Inherited genetic variants associated with occurrence of multiple primary melanoma

Abstract

Recent studies including genome-wide association studies have identified several putative low-penetrance susceptibility loci for melanoma. We sought to determine their generalizability to genetic predisposition for multiple primary melanoma in the international population-based Genes, Environment, and Melanoma (GEM) Study. GEM is a case-control study of 1,206 incident cases of multiple primary melanoma and 2,469 incident first primary melanoma participants as the control group. We investigated the odds of developing multiple primary melanoma for 47 single nucleotide polymorphisms (SNP) from 21 distinct genetic regions previously reported to be associated with melanoma. ORs and 95% CIs were determined using logistic regression models adjusted for baseline features (age, sex, age by sex interaction, and study center). We investigated univariable models and built multivariable models to assess independent effects of SNPs. Eleven SNPs in 6 gene neighborhoods (TERT/CLPTM1L, TYRP1, MTAP, TYR, NCOA6, and MX2) and a PARP1 haplotype were associated with multiple primary melanoma. In a multivariable model that included only the most statistically significant findings from univariable modeling and adjusted for pigmentary phenotype, back nevi, and baseline features, we found TERT/CLPTM1L rs401681 (P = 0.004), TYRP1 rs2733832 (P = 0.006), MTAP rs1335510 (P = 0.0005), TYR rs10830253 (P = 0.003), and MX2 rs45430 (P = 0.008) to be significantly associated with multiple primary melanoma while NCOA6 rs4911442 approached significance (P = 0.06). The GEM study provides additional evidence for the relevance of these genetic regions to melanoma risk and estimates the magnitude of the observed genetic effect on development of subsequent primary melanoma.

Introduction

Clinically, melanoma is one of a small group of cancers where patients are at increased risk of potentially life-threatening subsequent primaries (1, 2) but the underlying genetic predispositions to multiple primaries are relatively unexplored. Recent genome wide association studies (GWAS) and candidate pathways studies have identified several low-penetrant genetic variants associated with cutaneous melanoma (3, 4). The majority of these variants are in gene regions associated with fair pigmentation, such as TYRP1, TYR, HERC2/OCA2, SLC45A2, and ASIP; nevi, such as PLA2G6, MTAP, and NID1; or both, such as IRF4 (4–12). More recent GWAS have identified melanoma risk-associated variants in genes, including ATM, MX2, PARP1, ARNT, and CASP8, which may not be associated with phenotypic risk (8, 13). However, the risk associated with these low-penetrant genetic variants in relation to multiple primary melanomas has rarely been evaluated.

We studied these variants in the Genes, Environment and Melanoma (GEM) Study, a large, international population-based case-control study in which the ‘cases’ are patients with multiple primary melanoma (MPM) and the ‘controls’ are patients with single primary melanoma (SPM) (14, 15). Participants’ germline DNAs were genotyped for 47 polymorphisms from 21 distinct genomic regions. We compared the odds of carrying the genotypes and haplotypes in MPM relative to SPM patients in univariable and multivariable analyses and assessed effect modification.

Materials and Methods

Study Population

The GEM Study is a population-based case-control study that enrolled 1,206 cases diagnosed with MPM (a second or higher order invasive or in situ primary melanoma) between 1998 and 2003 and 2,469 controls diagnosed with invasive SPM in 2000. In situ melanomas were eligible as MPM in order to take into account surveillance when the patient had a previous invasive melanoma. Patients were recruited from eight population-based cancer registries in the United States (New Jersey, North Carolina, California), Australia (New South Wales, Tasmania), Canada (Ontario, British Columbia), and Italy (Turin), and one hospital center in Michigan. GEM recruitment procedures and data collection have been described (14, 15). The Institutional Review Boards of all participating institutions approved the protocol; informed consent was obtained from each participant.

Demographic and melanoma risk factors, including hair and eye color, ability to tan, and number of back nevi, were collected from telephone interview and self-administered questionnaire. Using a glossy-colored guide to aid in differentiating between nevi and other skin lesions, subjects had the nevi on their backs counted by a family member or friend; and back nevi counts were categorized as 0–10 or >10 for this article. Back nevus counts were significantly correlated with whole-body nevus diagrams in GEM (data not shown). A phenotypic index variable was derived from additively combining: hair color (black or dark brown=0; light brown or blond=1; red=2), eye color (black or brown=0; green, grey, or hazel=1; blue=2), and ability to tan in response to sun exposure (deeply or moderately=0; occasionally or none=1). Those with index scores of 0 or 1 were defined as very low/low, 2 as medium low, 3 as medium high, and 4 or 5 as high/very high risk.

Genotyping

DNA was collected from buccal swab kits. Single nucleotide polymorphisms (SNPs) were genotyped using the MassArray iPLEX platform (Sequenom Inc, San Diego, CA) with quality control measures as previously reported (16). Two SNPs of interest were not compatible with the Sequenom design, and proxy SNPs (r² > 0.95) were chosen (1000 Genomes, CEU population; Proxy SNP; Broad Institute).

Statistical Analyses

Logistic regression models were used to estimate the odds ratio (OR) and 95% confidence interval (95% CI) for each SNP assuming an additive model of inheritance of the variant allele. All models were adjusted for baseline features: age, sex, an age by sex interaction, and study center. For each locus with multiple associated SNPs, we applied stepwise logistic regression to determine the SNP with the strongest association from among the significantly associated SNPs, keeping baseline variables fixed.

For the genes with at least two SNPs genotyped, we determined their haplotype blocks using the Haploview software algorithm (17) based on the pair-wise linkage disequilibrium information of the GEM population in combination with the HapMap CEU population (18). Within each haplotype block, we inferred the haplotypes in terms of probabilities for each individual from the SNP genotype input data using the PHASE algorithm, a Bayesian method in which the prior was chosen to approximate the coalescent process (19). For each haplotype block, the haplotype associations with MPM were assessed through haplotype trend regressions (HTR) (20). The HTR method effectively took into account the haplotype phase uncertainty and reduced bias by incorporating the inferred individual haplotype probabilities as the predictor variables in the regression models. Haplotypes with low estimated frequencies (< 0.01) were grouped together, reducing the number of haplotype categories and increasing the efficiency and power of haplotype analysis. Each haplotype or grouped rare haplotypes were then compared to the most common haplotypes in our study population.

Genotype and haplotype associations with phenotypic index were estimated using multinomial models and with back nevi using logistic regression models. In subsequent analyses, we limited the participants to those with no missing data for phenotypic index, back nevus counts, genotypes, and haplotypes of interest. Baseline models for genotypes and haplotypes were adjusted for baseline features; we then also adjusted the models for phenotypic index and back nevi; and finally also included all genotypes and haplotypes of interest in a multivariable model. Further, in exploratory stratified analyses, we assessed effect measure modification by phenotypic index and back nevus counts. The likelihood ratio test was used to test interactions, comparing models with main effects to models with main effects and interaction terms. All statistical tests were two-sided with P < 0.05 considered statistically significant. All data were analyzed using R (http://www.r-project.org/) or SAS 9.3 (Cary, NC) programs.

Results

The SPM and MPM patients’ age, sex, race, number of back nevi, phenotypic index, and tumors’ Breslow thicknesses are in Table S1. Twelve non-Caucasian patients were excluded from analyses. Forty-seven SNPs within 21 genetic loci previously reported to be association with melanoma were genotyped. SNP locations, minor allele frequencies, numbers of cases and controls genotyped, and literature references are in Table S2. Proxy SNPs rs6735656 and rs12278954 were used, respectively, for CASP8 rs10931936 and ATM rs1801516 identified by Barrett et al. (8).

Eleven SNPs in 6 gene neighborhoods (TERT/CLPTM1L, TYRP1, MTAP, TYR, NCOA6, and MX2) were significantly (P < 0.05) associated with MPM compared to SPM using additive models adjusted for baseline features (Table 1). The MTAP and TYR loci each had more than one significantly associated SNP. MTAP rs1335510 and TYR rs10830253 were brought forward for subsequent analyses because they each were the only SNP that remained in the stepwise logistic regression model for their locus.

Table 1.

Association of genotypes with MPM (N = 1,206) compared with SPM (N = 2,469) in the GEM Study

Chromosome	Gene neighborhood	SNP	a/A	Additive Model
				Per allele OR (95% CI)a	P value
1	ARNT	rs7412746	C/T	0.97 (0.87–1.08)	0.56
1	PARP1	rs3219090	A/G	0.90 (0.80–1.01)	0.07
1	PARP1	rs2695238	C/G	0.96 (0.85–1.07)	0.43
1	NID1	rs3768080	G/A	0.94 (0.85–1.05)	0.25
1	NID1	rs10754833	C/T	0.93 (0.84–1.04)	0.20
2	CASP8	rs6735656b	G/T	1.08 (0.96–1.22)	0.21
2	CASP8	rs13016963	A/G	1.10 (0.99–1.22)	0.09
5	TERT	rs2242652	T/C	1.04 (0.91–1.18)	0.60
5	TERT	rs2853676	A/G	1.07 (0.95–1.19)	0.28
5	TERT	rs13356727	G/A	1.10 (0.98–1.22)	0.10
5	TERT;CLPTM1L	rs4975616	G/A	1.06 (0.94–1.18)	0.34
5	TERT;CLPTM1L	rs401681	T/C	1.14 (1.02–1.26)	0.02
5	SLC45A2	rs16891982	C/G	0.71 (0.45–1.12)	0.14
5	SLC45A2	rs35391	T/C	0.70 (0.39–1.28)	0.25
5	SLC45A2	rs26722	T/C	0.66 (0.36–1.23)	0.19
5	SLC45A2	rs13289	G/C	0.99 (0.88–1.10)	0.81
6	IRF4	rs12203592	T/C	1.10 (0.97–1.24)	0.14
6	IRF4	rs872071	A/G	0.96 (0.86–1.07)	0.43
9	TYRP1	rs1408799	T/C	0.90 (0.80–1.01)	0.07
9	TYRP1	rs2733832	C/T	0.87 (0.78–0.97)	0.01
9	MTAP	rs2218220	T/C	0.88 (0.79–0.98)	0.02
9	MTAP	rs1335510d	G/T	0.82 (0.73–0.91)	0.0003
9	MTAP	rs7023329	G/A	0.87 (0.78–0.96)	0.008
9	MTAP	rs10811629	G/A	0.85 (0.76–0.94)	0.002
11	CCND1	rs11604821	G/A	1.03 (0.92–1.15)	0.64
11	CCND1	rs1485993	T/C	1.03 (0.92–1.15)	0.57
11	CCND1	rs11263498	T/C	1.01 (0.91–1.13)	0.82
11	TYR	rs1042602	A/C	0.91 (0.82–1.02)	0.11
11	TYR	rs10765198	C/T	1.22 (1.10–1.36)	0.0003
11	TYR	rs1847142	A/G	1.21 (1.09–1.35)	0.0004
11	TYR	rs10830253d	G/T	1.23 (1.10–1.37)	0.0002
11	ATM	rs12278954c	A/C	0.91 (0.78–1.05)	0.20
15	OCA2	rs1800407	A/G	1.10 (0.92–1.32)	0.29
15	OCA2	rs1800401	T/C	1.11 (0.86–1.42)	0.43
15	HERC2	rs1129038	G/A	0.99 (0.87–1.13)	0.88
15	HERC2	rs12913832	A/G	1.00 (0.88–1.13)	0.97
20	ASIP	rs17305657	C/T	1.12 (0.95–1.33)	0.18
20	ASIP	rs4911414	T/G	1.10 (0.99–1.23)	0.08
20	PIGU	rs910873	A/G	1.16 (0.99–1.36)	0.07
20	PIGU	rs17305573	C/T	1.15 (0.96–1.36)	0.12
20	NCOA6	rs4911442	G/A	1.19 (1.03–1.38)	0.02
20	MYH7B	rs1885120	C/G	1.15 (0.97–1.36)	0.11
20	LOC647979	rs1204552	A/T	1.10 (0.91–1.32)	0.32
21	MX2	rs45430	G/A	0.87 (0.77–0.97)	0.01
22	PLA2G6	rs6001027	G/A	1.03 (0.92–1.16)	0.61
22	PLA2G6	rs132985	T/C	0.96 (0.87–1.07)	0.51
22	PLA2G6	rs738322	G/A	0.95 (0.85–1.06)	0.35

NOTE: Limited to Caucasians. Bold type indicates P values < 0.05.

Abbreviations: a, minor allele; A, major allele.

^aAdjusted for age at diagnosis, sex, age by sex, and study center.

^brs6735656 is a proxy for rs10931936 (r²=0.965).

^crs12278954 is a proxy for rs1801516 (r²=1.00).

^dThe MTAP and TYR loci each had more than one SNP with a significant association (P value<.05). SNPs with the strongest association in stepwise logistic regression models are noted.

Of the haplotypes examined (Tables 2 and S3), a PARP1 haplotype (rs3219090, rs2695238) was significantly (P = 0.03) associated with MPM, as were haplotypes in MTAP and TYR, when adjusting for baseline features. However the statistical significance of the MTAP and TYR haplotype associations with MPM were weaker than the respective single MTAP rs1335510 and TYR rs10830253 SNP associations (Table S3). Thus, only the PARP1 haplotype was selected for further analysis.

Table 2.

Association of a PARP1 haplotype with multiple primary melanoma in the GEM study

Gene Haplotype	SNP		Haplotype frequency	OR (95% CI)a, b	Global P value
	1	2
PARP1: rs3219090 (SNP1), rs2695238 (SNP2)
1	G	G	0.672	1.00 (reference)	0.03
2	A	C	0.300	0.93 (0.83–1.05)
3	A	G	0.017	0.53 (0.34–0.81)
4	G	C	0.011	0.93 (0.56–1.56)

^aThe reference category is the haplotype with the highest frequency.

^bAdjusted for age, sex, age by sex, and study center.

Several of the MPM-associated genotypes were associated with phenotypic index or back nevus counts (Table S4), indicating that these SNPs are correlated with these phenotypes and may increase risk of MPM via these phenotypes, which are known melanoma risk factors.

When limiting the dataset to patients with no missing data for genotypes, haplotype, or traits of interest (Table 3), the TERT/CLPTM1L, TYRP1, MTAP, TYR, NCOA6, and MX2 genotypes, but not the PARP1 haplotype (P = 0.14), remained significantly associated with MPM after adjusting for baseline features. After additionally adjusting for phenotypic index and back nevi, the ORs did not appreciably change; although, the association with NCOA6 rs4911442 became insignificant (P = 0.07). In a multivariable model further adjusting for genotypes and the PARP1 haplotype, the TERT/CLPTM1L, TYRP1, MTAP, TYR, and MX2 genotypes remained significant, but the NCOA6 genotype (P = 0.06) and PARP1 haplotype (P = 0.22) did not; none of the ORs appreciably changed.

Table 3.

Genotype and haplotype associations with MPM (N = 1,074) compared with SPM (N = 2,137) for participants with no missing data in the GEM Study

Gene neighborhood	SNP(s)	Model	OR (95% CI)a	P value	OR (95% CI)b	P value	OR (95% CI)c	P value
Genotypes
TERT;CLPTM1L	rs401681	ADD	1.21 (1.08–1.35)	0.001	1.20 (1.07–1.35)	0.0006	1.19 (1.06–1.33)	0.004
TYRP1	rs2733832	ADD	0.83 (0.74–0.94)	0.0020	0.85 (0.75–0.95)	0.0009	0.85 (0.76–0.96)	0.006
MTAP	rs1335510	ADD	0.80 (0.71–0.89)	0.00010	0.80 (0.71–0.90)	0.0004	0.81 (0.72–0.91)	0.0005
TYR	rs10830253	ADD	1.23 (1.09–1.38)	0.0005	1.19 (1.06–1.33)	0.003	1.20 (1.06–1.34)	0.003
NCOA6	rs4911442	ADD	1.18 (1.01–1.38)	0.03	1.16 (0.99–1.35)	0.07	1.16 (0.99–1.35)	0.06
MX2	rs45430	ADD	0.87 (0.77–0.97)	0.02	0.87 (0.77–0.98)	0.02	0.85 (0.75–0.96)	0.008
Haplotype
PARP1	rs3219090; rs2695238	ACd	0.97 (0.85–1.09)	0.14	0.99 (0.87–1.12)	0.22	0.99 (0.88–1.13)	0.22
PARP1	rs3219090; rs2695238	AGd	0.52 (0.29–0.92)		0.55 (0.31–0.97)		0.55 (0.31–0.97)
PARP1	rs3219090; rs2695238	GCd	1.13 (0.64–1.97)		1.09 (0.62–1.91)		1.10 (0.62–1.95)

NOTE: Participants were included who had no missing data for phenotypic index, back nevus counts, or genotypes or haplotyes included in the table.

Abbreviations: ADD; additive.

^aAdjusted for age at diagnosis, sex, age by sex, study center.

^bAdjusted for age at diagnosis, sex, age by sex, study center, phenotypic index, and back nevi (0–10, >10).

^cAdjusted for age at diagnosis, sex, age by sex, study center, phenotypic index, back nevi (0–10, >10), and all genotypes/haplotypes in the table.

^dReference category determined by the haplotype with the highest frequency.

In an exploratory stratified analysis adjusted for baseline features, no evidence was found of effect modification by phenotypic index or back nevus counts on the association of genotypes with MPM (Table S5). However, there was evidence of effect modification by back nevi, but not phenotypic index, on the association between the PARP1 haplotype and MPM (P for interaction = 0.01). The PARP1 haplotype AG was negatively associated (P = 0.02) with MPM when 0–10 back nevi were present, while both the AG (P = 0.03) and AC (P = 0.01) haplotypes were negatively associated with MPM when >10 back nevi were present.

Discussion

In the international GEM study, we found that SNPs in TERT/CLPTM1L, TYRP1, MTAP, TYR, NCOA6, and MX2 and a PARP1 haplotype were associated with the occurrence of MPM. TERT/CLPTM1L rs401681, TYRP1 rs1408799, MTAP rs1335510, TYR rs10830253, and MX2 rs45430 were associated with MPM independently of each other and of phenotypic index and back nevi. NCOA6 rs4911442 and the PARP1 haplotype were not significant in the multivariable model, possibly as a result of diminished statistical power as there was little change in the odds ratios. There was no evidence for effect modification of SNP associations with MPM by patient phenotype; however, back nevi did modify the association of the PARP1 haplotype with MPM.

The single SNP associations reported in GEM are in the same direction as those reported in the literature (Table S2). In a recent large meta-analysis, Chatzinasiou et al. found variants in 8 of the 21 low-penetrant loci examined in our study to have strong epidemiological credibility (Venice criteria overall grade, A) as associated with melanoma, although this study did not specify whether the associations were specific to MPM, SPM or both (21). Variants in each of these loci reached significance in our study except for PIGU rs910873 (P = 0.07), MYH7B rs1885120 (P = 0.11) and SLC45A2 rs16891982 (P = 0.14), the ORs of which were in the same direction as previously reported. We are not aware of another group examining the PARP1 rs3219090, rs2695238 haplotype. Pena-Chilet et al. (22) found a similar trend towards protection from melanoma with a PARP1 haplotype containing the minor alleles of rs1136410 and rs3219090.

We know of only one other study addressing associations with MPM of common genetic variants in the loci discussed here. Helsing et al. (23) found no association of ASIP rs1015362 and rs4911414, TYR rs1126809, and TYRP1 rs1408799 with MPM in patients identified through the Norwegian Cancer Registry compared to melanoma-free blood donors. Although this design differs from GEM, we also found no association of ASIP rs4911414 and TYRP1 rs1408799 with MPM. We did not genotype ASIP rs1015362 and TYR rs1126809.

Strengths of our study include its large size, population-based case ascertainment, homogeneous questionnaire administration with respect to cases and controls, phenotypic trait adjustments, multivariable model, and examination of risk stratified by phenotypic traits. There may be a tendency for effect estimates of individual risk factors to be attenuated in our high baseline risk population (15). As evidence, several genotypes previously reported as associated with melanoma risk (PARP1 rs3219090, CASP8 rs13016963, TYRP1 rs2733832, and PIGU rs910873) had borderline associations (0.10 < P ≥ 0.05) in the same direction as the literature in GEM. However, the GEM study estimates are more relevant for survivors and their risk of subsequent melanoma than studies investigating lower risk populations. A limitation is that GEM may not have had sufficient power to detect associations of SNPs with lower minor allele frequencies (e.g. SLC45A2 rs16891982, MAF=0.017).

Subsequent melanomas are a major problem for melanoma patients but few studies have explored their genetic predisposition. Our results provide evidence that several putative low-penetrance susceptibility loci for melanoma are generalizable to risk of subsequent melanoma. Also, validation of genetic associations in the large international population-based GEM study adds further credibility that these loci are melanoma risk-associated. Knowledge of genetic risk factors for subsequent melanoma could inform screening algorithms, future risk estimation modeling, and future prevention studies for melanoma survivors.