Addressing Tumor Heterogeneity by Sensitizing Resistant Cancer Cells to T cell–Secreted Cytokines

Tumor heterogeneity is a major cause of therapy resistance. Genetic instability gives rise to diverse populations of tumor cells, and highly active therapies induce a strong selection pressure that frequently results in the outgrowth of resistant clones. Tumor cells can also shift between functional/transcriptional states, creating a second layer of heterogeneity (1–3). T cells serve as the central effector cells of cancer immunotherapies and have the ability to kill tumor cells following recognition of tumor peptides presented by MHC-I proteins. Recognition of such MHC-I–peptide complexes induces T-cell receptor (TCR) activation, triggering the targeted release of cytotoxic granules containing perforin and granzymes (4). Cytotoxic T cells exert a powerful selection pressure that can result in the outgrowth of tumor cells that have downregulated or lost MHC-I expression. Loss of MHC class I expression is common in some cancer types. For example, in diffuse large B-cell lymphoma (DLBCL), 50% of tumors were found to lack surface MHC-I expression (5). Somatic inactivation of B2M or HLA-I loci was found in 80% of MHC-I–negative cases. Also, a thorough analysis of non–small cell lung cancer demonstrated loss of heterozygosity among MHC class I genes in 40% of tumors. Allele-specific HLA loss was associated with high neoantigen burden and upregulation of cytolytic activity and PD-L1 expression, consistent with immune evasion (6). TCR recognition of tumor cells by CD8 T cells is dependent on continued MHC-I protein expression, and a single mutation, for example in the B2M gene, can render tumor cells invisible to the TCR (7).

Here, we investigated which alternative pathways could enable targeting of MHC-I–deficient tumor cells by activated T cells. Engagement of such alternative pathways could also be relevant for addressing the general issue of tumor heterogeneity. In a genetic screen, we observed that CD8 T cells not only killed MHC-I expressing tumor cells but also significantly reduced the number of MHC-I–deficient tumor cells added as negative controls to such cocultures (8). This finding could be explained by expression of cell-surface proteins (such as Fas ligand) or cytokines (such as IFNγ and TNFα) by activated CD8 T cells. Fas ligand is expressed by activated T cells and can induce apoptosis by Fas-expressing target cells (9). IFNγ plays an important role in tumor immunity because it induces upregulation of many genes in the MHC-I antigen presentation pathway. It has also been shown to exert an antiproliferative effect on tumor cells (10). TNFα typically promotes tumor cell survival through NF-κB activation but can alternatively induce cell death, dependent on a complex signal transduction pathway (11). The membrane-tethered TNF receptor signaling complex (Complex I) induces NF-κB activation, but alterations in signaling can result in the dissociation of RIPK1 and assembly of a proapoptotic complex composed of FADD, RIPK1, and caspase-8 (Complex II; ref. 12).

We therefore performed a genome-scale genetic screen in MHC-I–deficient tumor cells that were cocultured with MHC-I–expressing tumor cells and CD8 T cells. Remarkably, we discovered that the inactivation of many genes resulted in depletion of MHC-I–deficient tumor cells in the presence of activated T cells. A genetic dependency analysis demonstrated that depletion of MHC-I–deficient tumor cells was dependent on the receptors for TNFα, IFNγ, and type 1 interferons. Inactivation of both Rnf31 (TNF signaling) and Atg5 (autophagy) efficiently sensitized MHC-I–deficient tumor cells to apoptosis by major T cell–derived cytokines. We established the in vivo significance of these findings by demonstrating that tumors with a substantial population of MHC-I–deficient cells could be controlled by CD8 T cells. Furthermore, small-molecule inhibitors targeting these pathways strongly sensitized tumor cells to apoptosis by IFNγ and TNFα. Importantly, such apoptotic tumor cells were efficiently cross-presented by cross-presenting dendritic cells (cDC1) to CD8 T cells, explaining greatly increased tumor infiltration by T cells that expressed TNFα and IFNγ. Targeting of these pathways thus resulted in apoptosis of MHC-I–deficient tumor cells by T cell–derived cytokines, enhanced cross-presentation of such apoptotic tumor cells by cDC1s, and increased tumor infiltration by TNFα- and IFNγ-secreting T cells.

Coculture of Ova-specific OT-I CD8 T cells with a mixture of B16-Ova and B2m^−/− B16 tumor cells showed not only the expected depletion of B16-Ova cells by OT-I T cells but also a substantial reduction in the number of surviving MHC-I–deficient (B2m^−/−) tumor cells (Supplementary Fig. S1A). We therefore investigated which pathways could sensitize MHC-I–deficient tumor cells to apoptosis in the presence of activated CD8 T cells. We inactivated the B2m gene by CRISPR/Cas9 in a melanoma B16F10-Cas9 clone with a high Cas9 editing efficiency. B2m^−/− B16F10-Cas9 cells were then transduced with a genome-scale guide RNA (gRNA) library in a lentiviral vector. Edited tumor cells were mixed with B16-Ova cells at a 1:1 ratio and selected by 1-day coculture with OT-I T cells; a control population was maintained in the absence of OT-I T cells. The representation of all gRNAs in B2m^−/− B16 tumor cells was then quantified by Illumina sequencing of the gRNA cassette (Fig. 1A). The specificity of gRNA enrichment or depletion was determined by comparing gRNA representation in tumor cells in the presence or absence of OT-I T cells.

Key genes in the IFNγ and type I interferon signaling pathways were identified among the enriched gRNAs in the screen (Fig. 1B; Supplementary Table S1), including Jak1, Jak2, Stat1, Ifngr2, Ifngr1, Ifnar1, Ifnar2, and Irf9, indicating that inactivation of these cytokine signaling pathways reduced killing of B2m^−/− B16 tumor cells in the presence of activated T cells. gRNAs for Pvr were also enriched in the screen. PVR (CD155) is expressed on the surface of tumor cells and binds to both activating (CD226) and inhibitory (TIGIT, CD96) receptors on T cells and NK cells (13).

Analysis of depleted gRNAs revealed a number of resistance pathways to MHC-I–independent T-cell killing (Fig. 1C). We identified resistance genes in innate sensing pathways (Adar), regulators of interferon signaling (Ptpn2, Socs1, and Irf1), the ubiquitin-proteasome system (Nedd8, Ube2n, and Nploc4), autophagy (Vps4b, Atg5, Atg7, Ei24, and Atg16l1), the TNFα signaling pathway (Rnf31, Map3k7, Golt1b, Traf2, Otulin, and Ripk1), and the glycosylphosphatidylinositol (GPI) anchor attachment pathway (Gpaa1, Pigt, Pigu, Pigs, and Pigk). Notably, genes encoding all five components of the GPI transamidase complex, which catalyzes the final step of GPI anchor attachment to proteins, were depleted in the screen, providing evidence that this complex conferred resistance to MHC-I–independent T-cell killing.

We next investigated which resistance pathways were relevant for T cell–mediated killing of both MHC-I–expressing and MHC-I–deficient tumor cells because targeting of such pathways could sensitize both tumor cell populations. We compared depleted gRNAs in the MHC-I–independent T-cell killing screen reported here against a screen with MHC-I–expressing B16-Ova tumor cells that we reported previously (Fig. 1D; Supplementary Fig. S1B; ref. 8); a number of these genes were also previously identified by other groups in T-cell coculture screens and in vivo screens with MHC-I–positive tumor cells (Fig. 1E; refs. 14–18). We identified 95 genes for which gRNAs were depleted in both B2m-knockout (KO) and B2m-wild-type (WT) in screens with B16F10 tumor cells. Pathway enrichment analysis of these genes highlighted three major resistance pathways: attachment of GPI anchor to proteins, regulation of TNF signaling, and autophagy (Fig. 1F). Analysis of gRNAs enriched in both B2m-KO and B2m-WT screens identified the IFNγ signaling pathway (Supplementary Fig. S1C and S1D). The finding that many genes from previous B2m-WT screens also scored in the B2m-KO screen indicates that the effect of these gene KOs is not dependent on direct tumor cell killing through the perforin–granzyme pathway but rather involves alternative mechanisms as discussed in detail below.

We hypothesized that several of the discovered resistance pathways could be functionally connected and potentially provide opportunities for synergistic targeting of B2m-KO tumor cells. We therefore performed a genetic codependency screen in which we individually inactivated 10 genes in B16 B2m^−/− Cas9 cells that could be critical for T cell–mediated killing in a perforin/granzyme-independent manner, including Ifngr1, Ifnar1, Tnfrsf1a (cytokine signaling), Cgas, Mavs (innate immune signaling), Pvr, Fas (surface receptors), Gne (glycosylation), Mlkl, and Casp3 (cell death pathways; Fig. 2A; Supplementary Fig. S2A–S2E). Each of these gene-KO cell lines and the B16 B2m^−/− Cas9 cell line edited with a control gRNA were then transduced with a minipool gRNA library representing the identified genes from the primary screen (205 and 20 genes for depleted and enriched gRNAs, respectively, as well as 20 TNF receptor superfamily genes). Edited tumor cells were mixed at a 1:1 ratio with B16-Ova cells and selected for 48 hours with Ova-specific OT-I T cells. The representation of all gRNAs in the minipool library was then compared among the entire panel of KO cell lines (Fig. 2B; Supplementary Table S2).

This strategy revealed that TNF signaling genes (Rnf31 and Otulin) and an autophagy gene (Vps4b) were dependent on Ifngr1, Ifnar1, and Tnfrsf1a (Fig. 2B–D) because gRNA depletion for these genes was attenuated in Ifngr1-, Ifnar1-, and Tnfrsf1a-KO cell lines compared with the control-KO cell line. Protein–protein interaction network analysis further suggested connections between IFN signaling, TNF signaling, and autophagy modules (Fig. 2C), consistent with the genetic dependency analysis. Focused analysis of genes representing these major pathways showed that gRNA depletion of autophagy genes (Vps4b, Atg5, Atg10, and Ei24) and expectedly also TNF signaling genes (Rnf31, Otulin, Cflar, Ripk1, and Traf2) was attenuated in Tnfrsf1a-KO compared with control-KO cells. Depletion of gRNAs for several of these TNF signaling and autophagy genes was also attenuated in Ifngr1-KO and Ifnar1-KO compared with control-KO cells (Fig. 2D). These data indicated that cytokine signaling and autophagy pathways were relevant for MHC-I–independent killing of B2m-KO tumor cells in the presence of activated T cells.

We selected Rnf31 and Atg5 as representative members of the TNF signaling and autophagy pathways because gRNAs targeting these genes were strongly depleted in both primary and validation screens (Figs. 1C and E, and 2B and D). Rnf31 encodes the catalytic subunit of the LUBAC complex, which attaches linear ubiquitin chains to key proteins of the TNF receptor signaling complex, including RIPK1 and NEMO; this catalytic function of the LUBAC complex is important for prosurvival signaling induced by TNF (19–21). Atg5 encodes a core member of the autophagy pathway (22).

We developed a coculture assay in which a 1:1 mixture of B16-Ova (ZsGreen) and B2m^−/− B16 (mCherry) cells were cocultured with OT-I T cells. Genes of interest were selectively inactivated in B2m^−/− B16 cells to evaluate MHC-I–independent killing in the presence of activated T cells. We quantified these results as (i) the ratio of surviving B2m^−/− experimental-KO to B16-Ova tumor cells and (ii) the number of surviving B2m^−/− experimental-KO relative to B2m^−/− control-KO cells (Fig. 3A and B; Supplementary Fig. S3A–S3C). Inactivation of either Rnf31 or Atg5 genes reduced the number of surviving B2m^−/− cells in the presence of activated T cells (Fig. 3B and C). In contrast, the ratio of experimental-KO to control-KO tumor cells remained constant in the absence of T cells (Fig. 3B). We also evaluated Pigs, which encodes a component of the GPI anchor attachment complex (Fig. 3C; Supplementary Fig. S3D). Although we were able to confirm the findings from the primary screen, Pigs KO showed less synergy with Atg5 KO compared with Rnf31 (Fig. 3C). These data are consistent with previous studies demonstrating that inactivation of several genes in the TNFR1 signaling pathway could sensitize MHC-I–expressing tumor cells to TNFα and the identification of autophagy genes in previous tumor–T cell coculture screens (14–16).

We observed a striking synergistic effect when we inactivated both Rnf31 and Atg5 in B2m^−/− B16F10 cells as well as Py8119 and MC38 tumor cells that represent models of triple-negative breast cancer (TNBC) and colon cancer, respectively (Fig. 3B–F; Supplementary Fig. S3E–S3G). Supernatants from activated T cells greatly reduced the number of live Rnf31-KO cells and Rnf31/Atg5-double KO (dKO) compared with control-KO tumor cells (Fig. 3G), consistent with a major role of secreted molecules in MHC-I–independent T-cell killing. The genetic dependency screen indicated that the effect of Rnf31 deficiency was dependent on the receptors for IFNγ (Ifngr1 KO) and TNFα (Tnfrsf1a KO). We used three complementary approaches to further investigate the contribution of both IFNγ and TNFα to killing of Rnf31/Atg5-dKO tumor cells: (i) treatment of tumor cells with recombinant cytokines, (ii) antibody-mediated neutralization of one or both cytokines in cocultures with T cells, or (iii) inactivation of one or both cytokine receptor genes in B2m^−/− tumor cells (Fig. 3H–J). Treatment with IFNγ plus TNFα for 72 hours greatly diminished the number of live Rnf31/Atg5-dKO compared with control-KO B2m^−/− tumor cells. This cytokine combination also reduced the number of surviving single-KO tumor cells, but the effect was substantially weaker compared with Rnf31/Atg5-dKO cells (Fig. 3H). Consistent with these results, neutralization of either IFNγ or TNFα substantially reversed the sensitization effect of Rnf31/Atg5-dKO in B2m^−/− cells in a coculture assay with B16-Ova and OT-I T cells (Fig. 3I). Inactivation of IFNGR (Ifngr1 KO) or TNFR1 (Tnfrsf1a KO) in B2m^−/− tumor cells reversed the sensitization effect of Rnf31/Atg5 deficiency when these B2m^−/− tumor cells were cocultured with B16-Ova tumor cells and OT-I T cells (Fig. 3J; Supplementary Fig. S3H). These data demonstrated that the inactivation of Rnf31 and Atg5 greatly sensitized B2m^−/− tumor cells to T cell–derived cytokines and that effective depletion of these tumor cells involved both IFNγ and TNFα rather than merely TNFα.

We also investigated a potential association of TNFα and autophagy pathways with survival of patients with melanoma. RNA-sequencing (RNA-seq) data from B16 control-KO, Atg5-KO, and Rnf31/Atg5-dKO cells (Supplementary Fig. S4) were used to define the Atg5-KO and Rnf31/Atg5-dKO gene signature based on the top 200 upregulated and 200 downregulated genes (Supplementary Table S3). Both Atg5-KO and Rnf31/Atg5-dKO signatures were associated with longer survival in a melanoma dataset from The Cancer Genome Atlas as well as several melanoma datasets with immune-checkpoint blockade treatment (Fig. 3K). Such an association was observed in tumors with low or high MHC-I expression (Supplementary Fig. S5A). Furthermore, this effect was maintained even when interferon-inducible and cytokine genes were excluded from these signatures (Supplementary Fig. S5B).

We next investigated which pathways were responsible for cytokine-mediated cell death of Rnf31/Atg5-deficient B2m^−/− tumor cells. Addition of an apoptosis inhibitor, Z-VAD-FMK, substantially reversed the effect of Rnf31/Atg5 deficiency, whereas a necroptosis inhibitor, necrostatin-1, had little effect (Fig. 4A). Consistent with a minor role of necroptosis, inactivation of the gene encoding a major necroptosis mediator, Mlkl, only slightly reversed the effect of Rnf31/Atg5 deficiency (Supplementary Fig. S6A). TNF signaling can shift from membrane-tethered Complex I (NF-κB activation, prosurvival) to cytosolic Complex II (caspase-8 activation), which determines whether TNFα signaling induces cell survival or apoptosis (12). Key proteins of Complex II are caspase-8, an initiator caspase, as well as RIPK1 and FADD. Activated caspase-8 was indeed detected in Rnf31-KO or Rnf31/Atg5-dKO tumor cells following treatment with activated T-cell supernatant or recombinant TNFα (Fig. 4B and C). Notably, there was no activation of caspase-8 when control-KO tumor cells were treated with activated T-cell supernatant or recombinant cytokines. Coculture of Rnf31/Atg5-dKO tumor cells with B16-Ova and OT-I T cells resulted in the accumulation of a substantial percentage of tumor cells positive for both caspase-8 and caspase-3 (Fig. 4D). Western blot analysis revealed that activated, cleaved caspase-8 strongly accumulated in Rnf31/Atg5-dKO and to a lesser extent in Rnf31-KO and Atg5-KO tumor cells during 24-hour treatment with TNFα, whereas activated caspase-8 was undetectable in TNFα-treated control-KO tumor cells (Fig. 4E). The critical role of caspase-8 was underscored by inactivation of Casp8, which reversed the sensitizing effect of Rnf31/Atg5-dKO regardless of whether edited tumor cells were cocultured with T cells or treated with TNFα (Fig. 4F and G). Complex II represents a high-molecular-weight complex organized as a helical array of repeating copies of RIPK1 and caspase-8 (23). We therefore hypothesized that autophagy was critical for the timely removal of this large cytosolic protein complex. Confocal microscopy demonstrated that cleaved caspase-8 and LC3B were colocalized in Rnf31-KO cells, consistent with the hypothesis that activated caspase-8, along with other components of Complex II, was degraded by autophagy (Fig. 4H; Supplementary Fig. S6B). In addition, RNA-seq and RT-qPCR analyses revealed that Atg5-KO cells expressed higher levels of interferon-stimulated genes, including Isg15 and Oas1, compared with control-KO cells; the expression of interferon-stimulated genes was augmented by TNFα and IFNγ, consistent with constitutive activation of interferon signaling in autophagy-deficient tumor cells (Supplementary Figs. S4 and S6C). These findings demonstrated how targeting of the TNF signaling and autophagy pathways sensitized tumor cells to cell death by T cell–derived cytokines.

Next, we investigated whether a tumor with a substantial fraction of resistant cells (B2m^−/−) could be eliminated by T cells when Rnf31 and/or Atg5 genes were inactivated in the B2m^−/− population. To mimic a clinically relevant setting in which a cancer is composed of both MHC-I–expressing and –deficient tumor cells, B16-Ova cells and B16 B2m^−/− experimental-KO cells were mixed at a 4:1 ratio and inoculated subcutaneously. When tumors were palpable (day 7), activated OT-I T cells were transferred and treatment with a PD-1 antibody was initiated (Fig. 5A). In this experiment, B16-Ova cells were killed by transferred OT-I T cells in an antigen-specific manner, thus exposing neighboring B2m^−/− experimental-KO cells to a cytokine gradient formed by activated OT-I T cells. As expected, tumors composed entirely of B16-Ova cells were eliminated by this treatment regimen. However, tumors formed by a mixture of B16-Ova and B2m^−/− control-KO cells grew aggressively (Fig. 5B–D). Tumors formed by a mixture of B16-Ova and B2m^−/−Rnf31-KO cells showed significantly slower tumor growth compared with a mixed tumor containing B2m^−/− control-KO cells, resulting in survival of 33% of mice at day 50 (Fig. 5D). Aggressively growing tumors were entirely composed of B2m^−/− cells, confirming the central role of these cells in conferring therapy resistance (Fig. 5E). Strikingly, all tumors formed by a mixture of B16-Ova and B2m^−/−Rnf31/Atg5-dKO cells were eradicated by OT-I and anti–PD-1 antibody treatment and did not recur, resulting in the survival of 100% of mice by day 50 (Fig. 5B–D).

Adoptive T-cell therapy without PD-1 blockade also resulted in a substantial survival benefit against tumors composed of a mixture of B16-Ova and B2m^−/−Rnf31/Atg5-dKO cells, but not against tumors containing B16-Ova and B2m^−/− cells. Inclusion of a PD-1 mAb resulted in additional therapeutic benefit. Without PD-1 treatment, tumors recurred in ∼50% of mice at a late time point, even for B16-Ova tumors that did not contain a B2m^−/− population. In contrast, all mice treated with OT-I T cells and a PD-1 mAb were tumor-free at the end of the experiment, even if they had been implanted with tumors composed of B16-Ova and B2m^−/−Rnf31/Atg5-dKO cells (Supplementary Fig. S7A–S7C). Tumors that recurred in B16-Ova–inoculated mice following OT-I T-cell therapy lacked ZsGreen expression, a marker of the Ova expression cassette (Supplementary Fig. S7D).

To further evaluate these in vivo findings, we tested Py8119 tumors, a model of TNBC. Py8119-Ova cells successfully engrafted and formed palpable tumors with a maximum size at ∼day 7, but spontaneously began to regress around day 10 (Fig. 5F). CD8 T-cell depletion demonstrated that regression of Py8119-Ova tumors was T-cell dependent (Supplementary Fig. S8A); T cell–mediated rejection is likely explained by the introduced Ova antigen because Py8119 cells without Ova expression form aggressive tumors with limited responsiveness to immune-checkpoint blockade (24). Killing of Py8119-Ova tumor cells by endogenous CD8 T cells provided a unique opportunity to evaluate genes of interest in B2m^−/− cells without other interventions such as T-cell transfer or PD-1 antibody treatment. As expected, tumors formed by a 4:1 mixture of Py8119-Ova and B2m^−/− control-KO cells grew aggressively, which was also not slowed by inactivation of Atg5 in the B2m^−/− population (Fig. 5F and G). Mice with mixed tumors containing B2m^−/−Rnf31-KO cells showed slower tumor growth and survived longer than mice with mixed tumors containing B2m^−/− control-KO cells. Importantly, growth of mixed tumors containing a B2m^−/−Rnf31/Atg5-dKO population was greatly reduced, resulting in the survival of 87.5% of mice by day 50 (Fig. 5F and G).

We also used a B2m^+/+ tumor setting to study the effects of Rnf31 and Atg5 deficiency on tumor growth and survival (Supplementary Fig. S8B–S8D). B16 B2m^+/+ control-KO cells or B16 B2m^+/+ experimental-KO cells were injected subcutaneously, and mice were treated with isotype control mAb, CD8 depleting mAb, or PD-1 mAb. In the isotype control group, Rnf31-KO and Rnf31/Atg5-dKO tumors showed significantly slower growth compared with control-KO tumors. PD-1 antibody treatment further slowed the growth of both Rnf31-KO and Rnf31/Atg5-dKO tumors. When CD8 T cells were depleted, all four groups (control-KO, Rnf31-KO, Atg5-KO, and Rnf31/Atg5-dKO tumors) showed similar tumor growth kinetics, indicating that slowed growth of Rnf31-KO and Rnf31/Atg5-dKO tumors was dependent on CD8 T cells. Both Rnf31-KO and Rnf31/Atg5-dKO groups showed a survival benefit in isotype control and PD-1 antibody treatment groups compared with control-KO tumors. PD-1 antibody treatment extended survival in the Rnf31/Atg5-dKO group, with 57.1% of mice alive at day 50. These data demonstrated that targeting of Rnf31 and Atg5 enabled T cell–mediated immunity against tumors containing a substantial population of resistant cells with a B2m mutation.

We next investigated how the tumor microenvironment was modified by inactivation of Rnf31 and/or Atg5 genes in MHC-I–expressing tumor cells. B16 experimental-KO tumor cells (5 × 10⁵) were implanted subcutaneously, and on day 14, infiltrating immune cells were analyzed by flow cytometry. The number of infiltrating CD8 T cells and CD4 T cells was significantly increased in Rnf31/Atg5-dKO compared with ­control-KO tumors (Supplementary Fig. S9A and S9B). Particularly notable was a major increase in the number of IFNγ- and TNFα-producing CD4 and CD8 T cells (Fig. 6A; Supplementary Fig. S9C–S9H). Similarly, the total number of NK cells, as well as the number of NK cells secreting IFNγ or TNFα, was significantly increased in Rnf31/Atg5-dKO compared with control-KO tumors (Supplementary Fig. S9I–S9K). These data indicated that inactivation of Rnf31 and Atg5 genes not only sensitized tumor cells to IFNγ and TNFα secreted by activated T cells but also substantially increased tumor infiltration by T cells and NK cells that secreted these cytokines.

We next investigated which immunologic mechanisms could explain the enhanced infiltration of MHC-I–expressing Rnf31/Atg5-dKO tumors by IFNγ- and TNFα-producing T cells. Analysis of myeloid cell populations revealed an increased ratio of cDC1s to cDC2s in Rnf31/Atg5-dKO compared with ctrl-KO tumors (Supplementary Fig. S9L and S9M). This finding is relevant because cDC1s play an essential role in presenting antigens from apoptotic tumor cells to CD8 T cells and also CD4 T cells (25).

We hypothesized that Rnf31- and/or Atg5-deficient tumor cells may be more efficiently cross-presented to CD8 T cells because exposure to T cell–derived cytokines would generate apoptotic fragments that could be efficiently taken up by cDC1s. We set up a cross-presentation assay in which apoptosis was induced in B2m^−/− experimental-KO tumor cells by T-cell cytokines (TNFα, 10 ng/mL, or IFNγ, 50 ng/mL). Importantly, tumor cells were then extensively washed to prevent carryover of TNFα or IFNγ to cocultures with cDC1s. Treated tumor cells were then cocultured with freshly isolated XCR1⁺ cDC1s, followed by coculture of cDC1s with CFSE-labeled OT-I T cells (Fig. 6B). Analysis of tumor cells by annexin V staining (prior to incubation with cDC1s) showed that IFNγ and to a lesser extent TNFα efficiently induced tumor cell apoptosis, in particular in Rnf31/Atg5-dKO tumor cells but also in Rnf31-KO cells (Fig. 6C). Consistent with these findings, IFNγ- and TNFα-pretreated Rnf31/Atg5-dKO tumor cells were efficiently cross-presented by cDC1s as measured by proliferation (CFSE dilution) of OT-I T cells (Fig. 6D). These cytokines also enhanced cross-presentation of Rnf31-KO tumor cells, yet to a lesser extent. Collectively, these results demonstrated that Rnf31/Atg5 deficiency sensitized B2m^−/− tumor cells to T cell–derived cytokines and induced their apoptosis. Apoptotic Rnf31/Atg5-dKO cells were efficiently cross-presented by cDC1s to CD8 T cells, resulting in greater tumor infiltration by T cells secreting IFNγ and TNFα. These cellular interactions thus formed a positive feedback loop: Apoptosis of B2m^−/−Rnf31/Atg5-dKO cells by T cell–derived cytokines enhanced cross-presentation of antigens from apoptotic tumor cells to CD8 T cells, resulting in greater tumor infiltration by effector T cells secreting IFNγ and TNFα.

Next, we investigated whether these pathways could be targeted pharmacologically. cIAP1 and cIAP2 (cIAP1/2) play a critical role as direct upstream regulators of RNF31 in the TNFR1 signaling pathway. cIAP1/2 recruit LUBAC, a complex composed of RNF31 (HOIP), HOIL-1, and SHARPIN (21). We therefore tested whether a cIAP inhibitor (birinapant) had a similar sensitizing effect on B2m^−/− tumor cells as Rnf31 deficiency (26). VPS34 is a class III phosphoinositide 3 kinase (PIK3C3) that plays a central role in autophagy, and a high-affinity small-molecule inhibitor has been developed (SAR405; ref. 27). Proliferation or survival of tumor cells was not significantly affected by the cIAP inhibitor (500 nmol/L), the VPS34 inhibitor (500 nmol/L), or the combination of both inhibitors (500 nmol/L each). However, both individual inhibitors and in particular the combination of cIAP and VPS34 inhibitors reduced the number of surviving B2m^−/− tumor cells following incubation with activated T-cell supernatant (Fig. 7A). In an MHC-I–independent T-cell killing assay, the combination of cIAP and VPS34 inhibitors greatly decreased the number of surviving B2m^−/− tumor cells (Fig. 7B). Also, the combination of cIAP and VPS34 inhibitors substantially increased the percentage of B16F10 or Py8119 tumor cells with activated caspase-3 following treatment with TNFα or the combination of TNFα and IFNγ (Fig. 7C and D). Consistent with these findings, the combination of cIAP and VPS34 inhibitors substantially diminished the number of surviving B2m^−/− tumor cells following treatment with the combination of TNFα and IFNγ (Fig. 7E; Supplementary Fig. S10A and S10B).

We next investigated whether the combination of cIAP and VPS34 inhibitors could treat tumors containing a resistant population of B2m^−/− cells. Py8119-Ova and Py-B2m^−/− tumor cells were mixed at a 4:1 ratio, and 5 × 10⁵ cells were orthotopically injected into the mammary fat pad. On day 6 when tumors were palpable, mice were treated daily with cIAP inhibitor (10 mg/kg daily by i.p. injection), VPS34 inhibitor (10 mg/kg daily by oral gavage), or the combination of both drugs. Treatment with single agents (cIAP inhibitor or VPS34 inhibitor) resulted in modest but statistically significant slowing of tumor growth. Notably, treatment with the combination of cIAP and VPS34 inhibitors slowed tumor growth compared with solvent and single-inhibitor treatment groups (Fig. 7F). Mice treated with the combination of cIAP and VPS34 inhibitors survived significantly longer than mice treated with solvent or single agents (Supplementary Fig. S10C). Histopathologic analysis showed no evidence of toxicity or immune infiltration in major organs of non–tumor bearing mice treated for 3 weeks with the same regimen (Fig. 7G; Supplementary Fig. S11A). Also, there were no differences in the weight of mice across the four treatment groups. A human melanoma cell line, A375, also showed reduced survival when treated with individual inhibitors and in particular the combination of cIAP and VPS34 inhibitors following incubation with TNFα (Supplementary Fig. S11B). Treatment with IFNγ plus TNFα for 72 hours significantly diminished the number of live RNF31/ATG5-dKO compared with parental A375 tumor cells (Supplementary Fig. S11C). These data demonstrated that TNF signaling and autophagy pathways could be targeted pharmacologically to sensitize B2m^−/− tumor cells to T cell–derived cytokines and to slow the growth of tumors containing a substantial population of resistant B2m-deficient cells. Currently available pharmacologic inhibitors may not be optimal for this purpose, but this work may motivate the development of improved inhibitors for cytokine-mediated elimination of resistant tumor cells.

MHC-I–deficient tumor cells cannot be detected by the TCR of cytotoxic CD8 T cells, but surprisingly, inactivation of many different genes enabled depletion of such resistant tumor cells in the presence of activated T cells. Genetic interaction screens and functional experiments highlighted the role of T cell–secreted cytokines in MHC-I–independent killing of tumor cells, in particular IFNγ and TNFα. Inactivation of both Rnf31 (TNF signaling) and Atg5 (autophagy) genes strongly sensitized MHC-I–deficient tumor cells to apoptosis induced by T cell–secreted cytokines. Tumors containing both MHC-I–expressing and MHC-I–deficient tumor cells could be controlled by T cells when Rnf31 and Atg5 were inactivated in the MHC-I–deficient tumor cell population. This cytokine-mediated apoptosis pathway could thus inhibit the outgrowth of tumor cell clones resistant to cytotoxicity triggered by TCR recognition of tumor peptides presented by MHC-I.

Why does inhibition of autophagy induce tumor cell death by TNFα and IFNγ secreted by activated T cells? Inactivation of the Rnf31 gene or inhibition of cIAPs by a small-molecule inhibitor shifts TNFα signaling from a prosurvival to a cell death pathway and sensitizes tumor cells to TNFα-induced apoptosis (28–31). Rnf31 encodes HOIP, the catalytic subunit of the LUBAC complex that attaches linear ubiquitin chains to RIPK1 and IKKγ, a key step in prosurvival signaling through NF-κB (19, 20, 32). IFNγ induces the expression of RIPK1, which augments TNFα-mediated survival signaling (33). In the absence of LUBAC-mediated ubiquitination, RIPK1 (which is increased by IFNγ) dissociates from the membrane-tethered TNF receptor (Complex I), and instead forms a soluble, cytosolic cell death complex with FADD and caspase-8 (Complex II; ref. 12). Importantly, Complex II is a high-molecular-weight, soluble cytosolic complex (12), and a cryogenic electron microscopy structure has shown that it contains many copies of caspase-8 assembled along a helical axis (23). Western blot analysis demonstrated that activated caspase-8 strongly accumulated in Rnf31/Atg5-deficient tumor cells, whereas it was undetectable in control-KO cells. We also observed colocalization of activated caspase-8 with the LC3B, supporting our hypothesis that this cell death complex needs to be cleared by autophagy.

Inhibition of autophagy also enhanced the effects of IFNγ, and the combination of both TNFα and IFNγ signaling was particularly effective in inducing apoptosis of Rnf31/Atg5-deficient or cIAP + Vps34 inhibitor–treated tumor cells. RT-qPCR and RNA-seq analysis demonstrated that Atg5-KO tumor cells showed constitutive expression of interferon-stimulated genes, and their expression was enhanced following IFNγ or TNFα treatment. The cGAS–STING pathway detects cytosolic double-stranded DNAs and strongly induces a type 1 interferon response in tumor cells and virus-infected cells (34). Interestingly, STING signaling induces autophagy, which facilitates clearance of cytosolic viral particles (35). This mechanism also induces clearance of cytosolic micronuclei, a major substrate for cGAS activation in tumor cells (36). These findings indicate that autophagy attenuates interferon signaling by clearing cytosolic stimulators of the cGAS–STING pathway. Although previous studies have emphasized sensitization of MHC-I–expressing tumor cells to TNFα based on the well-established balance of Complex I versus Complex II signaling (12, 14, 15), we demonstrate that efficient MHC-I–independent killing of B2m^−/− tumor cells requires the action of both IFNγ and TNFα. Autophagy has also been reported to cause internalization and lysosomal degradation of MHC-I molecules in pancreatic cancer, and inhibition of autophagy can therefore also enhance T cell–mediated immunity against MHC-I–expressing tumor cells (37). A recent study demonstrated synergistic activity of Vsp34 inhibitors with PD-1/PD-L1 pathway inhibition in murine tumor models (27, 38). Autophagy inhibition may also modulate the functional state of immune cell types in the tumor microenvironment, including macrophages and DCs.

The approach presented here may be particularly relevant in T cell–infiltrated tumors characterized by substantial heterogeneity. Targeting of TNF signaling and autophagy pathways may also enhance the antitumor activity of NK cells because these cytotoxic lymphocytes also secrete IFNγ and TNFα, although this aspect was not directly addressed in this study. The strategy may also be relevant for tumors in which other immune cell populations, including macrophages, secrete substantial quantities of TNFα. Activated T cells express several costimulatory receptors belonging to the TNFRSF family, including OX40 and 4-1BB, which activate the NF-κB2 pathway. cIAP1 and 2 inhibit the NF-κB2 pathway, and cIAP inhibitors thus enhance costimulatory signaling in both human and murine T cells (39, 40). In a phase I clinical trial, the cIAP inhibitor birinapant was found to be well tolerated at a dose of 47 mg/m² (41). A number of clinical trials are ongoing to evaluate cIAP inhibitors in combination with other immunotherapy drugs, including PD-1, PD-L1, and LAG-3 mAbs (40). Potential side effects of the combination of cIAP plus autophagy inhibitors will need to be carefully considered; chronic inflammatory or autoimmune diseases characterized by infiltration by TNFα- and IFNγ-secreting T cells should be considered as contraindications. The tumor selectivity of the combination regimen could potentially be enhanced by nanoparticle-mediated delivery of these small-molecule drugs.

Cross-presentation of tumor antigens by DCs is essential for priming of CD8 T-cell responses (42). Interestingly, inactivation of Rnf31 and in particular Rnf31/Atg5 enhanced DC cross-presentation of apoptotic tumor cells pretreated with TNFα or IFNγ. In this experimental setting, B2m-deficient tumor cells could not directly present the antigen to CD8 T cells. Targeting of these pathways may therefore not only sensitize MHC-deficient tumor cells to apoptosis by T cell–secreted cytokines but also enhance priming of antitumor T-cell populations.

B16F10, Py8119, MC38, and A375 parental cell lines were purchased from ATCC. B16-Ova-ZsGreen cells were described previously, and Py8119-Ova-ZsGreen and MC38-Ova-ZsGreen were generated by the same method (43). An mCherry expression vector was generated by replacing the corresponding eGFP sequence in pLKO.3G (Addgene plasmid #14748). mCherry⁺ cells were generated by lentiviral transduction of the parental line and sorted using an Aria III flow cytometer (BD Biosciences) to purity in order to establish the cell line. All cell lines except for B16F10-Cas9 were grown in complete DMEM (10% FBS and 50 U/mL of penicillin/streptomycin). B16F10-Cas9 cells were maintained in complete DMEM plus blasticidin (2.5–5 μg/mL) to maintain Cas9 expression. Cell lines were confirmed to be negative for Mycoplasma contamination using the ATCC Universal Mycoplasma Detection Kit.

Cells were edited by electroporation with ribonucleoprotein complexes composed of Cas9 protein with bound gRNAs. gRNAs (20 μmol/L, designed using the Genomic Perturbation Platform CRISPick, Broad Institute) were mixed at an equimolar ratio with Cas9 protein (QB3 MacroLab, University of California, Berkeley). Editing of genes in tumor cells was performed by nucleofection using the SF cell line 96-well nucleofector kit and 6 × 10⁴ tumor cells per nucleofection reaction (Lonza Bioscience). Gene editing efficiency was determined by DNA sequencing and subsequent TIDE analysis, as well as Western blot analysis. Anti-cGAS mAb (clone D3O8O, Cell Signaling Technology; #31659S), anti-MAVS antibody (Cell Signaling Technology; #4983S), anti–caspase-3 antibody (Cell Signaling Technology; #9662S), anti-MLKL antibody (Cell Signaling Technology; #37705S), anti-ATG5 mAb (clone D5F5U, Cell Signaling Technology; #12994S), and anti-PIGS mAb (clone EPR11275[B], Abcam; #ab157211) were used in Western blot. Inactivation of Rvr, Ifnar1, Tnfrsf1a, and B2m was confirmed by flow cytometry analysis of surface PVR, IFNAR1, TNFR1, and H-2Kd/H-2Dd after staining with mAbs for these surface molecules [Clones; TX56 (BioLegend), MAR1-5A3 (Thermo Fisher), HM104 (BioLegend), 34-1-2S (BioLegend), respectively]. Crispr RNAs (crRNA) were generated by IDT. Sequences of crRNAs were: Control, ACTTTGCGCTTACATAGCAG; Rnf31, CTACCTCAACACCCTATCCA; Atg5, AAGAGTCAGCTATTTGACGT; B2m, ATTTGGATTTCAATGTGAGG; Ifngr1, GGTATTCCCAGCATACGACA; Ifnar1, TTCAGCAGAATATCGAACGT; Tnfrsf1a, AGTTGCAAGACATGTCGGAA; Cgas, GCGAGGGTCCAGGAAGGAAC; Mavs, CCAGCAACCAAAGTCACCAC; Pvr, AACTTAAGTGTAGAAGACGA; Fas, CAAACTTAGGACTTACCAAG; Gne, TTGCAGCTCAAAGATATATG; Mlkl, GACTTCATCAAAACGGCCCA; Casp3, AACCTCAGAGAGACATTCAT; RNF31, CCACCGTGCTGCGAAAGACA; ATG5, AAGAAGACATTAGTGAGATA. The control gRNA was selected based on the following criteria: (i) targeting of an intergenic region of the mouse genome and (ii) minimal enrichment or depletion in a prior genetic screen (8), indicative of a negligible influence on cell survival and proliferation.

Six- to 8-week-old female mice were used for all experiments. C57BL/6 mice (JAX stock #000664) and OT-I mice (C57BL/6-Tg(TcraTcrb)1100Mjb/J: JAX stock #003831) were purchased from The Jackson Laboratory. Colonies for each strain of mice were maintained in the same animal facility. Mice were housed in pathogen-free conditions and in accordance with the animal care guidelines from the Dana-Farber Cancer Institute (DFCI) standing committee on animals and the NIH. Animal protocols were approved by the DFCI Institutional Animal Care and Use Committee.

OT-I TCR transgenic mice were purchased from The Jackson Laboratory (stock #003831). CD8 T cells were isolated from spleens and lymph nodes from OT-I TCR transgenic mice using the EasySep mouse CD8⁺ T-cell isolation kit (STEMCELL #19753) according to the manufacturer's protocol. Freshly isolated CD8 T cells were cultured in complete RPMI 1640 media (10% FBS, 20 mmol/L HEPES, 1 mmol/L sodium pyruvate, 0.05 mmol/L 2-mercaptoethanol, 2 mmol/L L-glutamine, and 50 U/mL penicillin and streptomycin). T cells were stimulated with anti-CD3/CD28 beads (Thermo Fisher Scientific #11452D) at a bead-to-cell ratio of 1:2. Recombinant mouse IL2 (BioLegend, #575406) was added to the culture at 20 ng/mL. T cells were used for coculture with B16F10 cells after at least 8 days of in vitro culture.

Mouse CRISPR Brie lentiviral pooled libraries (Broad Institute of MIT and Harvard) consisting of 79,637 gRNAs were cotransfected with packaging plasmids (psPAX2 #12260 and pCMV-VSV-G #8454) into HEK293T cells using LT-1 transfection reagent (Mirus; cat. #MIR2305) following the manufacturer's protocol. Library DNA (37 μg), psPAX2 DNA (46 μg), and VSV-G DNA (4.62 μg) were mixed and transfected into HEK293T cells in a T162 flask. Six hours after transfection, media were removed and replaced with 60 mL of virus production media (DMEM supplemented with 20% of FBS). Forty-eight hours after transfection, lentiviral media were harvested and stored at −80°C.

B16F10 B2m^−/− Cas9 cells (1 × 10⁶) were plated per well of a 6-well plate. B16F10 B2m^−/− Cas9 cells were infected with different amounts of lentivirus overnight in the presence of 8 μg/mL of polybrene. The following day, 10⁵ infected B16F10 B2m^−/− Cas9 cells from each condition were seeded per well into a 6-well plate (in duplicates). Twenty-four hours following infection, puromycin (1 μg/mL) was added. Forty-eight hours later (after uninfected cells had died), infected cells in each well were counted. We calculated the percentage of cell survival for each viral concentration as described previously (8). The multiplicity of infection (MOI) for the screen was 0.06.

A total of approximately 10⁸ B16F10 B2m^−/− Cas9 cells and 10⁸ B16F10-Ova cells were prepared for each of the three replicates of the screen: (i) 1.6 × 10⁷ of edited B16F10 B2m^−/− Cas9 cells were mixed with 1.6 × 10⁷ of B16F10-Ova cells and cultured without T cells (control condition); (ii) 1.6 × 10⁷ of edited B16F10 B2m^−/− Cas9 cells were mixed with 1.6 × 10⁷ of B16F10-Ova and cocultured with 1.06 × 10⁷ OT-I T cells [effector-to-target ratio (E:T) of 1:3, experimental condition]. After incubation for 24 hours, genomic DNA was extracted from cells using the Blood and Cell Culture DNA Maxi/Midi Kit (Qiagen; #13362,13343) following the manufacturer's protocol. The Genetic Perturbation Platform at the Broad Institute of MIT and Harvard performed PCR amplification of the gRNA cassette for Illumina sequencing of gRNA representation (8). Protocols for PCR and Illumina sequencing are available online (http://portals.broadinstitute.org/gpp/public/resources/protocols).

The FASTQ reads and single-cell RNA sequences in the Brie library were loaded into MAGeCK v0.5.9 (Model-based Analysis of Genome-wide CRISPR–Cas9 Knockout; ref. 44), which performs read alignment, data quality control, and data normalization based on control single-guide RNAs (sgRNA; “norm-method” parameter was set to “control” and a list of 994 nontargeting sgRNAs were input to the parameter “control-sgrna”). MAGeCK further compared the experimental condition (with OT-1 coculture) and the control condition (without OT-I coculture) and identified potential regulators of MHC-I–independent T-cell killing.

To identify genes regulating both MHC-I–dependent and –independent T-cell killing, we integrated the CRISPR screen performed in B2m^−/− B16F10 cells and our previous CRISPR screens performed in B2m-WT B16F10 cells (8). Specifically, we overlapped genes regulating MHC-I–independent T-cell killing (P < 0.01 and fold change >2), and common regulators of MHC-I–dependent T-cell killing from two previous CRISPR screens (fold change >2). CRISPR screens from Kearney and colleagues (15) and Manguso and colleagues (17) were used to validate the role of overlapped genes in vitro and in vivo. Pathway enrichment analysis was performed for overlapped genes and context-specific regulators using MAGeCKFlute (45), which performs overrepresentation test on biological processes from the Gene Ontology database (46, 47).

Each of the following 10 genes—Ifngr1, Ifnar1, Tnfrsf1a (cytokine signaling), Cgas, Mavs (innate immune signaling), Pvr, Fas (surface receptors), Gne (glycosylation), Mlkl, and Casp3 (cell death pathways)—was individually inactivated in B16F10 B2m^−/− Cas9 cells. Efficient gene inactivation was confirmed by Western blot or TIDE analysis (>78.9% efficiency). Top hits from the primary genome-scale screen were selected based on an FDR < 0.05 for codependency screens. A minipool library was synthesized for the top 205 depleted and top 20 enriched genes from the primary screen, as well as 20 TNF receptor superfamily genes. This library was synthesized by the Genetic Perturbation Platform at the Broad Institute and included 1,470 gene-targeting gRNAs (6 gRNAs/gene) and 1,000 control gRNAs for data normalization. Lentivirus for the mini-pool gRNA library was produced as described above, and a low MOI was used for the validation screen (MOI = 0.06). Genetic codependency screens were performed as described above for the genome-scale screen with a representation of 1,000 (cell number/gRNA): (i) 2.5 × 10⁶ of edited B16F10 B2m^−/− Cas9 experimental-KO cells were mixed with 2.5 × 10⁶ of B16F10-Ova cells and cultured without T cells (control condition); (ii) 2.5 × 10⁶ of edited B16F10 B2m^−/− Cas9 experimental-KO cells were mixed with 2.5 × 10⁶ of B16F10-Ova cells and cocultured with 1.6 × 10⁶ OT-I T cells (E:T of 1:3; experimental condition, duplicate) for 2 days. Genomic DNA was extracted from cells, and gRNA representation was quantified as described above. The codependency screens were also analyzed using MAGeCK v0.5.9, which normalized the data based on control gRNAs. We evaluated the effect of gene KO for each edited tumor cell line by comparing gRNA abundance in the T-cell coculture condition versus a control condition without T-cell coculture. To compare the gRNA enrichment scores across conditions, we normalized the logarithm fold changes in all conditions to the same range using the min–max normalization method. We then evaluated the effect of each of the 10 gene KOs by computing the deviation of gRNA enrichment in each KO cell line versus the control-KO cell line. A deviation score greater than the 2-fold SD of the normalized gRNA enrichment score was considered significant. Protein–protein interaction networks of the selected genes were downloaded from STRING (48) and further visualized by the ggraph R package (https://cran.r-project.org/web/packages/ggraph/index.html).

We established an Atg5-KO signature and an Rnf31/Atg5-dKO signature by extracting the top 200 upregulated and 200 downregulated genes from an RNA-seq experiment conducted with B16F10 cells and taking the normalized DESeq2 Wald statistics as weights. We evaluated the Atg5-KO and Rnf31/Atg5-dKO signature scores by computing the weighted average expression of the signature genes. Finally, we tested the association of the Atg5-KO signature score or the Rnf31/Atg5-dKO signature score with melanoma patient survival using Cox proportional hazards regression analysis, with age, gender, race, and tumor purity regressed out.

B2m^−/− control-KO mCherry⁺ tumor cells (2.5 × 10⁴/well) or B2m^−/− experimental-KO mCherry⁺ tumor cells (2.5 × 10⁴/well) were cocultured with B16-Ova ZsGreen⁺ tumor cells (2.5 × 10⁴/well) in complete RPMI 1640 media in a 24-well plate with or without activated Ova-specific OT-I T cells [E:T = 1:3 (1.6 × 10⁴ OT-I cells) or 1:6 (0.8 × 10⁴ OT-I cells)]. After incubation for 48 hours, all floating and adherent cells were recovered with TripLE Express Enzyme (Thermo Fisher). Cells were stained with Zombie UV (BioLegend) and Brilliant Violet 785 anti-CD8α (53-6.7, BioLegend) and analyzed using a CytoFLEX LX flow cytometer (Beckman Coulter). Sensitivity of B2m^−/− experimental-KO cells to MHC-I–independent T-cell killing was evaluated by comparing (i) the ratio of live B2m^−/− experimental-KO to B16-Ova cells with the ratio of live B2m^−/− control-KO to B16-Ova cells or (ii) the number of live B2m^−/− experimental-KO cells to B2m^−/− control-KO cells.

In some experiments, carboxyfluorescein succinimidyl ester (CFSE; Thermo Fisher) and CellTrace Violet (CTV; Thermo Fisher) were used instead of mCherry and ZsGreen to identify tumor cell populations. Tumor cells were washed twice with PBS, stained with 1 ng/mL of CFSE or CTV in PBS for 3 minutes at room temperature, and washed with complete DMEM before addition to cocultures. At the end of the assay, cells were stained with Zombie NIR (BioLegend) and Brilliant Violet 785 anti-CD8α before analysis by flow cytometry. For analysis of cleaved caspase-3 and -8, tumor cells were intracellularly stained with PE-conjugated cleaved caspase-8 mAb (clone D5B2, Cell Signaling Technology) and Alexa Fluor 647–conjugated cleaved caspase-3 mAb (clone C92-605, BD Biosciences) diluted 1:150 using Fixation/Permeabilization Kit (BD Biosciences).

In cytokine blocking experiments, IFNγ-blocking antibody (25 μg/mL; Bio X cell, clone XMG1.2), TNFα-blocking antibody (25 μg/mL; Bio X cell, clone XT3.11), or the combination of both antibodies was added to mixed tumor cell cultures before the addition of OT-I T cells. In experiments with small-molecule inhibitors, tumor cells were pretreated with solvent (12% SBE-β-CD, MedChemExpress), cIAP inhibitor (birinapant, 500 nmol/L, MedChemExpress), VPS34 inhibitor (SAR405, 500 nmol/L, MedChemExpress), or the combination of both inhibitors (500 nmol/L each) for 48 hours before coculture. B16 B2m^−/− control-KO mCherry⁺ tumor cells (2.5 × 10⁴/well) were cocultured with B16-Ova ZsGreen⁺ tumor cells (2.5 × 10⁴/well) in complete RPMI 1640 media in a 24-well plate with or without activated Ova-specific OT-I T cells [E:T = 1:6 (0.8 × 10⁴ OT-I cells)]. The same concentrations of solvent, cIAP inhibitor, VPS34 inhibitor, or the combination of both inhibitors were added to T-cell coculture assays, and live tumor cells were analyzed 48 hours later.

B16 control-KO cells were labeled with CTV, and B16 control-KO, B16 Rnf31-KO, B16 Atg5-KO, and B16 Rnf31/Atg5-dKO cells were labeled with CFSE as described above. CTV-labeled B16 control-KO cells (2.5 × 10⁴/well) were cocultured with CFSE-labeled B16 control-KO cells or CFSE-labeled B16 experimental-KO cells (2.5 × 10⁴/well) in complete RPMI 1640 media in a 24-well plate with or without the addition of activated Ova-specific OT-I T cells (E:T = 1:6 [0.8 × 10⁴ OT-I cells]) plus 0.2 ng/mL of Ova-derived SIINFEKL peptide (Sigma-Aldrich). After incubation for 48 hours, all floating and adherent cells were recovered with TripLE Express Enzyme (Thermo Fisher). Cells were stained with Zombie NIR (BioLegend) and Brilliant Violet 785 anti-CD8α (53-6.7, BioLegend) and analyzed using a CytoFLEX LX flow cytometer (Beckman Coulter). Sensitivity of exp-KO cells to MHC-I–dependent T-cell killing was evaluated by comparing the ratio of live CFSE⁺ experimental-KO to CTV⁺ control-KO cells.

OT-I T cells were activated with anti-CD3/28 beads (bead-to-T cell ratio = 1:2) and 20 ng/mL of IL2 for 6 days. B16F10 tumor cells (1.5 × 10⁶) were plated with activated OT-I T cells (1.5 × 10⁶) and 0.5 ng/mL of SIINFEKL peptides in a 6-well plate in 2 mL of complete DMEM. Supernatant was collected, passed through a 20-μm filter, and stored at −80°C. Supernatant from B16F10 culture without OT-I T cells and SIINFEKL peptides was used as the control supernatant.

A Celigo image cytometer (Nexcelom) was used to study the survival of tumor cells after incubation with activated T-cell supernatant or T-cell cytokines (IFNγ and/or TNFα). B16 B2m^−/− control-KO or B16 B2m^−/− experimental-KO cells (5,000 cells/well) were added per well in flat-bottom 96-well plates (6 replicates/group). TNFα (7.5 ng/mL, PeproTech), IFNγ (50 ng/mL, PeproTech), or both cytokines were added to the culture and incubated for 72 hours. To A375 cells, human TNFα (50 ng/mL, PeproTech) or human TNFα (7.5 ng/mL) plus human IFNγ (25 or 50 ng/mL, PeproTech) was added. In activated T-cell supernatant assays, tumor cells were cultured in 75 μL of activated T-cell supernatant (or control supernatant) plus 25 μL of complete DMEM for 24 hours. Following 24 (supernatant experiments) or 72 (cytokine experiments) hours of culture, the media were removed and 100 μL of complete DMEM was slowly added along the wells so as not to disturb the adherent tumor cells. Live adherent tumor cells were then counted using a Celigo image cytometer (Nexcelom, PerkinElmer). In Fig. 7E and Supplementary Fig. S11B and S11C, adherent mCherry⁺ cells were counted as live cells. In the apoptosis and necroptosis inhibitor assays, 50 μmol/L of the apoptosis inhibitor Z-VAD-FMK (Selleck) or 10 μmol/L of the necroptosis inhibitor necrostatin-1 (Sigma-Aldrich) was added to cocultures of tumor cells and T cells. In experiments in which cleaved caspase-8 expression was examined, tumor cells were intracellularly stained with PE-conjugated anti–cleaved caspase-8 mAb (clone D5B2, Cell Signaling Technology) diluted 1:150 using Fixation/Permeabilization Kit (BD Biosciences). Pharmacologic sensitization to activated T-cell supernatant (or recombinant cytokines) was examined using solvent control (12.5% SBE-β-CD), cIAP inhibitor (birinapant, 500 nmol/L, MedChemExpress), VPS34 inhibitor (SAR405, 500 nmol/L, MedChemExpress), or cIAP inhibitor plus VPS34 inhibitor (500 nmol/L each) added to B16 B2m^−/− control-KO cells (2,500 or 5,000 cells/well) or PY8119 B2m^−/− control-KO cells, which were preincubated with the same concentrations of inhibitors for 48 hours. cIAP inhibitor (0, 12.5, 25, or 75 mmol/L) and/or VPS34 inhibitor (500 nmol/L) were added to A375 cells. Inhibitors were solubilized to 4 mg/mL in aqueous 12.5% Captisol (SBE-β-CD; MedChemExpress). Solutions were sonicated in a water bath, and 25 to 30 mmol/L HCl was added to increase solubility. Solutions were then neutralized by the addition of NaOH and adjusted to pH 7.5 using pH strips.

B16 B2m^−/− control-KO, B16 B2m^−/−Rnf31-KO, B16 B2m^−/−Atg5-KO, and B16 B2m^−/−Rnf31/Atg5-dKO cells were cultured overnight at a density of 1 × 10⁶ cells/mL with 20 ng/mL mouse TNFα (PeproTech), and then scraped and washed twice with PBS. Pellets were solubilized in RIPA buffer (Thermo Fisher) with protease inhibitors (Sigma-Aldrich) and clarified with Qiashredders (Qiagen). Lysates were assayed using the BCA Protein Assay (Thermo Fisher), and 30 μg total protein was loaded per lane onto 12% NuPAGE Novex Bis-Tris mini gels (Thermo Fisher). Gels were transferred to Sequi-Blot PVDF membranes (Bio-Rad) and probed using mouse cleaved caspase-8 polyclonal antibody (Cell Signaling Technology; #9429), followed by anti-rabbit HRP. Blots were incubated in Western Lightening Plus-ECL substrate (PerkinElmer). Luminescence was captured using a ChemiDoc MP system (Bio-Rad). Blots were then reprobed with GAPDH-HRP (Cell Signaling Technology; #8884).

B16F10 Rnf31-KO Cells (2.5 × 10⁶) were plated on poly-d-lysine (Gibco)–coated glass slides placed in a 100-mm dish in 10 mL of complete DMEM. Cells were treated for 4.5 hours with TNFα (100 ng/mL), fixed with ice-cold methanol for 15 minutes at −20°C, washed twice with PBS, and blocked with 3% rat serum in PBS for 15 minutes at 4°C. Cells were then stained with primary antibodies for 15 minutes at 4°C, washed twice with 3% rat serum in PBS, stained with secondary antibodies and DAPI (25 μg/mL, Life Technologies) for 15 minutes at 4°C, and washed 3 times with 3% rat serum in PBS. Primary and secondary antibodies were diluted in blocking buffer. Coverslips were mounted using Prolong Glass Antifade Mounting Medium (Thermo Fisher). Imaging was performed using a Zeiss LSM 980 confocal microscope. Images were captured using Zen Software. Three independent experiments with three biological replicates per group were performed. For cleaved caspase-8 detection, anti–cleaved caspase-8 rabbit mAb (1:200 dilution; clone D5B2, Cell Signaling Technology; #8592) and secondary antibody Alexa Fluor 647-goat anti-rabbit IgG (5 μg/mL, Southern Biotech) were used. Alexa Fluor 488 anti-LC3B antibody (1:100 dilution; clone EPR18709, Abcam, #ab225382) was used to detect LC3B.

Total RNA was extracted using the RNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. Extracted RNA (1 μg) was transcribed into cDNA using SuperScript IV VILO master mix with ezDNase enzyme according to the manufacturer's protocol (Thermo Fisher). The cDNA samples were diluted and used for real-time qPCR (RT-qPCR). TaqMan master mix (Thermo Fisher) and gene-specific primers (Cxcl10, Mm00445235_m1; Madcam1, Mm00522088_m1; Isg15, Mm01705338_s1; Gapdh, Mm99999915_g1; Thermo Fisher) were used for PCR amplification and detection using a QuantStudio 6 Flex Real-Time PCR System (Thermo Fisher). The RT-qPCR data were normalized to Gapdh mRNA.

Female C57BL/6J (The Jackson Laboratory; #000664) mice ages 6 to 9 weeks were purchased from The Jackson Laboratory. B16-Ova-ZsGreen cells and B16 B2m^−/− experimental-KO-mCherry cells (B2m^−/− control-KO, B2m^−/−Rnf31-KO, B2m^−/−Atg5-KO, or B2m^−/−Rnf31/Atg5-dKO) were mixed at a 4:1 ratio, and tumor cells (4 × 10⁵) were subcutaneously inoculated in 50 μL of PBS into the left flank of syngeneic C57BL/6J mice. When tumors were palpable (day 7), activated OT-I T cells (3 × 10⁶) were intravenously administered into the tail vain, and treatment with PD-1 mAb (Bio X Cell, clone 29F.1A12, #BE0273, 200 μg twice weekly) or rat IgG2a isotype control mAb (Bio X Cell, clone 2A3, #BE0087, 200 μg twice weekly) was initiated. Endpoints were tumor diameter of >20 mm, tumor ulceration, or death of mice. Tumor size was recorded twice per week. When tumor cell composition was analyzed, tumor-bearing mice were euthanized on days 25 to 30. Tumors were resected and cut into small pieces using sterile scalpels in serum-free RPMI 1640 media (Thermo Fisher; #11875093). Tumor fragments were transferred into 50 mL Falcon tubes with RPMI media (5 mL of digestion medium per 0.25 g tumor) containing 160 U/mL collagenase type IV (Gibco) and 80 U/mL DNase I (Sigma). Samples were incubated at 37°C on a shaker platform (250 rpm) for 30 minutes. Following enzymatic digestion, the reaction was quenched by washing cells with FACS buffer and centrifugation at 300 × g for 6 minutes to pellet the cells. The cells were resuspended in FACS buffer and passed through 70-μm cell strainers. Cells were stained for 10 minutes with 100 μL of Zombie UV dye (BioLegend) diluted 1:200 in PBS. The ratio of live ZsGreen⁺ cells to live mCherry⁺ cells was analyzed using a CytoFLEX LX flow cytometer.

Female C57BL/6J (The Jackson Laboratory; #000664) mice ages 6 to 9 weeks were purchased from The Jackson Laboratory. Py8119-Ova-ZsGreen cells and PY8119 B2m^−/− experimental-KO-mCherry cells (B2m^−/− control-KO, B2m^−/−Rnf31-KO, B2m^−/−Atg5-KO, or B2m^−/−Rnf31/Atg5-dKO) were mixed at a 4:1 ratio, and tumor cells (5 × 10⁵) were injected in 50 μL PBS orthotopically into the mammary fat pads of syngeneic C57BL/6J mice. Endpoints were tumor diameter of >20 mm, tumor ulceration, or death of mice. Tumor size was recorded twice per week. In the CD8 T-cell depletion experiment, mice received i.p. injection of CD8β mAb (Bio X Cell, clone 53–5.8 #BE0223) or isotype control mAb (Bio X Cell, clone HRPN #BP0088) at a dose of 100 μg in 100 μL PBS (day −1, day 0, and then twice weekly).

Female C57BL/6J (The Jackson Laboratory; #000664) mice ages 8 weeks were purchased from The Jackson Laboratory. B16F10 control-KO cells or B16F10 experimental-KO cells (B16 Rnf31-KO, B16 Atg5-KO, or B16 Rnf31/Atg5-dKO; 4 × 10⁵ in 50 μL of PBS) were subcutaneously inoculated into the left flank of mice. When tumors were palpable (day 7), mice with similar tumor burden were randomized to treatment groups. Mice were treated with PD-1 mAb (clone 29F.1A12, #BE0273, 200 μg/injection) or rat IgG2a isotype control antibody (anti-trinitrophenol, clone: 2A3, 200 μg/injection) starting on day 7 after tumor inoculation and then twice weekly. In the CD8 T cell–depletion group, mice received i.p. injection of CD8β mAb (Bio X Cell, clone 53–5.8 #BE0223, 100 μg in 100 μL PBS) on day −1, day 0, and then twice weekly. Endpoints were tumor diameter of >20 mm, tumor ulceration, or death of mice. Tumor size was recorded twice per week.

Tumors were excised on day 14 following inoculation, weighed, and cut into small pieces using sterile scalpels in serum-free RPMI 1640 media (Thermo Fisher; #11875093). Tumor fragments were transferred into 50 mL Falcon tubes with RPMI (5 mL digestion medium per 0.25 g tumor) containing 160 U/mL collagenase type IV (Gibco) and 80 U/mL DNase I (Sigma). Samples were incubated at 37°C on a shaker platform (250 rpm) for 30 minutes. Following enzymatic digestion, the reaction was quenched by washing cells with FACS buffer and centrifugation at 300 × g for 6 minutes to pellet the cells. The cells were resuspended in FACS buffer and passed through 70-μm cell strainers. CD45⁺ cells were isolated using CD45 TIL Microbeads (Miltenyi Biotec) following the manufacturer's protocol. For flow cytometry analysis, cells were stained for 10 minutes with 100 μL of Zombie UV dye (BioLegend) diluted 1:200 in PBS. Cells were washed once with FACS buffer and preincubated with TruStain FcX anti-mouse CD16/32 antibody (clone 93, BioLegend) at 1 μg per 10⁶ cells in 100 μL PBS for 5 minutes on ice before immunostaining.

For flow cytometry analysis of T cells and NK cells, cells were incubated with a combination of fluorochrome-conjugated antibodies to the following surface markers (from BioLegend unless otherwise indicated): CD3ε (17A2, BioLegend/BD Biosciences), CD4 (RM4-5), CD8α (53-6.7), CD45 (30-F11), and NK1.1 (PK13). Cells were stained in FACS buffer (total volume of 100 μL) for 15 minutes at room temperature, washed with FACS buffer, and subsequently fixed in Fixation Buffer (BD Biosciences) or fixed and permeabilized using the FoxP3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific) when intracellular staining was required. Fluorochrome-conjugated antibodies directed against the following intracellular markers were used: granzyme B (NGZB, eBioscience), IFNγ (XMG1.2, BioLegend), and TNFα (MP6-XT22, BioLegend). For analysis of intracellular cytokines in T cells, tumor single-cell suspensions were incubated with 1 × Leukocyte Activation Cocktail with BD GolgiPlug for 4 hours in complete RPMI before surface and intracellular staining.

For flow cytometry analysis of myeloid cells, cells were incubated with a combination of fluorochrome-conjugated antibodies to the following surface markers (from BioLegend unless otherwise indicated): CD3ε (clone 17A2), CD19 (clone 6D5), NK1.1 (clone PK136; dump channel markers), F4/80 (clone BM8), Ly6c (clone HK1.4), CD11b (M1/70), CD11c (clone N418), IA/E (clone M5/114.15.2), CD103 (clone 2E7), and CD8α (53–6.7). All antibodies for surface markers were used at a dilution of 1:200. Stained and fixed cells were resuspended in 300 μL of FACS buffer and stored at 4°C protected from light until analysis within 24 hours. Samples were analyzed using an LSRFortessa X-20 flow cytometer, and FACSDiva software was used for data acquisition. Data were analyzed using FlowJo version 10.7.1 (TreeStar).

B16-Ova B2m^−/− experimental-KO tumor cells (control-KO, Rnf31-KO, Atg5-KO, or Rnf31/Atg5-dKO) were added to 24-well plates (5 × 10⁴/well) in complete DMEM. When cells were attached, TNFα (10 ng/mL) or IFNγ (50 ng/mL) was added to wells. Cells were cultured for 48 hours following the initiation of cytokine treatment, and the percentage of apoptotic cells was then quantified by labeling with FITC Annexin V (BioLegend) according to the manufacture's protocol. Annexin V⁺ cells were analyzed using an LSRFortessa X-20 flow cytometer.

In parallel, XCR1⁺ cDC1s were harvested from spleens of WT C57BL/6 mice using mouse anti-XCR1 microbeads (Miltenyi Biotec) according to the manufacture’s protocol. The enriched DCs were seeded into 96-well U-bottom plates and treated with 5 ng/mL TNFa for 2 hours (2 × 104/well). In the meantime, pretreated tumor cells were harvested and washed 3 times with 1× PBS. Tumor cells (1 × 10⁴) were added to the plate and cocultured with DCs for 6 hours. During this time, CD8⁺ T cells from spleens and lymph nodes of OT-I mice were enriched by magnetic separation using the EasySep Mouse CD8⁺ T-cell Isolation Kit (STEMCELL Technologies) according to the manufacturer's instructions. Enriched CD8⁺ T cells were then labeled with CFSE (Life Technologies). CFSE-labeled CD8⁺ T cells were added to the plate and cocultured with DCs at a 5:1 ratio of T cells to cDC1s; 72 hours later, cells from each well were harvested and stained with CD11c (BioLegend; #117308), CD8 (BioLegend; #100712), and viability dyes (BioLegend; #423108) and analyzed by flow cytometry. T-cell proliferation was evaluated by CFSE dilution.

Py8119-Ova-ZsGreen cells and Py8119 B2m^−/− experimental-KO-mCherry cells were mixed at a 4:1 ratio, and tumor cells (3.5 × 10⁵) were inoculated into a mammary fat pad of 8-week-old female C57BL/6J mice. When tumors were palpable (day 7), mice were treated with either solvent (12% SBE-β-CD, 100 μL daily, i.p., and 50 μl, oral gavage; MedChemExpress), cIAP inhibitor (birinapant, 10 mg/kg daily, i.p. injection, MedChemExpress), VPS34 inhibitor (SAR405, 10 mg/kg daily, oral gavage, MedChemExpress), or the combination of both drugs. Endpoints were tumor diameter of >20 mm, tumor ulceration, or death of mice. Tumor size was recorded twice per week.

B16 control-KO, B16 Rnf31-KO, B16 Atg5-KO, B16 Rnf31/Atg5-dKO tumor cells (5 × 10⁵/well) were plated in 6-well plates and treated with and without TNFα (2 ng/mL, PeproTech) for 24 hours. Three biological replicates were used to extract total RNA using the RNeasy Plus Micro Kit (Qiagen; #74034) according to the manufacturer's protocol. Library preparation and sequencing were performed by Genewiz. The RNA-seq data were analyzed using RSEM (49), which aligns sequencing reads to the GRCm38 reference genome using STAR (50) and counts the reads in each gene. Differential expression analysis between gene-KO cells and control cells was performed using DESeq2 (51), and then pathway enrichment analysis was performed using gene set enrichment analysis (52) implemented in the clusterProfiler R package (53).

Female C57BL/6J mice were treated for 21 days with either solvent (12% SBE-β-CD, 100 μL daily, i.p., and 50 μL, oral gavage; MedChemExpress), cIAP inhibitor (birinapant, 10 mg/kg daily, i.p. injection, MedChemExpress), VPS34 inhibitor (SAR405, 10 mg/kg daily, oral gavage, MedChemExpress), or the combination of both drugs. Following treatment, mice were euthanized for whole animal immersion fixation in Bouin's solution (Sigma-Aldrich). The chest, abdomen, and skin surrounding the neck of mice were opened prior to immersion. After 72 hours of fixation, tissues were decalcified, embedded in paraffin, and sectioned with a microtome. Sections were stained with hematoxylin and eosin. Full necropsy and histologic analysis of major organs, including the liver, lung, kidney, and heart, were performed. Images were acquired on a Nikon Ti-E widefield light microscope.

Statistical analyses were performed using GraphPad Prism 9 software. Each experiment was repeated as indicated in the respective figure legend. One-way ANOVA with Tukey multiple comparisons test, two-way ANOVA with Tukey multiple comparisons test, or unpaired two-sided Mann–Whitney test was used as indicated for comparisons between the groups. Tumor growth kinetics were analyzed using two-way ANOVA with Dunnett post hoc test. For comparison of mouse survival curves, a log-rank (Mantel–Cox) test was used. All P values are two-sided, and statistical significance was evaluated at the 0.05 level.

All transcriptomic data generated by RNA-seq have been deposited at the NCBI Gene Expression Omnibus (GEO) with accession number GSE213078.

Y. Ito reports grants from the Kyoto University Foundation and the Japanese Society for the Promotion of Science (JSPS) during the conduct of the study, as well as personal fees from Asahikasei Pharma outside the submitted work. X.S. Liu reports grants from the NIH during the conduct of the study; conducted the work while on the faculty at DFCI; and is currently a board member for and CEO of GV20 Therapeutics. K.W. Wucherpfennig reports grants from Novartis and personal fees from TScan Therapeutics, SQZ Biotech, Bisou Bioscience Company, DEM BioPharma, Immunitas Therapeutics, and Nextech Invest outside the submitted work, as well as a patent for MICA vaccine licensed to Novartis. No disclosures were reported by the other authors.

Y. Ito: Conceptualization, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. D. Pan: Conceptualization, formal analysis, validation, investigation, methodology, writing–review and editing. W. Zhang: Data curation, formal analysis, investigation, methodology, writing–original draft. X. Zhang: Investigation, methodology. T.Y. Juan: Formal analysis, investigation, methodology. J.W. Pyrdol: Investigation, methodology. O. Kyrysyuk: Investigation, methodology. J.G. Doench: Data curation, software, formal analysis, writing–review and editing. X.S. Liu: Data curation, software, formal analysis, investigation, methodology. K.W. Wucherpfennig: Conceptualization, formal analysis, supervision, funding acquisition, writing–original draft, writing–review and editing.

We thank Sushil Kumar for help with RT-qPCR experiments, Soumya Badrinath for establishing protocols for analysis of tumor-infiltrating lymphocytes, and members of Wucherpfennig and Glimcher labs at DFCI for helpful discussion. This work was supported by NIH grants R01 CA238039, R01 CA251599, P01 CA163222, P01 CA236749 (to K.W. Wucherpfennig), and R01 CA234018 (to X.S. Liu and K.W. Wucherpfennig) and the Kyoto University Foundation (to Y. Ito) and Grants-in-Aid by JSPS for Research Activity Start-up (22K20758; to Y. Ito). K.W. Wucherpfennig is a member of the Parker Institute for Cancer Immunotherapy and the Ludwig Center at Harvard Medical School. We thank Dana-Farber/Harvard Cancer Center for the use of the Rodent Histopathology Core, which provided the histopathologic interpretation. Dana-Farber/Harvard Cancer Center is supported in part by NCI Cancer Center Support Grant NIH 5 P30 CA06516. Cartoons were created with BioRender.com.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.