PAK4 inhibition improves PD-1 blockade immunotherapy

Lack of tumor infiltration by immune cells is the main mechanism of primary resistance to programmed cell death protein 1 (PD-1) blockade therapies for cancer. It has been postulated that cancer cell-intrinsic mechanisms may actively exclude T cells from tumors, suggesting that the finding of actionable molecules that could be inhibited to increase T cell infiltration may syn-ergize with checkpoint inhibitor immunotherapy. Here, we show that p21-activated kinase 4 (PAK4) is enriched in non-respond-ing tumor biopsies with low T cell and dendritic cell infiltration. In mouse models, genetic deletion of PAK4 increased T cell infiltration and reversed resistance to PD-1 blockade in a CD8 Tcell-dependent manner. Furthermore, combination of anti-PD-1 with the PAK4 inhibitor KPT-9274 improved anti-tumor response compared with anti-PD-1 alone. Therefore, high PAK4 expres-sion is correlated with low T cell and dendritic cell infiltration and a lack of response to PD-1 blockade, which could be reversed with PAK4 inhibition.mmune checkpoint blockade therapies have significantly altered To address this issue, we compared the transcriptional landscapethe current landscape of cancer treatment1. Programmed cell of tumor biopsies from patients with advanced melanoma treatedIdeath protein 1 (PD-1) blockade induces major and durable anti- with anti-PD-1 immunotherapy. Here, we report on p21-activatedtumor response by releasing the PD-1/programmed death-ligand 1 kinase 4 (PAK4) as an actionable target that could be inhibitedcheckpoint that blocks the effector functions of anti-tumor T cells2. in combination with immune checkpoint blockade therapies toHowever, this approach has limited activity in patients with cancers increase immune cell infiltration and overcome primary resis-that lack pre-existing immune cell infiltration, which is the primary tance to these therapies. PAK4 is a kinase known to be involved inmechanism of resistance to PD-1 blockade therapy3–6. Exclusion of tumorigenesis that directly binds and phosphorylates a specific sitetumor infiltration by T cells could be mediated by several mecha- in β-catenin to activate WNT signaling15–18. Our work shows that:nisms that result in failure to attract or retain antigen-specific (1) PAK4 expression is enriched in non-responding tumor biopsiesT cells in tumors, such as a lack of antigenic mutations, alterations with low immune cell infiltration; (2) genetic and pharmacologicin the antigen processing machinery, loss of human leukocyte anti- PAK4 inhibition improve response to PD-1 blockade in vivo; andgen expression, and disruption of the interferon (IFN) signaling (3) this provides a novel therapeutic strategy that may improve thepathway that is needed to amplify the anti-tumor T cell response7. efficacy of immune checkpoint inhibitor therapies.

Results
Resistance to PD-1 blockade is associated with lack of immunecell infiltration. To identify drivers of resistance to immunother-apy, we generated transcriptome data from biopsies of 41 patientswith advanced melanoma treated with PD-1-blocking antibody.We sequenced a total of 27 baseline and 33 on-treatment biopsies,including 14 non-responding and 13 responding samples (Fig. 1a andSupplementary Table 1). We removed two samples because CD8Agene expression did not agree with CD8 protein levels measuredusing immunohistochemistry (IHC) (Supplementary Fig. 1a–c),and four samples based on their outlier keratinocyte biomarkergene expression of KRT15 and KRT5, which indicated that these biopsies mostly consisted of keratinocytes and did not have enough melanoma content19 (Supplementary Fig. 1d,e). On-treatment biop-sies taken from patients with a response to PD-1 blockade showed increased expression of a CD8 T cell effector signature including CXCL9, CXCL10, GZMB, PRF1, GZMA, CD8A, TBX21, IFNG and TNF (Fig. 1b; P = 1 × 10−4), consistent with previous data2–5. Paired t-tests with matched samples also confirmed that only biopsies from patients with a clinical response to PD-1 blockade exhibited signifi-cant increases in the expression of markers of immune response (Extended Data Fig. 1). We applied gene set enrichment analysis (GSEA) using the Gene Ontology gene sets to demonstrate that, unlike non-responding biopsies, the genes significantly increased in on-treatment responding biopsies were enriched in signatures associated with an adaptive immune response (Fig. 1c). We further identified immune genes that were upregulated in on-treatment responding biopsies relative to on-treatment non-responding biop-sies. As expected, CD8A, TNF, GZMA and IFNG, among other immune genes, were expressed at higher levels in biopsies from patients who responded to PD-1 blockade therapy (Fig. 1d and Extended Data Fig. 1). To estimate the relative abundances and diversity of the different immune cells present in the tumor biop-sies, we performed RNA sequencing (RNA-Seq)-based immune cell deconvolution using the microenvironment cell populations coun-ter (MCP-counter)20. Responding biopsies were significantly infil-trated with T cells, CD8 T cells, myeloid dendritic cells and natural killer cells compared with non-responding tumor biopsies (Fig. 1e and Extended Data Fig. 1). Altogether, on-treatment biopsies from patients with a response to therapy present the characteristic fea-tures of an adaptive immune response, while on-treatment biopsies from patients without a response mostly lack T cell infiltration.PAK4 expression is enriched in poorly infiltrated tumor samples and constitutes a potential target to improve PD-1 blockade immunotherapies.

Because immune cell exclusion was a com-mon factor among non-responding biopsies, we sought to deter-mine tumor-intrinsic drivers of T cell exclusion by comparing immune-infiltrated tumor biopsies with immune-excluded ones. Differential gene expression analysis revealed that only 18 overlap-ping genes were enriched in biopsies with both a low dendritic cell score and a low T cell score (log2[fold change] > 1 and false discov-ery rate < 5 × 10−5; Supplementary Table 2c). Among these genes, we were interested in studying an actionable gene whose function could be blocked by a drug. PAK4 stood out among the list of 18 genes as its expression was consistently higher in tumor biopsies with low infiltration with dendritic cells (adjusted P value (q) < 0.0001) and T cells (q < 0.0001) (Fig. 2a,c and Supplementary Table 2a), as well as in tumor biopsies with low expression of CD8A, TNF and IFNG (Fig. 2c and Supplementary Table 2b). The correlation with low intratumoral T cell and dendritic cell infiltration was validated in a published cohort of 99 biopsies analyzed by RNA-Seq5 (Fig. 2b).Furthermore, tumors with high expression of PAK4 were enriched and positively correlated with a signature of immune cell exclusion reported by Jerby-Arnon et al.6, based on analysis of 33 melanoma biopsies using single-cell RNA-Seq (Extended Data Fig. 2a,b). PAK4 is a serine/threonine kinase that functions downstream of the small GTPases CDC42 and RAC, and plays an important role in several signaling pathways involved in tumorigenesis, including a known function of phosphorylating β-catenin, and shuttling with it into the nucleus to activate the WNT/β-catenin pathway15–18,21. This func-tion of PAK4 seemed relevant based on previous work by Spranger et al.11 showing that tumor-intrinsic β-catenin signaling can impair T cell infiltration in melanoma.PAK4 negatively correlated with immune markers of an active CD8 T cell response, including CD8A, TNF, GZMA and PRF1, as well as with transcriptome signatures of different immune cell populations, such as T cells, CD8 T cells, cytotoxic T cells and den-dritic cells, in both our cohort and the Riaz et al.5 validation cohort (Fig. 2d and Extended Data Fig. 2c). To determine whether PAK4 was expressed by melanoma cancer cells, we performed IHC on on-treatment tumor biopsies. Indeed, PAK4 co-localized with the melanoma marker S100 (Fig. 2e and Extended Data Fig. 2d). In addition, IHC analysis validated the inverse correlation between PAK4 and CD8 T cell infiltration observed by RNA-Seq (Fig. 2e).

Overall, our data suggest that tumor-intrinsic PAK4 expression is associated with a lack of immune cell infiltration, and constitutes a potential target to overcome PD-1 blockade resistance.PAK4 expression correlates with WNT/β-catenin pathway acti-vation in melanoma tumor biopsies and regulates its activation in vitro. Given the evidence relating WNT signaling, immune infiltration and a lack of response to checkpoint blockade immuno-therapies in melanoma and other solid tumors, and consistent with the known relationship between PAK4 and WNT signaling15–18,21, we further investigated the role of PAK4 in the β-catenin path-way using clinical tumor samples. Tumor biopsies with high PAK4 expression had increased levels of MYC and CTNNB1 compared with tumor biopsies with low PAK4 expression (Fig. 3a). Tumors with high expression of PAK4 were also enriched for and positively correlated with a previously reported WNT signature that includes APC, MYC, CTNNB1, DKK2 and VEGFA12 (Fig. 3b). Furthermore, IHC analysis of the on-treatment tumor biopsies also showed that β-catenin co-localized with PAK4 (Fig. 3c). Of note, the PAK4 overlap with β-catenin was higher in the two tumor biopsies with low T cell infiltration, suggesting that there may be a dual require-ment of β-catenin and PAK4 to induce a T cell-excluded phenotype (Extended Data Fig. 2e).To directly investigate the impact of PAK4 deletion on WNT signaling, we first generated PAK4 knockout (KO) sublines of the murine melanoma B16 using CRISPR–Cas9 (three sublines: B16 KO 6.2, B16 KO 8.1 and B16 KO 8.2; Extended Data Fig. 3a–d). To quantify WNT signaling activation, B16 PAK4 KO cells wereas in the Riaz et al.5 validation cohort (b; n = 50 biopsies; q = 1.59 × 10−11). c, PAK4 expression was also enriched in samples with low T cell infiltration (q = 2.74 × 10−7), and low expression of CD8A (q = 9.08 × 10−9), TNF (q = 6.67 × 10−12) and IFNG (q = 1.9 × 10−6) (n = 15 biopsies per group for each comparison).

In a–c, P values were calculated using the negative binomial generalized linear model fitting and Wald significance test, while q values were obtained by applying the Benjamini–Hochberg method. d, PAK4 expression negatively correlates with log2[FPKM] expression of the known immune markers CD8A (r = −0.54; P = 7.95 × 10−6), TNF (r = −0.69; P = 1.12 × 10−9), GZMA (r = −0.59; P = 7.95 × 10−7) and PRF1 (r = −0.41; P = 6.20 × 10−4), as well as the different immune populations assessed using MCP-counter: T cells (r = −0.62; P = 1.04 × 10−7), CD8 T cells (r = −0.55; P = 5.25 × 10−6), cytotoxic lymphocytes (r = −0.46; P = 1.90 × 10−4) and dendritic cells (r = −0.49; P = 6.60 × 10−5) (n = 60 biopsies for all correlations). Correlations were calculated applying Pearson’s correlation coefficient test. e, Images from biopsies of two representative patients of non-responding/low T cell infiltration (top) and responding/high T cell infiltration (bottom). Slides were stained with S100, PAK4 and CD8. The results showed co-localization of PAK4 and S100, and validation of the exclusivity between PAK4 and CD8 expression. Scale bars: 100 µm.transfected with the Topflash luciferase reporter, which is under the control of consensus T cell factor-binding sites22,23. Whereas Wnt-3a exposure induced the Topflash luciferase activity in B16wild-type (WT) CRISPR control cells, the induction of Topflash luciferase activity by Wnt-3a was reduced in B16 PAK4 KO cells (Fig. 3d and Extended Data Fig. 4a). Rescuing WT PAK4 expressiongeometric mean of the following WNT-related genes: APC, MYC, CTNNB1, DKK2 and VEGFA. In a and b, P values were determined by two-sided Welch’s t-test (a and b (left)) and Pearson’s correlation coefficient (b (right)) (**P < 0.01; ***P < 0.001; ****P < 0.0001). From top to bottom, box plots show the maximum, third quartile, median, first quartile and minimum values. c, Images from biopsies of two representative patients of non-responding/low T cell infiltration (top) and responding/high T cell infiltration (bottom). Slides were stained with β-catenin, PAK4 and CD8. Scale bars: 100 µm. d,e, Topflash WNT activity assays indicating that B16 PAK4 KO cells failed to upregulate WNT signaling as high as PAK4 CRISPR control WT cells (d; P = 0.0054 for B16 WT CRISPR control (CC) versus B16 KO 6.2; P = 0.0026 for B16 WT CC versus B16 KO 8.1; P = 0.0033 for B16 WT CC versus B16 KO 8.2), while rescuing PAK4 expression increased basal WNT activity (e; P = 0.0002 for B16 KO 6.2 versus B16 KO 6.2 rescue; P < 0.0001 for B16 KO 8.1 versus B16 KO 8.1 rescue; P = 0.0004 for B16 KO 8.2 versus B16 KO 8.2 rescue) (n = 3 technical replicates per group). The results are representative of three independent experiments.

In d and e, data represent means ± s.e.m. and the results were compared by two-tailed unpaired t-test. f, Immunoblot for β-catenin S675 phosphorylation. Phosphorylation levels were decreased in B16 PAK4 KO cells compared with PAK4 WT cells and restored in PAK4 rescue cell lines. The results are representative of three independent experiments. Source data are available for d–f.a, Pan-cancer analysis using TCGA transcriptome data shows the negative correlation between PAK4 expression and T cell (blue), cytotoxic T cell (red) and dendritic cell scores (yellow) across 32 tumor types (the sample size for each cancer type and the associated P value for each correlation can be found in Supplementary Table 3). Correlations were evaluated using Spearman’s correlation coefficient. DLBC, diffuse large B-cell lymphoma. b,c, On-treatment non-responding biopsies (n = 14) have higher levels of log2[FPKM] PAK4 expression compared with responding biopsies (b; n = 13; P = 4.72 × 10−3), and are enriched in gene signatures related to known oncogenic signatures involved in immune cell exclusion, as observed by GSEA using Gene Ontology gene sets as targets (c). From top to bottom, box plots in b show the maximum, third quartile, median, first quartile and minimum values, and the P value was determined by two-sided Welch’s t-test (**P < 0.01).in B16 PAK4 KO cells increased baseline WNT activity (Fig. 3e and Extended Data Fig. 3e), although PAK4 deletion did not affect WNT activity at steady state (Extended Data Fig. 4b). In addition, PAK4 deletion decreased nuclear β-catenin phosphorylation at serine 675 (S675), which was restored in PAK4 rescue cell lines (Fig. 3f). Of note, neither PAK4 deletion nor overexpression affected the levels of β-catenin nuclear protein (Extended Data Fig. 4c). Moreover, PAK4 inhibition with the dual PAK4 and nicotinamide phosphoribosyl-transferase (NAMPT) inhibitor KPT-9274 (refs. 24–27) recapitulated the results observed with the B16 PAK4 KO clones as it diminished nuclear S675 β-catenin phosphorylation and decreased sensitiv-ity to Wnt-3a, while it did not affect WNT activity at steady state, nor nuclear β-catenin protein levels (Extended Data Fig. 5a–c).

Furthermore, B16 PAK4 KO cell lines decreased tyrosinase expression (Extended Data Fig. 5d) and lost their pigmentation when cultured over time (Extended Data Fig. 5e)—a phenotype that is consistent with the suggested role of PAK4 in melanogenesis and the β-catenin/MITF (microphthalmia-associated transcription fac-tor) pathway17. Taken together, our results validate the association between PAK4 expression and WNT/β-catenin pathway activation, and provide evidence that genetic deletion and pharmacological inhibition of PAK4 impair Wnt/β-catenin pathway signaling in vitro.PAK4 negatively correlates with immune cell infiltration across human cancers. We then investigated whether the association between PAK4 expression and the lack of T cell infiltration in mela-noma tumor biopsies could be expanded to other tumor types. To do so, we analyzed transcriptome data from 32 different cancer types in The Cancer Genome Atlas (TCGA), and calculated the cor-relation between PAK4 expression and T cell, cytotoxic T cell and dendritic cell scores generated using MCP-counter20 in all of the samples for each cancer type. In addition to cutaneous melanoma, we observed a negative correlation with T cell infiltration in the majority of cancer types (18 out of 32), including cancers that are notoriously resistant to anti-PD-1 therapy, such as prostate cancer, adrenocortical carcinoma, germ cell cancers and glioblastoma mul-tiforme (Fig. 4a and Supplementary Table 3). In line with published data, one of the strongest negative correlations was in pancreatic cancer, where a pan-PAK inhibitor had previously been shown to enhance anti-tumor immune response in a preclinical model28.Lack of response to PD-1 blockade is associated with increased PAK4 expression and is enriched in oncogenic pathways involved in immune cell exclusion. As PAK4 showed a strong inverse correla-tion with both dendritic cells and T cells in melanoma, we reasonedthat tumor biopsies from patients without a response to anti-PD-1 may have enriched PAK4 expression. Indeed, non-responding biopsies had higher levels of PAK4 transcripts (P = 0.004; Fig. 4b). We also investigated whether our cohort of tumor biopsies non-responding to PD-1 blockade therapy recapitulated known onco-genic mechanisms of T cell exclusion10.

To test this hypothesis, we compared on-treatment non-responding biopsies with responding biopsies, and applied GSEA using the curated gene sets. Signatures enriched in non-responding biopsies included gene sets related to WNT/β-catenin signaling and the WNT target gene MYC (Fig. 4c and Supplementary Table 4). Overall, biopsies from patients with-out a response to PD-1 blockade were enriched for PAK4 expression and gene signatures related to known oncogenic pathways involved in T cell exclusion10.Genetic KO of PAK4 sensitizes tumors to PD-1 blockade and increases immune cell infiltration. If PAK4 plays an active role in excluding tumor-specific T cells from the tumor microenvironment of melanoma biopsies, PAK4 inhibition would increase tumor-spe-cific T cell infiltration and hence sensitize tumors to PD-1 block-ade therapy. To test this hypothesis, we used the murine melanoma model B16, which exhibits primary resistance to PD-1 blockade29, lacks previous infiltration by tumor-specific lymphocytes30, andintrinsically expresses the immune resistance program defined by Jerby-Arnon6. To assess the anti-tumor efficacy of PD-1 blockade in the context of PAK4 deletion, we treated syngeneic C57BL/6 mice bearing B16 PAK4 KO or B16 WT tumors with a murine anti-PD-1 antibody. We observed the anti-tumor activity of PD-1 blockade only in melanoma tumors lacking PAK4 expression (Fig. 5a,b and Extended Data Fig. 6a,b). Of note, untreated B16 PAK4 KO tumors grew progressively, suggesting that although PAK4 deletion pro-vides sensitization to PD-1 blockade therapy, it is not sufficient by itself in the B16 model to trigger an anti-tumor immune response. In addition, restoring PAK4 protein levels in B16 PAK4 KO tumors resulted in the loss of PD-1 blockade anti-tumor efficacy (Fig. 5c and Extended Data Fig. 6c). To elucidate whether the observed response to anti-PD-1 was CD8 dependent, we depleted CD8 T cells in syngeneic C57BL/6 mice bearing B16 PAK4 KO tumors.

CD8 depletion completely abrogated the anti-tumor activity of mouse anti-PD-1, showing that PAK4 deletion sensitized melanoma B16 tumors to PD-1 blockade in a CD8 T cell-dependent manner (Fig. 5d and Extended Data Fig. 6d). These results suggest that genetic PAK4 deletion allows the infiltration of tumor-specific T cells that confer anti-tumor efficacy on PD-1 blockade.To validate whether PAK4 deletion facilitates immune cell infiltration, we performed immune profiling of tumor-infiltratingimmune cells using cytometry by time of flight (CyTOF), and iden-tified a total of 16 independent cell clusters (Fig. 6a). The T cell population was defined by three clusters, including a non-T regula-tory CD4 T cell cluster positive for CD3e, CD4, IFN-γ and Ki-67, a CD8 T cell cluster positive for CD3e, CD8a, Tbet and Ki-67, and a general T cell cluster positive for CD3e. A natural killer cluster positive for CD335 and CD161 was also identified. B16 PAK4 KO anti-PD-1-treated tumors presented increased infiltration of T and natural killer cells compared with B16 WT anti-PD-1-treated tumors (P = 0.049; Fig. 5e,f). Interestingly, untreated B16 PAK4 KO tumors already presented increased T and natural killer cell infiltration compared with B16 WT untreated tumors (P = 0.02; Fig. 5e,f), although we did not observe anti-tumor efficacy in the B16 PAK4 KO group (Fig. 6b). Consistently, B16 PAK4 KO tumors had increased levels of T cells regardless of treatment with murine anti-PD-1 (P = 0.009; Fig. 5g). Therefore, these data support the hypothesis that PAK4 depletion increases tumor-specific T cell infil-tration, which sensitizes tumors to PD-1 blockade.Pharmacological inhibition of PAK4 synergizes with PD-1 blockade immunotherapy. KPT-9274 is a dual PAK4 and NAMPT inhibitor24–27 currently in clinical trials. We tested whether treat-ment with KPT-9274 recapitulates the anti-tumor effects of genetic PAK4 deletion to sensitize B16 melanoma to murine anti-PD-1 therapy. Indeed, B16 murine melanoma tumors treated with anti-PD-1 in combination with KPT-9274 showed a stron-ger anti-tumor effect compared with anti-PD- 1 (P = 0.01; Fig. 7a) and KPT- 9274 monotherapy (P = 0.0007; Fig. 7a).

To expand the testing to other settings of partial anti-PD-1 therapy resistance, we used the MC38 mouse colon adenocarcinoma model, which is a model of a cancer with high tumor mutation burden and is partially sensitive to PD-1 blockade, but with ample margin for improvement as tumors grow progressively after a period of tran-sient response31,32. Consistent with being an immunogenic tumor model, and with PAK4 deletion per se facilitating T cell infiltra-tion (Fig. 5f,g), both MC38 WT tumors, treated with either a combination of KPT-9274 and anti-PD-1 or KPT-9274 alone, showed decreased tumor growth compared with the anti-PD-1 monotherapy group (Fig. 7b). We generated a PAK4 KO subline of MC38 through CRISPR–Cas9 gene editing (Extended Data Fig. 7a,b), and consistent with the results with KPT- 9274, MC38 PAK4 KO tumors achieved tumor regression even in the absence of anti-PD-1 therapy (Fig. 7c). Of note, MC38 PAK4 KO tumors only reached complete regression (n = 3) when treated with anti-PD-1, suggesting that PD-1 blockade also improves anti-tumorTcell response in the setting of partial response to anti-PD-1 therapy. In addition, we found that MC38 PAK4 KO clones were more sensitive to the anti-proliferative effects of tumor necrosis factor (TNF), which is consistent with the current literature33,34, and could contribute to the phenotype observed in this model (Extended Data Fig. 7c). Altogether, these data suggest that PAK4 inhibition synergizes with anti-PD-1 treatment.

Discussion
By studying the transcriptional landscape of biopsies of patients with melanoma treated with PD-1 blockade immunotherapy, we found that the expression of PAK4 is associated with immune exclu-sion and lack of clinical response. Genetic and pharmacological PAK4 inhibition altered WNT/β-catenin signaling, increased intra-tumoral T cell infiltration and improved the response to checkpoint blockade therapy in two mouse models. The negative correlation between PAK4 expression and T cell infiltration held true across several human cancers, including cancers notoriously resistant to PD-1 blockade, and hence expands the potential clinical applicabil-ity of the combined inhibition of PAK4 and PD-1.Finding novel molecular targets that could improve and overcome resistance to PD-1 blockade therapy remains one of the main challenges that needs to be tackled to increase the effi-cacy rate of cancer immunotherapies10. Our current work validates and builds on the fundamental knowledge that PD-1 blockade works by unleashing the immune breaks of a pre-existent tumor-specific T cell population2. The transcriptional characterization of melanoma tumor samples highlighted the common denomina-tor among biopsies of non-responding patients—the lack of a proper immune T cell infiltration of the tumor microenvironment. Therefore, interventions that increase the immunogenicity and the immune infiltration within the tumor microenvironment remain a top therapeutic priority7.The immune system exercises a selective pressure that shapes cancer evolution and results in the selection of malignant cells able to escape an immune cell attack. This has been termed can-cer immunoediting35. Cancer cells can exploit and rewire oncogenic signaling pathways that alter the immunogenicity of the tumor and confer an advantage against the immune system10.

Among the dif-ferent oncogenic signaling pathways, there is compelling evidence from several studies that associate an active WNT/β-catenin sig-naling pathway with immune exclusion and resistance to immune checkpoint blockade therapy11–13. In this context, among the list of differentially expressed genes in non-immune-infiltrated biopsies, the appeal of focusing our studies on PAK4 became more relevant given its previously reported involvement in the WNT/β-catenin pathway15,18. PAK4 deletion disrupted WNT signaling activity with-out altering β-catenin protein levels, which suggests that PAK4 may be regulating WNT activity by means other than the number of molecules that translocate into the nucleus, such as alteration of the interaction of β-catenin with other proteins that are important in regulating β-catenin transcriptional activity. Overall, our findings provide novel insights into the mechanism by which PAK4 impacts WNT signaling activity. However, it still remains necessary to fully elucidate how PAK4-induced WNT inhibition contributes to over-coming PD-1 blockade resistance.PAK4 overexpression or increased activity is associated with the development and progression of several tumor malignancies, including melanoma36, pancreatic cancer37 and prostate cancer38. In breast cancer, PAK4 messenger RNA levels are also often overexpressed39,and increasing AKT phosphorylation42–45.

Interestingly, PAK4 also presents kinase-independent functions. It has been reported that PAK4 acts as a scaffold to facilitate TNF receptor type 1-associated death domain (TRADD) protein binding to the TNF receptor to promote TNF survival activity33. Therefore, PAK4 protein inhibi-tion increases TNF-induced apoptosis—an observation that is in line with our results seen in the MC38 model34. The role of PAK4 and TNF signaling in PD-1 blockade sensitivity necessitates a better understanding and will need to be explored further in the future. In these studies, we primarily used the B16 murine melanoma model as it has previously been reported that it is poorly inflamed and resistant to checkpoint blockade29. Of note, we acknowledge that there may be additional mouse models that have not been used in this study and that could help expand our understanding of the role of PAK4 in immune cell exclusion, such as the BrafV600E/Pten−/−/ CAT-STA mice used in the Spranger et. al. study11.This particular mouse model has been used to study WNT-induced T cell exclusion in melanoma, and has provided evidence that the oncogenic WNT/ β-catenin pathway mediates dendritic cell exclusion, resulting in immune evasion. However, it is a model that relies on the in situ generation of multiple oncogene-driven primary skin cancers on induction of the driver oncogenes. This does not allow the genetic testing of PAK4 inhibition, unless a new transgenic mouse with an inducible PAK4 inhibition is created and cross-bred with the already BrafV600E/Pten−/−/CAT-STA triple genetically engineered mice.

In summary, this study presents a potential new therapeutic strategy to overcome PD-1 blockade resistance. Having analyzed three patient biopsy-derived RNA-Seq datasets, we conclude that PAK4 expression is enriched in poorly infiltrated tumor samples and constitutes a target to reverse PD-1 blockade resistance. Our pan-cancer correlation analysis with multiple cancer histologies having an anti-correlation of PAK4 expression and T cell infiltra-tion suggests that patients with different cancers could potentially benefit from dual PAK4 and PD-1 inhibition. The results from this study have led to the planning of a phase 1 clinical trial combining the anti-PD-1 nivolumab with the dual PAK4 and NAMPT inhibi-tor KPT-9274 (NCT02702492).Patients, tumor biopsies and response assessment. Tumor biopsies were collected under University of California, Los Angeles (UCLA) Institutional Review Board approvals 11-001918 and 11-003066 from 41 patients with metastatic melanoma treated with either pembrolizumab or nivolumab. All patients signed a written informed consent form. Samples were immediately stored in RNAlater (Ambion) or snap frozen in liquid nitrogen for subsequent RNA extraction.
was assessed for each biopsy independently by A.R. Complete patient clinical information can be found in Supplementary Table 1. RNA isolation and RNA-Seq analysis. We obtained a total of 66 tumor samples from which we extracted RNA using the AllPrep DNA/RNA Mini Kit (Qiagen) and mirVana miRNA Isolation Kit (Ambion). Poly-A selection was used for library construction, and samples were sequenced using the Illumina HiSeq 2500 platform with a read length of 2 × 100 at the UCLA Technology Center for Genomics and Bioinformatics. Raw FASTQ files were aligned to the hg19 genome using HISAT2 version 2.0.4 (ref. 46) with the default parameters, and counted with HTSeq version 0.6.1 (ref. 47) with the intersection-nonempty mode (ambiguous reads were counted if fully overlapping).

Raw counts were then normalized to fragments per kilobase of transcript per million mapped reads (FPKM). Two tumor biopsies were excluded from the analysis due to discordancy with previous IHC analysis (Supplementary Fig. 1a–c). Four tumor biopsies were excluded based on the expression of KRT15 and KRT5 (Supplementary Fig. 1d,e). A total of 60 tumor biopsies were considered for transcriptomic analysis. RNA-Seq-based cell deconvolution of tissue-infiltrating and stromal populations was performed using MCP-counter20 with the default settings, and immune cell infiltration was defined using the upper and lower quartile scores for each of the obtained immune cell populations. Differential gene expression was performed based on the negative binomial distribution with the DESeq2 package48 using the default settings (Wald significance test). Principal component analyses were also performed, using the DESeq2 package47, on prior normalization of raw reads using the variance-stabilizing transformation (vst) function. To identify enriched signaling pathways, we utilized GSEA with the following gene sets: C2 Curated Gene Sets and C5 Gene Ontology Gene Sets49. Pan-cancer correlation analysis between PAK4 expression and immune cell infiltration (calculated using MCP-counter as described above) was performed using gene expression data from 32 tumor types from the TCGA Research Network (http://cancergenome.nih.gov/).

Cell lines and PAK4 CRISPR–Cas9 KO and rescue. Murine B16 and MC38 cells were maintained in DMEM and RPMI medium respectively, supplemented with 10% fetal bovine serum, 100 U ml−1 penicillin and 100 µg ml−1 streptomycin at 37 °C under a humidified atmosphere of 5% CO2. The following single guide RNAs targeting PAK4 were used: forward: 5′-TTCGAGCACCGTGTACACAC-3′; reverse: 5′-GTGTGTACACGGTGCTCGAA-3′. These were cloned into the pSpCas9(BB)-2A-GFP vector (Addgene) as described in Zheng’s protocol50. Cells were then transfected with PAK4-single guide RNA plasmid using lipofectamine 3000 (Thermo Fisher Scientific), and green fluorescent protein-positive cells were collected and single-cell sorted 48 h after transfection at the UCLA Flow Cytometry Core. Genomic DNA was isolated for each clone (NucleoSpin Tissue XS; Macherey Nagel), and after PCR amplifying the PAK4 sequence, we used the tracking of indels by decomposition (TIDE)51 web tool to evaluate and confirm knockout efficiency (Extended Data Figs. 3a–c and 7a). PAK4 deletion was also validated by western blot, performed as described previously52. PAK4 antibody (Proteintech) immunoreactivity was assessed with an ECL-Plus Kit (Amersham Biosciences) and analyzed using the ChemiDoc MP system (Bio-Rad Laboratories) (Extended Data Figs. 3d and 7b). To restore PAK4 levels in PAK4 KO cells, we cloned the mouse PAK4 open reading frame into a lentiviral vector containing Thy1.1. 293T cells were used for lentiviral particle generation, and B16 PAK4 KO cells were transduced at 20% confluency.

Then, 24 h after transduction, the media was changed and cells were expanded and sorted based on Thy1.1 expression. PAK4 expression was then validated by western blot (Extended Data Fig. 3e). WNT activity assays. β-catenin protein levels and phosphorylation were investigated by western blot performed as described previously51 using the following antibodies: β-catenin (catalog number: 9587); phospho-β-catenin (S675; catalog number: 9567) and phospho-β-catenin (S33/37/T41; catalog number: 9561) (all from Cell Signaling Technology). Cytoplasm and nuclear extraction were performed with NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Fisher Scientific) following the manufacturer’s protocol. For the Topflash WNT activity assay, cells were plated in 24-well plates and co-transfected with pSV-β-galactosidase control vector (PR-E1081; Promega) along with either pTopflash (Addgene; catalog number: 12456) or pFopflash (Addgene; catalog number: 12457). Then, 24 h after transfection, cells were treated with Wnt-3a (R&D Systems) at 200 ng ml−1. After 8 h, cells were harvested using Reporter Lysis Buffer (Promega; catalog number: PR-E4030) and the luciferase activity was measured using a Bright-Glo Luciferase Assay System (Promega; catalog number: PR-E2610) and ATG-019 Beta-Glo Assay System (Promega; catalog number: PR-E4720). The luciferase activity was normalized to its corresponding Beta-Glo activity to account for transfection efficiency.