Reassessing the Association between Circulating Vitamin D and IGFBP-3: Observational and Mendelian Randomization Estimates from Independent Sources

This article is featured in Highlights of This Issue, p. 1387

Insulin-like growth factor binding protein 3 (IGFBP-3) is the most abundant circulating IGFBP and modulates the bioactivity of IGFs (1). Independent of IGFs, IGFBP-3 regulates cell proliferation, leading to increased interest in its role in carcinogenesis (2). A meta-analysis found that circulating IGFBP-3 was associated with increased odds of prostate cancer [OR for highest versus lowest quintile: 1.25; 95% confidence interval (CI),1.12–1.40; ref. 3]. This is in line with results from a nested case–control study within the Prostate Testing for Cancer and Treatment (ProtecT) trial, which found evidence that circulating IGFBP-3 was associated with increased prostate cancer odds (OR 1.28 per SD increase in IGFBP-3; 95% CI, 1.21–1.36; ref. 4). There is inconsistent evidence regarding the association between circulating IGFBP-3 and aggressive prostate cancer. However, a Mendelian randomization (MR) study concluded that the IGFs may be associated with more aggressive prostate cancer, in particular IGFBP-3 (5).

Active vitamin D [1,25-hydroxyvitamin D;1,25(OH)₂D] upregulates IGFBP-3 expression by binding to the vitamin D receptor which in turn binds to the vitamin D response element in the IGFBP-3 promoter (6, 7). In the context of this relationship, there is motivation to assess the relationship between IGFBP-3 and the inactive form of vitamin D [25-hydroxyvitamin D;25(OH)D], as it acts as a proxy for 1,25(OH)₂D and is the most likely form of supplementation (8, 9). Evidence from observational studies examining the direct association between circulating 25(OH)D and 1,25(OH)₂D with prostate cancer incidence and progression been inconclusive (10). However, there is evidence that SNPs in vitamin D pathway are associated with higher prostate cancer grade (11). Despite this, a recent MR study using SNPs [representing lower 25(OH)D] failed to provide evidence of a causal association between 25(OH)D and risk of seven types of cancers including prostate cancer (12). Currently, it is unlikely that 25(OH)D exerts large linear effects on prostate cancer, but small or nonlinear effects cannot be ruled out (13). Indeed, preclinical studies have suggested that a possible mechanism for a vitamin D effect could be through the regulation of IGFBP-3 (14–16).

Evidence from observational studies investigating the association between circulating 25(OH)D and IGFBP-3 has been conflicting. To compound this, observational studies can suffer from the effects of confounding, bias and reverse causation (17–20). Specifically, reverse causation could exist as IGFBP-3 can affect 25(OH)D through its effect on IGF-I and residual confounding might be present due to inadequate control for confounders such as BMI. Further examination of the relationship between 25(OH)D and IGFBP-3 is needed to clarify if vitamin D affects cancer via IGFBP-3.

MR, which uses genetic variants as proxies of an exposure, is a causal analysis method used to investigate causality between exposures and health outcomes. Associations between SNPs and outcomes can provide evidence of causation because they are not subjected to bias (i.e., reverse causation and confounding) found in observational studies (21). Recently, two meta-analyses of genome-wide association studies (GWASs) identified SNPs associated with circulating IGFBP-3 (22) and 25(OH)D (23), respectively. The SNPs associated with circulating IGFBP-3 and 25(OH)D provide a framework to assess the consequence of lifelong 25(OH)D on IGFBP-3 independent of other factors.

In this study, we aimed to re-assess the causal relationship between IGFBP-3 and prostate cancer. We also aimed to undertake both observational and one-sample MR analyses within ProtecT to assess the observational and causal relationship between 25(OH)D and IGFBP-3 and to compare these in order to assess the impact of potential bias, confounding or reverse causality. We also aimed to investigate whether there is a causal effect of 25(OH)D on IGFBP-3 using two-sample MR in independent data from the IGF working group of the CHARGE consortium.

This investigation had three components (Fig. 1): (i) two-sample MR analyses were conducted to reassess the causal relationship between IGFBP-3 and prostate cancer risk and advanced prostate cancer (Gleason ≥ 8, death from prostate cancer, PSA > 100 or metastasis) using summary statistics from the PRACTICAL consortium; (ii) linear regression analyses were conducted to investigate the observational relationship between circulating 25(OH)D and IGFBP-3 using individual level data from ProtecT; and (iii) one-sample MR analyses using individual level data from ProtecT and two-sample MR analyses (24, 25) using summary statistics from the IGFBP-3 GWAS were conducted to investigate the causal relationship between 25(OH)D and IGFBP-3.

Individual level analyses (observational and one-sample MR) were examined in a nested case–control study within the ProtecT trial (details of the cohort are described in the Supplementary Materials and Methods; refs. 26, 27). For the observational analyses, 1,446 cases and 1,449 controls with 25(OH)D and IGFBP-3 measured at diagnosis and before treatment were included. Circulating 25(OH)D in blood plasma was measured using tandem mass spectrometry, as described previously (28). Circulating 1,25(OH)₂D were measured in blood plasma using immunoassay, as previously described (29). Circulating IGFBP-3 in blood serum was measured using in-house radioimmunoassay, as previously described (Supplementary Materials and Methods; ref. 4). Seasonality in 25(OH)D and 1,25(OH)₂D levels were adjusted using the cosinor method as previously described (30). This method was chosen instead of adjusting models for season of blood draw, as it is a more accurate way to assess an individual's vitamin D status (31). Details of how potential confounders (BMI, exercise, smoking, alcohol consumption, family history of prostate cancer, history of benign prostatic hyperplasia, diabetes, social class, and ethnicity) were measured are provided in the Supplementary Materials and Methods. Genome-wide genotyping of participants was carried out using the Illumina Human660W array (Supplementary Materials and Methods). Altogether, 674 cases and 410 controls with genotype data and IGFBP-3 measured were included in the one-sample MR analyses.

A previous MR study (5) had investigated the association between the IGF axis with risk of prostate cancer by identifying SNPs associated with the IGFs from a GWAS for IGFBP-3 (n = 10,018; ref. 32) and candidate gene studies and using summary statistics from a prostate cancer risk GWAS (n = 22,898 cases and 23,054 controls) genotyped using the iCOGS array. A new GWAS for risk of prostate cancer (33) (n = 44,825 cases and 27,904 controls) genotyped using Oncoarray and an updated GWAS for IGFBP-3 (n = 18,995; ref. 22) were recently published. We updated the previous MR analysis for IGFBP-3 and risk of prostate cancer using the latest prostate cancer GWAS meta-analysis (33) and the latest IGFBP-3 GWAS meta-analysis (Supplementary Materials and Methods; ref. 22). For our MR analyses investigating the causal relationship between 25(OH)D and IGFBP-3, we used the following summary statistics: (i) IGFBP-3 GWAS (22) conducted by the IGF working group of the CHARGE consortium (n = 18,995); and (ii) 25(OH)D GWAS (n = 79,366; ref. 34) conducted by the SUNLIGHT consortium (Supplementary Materials and Methods).

The previous MR study (5) investigating the causal relationship between IGFs and prostate cancer used individual SNPs identified either by the discovery IGF GWAS (32) or by candidate gene studies. For our MR analyses, we constructed a new instrument for IGFBP-3 by selecting four independent IGFBP-3 associated SNPs (rs11977526, rs700753, rs1065656, rs4234798) identified by the updated IGFBP-3 GWAS (22). rs700753 was not present in the 1000 genomes imputation reference used in the prostate cancer GWAS. We identified a proxy SNP, rs700752, which was in high linkage disequilibrium (r² > 0.8) using SNAP (https://www.broadinstitute.org/mpg/snap/).

To-date, GWAS studies (23, 35, 36) have focused on identifying SNPs associated with 25(OH)D as it is more stable than 1,25(OH)₂D in the circulation. The latest 25(OH)D GWAS meta-analysis (34) conducted in Europeans identified two novel loci [SEC23A (rs8018720) and AMDHD1 (rs10745742)] with a genome-wide significant association with 25(OH)D and confirmed four previously identified loci (CYP2R1(rs10741657); DHCR7(rs12785878); GC(rs3755967); and CYP24A1(rs17216707); ref. 23) located in or near genes involved in 25(OH)D synthesis and metabolism. We constructed an instrument for 25(OH)D (allele score) by using the six 25(OH)D-associated SNPs. As sensitivity analyses, we constructed a synthesis (rs12785878 and rs10741657) and metabolism (rs3755967 and rs17216707) score using SNPs in genes involved in 25(OH)D synthesis and metabolism.

SNP-exposure (IGFBP-3) and SNP-outcome (prostate cancer) estimates for the four IGFBP-3–associated SNPs were combined using inverse-variance weighted (IVW) and maximum likelihood method to provide a weighted average of the causal estimates (37). The previous MR study found that SNPs associated with IGFBP-3 have pleiotropic effects on other biomarkers of the IGF pathway (5). The latest IGF GWAS found that two IGFBP-3–associated SNPs (rs700753 and rs1065656) were also associated with IGF-I at genome-wide significant levels (22). To account for the effect of the IGFBP-3–associated SNPs on IGF-I, we used multivariable MR, a new MR method that estimates causal effects using multiple SNPs associated with multiple exposures simultaneously (38, 39). Multivariable MR was conducted by regression of the SNP-prostate cancer estimates on SNP-IGFBP-3 and SNP-IGF-I estimates in a multivariable weighted regression model. As the IGF GWAS did not analyze other IGFs, the multivariable MR method could not take into account the association between SNPs and other biomarkers of the IGF pathway.

Observational associations between 25(OH)D and IGFBP-3 were assessed using linear regression for cases and controls combined into one cohort or stratified by case–control status. Additional analyses were adjusted for potential confounders (age, center, BMI, smoking, and diabetes status). Associations of 25(OH)D and IGFBP-3 with potential confounders (listed above) were estimated using linear regression. Potential nonlinear effects of 25(OH)D and IGFBP-3 were tested by carrying out linear regression with quadratic terms. Details of the analysis investigating the observational relationship between circulating 1,25(OH)₂D and IGFBP-3 in ProtecT are provided in the Supplementary Materials and Methods.

A weighted genetic risk score (GRS) for 25(OH)D was generated by taking the sum of genotypes (coded 0,1,2) multiplied by the strength of the effect of each SNP on 25(OH)D recorded by the 25(OH)D GWAS (23). Two-stage least squares analyses using the weighted GRS was used to obtain estimates of the associations between season-adjusted 25(OH)D and IGFBP-3 (40, 41). The properties of the 25(OH)D-associated SNPs as instruments were assessed in ProtecT by examination of: (i) first-stage F-statistic (measure of the strength of the association between the instrument and exposure) and (ii) associations of the SNPs with potential confounders. The first-stage F-statistic was obtained from the regression of 25(OH)D on the 25(OH)D genetic instrument (first stage of the two-stage least squares MR analysis). Sensitivity analyses were also performed by combining the estimates from SNPs involved in the synthesis and metabolism of vitamin D separately. We compared the instrumental variable estimates with those from observational analyses using the Durbin form of the Durbin–Wu–Hausman statistic (42).

The SNP-exposure [25(OH)D] and SNP-outcome (IGFBP-3) estimates for the six 25(OH)D-associated SNPs were combined using the IVW method (37). The IVW method (similar method to that used in the two-sample MR analysis of IGFBP-3 and prostate cancer) is equivalent to the two-stage least square analyses using data from ProtecT (43). Sensitivity analyses were performed by combining the estimates from SNPs involved in the synthesis and metabolism pathways separately. As only a small number of SNPs were included in our MR analyses, methods to test for pleiotropy such as MR-Egger regression (43), weighted median (44), and mode method (45) were not conducted as these lack power with small number of SNPs. A leave-one-out analysis, which repeats the MR analysis by leaving out each SNP in turn, was applied to assess whether any single SNP was driving the causal estimate. In the reverse direction, we investigated the causal effect of IGFBP-3 on 25(OH)D using the two-sample MR analyses (see Supplementary Materials and Methods).

Power for one-sample MR analyses to detect the causal effect of 25(OH)D on IGFBP-3 was estimated using the online tool (http://glimmer.rstudio.com/kn3in/mRnd/; ref. 46), specifying α = 0.05, R² = 0.05 (35) and using the observational estimates from the ProtecT sample set. Given a sample size of 1,134, in one-sample MR, a causal effect of 0.39 SD units per SD unit change in 25(OH)D was required to yield 80% power. For two-sample MR analysis of the same relationship, power to detect the causal effect of 25(OH)D on IGFBP-3 was estimated using the “pwr.r.test” function (implemented in the pwr package in R) specifying α = 0.05. The correlation between 25(OH)D and IGFBP-3 was obtained by taking the square-root of the variance in IGFBP-3 explained by the 25(OH)D associated SNPs. Given a sample size of 18,995 in the two-sample MR analysis, a causal effect needed to yield 80% power is 0.09 SD units. This method is not dissimilar to that suggested elsewhere noting that power in a two-sample procedure is limited by the direct SNP–outcome association (47). All analyses were performed in R (version 3.0.1), Stata version 14 (Stata Corp) and using MR-Base (www.mrbase.org).

In agreement with the previous MR of IGFBP-3 and prostate cancer (5), two-sample MR analyses using the latest prostate cancer GWAS found that the OR for prostate cancer per SD unit increase in circulating IGFBP-3 was 1.12 (95% CI, 0.91–1.36; P = 0.18) from IVW analyses. The multivariable estimate, which took into account the effect of the SNPs on IGF-I, was of similar magnitude to the IVW analyses and gave stronger evidence with an OR of 1.14 (95% CI, 1.02–1.28; P = 0.02; Supplementary Table S1). The OR for advanced prostate cancer per SD unit increase in IGFBP-3 was 1.16 (95% CI, 0.87–1.55; P = 0.19) from IVW analyses. The multivariable estimate was of similar magnitude to the IVW analyses, but gave stronger evidence with an OR of 1.22 (95% CI, 1.07–1.40; P = 0.004) (Supplementary Table S1).

Mean circulating IGFBP-3 levels were higher in the cases compared with the controls in ProtecT (4634.77 ng/mL vs. 4502.19 ng/mL, P for difference = 0.002). Mean season-adjusted circulating 25(OH)D levels did not differ between cases and controls (22.75 ng/mL vs. 22.71 ng/mL, respectively, P for difference = 0.87). Mean season-adjusted 1,25(OH)₂D levels did not differ between cases and controls (40.71 pg/mL vs. 42.06 pg/mL, respectively, P for difference = 0.12; Table 1). For cases and controls combined, circulating 25(OH)D and season-adjusted 25(OH)D were positively correlated with 1,25(OH)₂D levels (r = 0.24; P < 0.001 and r = 0.22; P < 0.001, respectively; Supplementary Table S2).

For cases and controls combined, an SD unit increase (SD = 7.8 ng/mL) in circulating 25(OH)D was associated with a 0.09 SD (95% CI, 0.05–0.13; P < 0.001) increase in circulating IGFBP-3 levels. A 0.09 SD unit increase in IGFBP-3 is equivalent to 93.3 ng/mL. There was no strong evidence for difference between cases and controls (Table 2). We found evidence that circulating 25(OH)D and IGFBP-3 are associated with potential confounding factors including age, BMI, and diabetes status (Supplementary Table S3). After adjustment for confounders, the effect (β = 0.09; 95% CI, 0.04–0.14; P < 0.001) was similar in magnitude to the unadjusted analysis. When stratified by case–control status, the precision of the estimates was reduced due to smaller sample sizes (Table 2). We found little evidence for a nonlinear effect of vitamin D on IGFBP-3 (P = 0.43).

For cases and controls combined, an SD unit increase in circulating 1,25(OH)₂D was associated with a 0.09 SD (95% CI, 0.04–0.14; P = 0.001) increase in IGFBP-3 (Table 2). We found evidence for association between 1,25(OH)₂D and potential confounders including BMI, family history of prostate cancer, smoking, and diabetes status (Supplementary Table S3). After adjustment for potential confounders, the effect (β = 0.09; 95% CI, 0.03–0.15; P = 0.003) was similar in magnitude to the unadjusted analysis. When stratified by case–control status, the precision of the estimates was reduced due to smaller sample sizes (Table 2).

An MR assumption is that there is a reliable association between the genetic instrument and the exposure (48). We validated this using data from ProtecT in which 1,416 individuals had both genotype data and 25(OH)D measured. In cases and controls combined, per unit increase in the allele score was associated with a 0.11 SD unit increase in 25(OH)D. There was no strong evidence for difference between cases and controls. For an individual with the maximum number of 25(OH)D increasing alleles (12), this corresponds to a 1.32 SD unit difference (on average) in 25(OH)D when compared with an individual with the minimum (0) number of 25(OH)D increasing alleles. This is higher in magnitude and is comparable to that of vitamin D supplementation which increases 25(OH)D by 0.45 SD units (35). In cases and controls combined, the metabolism score was associated with increased 25(OH)D (P < 0.001). For the metabolism score, there was no strong evidence for difference between cases and controls. In cases and controls combined, the synthesis score was found to be associated with increased 25(OH)D; however, the precision of the estimates was reduced when stratified by case–control status due to reduced sample size (Supplementary Table S4).

Using the allele score, each additional 25(OH)D-related allele was associated with a 0.05 pg/mL increase in 1,25(OH)₂D in cases and controls combined. There was no strong evidence for difference between cases and controls. In cases and controls combined, the metabolism score was also found to be associated with increased 1,25(OH)₂D (P = 0.005); however, the precision of the estimates was reduced when stratified by case–control status due to reduced sample size (Supplementary Table S4).

Another MR assumption is that the genetic instrument should not be associated with confounders (measured or unmeasured) of the exposure–outcome relationship (48). There was little evidence that the allele, metabolism, or synthesis score were associated with potential confounders (Supplementary Table S5).

The causal effect of 25(OH)D on IGFBP-3 was estimated using one-sample MR (Fig. 2). Using the allele score as an instrument for 25(OH)D, for cases and controls combined, the instrumental variable (IV) estimate of the causal relationship between a one SD increase in 25(OH)D and IGFBP-3 for cases and controls combined was −0.32 SD units (95% CI, −0.64 to −0.01; P = 0.04). When stratified by case–control status, the IV estimate for the controls was −0.10 (95% CI, −0.62–0.41; P = 0.69) and the IV estimate for the cases was −0.30 (95% CI, −0.79 to −0.02; P = 0.04). Sensitivity analyses based on the use of individual SNPs, synthesis, and metabolism score showed little evidence that 25(OH)D was causally associated with IGFBP-3 (Table 3; Fig. 3). The first-stage F-statistic using the allele score and metabolism score was 46.76 and 41.12, respectively. There was evidence of a departure of instrumental variable-derived estimates from those derived from observational analyses (Durbin–Wu–Hausman test P = 0.007; Table 3).

Using the six 25(OH)D-associated SNPs for the two-sample MR analysis, the IVW estimate between a one log-unit increase in 25(OH)D and IGFBP-3 was 0.11 SD units (95% CI, −0.10–0.31; P = 0.32; Fig. 3). In leave-one-out analyses, sequentially omitting each of the six 25(OH)D-associated SNPs provided similar causal estimates to IVW analysis (Supplementary Table S6).

Sensitivity analyses using SNPs involved in 25(OH)D synthesis found that the IVW estimate between a one log-unit increase in 25(OH)D and IGFBP-3 was 0.15 SD units (95% CI, −0.30–0.61; P = 0.51). Using SNPs involved in 25(OH)D metabolism, the IVW estimate between a one log-unit increase in 25(OH)D and IGFBP-3 was 0.07 SD units (95% CI, −0.17–0.31; P = 0.59; Fig. 3).

In the reverse direction, we investigated the causal effect of IGFBP-3 on 25(OH)D using the two-sample MR. The IVW estimate between a one SD increase in IGFBP-3 and 25(OH)D was 0.01 log-units (95% CI, −0.003–0.03; P = 0.10; Supplementary Table S7).

This study supported results from previous studies that the IGF pathway is associated with increased odds of prostate cancer and of advanced prostate cancer—in particular, IGFBP-3 (4, 5). Existing preclinical research has implicated 1,25(OH)₂D in the regulation of IGFBP-3 expression and in this context we set out to assess the causal relationship between the closest modifiable proxy for this (25(OH)D) and IGFBP-3. Using data from ProtecT, this study found observational evidence that 25(OH)D and 1,25(OH)₂D are associated with IGFBP-3 in the circulation. However, MR analyses using data from ProtecT were not suggestive of a causal effect for 25(OH)D. Using summary statistics from the new IGFBP-3 GWAS, two-sample MR analyses found little evidence of a causal relationship between circulating 25(OH)D and IGFBP-3.

The relationship of vitamin D and IGFBP-3 is complex due to the coexisting endocrine and autocrine/paracrine interactions (6). At the autocrine/paracrine level, several preclinical studies have demonstrated that 1,25(OH)₂D regulates IGFBP-3 expression (6, 7) and that the growth inhibitory effects of 1,25(OH)₂D on prostate cancer cells is altered by IGFBP-3 (14–16). At the endocrine level, evidence from observational studies examining whether 25(OH)D or 1,25(OH)₂D influence IGFBP-3 in the circulation have been conflicting, but inferences of causality are limited with observational studies due to effects of confounding and reverse causation (17–20). Our MR results suggesting that 25(OH)D does not have a causal effect on IGFBP-3 levels in the circulation are in agreement with the results from a randomized phase II trial of men with localized prostate cancer and high-grade prostatic intraepithelial neoplasia which found that supplementation with the vitamin D analog,1-alpha-hydroxyvitamin D2, did not have an effect on circulating IGFBP-3 levels (49). Another vitamin D supplementation trial conducted among women with high risk of breast cancer also found little evidence that circulating IGFBP-3 levels changed after supplementation with 25(OH)D (50). This does not preclude that increasing the availability of 25(OH)D with supplementation may lead to increased production of 1,25(OH)₂D within tissues that could then stimulate the local production of IGFBP-3 that is not detected in the circulation.

Evidence from this study corroborates that the IGF pathway is a risk factor for prostate cancer risk and progression (3–5), but challenges assumptions about the possible route from supplement-derived precursor vitamin D, through IGFBP-3 to cancer. This work sits in the broader context of a prospective nested case–control study which found evidence for interactions between plasma 25(OH)D and plasma IGF-I/IGFBP-3 ratio with regard to colorectal cancer risk (20). Another study of premenopausal women also reported that vitamin D supplementation was negatively associated with mammographic breast density, and that the association was stronger in those with high IGF-I and IGFBP-3 levels (18).

There could be several reasons why our MR findings differ from results from the observational studies. First, the point estimates are close to the null for our two sample MR analyses. It is possible that the true effect of 25(OH)D on IGFBP-3 could be small and undetectable using a causal analysis framework. Second, the genetic instruments for 25(OH)D may not be acting as a true proxy for 1,25(OH)₂D. Our analyses using the allele and metabolism score found that each additional 25(OH)D-related allele was associated with increased 1,25(OH)₂D levels for controls and cases combined (Supplementary Table S4). However, the F-statistic for instrumental analyses using allele, metabolism, and synthesis scores was lower for 1,25(OH)₂D compared with 25(OH)D (Table 3), indicating that they are weaker instruments for 1,25(OH)₂D. Our observational analyses in ProtecT found that 1,25(OH)₂D is associated with IGFBP-3 in the circulation; however, it is still unclear if a causal relationship exists between circulating 1,25(OH)₂D and IGFBP-3. Thirdly, given that vitamin D activating enzymes are present in many tissues and could influence 1,25(OH)₂D concentrations within tissues, including prostate tissues (51, 52), circulating 25(OH)D and 1,25(OH)₂D may not reflect the true bioavailability of 25(OH)D and 1,25(OH)₂D within tissues. Therefore, the endocrine relationship between 25(OH)D and IGFBP-3 may not be reflective of the autocrine/paracrine relationship.

Our study has several limitations. The SNPs associated with IGFBP-3 have been shown to be associated with more than one part of the IGF pathway. Although the multivariable MR method enables the determination of the causal effect of IGFBP-3 on prostate cancer independent of IGF-I, we cannot rule out associations with other members of the IGF pathway (5). As only a small number of SNPs were included in our MR analyses, methods to test for pleiotropy such as the MR-Egger (43), weighted median (44), and mode (45) method were not conducted as these lack power with small number of SNPs. As the SNPs associated with 25(OH)D only explained a small proportion of the variability in circulating 25(OH)D, large sample sizes are required to detect expected influences on IGFBP-3 levels and to be able to provide precise estimates for this effect. Although our point estimates are close to the null for the two-sample MR analyses, we cannot rule out the possibility that 25(OH)D may have small effects on IGFBP-3. Finally, our study population was limited to Europeans. Although population homogeneity eliminates population admixture as a potential confounder in our analyses, as 25(OH)D levels vary with sun exposure and skin color, the findings drawn from this study might not be applicable to other ethnic groups or individuals in different geographical locations.

This study confirmed results from previous studies that members of the IGF pathway, in particular IGFBP-3, has a causal effect on prostate cancer and advanced prostate cancer. Observationally, there is evidence that circulating 25(OH)D is positively associated with circulating IGFBP-3, but MR analyses indicated these findings were unlikely to be causal. Findings here are limited by the nature of instrumentation of both 25(OH)D and IGFBP-3 and the utility of circulating measures, but are important as they suggest that circulating 25(OH)D are unlikely to be causally related to circulating IGFBP-3, although vitamin D supplementation could affect local tissue levels.

No potential conflicts of interest were disclosed.

Conception and design: V.Y. Tan, K.M. Biernacka, R.M. Martin, C.M. Perks, N.J. Timpson, J.M.P. Holly

Development of methodology: N.J. Timpson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.M. Biernacka, R. Gilbert, R.C. Kaplan, Q. Qibin, R.M. Martin, the PRACTICAL consortium

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V.Y. Tan, K.M. Biernacka, T. Dudding, C. Bonilla, Q. Qibin, A. Teumer, C.M. Perks, N.J. Timpson, J.M.P. Holly

Writing, review, and/or revision of the manuscript: V.Y. Tan, K.M. Biernacka, T. Dudding, C. Bonilla, R. Gilbert, R.C. Kaplan, Q. Qibin, A. Teumer, R.M. Martin, C.M. Perks, the PRACTICAL consortium, N.J. Timpson, J.M.P. Holly

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V.Y. Tan, R. Gilbert, Q. Qibin, the PRACTICAL consortium

Study supervision: R.M. Martin, C.M. Perks, N.J. Timpson, J.M.P. Holly

This work was supported by CRUK (The Integrative Cancer Epidemiology Programme, C18281/A19169). V.Y. Tan, T. Dudding, C. Bonilla, N.J. Timpson and R.M. Martin work in the Medical Research Council Integrative Epidemiology Unit at the University of Bristol which is supported by the Medical Research Council and the University of Bristol (grant number MC_UU_12013/3). N.J. Timpson is a Wellcome Trust Investigator (202802/Z/16/Z) and works within the University of Bristol NIHR Biomedical Research Centre (BRC). T. Dudding is funded by The Wellcome Trust (grant number 201237/Z/16/Z), K.M. Biernacka is funded by Cancer Research UK (C11046/A10052) and by the National Institute for Health Research (NIHR) Bristol Biomedical Research Centre based at University Hospitals Bristol NHS Foundation Trust and the University of Bristol. Q. Qibin is supported by a Scientist Development Award (K01HL129892) from the NHLBI. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.