Report

Genomic correlates of response to immune checkpoint therapies in clear cell renal cell carcinoma

See allHide authors and affiliations

Science  16 Feb 2018:
Vol. 359, Issue 6377, pp. 801-806
DOI: 10.1126/science.aan5951

SNF'ing out antitumor immunity

Immune checkpoint inhibitors induce durable tumor regressions in some, but not all, cancer patients. Understanding the mechanisms that determine tumor sensitivity to these drugs could potentially expand the number of patients who benefit (see the Perspective by Ghorani and Quezada). Pan et al. discovered that tumor cells in which a specific SWI/SNF chromatin remodeling complex had been experimentally inactivated were more sensitive to T cell–mediated killing. The cells were more responsive to interferon-γ, leading to increased secretion of cytokines that promote antitumor immunity. Miao et al. examined the genomic features of tumors from patients with metastatic renal cell carcinoma who had been treated with immune checkpoint inhibitors. Tumors harboring inactivating mutations in PBRM1, which encodes a subunit of the same SWI/SNF complex, were more likely to respond to the drugs.

Science, this issue p. 770, p. 801; see also p. 745

Abstract

Immune checkpoint inhibitors targeting the programmed cell death 1 receptor (PD-1) improve survival in a subset of patients with clear cell renal cell carcinoma (ccRCC). To identify genomic alterations in ccRCC that correlate with response to anti–PD-1 monotherapy, we performed whole-exome sequencing of metastatic ccRCC from 35 patients. We found that clinical benefit was associated with loss-of-function mutations in the PBRM1 gene (P = 0.012), which encodes a subunit of the PBAF switch-sucrose nonfermentable (SWI/SNF) chromatin remodeling complex. We confirmed this finding in an independent validation cohort of 63 ccRCC patients treated with PD-1 or PD-L1 (PD-1 ligand) blockade therapy alone or in combination with anti–CTLA-4 (cytotoxic T lymphocyte-associated protein 4) therapies (P = 0.0071). Gene-expression analysis of PBAF-deficient ccRCC cell lines and PBRM1-deficient tumors revealed altered transcriptional output in JAK-STAT (Janus kinase–signal transducers and activators of transcription), hypoxia, and immune signaling pathways. PBRM1 loss in ccRCC may alter global tumor-cell expression profiles to influence responsiveness to immune checkpoint therapy.

Immune checkpoint inhibitors such as nivolumab extend the survival of a subset of patients with metastatic clear cell renal cell carcinoma (ccRCC) (1). Whether specific genomic features of ccRCC are associated with clinical benefit is unclear. In contrast to other human tumor types that respond to immunotherapy—such as non–small cell lung cancer (NSCLC), melanoma, and microsatellite-unstable colorectal adenocarcinoma—ccRCC harbors a low burden of somatic mutations (25). Melanoma and NSCLC typically harbor 10 to 400 mutations per megabase, and these genetic variants can generate tumor-specific antigens (neoantigens) that stimulate a strong antitumor immune response (14). By contrast, ccRCC harbors an average of only 1.1 mutations/Mb (6, 7), yet it ranks highly among tumor types in terms of immune cytolytic activity (8), immune infiltration score, and T cell infiltration score in the tumor microenvironment (9). These observations led us to hypothesize that distinct molecular mechanisms underlie the immunologically active tumor microenvironment and responsiveness to immune checkpoint therapy in patients with ccRCC.

As part of a prospective clinical trial (10), we first analyzed pretreatment tumors from 35 patients with metastatic ccRCC on a clinical trial of anti–programmed cell death 1 receptor (anti–PD-1) therapy (nivolumab). Whole-exome sequencing (WES) from paired tumor and normal tissue was performed to identify genetic correlates of clinical benefit. To validate the findings, we analyzed an independent cohort of 63 patients with metastatic ccRCC treated with therapies blocking PD-1 (e.g., nivolumab) or its ligand PD-L1 (e.g., atezolizumab) (Fig. 1A and table S1A) (11).

Fig. 1 Cohort consolidation and clinical characteristics of the discovery cohort.

(A) Sample inclusion and exclusion criteria and computational workflow. (B) Clinical stratification by degree-of-objective change in tumor burden (y axis) and duration of PFS (x axis). One patient (RCC_99) is not shown because of a lack of tumor response data. Patient RCC_50 (indicated by *) was classified as CB despite PFS < 6 months because there was continued tumor shrinkage after an initial period of minor tumor progression (see fig. S2). (C) Mutation burden in the discovery cohort by response group. (D) Ratio of subclonal to clonal mutations, as estimated by the ABSOLUTE method, by response group. ns, not significant.

Baseline clinical and demographic features in the discovery cohort have been previously described (10). The subset of patients with complete pretreatment molecular profiling did not differ substantially in clinical or demographic features from patients whose data did not meet technical quality-control standards (fig. S1, A and B, and supplemental methods) or from the larger published cohort (10). Given previous evidence suggesting that refined clinical stratifications are necessary to assess clinical benefit from immune checkpoint blockade (12), we defined a composite response endpoint incorporating response evaluation criteria in solid tumors (RECIST) (13), radiographic tumor shrinkage, and progression-free survival (PFS) (Fig. 1B and table S1B). Clinical benefit (CB) patients included those with complete response (CR) or partial response (PR) as defined by RECIST 1.1 (i.e., tumor shrinkage >30% from baseline) (13) or stable disease (SD), if they had any objective reduction in tumor burden lasting at least 6 months. This modification to include some patients with SD is intended to differentiate those patients with naturally indolent disease (i.e., slow tumor growth not surpassing 20% of baseline tumor size) from those with tumor response to immune checkpoint inhibitors (14). No clinical benefit (NCB) patients experienced progressive disease (PD), as determined by RECIST 1.1 and were discontinued from immunotherapy within 3 months. All other patients were termed “intermediate benefit” (IB). One patient in the discovery cohort was classified as CB despite PFS < 6 months because there was continued tumor shrinkage (67% decrease from baseline tumor size) after an initial period of minor tumor progression, and the patient had overall survival exceeding 32 months (fig. S2, A and B). Consistent with prior observations (1), the dose of nivolumab, patient gender, and baseline PD-L1 immunohistochemical staining from metastatic biopsies did not predict patient overall survival (OS) after initiation of anti–PD-1 therapy (P > 0.05 for all; log-rank test) (fig. S3).

Mean exome-wide target coverage in the discovery cohort was 128-fold for tumor sequencing and 91-fold for matched germline sequencing (tables S1A and S2A). Overall, nonsynonymous-mutation burden was moderate in the discovery cohort (median 82 per exome, range 45 to 157). The tumors of patients with CB and those with NCB showed similar mutation burdens and intratumoral heterogeneity (Fig. 1, C and D, and table S1, C and D). Mutations and copy-number alterations affecting antigen-presentation machinery and human leukocyte antigen (HLA) class I alleles were uncommon and were present in tumors of both CB and NCB patients (fig. S4, A and B).

We next focused our analysis on the mutations most likely to be functionally important. We applied the MutSig2CV algorithm (15) to identify genes recurrently mutated in the discovery cohort. Of these genes, we limited our search to highly deleterious variants, meaning known hotspot or putative truncating (frameshift insertion or deletion, nonsense, or splice-site) mutations. Of the seven recurrently mutated genes (Fig. 2A and table S1E) (6), PBRM1 (polybromo 1) was the only gene in which truncating, or loss-of-function (LOF) (11), mutations were enriched in tumors from patients in the CB versus NCB group [9/11 versus 3/13; Fisher’s exact test P = 0.012, q = 0.086, odds ratio for CB = 12.93, 95% confidence interval (CI) 1.54 to 190.8] (Fig. 2B and table S1F). In this cohort, all truncating PBRM1 alterations co-occurred with deletion of the nonmutated allele on chromosome 3p (Fig. 2A), resulting in complete LOF of PBRM1, and most of the mutations were predicted to be clonal (present in all tumor cells) by using the computational method ABSOLUTE (16) (table S1F). Prior large-scale sequencing studies have shown that PBRM1 LOF alterations occur in up to 41% of ccRCC tumors (17) and are commonly clonal events present in all or nearly all tumor cells (18). Patients whose tumors showed biallelic PBRM1 loss had significantly prolonged OS and PFS compared to patients without PBRM1 LOF (log-rank test P = 0.0074 and 0.029, respectively) (Fig. 2C and fig. S5), and they experienced sustained reductions in tumor burden (Fig. 2D).

Fig. 2 Analysis of tumor-genome features in discovery cohort reveals a correlation between PBRM1 LOF mutations and clinical benefit from anti–PD-1 therapy.

(A) Mutations in the discovery cohort. Patients are ordered by response category, with tumor mutation burden in decreasing order within each response category. Shown are all genes that were recurrently mutated at a significant frequency, as assessed by MutSig2CV analysis (table S1E). CNA, copy-number alteration. (B) Enrichment of truncating mutations in tumors from patients in the CB versus NCB groups. Red dashed line denotes q < 0.1 (Fisher’s exact test). Mutations in genes above the black dotted line are enriched in tumors of patients with clinical benefit from anti–PD-1 therapy, and mutations in genes below the line are enriched in tumors of patients with no clinical benefit. (C) Kaplan-Meier curve comparing overall survival of patients treated with anti–PD-1 therapy whose tumors did or did not harbor LOF mutations in PBRM1. See also fig. S5 for Kaplan-Meier curve comparing PFS of these patients. (D) Spider plot showing objective decrease in tumor burden in PBRM1-LOF (light blue) versus PBRM1-intact (yellow) tumors. Three patients with early progression on anti–PD-1 therapy and truncating mutations in PBRM1 (dark blue) had long and/or censored OS.

To evaluate the reproducibility of this finding, we then examined matched pretreatment tumor and germline genomic data from an additional 63 patients treated with anti–PD-1 or anti–PD-L1 [anti–PD-(L)1] therapy, either alone or in combination with anti–CTLA-4 (cytotoxic T lymphocyte-associated protein 4) therapy. Of these 63 patients, PBRM1 mutation status was derived from WES in 49 patients and from panel sequencing in 14 patients (Fig. 3, A and B, and table S2, A and B) (11). Tumors from CB patients were more likely to harbor truncating alterations in PBRM1 (17/27 versus 4/19, Fisher’s exact test P = 0.0071, odds ratio for CB = 6.10, 95% CI 1.42 to 32.64) (Fig. 3, C and D, and table S2C). Although we could not assess copy-number alterations in all samples in the validation cohort, the PBRM1 LOF mutations likely represented biallelic loss, as chromosome 3p deletions are nearly ubiquitous in ccRCC (6). Notably, one of the four NCB patients whose tumor showed a PBRM1 LOF mutation also had an alteration in B2M, which codes for a protein important in antigen presentation. This provides a potential explanation for the patient’s lack of clinical benefit from immune checkpoint blockade therapy despite having a truncating PBRM1 mutation.

Fig. 3 PBRM1 LOF mutations correlate with clinical benefit in a validation cohort of ccRCC patients treated with immune checkpoint inhibitors.

(A) Selection of the validation cohort. (B) Clinical outcomes in the validation cohort. Ten patients without posttreatment restaging scans (eight with clinical PD, two with SD, and one with PR) as well as 14 patients with targeted panel sequencing are not shown. (C) Proportion of tumors harboring PBRM1 LOF mutations in patients in the CB versus NCB groups. Error bars indicate SEM. *P < 0.05 (Fisher’s exact test). (D) Truncating alterations in PBRM1 and response to anti–PD-(L)1 therapies by sample. Colored boxes indicate samples with truncating mutations in PBRM1, and gray denotes samples without PBRM1 truncating mutations. Missense LOF denotes a missense mutation detected by targeted sequencing that was confirmed to be LOF by PBRM1 immunohistochemistry (see supplementary methods).

Although primary analyses excluded patients with IB owing to the unclear effect of immune checkpoint blockade therapy on patient outcomes in this group, the observed trend between PBRM1 mutation status and clinical benefit persisted with the inclusion of these patients as an intermediate phenotype. In both the discovery and validation cohorts, patients in the IB group had intermediate rates of PBRM1 LOF (82, 64, and 23% for CB, IB, and NCB in the discovery cohort and 63, 41, and 21% for CB, IB, and NCB in the validation cohort; Fisher-Freeman-Halton exact test P = 0.017 and 0.017). Additionally, although no difference in survival was observed between treatment-naive and previously treated patients in the discovery cohort (fig. S3), the PFS benefit conferred by PBRM1 LOF was more prominent in tumors from previously treated patients compared to those from patients receiving anti–PD-1 therapy as their first cancer therapy (P = 0.009) (fig. S6 and tables S1 and S2).

The PBRM1 gene codes for BAF180, a subunit of the PBAF subtype of the switch-sucrose nonfermentable (SWI/SNF) chromatin remodeling complex. The PBAF complex suppresses the hypoxia transcriptional signature in VHL−/− ccRCC (19, 20) (deficient in the tumor suppressor VHL), but its effects on tumor-immune interactions have not been thoroughly studied. To explore the potential effect of this complex on the immunophenotype of ccRCC, we analyzed previously reported whole-transcriptome sequencing (RNA-seq) data from A704 ccRCC cell lines with perturbations in the PBAF complex (20). Loss of BAF180 or the related PBAF subunit BRG1, encoded by the gene SMARCA4, prevents formation of the intact PBAF complex (20). We performed gene-expression analyses of BAF180-null (A704) cell lines versus PBAF–wild type (A704-BAF180) cell lines, as well as BRG1-null (A704-BAF180BRG1−/−) cell lines versus A704-BAF180 cell lines (Fig. 4A). Differential gene-expression analysis showed substantial overlaps (~50%) between the top 100 genes differentially expressed in A704 versus A704-BAF180 cell lines and A704-BAF180BRG1−/− versus A704-BAF180 cell lines (table S4). This reflects the fact that BAF180 is essential to the PBAF, but not the BAF, complex, whereas BRG1 is a required subunit of both. Thus, the BAF180-null and BRG1-null cell lines have some shared characteristics but are also biologically and phenotypically distinct.

Fig. 4 PBRM1 mutational status in ccRCC influences immune-related gene expression.

(A) GSEA was performed on PBAF-deficient (A704 and A704-BAF180BRG1−/−) versus PBAF-proficient (A704-BAF180) kidney-cancer cell lines using both hallmark and corresponding founder gene sets. GSEA enrichment plot shown for the KEGG cytokine-cytokine receptor interaction gene set in A704 versus A704-BAF180 (PBRM1 null versus wild type). Enrichment plot is similar for A704-BAF180BRG1−/− versus A704-BAF180 (BRG1 null versus wild type); see table S4. (B) GSEA was also performed on RNA-seq from pretreatment tumors in the discovery and validation cohorts of this study (n = 18 PBRM1-LOF versus n = 14 PBRM1-intact) by using the hallmark gene sets. Enrichment plots show increased expression of the hypoxia and IL-6–JAK-STAT3 gene sets in the PBRM1-LOF tumors.

Gene set enrichment analysis (GSEA) on 50 “hallmark” gene sets representing major biological processes (21) revealed five gene sets whose expression was significantly enriched in cell lines that were PBAF deficient. These included genes linked to IL-6 (interleukin-6)–JAK-STAT3 (Janus kinase–signal transducers and activators of transcription 3) signaling, TNF-α signaling via NF-κB, and IL-2–STAT5 signaling (Fig. 4A and table S5, A and B). As expected, the hallmark hypoxia gene set was upregulated in A704 versus A704-BAF180 cell lines [family-wise error rate (FWER) q = 0.071] (table S5A) (20). Across the more refined “founder” gene sets describing these five significantly enriched hallmark gene sets, the most strongly enriched gene set in PBAF-deficient cell lines was the Kyoto Encyclopedia of Genes and Genomes (KEGG) cytokine-cytokine receptor interaction gene set (FWER q = 0.0020 for A704 versus A704-BAF180, and q = 0.023 for A704-BAF180BRG1−/− versus A704-BAF180) (Fig. 4A and table S5, C to L). This gene set includes both immune-stimulatory (e.g., IL12, CCL21) and immune-inhibitory (e.g., IL10) genes, but Gene Ontology term analysis (11) showed that the genes most strongly enriched in PBAF-deficient cell lines were immune stimulatory (table S6). Previously reported GSEA analysis of untreated ccRCC from The Cancer Genome Atlas (TCGA) database and a murine model of PBRM1 loss also show amplified transcriptional outputs of hypoxia-inducible factor 1 (HIF1) and STAT3, involved in hypoxia response and JAK-STAT signaling, respectively, in PBRM1-mutant versus PBRM1–wild type states (19). GSEA analysis of RNA-seq from pretreatment tumors in the discovery and validation cohorts of this study (n = 18 PBRM1-LOF versus n = 14 PBRM1-intact) confirmed increased expression of the hypoxia and IL-6–JAK-STAT3 gene sets in the PBRM1-LOF tumors (Fig. 4B and tables S7, A and B, and S8). Given JAK-STAT3 pathway gene involvement in the interferon gamma (IFN-γ) signaling pathway and IFN-γ–dependent cancer immunostimulation (22), differential expression of these genes may affect the response of PBRM1-LOF patients to anti–PD-(L)1 therapy.

In addition to assessing tumor-intrinsic gene expression with GSEA, we further characterized the quality of the tumor-immune microenvironment in PBRM1-LOF versus PBRM1-intact ccRCC in three independent cohorts: patients from the TCGA database (6), an independent cohort of untreated ccRCC tumors (Sato) (23), and patient tumors from this study (table S8). In all three cohorts, tumors harboring LOF mutations in PBRM1 showed lower expression of immune-inhibitory ligands (e.g., CD276 and BTLA) (24) than those without PBRM1 mutations. This finding was somewhat unexpected, as high PD-L1 staining is associated with increased responsiveness to anti–PD-(L)1 agents in other cancer types (25, 26). However, the magnitudes of these differences were small and potentially confounded by differing degrees of tumor-stromal admixture (fig. S7, A to C) (9). We also examined LOF mutations in VHL, the most commonly-mutated gene in the TCGA ccRCC cohort. VHL mutation status did not correlate with immune-related gene expression (fig. S8), suggesting that observed differences in immune-related gene expression in the context of PBRM1 LOF may be specific to the PBRM1 gene.

In summary, we have shown that patients with metastatic ccRCC harboring truncating mutations in PBRM1 experienced increased clinical benefit from immune checkpoint therapy. This may be due to distinct immune-related gene expression profiles in PBRM1-mutant or PBAF-deficient tumor cells compared to their PBAF-intact counterparts, as shown by RNA-seq analyses in this study, though additional in vivo studies will be needed to further explore these findings. Given the high prevalence of PBRM1 LOF in ccRCC and of SWI/SNF alterations across all cancer types (more than 20%) (27), this finding has important implications as a molecular tool for considering immunotherapy responsiveness in ccRCC and across cancer types.

In vivo studies of mice harboring tumor clones with inactivation of PBRM1—or the related essential PBAF complex components ARID2 or BRD7—show that cells with PBAF loss are more sensitive to T cell–mediated cytotoxicity compared to their PBAF-intact counterparts (28). This finding lends a mechanistic basis to the results observed here and helps explain the conflicting results regarding PBRM1 mutation status as a prognostic variable in ccRCC (in the absence of immunotherapy) in prior studies (2937). Also, PBRM1 has previously been linked to longer PFS with vascular endothelial growth factor (VEGF)–targeted therapies (38). The observed interaction between PBRM1 status, prior treatment (largely with VEGF inhibitors), and response to immune checkpoint therapy in this study argues for further investigation of patient outcomes from sequential and combination treatment regimens that include anti–PD-(L)1. The relationship between PBRM1 LOF and clinical benefit from anti–PD-(L)1 therapies in ccRCC, as well as the immunological consequences of PBAF loss in other cancer types, merit further preclinical and prospective clinical validation.

Supplementary Materials

www.sciencemag.org/content/359/6377/801/suppl/DC1

Materials and Methods

Figs. S1 to S8

Tables S1 to S8

References (3957)

References and Notes

  1. Materials and methods are available as supplementary materials.
Acknowledgments: Funding: This project was supported by the Bristol-Myers Squibb II-ON consortium, the American Association for Cancer Research KureIt Grant for Kidney Cancer Research (E.M.V. and T.K.C.), and the Cancer Immunologic Data Commons (NIH U24CA224316). D.M. is a Howard Hughes Medical Institute Medical Research Fellow. T.K.C. is supported in part by the Dana-Farber/Harvard Cancer Center Kidney Specialized Program of Research Excellence (SPORE); the Kohlberg chair at Harvard Medical School; and the Trust Family, Michael Brigham, and Loker Pinard Funds for Kidney Cancer Research at the Dana-Farber Cancer Institute. T.H. is supported by the Gerstner Family Career Development Award, the National Cancer Institute (K12CA90628), and the U.S. Department of Defense (W81XWH-17-1-0546). Opinions, interpretations, conclusions, and recommendations are those of the author and are not necessarily endorsed by the U.S. Department of Defense. Author contributions: D.M., C.A.M., M.W.B., M.E.A., and E.M.V. performed genomic analyses. W.L. and W.G. performed the cell line experiments and generated the cell line genomic data. M.G., T.K.C., D.C., C.H., M.W.-R., M.H.V., and R.J.M. gathered the discovery cohort clinical and biological materials. S.S. contributed to immunohistochemistry. S.M.W., D.J.M., D.B., M.H.V., A.S., M.D.H., T.H., and C.N. collected the biological materials and clinical annotations for the validation cohort. A.T. contributed to project management. D.M., C.A.M., E.M.V., and T.K.C. prepared the initial draft of the manuscript. F.S.H., W.G.K., D.C., C.H., M.W.-R., A.S., M.H.V., R.J.M., T.K.C., and E.M.V. supervised the study. Competing interests: W.G.K. is a paid consultant for Agios, Fibrogen, Nextech Ventures, Peloton Therapeutics, Tracon, Third Rock Ventures, and serves on the Lilly Pharmaceuticals Board of Directors. E.M.V. is a paid consultant for Third Rock Ventures, Genome Medical, Inc., and Tango Therapeutics and receives research support from Bristol-Myers Squibb and Novartis. T.K.C. is a paid advisor for AstraZeneca, Bayer, Bristol-Myers Squibb, Cerulean, Foundation Medicine, Genentech, GlaxoSmithKline, Merck, Novartis, Peloton, Pfizer, Prometheus Labs, Roche, and Eisai. T.K.C. receives institutional research funding from AstraZeneca, Bristol-Myers Squibb, Exelixis, Genentech, GlaxoSmithKline, Merck, Novartis, Peloton, Pfizer, Roche, Tracon, and Eisai (for clinical trials). S.S. is a paid consultant for Merck and Bristol-Myers Squibb. R.J.M. is a paid consultant for Pfizer, Genentech-Roche, Novartis, Exelixis, and Eisai. T.H. is a paid consultant for Pfizer, Exelixis, and Roche. F.S.H. is a paid consultant for Bristol-Myers Squibb, Merck, Genentech, Novartis, Amgen, and EMD Serono. M.D.H. is a paid consultant for Bristol-Myers Squibb, Merck, Genentech-Roche, AstraZeneca, Mirati, Janssen, and Novartis. E.M.V., T.K.C., and D.M. are inventors on a patent application submitted by the Dana-Farber Cancer Institute that covers PBRM1 mutational status in tumors and response to immunotherapy. Data and materials availability: This study makes use of data generated by the Department of Pathology and Tumor Biology, Kyoto University (Sato cohort). The sequencing data are deposited in database of Genotypes and Phenotypes (dbGap) (accession number phs001493.v1.p1). The cell line transcriptome data are deposited in Gene Expression Omnibus (GEO) (accession number PRJNA371283).
View Abstract

Navigate This Article