VHL mutations predict survival of patients with renal cell carcinoma in response to the therapy of immune checkpoint inhibitors


 Purpose: To identify a predictive biomarker of immune checkpoint inhibitor (ICI) therapies for renal cell carcinoma (RCC). Methods: Survival analysis of mutations in a panel of 468 cancer-related genes in a cohort of 151 RCC patients who underwent the ICI treatment using Cox regression model univariate or multivariate analysis in publicly available datasets. Results: We found that VHL mutations were the only promising independent predictor for overall survival (OS) (HR=0.44, 95% CI=0.25-0.77 and P=0.004). More specifically, compared with 26 months of survival in the wildtype patients, metastatic RCC patients carrying truncated VHL mutations had a significantly longer survival time of nearly 70 months (HR=0.45, 95% CI=0.25-0.82 and P=0.008) in the presence of the ICI therapy. This survival benefit was also observed in another cohort of 35 patients with clear cell RCC from Dana-Farber Cancer Institute (DFCI): compared with 29 months in the wildtype patients, patients with VHL truncated mutations also had a longer median OS of 33 months (HR=0.59, 95% CI=0.24-1.44 and P=0.243). These observed survival benefits were independent of VHL expression and tumor infiltration immune cells in The Cancer Genome Atlas Kidney Renal Clear Cell Carcinoma (TCGA KIRC). Conclusion: VHL mutational inactivation may have an effect on the RCC response to ICI therapy, likely by the upregulation of PD-L1 via attenuating degradation of HIF-1α. To our knowledge, this is the first report of VHL mutations as an independent predictive biomarker for the ICI therapy in RCC, which, once validated by larger clinical trials, may help improve clinical decision-making in individualized treatment of RCC patients.


Introduction
RCC is one of the most common malignancies in the urinary system. Over 400,000 new patients suffer from RCC along with over 175,000 cancer-related deaths worldwide in 2018 [1] . Due to intrinsic resistance to the chemo-radiotherapy, the mainstay therapy for metastatic RCC (mRCC) has been limited to traditional therapies, such as cytokines or VEGF/mTOR targeting therapies. While these therapies have improved outcomes compared to the chemo-radiotherapy, they are still limited by toxicities or drug resistance that usually occur within the rst year after the treatment. As a result, the 5-year over survival of mRCC has remained < 8% for the past decades [2][3][4] . The immune checkpoint inhibitor (ICI) treatment, including anti-bodies targeting programmed cell death protein 1 (PD-1), programmed death-ligand 1 (PD-L1) or cytotoxic T-lymphocyte-associated protein 4 (CTLA4), provide survival bene ts only to a minority of mRCC with a durable anti-tumor effect; there is no obvious improvement of progression-free survival (PFS) in unselected mRCC patients, compared to the mTOR inhibitor therapy [3,5] , making it critical to identify sensitive mRCC patients who can bene t from the ICI treatments.
Recently, PD-L1 expression levels, defect mismatch repair (dMMR) or microsatellite instability high (MSI-H), tumor mutation burden (TMB) have positively demonstrated some predictive effects of the ICI treatment on various types of cancer [6][7][8] . However, to the best of our knowledge, there are few reliable predictive biomarkers for the indications of ICI treatment of mRCC, including TMB that has been proven to be incapable of providing potential survival bene ts [8] . As for PD-L1, which is expressed in less than 1% of the tumors in 75% of RCC patients, its role in predicting the treatment response is obviously trivial [9] . Hence, it is pivotal to identify alternative predictive biomarkers for the ICI treatment of mRCC. To this end, we analyzed the existing mutation pro les in tumors for the prediction of the ICI treatment outcomes, because mutations in several genes have been demonstrated to predict outcomes of cancer patients effectively [10] . Hence, in the present study, by utilizing the publicly available mutation data on various cancers, we aimed to nd mutations that may serve as potential predictors for overall survival (OS) in RCC patients.

Data information
We downloaded all the clinical information, mutations and RNA sequencing data of three independent datasets from the publicly available cBioPortal database (https://www.cbioportal.org), which included the

Initial analysis
In the initial analysis of ICI-related gene mutation data, we used the MSK-IMPACT dataset, considering all somatic mutation data that included missense, frame-shift, nonsense, nonstop, splice site, and translation start site changes, except for the synonymous mutations of the target panel containing 468 cancer-related genes [11] . These data were derived from the MSK-IMPACT assay that was applied to all the 151 mRCC patients who received ICI therapies. We rst divided the 151 mRCC patients into three groups according to the 468 genes panel tested: patients having non-mutated genes, patients with certain genes (e.g., VHL) with mutations, and patients with other genes with mutations, and then we performed univariate Cox proportional hazards regression analysis for these three groups to evaluate the association between gene mutation status and OS of mRCC patients. We also generated Kaplan-Meier survival curves to visualize the results. Those genes with a P value <0.05 for a mutation rate >0.05 were considered the candidates for the multivariate Cox regression analysis with adjustment for age, sex, TMB, and ICI regimes. Subsequently, we selected the patients with those genes that had truncated mutations (frameshift insertion, frameshift deletion, nonsense, splice-site) to be compared with those patients with the wildtype genes, and we then performed log-rank test to compare the differences in OS between these two groups of patients.
To replicate the ndings of those ICI-related genes present in mRCC patients included in the abovementioned patient cohorts, we used another DFCI cohort that had 35 metastatic ccRCC patients who had the similar genes detected for mutations and also received ICI therapies. To compare the survival between the patients carrying the genes with truncated mutations and the patients with wildtype genes, we performed the log-rank test to compare the differences in OS between two groups.

Further exploration
To ascertain whether those mutational genes present in the above-mentioned MSK-IMPACT cohort may affect the survival of RCC patients without ICI therapies, we analyzed the TCGA KIRC dataset from 451 ccRCC patients who had somatic mutation data but only received traditional treatments, including surgery, chemo-radiotherapy, cytokines and target therapy. Then, we compared OS among patients carrying all genes with mutations, patients carrying genes with truncated mutations, and patients carrying wildtype genes.
We further evaluated the effect of VHL expression on OS in ccRCC patients by using the TCGA-KIRC dataset by using the median VHL mRNA expression value obtained from RNA sequencing data as the cutoff value to divide the 534 ccRCC patients into two groups of low and high VHL mRNA expression, and we used the log-rank tests to compare the differences in OS between the two groups.
We also compared the differences in tumor-in ltrated immune cells between the groups of the patients carrying VHL mutational genes and patients carrying VHL wildtype genes in the 451 patients from the TCGA KIRC dataset by calculating the gene expression matrix using CIBERSORT online tools (https://cibersort.stanford.edu/) [12] .
All the statistical analyses were performed by using R language (version 3.5.1), and P values were twosided with a signi cance level of 0.05.

Results
The analysis owchart is shown in Figure 1. In the initial analysis, we used a cohort of 151 metastatic or unresectable RCC patients who underwent the ICI therapy (122 received anti PD-1/PD-L1 monotherapy, and the remaining 29 received a combination of anti CTLA-4 and anti PD-1/PD-L1 therapies). Clinical characteristics of the patents are presented in Table 1. The univariate Cox regression analyses of the three groups (i.e., patients with non-mutated genes, patients with certain genes with mutations, and patients with other genes with mutations) showed that mutations in 21 genes were signi cantly associated with OS, but the patients with the highest VHL mutation rate (70%) had the largest HR of 1.91, compared with the patients with mutations in other 20 genes (2%) (Supplement Table 1). Therefore, subsequent analyses focused on the VHL mutations. In Figure 2, we present survival curves of patients without mutations in any genes, patients with VHL mutations, and patients with mutations in genes other than VHL. In the following analysis, we merged the group of patients without mutations or with the VHL wildtype and the group of patients with mutations in genes other than VHL as one group to be compared with those with VHL mutations. In the univariate and multivariate Cox regression analyses for the association between the VHL mutation status and OS, we found that the VHL mutation status (70.2%, 106/151) was an independent predictor for OS in this patient cohort (HR = 0.52, 95% CI = 0.36-0.77, P = 0.001 and HR = 0.44, 95% CI = 0.25-0.77, P = 0.004 in univariate and multivariate analyses, respectively) ( Table 2). Speci cally, RCC patients with VHL mutations showed a signi cantly longer OS of 50 months, compared with 26 months in patients with the wildtype VHL or mutations in other genes ( Figure 3A). The detailed clinical information on the group of patients with VHL mutations and the other group is presented in Table 3. It appeared that VHL mutations occurred more often in ccRCC (102/121, 84.3%) than in non-ccRCC (n=4/30, 13.3%) (P < 0.001), and the VHL mutation group had a relatively higher mean TMB (4.70±2.46) than that (3.20±3.54) of the VHL wildtype group (P = 0.003). We then focused on truncated mutations, because they were considered functionally important mutations [13] . We found that RCC patients (n = 65) with VHL truncated mutations showed a signi cantly longer median OS of nearly 70 months, compared with 26 months in patients (n = 45) with the wildtype VHL (HR = 0.45, 95% CI = 0.25-0.82 and P = 0.008) ( Figure 3B).  **The results were in bold, if the P value was less than 0.05.
To further ascertain whether VHL mutations affected OS in RCC treated with traditional therapies, including surgery, chemo-radiotherapy, cytokines and target therapy, another 451 ccRCC patients from the TCGA KIRC dataset were divided into two groups by VHL mutation status with detailed information presented in Supplemental Table 2. The VHL mutation group (n= 225) showed no survival bene t, compared with the wildtype group (n = 226) ( Figure 3C). In addition, the VHL truncated mutation group (n = 126) also showed no signi cant survival bene t, compared with the wildtype group in this dataset ( Figure 3D). Furthermore, there was no association between the tumor in ltrated immune cells and the VHL mutation status in RCC in the TCGA KIRC dataset (Supplemental Table 3).
Additional analysis with the data from the TCGA KIRC dataset suggested that the VHL expression levels did not affect OS in ccRCC patients (n = 534) who did not receive any immunotherapies ( Figure 3E), of whom 267 patients had lower VHL expression, while other 267 patients had higher VHL expression.
To replicate the ndings in the MSK-IMPACT dataset that the VHL truncated mutation group had a longer median OS than the VHL wildtype group, who all received ICI therapies, we used the data from another DFCI cohort of 35 metastatic ccRCC patients, of whom those (n=16) with VHL truncated mutations also showed a longer median OS (33 months), compared with those (n=10) with the wildtype VHL (29 months), although it did not reach the statistical signi cance (HR = 0.59, 95% CI = 0.24-1.44 and P = 0.243) ( Figure 3F), likely due to a much smaller sample size (n = 35) of this small dataset.

Discussion
VHL, as a tumor suppressor, is mutated in over 50% ccRCC that constitutes the majority of RCC. In the present study, the ccRCC accounted for 80.1% RCC in the MSK-IMPACT patient cohort. The VHL gene was mutated in 84.3% and 51% of the ccRCC patients in the MSK-IMPACT cohort and the TCGA KIRC dataset, respectively. Although VHL mutations were signi cantly associated with TMB in the dataset we used, the TMB was shown to have a little value in predicating the response to ICI treatment [8] . Some studies suggested that the impact of immune cell in ltration on the ICI treatment was not negligible, despite this notion was not well established yet [14][15][16] . However, we observed in the present study that the immune cell in ltration rate was not signi cantly different between tumors with mutational and wildtype VHL.
Truncated mutations have been considered the important functional mutations, usually leading to the loss of gene functions [13,17] . It has been reported that inactivating mutations of VHL occurred in RCC may attenuate the degradation of HIF, thus leading to the hypoxia signal activation, which stimulates angiogenesis, cell growth and survival of renal cancer cells [18][19][20] . In addition, a previous study reported that the hypoxia induced the PD-L1 expression via upregulating HIF-1α expression in tumor-in ltrating myeloid cells [21] , while another study identi ed that overexpression of PD-L1 and PD-1 was induced in both human in vitro and murine models of hypoxia as well as by HIF-1α transfection [22] . Because mutational inactivation of VHL may attenuate the degradation of HIF, it is reasonable to speculate that the mutational inactivation of VHL could lead to the accumulation of the HIF-1α, thus leading to the PD-L1 up-expression in renal cancer cells, which then makes the cancer cells potentially targetable by the anti-PD-1/PD-L1 therapy. However, more experiments are needed to validate such a hypothesis.
In summary, we found in the present study that VHL mutations, the truncated mutations in particular as an independent predictor for OS, effectively predicted survival of mRCC patients in response to the ICI treatment. Considering other relevant ndings from previous studies, we believe that the survival bene t associated with VHL mutations could be due to the upregulation of PD-L1 by VHL mutational inactivation via HIF-1α, a possible mechanism that may explain the e cacy of immunotherapies in treating RCC. However, because studies with a limited sample size were used in the present study, additional lager studies or clinical trials as well as functional investigations are needed to con rm these results. The status of VHL mutations affect the OS of RCC patients.

Figure 3
The association of VHL mutations and expression with OS of RCC patients under ICI therapies or traditional therapies.
A: The patients (n=106) with VHL mutations showed a signi cantly longer OS than the wildtype population (n=45) in MSK-IMPACT RCC cohort underwent ICI treatment.