Inherited Genetic Variants Associated with Melanoma BRAF / NRAS Subtypes

Inherited Genetic Variants Associated with Melanoma BRAF/NRAS Subtypes Nancy E. Thomas, Sharon N. Edmiston, Irene Orlow, Peter A. Kanetsky, Li Luo, David C. Gibbs, Eloise A. Parrish, Honglin Hao, Klaus J. Busam, Bruce K. Armstrong, Anne Kricker, Anne E. Cust, Hoda Anton-Culver, Stephen B. Gruber, Richard P. Gallagher, Roberto Zanetti, Stefano Rosso, Lidia Sacchetto, Terence Dwyer, David W. Ollila, Colin B. Begg, Marianne Berwick and Kathleen Conway, on behalf of the GEM Study Group


INTRODUCTION
Genome-wide association studies and candidate pathway studies have identified low-penetrant genetic variants associated with melanoma risk. Many of these variants are in gene regions associated with pigmentation, such as TYRP1, TYR, HERC2/OCA2, SLC45A2, and ASIP; nevi, such as PLA2G6, MTAP, and NID1; or both, such as IRF4, whereas others are in genes, including ATM, MX2, PARP1, ARNT, and CASP8, not associated with melanoma-risk phenotypes (Amos et al., 2011;Barrett et al., 2011;Bishop et al., 2009;Fernandez et al., 2008;Gudbjartsson et al., 2008;Han et al., 2008;Jannot et al., 2005;Law et al., 2012;Macgregor et al., 2011;Nan et al., 2011;Zhang et al., 2012). In parallel but separate studies, we and Hacker et al. found that the increased number of nevi was associated with melanoma BRAF V600E and V600K subtypes (Hacker et al., 2016;Thomas et al., 2007Thomas et al., , 2017, and we found that BRAF V600E was associated with blond and/or light brown hair and BRAF V600K with less freckling (Thomas et al., 2017). MC1R has been inconsistently associated with BRAF V600E cases (Fargnoli et al., 2008;Hacker et al., 2010Hacker et al., , 2013Hacker et al., , 2016Landi et al., 2006;Thomas et al., 2010a). Our recent work indicates that MC1R variants are positively associated with BRAF V600E cases in people with darker phenotypes, but inversely associated with BRAF V600K cases with no significant effect modification by phenotype (Thomas et al., 2017). To our knowledge, the associations between melanoma BRAF/NRAS subtypes and other low-penetrant genetic variants besides MC1R have not been investigated.
We studied low-penetrant melanoma-risk variants in the Genes, Environment and Melanoma (GEM) Study, a large international population-based study of incident melanoma Millikan et al., 2006), in relationship to melanoma BRAF/NRAS subtypes. Participants' germline DNA was genotyped for 47 single-nucleotide polymorphisms (SNPs) from 21 distinct genomic regions, and their invasive primary melanomas were analyzed for BRAF and NRAS mutations. For each SNP, we used multinomial logistic regression to simultaneously assess its association with BRAF V600E, BRAF V600K, BRAF other, and NRASþ, compared with wild-type (WT; BRAFe/NRASe) melanoma adjusted for study features: age at diagnosis, sex, study center, and whether first-or higher-order primary melanoma.

Subject characteristics
In GEM, 1,223 participants of European origin had their incident cutaneous invasive primary melanomas analyzed for BRAF and NRAS mutations. The median age was 60 years; 59.5% were male; and 61.5% were from Australia and 38.5% from the United States (Table 1). These 1,223 melanomas (all from different patients) were from 908 patients (74.2%) who had only one melanoma at the time of recruitment (a first primary melanoma) and 315 patients (25.8%) who had more than one melanoma at the time of recruitment. For the latter group of patients, we retrieved and utilized for these analyses the 315 second-or higher-order primary melanomas that prompted their recruitment into GEM. The melanomas were 17.9% BRAF V600E, 5.6% BRAF V600K, 2.9% BRAF other, 13.5% NRASþ, and 60.2% WT. Each of these mutations was exclusive of the others. The predominant subtype was superficial spreading melanoma (67.9%). The median Breslow thickness was 0.70 mm. Of the 1,223 participants, 24.7% had 0e4, 21.7% had 11e25, and 17.9% had >25 nevi on their back. Blond or light brown was the natural hair color of a majority (63.3%) of the participants; 9.1% had red hair. A few freckles were found in 44.4% of the participants; 12.6% had many.
Of the haplotypes examined for the genes with at least two SNPs genotyped belonging to the same haplotype block, no haplotypes reached global significance Limited to individuals of European origin. Counts may not sum to the total number of study subjects due to missing data.  Table S2 online). Evaluation of the haplotype blocks and linkage disequilibrium patterns for IRF4 using both the GEM data and the Hapmap CEU population revealed that the two IRF4 SNPs were in different haplotype blocks; therefore, haplotype analysis did not apply to IRF4. We had previously found that a number of back nevi, hair color, and freckling were associated with BRAF/NRAS subtype (Thomas et al., 2017), indicating that these phenotypes could be mediators of genotype associations with BRAF/ NRAS subtype. To further explore this possibility, we examined, in participants with tumor BRAF/NRAS subtype analyzed, whether IRF4 rs12203592 or PLA2G6 rs132985 genotypes were associated with these phenotypes in models adjusted for study features (Supplementary Tables S3 online). rs12203592*T was significantly associated with fewer back nevi, darker hair color, and less freckling (all P global < 0.001). rs132985 was not significantly associated with back nevi, hair color, or freckling.
Next, we built multivariable models for each SNP that included these three phenotypes, first examining the association of each SNP with BRAF/NRAS subtype adjusted for study features, and then adding back nevi, hair color, and freckling, separately and then together to the models (Table 3). In the models for IRF4 rs12203592, the ORs for all three BRAF subtypes were attenuated as the phenotypes were progressively added to the model; in the model that included all factors, the global test for etiologic heterogeneity was not significant (P global ¼ 0.10). When examining the associations of PLA2G6 rs132985 with BRAF/NRAS subtypes, the OR of rs132985*T for BRAF V600E remained the same or similar in all the models. The OR of rs132985*T for BRAF other exon 15 mutations increased from 1.94 in the model adjusted for study features only to 1.99 in the fully adjusted model. The global tests for etiologic heterogeneity remained nominally significant.

DISCUSSION
Passing false discovery, IRF4 rs12203592*T was inversely associated with melanoma carrying BRAF V600E and V600K somatic mutations relative to WT melanoma. We, like others (Duffy et al., 2010a(Duffy et al., , 2010bHan et al., 2008;Zhang et al., 2013), found rs12203592*T to be associated with fewer nevi, darker hair color, and increased freckles. Previously, we reported that increased nevi were associated with BRAF V600E and V600K; lighter hair color with BRAF V600E; and decreased freckling with BRAF V600K compared with WT melanoma (Thomas et al., 2017). Hacker et al. (2016) also found increased nevi to be associated with BRAF V600E and V600K vs. WT melanoma. Thus, these associations are in the directions expected for potential mediation by these phenotypes of the associations of rs12203592 with BRAF subtypes. The IRF4 rs12203592 associations with BRAF subtypes were attenuated after adjustment for these three phenotypes, and these results suggest that a substantial portion of the impact of rs12203592 is mediated through these phenotypes or their underlying genotypes. Limited to 1,136 individuals of European origin who had no missing data for the IRF4 rs12203592 genotype or the phenotypes. The distribution of the melanoma subtypes was WT (n ¼ 682), BRAF V600E (n ¼ 199), BRAF V600K (n ¼ 64), BRAF other (n ¼ 33), and NRASþ (n ¼ 158). Participants with one or more data points missing for rs12203592 genotype (n ¼ 24), number of back nevi (n ¼ 20), hair color (n ¼ 3), or freckling (n ¼ 49) were excluded.
We report here positive associations of PLA2G6 rs132985*T with somatic BRAF V600E and BRAF other somatic mutations relative to WT melanoma. The literature supports that rs132985*C is positively associated with both nevus counts and melanoma risk (Duffy et al., 2017;Falchi et al., 2009;Fang et al., 2013;Kvaskoff et al., 2011). Falchi et al. (2009) explored whether nevus count mediated the association between rs132985 and melanoma and reported attenuation of the melanoma OR when adding nevus count to their model. Because of these findings and BRAF V600's known association with increased nevi (Hacker et al., 2016;Thomas et al., 2007Thomas et al., , 2017, it might have been expected that PLA2G6 rs132985*C would have been positively associated with BRAF subtypes. However, instead, we found a positive association with rs132985*T. Also, the associations of rs132985 with BRAF subtypes reported here in GEM were overall independent of the number of nevi, hair color, and freckling. Thus, the evidence provided here indicates that the associations of rs132985 with BRAF subtypes are not mediated by phenotypes. Possibly, the associations could instead be mediated by the reported apoptotic effects of the PLA2G6 gene (Akiba and Sato, 2004).
Our study's strengths are its population-based design, large sample size, and rigorous mutational analysis of BRAF and NRAS mutations. The study limitations include low numbers of BRAF V600K and BRAF other subtypes, limiting statistical power. It is possible that the power of the study was insufficient to detect the associations of some of the other SNPs tested with BRAF/NRAS subtypes. Also, we investigated a limited number of melanoma-riskeassociated genotypes and others remain to be tested in relationship to BRAF/NRAS subtypes (Duffy et al., 2017;Iles et al., 2013;Law et al., 2015). Our findings remain to be replicated.
Our results provide a link between the genetics of the person and somatic genetic data in melanoma to gain an understanding of how the genetics of both the person and the tumor interact. Our findings suggest roles for inherited IRF4 and PLA2G6 polymorphisms in the development of BRAF/ NRAS melanoma subtypes and that these roles have different underlying mechanisms. Although our work indicates that IRF4's associations are mediated by specific phenotypes, further investigation of factors underlying PLA2G6's associations with subtype is needed. Larger studies or pooled analyses and studies including more inherited melanoma-risk variants may provide additional insight into the development of melanoma molecular subtypes, further defining their risk factors and providing information that may lead toward improved prevention of this complex disease.

Study population
Details concerning the GEM study population, genotyping, and BRAF/NRAS mutational subtyping have been published previously (Begg et al., 2005;Gibbs et al., 2015Gibbs et al., , 2016Gibbs et al., , 2017Thomas et al., 2015Thomas et al., , 2017Vernali et al., 2017). Patient characteristics were collected via phone interviews and self-completed questionnaires (Kricker et al., 2007;Thomas et al., 2007Thomas et al., , 2010b. We collected patients' self-reported number of back nevi counted by a family member or friend, a measure that has been reported in other studies as predictive of total body nevus counts (Autier et al., 2001;English and Armstrong, 1994;English et al., 1988). GEM's 3,579 participants had first-or higher-order primary melanoma diagnosed between 1998 and 2003 in Australia, Canada, Italy, and the United States (Begg et al., 2004Millikan et al., 2006;Murali et al., 2012;Orlow et al., 2007). The institutional review board at each participating site approved the study protocol. Study participants provided written informed consent. Diagnostic slides were reviewed centrally for histopathologic criteria (Thomas et al., 2010b(Thomas et al., , 2013(Thomas et al., , 2014(Thomas et al., , 2015. We sought tissue sections from 2,116 participants' first-or higher-order incident invasive primary melanomas diagnosed in New South Wales (Australia), California, North Carolina, and Michigan. Of these 2,116 GEM participants, 1,227 (58%) had formalin-fixed, paraffin-embedded melanoma tissues obtained and analyzed for BRAF exon 15 (including codon 600) and NRAS exon 2 and 3 (including codons 61, 12, and 13) mutations using singlestrand conformational polymorphism analysis and radiolabeled sequencing of single-strand conformational polymorphism-positive samples (Thomas et al., 2007(Thomas et al., , 2015. Of these 1,227 patients, we limited the analyses presented here to the 1,223 participants of European origin. We previously reported the associations of BRAF/ NRAS subtypes with age, sex, tumor characteristics, and survival for 912 GEM first primaries (Thomas et al., 2015), with phenotype and MC1R for 1,227 participants (Thomas et al., 2017), and with age, sex, and phenotype in 214 GEM first primaries from North Carolina (Thomas et al., 2007) and 88 from Michigan (Poynter et al., 2006).

Statistical analysis
The melanomas were grouped as BRAF V600E, BRAF V600K, BRAF other (exon 15 mutations besides V600E and V600K), NRASþ (exon 2 or 3 mutation), or WT (wild-type negative for these mutations). Assuming an additive model of inheritance of the minor allele for each SNP, multinomial logistic regression models were used to estimate simultaneously the OR and 95% CI with BRAF V600E, BRAF V600K, BRAF other, and NRASþ, compared with WT (BRAFe/ NRASe) melanoma adjusted for study features: age at diagnosis (continuous), sex, study center, whether first or higher order primary. Some analyses were also adjusted for phenotypes. Statistical significance was assessed using Wald tests. The false discovery threshold adjusted for multiple comparisons was computed using a resampling method that takes into account the linkage disequilibrium information among SNPs evaluated and is less conservative than the classical Bonferroni procedure (He et al., 2013;Lin, 2005).
For the genes with at least two SNPs genotyped, we first determined their haplotype blocks using the Haploview software algorithm, as previously described (Gibbs et al., 2015). Each haplotype or grouped rare haplotypes were then compared with the most common haplotypes in our study population. Some associations could not be examined because of low genotype minor allele frequencies or infrequent haplotypes in some subtype categories and noted as nonestimable in the tables. For the two nominally significantly associated SNPs in the PLA2G6 locus, we applied stepwise logistic regression to determine the SNP with the stronger association keeping study features fixed.
We examined the relationship between the significantly associated genotypes and BRAF/NRAS subtype for potential mediation by phenotypes (back nevi, hair color, and freckling) associated in GEM with BRAF/NRAS subtype (Thomas et al., 2017). Using multinomial logistic regression adjusted for study features, we estimated the associations of the genotypes with these phenotypes limited to participants of European origin who had no missing data for the genotype, BRAF/NRAS subtype, or these phenotypes. Next, we used multivariable models for each SNP to examine its associations with BRAF/NRAS subtypes, adding the phenotypes separately and then together to models. All analytic models were adjusted for study features. All tests were two-sided with P < 0.05 considered statistically significant. All data were analyzed using SAS 9.4 (Cary, NC) or R (http://www.r-project.org/) programs.