New common variants affecting susceptibility to basal cell carcinoma

Cutaneous BCC is the most common cancer among people of European ancestry. The primary environmental risk factor for BCC is sun exposure, but genetics also has a substantial role. Some of the sequence variants that confer susceptibility seem to operate through their association with fair-pigmentation traits, common among Europeans, resulting in reduced protection from the damaging effects of UV radiation. Other sequence variants have no obvious role in pigmentation or UV susceptibility but instead seem to operate in the contexts of growth and differentiation of the basal layers of the skin^1–4.

We previously conducted a genome-wide SNP association scan (GWAS) for common BCC risk variants. We discovered susceptibility variants at 1p36, 1q42 and 5p15 (TERT-CLPTM1L)^1,4. Here we investigated 30 more high-ranking SNPs from the GWAS by genotyping them in an additional 903 Icelandic BCC cases and a case-control sample from Eastern Europe. SNPs at three loci were then selected for further investigation: rs11170164 in the keratin 5 (KRT5) gene, rs2151280 near the CDKN2A and CDKN2B genes and rs157935 on 7q32. These SNPs were further genotyped in case-control samples from Spain and the United States and proved to be significantly (P < 6 × 10⁻⁷) associated with BCC risk. An overview of the samples used in the study is presented in Supplementary Table 1; data for the 27 SNPs that were not studied further at this stage are listed in Supplementary Table 2. We also examined whether the three new BCC variants are concomitantly associated with squamous cell carcinoma (SCC), cutaneous melanoma (CM) and fair-pigmentation traits. We then used the same approach to investigate recently described variants in SLC45A2 and TERT-CLPTM1L^4–6.

After genotyping the Icelandic and non-Icelandic replication samples, a combined OR of 1.35 (P = 2.1 × 10⁻⁹) was detected for rs11170164[A], specifying a G138E substitution in KRT5 (Table 1). Because this P value is below the Bonferroni threshold for genome-wide significance (P = 1.6 × 10⁻⁷) and the association replicated consistently in different populations (OR = 1.49, P = 4.6 ×10⁻⁶ in the non-Icelandic samples), we conclude that the G138E substitution confers susceptibility to BCC. The KRT5 gene product K5 and its heterodimeric partner K14 are the major keratins of basal epithelial cells, forming the intermediate filament cytoskeletal network. This network is crucial for the structural integrity of the basal cell layer⁷.

Some sequence variants, particularly those controlling pigmentation trait variation, simultaneously affect risks of BCC, SCC and CM^1,3. Combining the Icelandic and US data for SCC revealed an OR of 1.25 (P = 0.012) for rs11170164[A] (Table 1), providing suggestive evidence of a concomitant risk of SCC. We also searched for an association with CM in 3,932 cases and 39,276 controls (Supplementary Table 1). Despite the large sample size, there was no evidence to indicate that G138E affects risk of CM (Table 1). We looked for evidence of association with eye color, hair color, freckles and sun sensitivity using self-reported data from approximately 6,200 Icelanders^8,9. There was no evidence of an association with any pigmentation trait (Supplementary Table 3) nor was there any difference in frequency of rs11170164[A] between individuals with BCC lesions at typically sun-exposed (head and arms) versus those with lesions at unexposed (trunk and legs) body sites (P = 0.09). Together, these data suggest that the G138E variant affects BCC susceptibility through mechanisms other than those related to obvious pigmentation characteristics.

We expanded our investigation of KRT5 to include six more exonic polymorphisms that are common in Europeans, including three additional nonsynonymous SNPs (nsSNPs). Two of the nsSNPs showed nominally significant signals: rs641615 (D197E) and rs11549949 (G543S) (Supplementary Table 4). For both, BCC risk was associated with the reference allele. The signal from G543S lost significance when multiple testing was taken into consideration. D197E gave a combined OR of 1.21 (P = 2.2 × 10⁻⁵) (Table 2). SNP rs641615 is in the same linkage disequilibrium (LD) block as rs11170164 (G138E) although the frequencies differ markedly (D′ = 1.00, r² = 0.03). In a multivariate analysis, the signal from D197E remained significant when adjusted for the effect of G138E (Table 2), indicating that the two substitutions affect risk independently. However, it remains possible that another unobserved variant that is in LD with both substitutions is responsible for the BCC susceptibility. The signal from G543S also retained its nominal significance when adjusted for the joint effects of G138E and D197E (Supplementary Table 4), raising the possibility, as yet unproven, of a third independent risk variant in KRT5.

Gly138 is highly conserved in vertebrates, and the Gly-to-Glu change is physicochemically nonconservative. We used a battery of predictive tests to evaluate whether G138E has an impact on K5 structure and function (Supplementary Table 5). Not all tests agreed, but the consensus was that G138E is probably damaging. Panther, for example, returned a probability of 96.6% that the substitution is deleterious. Evidence that D197E affects protein structure was less clear cut, whereas evidence for an effect of G543S was weak (Supplementary Table 5).

Some rare mutations affecting the basal cell keratins K5 and K14 cause epidermolysis bullosa (EB), a blistering condition in which the basal cell layer ruptures and breaks away from the underlying dermis¹⁰. Various forms of EB exist, ranging from mild blistering conditions to lethal failures of dermatogenesis. As with all keratins, K5 consists of a central α-helical rod flanked by nonhelical head and tail domains. EB mutations tend to cluster around the helix-initiating and helix-terminating regions at either end of the rod¹⁰. The G138E and D197E variants both occur in the helix-initiating region (Supplementary Fig. 1). The two variants were originally discovered during linkage searches for EB mutations but were discounted as neutral polymorphisms¹¹. However, subsequent reports indicate that when G138E and D197E occur as compound heterozygotes with high-penetrance EB mutations, they may be associated with a more extreme EB phenotype^12,13.

A second BCC predisposition locus was identified in the LD block on 9p21 containing the cyclin-dependent kinase inhibitor genes CDKN2A and CDKN2B, the ARF tumor suppressor encoded by CDKN2A and the noncoding RNA ANRIL^14,15 (Fig. 1 and Table 1). CDKN2A is well known for its involvement in familial melanoma¹⁶. Curiously, no risk of CM was associated with rs2151280, and we saw no evidence of an association with risk of SCC (Table 1). There was no association between rs2151280 and any of the pigmentation traits tested (Supplementary Table 3).

Previous reports indicated that rs10757278 in the CDKN2A/B LD block predisposes to coronary artery disease (CAD) (Fig. 1)¹⁷. Another SNP nearby (rs10811661) is associated with type 2 diabetes (T2D)^18–20. The LD relationships between these SNPs and the BCC SNP rs2151280 are shown in Supplementary Table 6. We genotyped rs10757278 and rs10811661 in the Icelandic BCC case-control set. Similarly, we investigated our CAD and T2D GWAS data^17,21 for associations with rs2151280. The CAD SNP rs10757278 and the T2D SNP rs10811661 showed no association with BCC risk, and the BCC SNP rs2151280 was not associated with risk of T2D (Supplementary Table 7). The rs2151280[C] allele showed a nominal association with reduced risk of CAD (OR = 0.91, P = 1.3 × 10⁻³) but this signal disappeared after adjustment for rs10757278 (residual P = 0.42; Supplementary Table 7). The simplest interpretation of these findings is that three distinct signals exist conferring separate risks of BCC, CAD and T2D. This implies that, in addition to the high-penetrance melanoma susceptibility variants in CDKN2A, at least three independent causative variants are present at 9p21, each having a quite distinct phenotypic effect.

A third new BCC susceptibility locus was tagged by rs157935 on 7q32 within fragile site FRA7H²². On replication, the T allele gave an OR of 1.23 (P = 5.7 × 10⁻¹⁰) (Table 1). The SNP showed no association with CM, SCC or pigmentation traits (Table 1 and Supplementary Table 3). The closest RefSeq gene is KLF14, encoding a Kruppel-like transcription factor that shows monoallelic maternal expression²³. KLF14 is 167 kb proximal to rs157935 and is separated from it by a region of high recombination. To investigate a possible effect of imprinting on the pattern of risk, we used the Icelandic genealogy and long-range phasing²⁴ to determine the parent-of-origin of the rs157935 alleles inherited by the Icelandic subjects who had been typed on the Illumina platform (see Online Methods). The relative risk of the T allele was estimated as 1.40 (P = 1.5 × 10⁻⁶) when inherited paternally but only 1.09 (and not significant) when inherited maternally (Table 1). As a direct test of whether the effect of the Tallele depends significantly on parent-of-origin, we observed that 237 heterozygous cases had inherited the T allele paternally whereas only 182 had inherited the T allele maternally (P = 9.6 × 10⁻³; see Online Methods). These results suggest that the increased BCC risk occurs mainly or exclusively through paternal transmission of the risk allele and is probably related to the imprinting at this locus. The rs157935 LD block contains microRNAs miR-29a and miR-29b-1, which include DNA methyltransferases among their targets²⁵. Further investigation of this locus is warranted, and any mechanistic model should consider that, although KLF14 is expressed from the maternal allele, the increased BCC risk seems to associate with the paternal allele.

A nsSNP in SLC45A2 (previously MATP) has recently been shown to confer susceptibility to melanoma in Spanish and French populations^5,6. The risk variant is the common reference allele of L374F (rs16891982) and is highly associated with fair pigmentation in Europeans²⁶. Here we confirmed the association with fair pigmentation using samples from Iceland, Eastern Europe, Spain and the United States, observing strong associations with all traits except red hair and freckles (Supplementary Table 3). We also confirmed the association with CM risk, noting that the OR was lower in Iceland than abroad (P_het for Icelandic versus non-Icelandic samples = 0.059, I² = 72%; Table 3). We have previously remarked that other pigmentation-associated variants tend to confer less relative risk of CM in Iceland than in other countries³. This may be a result of little sun exposure among Icelanders. In line with the notion that variants associated with both CM and pigmentation traits are expected to confer cross-risk of nonmelanoma skin cancers, we found that L374F also confers significant risk of BCC and SCC (Table 3). Again, we noted that the effect of the variant on BCC risk in Iceland was less than abroad (P_het for Icelandic versus non-Icelandic samples = 5.0 × 10⁻³, I² = 87%; Table 3).

We previously reported that rs401681 in the TERT-CLPTM1L locus affects risk of BCC and other cancers⁴. We also discovered evidence that the BCC risk allele rs401681[C] confers protection against CM. Here we sought to confirm this observation by typing additional CM case-control samples from Holland, Austria and Italy and larger control sample numbers from Iceland, Sweden and Spain (Table 4). Combined data from only the three new case-control sample sets revealed an OR for CM of 0.83 (P = 9.4 × 10⁻⁵) for the rs401681[C] allele, providing an independent confirmation of the initial finding (Table 4). Combining all the samples tested, the OR for CM was estimated as 0.86 (P = 5.0 × 10⁻⁸). Typing additional BCC samples from Spain and the United States also provided independent confirmation of the association with BCC, albeit with a considerably lower OR than observed in Iceland (Table 4). There was no evidence from the combined Icelandic and US sample data to support an association of rs401681 with SCC.

The rs401681[C] allele may be associated with an acceleration of the gradual shortening of telomeres with age⁴. Shorter telomeres in peripheral blood cells have been associated with an increased risk of BCC whereas longer telomeres are a risk factor for CM and nevi^27,28. This is probably related to the different roles of replicative senescence in basal keratinocytes and melanocytes²⁷. The inverse relationship between rs401681[C] and risks of BCC and CM is perhaps reflective of this difference.

To evaluate the impact of known common variants on the risk of BCC, we determined the joint population attributable risk (PAR; see Online Methods). Even without including the high-frequency SLC45A2 variant, we estimate a joint PAR of 74% for the BCC susceptibility loci we report here together with the previously known BCC susceptibility variants in ASIP, TYR, MC1R, 1p36, 1q42 and TERT-CLPTM1L^1,3,4 (Supplementary Table 8). Thus, these sequence variants have some role in most cases of BCC.