Metabolomics and Single-Nucleus Data From the Same MS Lesion
The circulating metabolome is known to be dysregulated in multiple sclerosis (MS), but the metabolome of the lesions themselves had not been studied. Ladakis and colleagues went to the tissue, and built the study around one design choice that makes the result interpretable: at every sampling site they took paired neighbouring samples for both metabolomics and single-nucleus RNA sequencing, so the metabolite profile and the cell profile come from the same piece of brain rather than from matched but separate specimens. That allows a question neither method answers alone, which is whether a given metabolite tracks with a given cell population. Seventeen samples from 5 people with secondary progressive MS and 8 samples from 6 controls went through metabolomic profiling, returning 783 metabolites, with paired transcriptomic and cell count data available for 19 of them.
The Sampling Scheme and the Chemistry
Frozen and paraffin-embedded mirror brain tissue blocks came from the Netherlands Brain Bank, five people with secondary progressive MS and three age- and sex-matched non-affected controls, plus three further controls from the Rocky Mountain Brain Bank. A 3 mm punch-biopsy tool took roughly 50 mg frozen samples from the edge of chronic active lesions, the core of chronic inactive lesions, and the periplaque white matter, with control specimens taken the same way from normal white matter. Metabolomics ran on a global panel using methanol protein precipitation followed by liquid chromatography with tandem mass spectrometry and gas chromatography with mass spectrometry, with spectra matched to a reference library and area under the curve used to measure each metabolite. The handling of the data is set out in detail: metabolites were normalized by extracted mass because one sample yielded 16 mg against 24 mg for the rest, 107 metabolites missing in 30% or more of samples were removed, remaining gaps were filled by k-nearest neighbours with three neighbours, and values were log transformed and scaled. An outlier check using a Euclidean distance sample network with a standardized connectivity measure, taking Zk below −3 as the cut, found none.
Lipids Separate Lesion From Control Tissue
Individual metabolites were compared using generalized estimating equation models adjusted for age and sex, with an unstructured correlation structure to handle multiple tissue samples from the same person. Against control white matter, MS lesions carried higher sphingosines (standardized mean difference 0.24, q = 2.88×10⁻⁵) and higher sphingomyelins and ceramides (0.26, q = 2.15×10⁻⁷), with lower lysophospholipids (−0.37, q = 1.86×10⁻⁷), energy metabolites (−0.31, q = 0.001), nucleotides (−0.28, q = 0.05) and monoacylglycerols (−0.23, q = 0.04). The periplaque white matter is where this gets interesting, because it looks normal and is not demyelinated, yet it already showed raised sphingomyelins and ceramides (0.09, q = 0.05) alongside lowered energy metabolites (−0.29, q = 0.01) and lysophospholipids (−0.13, q = 0.05). Sensitivity analyses adjusting for postmortem interval gave similar results. One exception stands out against the pattern: while most sphingolipids were uniformly raised in MS tissue, sphingosine-1-phosphate was not significantly different.
A Gradient Running From Periplaque to Core
Weighted gene correlation network analysis, applied here to metabolites rather than genes, reduced the panel to 16 modules, which were then compared across five tissue groups by ANOVA with Tukey post hoc tests. In lesion cores, sphingomyelins and ceramides were higher than in control white matter, periplaque white matter, and both the chronic active and chronic inactive lesion edges, making the core the extreme of the series rather than one point among several. Cores also carried higher sphingosines and dipeptides, and lower unsaturated fatty acids, endocannabinoids and nucleotide metabolites than control white matter. Both edge types showed raised sphingomyelins and ceramides, with the chronic inactive edge additionally raising sphingosines. And every tissue taken from an MS brain, periplaque included, had decreased lysophospholipids against control white matter.
Pairing Each Metabolite to a Cell Population
With transcriptomic and cell count data from the same tissue in 19 samples, metabolites were ranked by their Spearman correlation with the proportional abundance of each cell subpopulation, and metabolite set enrichment analysis produced a normalized enrichment score for each pathway and each cell type, with pathways of three or fewer metabolites excluded. Two opposing blocks came out. Sphingolipids, diacylglycerols and various cell membrane lipid metabolites correlated positively with astrocyte and immune cell subpopulations and negatively with oligodendrocytes and premyelinating oligodendrocyte progenitor cells. Endocannabinoids, long-chain fatty acids, monoacylglycerols, lysophospholipids and pantothenate or coenzyme A metabolites ran the other way, negative with astrocytic and immune subpopulations and positive with oligodendrocytes and premyelinating progenitors. Repeating the analysis with the network-derived pathways gave similar findings.
One Factor That Spans Both Layers
Multiomics Factor Analysis was run across both datasets, with four factors chosen. Factor 1 accounted for 40% of gene variability and 12% of metabolite variability, and it ordered the samples by pathological stage on its own: low in control and periplaque white matter, rising through the chronic inactive edge, then the chronic active edge, and highest in the core. The mRNA expression pattern of the top genes correlated strongly with levels of the top metabolites. On the positive side of that factor, gene pathways enriched in MS lesions included circulatory system development (q = 2.7×10⁻⁶), cell-to-cell signalling (1.7×10⁻⁵) and synaptic signalling (3.5×10⁻⁴), with phosphatidylethanolamines (9×10⁻⁷), dipeptides (1.5×10⁻⁶) and sphingomyelins (1.6×10⁻⁴) as the matching metabolic pathways. On the negative side, downregulated in MS lesions and particularly in cores, sat ensheathment of neurons (6×10⁻⁷), oligodendrocyte differentiation (2.1×10⁻⁶), cellular lipid metabolic process (5.5×10⁻⁴) and phospholipid biosynthesis (0.003), with hexosylceramides (9.6×10⁻¹²), endocannabinoids (5.2×10⁻⁹), lysophospholipids (4.3×10⁻⁵) and monoacylglycerols (0.014). KEGG and Reactome analyses gave similar results.
Two Readings, and What the Design Cannot Settle
The authors put forward two interpretations rather than choosing between them. Ceramides and sphingomyelins, together with diacylglycerols and phospholipids, are major components of myelin and could simply be breakdown products; since myelin has been identified as an inhibitor of oligodendrocyte progenitor differentiation and maturation, that effect might be exerted through these very metabolites. The opposing block, correlated with promyelinating and anti-inflammatory cell profiles, could instead mark the resolution of inflammation and the transition into remyelination. Mouse lipidomics after lysolecithin-induced demyelination supports the second reading, with ceramides and phospholipids rising only in the acute phase while free fatty acids rose later during remyelination. Two therapeutic leads follow: fingolimod appears to act directly on ceramide production in the CNS in mouse Alzheimer models, which might add to its lymphocyte sequestration effect, and inhibiting neutral sphingomyelinase 2 restored the ceramide-sphingosine balance and improved behavioural outcomes in mouse models of HIV, so those inhibitors might be worth testing in MS. Set against that, the authors note the weak record of three omega-3 supplementation studies and of cannabinoids for symptom management, which they say could mean those metabolites are not suitable as therapeutic targets. Their limitations are listed plainly. The sample is small and larger cohorts are needed. Patients were untreated for MS in the three months before death and most were never treated, but interactions between other drugs and the metabolomic profile cannot be fully excluded. Homogenizing brain tissue means they cannot separate metabolite concentrations inside cells from those outside, nor assign cellular origin. It is not known whether the metabolomic changes drive the inflammatory reaction that leads to demyelination or are a byproduct of demyelination and myelin breakdown. And correlating metabolite levels with cell populations does not imply causation.
Disclaimer: This blog post is based on the cited study 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:
Ladakis, D. C., Pedrini, E., Reyes-Mantilla, M. I., Sanjayan, M., Smith, M. D., Fitzgerald, K. C., Pardo, C. A., Reich, D. S., Absinta, M., & Bhargava, P. (2024). Metabolomics of multiple sclerosis lesions demonstrates lipid changes linked to alterations in transcriptomics-based cellular profiles. Neurology: Neuroimmunology & Neuroinflammation, 11(3), e200219. https://doi.org/10.1212/NXI.0000000000200219
