8q24 clear cell renal cell carcinoma germline variant is associated with VHL mutation status and clinical aggressiveness

Raw genotyping data from Affymetrix 6.0 array were obtained from germline DNA for 420 ccRCC patients, as well as corresponding clinical and pathological data. Somatic mutations of the four most frequently-mutated genes in ccRCC (BAP1, PBRM1, SETD2 and VHL) were obtained from TCGA [9]. The clear cell A and clear cell B (ccA/ccB) gene expression subtype classification has been reproducibly shown to be associated with outcome [11, 12] and was obtained from [9]. The Mayo SSIGN score has also been reproducibly shown to be associated with outcome [13‐16]. The Mayo SSIGN score is derived from an additive model that contains tumor stage, tumor size, tumor grade and presence of necrosis, and was calculated as described previously [17].

Quality control was performed on the genotying data for the 420 ccRCC TCGA patients, including 95% call rate (zero germline variants failed), 95% sample call rate (zero samples failed), Hardy–Weinberg equilibrium (821 variants failed with p < 0.000001), minor allele frequency (MAF; 216,942 variants had MAF < 0.05), sex check (100% concordance) and population stratification (all Caucasian). Genotyping data were phased and imputed using the Michigan Imputation Server with the Haplotype Reference Consortium (release 1) as the reference population. The imputation quality for each of the 14 variants is provided in Additional file 1: Table S1. An additive logistic regression model was used to assess the association between each of the 14 variants and subtype, with genotype coded as having 0, 1, or 2 copies of the minor allele for observed data and dosage was modeled as continuous for imputed data. A general linear model was used to assess the association between age of diagnosis and each of the 14 variants. To account for multiple testing, p values < 0.004 were considered statistically significant (0.05/14 = 0.004).

Associations between known RCC germline variants and genes within two-to-five megabase (Mb) were evaluated using Hi-C data via the HUGIn web browser [18]. Additional genetic information, limited to what was available in HUGIn, was included such as frequently interacting regions (FIREs), topologically associating domain (TAD) boundary regions, and occupancy of histone marks H3K27ac, H3K4me1 and H3K4me3. Because RCC is thought to originate from the renal tubular epithelium, and mesodermal stem cells form the tubule of the kidney, analyses were performed using mesendoderm cells. Mesenchymal stem cells were also analyzed [19]; results were similar across the two cell lines. We also used Hi-C data from ccRCC cell line (VHL mutant) Caki2 (GSM2827127 and GSM2827128) [20], (VHL wild-type) embryonic kidney cell lines HEK293T (GSM1081530 and GSM1081531) and HEK293T RAD21cv that was treated with tobacco etch mosaic virus protease (GSM1081526 and GSM1081527) [21]. RAD21 is a core subunit of cohesin complex, which is known to play a role in mediating chromosomal loops. In RAD21cv (a RAD21-EGFP variant) cells, RAD21cv replaced endogenous RAD21 and was incorporated into the cohesin complex. For Caki2 and HEK293 data, reads were mapped with Bowtie 2 [22] and alignments from two replicates were combined. Chromatin interaction was identified with HOMER (https://​homer.​ucsd.​edu/​homer/​interactions/​) at 20-kb resolution, which takes into account the dependence of interaction frequency and linear distance along each chromosome. For each risk locus, we combined the virtual 4C plot generated by the HUGIn web browser with the interaction plot generated from internally-analyzed Hi-C data.

Table 1 describes the 420 TCGA ccRCC patients that were analyzed. Of the 376 patients with available whole exome sequencing data, 150 (40%) had a VHL mutation, 34 (9%) BAP1 mutation, 119 (32%) PBRM1 mutation and 48 (13%) SETD2 mutation. We also subtyped patients according to disease aggressiveness using pathological indices defined by the Mayo SSIGN score [13] as well as molecularly according to the ccA/ccB gene expression subtype [11, 12]. Of the 355 patients that had available pathology data to calculate the Mayo SSIGN score, 81 (23%) were classified as aggressive (SSIGN score > 8). Using ccA/ccB to classify aggressiveness, of the 352 patients who had data, 168 (48%) were poor prognosis (ccB) subtype.

Using a case-case design, we evaluated the association of each of the 14 RCC germline variants with known ccRCC acquired alterations in BAP1, PBRM1, SETD2 and VHL (Table 2). We observed a statistically significant association after adjusting for multiple comparisons between the 8q24 variant rs35252396 and VHL mutation (OR = 1.60, p = 0.0037). While not significant after adjusting for multiple testing, we also observed a candidate association between EPAS1 variant rs7579899 and SETD2 mutation (OR = 1.87, p = 0.012) (Table 2).

We observed a statistically significant association between the 8q24 variant rs35252396 and Mayo SSIGN score (OR = 1.92, p = 0.00094) (Table 2). However, we did not observe a statistically-significant association between the known germline variants and ccA/ccB gene expression subtype.

We did not observe a statistically-significant association between the known germline variants and age at diagnosis (Additional file 1: Table S2).

We evaluated the interaction of each of the known germline variants with putative target genes in mesendoderm cell lines and mesenchymal stem cells using publically-available Hi-C data. Hi-C identifies chromatin interactions to evaluate the three dimensional chromatin structures inside the nucleus, which may identify long-range interactions. Some germline variants demonstrated interactions with nearby genes: e.g., rs4381241 with FAF1 (518.5 kb away), rs57579899 with EPAS1 (16.8 kb away), rs12105918 with ZEB2/ZEB2-AS1 (~ 70 kb away), rs1800057 with ATM (50.2 kb away) and rs4903064 with DPF3 (81.4 kb away) (Fig. 1; Additional file 1: Figure S1). However, some of these variants showed additional long-range interactions that have not been reported previously: e.g., rs4381241 with CDKN2C and TTC39A, rs7579899 with PRKCE, rs12105918 with ARHGAP15, GTDC1 and TEX41, rs1800057 with CUL5, ACAT1, NPAT and EXPH5 and rs4903064 with RGS6. Other germline variants also demonstrated long-range interactions that have not been reported previously: e.g., rs10936602 with SEC62 and PHC3, rs67311347 with ENTPD3, CTNNB1 and ULK4, rs2241261 with PEBP4 and EGR3, rs74911261 with EXPH5, rs718314 with SSPN and ITPR2 and rs4765623 with FAM101A (Additional file 1: Figure S1). Additionally, rs35252396 demonstrated interactions with PCAT1 and PCAT2 in mesenchymal stem cells but not in mesendoderm cells (Fig. 2). Similarly, rs11813268 had interactions with OBFC1 in mesenchymal stem cells but not in mesendoderm cells. Finally, we failed to identify genes within ± 1 Mb whose promoters interacted with rs7105934 in either mesenchymal stem cells or mesendoderm cells. All identified interactions were further evaluated in independent cell lines: VHL-mutant ccRCC cell line (Caki) and VHL wild-type embryonic kidney cell lines (HEK293 and HEK293 RAD21cv). A large proportion of the interactions were also identified in ccRCC or embryonic kidney cell lines. For example, rs10936602 interaction with SEC62, rs2241261 interactions with PEBP4 and EGR3, rs1800057 and rs74911261 interactions with EXPH5, as well as rs718314 interaction with SSPN were confirmed in all three cell lines. Further, rs12105918 interaction with GTDC1, rs11813268 interaction with OBFC1 and rs4765623 interaction with FAM101A were only identified in the ccRCC cell line, while rs7579899 interaction with PRKCE, rs35252396 interaction with PCAT2, and rs4903064 interaction with DPF3 were only identified in the embryonic kidney cell lines.