Somatic Mutations as Preoperative Predictors of Metastases in Patients with Localized Clear Cell Renal Cell Carcinoma – an Exploratory Analysis

Abstract

Objective

Recurrent genomic alterations in clear cell renal cell carcinoma (ccRCC) have been associated with treatment outcomes; however, current pre-operative predictive models do not include known genetic predictors.

We aimed to explore the value of common somatic mutations in the preoperative prediction of metastatic disease among patients treated for localized ccRCC.

Materials and Methods

After obtaining institutional review board approval, data of 254 patients with localized ccRCC treated between 2005–2015 who underwent genetic sequencing was collected. The mutation status of VHL, PBRM1, SETD2, BAP1 and KDM5C were evaluated in the nephrectomy tumor specimen, which served as a proxy for biopsy mutation status. The Raj et. al. preoperative nomogram was used to predict the 12-year metastatic free probability (MFP).

The study outcome was MFP; the relationship between MFP and mutation status was evaluated with Cox-regression models adjusting for the preoperative nomogram variables (age, gender, incidental presentation, lymphadenopathy, necrosis, and size).

Results

The study cohort included 188 males (74%) and 66 females (26%) with a median age of 58 years. VHL mutations were present in 152/254 patients (60%), PBRM1 in 91/254 (36%), SETD2 in 32/254 (13%), BAP1 in 19/254 (8%), and KDM5C in 19/254 (8%).

Median follow-up for survivors was 8.1 years. Estimated 12-year MFP was 70% (95%CI: 63–75%). On univariable analysis SETD2 (HR: 3.30), BAP1 (HR: 2.44) and PBRM1 (HR: 1.78) were significantly associated with a higher risk of metastases. After adjusting for known preoperative predictors in the existing nomogram, SETD2 mutations remained associated with a higher rate of metastases after nephrectomy (HR:2.09, 95% CI:1.19–3.67, p=0.011).

Conclusion

In the current exploratory analysis, SETD2 mutations were significant predictors of MFP among patients treated for localized ccRCC. Our findings support future studies evaluating genetic alterations in pre-operative renal biopsy samples as potential predictors of treatment outcome.

Introduction

Traditionally, the use of percutaneous renal tumor biopsy was limited due to concerns about its diagnostic yield and minimal impact on clinical management [1]. Recent publications support the high diagnostic accuracy, sensitivity and specificity of renal tumor biopsy [1–5]. Subsequently, international guidelines suggest an increasing role for biopsy especially prior to ablative therapy and active surveillance [6]. Despite these recommendations, overall utilization rates are still low [7–9]. Furthermore, while molecular information obtained from renal mass biopsies may improve patient risk stratification, only one study used the biopsy cell cycle proliferation score to predict adverse surgical pathology [10].

Over the last decade, there has been a growing understanding of the molecular basis of renal cell carcinoma (RCC) in general, and especially clear cell renal cell carcinoma (ccRCC) [11]. Recurrent genomic alterations in ccRCC, have been associated with pathological and clinical outcomes following treatment [12–18]. Incorporation of molecular profiling may therefore help identify higher-risk tumors and improve management strategies [19, 20]. However, few studies evaluated the predictive capability of recurrent genomic alterations when accounting for known clinical and pathological predictors, especially in the pre-operative setting, reporting controversial results [18, 20–22].

In the current study we used molecular information obtained from nephrectomy specimens as a surrogate for molecular findings on preoperative renal biopsy to explore possible associations between genetic alterations in localized ccRCC and metastatic free probability (MFP) after nephrectomy when controlling for preoperative clinical and radiologic characteristics.

Materials and Methods

After obtaining Institutional Review Board approval, we reviewed a cohort of 3,599 patients who underwent a partial or radical nephrectomy between the years 2005 and 2015 for a clinically localized renal mass, without evidence of distant metastases on preoperative imaging. Clear cell RCC histology was identified in 2,010 patients. In the study cohort we included 254 patients who had a ccRCC tumor and underwent genetic sequencing of their primary tumor (Figure 1).

Figure 1 –

Flow-chart of patients selected for the study cohort

Patient demographic and clinical characteristics including age, gender, race, and mode of presentation (incidental, local, or systemic) were collected. Preoperative cross-sectional imaging studies of all patients were reviewed by dedicated radiologists (CD, OA). Consistent with the variables included in the Raj et al. preoperative nomogram, images were reviewed for the presence of lymphadenopathy (yes/no), necrosis (yes/no), and tumor size (cm). Patients were treated with either a radical or partial nephrectomy and their tumor specimen was reviewed by a dedicated genitourinary pathologist who confirmed the diagnosis of ccRCC.

Primary tumors from nephrectomy specimens were used to obtain genetic information. Molecular characterization of 97/254 tumors (38%) was performed using Sanger sequencing, 90/254 (35%) underwent MSK-IMPACT testing (MSKCC Integrated Mutation Profiling of Actionable Cancer Targets), a hybridization capture-based next-generation sequencing assay for targeted deep sequencing of all exons and selected introns of key cancer genes [23], and 67/254 (26%) underwent whole exome sequencing as part of the previously reported TCGA cohort (Figure 1) [11]. A panel of 5 genes (VHL, PBRM1, SETD2, BAP1 and KDM5C) was chosen for analysis based on their high mutational frequency and clinical significance in previous publications [11, 13–16, 22]. Mutation frequencies were calculated and visualized with a heatmap. Preoperative clinical characteristics were compared based on mutation status using the Fisher’s exact test and Wilcoxon Rank Sum test, where appropriate. Hypothesis tests were adjusted for multiple comparisons within each gene (adj-p).

Follow-up after nephrectomy consisted of serial imaging. The presence and location of metastases on imaging were noted. Metastatic free probability (MFP) was calculated from the time of nephrectomy until evidence of metastases. Patients who died or were metastases free at last follow-up were censored, as per the methodology used in the Raj et al. nomogram. MFP was estimated with the Kaplan-Meier method.

To evaluate the pre-operative risk for developing metastases, we used a preoperative nomogram reported by Raj et. al. for predicting the 12-year MFP of patients with localized renal tumors [24]. The nomogram includes age, gender, mode of presentation, evidence of lymphadenopathy, evidence of necrosis and tumor size based on preoperative imaging. The primary endpoint of the current study was 12-year MFP consistent with the endpoint used in the original nomogram publication. A linear predictor was calculated from the nomogram to assess each individual patients’ risk. Univariable Cox regression was used to assess the relationship between gene mutation status and the nomogram linear predictor and MFP. Genes significant at p<0.05 were included in a multivariable model with the nomogram linear predictor. Only significant genetic mutations were included in the multivariable analysis due to the limited sample size and number of events in the genomic cohort. Two-sided p-values less than 0.05 were considered statistically significant. We used SAS 9.4 (The SAS Institute, Cary, NC) for all statistical analyses.

Results

The study cohort included 188 male (74%) and 66 female (26%) patients with a median age of 58 years (IQR 51 – 67). Most patients presented incidentally (80%, n=204). On pre-operative imaging, median tumor size was 5.3cm (IQR, 3.6–8.1), most patients (78%, n=198) had evidence of tumor necrosis and 9% of patients (n=24) had enlarged lymph nodes (Table 1). Among the 24 patients with enlarged lymph nodes on preoperative imaging, 16 (67%) underwent lymphadenectomy, 6 of whom had pathologic lymph node involvement.

Table 1.

Clinical characteristics of patients with clear cell renal cell carcinoma and genetic information (n=254); numbers represent frequency with percent of total in parentheses unless otherwise specified

		N (%)
Age at surgery, years	Median (IQR)	58 (51–67)
Gender	Male	188 (74)
	Female	66 (26)
Race	White	235 (92.5)
	Black	8 (3.1)
	Asian	5 (2)
	Other	4 (1.6)
	Unknown	2 (0.8)
Presentation	Incidental	204 (80.3)
	Local	46 (18.1)
	Systemic	4 (1.6)
Lymphadenopathy	Yes	24 (9.4)
	No	230 (90.6)
Necrosis	Yes	198 (78)
	No	56 (22)
Maximum Dimension, cm	Median (IQR)	5.3 (3.6–8.1)
Fuhrman Grade	1 – 2	97 (38)
	3 – 4	153 (60)
	Unknown	4 (2)
T-Stage	T1	112 (44)
	T2	18 (7)
	T3	120 (47)
	T4	4 (2)
N-Stage	N0/ Nx	248 (98)
	N+	6 (2)
Surgical Margins	Negative	236 (93)
	Positive	18 (7)

IQR = interquartile range

VHL mutation was present in 152/254 patients (60%), PBRM1 in 91/254 (36%), SETD2 in 32/254 (13%), BAP1 in 19/254 (8%), and KDM5C in 19/254 (8%), (Figure 2). We found an association between PBRM1 mutations and a higher rate of necrosis (87% vs. 73%, adj-p=0.035) and larger tumor size (median 5.8cm vs. 4.6cm, adj-p=0.035). SETD2 mutations were associated with a higher rate of local (31% vs. 16%) and systemic (9 vs. 1%) presentation (adj-p=0.004). BAP1 mutations were associated with a higher rate of local (26% vs. 17%) and systemic (11% vs. 1%) presentation (adj-p=0.022), lymphadenopathy (37% vs. 7%, adj-p=0.004) and larger tumor size (median 7.7cm vs. 5.1cm, adj-p=0.022). We found no association between VHL and KDM5C mutation status and clinical characteristics. Among the 24 patients with enlarged lymph nodes on preoperative imaging VHL mutations were found in 13/24 patients (54%), PBRM1, SETD2 and BAP1 in 7/24 patients (29%) each, and KDM5C in none of the patients.

Figure 2 –

Gene heatmap with mutation frequency among the cohort of clear cell renal cell carcinoma patients who underwent genetic sequencing of their primary renal tumor (n=254)

Median follow-up for survivors was 8.1 years (IQR 5.0–10.1). By the end of follow up, 71 patients developed metastases; median MFP was not reached. The estimated 12-year MFP of the cohort was 70% (95% CI of 63% - 75%). Five-year metastases-free survival was 71% (95% CI 62%−77%) for patient with VHL mutations, 68% (57% - 77%) for PBRM1, 46% (28% - 62%) for SETD2, 46% (21% - 68%) for BAP1 and 63% (38% - 80%) for KDM5C.

On univariable analyses SETD2 (HR:3.30, 95% CI:1.94–5.59, p<0.001), BAP1 (HR:2.44, 95% CI:1.21–4.93, p=0.013) and PBRM1 (HR:1.78, 95% CI:1.11–2.83, p=0.016) were associated with a higher risk of metastases compared to wild-type patients (Table 2). No significant association was found between MFP and VHL (HR: 1.41, p=0.18) or KDM5C (HR: 1.58, p=0.22) mutations. Additionally, the nomogram linear predictor, which included age, gender, mode of presentation, evidence of lymphadenopathy, evidence of necrosis and tumor size based on preoperative imaging, was associated with metastatic free probability on univariate analyses (HR:2.62, 95% CI:2.1–3.27, p<0.001).

Table 2.

Univariable and multivariable Cox models for predictors of metastasis-free probability (N=254)

		Univariable				Multivariable
		N(E)	HR	[95% CI]	p-value	N(E)	HR	[95% CI]	p-value
VHL	Yes	152 (48)	1.41	[0.86 2.32] -	0.18		---
	No	102 (23)	REF				---
PBRM1	Yes	91 (34)	1.78	[1.112.83] -	0.016		1.41	[0.85 2.35] -	0.18
	No	163 (37)	REF				---
SETD2	Yes	32 (19)	3.30	[1.94 5.59] -	<.001		2.09	[1.19 3.67] -	0.011
	No	222 (52)	REF				---
BAP1	Yes	19 (9)	2.44	[1.21 4.93] -	0.013		0.83	[0.37 1.87] -	0.65
	No	235 (62)	REF				---
KDM5C	Yes	19 (8)	1.58	[0.76 3.31] -	0.22		---
	No	235 (63)	REF				---
Nomogram Linear Predictor*		254 (71)	2.62	[2.10 3.27] -	<.001		2.58	[2.01 3.30] -	<.001

^*The nomogram linear predictor includes the following factors: age, gender, mode of presentation, evidence of lymphadenopathy, evidence of necrosis and tumor size based on preoperative imaging.

The following equation from Raj et al, was used to calculate the value of the nomogram linear predictor: −3.1830084 − 0.00065242845*age + 0.10166342*gender + 0.56585476*presentation + 1.0072686*lymphadenopathy + 0.26592168*necrosis + 0.65408506*size - 0.0086883408*max(size-2, 0)**3 + 0.013366678*max(size-4.8, 0)**3-0.0046783373*max(size-10, 0)**3. Size was treated as a cubic spline.

N = Total # patients for level; E = # events for level; HR = hazard ratio; 95%CI = 95% confidence interval

In the multivariable model including the nomogram linear predictor, SETD2 remained a significant predictor of MFP (HR:2.09, 95% CI:1.19–3.67, p=0.011). However, PBRM1 (p=0.18) and BAP1 (p=0.65) were not significantly associated with MFP in the multivariable setting (Table 2). Adding genetic information to the linear predictor increased the AUC of the ROC curves at all time points (Figure 3). The AUC for the combined model was 0.83 (95% CI 0.78–0.88) at 3 years, 0.85 (95% CI 0.81–0.90) at 6 years. The AUC dropped to 0.68 (95% CI 0.51–0.85) at 12 years, possibly due to the fewer amount of patients at risk at this time point.

Figure 3 –

Time-dependent ROC curves for metastasis-free probability for patients with clear cell renal tumors and genetic data (n=254) when accounting for (A) the linear predictor alone, and (B) the linear predictor + genetic information. The diagonal line in each of the plots corresponds to random classification. The corresponding AUCs are listed under each plot

Discussion

In the current study we used genetic data obtained from nephrectomy specimens as a proxy for genetic information on renal mass biopsy to explore the possible predictive capability of recurrent genomic alterations within the pre-operative setting. Consistent with previous reports, we found an association between SETD2, BAP1 and PBRM1 and MFP on univariable analysis. Furthermore, SETD2 remained an independent predictor of outcome when adjusting for the metastatic probability based on a validated pre-operative nomogram.

Recent reports support the accuracy and utility of renal mass biopsies [1–5]. In a systemic review and meta-analysis, the median rate of diagnostic biopsies was 92% and the estimated sensitivity and specificity of diagnostic biopsies were 99.1% and 99.7%, respectively. The median overall complication rate across these studies was 8%, the majority of which were low grade complications [2]. Similarly, in series evaluating the utility of renal biopsy in small renal masses, the biopsy was diagnostic in 87%−90% of cases, and biopsy histology was concordant with surgical pathology in 86%−93% of cases [4, 5]. In light of these results, there is an increase in the use of renal mass biopsy, especially in patients who undergo non-surgical management, however, overall utilization rates remain low [7–9]. Unlike tumor histology, the median concordance rate between grading on biopsy and surgical specimen was 63% when using the four-tier Fuhrman grading system [2], therefore we did not include grade in our analysis of preoperative predictors.

Genetic sequencing of biopsy samples and identification of recurrent genetic alterations may improve risk stratification, increasing the utility of renal mass biopsy in treatment decision making [19, 20]. In a recent publication, Tosoian et al. reported that the cell cycle proliferation score, a gene expression classifier, obtained from preoperative renal biopsy was a significant predictor of adverse pathology at nephrectomy when added to a baseline model including age, sex, race, tumor size, biopsy grade and histology. However, data regarding the association between the cell cycle proliferation score and oncologic outcome was not available [10].

Multiple publications have shown recurrent genomic alterations, which may be identified on renal biopsy, are linked to pathological and clinical outcome in RCC [12–18]. Patients with clear cell RCC and a PBRM1 mutation were more likely to present at an advanced stage and those with a BAP1 mutation were more likely to have a higher Fuhrman grade [13, 17]. Furthermore, BAP1 and SETD2 mutations were associated with a worse cancer specific and overall survival [12–16, 18]. Mutations in KDM5C were associated with higher stage disease at presentation [13]. Among patients with small renal masses, KDM5C mutations were associated with inferior survival from either recurrence or death from disease [25]. Consistent with previous reports, among the five genes evaluated in the current cohort, we found univariable associations between BAP1, SETD2 and PBRM1 and both adverse preoperative clinical factors and a higher risk for metastases. We observed an association between KDM5C mutations and a higher rate of metastatic disease consistent with previous publications; however, this association did not reach statistical significance possibly due to the small number of patients with this mutation in our cohort (n=19).

Few studies evaluated the added benefit of genomic alterations to commonly used prognostic models for renal cell carcinoma [15, 18, 21, 22]. An initial report investigating the role of select genomic markers in a preoperative model, including tumor size and age, and the postoperative Mayo Clinic stage, size, grade, and necrosis (SSIGN) prognostic scoring system did not show a substantial increase in the predictive accuracy of these models when adding genetic information [21]. In a subsequent analysis on a larger cohort of patients, TP53 mutations were significantly associated with decreased cancer specific survival and SETD2 mutations showed a significant association with an increased risk of recurrence when controlling for the post-operative SSIGN model [22]. Similarly, BAP1 mutation, when assessed with immunohistochemical staining of tumor samples, was an independent predictor of poor prognosis both after adjusting for the UCLA integrated staging system, and among patients with a low-risk of adverse outcomes based on the Mayo clinic SSIGN score [15]. In an additional cohort, patients with SETD2 mutations were twice as likely to experience RCC-specific death than patients with tumor expressing the SETD2 gene. This finding was also true within the low-risk SSIGN group [18]. In patients with advanced or metastatic renal-cell carcinoma treated with first-line tyrosine kinase inhibitors, incorporating the mutation status of BAP1, PBRM1 and TP53 improved the predictive ability of the original MSKCC risk model for predicting overall survival and progression free survival [26]. Due to the high concordance between renal biopsy findings and surgical pathology in recent series, we used surgical pathology as a surrogate for the biopsy pathology in the current study. Within this proxy, SETD2 remained an independent predictor of MFP when controlling for the nomogram linear predictor. The addition of SETD2 status increased the AUC 2–3% beyond the existing preoperative nomogram, which was not optimized to include genetic information, and patients with SETD2 mutation had twice the hazard of metastases. Our findings suggest mutation status should be incorporated when developing future preoperative nomograms, with the aim of improving their predictive capabilities.

Intratumoral heterogeneity may pose an obstacle for the use of genetic information obtained from biopsies, as a single sample may not encompass the full genetic landscape of a tumor [27]. Furthermore, multi-regional biopsies are required to define tumor clonality, and identify the evolutionary subtype of the tumor which may also have predictive importance [28]. Similar to most studies evaluating genetic predictors, we used sequencing data from a single tumor region which underestimates intratumorally heterogeneity [27, 29]. In an attempt to evaluate the number of cores required to identify mutations within a tumor, Sankin et al reported that using 3 separate biopsy cores from different regions of the same tumor enables detection of mutations in PBRM1, SETD2, BAP1 and KDM5C in 90% of cases. According to the TracerX study, patients with small renal masses tend to have lower Intratumoral heterogeneity and in these cases obtaining two biopsy cores enables identification of nearly all subclonal driver events [28]. Thus, renal biopsies may be sufficient for characterizing small renal masses, but may prove more problematic for larger tumors [28]. Tosoian et al. compared biopsy- and nephrectomy-derived cell cycle proliferation scores, showing that both were well correlated (Pearson correlation: 0.55) with a similar predicted probability of adverse pathology, suggesting gene expression levels in biopsy samples represent those of the whole tumor [10]. We and other are currently working on ways to incorporate mutational clonality and evolutionary subtypes into single region analysis to improve predictive accuracy. In addition, further efforts to link genomic alterations to radiomic features may improve biopsy targeting of “genomically important” regions of a larger renal mass [30].

The study limitations include selection bias apparent by the relatively large tumors included within our cohort (median tumor size 5.3cm) and small proportion of the overall cohort with genomic information, thus limiting the generalizability of our findings. Only 1/82 patients treated for stage T1a disease had evidence of metastases during followup, precluding us from performing outcome analyses in this subgroup of patients. Considering these patient characteristics, our findings may be useful for identifying high risk patients suitable for inclusion in neoadjuvant trials for localized ccRCC rather than patients suited for active surveillance. We relied on pathologic and genetic information obtained at nephrectomy rather than renal biopsy; therefore, tumor heterogeneity may have affected our findings, especially when accounting for the large size of tumors included in our cohort. Additional studies performed on renal biopsies are required. Furthermore, the use of immunohistochemical staining to define mutation status should be evaluated, as they are more applicable for widespread use. Nevertheless, our initial findings are among the few to suggest an association between genetic data and outcome when controlling for known preoperative predictors of outcome, supporting the need for future studies to validate the role of histology and genetic data obtained from renal biopsies for the preoperative prediction of treatment outcome in patients with localized ccRCC.

Conclusions

Our exploratory analysis demonstrates that in patients with localized ccRCC, SETD2 status is an independent predictor of metastatic free probability when accounting for preoperative risk factors. If validated, molecular analysis of tumor biopsy specimens may play a role in treatment decision-making. Future studies using tumor biopsy samples are required to validate our initial findings within the pre-operative setting.

Highlights.

• SETD2, BAP1 and PBRM1 mutations are associated with metastases in clear cell renal tumors

• We used molecular data from nephrectomy specimens as a proxy for biopsy genetic data

• SETD2 is associated with metastases after adjusting for a validated preoperative nomogram

• Pending confirmatory studies, genetic alterations in renal biopsies may improve outcome prediction