CRISPR-GEMM Pooled Mutagenic Screening Identifies KMT2D as a Major Modulator of Immune Checkpoint Blockade

Checkpoint immunotherapy has achieved substantial success, showing clinical benefits across multiple tumor types with durable responses even in chemoresistant and metastatic cancers (1–4). However, the majority of patients do not respond to checkpoint immunotherapy (5, 6), indicating the importance of precision immunotherapy, where patients are stratified on the basis of functional and clinical evidence, subsequently receiving the treatments or combinations most likely to benefit them.

A multitude of approaches have been applied to understand the features associated with immunotherapy response (7, 8). These include whole-genome sequencing (7, 9, 10), proteomic analysis (11), single-cell transcriptomic analysis (12), in vitro cancer–immune cell cocultures (13, 14), and ex vivo/in vivo screens using cell lines in tumor transplant models (15). Several factors, including PD-L1 expression, tumor mutation burden (16), neoantigen burden (17), and immune infiltration status (18, 19), as well as certain oncogenic pathways (20), have been demonstrated to be correlated with immunotherapy response. In addition, many mechanisms have been described in primary or acquired resistance to immunotherapy (21, 22). For instance, tumors can foster the development of an immunosuppressive tumor microenvironment (23), or acquire new mutations that reduce immune recognition and apoptosis (24).

Despite these advances, our understanding of the genetic factors that dictate response to checkpoint immunotherapy remains incomplete. Analysis of patient cohorts can reveal associations with immune checkpoint blockade (ICB) response, but such studies cannot firmly establish causality. Current genetic screening approaches using in vitro or ex vivo cultured cell lines are confined by the mutation background and may miss subtle factors that influence ICB response in the complex immunologic setting of the tumor microenvironment. Genetically engineered mouse models (GEMM; ref. 25) can more precisely mimic the features of human cancers, because such tumors develop from cells within the native organs of fully immunocompetent animals, thereby preserving the immune microenvironment. Because of these features, GEMMs offer certain distinct advantages for the studies of tumor immunology. Although traditional GEMMs can target only a handful of genes at a time, CRISPR enables pooled targeting of multiple genes through somatic genome editing. We have previously developed CRISPR-GEMMs that enabled large-scale direct in vivo screening of functional tumor suppressors (26, 27). Using CRISPR-GEMMs, genetically complex tumors can be readily generated in individual mice that each reflect the genetic and cellular heterogeneity of human tumors, with the flexibility to target any desired sets of genes.

Here, we performed a CRISPR-GEMM screen of significantly mutated genes (SMG) in human cancers (28, 29), examining the effect of these mutations on ICB response. We specifically pinpoint Kmt2d deficiency as a major mediator of sensitivity to ICB therapy in diverse cancer types, suggesting its potential as a biomarker for patient stratification.

To perform a screen for genetic modulators of immunotherapy response in conditions that closely mimic human cancers, we developed a CRISPR-GEMM model of liver cancer in which AAV-CRISPR–mediated pooled mutagenesis drives autochthonous liver tumorigenesis in fully immunocompetent mice. We designed an AAV-CRISPR vector that expresses Cre recombinase under a liver-specific thyroxine-binding globulin (TBG) promoter, together with two single guide RNA (sgRNA) expression cassettes: one for a Trp53-targeting sgRNA, and the other as a backbone sgRNA designed for cloning and expression of additional specific sgRNA(s). We utilized the mouse tumor suppressor gene (mTSG) library, which targets the top 49 most frequently mutated tumor suppressor genes in the pan-cancer datasets of The Cancer Genome Atlas (TCGA) with 7 housekeeping genes as internal controls (26, 27). We cloned the mTSG library into the AAV-TBG-CRISPR vector, and generated AAVs carrying the pooled sgRNA library (Fig. 1A). To monitor liver tumorigenesis in vivo, we crossed LSL-Cas9–2A-EGFP (LSL-Cas9) mice with LSL-firefly luciferase (LSL-Fluc) mice to generate LSL-Cas9; LSL-Fluc mice. We then introduced the base vector AAVs (AAV-Vector), sgTrp53-delivering AAVs (AAV-sgTrp53) or mTSG library AAVs (AAV-mTSG) into the mice by intravenous injection.

We monitored the bioluminescence signals in the injected mice using the intravital imaging system (IVIS). We observed a steady increase of luciferase signal from day 40 to day 60, indicative of ongoing tumorigenesis (Supplementary Fig. S1A). On the basis of the IVIS data, we assigned the AAV-mTSG injected mice into 3 size-matched cohorts to receive PBS, anti–PD-1, or anti-CTLA4 treatment (Fig. 1A). All of the AAV-mTSG–injected mice treated with PBS died within 100 days (Fig. 1B), having developed large liver tumors with 100% penetrance. In sharp contrast, no mice died from tumors in the AAV-Vector– or AAV-sgTrp53–injected groups. Although the IVIS data suggested no significant change between ICB therapy and PBS treatment groups, anti–PD-1 (n = 11) or anti-CTLA4 (n = 11) treatment prolonged overall survival in comparison with mice receiving PBS treatment (n = 15; Fig. 1B; Supplementary Fig. S1B). We then harvested all liver lobes for genomic sequencing and histologic characterization. Endpoint histologic sections from vector-treated mice (n = 3) revealed no tumor lesions, whereas all mTSG-treated mice (n = 37) developed large and heterogenous liver tumors (Fig. 1C).

We processed the tumors for targeted analysis of the predicted sgRNA cut sites using molecular inversion probe (MIP) sequencing (27). Representative variants of insertions and deletions (indels) detected by MIP capture sequencing are shown at the cut sites of B2m sg3 (Fig. 1D), Arid1a sg4 (Fig. 1E), and Kmt2d sg3 (Fig. 1F). We summed the constituent variant frequencies of each sgRNA and charted the mutation landscape associated with each treatment (Fig. 1G). We then calculated the mean variant frequencies for each gene, grouping samples by treatment condition. By comparing the gene mutation frequencies in the different treatment groups, we uncovered genetic perturbations that were comparatively enriched or depleted with anti–PD-1 or anti-CTLA4 treatment (Fig. 1G and A). Comparing the anti-CTLA4–treated mice with PBS-treated mice, the mutation frequencies of B2m, Grlf1, Bcor, and Kdm5c were significantly increased, whereas the mutation frequencies of Arid1a were significantly decreased (Fig. 2B). Comparing anti–PD-1-treated mice with PBS-treated mice, knockout of B2m, Grif1, Vhl, Cdkn1b, or Bcor was correlated with anti–PD-1 resistance, whereas knockout of Kmt2d, Arid1a, Rnf43, or Atrx was associated with anti–PD-1 responsiveness (Fig. 2C).

From our screen data, Arid1a loss sensitized tumors to both anti-CTLA4 and anti–PD-1 treatment, whereas Kmt2d mutations showed the largest magnitude of sensitization to anti–PD-1. To individually validate these findings, we developed and utilized several genetic liver cancer models. We first crossed CAG-LSL-Myc transgenic mice with LSL-Cas9 mice, and then injected these mice with AAVs carrying TBG-Cre and sgTrp53 to drive ectopic Myc expression and Trp53 knockout in the liver (Fig. 2D). To interrogate the effects of concurrent Kmt2d or Arid1a mutations in the setting of Myc overexpression and Trp53 loss, we injected AAVs carrying sgTrp53+sgKmt2d or sgTrp53+sgArid1a into LSL-Myc;LSL-Cas9 mice. As a control, we injected sgTrp53+sgNTC AAVs. Mice bearing Myc+sgTrp53 tumors had a median survival of 155 days, with 50% (3/6) of mice dying within 4 months, and anti–PD-1 treatment did not show a significant survival benefit (median survival of 199 days; Fig. 2E). Additional Arid1a mutations on top of Myc+sgTrp53 resulted in more aggressive tumors, with 100% (9/9) of mice dying within 4 months (median survival of 83 days). Treatment with anti–PD-1 marginally prolonged survival (median survival 105 days), with 12.5% (1/8) of mice alive 400 days post-injection (Fig. 2F). Similar results were found using a liver cancer model with Trp53 and Apc knockout as the genetic background, with Arid1a-mutant tumors showing a significant response to anti–PD-1 (Supplementary Fig. S1C–S1E)

For mice bearing Myc+sgTrp53+sgKmt2d tumors, 8 of 9 (88.9%) mice died within 4 months (median survival of 73 days), indicating that KMT2D functions as a tumor suppressor in this context. Strikingly, anti–PD-1 therapy prolonged survival of mice bearing Myc+sgTrp53+sgKmt2d tumors, with >50% mice alive at 400 dpi (Fig. 2G). We collected endpoint liver tumors for histologic and IHC characterization of these tumors (Fig. 2H–K), observing that Arid1a- or Kmt2d-mutant liver tumors were more infiltrated with CD45⁺ immune cells (Fig. 2I) and CD3⁺ T cells (Fig. 2J), particularly after anti–PD-1 therapy. Upon anti–PD-1 treatment, F4/80⁺ macrophages were more abundant in Arid1a-mutant liver tumors, with similar trends in Kmt2d-mutant liver tumors (Fig. 2K). Collectively, these data demonstrate that Arid1a and Kmt2d encode functional tumor suppressors in the liver, and autochthonous liver tumors with mutations in Kmt2d are more likely to respond to anti–PD-1 treatment, validating the results from the initial screen.

To further assess the role of Kmt2d loss on the cancer–immune interactions in liver cancer, we first established a primary tumor cell line from autochthonous Myc+sgTrp53 liver tumors generated in C57BL/6J (B6) mice (MA1L cells). We then transduced these cells with either vector control or sgKmt2d(Supplementary Fig. S2A). As Kmt2d deficiency has previously been implicated in genome instability (30), we cultured MA1L-Vector and MA1L-sgKmt2d cells in vitro for approximately 100 days and then transplanted the cells into mice to investigate the impact of Kmt2d deficiency. MA1L-sgKmt2d cells formed larger tumors in immunocompromised Rag1^−/− mice compared with vector control, again indicating the role of KMT2D as a tumor suppressor (Supplementary Fig. S2B). In contrast, in immunocompetent C57BL/6J mice, the tumors formed by Kmt2d-mutant cells were eliminated more rapidly compared with the control (Supplementary Fig. S2C). Notably, the MA1L cell line was established from an endpoint liver tumor and therefore had likely accumulated genetic/epigenetic alterations that increased the immunogenicity of the MA1L cells. In addition, the immunogenicity of Cas9 in the CRISPR system may contribute to the rejection of the MA1L cells by C57BL/6J mice. Nevertheless, these data reaffirm that Kmt2d mutation sensitizes liver tumors to immune rejection.

KMT2D is highly mutated in multiple human cancer types, with an average mutation frequency of 4% to 8% across all patients with cancer, and more than 20% of patients with skin cancer and bladder cancer (Fig. 3A; Supplementary Fig. S2D). These mutations are often truncating mutations or putative driver missense mutations (Fig 3B; Supplementary Fig. S2D), supporting the general role of KMT2D as a tumor suppressor gene in humans. Given the prevalence of KMT2D mutations across diverse human cancers, we sought to investigate whether Kmt2d deficiency promotes anti–PD-1 responsiveness in other tumor types.

As KMT2D is highly mutated in human bladder cancers, we mutated Kmt2d in MB49 bladder cancer cells (MB49-sgKmt2d; Supplementary Fig. S2E) and transplanted the cells into C57BL/6J mice, with vector-transduced cells (MB49-Vector) as a control. When treated at an early stage, both MB49-sgKmt2d and MB49-Vector tumors responded to anti–PD-1 treatment (Fig. 3C). However, MB49-sgKmt2d tumors were comparatively more responsive to anti–PD-1 therapy (Fig. 3C). Accordingly, anti–PD-1 treatment significantly improved overall survival, with MB49-sgKmt2d tumor-bearing mice surviving slightly longer than mice with MB49-Vector tumors (Fig. 3D). Examining the tumor growth kinetics, we noted an early stage of immune elimination of the MB49 tumors around 10 days post-injection. We therefore investigated the responses of late-stage MB49 tumors to anti–PD-1 therapy. When treated at a later time point after the initial immune response, MB49-Vector tumors did not respond to anti–PD-1 treatment (Supplementary Fig. S2F), but Kmt2d-mutant tumors were still responsive to anti–PD-1 treatment (Supplementary Fig. S2G). Anti–PD-1 therapy showed a significant survival benefit in mice bearing MB49-sgKmt2d tumors, but not in mice with MB49-Vector tumors (Supplementary Fig. S2H).

Similarly, we mutated Kmt2d in E0771 triple-negative breast cancer cells, B16F10 melanoma cells, and Lewis lung cancer (LLC) cells (Supplementary Fig. S2I), and then transplanted them into C57BL/6J mice. For the orthotopic E0771 model, only the mice bearing Kmt2d-mutant tumors significantly benefited from the anti–PD-1 treatment (Fig. 3E and F; Supplementary Fig. S2J and S2K). Orthotopic tumors formed by vector-transduced B16F10 cells (B16F10-Vector) were resistant to anti–PD-1 treatment (Fig. 3G), but the addition of Kmt2d loss sensitized the tumors to anti–PD-1 treatment (Fig. 3H). Similarly, subcutaneous Vector-transduced LLC tumors were resistant to anti–PD-1 treatment, but Kmt2d-mutant LLC tumors partially responded to anti–PD-1 treatment (Fig. 3I and J).

To examine whether Kmt2d loss induces anti–PD-1 sensitivity by specifically reshaping the local microenvironment, we utilized a dual-tumor model in which we transplanted LLC-Vector cells in the left flanks and LLC-sgKmt2d cells in the right flanks of individual mice. We found that only the Kmt2d-mutant tumors responded to anti–PD-1, anti–PD-L1, or anti-CTLA4 treatment (Supplementary Fig. S3A–S3D), suggesting that Kmt2d loss sensitizes tumors to checkpoint therapy by altering the tumor microenvironment. Together, these results demonstrate that Kmt2d mutation promotes anti–PD-1 response in 4 additional tumor types (bladder cancer, triple-negative breast cancer, melanoma, and lung cancer), generalizing the findings from the CRISPR-GEMM liver cancer system (Fig. 2G).

To investigate the mechanisms underlying the enhanced antitumor response against Kmt2d-mutant MA1L liver tumors, we analyzed the tumor immune microenvironment by flow cytometry (Supplementary Fig. S4) at day 11 when tumor regression begins. We found that Kmt2d-mutant MA1L tumors had more CD45⁺ immune cells, CD4⁺ T cells, and macrophages compared with controls (Supplementary Fig. S5A). CD8⁺ T-cell infiltration also trended toward an increase in Kmt2d-mutant tumors, although this was not statistically significant. To examine whether Kmt2d knockout consistently promotes immune infiltration, we further analyzed the tumor microenvironment of control or Kmt2d-mutant MB49 bladder cancers. We similarly found increased infiltration of CD45⁺ immune cells, especially CD8⁺ T cells, in MB49-sgKmt2d tumors (Supplementary Fig. S5B).

Because the MA1L liver tumors and MB49 bladder tumors were often rejected after anti–PD-1 therapy, we used the anti–PD-1–resistant LLC model to further investigate the effect of anti–PD-1 therapy on Kmt2d-mutant tumors. At 19 days post-induction of the dual-tumor LLC model (Fig. 4A), we analyzed the immune context of tumors formed by LLC-sgKmt2d and LLC-Vector cells, with or without anti–PD-1 therapy (Fig. 4B; Supplementary Fig. S6A). We observed that Lewis-sgKmt2d tumors had increased infiltration of CD45⁺ immune cells compared with Lewis-Vector tumors, particularly after anti–PD-1 treatment (Fig. 4B; Supplementary Fig. S6B). In Kmt2d-mutant tumors, we found significantly increased infiltration of T cells, including CD4⁺ T cells, CD8⁺ T cells, and IFNγ⁺ CD8⁺ T cells (Fig. 4B), as well as increased antigen-presenting cells, such as dendritic cells and macrophages (Fig. 4C). These changes became apparent after anti–PD-1 treatment. No difference was observed in the abundance of neutrophils and regulatory T cells (Fig. 4C; Supplementary Fig. S6A). By further analyzing the polarizations of tumor-infiltrating macrophages, we found the macrophages were dominated by tumor-associated macrophage 1 (TAM1), although the abundances of TAM1 and TAM2 were both increased in Kmt2d-mutant tumors after anti–PD-1 treatment (Fig. 4C). Notably, tumor-infiltrating T cells and innate immune populations (i.e., monocytes, neutrophils, and macrophages) expressed PD-1 on their surface (Fig. 4D; Supplementary Fig. S6A). Similar results were obtained using the single-tumor LLC model (Supplementary Fig. S6B). These data suggest that the enhanced efficacy of anti–PD-1 therapy in Kmt2d-mutant tumors may be due to its effects on T cells as well as myeloid cells, in concordance with a recent observation in these cell types (31).

To assess the relevance of these findings in clinical cohorts, we evaluated the correlation of KMT2D expression and immune infiltration status across multiple human cancers. In the TCGA, KMT2D expression levels are negatively correlated with intratumoral macrophage abundance in 21 of 33 (63.6%) human cancer types (Fig. 4E and F). Similarly, we observed that, across multiple cancer types, KMT2D expression is negatively correlated with expression of the monocyte-macrophage marker CD14 and the cytotoxic T-cell markers GZMB and GZMA (Supplementary Fig. S7A–S7C). Collectively, these analyses indicate intratumoral macrophage and cytotoxic CD8⁺ T-cell abundance is increased in tumors with low KMT2D levels, substantiating the results from the FACS data.

We sought to unravel the mechanisms by which Kmt2d mutation leads to elevated immune infiltration. We first validated that CRISPR targeting of Kmt2d led to a loss of KMT2D protein (Fig. 5A and B), with decreased levels of H3K4me1 but not H3K4me3 (Fig. 5C and D; Supplementary Fig. S7D). To explore the consequences of KMT2D loss, we further analyzed the TCGA data and identified all genes that are significantly correlated with KMT2D expression. To pinpoint associations that are generalizable across multiple cancer types, we selected genes that were concordantly correlated with KMT2D in several independent cohorts (Supplementary Fig. S7E). Genes positively correlated with KMT2D are enriched for transcription, poly(A) RNA binding, UBL conjugation pathway, mRNA processing, DNA damage and repair, and ubiquitin-protein transferase activity (Fig. 5E). Genes negatively correlated with KMT2D expression are enriched in ribosomal protein, mitochondrion, oxidative phosphorylation, antigen processing and presentation by major histocompatibility complex class I (MHC-I), proteasome, and cellular oxidant detoxification (Fig. 5F). These results indicate that KMT2D has an important role in transcriptional regulation and DNA repair. Indeed, recent studies showed that KMT2D could prevent collisions between transcription and replication machineries (32), and Kmt2d mutation resulted in transcription stress and DNA breaks in replicating regions (30).

To assess DNA damage in Kmt2d-mutant and control cells, we used immunofluorescence assays to quantify nuclear γH2AX foci, a marker of unrepaired DNA lesions (Fig. 5G). As a control, we found Arid1a-mutant cells had significantly higher numbers of γH2AX foci compared with the control (Fig. 5H), consistent with the role of ARID1A in mismatch repair (MMR; ref. 33). Notably, we found Kmt2d-mutant cells also had significantly higher levels of γH2AX foci in MA1L liver cancer cells (Fig. 5G and H) and MB49 bladder cancer cells (Supplementary Fig. S8A and S8B). These differences were not solely due to Cas9-induced double-stranded breaks, because cells transduced with sgKmt2d similarly showed increased γH2AX foci compared with cells transduced with Aavs1 locus (Ppp1r12c)–targeting sgRNA (Supplementary Fig. S8C and S8D). To confirm that Kmt2d mutation leads to elevated DNA damage, we cultured the Kmt2d-mutant cells and vector control cells in vitro for 71 days and performed whole-exome sequencing. We found that the mutation burden of vector cells decreased over time, whereas the mutation burden of Kmt2d-mutant cells increased (Fig. 5I). This increase was not due to differences in cell proliferation, because cells transduced with sgKmt2d or sgArid1a proliferated at similar rates as Vector-transduced cells (Supplementary Fig. S8E and S8F).

To examine whether these findings are recapitulated in human cancer, we knocked out KMT2D in H1299 human lung cancer cells (Supplementary Fig. S8G), finding that KMT2D loss led to significantly higher levels of γH2AX and 53BP1 foci (Supplementary Fig. S8H and S8I). We then analyzed the TCGA datasets and found that the tumor mutation burden (TMB) of KMT2D-mutant tumors was indeed significantly higher than the TMB of KMT2D–wild-type tumors across multiple cancer types (Fig. 5J and K). Because KMT2D is highly mutated in human bladder cancers (Fig. 3A), we further analyzed the correlations of KMT2D mutation with TMB and anti–PD-1 responses using two cohorts of patients with bladder cancer (34, 35). In both cohorts, we found that KMT2D-mutant bladder cancers had significantly higher TMB (Fig. 5L and M), and were more likely to respond to ICB with anti–PD-L1 (Fig. 5N and O), especially in the TMB-high patients (Supplementary Fig. S9A and S9B). Thus, KMT2D mutation leads to elevated DNA damage and is correlated with higher TMB in multiple types of human cancers, as well as better responses to checkpoint immunotherapy (10).

As KMT2D is an epigenetic modifier associated with chromatin regulation and enhancer activation, we performed assay for transposase-accessible chromatin using sequencing (ATAC-seq) to examine the chromatin landscape of Kmt2d-mutant or control MA1L liver tumor cells (Fig. 6A), with or without IFNγ treatment. Correlation analysis revealed a robust clustering of sgKmt2d cells separately from control cells (Fig. 6B). Comparing Kmt2d-mutant to control cells (without IFNγ treatment), 10,791 sites were more accessible and 9,553 sites were less accessible (Fig. 6C). Motif analysis of the more accessible sites revealed enrichment for AP1 family factors, CTCF, and TCF3/TCF4 (Fig. 6D), whereas the less acessible sites were also enriched for AP1 family factors (Fig. 6E). The common enrichment for AP1-binding motifs suggests a global rewiring of AP1-driven programs upon perturbation of Kmt2d.

Upon IFNγ treatment, we observed large changes in chromatin accessibility. Interestingly, significantly fewer sites became less accessible in sgKmt2d cells compared with Vector cells after IFNγ treatment, whereas a similar number of sites became more accessible (Fig. 6F). These differentially accessible sites could be broadly classified into 6 clusters (Fig. 6G). Sites in Cluster 3 (989 sites) were less accessible in sgKmt2d cells compared with control cells at baseline prior to IFNγ, but became more accessible after IFNγ treatment to match the levels in control cells (Fig. 6G–I). Conversely, sites in Cluster 5 were more acessible in control cells compared with sgKmt2d cells prior to IFNγ treatment, but became less accessible after IFNγ treatment to a level similar to sgKmt2d cells. These findings indicate that IFNγ stimulation converges the chromatin landscapes of wild-type and Kmt2d-mutant cells (Supplementary Fig. S9C). Thus, Kmt2d deficiency systematically reshapes the chromatin accessibility of IFNγ-induced genes at baseline, and IFNγ stimulation partially normalizes these differences. Motif analysis reaffirmed the systematic rewiring of AP1 family factors in Kmt2d-mutant cells (Fig. 6J–M), further demonstrating that these alterations also influence chromatin changes in response to IFNγ treatment.

The association of KMT2D with transcription regulation and DNA damage (Fig. 5; Supplementary Fig. S8), along with the broad alterations in chromatin accessibility (Fig. 6), suggest that Kmt2d loss would lead to systematic transcriptional remodeling. To study the transcriptomic changes caused by Kmt2d mutation, we performed RNA sequencing (RNA-seq) on the MA1L liver tumor cells. Comparing with vector controls, we found that 753 genes were upregulated and 1,540 genes were downregulated in Kmt2d-mutant cells (Fig. 7A). As the ATAC-seq analyses had pointed to a rewiring of AP1 family factors, we investigated the expression of genes encoding different transcription factors that constitute AP1 dimers. In MA1L-sgKmt2d cells, we observed upregulation of Fosl2, Mafb, and Maf (Supplementary Fig. S9D), and downregulation of Fosl1, Maff, and Atf3 (Supplementary Fig. S9E). To explore the potential downstream consequences of increasing the relative abundance of FOSL2 within AP1 dimers, we examined FOSL2 chromatin immunoprecipitation sequencing (ChIP-seq) data from HepG2 human liver cancer cells. We found that FOSL2 binds near important regulators of tumor–immune interactions, including IFNGR1, IFNGR2, JAK1, and JAK2 (Supplementary Fig. S9F).

Analysis of the upregulated genes in Kmt2d-mutant cells revealed multiple enriched categories, including extracellular matrix, transcription regulation, focal adhesion, zinc-finger, ECM–receptor interaction, and WNT signaling (Fig. 7B). Conversely, Kmt2d-mutant cells showed downregulation of oxidoreductase genes and multiple metabolic pathways, including cholesterol metabolism, tricarboxylic acid (TCA) cycle, and lipid metabolism, as well as mitochondrion, lysosome, and innate immunity (Fig. 7C).

Genes encoding chemokines (Cxcl1, Cxcl5, Cxcl15) related to neutrophil recruitment were upregulated in Kmt2d-mutant cells compared with the vector control (Supplementary Fig. S10A and S10B). qPCR validation experiments confirmed the upregulation of Cxcl1 and Cxcl15 mRNA upon Kmt2d loss in both MA1L liver cancer cells and MB49 bladder cancer cells (Supplementary Fig. S10B and S10C). We then investigated whether these in vitro transcriptional changes were reflected in vivo. When profiling the chemokines in MA1L-derived tumors, we detected significantly higher levels of CCL2, CCL5, CCL22, and CXCL9 protein in MA1L-sgKmt2d tumors, but similar levels of CXCL1 and CXCL5 protein (Fig. 7D). These chemokine changes were notably distinct from the chemokine profiles of in vitro cultured MA1L cells (Supplementary Fig. S10D), suggesting that the upregulation of CCL5, CCL22, and CXCL9 in MA1L-sgKmt2d tumors in vivo was likely contributed by other immune cells that were recruited to the tumors. Thus, the increase of these chemokines may explain the increased recruitment of antigen-presenting cells and T cells in Kmt2d-mutant liver tumors (Supplementary Fig. S5). However, we did not detect significant changes of these chemokines in MB49 tumors (Supplementary Fig. S10E and S10F) or LLC tumors (Supplementary Fig. S10G–S10I). These differences could be caused by a multitude of complex factors, including differences in tumor stage, the cell types involved, and their genetic backgrounds.

We next used the RNA-seq data to predict the neoantigens in Kmt2d-mutant and control cells. We found that Kmt2d-mutant and control cells shared 96 predicted neoantigens. Notably, Kmt2d-mutant cells have an additional 56 predicted neoantigens that were not predicted in the control cells, whereas the control cells have 10 additional predicted neoantigens (Fig. 7E). In addition to neoantigens encoded within the canonical proteome, aberrant ribosomal products and alleged noncoding regions can serve as a major source of tumor antigens (36–39). A previous study suggested that Kmt2d mutation leads to transcriptional stress (30), and our analysis of the TCGA implicated KMT2D in transcription regulation and mRNA processing (Fig. 5E). We therefore assessed whether Kmt2d mutation affects RNA splicing and transcript quality by examining intronic retention rates in mRNA. Kmt2d-mutant cells had significantly increased intronic retention in the mRNA, with approximately 12% more intronic reads than vector control cells (Fig. 7F). This is consistent with a role for KMT2D in transcription regulation and mRNA processing. As many transposable elements (TE) such as endogenous retroviruses are often contained within introns, we then assessed the expression of TEs in Kmt2d-mutant versus control cells. We observed that 66 TEs were significantly upregulated in Kmt2d-mutant cells, whereas only 8 TEs were downregulated (Fig. 7G). These data suggested that Kmt2d mutation leads to transcriptional stress and dysregulated RNA splicing, leading to intron retention and heightened expression of TEs. In conjunction with the increased DNA damage and mutation burden upon KMT2D loss, these findings illuminate several sources of potential neoantigens in Kmt2d-mutant cells.

To generate antigenic peptides, coding transcripts must first be translated and the resultant proteins ubiquitinated for proteasome-mediated degradation. A number of genes involved in ubiquitination were transcriptionally upregulated in Kmt2d-mutant cells compared with vector control cells (Fig. 7H). To examine whether the levels of ubiquitinated proteins destined for proteasomal degradation are higher in Kmt2d-mutant cells, we performed immunoblot assays of ubiquitinated proteins with or without MG132, an inhibitor of proteasome degradation (Fig. 7I–M). Without MG132, Kmt2d-mutant cells had either higher or comparable levels of ubiquitinated proteins compared with control (Fig. 7I, J, L, and M; Supplementary Fig. S11A and S11B). When proteasome degradation was inhibited by MG132, we detected significantly higher levels of ubiquitinated proteins in Kmt2d-mutant MA1L and Kmt2d-mutant MB49 cancer cells compared with the corresponding controls (Fig. 7K and M), although we did not see the same trend in LLC cells (Supplementary Fig. S11A and S11B). Interestingly, when treated by IFNγ, both Kmt2d-mutant and control MB49 cells showed higher levels of ubiquitinated proteins (Supplementary Fig. S11C and S11D). These results indicated that Kmt2d-mutant cells generated more proteins, which were ubiquitinated and subjected to proteasomal degradation. This finding is further supported in patient cohorts, as proteasome-related genes were enriched among the genes negatively correlated with KMT2D expression in the TCGA (Fig. 5F).

Following protein ubiquitination and proteasomal degradation, the resultant peptides must be loaded onto MHC-I to be presented. However, the RNA-seq data revealed that several genes in the MHC-I family were downregulated in Kmt2d-mutant MA1L cells compared with vector control cells (Supplementary Fig. S11E), potentially dampening the presentation of potential antigens. Analyzing the levels of cell-surface MHC-I by flow cytometry, we found that the baseline levels of surface H2-K^b on Kmt2d-mutant cells were comparable to that of vector control (Fig. 7N; Supplementary Fig. S11F). When stimulated by IFNγ, the surface levels of total H2-K^b were significantly increased and Kmt2d-mutant cells exhibited even higher levels of H2-K^b than the control cells (Fig. 7N; Supplementary Fig. S11F). To explicitly test whether tumor antigens would be more efficiently presented in Kmt2d-mutant cells, we transduced the MA1L tumor cells with a phosphoglycerate kinase (PGK) promoter driving ovalbumin (OVA) to examine the presentation levels of H2-K^b-SIINFEKL. We did not observe any difference in the levels of surface H2-K^b-SIINFEKL in Kmt2d-mutant and vector control cells without IFNγ treatment (Fig. 7O). However, Kmt2d-mutant cells had significantly higher levels of H2-K^b-SIINFEKL than vector control cells when stimulated by IFNγ (Fig. 7O), indicating that Kmt2d-mutant cells respond strongly to IFNγ treatment by upregulating antigen presentation.

Taken together, these results demonstrate that Kmt2d mutation in tumor cells may lead to higher levels of neoantigens by causing DNA damage, increasing mutation burden, inducing intronic retention, and activating expression of TEs. Furthermore, Kmt2d-mutant cells are characterized by increased proteasomal degradation as well as increased IFNγ-stimulated antigen presentation. As a consequence, Kmt2d-mutant tumors exhibit elevated infiltration of PD-1⁺ T cells and macrophages, the latter of which may further amplify the antitumor effect of anti–PD-1 therapy by activating the adaptive immune system. As a histone methyltransferase, KMT2D likely has pleiotropic effects on tumors; at least in the context of the experiments shown here, a plausible mechanistic explanation is that multiple pathways perturbed by Kmt2d deficiency converge to potentiate response to anti–PD-1 immunotherapy (Fig. 7P).

In this study, we performed a mutagenesis screen in a CRISPR-GEMM liver tumor model to pinpoint genetic modulators of immunotherapy response. Distinct from previous studies using in vitro coculture and tumor transplantation models, our screening system utilizes an autochthonous tumor model that preserves the native microenvironmental context. Here, we systematically mapped the fitness of diverse mutations under immunotherapy treatments, demonstrating the power of this platform for interrogating causal relationships of specific mutations and response to ICB.

Identifying the molecular features that dictate response to immunotherapy has the potential to provide valuable guidance to clinicians. We found that mutations in the H3K4 methyltransferase Kmt2d potentiate response to anti–PD-1 therapy in diverse cancer types. Of note, the heightened anti–PD-1 response and elevated DNA damage caused by Kmt2d loss does not appear to be dependent on mutant p53. Among the tumor cell lines used in this study, E0771, B16F10, and MA1L cells are p53-deficient, whereas LLC and MB49 cells are p53-competent. In addition, a prior study that demonstrated that Kmt2d deficiency leads to DNA damage and genomic instability was performed using Trp53–wild-type mouse MEF cells (30). As KMT2D is a tumor suppressor gene that is recurrently mutated across multiple human cancers, the identification of KMT2D deficiency as a predictor of anti–PD-1 therapy response may have important implications for patient stratification and clinical decision-making.

Our study revealed that Kmt2d-mutant cancer cells exhibited an elevated level of DNA damage and higher mutation burden, and we further corroborated these findings in patient tumor datasets. These findings are consistent with the reported role of KMT2D in genome stability, as KMT2C- and KMT2D-dependent H3K4 methylation at replication forks was found to be involved in replication stress (40). We also found that Kmt2d deficiency led to compromised RNA splicing and activation of TEs. Intron retention and activation of TEs can potentially result in the presentation of immunogenic antigens, and we observed that Kmt2d-mutant cells exhibit increased protein ubiquitination, indicating increased proteasomal degradation. Given the increased mutation burden and aberrant transcription of Kmt2d-mutant cells, we speculate that upregulation of proteasome activity may be a compensatory response to the production of abnormal proteins in these cells. Importantly, the resultant increase in proteasomal degradation is further associated with increased IFNγ-stimulated antigen presentation in these cells, thus providing an explanation for the enhanced sensitivity to anti–PD-1 therapy in Kmt2d-mutant tumors.

Interestingly, when we rechallenged mice that had successfully rejected Kmt2d-mutant tumors with either Kmt2d-mutant or wild-type tumor cells, both types of tumors were rejected within 2 weeks (Supplementary Fig. S11G). Although the Kmt2d-mutant cells had accumulated more mutations, it is worth emphasizing that the Kmt2d-mutant and wild-type cells nevertheless share most of their mutations by virtue of their common background, and tumor rejection is often mediated by multifaceted immune responses against multiple antigens. In addition, Kmt2d-mutant tumors display higher levels of both unique and shared antigens after anti–PD-1 treatment, thereby promoting the development of immune memory against both unique and shared tumor antigens. We also note that the CRISPR components may have some degree of immunogenicity, thus contributing to the rejection of these cells by B6 mice; nevertheless, all experiments were conducted in parallel with mutant (gene-targeting sgRNAs) groups directly compared with controls (vector or nontargeting guide RNA).

We further found that Kmt2d-mutant cells had increased levels of the myeloid-recruiting cytokines Cxcl1 and Cxcl5 at both the RNA and protein level. However, when profiling the chemokines in primary tumors derived from these cells, we instead detected increased CCL2, CCL5, CCL22, and CXCL9 levels in Kmt2d-mutant tumors (Supplementary Fig. S11C), suggesting that the upregulation of these chemokines was due to host immune cells recruited to the tumors. These chemokines would promote the infiltration of antigen-presenting cells and T cells into the tumors. Indeed, Kmt2d-mutant tumors had significantly increased infiltration by antigen-presenting cells, CD4⁺ T cells, and CD8⁺ T cells, which all expressed high levels of PD-1. Recent work has demonstrated that PD-1 blockade promotes antitumor immunity not only through its action on T cells, but also by leveraging myeloid-derived innate immune cells toward antitumor function (31, 41). The elevated immune infiltration in Kmt2d-mutant tumors can therefore be explained by the elevated antigenicity of Kmt2d-mutant cells as well as elevated myeloid cell recruitment. However, a limitation of these data is that certain validation experiments were performed using subcutaneous transplantation models that may not accurately reflect the tumor microenvironment of the cancer's origin.

In summary, these data collectively demonstrate that Kmt2d loss sensitizes diverse tumor types to checkpoint blockade immunotherapy. This study showcases the power of CRISPR-GEMM models for interrogating complex molecular landscapes in native tumor microenvironments, enabling the dissection of immunotherapeutic responses. Given the prevalence of KMT2D mutations in diverse cancer types, our study could help identify a sizeable patient subpopulation that may have higher chances of being sensitive to ICB therapies such as PD-1 checkpoint blockade.

This study has received institutional regulatory approval. All recombinant DNA and biosafety work was performed under the guidelines of the Yale Environment, Health and Safety (EHS) Committee with an approved protocol (Chen-rDNA-15–45; Chen-rDNA-18–45). All animal work was performed under the guidelines of Yale University Institutional Animal Care and Use Committee with approved protocols (Chen-2015–20068; Chen-2018–20068). All human sample work was performed under the guidelines of Yale University Institutional Review Board with an approved protocol (HIC#2000020784).

The AAV-CRISPR vector was designed to express Cre recombinase under a liver-specific TBG promoter. Each vector has two sgRNA expression cassettes, with one of them encoding an sgRNA targeting Trp53, and the other as an open sgRNA expression cassette (double SapI sites for sgRNA cloning). We also designed a liver-specific AAV-CRISPR vector with only one sgRNA expression cassette as a control to study the impact of Trp53 knockout. The mTSG library was generated as described previously (26, 27), with more than 100× coverage to ensure proper representation of the library.

8.7 μg of AAV9 serotype plasmid, 10.4 μg of pDF6 helper plasmid, and 5.2 μg of AAV expression plasmid were added into 450 μL Opti-MEM and mixed well, then complexed with PEI, incubating at room temperature for 10 to 15 minutes before adding them drop-wise into HEK293FT cells at 80% to 90% confluency. Forty-eight to 72 hours post-transfection, the transfected cells were collected. AAVs were purified using chloroform extraction and titrated by qPCR assay (see Supplementary Methods).

Rosa26-LSL-Cas9–2A-EGFP (LSL-Cas9) mice were bred with C57BL/6J mice, FVB.129S6(B6)-Gt(ROSA)26Sortm1(Luc)Kael/J mice (LSL-Luc), or C57BL/6N-Gt(ROSA)26Sortm13(CAG-MYC,-CD2*)Rsky/J mice (LSL-Myc for short). Mixed-gender (randomized males and females) 8- to 12-week-old mice were used for experiments. For intravenous injection of AAVs, the mice were restrained in a rodent restrainer (Braintree Scientific). Tails were sterilized by 70% ethanol, and 100 to 200 μL of concentrated AAVs (∼1–2 × 10¹¹ GCs in total) were injected per mouse. All the mice survived the procedure. When tumor initiation was observed, the mice were randomly assigned into 3 groups to receive treatment of PBS, 8 mg/kg anti–PD-1, or 4 mg/kg anti-CTLA4 twice a week, for 5–6 doses at the indicated times.

After AAV injection, mice were imaged by IVIS each month. Briefly, mice were anesthetized by isofluorane, and then 100 to 150 μL of 30 mg/mL firefly d-luciferin potassium salt was intraperitoneally injected with approximately 150 mg/kg body weight. Ten to 15 minutes after injection, the mice were imaged for in vivo tumor growth using an IVIS machine (PerkinElmer). Relative tumor burden was quantified using LivingImage software (PerkinElmer).

We observed that the LSL-Cas9 mice receiving AAV-mTSG intravenous injections rapidly deteriorated in their body condition scores (due to tumor development in most cases). Mice with body condition score < 2 were euthanized, and the euthanasia date was recorded as the last survival date. Survival data was analyzed by standard Kaplan–Meier method, using GraphPad Prism. Statistical significance was assessed by log-rank test. Mice euthanized early in a healthy state were excluded from calculation of survival percentages.

The genomic DNA from frozen ground tissue was purified using DNeasy Blood & Tissue Kits (Qiagen) or standard DNA extraction protocol (see Supplementary Methods). The concentration was measured using a NanoDrop (Thermo Fisher Scientific).

Rosa26-LSL-Cas9–2A-EGFP knock-in mice were bred with C57BL/6N-Gt(ROSA)26Sortm13(CAG-MYC,-CD2*)Rsky/J mice (LSL-Myc for short) to obtain LSL-Myc; LSL-Cas9 mice. Mixed-gender mice of 7–12 weeks old were used for experiments. Autochthonous liver tumors with Myc overexpression and mutant Trp53 were induced by injecting 1–2*10¹¹ GCs of TBG-Cre AAVs carrying sgTrp53 + sgNTC into immunocompetent LSL-Myc;LSL-Cas9 mice. Autochthonous liver tumors with additional Arid1a or Kmt2d mutations were induced by injecting 1–2 × 10¹¹ GCs of AAVs carrying sgTrp53 + sgArid1a or sgTrp53 + sgKmt2d into LSL-MYC; LSL-Cas9 mice. Liver tumorigenesis was detectable 60 days after AAV injection, at which point the mice were randomly assigned into 2 groups to receive treatment of PBS or 8 mg/kg anti–PD-1, twice a week for 5 doses.

Liver tumors were collected and fixed in 10% neutral formalin for 2 to 5 days, then transferred into 70% ethanol. Hematoxylin and eosin staining or IHC staining of CD45, CD3, F4/80, or cytokeratin pan-cytokeratin were performed on 3- to 5-μm tissue sections using standard procedures at Yale Pathology Core Facility. To quantify CD45-, CD3-, and F4/80-positive cells, the slides of different regions of tumor samples were quantitatively scored using the IHC profiler in ImageJ software (42), and only the percentage distribution of high positive was regarded as positive staining.

HEK293FT cells were purchased from Thermo Fisher Scientific (catalog no. R70007). E0771 mouse triple-negative breast cancer cells were purchased from CH3 (catalog no. 940001). B16F10 mouse melanoma cells (catalog no. CRL-6475), mouse Lewis lung cancer carcinoma cells (catalog no. CRL-1642), and H1299 human lung carcinoma (non–small cell lung cancer) cells (catalog no. CRL-5803) were purchased from ATCC. MB49 mouse bladder cancer carcinoma cells were purchased from Sigma-Aldrich (catalog no. SCC148). MA1L and MA1NC cells were established from autochthonous liver tumors with Myc overexpression and Trp53 knockout mutation generated by the intravenous injection of sgTrp53-targeted AAVs into B6 background LSL-Myc;LSL-Cas9 mice. The cells tested negative for Mycoplasma contamination. All the purchased cell lines have been authenticated by the original vendors. All cells were cultured in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin, in a CO₂ cell incubator 37°C.

Autochthonous liver tumors with Myc overexpression and Trp53 knockout mutation were generated by the intravenous injection of Trp53-targeted AAVs into LSL-Myc;LSL-Cas9 mice. At the survival endpoint, the liver tumors were isolated and made into single-cell suspensions by digestion with collagenase IV after mincing into small pieces and passing through 40-μm cell restrainer. The cells were then cultured in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin. To knock out Kmt2d or Arid1a, the tumor cells were transduced with lentiviruses carrying sgRNAs targeting Kmt2d or Arid1a. The transduced cells were selected with 3 to 5 μg/mL puromycin, and the knockout of Kmt2d and Arid1a was confirmed by T7E1 assay.

Cas9-expressing MB49 cells, E0771 cells, B16F10 cells, and LLC cells were generated by transduction with lentiviruses carrying EFS-Cas9–2A-BlastR-WPRE and selected under 10 μg/mL blasticidin S. To knock out Kmt2d, these Cas9-expressing cells were transduced with lentiviruses carrying an Kmt2d sgRNA, and cells transduced with lentiviral vector or nontargeting sgRNA were used as a control. The transduced cells were selected with 3 to 5 μg/mL puromycin at 24 hours post-infection. To generate syngeneic mouse bladder tumors, 5 × 10⁶ of vector- or Kmt2d sgRNA- transduced MB49 cells were transplanted subcutaneously into the right flank of C57BL/6J mice. To generate orthotopic breast tumors, 2 × 10⁶ of vector- or Kmt2d sgRNA–transduced E0771 cells were transplanted into the fat pad of C57BL/6J mice. To generate melanoma or Lewis lung tumors in C57BL/6J mice, 2 × 10⁶ vector- or Kmt2d sgRNA–transduced B16F10 cells or LLC cells were subcutaneously transplanted into the right flank of C57BL/6J mice. Tumor growth was monitored and assigned into two groups to receive the treatment of 8 mg/kg anti–PD-1 or PBS twice a week at the indicated time.

To generate a dual-tumor model of LLC, 2 × 10⁶ of LLC-Vector and LLC-sgKmt2d cells were transplanted into the left flank and right flank of C57BL/6J mice, respectively. Tumor growth was monitored, and mice were assigned into two groups to receive 8 mg/kg anti–PD-1 or PBS twice a week at indicated times.

All antibodies for flow were purchased from BioLegend or eBioscience. Single-cell suspensions of tumors or spleens were prepared using a gentleMACS tissue dissociation system. All flow antibodies were used at 1:100 dilutions for staining unless otherwise noted. After staining, cells were centrifuged at 300–600 × g for 5 minutes, and washed twice with staining buffer before being analyzed or sorted on a BD FACSAria. The data was analyzed using the FlowJo software (v9.9.4 or v10.3). A previously reported strategy was used to define the immune populations in tumors (43).

The CRISPR knockout construct Lenti-U6-sgBsmBI-EFS-Puro-WPRE was generated. To clone sgRNA targeting individual genes, such as Kmt2d and Arid1a, the corresponding oligos were synthesized, annealed, and cloned into BsmBI linearized lentiviral knockout vectors. The following sgRNAs were used for Kmt2d. sgRNA1: GCCGGCTATGTCGGGCCTGT; sgRNA3: GTGTGTGAGACATGTGACAA. For Arid1a, sgRNA4: GACGCATGAGCCATTCTCCC. Lentiviruses were produced by cotransfecting the lentiviral CRISPR knockout plasmids, together with packing plasmids pMD2.G and psPAX2, into 80% to 90% confluent HEK293FT cells. The lentivirus-containing supernatants were collected at 48 and 72 hours post-transfection, aliquoted, and stored at −80°C before use. To knock out KMT2D in human lung cancer H1299 cells, the corresponding oligos of sgRNA1: GGTGGAAATTCCCGCCAACG; sgRNA2: AAATGGCTGTTGATCCCATG were synthesized, annealed, and cloned into BsmBI linearized lentiviral knockout vectors. After lentiviral production, the Cas9-transduced tumor cells were infected and selected with 3–5 μg/mL puromycin to obtain individual gene knockout cells. CRISPR mutagenesis was confirmed by extracting genomic DNA for T7E1 assays.

Vector, sgKmt2d, or sgArid1a transduced primary liver tumor cells (MA1L), MB49-Vector, MB49-sgKmt2d, LLC-Vector, and LLC-sgKmt2d cells were seeded into 6-well plates or 10-cm dishes and cultured for 24 hours. Then, 15 μmol/L MG132, 10 ng/mL IFNγ, or DMSO was added and incubated for 2 to 3 hours before harvesting the treated cells. The harvested cells were washed twice with ice-cold PBS, and then lysed with 1× RIPA buffer on ice for 15 minutes. Cell lysates were centrifuged at 12,000 × g for 15 minutes at 4°C and protein-containing supernatant was collected. Protein concentration was measured using a BCA assay (Abcam) and approximately 20 μg of protein from each sample were loaded into SDS-PAGE gels. After electrophoresis and transmembrane, immunoblot assays were performed with antibody against ubiquitin (clone Ubi-1, Sigma), with GAPDH being used as internal control. The relative levels of ubiquitinated protein were quantified by grayscale analysis.

2 × 10⁵ vector-, sgKmt2d-, or sgArid1a-transduced MA1L primary liver tumor cells (MA1L-Vector, MA1L-sgKmt2d, MA1L-sgArid1a), MB49-Vector, and MB49-sgKmt2d cells were seeded into 12-well plates. To test the effect of IFNγ on surface MHC-I or peptide–MHC-I presentation, 0, 5 ng/mL, or 10 ng/mL IFNγ were added and treated for 24 to 48 hours. The treated cells were collected and washed twice with 2% FBS in PBS. Then, the cells were stained with 1:100 diluted PE-H-2K^b/H-2D^b, and APC-SIINFEKL-H-2K^b for 30 minutes on ice and washed twice with 2% FBS in PBS before flow cytometry analysis. Samples were run on Attune NxT Flow Cytometer and the mean fluorescence intensities were quantified.

RNA from control and Kmt2d-mutant cells was extracted using TRIzol Reagent (Invitrogen) by following standard RNA extraction protocols. The first-strand cDNA of RNA was synthesized using SuperScript IV Reverse Transcriptase (Invitrogen). After normalizing the concentrations of cDNA with nuclease-free water, quantitative PCR (qPCR) was performed by adding designated TaqMan probe of genes of interest, and GAPDH was used as an internal positive control.

Cells in a 6-well plate or 10-cm dish were washed twice with ice-cold PBS. The cells were then lysed with 1× RIPA buffer on ice for 15 minutes, or nuclear protein purification using Nuclear Extraction Kit (Abcam). Cell lysates were centrifuged at 12,000 × g for 15 minutes at 4°C, and protein-containing supernatant was collected. Protein concentration was measured using a BCA Assay (Abcam) and 20 μg of protein in each sample was loaded into SDS-PAGE gel. After electrophoresis, proteins separated in gel were transferred into nitrocellulose membranes. Membranes were blocked at room temperature for 1 hour using 5% skim milk in TBST, followed by incubation with primary antibody in 4°C overnight. After washing three times with TBST, horseradish peroxidase–conjugated secondary antibody was added and incubated at room temperature for 30 to 60 minutes. The chemiluminescent substrate (Clarity Western ECL Substrate, Bio-Rad) was added on top of blot membrane according to the manufacturer's instructions. The signals were captured using a CCD camera-based imager (GE Healthcare).

Raw FASTQ reads were mapped to the mm10 genome using bwa mem v.0.7.17 (44). BAM files were sorted and indexed using SAMtools v1.3 (45). Indel variants were then called using SAMtools and VarScan v2.3.9 (46). All detected indels were filtered by requiring that each indel must overlap the ±3 bp window surrounding the predicted cut site of the closest sgRNA. We excluded variants at Rps19 sg5 because vector control samples were also found to have heterozygous mutations at this site.

The remaining indel variants were summed for each sgRNA site to obtain a mutation frequency table. To further filter detected variants, we employed a false discovery approach based on vector control samples. For each sgRNA in the library, we took the highest variant frequency across all the vector control samples and set this value as the minimum cutoff when filtering the mTSG samples. In addition, we further set a 5% variant frequency cutoff to ensure stringent detection of indels. The filtered sgRNA variant frequency table was then averaged by gene to obtain the gene-level frequency table. We then used the gene-level variant frequencies to determine enrichment or depletion of specific mutations in ICB-treated versus PBS-treated samples by two-tailed unpaired t test.

KMT2D mutation status was queried using cBioPortal using the OQL specifiers “MUT HOMDEL” for all mutations and deletions (47), or “DRIVER NONSENSE NONSTART NONSTOP FRAMESHIFT SPLICE TRUNC HOMDEL” for anticipated loss-of-function mutations and deletions. The different cancer types in the curated nonredundant set were consolidated on the basis of the tissue of origin.

To determine the association between KMT2D and tumor mutation burden, the cBioPortal was queried across the PanCancer TCGA cohorts. Tumor types with at least 5 KMT2D-mutant samples were considered for analysis. Statistical significance was assessed by two-tailed Mann–Whitney test.

RNA-seq count data from TCGA were downloaded from the GDC Data Portal and normalized to transcripts per million. The Spearman correlation between each gene and KMT2D was calculated, and P values were adjusted for multiple comparisons by the Benjamini–Hochberg method. We then tabulated the number of individual cancer types for which a given gene was concordantly correlated with KMT2D. Inferred cell type abundances in TCGA tumors were downloaded from the xCell website (http://xcell.ucsf.edu/). Correlations between KMT2D and cell type abundances were calculated in the same manner as with individual genes.

To obtain the pan-cancer gene sets that are positively or negatively correlated with KMT2D, we selected for genes that are concordantly correlated with KMT2D across multiple cancer types. On the basis of the empirical cumulative density function of the number of cancer types for which each gene was significantly correlated with KMT2D, we selected a cutoff that would select approximately the top 5% of genes (30+ cancer types among positively correlated genes, 21+ cancer types among negatively correlated genes). DAVID gene ontology analysis was performed on the resultant gene sets.

Raw FASTQ reads were mapped to the mm10 genome using the bwa mem function in BWA v.0.7.17. Mutations in MA1L cells were called using Strelka v2.9.2 by comparing with wild-type liver exomes from C57BL/6J mice.

Raw FASTQ reads were mapped to the mm10 genome using Bowtie v2 (48). ATAC-seq accessible regions were called using MACS2 (49). Accessible regions across all samples were combined and the read counts in each region were tabulated. Pairwise Spearman correlations were calculated using the read counts in each region. Differential accessibility analysis was performed using DESeq2 (50). Intersection of accessible regions and motif analysis was performed using HOMER (51).

Raw FASTQ reads were quantified to the mm10 transcriptome using Kallisto (52). Differential expression analysis was performed using Sleuth (53). DAVID gene ontology analysis was performed on genes with an adjusted P < 0.05. For neoantigen prediction, RNA-seq reads were aligned to the mm10 genome using STAR, then mutations were called using the RNA-seq mode of Strelka v2 (54). The resultant mutations were annotated using VEP, then neoantigens were predicted using pVACtools with H-2K^b and H-2D^b as the candidate MHC-I alleles (55).

To analyze expression of transposable elements, the raw FASTQ reads were first realigned using STAR (56) with modified settings (outFilterMultimapNmax 100, winAnchorMultimapNmax 100). Transposable elements were quantified by TEcount from TEToolkit. Differential expression was assessed using the raw counts of all genes and transposable elements with DEseq2.

FOSL2 ChIP-seq data in HepG2 cells were downloaded from the ENCODE database and visualized in the Integrative Genomics Viewer.

Sample size was determined according to the lab's prior work or similar studies in the literature.

In animal experiments, mice were randomized by sex, cage, and littermates. In vitro experiments were not randomized or blinded. Investigators were blinded in mouse experiments by labeling cages with generic identifiers. In next-generation sequencing data analysis, investigators were blinded for initial processing of the original data using key-coded metadata.

Data between two groups were analyzed using a two-tailed unpaired t test. Different levels of statistical significance were accessed on the basis of specific P values and type I error cutoffs (0.05, 0.01, 0.001, 0.0001). GraphPad Prism and R were used for analyses.

Codes used for data analysis or generation of the figures related to this study are available on GitHub (https://github.com/rdchow/immunoMIPS/).

All data generated or analyzed during this study are included in this article and its supplementary information files. Specifically, source data and statistics for non–high-throughput experiments such as flow cytometry, qPCR, protein experiments, and other molecular or cellular assays are provided in Supplementary Tables. Processed data for genomic sequencing (e.g., RNA-seq, exome sequencing, ATAC-seq) and other forms of high-throughput experiments are provided as processed quantifications in Supplementary Datasets. Raw sequencing data have been deposited to NIH Sequence Read Archive (SRA) or Gene Expression Omnibus: MIPS sequencing of mouse tissue and exome sequencing of MA1L cells (PRJNA634679); ATAC-seq and RNA-seq (GSE151227). Original cell lines are available at commercial sources listed in Supplementary Data. Genetically modified cell lines are available via the Chen lab. Most data, reagents, methods, computational codes, and materials that support the findings of this research are available from the corresponding author upon reasonable request. Some material used in the reported research may require requests to collaborators and agreements with other entities. Requests are reviewed by Yale University to verify whether the request is subject to any intellectual property or confidentiality obligations. Any material that can be shared will be released via a Material Transfer Agreement.

G. Wang reports grants from Cancer Research Institute (CRI Postdoctoral fellowship) during the conduct of the study. R.D. Chow reports grants from NIH/NCI during the conduct of the study. X. Dai reports grants from the Charles H. Revson Foundation during the conduct of the study. S. Chen is a cofounder, funding recipient, and scientific advisor of EvolveImmune Therapeutics, which is not related to this study. S. Chen reports grants from NIH (DP2CA238295, R01CA231112, U54CA209992, R33CA225498, RF1DA048811), Damon Runyon Cancer Research Foundation (DFS-13-15), Melanoma Research Alliance (16-003524, 412806), St-Baldrick's Foundation (426685), Breast Cancer Alliance, Cancer Research Institute (CLIP), American Association for Cancer Research (17-20-01-CHEN), The Mary Kay Foundation (017-81), The V Foundation (V2017-022), Ludwig Family Foundation, DoD (W81XWH-17-1-0235, W81XWH-20-1-0072), Sontag Foundation (distinguished scientist award), and Chenevert Family Foundation during the conduct of the study; and other from Dexter Lu (gift on brain cancer research), EvolveImmune Therapeutics (sponsored research agreements), Blavatnik Family Foundation (innovation awards); and grants from ACGT (pending grant) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

G. Wang: Data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft. R.D. Chow: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-original draft. L. Zhu: Formal analysis, validation, investigation, methodology. Z. Bai: Data curation, formal analysis, validation, investigation, methodology. L. Ye: Data curation, formal analysis, investigation, methodology. F. Zhang: Validation. P.A. Renauer: Formal analysis, visualization. M.B. Dong: Methodology, writing-review and editing. X. Dai: Validation. X. Zhang: Validation. Y. Du: Validation. Y. Cheng: Investigation. L. Niu: Investigation. Z. Chu: Validation. K. Kim: Validation. C. Liao: Validation. P. Clark: Resources, project administration. Y. Errami: Resources, validation, project administration. S. Chen: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, project administration, writing-review and editing.

We thank Drs. Charles Fuchs and Roy Herbst for discussion. We thank Sarah Slavoff and Zhenkun Na for their assistance in protein work. We thank all members in Chen laboratory, as well as various colleagues in Department of Genetics, Systems Biology Institute, Cancer Systems Biology Center, MCGD Program, Immunobiology Program, BBS Program, Cancer Center, Stem Cell Center, Liver Center, RNA Center, and Center for Biomedical Data Sciences at Yale for assistance and/or discussion. We thank the Center for Genome Analysis, Center for Molecular Discovery, Pathology Tissue Services, Histology Services, Electron Microscopy, High Performance Computing Center, West Campus Analytical Chemistry Core, West Campus Imaging Core, and Keck Biotechnology Resource Laboratory at Yale for technical support. S. Chen is supported by the Yale SBI/Genetics Startup Fund, NIH/NCI/NIDA (DP2CA238295, 1R01CA231112, U54CA209992-8697, R33CA225498, RF1DA048811), Damon Runyon Dale Frey Award (DFS-13-15), Melanoma Research Alliance (412806, 16-003524), St. Baldrick's Foundation (426685), Breast Cancer Alliance, Cancer Research Insitute (CLIP), the 2017 AACR NextGen Grant for Transformative Cancer Research (grant number 17-20-01-CHEN), The Mary Kay Foundation (017-81), The V Foundation (V2017-022), Ludwig Family Foundation, DoD (W81XWH-17-1-0235 and W81XWH-20-1-0072), Sontag Foundation, and Chenevert Family Foundation. G. Wang is supported by the CRI Irvington and RJ Anderson Postdoctoral Fellowships. X. Dai is supported by the Charles H. Revson Senior Postdoctoral Fellowship. R.D. Chow is supported by the NIH/NCI (T32GM007205, F30CA250249). M.B. Dong is supported by the Yale MSTP training grant from the NIH (T32GM007205). P.A. Renauer is supported by a Yale PhD training grant from the NIH (T32GM007499) and the Lo Fellowship of Excellence of Stem Cell Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.