Identification of a Novel Pathogenic Germline KDR Variant in Melanoma

Recently identified melanoma driver mutations have paved the way for rational development of effective targeted inhibitors that have increased survival rates for the metastatic patients whose tumors carry these specific mutations (1–4). Inhibition of the MAPK pathway in BRAF-mutated melanoma patients with BRAF and/or MEK inhibitors produces high response rates (1, 2, 5). However, approximately 30% of melanoma tumors lack activating mutations in BRAF or other somatic mutation drivers NRAS and KIT (6–8), and are designated as wild-type (WT) melanoma. While response rates to combination immune checkpoint blockade in BRAF WT tumors have been reported as high as 61% (9), few effective treatment options are available for WT patients who do not respond to immunotherapy and/or experience unacceptable toxicity, eventually leading to the necessity to stop treatment. As next-generation sequencing (NGS) technologies become increasingly economical (10, 11), low-frequency somatic driver mutations are being identified in multiple tumor types (12–14), facilitating new clinical trial schemes, such as the “basket trial,” in which patients are accrued for therapy according to a specific, targetable genetic alteration independent of tumor histotype (15).

When tumor–normal pairs are compared as a control to identify tumor-specific somatic mutations in NGS, nonsynonymous germline variants can be identified through tumor–normal subtractive analysis (16, 17). Recently, The Cancer Genome Atlas published a multilayer profiling of 333 cutaneous melanomas, revealing a catalog of potentially actionable somatic alterations. However, common germline variants were excluded based on their presence in the “normal” sample (18). In melanoma, germline variants have been described in high- (CDKN2A and CDK4), moderate- (MC1R), and low-penetrance susceptibility genes (19). There are around 50 low-penetrance melanoma susceptibility genes that are relevant to pigmentation/nevus count, the antitumor immune response, DNA-repair, metabolism, and vitamin D receptor polymorphisms. Data from a recent melanoma genome-wide association study (20, 21) and evidence from our group (22, 23) demonstrate their impact on tumor progression and clinical outcome. However, to our knowledge, no therapy targeting germline variants exists in melanoma.

In this study, we sought to examine the clinical utility of a targeted NGS approach to identify actionable somatic and germline variants with immediate relevance to the treatment of melanoma. We analyzed tumor and germline DNA from melanoma patients with a cancer-associated gene panel covering 2,800 COSMIC mutations within 50 known oncogenes and tumor suppressor genes. We identified potentially actionable somatic mutations in melanoma in WT tumors and characterized for the first time a pathogenic kinase insert domain receptor (KDR) germline variant (Q472H) in melanoma. Our data provide support for the concept of integrating germline DNA analyses to improve the personalized treatment of cancer patients.

We studied a pilot cohort of 34 stage III–IV and an independent cohort of 1,223 stage I-IV melanoma patients. All patients were enrolled in the New York University Interdisciplinary Melanoma Cooperative Group biorepository database from May 2002 to May 2012 and were prospectively followed according to an NYU Institutional Review Board–approved protocol and in accordance with the Federal Wide Assurance for the Protection of Human Subjects approved by the Department of Health and Human Services. Clinical variables assessed at the time of enrollment include age, gender, stage, histologic subtype, anatomic site, and treatment. Response to treatment, melanoma status, and survival information were prospectively obtained via active follow-up every 6 months.

Tumor genomic DNA was extracted from macrodissected, formalin-fixed paraffin-embedded (FFPE) melanoma tissue samples from the pilot cohort with the Qiagen QIAamp DNA FFPE Tissue Kit.

Germline genomic DNA was extracted from whole blood samples from melanoma patients with the Qiagen QIAamp DNA Blood Kit. Serum samples used for VEGF ELISA were collected from stage III/IV melanoma patients before metastasectomy. FFPE tissues used for CD34 immunohistochemical staining were obtained from metastatic melanoma samples. Primary established cell lines (WM 1575 and WM 3248) were purchased commercially from the Wistar Institute (Philadelphia, PA), where they were authenticated by short tandem repeat (STR) profiling (http://www.wistar.org/lab/meenhard-herlyn-dvm-dsc/page/melanoma-cell-str-profiles) and used in our laboratory for no longer than 6 months following resuscitation. Primary melanoma patient–derived cell cultures were developed previously in our laboratory and were tested and authenticated against the original tumor specimens from which they were derived (24).

Targeted sequencing: The AmpliSeq Cancer Hotspot Panel v2 assay (Life Technologies) was used to amplify DNA from melanoma tumors and matched germline (peripheral blood) according to manufacturer's instructions. The assay consists of a single ultraplex PCR reaction with primer sets for 207 amplicons. We used AmpliSeq v.2 chemistry, Ion Express barcode adapters, the Ion PGM (Personal Genome Machine; Life Technologies), 200 sequencing kit, and the Ion 318 Chip.

The sequencing data underwent a primary analysis using the Torrent Suite server. Optimized signal processing, base calling, and sequence alignment was reviewed to assess the quality and accuracy of the sequencing runs. Detection of SNPs and indels from ion-sequencing data was performed using Torrent Variant Caller software within the Torrent Suite. Secondary analysis (annotation) was performed by uploading VCF files into Ion-Reporter using the ion reporter uploader plugin. All variants identified were validated by Sanger sequencing.

A total of 1,223 germline DNA samples from the validation cohort were genotyped for the KDR Q472H variant using the MassARRAY iPLEX platform (Agena Bioscience). The variant was in Hardy–Weinberg equilibrium (P = 0.526).

IHC was performed on 26 FFPE metastatic melanoma tumor samples. Tissues were evaluated for CD34 expression using a CD34-specific rabbit monoclonal antibody (1:500; Abcam, #ab81289). For each sample, the intratumoral region showing the highest microvessel density (MVD) was selected, and the density of microvessels were scored in a blinded fashion in each sample at 20× magnification by an attending pathologist (Dr. Farbod Darvishian, New York University School of Medicine, New York, NY).

VEGF in cell line supernatants and patient sera was assessed by ELISA according to manufacturer's instructions (R&D Systems). All assays were performed in duplicate.

Four primary melanoma patient–derived cell cultures (10-230, 09-085, 09-241, and 11-161) and two established melanoma cell lines (WM 1575 and WM 3248) were treated with vehicle (PBS) or 10 μg/mL of VEGFR2 blocking antibody (R&D Systems, #MAB3572), loaded with 2 μmol/L carboxyfluorescein succinimidyl ester (CFSE) and cultured in complete media. After 3 days of culture, the CFSE concentration was assessed by flow cytometry as a measure of cell proliferation. Each experiment was performed in duplicate and repeated three times.

Cultured melanoma cells (2.5 × 10⁴) were seeded in serum-free medium containing vehicle (PBS) or 10 μg/mL of VEGFR2 blocking antibody (R&D Systems, #MAB3572) in the top chamber of a porous insert coated with Matrigel (Becton Dickinson) within a 24-well plate containing complete medium (chemoattractant). After 22 hours, noninvading cells that remained in the top chamber were carefully removed using a cotton swab. Cells adhering to the bottom of the filter were fixed with cold 4% PFA and stained with 0.1% crystal violet and 20% methanol. Inserts were scanned and four quarters were photographed for each insert and counted using an Axiovert 10 inverted microscope (Carl Zeiss).

The association analysis of germline variants with melanoma survival was performed using univariate and multivariate Cox proportional hazards model under co-dominant model (2 degree freedom test) for both pilot and validation cohorts. Multivariate analyses were stratified by tumor stage and adjusted by clinicopathologic covariates: age and thickness as continuous covariates; gender, ulceration status (present/absent), and anatomic site (axial/extremity) as dichotomous covariates; and histologic type as categorical covariates. Kaplan–Meier plots were generated in R, and the statistical significance of the difference of Kaplan–Meier curves was calculated by log-rank test in dominant model: comparing WT homozygotes (Q472) versus carriers of H472 allele (pooling Q472H heterozygotes and H472 homozygotes).

Contingency tables and χ² test were used to compare different types of metastases (skin/subcutaneous, lymph nodes, lung, liver, and brain) between patients with and without the KDR Q472H variant.

Two-tailed unpaired t test was used to compare KDR Q472H positive versus negative tumors according to: (i) microvessel density through CD34 immunohistochemical staining in FFPE metastatic melanoma tumor samples and (ii) VEGF expression in melanoma cell line supernatants and sera from melanoma patients.

Two-tailed paired t test was used to compare melanoma cells treated versus untreated with VEGFR2 blocking antibody according to: (i) cell proliferation and (ii) cell invasion. This comparison was made between melanoma cells that were KDR Q472H–positive and negative.

We used Ion Torrent NGS using a targeted panel of hotspot mutations from 50 genes implicated in different cancer types to (i) detect known actionable mutations in melanoma (e.g., BRAF V600) and (ii) identify infrequent somatic mutations that occur in other tumor types and could be actionable in melanoma. Tumor and matched germline DNA from a pilot cohort of 34 patients were sequenced by Ion Torrent (Table 1). The data were filtered to exclude all synonymous and intronic variants. A quality score cutoff of 100 and an allele frequency of >5 (with a sequencing coverage > 250×) were also applied to eliminate low-frequency artifacts (Figs. 1A; Supplementary Fig. S1), identifying a total of 66 somatic mutations in our pilot cohort (n = 34) that were absent in matched germline DNA. The majority of the somatic mutations were missense mutations (60/66, 91%); 5 of 66 were nonsense mutations (8%), and only one frameshift deletion mutation was identified. The characteristic C>T transversions associated with the UV signature of DNA damage were the most frequent changes observed (25) (Fig. 1B).

In the 36 tumors we analyzed from 34 patients (pilot cohort), we identified somatic mutations in BRAF, NRAS, and KIT genes at frequencies that were consistent with those described previously (6–8) BRAF was mutated in 14 of 36 samples (39%), NRAS was mutated in 10 of 36 samples (28%), and KIT was mutated in 2 of 36 samples (6%; Fig. 1C). The majority of BRAF mutations were V600E (7/14; 50%) and V600K (3/14; 21%). Less common but previously characterized BRAF (26–28), NRAS, and KIT mutations are summarized in Supplementary Fig. S2. Paired primary/metastatic tumor samples demonstrated several instances of molecular heterogeneity (Supplementary Fig. S3).

Nine of 36 tumor samples (26%) were negative for BRAF, NRAS, or KIT mutations and hence were designated as WT melanomas. Our results show that these melanomas harbored mutations in known oncogenes and tumor suppressors, including TP53, ERBB4, PIK3CA, NOTCH, EZH2, KRAS, HRAS, and RB1 (Fig. 1C and Supplementary Fig. S4; Table 2). Although these molecules have been implicated in the development and progression of other tumor types, their specific roles in melanoma are less clear. We analyzed genes that were recurrently mutated in at least two samples in our cohort (Supplementary Fig. S2).

In the second phase of our analysis, we examined 9 germline coding variants identified through normal–tumor subtractive analysis of the Ion Torrent sequencing of the 34 melanoma patients (Fig. 1D). We elected to study a KDR germline variant Q472H further, as it has previously been shown to play a role in multiple cancer types, including lung adenocarcinoma (29), and demonstrated a high frequency in our patient cohort (Fig. 1D) but remained unexamined in the context of melanoma. In an independent validation cohort of 1,223 germline DNA samples from melanoma patients (Table 1), the minor allele frequency (MAF) of KDR variant was comparable (36.9%) with that in our pilot cohort. In comparison, the overall MAF of the KDR Q472H variant (rs1870377) is 21% in the general population (1000 Genomes Project database; http://www.1000genomes.org/), ranging from 9% in the African-American population to 47% of the Asian population, with an MAF of 13% in the Caucasian-American population. No significant associations were observed between the variant and clinicopathologic variables (Supplementary Fig. S5).

Although we did not find a significant correlation between KDR Q472H and patient survival, other studies suggest that the impact of KDR on clinical outcome might depend on other KDR SNPs or the treatments patients received (30–33). Therefore, we sought to determine whether KDR Q472H could impact the melanoma phenotype. KDR plays a crucial role in mediating angiogenic endothelial cell responses via the VEGF pathway, and VEGF levels have been reported to correlate with tumor growth rate and MVD (29, 34). We performed IHC using a CD34 antibody to quantitate MVD within 26 metastatic melanoma FFPE samples (from 13 KDR WT homozygous patients and 13 heterozygous KDR Q472H patients, including patients from our pilot cohort). Our results showed nearly 2-fold higher MVD in tumor samples from patients with KDR Q472H variant compared with KDR WT patients, where the mean MVD was 44.8 microvessels in tumors from patients carrying Q472H variant compared with 23.9 microvessels in tumors from KDR WT patients (Fig. 2A; P = 0.01).

The KDR Q472H variant is an alteration leading to increased VEGFR2 phosphorylation. In addition, the existence of a positive autocrine loop between VEGFR2 and VEGF has been proposed to lead to higher levels of VEGF secretion that ultimately sustains enhanced angiogenesis (35, 36). Therefore, we hypothesized that the KDR Q472H variant might stimulate an increase in VEGF secretion in melanoma patients and thus drive tumor angiogenesis. We first used a panel of four primary, melanoma patient–derived cell cultures [09-085, 09-241, 10-230, and 11-161; ref. 24] and two established (WM 1575 and WM 3248) melanoma cell lines, heterozygous carriers of the H472 variant or WT (Q472) KDR. All primary cultures were derived from patient tumors that were sequenced in our pilot cohort study. Analysis of VEGF levels by ELISA revealed a 29.7-fold increase in VEGF levels in tissue culture supernatants from KDR Q472H cell lines (09-085, 09-241 and 10-230; mean VEGF expression = 968.5 pg/mL) compared with the supernatants from KDR WT cell lines (11-161, WM 1575 and WM 3248; mean VEGF expression = 32.5 pg/mL; Fig. 2B). We then compared VEGF levels in sera (collected prior to surgical tumor resection) between KDR Q472H melanoma patients (n = 36) and KDR WT patients (n = 34) using ELISA. Our results confirmed that VEGF levels are significantly higher in sera from KDR Q472H patients (mean VEGF level = 401.8 pg/mL) compared with sera from KDR WT patients (mean VEGF level = 192.35 pg/mL; Fig. 2B; P = 0.04).

We next assessed the proliferation of melanoma cells carrying KDR Q472H variant compared with KDR WT melanoma cells. Although our results showed that proliferation rates were increased in melanoma cells with the KDR Q472H variant, this effect did not reach statistical significance (Fig. 3A; P > 0.05). We reasoned that KDR inhibition might impact the proliferation and invasion of melanoma cell lines carrying the KDR Q472H variant. To test this, we treated KDR Q472H and KDR WT melanoma cell lines with a VEGFR2 blocking antibody, that neutralizes VEGFR2 activity by antagonizing the binding of VEGF, and assessed cell proliferation. We found that KDR blockade selectively inhibits the proliferation of melanoma cells with the KDR Q472H variant. Our results showed that KDR inhibition abrogated cell proliferation by approximately 50% in melanoma cells with the KDR Q472H allele (P = 0.02), whereas no significant growth inhibition was seen in KDR WT melanoma cells (Fig. 3A). Finally, pretreatment of KDR Q472H and KDR WT melanoma cells lines with VEGFR2 blocking antibody resulted in a 64% reduction in cell invasion as compared with KDR WT cells, in which no significant effect on cell invasion upon treatment was observed (Fig. 3B; P = 0.003). Together, these results suggest an increased dependence of KDR Q472H melanoma cells on VEGFR2 signaling.

In this study, we performed targeted NGS to analyze both somatic and germline variants in 50 cancer-associated genes across 36 melanoma samples from 34 patients. We detected several low-frequency novel actionable somatic mutations that warrant consideration for targeting through a "basket trial" scheme. Our data also identified a germline KDR Q472H variant in one third of melanoma patients that is associated with an angiogenic phenotype and which is selectively sensitive to targeted VEGFR2 inhibition.

We demonstrated that WT melanomas harbor somatic mutations that have been implicated in the progression of other cancer types and are potentially actionable, as inhibitors targeted against these mutations are in clinical development or have already been approved for treatment of other cancers (37, 38). Our study identified nine such genes that were mutated in at least two different samples: TP53, NOTCH, KRAS, SMAD4, HRAS, CDKN2A, CTNNB1, RB1, and SMARCB1. Hence, we confirm that the targeted sequencing approach allows cancer-associated somatic mutations to be identified in melanoma tumor specimens, building on large-scale sequencing efforts in this and other tumor types (25, 39). Observed at low frequencies that make individual melanoma specific clinical trials unfeasible, these data support the further development and application of the "basket trial" concept: clinical trials of therapies targeted against a tumor's mutational status independent of the tumor histotype (eg. ClinicalTrials.gov Identifier: NCT01219699 targeting PI3K).

Clinical sequencing of tumors to identify low-frequency somatic driver mutations is likely to uncover incidental germline variants that might be relevant to cancer risk, prognostication, or treatment response. There have been extensive discussions of the utility and ethical considerations of reporting these results (40). However, these discussions have mostly focused on findings that have clear medical value but that are secondary to the indication for ordering sequencing. Although the clinical sequencing practice currently limits the use of germline patient samples as a control to identify tumor-specific somatic mutations, the sequencing of “normal”–tumor pairs will have important clinical implications by increasing the potential to identify impactful germline variants (22, 23). First, the potential to unravel deleterious germline mutations known to be associated with a high risk of cancer (e.g. BRCA1/2, APC, and p53) support the necessity to address the ethical and clinical impact of such incidental findings. Notably, in our study we detected 9 rare germline variants including 2 high-risk genes associated with cancer predisposition (TP53 and APC). The potential utility of common genetic variants in personalized medicine is emerging with the wealth of accumulated data from large population-based whole-exome sequencing efforts (41, 42). Common germline variants have the potential for routine utilization and transformative clinical impact because of their distribution in a large proportion of patients. Second, the impact of germline variants in patients may affect the tumor microenvironment, which is considered essential for tumor development and progression, and may also impact the response to antitumor therapies and ultimately patient survival (43, 44). We have recently shown that an IL10 variant (rs3024493) has been significantly associated with melanoma overall survival and IL10 secretion from CD4⁺ T cells, a finding that may have implications for the antitumor immune response (23). In turn, this could modify determinants of tumor progression and therapy resistance, including angiogenesis, immunity, and remodeling of the extracellular matrix. As such, far from being merely incidental findings, the germline variants captured on targeted-sequencing panels are, in addition to somatic mutations, plausible candidates for better clinical and molecular stratification of melanoma patients, which would enable more effective personalized prognostication and treatment.

We demonstrated that the KDR Q472H germline variant is associated with an angiogenic melanoma phenotype, characterized by significantly increased tumor vascular density and VEGF secretion. KDR is a VEGF receptor (VEGFR2) critical for physiologic and tumor angiogenesis (45), and multiple KDR genetic variants have been implicated in multiple tumor types (29, 30, 46). Consistent with our findings, KDR has previously been correlated with increased endothelial cell angiogenesis via VEGF stimulation (47), and KDR Q472H has been shown to mediate VEGFR2 phosphorylation and enhanced angiogenesis (29). Our study extends these observations by showing that melanoma cells carrying the KDR Q472H variant not only secrete more VEGF but have a higher proliferative and invasive capacity. We demonstrate a dependence of KDR Q472H melanoma cell proliferation and invasion on KDR signaling, as melanoma cells that harbor KDR Q472H were sensitive to targeted inhibition of VEGFR2, an effect that was not observed in KDR WT cell lines. Antiangiogenic approaches to tumor therapy are used clinically, such as bevacizumab, a VEGF inhibitor that is FDA-approved for colorectal cancer and more recently for platinum-resistant epithelial ovarian cancer in combination with chemotherapy. However, to date no antiangiogenic treatment has been approved for melanoma, although ongoing clinical trials are investigating VEGFR and VEGF inhibitors in the adjuvant and metastatic melanoma settings (ClinicalTrials.gov Identifier: NCT00139360; refs. 48, 49). Our preclinical data suggest that the subset of melanoma patients whose tumors carry the KDR Q472H variant may respond better to VEGFR2 inhibition. Furthermore, our observations might in part explain the poor clinical responses of a general unselected population of melanoma patients to VEGF/VEGFR inhibition. For example, the addition of bevacizumab to carboplatin plus paclitaxel (BEAM trial; ref. 50) did not significantly improve the progression-free survival of melanoma patients. It is interesting to speculate that the clinical response to antiangiogenic therapy could be analyzed separately in melanoma patients carrying the KDR Q472H germline variant versus those patients who are KDR WT.

Our observations suggest that KDR Q472H does not influence prognosis in melanoma. Previous studies have shown mixed results on the impact of KDR SNPs on prognosis in other cancers including colorectal and non–small cell lung and hepatocellular carcinomas. These results suggested that the impact of KDR on clinical outcome might depend on other KDR SNPs or the treatments patients received (30, 33, 46). These conflicting reports indicate that the precise clinical significance of individual KDR variants is likely to be specific to individual tumor types and to depend on the treatment regimen. Importantly, the size of the cohort being studied may play a crucial role in determining the relevance of KDR variants to clinical outcome, as small patient subsets used in many of these prior studies might significantly account for the variability of the observed associations with survival.

In summary, we have used a clinically validated NGS assay to identify a novel germline variant, KDR Q472H, which promotes an angiogenic phenotype in melanoma. Furthermore, we identified several actionable somatic mutations in BRAF/NRAS/KIT WT melanoma, a finding that has implications for the molecular stratification and treatment of these tumors. We propose that targeted-sequencing approaches to identify clinically relevant gene variants should not focus solely on somatic mutations but instead should also include germline variants that might interact with somatic driver mutations to further promote tumor development and progression and could also represent novel therapeutic targets belonging to the tumor microenvironment.

No potential conflicts of interest were disclosed.

Conception and design: I.P. Silva, A. Salhi, K.M. Giles, E.M. Robinson, I. Osman

Development of methodology: I.P. Silva, A. Salhi, N. Ismaili, I. Osman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I.P. Silva, A. Salhi, N. Ismaili, E.M. Robinson, R.L. Shapiro, A. Pavlick, I. Osman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I.P. Silva, A. Salhi, K.M. Giles, M. Vogelsang, S.W. Han, N. Ismaili, E.M. Robinson, M.A. Wilson, A. Pavlick, J. Zhong, T. Kirchhoff, I. Osman

Writing, review, and/or revision of the manuscript: I.P. Silva, A. Salhi, K.M. Giles, N. Ismaili, E.M. Robinson, M.A. Wilson, R.L. Shapiro, A. Pavlick, J. Zhong, T. Kirchhoff, I. Osman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Salhi, K.P. Lui, E.M. Robinson

Study supervision: I. Osman

The authors thank Dr. Farbod Darvishian for technical assistance in performing the scoring of the FFPE immunohistochemical staining.

This work was supported by the NYU Cancer Institute NCI Cancer Center Support Grant (5 P30 CA 016087-27), the NIH (1R01CA155234 to I. Osman; 1R21CA184924-01 to T. Kirchhoff), The Chemotherapy Foundation (to I. Osman), the Gulbenkian Programme for Advanced Medical Education (to I.P. Silva), and the Marc Jacobs Campaign for Melanoma Research.

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.