Bidirectional Mendelian randomisation analysis of the relationship between circulating vitamin D concentration and colorectal cancer risk

Epidemiological evidence is consistent with a protective effect of vitamin D against colorectal cancer (CRC), but the observed strong associations are open to confounders and potential reverse causation. Previous Mendelian randomisation (MR) studies were limited by poor genetic instruments and inadequate statistical power. Moreover, whether genetically higher CRC risk can influence vitamin D level, namely the reverse causation, still remains unknown. Herein, we report the first bidirectional MR study. We employed 110 newly identified genetic variants as proxies for vitamin D to obtain unconfounded effect estimates on CRC risk in 26 397 CRC cases and 41 481 controls of European ancestry. To test for reserve causation, we estimated effects of 115 CRC‐risk variants on vitamin D level among 417 580 participants from the UK Biobank. The causal association was estimated using the random‐effect inverse‐variance weighted (IVW) method. We found no significant causal effect of vitamin D on CRC risk [IVW estimate odds ratio: 0.97, 95% confidence interval (CI) = 0.88‐1.07, P = .565]. Similarly, no significant reverse causal association was identified between genetically increased CRC risk and vitamin D levels (IVW estimate β: −0.002, 95% CI = −0.008 to 0.004, P = .543). Stratified analysis by tumour sites did not identify significant causal associations in either direction between vitamin D and colon or rectal cancer. Despite the improved statistical power of this study, we found no evidence of causal association of either direction between circulating vitamin D and CRC risk. Significant associations reported by observational studies may be primarily driven by unidentified confounders.

(MR) studies were limited by poor genetic instruments and inadequate statistical power. Moreover, whether genetically higher CRC risk can influence vitamin D level, namely the reverse causation, still remains unknown. Herein, we report the first bidirectional MR study. We employed 110 newly identified genetic variants as proxies for vitamin D to obtain unconfounded effect estimates on CRC risk in 26 397 CRC cases and 41 481 controls of European ancestry. To test for reserve causation, we estimated effects of 115 CRC-risk variants on vitamin D level among 417 580 participants from the UK Biobank. The causal association was estimated using the random-effect inverse-variance weighted (IVW) method. We found no significant causal effect of vitamin D on CRC risk [IVW estimate odds ratio: 0.97, 95% confidence interval (CI) = 0.88-1.07, P = .565]. Similarly, no significant reverse causal association was identified between genetically increased CRC risk and vitamin D levels (IVW estimate β: À0.002, 95% CI = À0.008 to 0.004, P = .543). Stratified analysis by tumour sites did not identify significant causal associations in either direction between vitamin D and colon or rectal cancer. Despite the improved statistical power of this study, we found no evidence of causal association of either direction between circulating vitamin D and CRC risk.

What's new?
Vitamin D is associated with outcomes of colorectal cancer (CRC), but a causal relationship hasn't been established. Previous studies could not rule out that predisposition to CRC reduces circulating vitamin D. Here, the authors report results from a bi-directional Mendelian randomization. Using 110 variants and 26,397 patients allowed for a statistically powerful analysis. They found no causal relationship in either direction: low vitamin D did not increase CRC risk, and elevated CRC risk did not reduce vitamin D levels. Previously observed associations, they suggest, may be driven by unknown confounders.

| INTRODUCTION
Population-based observational studies consistently demonstrate that lower circulating vitamin D levels are associated with a higher risk of colorectal cancer (CRC). 1 There has been increasing interest in establishing whether this relationship is causal, because it would provide the rationale for exploring dietary vitamin D supplements for CRC prevention. All randomised trials to date, such as the VITAL trial, 2 have been underpowered to detect effects on CRC risk given prohibitive cost and time to observe the required number of incident CRC cases. Mendelian randomisation (MR) is an alternative approach to investigate causality. 3 However, previous two MR studies 4,5 on vitamin D and CRC risk employed no more than six known variants associated with vitamin D, and therefore were hampered by poor genetic instruments (2.84% of vitamin D variance explained) which resulted in inadequate statistical power to detect small to modest causal effects. A recent large genome-wide association study (GWAS) including more than 400 000 Europeans from the UK Biobank cohort identified 143 independent vitamin D-related loci that could explain 5.7% to 10.5% of vitamin D variance, 6 allowing the development of a significantly improved genetic instrument and therefore, a better powered MR study. In addition, previous MR studies only investigated the vitamin D-CRC association, and the null findings therefore could not exclude reverse causation, namely that genetic predisposition to CRC could possibly cause decreased circulating vitamin D concentration. Here, we report a bidirectional MR using much improved genetic instruments to explore whether the observed associations between vitamin D and CRC risk are causal.

| MATERIALS AND METHODS
We adopted a two-sample MR approach to estimate causal effects in both directions. To analyse the causal effect of vitamin D on CRC risk, we created a genetic instrument using common variants (minor allele frequency > 5%) identified from a recent GWAS on circulating vitamin D concentration (P < 5 Â 10 À8 ) including 417 580 Europeans from the UK Biobank cohort. 6 Effect estimates of these variants along with standard errors (SEs) were extracted from the same study. We then assessed associations between these vitamin D variants and CRC risk by conducting a meta-analysis of effect estimates from 14 GWASs on We calculated the R 2 statistic to evaluate linkage disequilibrium (LD) among genetic variants. For any pair of variants in LD (R 2 > 0.2), we included the one with the smallest P value in relation to the exposure. In an attempt to control for horizontal pleiotropy, we excluded outlier variants using the MR-Pleiotropy Residual Sum and Outlier (MR-PRESSO) approach. 11 The strength of instruments was assessed by calculating F-statistics (F < 10 was deemed as a weak instrument).
Causal effect estimates were generated using the inversevariance weighted (IVW) method, which assumes all genetic variants as valid instruments. 12 We also conducted additional analyses  To evaluate any reverse causation effects, a total of 115 variants associated with CRC risk were used as instruments. Effects of these variants on vitamin D concentration were extracted from the previous GWAS and can be found in the Table S2. We found that genetically higher risk of CRC was not significantly associated with the vitamin D concentration of 417 580 participants from the UK Biobank cohort using the IVW estimator (β: À0.002, 95% CI = À0.008 to 0.004, P = .543, scatter plot in Figure 1B). Similarly, null causal effects of CRC risk on vitamin D concentration were observed, after Bonferroni correction, using either the median-based estimator (β: 0.001, 95% CI = À0.008 to 0.009, P = .850) or the MR-Egger method (β: 0.016, 95% CI = 0.001-0.030, P = .039). MR-Egger analysis revealed suggestive evidence for horizontal pleiotropic effects for the variants used as instruments (P = .011).
After controlling for possible pleiotropic effects of BMI, our multivariable MR analysis found no significant causal associations between vitamin D concentration and CRC risk in either direction (Table 1).
Regarding stratified analysis, we failed to observe any significant causal associations between vitamin D and colon or rectal cancer risk (Tables 1, S3 and S4). A total of 38 variants associated with colon cancer and 26 variants associated with rectal cancer were included as instruments to investigate site-specific reverse causation (Table S5), and we did not observe any significant causal effects of genetically higher risk of colon or rectal cancer on vitamin D level (Table 1).

| DISCUSSION
Since circulating vitamin D level is readily modifiable, there has been growing interest in proving causality in the association between vitamin D and CRC risk. Our study is an advance on prior knowledge in the field, where we created a markedly more extensive genetic instrument compared to our previous MRs (110 vs 6 variants). The generation of these genetic instruments, and the largest sample size to date (26 397 cases), has enabled us to conduct statistically powerful analyses. Our findings provide robust evidence that even small-to-modest causal effects of vitamin D on CRC risk are unlikely.
The strength of the genetic instrumental variable and our recent GWAS on CRC risk has also allowed us to perform the first ever bidi- In addition to enhanced power, strengths of this study also include that we adopted multiple MR methods such as MR-PRESSO and MR-Egger to detect possible pleiotropic effects of included genetic instruments, which could potentially violate basic MR assumptions. 21 Large sample size, aligned with data granularity, combined with the power of the genetic instruments that we employed, also enabled us to explore the subsidiary aim of understanding the contribution of BMI as a potential confounding, but modifiable effect. After controlling for BMI, our multivariable MR analysis found no significant impact of BMI on vitamin D-CRC risk association in either direction. To explore the possibility that combining colon and rectal cancer may underestimate the association between vitamin D and a specific cancer location, we conducted a stratified analysis using colon and rectal cancer specific instruments and reran the bidirectional MR analysis. This analysis also confirmed lack of association between vitamin D and either colon or rectal cancer, in either direction. The major limitation of this study is the limited access to individual-level data to aggregate a large enough sample size for further stratified analysis.
In conclusion, this is the first study to present bidirectional assessment of the association between vitamin D and CRC risk using MR. Our findings add to current knowledge by concluding that the observed associations between vitamin D and CRC risk are most likely driven by unknown confounders instead of possible reverse causation or small to modest effects that failed to have been detected by previous MR studies with limited statistical power. Future efforts may focus more on circulating vitamin D level as a predictive biomarker instead of a therapeutic target for CRC prevention.

CONFLICT OF INTEREST
The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT
The individual-level data sets used and/or analysed during the current study are available from the corresponding author on reasonable request. All summary statistics used in this study can be found in Supporting Information.

ETHICS STATEMENT
Study cohorts used were approved by the ethical review board at respective study centres in accordance with the tenets of the Declara-