Which Metabolites Cause MS? The First Metabolome-Wide Mendelian Randomization Study in MS

Metabolomics has produced a long list of molecules that differ between people with multiple sclerosis (MS) and healthy controls, and almost none of that work can say which direction the arrow points. Lipid levels respond to disease progression, so a raised measurement in a patient cohort may be a consequence rather than a cause. Ge and colleagues addressed this by running the first metabolome-wide Mendelian randomization (MR) study for MS, estimating the causal effect of 571 circulating metabolites on disease risk. Genetic instruments came from three published GWAS of the blood metabolome, covering 452, 123, and 249 metabolites in 7,824, up to 24,925, and 115,078 participants respectively. The outcome was the International Multiple Sclerosis Genetics Consortium discovery GWAS, 14,802 cases and 26,703 controls of European ancestry. Twenty-nine metabolites showed suggestive evidence of a causal association, concentrated in lipid metabolism, amino acid metabolism, and energy metabolism.

Two Instrument Thresholds, and Why the Authors Warn Against Reading Them as Repeats
The analysis required at least three independent instruments per metabolite so that heterogeneity and pleiotropy could be tested at all. Instruments were clumped at r² < 0.001 within a 10 Mb window, proxy SNPs were accepted at r² ≥ 0.8, and every metabolite cleared an F-statistic above 10, which rules out weak-instrument bias. The primary estimator was multiplicative random-effect inverse variance weighting, backed by MR-Egger, weighted median, weighted mode, MR-PRESSO, Cochran's Q, and the MR Steiger directionality test. The design choice that shapes the whole result set is the use of two significance cutoffs for instrument selection. At the conventional genome-wide threshold of P < 5 × 10⁻⁸, 404 metabolites had enough instruments. Relaxing to P < 1 × 10⁻⁶ added another 167, reaching 571. The authors state plainly that the two instrument sets overlap and therefore are not independent repeats of each other, a caveat that similar studies often leave out.

The Lipoprotein Subclass Decides the Sign, Not the Lipid
Ten of the 29 metabolites are lipids in specific lipoprotein subclasses, and they split cleanly by particle. In large VLDL, total cholesterol, phospholipids, and triglycerides were all associated with lower MS risk (OR 0.83, 0.80, and 0.81). Phospholipids in small VLDL ran the same way (OR 0.80), as did phospholipids in chylomicrons and the largest VLDL particles (OR 0.75). In very large HDL the direction reversed: total cholesterol at OR 1.20 (95% CI 1.04-1.40), cholesterol esters at OR 1.14, and phospholipids at OR 1.13. The same molecules, measured in different carriers, give opposite estimates. This also resolves an observational puzzle the authors raise. Gafson and colleagues had reported in 2018 that cholesterol, phospholipids, and triglycerides ran higher in the larger subclasses of both VLDL and HDL in relapsing-remitting patients. Ge and colleagues argue the VLDL half of that observation is likely reverse causation, since lipid levels may respond to progression, while the MR estimate for VLDL points the other way.

Serine Carries the Strongest Amino Acid Signal
Five of the 29 metabolites are amino acids. Serine produced the strongest estimate in the study, OR 1.56 (95% CI 1.25-1.95, P = 9.32 × 10⁻⁵), and the leave-one-out analysis showed the effect was not driven by any single SNP. Lysine followed at OR 1.18 (95% CI 1.01-1.38) and O-sulfo-L-tyrosine at OR 1.15 (95% CI 1.01-1.31). The two gamma-glutamyl amino acids point in opposite directions: gamma-glutamylleucine is protective at OR 0.80 (95% CI 0.67-0.96) while gamma-glutamylphenylalanine raises risk at OR 1.22 (95% CI 1.04-1.43). Readers should note that the paper's Discussion section states this pair the other way round, describing gamma-glutamylleucine as risk-increasing and gamma-glutamylphenylalanine as risk-decreasing; the Results text and Figure 2 agree with each other and with the confidence intervals above. The serine result connects to established MS biology through two routes. Wang and colleagues showed in 2019 that Epstein-Barr virus infection, a known MS risk factor, upregulates the import and biosynthesis of serine in B cells. Serine is also a precursor of phosphatidylserine and sphingomyelin, lipids central to myelin. The observational literature is less tidy: plasma serine is lower in relapsing-remitting patients but higher in secondary progressive patients.

Ketone Bodies Survived the Outlier Test and Still Point the Wrong Way
Acetoacetate (OR 2.47, 95% CI 1.14-5.34) and acetone (OR 2.45, 95% CI 1.02-5.90) produced the largest point estimates in the study, with confidence intervals wide enough that both nearly touch 1.0. Visual inspection of the leave-one-out plots suggested outliers for both, so the authors ran MR-PRESSO. For acetoacetate the global test found no significant outliers. For acetone, removing the outlier SNPs strengthened the estimate to OR 3.60 (95% CI 1.86-6.97, P = 0.007). Uridine behaved similarly, at OR 1.45 (95% CI 1.10-1.91, P = 0.008), matching case-control reports of higher serum and cerebrospinal fluid uridine in MS patients. The ketone result sits awkwardly against the clinical literature. The authors note that higher circulating ketone bodies may reflect a protective shift in energy metabolism and that ketogenic diets have shown suggestive benefits in MS, which is the opposite of what a risk-increasing odds ratio implies. They flag this as a reason to study ketone bodies in MS development rather than only in treatment, and leave the tension unresolved.

Six Findings Are Firmer Than the Other Twenty-Three
The 29 metabolites are not equally supported, and the paper's own tiering is the most useful thing to carry away from it. Six metabolites were nominally significant and directionally consistent under both instrument thresholds. The remaining 23 reached nominal significance under only one cutoff, and 14 of those 23 have genetic instruments only under the lenient P < 1 × 10⁻⁶ threshold, meaning they would not exist as MR results at all under conventional genome-wide selection. For the other nine, the estimates from both thresholds were directionally consistent and close together. Set against previous work, the novelty is narrower than the metabolite count suggests: most of these molecules had already been linked to MS status or severity in traditional epidemiological studies, and one, HDL cholesterol, had appeared in an earlier MR study. What is new is the direction testing, not the molecules. The MR Steiger test supported metabolite-to-MS causality rather than the reverse throughout.

What the Design Cannot Tell You About Timing or Subtype
Four limitations bound how far these estimates travel. The MR estimates reflect lifelong genetic exposure and carry no information about a critical window, which matters for a disease with a well-documented adolescent risk period. The analysis could not separate MS subtypes, because subtype-specific GWAS of sufficient size do not yet exist, so a single odds ratio is being asked to cover relapsing-remitting and progressive disease at once. The MR methods assume a linear relationship between exposure and outcome, which may not hold for metabolites with an optimal physiological range. And every participant was of European descent. Two further gaps come from the data rather than the design: metabolites without three usable instruments were dropped entirely, and some metabolites appeared in two of the source GWAS but reached significance in only one, which the authors attribute to differing cohort characteristics and sample sizes. The analysis scripts are public on GitHub, and a second author reproduced the key MR results independently before the paper went out.

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:
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
Studies named above (Gafson 2018; Lazzarino 2017; Sylvestre 2020; Rzepinski 2022; Wang 2019; Lin 2022; Almrahmi 2022) are cited as reported and discussed within this article; see its reference list for full details.