A Blood Test for Progressive MS? What the Metabolite Signature Actually Shows
Secondary progressive multiple sclerosis (SPMS) is diagnosed backwards. No validated imaging or biofluid marker separates it from relapsing-remitting disease (RRMS), so clinicians confirm the transition only after irreversible disability has already accumulated. That delay carries a cost, because many disease-modifying therapies are approved for RRMS and very few for SPMS, meaning the treatment decision arrives late by design. Oppong and colleagues asked whether serum metabolomics could make the call earlier. They analysed sera from 52 patients with RRMS, 29 with SPMS, 80 healthy donors, and 30 disease controls with neuromyelitis optica, chosen because it shares symptoms with MS and is sometimes misdiagnosed as it. More than 250 metabolites were measured on a nuclear magnetic resonance platform and fed to seven machine learning models, then cross-checked against whole blood RNA sequencing in a subset of the same patients.
The Classifier Works, and It Leans on Clinical Variables
Boosted logistic regression separated SPMS from RRMS with an accuracy of 0.926, sensitivity of 0.897, specificity of 0.942, and an area under the ROC curve of 0.965. Logistic regression and random forest models correctly identified 26 of the 29 SPMS patients. Against the neuromyelitis optica controls the separation was sharper still, reaching an AUC of 0.993, which matters because that is the comparison a clinician actually confronts at diagnosis. The authors then did something many biomarker papers skip. Age and expanded disability status scale (EDSS) score appeared as top features in all seven models, so they reran the analysis without them. Accuracy fell from 88.9% to 80.3% and the AUC from 0.936 to 0.844. Metabolites alone still classify patients usefully, but the headline figures rest partly on two clinical variables that already encode how far the disease has progressed.
One Ketone Body Beat the Disability Scale
To turn the signature into something usable, the team calculated optimum cut-off values for each metabolite appearing in more than four models and built a points-based score. Five features survived: cholines, glutamine, saturated fatty acids, acetoacetate, and sphingomyelins. Trained on 70% of the patients and tested on the remaining 30%, the score separated RRMS from SPMS with an AUC of 0.9464. The discussion reports a further comparison. Acetoacetate on its own reached an AUC of 0.87, against 0.70 for EDSS and 0.83 for age. A single ketone body outperformed the disability scale that clinicians currently use to make this determination retrospectively. Since ketone bodies cross the blood-brain barrier and may be produced in the brain before elimination in blood, the authors argue serum monitoring could flag progression before disability accrues, while noting this has yet to be shown.
The Shift Runs From Glycolysis Toward Gluconeogenesis and Ketogenesis
Enrichment analysis pointed consistently at cellular respiration. Ketone body synthesis and degradation came out strongest (KEGG p = 1.27E-03), followed by glycerolipid metabolism (p = 1.41E-02), the citric acid cycle (p = 2.18E-02), pyruvate metabolism (p = 2.61E-02), and gluconeogenesis (p = 3.57E-02). The individual metabolite concentrations tell the same story with a clear direction. Lactate and glutamine, both tied to gluconeogenesis, were elevated in SPMS, as were the ketone bodies acetoacetate and beta-hydroxybutyrate and the citric acid cycle intermediate citrate. Alanine and pyruvate, both tied to glycolysis, were reduced. Lipids moved as well, with cholesterol esters in medium LDL, free cholesterol in small LDL and extra-large VLDL, saturated fatty acids, and linoleic acid among the top discriminating features. The pattern reads as cells in progressive disease turning away from glycolysis and toward gluconeogenesis and ketogenesis, which is what tissue under metabolic stress does when glucose handling falters.
The Transcriptome Agrees, but Thirteen People Produced It
Whole blood RNA sequencing identified 1,052 differentially expressed genes that clustered SPMS apart from RRMS, of which 948 were upregulated and 104 downregulated. Pathway analysis returned metabolism of RNA, cellular response to stress, metabolism of lipids, and cellular respiration, with 215 genes falling into those metabolic pathways, alongside immune pathways including TNFA signalling. The agreement with the metabolomics is real, and so is the caveat: this sequencing was performed on 8 patients with SPMS and 5 with RRMS. Thirteen people yielding over a thousand differentially expressed genes, 90% of them moving in the same direction, is a result to treat as hypothesis-generating rather than established. The authors handled this the right way by testing their pathway list against an independent dataset covering whole blood, myeloid cells, lymphocytes, and oligodendrocyte precursor cells from SPMS and RRMS patients, where overlapping pathways appeared.
A Network That Proposes a Mechanism Without Testing It
Combining all 1,052 genes with the metabolomic signature produced a network in which 30 genes interact with eight metabolites: creatinine, citrate, pyruvate, lactate, phenylalanine, tyrosine, glycine, and linoleic acid. The downregulated gene GOT2 (log2 fold change −0.617) sits at the junction of amino acid metabolism, aminoacyl-tRNA biosynthesis, lipid metabolism, pyruvate metabolism, gluconeogenesis, and ketogenesis, and has been linked to MS before. Upregulated genes in the lipid arm included SCD5, PLA2G12A, CARM1, and SLC25A1. From this the authors propose that the metabolic switch imposes metabolic stress, which triggers stress response pathways and subsequent neurodegeneration. The data are consistent with that sequence but cannot demonstrate it. Every sample here was collected at a single time point, so nothing in the design distinguishes a metabolic switch that drives neurodegeneration from one that follows it.
The Treatment Confound Has to Be Settled First
The limitation that most affects whether this becomes a clinical test is one the authors state plainly: patients with SPMS had been treated with disease-modifying therapies before their transition, whereas the RRMS cohort were recruited before first treatment. The two groups therefore differ in drug exposure as well as disease stage, and a serum metabolic difference could reflect either. The authors cite earlier work showing metabolites discriminate these subsets independent of treatment regimen, but this cohort cannot settle the question on its own. Other constraints stack on top: group sizes were unbalanced, age and EDSS differed between groups, the NMR platform measures fewer molecules than mass spectrometry, batch-to-batch variation is difficult to reproduce without appropriate controls, and serum metabolite levels shift with age, diet, sex, and hormonal status. The score needs validating in independent cohorts before anyone uses it. Two study designs would resolve the central ambiguity: treatment-matched RRMS and SPMS groups, or longitudinal sampling of RRMS patients through their transition, which is the only approach that can show whether acetoacetate rises before disability accrues or merely after.
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:
Oppong, A. E., Coelewij, L., Robertson, G., Martin-Gutierrez, L., Waddington, K. E., Dönnes, P., Nytrova, P., Farrell, R., Pineda-Torra, I., & Jury, E. C. (2024). Blood metabolomic and transcriptomic signatures stratify patient subgroups in multiple sclerosis according to disease severity. iScience, 27(3), 109225. https://doi.org/10.1016/j.isci.2024.109225
