What scRNA-seq Adds to Metabolite Claims in MS

Single-cell and single-nucleus RNA sequencing (scRNA-seq and snRNA-seq) answer one question well: which cell is doing something. Neither measures a metabolite. Three recent multiple sclerosis (MS) papers pair that strength with a metabolic claim of an unusual kind, because in none of them is the molecule under study being burned for energy. It is the raw material for a chemical tag on a transcription factor, or a danger signal that a receptor reads, or a nutrient that one cell strips from the cells around it. What repays attention across the three is the specific job the sequencing was given, because it differs each time: finding which cells touched each other, mapping a pathway across lesion regions, and prioritising a gene in a compartment nobody had examined.

scRNA-seq Used to Find Which Cells Touched
Lee and colleagues set out to work out what astrocytes and CD4 T cells do to each other, which is harder than it sounds because the interesting events happen on contact. They combined rabies barcode interaction detection followed by sequencing with scRNA-seq, co-culture systems and cell-specific CRISPR-Cas9 perturbation in living animals, and labelled intercellular contacts to pull out the CD4 T cells that had actually touched an astrocyte. Astrocytes carrying CD40 and MHC class II promoted T cell autoimmunity in the central nervous system, and direct contact between astrocyte and CD4 T cell strengthened pathogenic T helper 17 responses. The barcoding matters here more than it might appear: it turns "these two cells touched" into something measured rather than inferred from how close they sit in a section, which is the usual weakness of contact arguments built on imaging alone. So far this is immunology, with the sequencing doing the work of naming partners.

Where the Molecule Enters, Beyond What Sequencing Showed
The metabolic step appears when they turn the question around and ask what the T cell does to the astrocyte. Using subproteomic methods in living animals together with AlphaFold-Multimer predictions, they report that CD40L on the CD4 T cell activates CD40 on the astrocyte, and that this drives accumulation of lipid droplets marked by PLIN4. Those droplets supply acetyl-CoA, which is needed for the acetylation of the p65 subunit that switches NF-κB on, and NF-κB activation is what lets the astrocyte present antigen. The molecule is a two-carbon donor for a protein tag rather than fuel. Astrocytes positive for CD40, MHC class II and lipid droplets were then found in MS samples by single-nucleus RNA sequencing and immunohistochemistry, which is a presence check in human tissue; the causal chain itself was built in mice.

snRNA-seq Used as a Map Across Lesion Regions
Hu and colleagues gave snRNA-seq the opposite job, using it as a map rather than a partner finder, and reanalysed published human MS white matter datasets, covering periplaque tissue, inactive and active chronic lesion edges, lesion cores, relapsing-remitting white matter and healthy controls, between roughly 4,000 and 20,000 nuclei per group. Reactive astrocytes marked by complement C3 and GFAP expanded sharply at chronic active lesion edges and in active lesions. Components of the cGAS-STING pathway rose specifically at those edges while periplaque regions sat at control levels, so the activation was spatially confined rather than diffuse, and module scores at the single-cell level were higher there than at any other site. Both GFAP and C3 were higher inside the astrocytes from those edges than in control white matter, so the shift is not only a change in how many astrocytes are present but in what the individual cells express. The pathway in question senses double-stranded DNA, which makes the molecule an alarm ligand rather than a substrate.

From a Sensed Signal to a Secreted Protein
This is the one of the three that measured its molecule in people. Cell-free double-stranded DNA was raised by immunoassay in both cerebrospinal fluid and serum from MS patients against controls, and serum levels tracked disability. GFAP and C3 protein were also raised in patient fluid, with serum C3 correlating with disability scores across 34 patients at a coefficient of 0.6144. In culture, DNA mimicry with poly(dA:dT) on top of inflammatory cytokines amplified STING phosphorylation beyond either alone, and the downstream effector they name is clusterin, a secreted protein that drove oligodendrocyte apoptosis and disrupted myelin in slice cultures. The brain-penetrant STING inhibitor H-151 blocked clusterin release and improved animals both preventively and after disease was established.

scRNA-seq Turned on Treatment-Naive Bone Marrow
Xie and colleagues pointed scRNA-seq at bone marrow rather than brain, reanalysing data from treatment-naive MS patients and healthy controls with donor-aware proportion analysis, trajectory inference, module scoring, donor-level pseudobulk enrichment, expression-based transcription factor prioritisation and a virtual perturbation of ARG1. MS marrow showed myeloid skewing and a heterogeneous suppressor-like neutrophil state, and the regulatory analyses converged on an ARG1 network, with neutrophil degranulation the strongest enrichment after multiple-testing correction. They report a negative alongside it: donor-level subtype proportions were comparable between patients and controls. In animals, marrow suppressor cells and ARG1 protein both rose, and oral arginase inhibition given before onset went with lower clinical scores. Arginase consumes arginine, so here the molecule matters by its absence from neighbouring cells. Worth being clear about the weakest link: the perturbation of ARG1 in the human data is computational, and the wet-lab step that follows is in animals, so the human contribution is the transcriptional landscape and nothing more.

What the Sequencing Settles and What It Hands On
Reading the three together, the practical difference is what a perturbation has to do. If a metabolite is fuel, the intervention is a flux: more or less of it. If it is a tag donor, an alarm ligand or a withdrawn nutrient, the intervention is an enzyme or a receptor, and that is a gene. Each study ends up there: PLIN4 and CD40 in one, cGAS and STING with clusterin downstream in the second, ARG1 in the third, each with a tool already pointed at it. That is also the honest limit of what the sequencing did. scRNA-seq and snRNA-seq delivered the cell and, through enrichment and regulatory inference, the candidate gene; the chemistry arrived from immunoassays, western blots, droplet staining and animal pharmacology. Only one of the three put a number on the molecule itself in patients, and it did so with an immunoassay rather than with any sequencing.

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:
Lee, J.-H., Li, Z., Soto, J. S., et al., Prat, A., & Quintana, F. J. (2026). Antigen presentation by CD40+MHC-II+ astrocytes promotes CNS autoimmunity. Nature, 657(8132), 755-764. https://doi.org/10.1038/s41586-026-10860-6
Hu, S., Xiao, X., Cheng, X., et al., Zhao, Y., & Li, R. (2026). The cGAS-STING pathway drives astrocyte-mediated demyelination in multiple sclerosis through clusterin secretion. CNS Neuroscience & Therapeutics, 32(8), e71091. https://doi.org/10.1002/cns.71091
Xie, H., Fu, Y., Bi, Y., et al., Chen, B., & Ren, H. (2026). An ARG1-associated immunometabolic program in bone marrow PMN-MDSC-like cells in multiple sclerosis and experimental autoimmune encephalomyelitis. Frontiers in Immunology, 17, 1924483. https://doi.org/10.3389/fimmu.2026.1924483