Vitamin D: marker, cause or consequence of depression? An exploration using genomics

Background Observational studies suggest an association between circulating vitamin D and depression. Trials testing the effect of vitamin D supplementation on depression reported inconclusive findings. It remains unknown whether the vitamin D-depression association stems from shared etiology or from a direct causal relationship. We explored the nature of the association between 25-hydroxyvitamin D (25-OH-D) and major depressive disorder (MDD) exploiting data and statistical tools from genomics. Methods Results from the two largest GWAS on 25-OH-D (79,366 samples) and major depressive disorder (MDD; 135,458 cases and 344,901 controls) were applied to individual-level data (>2,000 subjects with measures of genotype, circulating 25-OH-D and DSM-IV lifetime MDD) and summary-level data analyses. A genetic association between 25-OH-D and MDD was tested by polygenic risk scores (PRS) and by estimating genetic correlation between traits. Two-sample Mendelian Randomization (2SMR) analyses tested the potential bidirectional causality between 25-OH-D and depression. Results In individual-level data, the 25-OH-D PRS was associated (p=1.4e−20) with 25-OH-D level, but not with lifetime MDD. Conversely, the MDD PRS was associated with MDD (p=2.3e−5), but not with 25-OH-D. In summary-level data analyses, the rg between the traits was low and not significant (−0.06, p=0.11). 2SMR analyses provided no evidence of a significant causal role of 25-OH-D for MD and vice versa. Conclusions The use of genomics tools indicated that shared etiology or direct causality between vitamin D concentrations and depression is unlikely: vitamin D may represent a marker rather than a cause, or consequence, of depression.


INTRODUCTION
Meta-analyses of large observational studies provided evidence of cross-sectional and longitudinal associations between circulating vitamin D and depression. 1 In >2,300 psychiatrically well-characterized participants from the Netherlands Study of Depression and Anxiety (NESDA), we previously showed 2 that concentrations of 25-hydroxyvitamin D (25-OH-D, the body reserve of vitamin D) were lower in patients with remitted and current depressive disorders as compared to healthy controls. Furthermore, in patients with current depression low 25-OH-D was associated with symptoms severity and chronicity of depressive disorders after 2 years. It is not trivial to disentangle whether this observational correlation stems from shared underlying aetiological roots or a direct causal relationship between vitamin D and depression. Alternatively, as highlighted by Ioannidis, 3 observational associations emerging in nutritional epidemiology may be the product of unresolved confounding. Nevertheless, several hypotheses based on preclinical data emerged about the potential impact of vitamin D on brain structure (vitamin D receptor is expressed in prefrontal cortex, amygdala and hippocampus) and pathophysiological processes relevant for mood. 4 Meta-analyses 5-7 pooling results from trials testing the effect of vitamin D supplementation on depression reported substantially inconclusive findings.
To move forward it is crucial to obtain a clearer knowledge of the exact nature of the vitamin D-depression relationship. Do both traits emerge from a common genetic liability? Does vitamin D truly exert a potential causal effect on depression, or is the effect in the reverse direction?
Genetic research has made enormous progress over the last years, and now provide unique opportunities to investigate shared risk and causality between traits applying new statistical tools and results from genome-wide association studies (GWAS). 8 In the latest GWAS 9 on 25-OH-D serum concentrations, the estimate of genetic correlation with depressive symptoms was not-significant, indicating lack of shared genetic risk. Nevertheless this result may have been affected by the different genetic architectures of depression (highly polygenic with many genetic variants with small effects) and 25-OH-D (few genetic loci with relatively larger effect sizes), or by the use of data for depression from GWAS 10 relatively underpowered as compared to those currently available.
In the present study, we leveraged on results from the two largest GWAS on 25-OH-D (SUNLIGHT consortium: 79,366 samples) 9 and MDD (Psychiatric Genomics Consortium (PGC): 135,458 cases and 344,901 controls) 11 . We estimated the degree of genetic overlap between the two traits using both individual-level data (in NESDA 2 ) and summary-level data analyses. Using summary data, Mendelian randomization analyses exploring the potential bidirectional causality between 25-OH-D and depression were also performed.

METHODS
All methods reported hereby are extensively described in eMaterials. Individual-level data analyses were based on up to 2,376 NESDA participants with measures of genotype, circulating 25-OH-D and DSM-IV lifetime diagnoses of MDD (1,750 cases). Polygenic risk scores (PRS) for 25-OH-D and MDD were built based on GWAS summary statistics from, respectively, SUNLIGHT consortium (79,366 samples) 9 and PGC (135,458 cases and 344,901 controls). 11 The PRS for increased 25-OH-D included the six independent genomewide significant SNPs, jointly explaining an appreciable proportion (38%) of SNPheritability in the discovery GWAS. 9 PRS for MDD was built based on the full polygenic signal from the PGC GWAS using LDpred method. 12 Same-and cross-trait associations of PRS with 25-OH-D concentrations and lifetime MDD diagnosis were estimated using regression models adjusted for sex and 10 ancestry-informative genetic principal components. The proportion of variance explained by GPRS was additionally estimated. In additional analyses focusing on MDD cases, the association between the two PRS and symptom severity measured by Inventory of Depressive Symptoms (IDS-SR 30 ) 13 was also estimated adjusting for sex and principal components.
Finally, summary-level data analyses were run using results from the GWAS of the two traits. Firstly, LD-score regression 14 was applied to GWAS summary statistics to estimate genome-wide genetic correlation between 25-OH-D and MD. Then, the six genome-wide significant SNPs for increased 25-OH-D were used as instrument in two-sample MR (2SMR) 15 testing the potential causal role of vitamin D on MDD using inverse-variance weighted fixed effects-meta-analysis (sensitivity analyses included median-and mode-mode base estimators). Opposite MR analyses testing the causal role of depression for 25-OH-D levels were additionally run using 37 independent genome-wide significant SNPs selected form the discovery GWAS 11 as instrument for MDD.

RESULTS
In analyses based on NESDA data, each SD increase in 25-OH-D levels was associated with lower odds of lifetime MDD (sex-adjusted OR=0.77,95%CIs=0.70-0.87, p=5.8e-6). The PRS for increased 25-OH-D was strongly associated (per SD increase: b=5.28, p=1.4e-20; full results in Table 1)  These results were complemented with summary-level data analyses. The estimate of genome-wide genetic correlation between 25-OH-D levels and MD was not significantly different from 0 (rg=-0.06,se= 0.04,p=0.11). Table 2

DISCUSSION
We examined the nature of the association between vitamin D and depression using data and statistical tools from genomics. Evidence of a common genetic base between the two traits, emerging from shared etiology or through a direct causal mechanism, that could potentially explain their phenotypic association was not found. Even in data from the NESDA cohort,  Results from linear (outcome: 25-OH-D) and binary logistic (outcome: MDD) regression analyses adjusted for sex and ten ancestry-informative genetic principal components. Unit of measure 25(OH)D: nmol/l (multiply by 0.4 to obtain ng/ml). 18 -leave-one-out SNP IVW analyses (eFigure1b and eFigure2b) indicated that the estimates were substantially not influenced by individual SNPs a eTable2 shows the estimated power for each of the 6 vitamin D-associated SNP to be found significantly (a=0.05) associated with MDD in PGC GWAS assuming a true causal link between 25-OH-D and MDD. None of the 6 SNPs, for which adequate statistical power was established, was found to be associated with MDD, additionally confirming that that the phenotypic link between Vitamin-D and MDD is not attributable to a causal effect of Vitamin-D on MDD liability.   Table 1