ARID1A suppresses R-loop mediated STING-Type I Interferon pathway activation of anti-tumor immunity

SUMMARY

Clinical trials have identified ARID1A mutations as enriched among patients who respond favorably to Immune Checkpoint Blockade (ICB) in several solid tumor types independent of microsatellite instability. We show that ARID1A loss in murine models is sufficient to induce anti-tumor immune phenotypes observed in ARID1A mutant human cancers, including increased CD8+ T cell infiltration and cytolytic activity. ARID1A deficient cancers upregulated an interferon (IFN) gene expression signature, the ARID1A-IFN signature, associated with increased R-loops and cytosolic single stranded DNA (ssDNA). Overexpression of the R-loop resolving enzyme, RNaseH2B, or cytosolic DNase, TREX1, in ARID1A deficient cells prevented cytosolic ssDNA accumulation and ARID1A-IFN gene upregulation. Further, the ARID1A-IFN signature and anti-tumor immunity were driven by STING dependent Type I IFN signaling, which was required for improved responsiveness of ARID1A mutant tumors to ICB treatment. These findings define a molecular mechanism underlying anti-tumor immunity in ARID1A mutant cancers.

Graphical Abstract

In Brief

ARID1A loss in tumor cells triggers anti-tumor immunity via R-loop derived cytosolic DNA and activation of the STING-Type I Interferon pathway, defining a molecular mechanism underlying ARID1A mutant anti-tumor immunity and immunotherapy response observed clinically.

INTRODUCTION

Immune checkpoint blockade (ICB) is a class of cancer immunotherapy that unleashes the immune system to attack cancer by blocking the interaction of inhibitory receptors on T cells with their ligands. ICB has resulted in durable, even curative, anti-tumor immune responses; however, in the most types of cancer, fewer than half of all patients respond favorably to ICB¹. Thus, considerable clinical and pre-clinical efforts have been made to identify genetic and molecular biomarkers that can effectively predict ICB patient response.

Genomic sequencing consortia studies have revealed mutations in SWI/SNF complex subunits in ~20% of all human cancers^2,3. The SWI/SNF complex is a multi-subunit chromatin remodeling complex that uses energy derived from ATP hydrolysis to slide DNA along nucleosomes to regulate chromatin accessibility, transcription, and DNA repair⁴. The most frequently mutated variant of SWI/SNF complexes in cancer is the canonical BAF (cBAF) complex which is defined by incorporation of unique subunits ARID1A/ARID1B/DPF2⁴. The cBAF subunit ARID1A is the most frequently mutated SWI/SNF subunit gene and among the most frequently mutated tumor suppressor genes in human cancer². Mutations in ARID1A are predominantly inactivating resulting in loss of protein expression and occur in ~8–60% of endometrial, ovarian clear cell, colon, gastric, liver, pancreatic cancers^2,3,5.

Recently, ICB clinical trials have identified ARID1A mutations as significantly enriched among immunotherapy responders in a diverse array of solid tumor types^6–11. Previously published pre-clinical research reported that ARID1A loss results in increased DNA mutability and response to ICB due to loss of an ARID1A dependent interaction between the cBAF complex and the mismatch repair (MMR) complex¹². However, recent clinical trials of ICB have shown ARID1A mutations are enriched among ICB responders in diverse tumor types in a manner independent of microsatellite instability^6–8,13. Moreover, ARID1A mutation has been reported to endow an added benefit in overall survival following ICB in bladder cancer patients with high tumor mutational burden (TMB)⁸ and in pancreatic cancer patients receiving ICB in the absence of high TMB⁷. On the other hand, a meta-analysis on SWI/SNF mutations in a pan-cancer setting following ICB treatment found no clinical benefit associated with SWI/SNF loss of function mutations after correcting for high TMB¹⁴. Collectively, this literature highlights the need to understand whether ARID1A mutation is truly causative in promoting anti-tumor immunity or simply associated with anti-tumor immunity due to other features present in tumors harboring ARID1A mutation.

In this study, we sought to investigate the effects of ARID1A loss on anti-tumor immunity and ICB response in isogenic syngeneic tumor models to clarify the relationship between ARID1A mutation and tumor immune surveillance. We found that CRISPR/Cas9-mediated Arid1a deletion was sufficient to induce anti-tumor immunity in murine tumor models, characterized by a robust anti-tumor T cell response. ARID1A deficiency or cBAF inhibition induced a cell intrinsic Type I IFN gene expression program (ARID1A-IFN), triggered by STING dependent cytosolic DNA sensing resulting from increased R-loops, cytosolic single stranded DNA (ssDNA), and cytosolic RNA:DNA hybrids. Finally, we show that anti-tumor immunity is driven by STING and Type I IFN, which is also required for enhanced response of ARID1A mutant tumors to ICB. These findings represent a molecular mechanism by which ARID1A mutation can promote ICB response and may have translational utility in improving patient selection for ICB and by providing a rationale for use of cBAF inhibitors to augment ICB efficacy.

RESULTS

ARID1A Loss in Murine Tumor Models Recapitulates Human ARID1A Mutant Anti-Tumor Immunity

To investigate the effects of ARID1A loss on anti-tumor immunity, we utilized CRISPR-Cas9 genetic engineering to generate ARID1A deficient (sgArid1a) and CRISPR-Cas9 control (sgScramble) B16F10 melanoma and MC38 colon cancer syngeneic mouse cell lines (Figure S1A). Following subcutaneous injection of B16F10 or MC38 cancer cells into the flanks of C57BL/6J mice, we observed significantly slower tumor growth and decreased tumor weight of sgArid1a tumors compared to sgScramble tumors (Figure 1A–D and S1B). This result was not due to a difference in cell proliferation rate as sgArid1a and sgScramble cells grow at nearly identical rates (Figure S1C–D). Flow cytometry immune profiling revealed a significant increase of CD45+ tumor infiltrating immune cells in sgArid1a B16F10 tumors (Figure 1E). Among CD45+ cells, sgArid1a tumors contained significant increases in CD8+ cytotoxic T cells, natural killer cells (‘NK cells’), conventional class I dendritic cells (‘cDC1’), and regulatory T cells (‘Tregs’) (Figure 1F–I). sgArid1a tumors recapitulate increases in the percentage of T and NK cell subsets in ARID1A mutant human cancers as assessed by Cibersort immune estimate analysis¹⁵ of RNA-seq from The Cancer Genome Atlas (TCGA) (Figure S1E). Moreover, we observed increased CD8+ T cell infiltration in human endometrial cancer histological samples with ARID1A protein loss or confirmed mutation compared to non-mutant ARID1A positive patient samples (Figure 1J and S1F). Strikingly, an increased percentage of CD8+ T cells in sgArid1a tumors expressed immunostimulatory cytokines such as IFNγ and TNFα and the cytotoxic effector molecule Granzyme B (GZBM) (Figure 1K–L). Correspondingly, human ARID1A mutant cancers displayed significantly elevated cytolytic scores, which is a proxy of immune cell cytotoxicity¹⁶ (Figure 1M). An increased percentage of CD8+ T cells infiltrating sgArid1a tumors were positive for T cell activation and proliferation markers PD1 and Ki67 (Figure 1N–O). Further, there was an increased number of stem-like SLAMF6+ TIM3- Progenitor Exhausted (ProEx) CD8+ T cells in sgArid1a tumors (Figure 1P), as well as a higher percentage and number of TIM3+ SLAMF6- Terminal exhausted T cells (‘TermEx’) (Figure 1Q). Thus, sgArid1a tumors contain increased cellular density of ProEx and TermEx CD8+ T cells which perform critical anti-tumor functions such as self-renewal and cytotoxicity, respectively¹⁷. Additional immune cell functional differences include increased proliferation (Ki67) and effector cytokine production in CD4+ T cells (TNFα, IFNγ) and NK cells (IFNγ, GZMB) infiltrating sgArid1a tumors relative to sgScramble tumors (Figure 1R–S, Figure S1G–H).

Figure 1. ARID1A loss is sufficient to induce anti-tumor immunity in murine tumor models and recapitulates features of human ARID1A mutant anti-tumor immunity.

(A-B) Tumor growth curves of sgScramble and sgArid1a B16F10 and MC38 tumors. Data are represented as mean ± SEM. (C-D) Tumor weights on Day 15 or Day 17 post injection for B16F10 tumors or MC38 tumors. (E-I) Tumor infiltrating immune cell profile in sgScramble and sgArid1a B16F10 tumors. (J) Human endometrial cancer IHC samples with representative images of ARID1A IHC and multiplexed fluorescence IHC for CD8α (yellow) and DAPI (blue) in patient samples with ARID1A protein intact or ARID1A protein loss. MMR status assessed via IHC (scale bar = 100μM). (K-L) Quantification and flow cytometry contour plots of CD8+ T cell TNFα and IFNγ or Granzyme B and IFNγ staining following ex vivo PMA/ionomycin stimulation from sgScramble and sgArid1a B16F10 tumors. (M) TCGA cohort cytolytic scores in ARID1A mutant and non-mutant cancers. (N) Ki67 staining in CD8+ T cells from spleen, sgScramble, or sgArid1a B16F10 tumors. (O) PD1 staining in CD8+ T cells from sgScramble and sgArid1a B16F10 tumors. (P) CD8+ T cell exhaustion subsets quantified by cells per gram of B16F10 tumor. (Q) CD8+ T cell exhaustion subsets in sgScramble and sgArid1a B16F10 tumors. (R) Tumor infiltrating CD4+ T cell IFNγ and TNFα staining following ex vivo PMA/ionomycin stimulation. (S) Tumor infiltrating NK cell IFNγ and Granzyme B staining following ex vivo PMA/ionomycin stimulation. (T-U) Comparisons of tumor growth curves of sgScramble and sgArid1a B16F10 and MC38 tumors in Rag1 −/− and C57BL/6J mice. Data are represented as mean ± SEM. All data are represented as mean ± SD unless otherwise noted. Statistical analysis in Fig 1L was performed using Welch’s Two sample t-test.

Supporting the essential role of T cells in mediating ARID1A mutant anti-tumor immunity, we observed that the tumor growth deficiencies of sgArid1a B16F10 and MC38 tumors relative to sgScramble tumors was rescued in Rag1 knockout mice which lack an adaptive immune system (Figure 1T–U). Together, these results demonstrate that ARID1A loss is sufficient to induce anti-tumor immunity in murine tumor models which are characterized by a robust T cell response and recapitulate features of anti-tumor immunity in ARID1A mutant human cancers.

Bulk and Single Cell Transcriptomic Analyses of ARID1A Deficient Tumors Reveals Enrichment of IFN Gene Expression Signatures and CD8+ T cell Hyperexpansion

To further characterize the effects of ARID1A loss on the tumor-immune microenvironment (TME), we utilized bulk and single cell RNA-sequencing (RNA-seq and scRNA-seq) on tumors or tumor infiltrating CD8+ T cells. Bulk RNA-seq of sgScramble and sgArid1a B16F10 tumors further illuminated the immunogenic features of the ARID1A deficient TME that are conducive to supporting robust adaptive T cell immunity. For example, sgArid1a B16F10 tumors showed upregulation of antigen presentation and processing genes (Nlrc5, H2-K1, B2M) and immunogenic chemokines (Cxcl9, Cxcl13), which are also upregulated in human ARID1A mutant cancers in TCGA (Figure 2A, Figure S2A). Moreover, we observed downregulation of immunosuppressive genes Arg1, Ptgs2, and Nos2 in sgArid1a tumors (Figure 2A). Gene set enrichment analysis (GSEA) of gene signatures enriched in sgArid1a tumors revealed the top enrichments as IFN Alpha and IFN Gamma Responses (Figure 2B), pathways that were similarly enriched in TCGA gene expression data for human ARID1A mutant colorectal, stomach, cholangiosarcoma, liver, uterine, kidney, and pancreatic cancers (Figure 2C), but not melanoma or breast cancer. These data reveal sgArid1a tumors, like many ARID1A mutant human tumors, exhibit upregulation of immunogenic gene programs associated with the IFN Alpha and Gamma response pathways.

Figure 2. ARID1A loss induces a tumor transcriptome dominated by IFN response and is associated with altered CD8+ T cell states and T cell receptor clonality types.

(A) Heatmap of differentially expressed genes in sgArid1a and sgScramble B16F10 tumors. (B) GSEA Hallmarks enriched in sgArid1a B16F10 tumors. (C) GSEA curves of IFN Alpha and Gamma Response gene sets comparing ARID1A mutant versus non-mutant cancers from TCGA. (D) UMAP of CD8+ T cells sorted from sgScramble and sgArid1a tumors and subjected to scRNA-sequencing. (E-F) Cluster density and quantification as proportion of UMAP. (G) Anti-PD1 Response Signature gene set expression projected on UMAP. (H) GSEA Hallmarks enriched in C2 effector-like CD8+ T cells from sgArid1a tumors compared to sgScramble tumors. (I) CD8+ T cell metagene ISG score composed of the composite expression of genes from IFN Alpha and Gamma Response GSEA leading edge in C1–C5 projected on UMAP. (J) T cell receptor clonality classes projected onto UMAP with stacked bar plot quantification of classes. (K) Quantitation of the proportion of individual TCR clonotypes in sgArid1a and sgScramble tumors.

To investigate the anti-tumor CD8+ T cell response in ARID1A deficient tumors at single cell resolution, we performed scRNA-seq paired with T cell receptor sequencing (TCR-seq) on CD8+ T cells sorted from sgArid1a or sgScramble B16F10 tumors. Tumor infiltrating CD8+ T cells clustered into five clusters based on well-established marker genes: Naïve-like (Sell, Ccr7, Il7r), Effector-like (Irf7, Ccl5, Cxcr3), Exhausted (Tox, Havcr2), Innate-like (Fcer1g, Klra7), and Proliferating (Mki67, Top2a) (C1–C5) (Figure 2D, Figure S2B–C). We found that the majority of CD8+ T cells isolated from sgScramble tumors clustered in the C1 Naïve-like cluster while CD8+ T cells from sgArid1a tumors showed increased density in the C2 Effector-like cluster, C3 Exhausted cluster, C4 Innate-like cluster, and C5 Proliferating cluster (Figure 2E–F). The altered cellular states of CD8+ T cells in sgArid1a tumors confirms our flow cytometry analysis and reveals insights into CD8+ T cell differentiation states in our system.

Notably, we observed that CD8+ T cells from sgArid1a tumors displayed increased expression of a CD8+ T cell specific gene signature of anti-PD1 therapy response¹⁸ compared to sgScramble tumors (Figure 2G). The highest expressers of the anti-PD1 therapy response signature resided in the C2 effector-like cluster. GSEA analysis in C2 CD8+ T cells from sgArid1a tumors revealed top enrichments in IFN Alpha and Gamma Response signatures (Figure 2H). Indeed, we observed a striking enrichment in expression of an IFN stimulated gene (ISG) metagene composed of leading-edge genes from our GSEA analysis in CD8+ T cells from sgArid1a tumors, particularly in the C2 cluster (Figure 2I). This CD8+ T cell ISG enrichment may provide support for intra-tumoral T cell expansion in sgArid1a tumors as Type I IFN has been reported to promote T cell expansion in viral infection¹⁹ and CAR-T cells expressing a degradation resistant form of the Type I IFN receptor, IFNAR, show improved anti-tumor function and persistence in tumors²⁰.

Single cell TCR-seq analysis revealed that CD8+ T cells isolated from sgArid1a contained TCR clonotypes that were ‘hyperexpanded’ (>100 copies), in contrast to CD8+ T cells from sgScramble tumors which primarily consisted of single TCR clonotypes (Figure 2J–K). Specifically, C2 Effector-like, C3 Exhausted, and C5 Proliferating CD8+ T cells in sgArid1a tumors showed enrichment of T cells from large and hyperexpanded TCR clonotypes (Figure 2J), including clonotypes shared with sgScramble tumors and unique to sgArid1a tumors (Figure 2K). Together, these single cell CD8+ T cell gene expression and TCR repertoire data provide further insight into the sgArid1a mutant anti-tumor T cell response.

ARID1A Loss or cBAF Inhibition Induces a Cancer Cell Intrinsic IFN Response Signature that is Associated with Anti-Tumor Immunity and Immunotherapy Response in Patients

To investigate cancer cell intrinsic changes in gene expression that drive tumor immunity upon ARID1A deletion, we performed RNA-seq on sgArid1a B16F10 and MC38 cell lines as well as Arid1af/f:Actin-CreERT2 (Arid1a−/−) mouse embryonic fibroblasts (MEFs) cultured in vitro. GSEA analyses revealed that ARID1A loss results in upregulation of genes in the IFN Alpha and Gamma Responses (Figure 3A–3B, Figure S3A). We confirmed upregulation of ISG mRNA levels via qPCR in an additional sgArid1a engineered CT26 mouse colon cancer cell line (Figure S3B–C). As an orthogonal approach, we treated B16F10 cells with a cBAF small molecule inhibitor, ‘BD98’^21,22, the SMARCA4/2 ATPase inhibitor, BRM014²³, or a SMARCA4/2 degrader, ACBI1²⁴ for 72 hours. Similar to ARID1A genetic deletion, BD98 treatment resulted in robust induction of IFN Alpha and Gamma Response genes (Figure 3C); ACBI1 and BRM014 also induced ISG mRNAs and proteins (Figure S3D–F).

Figure 3. ARID1A loss or cBAF inhibition induces a cancer-cell intrinsic IFN gene expression signature associated with anti-tumor immunity and immunotherapy response.

(A-C) GSEA of Hallmark gene sets enriched in sgArid1a B16F10, sgArid1a MC38, and BD98 treated B16F10s, respectively. (D) Venn diagram of significantly upregulated genes in sgArid1a B16F10, sgArid1a MC38, and Arid1a^−/− MEFs with 57 commonly upregulated ISGs used to construct the ARID1A-IFN signature. (E) GSEA curves of ARID1A-IFN signature enrichment comparing ARID1A mutant vs non-mutant cancers from TCGA. (F) Pie chart of ARID1A mutation types in TCGA cohorts analyzed in this manuscript. (G) Tumor RPPA ARID1A levels between non-mutant, ARID1A nonsense/FS deletion, and ARID1A missense mutations in Colon, Stomach, and Uterine TCGA cohorts. (H-I) ARID1A-IFN signature and immune cytolytic scores between non-mutant, ARID1A nonsense/FS deletion, and ARID1A missense mutations in Colon, Stomach, and Uterine TCGA cohorts. (J) Survival pot of colon cancer patients stratified by top and bottom 20% of patients for ARID1A-IFN signature score. (K) ARID1A-IFN signature scores for non-mutant and ARID1A mutant colon cancer patients stratified by survival status at 100 months. (L) ARID1A-IFN signature scores association with Anti-PD1 response in stomach cancer patients across RECIST clinical response groups. Statistical analysis in Fig 3G–I were performed using Wilcoxon Rank Sum Test, Fig 3J log-rank test survival test, Fig 3K via two-way ANOVA, and Fig 3L via one-way ANOVA.

We constructed the “ARID1A-IFN signature” by taking ISGs that were commonly upregulated in ARID1A deficient B16F10, MC38, and MEF cell lines (Figure 3D), which resulted in a list of 57 ISGs including genes involved with antigen presentation (Tap1, Psmb8, Psmb9), IFN responsive transcription factors (Stat1, Stat2, Irf9), and T cell recruitment (Cxcl10). We next sought to determine whether the ARID1A-IFN signature is upregulated in ARID1A mutant human tumors and define its association with ICB response. Indeed, the ARID1A-IFN signature is enriched in ARID1A mutant human tumors in TCGA cohorts (Figure 3E). Among ARID1A mutations in cancer, the majority are either nonsense or frameshift deletions that are predicted to result in loss of function, and 20% are missense mutations whose impact on protein expression and function is unclear (Figure 3F). To investigate how different types of ARID1A mutations associate with our phenotypes of interest, we utilized TCGA multi-omics data for colorectal, stomach, and uterine cancers as ARID1A is frequently mutated in these cohorts and they include diverse types of ARID1A mutations (e.g. both nonsense/frameshift and missense). We observed that nonsense/frameshift deletions, but not missense mutations, were associated with decreased ARID1A protein levels (Figure 3G). Moreover, we observed significant increases in the ARID1A-IFN signature score and immune cytolytic score in nonsense/frameshift deletions, but not missense mutations, compared to non-mutant across tumor types (Figure 3H–I). Consistent with this, we found that cancer cell lines in the cancer cell line encyclopedia (CCLE) in the bottom quartile for ARID1A expression show increased expression of Type I IFNs, antigen presentation, and chemokines as well as many ARID1A-IFN signature genes compared to cell lines in the top quartile (Figure S3G–H). These data indicate that tumors harboring ARID1A mutations or epigenetic perturbations leading to loss of ARID1A expression are more likely to have ARID1A-IFN gene induction and associated anti-tumor immunity.

To assess the association between the ARID1A-IFN signature and clinical outcomes, we analyzed the Atlas and Compass of Immune-Cancer-Microbiome interactions in Colon Cancer (AC-ICAM) dataset²⁵. When examining all colon cancer patients in AI-ICAM, we found that the patients in the top quintile for ARID1A-IFN signature score had significantly improved overall survival compared to patients in the bottom quintile for ARID1A-IFN signature score (Figure 3J). Moreover, when we examined the ARID1A-IFN signature scores of non-mutant versus ARID1A mutant patients stratified by overall survival at 100 months, we discovered a selective and significant association with a high ARID1A-IFN signature score and long-term survival in ARID1A mutant patients but not in non-mutant patients (Figure 3K). Finally, we found a significant association with high ARID1A-IFN signature and complete responder patients as defined by Response Evaluation Criteria in Solid Tumors (RECIST) in a transcriptomics dataset of anti-PD1 ICB treatment in stomach cancer²⁶ (Figure 3L). Collectively, these bioinformatic analyses provide insights into which ARID1A mutations are most likely to contribute to an inflammatory TME and demonstrate significant associations between high ARID1A-IFN signature and beneficial clinical outcomes, including in the ICB setting.

ARID1A Loss Results in Type I IFN Pathway Activation and Primed Response to IFN-γ

Overlap between the ARID1A-IFN signature and IFN response pathway genes suggested that ISG induction following ARID1A loss is likely dependent on IFN signaling. To determine which IFN pathway is being activated in ARID1A deficient cancer cells, we utilized neutralizing antibodies to block the two main IFN signaling pathways in non-immune cells, Type I and Type III IFN. IFNAR blockade in sgArid1a or BD98 treated B16F10 cells reduced upregulation of ISG transcripts and ISG proteins, including STAT1, RIG-1, and ISG15, to control levels (Figure 4A–C). In contrast, Type III IFN blockade had no effect on upregulated ISGs in sgArid1a cells (Figure S4A). Consistent with IFNAR dependence, we found that the JAK kinase inhibitor, ruxolitinib, reduced upregulated ISG mRNA and protein in sgArid1a B16F10s to sgScramble levels (Figure S4B–C). Moreover, BD98 mediated induction of an IFN stimulated response element (ISRE)-GFP reporter in MEFs was completely dependent on JAK-STAT signaling (Figure S4D).

Figure 4. ARID1A loss activates Type I IFN pathway and enhanced response to IFNγ.

(A) Heatmaps of ISGs commonly upregulated in sgArid1a or BD98 treated B16F10 cells and their expression following anti-IFNAR treatment in each condition. (B-C) Western blots of ISGs RIG-I, STAT1, and ISG15 in sgArid1a (B) and BD98 treated (C) B16F10 cells with or without anti-IFNAR treatment. (D-E) qPCR for upregulated ISGs in sgArid1a B16F10s (D) or sgArid1a MC38s (E) with or without treatment with listed IFN blocking antibodies normalized to sgScramble mRNA levels. All sgArid1a to sgScramble comparisons are statistically significant at p >0.01 confidence and all sgArid1a vs sgArid1a + IFN blocking antibody comparisons significance results are shown above the IFN blocking antibody condition bar. (F) Median fluorescence intensity (MFI) quantification and representative histograms for ISRE-GFP signal in vehicle, BD98 treated, or BD98 treated + IFN blocking antibodies in ISRE-GFP reporter MEFs. (G) IFNβ ELISA results from sgScramble and sgArid1a B16F10 and MC38 cell supernatants. (H) IFNβ ELISA results from vehicle or BD98 treated MEFs supernatants. (I) Clustered heatmap of gene expression for IFNγ responsive genes in untreated or IFNγ treated sgScramble or sgArid1a B16F10 cells. (J) Heatmap of selected genes whose expression is higher in sgArid1a B16F10 cells following IFNγ treatment. (K-L) Flow cytometry histograms and median fluorescence intensity quantifications of MHC Class I and II in sgScramble or sgArid1a B16F10 cells with or without IFNγ treatment. (M) Quantification of percentage of dead B16F10s following co-culture assay between P14 CD8+ T cells and GP33 pulsed B16F10s with or without IFNγ pre-treatment of B16F10s. All data are represented as mean ± SD. Statistical analysis in Fig 4D–E was performed using two-way ANOVA and Fig 4F statistical analysis using one-way ANOVA.

We then tested the contribution of IFN beta (IFNβ) and six of thirteen IFN alphas (IFNα) to ISG induction in ARID1A deficient cells using neutralizing antibodies. In sgArid1a B16F10 and MC38 cancer cell lines, we observed that blockade of IFNβ, but not IFNα, partially recapitulated the effects of IFNAR blockade (Figure 4D–E). BD98 induced activation of an ISRE-GFP reporter in MEFs was similarly completely rescued by IFNAR blockade, partially reduced by IFNβ blockade and unaffected by IFNα blockade (Figure 4F). These results implied that ARID1A deficiency induces IFNβ, and indeed, we observed increased Ifnb1 mRNA in sgArid1a B16F10 and MC38 cells, albeit insignificant (Figure S4E), as well as significant Ifnb1 induction in Arid1a−/− and BD98 treated MEFs (Figure S4F–G). IFNβ was also significantly increased in the supernatants of sgArid1a MC38s and BD98 treated MEFs compared to controls (Figure 4G–H), but undetectable in B16F10 supernatants of either genotype. Consistent with a low level of Type I IFN signaling, there was no detectable induction of phosphorylated STAT1/2 downstream of IFNAR in sgArid1a B16F10 and MC38 cells (Figure S4H–I). Collectively, these results suggest that ISG induction in ARID1A deficient cells is dependent on IFNAR activation, partially via IFNβ.

As all three members of the IFN Stimulated Gene Factor 3 Complex or ISGF3 (STAT1, STAT2, IRF9) activated downstream of IFNAR are upregulated in the ARID1A-IFN signature, we wondered if ARID1A deficient cells might have enhanced genomic binding of ISGF3. Indeed, ISGF3 subunits were more strongly bound in Arid1a−/− MEFs compared to wildtype MEFs, including at the ISRE motif (Figure S5A–B), and ~70% of ISGs in the ARID1A-IFN signature showed enhanced binding of at least two ISGF3 subunits (Figure S5C–D). However, Ifnb1 and over 90% of the ARID1A-IFN signature genes showed no significant changes in chromatin accessibility by ATAC-seq (Figure S5E–G), suggesting that the mechanism of ARID1A-IFN gene induction is likely not mediated by epigenetic derepression of ISGs or Ifnb1.

We reasoned that ARID1A deficient cells might have an enhanced cellular response to IFNγ given that STAT1 is upregulated in sgArid1a cells. We therefore treated sgArid1a B16F10 cells in vitro with or without IFNγ for 24 hours followed by RNA-seq. We found that genes upregulated by IFNγ fell into two clusters; cluster 1 was uniformly upregulated in sgArid1a and sgScramble cells, while cluster 2 was predominantly elevated in sgArid1a cells at baseline and primed for hyperactivation following IFNγ stimulation (Figure 4I). The primed cluster included genes involved in immune cell recruitment (Cxcl10, Il18, Ccl5) and antigen presentation, including MHC Class I, MHC Class II, and immunoproteasome genes (Figure 4J). Additionally, MHC Class I and MHC Class II cell surface expression was elevated at baseline and hyper-induced following IFNγ stimulation in sgArid1a B16F10 and MC38 cells via flow cytometry (Figure 4K–L, Figure S4J–K). To test if the sgArid1a enhanced response to IFNγ has a functional effect on tumor killing by T cells, we cultured P14 TCR transgenic CD8+ T cells pre-activated by anti-CD3/CD28 and B16F10 cells pulsed with the P14 cognate antigen, GP33 peptide, with or without pre-treatment of B16F10 cells with IFNγ. Strikingly, sgArid1a B16F10s were more susceptible to T cell killing, particularly in the IFNγ stimulated condition, where approximately 2.5 times as many sgArid1a B16F10s were killed compared to sgScramble (Figure 4M). From these data, we posit that elevated basal chemokine expression in ARID1A deficient tumor cells likely recruits T cells into the tumor, wherein T cell derived IFNγ further induces primed tumor cell antigen presentation complexes resulting in increased T cell killing of ARID1A deficient tumor cells.

ARID1A Loss or cBAF Inhibition Induce R-loop Derived Cytosolic DNA that Drives the ARID1A IFN Signature

To determine the molecular events upstream of Type I IFN induction in ARID1A deficient cancer cells, we began investigating endogenous pathogen associated molecular patterns (PAMPs) known to induce Type I IFN via viral mimicry. First, we considered that ARID1A loss may result in the de-repression of endogenous retroviruses (ERVs) prompting double stranded RNA (dsRNA) sensing and induction of IFNs as has been reported with other SWI/SNF complex subunits^27,28. However, sgArid1a and BD98 treated B16F10 cells showed a significant reduction of cytosolic dsRNA relative to sgScramble cells, despite robust detection of dsRNA following transfection of Poly(I:C) (Figure S6A). Additionally, we observed an overall decrease in ERV expression via Total-RNA sequencing analyses in sgArid1a B16F10 (Figure S6B). These data suggest that derepression of ERVs is unlikely to drive the ARID1A-IFN signature.

We next investigated viral mimicry in the form of R-loop derived cytosolic ssDNA, which is known to trigger Type I IFN^29–31, as ARID1A loss has recently been associated with accumulation of R-loops^32–35. To assay for R-loops in ARID1A deficient cells, we purified a catalytically inactive RNaseH1 mutant protein with GFP tag (dRNH1-GFP) that binds to RNA:DNA hybrids and displays superior RNA:DNA hybrid specificity to the widely used S9.6 antibody^36,37. We discovered that sgArid1a B16F10 as well as BD98 or BRM014 treated B16F10 showed significantly elevated levels of nuclear R-loops and cytoplasmic RNA:DNA hybrids (Figure 5A). The dRNH1-GFP signal in sgArid1a B16F10 was sensitive to RNaseH treatment, confirming the specificity of this reagent for RNA:DNA hybrids (Figure S6C). In addition, sgArid1a B16F10 and BD98, BRM014, or ACBI1 treated B16F10 exhibited significantly elevated levels of cytosolic ssDNA (Figure 5B, S6D), which were abolished following transfection of S1 nuclease which degrades ssDNA, but not dsDNA (Figure S6E). Collectively, these findings support recent reports implicating SWI/SNF complexes in R-loop resolution^32–35 and demonstrate that ATP dependent chromatin remodeling is required for R-loop resolution, as BRM014 ATPase inhibition phenocopies ARID1A deletion and cBAF inhibition.

Figure 5. ARID1A loss or cBAF inhibition induces R-loop derived cytosolic DNA that drives the ARID1A IFN Signature.

(A) Immunofluorescence of dRNH1 and DAPI and quantification of dRNH1-GFP (RNA:DNA hybrids) in B16F10 in sgScramble and sgArid1a B16F10s or B16F10 cells treated with BD98 or BRM014 for 72hrs (scale bar = 5μM). (B) Immunofluorescence of ssDNA and DAPI and quantification of cytosolic ssDNA in sgScramble and sgArid1a B16F10s or B16F10 cells treated with BD98 or BRM014 for 72hrs (scale bar = 20μM). (C) Immunofluorescence of dRNH1 and DAPI and quantification of dRNH1 in sgScramble and sgArid1a B16F10 cells with or with overexpression of TREX1 or RNASEH2B (scale bar = 5μM). (D) Immunofluorescence of ssDNA and DAPI and quantification of cytosolic ssDNA in sgScramble and sgArid1a B16F10 cells with or with overexpression of TREX1 or RNASEH2B. (E) Western blots of ISGs STAT1 and ISG15 induced in sgArid1a B16F10s with or without overexpression of TREX1 or RNASEH2B (scale bar = 10μM). (F) Immunofluorescence of ssDNA and DAPI and quantification cytosolic ssDNA in MEFs treated with vehicle or BD98 with or without overexpression of TREX1 or RNASEH2B (scale bar = 30μM). (G) Western blots of ISGs RIG-1, STAT1, and ISG15 induced with BD98 treatment in MEFs with or without overexpression of TREX1 or RNASEH2B. (H) IFNβ ELISA using supernatants from MEFs treated with vehicle or BD98 with or without overexpression of TREX1 or RNASEH2B. Data are represented as mean ± SD. Vehicle and BD98 data is identical to Figure 4H. (I) Western blots of ISGs RIG-1, STAT1, and ISG15 induced with in sgArid1a MC38s with or without overexpression of TREX1 or RNASEH2B. (J) IFNβ ELISA using supernatants from sgScramble or sgArid1a MC38s treated with or without overexpression of TREX1 or RNASEH2B. Data are represented as mean ± SD. sgScramble and sgArid1a data is identical to Figure 4G. Statistical analyses for Figures 5A–D, 5F, 5H, and 5J were performed via one-way ANOVA.

We performed a BD98 time-course and found close concordance between the appearance of cytosolic ssDNA and STAT1 induction following BD98 treatment with maximal induction at 72 hours (Figure S6F–G), suggesting that cytosolic ssDNA³⁸ could be an inflammatory trigger upon ARID1A deficiency. To further investigate the relationship between R-loops, cytosolic ssDNA, and the ARID1A-IFN signature, we overexpressed the cytosolic DNase TREX1 or the RNA:DNA hybrid resolving nuclease, RNASEH2B, in ARID1A deficient or BD98 treated cells. As expected, TREX1 overexpression significantly reduced elevated levels of cytosolic RNA:DNA hybrids (Figure 5C), while RNASEH2B overexpression significantly reduced elevated levels of both nuclear and cytosolic RNA:DNA hybrids in sgArid1a B16F10s (Figure 5C). Overexpression of TREX1 and RNASEH2B also reduced cytosolic ssDNA in sgArid1a B16F10 and BD98 treated MEFs (Figure 5D, 5F). Critically, in ARID1A deficient cancer cells and BD98 treated MEFs, TREX1 or RNASEH2B overexpression reduced upregulated ISGs and IFNβ secretion (Figure 5E, 5G–J, S6H–I). These data thus support a model whereby R-loop derived cytosolic DNA species drive expression of the ARID1A-IFN signature.

ARID1A-IFN Signature and Anti-Tumor Immunity is Dependent on STING Cytosolic DNA Sensing

We next investigated cGAS-STING cytosolic DNA sensing in relation to the ARID1A-IFN signature and anti-tumor immunity as this pathway has previously been reported to be activated by cytosolic ssDNA and RNA:DNA hybrids^38–40. We generated sgArid1a B16F10 cells with CRISPR-Cas9 mediated cGAS (‘Cgas dKO’) or STING knockout (‘Sting dKO’) (Figure 6A). In Cgas and Sting dKO cell lines, ISG mRNA and protein upregulation in sgArid1a B16F10 was restored to sgScramble levels (Figure 6A–B). Notably, cGAS- and STING-dependent genes in sgArid1a B16F10 were enriched in Type I IFN response genes (Figure 6C). ISG protein expression in both sgArid1a B16F10 and MC38 cells was also reduced by STING or TBK1 inhibition with H-151 and BX-795, respectively (Figure S7A–B). BD98 failed to induce ISGs in Cgas−/− and Sting−/− MEFs despite robust activation in wild-type MEFs (Figure S7C–D). Moreover, the significant induction of IFNβ secretion in sgArid1a MC38s and BD98 treated MEFs was abolished upon STING pathway perturbations (Figure S7E–F). Supporting low level activation of the STING pathway in ARID1A deficient cells, we observed a modest but significant increase in phospho-TBK1 via both microscopy and western blot in sgArid1a B16F10, which could be enhanced via STING agonist treatment (Figure S7G–H). Additionally, activation of the STING pathway was required for the IFNγ mediated priming of MHC-Class I and II in sgArid1a B16F10 cells (Figure S7I). Together, these data demonstrate that the cGAS-STING cytosolic DNA sensing pathway is required for induction of ISGs and IFNβ following either ARID1A genetic deletion or cBAF inhibition.

Figure 6. ARID1A loss induced ISG induction and anti-tumor immunity is dependent on cGAS-STING cytosolic DNA sensing Pathway.

(A) Western blot of ARID1A, cGAS, STING, and ISG15 in sgScramble sgArid1a, sgArid1a/sgCgas (cGAS dKO) and sgArid1a/sgSting (Sting dKO) B16F10s. (B) Heatmap of genes induced in sgArid1a that are dependent on cGAS-STING. (C) Over-representation analysis of sgArid1a induced cGAS-STING dependent genes against Hallmark gene sets. (D) B16F10 tumor growth curves comparing sgScramble, sgArid1a, and Sting dKO genotypes. Data are represented as mean ± SEM. (E-F) Tumor infiltrating immune populations whose significant increase in sgArid1a tumors are rescued in Sting dKO tumors. (G-H) Quantification and flow cytometry contour plots of CD8+ T cell TNFα and IFNγ or Granzyme B and IFNγ staining following ex vivo PMA/ionomycin stimulation in listed tumor genotypes. (I) Quantification and flow cytometry contour plots of CD4+ T cell IFNγ and TNFα staining following ex vivo PMA/ionomycin stimulation in listed tumor genotypes. All data are represented as mean ± SD unless otherwise noted. Statistical analysis in Fig 6E and Fig6E–I were performed via one-way ANOVA.

To probe the requirement of STING cytosolic DNA sensing in ARID1A deficient anti-tumor immunity, we injected mice with sgScramble, sgArid1a, and Sting dKO B16F10 and observed a significant rescue of sgArid1a tumor growth with added Sting deficiency (Figure 6D). This was especially notable as we observed slower proliferation in vitro of Sting dKO B16F10 relative to sgScramble and sgArid1a B16F10 (Figure S7J). Flow cytometry immunoprofiling revealed a significant reduction in CD45+ and NK cells in Sting dKO tumors (Figure 6E–F). Strikingly, the increased percentage of TNFα/IFNγ+ CD8+ and CD4+ T cells and GZMB/IFNγ+ CD8+ T cells in sgArid1a tumors was completely rescued to sgScramble levels in Sting dKO tumors (Figure 6G–I). Cancer cells are known to silence immunogenic regulators to mediate immune escape, and indeed, we found that ARID1A mutant cancer cell lines have significantly downregulated STING and JAK1 protein compared to non-mutant cancer cell lines (Figure S7K). Collectively, these data demonstrate that ARID1A deficient cancer cell STING is required for increased immune cell recruitment into the tumor, increased T cell functionality, and slower tumor growth phenotype, and that some ARID1A mutant cancers downregulate STING as a mechanism of immune evasion.

ARID1A Deficient Anti-Tumor Immunity and Immunotherapy Response is Type I IFN Dependent

To explore the requirements of anti-tumor immunity and immunotherapy responsiveness in relation to R-loop driven STING-Type I IFN signaling in ARID1A deficient cancer cells, we engrafted sgArid1a and sgScramble B16F10 tumors into mice treated systemically with an anti-IFNAR blocking or isotype control antibody. We observed a significant rescue of tumor growth in sgArid1a tumors treated with IFNAR blockade (Figure 7A), accompanied by a significant reduction in tumor infiltrating immune cells, total and GZMB/IFNγ+ NK cells (Figure 7B–D), and TNFα/IFNγ+ CD8+ and CD4+ T cells (Figure 7E–F). These data are consistent with the Sting dKO tumor growth and immune-profiling data and further demonstrate that Type I IFN signaling is required for robust anti-tumor immunity in this ARID1A deficient tumor model.

Figure 7. ARID1A deficient anti-tumor immunity and immune checkpoint blockade response is Type I IFN dependent.

(A) Tumor growth curves in sgArid1a and sgScramble B16F10 with or without anti-IFNAR treatment. Data are represented as mean ± SEM. (B-C) Quantification of tumor infiltrating immune populations whose significant increase in sgArid1a tumors are rescued in sgArid1a tumors treated with anti-IFNAR. (D) Quantification of GZMB/IFNγ+ NK cells in sgArid1a or sgScramble tumors with and without anti-IFNAR treatment. (E) Quantification of TNFα/IFNγ positive CD4+ T cells in sgArid1a or sgScramble tumors with and without anti-IFNAR treatment. (F) Quantification and flow cytometry contour plots of TNFα/IFNγ positive CD8+ T cells in sgArid1a or sgScramble tumors with and without anti-IFNAR treatment. (G) Individual B16F10 tumor growth curves of sgScramble and sgArid1a tumors treated with anti-PDL1 and anti-CTLA4 or sgArid1a tumors treated with anti-IFNAR, anti-PDL1, and anti-CTLA4. (H) Kaplan-Meier survival curve of mice flank injected with sgScramble or sgArid1a B16F10 tumors treated with anti-PDL1 and anti-CTLA4 or sgArid1a tumors treated with anti-IFNAR, anti-PDL1, and anti-CTLA4. (I) Individual CT26 tumor growth curves of sgScramble and sgArid1a tumors treated with anti-PDL1 or sgArid1a tumors treated with anti-IFNAR and anti-PDL1. (J) Kaplan-Meier survival curve of mice flank injected with sgScramble or sgArid1a CT26 tumors treated with anti-PDL1 or sgArid1a tumors treated with anti-IFNAR and anti-PDL1. (K) Individual MC38 tumor growth curves of sgScramble and sgArid1a tumors treated with anti-PDL1 or sgArid1a tumors treated with anti-IFNAR and anti-PDL1. (L) Kaplan-Meier survival curve of mice flank injected with sgScramble or sgArid1a MC38 tumors treated with anti-PDL1 or sgArid1a tumors treated with anti-IFNAR and anti-PDL1. Statistical analysis in Fig 7B–F was performed via two-way ANOVA. All data are represented as mean ± SD unless otherwise noted. Statistical analysis in Fig 7H, 7J, and 7L were performed using the log-rank survival test.

To test the impact of our findings in the context of a clinically relevant therapeutic intervention, we treated sgScramble or sgArid1a B16F10, CT26, and MC38 tumors with ICB immunotherapy +/− IFNAR blockade. In the B16F10 tumor model which is known to be poorly responsive to ICB^41,42, we observed a significantly improved ICB response in sgArid1a tumors compared to sgScramble tumors, which was significantly blunted by IFNAR blockade (Figure 7G–H). We observed similar results in the CT26 colorectal tumor model, which is also poorly responsive to ICB¹⁸ (Figure 7I–J), and the MC38 colorectal tumor model, which is ICB responsive^18,41 (Figure 7K–L). These data demonstrate that Type I IFN signaling is required for improved ARID1A deficient tumor immunotherapy responses and thereby connect our findings of the ARID1A deficient R-loop driven STING-Type I IFN signaling axis to the clinical efficacy of ICB immunotherapies in ARID1A mutant tumors.

DISCUSSION

In this study, we demonstrate that ARID1A loss in tumor cells induces R-loops which give rise to cytosolic DNA species that activate STING-Type I IFN signaling, inducing an ARID1A-IFN gene expression signature that promotes anti-tumor immunity. We demonstrate that ARID1A loss is sufficient to induce anti-tumor immunity and ICB response in murine tumor models and that these models recapitulate features of human ARID1A mutant tumors such as enhanced T cell infiltration and inflammatory cytokine expression. These data provide insight into the molecular mechanisms that regulate the enhanced responsiveness of ARID1A mutant human tumors to ICB observed in clinical trials. Further, our results predict that ARID1A mutant human tumors exhibiting the ARID1A-IFN signature will be responsive to ICB immunotherapy. As ARID1A is mutated in a myriad of tumor types with diverse mutational landscapes, not all ARID1A mutant human tumors display the ARID1A-IFN gene signature we describe here. Future work will determine in what context ARID1A mutations induce the ARID1A-IFN signature and whether the molecular features identified here have predictive utility in stratifying patients with ARID1A mutant tumors as ICB responsive versus non-responsive. Intriguingly, R-loops are associated with DNA replication stress, which was recently reported as a biomarker of ICB response in non-hypermutated cancers³⁰. Thus, the ARID1A-IFN signature we discovered originating from R-loops may represent a fundamental mechanism by which tumor cell replication stress can trigger an anti-tumor immune response, even in the absence of ARID1A mutation or hyper-mutated phenotypes.

By utilizing scRNA/TCR sequencing and flow cytometry immune profiling, we demonstrate there is an increase in CD8+ T cell tumor infiltration, TCR clonotype expansion, and altered gene expression states in sgArid1a tumors compared to sgScramble tumors. In particular, CD8+ T cells in sgArid1a tumors exhibited a shift from Naïve like and towards more Effector like and Exhausted clusters that perform key anti-tumor functions such as self-renewal following ICB and cytotoxicity, respectively¹⁷. Moreover, Effector like CD8+ T cells from sgArid1a tumors displayed robust ISG expression, which has been associated with T cell expansion in anti-tumor immunity²⁰ and viral infection^19,43. The association of these CD8+ T cell signatures with ICB response and T cell expansion, respectively, suggest that these T cell states contribute to the robust anti-tumor immune response in sgArid1a tumors. TCR clonotype analyses on CD8+ T cells revealed that sgArid1a tumors drive hyperexpansion of clonotypes shared with CD8+ T cells in sgScramble tumors and unique to CD8+ T cells in sgArid1a tumors. Future work will determine whether ARID1A deficiency increases the rate of tumor neoantigen generation and presentation in tumors.

Our findings that ARID1A loss results in R-loop accumulation are consistent with recent reports demonstrating that SWI/SNF complexes are required for effective R-loop resolution^32–35. The ATP dependent chromatin remodeling function of SWI/SNF was required for the resolution of R-loops, suggesting that cBAF may remodel chromatin to facilitate binding of R-loop resolution factors as has been recently reported^32–35. Furthermore, we discovered that ARID1A mutant cancer cell lines in the DepMap RNAi screening project are highly dependent on genes annotated as R-loop regulators or interactors⁴⁴ (Figure S7L–M). These data suggest R-loop accumulation could be a widely relevant consequence of ARID1A mutations and could render ARID1A mutant cancers selectively vulnerable to inhibition of R-loop resolution factors, as has been shown for ATR inhibitors^22,45,46.

We demonstrate that ARID1A loss induced R-loops and resulting STING dependent cytosolic DNA sensing are the major effectors of the ARID1A-IFN signature. This suggests that the roles of ARID1A in enhancer function or DNA repair are secondary to its role in R-loop resolution with regard to the ARID1A-IFN signature. For example, we observed no significant changes in chromatin accessibility following ARID1A loss at IFN or ISG loci in B16F10 cells or HCT116 cells⁴⁷. Accordingly, ARID1A deficient cancer cells have increased cytosolic ssDNA and RNA:DNA hybrids, both of which have been reported to activate cGAS-STING^38–40. Further work is needed to determine the relative contribution of these two cytosolic nucleic acid species in triggering cGAS-STING activation and to delineate how ssDNA and RNA:DNA hybrids are liberated from R-loops allowing for cytoplasmic release. Finally, we hypothesize that the R-loop triggered response we describe here is not mutually exclusive with previous work reporting that hypermutability in ARID1A mutant cancers promotes ICB response¹², but that these mechanisms may act synergistically in promoting an optimal ICB response in ARID1A mutant cancers.

Interestingly, our work suggests that relatively modest increases in tumor cell derived Type I IFN are sufficient to trigger anti-tumor immunity. Indeed, although we observed significant increases in IFNβ secretion and a requirement for IFNAR and JAK-STAT signaling for induction of the ARID1A-IFN signature, we did not detect phosphorylated STAT1/2, which is associated with canonical Type I IFN signaling for example during viral infection, but based on our work is not a conclusive assay for the presence of Type I IFN signaling. We also note that tumor IFN gene expression signatures have recently been shown to promote tumor-immune evasion^48,49. Notably, the ARID1A-IFN signature bears less overlap with IFN signatures that are associated with immunotherapy resistance^48,50,51 compared to IFN signatures associated with PAMPs^52–54 known trigger anti-tumor immunity. Nevertheless, future studies are needed to understand how IFN promotes anti-tumor immunity in the ARID1A mutant TME. Our studies suggest that IFNAR signaling on tumor cells contributes to the surveillance of ARID1A mutant tumors as anti-IFNAR blockade in vitro reduces expression of ISGs associated with lymphocyte chemoattraction and antigen presentation. Additionally, CD8+ T cell ISG expression revealed in scRNA-seq data suggest that CD8+ T cells also respond to Type I IFN in the ARID1A mutant TME. Delineating the respective contributions of IFNAR signaling on tumor and non-tumor cells in the ARID1A mutant TME will help elucidate the effectiveness of IFN in ARID1A mutant tumor surveillance and ICB sensitivity.

Finally, we demonstrate that available SWI/SNF inhibitors phenocopy ARID1A mutation with respect to induction of the ARID1A-IFN signature. Thus, it is conceivable that cBAF inhibitors could be used to convert immunologically cold tumors to more immune infiltrated hot tumors by inducing R-loop driven ARID1A-IFN induction. cBAF inhibitors may also have an added benefit of boosting the anti-tumor activity of CD8+ T cells as CD8+ T cell intrinsic ARID1A loss has recently been reported to prevent exhaustion phenotypes and enhance memory characteristics and anti-tumor function^55–58. Thus, cBAF inhibition could effectively serve as a one-two punch increasing tumor cell immunogenicity and enhancing CD8+ T cell function. This will be an intriguing therapeutic avenue as pharmacological inhibitors of the SWI/SNF complex are actively being developed for the treatment of various tumor types and have been reported to have minimal baseline toxicity^59,60. However, SWI/SNF inhibition carries concerns such as off target toxicity and potential tumorigenic effects which should be carefully addressed in future studies. Collectively, these findings represent a previously unappreciated molecular mechanism by which ARID1A mutation can promote anti-tumor immunity and may have translational utility in improving patient selection for ICB as well as providing a rationale for cBAF inhibitors to augment ICB efficacy.

LIMITATIONS OF STUDY

While we observe strong concordance between our sgArid1a models and human ARID1A mutant cancer with regard to increased IFN response signature, increased T cell infiltration, and improved ICB response, we note that syngeneic tumor models are not able to address questions surrounding the effects of ARID1A mutations that occur naturally during the course of tumor evolution. Thus, while syngeneic models are an effective method to mechanistically understand the ICB response in ARID1A mutant cancers, questions on how the inflammatory effects of ARID1A mutation shape tumor evolution will need to be addressed in future studies.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Diana C. Hargreaves (dhargreaves@salk.edu).

Materials Availability

This study did not generate new unique reagents.

Data and Code Availability

• All sequencing data generated in this paper can be accessed from GEO and are publicly available as of the date of publication. Accession numbers are listed in the key resources table.

• All original code has been deposited at Zenodo and are available at https://github.com/mbmaxwell. Zenodo DOI is listed in the key resources table.

• Additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REGENT OR RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-mouse CD45	BioLegend	103130
Anti-mouse CD8a	BD Biosciences	563786
Anti-mouse CD4	BD Biosciences	612843
Anti-mouse NK1.1	BioLegend	108708
Anti-mouse CD3e	eBiosciences	16-0031-82
Anti-mouse TIM3	BioLegend	119727
Anti-mouse SLAMF6	BioLegend	134606
Anti-mouse PD1	BioLegend	114117
Anti-mouse FOXP3	eBiosciences	11-5773-82
Anti-mouse Ki67	BD Biosciences	566109
Anti-mouse TNFα	BD Biosciences	563944
Anti-mouse IFNγ	BioLegend	505826
Anti-mouse Granzyme B	BioLegend	515403
Anti-mouse CD11b	BioLegend	101230
Anti-mouse CD11c	BioLegend	117310
Anti-mouse Ly6G	BioLegend	127624
Anti-mouse Ly6C	BioLegend	128007
Anti-mouse CD24	BioLegend	101805
Anti-mouse CD103	BD Biosciences	749393
Anti-mouse F4/80	BD Biosciences	565614
Anti-mouse MHC Class I	BioLegend	114718
Anti-mouse MHC Class II	BioLegend	107643
Anti-mouse IFNAR	Leinco	I-401
Anti-mouse IgG1	Leinco	I-536
Anti-mouse PDL1	Bio X Cell	BE0101
Anti-mouse IgG2b	Bio X Cell	BE0086
Anti-mouse CTLA4	Leinco	C2856
Anti-mouse IgG2b	Leinco	M1415
Anti-IFNα	Leinco	I-1183
Anti-IFNβ	Leinco	I-1182
Anti-IFNλ2/3	R&D Systems	MAB17892
Anti-Rat IgG2b	R&D Systems	MAB0061
Anti-mouse/human ARID1A	Santa Cruz Biotechnology	sc-32761
Anti-mouse/human SMARCA4 (BRG1)	Abcam	Ab110641
Anti-mouse STAT1	Cell Signaling Technology	9172
Anti-mouse RIG-I	Cell Signaling Technology	3743
Anti-mouse ISG15	Santa Cruz Biotechnology	sc-166755
Anti-mouse STAT2	Cell Signaling Technology	72604
Anti-mouse IRF9	Millipore Sigma	MABS1920
Anti-mouse cGAS	Cell Signaling Technology	31659
Anti-mouse STING	Cell Signaling Technology	13647
Anti-mouse Phospho-TBK1 (S172)	Cell Signaling Technology	5483
Anti-mouse Phospho-STAT1 (Tyr701)	Cell Signaling Technology	7649S
Anti-mouse Phospho-STAT2 (Tyr689)	Millipore Sigma	07–224
Anti-mouse α-Tubulin	Cell Signaling Technology	2144
Anti-mouse β-Actin	Cell Signaling Technology	3700
Anti-Single Stranded DNA	Millipore Sigma	MAB3868
Anti-Double Stranded RNA (J2)	Millipore	MABE1134
Anti-CD8a (IHC)	ThermoFisher Scientific	MA5–13473
Chemicals, peptides, and recombinant proteins
H-151	Invivogen	inh-h151
BX-795	Invivogen	Tlrl-bx7
Ruxolitinib	Invivogen	Tlrl-rux
Recombinant Murine IFNγ	Peprotech	315–05
Recombinant Murine IFNα	BioLegend	752804
BD98	Laboratory of Dr. Emily Dykhuizen	NA
ACBI1	Selleck Chemicals	S9612
BRM014	MedChemExpress	HY-119374
diAZBI	Invivogen	Tlrl-diabzi-2
Poly(I:C) HMW	Invivogen	Tlrl-pic
Bredfeldin A	BioLegend	420601
Foxp3/Transcription factor staining buffer set	ThermoFisher Scientific	00-5523-00
10X Permeabilization buffer	ThermoFisher Scientific	00-8333-56
DNase I	Roche	11284932001
Liberase	Roche	5401020001
Zombie Red	BioLegend	423110
Ionomycin	Stem Cell Technologies	73724
PMA	Sigma	P1585
Puromycin	ThermoFisher Scientific	A1113803
DAPI	Sigma	D9542
S1 nuclease	ThermoFisher Scientific	EN0321
RNaseH	New England Bio Labs	M0297L
Rnase A	ThermoFisher Scientific	EN0531
Proteinase K	ThermoFisher Scientific	17916
ProLong Glass Mounting Media	ThermoFisher Scientific	P36980
Ultracomp beads	ThermoFisher Scientific	01-2222-42
Precision count beads	BioLegend	424902
Fc receptor blocking antibody	Tonbo	70–0161
RBC lysis buffer	BioLegend	420301
Percoll	GE Healthcare	17-0891-01
Bis-Tris gels	Life Technologies	NW04120BOX
Quick-RNA miniprep kit	Zymo	R1055
Amicon Ultra-15	Millipore Sigma	UFC901024
Amicon Ultra-0.5	Millipore Sigma	UFC500396
Pierce Pro-Ject Protein Transfection Reagent	ThermoFisher Scientific	89850
Protein A bead	ThermoFisher Scientific	10002D
Protein G bead	ThermoFisher Scientific	15027866
Tn5 transposase	Illumina	15027865
1x Tagment DNA buffer	Illumina	15027866
AMPure XP Beads	Beckman Coulter	A63882
Critical Commercial Assays
Mouse IFN beta ProQuantum Immunoassay Kit	Thermo Scientific	A47435
CellTiter-Glo	Promega	G756A
Akoya 4 color IHC Kit	Akoya Biosciences	NEL810001KT
VECTASTAIN® Elite ABC-HRP Kit	Vector Laboratories	PK-6200
Chromium Next GEM Single Cell 5’ Reagent Kit v2	10X Genomics	PN-1000256
Chromium Single Cell Mouse TCR Amplification Kit	10X Genomics	PN-1000254
Deposited data
RNA-seq	This paper	GEO: GSE217803, Zenodo: 10.5281/zenodo.10936801
scRNA/TCR-seq	This paper	GEO: GSE217806, Zenodo: 10.5281/zenodo.10936801
ChIP-seq	This paper	GEO: GSE217805, Zenodo: 10.5281/zenodo.10936801
ATAC-seq	This paper	GEO: GSE217804, Zenodo: 10.5281/zenodo.10936801
Experimental models: Cell lines
B16F10 melanoma cells	ATCC	CRL-6475
MC38 colon cancer cells	Laboratory of Dr. Gerald Shadel	NA
CT26 colon cancer cells	ATCC	CRL-2638
Arid 1af/f:Actin-CreERT2 MEF cells	Hargreaves Lab	NA
Wildtype, Cgas−/−, Sting−/− MEF cells	Laboratory of Dr. Daniel Stetson	NA
Experimental models: Organisms/Strains
C57BL/6J mice	Jackson Labs	000664
B6.129S7-Rag1tm1Mom/J (Rag1−/− mice)	Jackson Labs	002216
Recombinant DNA
PX458-Cas9-GFP plasmid	Addgene	48138
dRNH1-GFP plasmid	Addgene	174448
TREX1 plasmid	Addgene	27218
RNASEH2B plasmid	Addgene	108697
N103 lentiviral construct	Laboratory of Dr. Nate Hathaway	Headley et al., 2019
N103-TREX1	This study	NA
N103-RNASEH2B	This study	NA
Software and algorithms
R v4.05	CRAN	https://cran.r-project.org/
Homer	Heinz et al., 2010	http://homer.ucsd.edu /homer/
STAR V2.5	Dobin et al., 2013	https://github.com/alexdobin/STAR
DESeq2 v1.12.4	Love et al., 2014	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
edgeR	Robinson et al., 2010	https://bioconductor.org/packages/release/bioc/html/edgeR.html
GSEA	Subramanian et al., 2005	http://software.broadinstitute.org/gsea/ index.jsp
CellRanger v3.02	NA	https://github.com/10XGenomics/cellranger
Cibersort	Chen et al., 2018	https://cibersortx.stanford.edu/
ImageJ	Open source	https://imagej.net/software/fiji/
CellProfiler	Stirling et al., 2021	https://cellprofiler.org/
FlowJo v10.8.1	NA	https://www.flowjo.com/solutions/flowjo/downloads
ImageStudio	NA	https://www.licor.com/bio/image-studio/
QuPath	Bankhead et al, 2017	https://qupath.readthedocs.io/en/stable/index.html
ggplot2	Wickham et al., 2016	https://ggplot2.tidyverse.org/
dplyr	Wickham et al., 2023	https://dplyr.tidyverse.org/
ComplexHeatmap	Gu et al., 2022	https://jokergoo.github.io/ComplexHeatmap-reference/book/
ClusterProfiler	Yu et al., 2012	https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html
TCGAbiolinks	Mounir et al., 2019	https://bioconductor.org/packages/release/bioc/html/TCGAbiolinks.html
Immunedeconv	Sturm et al., 2019	https://github.com/omnideconv/immunedeconv
WebGestaltR	Liao et al., 2019	https://rdrr.io/cran/WebGestaltR/man/WebGestaltR.html
Seurat	Hao and Hao et al., 2021	https://satijalab.org/seurat/
Survival	Therneau, 2021	https://cran.r-project.org/web/packages/survival/index.html
Survminer	Kassambara, 2021	https://cran.r-project.org/web/packages/survminer/index.html
scDataviz	Bligh et al., 2020	https://github.com/kevinblighe/scDataviz
ggpubfigs	Steenwyk et al., 2021	https://github.com/JLSteenwyk/ggpubfigs
Nebulosa	Alquicira-Hernandez et al., 2021	https://github.com/powellgenomicslab/Nebulosa
scRepertoire	Borcherding et al., 2022	https://github.com/ncborcherding/scRepertoire

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell lines

B16F10 melanoma, MC38 colon cancer, CT26 colon cancer, and MEF cell lines were cultured in DMEM (Corning), 10% FBS, and 1% Penicillin/Streptomycin. All cell lines used in this study are mouse cell lines. All cells were grown at 37°C with 5% CO₂. All cell lines were negative for mycoplasma when tested by MycoAlert Mycoplasma Detection Kit (Lonza, Basel, Switzerland).

Mice

C57BL/6J or B6-Rag1−/− mice were purchased from Jackson Laboratories. Male mice were used in this study and were six to eight weeks of age at the beginning of tumor studies. Animals were housed in specific-pathogen-free facilities at the Salk Institute. All experimental studies were approved and performed in accordance with guidelines and regulations implemented by the Salk Institute Animal Care and Use Committee.

Human patient samples

Human endometrial cancer formalin fixed paraffin embedded (FFPE) slides were obtained from the UCSD Moore’s Cancer Center in accordance with IRB-approved study (#181313). Seven FFPE slides were analyzed in this study. Clinical variables such as age, ethnicity, and disease stage were not controlled for. Patient characteristics are indicated in Table S4.

METHOD DETAILS

CRISPR-Cas9 genetic engineering

Gene targeting by CRISPR/Cas9 was achieved by transfection of pSpCas9(BB)-2A-GFP PX458 (Addgene, 48138) with guide RNA sequences cloned into the Cas9 vector (Table S3). Successful targeting of genes of interest following selection was confirmed by sanger sequencing and immunoblotting. Single cell ARID1A deficient clones and ARID1A deficient pools were generated to confirm phenotypes of interest were a result of ARID1A deficiency and not clonal variation. Cgas or Sting1 targeting sgRNAs were transfected into sgArid1a B16F10 cells to generate Cgas dKO and Sting dKO cell lines. Scramble non-targeting controls were transfected into existing sgScramble or sgArid1a B16F10 cells in parallel as controls for Cgas and Sting dKO.

Cell culture chemical, antibody, and cytokine treatments

The following doses and treatment times were used for the listed compounds, antibodies, and cytokines unless otherwise stated: BD98 (10μM) for 72 hours, BRM014 (1μM) for 72 hours, ACBI1 (1μM) for 72hrs, H-151 (10μM) for 48 hours, BX-795 (1μM) for 24 hours, Ruxolitinib (250nM) for 72 hours, Tamoxifen (1μM) for 24 hours, diAZBI (20uM), anti-IFNAR or isotype control antibodies (10μg/mL) for 72 hours, anti-IFNα (30μg/mL) for 72 hours, anti-IFNβ (30μg/mL) for 72 hours, anti-IFNλ3 or isotype control antibodies (10μg/mL) for 72 hours, murine IFNγ (1ng/mL) for 24 hours, and murine IFNα (200ng/mL) for 30 minutes.

Murine tumor studies

For all of our tumor models, tumor cells we engrafted subcutaneously (5 × 10⁵ cells unless otherwise stated) into the flanks of 6–8-week-old age matched male C57BL/6J or B6-Rag1−/− mice. Tumor volume was calculated using the formula: Volume = 0.5 × Length × Width × Width using measurements collected with digital calipers. Mice inoculated with B16F10 tumors were euthanized whenever tumors reached 20mm in any direction and MC38 and CT26 inoculated mice were euthanized whenever tumors reached 1000mm3 in accordance with a Salk Institute approved IACUC protocol. Tumors were dissociated by chopping followed by incubation with Liberase (Roche, 5401020001) and DNase I (Roche, 11284932001), homogenized and filtered through a 100uM filter, then depleted of red blood cells via RBC lysis buffer (BioLegend, 420301). For B16F10 tumors, an additional step for obtaining a single cell suspension was performed via centrifugation in 47% Percoll solution.

In vivo antibodies treatments were administered via intraperitoneal injection and diluted in PBS. IFNAR blockade in the absence of immunotherapy in the B16F10 model was achieved via injection of anti-IFNAR at 250μg/mouse on days 3, 6, 9, 12, and 14 post tumor cell injection. For B16F10 ICB experiments we utilized the following antibody doses and dosing schedule: anti-PDL1 (10F.9G2, Bio X Cell) at 200μg/mouse, anti-CTLA4 at 200μg/mouse for initial injection and 100μg/mouse for subsequent injections (9D9, Leinco), and anti-IFNAR (MAR1–5A3, Leinco) at 250μg/mouse with treatment beginning on day 3 post tumor engraftment and continuing twice weekly until completion of the experiment. For MC38 and CT26 ICB experiments we used anti-PDL1 (10F.9G2, Bio X Cell) at 200μg/mouse and anti-IFNAR (MAR1–5A3, Leinco) at 250μg/mouse with treatment beginning on day 7 and day 2, respectively, post tumor engraftment and continued twice weekly until completion of the experiment. Statistical analysis was performed in GraphPad Prism

Flow cytometry Immunophenotyping

Cell surface staining of single cell suspensions from tumors or spleen was performed using flurophore-conjugated antibodies (BioLegend, BD Biosciences, eBioscience). All samples were stained with Zombie Red (BioLegend, 423110) and incubated with Fc receptor blocking antibody (Tonbo, 70–0161). Intracellular cytokine staining was performed by fixing and permeabilizing with the eBioscience Foxp3/Transcription Factor Staining kit (Thermo Fisher, 00-5523-00). UltraComp eBeads Compensation Beads (Thermo Fisher, 01-2222-42) and Precision Count Beads (BioLegend, 424902) were used for compensation and cell counting, respectively.

Ex vivo lymphocyte stimulation for cytokine staining was performed by incubating tumor-immune or splenocyte single cell suspensions with 1uM Ionomycin (Stem Cell Technologies, 73724) and PMA (Sigma, P1585) for 4 hours at 37°C in the presence of Brefeldin A (BioLegend, 420601).

RNA-Seq sample preparation and analysis

For bulk tumor RNA-seq, tumors were isolated from mice, flash frozen in liquid nitrogen, crushed with a mortar and pestle in liquid nitrogen to make powder, 25 mgs of frozen tumor were homogenized in RNA-lysis buffer (Zymo), and purified with the Zymo Research Quick-RNA miniprep kit. For in vitro RNA-seq experiments, RNA from 5 × 10⁵ cultured cells was extracted and purified with the Zymo Research Quick-RNA miniprep kit according to the manufacturer’s instructions. RNA-Seq libraries were prepared using Illumina TruSeq Stranded mRNA kit following the manufacturer’s instructions with 5 μg of input RNA.

Single-end 51-bp reads were aligned to the M. musculus mm10 genome using STAR v2.5.3a⁶¹ with default parameters. RNA expression was quantified as raw integer counts using analyzeRepeats.pl in HOMER⁶² using the following parameters: -count exons -condenseGenes. Differentially expressed genes (DEGs) were identified with getDiffExpression.pl in HOMER using edgeR⁶³ (cut-offs were set at log2 FC = 0.585 and false discovery rate (FDR) at 0.05). GSEA⁶⁴ was performed against HALLMARK gene sets. All transcriptomic heatmaps were generated using the ComplexHeatmap R package⁶⁵. GSEA and ORA for RNA-seq datasets generated by this manuscript were performed using the ClusterProfiler⁶⁶ R package.

Dual single cell RNA and T cell receptor sequencing

Tumor infiltrating CD8+ CD3+ T cells were FACS sorted from sgScramble or sgArid1a B16F10 tumors and loaded into a 10X ChIP K aiming for 20,000 CD8+ T cells from each tumor genotype. scRNA gene expression and TCR libraries were prepared with the 10x Genomics Chromium Next GEM Single Cell 5’ Reagent Kit v2 (PN-1000256) and 10x Genomics Chromium Single Cell Mouse TCR Amplification Kit (PN-1000254), respectively, with indexes from 10x Genomics Dual Index Kit TT Set A (PN-1000215).

FASTQ sequencing output files were demultiplexed and aligned to the mm10 genome using Cell Ranger (version 6.0.1). Downstream data processing was performed in R (version 3.6.1) with the Seurat package⁶⁷ (version 4.0.2). The data was subset to remove cells with mitochondrial gene expression greater than 10% before normalization and scaling. Clustering was performed using the Seurat FindClusters command with a resolution of 0.20 to prevent over-clustering. A small CD8+ dendritic cell cluster and a cluster of apoptotic cells expressing high levels of mitochondrial genes were excluded from downstream analyses. PCA was performed on the normalized scaled data and 13 dimensions were chosen as input to generate a two-dimensional visualization using RunUMAP (Seurat) which were visualized using DimPlot (Seurat) and the color-blind friendly palette from the ggpubfigs R package⁶⁸. Cell density was plotted using contourPlot from the scDataviz⁶⁹ R package (version 1.3.1). Significantly differentially expressed cluster marker genes were delineated with the FindMarkers (Seurat) command and visualized with Nebulosa R package⁷⁰ (version 1.0.2), FeaturePlot (Seurat), and DoHeatmap (Seurat). Gene signatures of interest were visualized using the AddModuleScore and FeauturePlot Seurat commands. The anti-PD1 response signature gene list was previously described. The CD8+ T cell ISG score was constructed from Leading Edge genes from GSEA IFN Alpha or Gamma Hallmark gene sets that were enriched in clusters 1–5 (Table S1). Differential expression analyses was performed between sgArid1a and sgScramble tumor genotypes in every cluster with Seurat::FindMarkers() (v4.2), requiring the genes to be detected in at least 5% of cells. To prepare the input ranked gene list for GSEA, the resulted genes in each cluster were ordered by scores that calculated as −log10(FDR)*sign[log2(FC)]. GSEA was performed with WebGestalt⁷¹ using HALLMARK gene sets with FDR <0.05 as the significance threshold, protein coding genes as the reference list, a minimum number of genes in a category of 5, permutation time of 1000 and visualizing as a barplot with normalized enrichment scores.

TCR analysis was performed using the scRepertoire R package⁷² (version 1.5.3). TCR clonotype expansion was visualized using clonalHomeostasis (scRepertoire). We compared the presence of individual clonotypes across treatments using scatterClonotype (scRepertoire). TCR was combined with gene expression to investigate cluster-based clonotype expansion using combineExpression (scRepertoire) and DimPlot (Seurat).

Total RNA-Seq ERV expression analysis

Total RNA-seq reads were aligned to the mouse mm10 reference genome using STAR with the flags – winAnchorMultimapNmax 200 –outFilterMultimapNmax 100. Tag directories were then generated, raw counts of repeat elements were counted, and repeat element differential expression analysis was performed using DEseq2⁷³. Differentially expressed ERVs were plotted in ggplot2 after filtering out repeats with a log2FC of zero or an adjusted p values of greater than .95 to save computing time.

ATAC-Seq and analysis

The Omni ATAC-seq protocol was employed as previously described⁷⁴. Briefly, 50,000 cells were washed with cold PBS, collected by centrifugation then lysed in ATAC resuspension buffer (10 mM Tris-HCl, pH 7.4, 10 mM NaCl, 3 mM MgCl) containing 0.1% NP40, 0.1% Tween-20, and 0.01% Digitonin. Nuclei were then pelleted and incubated in transposition mix containing Tn5 transposase (Illumina). Purified DNA was then ligated with adapters, amplified and size selected for sequencing. Library DNA was sequenced with paired end 42 bp reads.

Paired-end 42-bp or paired-end 75-bp reads were aligned to the M. musculus mm10 genome using STAR⁶¹ with default parameters. ATAC-seq peaks were called using the HOMER findPeaks program using parameters for DNAse-seq (-style dnase). Peaks were called when enriched >4.0-fold over local tag counts. Differentially accessible regions were identified using edgeR⁶³ by calling getDifferentialPeaksReplicates.pl in HOMER with fold change ≥2.0 or ≤−2.0, FDR < 0.05. Peaks sets were annotated with HOMER.

ChIP-Seq and analysis

Cells were harvested and crosslinked in 1% formaldehyde for 10 min before quenching with glycine for 5 min on ice. Cells were pelleted by centrifugation and snap frozen in dry ice before storage at −80°C. Pellets were thawed on ice and resuspended in rinse buffer 1 (50 mM HEPES pH 8.0, 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% NP40, 0.25% Triton X100), collected by centrifugation then resuspended in rinse buffer 2 (10 mM Tris pH 8.0, 1 mM EDTA, 0.5 mM EGTA, 200 mM NaCl). Cells were washed and resuspended in shearing buffer before sonication using Covaris E220 (0.1% SDS, 1 mM EDTA, pH 8, 10 mM Tris HCl, pH 8). For STAT1, STAT2, and IRF9 ChIPs, cells were sheared for 8 mins at 140W with 5% duty factor. DNA was then made up to 1x IP buffer (50 mM HEPES/KOH pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X100, 0.1% DOC, 0.1% SDS, supplemented with protease inhibitors) and 3 mg antibody added for overnight incubation with rolling at 4°C. Antibody bound DNA was recovered using a 1:1 mixture of Protein A and Protein G beads, washed and treated with Proteinase K and RNAseA. Purified ChIP DNA was then used for library generation for ChIP-seq.

Paired-end 75-bp reads were aligned to the M. musculus mm10 genome using STAR (Dobin et al., 2013) with default parameters. ChIP-seq peaks were called using findPeaks within HOMER using - style factor and an -i input control. Differential ChIP peaks were called using getDiffExpression.pl with fold change ≥ 1.5 or ≤ 1.5, Poisson P value < 0.0001. ChIP-seq peaks were annotated by mapping to the nearest transcription start site (TSS) using the annotatePeaks.pl program. Histograms were generated using Homer annotatePeaks.pl with parameters -size 2000 and -hist 10. Heatmap matrices showing overlap of binding sites were generated using Homer annotatePeaks.pl with parameters -size 6000 and -hist 25 and -ghist. Heatmap visualization were generated in TreeView (https://bitbucket.org/TreeView3Dev/treeview3/wiki/Downloads) by inputting heatmap matrix file after removing the header row of the heatmap matrix file.

TCGA transcriptomic analyses

For each TCGA cohort, primary tumor samples were grouped by whether an ARID1A mutant was present or not using the TCGAbiolinks and maftools R packages. Raw STAR-counts were downloaded for each cohort using the TCGAbiolinks R package. To deconvolute and compare the immune cell composition in two groups of every cohort, we applied method CIBERSORT to estimate cell fraction based on their bulk gene expression (FPKM values) profiles, which is implemented in R immunedeconv::deconvolute() (v2.1)⁷⁵. P values were calculated using Wilcoxon rank-sum test. Cytolytic score was calculated as the geometric mean of the cytotoxic genes PRF1 and GZMA FPKM values as previously described¹⁶ in ARID1A mutant or non-mutant cancers within each of the TCGA cohorts analyzed. P values were calculated using the Wilcoxon rank-sum test and cyt score values were transformed using log1p function before plotting. ARID1A-IFN signature score was calculated as the mean FPKM values of the human orthologs of the ARID1A-IFN signature gene set. P values were calculated using the Wilcoxon rank-sum test.

For DESeq2, raw STAR counts were used as input for differential expression analysis using the R package, DESeq2 (v1.24)⁷³ between ARID1A mutant and non-mutant samples within each TCGA cohort analyzed. The ranked gene list with scores was generated for GSEA from DESeq2 generated FDR and FC, of which scores were calculated as −log10(FDR)*sign[log2(FC)]. GSEA was performed with WebGestalt R package⁷¹ using HALLMARK human gene sets and the human orthologs of the ARID1A-IFN signature gene set, protein coding genes as the reference list, a minimum number of genes in a category of 5, permutation time of 1000. Statistical analysis was performed in R.

ARID1A-IFN Signature association with AC-ICAM cohort survival

mRNA normalized counts and mutation data for AC-ICAM were downloaded from cBioPortal. To assess the association with ARID1A-IFN signature score and patient survival, patients were stratified into high (top 20%) and low (bottom 20%) groups for the ARID1A-IFN signature score followed by log-rank survival test of these patients. We also assessed the association of ARID1A-IFN signature score on survival at 100 months by stratifying patients as non-mutant-dead, non-mutant-alive, ARID1A mutant- dead, and ARID1A mutant-alive followed by plotting each patient’s mean ARID1A-IFN signature score across the four groups and performing two-way ANOVA test. Statistical analysis was performed in R.

ARID1A-IFN Signature association with ICB response

For the ARID1A-IFN signature score association with ICB response in stomach cancer analysis, we first downloaded expression (FPKM) and clinical data from the Tumor Immunotherapy Gene Expression Resource (http://tiger.canceromics.org/#/download) for the STAD-PRJEB25780 dataset. The mean FPKM value for all ARID1A-IFN signature genes was then calculated in each patient. We then assessed association with ICB response by performing one-way ANOVA test between all ICB response groups based on their ARID1A-IFN signature scores. Statistical analysis was performed in R.

Cancer Cell Line Encylopedia transcriptomic analysis

The RNAseq TPM gene expression data of cancer cell lines were downloaded from CCLE (Cancer Cell Line Encyclopedia⁷⁶; https://sites.broadinstitute.org/ccle/datasets). Cancer cell lines were sorted according to the expression of ARID1A, and the top and bottom quartiles were compared for expression of human orthologs of the ARID1A-IFN signature genes, Type I IFN genes, and the IFN Alpha and Gamma Response Hallmark gene sets. P values were calculated using Wilcoxon rank-sum test.

DepMap genetic dependency and proteomics analysis

ARID1A mutant cancer cell lines in the DepMap online portal were identified by downloading ARID1A the Mutation table in the Characterization tab of the ARID1A DepMap gene page (n=306 mutations) which represents 222 ARID1A mutant cell lines. Differences in genetic dependency between ARID1A mutant vs non-mutant cancer cell lines were then assessed by performing a two-class comparison between ARID1A mutant cancer cell lines vs all other cell lines in the DepMap RNAi screen using the DepMap custom analysis tool (https://depmap.org/portal/interactive/custom_analysis). ARID1A mutant dependencies were classified by having effect sizes of less than or equal to −0.1 and Q-values of less than or equal to 0.05. To assess if R-loop regulator genes were enriched among ARID1A mutant genetic dependencies, we annotated genes in the ARID1A mutant dependencies group and background list (all other genes in two-class comparison results) as R-loop regulator gene or not using the R-loopBase database (https://rloopbase.nju.edu.cn/) plus SWI/SNF genes as a reference list for R-loop regulators and then performed Fisher’s Exact Test.

Differentially abundant proteins between ARID1A mutant vs non-mutant cancer cell lines were also assessed by performing a two-class comparison between ARID1A mutant cancer cell lines vs all other cell lines in the CCLE Proteomics dataset using the DepMap custom analysis tool. Differentially abundant proteins were classified by having effect sizes of less than or equal to −0.1 and Q-values of less than or equal to 0.05.

P14 CD8+ T-cell isolation and co-culture assay with GP33 pulsed cancer cells

Splenocytes from P14+ mice were stained in 500μL PBS+2%FBS+1mM EDTA with 1% normal rat serum (STEMCELL technologies) with 2 μg/mL of the following antibodies); biotin anti-B220 (clone RA3–6B2), biotin anti-CD11c (clone N418), biotin anti-CD11b (clone M1/70), biotin anti-CD4 (clone GK1.5 or RM4–5), biotin anti-CD49b (clone DX5), biotin anti-TCRγδ (clone GL3), and 5μg/mL biotin anti-Ter119 (clone TER-119) (all from Biolegend). Negative selection for CD8+ cells was performed using Biolegend MojoSort Streptavidin Nanobeads (#480016) using 25μL per sample, followed by RBC-lysis (Biolegend). At a concentration of up to 1.5 million cells per well, negatively-selected CD8+ P14+ cells were plated onto 24 well TC-treated plates pre-coated with goat anti-hamster IgG (AffiniPure Goat Anti-Armenian Hamster IgG (H+L) (Jackson ImmunoResearch cat#127-005-099) at 30μg/mL in PBS in 500mL per well of RPMI+L-glutamine+10%FBS+50μM B-Me supplemented with 2 μg/mL purified hamster anti-mouse CD3 (NA/LE [no azide; low endotoxin) (BD cat#553057), 0.5 μg/mL of purified hamster anti-mouse CD28 (NA/LE) (BD cat# 553294) and 10 ng/mL recombinant human IL-2 [rhIL-2] (Peprotech 200–02) for 18hrs.

B16F10 cells were treated with vehicle control or 1ng/mL of IFNγ for 24hrs, trypsinized and washed, then incubated in water bath for hr with 1μg/mL GP33 peptide (Genscript cat#RP20257), and agitated every 15mins. After 1hr, peptide was washed out via 3 PBS washes then cells were added (either 2K or 10K) to wells of a flat-bottom TC-treated 96-well plate containing 20K of pre-activated P14 CD8+ cells in (RPMI+L-glutamine+10%FBS+50μM B-Me). Where applicable, 10ng/mL of recombinant human IL-2 [rhIL-2] (Peprotech 200–02) was included. After 72hrs of co-culture, non-adherent cells were removed from the 96-well plate by pipetting and combined with adherent cells following trypsinization. After staining for cell surface markers for 15 minutes (CD45.1 - APC 1/200, CD8-BUV395 1/200 in 25μL volume per sample), an Annexin-V FITC kit (Biolegend) was used according to manufacturer’s instructions with modification of using 1μL of Annexin and 1μL of 7-AAD per sample. Staining was performed in 25μL of binding buffer per sample and an additional 100μL was added prior to acquisition of flow cytometry data. Statistical analysis was performed in GraphPad Prism.

Lentivirus preparation and transduction

Construction of TREX1 (Addgene #27218) and RNASEH2B (Addgene #108697) lentiviral expression vectors were achieved by cloning the into the N103 3x-FLAG lentiviral expression vector⁷⁷ and with a puromycin selection marker. HEK293T cells were transfected with the lentiviral constructs and packaging plasmids Md2G and psPAX2 using polyethylenimine-mediated transfection. Seventy-two hours post transfection, the media containing the virus was collected, filtered, and centrifuged in a Amicon Ultra-15 100K (Millipore Sigma) device at 4,000 ×g for 40 minutes at 4°C. Concentrated virus was collected from the Amicon device chamber and stored at −80°C until use. Transduction with concentrated virus was performed for 24 hours with 5μg mL-1 Polybrene. Forty-eight hours later, lentiviral construct expressing cells were selected with 2μg mL-1 of puromycin. Selection was maintained throughout the course of overexpression experiments.

IFNβ ELISA

Supernatants from cultured cell lines were collected, spun down at 500G for 3 mins to pellet any dead cells, and then concentrated 10X using Amicon Ultra 0.5 (Millipore Sigma) protein concentrators followed by flash freezing. Concentrated supernatants were then used as input for the Mouse IFN beta ProQuantum Immunoassay Kit (Thermo Scientific) following the kit protocol. Statistical analysis was performed in GraphPad Prism.

Immunoblotting

Protein samples were run on 4%–12% Bis-Tris gels (Life Technologies). After primary antibody incubation, which is typically done overnight at 4°C, blots were probed with 1:20,000 dilution of fluorescently labeled secondary antibodies in 2% BSA in TBST (1X Tris-buffered saline with 0.1% Tween-20) for an hour at room temperature (RT). Fluorescent images were developed using an Odyssey scanner and analyzed using Image Studio 2 software. Western blot panels containing ISG15 (except for Figures S3E and S7C) are composites as ISG15 blots were run on a separate gel/immunoblot from the same lysates. Original blot’s 15kd section, where ISG15 runs were discarded thus requiring us a second blot. Western blots in for phosphorylated and total STAT proteins in Figure S4F–G are composites. Since the phospho-STAT1/2 and total STAT1/2 antibodies are all from rabbit host species, we ran a blot for total STAT proteins and another for phosphorylated STAT proteins from the same lysates.

Quantitative PCR

Total RNA was extracted with Quick-RNA miniprep/microprep kit (Zymo), and 500ng total RNA per sample was reverse transcribed SuperScript IV VILO Master Mix (Thermo). For qPCR, cDNA samples were diluted 1:100 for a 10ul reaction system with 500nM of the indicated forward and reverse primers (Table S2) with the iTaq Universal SYBR Green Mastermix (Biorad). For each biological sample, three technical replicates were performed and normalized against the GAPDH threshhold cycle (Ct) value to calculate ΔCt. The ΔCt of each sample was then compared to the ΔCt of the control sample to generate the ΔΔCt value. Relative expression was then analysed using the 2−ΔΔCt method and the relative fold change was plotted with the control samples given a value of 1.0. Statistical analysis was performed in GraphPad Prism.

Immunofluorescence

Cells were cultured on glass coverslips, fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and blocked in 1X TBS containing 5% Normal Goat Serum. Coverslips were incubated with primary antibody overnight at 4°C followed by incubation with fluorescently conjugated secondary antibody for 1 hour at room temperature, counterstained with DAPI, then mounted on glass slides using ProLong Glass mounting media (Thermo Fisher). To verify specificity of ssDNA staining, live cells were transfected with S1 nuclease using ProJect protein transfection reagent for 4 hours at 37 degrees Celsius prior to staining.

dRNH1-GFP protein purification was performed as previously described. For dRNH1-GFP staining, cells were fixed and permeabilized in ice cold Methanol for 10 minutes on ice, blocked, and stained for dRNH1-GFP as previously described. To verify specificity of dRNH1-GFP, fixed coverslips were pre-treated with or without RNaseH from E. Coli for 4 hours at 37 degrees Celsius prior to dRNH1-GFP staining. All slides were imaged using a Zeiss LSM 880 confocal microscope with 63x objective and median fluorescence intensity was analyzed using CellProfiler⁷⁸. Statistical analysis was performed in GraphPad Prism.

Immunohistochemistry

Immunohistochemistry (IHC) was performed on FFPE endometrial cancer slides (n=7) obtained from the UCSD Moore’s Cancer Center in accordance with IRB-approved study (#181313). Patient MMR status was obtained from clinical testing for MMR protein expression via IHC. ARID1A mutation status in one patient was determined via Foundation Medicine next generation sequencing testing. Endometrial gland areas identified by a pathologist as cancer were analyzed for ARID1A expression and CD8+ T cell infiltration. Briefly, slides were dewaxed in xylene, rehydrated in ethanol, followed by antigen retrieval in Akoya Biosciences AR6 buffer, and blocking in Akoya Biosciences blocking buffer. Vectastain Elite ABC-HRP kit was used to detect ARID1A (PSG-3, Santa Cruz, 1:1000) expression in FFPE endometrial cancer slides. The Akoya Biosciences Opal 4 color IHC kit was used for multiplexed fluorescent IHC. CD8αantibody incubations were performed overnight at 1:200 and 1:400 dilutions, respectively. Analysis of CD8+ T cell infiltration as a percentage of cells within cancerous endometrial gland areas was performed using QuPath and statistical analysis performed in R⁷⁹.

CellTiter-Glo assay

Cells were plated as single cells at 200 cells per well in 96-well plates. Each condition tested was plated in at least triplicate. Cell Titer Glo Assay (Promega) was used to measure cell number via ATP luminescence. A reference measurement was taken on day 1 and subsequent measurements were taken every day after until experiment completion using a Tecan Infinite M1000 Pro plate reader to establish the proliferation rate of cells of interest.

QUANTIFICATION AND STATISTICAL ANALYSIS

Software for experimental quantifications are named in the respective method details sections. Statistical analyses were performed using the two-tailed, unpaired, Student’s t-test unless otherwise specified in figure legends. The P values were represented as follows: ns, not significant, *p<0.05, **p<0.01, ***p<0.001, **** p<.00001.

Supplementary Material

1

Figure S1. Additional characterization of syngeneic cancer cell lines and analysis of immune infiltration in human cancers, related to Figure 1.

(A) Western blots of ARID1A protein level in sgArid1a or sgScramble B16F10 and MC38 cell lines. (B) Kaplan-Meier survival curve of mice flank injected with sgScramble or sgArid1a B16F10 tumors. (C-D) Cell proliferation rates of sgArid1a or sgScramble B16F10 and MC38 cells in vitro. Data are represented as mean ± SEM. (E) Cibersort quantification of proportion of T cells and NK cells as a percent of immune cell estimate in ARID1A mutant or non-mutant cancers in indicated TCGA cohorts. Stars in ARID1A mutant or non-mutant stacked bar chart groups indicate significantly more of the subset in tumors in the indicated cancer genotype. (F) Human endometrial cancer immunohistochemistry samples with representative images of ARID1A IHC and multiplexed fluorescence IHC for CD8α (yellow) and DAPI (blue) in patient samples with quantification of percent of cells that are CD8+ in ARID1A positive or ARID1A loss/mutant samples. MMR status assessed via IHC (scale bar = 100μM). (G-H) Flow cytometry histogram of Ki67 staining gated on CD4+ T cells and NK cells from sgArid1a (red) or sgScramble (grey) tumors with quantification of percentage of positive cells. All data are represented as mean ± SD unless otherwise noted. Statistical analysis in Fig S1B was performed using the log-rank survival test while Fig S1E–F statistical analyses were performed via Wilcoxon Rank Sum Test.

2

Figure S2. Chemokine expression in human cancers and CD8+ T cell scRNA-seq cluster marker genes, related to Figure 2.

(A) mRNA expression data for CXCL9, CXCL10, CXCL13 in ARID1A mutant and non-mutant cancers in the indicated TCGA cohorts. (B) Heatmap of marker genes defining clustering for single cell RNA sequencing of B16F10 tumor infiltrating CD8+ T cells. (C) Nebulosa density plots of representative marker genes for CD8+ T cell clusters C1–C5. Statistical analysis in Fig S2A was performed via Wilcoxon Rank Sum Test.

3

Figure S3. Additional characterization of IFN response in ARID1A deficient or SWI/SNF inhibitor treated cells and ISG expression of ARID1A low expressing human cancer cell lines, related to Figure 3.

(A) GSEA curves of IFN Alpha and Gamma Response Pathways in the indicated genotypes or treatments. (B) Western blot for ARID1A and ISGs in sgScramble and sgArid1a CT26 cell lines. (C) Normalized ISG mRNA levels in sgScramble and sgArid1a CT26 cells readout via qPCR. (D) Normalized ISG mRNA levels in vehicle or ACBI1 treated B16F10 cells readout via qPCR. (E) Western blot of SMARCA4 and ISGs in B16F10 cells treated with ACBI1 or vehicle control. (F) Western blot of SMARCA4 and STAT1 in vehicle or BRM014 treated samples at the following doses: 30nM, 100nM, 300nM, 1uM, and 3uM. (G) Cancer Cell Line Encyclopedia expression data for IFN genes and ISGs comparing cell lines in the bottom and top quartile for ARID1A expression. (H) Percent of human orthologs of ARID1A-IFN signature that are significantly upregulated, downregulated, or not significantly changed in the bottom quartile of ARID1A expressing cell lines in the CCLE compared to the top quartile of ARID1A expressing cell lines. All data are represented as mean ± SD unless otherwise noted. Statistical analysis in Fig S3C and S3E were performed using two-way ANOVA and Fig S3F via the Wilcoxon Rank Sum Test.

4

Figure S4. Blockade of Type III IFN, JAK-STAT signaling and additional characterization of IFN phenotype, related to Figure 4.

(A) Western blot of ISGs RIG-I and STAT1 in sgScramble or sgArid1a B16F10 cells with or without anti-Type III IFN antibody treatment. (B) Heatmap of ARID1A-IFN signature ISGs induced in sgArid1a B16F10 cells with or without treatment of JAK inhibitor (ruxolitinib). (C) Western blot of ISGs RIG-I and STAT1 in sgScramble or sgArid1a B16F10 cells with or without JAKi treatment. (D) Median fluorescence intensity (MFI) quantification and representative histograms for ISRE-GFP signal in vehicle, BD98 treated, or BD98 treated + JAKi (ruxolitinib) in ISRE-GFP reporter MEFs. (E) Normalized Ifnb1 mRNA levels in sgScramble and sgArid1a B16F10 and MC38 cells readout via qPCR. (F-G) Normalized Ifnb1 mRNA levels in WT and Arid1a−/− or vehicle and BD98 treated MEFs, respectively. (H-I) Western blots of total and phosphorylated STAT proteins downstream of IFNAR in sgScramble and sgArid1a B16F10 and MC38 cells. (J-K) Quantification of MHC Class I and II via flow cytometry in sgScramble or sgArid1a MC38 cells with or without IFNγ stimulation. All data are represented as mean ± SD unless otherwise noted. Statistical analysis in Fig S4D was performed via one way ANOVA.

5

Figure S5. ISGF3 ChIP-seq and ATAC-seq data in wildtype and ARID1A deficient cells, related to Figure 4.

(A) Heatmap of merged of Arid1a−/− and WT MEFs ChIP binding sites for STAT1, STAT2, and IRF9. (B) Histograms of STAT1, STAT2, and IRF9 binding site normalized ChIP coverage genome-wide or at the ISRE motif in Arid1a−/− or WT MEFs. (C) Increased binding by STAT1/STAT2/IRF9 or STAT2/IRF9 at ARID1A-IFN signature genes in Arid1a−/− MEFs. (D) Representative gene tracks of ISGs showing increased STAT1, STAT2, and IRF9 binding in Arid1a−/− MEFs compared to WT MEFs. (E) Percent of the ARID1A-IFN signature genes which show no significant change (black), significant increase (red), or significant decrease (blue) in ATAC-seq peaks annotated to the nearest gene. (F-G) Gene tracks of ATAC-Seq coverage of ISG loci and Ifnb1 loci in sgArid1a or sgScramble B16F10 cells.

6

Figure S6. Additional characterization of PAMPs, R-loops, and ISG expression with or without overexpression of RNASEH2B or TREX1, related to Figure 5.

(A) Immunofluorescence of double stranded RNA (dsRNA) and DAPI counterstain with quantification of dsRNA intensity in the indicated groups (scale bar = 20μM). (B) Volcano plot of differentially expressed ERVs in sgArid1a compared to sgScramble B16F10 cells. (C) Immunofluorescence of dRNH1-GFP and DAPI with quantification of nuclear RNA:DNA hybrids in sgScramble and sgArid1a B16F10s with or without treatment of fixed and permeabilized cells with recombinant RNaseH (scale bar = 5μM). (D) Immunofluorescence of ssDNA and DAPI with quantification of cytosolic ssDNA in B16F10 cells treated with vehicle or ACBI1 for 72hrs (scale bar = 10μM). (E) Immunofluorescence of ssDNA and DAPI counterstain with quantification of ssDNA in sgScramble and sgArid1a B16F10s with or without protein transfection of S1 nuclease (scale bar = 10μM). (F) BD98 time-course experiment following cytosolic ssDNA levels with representative images and quantification (scale bar = 10μM). (G) BD98 time-course experiment following STAT1 protein levels via western blot with quantification. (H) Normalized ISG mRNA levels in vehicle or BD98 treated MEFs with or without overexpression of TREX1 or RNASEH2B readout via qPCR. (I) Normalized ISG mRNA levels in sgScramble or sgArid1a MC38s with or without overexpression of TREX1 or RNASEH2B readout via qPCR. All data are represented as mean ± SD unless otherwise noted. Statistical analysis in Fig S6A and S6G were performed via one way ANOVA and statistical analysis in S6C, S6J–K were performed via two-way ANOVA.

7

Figure S7. Additional characterization of STING pathway and silencing of STING pathway components in human ARID1A mutant cancer cell lines, related to Figure 6.

(A) Western blot assessing STAT1 expression following STING (H-151) or TBK1 inhibition (BX-795) in sgArid1a B16F10. (B) Western blot assessing ISG expression following STING (H-151) or TBK1 inhibition (BX-795) in sgArid1a MC38. (C-D) Western blot and qPCR assessing ISG expression following BD98 treatment in wildtype, Cgas−/−, or Sting−/− MEFs. (E) IFNβ ELISA from supernatants of the following groups in MC38 cells: sgScramble, sgArid1a, sgArid1a + STINGi (H-151), or sgArid1a + TBK1i (BX-795). (F) IFNβ ELISA from supernatants of vehicle or BD98 treated WT, Cgas−/−, or Sting−/− MEFs. (G) Immunofluorescence of phospho-TBK1 and DAPI in sgArid1a or sgScramble B16F10s with quantification. (H) Western blot of total and phospho-TBK1 in unstimulated or STING agonist (diAZBI, 20μM) treated cells in a timecourse experiment with quantification of the levels of pTBK1 in both genotypes normalized to sgScramble unstimulated (scale bar = 10μM). (I) Median fluorescence intensities of MHC Class I and II following IFNγ (1ng/mL 24hrs) treatment with or without STING inhibition (H-151) in sgArid1a B16F10s. (J) Cell proliferation rates of sgScramble, sgArid1a, and Sting dKO B16F10 cell lines assessed via Cell Titer Glo. Data are represented as mean ± SEM. (K) Volcano plot of differentially abundant proteins in ARID1A mutant vs non-mutant cancer cell lines in CCLE proteomics dataset. (L) Volcano plot of genetic dependencies in ARID1A mutant cancer cell lines (red) from DepMap-RNAi screening data with select R-loop regulator genes labeled in ARID1A mutant dependencies. (M) Over-representation of R-loop regulator genes among ARID1A mutant cancer cell line genetic dependencies vs non-dependency genes. All data are represented as mean ± SD unless otherwise noted. Statistical analysis in Fig S7D was performed via two-way ANOVA, Fig S7E–F and S7K analysis via one-way ANOVA, and S7M via Fisher’s Exact Test. An irrelevant lane in Figures S7A and S7B was digitally removed and is indicated by a black line.

Highlights.

• ARID1A loss is sufficient to induce anti-tumor immunity and immunotherapy response

• ARID1A loss or cBAF inhibition induces interferon (IFN) stimulated gene expression

• ARID1A-IFN signature is driven by R-loop derived cytosolic DNA and the STING pathway

• ARID1A mutant anti-tumor immunity and immunotherapy response is Type I IFN dependent