Pairing Single-Cell Data With Metabolites in MS
Single-cell RNA sequencing measures transcripts. Metabolites are not transcripts, and the distance between the two is where most claims about metabolism in multiple sclerosis (MS) quietly live. A cell that transcribes the whole glutathione machinery may still be short of glutathione; a pathway that lights up in an enrichment test has told you about potential, not flux. Three recent papers approach that gap from different distances. Chen and colleagues read metabolic potential from human tissue transcriptomes and then test the metabolites themselves in a separate cell system. Ahuja and colleagues run lipidomics and single-nuclei transcriptomics on the same cultures. Jiang and colleagues infer lipid metabolism from transcriptional programmes alone. Reading them together is mostly an exercise in watching how far each one has to reach.
Starting From Human MS Tissue, Then Leaving It
Chen and colleagues assembled their analysis from five sources: human MS optic-chiasm transcriptomics (GSE100297), RNA sequencing of glutamate-treated primary neurons (GSE255041), differential-expression summaries from R28 retinal precursor cells, a single-nucleus atlas of normal human optic nerve (GSE236566), and their own R28 experiments. The human MS tissue showed enrichment of unfolded-protein-response, oxidative-stress, apoptosis and ferroptosis programmes. The glutamate-treated neurons showed amino-acid stress, unfolded-protein response, oxidative stress and apoptosis, with ferroptosis enrichment that was concordant but weaker. The agreement between the two is partial rather than clean, and the paper reports it that way, which matters because glutamate exposure is being used here as a stand-in for what MS tissue is doing.
Which Cell Type Holds Which Metabolic Programme
The single-nucleus work gives the cell-type split, and it is worth being precise about whose nerve it came from: a normal human optic nerve, not an MS one. In that tissue, redox and lipid-peroxidation programmes were highest in myeloid cells, while hexosamine and O-GlcNAc gene programmes together with SLC7A11, the cystine-glutamate exchanger, were prominent in the astrocytic and stromal compartments. That division assigns two halves of one problem to two different cell populations: the cells most loaded with peroxidation machinery are not the cells carrying the main cystine import route. Any metabolite intervention aimed at the glutathione axis is therefore acting on astrocytic biology while the lipid damage is being read out somewhere else, and a transcript atlas is the only thing here that could have shown that.
Two Metabolites, Tested Separately, With Mixed Results
The metabolite experiments then run in R28 cells rather than in nerve tissue. Glucosamine at 2.5 mM shifted the glutamate dose-response curve rightward and preserved both the GSH/GSSG ratio and mitochondrial respiration, but it also lowered basal metabolic activity, mitochondrial membrane potential and ATP-based viability, and it raised global O-GlcNAc levels roughly 3.6-fold. Selenium behaved by dose rather than by direction: 1 and 5 µM preserved metabolic activity, 5 µM reduced lipid oxidation, and 20 µM raised basal lipid oxidation. The negative results are the informative ones. RSL3 and erastin cut viability sharply in the absence of glutamate, ferrostatin-1 largely restored it, and glucosamine did not. The selenium benefit against RSL3 shrank when glutamate was present, with an adjusted interaction P of 0.0342. The authors state outright that the GPX4 dependence of glucosamine-mediated protection remains unresolved, which is the honest reading of a compound that helps against one stress and fails against a direct ferroptosis trigger.
Measuring Lipids and Transcripts in the Same Material
Ahuja and colleagues sidestep the pairing problem by building a system they can run both assays on. Their human myelin spheres come from cocultured neurons and oligodendrocyte precursor cells derived from human pluripotent stem cells, and they report myelination as early as six weeks into coculture, assessed by confocal and electron microscopy, single-nuclei transcriptomics, lipidomics and electrophysiology on the same cultures. The structures matured into multilamellar, compacted myelin sheaths whose lipid compositions and transcriptomic profiles followed the temporal dynamics of human oligodendrocyte development in vivo. Having both readouts on one material is the methodological point: a lipid class and the transcripts for its synthesis either move together or they do not, with nothing borrowed from another dataset. The stated resemblance is to late fetal oligodendrocytes, which sets the ceiling on how far the system reaches toward MS, and the two challenges they applied, lysolecithin-induced demyelination and Rabies virus infection, strip myelin chemically or attack it virally rather than through autoimmunity. What it can answer is whether a lipid change precedes or follows a transcriptional change during myelin formation, which no human lesion sample can answer, because the lesion arrives already finished.
Lipid Metabolism Read Entirely From Transcripts
Jiang and colleagues sit at the far end of the range, with lipid metabolism inferred from pathway enrichment and no metabolite measured. Working in experimental autoimmune encephalomyelitis, they found that splenic and bone marrow neutrophils acquired both CD4 T cell suppressive and Th17-promoting activity, while lung neutrophils were predominantly suppressive, and they describe three subsets: a Ki67-defined N0, an Acvrl1-defined N1 and a CD300ld-defined N2. The mechanism they propose is that disease drives splenic extramedullary granulopoiesis, generating a heterogeneous neutrophil pool, and that the lung microenvironment then pushes CD300ld neutrophils toward a more immunosuppressive state through pathways associated with lipid metabolism. Analogous subsets turned up in rheumatoid arthritis and lupus models and in patients with MS. The patient data are the part that connects this to MS at all, and they establish that the subsets exist rather than that the lipid mechanism operates in people.
Three Distances From an Actual Measurement
Laid side by side, the three papers differ less in subject than in how much is measured and how much is inferred. Ahuja and colleagues measure lipids and transcripts on one material, and pay for it with a system that resembles late fetal development. Chen and colleagues measure metabolite effects directly but in retinal precursor cells, while the human transcripts that motivated the choice came from MS tissue and a normal optic nerve, and they leave the key mechanistic question open by name. Jiang and colleagues measure neither metabolite nor flux, and reach MS through subset correspondence in patients. None of the three is a metabolite measurement in MS tissue resolved to single cells, and the shape of what is missing is the same in all three.
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:
Chen, Y.-J., Chen, Y.-Y., Liu, S.-T., Wu, C.-J., & Chiu, Y.-L. (2026). Glucosamine and selenium differentially regulate glutathione, lipid peroxidation, and mitochondrial function during glutamate stress in retinal neural cells. Free Radical Biology & Medicine. https://doi.org/10.1016/j.freeradbiomed.2026.10.031
Ahuja, K., Ramezankhani, R., Wang, X., Burg, T., Amos, G., Neyrinck, K., Silva, A., Arekatla, G., Cassidy, E. E., Nami, F., Terryn, J., Wierda, K., Vints, K., Vandamme, N., Poovathingal, S., Lambrichts, I., Swinnen, J. V., Van Den Bosch, L., De Groef, L., Moons, L., Verfaillie, C., Neyts, J., Jochmans, D., & Chai, Y. C. (2026). Human myelin spheres for in vitro oligodendrocyte maturation, myelination and neurological disease modeling. Stem Cell Reviews and Reports, 22(3), 1420–1441. https://doi.org/10.1007/s12015-026-11061-4
Jiang, Q., Zhu, G., Si, W., Ma, X., Liu, Q., & Yang, G. (2026). Single-cell transcriptomics reveal tissue-specific neutrophil heterogeneity and functional reprogramming in autoimmunity. Journal of Autoimmunity, 164, 103637. https://doi.org/10.1016/j.jaut.2026.103637
