Asking Genetics Which Metabolites Matter in MS

Metabolomics in multiple sclerosis (MS) has a direction-of-effect problem. A metabolite measured in a patient's plasma can be a cause of the disease, a consequence of it, or a consequence of the drug being used to treat it, and a single cross-sectional measurement cannot separate those three. Genetic instruments offer one way around this, since the alleles that set a person's metabolite levels are fixed from conception and cannot have been moved by a disease that began decades later. Four papers from 2025 and 2026 work on this from different ends. Three run Mendelian randomization on metabolite genome-wide association study data and arrive at named metabolites for MS. The fourth deletes the gene whose enzyme consumes a metabolite and measures what happens to the tissue, which is the test the first three ask for.

A Metabolome-Wide Scan Across Eight Brain Disorders
Gilchrist and colleagues tested roughly 1,000 plasma metabolites and 300 metabolite ratios against four psychiatric disorders and four neurodegenerative diseases, MS among them. Instruments came from two non-overlapping Metabolon panels, 1,091 metabolites and 309 ratios in 8,299 individuals from the Canadian Longitudinal Study of Aging and 722 metabolites in 8,809 individuals from the NIHR UK Bioresource, with 431 metabolites meta-analysed to a sample of 17,038. Inverse-variance weighted MR gave 138 effects after per-outcome FDR correction, of which 85 passed a sensitivity rule requiring six methods to point the same way and at least four to stay significant. Odds ratios ran from 0.73 to 1.48, and no reverse effects survived FDR correction. The largest positive effect anywhere in the study was on MS, for the lipid 1-(1-enyl-oleoyl)-GPE (P-18:1) at OR 1.48. Among the MS-specific hits, uridine and dihydroorotate both raised risk, and they sit on the same pyrimidine route, uridine being an eventual product of dihydroorotate.

What Survived the Instrument Checks
Over half of the 85 effects turned out to rest on a single influential variant in leave-one-out analysis, leaving 41 polygenic metabolite-outcome pairs, seven of them with MS. Bayesian model averaging was then run per outcome to handle the genetic correlation between metabolites; 5-dodecenoate was dropped from the MS model for correlating above 95% with N-formylmethionine and uridine, and every metabolite retained for MS cleared a marginal inclusion probability of 0.1, which the authors read as separate causal contributions rather than one signal seen several times. Carnitine derivatives came out protective in both neurodegenerative models, adipoylcarnitine for MS and acetylcarnitine and propionylcarnitine for ALS, with acetylcarnitine holding the highest marginal inclusion probability of any metabolite tested at 0.893. Two further checks cut the other way. Colocalization for 5-methyluridine and MS returned PP.H3 of 0.63, suggestive of two distinct causal variants in linkage disequilibrium rather than one shared one. And polygenic scores for the prioritised metabolites predicted their own outcomes negligibly in the UK Biobank, with nothing surviving FDR correction over 23 tests.

Adding a Protein Upstream of the Metabolite
Yan and colleagues asked a narrower question with the same machinery. They instrumented 91 SCALLOP inflammatory proteins, tested them against MS in two outcome datasets, the IMSGC discovery set with 14,802 cases and 26,703 controls and a UK Biobank set with 1,356 cases and 395,209 controls, then ran a second step through 1,091 blood metabolites, 309 ratios, 233 circulating metabolic traits and 338 cerebrospinal fluid metabolites. Interleukin-7 held up in both cohorts, OR 1.40 (1.07 to 1.83) in IMSGC and 1.001 (1.000 to 1.002) in UK Biobank, with a weighted median of 1.43 and a Bayesian weighted estimate of 1.40, and reverse MR from MS to IL-7 was null. Six serum metabolites carried part of the IL-7 effect: taurodeoxycholate at 16.5%, sphingomyelin (d18:1/24:1, d18:2/24:0) at 10.0%, taurocholenate sulfate at 8.7%, anthranilate at 8.4%, the leucine-to-phosphate ratio at 8.4% and albumin at 3.9%. The authors report that most of those mediation intervals include zero, anthranilate being the exception at beta 0.028 (0.01 to 0.06). No mediation came through the cerebrospinal fluid metabolites.

The Same Design, an Independent Team, One Fluid
Mao and colleagues built the same two-step structure on the same IMSGC outcome, with circulating inflammatory proteins as exposure and cerebrospinal fluid metabolites as the candidate middle layer. Five inflammatory proteins came out associated with MS, three of them positively on genetic prediction. Ten cerebrospinal fluid metabolites were linked to MS and six were positively associated with disease on genetic prediction, with no reverse effect of MS on the metabolites. The mediation step then returned nothing: the authors state that cerebrospinal fluid metabolites do not serve as an intermediary factor in the pathway from inflammatory proteins to MS, and they put that negative in the conclusion rather than in a footnote. So two groups, working separately with different protein panels and different instrument thresholds, both found cerebrospinal fluid metabolites causally tied to MS and both failed to find them carrying an inflammatory protein's effect.

The Next Steps the Three Authors Name
Each paper is explicit about where its own design runs out. Gilchrist and colleagues note that they worked in blood, that metabolites measured in brain tissue or cerebrospinal fluid may give more direct insight into brain disorders, and that triangulation across tissue types is the obvious next move; they also write that the metabolites they prioritised would benefit from functional work in vitro or in vivo before anyone reads them as targets, and that polygenic scores calculated only in the UK Biobank inherit that cohort's selection bias. Yan and colleagues flag their screening threshold of p below 1 x 10^-5 without multiple-testing correction as a false-positive risk, note that mediation proportions of this size leave room for mediators they did not measure, and restrict their claims to European-ancestry samples. Gilchrist and colleagues also exclude the APOE region from their Alzheimer's analyses on MR grounds and say plainly what that costs them. Read together, the three MR papers converge on the same two requests: measure the brain compartment properly, and take the named metabolites into an experiment.

Deleting the Gene Whose Enzyme Consumes the Metabolite
Kerndl and colleagues ran that experiment on a different metabolite. Targeted metabolomics of mouse spinal cord and cerebrospinal fluid across experimental autoimmune encephalomyelitis phases gave 57 spinal cord and 15 cerebrospinal fluid disease-linked metabolites, and pathway enrichment pointed at arginine biosynthesis in both tissues: arginine down, citrulline and argininosuccinate up, the fall in cerebrospinal fluid arginine confirmed by ELISA alongside upregulation of the myeloid arginine transporter Slc7a2. Roughly half the infiltrating monocyte-derived cells were ARG1-positive and iNOS-negative, confirmed by CCR2 fate mapping, and MALDI imaging put ornithine and fumarate inside lesions. ARG1 immunohistochemistry on 18 MS and 7 control postmortem brains found little in control white matter and significantly more at the rim and centre of active lesions, in CD68-positive cells. Deleting Arg1 in myeloid cells then raised cerebrospinal fluid arginine, reversed the trajectory of many disease-linked metabolites including quinolinic and alpha-aminoadipic acid, raised reduced glutathione, expanded regulatory T cells with more IL-10 and TGF-beta, and improved disease. An arginine-free diet moved every one of those readouts the other way, with more demyelination on Luxol fast blue staining.

Disclaimer: This blog post is based on the cited studies 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:
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
Yan, W., Jianhong, W., Ping, G., & Linming, Z. (2025). Inflammation-metabolite crosstalk in multiple sclerosis: A mediation Mendelian randomization study of plasma inflammatory proteins, iron, and serum/cerebrospinal fluid metabolites. Science Progress, 108(3), 00368504251378619. https://doi.org/10.1177/00368504251378619
Mao, X., Lu, X., Liu, Y., Wu, H., Li, B., & Bi, X. (2025). Exploring the mediating role of cerebrospinal fluid metabolites in the pathway from circulating inflammatory proteins to multiple sclerosis: A Mendelian randomization study. Multiple Sclerosis and Related Disorders, 98, 106440. https://doi.org/10.1016/j.msard.2025.106440
Kerndl, M., Musiejovsky, L., Komljenovic, A., et al., Sharif, O., & Schabbauer, G. (2026). Monocyte infiltration induces CNS arginine catabolism to fuel neuroinflammation. Nature Immunology, 27(7), 1418–1432. https://doi.org/10.1038/s41590-026-02516-4