Genetics

This post covers four human studies from 2026 that read MS single-cell data through perturbation rather than description, with no animal model among them, and ranks them by what each design can license. It covers the Brain study using ocrelizumab as an in vivo probe, where a machine learning pipeline built for continuous cell states found a chronically activated proinflammatory-cytotoxic T cell state that disappears when B cells are depleted, then recovered the same state in aggressive relapsing patients, in CSF-enriched clonotypes including EBV-specific ones, and in chronically inflamed lesion rims, with natalizumab producing the concordant opposite shift. It covers the two-patient CAR-T preprint in which CD19-targeted cells expanded in CSF and oligoclonal bands disappeared despite peripheral B cell return. It covers the reanalysis of public CSF and blood B cell datasets finding a raised zinc transporter and metallothionein signature in MS, and the spatial transcriptomics of cortical lesion stages where chronic inactive lesions carried heat shock proteins HSPA1A and HSPA1B. It states which of the four are hypotheses and which is a finding.