Where the Vitamin D Benefit Stops in MS Risk
The vitamin D literature in multiple sclerosis (MS) has a dose problem rather than a direction problem. Observational work and Mendelian randomization both point toward low 25-hydroxyvitamin D raising risk, but supplementation trials have been inconsistent, the regimens tested range from high daily doses to prolonged courses, and some evidence suggests higher doses may lead to worse outcomes. Yu and colleagues approached this by stacking four layers of evidence on the same question: a prospective cohort to find the shape of the relationship, linkage disequilibrium score regression for shared genetic architecture, pleiotropy and colocalization analysis for shared loci, and bidirectional Mendelian randomization for direction of effect. The cohort layer runs on 413,594 UK Biobank participants, among whom 861 developed MS during follow-up.
The Cohort, and the Shape It Found
The selection chain is worth following because it determines what the curve describes. From 502,419 total participants, 488,181 had genetic data, and successive exclusions removed those of non-European ancestry, those missing a vitamin D measurement, those below the 1st or above the 99th percentile of values, and 839 people who already had MS at baseline, leaving 413,594. Cases were defined by ICD-10 code G35, and participants were followed from baseline until first MS diagnosis, death, loss to follow-up, or the end of 2021, with anyone diagnosed within six months of baseline dropped so that existing disease was not counted as new. Cox models ran in three nested adjustments, the last adding season of blood sampling and physical activity to age, sex, deprivation index, type 2 diabetes, cancer, osteoporosis, monocyte count and leukocyte count. Lower 25OHD tracked higher MS risk in all three. Restricted cubic splines then showed the relationship is not a straight line, with a p-value for nonlinearity of 0.024 across the cohort, and an L-shaped curve whose inflection point sits at 47.9 nmol/L. Below that level, falling 25OHD tracks rising MS risk; above it, the correlation is no longer significant.
Where the Sexes Diverge
The nonlinearity held in women at a p-value of 0.044 and did not reach significance in men at 0.3445. A sensitivity analysis that kept the extreme values left the association significant in women and not in men. The interaction term between sex and 25OHD came in at 0.052, just above the conventional cutoff, and the authors treat it as a trend rather than a result, pointing out that the subgroup analyses showed the inverse association running consistently in both sexes. The biology they propose for the divergence is specific: estrogen has been shown to enhance vitamin D receptor expression, which could strengthen the immune-regulatory properties of 25OHD in women, so women might reach optimal levels faster on supplementation while also being more exposed to over-supplementation, where excessive levels may paradoxically raise risk.
Shared Genetic Architecture
The genetic layers drew on a 25OHD GWAS of 401,460 white British participants, with SNP heritability for the trait estimated at 16.1%, and an MS GWAS of 76,755 patients and 243,649 controls, the two chosen from completely different cohorts to avoid false positives from sample overlap. Score regression returned a genetic correlation of −0.096 (P = 2.98×10⁻⁸) without the intercept term and −0.1379 (P = 1.86×10⁻⁸) with it, the intercept of 0.0134 included to absorb any residual overlap between the datasets. Cross-trait pleiotropy analysis then found 24 independent loci shared between the two traits, with no genomic inflation in the quantile-quantile plot. Four of the 24 carried strong colocalization evidence, meaning the same causal variant is likely to drive both signals: 6p24.3 near RREB1 at a posterior probability of 0.967, 19p13.2 near SMARCA4 and LDLR at 0.956, 6p22.2 near SCGN at 0.954, and 12q14.1 near CDK4 at 0.891.
What the Shared Genes Point At
Three gene-mapping approaches ran in parallel rather than in sequence, which makes their overlap informative. Lead SNPs mapped to 31 nearby genes, gene-based testing returned 70 pleiotropic genes, and expression quantitative trait loci in whole blood, EBV-transformed lymphocytes and spleen returned 129. Only three genes came through all three routes: HORMAD1, PDE4C and TRMT61A. The pathway picture was more uniform than the gene lists. All ten of the top enriched gene sets concerned lipid metabolism, led by the metabolic pathways of LDL, HDL and triglycerides, statin inhibition of cholesterol production, and high-density lipoprotein particle remodelling. Tissue enrichment put the top three at EBV-transformed lymphocytes, spleen and terminal ileum, stratified score regression found enrichment in B lymphocytes and T cell subsets, and single-cell gene-set analysis put the shared signal most strongly in CD4-positive alpha beta T cells, followed by memory B cells and mucosal invariant T cells.
The Causal Layer and a Mediation Estimate
Bidirectional two-sample Mendelian randomization used instruments at genome-wide significance with F-statistics checked for strength. Heterogeneity was present, so random-effects inverse variance weighting served as the primary result: an odds ratio of 0.733 (95% CI 0.622–0.864, P = 2.07×10⁻⁴), meaning higher genetically predicted 25OHD lowers MS risk. Seven methods agreed on both direction and rough magnitude, from MR-Egger at 0.774 to weighted median at 0.785, and the MR-Egger intercept returned P = 0.508, which excludes directional pleiotropy. The reverse direction was null throughout, with every method sitting between 0.999 and 1.012 and no p-value below 0.14, so MS does not appear to change 25OHD. Multivariable analysis adjusting for blood biochemistry measures, diabetes and body mass index left the association intact. Going further, the authors estimated causal effects of 25OHD on 280 immune cell phenotypes, then tested those cells against MS while adjusting for 25OHD, and found specific immune cell populations partially mediating the effect with mediation proportions of roughly 32% to 45%.
The Tension the Authors Name, and Their Limits
The two halves of the paper answer different questions, and the authors say why. Mendelian randomization assumes a linear relationship between exposure and outcome, and they note that nonlinear MR has been shown to be unreasonable, so the causal relationship cannot be inferred under nonlinear scenarios. The MR therefore supports benefit on average while the cohort shows where that benefit stops, and neither can check the other. Their reading of the trial literature runs with the cohort: a meta-study suggests supplementation may have a therapeutic effect but the appropriate dose is uncertain and higher doses may lead to worse outcomes, which they find consistent with an L-shaped association in which very high concentrations do not reduce risk and may slightly raise it. Among the genes with high colocalization evidence, they single out SMARCA4, which has existing reports linking it to 25OHD and has been identified as potentially regulating sex-specific immune responses and lipid metabolism in MS, as a candidate target deserving experimental investigation, while noting that RREB1 and SCGN have no direct link to either trait yet. Four limitations close the paper. The observational nature of genetic associations cannot definitively prove causation and may overlook confounding factors. The results come from a European ancestry cohort and need validation across diverse ethnicities. Undetected pleiotropic loci or pathogenic variants could muddy the interpretation. And the sex interaction, at a p-value of 0.052, needs larger samples before anyone can say whether sex modifies the effect of 25OHD on MS risk.
Disclaimer: This blog post is based on the cited study and is intended for informational purposes only. It is not intended to provide medical advice. Please consult with a healthcare professional for any health concerns.
Reference:
Yu, X.-H., Lu, H.-M., Li, J., Su, M.-Z., Li, X.-M., & Jin, Y. (2024). Association between 25(OH) vitamin D and multiple sclerosis: cohort, shared genetics, and causality. Nutrition Journal, 23, 151. https://doi.org/10.1186/s12937-024-01059-4
