The Effect on Melanoma Risk of Genes Previously Associated With Telomere Length

Telomere length has been associated with risk of many cancers, but results are inconsistent. Seven single nucleotide polymorphisms (SNPs) previously associated with mean leukocyte telomere length were either genotyped or well-imputed in 11108 case patients and 13933 control patients from Europe, Israel, the United States and Australia, four of the seven SNPs reached a P value under .05 (two-sided). A genetic score that predicts telomere length, derived from these seven SNPs, is strongly associated (P = 8.92x10-9, two-sided) with melanoma risk. This demonstrates that the previously observed association between longer telomere length and increased melanoma risk is not attributable to confounding via shared environmental effects (such as ultraviolet exposure) or reverse causality. We provide the first proof that multiple germline genetic determinants of telomere length influence cancer risk.

The ends of chromosomes are protected from instability by tandem nucleotide repeats, known as telomeres. Telomeres shorten both with age and following exposures associated with cancer risk, such as smoking and ultraviolet (UV) irradiation (1,2). Thus, telomere maintenance processes are natural candidates for explaining carcinogenesis. Telomere length has been associated with risk of various age-related diseases, including cancers (3,4). However, with inconsistent results between retrospective and prospective studies (4-7) and methodological concerns (8), conclusions have been at best cautious. The recognition that any reported association might be because of either reverse causation (the cancer itself or therapeutics affecting telomere length) (9) or shared environmental factors affecting both telomere length and cancer risk has meant that the question of a causal relationship remains unresolved.
There has, however, been consistency in studies of melanoma. Longer telomeres have been associated both with increased melanoma risk in a study involving 557 cases (10), and increased nevus number (2,11), a major risk factor for melanoma (12). A prospective study of 47 102 subjects (13) found no association between telomere length and overall cancer risk after adjusting for shared risk factors, although it did not account for potential differences in direction of effect by cancer site (14). However, alleles in the telomerase-coding gene TERT that predispose to shorter telomere length, increase the risk of most cancers but are protective for melanoma (Supplementary Materials, available online) (15). Additionally, high penetrance melanoma mutations have been reported in genes encoding components of the Shelterin complex (POT1), which is crucial for the maintenance and signaling function of telomeres (16): POT1 mutations resulted in longer telomeres (17).
The existence of genetic variants influencing both telomere length and cancer susceptibility would argue against either reverse causality or shared environmental effect (the latter affecting even prospective studies), explaining the association between telomere length and cancer risk. A recent meta-analysis (18) identified seven genome-wide statistically significant loci for mean leukocyte telomere length, five (TERC, TERT, NAF1, OBFC1, and RTEL1) containing known telomere-related genes, and two others (ZNF208 and ACYP2). Of these loci, other than TERT, only TERC and RTEL1 have been associated with risk of any disease (18)(19)(20)(21)(22). The study investigated the effect of the top SNP at each of the seven loci on risk of coronary artery disease (CAD) but, despite a huge sample size (>22 000 case patients and 64 000 control patients), no SNP was statistically significantly associated. A score based on genotypes at these loci and effect estimates from the telomere meta-analysis showed modest BRiEf coMMunicATion association with CAD risk (P = .01, associating shorter telomeres with increased risk). Another study of similar design (albeit smaller and more limited coverage) (23), found genome-wide statistical significance for association between mean telomere length and TERC, TERT, OBFC1, a novel locus at 3p14.4, and support for ACYP2, NAF1, and RTEL1. Of these, only TERT was associated with risk of breast, ovarian, and prostate cancer, while OBFC1 was associated with a subtype of ovarian cancer.
Given the potential role of telomere length in melanoma development, we investigated the variants identified by the telomere meta-analysis (18) in a genome-wide association study (GWAS) of melanoma. Our study consisted of 11 108 case patients and 13 933 control patients (Supplementary Table 1, available online) from Europe, Israel, the United States, and Australia. Written informed consent was obtained from each subject, and the investigations were performed after approval by the institutional review board for each recruiting center. As by far the biggest study of germline determinants of telomere length to date, we used the effect estimates for the seven SNPs from the telomere meta-analysis (18).
The estimated effect of these seven SNPs on telomere length (18) and their estimated effect on melanoma risk are surprisingly well correlated (Pearson's correlation = 0.92, P = .002, two-sided) ( Table 1; Supplementary  Figure 2, available online). For all but the least statistically significant telomere SNP (ACYP2), the allele associated with decreased telomere length is more frequent in control patients than melanoma case patients, consistent with a protective role for shorter telomeres in melanoma.
For each sample in our study, we constructed a genetic score predicting telomere length by calculating a weighted mean of genotype dosage across the seven telomere length-associated SNPs. The weights for each SNP were the age-and sex-adjusted effect estimates (log odds ratios) from the telomere meta-analysis (18). We then used this score in a logistic regression of melanoma risk (Supplementary Materials, available online).
We found a strong association between increased telomere score and increased risk of melanoma (P = 8.92 × 10 −9 ) that was consistent across geographic regions ( Figure 1). Categorizing telomere score into quartiles, we observed a linear effect on melanoma risk; those in the highest quartile are estimated to be at 1.29 times the risk of melanoma of those in the lowest quartile (Supplementary Figure 3, available online).
Thus, several previously-identified telomere-associated SNPs, as well as a score based on their combined effect, are associated with melanoma risk. The fact that the telomere-associated SNP is often far less statistically significant than the strongest melanoma-associated SNP at several loci is likely in part because the telomere GWAS data are imputed from a reference panel with far fewer SNPs (Supplementary Materials, available online), so the effect of these telomere loci on melanoma risk will be underestimated here. Indeed, given the large number of genetic variants that are not able to be imputed and the possibility that several genetic variants could be responsible for the signal at a single locus, it is unlikely that the top SNP identified is a functional variant, and so the effect of the locus on both melanoma risk and telomere length is likely underestimated.
Previous studies have found at best a weak association between telomere-associated loci and disease risk. This highly statistically significant association confirms the hypothesis that the genetic factors underlying telomere length have an especially strong influence on melanoma risk and that, unusually, longer telomere length  .86 * Telomere association information and minor allele frequency taken from telomere length genome-wide association study (18). MAF = minor allele frequency; SNP = single nucleotide polymorphism.
† Two-sided P values from meta-analysis of results from SNPTEST2 (24) using gene dosage and assuming an additive model.
predisposes an individual to melanoma. These seven loci explained in total only 1.2% of the variation in telomere length (18), and the combined score presented here explains only 0.14% of the variation in melanoma risk (McFadden's pseudo-r 2 ). This order of magnitude is unsurprising for such a score. To put these values in context, the most statistically significant single SNP for melanoma risk in the Leeds data set is rs258322, near MC1R (the red hair gene); it explains 5.7% of the variation in pigmentation but only 1.29% of the variation in melanoma risk. The biggest limitation of the present study is that it only considers the effect on melanoma risk of the seven most statistically significant loci from the telomere length GWAS, as these are the only ones for which results are publicly available. Ideally we would have included a larger number of potentially telomere-associated SNPs, rather than just those reaching genomewide statistical significance.
Our findings do not imply that telomere length acts directly on cancer risk and could reflect pleiotropic effects of telomere-length loci (such as the ease with which telomerase is reactivated in a melanocytic nevus). However, a mechanism for melanoma has been proposed, namely that longer telomeres increase the duration of proliferation of cells in a melanocytic nevus (11). If senescence is delayed in melanocytes, this could allow further mutations to occur, increasing the chance of malignancy (10). This is the first time that a strong association between multiple telomereassociated loci and any disease risk has been established.