Genetic analysis of the vitamin D receptor gene in two epithelial cancers: melanoma and breast cancer case-control studies

Three public databases were used to collect information about SNPs in VDR: NCBI http://www.ncbi.nlm.nih.gov, Ensembl http://www.ensembl.org, and HapMap http://www.hapmap.org. Four SNPs were considered for inclusion because they have been widely analysed in previous epidemiological studies. All had minor allele frequency (MAF) greater than or equal to 10%. Two are located on exons, one is in the putative promoter region and the other in the 3'utr region. The two coding SNPs selected have been reported to be associated with breast cancer, in previous studies [11, 20, 21].

Genotyping was carried out using the TaqMan platform following the manufacturer's instructions. SNPs assays were designed using Applied Biosystems Assay-by-Design and Assay-on-Demand probes (Applied Biosystems, Foster City, CA, USA) (provided upon request). The genotype of each sample was automatically determined by measuring final allele-specific fluorescence in the ABI Prism 7900HT Detection System, using the SDS 2.1 software for allele discrimination (Applied Biosystems, Foster city, USA).

As a quality control measure, we included at least 2 sample duplicates and 1 non-template sample per 96-well plate. Genotypes were scored by two different personnel in the laboratory. We obtained a concordance rate of 100% for all four SNPs studied.

For all polymorphisms studied, Fisher's exact test was used both to test for deviations from Hardy-Weinberg equilibrium (HWE) among controls and to compare differences in the MAF distributions between cases and controls.

In order to assess associations between genotypes, haplotypes and cancer risk, several analyses were performed. Genotype-related odds ratios (ORs), their corresponding 95% confidence intervals (CIs) and associated p-values were estimated via unconditional logistic regression. This was done for each of heterozygotes and minor-allele homozygotes relative to common-allele homozygotes, as well as under an additive model, in the latter case estimating an effect per copy of the minor allele carried. Known or suspected risk factors for BC (age, number of live births, age at menarche, and menopause status) and MM (eye colour, hair colour, skin colour, number of nevi, lentigines, and childhood sunburns) were evaluated for potential confounding effects by including them in multivariate analyses.

Among BC cases only, the presence of metastastic disease at diagnosis (yes versus no), tumour histology (invasive versus in situ), tumour grade (grade > 1 versus grade 1), tumour size (> 2 cm versus = 2 cm), nodal involvement (yes versus no), estrogen receptor status (positive versus negative) and progesterone receptor status (positive versus negative), were used in the analysis. For MM cases-only analyses, the prior diagnosis of MM (yes versus no), phototype (I/II versus III/IV), tumour location (head/neck/trunk versus extremities) and tumour depth (T2/T3/T4 versus T0/T1) were considered as the outcome variables.

SPSS v11.0 was used to carry out these analyses. All p-values were two-sided and those less than 0.05 were considered statistically significant

Allelic frequencies for each SNP and the p-value for their comparison between cases and controls are presented in Table 1. We found no evidence of departure from Hardy-Weinberg equilibrium for any of the four SNPs genotyped (all p-values > 0.05). Results from univariate and multivariate genotype analysis are shown in Table 2.

We observed evidence of differences in minor allele frequency (MAF) between BC cases and controls for the synonymous change rs731236 (TaqI) (p = 0.028). The estimated OR per minor allele (C) in this SNP was 0.84 (95%CI 0.71–0.99, p = 0.034). This per-allele OR estimate was not substantially different in the multivariate analysis adjusting for age, number of live births, age at menarche, and menopause status (OR per allele = 0.85, 95% CI 0.69–1.03, p = 0.102). Regarding the SNP rs2228570 (FokI) (Met1Thr, formerly known as rs10735810), weak evidence of differences in MAF between BC cases and controls was observed (p = 0.080). The estimated OR per minor allele in this SNP was 1.17 (95%CI 0.98–1.40, p = 0.081), whereas the per-allele OR estimated in the multivariate analysis adjusting for potential confounding factors was higher, and statistically significant (OR per allele = 1.26, 95%CI 1.02–1.57, p = 0.036).

In general, previous studies have found no evidence of association with BC for rs731236 (TaqI) and rs2228570 (FokI) [9, 11, 22‐27]. All of these studies had limited statistical power to detect a moderate association. Population stratification may be another explanation for the lack of consistency in results. However, two studies with marginal statistically significant results for rs731236 (TaqI) reported contradictory results [11, 24], whereas studies using larger sample sizes from Caucasian populations have shown risk effect of rs2228570 (FokI) consistent with that detected in the present study [20, 21]. An association with rs2228570 (FokI) was observed after adjustment for established risk factors including those used in the present study. This difference may be due to the tight relationship between VDR protein function and the hormonal aspect of BC aetiology such as menarche, parity and menopause [28, 29].

In the case of MM, we did not observe any evidence of association with rs731236 (TaqI) (OR per allele = 1.08, 95%CI = 0.84–1.40, p = 0.55) or rs2228570 (FokI) (OR per allele = 0.91 95%CI = 0.70–1.19, p = 0.49). These results are consistent with the findings of smaller sample size studies [13, 30, 31] although other studies reported a protective tendency of rs731236 (TaqI) [12, 32]. Only two studies reported an association of rs2228570 (FokI) and MM risk in North-European populations, one of them being a larger sample size study [12, 32].

The rs731236 (TaqI) SNP is in linkage disequilibrium with other polymorphisms in the 3' extreme of the gene in Caucasian populations. Functional studies of these polymorphisms have evaluated their putative implication in the regulation of transcription, translation or RNA processing, but no consistent results were obtained [33]. However, functional studies of rs2228570 (FokI) have suggested a loss of the reported VDR benefits induced by the minor allele, due to its location in the first codon of the protein (Met1Thr). That is, the minor A allele appears to be associated with the use of an alternate start codon, which triggers a longer and less potent transcriptional activator protein form [7], which is consistent with it being associated with increased risk of BC. We did not observe evidence of association for any other SNP in BC or MM.

We assessed whether VDR SNPs were associated with various clinical and phenotypic characteristics using cases and controls combined. We tested for associations with tumour characteristics among cases only for each disease. Results are summarised in Table 3. The rare allele in the non-synonymous SNP rs2228570 (TaqI) appeared to be strongly associated with the absence of childhood sunburns (OR per allele = 0.65, 95% CI 0.49–0.86, p = 0.003), and this was maintained among controls only (OR per allele = 0.63). There was also weak evidence that it is associated with a prior diagnosis of MM in MM patients (p = 0.060). We also observed marginally significant associations for the 3'utr SNP rs739837 (BglI) with fair skin colour (p = 0.048) and with Fitzpatrick's phototype I/II (0.070). Finally, the VDR promoter SNP rs4516035 was associated with tumours located in the head-neck or trunk (p = 0.020). No other associations were observed for BC or MM. Although we did not detect a significant effect of VDR SNPs directly on MM, the associations identified with MM phenotypic characteristics suggest that VDR SNPs may modulate MM susceptibility.

The strength of our study is the ability to control for the many established risk factors for BC and MM, although we recognize that there was potential for misclassification of phenotypic characteristics due to the subjective nature of the phenotypic attributes considered. Controls participated on a volunteer basis which may have introduced some selection bias. However, the fact that they were frequency matched to cases on age and sex for melanoma and that breast cancer controls were selected so that their age range was comparable to that of cases and that the variable of primary interest was genetic would have kept such bias to a minimum. It should be noted that, the sample size of both studies was relatively limited and so associations can not be ruled out for rs731236 (TaqI), and rs2228570 (FokI), particularly in MM. The association of rs731236 (TaqI) and BC, was not statistically significant under a multivariate model. However, the estimated relative risk did not change substantially, indicating that the increase in p-value was due to the reduced sample size (with available covariate data), rather than due to confounding. Finally, conclusions are based on nominal p-values at 5% statistical significance and therefore require replication in independent studies.