It is poorly understood how the tumor immune microenvironment influences disease recurrence in localized clear-cell renal cell carcinoma (ccRCC). Here we performed whole-transcriptomic profiling of 236 tumors from patients assigned to the placebo-only arm of a randomized, adjuvant clinical trial for high-risk localized ccRCC. Unbiased pathway analysis identified myeloid-derived IL6 as a key mediator. Furthermore, a novel myeloid gene signature strongly correlated with disease recurrence and overall survival on uni- and multivariate analyses and is linked to TP53 inactivation across multiple data sets. Strikingly, effector T-cell gene signatures, infiltration patterns, and exhaustion markers were not associated with disease recurrence. Targeting immunosuppressive myeloid inflammation with an adenosine A2A receptor antagonist in a novel, immunocompetent, Tp53-inactivated mouse model significantly reduced metastatic development. Our findings suggest that myeloid inflammation promotes disease recurrence in ccRCC and is targetable as well as provide a potential biomarker-based framework for the design of future immuno-oncology trials in ccRCC.
Improved understanding of factors that influence metastatic development in localized ccRCC is greatly needed to aid accurate prediction of disease recurrence, clinical decision-making, and future adjuvant clinical trial design. Our analysis implicates intratumoral myeloid inflammation as a key driver of metastasis in patients and a novel immunocompetent mouse model.
The majority of patients (∼75%) with clear-cell renal cell carcinoma (ccRCC) initially present with localized disease for which surgical resection followed by active surveillance constituted the standard of care prior to the recent FDA approval of adjuvant pembrolizumab (1). Up to 40% to 50% of these patients, however, will subsequently develop disease recurrence (2, 3). Clinicopathologic risk stratification systems such as the Mayo Clinic's stage, size, grade, and necrosis (SSIGN) and the UCLA integrated staging system (UISS) have been developed to help identify patients at high risk of relapse (2, 3). Additionally, three validated RNA gene signature–based tools, ClearCode34 (4), the cell-cycle proliferation (CCP) score (5), and the Recurrence Score (6), can also predict recurrence after nephrectomy. There is a growing understanding that ccRCC is composed of highly heterogeneous subtypes that can be distinguished by genomic and transcriptomic features reflective of the stromal and immune microenvironment (7). Indeed, alterations in frequent ccRCC driver genes such as BAP1, SETD2, and TP53 are correlated with worse recurrence-free and cancer-specific survival (8–11), and these mutations can be associated with distinct tumor microenvironments (TME; refs. 12, 13). Furthermore, biomarker analyses of pretreatment samples from recent clinical trials have revealed correlations between TME gene signatures related to angiogenesis or tumor-infiltrating immune cells and outcomes to systemic therapies targeting tumor vasculature and immune-checkpoint receptors, respectively (7, 14–16). We hypothesized that such microenvironmental features might also influence the metastatic development of localized disease and identification of these factors may support the design of future mechanism-based adjuvant trials.
Utilizing multimodal profiling of the placebo arm of a large adjuvant ccRCC trial (17), we find that metastatic recurrence in ccRCC is largely independent of T-cell response and rather is associated with a myeloid inflammatory state, which can be linked to TP53 inactivation across numerous data sets. We further develop a first-of-a-kind immunocompetent, metastatic, Tp53-inactivated ccRCC mouse model and demonstrate that targeting myeloid populations via adenosine A2A receptor antagonism impedes metastatic development.
Differential Gene Expression Analysis Identifies Oncogenic and Immune Pathways Associated with Disease Recurrence
A total of 236 of 769 patients from the placebo arm of the PROTECT (17) trial (adjuvant pazopanib vs. placebo for high-risk localized ccRCC) had clinical, pathologic, IHC, and tumor whole-genome microarray data available for analysis. Demographic and clinicopathologic characteristics of the patient population are listed in Supplementary Table S1. We performed differentially expressed gene (DEG) analysis in patients who developed disease recurrence (n = 88) versus those who did not (n = 148). We identified 765 DEGs in tumors associated with disease recurrence (Fig. 1A; Supplementary Table S2). Gene set enrichment analysis (GSEA) demonstrated several oncogenic pathways known to be upregulated in ccRCC enriched in patients with disease recurrence, including epithelial-to-mesenchymal transition (EMT), MYC targets, G2–M checkpoint, and MTORC1 signaling (Fig. 1B). Multiple immune and inflammatory signaling pathways in tumors of patients with disease recurrence were also upregulated, including interferon gamma, TNFα, inflammatory response, and IL6 (Fig. 1B). Unbiased Ingenuity Pathway Analysis (IPA) revealed IL6 as a key component among the molecular pathways enriched in said tumors (Fig. 1C).
Immune Deconvolution Links Myeloid Inflammation and the IL6 Pathway to Metastasis
Recent single-cell RNA sequencing (scRNA-seq) analysis of ccRCC tumors revealed enrichment of IL6 expression primarily in myeloid, vascular endothelial, and CA9+ tumor cell populations (18). We, therefore, performed immune deconvolution using gene signatures reflective of myeloid inflammation that have recently been associated with responses to systemic therapy in the metastatic disease setting (15, 16, 19, 20) to our cohort. Both the adenosine and myeloid signatures were significantly enriched in tumors of patients with subsequent disease recurrence (Fig. 2A). Adenosine signature groups defined by quartiles were significantly and adversely associated with median disease-free survival (DFS; log-rank test, P = 1.79E–08; Supplementary Fig. S1A) and median overall survival (OS; log-rank test, P = 3.71E–06; Supplementary Fig. S1B), with scores in the highest quartile demonstrating the worst DFS [hazard ratio (HR) 5.37; 95% confidence interval (CI), 2.73–10.6; P = 1.00E–6; Q1 as reference] and OS (HR 5.39; 95% CI, 2.04–14.2; P = 6.8E−04; Q1 as reference). Similarly, patients who harbored higher myeloid scores demonstrated decreased DFS (log-rank test, P = 2.20E–08; Supplementary Fig. S1C) and worse OS (log-rank test, P = 7.69E–05; Supplementary Fig. S1D), with the highest quartile scores associated with the worst DFS (HR 7.09; 95% CI, 3.28–15.3; P = 6E–07) and OS (HR 6.39; 95% CI, 2.18–18.7; P = 7.1E–04; Q1 as reference). Given the similarities between these two signatures (Supplementary Fig. S1E), we sought to determine which of the included target genes (nine in total with five shared and four private) most strongly associated with survival outcomes in ccRCC. To this end, we performed a Cox regression analysis on expression levels of the nine genes comprising the adenosine and myeloid signatures in The Cancer Genome Atlas (TCGA) pan-cancer data set (21) with respect to OS. Out of the 32 malignancies examined in the TCGA data set, 17 had an expression of at least one gene significantly associated with OS (Fig. 2B). Within the TCGA-kidney renal clear-cell carcinoma (KIRC) cohort, we found upregulation of IL1β, CXCL1, CXCL2, CXCL3, CXCL5, CXCL8, and IL6 was associated with significantly decreased OS (Fig. 2C). These genes were then combined to create a new composite myeloid inflammatory signature, termed MSK inflammatory (MSKI; Supplementary Fig. S1E; Fig. 2D). As expected, MSKI single-sample GSEA (ssGSEA) scores grouped by quartile were significantly associated with OS (log-rank test, P = 1.53E–07) in the TCGA-KIRC cohort (Supplementary Fig. S1F). MSKI scores were also associated with worse DFS (log-rank test, P = 5.44E–03; Fig. 2E), with the worst outcomes seen in tumors with the highest quartile of enrichment scores (DFS: HR 2.38; 95% CI, 1.24–4.59; P = 9.45E–03; OS: HR 3.08; 95% CI, 1.96–4.85; P = 1.1E–06; Q1 as reference). Within the PROTECT cohort, the MSKI gene signature was significantly enriched in recurrent versus nonrecurrent tumors (Fig. 2F), with higher enrichment scores associated with worse DFS (log-rank test, P = 4.00E–11; Fig. 2G) and OS (log-rank test, P = 4.76E–07; Supplementary Fig. S1G). We then applied the MSKI signature to an external validation cohort of 84 patients with localized, untreated ccRCC, termed the MOFFITT cohort (see Supplementary Table S3 for clinicopathologic demographics), and confirmed a significant association with DFS (log-rank P = 0.0199; Fig. 2H) as well as a nonsignificant OS trend (log-rank P = 0.108; Supplementary Fig. S1H).
Lymphoid Infiltration and Exhaustion Markers Fail to Predict Disease Recurrence
Intratumoral CD8+ cell infiltration has been associated with conflicting outcomes in patients with metastatic (22, 23) and localized ccRCC (24). Gene signatures reflective of lymphoid infiltration (18) including effector T-cell (Teff; ref. 15) and JAVELIN Renal 101 Immuno (Javelin; ref. 16) have been associated with response rates to immune-checkpoint blockade (ICB) + anti-VEGF therapies in metastatic ccRCC (mccRCC). However, deconvolved scores for Teff (Mann–Whitney test, P = 0.2099) and Javelin (Mann–Whitney test, P = 0.2014) signatures did not significantly differ between recurrent and nonrecurrent tumors in the PROTECT cohort (Fig. 3A). Quantitative IHC demonstrated these effector signatures were significantly correlated with intratumoral CD8+ cell counts (Fig. 3B), which also did not differ between tumors with/without subsequent disease recurrence in the PROTECT (Mann–Whitney test, P = 0.1453; Fig. 3C) and MOFFITT (Mann–Whitney test, P = 0.27; Supplementary Fig. S2A) cohorts.
Further, we found no association between Teff levels (Teffhi vs. Tefflo) and DFS (log-rank test, P = 0.1449; Supplementary Fig. S2B) or OS (log-rank test, P = 0.6243; Supplementary Fig. S2C). In the biomarker analysis of the IMmotion150 trial (15), the efficacy of PD-L1–directed monotherapy in Teffhi tumors was limited in the presence of high myeloid signatures, suggestive of myeloid-mediated T-cell suppression. In a Kaplan–Meier analysis of Teff and myeloid subgroups (Teffhi vs. Tefflo; Myeloidhi vs. Myeloidlo), we found the myeloid scores to be the primary driver of outcomes (DFS: log-rank test, P = 1.19E–07; Supplementary Fig. S2D; OS: log-rank test, P = 1.58E–04; Supplementary Fig. S2E), irrespective of high and low Teff scores.
We also analyzed IHC data for patients in the PROTECT (n = 236) and MOFFITT (n = 84) cohorts to assign distinct patterns of CD8+ cell infiltration including “inflamed,” “desert,” and “excluded.” Kaplan–Meier analysis revealed no association between these CD8+ infiltration patterns and DFS in the PROTECT (log-rank test, P = 0.1603; Fig. 3D) and MOFFITT (log-rank test, P = 0.76; Fig. 3E) cohorts. Coexpression of inhibitory immune-checkpoint molecules on CD8+ cells, including PD-1, TIM3, and LAG3, reflects an “exhausted” phenotype that has been associated with worse disease outcomes (25, 26). Therefore, we compared immune-checkpoint (Fig. 3F) and immunosuppression (ref. 27; Supplementary Fig. S2F) scores but found no significant difference between recurrent and nonrecurrent tumors in the PROTECT cohort. IHC markers of exhaustion including TIM3 (Mann–Whitney test, P = 0.4309; Fig. 3G) and LAG3 (Mann–Whitney test, P = 0.0734; Fig. 3H) expression also did not correlate with disease recurrence. There was no significant association between tumor PD-L1 status and disease recurrence (Fig. 3I). We noted a strong correlation between the immune-checkpoint scores and Teff scores (Spearman, r = 0.71; P = 5.69E–38; Fig. 3J), but none between the immune-checkpoint scores and MSKI scores (Spearman, r = 0.004; P = 0.9512; Fig. 3K). Accordingly, on subgroup analysis, the immune-checkpoint scores (Fig. 3L) and LAG3 IHC scores (Supplementary Fig. S2G) were only influenced by the degree of lymphocytic (Teffhi vs. Tefflo) but not myeloid (MSKIhi vs. MSKIlo) infiltration. Together, these data indicate that lymphocytic infiltrates were not associated with disease recurrence.
The Angiogenic Pathway Is Downregulated in Recurrent Tumors
The ubiquitous loss-of-function alterations of the von Hippel–Lindau (VHL) gene in ccRCC result in stabilization of hypoxia-inducible factors (HIF) and subsequent transcriptional upregulation of proangiogenic genes. The study of this canonical pathway in ccRCC led to the successful development of antiangiogenic-based therapies, and, indeed, upregulation of angiogenesis genes was associated with added antitumor effect from tyrosine kinase inhibitor (TKI) therapy in the metastatic setting (7, 15, 16, 19).
In our cohort, we found that patients with lower angiogenesis scores had increased disease recurrence (Mann–Whitney test, P = 4.28E–03; Supplementary Fig. S3A) and on Kaplan–Meier analysis displayed inferior DFS (log-rank test, P = 4.85E–03; Q1: HR 3.02; 95% CI, 1.57–5.79; P = 8.91E–04; Q4 as reference; Supplementary Fig. S3B) and shorter OS (log-rank test, P = 9.58E–04; Q1: HR 4.10; 95% CI, 1.66–10.2; P = 0.0022; Q4 as reference; Supplementary Fig. S3B). Similarly, in the TCGA-KIRC cohort, we found that a lower angiogenesis score was associated with decreased OS (log-rank test, P = 1.65E–6) but not DFS (log-rank test, P = 0.607; Supplementary Fig. S3C). With a notably smaller sample size, we did not detect significant associations between angiogenesis scores and DFS (log-rank P = 0.114) or OS (log-rank P = 0.0601) in the MOFFITT cohort (Supplementary Fig. S3D). However, there was a visible trend toward worse DFS and OS in tumors with the lowest quartile of angiogenesis score.
Integrated Clinical and Genomic Models Reveal Independent Predictive Value of Myeloid Inflammation on Clinical Outcomes
Given our univariate findings for gene signatures reflecting myeloid inflammation (prometastasis) and tumor angiogenesis (antimetastasis) gene signatures with respect to disease outcomes, we sought to integrate both TME features into and put them in context with established clinical and genomic strata. On univariate analysis, the MSKI score had no significant associations with tumor stage or grade (Kruskal–Wallis test, P = 0.1649, Mann–Whitney test, P = 0.099, respectively; Fig. 4A); however, angiogenesis scores were significantly lower in high- versus low-grade tumors (Mann–Whitney test, P = 4.41E–05; Fig. 4B), again without correlation with stage. We failed to detect associations between either MSKI or angiogenesis score and modified prognostic UISS scores (Supplementary Fig. S4A and S4B). Next, we applied the validated genomic-based risk scores ClearCode34 (4) and the cell-cycle proliferation (CCP) score (5) to the PROTECT cohort. Consistent with previous findings, the ClearCode34 poor risk group (ccB) had significantly shorter DFS than those in the good risk group (ccA; log-rank test, P = 1.53E–06; Supplementary Fig. S4C), and patients with high CCP scores from their tumors had shorter DFS than those with low CCP scores (log-rank test, P = 2.15E–04; Supplementary Fig. S4D). We noted significantly higher MSKI and lower angiogenesis scores in ccB versus ccA tumors (Mann–Whitney test, P = 4.33E–10 and P = 8.01E–13, respectively; Fig. 4C). The MSKI scores were noted to have a positive correlation with CCP scores (Spearman, r = 0.35; P = 3.86E–08; Fig. 4D) and a negative correlation with angiogenesis scores (Spearman, r = −0.15; P = 0.0179; Fig. 4D). Angiogenesis scores exhibited a negative correlation with CCP scores (Spearman, r = −0.39; P = 8.0E–10; Fig. 4D). Taken together, tumors from patients with subsequent relapse of disease were more likely to be characterized as ccB group and had higher MSKI scores but lower angiogenesis scores. We then integrated these four variables into a multivariate Cox regression model and found that only the MSKI gene signature remained significantly associated with DFS (HR 1.28; 95% CI, 1.16–1.40; P = 5.69E–07; Fig. 4E) and OS (HR 1.23; 95% CI, 1.09–1.39; P = 7.81E–04; Fig. 4F). We also used 2-year time point receiving operating characteristic (ROC) curves and C-indices to assess the classification for both OS and DFS multivariable models. Models were run with and without MSKI as a covariate in order to determine the additive effects of MSKI, and we found that the addition of MSKI improved the C-index from 0.6 to 0.73 for DFS (Fig. 4G) and 0.66 to 0.72 for OS (Fig. 4H).
Cell Populations Underlying and Intratumoral Heterogeneity of the MSKI and Angiogenesis Gene Signatures
To determine which intratumoral cell populations were contributing to the prognostic gene signatures in our cohort, we performed fluorescence-activated cell sorting (FACS) and RNA-seq on 127 samples of 15 immune populations and non-CD45 cells isolated from ccRCC tumors, normal kidney, and peripheral blood from a total of 21 patients (Supplementary Fig. S5 for flow gating strategy). In line with the cell populations underlying the adenosine and myeloid signatures reported in a recent scRNA-seq analysis (18), we observed the highest MSKI median enrichment scores in myeloid cell populations, including dendritic cells, neutrophils, macrophages, and monocytes (Fig. 5A).
Multiregional tumor profiling studies have revealed a striking degree of intratumoral heterogeneity (ITH) in genomic alterations that have been associated with metastatic potential and clinical outcomes (28–30). We therefore wondered whether our transcriptomic signatures were also vulnerable to ITH. To explore this, we analyzed a cohort of 29 patients with ccRCC undergoing surgical resection at the Memorial Sloan Kettering Cancer Center (MSKCC) in which each patient had a median of five regions per tumor (range, 2–7) collected for analysis. We first performed immunofluorescence (IF) staining in order to validate the MSKI and angiogenesis signatures at the protein level. We found strong positive correlations between the number of cells expressing the pan-macrophage marker CD68 and the MSKI signature (Spearman, r = 0.67; P = 6.24E–05; Fig. 5B) and significantly higher CD68 IF cell counts in MSKIhi versus MSKIlo tumors (Mann–Whitney test, P = 7.64E–04; Fig. 5C). This suggests that although the MSKI signature is expressed by most myeloid cells (Fig. 5A), the signature may be predominantly capturing tumor-associated macrophages (TAM) due to their high abundance in tumors. As expected, staining of the epithelial marker CD31 was also strongly correlated with angiogenesis scores (Spearman, r = 0.63; P = 2.31E–04; Fig. 5D), with higher CD31 IF scores in angiogenesis-high versus angiogenesis-low tumors (Mann–Whitney test, P = 7.64E–04; Fig. 5E).
We then explored the ITH of MSKI and angiogenesis scores by performing RNA-seq on this cohort. Overall, there was not a marked degree of ITH with respect to these microenvironmental features (Fig. 5F). We found that all regions analyzed in 48.3% and 58.6% of tumors were uniformly above or below the median enrichment score for the MSKI and angiogenesis signatures, respectively. Representative images of CD31 and CD68 IF and the corresponding angiogenesis and MSKI scores, respectively, for two multiregional tumor samples are highlighted in Fig. 5G.
Development of a Novel Electroporation-Derived ccRCC Syngeneic Model
We next sought to validate the importance of myeloid inflammation in RCC recurrence in a mouse model. Traditional germline-modified genetically engineered mouse models (GEMM) most accurately recapitulate natural malignant disease progression; however, it can take many months to both cross the various alleles sufficient for tumor formation and expand the experimental colony. Additionally, penetrance in RCC models is not optimal, and tumor onset time can be highly variable between mice, taking a minimum of 2 to 5 months following tumor-suppressor gene deletion or oncogene activation (31–33). Lastly, of the existing ccRCC GEMMs, only one has been reported to develop metastases, but this occurs at a very low frequency (31–34), thus limiting studies on mRCC.
To generate a more tractable and biologically relevant metastatic ccRCC model, the left kidney of wild-type (WT) C57BL/6 mice was externalized, followed by subcapsular injection of a 50 μL cocktail of plasmids encoding a transposase system to stably integrate EF-1α promoter–driven c-myc overexpression, transient expression of Cas9, and single-guide RNA (sgRNA) guides targeting Vhl, Rb1, and Tp53. We chose to model this genetic combination due to the extensive molecular similarities of a GEMM encompassing inducible kidney-specific deletion of Vhl, p53, and Rb1 (KVpR) to human ccRCC (32). Tp53 is also enriched in metastatic cohorts (7, 16, 19, 35) and is associated with a myeloid-high phenotype (7). Additionally, C-myc is amplified in 8% to 15% of human ccRCCs and is associated with the development of metastases (32, 36). Following plasmid injection, the kidney and injection site were immediately electroporated via electrode tweezers, and the wound was surgically closed. At 48.7 weeks after electroporation, a large tumor in the left kidney was detected by MRI, dissected, dissociated, and cultured to derive a syngeneic cell line (Fig. 6A and B). No overt metastases were detected.
Novel Syngeneic Tumors Molecularly Resemble ccRCC and Spontaneously Metastasize to Distant Organs
Gene editing of the parental tumor and derived cell line was confirmed by Sanger sequencing (Supplementary Fig. S6A). A 1-base pair (bp) deletion was detected in all three targeted genes at the expected cut site, confirming that the tumor was electroporation derived rather than spontaneous. Histologically, the parental tumor was high grade and highly dedifferentiated (Fig. 6C). Staining for the neuroendocrine markers synaptophysin and chromogranin A was negative (Supplementary Fig. S6B) despite areas of histologic large-cell neuroendocrine features (Fig. 6C). Since ccRCC originates from kidney proximal tubule cells (37) and VHL loss is used as a clinical diagnostic marker of ccRCC in poorly differentiated tumors, we confirmed by IHC that the tumor was epithelial (pan-CK+) and of kidney (Pax8+) origin (Fig. 6D). Two of three proximal tubule markers (CD10+, Megalin+, Aquaporin 1−) were present, and deletions at the protein level of VHL [via downstream upregulation of HIF1α, carbonic anhydrase 9 (CA9), and Glut1], p53, and Rb1 plus myc overexpression were confirmed (Fig. 6D). We successfully derived a cell line from this tumor, LVRCC67, which grows rapidly in vitro and retains these molecular markers (Fig. 6A and E).
Whole-exome sequencing of LVRCC67 confirmed CRISPR-mediated 1-bp deletion in Vhl, Tp53, and Rb1 and failed to detect additional oncogenic events (Supplementary Table S4). However, there were large deletions in chromosome 10 as well as substantial copy-number alterations in small regions of chromosome 13 (Supplementary Fig. S6C and Supplementary Table S5). When injected subcutaneously into WT C57BL/6 mice, LVRCC67 tumors retained their dedifferentiated histology (Supplementary Fig. S6D). However, unlike the parent tumor, metastases to the lung (Fig. 6F) and/or liver (Fig. 6G) were readily observed in 80% (4/5) of mice and these metastatic nodules also exhibited dedifferentiated morphologies.
Syngeneic Tumors Resemble an Aggressive Stromal/Proliferative Human ccRCC Molecular Subtype
A recent study analyzing 823 advanced ccRCC tumors identified seven molecular subtypes with distinct biological signatures (7) and differential prognosis (38). In order to compare our mouse model with these molecular clusters, we performed bulk RNA-seq of subcutaneous LVRCC67 tumors and healthy normal kidney cortex tissue and looked for enrichment of hallmark gene set signatures from the Molecular Signatures Database (MSigDB; ref. 39) as well as TME gene signatures. We observed that our mouse tumors had strong upregulation of gene signatures related to cell cycle (MYC targets, G2–M checkpoint, and E2F targets), EMT, and stroma, as well as genes associated with anabolic metabolism (higher FAS/pentose phosphate, reduced oxidative phosphorylation, and reduced fatty acid metabolism; Fig. 6H; Supplementary Fig. S6E). Tumors were angiogenic, consistent with Vhl loss–mediated pseudohypoxia and known roles of p53 in regulating angiogenesis (40). We also observed enrichment of myeloid and MSKI signatures, together suggesting the model strongly resembles one of the seven molecular subtypes (the stromal/proliferative molecular cluster 6, referred to as SPMC) reported in the above-mentioned study of human ccRCC (7). Our model also aligns well genomically with this molecular subtype, as TP53 mutations were highest in this cluster (29%; ref. 7). Although CDKN2A/B alterations were enriched in SPMC tumors but not in our mouse tumors (Supplementary Tables S4 and S5), this is consistent with the finding that CDKN2A/B and TP53 mutations were largely nonoverlapping (7).
Patients with SPMC tumors have a particularly poor prognosis compared with other molecular subtypes (38). In line with this, pathways associated with this aggressive tumor type (EMT, cell cycle, myeloid) also seemed to be enriched in recurrent tumors within the PROTECT cohort (Fig. 1B). We, therefore, clustered PROTECT cohort tumors by recurrence status and MSKI gene expression quartiles and looked specifically at the expression of gene signatures used to define the seven molecular subtypes described in the IMmotion analysis (Supplementary Fig. S7). Indeed, the gene expression profiles defining SPMC tumors (described above for mouse tumors) were enriched in recurrent tumors and within the highest MSKI quartile. Our data suggest that the aggressiveness of this analogous molecular subtype may be driven by a myeloid inflammatory program.
Given the enrichment of TP53 alterations within SPMC tumors and the p53-driven foundation of our metastatic mouse model, we wondered whether there was also an association between TP53 mutations and MSKI score, which predicts recurrence. Although genomic data were unavailable for the PROTECT cohort, we confirmed in two previously described clinical cohorts (19, 41) that TP53 mutations were indeed associated with significantly higher MSKI scores (JAVELIN101 P = 0.006, COMPARZ P = 0.033; Supplementary Fig. S8A and S8B).
Adenosine Receptor but Not PD-1 Inhibition Attenuates ccRCC Metastasis
We next leveraged our mRCC model to test whether targeting adenosine/myeloid-mediated inflammation and the MSKIhi environment could prevent metastases. To do this, mice were injected subcutaneously with LVRCC67 cells followed by daily constitutive treatment from day 2 with ciforadenant (CPI-444; ref. 14), a small-molecule, high-affinity selective inhibitor of the immunosuppressive adenosine A2A receptor (A2AR), or twice weekly treatment with anti–PD-1. Primary tumor volumes were measured, and mice were sacrificed on day 35 to assess metastatic burden.
CPI-444 significantly reduced primary tumor growth [Fisher least significant difference (LSD) test, P = 1.4E–05], whereas anti–PD-1–treated mice had a heterogeneous response that was not overall statistically significant (Fisher LSD test, P = 0.0858; Fig. 7A). Upon dissection, only 27% (4/15) of CPI-444–treated mice, compared with 65% (9/14) of control and 71% (5/7) of anti–PD-1–treated mice, had macroscopic metastases (Fisher exact test, control compared with CPI-444 P = 0.065; Fig. 7B). Additionally, CPI-444– but not anti–PD-1–treated mice had a lower abundance of metastatic nodules per mouse (Wilcoxon rank-sum test, control versus CP444 P = 0.024 or anti–PD-1 P = 0.91; Fig. 7C).
Due to insufficient metastatic tumor tissue, we used primary tumors to assess drug-related changes to the immune TME using flow cytometry (Fig. 7D; Supplementary Fig. S9A and S9B). Control tumors were myeloid dominated, mostly by TAMs that constitute almost 50% of CD45+ cells. Other myeloid cells included low levels of monocytes and neutrophils. Importantly, tumor-infiltrating CD8+, CD4+, and regulatory T cells (Treg) were found in low abundance. Together, this validates our RNA-seq data that suggest LVRCC67 tumors are myeloid-rich and T cell–low, similar to SPMC human ccRCCs.
As expected, we found that overall CD45+ cell infiltration was significantly enhanced by both CPI-444 and anti–PD-1 treatment (Fig. 7D). We observed a striking increase in natural killer (NK) cells following CPI-444 and to a lesser degree following anti–PD-1 (Fig. 7D). Adenosine is known to affect the proliferation, maturation, and cytotoxicity of NK cells (42) and may therefore be an important mechanism of tumor control. Flow-sorted populations show that NK/innate lymphocyte cell populations can also express the adenosine signature, but at lower levels than myeloid cells (Fig. 5A).
Both drugs significantly increased the percentage of CD11b+Ly6C+ monocytes, suggesting enhanced recruitment from blood; however, total CD11b+F4/80+CD80+ TAMs were significantly reduced (Wilcoxon rank-sum test, P = 0.00150 and P = 0.0433 for CPI-444 or anti–PD-1 versus control, respectively; Fig. 7D). Further, TAM phenotyping showed a reduction of M2-like CD206+ TAMs and enrichment of M1-like TAMs that express MHC-II and IRF8, a transcription factor necessary for antigen presentation (43), in line with adenosine-mediated immunosuppression of TAMs (42). The majority of CD8+ T cells were neither activated (PD-1+granzyme B+) nor exhausted (PD-1+granzyme B−), suggestive of blood circulating or bystander naive populations (Fig. 7D). In accordance with a T cell–cold tumor, we did not see substantial and consistent changes in T-cell activation, exhaustion, or proliferation following CPI-444 or anti–PD-1 treatment (Fig. 7D).
We noticed that within the control group, some mice were more prone to metastases than others and wondered whether there were underlying immune TME differences. We observed the same upward shifts in NK cells and M1-like TAMs in control mice without metastases and no T cell–associated changes (Supplementary Fig. S9C). Taken together, our preclinical data support a T cell–independent, TAM- and NK-mediated mechanism of recurrence, consistent with our clinical data, and suggest that the drugs may be enhancing preexisting antimetastatic immunity.
In this report, we describe the results of a comprehensive transcriptomic analysis integrating clinicopathologic, flow cytometric, and IF data from patients with high-risk localized ccRCC receiving placebo on a randomized phase III trial (PROTECT), the TCGA, and smaller, independent cohorts. Our unbiased IPA core analysis identified IL6 as a key component of the various upregulated molecular pathways in recurrent versus nonrecurrent tumors. Elevated serum and intratumoral IL6 have been associated with higher rates of lymphatic and distant metastases (44, 45) and worse disease outcomes in ccRCC (46–50). In human RCC samples, myeloid-derived cytokines are positively correlated with intratumoral myeloid-derived suppressor cells, and blockade of IL1β and IL6 in mice decreased tumor myeloid-derived suppressor cell infiltration, increased antitumor T-cell responses, and decreased tumor growth (51, 52).
We found a striking association between gene signatures representative of myeloid inflammation and disease outcomes after nephrectomy that was validated in two independent cohorts and maintained significant prognostic value on multivariate analysis. Our findings are supported by a recent study identifying an association of C1Q+TREM2+APOE+ TAMs with recurrence in ccRCC in a small cohort (53). Since our MSKI signature captures but is not specific to this C1Q+ TAM population (18), future functional validation using higher resolution tools is needed. Although the basis of our signature is not novel, the genes comprising the adenosine and myeloid signatures were refined via TCGA analysis so that only the most relevant genes were combined to create the MSKI signature. This is the first reported application of these signatures in the localized disease recurrence setting.
Our analysis assessing tumor CD8 T-cell density, lymphocyte infiltration patterns, exhaustion states, and tumor PD-L1 expression failed to detect an association with disease outcomes after nephrectomy. There have been conflicting data on the prognostic significance of intratumoral CD8 cell counts in ccRCC (24, 25, 54, 55), and our data add to the studies that did not find an association with tumor recurrence. A recent scRNA-seq study reported increasing interactions between M2-like TAMs and exhausted CD8 T cells with advancing ccRCC disease stage (26), and the IMmotion150 trial also suggested that myeloid cells can suppress ICB-induced antitumor T-cell responses in ccRCC (15). However, in the present study, myeloid abundance affected neither T-cell exhaustion nor the prognostic ability of Teff signatures. In line with this, Şenbabaoğlu and colleagues (56) reported two T cell–enriched clusters in the TCGA cohort (termed TCa and TCb), of which the TCb cluster had significantly greater macrophage infiltration and worse cancer-specific survival, whereas TCa clusters had similar survival to non- and poorly infiltrated groups, suggesting that T cells may not be prognostic. No TP53 mutations were reported in the 13 patients profiled by scRNA-seq (26), perhaps highlighting that metastatic progression of the SPMC tumor subset is not governed by T cells and potentially points toward myeloid regulation of earlier events in the metastatic cascade such as stromal remodeling or EMT (57).
Patients with highly angiogenic tumors in the PROTECT cohort were less likely to develop metastatic disease and had more favorable OS compared with poorly angiogenic tumors. Although we did not detect any significant associations between angiogenesis and disease recurrence in the TCGA-KIRC cohort, we noted poorly angiogenic tumors had significantly worse OS. Interestingly, analyses from recent clinical trials in mccRCC suggest that angiogenesis signatures may constitute a biomarker of response to anti-VEGF agents (7, 15, 16, 19). Taken together with our data showing enrichment of angio-low tumors in mccRCC patients, which are intrinsically less responsive to anti-VEGF agents, this may provide an explanation for the lack of efficacy of previous anti-VEGF TKI adjuvant trials.
In order to facilitate rapid basic and translational mRCC research, we generated a relevant, syngeneic model of mccRCC on a pure C57BL/6 background using electroporation to deliver CRISPR/Cas9 vectors to murine organs in vivo. These methods have been successful in initiating tumor development in a number of different cancer types, including liver, pancreatic, prostate, and ovarian (58–62). This is the first report of this approach in the kidney, and we experienced very low tumor penetrance. Tumor induction in murine kidneys has been extremely challenging, likely owing to its intrinsic low cell turnover, limited regenerative capacity, and tumor-suppressor rather than oncogene-driven nature (63, 64). Although we were fortunate to generate a syngeneic cell line, further optimization of the electroporation approach is greatly needed to enable rapid and flexible genome editing as originally intended. Nevertheless, our model offers a molecularly relevant alternative to the only available RCC syngeneic line, RenCa, which does not harbor Vhl loss, the most frequent initiating genomic event in ccRCC (64). LVRCC67 tumors transcriptomically resemble a TP53 mutation–enriched molecular subtype of human ccRCC characterized by high cell proliferation, EMT, myeloid cells, and poor prognosis. Encouragingly, the genomically matched KVpR autochthonous mouse model has a similar transcriptomic and immune profile (65), supporting that key aspects of the tumorigenesis process in our implantation model may be conserved.
Our data support a potential role for TP53 loss in driving metastases in ccRCC. Although there was no significant difference in TP53 genomic alterations in primary tumors between ccRCC patients with or without metastases in the recently described MSK-MET cohort (66), multiple IHC-based studies have shown TP53 alterations to be significantly correlated with poor DFS in RCC (67–73). We found striking transcriptional similarities between our p53-driven mouse model and human TP53-enriched myeloid-dominant SPMS tumors, as well as similar enriched pathways within recurrent tumors of the PROTECT cohort. We observed enrichment of the cell-cycle and EMT pathways in recurrent human and mouse tumors, which may reflect the reported roles of p53 in regulating these processes (74, 75).
We linked TP53 mutations with higher MSKI scores in the COMPARZ and JAVELIN101 cohorts. Several studies in autochthonous mouse models (75–77) suggest a direct role of tumor cell p53 loss in immunosuppressive myeloid recruitment, differentiation, and function. This may occur through enhanced release of myeloid chemokines from p53-null tumor cells, including CXCL17, IL6, CXCL3, and CXCL5, three of which contribute to the MSKI gene signature (76, 77). Additionally, IL1β (also a component of the MSKI signature) release from TAMs can be induced by p53 loss–mediated Wnt autocrine signaling, which subsequently promotes γδ T cell–mediated expansion of systemic cKIT+ immature neutrophils that facilitate metastasis, with no reported changes in CD8 and CD4 T cells (78). It is possible that this tumor cell p53–TAM–neutrophil axis also governs metastases in RCC, since a high neutrophil-to-lymphocyte ratio is a known prognostic factor (79).
Using our SPMC-like mouse model, we evaluated the efficacy of myeloid inhibition on disease recurrence. Fong and colleagues have shown that patients with high adenosine signatures have greater responses to the adenosine A2AR inhibitor CPI-444 (14). Indeed CPI-444 reduced primary and metastatic tumor burden in our MSKI-rich tumors. Importantly, the durability of response needs to be evaluated in future experiments in which the primary tumor is resected, enabling longer follow-up times.
Interestingly we observed a lack of efficacy and associated changes in T-cell phenotypes following anti–PD-1 treatment in mice. Molecular subgroup analyses indicate that ccRCC patients harboring the SPMC subtype of tumors are not responsive to first-line atezolizumab plus bevacizumab (38). This raises concerns that despite overall efficacy with anti–PD-1 monotherapy seen in the KEYNOTE-564 trial (1), this subtype may also preclude response to ICB in the adjuvant setting. However, whether SPMC tumors also respond poorly to other more efficacious ICB VEGF combination approaches remains to be determined, although there does not seem to be an association between TP53 mutations and response to avelumab and axitinib (16).
Our data suggest that targeting the adenosine pathway may be beneficial in this myeloid-high and angiogenesis-low setting. In our study, we used an adenosine A2AR antagonist, but adenosine signaling via the immunosuppressive A2B receptor or other components involved in adenosine regulation may constitute a resistance mechanism. Our mouse model provides a valuable means to test additional adenosine pathway inhibitors, which are currently in the early phases of clinical development (42), for RCC adjuvant therapy.
Our study has limitations. Although the PROTECT cohort received no prior systemic therapies, overall survival analyses may be confounded by the treatment given in the metastatic disease setting. We were not able to obtain sequencing data from patients enrolled on the treatment arm of the PROTECT trial. However, we do speculate that, based on our prior transcriptomic analysis from the COMPARZ trial (19), myeloid-high tumors would likely see worse outcomes compared with myeloid-low tumors in patients receiving pazopanib. In the MOFFITT cohort, the MSKI score was significantly associated with recurrence but only trending for OS, potentially due to the relatively smaller patient population, necessitating validation in larger independent cohorts. Although the PROTECT cohort has a skewed distribution of T stages with no pT1 tumors and consisting of 73.9% pT3a tumors, the MSKI signature was prognostic in the TCGA cohort, suggesting relevance beyond adjuvant trial populations.
In summary, this study highlights the crucial independent role of myeloid inflammation in the development of metastatic ccRCC. In contrast to the pervading paradigm, this axis does not seem to be reliant on CD8 T-cell effector function or regulated through well-established immune checkpoints. We posit that a substantial patient subgroup with TP53-altered, treatment-refractory ccRCC tumors may derive benefit from the blockade of pleiotropic immunosuppressive adenosine signaling, and we provide a valuable preclinical model to advance our understanding and treatment of this aggressive subtype. More generally, this work emphasizes the importance of molecular subtyping for patient selection in future adjuvant trials.
The phase III PROTECT (Pazopanib As Adjuvant Therapy in Localized/Locally Advanced RCC After Nephrectomy) study enrolled >1,500 patients with resected pT2 (high grade), ≥pT3, or N1 (any grade) ccRCC with 1:1 randomization to 1 year of pazopanib versus placebo. None of the patients had received any prior systemic therapy. As previously reported (17), the study failed to demonstrate DFS benefit of adjuvant pazopanib. All patients provided archival tumor tissue at the time of enrollment. We analyzed tumor tissue as well as demographic and clinical data from patients enrolled on the placebo arm of the trial, which was conducted in accordance with the Declaration of Helsinki. All patients provided written informed consent for participation in the clinical trial (17).
We obtained 84 tumor samples from 84 patients with RCC, from 2004 to 2018, through protocols approved by the institutional review board (H. Lee Moffitt Cancer Center and Research Institute's Total Cancer Care protocol; Advarra IRB Pro00014441). Samples were obtained via surgical excision. None of the patients had received any prior systemic therapy. Written informed consent was obtained from all tissue donors. Patients were included if they had clinically localized ccRCC, provided written consent to the molecular characterization of their tissue, and did not receive any systemic therapy prior to tissue collection.
Patient tumor samples then underwent microdissection and bulk RNA-seq. The TruSeq RNA Exome kit (Illumina, Inc.) for 50 million 100-bp paired-end reads was utilized, and RNA-seq reads were aligned to the human reference genome in a splice-aware fashion using Spliced Transcripts Alignment to a Reference (STAR; ref. 80), allowing for accurate alignments of sequences across introns. Aligned sequences were assigned to exons using the HTSeq package (81) to generate initial counts by region. Normalization, expression modeling, and difference testing were performed using DESeq2 (82).
This cohort was composed of 21 patients undergoing partial or radical nephrectomy with pathologic features consistent with those required for eligibility on the PROTECT trial (17). Briefly, patients were >18 years of age with nonmetastatic clear-cell or predominantly clear-cell ccRCC histology, underwent partial or radical nephrectomy, and had intermediate- or high-risk disease as defined by the SSIGN system. Formalin-fixed, paraffin-embedded (FFPE) tumor blocks were collected for histopathologic and IF analyses.
MSKCC Multiregional RNA-seq Cohort
This cohort consisted of two to seven tumor regions from 29 patients obtained from the operating room during nephrectomy. At the time of specimen extraction, samples 1 to 1.5 cm in the largest dimension were obtained by the treating surgeon (A.A. Hakimi) from spatially distinct tumor regions (at least 1 cm apart).
RNA-seq raw read sequences were then aligned against human genome assembly hg19 (Feb.2009/GRCh37; https://genome.ucsc.edu/cgi-bin/hgGateway?db=hg19) or mm10 (Dec.2011/GRCm38; https://genome.ucsc.edu/cgi-bin/hgGateway?db=mm10) for human or mouse samples respectively by STAR 2-pass (80) alignment. RNA-seq gene-level count values were computed using the R package GenomicAlignments (84) over aligned reads with UCSC KnownGene (85) in hg19 or mm10 as the base gene model. The Union counting mode was used, and only mapped paired reads were considered. Fragments per kilobase million (FPKM) values were then computed from gene-level counts using the fpkm function from the R package DESeq2 (82).
Two hundred thirty-six patients had microarray and clinicopathologic data available for analyses. RNA from the primary tumors was extracted by AltheaDx in 2013 using the Qiagen RNeasy FFPE Kit with a modified deparaffinization step. Gene expression profiles were derived via Affymetrix GeneChip HTA 2.0 (Affymetrix). Microarray RNA data were normalized to log2 value. Probes without corresponding gene symbols found were excluded from further analysis. For genes matched with multiple probes, the probe with maximum median absolute deviation was chosen for representing the expression of the gene. The log2-normalized expression values were used for immune deconvolution analyses.
RNA Expression–Based Gene Signatures
Gene signatures were defined as follows: angiogenesis (15): VEGFA, KDR, ESM1, PECAM1, ANGPTL4, and CD34; Teff (15): CD8A, EOMES, PRF1, IFNG, and CD274; myeloid (15): IL6, CXCL1, CXCL2, CXCL3, CXCL8, and PTGS2; Javelin (16): CD3G, CD3E, CD8B, THEMIS, TRAT1, GRAP2, CD247, CD2, CD96, PRF1, CD6, IL7R, ITK, GPR18, EOMES, SIT1, NLRC3, CD2, CD96, PRF1, CD244, KLRD1, SH2D1A, CCL5, XCL2, CST7, GFI1, KCNA3, and PSTPIP1; adenosine (14): CXCL1, CXCL2, CXCL3, CXCL5, CXCL6, CXCL8, IL1B, and PTGS2; MSKI: CXCL1, CXCL2, CXCL3, CXCL5, CXCL8, IL6, and IL1B; immune checkpoint: CD274, CTLA-4, HAVCR2, LAG3, PDCD1, PDCD1LG2, and TIGIT; and immunosuppression (27): CD274, IDO1, FASLG, CTLA-4, PDCD1, LAG3, HAVCR2, PDCD1LG2, IL10, TNF, TGFB1, IL12A, and PTGS2. ssGSEA takes the sample gene expression values as the input and computes an overexpression measure for the given gene list of immune cell type relative to all other genes in the transcriptome. The specific method of deriving angiogenesis, Teff, myeloid, and Javelin scores was according to their corresponding publications. Other TME feature gene enrichment scores were quantified using the ssGSEA implementation through the R package “gsva” (86, 87). ESTIMATE was used to calculate an immune score, which is the estimate of immune cell component in tumor tissue and is calculated through the “estimate” R package (88) based on a given gene expression profile in FPKM or normalized log2-transformed values.
Analysis of DEGs
The R package “limma” (version 3.32.10) was used for DEGs (89). Limma returned empirical Bayes moderated-t P values and adjusted P values (Q-value) to correct for multiple comparisons testing using the Benjamini–Hochberg method to control the false discovery rate. Genes with mean expression (log2 value) greater than 1, a nominal P value less than 0.05, and fold change greater than 20% were subjected to IPA.
IPA (Ingenuity) was used for running canonical pathway, upstream regulator, disease, and biofunction analyses over the genes differentially expressed between patients with a disease recurrence and those without disease recurrence.
Sorted cell populations were derived from 21 patients with renal cell carcinoma with a total of 127 unique samples representing 16 distinct flow-sorted populations. Of these, 106 were sorted from tumor tissues and 21 monocyte samples were sorted from peripheral blood (n = 20) or normal kidney (n = 1). For bulk RNA-seq, immune populations were sorted on Aria II (Becton Dickinson) into TRIzol LS and flash frozen. RNA was extracted with chloroform. Isopropanol and linear acrylamide were added, and the RNA was precipitated with 75% ethanol. Total RNA (0.649–1 ng) with RNA integrity numbers 6.8 to 10 underwent amplification using the SMART-Seq v4 Ultra Low Input RNA Kit (Clontech; cat. #63488). Amplified cDNA (15 ng) was used to prepare libraries with the KAPA Hyper Prep Kit (Kapa Biosystems, KK8504) using 8 cycles of PCR. Samples were barcoded and run on a HiSeq 2500 in 50 bp/50 bp paired-end run using the HiSeq3000/4000 SBS Kit (Illumina, Inc.) for 30 to 40 million paired reads. The raw FASTQ files were processed as described above.
IHC for the PROTECT Cohort
Evaluation of multiple exploratory markers (targets) using IHC analysis was done using the following primary antibodies: CD8 (C8/144B; Dako), CD68 (KP-1; Ventana), PD-L1 (28-8; Dako), TIM3 (D5D5RTM; Novartis), and LAG3 (17B4; Novus Biologicals). The Ventana BenchMark XT autostainer was used with the following detection systems: Ultraview DAB (CD8 and CD68), Optiview DAB with (TIM3) or without (LAG3) amplification. The Envision DAB detection system was used for PD-L1 staining using the Dako Link 48 platform. PD-L1 IHC positivity was determined as a percentage of stained tumor cells (tumor positivity score). Batch analysis of CD8 IHC, CD68 IHC, LAG3 IHC, and TIM3 IHC was used to determine the amount of IHC staining by measuring the area (marker area) of the IHC-stained cells or structures in the center and periphery of a tumor. All IHC-stained slides were evaluated by a pathologist using the universal immune cell marker scoring algorithm. The output of this analysis is the relative surface area (%) of the tumor with a marker-positive immune cell density that belongs to one of the eight data fields per IHC slide. There are four fields for intrastromal marker-positive immune cells and four fields for intraepithelial marker-positive immune cells. In addition, the immune phenotype was scored as “desert,” “inflamed,” or “excluded” in the IHC slides. CD8 cell density measures for the intratumoral and stromal compartments were divided into high and low based on a median cutoff. Excluded tumors were defined as high stromal and low intratumoral CD8 cell densities. Desert tumors were defined as low stromal and intratumoral CD8 cell densities. The rest, including high stromal/high intratumoral and low stromal/high intratumoral, were defined as inflamed. In addition, and only for TIM3, the percentage of marker-positive tumor cells was estimated.
IF for the MSKCC Cohort
The IF detections of CD31 and CD68 were performed at the Molecular Cytology Core Facility of MSKCC using the Discovery Ultra processor (Ventana Medical Systems, Roche-AZ). After 32 minutes of heat and Cell Conditioning 1 (CC1; Ventana; cat. #950-500) retrieval, the tissue sections were blocked first for 30 minutes in Background Blocking reagent (Innovex; cat. #NB306). A mouse monoclonal anti-CD31 antibody (Ventana-Roche; cat. #760-4378) was used in prediluted serum. The incubation with the primary antibody was done for 5 hours, followed by biotinylated anti-mouse secondary (Vector Labs, MOM Kit BMK-2202) in 5.75 μg/mL Blocker D, streptavidin–HRP, and TSA Alexa488 (Life Tech; cat. #B40932) prepared according to the manufacturer's instruction in 1:150 for 16 minutes. A mouse monoclonal IgG1 anti-CD68 antibody (Dako; cat. #M0814) was used in 0.02 μg/mL concentrations. The incubation with the primary antibody was done for 5 hours, followed by biotinylated goat anti-mouse secondary (Vector Labs, MOM Kit BMK-2202) in 5.75 μg/mL Blocker D, streptavidin–HRP, and CF 543 (Biotium; cat. #92172) prepared according to the manufacturer's instruction in 1:500 for 16 minutes. All slides were counterstained in 5 μg/mL 4′,6-diamidino-2-phenylindole (DAPI; Sigma; D9542) for 5 minutes at room temperature, mounted with antifade mounting medium Mowiol [Mowiol 4-88 (CALBIOCHEM code: 475904)], and coverslipped.
Slides were scanned with Pannoramic P250 Flash scanner (3DHistech) using 20×/0.8 NA objective lens. Digitized images were viewed using CaseViewer software (3DHistech), where two to three representative regions of interest (ROI), each 2 mm2 in area, were selected from each tumor and then exported as .tif files. These regions were then analyzed using ImageJ/Fiji software (v2.1.0; NIH; ref. 91).
IF for the MOFFITT Cohort
Prior to IF, tissue blocks were prepared from macrodissected RCC tumor specimens, and an experienced genitourinary pathologist (J. Dhillon) reviewed each slide obtained from FFPE tissue samples and annotated three spatially distinct ROIs from the tumor core and three ROIs from the stroma. The tumor-core ROIs studied were selected regions with high tumor cell content and without evidence of grossly necrotic tissue.
Tissue samples were then stained using the PerkinElmer OPAL 7 Color Automation IHC Kit (PerkinElmer) on the BOND RX Autostainer (Leica Biosystems). In brief, tissue slides were stained using antibodies against CD8 [CD8 (C8/144B) Mouse monoclonal antibody, Agilent Dako; ID: GA623; RRID: AB_2075537].
All subsequent steps, including deparaffinization, antigen retrieval, and staining, were performed using the OPAL manufacturer's protocol. Pan-cytokeratin (Pan-CK) and DAPI counterstaining were applied to all slides, and imaging was performed using the Vectra3 Automated Quantitative Pathology Imaging System (PerkinElmer). Importantly, Pan-CK was applied only after confirmation of tumor content by examination of hematoxylin and eosin images for each ROI. Multilayer TIFF images were exported from InForm (PerkinElmer) and loaded into HALO (Indica Labs) for quantitative image analyses (34). The size of the ROIs was standardized at 1,356 × 1,012 pixels with a resolution of 0.5 μm/pixel for a total surface area of 0.343 mm2. Using HALO, for each staining marker a positivity threshold within either the nucleus or cytoplasm was set, and the entire image set was analyzed. From this analysis, generated data included the total cell count, positive cell counts for CD8, fluorescence intensity of every individual cell, percentage of cells that were positive for CD8, and Cartesian coordinates for their location in the ROI image. Median cutoffs for CD8 cell density (cells per mm2) in the tumor core and stroma zones were determined. Tumors with high stromal and low tumor CD8 cell density were defined as excluded, low stromal and low tumor CD8 cell density were defined as desert, and the remainder were considered inflamed (i.e., high tumor and low stromal CD8 counts, high tumor and high stromal CD8 counts).
All mouse experiments were approved by the MSKCC Internal Animal Care and Use Committee. Mice were maintained under specific pathogen–free conditions, and food and water were provided ad libitum. Mice were purchased from The Jackson Laboratory.
Electroporation-Derived Syngeneic Model
To generate electroporation-derived tumors, sgRNAs targeting Vhl (CCCGGTGGTAAGATCGGGTA), Tp53 (ACCCTGTCACCGAGACCCC), and Rb1 (TGCGCGGGGTCGTCCTCCCG) were cloned into a pX330 CRISPR/Cas9 vector (Addgene #42230). A pX330 vector containing two sgRNA cassettes was cloned using the previously described approach (61). These recombinant pX330 plasmids (20 μg each) were mixed with a c-myc-PT3EF1a transposon vector (5 μg) and a Sleeping Beauty transposase (SB13; 1 μg) in sterile saline for injection. All plasmids were purified using low endotoxin kits (Plasmid Plus Kits; Qiagen). The SB13 and c-myc-PT3EF1a transposon vector were gifts from Josef Leibold (MSKCC) via Xin Chen (University of California, San Francisco; Addgene plasmid #92046). The pX330 vector was a gift from Feng Zhang of the Broad Institute (Addgene plasmid #42230).
For surgery, mice were anesthetized using isofluorane and given the analgesics buprenorphine, meloxicam, and bupivacaine according to institutional guidelines. With the mouse in right lateral recumbency, a ∼1 cm left paracostal (vertical) incision was made through the skin and peritoneum in the flank just caudal to the last rib. The kidney was gently exteriorized using blunt forceps. A 50-μL plasmid solution was injected using a 30-G needle inserted at a shallow angle just below the renal capsule so that a visible bleb was formed between the renal capsule and parenchyma. The injection site was electroporated using the NEPA21 electro-kinetic transformation system (Nepa Gene, Bulldog Bio) with CUY661-3 × 7 electrode tweezers using two consecutive positive poring pulses with 50 V, 30 ms duration, 450 ms interval, and 10% decay rate, followed by five consecutive positive transfer pulses with 30 V, 50 ms duration, 450 ms intervals, and 40% decay rate, followed by five consecutive negative transfer pulses with the same settings. The kidney was replaced in the abdominal cavity, and the wounds were opposed. Mice were monitored for tumor development by manual palpation and confirmed by ultrasound and MRI imaging.
The LVRCC67 cell line was derived by dissociation of the primary tumor using 0.2 mg/mL liberase TM (Roche) and 10 μg/mL DNase I (DN25, Sigma) in PBS at 37°C with shaking for 30 minutes. Cells were washed and passed through a 45-μm cell strainer and cultured in previously described K1 media (65). Cells were passaged twice weekly at a 1:8 dilution. Cells were tested for Mycoplasma using the MycoAlert kit (Lonza; latest date tested April 4, 2022), and early passage cells (P2–P5) were used in all experiments. Genome editing in the electroporation-derived tumor and cell line was confirmed by cut-site PCR and Sanger sequencing, followed by sequence deconvolution using the Synthego ICE CRISPR Analysis Tool (Synthego; https://ice.synthego.com/#/).
For the syngeneic transplantation model, 5 × 105 LVRCC67 cells in 100 μL of PBS were injected subcutaneously into both flanks of 6- to 8-week-old male C57BL/6 mice (The Jackson Laboratory). Mice were randomized by cages to receive either 100 mg/kg CPI-444 (4,000 mg/kg medicated chow; produced and irradiated by Research Diets, provided by Corvus Pharmaceuticals), control chow (Teklad 2018 standard diet), IgG isotype control (2A3, BioXcel, 100 μg), or anti–PD-1 (RMP1-14, BioXcel 100 μg) antibodies administered intraperitoneally in 100 μL of PBS twice weekly. Tumor volumes were measured twice weekly using digital calipers and calculated using the formula π/6 × (length × width)3/2. Mice were euthanized with carbon dioxide prior to necropsy. Metastatic tumor burden was determined by gross examination of the liver, lungs, kidney, and spleen.
IHC for Mouse Tumors
To characterize the LVRCC67 parental tumor, FFPE sections were dewaxed in xylene and rehydrated through graded ethanols into ddH2O, and antigen retrieval was performed using an autoclave for 15 minutes in preheated citrate-based antigen unmasking solution (except for chromogranin A, which required Tris/ETDA-based antigen retrieval solution; Vector Laboratories). Endogenous peroxidases were quenched using BLOXALL Blocking Solution (Vector Laboratories), and nonspecific binding was blocked using 2.5% normal serum matched to the species of the secondary antibody. For rabbit primary antibodies (Pax8, cat. #10336-1-AP, Proteintech, 1:1,000; HIF1α, clone D1S7W, Cell Signaling Technology, 1:400; Aqp1, #Ab15080, Abcam, 1:500; c-myc, clone Y69, Abcam, 1:100; CA9, clone EPR23055-5, Abcam, 1:500; Megalin, #Ab76969, Abcam, 1:1,000; Glut1, #Ab14683, Abcam, 1:500; Synaptophysin, clone YE269, Abcam, 1:400; chromogranin A, clone EPR22537-248, 1:1,000), the ImmPRESS Excel Amplified Polymer Staining Kit was used as recommended by the manufacturer. For mouse primary antibodies (Pan-CK, clone AE1/AE3, 1:1,000; p53, clone 1C12, Cell Signaling Technology, 1:100; Rb1, clone G3245, BD Pharmingen, 1:100), the VECTASTAIN Elite ABC-HRP Universal Kit was used according to the manufacturer's instructions with biotin/avidin block but also included an additional 1-hour block with mouse-on-mouse block [Goat F(ab) Anti-Mouse IgG H&L; 1:50 dilution; Abcam] prior to the addition of primary antibodies. For goat primary antibodies (CD10; cat. #PA5-47075; Invitrogen; 1:1,000), the VECTASTAIN Elite ABC-HRP Kit was used, but biotinylated rabbit-anti-goat secondary (1:200; Vector Laboratories) was used in place of the biotinylated universal antibody. Finally, slides were dehydrated through a reverse ethanol gradient into xylene, mounted, and coverslipped in Cytoseal XYL (Thermo Scientific). Representative brightfield images were captured using a Zeiss microscope fitted with an AxioCam HRC Camera.
For the validation of gene deletions and kidney cell origin of the LVRCC67 cell line, whole-cell extracts were prepared using CellLytic M (Sigma) and sonication for 10 minutes in a 4°C water bath. After centrifugation at 21,000 rpm, supernatants were mixed with 6× loading dye containing 2-mercaptoethanol and run on precast Novex 4% to 12% Bis-Tris gels (Invitrogen) for 90 minutes at 130 V using the Bio-Rad Transblot system. Proteins were transferred onto methanol-activated PVDF membranes for 1 hour using 100 V. Membranes were blocked with 5% milk and incubated at 4°C overnight in 5% BSA containing primary antibodies (Pax8, cat. #10336-1-AP, Proteintech 1:2,500; HIF1α, clone D1S7W, Cell Signaling Technology, 1:1,000; c-myc, clone Y69, Abcam, 1:1,000; Rb1, cat. #9301S, Cell Signaling Technology, 1:500; p53, clone 1C12, Cell Signaling Technology, 1:1,000; CD10, cat. #PA5-47075, Invitrogen, 1:2,000; HSP-90, clone C45G5 Cell Signaling Technology, 1:3,000; β-actin, clone D6A8, 1:5,000), followed by secondary antibody incubation the following day for 1 hour at room temperature.
For mouse RNA-seq, three normal kidney cortexes and four snap-frozen tumors collected at experimental endpoint were submitted to the Integrated Genomics Core for isolation of RNA and library preparation using Illumina TruSeq stranded mRNA (polyA); RNA was sequenced to a depth of 40 to 50 million 100-bp paired-end reads per sample. RNA-seq raw read sequences were then aligned as described above.
Resected tumors were collected in DMEM + 10% FBS on ice, washed in PBS, and digested in PBS containing 0.2 mg/mL Liberase TM (Roche) and 10 μg/mL DNase I (DN25, Sigma) at 37°C with shaking for 30 minutes. The digestion reaction was quenched using an equal volume of 10 mmol/L EDTA PBS, followed by straining through 70-μm and 45-μm cell strainers and red blood cell lysis (Ack lysis buffer; Gibco). Single-cell suspensions were stained with viability dye (Ghost Dye Violet 510; Tonbo Bioscience) and Fc block (TruStain FcX; BioLegend) followed by surface antibodies. Surface antibodies were fixed, and cells were permeabilized to allow subsequent staining of intracellular antibodies using the Foxp3/transcription factor staining buffer kit (Tonbo Bioscience) according to the manufacturer's instructions. The following antibodies were used: CD80-FITC, 16-10A1, 1:100; CD206-PCPCy5.5, C068C2, 1:200; Ly6C-APC-Cy7, AL-21, 1:300; IRF8-PE, V3GYWCH, 1:200; Ly6G-PE-Cy7, 1A8, 1:800; F4/80-PE-CF594, T45-2342, 1:200; MHC-II-BV605, M5/114.15.2, 1:300; CD45-BV650, 30-F11, 1:500; CD11b-BV711, M1/70, 1:400; PD-1-FITC, RMP1-30, 1:200; TCRγδ-PerCPeFluor710, eBioGL3, 1:200; granzyme B-APC, GB11, 1:100; CD8α-APC-Cy7, 53-6.7, 1:200; TCRβ-PE, eBio4B10, 1:100; NK1.1-PE-Cy7, PK136, 1:300; CD4-PE-TexasRed, RM5-5, 1:200; Foxp3-BV421, FJK-16S, 1:200; CD45-BV605, 30-F11, 1:200; and Ki-67-BV650, B56, 1:75. Stained single-cell suspensions were analyzed using the Fortessa flow cytometric analyzer (BD) and FlowJo software (Treestar). The gating strategies are shown in Supplementary Fig. S5.
DNA from the LVRCC67 cell line pellet or normal kidney of a WT C57BL/6 mouse was extracted and sequenced at 150× or 100×, respectively. Libraries were prepared using Roche's KAPA Hyperchemistry and Twist's exome capture by the Integrated Genomics Operation at MSKCC. FASTQ files were aligned to mm10 reference genome with bwamem v.0.7.15. Further base-quality score recalibration and duplicate-read removal were performed using the Genome Analysis Toolkit (GATK) version 126.96.36.199 following raw reads alignment guidelines (92). Somatic mutations were called with VarScan v.2.4.3, Strelka v.2.9.10, Platypus 0.8.1, Mutect2 (part of GATK 188.8.131.52), and SomaticSniper 184.108.40.206 (for single-nucleotide variants only). For single-nucleotide variants and indels, the following filtering parameters were applied: called by at least 2 out of 5 callers, covered by >50 reads in tumors and >7 reads in normals, with >10 reads supporting the mutation, mutant allele fraction ≥10%, and <1% reads supporting the mutation in the normal (Tcov >10 and Tac >4 and Taf ≥0.04 and Ncov >7 and Naf ≤0.01). Variants with Tcov <20 or Tac <4 are marked as low_confidence. Indels reside in the blacklisted regions (https://www.encodeproject.org/annotations/ENCSR636HFF/), and low mappability regions (such as repeat maskers) are excluded. Copy-number variations were assessed using FACETS v.0.6.0 (93). The fraction of the copy number–altered genome was defined as the fraction of the genome with either nondiploid copy number or evidence of loss of heterozygosity.
Differences between groups were tested using Fisher exact test and χ2 test (SAS 9.4, SAS Institute Inc.) for categorical variables and the Wilcoxon rank-sum test (SAS 9.4 and R package stats version 4.1.0) for continuous variables between sample groups. The Kruskal–Wallis test (R package “stats” and SAS 9.4) was used for comparisons between three or more groups of continuous variables. Patients who were lost to follow-up or alive at the time of the study end date were treated as censored events. Survival curves were calculated according to the Kaplan–Meier method, and differences between curves were assessed using the log-rank test. The HR estimates and 95% CIs were determined by the Cox proportional hazards regression modeling (SAS 9.4) and MatSurv (94). All statistical analyses were post hoc; statistical significance was set at P < 0.05.
The data from the PROTECT cohort are available on request from the relevant Data Access Committee (e.g., https://ega-archive.org/studies/EGAS00001006344). Access to the TCGA cohort clinicopathologic, genomic, and transcriptomic data is detailed in the “TCGA-KIRC Cohort” section. Data from the MSKCC and MOFFITT cohorts are available upon request from the corresponding authors.
J. Leibold reports grants from the German Research Foundation (DFG) and a Shulamit Katzman Endowed Postdoctoral Research Fellowship during the conduct of the study. M. Golkaram is a shareholder in and an employee of Illumina, Inc. M. Marker reports other support from Novartis Pharmaceuticals during the conduct of the study. A. Reising reports other support from Novartis Pharmaceuticals outside the submitted work. A. Savchenko is a full-time Novartis employee. J. Millholland reports other support from Novartis outside the submitted work. Y.-B. Chen reports grants from the NIH during the conduct of the study. B.J. Manley reports personal fees from Merck outside the submitted work. V. Makarov reports a patent for EP3090066A2, determinants of cancer response to immunotherapy, issued. D.B. Solit reports personal fees from Pfizer, Loxo/Lilly Oncology, Vividion Therapeutics, Scorpion Therapeutics, Fore Therapeutics, and BridgeBio outside the submitted work. T.A. Chan reports other support from Gritstone, grants and personal fees from Illumina, Inc. and Nysnobio, and grants from PGDx, Pfizer, and Innovenn during the conduct of the study. R.J. Motzer reports personal fees from Novartis and grants from Novartis during the conduct of the study, as well as personal fees from Genentech/Roche, Pfizer, Eisai, Exelixis, EMD Serono, Merck, Aveo, and Calithera, and grants from Genentech/Roche, Pfizer, Eisai, Exelixis, Merck, Aveo, and Bristol Myers Squibb outside the submitted work. M.H. Voss reports personal fees from Pfizer, Exelixis, Eisai, Merck, Calithera, Aveo, Roche/Genentech, Affimed, Oncorena, and MicuRx outside the submitted work. A.A. Hakimi reports personal fees from Merck (advisory board) outside the submitted work. No disclosures were reported by the other authors.
P.M. Rappold: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. L. Vuong: Conceptualization, resources, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. J. Leibold: Resources, data curation, formal analysis, visualization, methodology, writing–review and editing. N.H. Chakiryan: Resources, data curation, software, formal analysis, visualization, methodology, writing–review and editing. M. Curry: Conceptualization, data curation, software, formal analysis, methodology, writing–review and editing. F. Kuo: Conceptualization, data curation, software, formal analysis, validation, methodology, writing–review and editing. E. Sabio: Data curation, formal analysis, investigation, methodology, writing–review and editing. H. Jiang: Data curation, formal analysis, visualization, methodology, writing–review and editing. B.G. Nixon: Resources, data curation, software, formal analysis, supervision, validation, visualization, methodology. M. Liu: Conceptualization, resources, data curation, software, formal analysis, supervision, validation, methodology. A.E. Berglund: Conceptualization, data curation, software, formal analysis, supervision, validation, investigation, methodology. A.W. Silagy: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology. E.A. Mascareno: Data curation, formal analysis, supervision, investigation, visualization, methodology. M. Golkaram: Data curation, formal analysis, supervision, investigation, visualization, methodology, writing–review and editing. M. Marker: Data curation, formal analysis, supervision, investigation, methodology, writing–review and editing. A. Reising: Data curation, formal analysis, supervision, investigation, methodology, writing–review and editing. A. Savchenko: Data curation, formal analysis, supervision, investigation, methodology, writing–review and editing. J. Millholland: Data curation, formal analysis, supervision, investigation, methodology, writing–review and editing. Y.-B. Chen: Data curation, formal analysis, supervision, methodology, writing–review and editing. P. Russo: Conceptualization, data curation, formal analysis, supervision, methodology, writing–review and editing. J. Coleman: Conceptualization, data curation, formal analysis, supervision, methodology, writing–review and editing. E. Reznik: Conceptualization, data curation, formal analysis, supervision, methodology, writing–review and editing. B.J. Manley: Conceptualization, resources, data curation, formal analysis, supervision, methodology, writing–original draft, project administration, writing–review and editing. I. Ostrovnaya: Conceptualization, resources, data curation, software, formal analysis, supervision, methodology, writing–original draft, project administration, writing–review and editing. V. Makarov: Conceptualization, resources, data curation, software, formal analysis, supervision, methodology, writing–original draft, project administration, writing–review and editing. R.G. DiNatale: Resources, data curation, software, formal analysis, visualization, methodology. K.A. Blum: Resources, data curation, formal analysis, visualization, methodology. X. Ma: Resources, data curation, formal analysis, supervision, visualization, writing–review and editing. D. Chowell: Resources, data curation, formal analysis, supervision, writing–review and editing. M.O. Li: Resources, data curation, formal analysis, supervision, writing–review and editing. D.B. Solit: Resources, data curation, formal analysis, supervision, writing–review and editing. S.W. Lowe: Resources, data curation, software, formal analysis, supervision, writing–review and editing. T.A. Chan: Resources, data curation, software, formal analysis, supervision, writing–review and editing. R.J. Motzer: Conceptualization, resources, data curation, software, formal analysis, supervision, methodology, writing–original draft, project administration, writing–review and editing. M.H. Voss: Conceptualization, resources, data curation, formal analysis, supervision, methodology, writing–original draft, project administration, writing–review and editing. A.A. Hakimi: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing–original draft, project administration, writing–review and editing.
We thank Corvus Pharmaceuticals for kindly providing the CPI-444 medicated and control chow, as well as the valuable services and advice received from the integrated genomics, flow cytometry, small animal imaging, and molecular cytology core facilities at MSKCC.
This study was supported by NIH 5P30CA008748-56 MSK Cancer Center Support Grant (A.A. Hakimi); NIH 5R01CA258886 (A.A. Hakimi); Department of Defense KC180165 (A.A. Hakimi); Cycle for Survival (A.A. Hakimi); Weiss Family Funds (A.A. Hakimi); the Oncology Research Information Exchange Network (ORIEN) Avatar Project in collaboration with the H. Lee Moffitt Cancer Center and Research Institute and supported by the Tissue Core, Biostatistics, and Bioinformatics Shared Resource Core Facilities at the H. Lee Moffitt Cancer Center and Research Institute; an NCI-designated Comprehensive Cancer Center grant (P30-CA076292; to B.J. Manley); OMF CA233944 (S.W. Lowe); the German Research Foundation (DFG) under Germany's Excellence Strategy (EXC 2180-390900677; J. Leibold); and a Shulamit Katzman Endowed Postdoctoral Research Fellowship (J. Leibold).
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.
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).