@sanger.ac.uk. Funding for the project was provided by the 10.13039/100010269Wellcome Trust. Website: https://www.deciphergenomics.org/. This project has been made possible in part by the Chan Zuckerberg Foundation (2019-202666) to M. Noseda and the British Heart Foundation and Deutsches Zentrum fur Herz-Kreislauf-Forschung (BHF/DZHK: SP/19/1/34461) to M.Noseda. M. Noseda and L. Mach were supported by the Rosetrees Trust Intermediate Project Grant (PGS23/100028) British Heart Foundation Centre of Research Excellence (RE/24/130023) and NIHR Imperial B­­iomedical Research Centre. L. Mach was further supported by British Heart Foundation Clinical Research Training Fellowship (FS/CRTF/23/24444), British Society for Heart Failure Research Fellowship, and Alexander Jansons Myocarditis UK. We thank Drs. Delphine Bichet and Florian Lesage (Universite de Nice Sophia Antipolis, France) for sharing the hTREK-1a plasmid. We thank Drs. Mohamed Hosny and Magdi Yacoub (Magdi Yacoub Foundation, Egypt) for reviewing the phenotype of affected individuals recruited at their center. R.T. is supported by the Canada Research Chairs program and the Philippa and Marvin Carsley Chair in cardiovascular genetics. The PNC “Hub Life Science- Diagnostica Avanzata (HLS-DA), PNC-E3-2022-23683266– CUP: C43C22001630001” is funded by the Italian Minister of Health. The support of Italian Ministry of Education and Research (MUR) “Dipartimenti di Eccellenza Program 2023–2027” - Dept of Pathophysiology and Transplantation, University of Milan to D.R. and G.P.C. is gratefully acknowledged. This work was promoted within the European Reference Network (ERN) for Rare Neuromuscular Diseases. We thank Sara Teichmann for sharing data on scRNA-seq in human hearts. The graphical abstract was created in https://BioRender.com. L. Monserrat is a shareholder in Dilemma Solutions SL. D.A.C. is an employee of and may own stock in GeneDx. H.M.A., V.T., G.S., E.V.I., H.H., D.F.G., A.T.S., and K.S. report employment at deCODE Genetics during the conduct of the study. C.E. reports grants from Abbott Diagnostics and Novo Nordisk outside the submitted work. K.U.K. reports research support from Intermountain Foundation during the conduct of the study. L.N. reports a stock option grant from Culmination Bio. H.B. reports lecture fees from Amgen, MSD, Sanofi Avensis, Bristol Myers Squibb, and Pfizer; grants from Novo Nordic Foundation; and another from Novo Nordic Foundation (shares) outside the submitted work.
PMC12105127
Chemical profiling and anticancer activity of Alnus incana dichloromethane fraction on HeLa cells via cell cycle arrest and apoptosis
Cervical cancer remains a global health challenge with persistently high incidence and mortality rates despite advancements in conventional treatments. The therapeutic potential of natural products has gained attention, particularly for their selective cytotoxicity and ability to modulate cancer pathways. Alnus incana (L.) Moench, a species-rich in bioactive compounds, shows potential as an anticancer agent; however, the cytotoxic effects of its leaves dichloromethane (DCM) extract remain underexplored. This study investigates the DCM fraction’s cytotoxicity on various cancer cell lines, with a primary focus on HeLa cells. The cytotoxic effects of the A. incana DCM fraction were evaluated in a dose-dependent manner using the MTT assay on several cancer cell lines, with particular emphasis on HeLa cells. Flow cytometry was used to assess cell cycle arrest and apoptosis, while RT-qPCR quantified changes in the expression of apoptotic markers (Bax, Bcl-2, and p53). Chemical composition analysis was conducted using gas chromatography-mass spectrometry/flame ionization detection (GC-MS/FID) to identify the major bioactive compounds within the fraction. The DCM fraction exhibited dose-dependent cytotoxicity in HeLa cells, with an IC50 value of 135.6 µg/mL and a selectivity index (SI) of 2.72 relative to normal cells. Flow cytometry analysis revealed G0/G1 cell cycle arrest, significantly hindering progression through the S and G2/M phases. Moreover, there was a significant increase in both early and late apoptotic cell populations, correlating with the upregulation of pro-apoptotic genes (Bax and p53) and the downregulation of the anti-apoptotic gene Bcl-2. The chemical analysis identified 22 compounds in the unsaponifiable fraction, chiefly terpenoids such as phytol (65.74%). The saponifiable fraction presented a balanced composition of saturated (48.69%) and unsaturated (51.29%) fatty acids, with palmitic acid, linolenic acid, and linoleic acid as the predominant compounds. While the DCM fraction’s relatively high IC50 value may limit its utility as a standalone treatment, its ability to induce cell cycle arrest and apoptosis demonstrates its promise as a co-therapeutic agent with conventional anticancer drugs. Further research is essential to elucidate its precise mechanisms of action and to evaluate its efficacy in combination therapies, potentially advancing its role in cervical cancer treatment. Keywords: Alnus incana, Cervical cancer, HeLa cells, Cytotoxicity, Cell cycle arrest, ApoptosisCancer, in its diverse forms, remains a significant global health burden, characterized by high incidence and mortality rates. Cervical cancer, a primary type of gynecological cancer predominantly caused by high-risk human papillomavirus (HPV) infection, is a striking example of this burden. Additional risk factors, including HIV (human deficiency virus), Chlamydia trachomatis, smoking, and hormonal influences (e.g. high parity and long-term oral contraceptive use), amplify its aggressiveness and contribute to its widespread impact on women’s health worldwide . Despite advancements in treatment, which typically combines platinum-based chemotherapy (e.g. cisplatin) with adjuvant radiation [2, 3], cervical cancer remains the fourth most common cancer in women globally, with an estimated 660,000 new cases and 350,000 deaths in 2022 . This ongoing challenge underscores the urgent need for adjunctive therapies that selectively target cancer cells while minimizing toxicity to healthy tissues and potentially improving patient quality of life [4, 5]. Natural products, known for their diverse bioactive compounds and favorable safety profiles, are emerging as promising candidates to fill this therapeutic gap. Many exhibit anticancer properties through modulation of key cancer pathways, including the AKT/mTOR pathway, apoptosis, and cell cycle regulation [6, 7]. Alnus incana (L.) Moench (A. incana), a member of the Betulaceae family, belongs to a genus with a rich history of traditional medicinal use for various ailments, including cancer . This genus, comprising approximately 35 species, is known for producing a wide array of bioactive constituents, including triterpenoids, steroids, flavonoids, phenolics, tannins, diarylheptanoids, and other potentially beneficial compounds [8, 9]. While studies have evaluated cytotoxic activity in A. incana water and methanol extracts [10, 11], the cytotoxic potential of the dichloromethane (DCM) extract remains unexplored. Interestingly, Rashed et al. reported superior cytotoxic activity in the DCM extract of Alnus rugosa, a subspecies closely related to A. incana , suggesting that A. incana DCM extract may also possess similar potent anticancer properties with potential benefits for gynecological cancer treatment. Inducing apoptosis and cell cycle arrest are well-established strategies in cancer treatment, particularly gynecological cancers, as they enable selective targeting of tumor cells while reducing harm to reproductive and other healthy tissues . Apoptosis, a controlled process of cell death, is frequently disrupted in cancer cells, enabling them to evade elimination and continue proliferation. Therapeutic approaches that target apoptotic pathways, such as BCL-2 inhibitors that block survival signals and TRAIL analogues that trigger death receptors, aim to reinstate the apoptotic response in resistant tumor cells. Additionally, drugs targeting key regulators of apoptosis, such as the p53 and ISR signaling pathways, show promise in restoring normal cell death processes and limiting tumor growth . Similarly, inducing cell cycle arrest can prevent cancer cells from dividing, effectively halting tumor progression. By disrupting the cell cycle at critical checkpoints, these therapeutic agents compel cancer cells to pause, leading to eventual cell death. This approach has shown effectiveness across a range of cancers, with several cell cycle-targeting drugs demonstrating clinical success . Recent studies highlight the efficacy of various natural extracts in inducing these mechanisms, offering promising avenues for cancer therapeutic development . Investigating natural products may unveil novel compounds that selectively disrupt these critical pathways, providing solutions to the persistent challenges of toxicity and drug resistance faced by current treatments, and potentially reducing chemotherapy-induced side effects, which could enhance overall patient well-being. Therefore, this study aimed to investigate the cytotoxic effects of the DCM fraction of A. incana leaves on cervical cancer cells, particularly HeLa cells, a widely used model for gynecological cancers. The study examined the fraction’s impact on cell cycle progression, induction of apoptosis, and the expression of key genes (Bax, p53, Bcl-2) related to cell death pathways. Additionally, a comprehensive chemical profiling of the fraction was conducted to identify the bioactive constituents that may contribute to its cytotoxic properties. This research not only enhances understanding of potential adjunctive treatments for cervical cancer but also contributes to the broader field of complementary approaches in gynecological oncology, aiming to improve patient outcomes, quality of life, and overall well-being. The A. incana DCM fraction used in this study was obtained from our previous work . In brief, A. incana (L.) Moench leaves were collected with permission from Al Zoharia Research Garden, Cairo, Egypt, following institutional, national, and international guidelines. Dr. Mamdouh Shokry, a botanist at Al Zoharia Research Garden, validated and authenticated the plant material. A voucher specimen was deposited at the herbarium of the Pharmacognosy Lab, Faculty of Pharmacy (Girls), Al-Azhar University, under the number AR-2016. The leaves were extracted with 100% methanol, followed by liquid-liquid partitioning using solvents of increasing polarity. The DCM fraction, representing the lipoidal portion, was selected for further investigation in this study. Human cancer cell lines, HepG2 (liver), MCF-7 (breast), HCT116 (colon), and HeLa (cervical), along with the non-cancerous HSF (human skin fibroblasts) cell line were obtained from American Type Cell Culture Collection (ATCC) and maintained as adherent monolayers at VACSERA cell culture library (Giza, Egypt). All cells were cultured in RPMI-1640 medium (Lonza GmbH, Cologne, Germany) supplemented with 10% heat-inactivated fetal bovine serum (FBS; BIOWEST, Bradenton, FL, USA), 100 U/mL penicillin (Company), and 2 mg/mL streptomycin (Company). Cells were incubated at 37 °C in a humidified atmosphere with 5% CO2, and regular subculturing was performed under aseptic conditions to maintain exponential cell growth. Cell viability was assessed using the MTT assay as described by . Briefly, this colorimetric method measures the reduction of a yellow tetrazolium salt (MTT) to a purple formazan product by metabolically active cells. Briefly, cells were seeded at a density of 50,000 cells/well in 100 µL of the media in a 96-well plate and incubated for 24 h. Afterward, serial concentrations of the A. incana DCM fraction (400–3.1 µg/mL) and doxorubicin (20–2.5 µg/mL, positive control) were added, followed by a 48-hour incubation. Four wells were used for each condition, and the experiment was repeated independently three times to ensure reproducibility. Doxorubicin was selected as the positive control due to its well-established efficacy and its use as a benchmark in cancer research. Its potent cytotoxic effects make it a standard reference for evaluating the antiproliferative activity of novel compounds. Next, 10 µL of MTT solution was added to each well, and the plate was incubated for 4 h at 37 °C. The formed formazan crystals were dissolved in 100 µL DMSO with agitation, and absorbance was measured at 570 nm using an ELX800 UV universal microplate reader. Cell viability was calculated as a percentage of the control using the formula: (Mean absorbance of treated sample / Mean absorbance of control) × 100. Dose-response curves were generated, and IC50 (half-maximal inhibitory concentration) was determined using Prism software. The selectivity index (SI) was calculated as the ratio of the IC50 value for normal cells to the IC50 value for cancer cells, with SI > 1 indicating selectivity towards cancer cells. Higher SI values suggest greater therapeutic selectivity and potential efficacy . To assess the impact of A. incana DCM fraction on cell cycle progression, propidium iodide (PI) staining followed by flow cytometry was employed as outlined previously . PI (Company) is a fluorescent dye that binds to DNA allowing differentiation of cell cycle phases (G0/G1, S, G2/M) based on their DNA content. Briefly, HeLa cells (2 × 10 cells/mL) were cultured overnight and then treated with the DCM fraction at its IC50 concentration (135.6 µg/mL) for 24 h. After treatment, cells were fixed, stained with a PI/RNase solution, and analyzed using a FACSCalibur Scan flow cytometer (BD Biosciences) to determine cell cycle distribution. To determine the mode of cell death induced by A. incana DCM fraction, the Annexin V-FITC apoptosis detection kit (BioVision Inc., CA, USA, K101-25) was used following the manufacturer’s protocol. This assay detects phosphatidylserine translocation, a hallmark of early apoptosis, by Annexin V-FITC staining. PI, a viability dye excluded by healthy and early apoptotic cells with intact membranes, stains the DNA of late apoptotic and necrotic cells with compromised membranes, allowing for their identification. Briefly, HeLa cells (2 × 10 cells/mL) were cultured for 24 h and treated with the extract at its IC50 concentration (135.6 µg/mL) for an additional 48 h. Following treatment, cells were stained with Annexin V-FITC and PI. Apoptosis was quantified by flow cytometry using a BD FACSCalibur Scan system (BD Biosciences). For accurate quantification of apoptotic populations, a specific gating strategy was applied. Cells were initially gated based on forward scatter (FSC) versus side scatter (SSC) to exclude debris, followed by gating on FSC-Height (FSC-H) versus FSC-Area (FSC-A) to exclude doublets. Apoptotic populations were classified by gating on Annexin V-FITC versus PI as follows: live cells (Q4) were Annexin V-/PI-, early apoptotic cells (Q3) were Annexin V+/PI-, late apoptotic/necrotic cells (Q2) were Annexin V+/PI+, and dead/necrotic cells (Q1) were Annexin V-/PI+. The percentage of cells in each quadrant was calculated to assess the extent of apoptosis and necrosis. Total RNA was isolated from both treated and untreated HeLa cells using the RNeasy Mini Kit (Qiagen, 74104) following the manufacturer’s protocol. To evaluate the effects of A. incana DCM extract (135.6 µg/mL, IC50) on gene expression, complementary DNA (cDNA) was synthesized from 1 µg of RNA using the iScript™ One-Step RT-PCR Kit (Bio-Rad, Hercules, CA). Primers for the target genes (Table 1) were verified with NCBI Primer-Blast. Relative expression levels were normalized to GAPDH as the housekeeping gene and calculated using the 2 method. The PCR conditions included an initial denaturation at 50 °C for 10 min, and 95 °C for 5 min, followed by 30–45 cycles of 95 °C for 10 s and 55–60 °C for 30 s. A melting curve was generated by gradually increasing the temperature from 60 °C to 95 °C after the final cycle. Primer pair sequences used for qRT-PCR analysis F 5’- ATCGCCCTGTGGATGACTGAGT − 3’ R 5’- GCCAGGAGAAATCAAACAGAGGC-3’ F 5’- CCTCAGCATCTTATCCGAGTGG − 3’ R 5’- TGGATGGTGGTACAGTCAGAGC − 3’ F 5’-TCAGGATGCGTCCACCAAGAAG − 3’ R 5’-TGTGTCCACGGCGGCAATCATC − 3’ F 5’- GTCTCCTCTGACTTCAACAGCG − 3’ R 5’- ACCACCCTGTTGCTGTAGCCAA-3’ The A. incana DCM fraction was subjected to comprehensive chemical analysis using GC-MS/FID detection. GC-MS was employed to identify unsaponifiable compounds, while GC-FID was utilized for fatty acids characterization. For GC-MS analysis, the DCM fraction underwent saponification with ethanolic potassium hydroxide, followed by extraction of the unsaponifiable matter with toluene. The extract was injected into a GC-MS system (Agilent 7890B) equipped with a DB-5MS column (30 m x 0.25 mm ID, 0.25 μm film thickness). Helium served as the carrier gas at a flow rate of 3.0 mL/min in splitless mode, with an injection volume of 1 µL. The temperature program was as follows: 40 °C for 1 min, increased by 10 °C/min to 200 °C (held for 1 min), then increased by 20 °C/min to 220 °C (held for 1 min), and finally increased by 30 °C/min to 320 °C (held for 10 min). The injector and detector temperatures were set at 250 °C and 320 °C, respectively. Mass spectra were obtained using electron ionization (70 eV) in the m/z range of 50–600, with compound identification performed by comparing spectra to the Wiley and NIST Mass Spectral Libraries. For fatty acid analysis, the saponifiable matter was derivatized into fatty acid methyl esters (FAMEs) using a 2 M potassium hydroxide solution in methanol. The FAMEs were injected into a GC system (Agilent 7890B) equipped with a Zebron ZB-FAME column (60 m x 0.25 mm ID, 0.25 μm film thickness). Hydrogen was used as the carrier gas at a flow rate of 1.8 mL/min (split 1:50), with an injection volume of 1 µL. The temperature program began at 100 °C for 3 min and increased by 2.5 °C/min to 240 °C (held for 10 min). The injector and flame ionization detector (FID) were set at 250 °C and 285 °C, respectively. Fatty acids were identified by comparing the retention times of FAME peaks with those of known standards. Statistical analysis was performed using SPSS version 24.0. Differences between the means of two independent groups were analyzed using an independent t-test, with p < 0.05 (*) considered statistically significant, indicating meaningful differences in the data. To evaluate the cytotoxic potential and selectivity of the A. incana DCM fraction, MTT assay was performed on four cancer cell lines (HepG2, MCF-7, HCT116, and HeLa) along with a normal cell line (HSF). The DCM fraction exhibited a dose-dependent inhibition of cell proliferation across all tested cancer cell lines, with IC50 values ranging from 135.6 to 237.1 µg/mL (Fig. 1; Table 2). To assess the selectivity, the Selectivity Index (SI), calculated as the IC50 ratio in normal cells to cancer cells, was determined (Table 2). Based on the IC50 and SI values, the HeLa cell line was selected for further investigation due to its favorable response. Furthermore, the IC50 and SI of doxorubicin against HeLa cells were also determined as a positive control (Fig. 1; Table 2). Cytotoxic effects and IC50 values of the A. incana DCM fraction and doxorubicin on various cell lines over 48 h, as determined by the MTT assay. Panels (a–e) show the cytotoxic effects of the A. incana DCM fraction on (a) normal human skin fibroblasts (HSF), (b) cervical cancer cells (HeLa), (c) breast cancer cells (MCF-7), (d) liver cancer cells (HepG2), and (e) colon cancer cells (HCT-116). Panels (f) and (g) illustrate the cytotoxic effects and IC50 values of doxorubicin on (f) HSF and (g) HeLa cells after 48 h IC50 values and selectivity index (SI) of A. incana DCM fraction and doxorubicin against various cell lines To investigate the effect of the A. incana DCM fraction on cell cycle progression, HeLa cells were treated with the IC50 concentration (135.6 µg/mL) of the fraction for 48 h. Cell cycle distribution was assessed by flow cytometry using Propidium Iodide (PI) staining, which binds to DNA and allows differentiation of cell cycle phases (G0/G1, S, G2/M) based on DNA content. As illustrated in Fig. 2, a significant increase (p < 0.05) in the G0/G1 phase population (58.22%) compared to the control (47.42%) was observed, indicating cell cycle arrest at the G0/G1 checkpoint. Concomitantly, significant decreases were observed in both the S phase (29.38%) and G2/M phase (12.4%) populations relative to untreated cells (35.76% and 16.82%, respectively). These results suggest that the DCM fraction of A. incana exerts cytostatic effects on HeLa cells by inducing cell cycle arrest. Flow cytometric analysis of cell cycle distribution in HeLa cells treated with the A. incana DCM fraction. HeLa cells were treated with the IC50 concentration (135.6 µg/mL) of the A. incana DCM fraction for 48 h, and cell cycle phase distribution was analyzed via flow cytometry. Panels: (a) Control (untreated) HeLa cells, (b) HeLa cells treated with the DCM fraction, and (c) A graphical comparison of cell percentages across each cell cycle phase between treated and untreated groups HeLa cells treated with the IC₅₀ concentration (135.6 µg/mL) of the A. incana DCM fraction for 48 h were assessed for apoptosis and necrosis using Annexin V-FITC and PI staining analyzed by flow cytometry. As illustrated in Fig. 3, treatment with the A. incana DCM fraction resulted in a statistically significant increase (p < 0.05) in early and late apoptotic, as well as necrotic cells, compared to the untreated control group. Specifically, the percentage of early apoptotic cells increased from 0.37 to 14.32%, and late apoptotic cells from 0.09 to 4.39%. Necrosis was also notably induced, with the proportion of necrotic cells rising from 1.15 to 2.84%. Flow cytometric analysis of apoptosis and necrosis in HeLa cells treated with the A. incana DCM fraction. Detection of apoptotic and necrotic cell populations was conducted using Annexin V-FITC and PI dual staining after treating HeLa cells with the IC50 concentration (135.6 µg/mL) of the A. incana DCM fraction for 48 h. Quadrant distribution: upper left (necrotic cells), lower left (viable cells), lower right (early apoptotic cells), and upper right (late apoptotic cells). Panels: (a) Control (untreated) HeLa cells, (b) DCM-treated HeLa cells, and (c) Graphical representation comparing cell percentages in each quadrant between treated and untreated cells To investigate the molecular mechanisms underlying the apoptotic effects of the A. incana DCM fraction, reverse transcription-quantitative polymerase chain reaction (RT-qPCR) was employed to analyze the expression of key apoptosis-related genes in HeLa cells. Treatment with the IC50 concentration of the DCM extract significantly upregulated the expression of the pro-apoptotic genes p53 and Bax, with fold change values of 2.60 and 2.96, respectively. In contrast, the expression of the anti-apoptotic gene Bcl-2 was significantly downregulated, with a fold change value of 0.6467 (Fig. 4). Gene expression of (a) Pro-apoptotic genes p53 and Bax, and (b) Anti-apoptotic gene Bcl-2 in HeLa cells treated and untreated with the A. incana DCM fraction The chemical analysis of the Alnus incana leaves’ DCM fraction, conducted using GC-MS for unsaponifiable matter and GC-FID for saponifiable matter, revealed a diverse composition of compounds (Fig. 5). The unsaponifiable fraction was predominantly composed of terpenoids (75.07%), with phytol—a diterpene chlorophyll compound known for its potential therapeutic properties—as the most abundant component (65.74%). Other significant terpenoids included isophytol, squalene, lupeol, and lup-20(29)-en-3-one. Additionally, the fraction contained hydrocarbons, alcohols, ketones, esters, and sterols like γ-sitosterol (Table 3). The saponifiable fraction primarily consisted of fatty acids, nearly equally divided between saturated (48.69%) and unsaturated (51.29%) types. Among the saturated fatty acids, palmitic acid (31.27%) was the most abundant, while linolenic acid (26.98%) and linoleic acid (11.21%) were the major unsaturated fatty acids (Table 4). These findings underscore the substantial presence and potential pharmacological relevance of both terpenoids and fatty acids in the DCM fraction of A. incana leaves. Chromatographic analysis of Alnus incana leaves’ DCM fraction: (a) GC-MS chromatogram for unsaponifiable matter, and (b) GC-FID chromatogram for saponifiable matter (fatty acids) Chemical profile of the unsaponifiable matter in the DCM fraction of A. incana leaves Chemical profile of the saponifiable matter in the DCM fraction of Alnus incana leaves This study investigated the cytotoxic potential of the DCM fraction from the leaves of the medicinal plant A. incana. Given the documented cytotoxic activity of the DCM extract from the related subspecies A. rugosa , we hypothesized that A. incana might exhibit similar effects. Although previous studies confirmed the cytotoxic properties of its methanol extract , the effects of the DCM fraction remain unexamined. To address this gap, the fraction’s cytotoxicity was evaluated against four human cancer cell lines: HepG2, HCT116, HeLa, and MCF-7, representing diverse cancer types. The A. incana DCM fraction demonstrated IC50 values ranging from 135.6 to 273.1 µg/mL, with HeLa cells (a cervical cancer model) showing the highest sensitivity (IC50 = 135.6 µg/mL). Significantly, the DCM fraction exhibited a selectivity index (SI) of 2.72 against HeLa cells, indicating preferential cytotoxicity toward cancer cells over normal cells and highlighting its therapeutic potential. Although the IC50 values for A. incana were higher than those reported for the DCM extract of the related subspecies A. rugosa , this difference may be attributed to variations in the concentration or composition of active compounds in each extract. Despite this, both species showed their most potent cytotoxic effects against HeLa cells, suggesting a potential shared mechanism targeting cervical cancer cells specifically. Additionally, prior studies have reported that the methanol extract of A. incana leaves has a pronounced effect on HeLa cells (IC50 of 68.5 µg/mL) . The selective cytotoxicity observed in both A. incana and A. rugosa DCM extracts, alongside the methanol extract’s activity, indicates the presence of bioactive compounds within these species that may preferentially disrupt HeLa cell function. This selective activity underscores the potential of Alnus species as promising candidates for developing targeted therapies against gynecological cancers, particularly cervical cancer. Given the substantial cytotoxic response observed, further investigation into the active compounds and mechanisms of action of the DCM fraction of A. incana leaves is warranted. To explore the mechanisms underlying this cytotoxicity, we investigated the effects of A. incana DCM fraction on cell cycle progression and apoptosis in HeLa cells. The results revealed a significant accumulation of cells in the G0/G1 phase, indicating effective cell cycle arrest. This is a crucial finding science cell cycle arrest at key checkpoints is a well-established therapeutic strategy for inhibiting cancer cell proliferation. The mammalian cell cycle comprises four distinct phases: G1, S, G2, and M, where RNA and protein synthesis occur in G1 and G2, DNA replication in S, and chromosome segregation in M . By halting cells in the G0/G1 phase, the A. incana DCM fraction appears to block progression to DNA synthesis, which could lead to senescence or cell death. Several plant-derived anticancer agents, such as taxanes and vinca alkaloids, function by disrupting key cell cycle regulators . In addition to inducing cell cycle arrest, the DCM fraction significantly promoted apoptosis, a fundamental mechanism in cancer treatment. Apoptosis, or programmed cell death, is triggered by many natural cytotoxic agents, both clinically approved and experimental, and is essential for eliminating cancer cells . Flow cytometric analysis using Annexin V and propidium iodide staining revealed a marked increase in apoptotic cell death (40.67-fold increase, 18.71% vs. 0.46% in the control) induced by the A. incana DCM fraction. Early apoptosis rose to 14.32% from 0.37%, and late apoptosis increased to 4.39% from 0.09%. Necrosis also increased by 2.5-fold (2.84% vs. 1.15% in the control), indicating a broad cytotoxic effect on HeLa cells. These findings reinforce the potential of A. incana DCM fraction as an effective anticancer agent by simultaneously inducing cell cycle arrest and apoptosis. To elucidate the molecular mechanisms behind these effects, we evaluated the expression of apoptotic regulators, including p53, Bcl-2, and Bax, through RT-PCR. The intrinsic apoptotic pathway is regulated by the balance between pro-apoptotic Bax and anti-apoptotic Bcl-2, where Bcl-2 stabilizes the mitochondrial membrane and prevents cytochrome c release, whereas Bax promotes apoptosis by facilitating cytochrome c release . p53, a critical tumor suppressor, enhances Bax expression, thus its stimulation tipping the balance towards apoptosis . In our study, the A. incana DCM fraction significantly upregulated Bax and p53 while downregulating Bcl-2, resulting in an elevated Bax/Bcl-2 ratio, a key indicator of apoptosis activation. Previous research reported that elevated p53 expression and a higher Bax/Bcl-2 ratio correlate with increased tumor sensitivity to anticancer therapies , further supporting the potential of the A. incana DCM fraction as a therapeutic agent. Given these findings, it is essential to contextualize the anticancer activity of the A. incana DCM fraction by comparing it to established cervical cancer treatments, such as cisplatin and paclitaxel. Cisplatin exerts its cytotoxic effects primarily through DNA cross-linking, causing extensive DNA damage that leads predominantly to G2/M cell cycle arrest and subsequent apoptosis induction mainly via p53-dependent pathways [24, 25]. Despite its efficacy, cisplatin is often limited by severe adverse effects, including nephrotoxicity, neurotoxicity, and drug resistance . Paclitaxel, another cornerstone treatment, acts by stabilizing microtubules, causing prolonged G2/M mitotic arrest and promoting apoptosis through a p53-independent pathway involving cyclin B1/CDC2 activation and Bcl-2 phosphorylation [24, 26]. However, paclitaxel similarly has clinical limitations, including dose-dependent toxicities such as peripheral neuropathy and myelosuppression . In contrast, our findings demonstrate that the A. incana DCM fraction distinctly arrests cells at the G0/G1 phase and promotes apoptosis via a p53-dependent pathway. These unique molecular characteristics suggest a complementary role alongside traditional chemotherapeutics, potentially reducing toxicities and overcoming drug resistance, despite its relatively high IC50. The chemical profile of the A. incana DCM fraction, analyzed via GC-MS/FID detection, revealed a high content of phytol (65.74%) in unsaponifiable matter, a compound known for its broad-spectrum biological activities, particularly in cancer therapy. Phytol, a diterpenoid derived from chlorophyll, has been widely recognized for its anti-inflammatory and anticancer properties, including the modulation of cell cycle and apoptosis pathways . Previous studies have demonstrated its ability to induce apoptosis across various cancer cell lines, including lung, colon, and breast cancers, primarily through the disruption of mitochondrial membrane potential and subsequent activation of the intrinsic apoptotic pathway [28, 29]. Our findings, which show significant G0/G1 cell cycle arrest and apoptosis in HeLa cells, align with these established mechanisms of action, further validating phytol’s role as a potent anticancer agent. In addition to phytol, the saponifiable matter of the DCM fraction contained palmitic acid (31.27%) and linolenic acid (26.98%) as the major components. Both fatty acids are well-documented for their apoptosis-inducing capabilities. Palmitic acid, for instance, has been reported to upregulate pro-apoptotic proteins such as Bax and p53, while downregulating anti-apoptotic Bcl-2, leading to enhanced apoptosis via caspase activation in colorectal and breast cancer models . Similarly, linolenic acid exerts anticancer effects by modulating the Bcl-2 family proteins, increasing pro-apoptotic Bax expression, and inducing endoplasmic reticulum (ER) stress, which contributes to cancer cell death. Linolenic acid has also been shown to inhibit the PI3K/Akt signaling pathway and suppress fatty acid synthase (FASN), an enzyme overexpressed in many cancers . Our study’s findings of increased Bax expression and a heightened Bax/Bcl-2 ratio in HeLa cells are consistent with these mechanisms, suggesting that palmitic and linolenic acids likely contribute to the cytotoxic effects observed. Despite these promising results, the study possesses limitations typical of in vitro investigations. Notably, the biological complexity of tumor microenvironments, such as gene expression variability, signaling heterogeneity, immune interactions, and stromal influences, cannot be adequately mimicked using isolated cell lines like HeLa. Additionally, the relatively high IC50 necessitates exploration of advanced drug delivery systems or combination therapies. Future research should incorporate additional cancer cell lines, normal cell toxicity assessments, patient-derived xenografts (PDX), and innovative delivery strategies, such as nanoparticle encapsulation, to enhance efficacy, reduce toxicity, and facilitate clinical translation. These steps are crucial for validating the preclinical promise and advancing the clinical applicability of the Alnus incana DCM fraction. This study demonstrates that the A. incana DCM fraction, rich in bioactive compounds such as phytol, palmitic acid, and linolenic acid, exerts notable cytotoxic effects on various cancer cell lines, particularly HeLa cells. These effects are mediated through inducing G0/G1 cell cycle arrest and apoptosis via modulation of Bax, Bcl-2, and p53 pathways. Although the relatively high IC50 value (135.6 µg/mL) limits its potential as a standalone therapy, the fraction’s selective cytotoxicity toward cancer cells suggests a potentially favorable safety profile. Its unique ability to activate apoptotic and cell-cycle arrest pathways supports its promise as an adjunctive therapy, particularly in combination with conventional chemotherapeutic agents, which may improve treatment outcomes by reducing associated toxicities or overcoming resistance. Further in-depth mechanistic studies, along with rigorous in vivo evaluations using clinically relevant models such as patient-derived xenografts, are essential to validate these promising findings and fully realize the clinical potential of this natural fraction. Not applicable. American Type Culture Collection B-cell Lymphoma (e.g., BCL-2, a family of apoptosis regulators) Dichloromethane Dimethyl Sulfoxide Deoxyribonucleic Acid Endoplasmic Reticulum Fatty Acid Methyl Esters Fatty Acid Synthase Fetal Bovine Serum Flame Ionization Detector Fluorescein Isothiocyanate Glyceraldehyde-3-phosphate Dehydrogenase (housekeeping gene) Gas Chromatography Human Immunodeficiency Virus Human Papillomavirus Michigan Cancer Foundation (MCF-7 breast cancer cell line) Mass Spectrometry 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (assay reagent) National Center for Biotechnology Information National Institute of Standards and Technology Polymerase Chain Reaction Propidium Iodide Ribonucleic Acid Reverse Transcription (in qRT-PCR) Selectivity Index Statistical Package for the Social Sciences Tumor Necrosis Factor-Related Apoptosis-Inducing Ligand Ultraviolet (UV absorbance measurement) Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB). Data is provided within the manuscript. A. incana (L.) Moench leaves were collected with permission from Al Zoharia Research Garden, Cairo, Egypt, following institutional, national, and international guidelines. All authors consent for the publication. The authors declare no competing interests. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Data is provided within the manuscript.
PMC10287567
An integrated cell atlas of the lung in health and disease
Single-cell technologies have transformed our understanding of human tissues. Yet, studies typically capture only a limited number of donors and disagree on cell type definitions. Integrating many single-cell datasets can address these limitations of individual studies and capture the variability present in the population. Here we present the integrated Human Lung Cell Atlas (HLCA), combining 49 datasets of the human respiratory system into a single atlas spanning over 2.4 million cells from 486 individuals. The HLCA presents a consensus cell type re-annotation with matching marker genes, including annotations of rare and previously undescribed cell types. Leveraging the number and diversity of individuals in the HLCA, we identify gene modules that are associated with demographic covariates such as age, sex and body mass index, as well as gene modules changing expression along the proximal-to-distal axis of the bronchial tree. Mapping new data to the HLCA enables rapid data annotation and interpretation. Using the HLCA as a reference for the study of disease, we identify shared cell states across multiple lung diseases, including SPP1 profibrotic monocyte-derived macrophages in COVID-19, pulmonary fibrosis and lung carcinoma. Overall, the HLCA serves as an example for the development and use of large-scale, cross-dataset organ atlases within the Human Cell Atlas.Rapid technological improvements over the past decade have allowed single-cell datasets to grow both in size and number. This has led consortia, such as the Human Cell Atlas, to pursue the generation of large-scale reference atlases of human organs. To advance our understanding of health and disease, such atlases must capture variation between individuals that is expected to impact the molecular phenotypes of the cells in a tissue. Whereas the generation of atlases at this scale by single research groups is currently not feasible, integrating datasets generated by the research community at large will enable capture of the diversity of the cellular landscape across individuals. Several foundational studies have started to map the cellular landscape of the healthy human lung. These studies each have a specific bias due to their choice of experimental protocol and technologies, and are therefore not tailored to serve as a universal reference. The studies moreover include only a limited number of samples and individuals, thus lacking the scale and diversity to capture the full cellular heterogeneity present within the lung as well as across individuals. Integrated single-cell atlases provide novel insights not obtained in individual studies. Recent reference atlases have led to the discovery of unknown cell types, the identification of marker genes that are reproducible across studies, the comparison of animal and in vitro models with human healthy and diseased tissue and patient stratification for disease endotypes. However, many currently available integrated atlases are limited in the number of human samples, datasets or cell types per organ, as well as donor metadata (for example, age, body mass index (BMI) and smoking status), or focus mainly on a specific disease. These limitations constrain the potential of atlases to serve as a reference, as they fail to represent and catalog the diversity of cellular phenotypes within the healthy organ and across individuals. Moreover, when integrating data from different sources, it is paramount to correctly separate technical biases from biologically relevant information. Yet, the majority of existing atlases have not assessed the quality of their data integration. Nonetheless, successful integration of the available datasets into a single tissue atlas is a critical step in achieving the goals of the Human Cell Atlas. In this resource, we present an integrated single-cell transcriptomic atlas of the human respiratory system, including the upper and lower airways, from published and newly generated datasets (Fig. 1). The Human Lung Cell Atlas (HLCA) comprises data from 486 donors and 49 datasets, including 2.4 million cells, which we re-annotated to generate a consensus cell type reference. The HLCA expands our understanding of the healthy lung and its changes in disease and can be used as a reference for analyzing future lung data. Together, we provide a roadmap for building and using comprehensive, interpretable and up-to-date organ- and population-scale cell atlases.Fig. 1HLCA study overview.Harmonized cell annotations, raw count data, harmonized patient and sample metadata and sample anatomical locations encoded into a CCF were collected and generated as input for the HLCA core (left). After integration of the core datasets, the atlas was extended by mapping 35 additional datasets, including disease samples, to the HLCA core, bringing the total number of cells in the extended HLCA to 2.4 million (M). The HLCA core provides detailed consensus cell annotations with matched consensus cell type markers (top right), gene modules associated with technical, demographic and anatomical covariates in various cell types (middle right), GWAS-based association of lung conditions with cell types (middle right) and a reference projection model to annotate new data (middle right) and discover previously undescribed cell types, transitional cell states and disease-associated cell states (right, bottom). Harmonized cell annotations, raw count data, harmonized patient and sample metadata and sample anatomical locations encoded into a CCF were collected and generated as input for the HLCA core (left). After integration of the core datasets, the atlas was extended by mapping 35 additional datasets, including disease samples, to the HLCA core, bringing the total number of cells in the extended HLCA to 2.4 million (M). The HLCA core provides detailed consensus cell annotations with matched consensus cell type markers (top right), gene modules associated with technical, demographic and anatomical covariates in various cell types (middle right), GWAS-based association of lung conditions with cell types (middle right) and a reference projection model to annotate new data (middle right) and discover previously undescribed cell types, transitional cell states and disease-associated cell states (right, bottom). To build the HLCA, we collected single-cell RNA sequencing (scRNA-seq) data and detailed, harmonized technical, biological and demographic metadata from 14 datasets (11 published and three unpublished). These datasets include samples from 107 individuals, with diversity in age, sex, ethnicity (harmonized as detailed in Methods), BMI and smoking status (Fig. 2a). Cells were obtained from 166 tissue samples using a variety of tissue donors, sampling methods, experimental protocols and sequencing platforms (Supplementary Tables 1 and 2). Anatomical locations of the samples were projected onto a one-dimensional (1D) common coordinate framework (CCF), representing the proximal (0) to distal (1) axis of the respiratory system, to standardize the anatomical location of origin (Fig. 2a and Supplementary Tables 2 and 3).Fig. 2Composition and construction of the HLCA core.a, Donor and sample composition in the HLCA core for demographic and anatomical variables. Donors/samples without annotation are shown as not available (NA; gray bars) for each variable. For the anatomical region CCF score, 0 represents the most proximal part of the lung and airways (nose) and 1 represents the most distal (distal parenchyma). Donors show diversity in ethnicity (harmonized metadata proportions: 65% European, 14% African, 2% admixed American, 2% mixed, 2% Asian, 0.4% Pacific Islander and 14% unannotated; see Methods), smoking status (52% never, 16% former, 15% active and 17% NA), sex (60% male and 40% female), age (ranging from 10–76 years) and BMI (20–49; 30% NA). b, Overview of the HLCA core cell type composition for the first three levels of cell annotation, based on harmonized original labels. In the cell type hierarchy, the lowest level (1) consists of the coarsest possible annotations (that is, epithelial (48% of cells), immune (38%), endothelial (9%) and stromal (4%)). Higher levels (2–5) recursively break up coarser-level labels into finer ones (Methods). Cells were set to ‘none’ if no cell type label was available at the level. Cell labels making up less than 0.02% of all cells are not shown. Overall, 94, 66 and 7% of cells were annotated at levels 3, 4 and 5, respectively. c, Cell type composition per sample, based on level 2 labels. Samples are ordered by anatomical region CCF score. d, Summary of the dataset integration benchmarking results. Batch correction score and biological conservation score each show the mean across metrics of that type, as shown in Supplementary Fig. 1, with metric scores scaled to range from 0 to 1. Both Scanorama and fastMNN were benchmarked on two distinct outputs: the integrated gene expression matrix and integrated embedding (see output). The methods are ordered by overall score. For each method, the results are shown only for their best-performing data preprocessing. Methods marked with an asterisk use coarse cell type labels as input. Preprocessing is specified under HVG (that is, whether or not genes were subsetted to the 2,000 (HVG) or 6,000 (FULL) most highly variable genes before integration) and scaling (whether genes were left unscaled or scaled to have a mean of 0 and a standard deviation of 1 across all cells). EC, endothelial cell; NK, natural killer; Bioconserv., conservation of biological signal. a, Donor and sample composition in the HLCA core for demographic and anatomical variables. Donors/samples without annotation are shown as not available (NA; gray bars) for each variable. For the anatomical region CCF score, 0 represents the most proximal part of the lung and airways (nose) and 1 represents the most distal (distal parenchyma). Donors show diversity in ethnicity (harmonized metadata proportions: 65% European, 14% African, 2% admixed American, 2% mixed, 2% Asian, 0.4% Pacific Islander and 14% unannotated; see Methods), smoking status (52% never, 16% former, 15% active and 17% NA), sex (60% male and 40% female), age (ranging from 10–76 years) and BMI (20–49; 30% NA). b, Overview of the HLCA core cell type composition for the first three levels of cell annotation, based on harmonized original labels. In the cell type hierarchy, the lowest level (1) consists of the coarsest possible annotations (that is, epithelial (48% of cells), immune (38%), endothelial (9%) and stromal (4%)). Higher levels (2–5) recursively break up coarser-level labels into finer ones (Methods). Cells were set to ‘none’ if no cell type label was available at the level. Cell labels making up less than 0.02% of all cells are not shown. Overall, 94, 66 and 7% of cells were annotated at levels 3, 4 and 5, respectively. c, Cell type composition per sample, based on level 2 labels. Samples are ordered by anatomical region CCF score. d, Summary of the dataset integration benchmarking results. Batch correction score and biological conservation score each show the mean across metrics of that type, as shown in Supplementary Fig. 1, with metric scores scaled to range from 0 to 1. Both Scanorama and fastMNN were benchmarked on two distinct outputs: the integrated gene expression matrix and integrated embedding (see output). The methods are ordered by overall score. For each method, the results are shown only for their best-performing data preprocessing. Methods marked with an asterisk use coarse cell type labels as input. Preprocessing is specified under HVG (that is, whether or not genes were subsetted to the 2,000 (HVG) or 6,000 (FULL) most highly variable genes before integration) and scaling (whether genes were left unscaled or scaled to have a mean of 0 and a standard deviation of 1 across all cells). EC, endothelial cell; NK, natural killer; Bioconserv., conservation of biological signal. Consensus definitions of cell types based on single-cell transcriptomic data across studies—particularly of transitional cell states—are lacking. To enable supervised data integration and downstream integrated analysis, we harmonized cell type nomenclature by building a five-level hierarchical cell identity reference framework (Methods, Supplementary Table 4 and Fig. 2b). We then unified cell type labeling across datasets by mapping the collected cell identity labels for every dataset as provided by the data generator to the hierarchical reference framework, showing varying cell type proportions per sample (Fig. 2c). To optimally remove dataset-specific batch effects, we evaluated 12 different data integration methods on 12 datasets (Fig. 2d and Supplementary Fig. 1) using our previously established benchmarking pipeline. We used the top-performing integration method, scANVI, to create an integrated embedding of all 584,444 cells of 107 individuals from the collected datasets: the HLCA core (Fig. 3a).Fig. 3The HLCA core conserves detailed biology and enables consensus-driven annotation.a, A UMAP of the integrated HLCA, colored by level 1 annotation. b, Cluster label disagreement (label entropy) of Leiden 3 clusters of the HLCA. The HLCA was split into three parts (immune, epithelial and endothelial/stromal) for ease of visualization. Cells from every cluster are colored by label entropy. Clusters with less than 20% of cells annotated at level 3 are colored gray. c, Cell type label composition of the immune cluster with the most label disagreement (left), with original labels (middle left) and matching manual re-annotations (middle right). A zoom-in on the UMAP from b shows the final re-annotations (right). d, UMAPs of the immune, epithelial and endothelial/stromal parts of the HLCA core with cell annotations from the expert manual re-annotation. e, Percentage of cells originally labeled correctly, mislabeled or underlabeled (that is, only labeled at a coarser level) compared with final manual re-annotations. The percentages were calculated per manual annotation, as well as across all cells (right bar). f, UMAP of HLCA clusters annotated as rare epithelial cell types (that is, ionocytes, neuroendocrine cells and tuft cells). Final annotations, original labels and the study of origin are shown (top), as well as the expression of ionocyte marker FOXI1, tuft cell marker LRMP and neuroendocrine marker CALCA (bottom). g, Log-normalized expression of the migratory dendritic cell marker CCR7 in cells identified during re-annotation as migratory dendritic cells, versus other dendritic cells. AT, alveolar type; DC, dendritic cell; FB, fibroblast; Mph, macrophage; MT, metallothionein; SM, smooth muscle; SMG, submucosal gland; TB, terminal bronchiole. a, A UMAP of the integrated HLCA, colored by level 1 annotation. b, Cluster label disagreement (label entropy) of Leiden 3 clusters of the HLCA. The HLCA was split into three parts (immune, epithelial and endothelial/stromal) for ease of visualization. Cells from every cluster are colored by label entropy. Clusters with less than 20% of cells annotated at level 3 are colored gray. c, Cell type label composition of the immune cluster with the most label disagreement (left), with original labels (middle left) and matching manual re-annotations (middle right). A zoom-in on the UMAP from b shows the final re-annotations (right). d, UMAPs of the immune, epithelial and endothelial/stromal parts of the HLCA core with cell annotations from the expert manual re-annotation. e, Percentage of cells originally labeled correctly, mislabeled or underlabeled (that is, only labeled at a coarser level) compared with final manual re-annotations. The percentages were calculated per manual annotation, as well as across all cells (right bar). f, UMAP of HLCA clusters annotated as rare epithelial cell types (that is, ionocytes, neuroendocrine cells and tuft cells). Final annotations, original labels and the study of origin are shown (top), as well as the expression of ionocyte marker FOXI1, tuft cell marker LRMP and neuroendocrine marker CALCA (bottom). g, Log-normalized expression of the migratory dendritic cell marker CCR7 in cells identified during re-annotation as migratory dendritic cells, versus other dendritic cells. AT, alveolar type; DC, dendritic cell; FB, fibroblast; Mph, macrophage; MT, metallothionein; SM, smooth muscle; SMG, submucosal gland; TB, terminal bronchiole. A large-scale integrated atlas provides the unique opportunity to systematically investigate the consensus in cell type labeling across datasets. To identify areas of consensus and disagreement, we iteratively clustered the HLCA core and investigated donor diversity and cell type label agreement in these clusters using entropy scores (see Methods). Most clusters contained cells from many donors (Extended Data Fig. 1a). Clusters with low donor diversity (n = 14) were largely immune cell clusters (n = 13), representing donor- or donor group-specific phenotypes. Similarly, a high diversity of (contradictory) cell type labels (high label entropy) can identify both annotation disagreements between studies and clusters of doublets (Methods). Most clusters (61 out of 94) showed low label entropy, suggesting overall agreement of coarse cell type labels across datasets (Fig. 3b). The remaining 33 clusters exhibited high label entropy, highlighting cellular phenotypes that were differently labeled across datasets (Fig. 3b). For example, the immune cluster with the highest label entropy contained many cells that were originally mislabeled as monocytes and macrophages but were actually type 2 dendritic cells (Fig. 3c and Extended Data Fig. 1b). Thus, populations with high label entropy identify mislabeled cell types, indicating the need for consensus re-annotation of the integrated atlas. As a first step to achieve such a consensus on the diversity of cell types present in the HLCA core, we performed a full re-annotation of the integrated data on the basis of the original annotations and six expert opinions (consensus annotation; Methods and Fig. 3d). Each of the 61 annotated cell types (Supplementary Table 5) was detected in at least four datasets out of 14, often in specific parts of the respiratory system, and different cell types showed varying fractions of proliferating (MKI67) cells (Extended Data Fig. 2a–c). While our consensus cell type annotations partly correspond to original labels (41% of cells), there were also refinements (28%) and substantial re-annotations (31%; Fig. 3e and Supplementary Fig. 2). To robustly characterize the cell types, we established a universal set of marker genes that generalizes across individuals and studies (Methods, Extended Data Fig. 3 and Supplementary Table 6). The fully re-annotated HLCA core thus combines data from a diverse set of studies to provide a carefully curated reference for cell type annotations and marker genes in healthy lung tissue. Rare cell types, such as ionocytes, tuft cells, neuroendocrine cells and specific immune cell subsets, are often difficult to identify in individual datasets. Yet, combining datasets in the HLCA core provides better power for identifying these rare cell types. Ionocytes, tuft and neuroendocrine cells make up only 0.08, 0.01 and 0.02% of the cells in the HLCA core according to the original labels, and were originally identified in only seven, two and four datasets out of 14, respectively. Despite their low abundance, these cells formed three separate clusters of the HLCA core (Fig. 3f). Our re-annotation increases the number of datasets in which these cells are detected up to threefold and identifies both cells falsely annotated as monocytes, tuft cells or neuroendocrine cells, as well as originally undetected rare cells (Fig. 3f and Supplementary Fig. 3a). Importantly, other integration methods tested during our benchmarking, such as Harmony and Seurat’s RPCA, failed to separate these rare cells into distinct clusters (Supplementary Fig. 3b). We were further able to detect six cell identities that were not previously found in the human lung or were only recently described in individual studies. These cell types include migratory dendritic cells (n = 312 cells, expressing CCR7, LAD1 and COL19), hematopoietic stem cells (n = 60, expressing SPINK2, STMN, PRSS57 and CD34), highly proliferative hillock-like epithelial cells not previously reported in adult human lung (n = 4,600, expressing KRT6A, KRT13 and KRT14), the recently described alveolar type 0 cells (n = 1,440, expressing STFPB, SCGB3A2, SFTPC and SCGB3A1) and the closely related preterminal bronchiole secretory cells (n = 4,393, expressing SFTPB, SCGB3A2, SFTPC and SCGB3A1, together with alveolar type 0 cells called transitional club-AT2 cells) and a subset of smooth muscle cells (n = 335) that to our knowledge have not previously been described (Fig. 3d,g and Extended Data Fig. 4a–f). These smooth muscle cells, predominantly found in the airways, express canonical smooth muscle markers (CNN1 and MYH11) and also uniquely and consistently express FAM83D across datasets (Extended Data Fig. 4e,f). The HLCA core thus enables improved detection and identification of rare cell types, as well as the discovery of unknown cell types. Demographic and other metadata covariates affect cellular transcriptional phenotypes. Better insight into the impact of these covariates (for example, sex, BMI and smoking) on cell type gene expression can shed light on the contribution of these factors to progression from healthy to diseased states. In addition, technical covariates such as ribosomal and mitochondrial genes exhibit batch-specific variation in expression (Methods and Supplementary Table 7). The diversity in demographics (for example, smoking status, age, harmonized ethnicity and BMI) and experimental protocols represented in the HLCA core enables us to explore the contribution of each technical or biological covariate to cell type-specific gene expression variation (Methods and Supplementary Fig. 4). For many cell types, anatomical location is the biological variable explaining most of the variance between samples (Fig. 4a). Furthermore, sex is most associated with transcriptomic variation in lymphatic endothelial cells, whereas BMI is most associated with variation in B and T cells, harmonized ethnicity in transitional club-AT2 cells and smoking status in innate lymphoid/natural killer cells. Furthermore, for several cell types (for example, mast, AT1 and smooth muscle cells), the tissue dissociation protocol explains most of the variance of all technical as well as biological covariates recorded. These associations provide a systematic overview of the effects of biological and technical factors on the transcriptional state of lung cell types.Fig. 4Demographic and technical variables driving interindividual variation.a, Fraction of total inter-sample variance in the HLCA core integrated embedding that correlates with specific covariates. Covariates are split into technical (left) and biological covariates (right). Cell types are ordered by the number of samples in which they were detected. Only cell types present in at least 40 samples are shown. Tissue sampling method represents the way a sample was obtained (for example, surgical resection or nasal brush). Donor status represents the state of the donor at the moment of sample collection (for example, organ donor, diseased alive or healthy alive). The heatmap is masked gray where fewer than 40 samples were annotated for a specific covariate or where only one value was observed for all samples for that cell type. b, Selection of gene sets that are significantly associated with anatomical location CCF score, in different airway epithelial cell types. All gene set names are Gene Ontology biological process (GO: BP) terms. Sets upregulated toward distal lungs are shown in green, whereas sets downregulated are shown in blue. The full name of the term marked by an asterisk is ‘Antigen processing and presentation of exogenous peptide antigen via MHC-I’. c, Cell type proportions per sample, along the proximal-to-distal axis of the respiratory system. The lowest and highest CCF scores shown (0.36 and 0.97) represent the most proximal and most distal sampled parts of the respiratory system, respectively (trachea and parenchyma), excluding the upper airways. The dots are colored by the tissue dissociation protocol and tissue sampling method used for each sample. The boxes show the median and interquartile range of the proportions. Samples with proportions more than 1.5 times the interquartile range away from the high and low quartile are considered outliers. Whiskers extend to the furthest nonoutlier point. n = 23, 19, 9 and 90 for CCF scores 0.36, 0.72, 0.81 and 0.97, respectively. d, Selection of gene sets significantly up- (green) or downregulated (blue) with increasing BMI, in four different cell types. For b and d, P values were calculated using correlation-adjusted mean-rank gene set tests (Methods) and false discovery rate corrected using the Benjamini–Hochberg procedure. IL-1, interleukin-1; MHC-I, major histocompatibility complex class I; TNF, tumor necrosis factor. a, Fraction of total inter-sample variance in the HLCA core integrated embedding that correlates with specific covariates. Covariates are split into technical (left) and biological covariates (right). Cell types are ordered by the number of samples in which they were detected. Only cell types present in at least 40 samples are shown. Tissue sampling method represents the way a sample was obtained (for example, surgical resection or nasal brush). Donor status represents the state of the donor at the moment of sample collection (for example, organ donor, diseased alive or healthy alive). The heatmap is masked gray where fewer than 40 samples were annotated for a specific covariate or where only one value was observed for all samples for that cell type. b, Selection of gene sets that are significantly associated with anatomical location CCF score, in different airway epithelial cell types. All gene set names are Gene Ontology biological process (GO: BP) terms. Sets upregulated toward distal lungs are shown in green, whereas sets downregulated are shown in blue. The full name of the term marked by an asterisk is ‘Antigen processing and presentation of exogenous peptide antigen via MHC-I’. c, Cell type proportions per sample, along the proximal-to-distal axis of the respiratory system. The lowest and highest CCF scores shown (0.36 and 0.97) represent the most proximal and most distal sampled parts of the respiratory system, respectively (trachea and parenchyma), excluding the upper airways. The dots are colored by the tissue dissociation protocol and tissue sampling method used for each sample. The boxes show the median and interquartile range of the proportions. Samples with proportions more than 1.5 times the interquartile range away from the high and low quartile are considered outliers. Whiskers extend to the furthest nonoutlier point. n = 23, 19, 9 and 90 for CCF scores 0.36, 0.72, 0.81 and 0.97, respectively. d, Selection of gene sets significantly up- (green) or downregulated (blue) with increasing BMI, in four different cell types. For b and d, P values were calculated using correlation-adjusted mean-rank gene set tests (Methods) and false discovery rate corrected using the Benjamini–Hochberg procedure. IL-1, interleukin-1; MHC-I, major histocompatibility complex class I; TNF, tumor necrosis factor. To better characterize how biological variables affect cellular phenotypes, we modeled their cell type-specific effects on the transcriptome at the gene level (Methods). Sex-related differences in lymphatic endothelial cells are dominated by differential expression of genes located on the X and Y chromosomes, but also include a decrease in IFNAR1 in females (Supplementary Table 8), which may be linked to differential interferon responses between the biological sexes. We furthermore found cell type-specific programs that change with proximal (low CCF score) to distal (high CCF score) location along the respiratory tract (Supplementary Tables 8 and 9). For instance, oxidative phosphorylation (including cytochrome c oxidase genes such as COX7A1), antigen presentation by major histocompatibility complex class I molecules (including proteasome and protease subunit genes such as PSMD14 and PSMB4), signaling by interleukin-1 and tumor necrosis factor α, as well as planar cell polarity, were downregulated toward more distal locations in secretory, multiciliated and basal cells (Fig. 4b). Some gene programs were specific for a subset of airway epithelial cell types (for example, cornification and keratinization, which were programs that were downregulated in distal multiciliated and secretory cells; including genes such as KRT8 and KRT19). The changes in airway epithelial cell states toward the terminal airways are further illustrated by increased expression of developmental pathway genes such as NKX2-1, NFIB, GATA6, BMP4 and SOX9 in multiciliated cells along the proximal-to-distal axis (Fig. 4b), whereas basal cells decrease in number (Fig. 4c). Similarly, several cell types display transcriptomic changes in donors with increasing BMI (Fig. 4d and Supplementary Tables 8 and 9). AT2 cells, secretory cells and alveolar macrophages exhibit downregulation of a range of biological processes (Supplementary Fig. 5), including cellular respiration, differentiation and synthesis of peptides and other molecules. In secretory cells, a downregulation of the insulin response pathway is also associated with higher BMI, in line with the insulin resistance observed in donors with obesity. In alveolar macrophages, inflammatory responses involving JAK/STAT signaling (previously associated with obesity-induced chronic systemic inflammation) are associated with higher BMI. In contrast, in plasma cells, high BMI is associated with downregulation of gene sets associated with immune response and upregulation of gene sets associated with cellular respiration, the cell cycle and DNA repair. This is consistent with obesity being a known risk factor for multiple myeloma—a plasma cell malignancy. Thus, the HLCA enables a detailed understanding of the effects of anatomical and demographic covariates on the cellular landscape of the lung and their relation to disease. Biological and technical factors can also affect cell type proportions. Indeed, all cell types show changes in abundance as a function of anatomical location (Fig. 4c and Extended Data Fig. 5). For example, ionocytes are present at comparable proportions in the airway epithelium, from the larger lower airways (CCF score = 0.36) down to the distal lobular airways (CCF score = 0.81), while being largely absent in the lung parenchyma (CCF score = 0.97). In contrast, neuroendocrine cells are predominantly observed in the larger lower airways but are absent from more distal parts of the bronchial tree (Fig. 4c). In some cases, these proportions are highly dependent on the tissue sampling method and the dissociation protocol used (for example, for smooth muscle FAM83D cells; Extended Data Fig. 5). These observations shed light on the effects of biological and technical factors on the abundance of cell types in different parts of the lung and can help guide important choices in study design. The HLCA core contains an unprecedented diversity of donors, sampling protocols and cell identities, and can serve as a transcriptomic reference for lung research. New datasets can be mapped to this reference to substantially speed up data analysis by transferring consensus cell identity annotations to the new data. We tested this on a recently released multimodal lung dataset (Methods, Fig. 6a and Extended Data Fig. 6). Overall, the transferred labels were correct in the majority of cases, with 68% of the cells correctly labeled, 14% of labels incorrectly labeled and 18% set to unknown due to highly uncertain labeling (Fig. 5b and Methods). Uncertain labels were observed specifically in continuous transitions from one cell type to another and among cellular identities not present in the HLCA core, including rare cell identities (erythrocytes (n = 328), chondrocytes (n = 42), myelinating Schwann cells (n = 7), nonmyelinating Schwann cells (n = 29) and nerve-associated fibroblasts (n = 66); Fig. 5b and Extended Data Fig. 6d). Taken together, these results show that the HLCA core can be used for highly detailed annotation of new datasets, while allowing for the identification of unknown cell types in these datasets based on label transfer uncertainty.Fig. 5The HLCA core serves as a reference for label transfer and data contextualization.a, UMAP of the jointly embedded HLCA core (gray) and the projected healthy lung dataset (colored by label transfer uncertainty). HLCA cell types surrounding regions of high uncertainty are labeled. b, Percentage of cells from the newly mapped healthy lung dataset that are annotated either correctly or incorrectly by label transfer annotation or annotated as unknown, split by original cell type label (number of cells in parentheses). Cell type labels not present in the HLCA are boxed. c, Top, percentage of cells derived from tumor tissue, per endothelial cell cluster from the joint HLCA core and lung cancer data embedding. Only clusters with at least ten tumor cells are shown. Clusters are named based on the dominant HLCA core cell type annotation in the cluster. Middle, box plot showing the expression of EDNRB in endothelial cell clusters, split by tissue source. Bottom, as in the middle plot but for the expression of ACKR1. Numbers of cells per group were as follows: 6,574 (endothelial cell aerocyte capillary), 7,379 (endothelial cell arterial (I)), 10,906 (endothelial cell general capillary (I)), 3,440 (endothelial cell general capillary (II)), 2,859 (endothelial cell general capillary (III)), 6,318 (endothelial cell venous pulmonary) and 7,161 (endothelial cell venous systemic). d, Association of HLCA cell types with four different lung phenotypes based on previously performed GWASs. The horizontal dashed lines indicate a significance threshold of α = 0.05. P values were calculated using linkage disequilibrium score regression (Methods) and multiple testing corrected with the Benjamini–Hochberg procedure. e, Cell type proportions in lung bulk expression samples as estimated from HLCA-based cell type deconvolution, comparing controls (n = 281) versus donors with severe COPD (GOLD stage 3/4; n = 83). f, UMAP of fibroblast-dominated clusters from the jointly embedded HLCA core and mapped healthy lung dataset, colored by spatial cluster, with cells outside of the indicated clusters colored in gray. For all boxplots, the boxes show the median and interquartile range. Data points more than 1.5 times the interquartile range outside the low and high quartile are considered outliers. In c, these are not shown (see Supplementary Fig. 6 for full results), whereas in e, they are shown. Whiskers extend to the furthest nonoutlier point. corr., corrected; FVC, forced vital capacity; MAIT cells, mucosal-associated invariant T cells; NKT cells, natural killer T cells. a, UMAP of the jointly embedded HLCA core (gray) and the projected healthy lung dataset (colored by label transfer uncertainty). HLCA cell types surrounding regions of high uncertainty are labeled. b, Percentage of cells from the newly mapped healthy lung dataset that are annotated either correctly or incorrectly by label transfer annotation or annotated as unknown, split by original cell type label (number of cells in parentheses). Cell type labels not present in the HLCA are boxed. c, Top, percentage of cells derived from tumor tissue, per endothelial cell cluster from the joint HLCA core and lung cancer data embedding. Only clusters with at least ten tumor cells are shown. Clusters are named based on the dominant HLCA core cell type annotation in the cluster. Middle, box plot showing the expression of EDNRB in endothelial cell clusters, split by tissue source. Bottom, as in the middle plot but for the expression of ACKR1. Numbers of cells per group were as follows: 6,574 (endothelial cell aerocyte capillary), 7,379 (endothelial cell arterial (I)), 10,906 (endothelial cell general capillary (I)), 3,440 (endothelial cell general capillary (II)), 2,859 (endothelial cell general capillary (III)), 6,318 (endothelial cell venous pulmonary) and 7,161 (endothelial cell venous systemic). d, Association of HLCA cell types with four different lung phenotypes based on previously performed GWASs. The horizontal dashed lines indicate a significance threshold of α = 0.05. P values were calculated using linkage disequilibrium score regression (Methods) and multiple testing corrected with the Benjamini–Hochberg procedure. e, Cell type proportions in lung bulk expression samples as estimated from HLCA-based cell type deconvolution, comparing controls (n = 281) versus donors with severe COPD (GOLD stage 3/4; n = 83). f, UMAP of fibroblast-dominated clusters from the jointly embedded HLCA core and mapped healthy lung dataset, colored by spatial cluster, with cells outside of the indicated clusters colored in gray. For all boxplots, the boxes show the median and interquartile range. Data points more than 1.5 times the interquartile range outside the low and high quartile are considered outliers. In c, these are not shown (see Supplementary Fig. 6 for full results), whereas in e, they are shown. Whiskers extend to the furthest nonoutlier point. corr., corrected; FVC, forced vital capacity; MAIT cells, mucosal-associated invariant T cells; NKT cells, natural killer T cells. Single-cell studies of disease rely on adequate, matching control samples to allow correct identification of disease-specific changes. To demonstrate the ability of the HLCA core to serve as a comprehensive healthy control and contextualize disease data, we mapped scRNA-seq data from lung cancer samples to the HLCA core (Methods and Extended Data Fig. 7a–c). Using HLCA label transfer, we correctly identified cell states missing from the HLCA core as unknown (cancer cells and erythroblasts). The remaining cells were annotated correctly in 77%, incorrectly in 1% and as unknown in 22% of cases (Extended Data Fig. 7d–g). A finding of the original study was the separation of endothelial cells into tumor-associated and normal cells. Clustering of the projected dataset with the HLCA reference showed that cells expressing the suggested tumor-associated marker ACKR1 were also abundant in healthy tissue from the HLCA core, specifically in venous endothelial cells (both pulmonary and systemic, Fig. 5c and Supplementary Fig. 6a–c). This suggests that ACKR1 is a general marker of venous endothelial cells rather than a tumor-specific endothelial cell marker. Similarly, the reported normal endothelial cell marker EDNRB characterizes aerocyte capillary endothelial cells, both in tumor and in healthy tissue (Fig. 5c and Supplementary Fig. 6d). As endothelial cell numbers in the original study were low, correctly identifying and distinguishing these cell types without a larger healthy reference is challenging. Thus, by serving as a comprehensive healthy control, the HLCA prevents misinterpretation of limitations in sampling and experimental design as meaningful differences between healthy and diseased tissue. In addition, the HLCA can provide context to the results of large-scale genetic studies of disease. Genome-wide association studies (GWASs) link disease with specific genomic variants that may confer an increased risk of disease. Previous studies have linked such variants to cell type-specific mechanistic hypotheses, which are often lacking in the initial association study. Yet, these studies fail to include all known lung cell types in their cell type reference. To demonstrate the value of the HLCA core in contextualizing genetic data, we mapped association results from four GWASs of lung function or disease to the HLCA core cell types, by testing significant enrichment of both weakly and strongly disease-associated variants in regions of genes that characterize each cell type (Fig. 5d, Supplementary Fig. 7 and Methods). We show that genomic variants linked to lung function (forced vital capacity) are associated with smooth muscle (adjusted P value (Padj) = 0.07), alveolar fibroblasts (Padj = 0.07), peribronchial fibroblasts (Padj = 0.07) and myofibroblasts (Padj = 0.07), suggesting that these fibroblast subtypes play a causative role in inherited differences in lung function. We further find a significant association of lung T cells with asthma-associated single-nucleotide polymorphisms (SNPs) (Padj = 0.005). Lung adenocarcinoma-associated variants trend towards AT2 cells (Padj = 0.18) and myofibroblasts are significantly associated with chronic obstructive pulmonary disease (COPD) GWAS SNPs (Padj = 0.04). Thus, by linking genetic predispositions to lung cell types, the HLCA core serves as a valuable resource with which to improve our understanding of lung function and disease. Finally, the HLCA can be used as a reference for cell type deconvolution of bulk RNA expression samples, which have been shown to reflect cell type proportions more accurately than scRNA-seq datasets. Inferring cell type proportions from bulk RNA samples from nasal brushings and bronchial biopsies using the HLCA core (Supplementary Table 10, Supplementary Fig. 8a and Methods) revealed no significant cell type compositional changes associated with corticosteroid inhalation or asthma, respectively (Supplementary Fig. 8b,c and Supplementary Table 11). In contrast, we find that the proportion of capillary endothelial cells in lung resection tissue from the Lung Tissue Database is higher in samples from patients with severe COPD (GOLD stage 3 or 4) than in those from non-COPD controls matched for age and smoking history (Padj = 0.0004). Conversely, alveolar and interstitial macrophages, AT2 cells and dendritic cells decrease in proportion (Fig. 5e, Supplementary Fig. 8d and Supplementary Table 11; Padj = 0.0007, 0.0003, 0.005 and 3.21 × 10, respectively). Finally, smooth muscle shows the largest shift in proportion, increasing significantly in patients with severe COPD (P = 1.85 × 10) in line with previous work. As deconvolution of bulk samples using the HLCA can reveal disease-specific changes in cell type composition, we provide publicly available preprocessed cell type signature matrices based on the HLCA core (https://github.com/LungCellAtlas/HLCA). As knowledge of cell types in the lung expands, and the sizes of newly generated datasets increase, annotations in the HLCA core will need to be further refined. The HLCA and its annotations can be updated by learning from new data projected onto the reference. We simulated such an HLCA update using the previously projected healthy lung dataset, specifically focusing on the cell identities that were distinguished based on their tissue location in matched spatial transcriptomic data (spatially annotated cell types). These cell identities were present at very low frequencies (median: 0.005% of all cells; Supplementary Fig. 9a). Both spatially annotated mesenchymal cell types with more than 40 cells (immune-recruiting fibroblasts and chondrocytes) and two rare cell types (myelinating Schwann cells and perineurial nerve-associated fibroblasts) were recovered in distinct clusters (spatially annotated clusters), and three of these (all except chondrocytes) also contained cells from the HLCA core, thereby enabling a refinement of existing HLCA core annotations using the spatial context from the projected dataset (Fig. 5f and Supplementary Fig. 9b,c). In this manner the HLCA core and its annotations can be refined by mapping new datasets to the atlas and incorporating annotations from these new datasets into the reference. To extend the atlas and include samples from lung disease, we mapped 1,797,714 cells from 380 healthy and diseased individuals from 37 datasets (four unpublished and 33 published) to the HLCA core using scArches, bringing the HLCA to a total of 2.4 million cells from 486 individuals (Fig. 6a and Supplementary Table 1). Label transfer from the HLCA core to the newly mapped datasets enabled detailed cell type annotation across datasets even for rare cells, including 2,048 migratory dendritic cells identified across 28 datasets with label transfer, whereas this cell type was originally labeled in only two of 12 labeled datasets (Extended Data Fig. 8).Fig. 6The extended HLCA enables the identification of disease-associated cell states.a, UMAP of the extended HLCA colored by coarse annotation (HLCA core) or in gray (cells mapped to the core). b, Uncertainty of label transfer from the HLCA core to newly mapped datasets, categorized by several experimental or biological features. Categories with fewer than two instances are not shown. The numbers of datasets per category were as follows: 30 cells, 7 nuclei, 23 healthy, 5 IPF, 3 CF, 3 carcinoma, 4 ILD, 8 surgical resection, 7 donor lung, 12 lung explant, 6 bronchoalveolar lavage fluid, 4 autopsy, 9 10x 5′, 31 10x 3′, 4 Drop-Seq and 3 Seq-Well. c, Bottom, mean label transfer uncertainty per mapped healthy lung sample in the HLCA extension, grouped into age bins and colored by study. The numbers of mapped samples per age bin were as follows: 43 for 0–10 years, 33 for 10–20 years, 31 for 20–30 years, 23 for 30–40 years, 19 for 40–50 years, 12 for 50–60 years, 9 for 60–70 years, 8 for 70–80 years and 2 for 80–90 years. Top, bar plot showing the number of donors per age group in the HLCA core. d, Violin plot of label transfer uncertainty per transferred cell type label for a single mapped IPF dataset, split into cells from healthy donors (blue) and donors with IPF (orange). e, Uncertainty-based disease signature scores among alveolar fibroblasts and alveolar macrophages, split into cells from control donors (n = 10,453 and 1,812, respectively), and low-uncertainty cells (n = 1,419 and 200, respectively) and high-uncertainty cells (n = 1,172 and 162, respectively) from donors with IPF. f, UMAP embedding of alveolar fibroblasts (labeled with manual annotation (core) or label transfer (five IPF datasets)) colored by Leiden cluster. g, Composition of the clusters shown in f by study, with cells from control samples colored in gray. h, Expression of marker genes for IPF-enriched cluster 0 per alveolar fibroblast cluster. Cluster 5 was excluded as 96% of its cells were from a single donor. i, UMAP of all MDMs in the HLCA, colored by Leiden cluster. j, Composition of the MDM clusters from i by disease. k, Expression of cluster marker genes among all MDM clusters excluding donor-specific clusters 5 and 6. For h and k, mean counts were normalized such that the highest group mean was set to 1 for each gene. For b, c and e, the boxes show the median and interquartile range. Data points more than 1.5 times the interquartile range outside the low and high quartile are considered outliers. Whiskers extend to the furthest nonoutlier point. BALF, bronchoalveolar lavage fluid; CF, cystic fibrosis; Drop-Seq, droplet sequencing; ILD, interstitial lung disease; Mph, macrophages; SM, smooth muscle; uncert., uncertainty. a, UMAP of the extended HLCA colored by coarse annotation (HLCA core) or in gray (cells mapped to the core). b, Uncertainty of label transfer from the HLCA core to newly mapped datasets, categorized by several experimental or biological features. Categories with fewer than two instances are not shown. The numbers of datasets per category were as follows: 30 cells, 7 nuclei, 23 healthy, 5 IPF, 3 CF, 3 carcinoma, 4 ILD, 8 surgical resection, 7 donor lung, 12 lung explant, 6 bronchoalveolar lavage fluid, 4 autopsy, 9 10x 5′, 31 10x 3′, 4 Drop-Seq and 3 Seq-Well. c, Bottom, mean label transfer uncertainty per mapped healthy lung sample in the HLCA extension, grouped into age bins and colored by study. The numbers of mapped samples per age bin were as follows: 43 for 0–10 years, 33 for 10–20 years, 31 for 20–30 years, 23 for 30–40 years, 19 for 40–50 years, 12 for 50–60 years, 9 for 60–70 years, 8 for 70–80 years and 2 for 80–90 years. Top, bar plot showing the number of donors per age group in the HLCA core. d, Violin plot of label transfer uncertainty per transferred cell type label for a single mapped IPF dataset, split into cells from healthy donors (blue) and donors with IPF (orange). e, Uncertainty-based disease signature scores among alveolar fibroblasts and alveolar macrophages, split into cells from control donors (n = 10,453 and 1,812, respectively), and low-uncertainty cells (n = 1,419 and 200, respectively) and high-uncertainty cells (n = 1,172 and 162, respectively) from donors with IPF. f, UMAP embedding of alveolar fibroblasts (labeled with manual annotation (core) or label transfer (five IPF datasets)) colored by Leiden cluster. g, Composition of the clusters shown in f by study, with cells from control samples colored in gray. h, Expression of marker genes for IPF-enriched cluster 0 per alveolar fibroblast cluster. Cluster 5 was excluded as 96% of its cells were from a single donor. i, UMAP of all MDMs in the HLCA, colored by Leiden cluster. j, Composition of the MDM clusters from i by disease. k, Expression of cluster marker genes among all MDM clusters excluding donor-specific clusters 5 and 6. For h and k, mean counts were normalized such that the highest group mean was set to 1 for each gene. For b, c and e, the boxes show the median and interquartile range. Data points more than 1.5 times the interquartile range outside the low and high quartile are considered outliers. Whiskers extend to the furthest nonoutlier point. BALF, bronchoalveolar lavage fluid; CF, cystic fibrosis; Drop-Seq, droplet sequencing; ILD, interstitial lung disease; Mph, macrophages; SM, smooth muscle; uncert., uncertainty. Out of 37 new datasets, 27 were observed to map well to the HLCA, as evaluated by the mean label transfer uncertainty score (Fig. 6b, Supplementary Fig. 10a and Methods). The remaining ten datasets were often from coronavirus disease 2019 (COVID-19) studies or, unlike the HLCA core, contained pediatric samples (Fig. 6b,c and Supplementary Fig. 10b). In these datasets, higher uncertainty values may be attributable to true biological differences between the mapped data and the HLCA core adult, healthy lung samples. Overall, the successfully mapped datasets include disease samples, as well as single-nucleus and single-cell data from multiple chemistries (Fig. 6b), demonstrating the potential of the HLCA core as a universal reference. Pulmonary diseases are characterized by the emergence of unique disease-associated transcriptional phenotypes. We observed higher levels of label transfer uncertainty in datasets from diseased lungs (Fig. 6b, condition), possibly flagging cell types changed in response to disease. Specifically, labels of alveolar fibroblasts and alveolar macrophages, which interact to form a dysregulated cellular circuit in idiopathic pulmonary fibrosis (IPFs), are transferred with higher uncertainty in IPF samples than in samples from healthy controls from the same dataset (Fig. 6d and Extended Data Fig. 9a,b). Furthermore, uncertainty scores separate cells—derived from donors with IPF—within these cell types into more and less affected subsets: the genes more highly expressed in the high-uncertainty subset are also lowly expressed in healthy samples (Fig. 6e). Genes downregulated in high-uncertainty IPF macrophages are associated with homeostatic functions of tissue-resident alveolar macrophages and lipid metabolism (PPARG, FABP4 and others), while upregulated genes are associated with extracellular matrix remodeling and scar formation in the context of lung fibrosis (SPP1, PLA2G7 and CCL2; Supplementary Tables 12 and 13 and Extended Data Fig. 9b,c). Thus, the HLCA core can be used to annotate new data, identify previously unreported populations, and—using label transfer uncertainty scores—help to detect disease-affected cell states and corresponding gene expression programs. This vastly speeds up analysis and interpretation of new data, automatically prioritizing the most relevant populations. Automated mapping of new data to the HLCA core can be done by any user via an interactive web portal (https://github.com/LungCellAtlas/HLCA) or using code tutorials as provided online. Similar to healthy cellular states, the HLCA can provide insight into disease-specific states that are consistent across demographics and experimental protocols. To demonstrate this, we determined which cell types are consistently affected by IPF across datasets, extending the previous IPF analysis to five independent datasets. We found that cells labeled as alveolar fibroblasts consistently show high uncertainty levels in IPF samples compared with controls across all mapped IPF datasets that include controls (Extended Data Fig. 10a). Clustering of alveolar fibroblasts from the HLCA core and all IPF datasets shows that cells from patients with IPF predominantly cluster together in a single cluster (Fig. 6f,g and Extended Data Fig. 10b) characterized by high expression of genes previously associated with IPF (CCL2, COL1A1, CTHRC1 and MMP19), as well as further fibrosis-associated markers (SERPINE1, an inhibitor of extracellular matrix breakdown, and HIF1A, a chronic hypoxia response gene; Fig. 6h and Supplementary Table 14). These marker genes are consistently expressed across datasets (Extended Data Fig. 10c), confirming that the identification of this IPF-specific alveolar fibroblast state is reproducible. The HLCA contains data across more than ten lung diseases, providing the unique opportunity to discover cellular states shared across diseases. Discovering such common diseased cellular states could improve our understanding of lung diseases and accelerate the identification of effective treatments. For example, profibrotic SPP1 monocyte-derived macrophages (MDMs) have previously been reported in COVID-19, IPF and cancer. To test whether similar cross-disease MDM states could be discovered in the HLCA, we performed clustering of all MDMs from the HLCA (Fig. 6i). We identified four main MDM subtypes (Methods and Supplementary Table 15), each showing distinct gene expression and disease enrichment patterns, and representing different stages of monocyte-to-MDM differentiation and adaptation to the disease microenvironment. First, an early and inflammatory MDM state was observed that was high in the expression of CCL2, a gene involved in the recruitment of immune cells. This cluster predominantly contained cells from bronchoalveolar lavage fluid samples collected early during the course of COVID-19 pneumonia (cluster 2; IL1RN and S100A12; Fig. 6i–k and Extended Data Fig. 10d–h). We further observed an MDM subset expressing inflammation and phagocytosis-associated genes (cluster 4; CCL18, IL18, C1QA and TREM2) and enriched for samples from patients with COVID-19 pneumonia, as well as samples from patients with lung carcinoma (Fig. 6i–k and Extended Data Fig. 10d–h). A third MDM subset represented a more differentiated MDM phenotype, as indicated by the expression of MARCO and MCEMP1, dominated by cells from nondiseased samples (cluster 3; Fig. 6i–k and Extended Data Fig. 10d,f). The final MDM subset was dominated by IPF samples. Interestingly, this cluster was also enriched for cells from patients who died late in the course of COVID-19 and developed post-COVID-19 lung fibrosis, as well as cells from patients with lung carcinoma (cluster 0; Fig. 6i–k and Extended Data Fig. 10g–i). This multidisease cluster is marked by high expression of SPP1, LPL and CHIT1—markers that have been shown to play a causal role in the development of lung fibrosis (Fig. 6k), one of which (CHIT1) is currently being investigated as a therapeutic target for IPF. The expression of these markers is consistent across diseases and studies (Extended Data Fig. 10f), suggesting that also in cancer and late-stage COVID-19 samples a subset of MDMs adopt a fibrosis-associated phenotype. Together, this analysis shows that the HLCA enables a better understanding of cellular states shared between diseases and thereby has the potential to accelerate the discovery of effective disease treatments. In this study, we built the HLCA: an integrated reference atlas of the human respiratory system. While previous studies have described the cellular heterogeneity within the human lung, study-specific biases due to experimental design and a limited number of sampled individuals constrain their capacity to capture population variation and serve as a universal reference. The HLCA integrates data from 49 datasets to produce such a reference of 2.4 million cells, covering all major lung scRNA-seq studies published to date. The core of this atlas consists of a fully integrated healthy reference of 14 datasets with 61 cell identities, including rare and novel cell types, representing a data-derived consensus annotation of the cellular landscape of the human lung. We leveraged the unprecedented complexity of the HLCA to recover cell type-specific gene modules associated with covariates such as lung anatomical location, age, sex, BMI and smoking status. By projecting data onto the HLCA, we showed that the HLCA enables a fast and detailed annotation of new datasets, as well as the identification of unique, disease-associated cell states and cell states common to multiple diseases. The HLCA is publicly available as a resource for the community, together with an online platform for automated mapping of new data. Taken together, the HLCA is a universal reference for single-cell lung research that promises to accelerate future studies into pulmonary health and disease. The ultimate goal of a human lung cell atlas reference is to provide a comprehensive overview of all cells in the healthy human lung, as well as their variation from individual to individual. Despite its overall diversity, the HLCA is limited by the biological, demographic and experimental diversity in the foundational single-cell studies. For example, 65% of the HLCA core data are from individuals of European harmonized ethnicity, highlighting the urgent need for diversification of the population sampled in lung studies. Moreover, ethnicity metadata were based on self-reports and harmonized across datasets, which is an imperfect approach to representing the diversity of the atlas. SNP-based inference of genetic ancestry constitutes a more objective and therefore preferable approach to the grouping of individuals based on genetic background and would aid in better assessing the genetic diversity captured in the atlas. Overall, more diverse samples will enrich the atlas, diversify captured cell identities and improve the quality of the HLCA as a reference for new datasets. Such a reference will also enable comparison with model systems such as mice, cell lines or organoids, although further method development may be required to map across diverse in vitro and clinical datasets. The constituent datasets of the HLCA vary widely in experimental design, such as the sample handling protocol or single-cell platform used, causing dataset-specific batch effects. The quality of the HLCA hinges on the choice of data integration method, with methods such as Seurat’s RPCA and Harmony failing to correctly group rare cell identities into separate clusters. Nevertheless, also in the HLCA, certain subsets of T cells (regulatory T cells and γδ T cells) could not be identified as separate clusters, showing the limitations of the current HLCA in capturing cellular heterogeneity for a subset of immune cell types. Mapping additional datasets with high-resolution annotations (for example, derived from multimodal data) could provide the power to detect these cell identities in the atlas. Indeed, the HLCA must be viewed as a live resource that requires continuous updates. While we showed that mapping new, spatially annotated data to the HLCA core can refine HLCA annotations, this new knowledge must be consolidated by regular updates of the HLCA with new datasets (including epigenomic, spatial and imaging data) and refinements of HLCA annotations based on additional expert opinions. Thereby, the HLCA can serve as a community- and data-driven platform for open discussion on lung cell identities as the respiratory community progresses in charting the cellular landscape of the lung. In this process, we envision that the HLCA will be completed in two phases: first on the level of cellular variation (when no new consensus cell types can be found) and then in the description of individual variation (when population diversity is fully represented). Taken together, the HLCA provides a central single-cell reference of unprecedented size. It offers a model framework for building integrated, consensus-based, population-scale atlases for other organs within the Human Cell Atlas. The HLCA is publicly available, and combined with an open-access platform to map new datasets to the atlas, this resource paves the way toward a better and more complete understanding of both health and disease in the human lung. Ethics approval information per study was as follows. For the pooled data from refs. , approval was given by the Vanderbilt Institutional Review Board (IRB) (numbers 060165 and 171657) and Western IRB (number 20181836). All samples were collected from declined organ donors who were also consented for research. For ref. , the study was approved by the Comité de Protection des Personnes Sud Est IV (approval number 17/081). Informed written consent was obtained from all participants involved. For Jain_Misharin_2021 (A.V.M., M.J. and N.S.M., newly generated dataset), the protocol was approved by the Northwestern University IRB (STU00214826). Written informed consent was obtained from all study participants. For ref. , patient tissues were obtained under a protocol approved by Stanford University’s Human Subjects Research Compliance Office (IRB 15166). Informed consent was obtained from each patient before surgery. For ref. , healthy control lungs were obtained under a protocol approved by the University of Pittsburgh Committee for Oversight of Research and Clinical Training Involving Decedents (CORID protocol 718) and following rejection as candidate donors for transplant (IRB STUDY 19100326). For ref. , tissue samples were obtained from the Cambridge Biorepository for Translational Medicine (CBTM) with approval from the National Research Ethics Services (NRES) Committee of East of England—Cambridge South (15/EE/0152). Tissue samples were obtained with informed consent from the donor families. For ref. , the protocol was approved by the Northwestern University IRB (STU00212120). Written informed consent was obtained from all individuals in the study. For the pooled data from ref. and associated unpublished data, the protocol was approved by the IRB (Algemeen Beoordelings- en Registratieformulier number NL69765.042.19). Patients gave informed consent. For ref. , the National Jewish Health IRB approved the research under IRB protocols HS-3209 and HS-2240. Informed consent was obtained from authorized family members of all donors. For ref. , approval was given by the NRES Committee of East of England—Cambridge South (Research Ethics Committee (REC) reference: 15/EE/0152). Informed consent for use of the tissue was obtained from the donors’ families. For Barbry_unpubl (P.B., L.-E.Z., M.J.A., A.C., C.B. et al., newly generated dataset), the protocol was approved by the Centre Hospitalier Universitaire de Nice. Nasal and tracheobronchial samples were collected from patients with IPF after obtaining their informed consent. For ref. , approved was given by the IRB of Northwestern University (STU00212120, STU00213177, STU00212511 and STU00212579). For inclusion in this study, patients or their designated medical power of attorney provided informed consent. For Duong_lungMAP_unpubl (T.E.D., K.Z., X.S., J.S.H. and G.P., newly generated dataset), all postmortem human donor lung samples were obtained from the Biorepository for Investigation of Neonatal Diseases of the Lung (BRINDL), supported by the National Heart, Lung, and Blood Institute (NHLBI) LungMAP Human Tissue Core housed at the University of Rochester. Consent can be found on the repository’s website (brindl.urmc.rochester.edu/). For ref. , the study was conducted in accordance with the Declaration of Helsinki and Department of Health and Human Services Belmont Report. The use of biomaterial and data for this study was approved by the local ethics committee of the Medical Faculty Heidelberg (S-270/2001 and S-538/2012). All individuals gave informed consent for inclusion before they participated in the study. For ref. , human lung tissues were procured under each institution’s approved IRB protocol (numbers 00035396 (Cedars-Sinai Medical Center), 03-1396 (University of North Carolina at Chapel Hill), 1172286 (Cystic Fibrosis Foundation and WIRB-Copernicus Group Western IRB) and 16-000742 (University of California, Los Angeles)). Informed consent was obtained from lung donors or their authorized representatives. For ref. , the study was approved and monitored by the National Jewish Health IRB (FWA00000778). Written informed consent was obtained from all participants. For ref. , the study protocol was approved by the Partners Healthcare IRB (protocol 2011P002419). For ref. , lung tissue was obtained under a protocol approved by the University of Pittsburgh IRB (IRB STUDY 19100326) during transplantation surgery. For ref. , the study was conducted according to the principles expressed in the Declaration of Helsinki. Ethical approval was obtained from Ethics Committee Research UZ/KU Leuven (S63881). All participants provided written informed consent for sample collection and subsequent analyses. For ref. , approval was given by the NRES Committee of East of England—Cambridge South (15/EE/0152). The CBTM operates in accordance with UK Human Tissue Authority guidelines. Samples were obtained from deceased transplant organ donors by the CBTM with informed consent from the donor families. For ref. , ethical approval was given through the Living Airway Biobank, administered through the University College London Great Ormond Street Institute of Child Health (REC reference: 19/NW/0171; Integrated Research Application System (IRAS) project ID: 261511; North West Liverpool East REC), REC reference 18/SC/0514 (IRAS project ID: 245471; South Central Hampshire B REC; administered through the University College London Hospitals NHS Foundation Trust), REC reference 18/EE/0150 (IRAS project ID: 236570; East of England—Cambridge Central REC; administered through Great Ormond Street Hospital NHS Foundation Trust) and REC reference 08/H0308/267 (administered through the Cambridge University Hospitals NHS Foundation Trust), as well as by the local R&D departments at all hospitals. All of the study participants or their surrogates provided informed consent. For ref. , all protocols were reviewed and approved by the IRB at the Memorial Sloan Kettering Cancer Center (IRB protocol 14-091). Noninvolved lung, tumor tissues and metastatic lesions were obtained from patients with lung adenocarcinoma undergoing resection surgery at the Memorial Sloan Kettering Cancer Center after obtaining informed consent. For ref. , samples underwent IRB review and approval at the institutions where they were originally collected. Specifically, the Dana-Farber Cancer Institute approved protocol 13-416, the partners Massachusetts General Hospital and Brigham and Women’s Hospital approved protocols 2020P000804, 2020P000849 and 2015P002215, the Beth Israel Deaconess Medical Center approved protocols 2020P000406 and 2020P000418 and New York Presbyterian Hospital/Columbia University Irving Medical Center approved protocols IRB-AAAT0785, IRB-AAAB2667 and IRB-AAAS7370. Secondary analysis of samples at the Broad Institute was covered under Massachusetts Institute of Technology IRB protocols 1603505962 and 1612793224, or the Not Human Subjects Research protocol ORSP-3635. Donor identities were encoded at the hospitals before shipping to or sharing with the Broad Institute for sample processing or data analysis, respectively. For ref. , the study was approved by the local ethics committee of the Ludwig Maximilian University of Munich (EK 333-10 and 382-10). Written informed consent was obtained from all patients. For Schiller_2021 (H.B.S., J.G.-S., C.H.M., B.H.K., M.A. et al., newly generated dataset), the study was approved by the local ethics committee of the Ludwig Maximilian University of Munich (EK 333-10 and 382-10). Written informed consent was obtained from all patients. For Schultze_unpubl (J.L.S., C.S.F., T.S.K. and E.C., newly generated dataset), human lung tissue was available for research purposes following ethical approval from Hannover Medical School (ethical vote of the German Centre for Lung Research (DZL) number 7414, 2017). All patients in this study provided written informed consent for sample collection and data analysis. For ref. , samples were obtained under the Cells and Mediators IRB protocol (2003P002088). All individuals provided written informed consent. For ref. , the studies described were conducted according to the principles of the Declaration of Helsinki. The study was approved by the University of California, San Francisco IRB. Written informed consent was obtained from all individuals. For ref. , peripheral blood was obtained from healthy consenting adult volunteers by venipuncture through a protocol approved by the Columbia University IRB. All relevant ethical regulations for work with human participants were complied with. For ref. , donor lung samples were provided through the federal United Network for Organ Sharing via the National Disease Research Interchange and International Institute for the Advancement of Medicine and entered into the NHLBI LungMAP BRINDL at the University of Rochester Medical Center, overseen by the IRB as RSRB00047606. For for the pooled data from ref. and associated unpublished data, human lung tissue collection was approved by the Duke University IRB (Pro00082379) and University of North Carolina Biomedical IRB (03-1396) under exempt protocols. Consent was obtained to use human tissues for research purposes. For ref. , the study was approved by the local ethics committee at University Hospitals Leuven (B322201422081) and all of the relevant ethical regulations were complied with. Only patients with untreated, primary, nonmetastatic lung tumors who underwent lung lobe resection with curative intent and who provided informed consent were included in this study. For ref. , all of the research involving human participants was approved by the Northwestern University IRB. Samples from patients with COVID-19, viral pneumonia and other pneumonia, as well as controls without pneumonia, were collected from participants enrolled in the Successful Clinical Response in Pneumonia Therapy study STU00204868. All study participants or their surrogates provided informed consent. For ref. , the IRB of the University of Cincinnati College of Medicine approved all human-relevant studies. For ref. , the study was conducted according to the principles expressed in the Declaration of Helsinki. Ethical approval was obtained from the REC of Shenzhen Third People’s Hospital (2020-112). All participants provided written informed consent for sample collection and subsequent analyses. Further study details can be found in Supplementary Table 1. Several previously unpublished datasets were used for the HLCA and generated as follows. Participants recruited by the Pneumology Unit of Nice University Hospital were sampled between 1 and 15 December 2020. The full procedure, including patient inclusion criteria, is detailed at https://www.clinicaltrials.gov/ct2/show/NCT04529993. Nasal and tracheobronchial samples were collected from patients with IPF after obtaining their informed consent, following a protocol approved by the Centre Hospitalier Universitaire de Nice. The data were derived from the clinical trial registered at ClinicalTrials.gov under reference NCT04529993. This study was described as an interventional study instead of an observational study because the participants were volunteers and all assigned to a specific bronchoscopy not related to routine medical care. Participants were prospectively assigned to a procedure (bronchoscopy) according to a specific protocol to assess our ability to sample the airway. No other procedures were included in this study. Metadata of the donors’ sex was based on self-report. The libraries were prepared as described in Deprez et al. and yielded an average of 61,000 ± 11,000 cells per sample, with a viability above 95%. The single-cell suspension was used to generate single-cell libraries following the v3.1 protocol for 3′ chemistry from 10x Genomics (CG000204). Sequencing was performed on a NextSeq 500/550 sequencer (Illumina). Raw sequencing data were processed using the Cell Ranger 6.0.0 pipeline, with the reference genome GRCh38 and annotation using Ensembl98. For each sample, cells with fewer than 200 transcripts or more than 40,000 transcripts were filtered out, as well as genes expressed in fewer than three cells. Normalization and log transformation were done using the standard Scanpy pipeline. Principal component analysis (PCA) was performed on 1,000 highly variable genes (HVGs) to compute 50 principal components, and the Louvain algorithm was used for clustering. These clusters were then annotated by hand for each sample. Raw counts and the thus obtained cell annotations were used as input for the HLCA. Tumor-free, uninvolved lung samples (peritumor tissues) were obtained during tumor resections at the lung specialist clinic Asklepios Fachkliniken München-Gauting and accessed through the bioArchive of the Comprehensive Pneumology Center in Munich. The study was approved by the local ethics committee of the Ludwig Maximilian University of Munich (EK 333-10 and 382-10), and written informed consent was obtained from all patients. All fresh tissues from patients in a given time frame without any specific selection criteria were included, and only patients with obvious chronic lung disease as comorbidity based on their lung function parameters before tumor resection were excluded. Metadata of the donors’ sex were based on self-report. Single-cell suspensions for scRNA-seq were generated as previously described. In brief, lung tissue samples were cut into smaller pieces, washed with phosphate-buffered saline (PBS) and enzymatically digested using an enzyme mix composed of dispase, collagenase, elastase and DNAse for 45 min at 37 °C while shaking. After inactivating the enzymatic activity with 10% fetal calf serum (FCS)/PBS, dissociated cells were passed through a 70 µm cell strainer, pelleted by centrifugation (300g; 5 min) and subjected to red blood cell lysis. After stopping the lysis with 10% FCS/PBS, the cell suspension was passed through a 30 µm strainer and pelleted. Cells were resuspended in 10% FCS/PBS, assessed for viability and counted using a Neubauer hematocytometer. The cell concentration was adjusted to 1,000 cells per µl and ~16,000 cells were loaded on a 10x Genomics Chip G with Chromium Single Cell 3′ v3.1 gel beads and reagents (3′ GEX v3.1; 10x Genomics). Libraries were prepared according to the manufacturer’s protocol (CG000204_RevD; 10× Genomics). After a quality check, scRNA-seq libraries were pooled and sequenced on a NovaSeq 6000 instrument. The generation of count matrices was performed using the Cell Ranger computational pipeline (v3.1.0; STAR v2.5.3a). The reads were aligned to the GRCh38 human reference genome (GRCh38; Ensembl99). Downstream analysis was performed using the Scanpy package (version 1.8.0). We assessed the quality of our libraries and excluded barcodes with fewer than 300 genes detected, while retaining those with a number of transcripts between 500 and 30,000. Furthermore, cells with a high proportion (>15%) of transcript counts derived from mitochondrial-encoded genes were removed. Genes were considered if they were expressed in at least five cells. Raw counts of cells that passed filtering were used as input for the HLCA. All postmortem human donor lung samples were obtained from BRINDL, supported by the NHLBI LungMAP Human Tissue Core housed at the University of Rochester. Consent, tissue acquisition and storage protocols can be found on the repository’s website (brindl.urmc.rochester.edu/). Data were collected as part of the Human Biomolecular Atlas Program (HuBMAP). Metadata of the donor’s sex were based on self-report. For isolation of single nuclei, ten cryosections (40 µm thickness) from O.C.T.-embedded tissue blocks stored at −80 °C were shipped on dry ice and processed according to a published protocol. Single-nucleus RNA-seq was completed using 10x Chromium Single Cell 3’ Reagent Kits v3, according to a published protocol. Raw sequencing data were processed using the 10x Cell Ranger v3 pipeline and the GRCh38 reference genome. For downstream analysis, mitochondrial transcripts and doublets identified by DoubletDetection version 2.4.0 were removed. Samples were then combined and cell barcodes were filtered based on the genes detected (>200 and <7,500) and the gene unique molecular identifier (UMI) ratio (gene.vs.molecule.cell.filter function) using Pagoda2 (github.com/hms-dbmi/pagoda2). Also using Pagoda2 for clustering, counts were normalized to total counts per nucleus. For batch correction, gene expression was scaled to dataset average expression. After variance normalization, all significantly variant genes (n = 4,519) were used for PCA. Clustering was done at different k values (50, 100 or 200) using the top 50 principal components and the infomap community detection algorithm. Then, principal component and cluster annotations were imported into Seurat version 4.0.0. Differentially expressed genes for all clusters were generated for each k resolution using Seurat FindAllMarkers (only.pos = TRUE, max.cells.per.ident = 1000, logfc.threshold = 0.25, min.pct = 0.25). Clusters were manually annotated based on distinct differentially expressed marker genes. Raw counts and the thus obtained cell annotations were used as input for the HLCA. These data were a combination of published and unpublished data. In both cases, healthy volunteers were recruited for bronchoscopy at the University Medical Center in Groningen after giving informed consent and according to the protocol approved by the IRB (ABR number NL69765.042.19). Inclusion criteria and tissue processing were performed as previously described. In short, all donors were 20–65 years old and had a history of smoking <10 pack-years. Metadata of the donors’ sex were based on self-report. To exclude respiratory disease, the following criteria were used: absent history of asthma or COPD; no use of asthma- or COPD-related medication; a negative provocation test (concentration of methacholine that provokes a 20% decrease in the forced expiratory volume in 1 s (FEV1) > 8 mg ml); no airflow obstruction (FEV1/forced vital capacity ≥ 70%); and an absence of lung function impairment (that is, FEV1 ≥ 80% predicted). All donors underwent a bronchoscopy under sedation using a standardized protocol. Nasal brushes were obtained from the lateral inferior turbinate in a subset of the volunteers immediately before bronchoscopy using a Cyto-Pak CytoSoft nasal brush (Medical Packaging Corporation). Six macroscopically adequate endobronchial biopsies were collected for this study, located between the third and sixth generation of the right lower and middle lobe. Bronchial brushes were obtained from a different airway at similar anatomical locations using a Cellebrity bronchial brush (Boston Scientific). Extracted biopsies and bronchial and nasal brushes were processed directly, with a maximum of 1 h delay. Bronchial biopsies were chopped biopsies using a single-edge razor blade. A single-cell solution was obtained by tissue digestion using 1 mg ml collagenase D and 0.1 mg ml DNase I (Roche) in Hanks’ Balanced Salt Solution (Lonza) at 37 °C for 1 h with gentle agitation for both nasal brushes and bronchial biopsies. Single-cell suspensions were filtered and forced using a 70 µm nylon cell strainer (Falcon), followed by centrifugation at 550g and 4 °C for 5 min and one wash with PBS containing 1% bovine serum albumin (BSA; Sigma–Aldrich). The single-cell suspensions used for 10x Genomics scRNA-seq analysis were cleared of red blood cells using a red blood cell lysis buffer (eBioscience) followed by live cell counting and loading of 10,000 cells per lane. We used 10x Genomics Chromium Single Cell 3′ Reagent Kits v2 and v3 according to the manufacturers’ instructions. Raw sequencing data were processed using the Cell Ranger 3.1.0-based HLCA pipeline, with the reference genome GRCh38 and annotation using Ensembl98. Ambient RNA correction was performed with FastCAR, using an empty library cutoff of 100 UMI and a maximum allowed contamination chance of 0.05, ignoring the mitochondrial RNA. Data were merged and processed using Seurat, filtering to libraries with >500 UMIs and >200 genes and to the libraries containing the lowest 95% of mitochondrial RNA per sample and <25% mitochondrial RNA, normalized using sctransform while regressing out variation correlating with the percentage of mitochondrial RNA per cell. In general, 15 principal components were used for the clustering, at a resolution of 0.5 to facilitate manual annotation of the dataset. Clusters in the final object that were driven by single donors were removed. Raw counts and cell annotations were used as input for the HLCA. Nasal epithelial samples were collected from healthy volunteers who provided informed consent at Northwestern Medicine in Chicago. The protocol was approved by the Northwestern University IRB (STU00214826). Healthy volunteers were recruited to match a cohort of patients with cystic fibrosis for the ongoing study at Northwestern University (with M.J. as the principal investigator). In both studies, A.V.M. did not influence participant recruitment and did not introduce biases in sample selection. Metadata of the donors’ sex were based on self-report. Briefly, donors were seated and asked to extend their neck. A nasal curette (Rhino-Pro; VWR) was inserted into either nare and gently slid in the direction of posterior to anterior ~1 cm along the lateral inferior turbinate. Five curettes were obtained per participant. The curette tip was then cut and placed in 2 ml hypothermosol and stored at 4 C until processing. A single-cell suspension was generated using the cold-active dispase protocol reported by Deprez et al. and Zaragosi and Barbry with slight modification. Specifically, ethylenediaminetetraacetic acid (EDTA) was omitted and cells were dispersed by pipetting 20 times every 5 min using a 1 ml tip instead of tritration using a 21/23 G needle. The final concentration of protease from Bacillus licheniformis was 10 mg ml. The total digestion time was 30 min. Following the wash in 4 ml 0.5% BSA in PBS and centrifugation at 400g for 10 min, cells were resuspended in 0.5% BSA in PBS and counted using a Nexcelom K2 Cellometer with acridine orange/propidium iodide reagent. This protocol typically yields ~300–500,000 cells with a viability of >95%. The resulting single-cell suspension was then used to generate single-cell libraries following the protocol for 5′ V1 (CG000086 Rev M; 10x Genomics) or V2 chemistry (CG000331 Rev A; 10x Genomics). Excess cells from two of the samples were pooled together to generate one additional single-cell library. After a quality check, the libraries were pooled and sequenced on a NovaSeq 6000 instrument. Raw sequencing data were processed using the Cell Ranger 3.1.0 pipeline, with the reference genome GRCh38 and annotation using Ensembl98. To assign sample information to cells in the single-cell library prepared from two samples, we ran souporcell version 2.0 for that library and two libraries that were prepared from these samples separately. We used common genetic variants prepared by the souporcell authors to separate cells into two groups by genotype for each library, and Pearson correlation between the identified genotypes across libraries to establish correspondence between genotype and sample. Cell annotations were assigned to cell clusters based on expert interpretation of marker genes for each cluster. Cell clusters were derived with the Seurat version 3.2 workflow in which samples were normalized with sctransform, 3,000 HVGs were selected and integrated and clusters were derived from 30 principal components using the Louvain algorithm with default parameters. Clusters with a low number of UMIs and high expression of ribosomal or mitochondrial genes were excluded as low quality. Raw counts and the thus obtained cell annotations were used as input for the HLCA. Human lung tissue wabus available for research purposes following ethical approval from Hannover Medical School (Nr. 7414, 2017). All patients in this study provided written informed consent for sample collection and data analyses. At Hannover Medical School, patients with lung cancer were recruited in the course of their operation (that is, surgical tumor resection was performed according to the ethical vote of the German Centre for Lung Research, ethical vote 7414 and data safety guidelines). There was no bias in patient recruitment since the samples were collected as fresh native tissue following surgical tumor resection and according to the availability of surplus adjacent nonmalignant lung tissue, which was resected in parallel to the tumor tissue. Metadata of the donors’ sex were based on self-report or reported by medical professionals during consenting. Fresh adjacent normal tumor-free lung tissues from patients with non-small cell lung cancer tumors were obtained by the Lung Research group (D. Jonigk, Pathology, Hannover Medical School) and processed for single-cell isolation immediately. Lung tissue was chopped with a scalpel and scissors and digested using BD Tumor Dissociation Reagent (BD Biosciences) for 30 min in a 37 °C water bath. The digestion was stopped with 1% FCS and 2 mM EDTA in PBS without Ca/Mg and cells were filtered over a 70 µm cell strainer (BD Falcon). Erythrocytes were removed using a human MACSxpress Erythrocyte Depletion Kit (Miltenyi Biotec) and cells were filtered using a 40 µm cell strainer (BD Falcon). The viability of the cells was assessed microscopically and by flow cytometry using a LIVE/DEAD Fixable Yellow Dead Cell Stain Kit (Invitrogen) and was ~84%. The single-cell suspension was processed for scRNA-seq and library preparation with the Seq-Well protocol. Library pools with fewer than 100 cells were discarded and merged into one object. The Seurat v3.2 pipeline was used to further analyze the data. Genes in fewer than five cells in the dataset, as well as the mitochondrial genes MT-RNR1 and MT-RNR2, were removed. Cells with fewer than 200 genes were discarded, whereas cells with <5% mitochondrial genes or <30% intronic reads were kept for further analysis. The data were log normalized and 2,000 variable genes were calculated for each sample for integration with Seurat’s Canonical Correlation Analysis algorithm. The data were scaled, 50 principle components were selected and the data were clustered with 0.6 resolution. Cluster annotation revealed a low-quality cluster that was subsequently removed and the process (the calculation of variable genes, calculation of 30 principal components, clustering with 0.4 resolution) was repeated. Raw counts of the cells that passed all filtering were provided as input for the HLCA. To accommodate data protection legislation, scRNA-seq datasets of healthy lung tissue were shared by dataset generators as raw count matrices, thereby obviating the need to share genetic information. Count matrices were generated using varying software (Supplementary Table 1). Previously published scRNA-seq data were partly realigned by the dataset generators: the raw sequencing data of four previously published studies were realigned to GRCh38 using Ensembl84 for the HLCA. For two of these studies, the Cell Ranger 3.1.0-based HLCA pipeline was used for realignment. For the remaining two, data were processed as previously described, but with the reference genome and genome annotation adapted to the HLCA (GRCh38; Ensembl84). All other datasets in the HLCA core were originally already aligned to GRCh38 (Ensembl84) except data from ref. (GRCh38; Ensembl93) (Supplementary Table 1). For ref. , the count matrices provided had ambient RNA removed, as described previously. For all of the datasets from the HLCA core, a preformatted sample metadata form was filled out by the dataset providers for every sample, containing metadata such as the ID of the donor from whom the sample came, the donor’s self-reported ethnicity, the type of sample, the sequencing platform and so on (Supplementary Table 2). Ethnicity metadata were based on self-reported ethnicity for live donors or retrieved from medical records or assigned by the organ procurement team in the case of organ donors, as collected in the individual studies. For donor ethnicity, the following categories of self-reported ethnicity were used during metadata collection: Black, white, Latino, Asian, Pacific Islander and mixed. To conform to pre-existing 1,000 Genomes ancestry superpopulations, these self-reported ethnicity categories were then harmonized with the superpopulation categories as follows: Black was categorized as African, white as European and Latino as admixed American, while keeping the category Asian (merging the superpopulations East Asians and South Asians as this granularity was missing from the collected self-reported ethnicity data) and keeping Pacific Islander, as this category did not correspond to any of the superpopulations but does constitute a separate population in HANCESTRO. We refer to the resulting categories as harmonized ethnicity. Both self-reported ethnicity (as collected) and harmonized ethnicity per donor are detailed in Supplementary Table 2. Cell type annotations from dataset providers were included in all datasets. For tissue dissociation protocols, protocols were grouped based on: (1) enzyme(s) used for tissue dissociation; and (2) the digestion time in cases where large differences were observed between protocols (that is, cold protease protocols were split into two groups: 30–60 min versus overnight). Patients with lung conditions affecting larger parts of the lung, such as asthma or pulmonary fibrosis, were excluded from the HLCA core and later added to the extended atlas. For the HLCA core, all matrices were gene filtered based on Cell Ranger Ensembl84 gene-type filtering (resulting in 33,694 gene IDs). Cells with fewer than 200 genes detected were removed (removing 2,335 cells and 21 extra erythrocytes with close to 200 genes expressed as these hampered SCRAN normalization; see below), along with genes expressed in fewer than ten cells (removing 5,167 out of 33,694 genes). To normalize for differences in total UMI counts per cell, we performed SCRAN normalization. Since SCRAN assumes that at least half of the genes in the data being normalized are not differentially expressed between subgroups of cells, we performed SCRAN normalization within clusters. To this end, we first performed total count normalization, by dividing each count by its cell’s total count and multiplying by 10,000. We then performed a log transformation using natural log and pseudocount 1. A PCA was subsequently performed. Using the first 50 principal components, a neighborhood graph was calculated with the number of neighbors set to k = 15. Data were subsequently clustered with Louvain clustering at a resolution of r = 0.5. SCRAN normalization was then performed on the raw counts, using the Louvain clusters as input clusters and with the minimum mean (library size adjusted) average count of genes to be used for normalization set to 0.1. The resulting size factors were used for normalization. For the final HLCA (and not the benchmarking subset), cells with abnormally low size factors (<0.01) or abnormally high total counts after normalization (>10 × 10) were removed from the data (267 cells in total). To harmonize cell type labels from different datasets in the HLCA core, a common reference was created to which original cell type labels were mapped (Supplementary Table 4). To accommodate labels at different levels of detail, the cell type reference was made hierarchical, with level 1 containing the coarsest possible labels (immune, epithelial and so on) and level 5 containing the finest possible labels (for example, naive CD4 T cells). Levels were created in a data-driven fashion, recursively breaking up coarser-level labels into finer ones where finer labels were available. To map anatomical location to a 1D CCF score that could be used for modeling, a distinction was made between upper and lower airways. First, an anatomical coordinate score was applied to the upper airways, starting at 0 and increasing linearly (with a value of 0.5) between each of the following anatomical locations: inferior turbinate, nasopharynx, oropharnyx, vesibula and larynx. The trachea received the next anatomical coordinate score using the same linear increment as in the upper airways (a score of 2.5). In the lower airways, the coordinate score within the bronchial tree was based on the generation airway, with the trachea being the first generation and the total number of generations assumed to be 23 (ref. ). The alveolar sac was assigned the coordinate score of the 23rd generation airway. The coordinate score of each generation airway was calculated by taking the log2 value of the generation and adding it to the score of the trachea. Using this methodology, the alveolus received an anatomical coordinate score of 7.02. To calculate the final CCF coordinate, the coordinate scores (ranging from 0 to 7.02) were scaled to a value between 0 (inferior turbinate) and 1 (alveolus). Samples were then mapped to this coordinate system using the best approximation of the sampling location for each of the samples of the core HLCA (Supplementary Table 3). For computational efficiency, benchmarking was performed on a subset of the total atlas, including data from ten studies split into 13 datasets (ref. was split into 10xv1 and 10xv2 data; ref. was split into 10xv2 and 10xv3 data; and the pooled data from ref. and associated unpublished data were split into two based on the processing site). The data came from 72 donors, 124 samples and 372,111 cells. Preprocessing of the benchmarking data included the filtering of cells (minimum number of total UMI counts: 500) and genes (minimum number of cells expressing the gene: 5). For integration benchmarking, the scIB benchmarking framework was used with default integration parameter settings unless otherwise specified. All benchmarked methods except scGen (that is, BBKNN, ComBat, Conos, fas tMNN, Harmony, Scanorama, scANVI, scVI and Seurat RPCA) were run at least twice: on the 2,000 most HVGs; and on the 6,000 most HVGs. Of these methods, all that did not require raw counts as input were run twice on each gene set: once with gene counts scaled to have a mean of 0 and standard deviation of 1; and once with unscaled gene counts. scVI and scANVI, which require raw counts as input, were not run on scaled gene counts. scGen was only tested on 2,000 unscaled HVGs. For HVG selection, first, HVGs were calculated per dataset using Cell Ranger-based HVG selection (default parameter settings: min_disp=0.5, min_mean=0.0125, max_mean=3, span=0.3, n_bins=20). Then, genes that were highly variable in all datasets were considered overall highly variable, followed by genes highly variable in all datasets but one, in all datasets but two and so on until a predetermined number of genes were selected (2,000 or 6,000 genes). For scANVI and scVI, genes were subset to the HVG set and the resulting raw count matrix was used as input. For all other methods, SCRAN-normalized (as described above) data were used. Genes were then subset to the precalculated HVG sets. For integration of gene-scaled data, all genes were scaled to have mean of 0 and standard deviation of 1. Two integration methods allowed for input of cell type labels to guide the integration: scGen and scANVI. As labels, level 3 harmonized cell type labels were used (Supplementary Table 4), except for blood vessel endothelial, fibroblast lineage, mesothelial and smooth muscle cells, for which we used level 2 labels. Since scGen does not accept unlabeled cells, cells that did not have annotations available at these levels (that is, cells annotated as cycling, epithelial, stromal or lymphoid cells with no further annotations; 4,499 cells in total) were left out of the benchmarking data. The dataset rather than the donor of the sample was used as the batch parameter. The maximum memory usage was set to 376 Gb and all methods requiring more memory were excluded from the analysis. The quality of each of the integrations was scored using 12 metrics, with four metrics measuring the batch correction quality and eight metrics quantifying the conservation of biological signal after integration (Supplementary Fig. 1; metrics previously described). Overall scores were computed by taking a 0.4:0.6 weighted mean of batch effect removal to biological variation conservation (bioconservation), respectively. Methods were ranked based on overall score (Supplementary Fig. 1). For integration of the data into the HLCA core, we first determined for which cases studies had to be split into separate datasets (which were treated as batches during integration). Reasons for possible splitting were: (1) different 10x versions used within a study (for example, 10xv2 versus 10xv3); or (2) the processing of samples at different institutes within a study. To determine whether these covariates caused batch effects within a study, we performed principal component regression. To this end, we preprocessed single studies separately (total count normalization to median total counts across cells and subsequent PCA with 50 principal components). For each study, we then calculated the fraction of the variance in the first 50 principal components that could be explained (PCexpl) by the covariate of interest (that is, 10x version or processing institute):[12pt] $$}_}} = _^ }( }} )}}_^ }( }_i} )}}$$=∑i=150varcov∑i=150varPCiwhere var(PCi|cov) is the variance in scores for the ith principal component across cells that can be explained by the covariate under consideration, based on a linear regression. Then, 10x version or processing institute assignments were randomly shuffled between samples and PCexpl was calculated for the randomized covariate. This was repeated over ten random shufflings and the mean and standard deviation of PCexpl were then calculated for the covariate. If the nonrandomized PCexpl was more than 1.5 standard deviations above the randomized PCexpl, we considered the covariate a source of batch effect and split the study into separate datasets. Thus, both Jain_Misharin_2021 and ref. were split into 10xv1 and 10xv2; ref. was split into 10xv2 and 10xv3; and ref. and its pooled unpublished data were not split based on 10x version nor on processing location. For integration of the datasets into the HLCA core, coarse cell type labels were used as described for integration benchmarking (AT1, AT2, arterial endothelial cell, B cell lineage, basal, bronchial vessel 1, bronchial vessel 2, capillary, multiciliated, dendritic, fibroblast lineage, KRT5KRT17 epithelial, lymphatic endothelial cell, macrophages, mast cells, megakaryocytes, mesothelium, monocytes, neutrophils, natural killer/natural killer T cells, proliferating cells, rare, secretory, smooth muscle, squamous, submucosal secretory, T cell lineage, venous and unlabeled), except cells with lacking annotations were set to unlabeled instead of being removed. scANVI was run on the raw counts of the 2,000 most HVGs (calculated as described above), using datasets as the batch variable to enable the conservation of interindividual variation. The following parameter settings were used: number of layers: 2; number of latent dimensions: 30; encode covariates: True; deeply inject covariates: False; use layer norm: both; use batch norm: none; gene likelihood: nb; n epochs unsupervised: 500; n epochs semi-supervised: 200; and frequency: 1. For the unsupervised training, the following early-stopping parameters were used: early stopping metric: elbo; save best state metric: elbo; patience: 10; threshold: 0; reduce lr on plateau: True; lr patience: 8; and lr_factor: 0.1. For the semisupervised training, the following early-stopping parameter settings were used: early stopping metric: accuracy; save best state metric: accuracy; on: full dataset; patience: 10; threshold: 0.001; reduce lr on plateau: True; lr_patience: 8; and lr_factor: 0.1. The integrated latent embedding generated by scANVI was used for downstream analysis (clustering and visualization). For gene-level analyses (differential expression and covariate effect modeling), uncorrected counts were used. To cluster the cells in the HLCA core, a nearest neighbor graph was calculated based on the 30 latent dimensions that were obtained from the scANVI output, with the number of neighbors set to k = 30. This choice of k, while improving clustering robustness, could impair the detection of very rare cell types. Coarse Leiden clustering was done on the graph with a resolution of r = 0.01. For each of the resulting level 1 clusters, a new neighbor graph was calculated using scANVIs 30 latent dimensions, with the number of neighbors again set to k = 30. Based on the new neighbor graph, each cluster was clustered into smaller level 2 clusters with Leiden clustering at a resolution of r = 0.2. The same was done for levels 3 and 4 and (where needed) 5, with k set to 15, 10 and 10, respectively, and the resolution set to 0.2. Clusters were named based on their parent clusters and sister clusters (for example, cluster 1.2 is the third biggest subcluster (starting at 0) of cluster 1). For visualization, a 2D UMAP of the atlas was generated based on the 30 nearest neighbors graph. To quantify cluster cell type label disagreement for a specific level of annotation, the label Shannon entropy was calculated on the cell type label distribution per cluster as[12pt] $$- _^k p( )}[ )} ],$$−∑i=1kpxilogpxi,where x1…xk are the set of labels at that annotation level and p(xi) is the fraction of cells in the cluster that was labeled as xi. Cells without a label at the level under consideration were not included in the entropy calculation. If <20% of cells were labeled at the level under consideration, the entropy was set to not available for the figures. The entropy of donors per cluster (that is, diversity of donors in a cluster) was calculated in the same way. To set a threshold for high label entropy, we calculated the label entropy of a hypothetical cluster with 75% of cells given one label and 25% of cells given another label, as a cluster with <75% of cells with the same label suggests substantial disagreement in terms of cell type labeling. Clusters with a label entropy above that level (0.56) were considered to have high label entropy. Six small clusters with high label entropy even at the coarsest level of annotation highlighted doublet populations (identified via simultaneous expression of lineage-specific marker genes; for example, expression of both epithelial (AT2) and stromal (smooth muscle) marker genes) not labeled as such in the original datasets. These clusters were removed from the HLCA core, bringing the total number of clusters to 94. To set a threshold for low donor entropy, we calculated the label entropy for a hypothetical cluster with 95% of cells from one donor and the remaining 5% of cells distributed over all other donors, as clusters with >95% of the cells from the same cluster could be considered single-donor clusters, possibly caused by remaining batch effects or by donor-specific biology that is difficult to interpret. Clusters with a donor entropy below that level (0.43) were considered clusters with low donor entropy. To determine how well rare cell types (ionocytes, neuroendocrine cells and tuft cells) were clustered together and separate from other cell types after integration, we calculated recall (the percentage of all cells annotated as a specific rare cell type that were grouped into the cluster) and precision (the percentage of cells from the cluster that were annotated as a specific rare cell type) for all level 3 clusters. Nested clustering on Harmony and Seurat’s RPCA output was done based on PCA of the corrected gene counts, recalculating the principal components for every parent cluster when performing clustering into smaller children clusters, with clustering performed as described above under ‘UMAP embedding and clustering’. The level 3 clusters with the highest sensitivity for each cell type are included in Supplementary Fig. 3b. Re-annotation of cells in the HLCA core was done by six investigators with expertise in lung biology (E.M., M.C.N., A.V.M., L.-E.Z., N.E.B. and J.A.K.) based on original annotations and differentially expressed genes of the HLCA core clusters. Annotation was done per cluster, using finer clusters where these represented specific known cell types or states rather than donor-specific variation. Annotations of cell identities were hierarchical (as was the harmonized cell type reference) and each cluster was annotated at the finest known level, whereafter coarser levels could automatically be inferred (for example, a cell annotated as a CD8 T cell was then automatically annotated as of T cell lineage at level 3, lymphoid cell lineage at level 2 and immune cell lineage at level 1). The number of cells per cell type is shown for all levels in Supplementary Table 5. Mislabeling of original cells was identified by comparing final annotations with original harmonized labels and checking whether these were contradictory (and not only done at different levels of detail). Out of 61 final cell types, 18 included mostly mislabeled cells, 12 of which were previously known cell types. Despite consisting of mostly mislabeled cells from the original datasets, individual experts agreed on the annotation of these cell types: for all previously known cell types with a high proportion of mislabeled cells, the majority of experts agreed on the final annotation for the atlas, or only differed in the granularity of annotation. Marker genes were calculated based on per-sample, per-cell-type pseudo-bulks, calculating the mean (normalized and log-transformed) expression per pseudo-bulk for every gene. Pseudo-bulks were only calculated for a sample if it had at least ten cells of the cell type under consideration. An exception was made for cell types with fewer than 100 cells in total, for which the minimum number of cells per sample was set to 3. Pseudo-bulks rather than cell-level counts were used to ensure equal weighing of every sample in the differential expression test, thus statistically testing cell type-specific changes in expression that were significant across samples rather than cells. As pseudo-bulks represent the mean of a repeated draw from a single distribution, based on the central limit theorem, we expect pseudo-bulk gene counts to be normally distributed, and a t-test was therefore used to test differential gene expression, comparing a single cell type with all other cell types in the atlas (marker iteration 1). To further filter out differentially expressed genes that were not consistently expressed across samples, we applied a filtering step to remove genes expressed in <80% of the pseudo-bulks, or genes expressed in <50% of cells per pseudo-bulk (with the filtering based on the mean across pseudo-bulks). Similarly, to ensure specificity of gene expression, additional filtering was done to remove genes expressed in >20% of other pseudo-bulks. For many cell types, marker genes unique to a single cell type across the entire atlas could not be found. To nonetheless collect a robust and unique set of marker genes for every cell type, a hierarchical approach was taken, subsetting the atlas to four compartments (epithelial, endothelial, immune and stromal, for each of which a marker set was calculated) before calculating cell type-specific marker genes and filtering on uniqueness only within the compartment (marker iteration 2). When necessary, a second subsetting step was done, now subsetting to the next coarsest cell type set within the compartment (for example, lymphatic endothelial cells) and repeating the same procedure (marker iteration 3). Finally, filtering criteria were loosened for the remaining cell types for which no unique markers could be found in any of the iterations (marker iterations 4 and 5). Exact filtering parameters per iteration can be found in Supplementary Table 16. For lymphatic endothelial cell subtypes, one subtype contained sufficient cells for only a single sample, hampering a pseudo-bulk-based approach. Therefore, lymphatic endothelial cell subset markers (mature, differentiating and proliferating) were chosen based on known literature, after checking consistency with expression patterns observed in the HLCA lymphatic endothelial cells. To quantify the extent to which different technical and biological covariates correlated with interindividual variation in the atlas, we calculated the variance explained by each covariate for each cell type. We first split the data in the HLCA core by cell type annotation, merging substates of a single cell type into one (Supplementary Table 5; includes the number of cells per cell type). For every cell type, we excluded samples that had fewer than ten cells of the sample. We then summarized covariate values per sample for every cell type depending on the variable, taking the mean across cells from a sample for scANVI latent components (integration results), UMI counts per cell and fractions of mitochondrial UMIs, while for all other covariates (for example, BMI and tissue sampling method) each sample had only one value; therefore, these values were used. Next, we performed principal component regression on every covariate, as described previously (see the section ‘Splitting of studies into datasets’), but now using scANVI latent component scores instead of principal component scores for the regression, yielding a fraction of latent component variance explained per covariate. Samples that did not have a value for a given covariate (for example, where the BMI was not recorded for the donor) were excluded from the regression. Categorical covariates were converted to dummy variables. Cell type–covariate pairs for which only one value was observed for the covariate were excluded from the analysis. Quantification of the correlation or dependence between variables within a cell type and identification of the minimum number of samples needed to control for spurious correlation are described below. To check the extent to which covariates correlated with each other, thereby possibly acting as confounders in the principal component regression scores, we determined dependence between all covariate pairs for every cell type. If at least one covariate was continuous, we calculated the fraction of variance in the continuous covariate that could be explained by the other covariate (dummying categorical covariates) and took the square root (equal to Pearson’s r for two continuous covariates). For two categorical covariates, if both covariates had more than two unique values, we calculated normalized mutual information between the covariates using scikit-learn, since a linear regression between these two covariates is not possible. To control for spurious correlations between interindividual cell type variation and covariates due to low sample numbers, we assessed the relationship between sample number and mean variance explained across all covariates for every cell type. We found that for cell types sampled in fewer than 40 samples the mean variance explained across all covariates showed a high negative correlation with the number of samples (Supplementary Fig. 4a). We reasoned that for these cell types correlations between interindividual variation and our covariates were inflated due to undersampling. Moreover, we note that at lower sample numbers technical and biological covariates often strongly correlate with each other across donors (Supplementary Fig. 4c). This might lead to the attribution of true biological variation to technical covariates, and vice versa, complicating the interpretation of observed interindividual cell type variation. Consequently, we consider 40 a recommended minimum number of samples to avoid spurious correlations between observed interindividual variation and tested covariates, and excluded results from cell types with fewer samples. To select cell types for which covariate effects could be confidently modeled at the gene level, we followed the same procedure for every cell type: we filtered out all genes that were expressed in fewer than 50 cells and all samples that had fewer than ten cells of the cell type. We furthermore filtered out datasets with fewer than two donors and refrained from modeling categories in covariates that had fewer than three donors in their category for that cell type. We encoded smoking status as a continuous covariate, setting never to 0, former to 0.5 and current to 1. Anatomical region was encoded into anatomical region CCF scores as described earlier. As we noted that changes from the nose to the rest of the airways and lungs were often independent from continuous changes observed in the lungs only, we encoded nasal as a separate covariate, setting samples from the nose to 1 and all others to 0. BMI and age were rescaled, such that the 10th percentile of observed values across the atlas was set to 0 and the 90th percentile was set to 1 (25 and 64 years for age, respectively, and 21.32 and 36,86 for BMI). To determine whether covariance between covariates was low enough for modeling, we calculated the variance inflation factor (VIF) between covariates at the donor level. The VIF quantifies multicollinearity among covariates of an ordinary least squares regression and a high VIF indicates strong linear dependence between variables. If the VIF was higher than 5 for any covariate for a specific cell type, we concluded that covariance was too high and excluded that cell type from the modeling. As many cell types lacked sufficient representation of harmonized ethnicities other than European, including harmonized ethnicity in the analysis simultaneously decreased the samples that could be included in the analysis to only those with ethnicity annotations; hence, we excluded harmonized ethnicity from the modeling. To model the effects of demographic and anatomical covariates (sex, age, BMI, harmonized ethnicity, smoking status and anatomical location of the sample) on gene expression, we employed a generalized linear mixed model. We used sample-level pseudo-bulks (split by cell type), since the covariates modeled also varied at the sample or donor level and not at the cell level. Modeling these covariates at the cell level (that is, treating single cells as independent samples even when they come from the same sample) has been shown to inflate P values. First, we split the lung cell atlas by cell type annotation, pooling detailed annotations into one subtype (for example, grouping all lymphatic endothelial cell subtypes into one) (Supplementary Table 5; includes the number of cells per cell type). Subsequent filtering, covariate encoding and exclusion of cell types due to covariate dependence are described above. Gene counts were summed across cells for every sample, within cell type. Sample-wise sums (that is, pseudo-bulks) were normalized using edgeR’s calcNormFactors function, using default parameter settings. We then used voom, a method designed for bulk RNA-seq that estimates observation-specific gene variances and incorporates these into the modeling. Specifically, we used a voom extension (differential expression testing with linear mixed models) that allows for mixed-effects modeling and modeled gene expression as:[12pt] $$}[ }} ] 1 + } + } + } + } + } + }\, } \\+ ( }} )$$~1+age+sex+BMI+smoking+nose+CCFscore+1∣datasetwhere dataset is treated as a random effect to correct for dataset-specific changes in expression and all other effects are modeled as fixed effects. Resulting P values were corrected for multiple testing within every covariate using the Benjamini–Hochberg procedure. To identify more systematic patterns across genes and changes happening at the gene set level, a gene set enrichment analysis was performed using correlation-adjusted mean-rank gene set tests. The analysis was performed in R using the cameraPR function in the limma package, with the differential expression test statistic. Gene Ontology biological process terms were tested separately for each comparison. These sets were obtained from MSigDB (version 7.1), as provided by the Walter and Eliza Hall Institute (https://bioinf.wehi.edu.au/MSigDB/index.html). To stratify GWAS results from several lung diseases by lung cell type, we applied stratified linkage disequilibrium score regression in single cells (sc-LDSC). sc-LDSC can link GWAS results to cell types by calculating, for each cell type, whether disease-associated variants are enriched in genomic regions of cell-type specific genes (i.e. the region of each gene and its surrounding base pairs), while taking into account the genetic signal of proximal linkage disequilibrium-associated regions. Here cell-type specific genes are defined as genes differentially expressed in the cell type of interest. In contrast with simple enrichment testing of only significantly disease-associated genes from a GWAS among genes differentially expressed in a cell type, this method takes into account all SNPs included in the GWAS. Thus, consistent enrichment of weakly disease-associated genes (that would not individually pass significance tests) in a cell type could still lead to a significant association between the disease and the cell type. In this way, sc-LDSC provides more statistical power to detect associations between cell types and heritable phenotypes such as lung diseases. To perform sc-LDSC on the HLCA, first a differential gene expression test was performed for every grouped cell type (Supplementary Table 5) in the HLCA using a Wilcoxon rank-sum test, testing against the rest of the atlas. The top 1,000 most significant genes with positive fold changes were stored as genes characterizing that cell type (cell type genes) and used as input for LDSC. Gene coordinates of cell type genes were obtained based on the GRCh37.13 genome annotation. For SNP data (names, locations and linkage-related information), the 1000 Genomes European reference (GRCh37) was used, as previously described. Only SNPs from the HapMap 3 project were included in the analysis. For identification of SNPs in the vicinity of cell type genes, we used a window size of 100,000 base pairs. Genes from X and Y chromosomes, as well as human leukocyte antigen genes, were excluded because of their unusual genetic architecture and linkage patterns. For linkage disequilibrium score calculation, a 1 cM window was used. Significance of the link between a phenotype and a cell type was calculated using LDSC. P values yielded by LDSC were corrected for multiple testing for every disease tested using the Benjamini–Hochberg correction procedure. As a negative control, the analysis was performed with a GWAS of depression and no cell types were found to be significant (Supplementary Fig. 7). The numbers of cases and controls per GWAS study were as follows: n = 2,668 cases and 8,591 controls for IPF; n = 35,735 cases and 222,076 controls for COPD; n = 11,273 cases and 55,483 controls for lung adenocarcinoma; n = 321,047 individuals for lung function; n = 88,486 cases and 447,859 controls for asthma; and n = 113,769 cases and 208,811 controls for depression (used as negative control). To enable deconvolution of bulk expression data on the basis of the HLCA, HLCA cell type signature matrices were generated. One generic-purpose signature matrix was created per sublocation of the respiratory system (that is, one parenchyma, one airway and one nose tissue matrix; Supplementary Table 10). Additionally, a script to generate custom reference sets from the HLCA data is provided together with the HLCA code on GitHub (https://github.com/LungCellAtlas/HLCA) to tailor the deconvolution signature matrix to any specific research question. Cell types were included in the bulk deconvolution signature matrix on the basis of cell proportions (constituting >2% of cells within samples of the corresponding tissue in the HLCA core). In addition, cell types were merged when they were deemed too transcriptionally similar. For each included cell type, 200 cells were randomly sampled from the HLCA core, while all cells were included for cell types with fewer than 200 cells present in the HLCA core. Cells were sampled from the matching anatomical location (for example, nose T cells rather than parenchymal T cells were used for the nose signature matrix). Signature matrices were constructed using CIBERSORTx (version 1.0) according to default settings, and no cross-platform batch correction was applied. The reference data were optimized by deconvolution of pseudo-bulk samples constructed from the HLCA core data, assessing deconvolution performance per included cell type based on the correlation of predicted proportions with ground truth composition (Supplementary Fig. 8a). The following cell types were included in the deconvolution: endothelial cell arterial, endothelial cell capillary, lymphatic endothelial cell, basal and secretory (merged), multiciliated lineage, AT2, B cell lineage, innate lymphoid cell (ILC) natural killer and T cell lineage (merged), dendritic cells, alveolar macrophages, interstitial macrophages, mast cells, fibroblast lineage, smooth muscle, endothelial cell venous and monocytes (for the parenchyma); basal resting and suprabasal (merged), multiciliated lineage, club, goblet, dendritic cells, hillock like and T cell lineage (for the nose); and endothelial cell venous, CD4 T cells, fibroblasts, smooth muscle, basal and secretory (merged), multiciliated lineage, endothelial cell capillary, interstitial macrophages, B cell lineage, natural killer cells, CD8 T cells, dendritic cells, alveolar macrophages, mast cells and monocytes (for the airway). Capillary endothelial cells and interstitial macrophages (airway) were excluded from statistical testing due to poor performance in the benchmark. Venous endothelial cells and monocytes (parenchyma), hillock-like cells and T cell lineage cells (nose) and B cell lineage cells, natural killer cells, CD8 T cells, dendritic cells, alveolar macrophages, mast cells and monocytes (airways) were excluded from statistical testing due to >60% zero proportions. The parenchymal signature matrix was used to deconvolve RNA expression data of samples from the Lung Tissue Database (GEO accession number GSE23546) using only lung tissue samples from patients with COPD GOLD stages 3 and 4 (n = 27 and 56, respectively) and matched controls (n = 281). The Lung Tissue Database dataset was run on the Rosetta/Merck Human RSTA Custom Affymetrix 2.0 microarray platform (HuRSTA-2a520709; GPL10379). As this platform has multiple probe sets for each gene, we focused on the probe sets that were derived from curated RefSeq records (with NM_ accession prefixes) when present to maximize the accuracy of the deconvolution. Where genes did not have probe sets based on curated RefSeq records or had multiple probe sets mapping to curated RefSeq records, the probe set with the highest average microarray intensity across samples was selected. Quantile normalization of the data and subsequent deconvolution were performed using CIBERSORTx. A Wilcoxon rank-sum test between control and GOLD stage 3/4 samples was performed to identify statistically significant compositional changes in advanced-stage COPD compared with control tissue. GOLD 3/4 and control samples were matched for age and smoking history. Cell types with >60% of samples predicted to have a proportion of zero were excluded from the Wilcoxon test, as the high number of tied ranks (zeros in both groups) would result in inflated P values. P values were multiple testing corrected using the Benjamini–Hochberg procedure. The same procedure was followed for a dataset of nasal brush bulk RNA-seq samples from asthmatic donors pre- and postinhalation of corticosteroids (n = 54 and 26, respectively) and a dataset of airway biopsy bulk RNA-seq samples from asthmatic donors and controls (n = 95 and 38, respectively). As these consisted of RNA-seq data, no quantile normalization was applied. To map unseen scRNA-seq and single-nucleus RNA-seq data to the HLCA, we used scArches, our transfer learning-based method that enables mapping of new data to an existing reference atlas. scArches trains an adaptor added to a reference embedding model, thereby enabling it to generate a common embedding of the new data and the reference, allowing reanalysis and de novo clustering of the joint data. The data to map were subsetted to the same 2,000 HVGs that were used for HLCA integration and embedding, and HVGs that were absent in the new data were set to 0 counts for all cells. Raw counts were used as input for scArches, except for the ref. dataset, for which ambient RNA removal was run previously on the raw counts. Healthy lung data were split into two datasets: 3′ and 5′ based. Lung cancer data were also split into two datasets: 10xv1 and 10xv2. The model that was learned previously for HLCA integration using scANVI was used as the basis for the scArches mapping. scArches was then run to train adaptor weights that allowed for mapping of new data into the existing HLCA embedding, using the following parameter settings: freeze-dropout: true; surgery_epochs: 500; train base model: false; metrics to monitor: accuracy and elbo; weight-decay: 0; and frequency: 1. The following early-stopping criteria were used: early stopping metric: elbo; save best state metric: elbo; on: full dataset; patience: 10; threshold: 0.001; reduce lr on plateau: True; lr patience: 8l and lr_factor: 0.1. To enable cross-dataset gene-level analysis, harmonization of gene names from different datasets (using different reference genome builds and genome annotations; Supplementary Table 1) was necessary. Both annotation sources (for example, Ensembl or RefSeq) and annotation versions (for example, Ensembl release 84 or Ensembl release 91) contribute to the variation between different gene naming schemes. Therefore, both annotation sources and versions, including outdated ones, need to be taken into account to enable the mapping of all gene names to a single naming scheme. For the harmonization of gene names, we aimed to map all original gene names to the target scheme HUGO Gene Nomenclature Committee gene name, corresponding to the Ensembl release 107 publication. To find the most likely match between each original gene name and a target gene name, we first downloaded Ensembl releases 79 to 107, which included for each release: (1) all Ensembl gene IDs from the downloaded release (for example, ENSG00000081237.21); (2) corresponding Ensembl transcript and protein IDs (for example, ENST00000442510.8 and ENSP00000411355.3); (3) matching Ensembl IDs from the previous release; (4) matching gene IDs from other genome annotation sources (for example, RefSeq); and (5) matching gene, transcript and protein identifiers from various external resources, such as UniProt, the HUGO Gene Nomenclature Committee and the Consensus Coding Sequence Project. We then constructed a graph, with each Ensembl ID, other genome annotation ID and external resource identifier represented by a single node per release. Nodes were then connected based on the matching (points 2–5) provided by Ensembl, weighing edges based on Ensembl similarity scores where available. For each original gene name from the HLCA datasets, the path with the lowest mean edge weight from that gene name to a gene name from the target names (Ensembl release 107) was selected to find the most likely matching gene name from the target (Supplementary Table 17). Genes for which no target could be found were excluded from downstream analysis. When multiple genes were matched with the same target gene name, counts from the original genes were summed. To identify the genes most commonly exhibiting batch-specific expression, the HLCA was split by cell type and a differential expression analysis was performed (based on a Wilcoxon rank-sum test) in each cell type, comparing cells from one dataset (batch) with those from all other datasets and repeating this for all datasets. Datasets that had fewer than ten cells of the cell type or fewer than three samples with cells of the cell type were excluded from the test. For each test, genes were filtered such that only genes that were significantly upregulated were retained. Next, the fraction of included datasets in which a gene was significantly upregulated in the cell type (affected dataset fraction) was calculated for all genes. To find genes that were often batch effect associated across many cell types, the mean of the affected dataset fractions was calculated across cell types for each gene. To perform label transfer from the HLCA core to the mapped datasets from the extended HLCA, we used the scArches k nearest neighbor-based label transfer algorithm. Briefly, a k nearest neighbor graph was generated from the joint embedding of the HLCA core and the new, mapped dataset, setting the number of neighbors to k = 50. Based on the abundance and proximity in a cell’s neighborhood of reference cells of different types, the most likely cell type label for that cell was selected. Furthermore, a matching uncertainty score was calculated based on the consistency of reference annotations among the k nearest neighbors of the cell of interest[12pt] $$u_}_c} = 1 - p( }_}\! .$$,y,Nc=1−pY=y∣X=c,Ncwhere uc,y,Nc is the uncertainty score for a query cell c with transferred label y; Nc is its set of k nearest neighbors; and p(Y = y|X = c, Nc) is the weighted (by edge weights) proportion of Nc with label y, as previously described. Thus, high consistency in HLCA core annotations leads to low uncertainty scores and low consistency (that is, mixing of distinct reference annotations) leads to high uncertainty scores. For label transfer to lung cancer and healthy, spatially annotated projected data (Fig. 5b and Extended Data Fig. 7g), cells with an uncertainty score above 0.3 were set to unknown. Disagreement between original labels and transferred annotations (that is, transferred annotations with high certainty but not matching the original label) in the data from ref. highlighted three different cases: annotations not included in the mapped data (for example, preterminal bronchiole secretory cells, which were labeled as club and goblet in the mapped data; these may not be incorrect label transfers but cannot be verified by label comparison alone); cell types that are part of a continuum, with cutoffs between cell types chosen differently in the reference than in the projected data (for example, macrophage subtypes); and cell types missing in the HLCA core that have high transcriptional similarity to other cell types that are present in the HLCA, which was observed for several finely annotated immune cell identities. For example, γδ T cells, ILCs, megakaryocytes, natural killer T cells and regulatory T cells were not annotated in the HLCA core, as these cell types could not be distinguished with confidence in the integrated object and were often lacking in the constituent datasets. Indeed, cell types from the T cell/ILC/natural killer lineages are known to be particularly difficult to annotate using transcriptomic data only. Therefore, cells annotated with these labels in the projected dataset were largely incorrectly annotated as CD4 T cells, CD8 T cells and natural killer cells through label transfer (Fig. 5b and Extended Data Fig. 6e) For the extended atlas, we calibrated the uncertainty score cutoff by determining which uncertainty levels indicate possible failure of label transfer. To determine the uncertainty score at which technical variability from residual batch effects impairs correct label transfer, we evaluated how label transfer performed at the level of datasets, as these predominantly differ in experimental design. To determine an uncertainty threshold indicative of possible failure of label transfer, we harmonized original labels for 12 projected datasets (one unpublished: Duong_lungMAP_unpubl) and assessed the correspondence between original labels with the transferred annotations. Only cells with level 3 or 4 original annotations were considered, as these levels represent informative annotations while not representing the finest detail. Level 5 annotations will often display high uncertainty levels due to high annotation granularity rather than remaining batch effects. To assess the optimal uncertainty cutoff for labeling a new cell as unknown, we used these results to generate a receiver operating characteristic curve. We chose a cutoff around the elbow point, keeping the false positive rate below 0.5 (uncertainty cutoff = 0.2; true positive rate = 0.879; false positive rate = 0.495) to best distinguish correct from incorrect label transfers (Supplementary Fig. 10a). False positives were either due to incorrect label transfer or incorrect annotations in the original datasets. Cells with an uncertainty higher than 0.2 were set to unknown. The ref. study of healthy lung included cell type annotations based on matched spatial transcriptomic data. Many of these annotations were not present in the HLCA core. To determine whether these spatial cell types could still be recovered after mapping to the HLCA core, we looked for clusters specifically grouping these cells. We focused on six spatial cell types: perineurial nerve-associated fibroblasts; endoneurial nerve-associated fibroblasts; immune-recruiting fibroblasts; chondrocytes; myelinating Schwann cells; and nonmyelinating Schwann cells. As these cell types were often present at very small frequencies, we performed clustering at different resolutions to determine whether these cells were clustered separately at any of these resolutions. We clustered at resolutions of 0.1, 0.2, 0.5, 1, 2, 3, 5, 10, 15, 20, 25, 30, 50, 80 and 100, with the number of neighbors set to k = 30 for resolutions under 25 and k = 15 for resolutions of 25 or higher, to enable the detection of smaller clusters. Minimum recall (the percentage of cells with the spatial cell type annotation captured in the cluster) and minimum precision (the percentage of cells from ref. in the cluster that had the spatial cell type annotation) were both set to 25%. The cluster with the highest recall was selected for every spatial cell type (unless this cluster decreased precision by >33% compared with the cluster with the second highest recall). If the precision of the next best cluster was doubled compared with the cluster with the highest recall and recall did not decrease by >20%, this cluster was selected. To learn disease-specific signatures based on label transfer uncertainty scores, cells from the mapped data with the same transferred label (either alveolar fibroblasts or alveolar macrophages) were split into low-uncertainty cells (<0.2) and high-uncertainty cells (>0.4), excluding cells between these extremes (for alveolar fibroblasts, n = 11,119 (<0.2) and n = 2,863 (>0.4); for alveolar macrophages, n = 1,770 (<0.2) and n = 577 (>0.4)). We then performed a differential expression analysis on SCRAN-normalized counts using a Wilcoxon rank-sum test with default parameters, comparing high- and low-uncertainty cells. The 20 most upregulated genes based on log-fold changes were selected after filtering out genes with a false discovery rate-corrected P value (using the Benjamini–Hochberg procedure) above 0.05 and genes with a mean expression below 0.1 in the high-uncertainty group. To calculate the score of a cell for the given set of genes, the average expression of the set of genes was calculated, after which the average expression of a reference set of genes was subtracted from the original average, as described previously. The reference set consists of a randomly sampled set of genes for each binned expression value. The resulting score was considered the cell’s disease signature score. To uncover the cell identities affected in IPF, label transfer uncertainty was analyzed for three mapped datasets from the extended HLCA that included both IPF and healthy samples. Cell types of interest were determined based on the largest increase in mean label transfer uncertainty in IPF compared with healthy samples, while checking for consistency in increments across the three datasets. This highlighted alveolar fibroblasts as the main cell type of interest. To find IPF-specific alveolar fibroblast states, all alveolar fibroblasts from the abovementioned datasets and two more IPF datasets (for which no healthy data were mapped, as these were already in the core) were clustered together with the alveolar fibroblasts from the HLCA core. For clustering, a k nearest neighbor graph was calculated on the joint scArches-derived 30-dimensional embedding space setting k = 30, after which the cells were clustered using the Leiden algorithm with a resolution of 0.3. The resolution was chosen such that datasets were not isolated in single clusters, thus avoiding clustering driven solely by dataset-specific batch effects. One cluster (cluster 5) was small (n = 460 cells) and displayed low donor entropy (0.17), indicating that it almost exclusively came from a single donor (96% of cells from HLCA core donor 390C). It was therefore excluded from further analysis. To perform differential gene expression analysis, gene counts were normalized to a total of 7,666 counts (the median number of counts across the HLCA) and then log transformed with a pseudocount of 1. Finally, a Wilcoxon rank-sum test was used on the normalized data to detect differentially expressed genes for cluster 0 (n = 6,765 cells versus a total of n = 14,731). The results were filtered such that genes expressed in <30% of cells of the cluster of interest were excluded, as well as genes that were expressed in >20% of cells outside of the cluster and genes with a multiple testing-corrected P value (using the Benjamini–Hochberg procedure) above 0.05 (Supplementary Table 14). To investigate whether the HLCA can be used to identify disease-associated cell states shared across multiple diseases, MDMs from the HLCA core, together with all cells from the mapped datasets labeled as MDMs based on label transfer, were jointly analyzed. Datasets and diseases with fewer than 50 MDMs were excluded from the analysis. The cells were subsequently clustered as described above for the cross-dataset IPF analysis. Finally, a Wilcoxon rank-sum test was used on the normalized data to detect differentially expressed genes per cluster (number of cells per cluster: n = 64,915 (cluster 0), 47,539 (cluster 1), 32,027 (cluster 2), 31,097 (cluster 3), 25,267 (cluster 4), 1,998 (cluster 5) and 307 (cluster 6)). The results were filtered as described above (Supplementary Table 15). The following tools and versions were used: R (version 4.1.1 for covariate modeling and version 4.0.3 for GSEA); edgeR (version 3.28.1); lme4 (version 1.1-27.1); LDSC (version 1.0.1); Limma (version 3.46.0); Scanpy (version 1.9.1); scArches (version 0.3.5); scIB (version 0.1.1); scikit-learn (version 0.24.1); and scvi-tools (scANVI; version 0.8.1). Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41591-023-02327-2.
PMC2063610
Loss of stem cell regenerative capacity within aged niches
This work uncovers novel mechanisms of aging within stem cell niches that are evolutionarily conserved between mice and humans and affect both embryonic and adult stem cells. Specifically, we have examined the effects of aged muscle and systemic niches on key molecular identifiers of regenerative potential of human embryonic stem cells (hESCs) and post-natal muscle stem cells (satellite cells). Our results reveal that aged differentiated niches dominantly inhibit the expression of Oct4 in hESCs and Myf-5 in activated satellite cells, and reduce proliferation and myogenic differentiation of both embryonic and tissue-specific adult stem cells (ASCs). Therefore, despite their general neoorganogenesis potential, the ability of hESCs, and the more differentiated myogenic ASCs to contribute to tissue repair in the old will be greatly restricted due to the conserved inhibitory influence of aged differentiated niches. Significantly, this work establishes that hESC-derived factors enhance the regenerative potential of both young and, importantly, aged muscle stem cells in vitro and in vivo; thus, suggesting that the regenerative outcome of stem cell-based replacement therapies will be determined by a balance between negative influences of aged tissues on transplanted cells and positive effects of embryonic cells on the endogenous regenerative capacity. Comprehensively, this work points toward novel venues for in situ restoration of tissue repair in the old and identifies critical determinants of successful cell-replacement therapies for aged degenerating organs.Embryonic stem cells (ESCs) are distinguished by their ability to self-renew and to differentiate into any other cell type via asymmetric cell divisions, in which one daughter cell maintains ‘stemness’ while the other daughter cell differentiates into a particular tissue type. ESCs, including those of human origin (hESCs), are derived from the blastocyst and can be propagated in vitro (Evans & Kaufman, 1981; Thomson et al., 1998; Wobus & Boheler, 2005). Their tremendous potential for organogenesis has created a great interest in using hESCs for replacing tissues and organs lost to disease, or old age (reviewed in Wobus & Boheler, 2005). As such, the use of hESCs is particularly important, due to the fact that adult organ stem cells are often limited in number, cell-fate plasticity, expansion capacity, telomere length, and lifespan (Mayhall et al., 2004). The general goal behind most cell-replacement approaches is to expand and then differentiate hESCs in vitro, thus producing a cell type of interest, such as neuronal, blood, endothelial, pancreatic, bone, and others. These differentiated cells are expected to replace their dysfunctional counterparts in vivo. The scope of disorders that can be potentially treated with a neoorganogenesis approach is large and includes many that are currently incurable, such as muscle atrophy, diabetes, Alzheimer's disease, Parkinson's disease, and other degenerative diseases that often accompany human aging (McDonald et al., 1999; Liu et al., 2000; Hori et al., 2002; Kim et al., 2002; Blyszczuk et al., 2003). While many studies have focused on the derivation, propagation, and in vitro differentiation of hESCs (reviewed in Hoffman & Carpenter, 2005; Wobus & Boheler, 2005), relatively few have examined the properties of these cells and their more differentiated progeny in the aged, as opposed to the young, systemic and local organ environments. Recently published data suggest that these extrinsic cues become altered with age in ways that preclude activation of organ stem cells (such as satellite cells), inhibit repair-specific molecular signaling (such as delta-Notch), and interfere with productive tissue repair (Conboy et al., 2003, 2005; Janzen et al., 2006; Krishnamurthy et al., 2006; Molofsky et al., 2006). Furthermore, at least two lines of evidence suggest that stem cell-based tissue-replacement therapies might be hindered in the elderly, because all cells along the developmental lineage (e.g., stem cells, more differentiated progenitor cells or even tissues containing a pool of precursors) might rapidly ‘age’ and fail to contribute to organ repair when introduced into the old organism in vivo. First, in heterochronic tissue-transplantation studies, the age of the host environment determined the regenerative outcome, as both young and old skeletal muscle explants containing differentiated and precursor cells effectively regenerated in young, but not in old animals (Zacks & Sheff, 1982; Carlson & Faulkner, 1989). Second, using parabiotically paired young and old mice, the regenerative potential of muscle and liver was shown to be influenced by the age of the systemic environment (Conboy et al., 2005). Thus, we sought to determine whether key molecular identifiers of stem cell properties, the rate of cell proliferation, and the myogenic capacity would be influenced by the age of extrinsic milieu, regardless of whether stem cells are embryonic or the more differentiated, muscle-specific satellite cells. Satellite cells are muscle stem cells situated in direct contact with myofibers, the differentiated muscle cells. When myofibers are damaged, quiescent satellite cells are activated to proliferate and then differentiate into fusion-competent myoblasts that continue to proliferate and can form primary cultures, but are also capable of producing new, multinucleated myofibers or myotubes in vitro and in vivo (Morgan et al., 2002; Collins et al., 2005; Wagers & Conboy, 2005). Activated satellite cells express myogenic markers, such as Myf5, M-cadherin, and Paired box gene 7 (Pax7); fusion-competent myoblasts express high levels of desmin, and de novo generated myofibers or myotubes express embryonic myosin heavy chain (eMyHC) and continue to express desmin (Schultz & McCormick, 1994; Wagers & Conboy, 2005). While desmin can be also present in smooth and cardiac muscle cells, the isolation of hind limb skeletal muscle with subsequent purification of myofibers away from all interstitial cells, as well as purification of associated muscle stem cells results in primary cultures that are uniformly of skeletal muscle lineage. Every desmin cell in such cultures is a fusion-competent myoblast, and is able to produce multinucleated myotubes after 48 h of culture in differentiation-promoting medium [Dulbecco's modified Eagle's medium (DMEM) with 2% horse serum]. Some of these myogenic cells fuse into myotubes, even in the mitogen-rich medium [(Opti-MEM (Invitrogen, Carlsbad, CA, USA) with 5–10% mouse serum or DMEM with 10% fetal bovine serum, FBS] (Conboy & Rando, 2002; Conboy et al., 2003; and see below). An experimental system was developed that (i) provided the ability to study the regenerative response of hESCs and of muscle stem cells in various heterochronic environments in vitro; and (ii) allowed examination of the effects of hESCs on muscle repair, in vivo, after transplantation into young vs. old hosts. This model allowed us to address both the negative effects of the aged niche on key stem cell properties and the positive effects of hESCs on the aged muscle-specific organ progenitor cells in vitro, and on the regenerative capacity of old muscle in vivo. The resulting data demonstrate that the composition of conserved extrinsic cues, regulating stem cell responses, becomes altered with age in ways that inhibit both hESCs and adult stem cell regenerative potential. Specifically, molecular markers of stem cell functionality, e.g. Oct4 (in hESCs) and Myf5 (in muscle stem cells), the rate of cell proliferation, and the capacity for myogenic differentiation are all dominantly inhibited by the aged systemic milieu, and by the old differentiated muscle tissue. However, while satellite cells are unable to deter the inhibitory affects of aged systemic and local niches, hESCs are capable of antagonizing the aged environments, thereby enhancing the regenerative potential of both young and old muscle stem cells in vitro and in vivo. Thus, a complex interplay between negative regulation of hESCs and adult muscle stem cells by the aged niche, and positive regulation of the host's regenerative responses by hESCs will likely determine the success of hESC-based cell-replacement therapies in the old. Previous work established that the upregulation of repair-specific molecular signaling mechanisms, such as Notch, and successful engagement of resident muscle stem cells in tissue repair are largely determined by the age of the systemic milieu, rather than by the cell-autonomous age of muscle cells, or by the differences in their numbers (Conboy & Rando, 2005; Conboy et al., 2005). Intriguingly, these experiments also hinted at a small but persistent inhibitory effect of the aged systemic milieu on the performance of young stem cells. Exploring this further, we found that young serum permits satellite cells to be myogenic, while old serum inhibits the satellite cell regenerative potential not only alone, but also when mixed with young serum, suggesting a dominant over-riding of ‘young’ serum factors (Fig. 1). Myofiber cultures, in which satellite cells have been activated by injury in vivo, were established from young (2–3 months) and old (22–24 months) C57-BL/6 male mice, as previously described (Conboy & Rando, 2002; Conboy et al., 2005). As previously shown, this method is well suited for the assessment of satellite cell regenerative myogenic capacity (Conboy & Rando, 2002; Wagers & Conboy, 2005). Isolated myofiber explants with associated satellite cells were cultured overnight in the presence of young or old serum (alone at 5% and 10%, and mixed at 5% young + 5% old); bromodeoxyuridine (BrdU) was added for the last 2 h of culture to measure the rate of cell proliferation. The effects of heterochronic systemic milieu on myogenic potential were examined as generation of proliferating myoblasts that express desmin and Myf5, and that spontaneously form multinucleated nascent myotubes. As shown in Fig. 1A and quantified in Fig. 1B, the age of sera clearly determined satellite cell regenerative potential and old serum strongly inhibited the myogenic potential of young satellite cells either when present alone, or when mixed with young sera. Similar data was obtained by using another myogenic marker, Pax7 (Supplementary Fig. S1). Additionally, there were two to three times fewer total cells generated in the presence of aged serum (not shown). The age of sera determined the regenerative potential of satellite cells. (A) Young satellite cells were cultured either in 5% or 10% young (Young), 10% old (Old), or in a 5%+ 5% mouse sera combination (young + old). Cells were analyzed by immunofluorescence microscopy, using anti-BrdU (red), antidesmin (green) or anti-Myf5 antibodies (green, small panels). Similar results are shown for Pax7 immunodetection (Supplementary Fig. S1). Hoechst (blue) labeled nuclei. (B) Three independent experiments were quantified [300 young myofibers per experiment] as percentage of desmin/Myf5/BrdU de novo generated cells for each age and culture condition. On average, two to three fewer cells were generated when cultured in the presence of old. Shown are identical microscope fields at ×40 magnification. At least three independent experiments produced similar results. (*) indicates P≤ 0.001 as compared to young sera. Importantly, it was not simply the dilution of young serum factors that resulted in diminished myogenic capacity when young and old sera were mixed, because young sera promoted robust myogenesis both at 10% and 5%. Thus, old serum factors dominantly inhibited the myogenic capacity of young satellite cells even in the presence of young serum. This observation suggests that satellite cells of young mice engage in efficient myogenic responses, in part, because the inhibitory influence of old circulatory milieu is absent. These data reveal that the regenerative potential of young muscle stem cells is determined by the age of the systemic milieu, prompting us to investigate whether hESCs would similarly succumb to inhibitory factors present in the aged circulation. To determine the effects of aged serum on stem cell self-renewal/pluripotency, we analyzed hESC expression of Oct4 and studied the rate of hESC proliferation, by assessing BrdU incorporation (Fig. 2) and Ki67 expression (Supplementary Fig. S2). Specifically, these determinants of hESC regenerative potential were examined in the presence of heterochronic (young vs. old) mouse sera added to typical hESC medium, e.g., MEF-conditioned medium (MCM). Oct4 is expressed by self-renewing, pluripotent ESCs in culture, by the totipotent inner cell mass of the blastocyst and by the germ cells (Nichols et al., 1998; Pesce et al., 1999). Most cells in control cultures or young conditions expressed high levels of this marker of ‘stemness’, and maintained their normal phenotype and morphology throughout the various co-culture experiments performed in this study (see below). The regenerative potential of embryonic stem cells was negatively affected by aged mouse sera. (A) hESCs were cultured in MCM with 10% young (young) or old (old) mouse serum, or in three control media: MCM without mouse sera; GM (myoblast medium of Ham's F10 with 20% FBS) and DMEM/FBS (hESC differentiation medium of DMEM with 10% FBS). BrdU was added for the last 2 h of culture to measure the rate of cell proliferation. Immunodetection assays were performed for BrdU (red), Oct4 (red), and Ki67 (Supplementary Fig. S2). Hoechst (blue) labels nuclei. A high rate of hESC proliferation and Oct4 expression is displayed in all control media and in the presence of young mouse serum. In contrast, hESC proliferation and Oct4 expression are inhibited in the presence of old mouse serum, either alone or when mixed with young serum. MCM with mouse sera at 5% gave results similar to those observed with 10% young mouse sera or in control media (Supplementary Fig. S3). (B) Three independent experiments yielded similar results and were quantified as percentage of BrdU and Oct4 cells for each culture condition. * indicates P < 0.001 as compared to young serum. Importantly, at 10% aged serum dramatically inhibited the self-renewal and proliferative potential of hESCs, as judged by highly diminished Oct4 expression and a lack of BrdU incorporation. Again, the inhibitory factors in the aged milieu were dominant over the young, as evidenced by a decline in Oct4 expression, the low rate of BrdU incorporation, and Ki67 expression in young and old mixed environments (5% young + 5% old sera in MCM). Similar to the data shown for adult stem cells (ASCs) (Fig. 1), it was not simply a dilution of young serum factors as hESCs robustly proliferated and expressed high levels of Oct4 when cultured with 5% young sera in MCM (Supplementary Fig. S3). Quantification of multiple independent experiments has demonstrated that hESC expression of Oct4 and BrdU incorporation have been reduced by two- to threefold in the aged milieu (Fig. 2B). As expected, hESCs cultured in control media, including MCM alone that does not contain either young or old serum, also displayed a high rate of proliferation and Oct4 expression (Fig. 2, control medium). Additionally, in this experimental set-up there was no general inhibitory effect of sera per se on hESC proliferation and Oct4 expression, as 10% young mouse sera (young) and 10–20% of FBS (growth medium and DMEM/FBS) allowed for a high rate of cell proliferation and for uniformly high Oct4 levels (Fig. 2). When instead of immediate exposure to aged mouse serum, hESCs were first cultured overnight in MCM, these cells were no longer susceptible to the negative effects of old systemic milieu (Fig. 3), suggesting that hESC-produced factors established an embryonic microniche that may provide temporary protection from the aged environment. It appears that satellite cells do not have such anti-aging ability, because despite an initial activation in entirely young environments, e.g., after muscle injury to young muscle, isolated satellite cells remain susceptible to inhibition by the old mouse serum (Figs 1 and 4C). Similarly, culturing satellite cells isolated from noninjured muscle in growth-promoting medium for 1–2 days does not protect against the inhibitory affects of aged systemic milieu (not shown). Embryonic stem cells produce youthful microniche in culture. (A) As opposed to immediate exposure to old mouse serum after passaging (10% old), preculturing of hESCs for 24 h in feeder-free conditions, e.g., Matrigel™ + MCM, prior to replacing MCM with MCM + 10% old mouse sera, resulted in continuously high BrdU incorporation and Oct4 expression (embryonic microniche + 10% old). BrdU was added for the last 2 h of culture to measure the rate of cell proliferation. Immunodetection of BrdU and Oct4 (both in red) was performed as described in Experimental procedures. Hoechst (blue) labels nuclei. (B) Three independent experiments yielded similar results and were quantified as percentage of BrdU/Oct4 for each condition. * indicates P < 0.001 as compared to ‘old + MCM’. Aged muscle niche inhibits the regenerative potential of hESCs and satellite cells. (A) Immunodetection of a mouse-specific M-cadherin (green) or desmin (red; both human and mouse proteins are detected) revealed that hESCs underwent muscle lineage differentiation when co-cultured with young, but not old myofibers. The myogenic progeny of hESCs appears M-cadherin/desmin (white arrow in young), as opposed to M-cadherin/desmin hESCs that lack myogenic commitment (white arrow in old). M-cadherin/desmin cells are the myogenic progeny of mouse satellite cells (yellow arrows). To assess the effects of secreted factors produced by young vs. old myofibers on the rate of hESC proliferation, transient, 2 h BrdU incorporation was examined in hESCs cultured for 48 h with supernatants produced by heterochronic myofiber explants (See Experimental procedures for details). As compared to young myofiber-derived supernatants (young myofiber supernant), exposure to old myofiber-derived supernatants (old myofiber supernant) inhibited hESCs proliferation, as judged by BrdU immunodetection (red). As expected, the rate of hESCs proliferation was high in control media (shown in Fig. 2). Hoechst (blue) labels nuclei in all experiments. Quantification of desmin/BrdU hESCs in direct myofiber cocultures, or with muscle supernatants, is shown in (B). * indicates P≤ 0.001 as compared to young. (C) Transwell co-cultures between purified young satellite cells and myofibers isolated from uninjured young (young myofiber) and old (old myofiber) muscle demonstrated that satellite cell regenerative myogenic capacity was inhibited by the aged differentiated muscle. Myogenic potential was determined by the ability of satellite cells to generate proliferating desmin myoblasts (immunodetection shown in green) and by rate of proliferation (2 h BrdU incorporation; immunodetection shown in red). (D) Satellite cell regenerative potential was quantified as percentage of desmin/BrdU cells for transwell co-cultures with young or old uninjured myofibers (i.e., RM, resting muscle). n = 3; * indicates P≤ 0.05 as compared to young. Comprehensively, these data establish that the inhibition of stem cell regenerative potential by the aged systemic milieu is conserved between species (mouse vs. human) and cell types (adult vs. embryonic stem cells). As summarized in Table 1, aged mouse sera similarly affected the expression of key molecular identifiers of both embryonic and adult stem cells, e.g., Oct4 in hESCs and Myf5 in mouse ASCs. As expected, adult mouse stem cells did not express Oct4, and hESCs did not express Myf5 in these experimental conditions (not shown). Moreover, aged systemic milieu had similar inhibitory effects on proliferation of hESCs and ASCs, suggesting that not only the regenerative capacity, but also the presence and expansion of stem cells will be significantly restricted in aged organs. Intriguingly, prolonged culturing of hESCs in their preferred in vitro conditions enables generation of an embryonic microniche that antagonizes the inhibitory influences of aged circulatory factors. Conservation of stem cell aging in the systemic environment Quantified results from Figs 1, 2 are summarized and presented as mean percentages from experimental replicates ± SE. Rate of proliferation (BrdU) and cell-fate identifier (Oct4 or Myf5) are shown for both ESCs and ASCs cultured in heterochronic systemic conditions of 10% young (young), 10% old (old) or in 5%+ 5% mouse sera combination (young + old). Results for 5% young mouse sera are very similar to those for 10% young mouse sera and are shown in Fig. 1 (ASCs) and Supplementary Fig. S3 (hESCs). After establishing that the aged systemic niche negatively affects the regenerative capacity of hESCs and of ASCs, we then assessed whether myogenic potential and the rate of cell proliferation would be inhibited in hESCs and ASCs by the aged local muscle niche. Myofibers with associated satellite cells were isolated from young and old injured muscle, and were directly co-cultured with hESCs in typical hESC differentiation medium (DMEM/FBS). Similar to Fig. 1, the myogenic potential in these co-cultures was assayed by the expression of desmin, which is present in both fusion-competent myoblasts and newly formed myotubes. To analyze whether hESCs, mouse myogenic progenitor cells or both could express desmin in direct co-cultures, we costained these cells with a mouse-specific antibody to a myogenic marker, M-cadherin, which does not react with human protein, and a desmin-specific antibody that recognizes both mouse and human proteins. As shown in Fig. 4A, hESCs underwent myogenic differentiation in co-cultures with young myofibers (M-cadherin/desmin mononucleated cells, white arrow in young). These myogenic progeny of hESCs in co-cultures with young myofibers could be of skeletal, smooth or cardiac muscle lineages (Debus et al., 1983; Fischman & Danto, 1985; Schultz & McCormick, 1994). As expected, the young mouse muscle progenitor cells (M-cadherin/desmin) were more advanced in their degree of myogenic differentiation, which was of skeletal muscle lineage, as judged by the formation of large, multinucleated de novo myotubes (yellow arrow in young). In addition to the myogenically differentiated human cells, co-cultures with young myofiber explants also contained some small undifferentiated hESC colonies, as determined by immunoreactivity to a human-specific antibody to the nuclear mitotic apparatus protein, NuMA and Oct4 expression (Supplementary Fig. S4). In contrast, when co-cultured with the aged mouse myofibers, only mouse cells appeared desmin (Fig. 4A, yellow arrow in old). These aged myogenic cells were of skeletal muscle lineage, based on spontaneous generation of multinucleated myotubes (see Fig. 5B) and based on induced differentiation into myotubes in DMEM + 2% horse serum (not shown). Importantly, the myogenic differentiation of hESCs failed in the aged co-cultures (Fig. 4A, white arrow in old). Furthermore, colonies of hESCs in co-cultures with aged myofibers typically differentiated into cells with fibroblast morphology, which lacked Oct4 expression (not shown). Spontaneous production of desmin myogenic cells in control hESC cultures without myofibers, or with young/old mouse sera was less than 0.1% (not shown). In vitro co-culture with hESCs enhanced myogenesis of mouse cells. (A) 1 × 10 hESCs or control hMSCs were co-cultured with 5 × 10 primary mouse myoblasts. hESCs expressing Oct4 (immunodetection shown in red) dramatically enhanced myotube formation of co-cultured mouse myoblasts (immunodetection of eMyHC is shown in green), as compared to co-cultures between mouse myoblasts and human mesenchymal stem cells (Mb + hMSCs) or myoblasts alone (Mb alone). Experiments were carried out in myoblast differentiation medium. Hoechst (blue) labels nuclei throughout this figure. (B) 1 × 10 hESCs or control hMSCs were co-cultured with young or old myofiber-associated satellite cells, as described in Experimental procedures. Co-culture with hESCs (myofiber + hESC), but not hMSCs (myofiber + hMSC) or control medium (DMEM/FBS), greatly enhanced the myogenic potential of both young and old myofiber-associated satellite cells, based on immunodetection of percentage of desmin de novo generated myoblasts and multinucleated myotubes. These experiments were carried out in GM. Shown are myogenic responses of mouse cells only, judged by lack of immunoreactivity to human-specific/hESC-specific antigens, such as NuMA and Oct4; and presence of mouse-specific immunoreactivity, e.g., M-cadherin (not shown). Both young and old myofiber associated satellite cells exhibited considerable myogenic improvement over control conditions. n = 3. In concert with the conservation of inhibitory affects of aged systemic niche, the negative influence of local muscle niche was also found to be conserved in its inhibition of hESC and ASC regenerative responses. Specifically, the myogenic capacity (generation of desmin myoblasts) was inhibited in young satellite cells co-cultured in a transwell system with aged myofibers (Fig. 4B). In addition, hESC and ASC proliferation (BrdU incorporation) was also inhibited by aged differentiated muscle (Fig. 4A,C). These data suggest that not only systemic but also local organ niches would inhibit key stem cell properties, e.g., myogenic capacity and the rate of proliferation in the aged organism. The conserved inhibitory influences of the differentiated muscle niche on hESC and ASC regenerative responses are summarized in Table 2. Conservation of stem cell aging in the local organ niche Quantified results from Fig. 4 are summarized and presented as mean percentages from experimental replicates ± SE. Rate of proliferation (BrdU) and myogenic differentiation (desmin) are shown for both ESCs and ASCs, in the presence of young vs. old differentiated muscle environments (young myofiber or old myofiber). While hESC properties were inhibited by aged differentiated muscle, the myogenic potential of aged satellite cells seemed to be enhanced by co-cultures with hESCs (Fig. 4A). Therefore, we further explored the enhancing and rejuvenating effects of hESCs on myogenic potential in vitro and in vivo, using human mesenchymal stem cells (hMSCs) as a negative control. First, we examined the effects of hESCs on myotube generation by co-culture with primary myoblasts freshly derived from activated-by-injury satellite cells (Conboy et al., 2003). As shown in Fig. 5A (Mb + hESC), primary myoblasts underwent very rapid and robust nascent myotube formation, when co-cultured with hESCs for 48 h in myoblast differentiation medium. Namely, remarkably large fused myotubes containing approximately 50–70 nuclei formed around hESCs colonies (Fig. 5A). In contrast, when co-cultured with hMSCs, myotube formation was no greater than in myoblast cultures alone (Fig. 5A, Mb + hMSC and Mb alone). Encouraged by these data, we analyzed the myogenic potential of young and old satellite cells co-cultured with hESCs for 48 h. As shown in Fig. 5B, hESCs conferred a much-enhanced myogenic capacity on both young and, importantly, old myofiber-associated satellite cells (rapid formation of desmin myogenic cells, many of which formed de novo multinucleated myotubes). Control co-cultures of these satellite cells with hMSCs displayed no enhanced myogenicity. In summary, while the myogenic potential (production of desmin fusion-competent cells) was more pronounced in young vs. old myofiber-associated satellite cells under all experimental conditions, a finding that is consistent with previous data (Conboy et al., 2003), a clear increase in myogenic potential of old satellite cells was noted in co-cultures with hESCs, as compared to control cultures devoid of hESCs (Fig. 4A,B). Interestingly, in addition to the rejuvenating effects of direct co-cultures shown in Fig. 5, soluble factors present in hESC-conditioned culture supernatants were also able to enhance myogenesis of aged satellite cells (Supplementary Fig. S5). Thus, in agreement with the notion that an established embryonic microniche antagonizes the inhibitory effects of the aged environment on stem cell responses (Fig. 3), the hESC-produced factors enhanced myogenic capacity of even old mouse satellite cells. Establishing that hESC-produced factors enhance adult myogenesis and rejuvenate the regenerative capacity of even aged satellite cells in vitro prompted us to examine whether the regeneration of old injured muscle will be improved by hESC transplantation in vivo. Additionally, based on the data shown above, we speculated that even if the host's repair capacity is improved, hESCs themselves will not be efficiently maintained or expanded in the context of old systemic and local organ environments, and will not directly contribute to the repair of aged skeletal muscle. To test these hypotheses, we injected 5 × 10 hESCs or control hMSCs into the tibialis anterior (TA) and gastrocnemius muscles of young and old mice at 24 h after cardiotoxin-induced injury, when activation/proliferation of endogenous satellite cells normally begins (Conboy et al., 2003, 2005; Wagers & Conboy, 2005). To avoid immune response against hESC antigens, mice were immunosuppressed using FK506 (Ito & Tanaka, 1997; Dumont, 2000). Muscle was isolated 5 days post-injury, when nascent differentiated myofibers normally replace the damaged tissue (Conboy et al., 2003), and 10 µm cryosections were analyzed for the success in tissue repair using hematoxylin and eosin (H&E) histochemistry and eMyHC immunodetection. H&E analysis reveals newly formed myofibers, based on their smaller size and centrally located nuclei. Additionally, de novo myofibers in the damaged area appear positive for eMyHC, while undamaged myofibers remain negative. As shown in Fig. 6A and quantified in 6B, injection of hESCs significantly enhanced regeneration of skeletal muscle. Remarkably, this positive embryonic effect was especially pronounced in old tissue. Skeletal muscle regeneration following hESC transplantation is a balance between the inhibitory influence of aged niches and the rejuvenating effects of hESCs. Young and old tibialis anterior and gastrocnemius muscles were injured by cardiotoxin injection. hESCs or hMSCs were transplanted at the site of injury and were analyzed by cryosectioning at Day 5 after injury (as described in Experimental procedures). (A) Newly regenerated myofibers were detected using eMyHC-specific antibody (green) and staining with H&E. In H&E staining, newly regenerated areas contain smaller, immature myofibers with centrally located nuclei. Uninjured myofibers are much larger, by comparison, with peripherally restricted nuclei. Poorly regenerated areas lack new myofibers and contain areas of fibrosis and inflammation. eMyHC immunodetection is specific for regenerating areas of muscle only. Both assays showed dramatic enhancement of muscle regeneration in ‘old + hESC’ vs. ‘old + hMSC’. Regeneration improvement was also seen in young + hESC, as compared to young + hMSC. (B) Quantification of muscle regeneration was performed by analyzing the density of newly formed myofibers per mm of injury site, which is the volume that typically covers the whole injured area. Multiple, 10 µm H&E sections were examined through the entire volume of injury in multiple, independently injured muscles. n = 20; * indicates P < 0.001 (‘old + hMSC’ compared to young + hMSC and ‘old + hMSC’ compared to ‘old + hESC’. (C) H&E and immunofluoresence staining for Oct4, and a human-specific antibody to NuMA, revealed the failure of hESCs to expand or persist in old, but the presence of hESCs in young muscle at 5 days post-transplantation. Hoechst (blue) labels nuclei. Importantly, such enhanced and rejuvenated muscle repair stems from an indirect induction, as hESCs themselves (or control hMSCs) did not physically contribute to the mouse myofibers, as judged by near absence (less than 0.1%) of human-specific NuMA nuclei in de novo desmin myofibers, analyzed through multiple injury sites. An example of one regenerated myofiber from young muscle injected with hESCs, with NuMA nucleus in a field of NuMA/desmin mouse myofibers, is shown in Supplementary Fig. S6. No such NuMA/desmin myofibers were detected in aged regenerated muscle (not shown). In agreement with the in vitro data, establishing that aged systemic and local niches inhibit hESC proliferation and Oct4 expression (Figs 2 and 4 and Supplementary Fig. S2), hESCs failed to expand or even persist in old muscle, as judged by the absence of NuMA/Oct4 hESC-derived cells in the aged tissue. In contrast, colonies of Numa/Oct4 hESC-derived cells that did not undergo myogenic differentiation were easily detected in young regenerating muscle (Fig. 6C). This finding validates several technical aspects of these experiments, and confirms the contrasting effects of young and old systemic and local organ niches on hESC self-renewal. These data further confirm and extrapolate our findings and demonstrate that when exposed to both aged systemic and local organ niches, hESCs fail to persist and do not contribute to tissue repair directly. At the same time, these embryonic cells indirectly but significantly improve the repair of aged injured muscle in vivo. The data presented here establish for the first time that both the local environment of old differentiated organ, e.g., skeletal muscle and the systemic milieu dramatically affect the regenerative potential of both hESCs and mouse post-natal myogenic progenitor cells. Not only are the factors promoting myogenic differentiation and proliferation of hESCs likely to become depleted with age, but the aged systemic and local organ niches are likely to contain dominant inhibitors of ASC and hESC regenerative potential (Figs 1, 2, and 4, summarized in Tables 1 and 2). Importantly, the similar inhibitory effects of old mouse serum and old myofibers on satellite cell (Figs 1 and 4C) and hESC (Figs 2 and 4A) proliferation and regenerative capacity suggest the conservation of elements in age-specific extrinsic regulatory mechanisms between evolutionarily distinct species and stem cell types. Additionally, a similarity in the inhibitory properties between systemic and local organ niches is also of interest and may indicate that molecules produced by old tissues have circulatory/endocrine activity; and/or that age-specific systemic inhibitory components become deposited in the old tissues. Humans display broad genetic polymorphisms and behavioral variations, which makes the identification of age-specific molecular changes complicated. In contrast, laboratory mice are genetically and environmentally controlled. Establishing that age-specific signals, regulating stem cell responses, are evolutionarily conserved and soluble enables the formation of rational approaches for the identification and characterization of the inhibitors involved, and for revealing the precise timing of their first appearance in serum and differentiated tissues with advancing age. Significantly, these experiments have also revealed that not only are hESCs able to protect themselves against the negative influences of aged mouse sera (Fig. 3), but these cells also produce factors that dramatically enhance the myogenic capacity of primary myoblasts and young and old satellite cells (Fig. 5), and also significantly improve repair of young and old injured muscle in vivo (Fig. 6). Identification of these embryonic factors would allow us to potentially enrich the arsenal of therapeutic tools for combating age-specific degenerative disorders. The interactions between hESCs and heterochronic differentiated niches, initially identified in vitro, have been confirmed by in vivo experiments. Namely, while the regenerative capacity, or presence, of hESCs is greatly restricted in aged, as compared to young skeletal muscle (where transplanted cells experience both old systemic and local environments), embryonic cells indirectly enhance and rejuvenate muscle repair when introduced at the time of muscle stem cell activation in the host, e.g., at Day 1 after the injury (Fig. 6). It remains to be determined whether the percentage of hESCs direct contribution to desmin myofibers in young muscle will be increased by transplanting these cells at a different time-point after muscle injury, e.g., at Days 3–5 (as in co-cultures with myofibers pre-injured for 3 days, Fig. 4A). In any case, the virtual lack of hESC and hMSC direct contribution to the newly regenerated skeletal muscle, when small numbers of these cells were injected into injured tissue, is completely consistent with the body of previous data demonstrating that myofiber-associated satellite cells conduct rapid and robust muscle repair and greatly outnumber injected human cells (Collins et al., 2005; Wagers & Conboy, 2005); that compared to muscle-specific satellite cells, the myogenic differentiation of hESCs in vitro remains very small (Fig. 5, Table 2), and that control hMSCs are not normally myogenic unless these cells overexpress exogenous constitutively active domain of Notch (Dezawa et al., 2005). Intriguingly, the failure of hESCs to strive in old skeletal muscle might represent a therapeutically desirable outcome. For example, while in young tissue hESC derivatives putatively would go on to produce teratomas, it is unlikely that teratoma formation would occur after hESC transplantation into aged skeletal muscle. Thus, the indirect beneficial effects of hESCs on tissue repair are unlikely to be compromised by the oncogenic properties of these embryonic cells in the context of old skeletal muscle. Comprehensively, the results of this work increase our understanding of aging as a process, reveal evolutionary conserved age-specific interactions between stem cells and their differentiated niches, and suggest novel therapeutic approaches for improving the regenerative responses of endogenous or transplanted stem cells in old individuals. Young (2–3 months), C57-BL/6 male mice were obtained from pathogen-free breeding colonies at Jackson Laboratories (Bar Harbor, ME, USA). Aged 22–24 months C57-BL/6 male mice were obtained from the National Institute on Aging (NIH). Animals were maintained in the North-West Animal Facility of the University of California, Berkeley, CA, USA, and handled in accordance with the Administrative Panel on Laboratory Animal Care at UC Berkeley. Myofiber cultures, in which satellite cells were activated by in vivo injury, were set up as previously described (Conboy & Rando, 2002; Conboy et al., 2005). Briefly, mice were injured by direct injection with 5 ng cardiotoxin (CTX-1) (Sigma, St Louis, MO, USA) into the tibialis anterior and gastrocnemius muscles using a 28-gauge needle. After 1–5 days post-injection, injured or uninjured muscle tissue was dissected out. Once isolated, whole muscle was prepared for cryosectioning (see below) or myofiber fragments were obtained from hind limb muscles by enzymatic digestion (see below), trituration, and multiple sedimentation and washing procedures. Additionally, blood was collected from mice for the isolation of sera. Briefly, blood cells were coagulated at 37 °C for 15’ and then were centrifuged repeatedly at 5900 g, 4 °C in a microfuge for 3’ to isolate sera. Mixtures of young and old sera were made 1 : 1. For example, in 5%+ 5% conditions, 50 µL of young and 50 µL old serum were added to 900 µL of culture medium (Opti-MEM or MCM, see co-culture procedures below). Explant and primary cell cultures were generated from C57-BL/6 mice, as previously described (Conboy & Rando, 2002; Conboy et al., 2003). Dissected gastrocnemius and tibialis anterior muscles underwent enzymatic digestion at 37 °C in DMEM (Invitrogen)/Pen-Strep (Invitrogen)/0.2% Collagenase Type IIA (Sigma) solution. Isolated fibers were resuspended in GM (Ham's F10 nutrient mixture (Mediatech, Inc., Herndon, VA, USA), 20% FBS (Mediatech), 5 ng mL bFGF (Chemicon, Temecula, CA, USA) and 1% Pen-Strep, and cultured on ECM-coated (BD Biosciences, San Jose, CA, USA) plates (diluted 1 : 500 in PBS). Cultures of primary myoblasts were derived from isolated fibers, through repeated passaging, and were maintained in GM. Myoblast differentiation medium [DMEM, supplemented with 2% horse serum (Mediatech)] was used to promote rapid formation of myotubes from cultured myoblasts (Morgan & Partridge, 2003). The federally approved hESC line, H7 (NIH no. WA07, obtained from WiCell Research Institue, Madison, WI, USA), was used in accordance with the UC Berkeley and UC San Francisco Committee on Human Research guidelines, and in accordance with NIH guidelines. To propagate hESCs, routine culturing and maintenance was performed using standard in vitro conditions for both feeder-dependent and feeder-free cultures (Geron Corporation, 2002). Briefly, hESCs grown on MEFs were cultured in standard hESC medium [Knockout™ DMEM, 20% KSR, 1% NEAA, 1 mm l-glutamine (Invitrogen), 0.1 mmβ-mercaptoethanol (Sigma)] and were supplemented with 4 ng mL hbFGF (Invitrogen). Feeder-free hESC cultures were maintained in MEF-conditioned hESC medium (MCM), 4 ng mL hbFGF. Differentiation medium for hESCs (DMEM/FBS) was made by replacing KSR with 20% FBS (Hyclone, Logan, UH, USA). hMSCs were maintained in mesenchymal stem cell GM, MSC-GM™ and were cultured according to supplier recommendations (Cambrex Walkersville, MD, USA). hESCs and hMSCs were typically seeded onto chambered slides coated with a 3% GFR Matrigel™ (BD Biosciences) substrate in PBS. Cells were typically incubated for 48 h at 37 °C, 5% CO2, under the various experimental conditions employed, then were fixed with 70% EtOH/PBS at 4 °C. hESCs and hMSCs were analyzed 24–48 h after experimental treatments, during which no apoptosis-related differences in cell numbers were observed. Heterochronic systemic cultures were established by culturing myofiber explants (in GM) or hESCs (in MCM) in the presence of young, old or young + old sera for 48 h (Figs 1 and 2 and Supplementary Figs S1–3). In such cultures, hESCs were passaged immediately prior to sera exposure. In contrast, preculturing of hESCs for 24 h in MCM, prior to replacing MCM with MCM + 10% old mouse sera was done for embryonic microniche experiments (Fig. 3). For heterochronic local organ niche cultures, hESCs were co-cultured directly with myofiber explants for 48 h in GM, or were cultured in the presence of supernatants derived from cultured myofiber explants for 48 h (Figs 4A and 5). Specifically, 1 × 10 hESCs or control hMSCs were co-cultured with identical volume, e.g., 100 µL, of young or old myofiber fragments with their associated satellite cells (Fig. 5). In experiments shown in Supplementary Fig. S5, culture-conditioned supernatant produced by hESCs grown in MCM was used as a medium in which 1 × 10 of myofiber-associated young or old satellite cells were cultured for 48 h. In direct co-cultures, mouse vs. human cells were distinguished by immunodetection with human-specific/hESC-specific and mouse-specific antibodies (Supplementary Fig. S4 and see below). To prepare muscle supernatants, explants were cultured for 24 h in GM and cellular debris was removed from conditioned media by multiple rounds of centrifugation. The absence of cells was confirmed by microscopic examination. To mimic the local organ niche for satellite cell assays (Fig. 4B), 1.0 µm transwell (Corning, NY, USA) co-cultures of uninjured explants with activated satellite cells were established. Activated-by-injury (24 h post-injury) satellite cells were seeded onto ECM-coated 12-well plates in Opti-MEM (Invitrogen) and 5% FBS. Transwells were placed over satellite cells and contained isolated myofiber explants from uninjured young or old muscle (i.e., resting muscle). Satellite cells were cultured for 72–96 h in the presence of myofiber explants and were fixed for immunodetection, as described above. hESCs were grown on MEFs and expanded in 6-well plates. Cells were treated with 1 mg mL Collagenase Type IV (Invitrogen) for 5–10 min, were washed and then incubated with 0.5 mg mL Dispase (Invitrogen) to lift only human cell colonies. Isolated hESCs were washed several times and resuspended in 100 µL hESC medium. Similarly, hMSCs were expanded in 6-well plates, lifted with Trypsin/EDTA (Invitrogen), washed and resuspended in 100 µL hESC medium. Approximately 5 × 10 hESCs or hMSCs were injected into 24 h post-injured gastrocnemius and tibialis anterior muscles of young and old mice, using a 21-gauge needle. Immunosuppression of animals was achieved by intraperitoneal injection of 1 mg kg FK506 (Sigma) at 48 h prior to cell transplantation, and on each day following transplantation. To assay the affects of heterochronic local and systemic environments on stem cell regenerative potential, hESC, hMSC, and myofiber-derived precursor cell cultures were fixed with 70% EtOH/PBS at 4 °C, and were analyzed by indirect immunofluorescence. Combinations of antibodies were used to co-stain cultures and histosections, in order to determine the percentages of cells that proliferated or differentiated and to distinguish hESCs from mouse cells. Antibodies to the myogenic transcription factors, Myf5/Pax7, the intermediate filament protein, desmin, and the marker of newly formed myotubes, eMyHC, were used to reveal commitment to myogenic differentiation. Cell commitment to this differentiation program was assessed by the efficiency of myotube formation, estimated by the number of nuclei per myotube. Ki67, a cell cycle related nuclear protein consistently absent in quiescent cells, was used as a marker for proliferation. Whereas Ki67 appears in all active phases of the cell cycle, BrdU staining allowed exclusive detection of cells in S-phase, thereby enabling accurate quantification of DNA synthesis. In select cultures, 10 µm BrdU was added for 2 h prior to fixation. BrdU-specific immunostaining required nuclear permeabilization with treatment of 4N HCl. hESCs were distinguished from mouse cells by using a species-specific antibody to the cell-surface marker M-cadherin for murine and the nuclear marker NuMA for human cells. Antibodies to Oct4 were used as a marker of hESC self-renewal/pluripotency. Following permeabilization in PBS, +1% FBS, +0.25% Triton X-100, cells were incubated with primary antibodies (concentration determined as per manufacturer's recommendations) for 1 h at room temperature in PBS, +1% FBS, washed several times, and then incubated with fluorophore-conjugated, species-specific secondary antibodies (diluted 1 : 500 in PBS + 1% FBS) for 1 h at room temperature. For histological analysis, dissected muscle was treated in a 25% sucrose/PBS solution, frozen in OCT compound (Tissue Tek) and cryosectioned at 10 µm. Immunostaining was performed in the manner described above, or H&E staining of cryosections was performed. Nuclei were visualized by Hoechst staining for all immunostains. Samples were analyzed at room temperature by using a Zeiss Axioscope 40 fluorescent microscope, and imaged with an Axiocan MRc camera and AxioVision software. All images depict identical microscope fields at ×20 magnification, unless otherwise noted. Antibodies to Oct4 (ab18976), BrdU (BU1/75 (ICR1), and Ki67 (ab15580) were purchased from Abcam (Cambridge, MA, USA). Antibody to M-cadherin (clone 12G4) was acquired from Upstate Biotechnology (Lake Placid, NY, USA), and NuMA antibody (Catalog number NA09L) from EMD Biosciences (San Diego, CA, USA). Antibody to developmental eMyHC (clone RNMy2/9D2) was acquired from Vector Laboratories (Burlingame, CA, USA). Myf5 (GTX77876) and Pax7 (GTX77888) antibodies were obtained from GeneTex (San Antonio, TX, USA). Desmin antibodies (clone DE-U-10 and Catalog number D8281), BrdU labeling reagent and FK506 (Catalog number F4679) were obtained from Sigma. Fluorophore-conjugated secondary antibodies (Alexa Fluor) were obtained from Molecular Probes (Eugene, OR, USA). A minimum of three replicates were undertaken for each experimental condition. Quantified data are presented as means ± SE. Significance testing was performed using one-way analysis of variance (anova) to compare data from different experimental groups. P values of < 0.05 were considered as statistically significant.
PMC12819543
Evaluation of agmatine’s anti-cancer efficacy in Caco-2 colorectal adenocarcinoma cells
This study aimed to evaluate the potential effects of agmatine on cell viability, migration, invasion, apoptosis, and the expression of the ABCB1, ABCC1, and ABCG2 genes in the Caco-2 colon cancer cell line. Agmatine efficacy was assessed thruogh proliferation, migration, and invasion assays at various concentrations. The apoptotic index was determined using apoptosis-related markers (Bax, Bcl-2, Csp-3) via apoptosis assays, quantitative real-time PCR (qRT-PCR), and Western blot analysis. Expression levels of the ABCG2, ABCB1, and ABCC1 genes were measured by qRT-PCR in agmatine-treated Caco-2 cells. Oxidative stress markers, including glutathione peroxidase (GPx) and catalase (CAT), were evaluated by qRT-PCR. Cell viability analysis revealed that agmatine exerted its most pronounced effects at 72 h, with significant reductions at concentrations of 6, 7.3, and 9 mM in Caco-2 cells and 6, 6.25, and 9 mM in L929 cells (p < 0.05). At these concentrations, migration and invasion assays showed dose-dependent decreases in cell motility and invasiveness in Caco-2 cells. Apoptosis analysis demonstrated a significant increase in the apoptotic index with rising agmatine concentrations. Significant decreases in GPx and CAT were observed in all three agmatine-treated Caco-2 groups compared to untreated controls (p < 0.01). However, the expression levels of ABCG2, ABCB1, and ABCC1 showed no significant changes following agmatine treatment (p > 0.05). These findings indicate that agmatine exerts antiproliferative, anti-migratory, anti-invasive, and pro-apoptotic effects in Caco-2 colon cancer cells, potentially through the modulation of apoptosis- and oxidative stress–related pathways. The lack of significant impacts on ABC transporter gene expression suggests that agmatine may be a promising candidate molecule for further translational studies in colorectal cancer.Worldwide, the incidence of colorectal cancer (CRC) is increasing each year, and is becoming a significant public health concern . Currently, chemotherapy protocols based on the combination of fluorouracil (5-FU) and oxaliplatin (OXA) constitute one of the backbones of CRC treatment. Nevertheless, because currently used therapeutic agents also penetrate healthy tissues, patients may experience many kinds of side effects . Yet, all these treatment options may not be equally effective for every patient and can entail serious side effects as well as resistance-related challenges . Consequently, it is of great importance to explore new therapeutic and supportive agents for CRC therapy. Polyamines can interact with DNA, RNA, ATP, proteins, and phospholipids under physiological conditions . Polyamine synthesis is essential for sustaining the continuous proliferation characteristic of cancer cells. Notably, the genes encoding two rate limiting enzymes in polyamine biosynthesis, ornithine decarboxylase (ODC; encoded by ODC1) is often dysregulated in various cancers . Agmatine has been reported to confer protective effects against ischemic injury and chronic neuropathic pain in mammals . It has also been identified as having neuroprotective and antioxidant properties in pathophysiological conditions such as anxiety disorders and depression [7, 8]. In contrast to the proliferative effects of the aforementioned polyamines, the polyamine agmatine has been shown to reduce ODC activity and the intracellular uptake of polyamines . Furthermore, emerging evidence suggests that agmatine exerts antiproliferative and pro-apoptotic effects in various cancer models such as prostate hepatocellular carcinoma and glioma primarily by interfering with polyamine metabolism and modulating key signaling pathways [9–11]. In colorectal cells, agmatine has been shown to inhibit cell proliferation and induce cell cycle arrest, potentially through the suppression of ODC activity and the disruption of polyamine homeostasis [11, 12]. These findings highlight agmatine’s potential to serve as a selective therapeutic against in cancer, particularly in tumors with high polyamine requirements, while sparing normal cells in which it exhibits protective and antioxidant properties . Collectively, these findings position agmatine as a promising therapeutic candidate for colorectal cancer, warranting further investigation into its clinical potential . One of the principal mechanisms underlying resistance to anticancer drugs is multidrug resistance (MDR), in which cancer cells overexpress the ATP-binding cassette (ABC) transporter proteins that actively secrete various chemotherapeutic agents from the cell . In particular, the transporters encoded by ABCB1 (MDR1/P-gp), ABCC1 (MRP1), and ABCG2 (BCRP) play crucial roles in pumping numerous chemotherapy drugs out of the cell, thus limiting intracellular drug accumulation and enabling tumor cell survival [16, 17]. This study aimed to examine whether agmatine has potential regulatory effects on cell viability, migration and invasion, apoptosis, reactive oxygen species (ROS), and the gene expression levels of ABCB1, ABCC1, and ABCG2 transporters, which play important roles in drug resistance in the human colon carcinoma cell line Caco-2. The Caco-2 and L929 cell lines were obtained from the American Type Culture Collection (Manassas, VA, USA). Caco-2 cells were cultured in high-glucose DMEM (EcoTechBio, Türkiye) mixed with 10% fetal bovine serum (FBS) Capricorn (Ebsdorfergrund, Germany) and 1% penicillin/streptomycin (Gibco Waltham, USA) at 37 °C and 5% CO₂. All experiments were performed using cells at passages 17–27. Experiments were performed in triplicate. Untreated Caco-2 cells were used as the control group in all experiments. Caco-2 and L929 cells were seeded in 96-well plates (20 × 10³ cells/well, 100 µl medium) and incubated for 24 h. Caco-2 cells were treated with agmatine at concentrations of 1, 3, 6, 9, 12, 15, 18, and 21 mM, while L929 cells were treated with 6, 9, 12, 15, 18, and 21 mM for 24, 48, and 72 h. Cell viability was assessed using the MTT assay. Briefly, 5 mg/mL MTT solution was added to the cells and incubated for 4 h. The medium was then removed, and 100 µL of DMSO was added to each well and incubated for 30 min. Absorbance was measured at 570 nm. All experiments were performed in triplicate. Transwell assays were performed using 8 μm pore inserts (Corning, NY, USA). For migration, 2 × 10⁴ Caco-2 cells were seeded in the upper chamber with serum-free medium and treated with agmatine (6, 7.3, and 9 mM) for 72 h, while 750 µl supplemented medium was added to the lower chamber. Non-migrated cells were removed, and migrated cells were fixed (4% paraformaldehyde, 100% methanol), stained with 0.1% crystal violet, and counted in three random fields under an inverted microscope. For invasion, 2 × 10⁴ cells were seeded in Matrigel-coated upper chambers, incubated for 24 h, and then treated with agmatine (6, 7.3, and 9 mM) for 72 h. The lower chamber contained medium with 10% FBS as a chemoattractant. Non-invasive cells were removed, and invasive cells were fixed, stained, and quantified as above. Caco-2 and L929 cells were seeded in 96-well plates (2 × 10⁴ cells/well) and incubated for 24 h. After PBS wash, Caco-2 cells were treated with agmatine (6, 7.3, and 9 mM) and L929 (6 mM, 6.25 mM, and 9 mM) for 72 h. Cells were then stained with DAPI (1.8 µl, 1.11 mg/ml) and propidium iodide (PI;1.17 µl, 0.85 mg/ml) for 25 min at 37 °C in the dark, washed with PBS, and analyzed for apoptosis using an fluorescence microscope. Caco-2 cells (2 × 10⁵/well) were seeded in 6-well plates and incubated for 24 h, then treated with agmatine (6, 7.3, and 9 mM) for 72 h. Cells were harvested with 0.25% trypsin-EDTA, pelleted (300 rpm, 5 min), resuspended in 5% TCA, and incubated at 4 °C for 15 min. After centrifugation (12,000×g, 10 min, 4 °C), supernatants were collected. For the assay, aliquots were mixed with 2% ninhydrin (Bio Basic Inc., Canada) and heated at 100 °C for 5 min, and absorbance was measured at 440 nm. Total RNA was extracted using Trizol reagent (BioBasic, Markham, Canada) according to the manufacturer’s instructions. The concentration and purity of RNAs were determined with spectrophotometry using NanoDrop (Thermo Fisher Scientific, Maryland, USA). cDNA synthesis was conducted using the OneScript Plus cDNA Synthesis Kit (Applied Biological Materials Inc., Richmond, BC, Canada) according to the manufacturer’s instructions. Quantitative real-time PCR (qRT-PCR) was performed using the BlasTaq™ 2X qPCR Master Mix Kit (Applied Biological Materials Inc., BC, Canada) according to the manufacturer’s instructions. The relative expression levels of ABCC1, ABCB1, ABCG2, Bax, Bcl-2, Caspase-3 (Csp-3), glutathione peroxidase (GPx) and catalase (CAT), with beta-actin (SenteBioLab, Ankara, Türkiye) as an internal control, were evaluated with qRT-PCR in a T100™ Thermal Cycler (Bio-Rad, USA). Caco2 and L929 cell lines were seeded with a density of 3 × 10 cells/well in a 6-well plate. The following day, Caco2 cells were treated with 6 mM, 7.3 mM, and 9 mM agmatine, and L929 cells were treated with 6 mM, 6.25 mM, 9 mM agmatine. After 72 h, whole-cell lysates were prepared using the Mammalian Total Protein Extraction Kit (SERVA, Germany), and protein concentrations were determined using the BCA Protein Assay Macro Kit (SERVA, Germany) according to the manufacturer’s instructions. Thirty µg proteins were run on 12% SDS-PAGE and transferred by using the semi-dry system to a 0.22-µm PVDF membrane. Membranes were blocked in 5% non-fat milk for 1 h and incubated with primer antibodies against Bcl-2 (CST, USA), BAX (CST, USA), cleaved Csp-3 (CST, USA) and GAPDH (AbClonal, USA) overnight at 4 °C. The following day, membranes were washed three times with 1X TBS-T and probed with HRP-conjugated secondary antibodies (Anti-Rabbit, IgG, AbClonal, USA) for 2 h at room temperature. After incubation, membranes were washed three times with 1X TBS-T and protein bands were detected with Clarity ECL substrate (Bio-Rad, USA) using the Chemi-Doc MP Imaging System (Bio-Rad, USA). Signals were normalized to housekeeping GAPDH, and intensities were calculated using Image Lab Software (version 6.1). All performed experiments were repeated three times, and the graphical presentations were performed using GraphPad Prism 10.0 (San Diego, CA, USA). Data are presented as the means ± standard deviation and were analyzed using SPSS 22.0 software (Chicago, IL, USA). Target gene expression was normalized to the corresponding B-actin value, and ΔCt values were calculated accordingly. The normality of ΔCt distributions was assessed using the Shapiro-Wilk test. As the ΔCt values did not deviate significantly from a normal distribution, parametric statistical tests were applied. Group comparisons were performed using one-way analysis of variance. When a significant overall group effect was detected, post hoc pairwise comparisons were conducted using Tukey’s honestly significant difference test. Statistical analyses were carried out on ΔCt values, whereas relative gene expression levels were calculated using the 2⁻ΔΔCt method for graphical presentation and descriptive purposes. A two-sided p value of < 0.05 was considered statistically significant. A schematic representation of the anti-tumor effects of agmatine in Caco-2 cells is illustrated in Fig. 1A. As shown in Fig. 1B, the viability of both Caco-2 and L929 cells decreased in a concentration-dependent manner with increasing agmatine concentrations. The half-maximal inhibitory concentration (IC₅₀) of agmatine at 72 h were determined to be 7.3 mM for Caco-2 cells and 6.25 mM for L929 cells. Overall, a significant increase in inhibition rate was observed in Caco-2 cells following agmatine treatment. Therefore, agmatine doses corresponding to concentrations below and above the IC₅₀ values (6, 7.3, and 9 mM for Caco-2 cells; 6, 6.25, and 9 mM for L929 cells) were selected for subsequent experiments. Fig. 1C presents the logarithmic representation of the MTT assay results for the cells.The effects of agmatine on the migration and invasion of Caco-2 cells were evaluated using transwell migration and Matrigel invasion assays. As shown in Fig. 2A the number of Caco-2 cells migrating through the membrane into the lower chamber was significantly and dose-dependently reduced in the agmatine-treated groups compared with control groups. When Caco-2 cells were treated with agmatine at concentrations of 6, 7.3, and 9 mM for 72 h, their relative migration rates decreased from 61.2% to 46.3% and further to 30.02%, respectively (Fig. 2A; p < 0.05). Similarly, agmatine was observed to inhibit Caco-2 cell invasion in a dose-dependent manner. At 72 h, treatment with 6, 7.3, and 9 mM agmatine resulted in a reduction of the relative invasion rate from 54.6% to 33.4% and further to 18.8%, respectively (Fig. 2B; p < 0.05). Collectively, both assays demonstrated that the numbers of migrated and invaded Caco-2 cells were significantly decreased in a dose-dependent manner in agmatine-treated groups (6, 7.3, and 9 mM) compared with untreated Caco-2 cells (control). The effects of agmatine on the apoptosis of Caco-2 and L929 cells were evaluated by fluorescence microscopy using DAPI and PI staining. DAPI/PI staining revealed that after agmatine treatment with 6, 7.3, or 9 mM in Caco-2 and 6, 6.25, or 9 mM in L929 for 72 h, apoptosis rates were increased dose-dependently when compared to the control cells (Fig. 3A). In Caco-2 cells treated with 6, 7.3, or 9 mM agmatine for 72 h, the apoptotic indices were 46% (ns), 50% (p < 0.05), and 36% (ns), respectively, compared to the untreated control. In contrast, agmatine treatment in L929 cells induced a dose-dependent increase in the apoptotic index, reaching 40% (ns), 46% (ns), and 66% (p < 0.001) at 6, 6.25, and 9 mM concentrations, respectively (Fig. 3B). Fig. 1(A) Schematic figure of anti-tumor effect of agmatine in Caco-2 cells. (Figure was created using BioRender under an academic license). (B) Cell viability (%) of agmatine on the Caco-2 and L929 cells. Caco-2 was treated with different concentrations of agmatine 1, 3, 6, 9, 12, 15, 18, and 21 mM and L929 cells were treated with 6, 9, 12, 15, 18, and 21 mM for 24, 48, and 72 h. (C) Logarithmic representation of the MTT assay results for the L929 and Caco-2. Error bars show standard error. * also indicates that statistically significant differences (*p < 0.05, **p < 0.01, ***p < 0.001) (A) Schematic figure of anti-tumor effect of agmatine in Caco-2 cells. (Figure was created using BioRender under an academic license). (B) Cell viability (%) of agmatine on the Caco-2 and L929 cells. Caco-2 was treated with different concentrations of agmatine 1, 3, 6, 9, 12, 15, 18, and 21 mM and L929 cells were treated with 6, 9, 12, 15, 18, and 21 mM for 24, 48, and 72 h. (C) Logarithmic representation of the MTT assay results for the L929 and Caco-2. Error bars show standard error. * also indicates that statistically significant differences (*p < 0.05, **p < 0.01, ***p < 0.001) Fig. 2Agmatine inhibited invasion and migration of Caco-2 cells. (A)After 72 h of agmatine treatment, the effect on the migration ability of Caco-2 cells was evaluated by transwell migration assay. (B) The relative migration ratios of migration cells. (A) After 72 h of agmatine treatment, the effect on the invasion ability of Caco-2 cells was evaluated by transwell invasion assay. (B) The relative invasion ratios of migration cells. Data were obtained from triplicate experiments and expressed as the means ± standard deviation. *p < 0.05 Agmatine inhibited invasion and migration of Caco-2 cells. (A)After 72 h of agmatine treatment, the effect on the migration ability of Caco-2 cells was evaluated by transwell migration assay. (B) The relative migration ratios of migration cells. (A) After 72 h of agmatine treatment, the effect on the invasion ability of Caco-2 cells was evaluated by transwell invasion assay. (B) The relative invasion ratios of migration cells. Data were obtained from triplicate experiments and expressed as the means ± standard deviation. *p < 0.05 Fig. 3Agmatine induced apoptosis in Caco-2 and L929 cells. (A) The apoptosis of Caco-2 cells (6, 7.3, and 9 mM) and L929 (6, 6.25 and 9 mM) after agmatine treatment was observed by DAPI and PI staining for 72 h. (B) The apoptotic index (%) values of Caco-2 and L929 cells. (C) Relative expression levels of Bax, Bcl-2 and Csp-3 in agmatine-treated Caco-2 and L929 cells versus control groups. Data were obtained from triplicate experiments and expressed as the means ± standard deviation. *p < 0.05, **p < 0.01, ***p < 0.001 Agmatine induced apoptosis in Caco-2 and L929 cells. (A) The apoptosis of Caco-2 cells (6, 7.3, and 9 mM) and L929 (6, 6.25 and 9 mM) after agmatine treatment was observed by DAPI and PI staining for 72 h. (B) The apoptotic index (%) values of Caco-2 and L929 cells. (C) Relative expression levels of Bax, Bcl-2 and Csp-3 in agmatine-treated Caco-2 and L929 cells versus control groups. Data were obtained from triplicate experiments and expressed as the means ± standard deviation. *p < 0.05, **p < 0.01, ***p < 0.001 The expression levels of the apoptosis-related genes, Bax, Bcl-2 and Casp-3 were evaluated in Caco-2 cells across the control, 6 mM, 7.3 mM, and 9 mM groups using ΔCt values normalized to B-actin (Fig. 3C). For Bax, the mean ΔCt value in the control group was 0.35 ± 0.02, compared with 1.01 ± 0.19 in the 6 mM group, 1.18 ± 0.07 in the 7.3 mM group, and 1.12 ± 0.13 in the 9 mM group, indicating a significant increased difference in the 6 mM, 7.3 mM, and 9 mM groups compared to the control (p < 0.001, p < 0.01, p < 0.01). When the 6 mM groups were compared to the control group, the expression levels of Bcl-2 increased in Caco-2. However, expression in the 7.3 and 9 mM groups decreased relative to the 6 mM group cells. The overall difference between groups was statistically significant without 9 mM, compared with the control (p < 0.01, p < 0.01, ns). The mean ΔCt value of Csp-3 was increased in the 7.3 mM, and 9 mM groups compared to the control group, with a significant group effect (p < 0.05, p < 0.001). qRT-PCR analysis revealed dose-dependent upregulation of Bcl-2 and downregulation of Bax and Csp-3 expression in agmatine-treated (6, 6.25, 9 mM) L929 cells, resulting in significant differences relative to the control (Fig. 3C). To investigate the effects of agmatine-induced ROS formation in Caco-2 and L929 cells, GPx and catalase (CAT) mRNA levels were assessed using qRT-PCR. In contrast to the control group, a significant decreases in GPx and CAT expression were observed in all three agmatine-treated groups of Caco-2 cells compared to the untreated controls. In L929 cells, however, GPx and CAT mRNA levels increased in a dose-dependent (6.25 and 9 mM) and linear manner relative to the control group (Fig. 4A and B). Fig. 4Relative Expression Levels of GPx, CAT, ABCG2, ABCC1, and ABCB1 in L929 and Caco-2 Cells Treated with Agmatine. (A) The expression levels of GPx and CAT in L929 cells were modulated by agmatine in a dose-dependent manner (B) The expression levels of GPx and CAT in Caco-2 cells were modulated by agmatine in a dose-dependent manner C) Relative expression levels of ABCG2, ABCC1, and ABCB1 in Caco-2 cells treated with 7.3 mM agmatine for 72 h compared to control. Data were obtained from triplicate experiments and expressed as the means ± standard deviation. *p < 0.05, **p < 0.01, ***p < 0.001 Relative Expression Levels of GPx, CAT, ABCG2, ABCC1, and ABCB1 in L929 and Caco-2 Cells Treated with Agmatine. (A) The expression levels of GPx and CAT in L929 cells were modulated by agmatine in a dose-dependent manner (B) The expression levels of GPx and CAT in Caco-2 cells were modulated by agmatine in a dose-dependent manner C) Relative expression levels of ABCG2, ABCC1, and ABCB1 in Caco-2 cells treated with 7.3 mM agmatine for 72 h compared to control. Data were obtained from triplicate experiments and expressed as the means ± standard deviation. *p < 0.05, **p < 0.01, ***p < 0.001 The relative expression levels of the ABCG2, ABCC1, and ABCB1 genes were evaluated with qRT-PCR after 7.3 mM agmatine treatment of Caco-2 cells. qRT-PCR data revealed no significant differences between the ABCG2, ABCC1, and ABCB1 expression levels and agmatine-treated Caco-2 cells (p > 0.05; Fig. 4C). Western blot analyses performed to assess apoptotic pathways (Bax, Bcl-2, and Csp-3) in Caco-2 cells treated with agmatine at concentrations of 6 mM, 7.3 mM, and 9 mM revealed no detectable bands for these proteins, despite showing clear and consistent GAPDH bands as the internal loading control (data not shown). The absence of detectable bands may indicate that agmatine does not strongly activate the apoptotic pathway at the protein level in this cell model or that the experimental conditions (e.g., insufficient lysate loading or low basal expression of these apoptotic proteins) were inadequate to yield a visible signal on the blot. To the best of our knowledge, the present study is the first to demonstrate the anti-proliferative, anti-migration, anti-invasive, and anti-apoptotic effects of agmatine in the Caco-2, a model widely used for its enterocyte-like differentiation properties . Colon epithelial cells are exposed to numerous dietary compounds and metabolites that can affect cell physiology. Among these, polyamines are found in plant- and animal-based foods and fermented food products . Polyamines are also produced by luminal colon bacteria. It has been reported that agmatine reduces cell proliferation in the HT-29 cell lineand that treating HT-29 cells with agmatine at a concentration of 1 mM reduces cell proliferation and leads to complete inhibition at a dose concentration of 5 mM . In the literature, it has been shown that agmatine affects ADC-dependent cell apoptosis in the CRC cell line (HCT116) . Agmatine produces a concentration-dependent inhibition of proliferation in six colorectal cancer cell lines, while colon carcinoma tissue samples displayed significantly lower agmatine levels compared to normal tissue . Furthermore, agmatine demonstrated antiproliferative effects in tumor cells derived from colonic, hepatic, and neuronal origins . Plasma agmatine levels in prostate cancer patients have been found to be significantly lower compared to healthy individuals; agmatine inhibits proliferation in prostate cancer cells and has been proposed as a potential biomarker and therapeutic agent . In another study, agmatine produced a dose-dependent antiproliferative effect in MCF-7 breast cancer cells and SK-MG-1 glioma cells, while also suppressing polyamine uptake [10, 22]. Our results demonstrate that agmatine at concentrations of 6 mM, 7.3 mM, and 9 mM significantly inhibits ROS formation in Caco-2 cells at 72 h, suggesting that agmatine possesses the ability to alleviate oxidative stress and prevent ROS-induced oxidative damage. Agmatine has been shown to reduce ROS production by activating the Nrf2/HO-1 and PI3K/Akt pathways in LPS-induced BV-2 microglia and RAW 264.7 macrophage cells, while increasing GPx activity and glutathione levels and inducing a “pre-adaptive response” . In a diabetes mellitus model, agmatine treatment has been demonstrated to elevate GPx and catalase levels in leukocytes and to reduce oxidative/nitrosative stress . Additionally, agmatine has been reported to inhibit tumor cell proliferation by modulating polyamine metabolism in HT-29 cells, with agmatine levels being significantly lower in colon cancer tissues compared to normal tissue, suggesting a potential association with polyamine dysregulation . Another study by Park et al. examined the protective effects of agmatine against cisplatin-induced cell death in an auditory cell line. The results indicated that agmatine inhibits apoptosis by suppressing Bax and Csp-3 expression, though it does not directly decrease ROS levels . These protective effects are thought to be associated with the regulatory role of agmatine on oxidative stress and inflammation . To the best of our knowledge, this is the first study to evaluate ABC genes in agmatine treated Caco-2 cancer lines. A previous study has investigated the relationships between ABCG2, ABCC1, and ABCB1 genes in two different colon cancer cell lines (Caco-2 and HT-29) . In the HT-29 cell line, a significant difference was found among the ABCB1, ABCG2, and ABCC1 genes. It was shown that there was high up regulation of ABCB1 and ABCG2 but low upregulation of the ABCC1 compared to the other genes. In the Caco-2 cell line, there was shown to be significant difference in all three genes; as a result, suggesting that the upregulation of expression of ABC genes is directly related to cancer cell line . ABCB1, ABCC1, and ABCC2 mRNA levels were compared between Caco-2 cells and other colorectal cancer cell lines (HCT-15, LoVo, DLD-1, HCT-116, SW620). The results showed that ABCB1 and ABCC2 expression were higher in Caco-2 cells, whereas no significant differences were observed in HT-29-like cell lines . In another study that included the Caco-2 and HT-29 cell lines, the promoter methylation levels of ABCB1, ABCC1, and ABCG2 were investigated. The ABCC1 promoter exhibited low methylation in both cell lines, while differences were observed in the methylation patterns of ABCB1 and ABCG2 . According to our current results, no significant difference was found in the Caco-2 cell line that was treated with agmatine. Our findings indicate that the major ABC transporters ABCG2, ABCC1, and ABCB1 are not modulated by agmatine treatment in colon cancer cells and do not mediate resistance to agmatine in Caco-2 tumor cells. According to the mRNA expression levels of ABCB1, ABCC1, and ABCG2 transporters in colorectal cancer patients, there is a significant difference in the expression levels of these ABC transporters. The expression levels were found to be significantly higher in cancer patients. These findings showed that the expression of ABC transporters is a potential biomarker for the diagnosis of CRC patients [30, 31]. This study included no vehicle control or known cytotoxic positive control agents, such as cisplatin or doxorubicin. In this context, the absence of a positive cytotoxic control limits the direct comparison of our findings with the standard chemotherapeutic agents currently used in clinical practice. However, the primary objective of this study was to investigate the intrinsic biological effects of agmatine on CRC cells. In this study, the inclusion of only the Caco-2 cell line among CRC cell lines. On the other hand, the inclusion of only the Caco-2 cell line, without testing other colon cancer cell lines, represents one of the limitations of this current work. The Caco-2 cell line was selected because it is a widely used and well-characterized in vitro model in CRC research, and it is particularly suitable for investigating cellular processes such as proliferation, migration, and invasion. However, the inclusion of additional colorectal cancer cell lines with distinct genetic and molecular characteristics would have enhanced the translational relevance of the findings. One other limitation of the present study is that Western blotting failed to detect bands corresponding to Bax, Bcl-2, and Csp-3 in agmatine-treated Caco-2 cells, despite successfully visualizing of the internal control GAPDH. Consequently, changes in the expression of these proteins could not be confirmed at the protein level, limiting the validation of observations potentially made at the transcriptional or other levels. These findings highlight agmatine as a promising candidate for further investigation in colorectal cancer therapy, particularly through the modulation of proliferation, invasion, apoptosis, and oxidative stress pathways. The lack of effects on ABC transporters further supports its potential as a selective and non-MDR-related agent. Future studies including additional CRC cell lines, in vivo models, and comparisons with established chemotherapeutic agents will be essential for fully elucidating its therapeutic potential and mechanisms of action.