Metabolites and Multiple Sclerosis: What Genetic Studies Agree and Disagree On

Genome-wide association studies have mapped more than 200 risk variants for multiple sclerosis (MS), and almost all of them sit in immune genes. Metabolism is missing from that map, which leaves an open question: do circulating metabolites sit upstream of MS, or do they shift only after the disease begins? Four genetic studies published between 2023 and 2025 attack the question with Mendelian randomization instead of patient sampling, and they disagree in useful ways. Ge and colleagues screened 571 circulating metabolites against the International Multiple Sclerosis Genetics Consortium (IMSGC) dataset. Song and colleagues screened 1,400 serum metabolites against five autoimmune diseases. Gilchrist and colleagues tested roughly 1,000 plasma metabolites and 300 metabolite ratios across four psychiatric and four neurodegenerative disorders, then filtered the survivors through colocalisation and polygenic scoring. Zhang and colleagues took the narrow route and tested vitamin D metabolite subtypes and calcium. Taken together they narrow a thousand candidate molecules down to a handful, and one of the survivors fails on a test the other papers never ran.

What Mendelian Randomization Settles, and What It Leaves Open
Mendelian randomization uses genetic variants as instruments for an exposure. Because alleles are assigned at conception, the design is far less vulnerable to reverse causation than measuring metabolites in patients who already have MS and are already on treatment. That matters here, because disease-modifying therapies alter lipid and amino acid profiles directly. The design has a specific weakness worth tracking across these four papers: they are not statistically independent. Ge and Zhang both used the same IMSGC outcome data, 14,802 cases and 26,703 controls, and Gilchrist drew on the same well-powered MS GWAS. Agreement between them is therefore weaker evidence than it looks, because the outcome side of the equation is shared. Where the studies differ is on the exposure side, in which metabolite panel they used, how strict the instrument threshold was, and whether they checked for confounding by linkage disequilibrium. Those choices drive most of the disagreement below.

Amino Acids, Ketone Bodies, and a Lipid Result That Flips With Particle Size
Ge and colleagues found 29 metabolites with suggestive causal associations. Genetically predicted serine raised MS odds by 56% (OR 1.56, 95% CI 1.25-1.95), lysine by 18% (OR 1.18, 95% CI 1.01-1.38). The two ketone bodies more than doubled the estimate: acetone at OR 2.45 (95% CI 1.02-5.90) and acetoacetate at OR 2.47 (95% CI 1.14-5.34), though both confidence intervals nearly touch 1.0 and rest on few instruments. The lipoprotein results split by particle. Total cholesterol and phospholipids in large VLDL particles were associated with lower MS risk (OR 0.83 and OR 0.80). The same two lipid measures in very large HDL particles ran the other way (OR 1.20 and OR 1.13). The molecule is identical; the particle carrying it determines the sign of the effect. Any study reporting "cholesterol" as a single number is averaging two opposing signals into noise.

Pyrimidine Metabolism Turns Up Twice, From Different Panels
Song and colleagues applied false discovery rate correction to 1,400 serum metabolites and kept three signals for MS, all of them nucleotide-related. 5-Methyluridine, also called ribothymidine, carried OR 1.191 (95% CI 1.086-1.307). 2'-Deoxyuridine carried OR 1.337 (95% CI 1.127-1.586). The ratio of S-adenosylhomocysteine to 5-methyluridine ran protective at OR 0.771 (95% CI 0.649-0.916). Cochran's Q and MR-Egger detected neither heterogeneity nor horizontal pleiotropy. A different metabolite panel produced the same theme: Gilchrist and colleagues recovered risk-increasing effects for both uridine and dihydroorotate on MS. Dihydroorotate sits directly upstream of uridine in pyrimidine biosynthesis, and it is the substrate of dihydroorotate dehydrogenase, the enzyme that teriflunomide inhibits in licensed MS treatment. Two independent panels converging on a pathway that an approved drug already targets is a stronger result than either paper alone.

Colocalisation Separates a Shared Cause From Linkage Disequilibrium
Gilchrist and colleagues ran 10,327 causal tests and let almost none through unchallenged. Of 138 associations surviving FDR correction, 85 passed sensitivity analysis, and once effects driven by single influential variants were removed, 41 remained, seven of them for MS. The largest positive effect anywhere in the study was a plasmalogen lipid on MS: 1-(1-enyl-oleoyl)-GPE (P-18:1) at OR 1.48 (95% CI 1.33-1.65). Adipoylcarnitine ran protective, consistent with lower serum carnitine tracking with fatigue in patients with MS. Then colocalisation reversed one of the findings above. At the chromosome 22 locus containing the 5-methyluridine instruments, the test returned PP.H3 = 0.63, suggestive evidence for two distinct causal variants rather than one shared between metabolite and disease. That pattern means the MR estimate can arise from linkage disequilibrium, violating the no-horizontal-pleiotropy assumption. Song's headline MS metabolite passed every conventional sensitivity check and still failed this one. The same analysis showed 29 lipid associations clustering at the FADS gene locus on chromosome 11, all of them with psychiatric rather than neurological outcomes.

Vitamin D Metabolites Lower Risk; Calcium Yields No Usable Estimate
Most MR work on vitamin D uses one number, total 25-hydroxyvitamin D. Zhang and colleagues broke it into subtypes and added calcium as a comparison. Genetically higher total 25(OH)D was associated with reduced MS risk (OR 0.81, 95% CI 0.70-0.94, P = 4.0e-03, n = 417,580). The subtypes tracked with it: 25(OH)D3 at OR 0.85 (95% CI 0.76-0.95) and C3-epi-25(OH)D3 at OR 0.85 (95% CI 0.74-0.98), each from GWAS data on 40,562 participants. Calcium produced OR 2.85 with a confidence interval of 0.42 to 19.53 (P = 0.285) despite an exposure sample of 305,349. An interval that wide rules nothing in or out, and calling it a null result overstates what the data support. The reasonable reading is that the vitamin D estimates are precise enough to act on and the calcium estimate is not informative at any sample size these instruments can reach.

Where These Four Studies Are Weakest
All four analyses used European-ancestry data only, so none of the effect sizes transfer to other populations without testing. Instrument thresholds were often lenient: 62% of the associations surviving Gilchrist's sensitivity filters came from instruments clumped at P < 5e-06 rather than the genome-wide 5e-08, which raises the chance of pleiotropic instruments slipping through. Polygenic scores built from the prioritised metabolites showed negligible predictive performance in the UK Biobank, so none of these molecules currently works as a risk marker in an individual patient. No metabolite in any of the four studies has been tested functionally in vitro or in vivo. Two concrete steps follow. Colocalisation should become a required filter before any metabolite is described as causal for MS, since it removed a signal that standard sensitivity analyses cleared. And the pyrimidine finding is testable now with measured metabolite panels rather than genetic proxies, in patients on and off dihydroorotate dehydrogenase inhibition.

Disclaimer: This blog post is based on the cited research articles 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.

References:
Ge, A., Sun, Y., Kiker, T., Zhou, Y., & Ye, K. (2023). A metabolome-wide Mendelian randomization study prioritizes potential causal circulating metabolites for multiple sclerosis. Journal of Neuroimmunology, 379, 578105. https://doi.org/10.1016/j.jneuroim.2023.578105
Song, S., Zhang, Q., & Yu, J. (2024). A Mendelian randomization study investigating the causal relationships between 1400 serum metabolites and autoimmune diseases. Heliyon, 10(14), e34560. https://doi.org/10.1016/j.heliyon.2024.e34560
Gilchrist, L., Mutz, J., Hysi, P., Legido-Quigley, C., Kõks, S., Lewis, C. M., & Proitsi, P. (2025). Evaluating metabolome-wide causal effects on risk for psychiatric and neurodegenerative disorders. BMC Medicine, 23(1), 326. https://doi.org/10.1186/s12916-025-04129-4
Zhang, Y., Liu, H., Zhang, H., Han, Z., Wang, T., Wang, L., & Liu, G. (2023). Causal association of genetically determined circulating vitamin D metabolites and calcium with multiple sclerosis in participants of European descent. European Journal of Clinical Nutrition, 77(4), 481-489. https://doi.org/10.1038/s41430-023-01260-4