|Home | About | Journals | Submit | Contact Us | Français|
Recent genome-wide association studies (GWAS) identified genetic loci associated with pigmentation, nevi, and skin cancer. We performed a review and meta-analysis of GWAS results, grouping them into four categories: (1) loci associated with pigmentation (hair, eye and/or skin color), cutaneous UV-response (sun sensitivity and/or freckling), and skin cancer; (2) loci associated with nevi and melanoma; (3) loci associated with pigmentation and/or cutaneous UV-response, but not skin cancer; and (4) loci distinctly associated with skin cancer, mostly basal cell carcinoma (BCC), but not pigmentation or cutaneous UV-response. These findings suggest at least two pathways for melanoma development (via pigmentation and via nevi), and two pathways for BCC development (via pigmentation and independent of pigmentation). However, further work is necessary to separate the association with skin cancer from the association with pigmentation. As with any GWAS, the identified loci may not include the causal variants and need confirmation by direct genome sequencing.
An important goal in improving our understanding of skin cancer is to identify mechanisms accounting for increased inherited susceptibility. Basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) are the most common types of skin cancer, and melanoma accounts for the most skin cancer deaths. Several high-penetrance loci for melanoma and basal cell carcinoma have been identified in melanoma-prone families and in families with nevoid basal cell carcinoma syndrome (NBCCS), respectively, including cyclin-dependent kinase inhibitor 2A (CDKN2A coding for p16 and p14ARF) and cyclin-dependent kinase 4 (CDK4) for melanoma and the human homolog of the Drosophila segment polarity gene patched (PTCH1) for BCC. Other high-penetrance loci have been suggested for melanoma on chromosomes 1p22 and 1p36, but the causal genes are yet unidentified (Bale et al., 1989; Gillanders et al., 2003; and Hussein et al., 2003). CDKN2A alterations account for approximately 20%-40% of melanomas occurring in families with three or more affected individuals (Kefford et al., 1999 and Goldstein et al., 2006), and CDK4 and PTCH1-related NBCCS only account for a small percentage of melanomas (Goldstein et al., 2006) and BCCs (Epstein 2008), respectively. Thus, environmental and other genetic factors likely account for the remaining risk.
The remaining genetic risks may be due to low penetrance susceptibility genes, such as the melanocortin-1 receptor (MC1R) gene. First recognized as a gene affecting pigmentation in animals, MC1R variants are associated with pigmentary phenotypes in humans as well, including red hair, pale skin, freckling, and sun sensitivity (reviewed in Rees 2003 and Rees 2004). These phenotypes are known to be associated with skin cancer risk, and MC1R variants are indeed associated with melanoma, BCC, and SCC; interestingly, the risk associated with melanoma remains significant after adjustment for pigmentation, suggesting that MC1R contributes to melanoma risk beyond the pigmentary phenotype or that residual confounding from pigmentation cannot be completely ruled out (Bastiaens et al., 2001; Box et al., 2001; Kennedy et al., 2001; Landi et al., 2005; Liboutet et al., 2006; Matichard et al., 2004; Palmer et al., 2000; Raimondi et al., 2008; Stratigos et al., 2006; Valverde et al., 1996, reviewed in Gerstenblith et al., 2007 and Rees 2006).
Subsequent studies examined other pigmentation genes. Animal studies identified associations with pigmentation for the agouti signaling protein (ASIP) and solute carrier family 24, member 5 (SLC24A5) genes; candidate gene studies in humans confirmed these associations with pigmentation (Graf et al., 2005; Kanetsky et al., 2002; Lamason et al., 2005; and Norton et al., 2007). Studies of the four types of oculocutaneous albinism, disorders of melanin synthesis characterized by light skin, hair, and eye pigmentation among other features, have identified additional pigmentation genes, including tyrosinase (TYR), tyrosinase-related protein 1 (TYRP1), oculocutaneous albinism type 2 gene (OCA2, previously called P gene), and solute carrier family 45, member 2 (SLC45A2, previously called membrane-associated transporter protein, MATP) (reviewed in Tomita and Suzuki, 2004). Patients with certain types of oculocutaneous albinism experience sun sensitivity and are at increased risk of skin cancer, particularly nonmelanoma skin cancers; melanomas, often amelanotic, are much less commonly reported (Luande et al. 1985; Perry and Silverberg, 2001; Terenziani et al. 2003; and Yakubu and Mabogunje, 1993). In population-based studies using candidate-gene approaches, an SLC45A2 variant was associated with dark hair, dark skin, and protection from melanoma (Fernandez et al., 2008; Graf et al., 2005; Guedj et al., 2008; and Nakayama et al., 2002). OCA2 variants were associated with melanoma in other studies (Duffy, et al., 2010 and Jannot et al., 2005), while ASIP was found to modify melanoma risk in the presence of MC1R variants (Landi et al., 2005).
Genome-wide association studies (GWAS) were conducted to possibly identify other genomic loci associated with pigmentation and skin cancer. GWAS can identify common, low-penetrance susceptibility loci without prior hypotheses on the role of specific genes. We performed a review and meta-analysis of summary results from GWAS and replication studies examining skin cancers and their major risk factors, including pigmentation, cutaneous UV-response (including sun sensitivity and/or freckling), and nevi. Using large populations with mostly Caucasian individuals, GWAS confirmed associations already known, such as MC1R with pigmentation and skin cancer, and also identified chromosomal regions previously not known to be associated with pigmentation and/or skin cancer. By including data from multiple studies in one analysis, we can better capture loci that have significant associations with pigmentation traits and susceptibility to melanoma and non-melanoma skin cancers distinctly and in combination. This allows further understanding of the complex interactions of pigmentation, nevi, and skin cancer.
We identified twelve studies that performed genome-wide analysis examining pigmentation traits, cutaneous UV-response, nevi, melanoma, basal cell carcinoma, and squamous cell carcinoma in human populations during the time from the first reports in 2007 through November 1, 2009 (Tables (TablesIaIa and andIb)Ib) (Bishop et al., 2009; Brown et al., 2008; Falchi et al., 2009; Han et al., 2008; Kayser et al., 2008; Nan et al., 2009b; Rafnar et al., 2009; Stacey et al., 2008; Stacey et al., 2009; Stokowski et al., 2007; Sulem et al., 2007; and Sulem et al., 2008). All studied Caucasian individuals except one that examined a South Asian population (Stokowski et al., 2007). In addition to searching for novel susceptibility loci, several of these studies also investigated the associations of previously identified single nucleotide polymorphisms (SNPs) with additional phenotypes (Bishop et al., 2009; Stacey et al., 2009; and Sulem et al., 2008). We also report data from four studies that did not conduct novel GWAS, but expanded or replicated findings from GWAS (Duffy et al., 2010 [epub August 2009]; Gathany et al., 2009; Gudbjartsson et al., 2008; and Nan et al., 2009a). A description of these studies is reported in Table Ic. An overview of the SNPs significantly associated with pigmentation traits including hair, eye, and skin color identified in GWAS and replicated within the same studies are presented in Table IIa. SNPs significantly associated with BCC or SCC identified in GWAS and replicated in the same or another study are presented in Table IIb. We also present detailed data for each significant SNP identified or genotyped in GWAS for hair color, eye color, skin color, sun sensitivity, freckling, nevi, and melanoma and replicated in at least one additional sample population from any study (Supplemental Tables Ia-g). SNPs with significant novel associations identified in the GWAS or replication studies are also included even if found in only one sample population; novel associations are shown in bold.
The significant loci from the GWAS were grouped into four categories: (1) loci associated with pigmentation (hair, eye, and/or skin color), cutaneous UV-response (sun sensitivity and/or freckling), and skin cancer; (2) loci associated with nevi and melanoma; (3) loci associated with pigmentation and/or cutaneous UV-response, but not skin cancer; and (4) loci associated with skin cancers, mostly basal cell carcinoma (BCC), but not pigmentation, or cutaneous UV-response. Table III shows overall associations with pigmentation, sun sensitivity, freckling, nevi, and skin cancer phenotypes for the genes located in the chromosomal regions identified by GWAS and replication studies; both significant and null associations are included for a complete synthesis of all results reported to date.
GWAS confirmed the known association of SNPs in MC1R, ASIP, TYR, TYRP1, SLC45A2, and OCA2 gene regions with pigmentation factors. SNPs in MC1R were associated with hair color (both red and light hair color), sun sensitivity, and freckling in the GWAS (Table IIa; Supplementary Tables Ia, b, e). In replication studies, Gudbjartsson et al. (2008) confirmed the previously-known associations of melanoma and BCC with MC1R,, and Duffy et al. (2010) confirmed associations with melanoma for two MC1R RHC variants, R151C and R160W (Table IIb; Supplementary Table Ig). The associations were not significant in Duffy et al. (2010), however, after adjustment for pigmentation.
SNPs in ASIP were associated with red hair color, sun sensitivity, and freckling (Table IIa; Supplementary Tables Ib, e). SNPs around MC1R and ASIP gene regions were the only ones significantly associated with red versus non-red hair color. A novel association with melanoma was found in the GWAS for loci around ASIP (Supplementary Table Ig). Replication studies further tested the association of ASIP SNPs with melanoma. In Gudbjartsson et al. (2008) and in a subset of cases and controls with 100% Northern European ancestry in Duffy et al. (2010), ASIP SNPs were associated with melanoma even after adjustment for pigmentation; Gudbjartsson et al. (2008) tested the ASIP haplotype (AH) rs1015362 and rs4911414, and Duffy et al. (2010) tested rs4911442. In Duffy et al. (2010), there was a suggestive pattern of interaction for MC1R and ASIP, as previously observed (Landi et al., 2005). In a third replication study, Nan et al. (2009a), tested the association of melanoma with three ASIP SNPs (rs1015362, rs4911414, and rs6058017) and AH (rs1015362 and rs4911414). The rs1015362 and rs6058017 SNPs and the ASIP haplotype had null associations with melanoma; rs4911414 had a suggestive association with melanoma risk but was not statistically significant.
Whereas Gudbjartsson et al. (2008) identified a novel association for BCC with the ASIP haplotype, Nan et al. (2009a) tested the association of rs1015362, rs4911414, rs6058017, and AH with BCC and SCC (Table IIb) and found that while rs1015362 and AH were not associated, rs4911414 and rs6058017 were nearly significantly associated with risk of BCC (in opposite directions); however, neither was significant after Bonferroni correction for multiple testing. The rs4911414 SNP was associated with SCC and there a suggestive association of rs6058017 with SCC, but these associations were not significant after Bonferroni correction (Nan et al. 2009a).
In GWAS, SNPs in TYR were associated with eye color, skin color, sun sensitivity, and freckling; in addition, a novel association with melanoma was found (Tables IIa; Supplementary Tables Ic-e, g). The association of TYR with melanoma was tested in replication studies. In Gudbjartsson et al. (2008), the TYR SNP rs1126809 (R402Q) was associated with melanoma after adjustment for pigmentation factors. In two other replication studies, however, although the direction of association was the same, this TYR SNP and rs1042602 (S192Y) were not associated with melanoma when adjusted for pigmentation factors (Duffy et al., 2010 and Nan et al., 2009a). In one replication study, the TYR SNP rs1126809 was significantly associated with BCC (Gudbjartsson et al., 2008) (Table IIb), while in Nan et al. (2009a) the association with BCC followed the same direction but was not significant. The TYR SNP rs1042602 was marginally associated with SCC, but the association was further diminished after Bonferroni correction (Nan et al., 2009a).
SNPs in TYRP1 were associated with eye color, and a novel association with melanoma was found in GWAS (Tables IIa; Supplementary Table Ic, g). Three replication studies also tested this association. In Gudbjartsson et al. (2008), the TYRP1 SNP rs1408799 was significantly associated with melanoma risk, even after adjustment for pigmentation factors in the Icelandic sample. In Nan et al. (2009a) and Duffy et al. (2010), the association of the same TYRP1 SNP with melanoma was confirmed, but the association was not statistically significant after Bonferroni correction for multiple comparisons in Nan et al. (2009a) and after adjustment for hair, eye, and skin color in Duffy et al. (2010).
In GWAS, SNPs in SLC45A2 were associated with light hair and skin color and sun sensitivity (Table IIa; Supplementary Tables Ia, d, e). Stacey et al. (2009) confirmed a previously identified association with melanoma. These results were confirmed in Duffy et al. (2010), with three SNPs showing significant associations with melanoma; after adjustment for pigmentation and Bonferroni correction, the SLC45A2 SNP rs16891982 remained significantly associated with melanoma. Nan et al. (2009a) also tested the association of SLC45A2 SNPs and melanoma; one SNP, rs13289, was associated with melanoma risk, but the association was not significant after Bonferroni correction. Stacey et al. (2009) identified novel associations of the SLC45A2 SNP rs16891982 with BCC and SCC (Table IIb; Supplementary Table Ig); in this study, the SLC45A2 SNP rs16891982 was associated with pigmentation factors, but results of the association with skin cancer were not adjusted for pigmentation. A subsequent replication study found no association for three SLC45A2 SNPs with either BCC or SCC (Nan et al., 2009a).
OCA2 SNPs were associated with eye color in GWAS (Table IIa; Supplementary Table 1c), confirming previous studies demonstrating that variation in eye color is linked to the OCA2 region (Duffy et al., 2007). OCA2 SNPs were not associated with melanoma in any GWAS. In a replication study, the OCA2 SNP rs1800407 was associated with melanoma but was not statistically significant after Bonferroni correction for multiple testing (Duffy et al., 2010). Other studies found an association with melanoma for this SNP and other OCA2 SNPs, but these studies were small (Fernandez et al., 2009 and Jannot et al., 2005). In Nan et al. (2009a), this OCA2 SNP was associated with BCC, but the association was not statistically significant after Bonferroni correction.
Two loci around the methylthioadenosine phosphorylase (MTAP, near CDKN2A/CDKN2B) and phospholipase A2, group VI (PLA2G6) genes were associated with nevi and melanoma (Supplementary Table If, g). The association between these genes and melanoma was no longer significant after adjustment for nevus count, suggesting that susceptibility loci for nevus count were mediating melanoma risk in this population. Nevus count has been associated with the 9p21 locus containing MTAP and CDKN2A in prior linkage studies (Falchi et al., 2006 and Zhu et al., 2007). GWAS did not confirm variants or loci previously associated with nevi by other methods, such as genome-wide linkage and candidate gene analysis on chromosomes 1, 2, 4, 6, 8, 16, 17 (Zhu et al., 2007), 5q31-32, and 2p24 (Falchi et al., 2006), and OCA2 and myosin VIIA (MYO7A) (Fernandez et al., 2009).
The loci associated with pigmentation and/or cutaneous UV-response but not melanoma or BCC included five genes, namely solute carrier family 24, member 4 (SLC24A4), two-pore segment channel 2 (TPCN2), interferon regulatory factor 4/exocyst complex component 2 (IRF4/EXOC2), kit ligand (KITLG), and hect domain and RCC1-like domain 2 (HERC2) (Table IIa). Three of these, SLC24A4, TPCN2, and IRF4, are novel potential ‘pigmentation genes.’ Loci in and around SLC24A4 were associated with hair and eye color and sun sensitivity (Supplementary Tables Ia, c, e). The association of SLC24A4 SNPs with light hair color was replicated by Duffy et al. (2010), and a new association with blue eye color was reported in that study. SLC24A4 is in the same solute carrier family as the other ‘pigmentation genes’ SLC45A2 and SLC24A5, which are potassium-dependent sodium/calcium exchangers (Lamason et al., 2005 and Sulem et al., 2007). SNPs in SLC24A5 were associated with skin color but were not tested for association with skin cancers in these GWAS (Supplementary Table Id). In a replication study, SNPs around SLC24A5 were not associated with melanoma, BCC, or SCC (Nan et al. 2009a). In GWAS, SNPs around TPCN2 were associated with hair color only (Supplementary Table Ia). Interestingly, TPCN2 encodes a protein also involved in calcium transport (Sulem et al., 2008). The relationship between solute transport and pigmentation, however, is not well-defined.
Unlike other ‘pigmentation genes’ that associate with light (or dark) pigmentation traits uniformly (i.e. light hair, light eye, light skin color and poor tanning), the IRF4 minor variant rs12203592 [T] was associated with dark hair color but light eye and skin color and sun sensitivity in GWAS (Table IIa; Supplementary Tables Ia, c-e). The locus on chromosome 6p25.3 near IRF4 (between IRF4 and EXOC2 (exocyst complex component 2)) was also associated with freckling in another GWAS (Supplementary Table Ie). Two studies further tested these associations (Duffy et al., 2010 and Gathany et al., 2009). Both studies replicated the significant association with dark hair color; Duffy et al. (2010) also replicated the significant associations with light eye and skin color for IRF4, but these associations were not significant in Gathany et al. (2009) perhaps due to the smaller sample size (Supplementary Tables Ia, c, d). A meta-analysis for the association of the IRF4 SNP rs12203592 with hair color confirmed this finding (Supplementary Table Ia). This observation suggests that for this locus, the mechanism of pigmentation control for hair color may differ from that for eye and skin color. Gudbjartsson et al. (2008) found that the association of melanoma and BCC with the chromosome 6p25.3 locus was not significant. In Duffy et al. (2010), the IRF4 SNP rs12203592 was nearly significantly associated with melanoma in the Northern European subset but was not significant after adjustment for pigmentation (Supplementary Table Ig). The IRF4 gene encodes a B cell proliferation/differentiation protein and is a member of the interferon regulatory factor family of transcription factors, which are involved in regulating gene expression in response to interferon and other cytokines. IRF4 is also called MUM1 and was used in one study to detect melanocytic lesions (nevi and melanoma) pathologically (Sundram et al., 2003). In another study, it was found to stain hematolymphoid neoplasms and melanomas but not breast, prostate, or GI tumors (Natkunam et al., 2001).
In GWAS, SNPs around KITLG loci were associated with hair color only (Supplementary Table Ia). Prior to GWAS, KITLG was known to have a role in pigmentation. KITLG functions in melanogenesis and was found to affect pigmentation in animal models (Hultman et al., 2007; reviewed in Wehrle-Haller, 2003). The association of KITLG polymorphisms with pigmentation in humans was confirmed in a study of skin color (Miller et al., 2007). Furthermore, gain of function mutations in KITLG were recently found to cause Familial Progressive Hyperpigmentation, an autosomal dominant syndrome characterized by hyperpigmented patches of the skin that expand and increase with age (Wang et al., 2009).
SNPs in HERC2, a gene located close to OCA2, were associated with hair, eye, and skin color and sun sensitivity (Supplementary Tables Ia, c-e). Different SNPs from the same locus were also associated with eye color in a candidate gene study (Eiberg et al., 2008). Previously, OCA2 variants were thought to be the primary determinants of blue eye color; however, the HERC2 SNP rs12913832 predicted eye color significantly better than any OCA2 haplotype (Sturm et al., 2008). Variants in HERC2 are thought to lead to a decrease in expression of the adjacent OCA2 gene, especially within iris melanocytes (Sturm et al, 2008). Duffy et al. (2010) found no association with melanoma for HERC2.
Four novel loci identified from GWAS were distinctly associated with BCC and one with BCC and melanoma, but not with pigmentation, sun sensitivity, freckling, or other skin cancers. The loci associated with BCC only are located around the following genes: peptidylarginine deiminase, type VI (PADI6), ras homolog gene family, member u (RHOU), kruppel-like factor 14 (KLF14), and keratin 5 (KRT5) (Table IIb). The mechanisms by which these loci increase the risk for BCC are unknown, but these findings suggest a pathway or pathways independent of pigmentation and sun sensitivity. Cyclin-dependent kinase inhibitor 2A/B (CDKN2A/B) was associated with BCC but not melanoma in GWAS. However, CDKN2A is a high penetrance melanoma susceptibility gene, as discussed previously, and in some families with CDKN2A mutations, there is an increased risk of pancreatic cancer, breast cancer, and neural tumors in addition to melanoma (Bahuau et al., 1998; Borg et al., 2000; Goldstein et al., 2006; Hewitt et al., 2002; Petronzelli et al., 2001; Randerson-Moor et al., 2001; and Rizos et al., 2001).
In GWAS, a SNP around telomerase reverse transcriptase-CLPTM1-like protein (TERT-CLPTM1L) was also associated with an increased risk of BCC and a protective effect on melanoma. The TERT-CLPTM1L locus is particularly interesting in that it is associated with multiple malignancies, including glioma and cancer of the lung, urinary bladder, prostate, cervix, and pancreas (Landi et al., 2009; Petersen et al., 2010; Rafnar et al., 2009; Shete et al., 2009; and Wang et al., 2008). While this locus is associated with increased risk of these malignancies, it is protective for melanoma in cases from Iceland, Sweden, and Spain (Rafnar et al., 2008 and Stacey et al., 2009) (Supplementary Table Ig). Variants in this locus are hypothesized to affect telomere length (Rafnar et al., 2009); in one study, short telomeres were associated with an increased risk of BCC while long telomeres were associated with an increased risk of melanoma (Han et al., 2009). Further exploration of the functional correlates of this locus is required to understand the responsible mechanisms for these associations.
We performed a meta-analysis across the GWAS and replication studies to refine previous associations and potentially identify new loci for a given phenotype. When phenotypes were defined uniformly across the GWAS, as for red hair color and melanoma, the results for SNPs identified in more than one GWAS and/or replication study were combined in a meta-analysis (Supplementary Tables Ib, g). Although few new SNPs reached genome-wide significance in the meta-analysis and previous associations were refined, no new loci emerged as significant in the meta-analysis for either phenotype. Forest plots for the melanoma meta-analysis are shown for SNPs with significant associations in at least three studies in Figure 1. The meta-analysis was limited by several factors. First, the phenotypic definitions varied across studies. For example, non-red hair color, eye color, skin color, and sun sensitivity were defined differently across GWAS, limiting our ability to combine results for these phenotypes. Second, manuscripts reported only major GWAS results, often limited to statistically significant associations, precluding the possibility of combining results that could have reached significance in a meta-analysis. Thus, using the available published data and collaboration with the primary investigators of each study, we were able to combine results only for red hair color and melanoma.
In the meta-analysis for red hair color (Supplementary Table Ib), the strongest signals for red versus non-red hair color were attributed to loci on chromosome 16 near MC1R, particularly the well-known MC1R ‘red hair color’ (RHC) alleles (R151C, R160W, and D294H) known to result in diminished function of the protein (reviewed in Garcia-Borron et al., 2005). The melanoma meta-analysis also includes replication studies, which tested SNPs identified in previous GWAS for associations with melanoma. In the melanoma meta-analysis, there is some redundancy of the study populations for several SNPs as noted in Supplementary Table Ig, which could not be quantified. For example, both Bishop et al. (2009) and Falchi et al. (2009) replicated GWAS findings in samples from the Leeds melanoma case-control study; since Falchi et al. (2009) used more Leeds samples, we excluded the replication data from Bishop et al. (2009) in the meta-analysis calculations. Also, Brown et al. (2008) and Bishop et al. (2009) used samples from the Q-MEGA study, and the redundancy could not be corrected; therefore, for SNPs with data from these studies, the meta-analysis p-values may be slightly reduced. For example, SNPs around ASIP had the strongest associations with melanoma (Figure 1), but the p-values could be affected by the inclusion of the same subjects in some of the studies contributing to the meta-analysis.
Genome-wide association studies of pigmentation, sun sensitivity, and skin cancer phenotypes have identified new loci associated with these phenotypes, some of which are shared and some of which are distinct. There are two main subsets of SNPs associated with melanoma risk, one associated with melanoma and pigmentation and one associated with melanoma and nevus count. SNPs in these subsets did not overlap, suggesting that there may be unique pathways to melanoma development. There also appears to be at least two pathways to BCC development: a pigmentation-dependent pathway and a pigmentation-independent pathway. Interestingly, all of the SNPs associated with both BCC and melanoma, except the one around TERT-CLPTM1L, are also associated with pigmentation, including SLC45A2, TYR, MC1R, and ASIP, suggesting that BCC and melanoma share a common pathway to development via pigmentation. An alternative possibility is that these SNPs are associated with pigmentation traits that predispose to skin cancer (melanoma and BCC) but that they do not themselves mechanistically lead to the development of skin cancer. A major limitation for the GWAS reviewed in this manuscript is that all but two skin cancer GWAS (Stacey et al., 2008 and Stacey et al., 2009) did not examine or adjust for skin cancer risk factors when reporting risk loci. Replication studies that adjusted for pigmentation did not consistently confirm associations for some SNPs with melanoma and BCC, including TYRP1, TYR, and ASIP (Duffy et al., 2010 and Nan et al., 2009a). Thus, future studies of these genes should focus on whether subjects carrying these SNPs are at increased risk of melanoma and/or BCC risk through pigmentation alone or through pathways independent of pigmentation.
In the GWAS of nevus count, it did appear that melanoma risk was mediated by nevi, since the association between SNPs and melanoma risk was diminished following adjustment for nevus count (Falchi et al., 2009). In this study, the number of nevi was not associated with frequency of sunburns or skin type; however, other pigmentation factors were not tested for association with nevus count. Similarly, SNPs found associated with nevus count were not tested for association with pigmentation or cutaneous UV-response. Therefore, we cannot definitively conclude that the association between nevus count and melanoma is independent of pigmentation or cutaneous UV-response. In future studies, pigmentation factors and other risk factors for melanoma should be fully accounted for so as to conclusively determine whether the genetic associations with skin cancer are direct or are mediated by pigmentation, nevi, or both. Further studies of the SNPs associated with BCC but not pigmentation may uncover novel pathways for BCC development. Finally, SNPs associated with pigmentation but not other phenotypes in these GWAS should be further examined and tested for skin cancer associations in additional populations.
GWAS have been a powerful tool for identifying novel loci associated with pigmentation and skin cancer phenotypes. Interestingly, many novel associations were found when significant loci identified in one GWAS were replicated for a different phenotype. It was difficult to perform a meta-analysis given varying phenotypic definitions of pigmentation and sun sensitivity traits, suggesting that future studies should employ consistent definitions and categorizations, when possible, to allow for data harmonization.
GWAS do have two major intrinsic limitations; first, only a relatively small number of SNPs are contained on the chip used for genotyping, leaving large numbers of SNPs untested for associations. Second, the identified loci may not be the causal variants, so associated SNPs are not necessarily causally related to the examined phenotype. Studies with MC1R demonstrate these limitations. The MC1R RHC alleles are not on the common platforms that were used for these GWAS. As noted for GWAS of red hair color, in all GWAS reviewed here, a signal on chromosome 16 near MC1R prompted further genotyping of the RHC alleles and multivariable analyses demonstrating that the signals were due to RHC alleles (Han et al., 2008; Nan et al., 2009b; Sulem et al., 2007; and Sulem et al., 2008). Interestingly, some of the MC1R SNPs identified in GWAS were not in linkage disequilibrium with the RHC alleles, suggesting that they acted independently; however, in multivariable analyses, only the association with RHC alleles remained significant. Therefore, GWAS results must be interpreted with caution: significant GWAS findings do not necessarily identify the causal variants or SNPs in linkage disequilibrium with the causal variants. Supporting this cautionary note is a recent study showing that the most significant SNP from a GWAS of a well-studied disease, sickle cell anemia, was located 9kb from the known causal variant; furthermore, these authors suggest that rare causal mutations may create ‘synthetic’ associations in GWAS that are credited to common variants (Dickson et al., 2010). This signifies the importance of prior knowledge of risk factors (like MC1R for melanoma), even in an apparent unbiased, or agnostic, approach and indicates that other tools such as direct genomic sequencing examining rarer variants should be considered in future studies.
In conclusion, although there are limitations, GWAS have provided important information regarding loci associated with pigmentation phenotypes and skin cancer. These data can guide future studies using sequencing and other techniques to identify causal variants, and generate hypotheses regarding biological mechanisms and functional consequences of the identified variants.
Using the terms “genome wide melanoma,” “genome wide sun sensitivity,” “genome wide pigmentation,” genome wide nevus,” “genome wide nevi,” “genome wide basal cell carcinoma,” “genome wide squamous cell carcinoma,” and “genome wide skin cancer,” we conducted a PubMed literature search and identified 12 studies that performed genome-wide analyses examining pigmentation and cutaneous UV-response traits, nevi, melanoma, basal cell carcinoma, and squamous cell carcinoma of the skin in human populations during the time from the first study (Stokowski et al. 2007) through November 1, 2009. We also identified four studies that expanded upon or replicated findings from GWAS within the same time frame. We reported associations that were significant at a GWAS level corresponding to a p-value<1×10−7 and replicated in at least one other sample. In studies replicating GWAS findings, a p-value<0.05 was considered significant unless otherwise specified in the study. Given different phenotypic definitions, a meta-analysis was only possible for red hair color and melanoma phenotypes.
For each given SNP associated with red hair color and/or melanoma, a meta-analysis was performed to combine odds ratios for a reference allele weighted by the estimation certainty under a fixed effects model. In all of the studies except those using pooling, the test statistics were corrected for population stratification using principal component analysis or multidimensional scaling. For all studies reviewed in this manuscript, with the exception of Gudbjartsson et al. (2008), Rafnar et al. (2009), Stacey et al. (2008), Stacey et al. (2009), Sulem et al. (2007), and Sulem et al. (2008), there was no strong evidence of cryptic relatedness or substructure based on genome control (GC) values; hence, we have used the original odds ratios (ORs) and p-values in the meta-analysis. The studies performed by deCODE Genetics (Gudbjartsson et al., 2008; Rafnar et al., 2009; Stacey et al., 2008; Stacey et al., 2009; Sulem et al., 2007; and Sulem et al., 2008) reported p-values adjusted by GC values because of strong evidence of cryptic relatedness and substructure. We have used the reported, GC-corrected test results in the meta-analysis.
Many papers reported ORs, confidence interval (CIs) of ORs, and p-values. In some cases, studies used different reference alleles for the same SNP; the reference alleles are listed in the tables to help interpret the direction of association. Where original studies reported different alleles, the reference alleles used for the meta-analyses are specified in bold in the “Meta-analysis” column of the Tables (Supplementary Tables Ib, g). The natural weights used for combining ORs are their standard deviations (SDs), which were derived from the reported CIs. The approach is appropriate if the p-values were derived from Wald-tests based on ORs and their SDs. However, in our situation, the p-values in some papers were computed based on the likelihood ratio statistics (LRT) (Rafnar et al., 2009; Stacey et al., 2008; Stacey et al., 2009; and Stokowski et al., 2007). The LRT produces much more accurate p-values than the Wald test statistic for studies with a very high proportion of controls. In this case, the meta-analysis based on ORs produces incorrect p-values. Instead, we performed a meta-analysis by combining the reported p-values in four steps:
The meta-analysis plots reporting the association between SNPs and melanoma risk (Figure 1) were obtained using the R statistical package “rmeta.”
We computed Cochran’s Q statistic (Cochran, 1954) and I2 statistic (Higgins and Thompson, 2002) to quantify the heterogeneity effect across studies. Under the null hypothesis of no heterogeneity, the Cochran’s Q statistic follows a χ2 distribution with N-1 degree of freedom, where N is the number of studies to be combined. For SNPs with suggestive evidence of heterogeneity (I2>0%), we performed a meta-analysis under a random effects model (Higgins and Thompson, 2002). As expected, the p-values under the fixed effects model and the random effects model are very different when there is strong evidence of heterogeneity.
Our analysis included only reported results for most of the reviewed SNPs. For a given SNP reported here, one or more GWAS may have failed to report the test results because the SNP was not genotyped, because the association between the SNP and the disease had been previously demonstrated to be significant, or because the SNP was not among the most significant in that study. For the latter two cases, potential reporting bias could be present in this analysis. However, for the SNPs reported in this manuscript, since existing data have provided very strong evidence of association, the potential reporting bias is unlikely to change the conclusion. Another limitation is that not all studies included in the meta-analysis reported data on all of the same SNPs; thus, we could only combine the SNPs that were reported in more than one study. For example, there were six studies that performed melanoma GWAS (Bishop et al., 2009; Brown et al., 2008; Falchi et al., 2009; Rafnar et al., 2009; Stacey et al., 2008; and Stacey et al., 2009), and four replication studies (Duffy et al. 2009; Gathany et al. 2009; Gudbjartsson et al. 2008; and Nan et al. 2009a); however, for most SNPs, we could combine data across only a few of the studies since these studies did not report data on all of the same SNPs. Only SNPs with novel and/or significant associations for melanoma are included in Supplementary Table Ig, and only SNPs reported in at least three studies are included in Figure 1. For the red hair color meta-analysis, only two GWAS examined this phenotype and were thus included, Sulem et al. (2008) and Han et al. (2008). The combined results across the two studies are reported (Supplementary Table Ib). Both studies reported many of the same SNPs. However, due to the fact that for some SNPs only p-values were reported, the direction of the association could not be inferred; therefore, the meta-analysis for red hair color was possible for most but not all reported significant SNPs.
This study was supported by the Intramural Research Program of NIH, National Cancer Institute, Division of Cancer Epidemiology and Genetics. The authors thank Drs. David Duffy, Hongmei Nan, Jiali Han, Mario Falchi, and Patrick Sulem for sharing data that allowed meta-analysis and refined associations for specific loci, and Barbara Rogers, William Wheeler, and Sara De Matteis for help with the graphical items.