Associations of vitamin D pathway genes with circulating 25-hydroxyvitamin-D, 1,25-dihydroxyvitamin-D, and prostate cancer: a nested case–control study

Vitamin D pathway single nucleotide polymorphisms (SNPs) are potentially useful proxies for investigating whether circulating vitamin D metabolites [total 25-hydroxyvitamin-D, 25(OH)D; 1,25-dihydroxyvitamin, 1,25(OH)2D] are causally related to prostate cancer. We investigated associations of sixteen SNPs across seven genes with prostate-specific antigen-detected prostate cancer. In a nested case–control study (within the ProtecT trial), we estimated odds ratios and 95 % confidence intervals (CIs) quantifying associations between SNPs and prostate cancer. Subgroup analyses investigated whether associations were stronger in men who had high/low sun exposure [a proxy for 25(OH)D]. We quantified associations of SNPs with stage (T1–T2/T3–T4) and grade (<7/≥7). Multiple variant scores included SNPs encoding proteins involved in 25(OH)D synthesis and metabolism. We included 1,275 prostate cancer cases (141 locally advanced, 385 high grades) and 2,062 healthy controls. Vitamin D-binding protein SNPs were associated with prostate cancer (rs4588-A: OR 1.20, CI 1.01, 1.41, p = 0.04; rs7041-T: OR 1.19, CI 1.02, 1.38, p = 0.03). Low 25(OH)D metabolism score was associated with high (vs low) grade (OR 0.76, CI 0.63, 0.93, p = 0.01); there was a similar association of its component variants: rs6013897-A in CYP24A1 (OR 0.78, CI 0.60, 1.01, p = 0.06) and rs10877012-T in CYP27B1 (OR 0.80, CI 0.63, 1.02, p = 0.07). There was no evidence that associations differed by level of sun exposure. We found some evidence that vitamin D pathway SNPs were associated with prostate cancer risk and grade, but not stage. There was no evidence of an association in men with deficient vitamin D (measured by having low sun exposure).


Introduction
Prostate cancer is the most common male cancer in industrialized countries but knowledge of modifiable risk factors is limited. Metabolites of vitamin D [total 25-hydroxyvitamin-D, 25(OH)D; 1,25-dihydroxyvitamin, 1,25(OH) 2 D] control cellular growth and differentiation [1,2], and administration of vitamin D analogues inhibits the progression of prostate cancer in animal models [3,4] and in phase II trials [5]. In line with ecological studies and our understanding of the biological actions of vitamin D, other epidemiological studies have shown inverse associations of circulating total 25(OH)D with prostate cancer risk [1,6,7]. Overall, however, the evidence is inconsistent, with our recent meta-analyses finding little evidence that either increased life course sun exposure, dietary vitamin D, or circulating 25(OH)D or 1,25(OH) 2 D are associated with prostate cancer risk [8][9][10]. In line with some other studies [10], recent results from the cohort reported in this paper found a two-fold increased risk of more aggressive (locally advanced stage and/or high grade) prostate cancers in men with deficient (vs adequate) circulating 25(OH)D [11], but no association of 1,25(OH) 2 D with prostate cancer risk, stage, or grade [12].
In Mendelian randomization analyses, genetic variants can be used as proxy measures of nutritional exposure to determine the unconfounded and unbiased effect of modifiable risk factors on disease outcomes since they are not subject to the biases commonly found in observational studies (e.g., reverse causation, recall bias, confounding) [13,14]. SNPs that are located in genes related to vitamin D metabolism and signalling could lend support to the hypothesis that vitamin D is related to prostate cancer and thus improve the case for a causal relationship. Several genetic variants regulate or influence the levels or actions of our exposure of interest [circulating 25(OH)D and 1,25(OH) 2 D]. Thus, if the intermediate phenotype is causally associated with the outcome (prostate cancer) then we would expect the genetic variants to be associated with the outcome, to the extent that the genetic variants affect the intermediate phenotype. There are a number of single nucleotide polymorphisms (SNPs) involved in the vitamin D pathway which are potentially useful proxies for investigating whether circulating vitamin D is causally related to prostate cancer: vitamin D 25-hydroxylase enzyme (CYP2R1) converts provitamin D (from sun exposure or dietary intake) into circulating 25(OH)D [15]; 1-ahydroxylase (CYP27B1) converts 25(OH)D into 1,25(OH) 2 D (the active form of the hormone) [16]; circulating 1,25(OH) 2 D is degraded by 24-hydroxylase (CYP24A1) [16] to 24,25(OH) 2 D; vitamin D-binding protein (VDBP or GC) is the major carrier of 25(OH)D and 1,25(OH) 2 D, transporting the metabolites to the target tissue [17]; and the vitamin D-receptor gene (VDR) is a key mediator of the biological actions of 1,25(OH) 2 D [2]. VDR SNPs have not been found to be associated with circulating vitamin D levels in previous studies, but we include these SNPs in our analysis as components of the vitamin D pathway that could have an influence on cancer, including prostate cancer, despite not being associated with circulating vitamin D levels [18]. Recent genome-wide association studies (GWAS) have uncovered robust associations of 25(OH)D concentration with polymorphisms in the genetic variants in VDBP, CYP2R1, CYP24A1, and a region on chromosome 11 encompassing the genes for the 7-dehydrocholesterol reductase (DHCR7) and NAD synthetase 1 (NADSYN1) [19,20]. Within this region, DHCR7 is the most obvious candidate encoding an enzyme that catalyzes the conversion of 7-dehydrocholesterol (a precursor of vitamin D) to cholesterol.
We investigated associations of vitamin D pathway polymorphisms with PSA-detected prostate cancer, overall and stratified by stage and grade, in a large UK-wide population-based case-control study [21]. Few studies involve exclusively PSA-detected prostate cancer, an important factor in the PSA-era due to the increasingly earlier detection of localized disease. We hypothesized that polymorphisms that reflect lower levels or cellular uptake of total 25(OH)D or 1,25(OH) 2 D are associated with an increased risk of prostate cancer and that the association is stronger for locally advanced versus localized, and highgrade versus low-grade cancer. Given previous reports [10,11,[22][23][24], we also investigated the possibility of a geneenvironment interaction, i.e., whether the association is stronger in men who have a low level of sun exposure (as a proxy for deficient vitamin D status). We hypothesized that, among men with deficient vitamin D levels, those who have a genotype which improves cellular vitamin D status would have a lower risk of prostate cancer than men without this genotype, whereas this association would not be as strong in men with sufficient levels.

Participants
The study is nested within a multi-center randomized controlled trial of treatments for localized disease: the Prostate Testing for cancer and Treatment (ProtecT) study [11,21]. During recruitment to the ProtecT study (between 2001 and 2009), over 100,000 men aged 50-69 years at 337 general practices in nine UK centers (Birmingham, Bristol, Cambridge, Cardiff, Edinburgh, Leeds, Leicester, Newcastle, Sheffield) were offered a PSA test at a community-based 'prostate check clinic', and those with raised levels (C3 ng/mL) were offered diagnostic biopsy. Detected tumors were all histologically confirmed and clinically staged using the TNM system [25]. Cancer stages T1-T2 were categorized as 'localized' and T3-T4 as 'locally advanced' as there were very few T4/metastasized tumors. 'High-grade' tumors were defined as a Gleason score C7 and low-grade tumors as Gleason score \7, after review of biopsy cores by a specialist uropathologist.

Case-control selection
A total of 1,914 cases and 48,692 controls were potentially eligible for selection for the current study (based on men recruited between 2003 and 2008), had provided a plasma sample, and consented to prostate cancer research. We randomly selected one stratum-matched control for each case from those men who had provided a non-fasted blood sample at the prostate check clinic. Controls were randomly selected from the same stratum-i.e., 5-year age band (age at PSA test) and GP/family practice-as cases. Prostate check clinics were held over consecutive weeks at each GP practice, and so matching cases and controls by GP also matches on time and season of blood draw. All participants in the ProtecT prostate check clinics who had no evidence of prostate cancer were eligible for selection as controls; that is, men with a PSA test\3 ng/mL, or a raised PSA (C3 ng/mL) combined with at least one negative biopsy and no subsequent prostate cancer diagnosis during the follow-up protocol for negative biopsies. All men provided written informed consent prior to inclusion in the study. Trent Multicentre Research Ethics Committee (MREC) approved the ProtecT study (MREC/01/4/025) and the associated ProMPT study which collected biological material (MREC/01/4/061).

Vitamin D pathway genes and vitamin D assays
The following genes were genotyped in ProtecT participants as part of a genetic association study examining the effect of 70 diet/nutrition relevant SNPs on prostate cancer risk [18,26]: VDR (ApaI: rs7975232, BsmI: rs1544410, FokI: rs10735810, TaqI: rs731236, Cdx2: rs11568820); VDBP (rs4588, rs7041); and CYP27B1 (rs10877012). DNA extraction was performed by Tepnel (http://www.tepnel. com), and genotyping was undertaken by KBioscience Ltd (www.kbioscience.co.uk), who use their own form of competitive allele-specific PCR (KASPar) and Taqman TM , for SNP analysis. Samples with more than 10 % genotype failure (7 SNPs) were defined as having poor DNA quality (2.6 %) and dropped from further analysis. Genotyping was repeated in 10 % of the study samples (with independent assessment) and for 99.98 % of those samples there was exact agreement between the two.
Circulating concentrations of total 25(OH)D (ng/mL) and 1,25(OH) 2 D (pg/mL) were measured in blood plasma collected at the prostate check clinic, prior to diagnosis, as described previously [11,28]. Briefly, 25(OH)D 2 and 25(OH)D 3 samples were measured using tandem mass spectrometry, in 31 batches over a period of approximately 3 months [11], and 1,25(OH) 2 D samples were quantified by immunoassay [28]  Vitamin D pathway scores Multiple variant allele scores were created based on SNPs found to be associated with vitamin D status in prior studies [29] and in the current study [30]. Two scores were calculated by summing up all appropriate SNPs in each individual: (1) Synthesis score: genes encoding proteins involved in 25(OH)D synthesis (CYP2R1 rs10741657, DHCR7 rs12785878) [29]; and (2) Metabolism score: genes encoding proteins involved in 25(OH)D metabolism (CYP24A1 rs6013897, CYP27B1 rs10877012) [29]. Each SNP genotype was coded as 0, 1, or 2 depending on the number of risk alleles the individual carries and their effects on vitamin D levels calculated so that an increasing score indicates decreasing levels of vitamin D. If there were missing SNP data, the individual was given a missing score.

Covariates
Measures of height, weight, weekly exercise, smoking status, family history of prostate cancer, history of benign prostatic hyperplasia, diabetes, occupational social class, and self-reported ethnicity were collected at the time of the initial PSA test [31], prior to knowledge of the PSA level or diagnosis in 85 % of men. We calculated body mass index (BMI; kg/m 2 ), which represents general adiposity. A measure of ''intense sun exposure'' was derived by summing time spent sunbathing, on holiday, and in foreign countries, from birth up until 2 years prior to the prostate clinic [8]. Missing answers were considered as zero; however, scores were not calculated if more than half of answers were missing. These sun exposure questions have been analyzed in detail previously, and further details of their derivation are published [8].
To avoid bias caused by complete case analysis [32], we multiply imputed all missing covariate values (i = 10) using chained equations [33], assuming those values could be predicted without bias from the observed relationships between covariates and the outcome measure, and substituting imputed values for missing values. The proportion of missing values per covariate was: age-group 0 %, ethnicity 0.3 %, BMI 28 %, smoking 26 %, family history of prostate cancer 11 %, history of BPH 2 %, diabetes 30 %, social class 6 %, and intense sun exposure 48 %.

Statistical analysis
Vitamin D pathway SNPs and scores and circulating 25(OH)D and 1,25(OH) 2 D Genotypes were checked for deviation from Hardy-Weinberg equilibrium using the hwsnp function implemented in Stata (Stata Corporation, College Station, Texas). Linear regression was used to examine the association of 25(OH)D and 1,25(OH) 2 D with individual SNPs and genetic scores, assuming an additive genetic model. Analyses were adjusted by age, study center, and season of blood draw.

Vitamin D pathway SNPs and scores and prostate cancer risk
To allow for the matched sets of cases and controls, conditional logistic regression was used to estimate odds ratios (OR) and 95 % confidence intervals (CIs) quantifying the association between exposure and all prostate cancers. The model included the case-control matching variables, age, and GP/family practice. A case-only analysis used unconditional logistic regression, controlling for age, study center, and season of blood draw (i.e., to reflect the matching variables), to quantify associations of SNPs with prostate cancer stage (locally advanced vs localized) and grade [high (C7) vs low (\7)]. A case-only analysis was used as all cases have undergone biopsy, therefore removing potential detection bias which could otherwise occur through misclassification of control status because of imperfect sensitivity of the PSA test [34]. SNPs were included as single variants, and effects were estimated per change in allele.

Gene-environment interaction of SNPs and scores and levels on prostate cancer
The association of SNPs with prostate cancer was repeated, stratified by level of sun exposure. Since serum vitamin D level is an effect common to vitamin D pathway SNPs and confounders, attempting to estimate this association using serum vitamin D levels would have resulted in biased estimates which may have led to spurious associations between SNPs and prostate cancer risk (this is known as collider bias [35], Fig. 1). Instead, we used sun exposure as a proxy for high or low vitamin D level due to environmental factors, assuming that men with low sun exposure will tend to have lower vitamin D levels and that sun exposure was not associated with vitamin D pathway SNPs. The association of sun exposure with serum vitamin D level and SNPs were tested using t tests and Chi-squared tests, respectively.
A likelihood ratio test, comparing the main effects model with the model including an interaction term between the SNP and level of sun exposure, was used to calculate a p value for interaction. A dichotomized indicator of sun exposure was created based on levels above and below the median level of sun exposure.

Population stratification
The top 10 principal components (PCs) that reflect the population's genetic structure were estimated according to Price et al. [36] from genome-wide SNPs genotyped, imputed and cleaned as described above. All 10 PCs were included as covariates in all regression models to account for confounding by population stratification.
All analyses were carried out in Stata 12 (StataCorp, 2012. College Station, TX). We used ice for multiple imputation with chained equations [33] for imputing missing data. All tests of statistical significance were twosided.

Characteristics of study participants
The current analysis includes 1,275 prostate cancer cases [1,131 (88.7 %) localized, 141 (11.1 %) locally advanced, 3 (0.24 %) missing stage; 887 (69.6 %) low grade, 385 (30.2 %) high grade, 3 (0.24 %) missing grade], and 2,062 controls that have at least one available SNP. 39.3 % of participants had data available on all SNPs, and 43 % were missing data on three or fewer SNPs. No man was missing more than 10 SNPs. Of these, 926 cases and 872 controls had an available 25(OH)D measurement and 779 cases and 737 controls had an available 1,25(OH) 2 D measurement (Fig. 2). The mean age of cases was 62.6 years and of controls was 61.7 years. As expected, the mean PSA level in cases was higher than in controls (9.5 vs 1.0 ng/mL). There were no substantial differences in baseline characteristics between cases and controls, except that more cases had a family history of prostate cancer versus controls (8.2 vs 5.6 %) and more cases had a normal (18.5-25) BMI (30.1 vs 25.5 %). Of the 99.3 % of subjects, who had recorded ethnicity, 98.9 % self-identified as white.
The mean (SD) 25(OH)D concentration in cases was 23.7 ng mL (8.7) and in controls was 23.5 ng/mL (8.7) (p for difference = 0.62). The mean (SD) 1,25(OH) 2 D concentration in cases was 40.6 pg.mL (18.4) and in controls was 40.9 pg/mL (18.1) (p for difference = 0.74). There were no differences between mean storage times of blood samples between cases and controls. Only VDR SNP rs7975232 was out of Hardy-Weinberg equilibrium in controls (p = 0.01).

Vitamin D pathway SNPs and covariates
For the most part, there was no evidence of associations between vitamin D pathway SNPS and the covariates (data not shown). There was evidence of an association of VDBP rs4588-A with BMI (p = 0.01), CYP27B1 rs703842-G with smoking status (p = 0.02), diabetes (p = 0.01) and social class (p = 0.01) and CY27B1 rs10877012-T with smoking status (p = 0.01), and diabetes (p = 0.01). These   Table 2). There was no convincing evidence that either the vitamin D pathway SNPs or the two scores were associated with stage (Table 3).There was evidence that the metabo- Vitamin D pathway SNPs, sun exposure and prostate cancer risk The mean 25(OH)D level in men who had below the median sun exposure was 22.1 ng/nL and who had above the median sun exposure was 24.6 ng/mL (p for difference \0.001). None of the vitamin D pathway SNPs were associated with sun exposure (data available on request). There was no evidence of an association between any SNPs or scores and prostate cancer risk within men with below the median sun exposure (all p interaction [0.04) ( Table 4).

Population stratification
All analyses were adjusted for population stratification. There was no evidence of an association between each SNP and score with the principal components used in this adjustment, indicating that population stratification was not likely to have affected our results (Supplementary Table 1).

Discussion
This study, of 1,275 prostate cancer cases and 2,062 healthy controls from the ProtecT study, investigated associations of sixteen vitamin D pathway polymorphisms with PSA-detected prostate cancer risk and, in cases, with stage and Gleason grade. There was evidence that two SNPs in VDBP, representing low 25(OH)D levels, were associated with increased prostate cancer risk and that a score measuring metabolism (indicating low 25(OH)D levels) and its component variants were associated with high Gleason grade. There was no other convincing evidence that vitamin D pathway SNPs were associated with prostate cancer risk, stage, or grade. There was no evidence that associations differed by level of sun exposure.
We validated the use of GWAS-identified SNPs in ProtecT as proxies for serum 25(OH)D as well as of scores including these variants, confirming associations reported in previous studies. The synthesis score appeared to be a reasonably strong instrument, although the scores explained less than 1 % of the trait variance for circulating 25(OH)D and 1,25(OH) 2 D. Four SNPs in VDBP (rs2282679, rs4588, rs7041, rs1155563) were strong instruments for 25(OH)D, and one SNP in VDBP (rs1155563) was a strong instrument for 1,25(OH) 2 D) (those SNPs with F C 10), explaining approximately 2 % of the variability.
Results from the Health Professionals Follow-up Study (HPFS) found that variants in CYP27A1 (p = 0.02) and VDR (12 SNPs, p = 0.01), and a score made up of seven , a pooled analysis of 10,000 cases and 11,000 controls, found that genetic variants near CYP24A1 were associated with a decreased risk of aggressive prostate cancer (p trend \0.001), and a score made of four genes thought to predict circulating levels of 25(OH)D (VDBP, CYP24A1, CYP2R1, DHCR7) was related to both overall and aggressive prostate cancer [38]. There was no association between prostate cancer risk and the other SNPs. Variants in the VDR have been associated with advanced stage or high Gleason grade [18,39], with a recent meta-analysis of 13 studies, including data from this study, finding an association between three VDR polymorphisms (ApaI, BsmI and TaqI) and prostate cancer grade [18]. However, VDR SNPs have not been consistently associated with 25(OH)D or 1,25(OH) 2 D concentrations or with prostate cancer risk. A recent meta-analysis of 34 studies found no evidence of an association of VDR BsmI and FokI with prostate cancer risk [40]. VDR Cdx2 AA genotype was associated with prostate cancer in men with low 25(OH)D (B15 ng/mL) (p interaction = 0.02) and with aggressive and high-grade prostate cancer in men with low 25(OH)D and 1,25(OH) 2 D (p interaction = 0.04 and 0.01, respectively) compared with men with normal levels [24]. Results from the Physician's Health Study [22] found no associations of VDR BsmI or TaqI [42].
Our recent epidemiological study from the same cohort found a two-fold increased risk of more aggressive (locally advanced stage and/or high grade) prostate cancers in men deficient in circulating 25(OH)D (\12 ng/mL) [11]. However, none of the SNPs expected to modulate 25(OH)D were associated with prostate cancer risk, stage, or grade. SNPs in the VDR, a key mediator of the biological actions of 1,25(OH) 2 D [2], were not associated with stage or grade. Our recent study found no association between 1,25(OH) 2 D and prostate cancer risk,stage, or grade [12]. Circulating levels of 1,25(OH) 2 D are tightly regulated [1]. Biological evidence shows that the prostate can locally convert 25(OH)D to 1,25(OH) 2 D [43,44], although prostate cancer tissue has a reduced ability to locally convert 25(OH)D to 1,25(OH) 2 D [45]. VDR is present in the prostate gland [43,46], so VDR status may better indicate local 1,25(OH) 2 D status than circulating levels. Circulating levels may not be a good indicator of what is happening at the cellular level. Recent results from the Alpha-Tocopherol, Beta-carotene Cancer Prevention (ATBC), study found that their observed association between circulating vitamin D and prostate cancer risk was made stronger when vitamin D-binding protein concentrations were also elevated [47]. This suggests that VDBP may modulate the impact of vitamin D status on prostate cancer, even though the SNPs were not directly associated with prostate cancer risk overall. Two VDBP SNPs were associated with prostate cancer risk in the current study, although there was no evidence of an interaction when stratified by sun exposure (p interaction = 0.12).

Strength and limitations of our study
Our study includes a large sample, with more men with prostate cancer and matched controls than previous studies, about which we have extensive information recorded. All of our men were resident in the UK and 99 % of our subjects self-reported their ethnicity as white. Since the decision to biopsy was based on PSA level, some of the controls with PSA \3 ng/mL will have unidentified prostate cancer [34] (misclassification bias) but this would not affect our analysis of locally advanced versus localized cancers (as all cancers were biopsy confirmed). Any misclassification of cancer status is likely to be non-differential with respect to vitamin D pathway polymorphisms, at most moderately attenuating any effect-estimates [48]. Our caseonly comparison uses 25OHD and 1,25(OH) 2 D concentrations measured in men diagnosed with locally advanced/ high-grade cancer versus men diagnosed whilst their tumor was localized/low grade. We categorized cancer stages T3-T4 as 'locally advanced' as there were very few T4/ metastasized tumors [*6 % have distal metastasis (T4 or M1)]. Circulating 25(OH)D and 1,25(OH) 2 D concentrations were measured at one laboratory, in as few batches and in as short a time frame as possible (thus attenuating any potential technical errors of measurement). Circulating vitamin D levels were measured in plasma collected at the prostate check clinic prior to diagnosis, with measurement of vitamin D concentrations blind to diagnosis. The study is population-based and thus subject to little selection bias. Circulating levels of 25(OH)D may vary by season of tissue collection [49], which is not a problem when analyzing vitamin D pathway SNPs. It is possible that we are studying a relatively healthy population, within which there is not enough variation in vitamin D status to be able to detect an affect  [50], so we may be using the wrong instruments to examine the effects of vitamin D status. Even though our sample was large, because the genetic variants explain a small proportion of the variability in circulating vitamin D levels, we would need an even larger sample to find robust evidence of an effect on prostate cancer and to be able to provide accurate estimates for this effect.
Our Mendelian randomization (MR) approach is more reliable than results from observational studies and can be used to strengthen the evidence of causality since genotypes are unlikely to be affected by confounding or reverse causation and are not subject to high levels of measurement error. Our SNPs and genetic scores were mostly not associated with the confounders, thus satisfying one of the main assumptions of MR. However, we did find evidence of an association of rs4588 with BMI and rs10877012 with smoking. These associations may be chance findings, but, if real, may pose a problem to our inference of causality, as BMI has been associated with prostate cancer [51]. All SNPs, except for the VDR SNPs, were assumed to be strong instruments based on previously published data, although in our control population only the VDBP SNPs were validated as such. Another important consideration in MR is that the instrument (i.e., the genetic score) should be associated with the outcome of interest (i.e., prostate cancer) only via the exposure (i.e., circulating vitamin D levels). For this assumption to hold the SNPs included in the genetic score cannot have pleiotropic effects on prostate cancer. This means that a genetic variant with biological pleiotropy will additionally affect prostate cancer via phenotypes unrelated to circulating vitamin D level. Since we cannot test the assumption of no effect of the instrument on the outcome via pathways other than through the exposure of interest, it is not possible to completely rule out pleiotropic influences on our results.

Conclusion
Our study found evidence that two SNPs in vitamin D-binding protein were associated with prostate cancer risk (rs4588-A and rs7041-T). A score measuring metabolism, and its component variants (rs6013897-A in CYP24A1 and rs10877012T in CYP27B1), were associated with Gleason grade (high grade vs low grade). There was no association of other vitamin D pathway polymorphisms being associated with overall prostate cancer risk, stage, or grade. There was no evidence of an association in men with deficient vitamin D (measured by having low sun exposure). distribution, and reproduction in any medium, provided the original author(s) and the source are credited.