Three Ways MS Genetics Studies Use Single-Cell Sequencing

Single-cell and single-nucleus RNA sequencing now turn up in almost every multiple sclerosis (MS) genetics paper, and it is tempting to read that as one method doing one job. It is not. A genetic association is a statement about populations; a cell is a statement about one body. The step between the two can be taken in either direction, and the direction chosen decides what the study is able to claim at the end. Three recent MS papers take it three different ways. One finds a variant first and then asks single-nucleus data where that variant acts. One finds cell states first and then asks the genetics whether those states were predictable from inherited risk. One measures nothing of its own and joins two public archives with statistics.

A Variant Found First, Then a Clinical Outcome
Giordano and colleagues started from a biological guess rather than a genome-wide sweep. Smoldering inflammation, the chronic compartmentalised immune process at the edge of chronic active lesions, is a driver of disability progression, so they tested genetic variation in genes tied to that process across two independent MS cohorts, a discovery set and then an independent confirmation set, totalling 2,817 patients. A locus in HIF1A came out associated with a more favourable disease course at over 20 years from onset. The harder part of any prognostic variant is showing the allele corresponds to something measurable, and here it did, in further independent cohorts: carriers of the protective HIF1A allele had lower paramagnetic rim lesion volume on MRI, lower neurofilament in both plasma and cerebrospinal fluid, and, in post-mortem progressive MS tissue, reduced microglial and macrophage inflammation with less axonal injury.

Single-Nucleus Data Asked Where the Allele Acts
The sequencing enters at that point, and the question put to it is narrow. Integrating single-nucleus RNA sequencing with spatial transcriptomics, the authors report that the HIF1A variant modulates gene expression in a cell-type-specific and context-dependent manner in the MS brain. Cell-type-specific and context-dependent is the entire reason the single-nucleus layer is there rather than a bulk measurement. Bulk RNA from MS brain tissue returns one number per sample, so a variant whose effect runs one way in one cell population and the other way in another, or appears only in one lesion region, averages out to nothing. The single-nucleus and spatial pair is what keeps the signal from cancelling itself, and it is what lets a prognostic allele become a candidate mechanism. The authors frame the result as a route from a genetic finding towards strategies aimed at progression.

Cell States Found First, Genetics Brought In After
King and colleagues ran the comparison the other way round, with the genetics arriving last. From longitudinal blood samples of patients with relapsing-remitting MS, they combined single-cell RNA sequencing, bulk transcriptomics, flow cytometry and targeted viral PCR into a time-resolved atlas of immune perturbations around relapse. A reproducible pre-relapse signature appeared in monocytes and B cells up to three months before clinical onset, enriched for host genes that respond to Epstein-Barr virus (EBV) lytic reactivation factors. The targeted PCR confirmed raised EBV LMP-1 transcripts in pre-relapse B cells, and flow cytometry showed expansion of CD11c atypical B cell populations displaying the EBV protein gp350. Only then does the genetics enter, as an annotation: the pre-relapse modules overlapped MS genome-wide association study risk loci and EBNA-2-bound enhancers, which the authors read as inherited susceptibility and EBV-responsive programmes operating through shared regulatory elements. They state the open question plainly: how this peripheral activation relates to central nervous system lesion formation remains to be established.

Two Public Archives Joined, Nothing Measured
Yan and colleagues generated no data at all, in what has become the commonest shape of an MS genetics paper. The expression side came from two public datasets: bulk RNA sequencing of seven chronic active MS lesions against ten healthy-donor white matter samples, and single-nucleus RNA sequencing of six MS plaque lesions against twelve white matter samples from six healthy donors. Nuclei with fewer than 200 genes or over 10 percent mitochondrial content were dropped, then clustered and annotated against the Human Cell Atlas, with microglia called on CSF1R and P2RY12. The bulk comparison yielded 428 genes up and 209 down; the microglia from the single-nucleus data yielded 721 up. The genetics side came from eQTLGen cis-eQTLs covering 15,695 genes in 31,684 participants, and MS summary statistics from the International Multiple Sclerosis Genetics Consortium on 115,803 individuals of European ancestry.

What Survived the Filters and What Did Not
Mendelian randomisation returned 996 genes with an association; intersecting those with the two expression lists left seven, five of which were kept. Higher expression of ARHGAP25 (odds ratio 1.45), HLA-DRB1 (2.24), MERTK (1.10), MS4A6A (1.16) and SYK (1.13) went with higher MS risk. The two dropped genes are instructive: SAMSN1 (0.74) and XBP1 (0.73) came out protective, at odds with the expression datasets, and the authors removed them on that ground. Cochran's Q found heterogeneity for HLA-DRB1 and SYK but not for the other three. Summary-data Mendelian randomisation supported a causal link for HLA-DRB1, though the HEIDI test fell below 0.05 in brain tissue while staying above it in whole blood, which the authors read as pleiotropy in the brain. Colocalisation put the posterior probability of a shared causal variant at 100.00 percent for HLA-DRB1 and 97.93 percent for SYK. The stated limits are the design's own: public datasets that may not cover the genetic diversity of MS patients, and functional validation still outstanding.

What Each Design Can and Cannot Settle
Set side by side, the three show what the single-cell layer can be asked for. When the variant comes first, single-nucleus and spatial data answer where: which cell, which lesion region, under what conditions. Without that answer a prognostic allele stays a correlation with a clinical course. When the cell state comes first, the genetics answers whether that state was foreseeable from inherited risk, and an overlap with risk loci and virus-bound enhancers argues for shared regulatory elements, not for any one patient's alleles. When neither is measured in the same people, single-cell data works as a filter on a gene list, and the gene that survives is the one whose expression, eQTL and association evidence already agree, here HLA-DRB1 on every metric run. All three finish at a gene with a tool pointed at it, and all three leave the same step open: the allele, the cell and the clinical course measured in the same person.

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:
Giordano, A., Stridh, P., Preziosa, P., et al., Kockum, I., & Esposito, F. (2026). A HIF1A variant impacts long-term disability and smoldering inflammation in multiple sclerosis. Acta Neuropathologica, 151(1), 12. https://doi.org/10.1007/s00401-026-02984-w
King, D. A., Saxena, S., Caefer, D., et al., Gewurz, B. E., & Chitnis, T. (2026). EBV reactivation priming of the peripheral immune system in multiple sclerosis relapse. Nature Medicine (online ahead of print). https://doi.org/10.1038/s41591-026-04665-3
Yan, W., Wen, J., & Jianhong, W. (2025). Integrating genomic, eQTL, and Mendelian randomization analyses to identify microglial drug targets in multiple sclerosis. Journal of Cellular and Molecular Medicine, 29(22), e70754. https://doi.org/10.1111/jcmm.70754