A Modest but Durable Vitamin D Effect on MS Risk

Observational studies have linked low serum 25-hydroxyvitamin D to multiple sclerosis (MS) for decades, but the association has never settled the question of direction. MS could be lowering vitamin D rather than the other way round, and sun exposure, obesity and socioeconomic position could be producing both. Rhead and colleagues addressed that with Mendelian randomization in two independent populations: 1,056 MS cases against 9,015 controls among white non-Hispanic members of Kaiser Permanente Northern California, and 6,335 cases against 5,762 controls in Sweden, for 7,391 and 14,777 in the pooled analysis. The instrument is three SNPs associated with serum 25-hydroxyvitamin D at genome-wide significance, sitting in GC, CYP2R1 and the NADSYN1/DHCR7 region. What separates this from most Mendelian randomization is that they had individual-level data, which allows adjustments that summary statistics cannot support.

The Result, and What One Unit Means
In California the odds ratio was 0.79 (95% CI 0.64–0.99, P = 0.04) and in Sweden 0.86 (0.76–0.98, P = 0.03), with the meta-analysis at 0.85 (0.76–0.94, P = 0.003) and no heterogeneity between them (I² = 0.0%, P = 0.51). Two populations from regions with different sunlight exposure agreeing is a genuine strength. The scale needs unpacking, though, because the odds ratio is per unit of the instrument, which is a weighted allele count rather than anything measured in a patient. Working from the allele frequencies and weights in their Table 1, the instrument runs from 0 to 1.62 with an expected mean of 1.059, which matches the 1.026 to 1.056 they report, and a standard deviation of about 0.31. A one-unit increase therefore spans roughly three standard deviations, and the effect per standard deviation works out at about 5% lower odds. The finding is real and it is modest, and the distinction matters for anyone reading 0.85 as the effect of taking a supplement.

The Adjustment Set Is What Separates This From Earlier Attempts
Models controlled for sex, year of birth, ever smoking, college graduation, body mass index at 18 to 20 years or in the 20s, genetic ancestry, HLA-DRB1*15:01 carrier status, and a weighted genetic risk score built from 110 non-HLA MS risk variants, with the Swedish analyses adding region of residence and study type. The last two covariates are the ones that matter most, because including both HLA status and a polygenic score of known MS variants demonstrates that the instrument's effect does not run through MS genetic background. The adjustment was doing real work: cases and controls differed on nearly everything measured, with ever smoking at 50% against 32% in California and HLA-DRB1*15:01 carriage at 53% against 27%. The authors report that adjustment lowered both P values and odds ratios relative to the unadjusted models.

They Tested the Assumptions Rather Than Asserting Them
Mendelian randomization rests on three assumptions, and this study checked the two that can be checked. For independence from confounders, they tested the instrument against year of birth, sex, smoking, college graduation, obesity, HLA-DRB1*15:01 status, the polygenic risk score, and study type, finding no association with any, and they disclose the single exception rather than omitting it: in the Swedish population the instrument was associated with two of six geographic regions. For the link between instrument and exposure, they returned to 2,077 Swedish participants whose serum had been collected at enrolment and showed the instrument predicted measured 25-hydroxyvitamin D at P < 0.0001 among both cases and controls. That step confirms the instrument works in the population being studied rather than only in the genome-wide study it was borrowed from.

Running Each SNP Separately Reveals Something
Splitting the instrument into its three SNPs and running each as a separate instrument, all six estimates pointed the same direction, ranging from 0.92 to 0.94 in California and 0.84 to 0.99 in Sweden, which is what this test exists to show. Two details deserve carrying forward. None of the three SNPs reached significance on its own in the California cohort, at P = 0.11, 0.26 and 0.27, so the combined instrument is carrying the result there. And rs2282679 in GC, which holds the largest weight in the instrument at β = 0.38, performs worst in both populations, at P = 0.26 and P = 0.80. The variant contributing most to the score contributes least to the effect.

The Negative Results, and the Pleiotropy the Authors Name
The instrument showed no association with the Multiple Sclerosis Severity Score or with age at onset in either population. That draws a boundary worth keeping: this supports vitamin D acting on whether MS develops, not on how it behaves afterwards, and the two questions are often discussed as if one answer covered both. The authors also name their own pleiotropy risk rather than gesturing at it. rs3829251 sits upstream of DHCR7, which functions in cholesterol synthesis, and cholesterol metabolism has itself been associated with MS, so pleiotropy cannot be excluded for that variant. They raise canalization as well, meaning the buffering of genetic variants against changes in 25-hydroxyvitamin D, noting that if it operates the association would be biased toward the null rather than away from it.

What the Design Cannot Reach
The study covers non-Hispanic whites only, which the authors flag as particularly relevant given evidence that 25-hydroxyvitamin D may not be associated with MS in Hispanic populations. The analysis assumes a linear relationship and additive effects of the three alleles on 25-hydroxyvitamin D. The severity analysis used a cross-sectional score and could not account for disease-modifying therapy. The third Mendelian randomization assumption, that the instrument affects the outcome only through the exposure, stays not fully testable, as they state directly. And the overlap with the earlier summary-level Mendelian randomization deserves noting: 2,812 Swedish participants appear in both studies, so the two are not fully independent confirmations of one another. What this paper establishes is narrower and firmer than its title suggests. Among white non-Hispanic people in two regions with different sunlight, genetically predicted lower 25-hydroxyvitamin D raises MS odds by an amount that survives adjustment for smoking, obesity, schooling, HLA type and polygenic MS risk. Whether raising an individual's vitamin D lowers their risk by a useful amount is the next question, and a different one.

Disclaimer: This blog post is based on the cited research article 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:
Rhead, B., Bäärnhielm, M., Gianfrancesco, M., Mok, A., Shao, X., Quach, H., Shen, L., Schaefer, C., Link, J., Gyllenberg, A., Hedström, A. K., Olsson, T., Hillert, J., Kockum, I., Glymour, M. M., Alfredsson, L., & Barcellos, L. F. (2016). Mendelian randomization shows a causal effect of low vitamin D on multiple sclerosis risk. Neurology: Genetics, 2(5), e97. https://doi.org/10.1212/NXG.0000000000000097