Xinqidi,Your biological research partner
Xinqidi Biotech Co.,Ltd,Wuhan,China 2008-2021
R&D 13th year

Loss of polycomb repressive complex 1 activity and chromosomal instability drive uveal melanoma prog

Issuing time:2021-09-18 09:52

Abstract

Chromosomal instability (CIN) and epigenetic alterations have been implicated in tumor progression and metastasis; yet how these two hallmarks of cancer are related remains poorly understood. By integrating genetic, epigenetic, and functional analyses at the single cell level, we show that progression of uveal melanoma (UM), the most common intraocular primary cancer in adults, is driven by loss of Polycomb Repressive Complex 1 (PRC1) in a subpopulation of tumor cells. This leads to transcriptional de-repression of PRC1-target genes and mitotic chromosome segregation errors. Ensuing CIN leads to the formation of rupture-prone micronuclei, exposing genomic double-stranded DNA (dsDNA) to the cytosol. This provokes tumor cell-intrinsic inflammatory signaling, mediated by aberrant activation of the cGAS-STING pathway. PRC1 inhibition promotes nuclear enlargement, induces a transcriptional response that is associated with significantly worse patient survival and clinical outcomes, and enhances migration that is rescued upon pharmacologic inhibition of CIN or STING. Thus, deregulation of PRC1 can promote tumor progression by inducing CIN and represents an opportunity for early therapeutic intervention.

Introduction

Uveal Melanoma (UM), a lethal eye cancer of adults and the second most common subtype of melanoma, is characterized by striking variability in metastatic tendency1,2,3. Once metastases are detected, median survival is less than twelve months4,5,6,7,8,9. Therefore, identifying patients at high-risk for metastasis and developing ways to intervene are critical priorities. Highly metastatic UM tumors differ from their more indolent counterparts in at least three ways1,3. First, they tend to have an “epithelioid” morphology with enlarged nuclei10,11. Second, they are often monosomic for chromosome 32,12,13, and frequently harbor mutations in the BAP1 gene (located on chromosome 3)2,14—a component of the polycomb repressive deubiquitinase (PR-DUB) complex that hydrolyzes ubiquitin at lysine 119 of the repressive Histone 2A (H2AK119)15,16. Third, they exhibit a distinctive gene expression signature2,17. A clinically-validated 12-gene signature—representative of the transcriptional changes that distinguish the two prognostic groups—is often used to identify patients with low-risk UM (Gene expression profile 1, GEP1) and high-risk tumors (Gene expression profile 2, GEP2)18,19 in the clinical setting. Currently, it is not known whether high-risk and low-risk UMs are fundamentally distinct disease subtypes or whether genetic and/or epigenetic changes in a subpopulation of tumors cells can lead to evolution from a relatively indolent to an aggressive UM (Fig. 1a). Identifying the underlying molecular basis for such a transition would provide insight into the molecular pathogenesis of UM and yield a critical opportunity for early therapeutic intervention. Here, we demonstrate that UM progression is driven by loss of PRC1 in a subpopulation of tumor cells, leading to transcriptional de-repression of PRC1-target genes and mitotic chromosome segregation errors. Hence, tumor stratification based on bulk transcriptional profiling of an inherently heterogeneous tumor is likely biased by detection of the most common tumor cell subpopulation.

Fig. 1: Phenotypic continuum of disease progression in primary UM.
figure1

a Distinguishing features of uveal melanoma (UM) with good (blue) and poor (red) prognosis; highlighting two potential models of disease progression. GEP gene expression profile, TCGA the Cancer Genome Atlas. b Patient tissue profiling (metadata summarized in Supplementary Table 1). Immediately following enucleation, six primary tumor specimens were obtained for clinical prognostic and single cell transcriptional profiling. Targeted sequencing using the MSK-IMPACT platform was performed on the formalin-fixed, paraffin embedded enucleation specimens. c Bulk GEP classification assigned to each patient according to the DecisionDx test (Castle Biosciences) (box). Individual tumor cells were likewise assigned to GEP1 (blue) vs. GEP2 (red) clinical prognostic groups according to their average expression of the GEP prognostic gene signatures using a two-component Bayesian Gaussian Mixture Model (BGMM, “Methods”). The fraction of individual tumor cells assigned to GEP1 (blue) and GEP2 (red) per patient is visualized in the bar graphs, where bootstrapping was used to correct for number of cells per patient (bar, mean; whiskers, 95% confidence intervals, 500 tumor cells sampled over n = 20 random subsets of the data). Asterisks, highlight a patient in which there was a discrepancy between DecisionDx bulk classification and the majority GEP classification prescribed by our single cell analysis of resected tumors. Notably, this patient (MSK-UM06) experienced metastatic progression and succumbed to their disease within six months of diagnosis (Supplementary Table 1). Intra-tumoral prognostic heterogeneity was validated in a second, independent cohort recently published by Durante et al.24. d Force-directed layout of all patient tumor cells colored by z-normalized imputed expression of the average GEP2 gene signature. e Individual tumor cells were likewise assigned to one of the four TCGA molecular subtypes of UM according to their average expression of characteristic genes using a four-component BGMM (Methods) in our cohort and a second, independent cohort recently published by Durante et al.24. The fraction of individual tumor cells assigned to TCGA subtype 1 (dark blue), subtype 2 (light blue), subtype 3 (pink) and subtype 4 (red) per patient is visualized in the bar graphs, where bootstrapping was used to correct for number of cells per patient (bar, mean; whiskers, 95% confidence intervals, 500 tumor cells sampled over n = 20 random subsets of the data). f Violin plots showing the distribution of intra-patient phenotypic volume (defined as the pseudo-determinant of the gene expression covariance matrix, detailed in “Methods”), controlled for number of cells and labeled by patient status (alive vs. deceased). Distributions represent 100 random subsamples from the data (n = 150 cells per patient). Overlaid bar and whisker plots reflect the mean and interquartile range.

Results

Single cell transcriptional landscape of uveal melanoma tumors

To resolve intra-tumoral heterogeneity and gain insights into tumor evolution that cannot be resolved by bulk tumor profiling20,21, we profiled the transcriptomes of 17,074 individual cells obtained from six freshly enucleated UM specimens from different prognostic categories (Fig. 1b). Immediately after enucleation, tumor specimens were obtained and sent for GEP clinical prognostication testing (DecisionDx) to assign individual patients to GEP prognostic classes19. According to DecisionDx, four of the six tumors were classified as high-risk GEP2 whereas the remaining two tumors were classified as low-risk GEP1 (Supplementary Table 1). Formalin-fixed, paraffin embedded specimens were also sent for targeted exome sequencing using the MSK-IMPACT platform22 (Fig. 1b, Supplementary Table 1). Fresh tumor samples analyzed with single-cell RNA sequencing (scRNA-seq) were derived from enucleation specimens without enrichment for a specific cell type, such that a piece of the viable tumor and its microenvironment were sampled in an unbiased manner. The library size, complexity, and viability metrics were of high quality (Supplementary Fig. 1a–c) and largely consistent across patients (Supplementary Fig. 1d, e). All scRNA-seq data were merged and normalized to create a global cell atlas, clustering23 of which revealed 27 cell types and states spanning retinal, immune and cancer cells (Supplementary Fig. 1f, g). Retinal and immune cell types were highly reproducible across patients, whereas patient-specific cell states were observed within tumor cell populations (Supplementary Fig. 1h–j). The majority of cells analyzed were tumor cells, expressing genes consistent with a melanocytic cell of origin (Supplementary Fig. 1i) and showing significant chromosome copy number alterations, including those canonically associated with UM (Supplementary Fig. 2a, b). In addition, we obtained single cell RNA sequencing data from an independent cohort of eight primary UM to validate key findings24.

Intratumoral phenotypic heterogeneity in UM tumors

Individual tumor cells were assigned to a prognostic class based on their average imputed expression of GEP1 and GEP2 discriminate genes using a two-component Bayesian Gaussian Mixture Model (Supplementary Fig. 3a, b, Methods). Strikingly, five tumors contained a heterogeneous admixture of cells that resembled both prognostic classes to varying extent (Fig. 1c, MSKCC). Likewise, the same classifier applied to the independent cohort of eight primary UM24, also revealed that the majority of primary tumors (7 out of 8) contain a heterogeneous admixture of tumor cells that resemble both prognostic classes to varying degrees, even when controlling for sampling differences across patients (Fig. 1c, Durante et al.24). Rather than observing two distinct states, we detected a continuum from a GEP1-like to a GEP2-like phenotype at the level of individual cells (Fig. 1d). Consistent with the previously described relationship between BAP1 loss and aggressive phenotype2,3,14,25,26,27,28,29,30,31, the GEP1-to-GEP2 transition was highly correlated with reduced BAP1 expression (Supplementary Fig. 3c) and the monosomy 3-associated gene expression signature17 (Supplementary Fig. 3d). Notably, GEP classification based on bulk transcriptional profiling (DecisionDx) was discordant with the majority GEP class detected by scRNA-seq in one out of six cases in the MSK dataset (Fig. 1c and Supplementary Table 1). A tumor (MSK-UM06) was classified as GEP1 based on bulk transcriptional sampling, yet harbored predominantly GEP2 cells (96.1%) by single cell analysis. This patient experienced metastatic progression and succumbed to disease within 6 months of diagnosis, in line with an abundance of aggressive tumor cells and highlighting the limitations of tumor stratification based on bulk transcriptional profiling of an inherently heterogeneous tumor.

While GEP classification is widely used clinically for diagnostic purposes, it has both technical and biological limitations. Detailed analysis of the TCGA uveal melanoma cohort (integrative analysis of UM transcriptomes, methylomes and genomic copy number data, n = 80 patients) has revealed the existence of four molecularly distinct biological and prognostic subsets of UM2. To likewise assess intra-tumor heterogeneity in the context of these molecularly distinct subsets, a four-component Bayesian Gaussian Mixture Model was applied to probabilistically assign individual tumor cells to each subtype based on their average imputed expression of characteristic genes (Fig. 1e and Supplementary Fig. 3e, f, Methods). Individual tumor cells promiscuously expressed markers associated with multiple TCGA subtypes (Supplementary Fig. 3g) and nearly all tumors across both cohorts24 showed substantial intra-tumoral heterogeneity when tumor cells were assigned to their maximally probable subtype (Fig. 1e). Therefore, and regardless of how risk levels are defined, individual UM tumors contained a heterogeneous admixture of cells that resembled different molecular subtypes to varying degrees. To capture this intra-patient cell state complexity—independent of clinical gene expression signatures—we applied a phenotypic volume metric32 across all variably expressed genes expressed by tumor cells within each patient. Notably, the two patients exhibiting the highest levels intra-tumoral phenotypic complexity succumbed to metastatic disease during the course of this study (Fig. 1f). Such intratumor heterogeneity - which cannot be resolved by bulk sequencing—suggests a model of UM progression; whereby cells within the primary tumor exist along various stages of an evolutionary continuum from an indolent towards a more aggressive phenotype.

The unexpected phenotypic progression underlying this cell state diversity (Fig. 1d) motivated us to apply archetypal analysis33 for unbiased, genome-wide transcriptomic characterization of tumor phenotypic states34. Analysis revealed 8 tumor cell archetypes, which are labeled on the force directed layout in Supplementary Fig. 4a. When individual tumor cells were assigned to their nearest archetype (“Methods”), each patient showed accumulation of multiple phenotypic states defining disease progression (Supplementary Fig. 4b). A local neighborhood of cells around each archetype was used to characterize genes differentially expressed in these bounding phenotypic states (Supplementary Fig. 4c). Archetypes were distinguished by differential expression of key pathways related to inflammatory response programs, aneuploidy, chromatin modifications and UM prognostic classifications. (Supplementary Fig. 4d). Collectively, this suggests such processes may underlie the evolution of UM tumors from an indolent to an aggressive phenotype.

Loss of PRC1 activity defines high-risk UM

To better understand the molecular underpinnings of UM single cell heterogeneity and tumor progression, we sought experimental models that recapitulate its distinct biological and prognostic classes. Towards this, we performed RNA sequencing of five established UM cell lines from diverse genetic backgrounds35,36,37. Cell lines distinctly clustered based on expression of the 80 characteristic genes that define the four major TCGA-UM subtypes or the 12-gene module that defines the GEP prognostic groups19 (Supplementary Fig. 5a–c). We designated cell lines falling at the opposite end of the gene expression spectrum, 92.1 and MP38, as low-risk and high-risk UM cells, respectively (Fig. 2a and Supplementary Fig. 5a). Expectedly, high-risk UM cells, MP38, harbored BAP1 mutation35, and had no detectable BAP1 protein (Fig. 2b). This is in line with reports showing that BAP1 genomic loss is a defining feature of aggressive UM1,2,3,14.

Fig. 2: Loss of PRC1-mediated transcriptional repression in high-risk UM.
figure2

a Unbiased hierarchical clustering of UM cell lines, 92.1 and MP38, in biological triplicates, based on normalized FPKM values of the GEP 12-discriminant geneset19. b Western blot of BAP1 and PRC1 core ligases RING1 and RNF2 relative to actin in 92.1 and MP38 cells. Data representative of biological triplicates. c Immunofluorescence of ubiquitinated H2A (green) and DAPI (blue) in MSK-UM03 (greatest proportion of GEP1 tumor cells in the cohort) and MSK-UM01 (greatest proportion of GEP2 tumor cells in the cohort). Representative images from six biological specimens. dHeatmaps representing CUT&RUN intensities of H2AK119Ub normalized to IgG in 92.1 and MP38. Data representative of biological duplicates. e Expression of H2AK119Ub target genes for individual tumor cells ranked by average imputed expression of the GEP2 gene signature (gene signatures annotated in Supplementary Data File 1) in ascending order from left to right. For each gene, imputed expression was z-normalized across all cells and smoothed using a 20-cell moving average window. Top, filled area plot showing average expression of GEP2 signature genes across ranked tumor cells. f Expression of H2AK119Ub target genes across the 4 molecular TCGA subtypes. Statistical significance tested using one-way ANOVA; p = 7.3 × 10−6; n = 80. Bars, mean of average expression; error bars, standard error of the mean. g Overall survival of (n = 80) TCGA-UM patients with primary tumors stratified by high (top 50th percentile, n = 40) and low (bottom 50th percentile, n = 40) expression of the ‘H2AK119Ub targets’ geneset. Statistical significance tested using two-sided log-rank test.

BAP1 hydrolyzes monoubiquitin on H2AK119Ub, a transcriptional repressive histone posttranslational modification (PTM)15,16,38,39. To test whether BAP1 loss translated into reduced H2AK119Ub, we performed H2AK119Ub immunofluorescence. To our surprise, high risk UM cells exhibited lower H2AK119Ub compared to their low risk counterparts (Supplementary Fig. 6a, b). Genome-wide localization analysis for H2AK119Ub, using Cleavage Under Targets and Release Using Nuclease (CUT&RUN)40, was consistent with the data from immunostaining. High-risk UM cells, MP38, exhibited near-complete loss of H2AK119Ub deposition at target loci compared to their low-risk counterparts, 92.1 (Fig. 2d). We next validated this finding in patient samples and found that the patient with predominantly low-risk features (e.g., MSK-UM03) exhibited significantly higher H2AK119Ub staining compared to those with high-risk gene expression profiles (e.g., MSK-UM01) (Fig. 2c).

Ubiquitiylation of H2AK119 is mediated by PRC1 through its ubiquitin ligase activity15,41. PRC1 contains a conserved core which consists of a RING1 (Really Interesting New Gene 1) or RNF2, both of which possess E3 ubiquitin ligase activity, in addition to one of the six PcG ring-finger domain proteins (PCGF1-6)15. PRC1 core ligase activity is necessary to achieve transcriptional repression of target genes41,42. Consistent with a reduction in H2AK119Ub in high risk UMs, transcript and protein levels of the core PRC1 ligases, RING1 and RNF2, were indeed lower in MP38 high-risk UM cells (Supplementary Fig. 6c and Fig. 2b). Similarly, in the TCGA cohort, UM tumors with BAP1 loss exhibited significantly lower RING1 and RNF2expression levels compared to tumors with intact BAP1, respectively (Supplementary Fig. 6d, e). Of all PRC1 accessory components (PCGF1-6), four (PCGF2,3,6 and PCGF4, also known as BMI1) exhibited reduced expression in MP38 cells compared to 92.1 (Supplementary Fig. 6f). In addition, expression of JARID2, a PRC2 accessory protein that is distinguished by its binding affinity to H2AK119Ub43,44, was also reduced in both MP38 cells compared to 92.1, as well as in human tumors with BAP1 loss (Supplementary Fig. 6g, h). Similar to H2AK119Ub, genome-wide localization analysis demonstrated near-loss of PRC1 components, RING1, RNF2, and BMI1 deposition in MP38, compared to 92.1 (Supplementary Fig. 7a). On the contrary, genomic loci bound to trimethylated Histone 3 at Lysine 27 (H3K27me3), another transcriptional repressive histone PTM were similar between 92.1 and MP38 (Supplementary Fig. 7a). Collectively this suggests widespread reduction of PRC1 components in high-risk UM.

In line with these findings, elevated expression of either JARID2 or RING1 mRNA was significantly associated with prolonged overall survival and decreased risk of metastasis (Supplementary Fig. 7b, c). Both JARID2 and RING1 are located on chromosome 6p, in line with prior observations that in UM, 6p gain is associated with a good prognosis2,45,46. We then asked whether low JARID2 and RING1 levels confer poor prognosis independent of 6p gain. The association between either JARID2 or RING1 and UM metastasis or death was assessed using logistic regression where 6p gain was a co-variate. JARID2 levels were inversely correlated with survival (Odds ratio (OR), 0.50; 95% confidence interval (CI), 0.31–0.74), and metastasis (OR, 0.69; CI, 0.50–0.91). Similarly, RING1 levels were inversely correlated with survival (OR, 0.61; CI, 0.37–0.89) and metastasis (OR, 0.65; CI, 0.43–0.89), indicating that higher expression levels of JARID2 or RING1 confer good prognosis independent of 6p gain.

We then asked whether a loss of PRC1 and H2AK119Ub translates into loss of transcriptional repression of target genes. We curated a gene list from genomic peaks of H2AK119Ub in UM cells (H2AK119Ub targets). These genes were highly expressed in patient tumor cells with increasing GEP2-like features (Fig. 2e and Supplementary Fig. 7d) (MSKCC and Durante et al.24), exhibited progressive de-repression across the 4 molecular TCGA subtypes (Fig. 2f), and foretold poor prognosis (Fig. 2g, TCGA). We interrogated the expression levels of genes that are bound to H3K27me3, excluding H2AK119Ub targets, and found that their levels were similar across the 4 molecular TCGA subtypes and did not correlate with overall survival (Supplementary Fig. 7e, f). Collectively, these results indicate that loss of H2AK119Ub and PRC1-mediated transcriptional repression is a feature of high-risk UM—despite concurrent loss of BAP1.

PRC1 inhibition phenocopies UM progression

To determine if loss of PRC1-mediated transcriptional repression underlies the transition from low-risk to high-risk UM, we used PRT4165 (thereafter referred to as PRT), a specific inhibitor of RING1 and RNF2 ligase activity, which is necessary to maintain PRC1-mediated target gene repression41,42. PRT has been shown to abolish ubiquitylated H2AK119 within 1 h of treatment47. PRT-treatment of low-risk 92.1 UM cells reduced H2AK119Ub levels after 2 h of treatment to those seen in MP38 cells (Supplementary Fig. 8a–d). On the other hand, treatment of high-risk MP38 UM cells, which have already adapted low basal activity of PRC1, had no impact on H2AK119Ub levels (Supplementary Fig. 8a, b). Importantly, PRC1 inhibition led to a profound transcriptional change in low-risk UM cells (92.1 and Mel202) while having minimal impact on gene expression in high-risk UM cells (MP41, MP46, and MP38) (Fig. 3a). Furthermore, treatment with PRT resulted in transcriptional upregulation of GEP2 signature genes and downregulation of GEP1 signature genes after just 24 h of treatment, in low-risk—but not high-risk—UM cells (Fig. 3b). On the contrary, inhibition of the H3K27 methyltransferase, EZH2, using the clinical grade inhibitor, EPZ, did not lead to transcriptional alterations in GEP1/2 signatures (Supplementary Fig. 8e). The majority (80.8%) of genes differentially expressed upon PRT treatment in low-risk UM cells were also differentially expressed between high-risk and low-risk UM cells at baseline (Fig. 3c).

Fig. 3: Evolution of aggressive UM phenotype triggered by PRC1 inhibition.
figure3

a Volcano plots showing differentially expressed genes upon PRT-treatment across five UM cell lines ranked according to their GEP1/GEP2 score (see Supplementary Fig. 5a–c). Fold change for individual genes is shown as a function of significance, -log10(FDR). Genes with FDR value less than 0.05 and fold change greater or less than 1, were highlighted in green and red, respectively. Genes not meeting significance (FDR > 0.05) are shown as gray. Number of genes meeting significance (FDR < 0.05) are annotated per each cell line. b Ratio of GEP1/GEP2 average gene signature expression (gene signatures annotated in Supplementary Data File 1) in low-risk UM cells (92.1) and high-risk UM cells (MP38) upon 24 h DMSO or PRT-treatment; FPKM values obtained from bulk RNA-seq are reported (bar, mean; circles, biological duplicates). c Venn diagram of differentially expressed genes (DEG) upon PRT treatment (red) and between high-risk UM cells (MP38) and low-risk UM cells (92.1) (green). d A schematic showing transcriptional de-repression upon pharmacologic inhibition of PRC1 using PRT (“PRT-geneset” annotated in Supplementary Data File 1). e Expression of the PRT-geneset across individual tumor cells ranked by average imputed expression of the GEP2 gene signature (gene signatures annotated in Supplementary Data File 1) in ascending order from left to right. For each gene, imputed expression was z-normalized across all cells and smoothed using a 20-cell moving average window. Top, filled area plot showing average expression of the GEP2 signature across ranked tumor cells. f Average expression of genes upregulated upon PRT-treatment of 92.1 cells (‘PRT-geneset’) across the 4 molecular TCGA subtypes. Statistical significance tested using one-way ANOVA; p = 9.6 × 10−9; n = 80. Bars, mean of average expression; error bars, standard error of the mean. g Overall survival of (n = 80) TCGA-UM patients with primary tumors stratified by high (top 50th percentile, n = 40) and low (bottom 50th percentile, n = 40) average expression of the “PRT-geneset”. Statistical significance tested using two-sided log-rank test.

Concordantly, genes upregulated upon pharmacologic inhibition of PRC1 in low-risk UM cells (fold change > 1 and Bonferroni <0.05, “PRT-geneset” defined in Fig. 3d) were highly expressed in patient tumor cells with increasing GEP2-like features (Fig. 3e and Supplementary Fig. 8f). These genes exhibited progressive de-repression across the 4 molecular TCGA subtypes (Fig. 3f) and foretold poor prognosis (Fig. 3g, TCGA), underscoring that loss of PRC1-mediated transcriptional repression is a feature of high-risk UM. Two of the 4 genes that define the GEP2 profile, ECM1 and HTR2B, were among the most differentially expressed genes in the PRT-geneset (n = 28) (Supplementary Data File 1).

In addition to transcriptional changes, PRT treatment led to profound morphologic alterations that have long been associated with high-risk UM tumors10,11; mainly enlarged nuclei and an epithelioid morphology (Fig. 4a, b). In line with the drug’s specificity to low-risk UM cells, characterized by elevated PRC1 activity, treatment of high-risk, PRC1-defecient MP38 cells with PRT had no effects on nuclear size and morphology (Supplementary Fig. 8g). Similarly, PRC1 inhibition also impaired growth of low-risk 92.1 cells, but not high-risk MP38 cells (Supplementary Fig. 8h). Notably, 92.1 cells treated with PRT grew at a slower rate, similar to high-risk UM cells, however reduced growth rates of PRT-treated 92.1 cells quickly rebounded to baseline following PRT withdrawal (Supplementary Fig. 8i). Collectively, these results suggest that loss of PRC1 activity recapitulates a transition from low-risk to high-risk UM—linking critical genetic, epigenetic, and cytologic features.

Fig. 4: Morphological changes and nuclear enlargement induced by PRC1 inhibition.
figure4

a Top panel, bright field light microscopy of low-risk UM cells (92.1) upon long-term DMSO and PRT-treatment. Bottom panel, DAPI staining of low-risk UM cells (92.1) after 48 h of PRT, and DMSO treatment. Images representative of experimental triplicates. b Violin plots showing the distribution of nuclei size in low-risk UM cells (92.1) upon PRT (n = 152) and DMSO (n = 163) treatment for 48 h. Lines distinguish interquartile ranges; width reflects observed nuclei number. Statistical significance tested using two-sided unpaired Student’s t test, p = 4.9 e−36. Source data are provided as a Source Data file.

Widespread CIN and inflammatory signaling in high-risk UM

To further explore the molecular underpinning of this transition, we performed gene set enrichment analysis (GSEA) comparing DMSO and PRT-treated low-risk UM cells. In addition to PRC1 target genes, PRT treatment led to significant upregulation of pathways related to inflammation, such as NF-κB (Normalized enrichment score (NES), 2.8; FDRq <0.001), IL6/STAT3 (NES, 1.8; FDRq, 0.02) and epithelial-to-mesenchymal transition (EMT) (NES, 2.0; FDRq, 0.04). Conversely, there was downregulation of pathways involved in the cell cycle and mitotic spindle assembly (Fig. 5a). Differentially expressed pathways upon PRT treatment were reminiscent of those seen enriched in high-risk MP38 as compared to low-risk 92.1 UM cells, with the latter exhibiting upregulation of inflammation and EMT-related gene sets (Supplementary Fig. 9a).

Fig. 5: Increased chromosome segregation errors upon PRC1 inhibition.

a Volcano plots showing the top differentially expressed pathways in low-risk UM cells (92.1) treated with PRT vs DMSO for 24 h; evaluated across biological triplicates. Normalized enrichment score (NES) for selected genesets shown as a function of significance, -log10(FDRq); FDRq, Bonferroni corrected p value. An unfiltered list of all significant gene signatures is provided in Supplementary Data File 5. b Examples of UM cells in anaphase stained for DAPI (blue) and centromeres (red); demonstrating different patterns of chromosome segregation errors. Additional patterns of missegregation are shown in Supplementary Fig. 9b. Representative images from biological triplicates. c Abundance of chromosome segregation error patterns during anaphase in low-risk UM cells (92.1) upon PRT-treatment as a function of time. Statistical significance tested using two-sided unpaired Student’s t test. Source data are provided as a Source Data file. d Abundance of chromosome segregation error patterns during anaphase across UM cell lines, arranged from left to right based on their GEP2 score. Source data are provided as a Source Data file. e Western blot of H2AK119Ub, STING, RING1, and RNF2 relative to Actin in Mel285 cells upon RING1 and RNF2 knockout. Data representative of biological triplicates. f Abundance of chromosome segregation error patterns during anaphase in UM cells (Mel285.Cas9) upon RNF2 knockout. Statistical significance tested using two-sided unpaired Student’s t test. Source data are provided as a Source Data file. Stacked bars, mean of each missegregation pattern; error bars, standard error of the mean across experimental triplicates (c–f).

Given increased aneuploidy in high-risk primary UM2 and increased inflammatory signaling (Fig. 5a) we observed in aggressive UM, as well as in other studies2,25,26,48,49,50,51,52, we set out to determine whether changes in PRC1 could potentially alter genomic stability. We have recently shown that errors in chromosome segregation—a defining feature of cancer cells with chromosomal instability (CIN)—can generate micronuclei, which, upon rupture, expose their enclosed genomic double-stranded DNA (dsDNA) to the cytosol53,54,55,56. This leads to aberrant activation of the cGAS-STING cytosolic dsDNA signaling pathway and downstream inflammatory signaling mediated by noncanonical NF-κB activation, as well as upregulation of EMT and migratory pathways that promote metastatic progression56. We hypothesized that loss of PRC1 function may trigger CIN, leading to enhanced migration and inflammatory signaling characteristic of high-risk UM. Indeed, PRT-treatment of low-risk 92.1 cells led to increased frequency of mitotic chromosome segregation errors, as evidenced by the preponderance of lagging chromosomes and acentric chromatin fragments in PRT-treated cells (Fig. 5b, c and Supplementary Fig. 9b). Importantly these defects emerged 24–48 h after drug treatment (Fig. 5c) arguing against a direct, short-term effect on the mitotic spindle and instead favoring a transcriptional response. Importantly, and in line with our earlier observations demonstrating PRT’s specificity to UM cells with intact PRC1, PRT treatment of high-risk MP38 cells showed no effect on chromosome segregation errors (Supplementary Fig. 9c). Correspondingly, intrinsic rates of chromosome missegregation correlated with the ratio of GEP2/GEP1 gene expression (Fig. 5d). Unlike PRC1, Inhibition of EZH2—the core catalytic component of the PRC2 complex—did not lead to an increase in chromosome segregation defects (Supplementary Fig. 9d). We then complemented these pharmacologic modulations with genetic manipulation of core PRC1 ligases. First, we established a CRISPR-Cas9 knockout system in Mel285, a low-risk uveal melanoma cell line which expresses BAP157. Knockout of either RING1 or RNF2 resulted in a significant reduction in H2AK119Ub levels (Fig. 5e), and loss of core PRC1 components dramatically increased chromosome segregation defects observed during anaphase (Fig. 5f).

To test whether increased chromosome missegregation during mitosis leads to the formation of micronuclei, we assessed the frequency of micronuclei in the various UM cell lines and found rates of micronuclei mirrored those of chromosome missegregation. High-risk UM cells had significantly higher rates of micronuclei compared to their low-risk counterparts (Fig. 6a). Similarly, PRT treatment of 92.1 cells—but not of high-risk MP38 cells—led to a significant increase in micronuclei (Fig. 6b, c and Supplementary Fig. 9e), whereas treatment with an EZH2 inhibitors had no impact on their formation (Supplementary Fig. 9f). Likewise, genetic knockout of RNF2 resulted in a significant increase in micronuclei frequency (Fig. 6d). Importantly, these micronuclei frequently co-localized with cGAS (Fig. 6e), indicative of cytosolic exposure of their enclosed genomic dsDNA and activation of cytosolic dsDNA sensing, cGAS-STING pathway.

Fig. 6: Cytosolic DNA exposure in high-risk UM.
figure6

a Baseline micronuclei frequency across UM cell lines, arranged from left to right based on their GEP2 score. Bar represents median; points, measured frequency of micronuclei per high-power field evaluated across three experimental replicates. Statistical significance tested using two-sided student t test; p = 2.8 × 10−8. Source data are provided as a Source Data file. b Micronuclei frequency in low-risk UM cells (92.1) upon PRT treatment as a function of time. For (a, b), bar represents median; points, measured frequency of micronuclei per high-power field evaluated across three experimental replicates. Statistical significance tested using two-sided student t test; p = 7.7 × 10−8. Source data are provided as a Source Data file. c Low-risk UM cells (92.1) stained for DAPI (white); showing increased nuclear size and micronuclei formation upon PRT-treatment. Images representative of experimental triplicates. d Micronuclei frequency in UM cells (Mel285.Cas9) upon RNF2 knockout. Bar represents median; points, measured frequency of micronuclei per high-power field evaluated across three experimental replicates. Statistical significance tested using two-sided student t test; p = 7.7 × 10−6. Source data are provided as a Source Data file. e An example of a cGAS (green) localization to micronuclei (blue). Representative image from biological triplicates.

In line with these findings, genetic knockout of either RING1 or RNF2 resulted in increased STING levels (Fig. 5e). We also observed upregulation of CGAS, STING (TMEM173) mRNA and downstream inflammatory response pathway effectors induced by CIN, including noncanonical NF-κB targets in patient tumor cells ranked according to the average GEP2 gene expression signature (Fig. 7a). These target genes are upregulated in response to chronic STING activation in cancer cells with CIN56. We validated these findings in an independent cohort24, where we likewise observed upregulation of CGAS, TMEM173 and downstream inflammation-related mRNAs, including non-canonical NF-κB targets, in high-risk tumor cells (Supplementary Fig. 10a). Concordantly, STING and cGAS mRNA expression levels were elevated in TCGA-UM tumors with genomic copy number loss in BAP1 compared to BAP1-intact tumors (Supplementary Fig. 10b). To further validate these findings in human tumor samples, we performed immunofluorescence imaging of STING and found elevated expression of tumor cell-intrinsic STING in the predominantly high-risk tumor (e.g., MSK-UM01) as compared to the predominantly low-risk tumor (e.g., MSK-UM03), where minimal STING expression was mainly restricted to the stromal compartment (Fig. 7b). In the TCGA cohort, TMEM173 expression was highly prognostic, whereby high STING levels predicted metastasis and reduced overall survival (Fig. 7c, TCGA). All together, these results suggest that CIN-induced cytosolic dsDNA signaling downstream of PRC1 loss, is a feature of high-risk UM.

Fig. 7: Cell-intrinsic inflammation in high-risk UM.

a Expression of key cytosolic nucleic acids sensors, intermediate signaling adapters, executioners, interferon stimulated genes (ISGs) and CIN-induced non-canonical NF-kB targets (shown in red)56for all patient tumor cells ranked by average imputed expression of the GEP2 gene signature (gene signatures annotated in Supplementary Data File 1) in ascending order from left to right; genes are clustered using an Euclidean distance metric. For each gene, imputed expression was z-normalized across all cells and smoothed using a 20-cell moving average window. Top, filled area plot showing average expression of the GEP2 signature across ranked tumor cells. b Immunofluorescence of STING (green) and DAPI (blue) in MSK-UM03 (greatest proportion of GEP1 tumor cells in the cohort) and MSK-UM01 (greatest proportion of GEP2 tumor cells in the cohort). Representative images from six biological specimens. c Overall survival (left) and UM-related metastasis (right) of (n = 80) TCGA-UM patients with primary tumors stratified by expression of high (top 50th percentile, n = 40) and low (bottom 50th percentile, n = 40) STING. Statistical significance tested using two-sided log-rank test; p (left) = 2.2 × 10−5.

PRC1 inhibition promotes CIN and STING-dependent migration

Chronic activation of noncanonical NF-κB and inflammatory pathways downstream of cGAS-STING in cancer cells with CIN has been shown to promote migration and metastasis56. We thus asked whether PRC1 loss could promote a migratory phenotype mediated by downstream CIN induction. Indeed, PRT-treatment enhanced the migration of low-risk UM cells, but not high-risk UM cells (Fig. 8a and Supplementary Fig. 10c). To determine whether the migratory phenotype induced by PRC1 inhibition is mediated through CIN and STING, we employed pharmacologic modulators of CIN and STING. CIN was suppressed by de-stabilizing kinetochore-microtubule attachments using UMK57, a small molecule that has been proposed to potentiate the activity of MCAK, a kinesin-13 protein whose microtubule-destabilizing activity at the centromere is critical for faithful chromosome segregation58,59,60. Treatment with UMK57 completely rescued the increase in chromosome missegregation seen upon PRT treatment of 92.1 low-risk UM cells and significantly reduced their migration (Supplementary Fig. 11a and Fig. 8a). We next used H151, a small molecule covalent inhibitor of STING that blocks its activation-induced palmitoylation61. Treatment with H151 also rescued the migratory phenotype seen upon PRC1 inhibition in low-risk 92.1 cells (Fig. 8b), indicating that CIN- and STING-mediated migratory effects are downstream of PRC1 loss. These drug treatments inhibited migration more than the DMSO control, which we attribute to low levels of basal missegregation rates even in low-risk DMSO-treated 92.1 cells.

Fig. 8: PRC1 inhibition promotes CIN and STING-dependent migration.
figure8

a Wound scratch assay to assess migratory potential of low-risk UM cells (92.1) upon treatment with DMSO, PRT, UMK57 (CIN inhibition) or PRT and UMK57. Data obtained from three experimental replicates. Data points, mean; error bars, standard error of the mean. Source data are provided as a Source Data file. b Wound scratch assay to assess migratory potential of low-risk UM cells (92.1) upon treatment with DMSO, PRT, H151 (STING inhibition) or PRT and H151. Data obtained from three experimental replicates. Data points, mean; error bars, standard error of the mean. Source data are provided as a Source Data file. c Ratio of GEP1/GEP2 average gene signature expression in low-risk UM cells (92.1, blue) and high-risk UM cells (MP38, red) upon 48-h treatment with PRT, reversine (CIN induction) or UMK57; FPKM values obtained from bulk RNA-seq are reported (bar, mean; bars, standard deviation). Biological triplicates. d Schematic of proposed model: Loss of PRC1 ligase activity leads to transcriptional de-repression of target genes contributing to GEP2 phenotype; this concomitantly promotes nuclear enlargement and morphological changes toward an epithelioid phenotype, and enhances migration through CIN-induced STING signaling. Pharmacologic modulators shown in yellow; (PRT, PRC1 inhibition; UMK57; CIN suppression; H151, STING inhibition). CIN, chromosomal instability.

Finally, we set out to determine the relative contributions of CIN to the transcriptional profile that defines high-risk UM. We performed RNA-sequencing on low-risk 92.1 cells with and without CIN induction through reversine treatment, a potent inhibitor of the mitotic kinase Mps162 in the absence of PRC1 inhibition. Pharmacologic CIN induction led to an expected increase in chromosome missegregation and micronuclei formation (Supplementary Fig. 11a, b), as well as transcriptional changes accounting for half of genes (50.4%) that were differentially expressed upon PRC1 inhibition (Supplementary Fig. 11c), including upregulation of inflammatory and migratory pathways (Supplementary Fig. 11d). However, it had no effect on the expression of GEP prognostic genes (Fig. 8c). Accordingly, CIN suppression using UMK57 in high-risk MP38 cells, or in 92.1 cells treated with PRT, did not alter the GEP-specific gene expression signature, despite leading to a decrease in chromosome missegregation and micronuclei formation (Fig. 8c and Supplementary Fig. 11a, b). Collectively, these findings indicate that while the migratory phenotype induced by PRC1 loss is mediated through CIN, canonical GEP-related genes are likely under direct control of PRC1, independent of CIN (Fig. 8d).

Discussion

Our results provide mechanistic insight into UM progression and argue against a model of independent clonal origins, implying that UM with good prognosis, left untreated, may evolve over time to acquire a more aggressive phenotype. Tumor stratification based on bulk transcriptional profiling of an inherently heterogeneous tumor is likely biased by detection of the most common tumor cell subpopulation or the region sampled during fine needle aspiration. Surprisingly, we observed that UMs with BAP1 loss exhibited de-repression—rather than enhanced repression—of PRC1 target genes, and loss—rather than gain—of H2AK119Ub. While BAP1 hydrolyzes ubiquitin on H2AK119Ub, thereby counteracting PRC1 action, its role in modulating the expression of genes under Polycomb regulation is less defined. For instance, in human cells, BAP1 appears to safeguard against gene silencing mediated by PRC116, whereas in Drosophila, loss of BAP1 homologue, Calypso, or other components of PR-DUB, was also found to lead to de-repression of Polycomb target genes38,63. It is possible that an intricate balance between H2A ubiquitination and de-ubiquitination may be necessary to achieve target gene repression64 and that BAP1 loss might be necessary in order to maintain very low basal levels of H2AK119Ub, which might be required for cellular viability. BAP1 function may also depend on cellular context. For instance, BAP1 loss mediates apoptosis in fibroblasts, liver, and pancreatic cells but not in melanocytes and mesothelial cells65. Our results show that, in addition to BAP1 loss, high-risk tumors exhibited loss of core PRC1 components and decreased H2A ubiquitination, indicating that BAP1 loss in UM reflects broader dysfunction in loss of PRC1-mediated transcriptional repression. Our results are in line with other studies that have demonstrated aberrant expression of epigenetic modifiers in high risk UM66, significant intra-tumoral spatial heterogeneity in H2AK119Ub immunostaining of UM, and reduced H2AK119Ub staining in UM compared to normal choroid67.

More generally, this work links two important hallmarks of tumor progression by demonstrating a functional link between epigenetic reprogramming and CIN, both of which have been implicated in tumor progression and metastasis56,68. In addition to its role in transcriptional regulation, Polycomb group proteins interact to form higher order chromatin structure. Consequently, their disruption changes the three-dimensional genome topology and may contribute to CIN15,69,70,71. In mouse fibroblasts, loss of canonical PRC1 component, CBX2, induces CIN72. Indeed, we found that inhibition of PRC1—but not PRC2—promotes widespread CIN in UM cells. We also found that PRC1 inhibition promotes a transition toward an epithelioid morphology characterized by nuclear enlargement, thereby linking epigenetic, transcriptional and histological features of UM. In line with our findings, nuclear enlargement has recently been shown to occur secondary to RING1B(RNF2) knockdown in mouse embryonic stem cells69.

It is crucial to identify pathways downstream of PRC1 loss that contribute to metastasis and which can be targeted for therapeutic benefit. We recently demonstrated that CIN can drive metastasis in a tumor cell-autonomous manner, through aberrant activation of the cytosolic DNA-sensing cGAS-STING pathway56. Here, we found that PRC1 inhibition triggers chromosome segregation errors that promote pro-metastatic chronic inflammatory signaling. While the migratory phenotype induced by PRC1 inhibition was suppressed in the presence of small molecule inhibitors of either CIN or STING, canonical GEP-related gene sets that define aggressive UM in the clinic appear to be under direct PRC1 control Suppression of CIN was insufficient to alter their expression levels. This suggests that PRC1 deregulation has both CIN- dependent and independent mechanisms of promoting UM progression. Nonetheless, given its functional consequences on cell migration, modulation of pathways downstream of PRC1, like CIN and the cGAS-STING pathway, may still represent a promising target that could be exploited to suppress UM progression and metastasis. By uncovering key steps involved in UM progression, our work highlights an opportunity for earlier therapeutic intervention to suppress tumor evolution toward a lethal metastatic phenotype.


Article classification: Biological abstract
Share to:
Tel:+86-027-87610298
Tel:+86-027-87610297
Add:Room A11-329, 1st Floor, No.1, SBI Venture Street, Optics Valley, East Lake
New Technology Development Zone, Wuhan, China.
Certificate NO.:U18Q28010569R0S