MC38, H2023, and HEK293T cells were obtained from either the National Cancer Institute (NCI) or the American Type Culture Collection (ATCC). Pmel-1 T cells used in in vitro assays and luciferase-expressing pmel-1 T cells used for in vivo studies were generated as previously described.20 Given that H2023 is a human lung cancer cell line expressing both MAGE-B2 and HLA-A0201, we generated T cells that can recognize HLA-A2-restricted MAGE-B2 as previously described.21 Additional information for cell culture condition, generation of tumor-reactive T cells and genetically modified tumor cell lines was provided in the online supplemental methods.

Pmel-1 TCR/Thy1.1 transgenic mice on a C57BL/6 background were a general gift provided by Dr Nicholas Restifo (NCI) and in-house bred. C57BL/6 mice, C57BL/6 albino mice and nude mice were purchased from the Charles River Frederick Research Model Facility. All mice were maintained in a specific pathogen-free barrier facility at either the University of Texas MD Anderson Cancer Center (MDACC) or the University of Houston. Mice were handled in accordance with protocols approved by Institutional Animal Care and Use Committees.

The mouse improved genome-wide knockout CRISPR Library version 2 consisting of 90,230 gRNAs targeted 18,424 genes was purchased from Addgene.22 Viral supernatants were generated as previously described. 1.5×10⁸ of MC38/gp100-Cas9 cells (MC38/GC cells) were transduced with pooled library lentivirus at low multiplicity of infection (MOI; ~0.15–0.2) and cultured with growth medium in the presence of 2 µg/mL puromycin to select gRNA-transduced cells. One week after selection, pooled gRNA-transduced tumor cells were seeded for overnight. For the T cell-treated groups, cultured Pmel T cells were added at an effector:target (E:T) ratio of 0.3:1 and 1:1 for 16 hours. For the non-T cell-treated groups, equal amount of T cell growth medium was added. After treatment, non-adhesive cells including T cells and dead tumor cells were removed by repeated wash using prewarmed phosphate-buffered saline (PBS). Additionally, tumor cells after 48-hour or 7-day puromycin selection were collected to determine the coverage of the genome-wide library in transduced cells before our functional immune screen. Triplication was performed for all treatment groups, and each sample includes at least 3×10⁷ of transduced tumor cells.

Genomic DNAs were extracted from tumor samples by Blood & Cell Culture DNA Maxi Kit (Qiagen). After genomic DNA isolation, gRNAs were amplified and barcoded with adaptation by nested PCR as previously described.22 The primers used for the PCR were listed in the online supplemental table 1. The PCR products were purified by using NucleoSpin Gel and PCR Clean-up Kit (Macherey-Nagel). Samples were then sequenced by Illumine NSG at MDACC Genomics Core.

Two independent assays by flow cytometry or real-time imaging were used to determine tumor sensitivity to T cell-mediated killing. For the flow cytometry-based cytotoxicity assay, tumor cells were mixed with paired tumor-reactive T cells at 37°C for 3 hours. The percentage of cleaved caspase-3⁺ cells in tumor cells was assessed by FACS Canto II (BD Biosciences). Additionally, we performed high-throughput time-lapse imaging microscopy in nanowell grids (TIMING) to real-time monitor T cell-mediated killing process at the single cell level as previously described.23 Detailed information for both assays was provided in the online supplemental methods.

Proteins were extracted by lysed tumor cells using RIPA lysis and extraction buffer (Thermo Fisher Scientific). The western blot analysis was used to determine the expression of total and phosphorated proteins. The intensity of protein bands was detected using ECL Plus (GE Healthcare Life Sciences) using the ChemiDoc Imaging System (Bio-Rad). The antibodies targeting Phospho-IκBα (Ser32) (clone 14D4, #9246), IκBα (clone 44D4, #4812), Phospho-RELA (Ser536, #3033), RELA (clone D14E12, #8242) and β-actin (8H10D10, #3700) were purchased from the Cell Signaling Technology and used for western blot analysis.

Additionally, 5×10⁵ of genetically modified MC38/gp100 tumors were subcutaneously injected into either nude or C57/BL6 mice. For the mice treated with immunomodulatory antibodies, 6 days after tumor inoculation, tumor-bearing mice were intraperitoneally administrated with either anti-PD-1 (BioLegend, clone 29F.1A12) or anti-OX40 (BioXcell, clone OX86) at the dose of 100 µg/dose, twice per week. Tumor sizes in all experimental mice were monitored by measuring the perpendicular diameters of the tumors every 3 days. To determine the in vivo effect of GSK2982772, a RIPK1-specific inhibitor, on the function of tumor-reactive CD8⁺ T cells, luciferase-expressing Pmel-1 T cells were transferred into C57BL/6 albino mice-bearing MC38/gp100 tumor as previously described.20 GSK2982772 (25 mg/kg; ChemieTek) suspended in vehicle (1% methylcellulose) was administered daily via oral gavage daily. All experiments were carried out in a blinded, randomized fashion.

Summary statistics (eg, mean, SEM) of the data are reported. Assessments of differences in continuous measurements between two groups were made using two-sample t-test posterior to data transformation (typically logarithmic, if necessary), or Mann-Whitney U test. Differences in tumor size among several treatments were evaluated using analysis of variance models with the Tukey’s multiple comparisons test. The log-rank test was used to assess the statistical significance of survival rate difference between two groups. A p value of less than 0.05 was considered significant. Graph generation statistical analyses were performed using the Prism software program (GraphPad Software), Tableau V.8.2 software program (Tableau Software), and R software programming language (V.3.1.0). The sample size for each experiment was chosen based on the study feasibility given its exploratory nature. Samples that did not meet proper experimental conditions were excluded from the analysis. Detailed information of bioinformatics analysis for the NGS results from the genome-wide gRNA screens and RNA-seq results in TCGA dataset was provided in the online supplemental Methods. The raw RNA-seq data in fastq format and calculated gene-level expression data have been deposited in the NCBI GEO database (GSE161496).

To functionally interrogate the contribution of tumor intrinsic pathways to tumor immune evasion, we developed a genome-wide CRISPR/Cas9-based immune screen using pooled gRNA libraries (figure 1A). This CRISPR/Cas9 immune screen employed a murine colon adenocarcinoma cell line MC38, expressing the melanoma antigen gp100 (MC38/gp100), paired with tumor-reactive murine Pmel T cells that can specifically recognize gp100 in an H2-D^b-restricted manner. This screen was based on the premise that the tumor cells expressing gene-specific gRNAs that target immune resistance genes would be significantly depleted on coincubation with tumor-reactive T cells, when compared with other gRNA-expressing cells (figure 1A). The MC38/gp100 cell line was first engineered to overexpress Cas9, which was confirmed by western blot analysis (online supplemental figure S1A). We then transduced Cas9-expressing MC38/gp100 (MC38/GC) cells with a murine genome-wide knockout CRISPR-gRNA library, which includes 90,230 sgRNAs targeting 18,424 genes.22 Infected tumor cells were then subjected to puromycin selection. To achieve a 50%–80% rate of T cell-mediated tumor killing, which is an optimal selection pressure for an in vitro dropout screen,7 we performed pilot studies to determine the cytotoxic effect of cultured Pmel T cells on MC38/GC tumor cells at different effector to target (E:T) ratios. Based on the results from the pilot experiment (online supplemental figure S1B), we selected the E:T ratios of 0.3:1 (ET0.3) and 1:1 (ET1) to represent conditions of weak and moderate T cell-selection pressure, respectively. The frequencies of gRNAs in pooled tumors among triplicate samples were highly correlated, thereby providing evidence of the consistency of our barcode sequencing results (online supplemental figure S1C). To remove gRNAs targeting genes that can broadly control cell proliferation and survival before our functional immune screens, pooled gRNA-expressing tumor cells were expanded in vitro for 7 days. Additionally, we referenced data from a previously reported genetic vulnerability screen24 to categorize the gRNA library into three subsets: essential genes (n=360), non-essential genes (n=927), and other genes (n=17 137). By comparing the distribution of the density of each gRNA group in pooled tumor samples, we observed that the overall distribution of the three gRNA groups in the reference samples was comparable. In contrast, a proportion of gRNAs in the essential group was significantly reduced in the samples after the 7-day expansion, but not in the non-essential and other groups (online supplemental figure S1D). This suggests that the 7-day expansion not only removed a series of gRNAs targeting key molecules for cell proliferation and survival, but also largely preserved the representation of a majority of gRNAs, particularly those in the non-essential and other groups, in pooled tumor cells before the immune screen.

Next, we determined the fold change (FC) in the frequency of each gRNA in tumor cells with and without T cell coincubation. Using an FC cut-off in gRNA frequency (Log₂FC<-1) and a statistically significant threshold of p<0.05, we found that 139 genes and 199 genes were significantly depleted in the ET0.3 and ET1 groups, respectively. Among them, 55 genes were identified under both conditions (figure 1B). As expected, more gRNAs in the essential group were significantly depleted in the samples under moderate selection pressure when compared with those under weak selection pressure (online supplemental figure S1E). We further performed gene ontology analysis using the list of depleted genes in the T cell-treated groups and identified several validated immunosuppressive pathways, such as PI3K-related pathways and the OXPHOS pathway (figure 1C).6 25 These results demonstrate that our optimized dropout screen can accurately recapitulate the contributions of tumor intrinsic factors to govern tumor sensitivity to T cell-killing. Importantly, our screen also led to the identification of novel immunological roles of a set of genes in traditional cancer signaling pathways, including molecules in the nuclear factor κB (NF-κB) pathway (Rela, Ikbkb, Ikbkg), tumor necrosis factor (TNF) pathway (Ripk1, Tnfaip3), and protein ubiquitination pathway (Psma1, Cul1) (figure 1D).

To evaluate the involvement of genes identified in our in vitro dropout screen in regulating tumor immune infiltration, we aligned these murine candidate genes with their human orthologs and found that 189 depleted candidates under moderate selection pressure (ET1) have matched human orthologs. Next, we extracted their RNA-seq-based expression profiles from the publicly available SKCM-TCGA dataset. Histologically reviewed lymphocyte infiltration scores (L scores) were also collected and used to determine immune infiltration at tumor sites. By using an L score cut-off of 3, we stratified patients with melanoma in the TCGA dataset into two groups, namely, the high lymphocyte infiltration (L score >3) and low lymphocyte infiltration (L score ≤3) groups. Among these candidates, 48 genes were significantly downregulated in melanomas with high lymphocyte infiltration (inflamed tumors) in comparison to those with poor lymphocyte infiltration (non-inflamed tumors) (figure 2A, table 1), underscoring their potential to modulate two rate-limiting steps of tumor immune response, namely, immune cell trafficking and tumor sensitivity to T cell-killing. These candidates were termed as dual immune resistance regulators (subsequently referred to as dual regulators). The remaining identified candidates whose expression levels were not associated with tumor immune infiltration were termed as cytotoxicity resistance regulators (subsequently referred to as cytotoxicity regulators). Additionally, 16 of the 48 genes were also identified under low T cell selection pressure (table 1). To confirm the distinct roles of two types of candidates in shaping TME, we selected one dual regulator and one cytotoxicity regulator that were top-ranked hits identified under both selection conditions to determine their association with a variety of immune gene expression signatures within tumor tissues. PRMT1, a member of the protein arginine N-methyltransferase family, was selected to represent dual regulators. As shown in figure 2B, increased PRMT1 expression was found in melanomas with poor immune cell infiltration. On the other hand, receptor‐interacting protein kinase 1 (RIPK1) whose expression levels are comparable between high and low immune infiltration groups (figure 2B), was chosen to represent cytotoxicity regulators. Given that the Cytolytic Activity Score (CYT) has been reported as a prognostic immune marker in patients with melanoma,16 we examined the correlation between the CYT and the expression of selected candidates, and found that the expression of PRMT1, but not RIPK1, was significantly higher in melanoma with low CYT (figure 2C). Besides melanoma, the analysis of pan-TCGA data including 33 types of cancers revealed that PRMT1 expression was also negatively correlated to CYT in patients with adrenocortical carcinoma (p=0.00204, r=−0.346), thyroid cancer (p=0.000104, r=−0.171) and sarcoma (p=0.0292, r=−0.135). In contrast, no meaningful negative correlation was observed between CYT and RIPK1 expression in tumor types across TCGA. Furthermore, we used the TCGA SKCM tumor purity calculated by previous studies26 27 to exclude the samples with tumor purity <0.5 from the correlation analysis to control for the confounding effect caused by poor tumor tissue purity. The negative correlation between PRMT1 expression and L score/CYT remains statistically significant in samples with relatively high tumor purity, whereas there is no significant correlation between RIPK1 expression and L score/CYT in the same set of tumors (online supplemental figure S2A,B). In addition to PRMT1, ALDOA, whose immunoregulatory role was reported in our previous study,7 was also identified as a dual regulator from the screen under moderate selection pressure. Similar to PRMT1, melanomas with high ALDOA expression displayed reduced L score and CYT (online supplemental figure S3A,B). Negative correlations between ALDOA and CYT were also found in a broad range of solid cancer types (online supplemental figure S3C). Moreover, we employed a computational deconvolution approach known as Cell-type Identification By Estimating Relative Subsets of RNA Transcripts (CIBERSORT)28 to characterize the cellular composition in the TME. Our analysis using the SKCM-TCGA dataset revealed a strong negative correlation between PRMT1 expression and the presence of CD8⁺ T cells in the TME (figure 2D). In addition to CD8⁺ T cells, tumors with high PRMT1 expression displayed low abundance of other types of antitumor effector cells, including CD4⁺ T helper cells, plasma cells, M1 macrophage and activated natural killer (NK) cells (figure 2D). However, the expression of RIPK1 only weakly correlates the presence of regulatory T cells and plasma cells in the TME (figure 2D). Taken together, integrating the clinical sample analysis with the genomic-wide genetic screen revealed a set of tumor intrinsic factors with the capability to control T cell-mediated tumor killing and/or shape the TME.

To validate the roles of PRMT1 and RIPK1 in regulating tumor response to T cell-mediated antitumor immune response, we knocked out the expression of PRMT1 or RIPK1 in both murine (MC38/GC) and human (H2023) tumor cells using two independent gRNAs and generated a set of tumor cell lines with either PRMT1 or RIPK1 knockout (figure 3A–D, upper panels). After coculturing with Pmel T cells, MC38/GC cells with stable expression of either Prmt1-specific or Ripk1-specific gRNAs were more sensitive to T cell-induced killing, which was determined by the percentage of tumor cells expressing cleaved caspase-3 (figure 3A,B, lower panels). Given that H2023 cells were derived from an HLA-A2 lung cancer patient expressing MAGE-B2, an identified tumor antigen, we generated human HLA-A2-restricted, MAGE-B2-specific TCR-T cells and evaluated the impact of PRMT1 and RIPK1 inhibition on susceptibility of H2023 to the apoptosis induced by these T cells. Consistent with murine tumor cells, knocking out of PRMT1 or RIPK1 significantly enhanced the apoptosis rate of H2023 cells after coincubation with paired tumor-reactive T cells (figure 3C,D, lower panels).

In addition to using flow cytometry-based cytotoxicity assays, we also leveraged our recently optimized live cell imaging method known as time-lapse imaging microscopy in nanowell grids23 to monitor T cell-induced killing of genetically modified individual tumor cells in real time. Equal numbers of MC38/GC tumor cells (red dye) and Pmel T cells (green dye) were labeled with lipophilic membrane dyes and co-incubated in the nanowell grids, and the induction of apoptosis was monitored using changes in morphology and annexin V staining. As shown in the representative images (figure 3E), the color and morphology of control tumor cells, which expressed a non-targeting gRNA (sgNC), remained unchanged within 3.75 hours after coincubation with T cells. However, annexin V-stained tumor cells were observed in Prmt1-specific gRNA-expressing tumor cells (sgPrmt1), within 3.75 hours after coincubation with T cells. Moreover, Ripk1-deficient tumor cells (sgRipk1) displayed morphological changes and became Annexin V staining positive as early as 2.5 hours after T cell treatment (figure 3E). These results confirmed that inhibition of Prmt1 or Ripk1 is sufficient to increase tumor sensitivity to T cell killing in vitro.

We further performed RNA-seq analysis to determine the transcriptomic profiles of sgPrmt1, sgRipk1 and sgNC tumor cells, and to identify the differentially expressed genes on Prmt1/Ripk1 knockout. By using gene set enrichment analysis (GSEA), we found that inhibiting Prmt1/Ripk1 can significantly alter the expression of genes in the apoptosis and IFN pathways, which could contribute to the enhanced sensitivity of sgPrmt1 and sgRipk1 tumor cells to T cell-mediated killing (online supplemental figure S4A–C). Although knocking out Prmt1 in MC38 tumor cells leads to major changes in the RNA level of multiple cytokines/chemokines, such as Vegfa, Ccl7 and Ccl9, no major changes in RNA level were observed for these cytokines/chemokines in sgRipk1 when compared with sgNC tumor cells (online supplemental figure S4D). Given that cytokines/chemokines are key immune regulators for immune cell trafficking, these results imply that PRMT1 and RIPK1 could play distinct roles in controlling tumor infiltration of immune cells.

Next, we examined the impact of genetic inhibition of Prmt1 and Ripk1 in murine tumor models. Although knocking out of Prmt1 or Ripk1 did not affect tumor growth in immune compromised mice (online supplemental figure S5), genetic inhibition of Prmt1 or Ripk1 expression in tumor cells moderately delayed their growth in immunocompetent mice (figure 3F,G), suggesting their extrinsic roles in regulating antitumor immune responses. We then challenged C57BL/6 mice with genetically modified MC38/GC tumors and tested whether inhibition of PRMT1 or RIPK1 could potentiate T cell-mediated antitumor responses in the in vivo cancer immunotherapy models described in online supplemental figure S6A. Anti-PD-1 and anti-OX40 treatments were selected to represent two distinct types of cancer immunotherapies, the antagonists of immune coinhibitory receptors and the agonists of immune costimulatory receptors, respectively. We found that the size of sgPrmt1 and sgRipk1 tumors were significantly shrunk after administration of anti-PD-1 antibody in relative to control tumors (figure 3F,G). Furthermore, anti-PD-1 treatment significantly extended the overall survival in the sgPrmt1 and sgRipk1 groups in related to the control group (figure 3H,I). Similar results were also obtained when sgPrmt1 and sgRipk1 tumors were treated with an OX40 agonist (online supplemental figure S6B,C). However, inhibition of PRMT1 or RIPK1 in tumors did not alter the survival of tumor-bearing mice in response to OX40 agonist treatment when compared with the control group (online supplemental figure S6D,E).

Previous studies have demonstrated that pharmaceutical inhibition of pathways with the capability to regulate both rate-limiting steps, such as the MAPK, PI3K, HSP90, and PRMT pathways, can synergize with the antitumor activity of ICB treatments.6 29–31 Therefore, here, we focused on validating the in vivo effects of the identified cytotoxicity regulators with bioactive compounds, such as RIPK1-specific inhibitors (RIPK1i), on T cell-mediated antitumor immune response, and evaluate the efficacy of combining RIPK1i with immunotherapy. GSK2982772 was recently developed as a first-in-class RIPK1i with an IC₅₀ value of 1 nM in an FP/ADP-Glo assay.32 As GSK2982772 prevents tumor necrosis factor (TNF)-induced necrotic cell death and blocks cytokine production in preclinical models,33 several phase IIa clinical studies have been conducted to test its therapeutic effects in patients with rheumatoid arthritis (NCT02858492), psoriasis (NCT02776033), or ulcerative colitis (NCT02903966). We first evaluated the target engagement of GSK2982772 at different concentrations based on the activation of the NF-κB pathway in response to TNFα stimulation. As shown in figure 4A, TNFα significantly increased the levels of p-IκBα and p-RELA, two markers for the activation of the NF-κB pathway, in MC38/gp100 tumor cells. In comparison to tumors receiving mock treatment, tumors pretreated with 5 µM or 10 µM GSK2982772 displayed significantly reduced levels of p-IκBα and p-RELA after TNFα stimulation (figure 4A), suggesting an efficient on-target effect of GSK2982772 starting at 5 µM. Compared with murine tumor cells, lower concentration of GSK2982772 pretreatment (ie, 2.5 µM) is sufficient to reduce the levels of p-IκBα and p-RELA in human H2023 tumor cells after TNFα stimulation (figure 4B). Our result is consistent with the previous report showing that GSK2982772 had a species selectivity.32 The in vitro cell growth assay showed that even a high concentration of GSK2982772 (10 µM) did not alter in vitro proliferation of both MC38/gp100 and H2023 tumor cells (figure 4C,D). However, GSK2982772 treatment significantly enhanced the apoptosis rate of MC38/gp100 and H2023 tumor cells after coincubation with their paired tumor-reactive T cells (figure 4E,F).

Next, we assessed the in vivo effect of GSK2982772 on T cell-mediated tumor killing and tumor infiltration of T cells using a murine ACT model (figure 5A). When we transferred luciferase-expressing Pmel T cells and used luciferase intensity at MC38/gp100 tumor sites to quantitate tumor infiltration of transferred T cells, we observed that the number of transferred T cells in mice treated with the RIPK1i was comparable to that in mice received mock treatment (figure 5B,C). As depicted in figure 5, RIPK1i treatment at a dose of 25 mg/kg potentiated the antitumor activity of transferred tumor-reactive T cells. These in vivo results further confirmed that the immunoregulatory role of tumor intrinsic RIPK1 is largely dependent on modulating the sensitivity of tumor cells to T cell killing, but not on altering the TME to facilitate T cell trafficking. Moreover, we examined antitumor effect of RIPK1i alone and in combination with immunomodulatory reagents (figure 5E). We found that RIPK1i alone moderately suppressed tumor growth in mice-bearing MC38/gp100 tumors (figure 5F). Interestingly, no significant synergistic antitumor effect was observed in mice treated with the combination of RIPK1i and anti-PD-1 (figure 5F). However, RIPK1i moderately enhanced the antitumor activity of anti-OX40 (figure 5G). Although the dual regulators tested in our previous studies displayed the potential to synergize with both immunomodulatory reagents, the results from RIPK1i imply that targeting cytotoxicity regulators on cancer immunotherapy can also be used to improve the efficacy of cancer immunotherapy.