Metabolic Drivers of Multiple Sclerosis: Insights from a Metabolome-Wide Mendelian Randomization Study
Multiple sclerosis is a chronic immune-mediated disorder of the central nervous system characterized by neuroinflammation, demyelination, axonal injury, and progressive neurodegeneration. Although genetic susceptibility and environmental exposures—including Epstein–Barr virus infection, smoking, obesity, and low vitamin D status—are recognized contributors, the molecular processes connecting these factors to disease initiation remain incompletely understood. Metabolomics has added an important dimension to multiple-sclerosis research by identifying alterations in amino acids, lipids, nucleotides, and energy-related metabolites in patients. However, conventional case-control metabolomic studies cannot reliably determine whether an altered metabolite contributes to disease development, results from established disease, reflects treatment exposure, or arises through an unmeasured confounding factor. The study by Ge and colleagues, entitled A Metabolome-Wide Mendelian Randomization Study Prioritizes Potential Causal Circulating Metabolites for Multiple Sclerosis, addresses this problem by systematically evaluating whether genetically predicted differences in circulating metabolite concentrations influence multiple-sclerosis risk. Rather than treating metabolic abnormalities merely as biomarkers of disease activity, the investigators sought to identify metabolites that may occupy an upstream position in pathogenesis and therefore warrant mechanistic or therapeutic investigation.
Mendelian Randomization as a Framework for Causal Inference
Mendelian randomization is a genetic epidemiological approach that uses inherited genetic variants as instrumental variables for an exposure, such as the circulating concentration of a metabolite. Because alleles are allocated at conception, their distribution is generally less affected by behavioral, clinical, or socioeconomic confounding than directly measured metabolite concentrations. In conceptual terms, this allocation resembles aspects of randomization in a clinical trial: individuals inherit variants associated with slightly higher or lower lifelong exposure levels, allowing investigators to test whether these genetically predicted differences are associated with disease risk. Valid inference nevertheless depends on three central assumptions: the genetic instruments must be associated with the metabolite, they must not share causes with the outcome, and they must affect multiple sclerosis principally through the metabolite rather than through independent biological pathways. The authors used a two-sample design, obtaining genetic associations with metabolites and multiple sclerosis from separate genome-wide association studies. This design is especially valuable when large individual-level datasets containing both metabolomic and neurological phenotypes are unavailable. Importantly, the resulting estimates represent the effects of genetically influenced, long-term differences in metabolite levels; they should not be interpreted as equivalent to the effects of short-term dietary supplementation, pharmacological intervention, or acute metabolic change.
A Metabolome-Wide Analytical Strategy
The analysis integrated three genome-wide association studies of the human blood metabolome, encompassing 452 metabolites in 7,824 participants, 123 metabolites in as many as 24,925 participants, and 249 metabolites in 115,078 participants. All contributing metabolomic cohorts were of European ancestry. Associations with multiple-sclerosis risk were obtained from an International Multiple Sclerosis Genetics Consortium dataset containing 14,802 cases and 26,703 controls. Genetic instruments were selected using both a conventional genome-wide significance threshold of P < 5×10 −8 and a more permissive threshold of P < 1×10−6. The second threshold expanded metabolome coverage and increased the number of available instruments, although the authors appropriately emphasized that analyses based on the two thresholds were not independent replications. Variants were pruned for linkage disequilibrium, proxy variants were used when appropriate, and metabolites were required to have at least three independent instruments. The primary causal estimate was generated with a multiplicative random-effects inverse-variance weighted model. Robustness was assessed using weighted-median, weighted-mode, MR-Egger, MR-PRESSO, heterogeneity testing, leave-one-out analyses, and the MR Steiger directionality test. The workflow diagram on page 3 clearly summarizes this sequence, from the three metabolite GWAS datasets through instrument selection, harmonization, primary analysis, sensitivity testing, and classification of consistent or suggestive findings.
Twenty-Nine Metabolites Were Prioritized
Using genome-wide significant instruments, the investigators generated Mendelian-randomization estimates for 404 metabolites. Relaxing the instrument-selection threshold increased the total to 571 metabolites. Across the 404 metabolites examined under both thresholds, effect estimates were reported to be highly concordant, supporting the internal consistency of the analytical strategy. Twenty-nine metabolites showed evidence compatible with a potential causal association with multiple-sclerosis risk. Six met the authors’ more stringent definition of consistency, demonstrating nominal statistical significance and the same direction of effect under both instrument-selection thresholds; 23 additional metabolites were classified as suggestive. Sensitivity analyses generally produced directionally similar estimates, although they did not always retain nominal significance, which is unsurprising because alternative Mendelian-randomization estimators often have lower statistical power. The Steiger analyses supported the direction from metabolite concentration to multiple sclerosis rather than the reverse direction. Nevertheless, these 29 findings should be viewed as prioritized candidates rather than definitive causal determinants. The study tested hundreds of exposures, and most reported associations were nominal rather than metabolome-wide significant after stringent correction for multiple comparisons. The principal scientific value therefore lies in narrowing a large metabolic search space to a biologically coherent set of candidates for replication, experimental validation, and prospective investigation.
Lipoprotein Subclasses Reveal Metabolic Specificity
Ten of the prioritized metabolites were lipid measurements assigned to particular lipoprotein subclasses, and their effects differed substantially according to particle size and composition. Genetically predicted total cholesterol, phospholipids, and triglycerides in large very-low-density lipoprotein particles were associated with lower multiple-sclerosis risk. Phospholipid concentrations in small VLDL and in chylomicrons or the largest VLDL particles also showed inverse associations. In contrast, total cholesterol, cholesterol esters, and phospholipids within very large high-density lipoprotein particles were associated with higher risk. This pattern is scientifically important because it challenges the oversimplified classification of HDL as uniformly protective and VLDL as uniformly harmful. Lipoprotein particles are heterogeneous molecular assemblies whose biological effects depend on size, lipid composition, apolipoprotein content, tissue trafficking, and inflammatory context. The forest plot on page 4 visually demonstrates this subclass-dependent divergence: several VLDL-related estimates fall below an odds ratio of one, whereas very-large-HDL lipid estimates fall above one. The findings are broadly compatible with earlier Mendelian-randomization evidence linking higher HDL cholesterol to multiple-sclerosis susceptibility, while adding a more granular metabolomic perspective. They also illustrate why total plasma lipid measurements may conceal disease-relevant signals located within specific lipoprotein fractions.
Amino Acids, Ketone Bodies, and Nucleotide Metabolism
Several non-lipid metabolites also emerged as notable candidates. Genetically predicted serine was associated with a 56% increase in the odds of multiple sclerosis per standard-deviation increase, while lysine and O-sulfo-L-tyrosine showed more modest positive associations. Serine is biologically compelling because it contributes to one-carbon metabolism and serves as a precursor for phosphatidylserine and sphingomyelin, lipids with important roles in cellular membranes and myelin biology. Epstein–Barr virus has also been reported to stimulate serine uptake and biosynthesis in B cells, suggesting a possible intersection between viral transformation, immune-cell metabolism, and multiple-sclerosis susceptibility. The ketone bodies acetoacetate and acetone showed comparatively large risk estimates, with odds ratios of approximately 2.5. Acetone remained significant after removal of outlying variants in MR-PRESSO analysis. Uridine, a nucleotide-related metabolite, was also positively associated with risk. These results require careful interpretation: elevated ketone bodies in patients could represent both a disease-associated metabolic disturbance and a compensatory response, while Mendelian randomization estimates the consequences of lifelong genetically influenced exposure rather than the clinical effects of a ketogenic diet. The plots on page 5 provide additional support for serine, acetoacetate, and acetone by comparing multiple estimators and examining whether individual variants disproportionately drive the results.
Scientific Strengths, Limitations, and Translational Outlook
The study’s principal strengths are its broad metabolome-wide scope, use of large GWAS datasets, application of two instrument-selection thresholds, exclusion of metabolites with insufficient genetic instruments, and extensive sensitivity analysis. All included instruments had F-statistics above 10, reducing concern about weak-instrument bias, and several independent Mendelian-randomization methods were used to evaluate consistency and horizontal pleiotropy. However, important limitations remain. Horizontal pleiotropy cannot be completely excluded; some metabolites could not be evaluated because fewer than three instruments were available; results occasionally differed between metabolomic GWAS datasets; and the analysis could not distinguish relapsing-remitting, secondary-progressive, or primary-progressive multiple sclerosis. Standard Mendelian-randomization models also assume approximately linear exposure-outcome relationships and provide limited information about critical developmental or preclinical windows. Furthermore, the restriction to participants of European ancestry limits generalizability and reinforces the need for multi-ancestry metabolomic GWAS. Clinically, the findings do not yet justify modifying dietary serine, lysine, lipid, or ketone intake. Instead, they identify metabolic pathways that deserve replication, functional experiments, longitudinal measurement before disease onset, and eventually target-specific intervention studies. The most defensible conclusion is that circulating lipid subclasses, amino acids, ketone bodies, and nucleotide-related metabolites may contribute to multiple-sclerosis susceptibility and provide a focused framework for studying the metabolic architecture of neuroimmunological disease.
Disclaimer: This blog post is based on the provided 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.
References:
Fitzgerald, K. C., Smith, M. D., Kim, S., Sotirchos, E. S., Kornberg, M. D., Douglas, M., ... & Bhargava, P. (2021). Multi-omic evaluation of metabolic alterations in multiple sclerosis identifies shifts in aromatic amino acid metabolism. Cell Reports Medicine, 2(10).
