Overcoming Genetically Based Resistance Mechanisms to PD-1 Blockade

Therapeutic efficacy of blocking negative immune regulation through the PD-1 checkpoint is limited by primary and acquired resistance (1–3). Understanding the underlying mechanisms of resistance may allow rational design of combinatorial therapies to overcome resistance. There is emerging data pointing to the loss-of-function mutations in JAK1 and JAK2 from the IFN signaling pathway and B2M from the antigen presentation pathway in patients who become resistant to immune checkpoint blockade therapy (4–10). The frequency of such mutations is low in the limited studies to date (11, 12), but these are defined mechanisms that provide biological evidence of how cancer cells try to escape from the antitumor immune response induced by anti–PD-1/PD-L1 therapies. Furthermore, five preclinical models based on CRISPR screens to study mechanisms of cancer escape from immune response also point to alterations in the antigen-presenting machinery and IFN signaling as the two dominant pathways involved in resistance (13–17). We acknowledge that the biology of these loss-of-function mutations is not straightforward, as several well-documented cases of patients with metastatic melanoma with B2M mutations at baseline were reported to respond to anti–PD-1 or anti-CTLA4 therapy (9, 18–21). In addition, it is envisioned that a subset of cases with primary and acquired resistance to PD-1 blockade therapy can obtain the similar resistant phenotypes through epigenetic changes in the same pathways, which are harder to study in patient-derived biopsies (1, 21–23). On the basis of this knowledge, we postulate that the full mechanistic understanding of these genetic acquired resistance mutations will allow the design of strategies aimed to overcome resistance.

The convergence of data on the IFNγ receptor signaling pathway suggests that the use of combinatorial approaches to reactivate the IFN pathway could overcome resistance driven by JAK mutations (24). For instance, Toll-like receptor 9 (TLR9) agonists that mimic bacterial DNA sequences stimulate both innate and adaptive immune responses by inducing the production of type I IFNs, in particular by plasmacytoid dendritic cells, a cell type with exquisite sensitivity to TLR9 agonists (25). On the other hand, it is well known that MHC class I–deficient tumors are susceptible to natural killer (NK) cell–dependent cytotoxicity (26–28). Without B2M on the tumor cell surface, HLA class I molecules are not stable and are unable to present antigens to CD8⁺ T cells. Hence, NK cells may be activated to recognize “missing self” (27).

To define the IFN signaling and antigen presentation mechanisms of resistance by JAK1, JAK2, and B2M loss-of-function mutations, we developed in vitro and in vivo models and studied the biological significance of these mutations leading to resistance to PD-1 blockade therapy. Using CRISPR/Cas9 genome editing, we created sublines lacking JAK1, JAK2, and B2M expression in four human melanoma cell lines and in the murine MC38 colon carcinoma, which is a well-characterized model with high mutational burden that responds well to anti–PD-1 therapy (29, 30), and validated key results in the widely used primary anti–PD-1 blockade–resistant B16 murine melanoma model (29). On the basis of the molecular understanding of these pathways, we hypothesized that a TLR9 agonist could initiate a potent type I IFN systemic response overcoming anti–PD-1 resistance in JAK1/2-deficient models and that a CD122 preferential IL2 pathway agonist could overcome resistance in B2M-deficient tumors by enhancing NK-cell and CD4 T-cell antitumor activity.

We generated and validated JAK1- and JAK2-knockout sublines of four human melanoma cell lines (M202, M233, M407, and M420), and B2M-knockout sublines of two human melanoma cell lines (M202 and M233) using CRISPR/Cas9 gene editing (Supplementary Figs. S1 and S2A). Next, we characterized the biology of JAK1/2-knockout cells in terms of response to IFNα, IFNβ, and IFNγ stimulation. Three of these four parental melanoma cell lines showed growth inhibition in response to direct in vitro treatment with the three IFNs, whereas M202 was mostly resistant with only a small amount of growth inhibition with IFNβ. All JAK1-knockout sublines were insensitive to all three IFNs, which is in line with the role that JAK1 plays in signaling downstream of both type I and II IFN receptors (31). As expected, because JAK2 is involved in only type II IFN receptor signaling, JAK2-knockout sublines were insensitive to IFNγ but remained sensitive to IFNα and IFNβ (Fig. 1A). We also evaluated the surface expression of PD-L1 and MHC class I after exposure to IFNs. In all JAK1-knockout sublines, PD-L1 and MHC class I surface expression did not increase after exposure to the three IFNs, whereas, as expected (5), the JAK2-knockout sublines did not respond to IFNγ but still responded to IFNα and IFNβ (Fig. 1B). In contrast, B2M knockout in the M202 and M233 cell lines did not affect the sensitivity to any of the three IFNs on cell growth inhibition and PD-L1 surface expression compared with the wild-type counterpart, but instead led to the loss of surface expression of MHC class I (Supplementary Fig. S2B–S2D). Therefore, JAK1-knockout cell lines lose the ability to respond to type I and II IFNs, JAK2-knockout cell lines lose the ability to respond to type II IFN, and B2M-knockout cell lines continue to respond to all three IFNs and lose expression of MHC class I.

To assess the impact of JAK1/2- and B2M-knockout mutations in the ability of T cells to recognize and attack tumor cells, we used HLA-A*02:01 MART1-positive M202 melanoma cells and set up a coculture using human T cells that were retrovirally transduced to express the F5 transgenic T-cell receptor (TCR) specific for MART1 (32, 33). MART1 TCR transgenic T cells had strong in vitro antitumor cytotoxicity against both JAK1/2-knockout MART1-positive melanoma cells comparable to the parental cells, with intact IFNγ production by these T cells in coculture experiments (Fig. 1C). This observation was consistent across different clones of the JAK1/2-knockout cell lines with similar ability to be recognized and killed by antigen-specific T cells (Supplementary Fig. S2E–S2G). On the other hand, B2M-knockout MART1-positive melanoma cells were not recognized by MART1 TCR transgenic T cells, exemplified by the lack of IFNγ production and cell killing when cocultured (Fig. 1C). Taken together, JAK1/2-knockout cell lines have intact T-cell cytotoxicity likely due to sufficient baseline levels of MHC class I expression allowing T-cell recognition. However, both T-cell recognition and cytotoxicity are lost when B2M was knocked out.

To investigate the change in response to IFNγ exposure, we cultured the four cell lines with and without JAK1/2- or B2M-knockout mutations with IFNγ. We harvested RNA after 6 hours of IFNγ stimulation for genome-wide transcriptome comparison with baseline by RNA sequencing (RNA-seq). We then filtered for gene sets involved in immune response and IFNγ signaling and obtained a total of 61 genes that had greater than a 2-fold change difference in expression in all parental cell lines upon IFNγ treatment. These genes were also upregulated in B2M-knockout sublines upon IFNγ exposure, but not in sublines with JAK1/2 knockout (Fig. 2). Overall, wild-type cell lines generally increased gene expression involved in antigen presentation machinery (such as B2M, HLA-A, HLA-B, HLA-C, HLA-F, HLA-G, HLA-DRB1, HLA-DRB5, CIITA, TAP1, TAP2, PSMB8, PSMB9, and PSMB10), IFNγ signaling (such as JAK1, JAK2, SOCS1, SOCS3, STAT1, STAT2, IRF1, IRF2, IRF7, and IRF9), and chemokines (CXCL9 and CXCL10). These changes were also observed in B2M-knockout sublines. However, the ability to upregulate these genes by IFNγ was lost in both JAK1/2-knockout sublines (Supplementary Fig. S3).

In summary, human melanoma cell lines with JAK1/2-knockout mutations do not respond to IFNs, as determined by upregulation of corresponding IFN-response target genes as well as surface expression of PD-L1 and MHC class I, but they can still be recognized by antigen-specific T cells leading to specific cytotoxicity provided they have constitutive baseline MHC class I expression. On the contrary, human melanoma cell lines with B2M-knockout mutations respond to IFNs with expression of corresponding IFN-response genes, but they are not recognized by antigen-specific T cells due to lack of surface MHC class I expression, and therefore are resistant to specific cytotoxicity.

There are very few syngeneic animal models that respond to PD-1 blockade therapy. The murine colon adenocarcinoma MC38 has been previously shown to partially respond to PD-1 blockade therapy (29, 30, 34). It is a carcinogen-induced cell line with high mutational burden which shows an increase in CD8⁺ T-cell infiltration after PD-1 blockade, and therefore it recapitulates highly mutated human cancers that respond to this therapy. To generate a mouse model of anti–PD-1 resistance, we created JAK1-, JAK2-, and B2M-knockout sublines of MC38 (Supplementary Fig. S4A–S4C). Once multiple knockout clones were selected and validated for each gene, we characterized their individual in vitro and in vivo growth curves, as well as the basal surface expression of PD-L1 and MHC class I, to select only the sublines [JAK1 single-guide RNA (sgRNA1) c5, JAK2 sgRNA2 c3, and B2M sgRNA2 c6] that behaved similarly to the parental cell line (Supplementary Fig. S4D and S4E). We found that the functional effects of IFN exposure in terms of growth inhibition and PD-L1 and MHC class I surface expression were broadly comparable with those of the human cell lines with the corresponding knockout genes. MC38 JAK1 knockout was insensitive to all three IFNs and MC38 JAK2 knockout was insensitive to only IFNγ (Fig. 3A–C).

To model in vivo resistance to PD-1 blockade, we injected MC38 wild-type or genetically modified MC38 sublines subcutaneously in the lower flank of the mice. When tumors became palpable (at day 5), mice received the first of four systemic injections of anti–PD-1 therapy or isotype control. In three replicate studies, we demonstrated that JAK1-, JAK2-, and B2M-knockout mutations result in the complete abrogation of the benefit of anti–PD-1 therapy (Fig. 3D and E; Supplementary Fig. S5A), as has been shown in patient biopsy–based studies (4–10, 35) and prior mouse models of B2M knockout (8).

To characterize the tumor-infiltrating immune cell populations in both MC38 wild-type and resistant tumors treated with either isotype or anti–PD-1 therapy, we performed cytometry by time-of-flight (CyTOF) analysis. A total of 19 independent cell clusters were identified using a panel with 28 markers (Fig. 4A; Supplementary Fig. S5B–S5D). The T-cell population was defined by four clusters, including a CD4 T-cell cluster positive for CD3e, CD4, and CD25, and three CD8 T-cell clusters positive for CD3e, CD8e, TBET, and PD-1. An NK-cell cluster positive for CD335 and CD161 was also identified (Fig. 4A and B). Using this approach, we analyzed MC38 tumors from mice after three doses of isotype or anti–PD-1 (day 13). Anti–PD-1–treated wild-type control MC38 tumors presented increased T-cell infiltration (P = 0.04) and a trend toward statistical significance (P = 0.07) of increased CD8⁺ T-cell infiltration compared with isotype control antibody-treated tumors, which is in line with data from prior studies (34). We did not observe significant changes in the CD4⁺ and NK-cell infiltration (Fig. 4C and D). However, anti–PD-1 did not induce any significant changes in T, CD4⁺ T, CD8⁺ T, and NK immune cell populations in MC38 JAK1-, JAK2-, and B2M-knockout tumors (Fig. 4C and D).

Given the limited change in the number of tumor-infiltrating CD8⁺ T cells with anti–PD-1 therapy in this model (34), we analyzed the change in functional phenotype of the CD8⁺ T cells including terminally CD8⁺ T exhausted cells (CD3e⁺, CD8a⁺, PD1⁺, TIM3⁺, EOMES⁺, and TBET⁺) and progenitor CD8⁺ T exhausted cells (CD3e⁺, CD8a⁺, PD1⁺, TBET⁺, and CD69⁺), within the CD8⁺ T cluster (CD3⁺CD8⁺; Fig. 4E). Interestingly, JAK1/2- and B2M-knockout tumors were enriched in terminally CD8⁺ T exhausted cells, which have been reported to be unable to respond to anti–PD-1 therapy (36–38), compared with MC38 wild-type tumors (Fig. 4F and G), in line with the observed lack of response to anti–PD-1 therapy in vivo.

Our results thus far demonstrate that the major effect of JAK1/2 knockout is the inability of cancer cells to respond and express an IFN-induced transcriptional profile that leads to downstream immune activation. In addition, JAK1/2-knockout tumors were less immune-infiltrated with functional CD8 T cells compared with wild-type tumors after anti–PD-1 treatment. Therefore, we investigated whether upstream activation with agents that trigger a type I IFN response in the tumor microenvironment could overcome resistance mediated by JAK1 and JAK2 loss. Type I IFN responses could be triggered by recognition of foreign DNA sequences through pattern recognition receptors such as TLR9 signaling. TLR9 stimulation of plasmacytoid dendritic cells, which are high producers of IFNα, could initiate a strong antitumor immune response (25) in tumors lacking JAK signaling because they can still be recognized and killed by antigen-specific T cells, as shown in Fig. 1. To this end, we bilaterally injected wild-type control and JAK1/2-knockout cells into mice, and when tumors became detectable mice were treated with intratumoral TLR9 agonist SD-101 injections into the right flank tumors. We observed antitumor effects in both the injected and contralateral noninjected left flank tumors, with or without systemic anti–PD-1 therapy. In three replicate experiments, the combined therapy of SD-101 and anti–PD-1 provided superior antitumor activity against JAK1- and JAK2-knockout tumors, overcoming resistance in both injected and contralateral noninjected sites (Fig. 5A; Supplementary Fig. S6A), as well as increased survival (Fig. 5B). The antitumor effect of intratumoral SD-101 was consistent when replicated using tumors from additional MC38 JAK1/2 CRISPR/Cas9 knockout sublines (Supplementary Fig. S6B).

To analyze the role of T and NK cells, we performed antibody-mediated CD4⁺ T-cell, CD8⁺ T-cell, and NK1.1⁺-cell depletion studies in mice with JAK1/2-knockout tumors treated unilaterally with intratumoral SD-101 or combination therapy with systemic anti–PD-1 therapy. Antibody-mediated depletion was confirmed by flow cytometric analysis of splenocytes (Supplementary Fig. S7A). In JAK1-knockout tumors, depletion of NK and CD4⁺ T cells, but not CD8⁺ T cells, partially abrogated the SD-101 antitumor activity (Fig. 5C; Supplementary Fig. S7B). In contrast, the CD8⁺ T-cell depletion completely ablated the effect of the combination therapy (Fig. 5D; Supplementary Fig. S7C). In JAK2-knockout tumors, depletion of CD4⁺ and NK cells resulted in a partial abrogation of antitumor activity, whereas CD8⁺ T-cell depletion resulted in complete abrogation of the antitumor efficacy of SD-101 alone and in combination with anti–PD-1 (Fig. 5C and D; Supplementary Fig. S7B and S7C).

To have the power to distinguish tumor-specific immune cell changes from combinatorial treatments, we integrated data from replicate mass cytometry experiments using an automated meta-analysis of CyTOF data, MetaCyto (39). We merged cytometry data across a wide range of cell surface and intracellular markers (Supplementary Tables S1 and S2) and identified immune cell populations using predefined cluster analysis (Supplementary Table S3). Anti–PD-1 therapy did not change the frequency of CD4⁺ and CD8⁺ T effector subsets in both JAK1/2-knockout tumors (Fig. 5E). These results are in line with Fig. 4 and suggest a lack of maintained T-cell activation in JAK1/2-knockout tumors. Treatment with single-agent SD-101 increased NK-cell mobilization in JAK1 knockout and also CD8⁺ and CD4⁺ T cells in JAK2-knockout tumors. Interestingly, the addition of anti–PD-1 treatment improved immune responses and increased the levels of CD3⁺ T, CD4⁺ T, CD8⁺ T, and B cells compared with SD-101 alone (Fig. 5F; Supplementary Fig. S8), in line with our depletion experiments (Fig. 5C and D).

As we documented the increase in T-cell infiltration when adding anti–PD-1 to intratumoral TLR9 administration, we next wanted to elucidate whether the combinatorial treatment increased expression of the T-cell chemoattracting chemokines CXCL9 and CXCL10 within the tumor microenvironment. Therefore, we performed qRT-PCR from MC38 wild-type control and knockout tumors after two doses of intratumoral SD-101 and three doses of anti–PD-1 compared with anti–PD-1 or isotype control. No differential expression of CXCL9 or CXCL10 was detected in JAK1/2-knockout tumors treated with anti–PD-1 comparable to isotype control. Interestingly, the combination of SD-101 with anti–PD-1 treatment produced higher levels of CXCL9 and CXCL10 than did tumors treated with anti–PD-1 or isotype control, in wild-type and JAK1/2-knockout tumors (Fig. 5G). Taken together, our results demonstrate that SD-101 is able to overcome resistance in both JAK1- and JAK2-knockout models by increasing infiltration of T cells and NK cells as a result of markedly increased levels of both CXCL9 and CXCL10 in the tumor microenvironment.

We reasoned that the absence of B2M and subsequent lack of surface expression of MHC class I may sensitize cancer cells to NK cells, as MHC class I is the major inhibitory ligand for NK-cell function (26–28). Therefore, we tested the CD122 preferential IL2 pathway agonist bempegaldesleukin (also known as NKTR-214; refs. 40, 41), administered either alone or in combination with anti–PD-1. In two replicate experiments, the systemic administration of bempegaldesleukin overcame therapeutic resistance to anti–PD-1 in B2M-knockout tumors (Fig. 6A; Supplementary Fig. S9A), with significantly longer survival compared with the control groups (Fig. 6B).

To assess the influence of T and NK cells in the bempegaldesleukin response, depletion studies were performed and confirmed by splenocyte flow cytometric analysis (Supplementary Fig. S7A). Depletion of CD8⁺ T cells did not affect the antitumor effect, as expected from the low MHC class I expression on the tumors. However, the depletion of either CD4⁺ T or NK cells abrogated the antitumor effect of bempegaldesleukin in the MC38 B2M-knockout tumors (Fig. 6C; Supplementary Fig. S9B).

Applying MetaCyto across a set of mass cytometry data, knockout of B2M in tumors resulted in the loss of MHC class I and the reduction of CD8⁺ T-cell infiltration (Fig. 6D and E). However, treatment with bempegaldesleukin was able to overcome resistance through increased infiltration of CD4⁺ T as well as NK cells, resulting in the rejection of B2M-knockout tumors (Fig. 6E). Taken together, our combined meta-analysis and depletion studies suggest that the immune compartment is distinct in the B2M knockout and wild-type tumors due to the lack of MHC class I expression. However, an effective antitumor immune response, consisting of CD4⁺ T cells and NK cells, can still be mounted in these tumors by treatment with bempegaldesleukin.

To validate the concepts underlying the combinatorial strategies to overcome resistance in JAK1/2- and B2M-knockout tumors, we used a second mouse tumor model, the B16 melanoma model, which exhibits primary resistance to PD-1 blockade (30). Functional effects of IFN exposure were consistent with their defective signaling: B16 JAK1 knockout did not upregulate PD-L1 and MHC class I surface expression to all three IFNs, and B16 JAK2 knockout to only IFNγ (Fig. 7A).

To evaluate the impact of the combinatorial treatment, TLR9 agonist plus anti–PD-1 in this model, we bilaterally implanted B16 wild-type control and JAK1/2-knockout cells into mice and treated with intratumoral SD-101 with or without systemic anti–PD-1 therapy, as we had done with the MC38 experiments. Combination of SD-101 with anti–PD-1 resulted in a significant growth delay in JAK1- and JAK2-knockout tumors, overcoming resistance in both injected and contralateral noninjected sites (Fig. 7B; Supplementary Fig. S10A). We also examined the effect of bempegaldesleukin alone or in combination with anti–PD-1 in B16 B2M-knockout tumors. Consistent with the results in the MC38 model, combination of bempegaldesleukin with anti–PD-1 overcame therapeutic resistance to anti–PD-1 in B2M-knockout tumors (Fig. 7C; Supplementary Fig. S10B), with a significant improved survival in B2M-knockout tumors compared with wild-type control.

Targeting mechanisms of resistance to immune checkpoint blockade is an area of high clinical need. In this study, we characterized the significance of JAK1/2 and B2M loss-of-function mutations in tumor cells as resistance mechanisms to anti–PD-1 therapy. Our functional data showed that JAK1/2-knockout mutations resulted in insensitivity to IFNγ-induced antitumor effects while maintaining the ability to be recognized and killed by T cells, and B2M knockout led to the lack of antigen presentation, cancer cell recognition, and cytotoxicity by T cells. This information allowed us to design combinatorial strategies to successfully overcome these anti–PD-1 resistance mechanisms in mouse models.

Defects leading to lack of sensitivity to the IFN signaling pathway have been described as a mechanism of tumor escape and immunoresistance (42–46). Baseline JAK1 or JAK2 mutations are infrequent in human cancers, representing less than 1% of all cases (4), with an increase in incidence in JAK1 mutations in endometrial carcinomas with microsatellite instability (MSI; refs. 7, 47, 48). Upon pressure from the immune system, some cancer cells may escape by developing an inactivating mutation in one JAK allele and losing the second wild-type allele, resulting in a homozygous loss-of-function mutation in the IFNγ response pathway. Such events have been documented in patients with melanoma treated with immune checkpoint blockade therapy with acquired resistance induced by mutations in JAK1 or JAK2 (5, 6), and in patients with metastatic cervical cancer receiving adoptive cell transfer therapy with a TCR specific for the E6 viral antigen mediated by the development of mutations in the IFNγ receptor 1 (46). These mechanisms of resistance have been modeled preclinically with different readouts. Using short hairpin RNA (shRNA) knockdown in human and murine melanoma cell lines, Luo and colleagues (49) concluded that silencing of JAK1 but not JAK2 was critical for the effects of IFNγ and anti–PD-1. However, given the long-standing evidence that both JAK1 and JAK2 are required for IFNγ receptor signaling (43, 45), it is likely that the differences in results may be due to the shRNA partial silencing approach. Furthermore, Williams and colleagues (50) used a murine model of melanoma expressing a strong foreign antigen to study the effects of JAK1 knockout in tumor implantation, showing that it resulted in decreased tumor outgrowth. This result is likely from the inability to express PD-L1 by the JAK1-knockout tumors, resulting in loss of adaptive immune resistance, and increased their clearance at the time of tumor implantation. Our data suggest that once the tumor cells with JAK1 knockout grow in vivo past the point of tumor implantation, then they become resistant to anti–PD-1 therapy due to the inability to respond to IFNγ and amplify the antitumor immune response.

The realization that IFN signaling by cancer cells is critical for anti–PD-1 response led us to hypothesize that changing the tumor microenvironment to have a strong IFN response by triggering pattern recognition receptors may overcome resistance due to lack of IFNγ signaling, and may improve the antitumor activity of immune checkpoint blockade therapy. Synthetic CpG-ODN agonists of TLR9 are being tested in the clinic in combination with the human anti–PD-1 antibody therapy. The combination of intratumoral SD-101 with pembrolizumab resulted in antitumor responses in patients with advanced melanoma who were refractory or resistant to prior anti–PD-1 therapy (51). In the same way, early clinical trials using two other TLR9 agonists, CMP-001 (52) and IMO-2125 (53), or the double-stranded RNA (poly I:C) BO-112 (54), in combination with immune checkpoint inhibitors, resulted in objective antitumor responses in patients who had previously progressed on treatment with anti–PD-1. Together, these clinical data validate our mouse model findings and suggest a strategic approach for overcoming resistance to PD-1 blockade therapy. We observed that the JAK1/2-knockout tumors were still being recognized by antigen-specific T cells and could lead to specific cytotoxicity independent of IFNγ signaling, which provides evidence that if a TLR9 agonist could reactivate type I IFN signaling in tumor microenvironment cells despite the cancer cells not being able to signal through the IFN receptor, then T cells will be able to exert antitumor activity.

Both JAK1/2-knockout tumors respond according to the mechanistic mode of action of SD-101 (55). Antitumor immune response involved in JAK1-knockout tumors was NK and CD4 T cell–dependent when treated with single-agent SD-101, but the addition of anti–PD-1 recovers the central role of CD8 T-cell effector, demonstrating that these tumors may sensitize to the synergistic effect of combined therapy. In contrast, JAK2-knockout tumors required mainly CD8⁺ T cells for tumor suppression, suggesting that JAK2 loss is a more favorable scenario for CD8⁺ T-cell responses.

Loss-of-function B2M mutations have been reported in microsatellite stable and MSI-high colorectal cancers (56), melanoma (9, 57), and lung cancers (58), and promote resistance to immunotherapy mediated by loss of antigen presentation (5, 8–10, 35). Some studies have found the loss of B2M in biopsies from patients who respond to anti–PD-1 (20, 21), with corresponding supportive results when knocking out B2M in mouse models of immunotherapy (8), indirectly suggesting to us that activation of MHC class I–independent CD4⁺ T cells or innate immune cells such as NK cells may contribute to clinical response (18). Bempegaldesleukin is a CD122 preferential IL2 pathway agonist that activates and expands effector CD4⁺ T, CD8⁺ T, and NK cells (40, 59). Combination therapy with bempegaldesleukin and nivolumab has shown encouraging objective responses in patients with treatment-naïve melanoma with PD-L1–negative tumors (60, 61). Consistent with our hypothesis, we obtained durable tumor responses with this combination in B2M-knockout tumors. In line with the lack of MHC class I antigen presentation, depletion studies suggested that NK and CD4⁺ T cells, which are not restricted by MHC class I, play a key role in the antitumor immunity in B2M-knockout tumors. Furthermore, Ardolino and colleagues (62) had also previously described that IL2 therapy could induce antitumor responses in a model of MHC class I–deficient mouse tumor. These studies provide strong mechanistic support for a strategy of IL2-based therapy with anti–PD-1 for treating patients with MHC class I–deficient tumors who have progressed on prior anti–PD-1 therapy, and may also enhance the antitumor activity in previously untreated tumors.

In conclusion, our study links the disruption of genes in the IFN receptor and antigen presentation pathways with mechanisms of resistance to PD-1 blockade therapy and may guide the design of effective combination therapies to overcome resistance. Even in the extreme setting of genetic resistance to PD-1 blockade by JAK1/2 loss-of-function mutations, resistance can be overcome by the intratumoral injection of a TLR9 agonist, whereas resistance through B2M loss can be overcome by an IL2 pathway agonist potently activating an antitumor CD4⁺ T-cell and NK-cell response. Our findings strongly support the testing of these rational combinatorial strategies in patients with such mechanisms of anti–PD-1 resistance.

Patient-derived melanoma cell lines were thawed and cultured at 37°C with 5% CO₂ in RPMI1640 medium supplemented with 10% FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, and 0.25 μg/mL amphotericin B. Cells were maintained and confirmed Mycoplasma- negative using MycoAlert Mycoplasma Detection Kit (Lonza). Cell lines were periodically sampled and used the GenePrint 10 System for cellular authentication, and matched with the earliest-passage cell lines. Cell lines were subject to experimental conditions after reaching two passages from thawing.

The human melanoma cell lines M202, M233, M407, and M420 were subjected to CRISPR/Cas9-mediated knockout of JAK1, JAK2, and B2M. MC38 and B16 murine cell lines were subjected to CRISPR/Cas9-mediated knockout of JAK1, JAK2, and B2M. Using the CRISPR design tool at https://www.welcome.ai/desktop-genetics, the sgRNAs targeted were designed (Supplementary Table S4).

Human melanoma M407 with CRISPR/Cas9-mediated JAK1 and JAK2 knockouts were generated by lentiviral transduction using particles encoding guide RNAs, a fully functional CAS9 cassette, GFP, and puromycin as selectable markers (Sigma-Aldrich), as described previously (5). All other sgRNAs were cloned into the pSpCas9(BB)-2A-GFP vector (Addgene; ref. 63) and then transformed into One Shot Stbl3 Chemically Competent E. coli (Invitrogen) and cultured overnight in Lysogeny Broth (LB) medium with ampicillin plates. Selected colonies were grown in LB overnight and DNA was isolated using the QIAprep Midiprep Kit (Qiagen). Plasmids were then sequence-verified using a U6 promoter primer forward 5′-GCCTATTTCCCATGATTCCTTC-3′. Cells were transfected using Lipofectamine 3000 as per the manufacturer's protocol (Thermo Fisher Scientific), and GFP-positive cells were collected and single-cell sorted 48–72 hours after transfection at the University of California, Los Angeles (UCLA) Flow Cytometry core. Genomic DNA was isolated for each clone (NucleoSpin Tissue XS, Macherey-Nagel) and then a 700-bp region containing the sgRNA was amplified by PCR using the HotStarTaq Master Mix (Qiagen). Disruption was confirmed by Sanger sequencing with tracking of indels by decomposition (TIDE; ref. 64) web tool and finally confirmed by Western blot analysis.

Cell proliferation and growth inhibition assays were performed by real-time live-cell imaging in an IncuCyte ZOOM (Essen Biosciences). Two thousand to five thousand cells were seeded in 96-well plates and cultured at 37°C. Twenty-four hours after plating, culture media were replaced with fresh media with or without 5,000 IU/mL IFNα (Merck Millipore, catalog no. IF007), 500 IU/mL IFNβ (Merck Millipore, catalog no. IF014), and 100 ng/mL IFNγ (BD Pharmingen, catalog no. 554616) in human melanoma cell lines and IFNα (Merck Millipore, catalog no. IF009), IFNβ (Merck Millipore, catalog no. IF011), and IFNγ (PeproTech, catalog no. 315-05) in mouse cell lines at the same concentrations. IFN concentrations were defined after dose-dependent growth inhibition optimization processes for all three IFNs (5).

Melanoma and murine cell lines were seeded into 6-well plates during day 1, targeting 70%–80% of confluence on the day of surface staining. The following day, culture media were replaced with fresh media with or without IFNα 5,000 IU/mL, -β 500 IU/mL, or -γ 100 ng/mL (same conditions as above) for 18 hours. During day 3, after incubation time, cells were trypsinized and incubated at 37°C for 2 hours with media containing the same concentrations of IFNα, IFNβ, or IFNγ. Then media were removed by centrifugation and cells were resuspended with 100% FBS and stained with Allophycocyanin (APC) anti–PD-L1 and PE-Cy7 anti–HLA-ABC in human melanoma cell lines or PE anti–PD-L1 and APC anti–MHC I in mice cell lines, on ice for 20 minutes. To continue, cells were washed once with 3 mL PBS and resuspended in 300 μL of PBS. Dead cell discriminator, 7-AAD, was added to samples prior to data acquisition by LSRII (Becton, Dickinson and Company). Data were analyzed using FlowJo Software (version 10.0.8r1, Tree Star Inc.). Experiments were performed at least twice for each cell line.

Parental human M202 melanoma cell line and established JAK1-, JAK2-, and B2M-knockout sublines were HLA-A*0201⁺MART1⁺ cell lines used to analyze recognition by TCR transgenic T cells with the use of in vitro coculture assays. Cells were cocultured with effector peripheral blood mononuclear cells (untransduced and MART1 F5–specific TCR) at an effector:target (E:T) ratio of 1:1, 2.5:1, and 10:1. Supernatants from six replicate wells for each condition were collected 24 hours post-coculturing and measured IFNγ release by ELISA (eBioscience). Cytotoxicity assays were conducted by real-time live-cell imaging in an IncuCyte ZOOM (Essen Biosciences) and expressed as percentage of cells that were killed by effector cells over period of coculture. Cell lines were stably transfected with a nuclear-localizing red fluorescent protein (NucLight Red Lentivirus EF1a Reagent, Essen Biosciences) to facilitate cell counts. All experiments were performed in a minimum of three independent runs. Graph production and statistical data were analyzed via Prism Software (GraphPad).

Melanoma cell lines were plated at 1.5–2.0 × 10⁶ cells per 10-cm dish or 2 × 10⁵ per 6-well plate for IFNγ treatment. Twenty-four hours after plating, culture media were replaced with fresh media with or without 100 ng/mL IFNγ (BD Pharmingen, catalog no. 554616). Cells were harvested 6 hours after the start of IFNγ treatment, and pellets were lysed in TRIzol reagent (Invitrogen, catalog no. 15596018) and stored at −80°C before RNA extraction.

Total RNA was extracted using a protocol that combined the TRIzol method and the RNeasy Mini Kit (Qiagen, catalog no. 74104). Briefly, the aqueous phase containing RNA from the TRIzol extraction method was mixed with an equal amount of 70% ethanol and loaded on a RNeasy mini column. The column was washed according to the kit manual and total RNA was eluted in 60 μL RNase-free water. RNasin RNase Inhibitor (Promega, catalog no. N2511) was added to a final concentration of 2 U/μL. Total RNA was submitted for RNA-seq to the UCLA Technology Center for Genomics & Bioinformatics.

Single-end reads 50 bp in length were mapped using HISAT2 (65) version 2.0.4 and aligned to the hg19 genome using default parameters. Reads were quantified using HTSeq (66) version 0.6.1 with the intersection-nonempty mode and counting ambiguous reads if fully overlapping. Raw counts were then normalized to fragments per kilobase of exon per million fragments mapped (FPKM) expression values. FPKM values were log₂ transformed with an offset of 1, and normalized by gene for heat-map visualization. Change in gene expression was quantified by calculating the log₂ fold change, comparing IFNγ treated with untreated cell lines. Genes were annotated as processes and pathways using the MSigDB (67), Kyoto Encyclopedia of Genes and Genomes (KEGG; ref. 68), and Reactome (69) gene sets. Heat maps were created using the R statistical language (https://www.R-project.org) and the ggplot2 (70) package.

Selected cell lines were maintained in 10-cm culture dishes and analyzed when 70%–80% confluent. Western blot analysis was performed as described previously (71). Primary antibodies included JAK1, JAK2, B2M, and GAPDH (all were obtained from Cell Signaling Technology). Immunoreactivity was analyzed with the ECL-Plus Kit (Amersham Biosciences Co.), using the ChemiDoc MP System (Bio-Rad Laboratories). Experiments were performed at least twice for each cell line.

C57BL/6 mice were bred and kept under defined flora pathogen–free conditions at the Association for the Assessment and Accreditation of Laboratory Animal Care–approved animal facility of the Division of Experimental Radiation Oncology, UCLA, and used under the UCLA Animal Research Committee protocol #2004-159-43I. The MC38 cell line was originally generated at the NCI Surgery Branch (originally labeled as Colo38), and was obtained from Dr. Robert Prins (Department of Neurosurgery, UCLA). The B16-F10 mouse melanoma cell lines were purchased from ATCC. The MC38 cell line, B16-F10 mouse melanoma cell line, and established knockout cell lines were cultured at 37°C with 5% CO₂ in DMEM (Invitrogen) supplemented with 10% FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, and 0.25 μg/mL amphotericin B. Cell lines were confirmed Mycoplasma-negative using MycoAlert Mycoplasma Detection Kit (Lonza) and periodically tested for authentication. For in vivo experiments, early-passage cell lines were used (less than 10 passages).

Antibodies for in vivo experiments: anti-mouse PD-1 (clone RMP1-14), anti-mouse CD8 (clone YTS 169.4, BE0117), anti-mouse CD4 (clone GK1.5, BE0003), anti-mouse NK1.1 (clone PK136, BE0036), and isotype control antibody (clone 2A3, BE0089), all from BioXCell. The CpG-C oligodeoxynucleotide SD-101 were obtained under a material transfer agreement with Dynavax. SD-101 was synthesized and purified by standard techniques as described previously (72). Bempegaldesleukin (NKTR-214; ref. 40) was provided by Nektar Therapeutics. Both were diluted in the recommended product formulation buffer for in vivo studies.

To establish subcutaneous tumors, 0.3 × 10⁶ of MC38 wild-type or established JAK1-, JAK2-, B2M-knockout cells per mouse were injected into the flanks of C57BL/6 mice. When tumors became palpable (day 5 or 6), four doses of 300 μg of anti–PD-1 or isotype control antibody were injected intraperitoneally every 3 days. Intratumoral treatment of SD-101 was used at 50 μg per injection. Bempegaldesleukin (NKTR-214) was used at 0.8 mg/kg every 9 days × two doses, intravenous (tail vein). For depletion studies, 300 μg of anti-CD8, 300 μg of anti-CD4, 300 μg of anti-NK1.1, or the combination were administered every 3 days starting the day before SD-101 or bempegaldesleukin until the end of the experiment. Splenocytes from control and depleted corresponding mice were taken to validate depletion efficacy. Tumors were followed from caliper measurements two or three times per week and tumor volume was calculated using the following formula: tumor volume = [(width)² × length]/2. Mean and SE of the tumor volumes per group was calculated.

MC38 wild-type or established JAK1-, JAK2-, and B2M-knockout tumor cells (0.3 × 10⁶) were implanted into the flanks of C57BL/6 mice. On day 13 following inoculation, tumors were harvested from mice at predefined treatment. Tumors were digested using the Tumor Dissociation Kit Mouse (Miltenyi Biotec). Spleens were dissociated and filtered with a 70-μm filter following digestion with the ACK Lysis Buffer (Lonza). Sample staining and data acquisition were performed as described previously (34) except that 3% paraformaldehyde was used and samples were not barcoded. A full list of immune markers used is described in Supplementary Table S2. Samples were analyzed using Fluidigm Helios Mass Cytometry System at the UCLA Flow Cytometry core. Samples were manually gated for cells, singlets and double expression of the viable CD45 single-cell–positive population (Supplementary Fig. S5B) using FlowJo software (version 10.4.2), and data files were analyzed using Cytofkit package (ref. 73; R version 3.5.1). To identify and annotate each of the clusters obtained, cluster median data were normalized, and a threshold of >0.5 was used to define positive immune markers. T-distributed stochastic neighbor embedding plots were generated by PhenoGraph clustering through cytofkiyShinyAPP from Cytofkit. We used FlowSom, an unsupervised automated algorithm which orders cells according to their phenotypic similarities. FlowSom clustered the T-CD8 cells into three branches and thus distinguished exhausted T-CD8 populations.

MetaCyto was performed using the default workflow for both flow and mass cytometry (CyTOF) data as described previously (39). By combining and clustering the markers for all datasets, MetaCyto was able to identify and track specific immune cell populations present across heterogeneous studies. We included the MetaCyto automated unsupervised clusters determined from the aggregated experiments and additionally supplied a list of predetermined functional clusters as listed in Supplementary Table S3 for the analysis. For each phenotype, we performed a regression analysis to estimate effect sizes and P values of the treatment relative to isotype in each immune cluster. Results were plotted via Prismv8 Software (GraphPad). MetaCyto is available as an R package on Bioconductor (http://bioconductor.org/packages/release/bioc/html/MetaCyto.html).

Total RNAs were extracted using the PureLink RNA Mini Kit (Invitrogen). CXCL9 and CXCL10 expression were measured by reverse transcription PCR following the manufacturer's protocol for the Power SYBR Green RNA-to-CT 1-Step Kit (Applied Biosystems) and using the following primers: CXCL9: 5′-CCCAATTGCAACAAAACTGA-3′ and 5′-AGTCCGGATCTAGGCAGGTT-3′ and CXCL10: 5′-AATCATCCCTGCGAGCCTAT-3′ and 5′-TTTTTGGCTAAACGCTTTCAT-3′.

G. Abril-Rodriguez is a consultant at Arcus Biosciences. K.M. Campbell has advisory board relationship with Geneoscopy LLC. B. Comin-Anduix is a consultant at Advarra. S. Hu-Lieskovan reports receiving speakers bureau honoraria from Amgen, BMS, Xencor, Genmab, and Merck. A. Ribas is a consultant at Amgen, Merck, Chugai, Novartis, and Sanofi, is a SAB member at RAPT, Imaginab, Isoplexis, Mapkure, Merus, Rgenix, Tango, Five Prime, Apricity, Compugen, and Cytomix, is a board member at Lutris, is a board member and consultant at Arcus and PACT, reports receiving commercial research grants from Agilent and Bristol-Myers Squibb, and has ownership interest (including patents) in Arsenal Bio. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D.Y. Torrejon, B. Comin-Anduix, S. Hu-Lieskovan, A. Ribas

Development of methodology: D.Y. Torrejon, G. Abril-Rodriguez, G. Parisi, A. Garcia-Diaz, A. Vega-Crespo, C.M. Lee, B. Berent-Maoz, B. Comin-Anduix, S. Hu-Lieskovan, A. Ribas

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.Y. Torrejon, A.S. Champhekar, A. Kalbasi, J.M. Zaretsky, G. Cheung-Lau, T. Wohlwender, P. Krystofinski, A. Vega-Crespo, P. Mascaro, S. Hu-Lieskovan, A. Ribas

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.Y. Torrejon, G. Abril-Rodriguez, A.S. Champhekar, J. Tsoi, K.M. Campbell, J.M. Zaretsky, C. Puig-Saus, C.S. Grasso, B. Comin-Anduix, S. Hu-Lieskovan, A. Ribas

Writing, review, and/or revision of the manuscript: D.Y. Torrejon, G. Abril-Rodriguez, A.S. Champhekar, J. Tsoi, K.M. Campbell, A. Kalbasi, J.M. Zaretsky, C. Puig-Saus, B. Comin-Anduix, S. Hu-Lieskovan, A. Ribas

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.Y. Torrejon, A. Vega-Crespo, B. Berent-Maoz, S. Hu-Lieskovan, A. Ribas

Study supervision: D.Y. Torrejon, S. Hu-Lieskovan, A. Ribas

This study was funded in part by the Parker Institute for Cancer Immunotherapy, NIH grants R35 CA197633 and P01 CA244118, the Ressler Family Fund, and support from Ken and Donna Schultz (all to A. Ribas). D.Y. Torrejon was supported by a Young Investigator Award from ASCO, a grant from the Spanish Society of Medical Oncology for Translational Research in Reference Centers, and the V Foundation-Gil Nickel Family Endowed Fellowship in Melanoma Research. It has been developed within the framework of a medical doctorate at the Autonomous University of Barcelona. G. Abril-Rodriguez was supported by the Isabel & Harvey Kibel Fellowship Award and the Alan Ghitis Fellowship Award for Melanoma Research. J. Tsoi and K.M. Campbell were supported by the NIH Ruth L. Kirschstein Institutional National Research Service Award #T32-CA009120 and the UCLA Tumor Immunology Training Grant (T32CA009120). G. Parisi was supported by the V Foundation-Gil Nickel Family Endowed Fellowship in Melanoma Research. J.M. Zaretsky was in the UCLA Medical Scientist Training Program supported by NIH training grant GM08042. S. Hu-Lieskovan was supported by a Conquer Cancer Foundation ASCO Career Development Award, a UCLA KL2 Translational Research Award, and a Melanoma Research Alliance Young Investigator Award. Flow and mass cytometry were performed in the UCLA Jonsson Comprehensive Cancer Center (JCCC) and Center for AIDS Research Flow Cytometry Core Facility that is supported by NIH awards P30 CA016042 and 5P30 AI028697, and by the JCCC, the UCLA AIDS Institute, and the David Geffen School of Medicine at UCLA. The purchase of the Helios mass cytometer that was used in this work was supported, in part, by funds provided by the James B. Pendleton Charitable Trust. We want to thank Dr. Cristiana Guiducci from Dynavax and Drs. Deborah Charych and Willem Overwijk from Nektar Therapeutics for helpful guidance in the performance of in vivo studies with SD-101 and bempegaldesleukin, respectively.

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.