Inactivation of Fbxw7 Impairs dsRNA Sensing and Confers Resistance to PD-1 Blockade

Immunotherapies, such as anti-CTLA4, anti–PD-1, or their combination, have revolutionized the treatment of patients with cancer (1). However, a key challenge to optimizing the opportunity provided by these therapies is the dramatically varied responses among different patients or even among different tumors in the same patient (2). Prior studies have identified decreased CD8⁺ T-cell infiltration (3), defects in IFN signaling (4, 5) or antigen presentation (6), as well as alteration of viral sensing pathways (7–9) as mechanisms leading to therapeutic resistance. These phenotypes can be altered because of oncogenic events in tumor cells, including activation of β-catenin (10, 11) and loss-of-function mutations in JAK1/2 (5) or in the tumor suppressor LKB1/STK11 (12). Nevertheless, these mechanisms collectively do not account for the majority of cases of immunotherapy resistance. Thus, the identification of additional molecular mechanisms of resistance has the potential to identify patients who are more likely to benefit from these treatments. Elucidation of resistance pathways could also enable rational therapeutic approaches that restore tumor immunity in genomically selected patient populations.

FBXW7 is a commonly mutated tumor suppressor in diverse tumor types. Missense mutations in FBXW7 are observed in about 6% of cancers, including endometrial, colon, cervical, stomach, skin, urothelial, lung, ovarian, testis, breast, pancreatic, renal, liver, prostate, brain, and thyroid cancers (13–15). Approximately 30% of human cancers also have deletions of chromosome 4q32, which includes the FBXW7 locus (15). Inactivating mutations or the genomic loss of FBXW7 disrupts the activity of an evolutionarily conserved SCF ubiquitin ligase complex (13, 16–18), leading to increases in cell proliferation and division proteins such as MYC, Cyclin E1, and JUN.

Although a role of FBXW7 in tumor immunity has not yet been shown, a recent report has described a function of FBXW7 in antiviral immunity through regulating the stability of RIG-I (encoded by DDX58; ref. 19). RIG-I and MDA5 (encoded by IFIH1) are two major viral nucleic acid sensors that defend against viral infection and other pathogens (20). Upon detection of double-stranded RNA (dsRNA) in tumor cells, RIG-I and MDA5 associate with MAVS, leading to the recruitment and autophosphorylation of TBK1. TBK1 phosphorylates the transcription factor IRF3, which triggers the expression of type I IFNs and proinflammatory cytokines, such as CXCL10. This pathway therefore activates innate immune responses in the tumor microenvironment (21).

In this study, we found that a loss-of-function mutation in FBXW7 was associated with resistance to PD-1 blockade in a patient with melanoma. Using an immunocompetent, anti–PD-1–sensitive melanoma mouse model, we found that Fbxw7 deletion or its mutation in tumor cells is sufficient to confer resistance to PD-1 blockade. Tumor-intrinsic Fbxw7 deficiency altered the tumor immune microenvironment by decreasing immune cell infiltration and diminished the activation of viral sensing and IFN signaling pathways in vivo. Fbxw7 was essential for the expression of RIG-I and MDA5, which are both required for Fbxw7-mediated dsRNA sensing. Finally, we have shown that restoration of dsRNA sensing in Fbxw7-deficient cells increased MHC-I expression and sensitized Fbxw7-deficient tumors to anti–PD-1. These findings provide insights into the function of the FBXW7 tumor suppressor gene in tumor immunity and suggest a therapeutic strategy to overcome resistance to PD-1 blockade in a genotype-selected group of patients.

To uncover oncogenic mutations that confer resistance to PD-1 blockade, we identified patients with metastatic melanoma who exhibited resistance of a single tumor site despite responses in other disease sites. One such patient, a 74-year-old man with diffusely metastatic melanoma (Fig. 1A), exhibited a complete response to pembrolizumab in all lesions within 11 months of treatment, except for a right adrenal mass, which did not respond. To identify tumor cell–intrinsic genomic changes associated with resistance to PD-1 blockade, we performed whole-exome sequencing and analysis on a pretreatment lesion (a cervical lymph node), the right adrenal resistant lesion, and a germline sample (peripheral blood mononuclear cells). We then used ABSOLUTE (22, 23) to determine allele fraction of called mutations and allelic copy-number information in the pretreatment and resistant samples. Overall, 1,583 somatic variants were shared between both tumors (Fig. 1B; Supplementary Table S1). Twenty-eight mutations were unique to the resistant adrenal tumor, whereas 26 mutations were unique to the pretreatment tumor, suggesting that the resistant lesion evolved from a precursor clone. As expected for melanoma, both the resistant and pretreatment tumors had a mutational signature consistent with UV exposure (Supplementary Fig. S1A). Both the pretreatment and resistant lesions had similar numbers of mutations, nonsynonymous mutations, and predicted neoantigens (Supplementary Fig. S1B). The copy-number profile between the pretreatment and resistant lesions was also similar (Supplementary Fig. S1C).

We evaluated the 28 mutations in the resistant tumor for known genomic mechanisms of resistance to immunotherapy. However, we found no somatic mutations in antigen presentation or IFN signaling pathways, which are correlated with anti–PD-1 resistance (4–6). We also used the Catalogue of Somatic Mutations in Cancer (COSMIC) database and several variant discovery engines (24–26) to predict oncogenic and deleterious mutations (Fig. 1C). The only known oncogenic mutation that distinguished the pretreatment and the resistant tumor was an arginine-to-cysteine mutation at amino acid 505 (R505C) in the tumor suppressor gene FBXW7 (Fig. 1B and C; Supplementary Table S2). The R505 mutation, the second most common mutation in FBXW7 observed in cancer (Fig. 1D), is associated with the increased expression of FBXW7 substrates and leads to dominant-negative phenotypes, suggesting that immunotherapy resistance could be associated with the loss of FBXW7 activity.

To test the possibility that FBXW7 inactivation leads to resistance to PD-1 blockade in melanoma, we developed a murine melanoma model lacking Fbxw7. Our model was based on D4M3A, a Braf-mutant, Pten-deleted melanoma murine cell line that is 98% genetically identical to C57BL/6 mice (27). D4M3A cells were modified ex vivo to express Cas9 (hereafter denoted D4C9), facilitating the rapid deletion of genes by CRISPR (28). Cells transduced with a control guide RNA (sgRNA) grew similarly in immunocompetent C57BL/6 and immunocompromised nude mice (Supplementary Figs. S2A and S2B and S9H and S9I). Anti–PD-1 treatment of immunocompetent mice with D4C9-sgCtrl tumors was associated with durable tumor control (>100 days), even after only three drug treatments (Supplementary Fig. S2C). Anti–PD-1 treatment had no impact on the survival of nude mice (Supplementary Fig. S2D).

To determine whether Fbxw7 is required for response to anti–PD-1 therapy, we generated D4C9 derivatives lacking Fbxw7. Three independent sgRNAs (each targeting all Fbxw7 isoforms) decreased FBXW7 protein with concomitant increases of the FBXW7 targets MYC and Cyclin E1 (Fig. 2A). There was no difference in the growth rate of Fbxw7-deficient and control D4C9 cells in vitro (Fig. 2B). However, Fbxw7-deficient tumors were resistant to anti–PD-1 treatment compared with isogenic matched wild-type tumors (Fig. 2C; Supplementary Fig. S9A–S9D). Animals with Fbxw7-deficient tumors also had significantly poorer survival after anti–PD-1 treatment compared with mice with wild-type tumors (Fig. 2D). To evaluate the specificity of these results, we expressed a sgRNA-resistant Fbxw7α cDNA in D4C9-sgFbxw7 cells (Supplementary Fig. S2E). Restoration of Fbxw7α increased the number of complete responders to anti–PD-1 as compared with Fbxw7-deficient tumors (Supplementary Figs. S2F and S9J–S9L). Although the survival of animals bearing tumors with Fbxw7 deletion was significantly poorer relative to control animals, mice with Fbxw7α-restored tumors survived similarly to controls (Supplementary Fig. S2G). To evaluate whether the R505C oncogenic mutation also confers resistance to anti–PD-1, we generated D4C9 cells expressing wild-type Fbxw7 or Fbxw7^R505C. Fbxw7^R505C expression induced the expected increase in MYC expression, consistent with its known dominant-negative effect (ref. 29; Fig. 2E). Although wild-type Fbxw7 significantly delayed tumor growth, expression of Fbxw7^R505C conferred resistance to PD-1 blockade (Fig. 2F and G; Supplementary Fig. S9E–S9G). Finally, to determine whether these findings applied to another cancer model, we generated an Fbxw7-deficient derivative of MC38, a colon carcinoma cell line syngeneic to C57BL/6 mice that is partially sensitive to anti–PD-1 treatment (ref. 30; Supplementary Fig. S2H). The deletion of Fbxw7 in this model also significantly diminished the response to anti–PD-1 treatment (Supplementary Fig. S2I–S2M). Together, these data demonstrate that loss of Fbxw7 activity confers resistance to PD-1 blockade.

To identify the mechanisms by which Fbxw7 deficiency impairs antitumor immunity, we used the NanoString nCounter System to measure the expression of immune-related mRNA transcripts in control and Fbxw7-deficient tumors before and after anti–PD-1 treatment (31). We found that Fbxw7 inactivation decreased a specific immune gene signature, including IFNγ-related genes, which has been shown to correlate with responses to pembrolizumab in patients with cancer (ref. 32; Fig. 3A; Supplementary Table S3). Profiling of Fbxw7 wild-type and deficient models also found that genes expressed in response to viral sensing or type I IFN stimulation were among the most significantly decreased genes in Fbxw7-deficient tumors compared with controls after anti–PD-1 treatment (Fig. 3B and C). To validate these findings, we compared the levels of two viral sensing targets in Fbxw7-deficient and control tumors by qPCR. Both Cxcl10 and Ifnb1 mRNA levels were significantly decreased in tumors lacking Fbxw7 compared with control tumors (Fig. 3D and E).

As viral sensing signaling pathways affect immune cell infiltration in tumors (7, 8), we evaluated the effects of Fbxw7 inactivation on the immune microenvironment. We measured the intratumoral abundance of immune cell populations in Fbxw7-deficient and control tumors from C57BL/6 mice after anti–PD-1 treatment. We found that loss of Fbxw7 significantly decreased global immune cell infiltration in tumors, altering both lymphocyte and myeloid cell infiltration (Supplementary Fig. S3A–S3C and gating strategy shown in Supplementary Fig. S10A). Fbxw7 inactivation led to diminished PD-1 blockade–induced CD8⁺ T-cell infiltration (Fig. 3F; Supplementary Fig. S3D–S3F), consistent with the established association of intratumoral CD8⁺ T-cell infiltration with the response to PD-1 blockade (3). We also found that the loss of Fbxw7 decreased the infiltration of dendritic cells and macrophages (Fig. 3G and H; Supplementary Fig. S3G). Macrophages trended toward a more immunosuppressive M2 phenotype following Fbxw7 inactivation (Supplementary Fig. S3H). Together, our results show that the resistance to PD-1 blockade caused by Fbxw7 inactivation is associated with an altered tumor immune microenvironment, including decreased viral sensing and antitumor immune cell infiltration.

We next used The Cancer Genome Atlas (TCGA) datasets to determine the relevance of our findings to human cancers. We used gene set enrichment analysis to identify significantly dysregulated gene sets in mutant FBXW7 compared with wild-type FBXW7 melanomas. Loss-of-function mutations in FBXW7 correlated with increased MYC signaling (Supplementary Fig. S3I and S3J) and diminished type I and type II IFN signaling (Fig. 3I and J; Supplementary Table S4). We also found that expression of FBXW7 strongly correlated with CD8⁺ T-cell infiltration in many human cancer types, including melanoma (Fig. 3K; Supplementary Table S5).

To examine the requirement of Fbxw7 for the activation of viral sensing pathways in tumor cells, we transfected cells with low molecular weight (LMW) poly(I:C), a synthetic dsRNA analog (7). Under these conditions, we observed no significant effect of LMW poly(I:C) on cell viability (Supplementary Fig. S4A). LMW poly(I:C) activated the TBK1–IRF3 signaling pathway (Fig. 4A) and induced Cxcl10 and Ifnb1 mRNA expression in control cells (Fig. 4B and C), but these phenotypes were strongly diminished by the genetic deletion of Fbxw7 (Fig. 4A–C). Conversely, exogenous wild-type Fbxw7 further increased LMW poly(I:C)-induced TBK1 activation, whereas Fbxw7^R505C suppressed it (Fig. 4D). Similarly, Cxcl10 and Ifnb1 mRNA expression were increased following wild-type Fbxw7 overexpression (Fig. 4E and F). We found that replication stress, which can increase dsRNA or double-stranded DNA (dsDNA) levels, was not observed upon wild-type Fbxw7 overexpression (Supplementary Fig. S4B). LMW poly(I:C) also induced MHC-I and PD-L1 cell-surface expression in control D4C9 cells, whereas Fbxw7 deletion significantly diminished these phenotypes (Fig. 4G and H; gating strategy shown in Supplementary Fig. S10B). Together, our results establish Fbxw7 as a novel positive regulator of dsRNA sensing, type I IFN production, and MHC-I expression in melanoma. To evaluate whether Fbxw7 regulates other nucleic acid sensing pathways, we treated control and Fbxw7-deficient cells with the STING agonist ADU-S100 (33). We found that the deletion of Fbxw7 did not affect ADU-S100–mediated TBK1/IRF3 activation (Supplementary Fig. S4C), or Cxcl10 and Ifnb1 mRNA expression (Supplementary Fig. S4D and S4E). The effects of this agonist were specific, as deletion of STING (encoded by Tmem173) blocked the induction of Cxcl10 and Ifnb1 mRNA (Supplementary Fig. S4F and S4G).

Recent studies have demonstrated that IFNγ signaling, which has also been linked to response to immunotherapy, leads to the activation of viral sensing pathways (7, 8). We observed that IFNγ stimulation also increased the abundance of endogenous dsRNA in D4C9 cells (Supplementary Fig. S5A and S5B). Therefore, we evaluated the effect of Fbxw7 deletion on IFN-regulated pathways. IFNγ was sufficient to activate the TBK1–IRF3 signaling pathway and to induce Cxcl10 and Ifnb1 mRNA expression in control D4C9 cells (Supplementary Fig. S5C–S5E). The IFNγ-mediated Cxcl10 expression was dependent on RIG-I and STING (Supplementary Fig. S5F). Conversely, Fbxw7 deletion decreased IFNγ-induced TBK1–IRF3 signaling and type I IFN production (Supplementary Fig. S5C–S5E). Inactivation of Fbxw7 also suppressed IFNγ-induced JAK1 and STAT1 phosphorylation (Supplementary Fig. S5G), without affecting JAK2 levels or IFNGR1/2 cell-surface expression (Supplementary Fig. S5G–S5I). Finally, Fbxw7 deletion was associated with a decrease in IFNγ-induced expression of MHC-I and PD-L1 (Supplementary Fig. S5G, S5J, and S5K) and type I IFN–induced MHC-I expression (Supplementary Fig. S5L).

To comprehensively evaluate the requirement of Fbxw7 in IFN signaling, we evaluated global gene expression after IFNγ stimulation of Fbxw7-deficient and control cells. Gene set enrichment analysis (34) was used to identify pathways enriched or depleted in Fbxw7-deficient cells. We found that two independent sgRNAs targeting Fbxw7 affected gene expression similarly (Fig. 4I). In both cases, we found that virus sensing, as well as type I and II IFN signaling pathways, were among the most significantly downregulated gene sets in Fbxw7-deficient cells compared with controls (Fig. 4J). Collectively, these results establish Fbxw7 as a regulator of the dsRNA sensing and IFN signaling pathways.

The inactivation of Fbxw7 leads to dysregulated MYC, which has been associated with altered tumor immunity. To test the possibility that dysregulated MYC underlies the altered viral sensing observed, we suppressed MYC in control and Fbxw7-deficient tumors. Consistent with previous findings (35), we found that MYC knockdown decreased the expression of PD-L1, as well as MHC-I. However, Myc deletion did not restore IFNγ-induced PD-L1 or MHC-I in Fbxw7-deficient tumors (Supplementary Fig. S6A), indicating that MYC is not required for Fbxw7-mediated IFN signaling.

To determine how Fbxw7 regulates dsRNA sensing, we generated D4C9 derivatives lacking Tbk1 or Mavs using CRISPR. As a control, we generated D4C9 derivatives lacking Tmem173 (Fig. 5A). Whereas expression of wild-type Fbxw7 was sufficient to enhance Ifnb1 production following transfection of dsRNA [LMW poly(I:C)] or dsDNA [poly(dA-dT)] analogues, we found that deletion of Tbk1 or Mavs significantly diminished wild-type Fbxw7-mediated Ifnb1 mRNA expression (Fig. 5B and C). Similar results were obtained upon transfection with dsRNA or dsDNA, consistent with the fact that dsDNA can be converted to dsRNA by RNA polymerase III (36). In contrast, the loss of Tmem173 did not decrease wild-type Fbxw7-mediated dsRNA or dsDNA sensing (Fig. 5B and C). These data confirm that FBXW7 does not regulate the STING-mediated dsDNA sensing pathway, and demonstrate that Tbk1 and Mavs are necessary for Fbxw7-dependent type I IFN production. Next, we functionally evaluated the requirement of RIG-I and MDA5, two upstream activators of MAVS, for Fbxw7-mediated dsRNA sensing. We generated D4C9 derivatives lacking Ddx58 or Ifih1 (Fig. 5D). Cells were then transfected either with LMW poly(I:C), preferentially recognized by RIG-I, or with high molecular weight (HMW) poly(I:C), preferentially recognized by MDA5 (37). We observed that loss of MDA5 or RIG-I strongly diminished dsRNA-induced, wild-type Fbxw7-mediated Ifnb1 mRNA expression (Fig. 5E and F).

In the context of antiviral responses, FBXW7 has been shown to promote RIG-I protein stability (19). Therefore, we evaluated the requirement of Fbxw7 for the expression of viral sensing proteins in tumor cells. The deletion of Fbxw7 led to diminished RIG-I and MDA5 protein expression without affecting MAVS, TBK1, and STING levels (Fig. 5G). In addition, wild-type Fbxw7 was sufficient to increase RIG-I and MDA5 protein levels (Fig. 5H). Together, our findings demonstrate that Fbxw7 is essential for the basal expression of viral sensors that are required for Fbxw7-mediated dsRNA sensing.

To determine whether Fbxw7 regulates RIG-I and MDA5 at the transcriptional or post-transcriptional level, we generated D4C9 cells expressing V5-tagged Ddx58 or Ifih1. We observed that Fbxw7 inactivation in these cells decreased exogenous RIG-I and MDA5 expression (Supplementary Fig. S6B), suggesting that Fbxw7 controls their expression at the posttranscriptional level. Song and colleagues (19) showed that FBXW7 mediates the degradation of SHP2, which promotes the degradation of RIG-I. However, we found that the deletion of Ptpn11 (encoding SHP2) did not restore RIG-I or MDA5 protein expression in Fbxw7-deficient cells (Supplementary Fig. S6C) and did not increase the stability of these proteins (Supplementary Fig. S6D).

To test whether altered dsRNA sensing in tumor cells causes resistance to anti–PD-1, we implanted D4C9 derivatives lacking Ddx58, Ifih1, or Mavs in C57BL/6 mice before treating animals with control IgG or anti–PD-1. We observed that loss of Ddx58, Ifih1, or Mavs impaired sensitivity of D4C9 tumors to PD-1 blockade (Fig. 6A; Supplementary Figs. S7A and S7B and S9M–S9P). Therefore, we hypothesized that the resistance of Fbxw7-deficient tumors to anti–PD-1 therapy is because of altered dsRNA sensing. To test this, we evaluated whether restoration of the dsRNA sensing pathway was sufficient to sensitize Fbxw7-deficient tumors to PD-1 blockade. As both MDA5 and RIG-I deletion diminished Fbxw7-mediated dsRNA sensing (see Fig. 5E and F), we evaluated whether the expression of Mavs could restore type I IFN production in Fbxw7-deficient cells. We generated Fbxw7-deficient and control D4C9 cells expressing either an empty vector, Egfp, or exogenous Mavs (Fig. 6B). Expressed Mavs formed aggregates (Supplementary Fig. S7C) and was sufficient to induce Cxcl10 and Ifnb1 mRNA expression (Fig. 6C and D) and MHC-I and PD-L1 cell-surface expression (Fig. 6E–G), consistent with its functional activation. Mavs overexpression did not affect cell viability (Supplementary Fig. S7D and S7E). The impact of Mavs overexpression was milder in Fbxw7-deficient cells compared with controls. This could be due to the required post-translational modification of viral sensors for their activation (38). To evaluate whether restoration of the dsRNA sensing pathway was sufficient to sensitize Fbxw7-deficient tumors to anti–PD-1, we then implanted control and Fbxw7-deficient D4C9 cells overexpressing Mavs or a control vector in C57BL/6 mice. Consistent with our data showing that wild-type Fbxw7 delays tumor growth (see Fig. 2F), we found that Mavs overexpression was sufficient to delay the growth of control and Fbxw7-deficient tumors. Importantly, Mavs overexpression sensitized Fbxw7-deficient tumors to PD-1 blockade (Fig. 6H; Supplementary Fig. S9Q–S9T) and prolonged the survival of anti–PD-1–treated, Fbxw7-deficient tumor-bearing mice (Fig. 6I). This was associated with significant changes in the tumor immune microenvironment, such as an increase in CD11c⁺ cells in both control tumors and Fbxw7-deficient tumors treated with anti–PD-1 (Supplementary Fig. S7F–S7H). Overall, our results suggest that intact dsRNA sensing promotes antitumor immunity and response to immunotherapy.

To further investigate whether restoration of IFN signaling was sufficient to sensitize Fbxw7-deficient tumors to PD-1 blockade, we overexpressed Irf1 in Fbxw7-deficient and control cells (Supplementary Fig. S7I). Irf1 expression was sufficient to increase the levels of PD-L1 and MHC-I expression (Supplementary Fig. S7J). Whereas Fbxw7-deficient tumors were resistant to PD-1 blockade, Fbxw7-deficient tumors overexpressing Irf1 were sensitive to this treatment, similar to control tumors (Supplementary Figs. S7K and S7L and S9U–S9W). Taken together, these findings show that Mavs and Irf1 overexpression similarly induce MHC-I and PD-L1 expression and sensitize Fbxw7-deficient tumors to PD-1 blockade. They also demonstrate that intact dsRNA sensing further affects antitumor immunity by delaying the growth of D4C9 tumors.

This study establishes, for the first time, a role for the tumor suppressor gene FBXW7 in antitumor immunity. Using a syngeneic, immunocompetent anti–PD-1–sensitive melanoma mouse model, we have demonstrated that Fbxw7 is required for dsRNA sensing and response to PD-1 blockade therapy (Supplementary Fig. S8). Importantly, our findings suggest that restoration of the viral sensing signaling pathway by overexpression of Mavs sensitizes Fbxw7-deficient tumors to anti–PD-1.

By functionally characterizing a single exceptional responder to pembrolizumab, we provided biological insight into mechanisms of immunotherapy responses, and identified Fbxw7 loss of function as a mechanism of resistance to PD-1 blockade. Although our work nominates Fbxw7 as a putative, clinically relevant biomarker of anti–PD-1 response, larger studies in patient populations are needed to evaluate its predictive value reliably. Similarly, the dysregulation of viral sensing pathways in large patient cohorts could be investigated.

Song and colleagues (19) have previously observed that FBXW7 regulates the stability of RIG-I in the context of viral stimulation. They found that FBXW7 mediates the degradation of SHP2, a negative regulator of RIG-I. Consistent with their findings, we observe that Fbxw7 is required for the expression of RIG-I. However, we found that Fbxw7 loss also affects the levels of MDA5, which was not seen in their model. It is notable that many regulators of MDA5 and RIG-I stability are shared, because of their significant protein similarity. Our findings suggest that SHP2 does not mediate the degradation of RIG-I. The mechanism by which FBXW7 regulates RIG-I and MDA5 in tumor cells remains to be fully elucidated. The identification of a direct modulator of their expression could enable rational therapeutics that restore dsRNA sensing and immunotherapy response in genomically defined patient populations.

In agreement with recent studies (8, 9, 39), our findings demonstrate the requirement of intact dsRNA sensing for MHC-I expression and response to immune checkpoint blockade. Furthermore, our in vivo experiments show that overexpression of Mavs is sufficient to delay the growth of tumors even in the absence of anti–PD-1 therapy. Therefore, it will be of therapeutic interest to comprehensively understand the mechanisms by which the activation of viral sensing pathways promotes tumor immunity.

Our findings have important therapeutic implications. First, they reinforce the value of activating the dsRNA sensing pathways to exert an antitumor effect in solid cancers. Therapeutic strategies such as synthetic TLR3, MDA5, or RIG-I agonists (40) are currently being tested in clinical trials for this purpose (41). Furthermore, we provide potential therapeutic strategies for sensitizing tumors with altered viral sensing to PD-1 blockade. However, our results suggest that such therapies may need to be selected for specific patient populations. For example, in Fbxw7-deficient tumors, combining anti–PD-1 with an agonist activating either a dsDNA or a putative dsRNA sensor that is not under the control of FBXW7, such as STING (42), may be advantageous. Reactivation of viral sensing signaling could also be achieved using viruses (43), or epigenetic modulators such as DNMT inhibitors (44).

Other genomic aberrations, beyond Fbxw7 loss of function, have been shown to alter viral sensing pathways and affect the efficacy of immune checkpoint inhibitors. Recent studies demonstrated that inactivation of the tumor suppressor STK11/LKB1 decreases STING expression and dsDNA sensing (45), whereas the loss of Stk11/Lkb1 in tumor cells promotes resistance to PD-1 blockade (12). Together, these findings suggest that the regulation of viral sensing pathways may be a common mechanism by which tumor suppressors affect tumor immunity. They provide further rationale for combining immune checkpoint blockade therapy with selected dsDNA or dsRNA agonists to overcome resistance to immunotherapy in genomically defined cancer patient populations.

We obtained prior, written informed consent from each patient for biopsy collection and analysis. The study was conducted in accordance with recognized ethical guidelines and approved by the Dana-Farber Cancer Institute Institutional Review Board (Boston, MA).

D4M3A and HeLa cells (gifts of David Fisher, Massachusetts General Hospital, Boston, MA) were cultured in DMEM supplemented with 5% FBS and penicillin–streptomycin. MC38 cells (gift of Robin Riley, NIH, Bethesda, MD) were grown in RPMI1640 supplemented with 5% FBS, penicillin–streptomycin, and glutamine. Cells were incubated at 37°C at 5% CO₂. Cells were tested monthly and found to be free of Mycoplasma using PCR-based screening (Applied Biological Materials). Experiments were performed within 25 cell passages after thawing.

Mouse IFNγ (Stem Cell Technologies) reconstituted in PBS 0.1% BSA was used at a final concentration of 0.1 ng/mL or 10 mg/mL. ADU-S100 (Chemietek) reconstituted in water was used at final concentration of 10 μmol/L. LMW-poly(I:C), HMW-poly(I:C), and poly(dA-dT) (InvivoGen) were reconstituted in sterile endotoxin-free water. These compounds were transfected at a final concentration of 1 μg/mL using the TransIT-LT1 Transfection Reagent (Mirus).

The lentiviral Cas9-blast (pXPR_101; Broad Institute of Harvard and MIT) vector was cotransfected with packaging plasmids PAX2 and pMD2.G into Lenti-X cells (Clontech) using TransIT-LT1. D4M3A and MC38 cells were infected with Cas9-blast lentivirus, and then selected with 10 μg/mL of blasticidin (InvivoGen). Single-cell clones were selected for high efficiency editing. sgRNA oligos (sequences in Supplementary Table S6) were annealed and cloned into LentiGuide-hygro (a derivative of LentiGuide-puro). Cas9-expressing clones were infected with lentivirus, followed by drug selection in 150 μg/mL hygromycin (InvivoGen). An early-passage, pooled hygromycin-resistant population of cells was used for all experiments.

The murine wild-type Fbxw7α cDNA in the pENTR1A vector was purchased from GenScript. The pDONR221 Egfp was purchased from Addgene (#25899). The empty vector control for viruses (pENTR4 no ccdB, Addgene #17424) was from Eric Campeau and Paul Kaufman (University of Massachusetts Medical School, Worcester, Massachusetts). The pCMV-SPORT6-Mavs was purchased from PlasmID. The murine Irf1 cDNA cloned in the pDONR vector was obtained from GeneCopoeia, Inc. Site-directed mutagenesis was performed using In-Fusion HD (Clontech). Primer sequences are provided in Supplementary Table S6. Expression constructs in the pLX304_zeo vector were generated using the Gateway cloning system. All plasmids and their derivatives were verified by sequencing. D4C9 cells were stably infected with lentivirus, and then selected with 400 μg/mL of Zeocin (InvivoGen). Murine Ddx58 and Ifih1 cDNAs were purchased from DNASU and Invivogen, respectively. cDNAs were cloned in the pLenti6.2_3xFLAG_V5_ccdB vector (Addgene #87072) using the Gateway cloning system and verified by sequencing. D4C9 cells were stably infected with lentivirus encoding the expression vectors.

Cells were seeded at day 0 and harvested at days 3, 6, and 9, or transfected with LMW poly(I:C) at day 1 and harvested at days 2 and 3. Trypan Blue was added to measure viability. Cell proliferation and viability were measured using the Vi-Cell XR (Beckman Coulter).

5 × 10⁵ D4M3A cells (or derivatives) or 1.2 × 10⁵ MC38 cells (or derivatives) were subcutaneously injected into both flanks of 6-week-old male C57BL/6 mice (Charles River Laboratories). Nine days after implantation, mice were randomized into two groups and treated with 200 μg of either control Rat IgG2α (BioXCell) or anti-PD-1 (clone 29F.1A12; ref. 46) by intraperitoneal injection at days 9, 12, and 15. Tumor volume was measured twice a week using a digital caliper. Individual growth curves are shown in Supplementary Fig. S9. Mice were sacrificed when tumor volume reached 1,000 mm³ and overall survival was monitored. All experiments were performed in compliance with federal laws and institutional guidelines and were approved by the Animal Care and Use Committee of the Dana-Farber Cancer Institute.

Please refer to Supplementary Table S7 for antibodies used.

SDS-PAGE, Western blots, and immunoprecipitation were conducted as described previously (47). Semi-denaturating detergent agarose gel electrophoresis (SDD–AGE) was performed according to published protocols (48) with minor changes. Cells were isolated in Buffer A (10 mmol/L Tris-HCl, pH 7.5, 10 mmol/L KCl, 1.5 mmol/L MgCl₂, 0.25 mol/L d-mannitol, and protease inhibitor cocktail) and homogenized using a 28-G needle syringe, prior to centrifugation at 700 × g for 10 minutes at 4°C. Supernatant was transferred and centrifuged at 10,000 × g for 30 minutes at 4°C. The pellet containing the crude mitochondria was resuspended in 4× sample buffer and loaded onto a 1.5% agarose gel. After electrophoresis in the running buffer (1× TBE, 0.1% SDS), the proteins were transferred to a nitrocellulose membrane (49).

Cells were treated with murine IFNγ or transfected with LMW poly(I:C) at the indicated concentrations. Twenty-four hours later, cells were stained with antibodies for the proteins of interest or with a control IgG, followed by FACS analysis. BD FACSCanto II or BD LSRFortessa was used for data acquisition and FlowJo was used for data analysis. The median fluorescence intensity was calculated using FlowJo. Gating strategy is provided in Supplementary Fig. S10.

Total RNA was extracted from tumors using the RNeasy Plus Mini Kit (Qiagen). Total RNA was extracted from cells using the TRizol reagent (Ambion). RT-PCR was performed using the iTaq Universal SYBR Green One-Step Kit (Bio-Rad), and amplification was measured with a LightCycler 96 (Roche). Expression levels were normalized to 18s. Primer sequences are provided in Supplementary Table S6.

Nine days after subcutaneous injection of tumor cells (see above), mice were randomized into two groups and either sacrificed (pretreatment) or treated with 200 μg of anti–PD-1 at days 9, 12, and 15 prior to tumor harvest at day 16 (post-treatment). Total RNA from tumors was submitted to the CAMD Research Core at Brigham and Women's Hospital (Boston, MA) for mRNA profiling using the nCounter Mouse PanCancer Immune Profiling Panel (NanoString Technologies). The analysis was done using the Advanced Analysis Module of nSolver.

D4C9-sgCtrl cells were plated on glass coverslips and treated with IFNγ (10 ng/mL) for 24 hours. Cells were fixed in 5% (v/v) formaldehyde and 0.25% (w/v) Triton X-100 and incubated overnight at 4°C with an anti-dsRNA antibody, prior to incubation with an anti-mouse IgG2α conjugated with Alexa Fluor 488, and staining with 6-diamidino-2-phenylindole (DAPI). Slides were imaged with a NIS-Element BR 4.60.00 (Nikon). Intensity of signal was measured using the NIS-Element BR 4.60.00 analysis software. Quantification of cell number was performed by ImageJ. Signal intensity was normalized by cell number.

Nine days after implantation, mice were randomized into two groups and either sacrificed (pretreatment) or treated with 200 μg of anti–PD-1 as above. Post-treatment tumors were obtained at day 16. Harvested tumors were processed as described previously (47). Analysis was performed by counting the number of CD3-, CD8-, and F4/80-positive cells per mm² of tumor.

Mice harboring tumors were treated with 200 μg of either control isotype or anti–PD-1 at days 13, 16, and 19 prior to sacrifice at day 20. Tumors were harvested, minced, and blended with the gentleMACS Dissociator (Miltenyi Biotec) and digested with the MACS Miltenyi Tumor Dissociation Kit (Miltenyi Biotec) according to the manufacturer's instructions. Tumor cells were washed with RPMI1640 medium and lysed with RBC Lysis Solution (Qiagen), prior to resuspension in FACS buffer: PBS (Life Technologies) containing 0.5% BSA and 2 mmol/L EDTA (Sigma-Aldrich). The Zombie Aqua Fixable Viability Kit (BioLegend) was applied to cells in combination with anti-mouse CD16/CD32 Blocking Antibody (BioLegend) for 20 minutes in the dark on ice, prior to incubation with primary antibodies for 1 hour in the dark on ice. Cells were fixed and permeabilized using the FOXP3/Transcription Factor Staining Buffer Set (eBioscience), according to the manufacturer's guidelines, and incubated with antibodies for intracellular antigens for 30 minutes in the dark at room temperature. Cells were washed, resuspended in FACS buffer, and analyzed using a BD LSRFortessa Flow Cytometer. Compensation was performed manually on BD FACSDiva using single-color and unstained controls. Signal threshold definition was defined using all-stained and unstained controls. Analysis was performed on FlowJo V10.5.0. Gating strategy is provided in Supplementary Fig. S10.

DNA extraction, sequencing, and whole-exome analysis were performed as described previously (50–52). Somatic nucleotide polymorphisms were identified by MuTect. Mutational clonality was estimated by ABSOLUTE, which uses allelic fraction of called mutations and allelic copy-number information to determine mutational clonality and overall tumor purity and ploidy. Clonal mutations were defined as those with estimated cancer cell fraction of 1 or those whose probability of being clonal exceeded the probability of being subclonal. For copy-number analysis, copy ratios were calculated for each captured target by dividing the tumor coverage by the median coverage obtained in a set of reference normal samples. Mutational signature deconvolution was conducted using a non-negative matrix factorization technique as described previously (52). Mutational signatures were chosen from those previously described in COSMIC (http://cancer.sanger.ac.uk/cosmic/signatures). For neoantigen prediction, the four-digit HLA type for each sample was inferred using Polysolver. Putative neoantigens were predicted by defining all novel amino acid 9-mers and 10-mers resulting from each somatic nonsynonymous point mutation and determining whether the predicted binding rank, a proxy for predicted binding affinity to the patient's germline HLA alleles, was <2% (53). Strong binders had a rank <0.5%, whereas weak binders had a rank between 0.5% and 2% using NetMHCpan (v3.0).

For analyzing correlation of FBXW7 with cytotoxic T lymphocytes, we calculated Spearman correlation and estimated statistical significance using TIMER. To evaluate the correlation of FBXW7 mutations with gene expression, we compared gene set enrichment analysis (hallmark gene set) on a ranked list of genes correlated with FBXW7-mutant melanomas versus nonmutated melanomas.

Control and Fbxw7-deficient cells were stimulated with mouse IFNγ 10 ng/mL or vehicle for 24 hours. Total RNA was extracted using the RNeasy Plus Mini Kit (Qiagen) following the manufacturer's protocol. Total RNA was submitted to the Molecular Biology Core Facility at Dana-Farber Cancer Institute for sequencing. Raw reads were aligned using Gene counts to produce count tables for each gene. Differential gene analysis was performed on gene raw counts with edgeR package bioConductor. Read count table was filtered so that each gene had a minimum of 1 count across all conditions. Other analyses including gene set enrichment analysis were performed using Bioconductor. Raw RNA sequencing will be deposited in the GEO database (accession number: GSE154891).

The results are presented as the mean ± SEM. Statistical significance was assessed using GraphPad Prism software. P < 0.05 was considered statistically significant; ns, statistically nonsignificant.

A. Lako is a senior scientist at Bristol-Myers Squibb. S. Rodig is a scientific advisory board member at Immunitas and reports receiving commercial research grants from Bristol-Myers Squibb, Merck, KITE/Gilead, and Affimed. G.J. Freeman has ownership interest (including patents) in Nextpoint, Triursus, AstraZeneca, Dako, Leica, Mayo Clinic, Novartis, Xios, IgM, iTeos, Roche, Merck MSD, Bristol-Myers Squibb, Merck KGA, and Boehringer-Ingelheim, and advisory board relationships with Roche, Bristol-Myers Squibb, Xios, Triursus, iTeos, Origimed, IgM, and Nextpoint. D.A. Barbie is a consultant at N of One/Qiagen, is a scientific advisory board member at Tango Bioscences, and reports receiving commercial research grants from Novartis, BMS, Lilly, and Gilead. F.S. Hodi is a consultant at Bristol-Myers Squibb, Genentech, Aduro, Verastem, Torque, Amgen, Pieris Pharmaceutical, Novartis, Merck, EMD Serono, Sanofi, Takeda, Surface, Compass Therapeutics, and Apriciy, reports receiving other commercial research support from Novartis (to institution) and Bristol-Myers Squibb (to institution), and has ownership interest (including patents) in MICA Related Disorders (to institution per institutional policies). E.M. Van Allen is a consultant at Tango Therapeutics, Genome Medical, Invitae, Ervaxx, and Janssen, reports receiving commercial research grants from Novartis, BMS, and Sanofi, and has ownership interest (including patents) in Ervaxx, Syapse, Genome Medical, and Tango Therapeutics. R. Haq is a consultant at Tango Therapeutics and reports receiving commercial research grants from Novartis and Bristol-Myers Squibb. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Gstalder, M. Shettigar, S. Rodig, W. Miles, E.M. Van Allen, R. Haq

Development of methodology: C. Gstalder, M. Shettigar, S.L. Carter, A. Lako, G.J. Freeman, D.A. Barbie, W. Miles, E.M. Van Allen, R. Haq

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Gstalder, B. Lutterbach, A.L. DeVine, C. Lin, P. Pancholi, E.I. Buchbinder, M.P. Manos, V. Rojas-Rudilla, E. Gjini, P.-H. Chen, A. Lako, S. Rodig, C.H. Yoon, F.S. Hodi, E.M. Van Allen, R. Haq

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Gstalder, D. Liu, D. Miao, M. Shettigar, A. Lako, S. Rodig, F.S. Hodi, W. Miles, E.M. Van Allen, R. Haq

Writing, review, and/or revision of the manuscript: C. Gstalder, D. Liu, D. Miao, B. Lutterbach, M. Shettigar, E.I. Buchbinder, G.J. Freeman, D.A. Barbie, F.S. Hodi, W. Miles, E.M. Van Allen, R. Haq

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.P. Manos, C.H. Yoon, R. Haq

Study supervision: S. Rodig, R. Haq

Other (collected patient samples): R. Brennick

R. Haq acknowledges funding from the Melanoma Research Foundation, the O'Connor-Macgregor Fund for Melanoma Research, and a Stand Up To Cancer (SU2C) Innovative Research Grant (grant no. #SU2C-AACR-IRG 16-17). SU2C is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the scientific partner of SU2C. E.M. Van Allen acknowledges funding from BroadNext10 and NIH R01CA227388. D. Liu was funded by the Damon Runyon Cancer Foundation Physician Scientist Training Grant, the Conquer Cancer Foundation, and the Society for Immunotherapy of Cancer–Bristol-Myers Squibb Postdoctoral Cancer Immunotherapy Translational Fellowship. G.J. Freeman acknowledges funding from NCI (P50CA101942). W. Miles was funded by the Damon Runyon-Rachleff Innovator Award. The authors would like to acknowledge the DFCI Oncology Data Retrieval System (OncDRS) for the aggregation, management, and delivery of the clinical and operational research data used in this project. We also thank Jennifer L. Guerriero and Adam N.R. Cartwright for help with multiparameter flow cytometry and immunologic aspects of this work. The content is solely the responsibility of the authors.

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