Year: 2026
Type: article
Abstract: Relapse-independent disability accumulation, or progression independent of relapse activity (PIRA), is an important clinical problem in multiple sclerosis, and it continues in many patients while peripheral inflammatory activity is controlled (1,2). In experimental autoimmune encephalomyelitis, Lee and colleagues report that astrocytes in the inflamed central nervous system can present antigen, and that this function depends on a lipid-handling programme (3). The question this finding opens is not whether the animal mechanism is credible, and it is not a choice between studying a cell state and studying its molecular pathway. The translational challenge is to establish whether treatment benefit depends on interrupting astrocytic antigen presentation, and therefore what evidence would allow that attribution. A reactive astrocyte state may still sustain disease: the relevant test is whether interrupting its antigen-presenting function changes pathology, not whether it appears first.In experimental autoimmune encephalomyelitis, lesion-associated astrocytes acquire an inducible state enriched for antigen-presenting and phagocytic programmes, including the MHC class II gene H2-ab1 (3). The experiments support a causal contribution of astrocytic antigen presentation to disease severity and implicate a PLIN4-associated metabolic programme: promoter-restricted (GfaABC1D) CRISPR-Cas9 inactivation of Cd40, H2-ab1 or Plin4 in astrocytes, delivered intracerebroventricularly 12 days after MOG35-55 immunization in C57BL/6J mice of both sexes, reduced disease severity, T-cell expansion and pathogenic Th17 cells; target engagement was verified at the level of CD40 and MHC-II expression, not at the edited locus, and off-target editing was not assessed. The chain described runs from T-cell-derived CD40L through astrocytic CD40 to a PLIN4-associated lipid-droplet metabolic programme in which ATP-citrate lyase (ACLY) generates acetyl-CoA, then to NF-κB activation, IL-6/IL-23, MHC class II-dependent antigen presentation, and pathogenic Th17 expansion. In cultured astrocytes, pharmacological ACLY inhibition reduced CD40-driven NF-κB activation, a readout one step upstream of antigen presentation. The acetylation of p65 at K310 that the study reports is a pathway-engagement readout; the same axis operates in other CNS antigen-presenting cells (3), and because the modification is shared with cytokine signalling it cannot show that the antigen-presenting function of astrocytes was suppressed. This is one study, and it has not been independently replicated; its readouts concern T-cell expansion and Th17 responses, features of relapse-type inflammation, and do not establish that the state operates in the compartmentalised inflammation of progression. The in vitro ACLY result should not be conflated with the promoter-restricted genetic evidence in EAE, and it does not establish that the same intervention would alter disease course.Human tissue provides an entry point, and no more. Lee and colleagues identified CD40+HLA-DRA+ astrocytes carrying lipid droplets in active and chronic active lesions, in their reanalysis of two published single-nucleus datasets with immunohistochemical validation (3). These astrocytes co-expressed CD74 with MHC class II, but HLA-DM and costimulatory molecules were not assessed, so their designation as functional antigen-presenting cells remains unproven; singlenucleus data cannot exclude ambient contamination. Three things have to be kept apart here: the marker combination, the antigen-presenting function it is taken to indicate, and any contribution to pathology. These findings identify a phenotypically similar astrocyte population in human lesions, but do not establish functional equivalence to the EAE state; co-localisation of this kind cannot show that these cells present antigen in vivo, that the state precedes tissue damage rather than follows it, or that it contributes to progression. Nor does it separate the astrocytic state from the chronic active lesion, whose association with disability is already established (4,5). The finding is a reason to look further, not a reason to conclude.The strongest competing explanation: the antigen-presenting astrocyte may be a reactive phenotype rather than a driver. Astrocytes sit within a lymphocyte-microglia-astrocyte axis in chronic active lesions, where they both respond to and amplify signals from other cells (6), and a cell that presents antigen in that environment may be doing so because of microglial activation or the lipid load released by ongoing demyelination. In that study CD40 and MHC class II expression was higher in microglia, dendritic cells and monocyte/macrophages than in astrocytes, and for Cd40, but not for H2-ab1, a myeloid-driven construct was more effective. Lipid-laden, MHC-II-positive astrocytes may also reflect passive uptake of myelin or cellular debris rather than antigen presentation. Driver and reactive readings need not exclude one another: a state that arises in response to injury can still sustain it. The useful question is therefore not whether the state appears first, but whether, against a background of established inflammation, it has an intervenable role in maintaining pathology. Cross-sectional co-localisation cannot establish direction. What would help is a functional experiment in which the astrocytic programme is interrupted or restored and lesion evolution is read out, interpreted alongside human observation.Measurement is the practical constraint, and the order in which evidence is acquired matters more than the number of candidate readouts. A tissue-level reference standard comes first, because biopsy and autopsy material can define what is being looked for but cannot be sampled repeatedly; a readout obtainable serially in patients comes next, followed by evidence on its origin and functional specificity, and only then a longitudinal test against clinical outcome. For now, paramagnetic-rim lesions (4) index lesion context rather than the astrocytic state. Candidate readouts have limits: slowly expanding lesions are a radiological construct not shown to index this state (7); TSPO PET is not astrocyte-specific and varies with methodology (8); glial fibrillary acidic protein (9) and chitinase-3-like 1 (10) are fluid markers of glial activation associated with progression, but neither identifies antigen presentation. Testing imaging candidates against tissue from the same lesions would be the most direct way to establish what a readout indexes. A bodyfluid marker would have to demonstrate both a central origin and a specific link to antigen presentation. Astrocytic lipid droplets are a tissue-level candidate feature; their use as a longitudinal biomarker, or their relationship to paramagnetic-rim lesions, remains to be established.For the clinical step, PIRA definitions and outcome ascertainment should be prespecified, drawing on the methodological review by Müller and colleagues (11), so that the result is less likely to reflect how progression was counted. A testable prediction is that a validated measure of this state would be associated with relapse-independent progression after accounting for inflammatory activity. Why this state should relate to progression rather than to acute lesion activity is itself an open question, and a longitudinal design would have to address it rather than assume it (12). Clinical association and cellular perturbation answer different questions; neither can substitute for the other. Causal structure should be stated and tied to measurement timing: age, disease duration, treatment, inflammatory activity and lesion load measured before the exposure are confounders, whereas tissue damage, atrophy and microglial activation measured after it lie on the causal path, so over-adjustment can conceal a real contribution. Loss of an adjusted association would weaken its incremental prognostic value, but would not by itself exclude a causal contribution.Selectivity is the harder problem; it is also the point at which a molecule-first strategy most easily misleads. ACLY generates acetyl-CoA in essentially all cells and sits at a branch point of fattyacid and cholesterol synthesis (13). An approved ACLY inhibitor already exists: bempedoic acid lowers LDL cholesterol (14), and its activation is largely hepatic (15), so it illustrates the distance between inhibiting a pathway and targeting a cell. CD40 is broadly expressed, including by microglia and macrophages in MS lesions (16); the systemic and thromboembolic effects that have constrained CD40-CD40L blockade (17) are not confined to antigen-presenting cells. That picture is incomplete without the second-generation agents: frexalimab, an anti-CD40L antibody, reduced new gadolinium-enhancing lesions in relapsing disease in a phase 2 trial (18), and a phase 3 trial of the same antibody against placebo is under way in non-relapsing secondary progressive multiple sclerosis (19). These are direct tests of the pathway in a progression-relevant population, and a positive result would establish that the pathway matters without showing that the benefit is astrocyte-directed. A central nervous system-penetrant Bruton's tyrosine kinase inhibitor has been tested in the same population, with a result that is likewise pathway-level rather than cell-level (20). Interrupting either would test a general metabolic or costimulatory pathway as much as the astrocytic state. State selectivity remains an aspiration until it is specified as requirements that a study could fail. We propose four, which are not of the same kind and are best read in sequence. The first two concern target engagement: effective exposure within lesions behind the blood-brain barrier, and suppression of antigen-presenting function in the relevant astrocytes, which requires a readout of that function rather than a shared upstream signal such as NF-κB activation or cytokine release. The third is a selectivity requirement: preservation of the metabolic, trophic and homeostatic functions that astrocytes provide to neurons and to oligodendrocytes, including the cholesterol metabolism on which oligodendrocyte survival and remyelination depend (21). Only the fourth bears on attribution: efficacy that is lost when the astrocytic programme is interrupted, as in an epistasis design in which a candidate agent is tested in animals whose astrocytic gene has already been inactivated: loss of benefit indicates dependence, and persistence indicates that the benefit arises elsewhere. Two caveats apply. Editing is partial, and in an already-attenuated animal the loss of drug benefit can reflect a floor effect rather than a shared pathway; non-additivity does not prove dependence. A dose-response or submaximal perturbation would separate the two.Thresholds for lesional exposure and for suppression of the readout cannot be fixed for an unspecified agent and assay; what can be required is that they are pre-specified and reported, and that the suppression required is anchored to the genetic benchmark (the effect achieved by the perturbation that produced benefit), so that a negative result is not confused with an underpowered one. This framework is our proposal, not a result of the study under discussion, and no agent has yet been shown to meet it. It does not follow that a pathway inhibitor is useless: a molecule-directed drug may be therapeutically useful without being demonstrably astrocytedirected. Such an intervention could complement existing immunotherapies, and the proposed mechanism might account for part of the residual pathology that persists after effective peripheral treatment. That remains a hypothesis: it is not evidence that anti-CD20 therapy fails, nor that astrocytes are the principal driver of progression. Whether this state is enriched in patients with progression despite B-cell depletion should be tested, not assumed, with controls matched for disease stage and lesion activity so that a treatment-related difference is not inferred from a stagerelated one (1,2).Two objections should be conceded in advance. The state may not be discrete or stable. Inducible states are often graded rather than binary (22), and overlapping programmes within one cell could make a threshold-based readout irreproducible. Broad molecular inhibition is worth pursuing: an effective inhibitor would test the pathway even without isolating the state. Separately, three failure modes should be distinguished, because they point to different decisions. A failure of readout limits what can be asked in patients but says nothing about the biology. A failure of selectivity limits therapeutic translation but leaves the mechanism standing. Only the third kind bears on the claim itself. If a sufficiently selective functional intervention suppresses astrocytic antigen presentation and pathology nonetheless progresses, the pathogenic contribution of this state in that setting is weakened, although redundancy among antigen-presenting cells, compensation, the timing of intervention and incomplete suppression would each have to be excluded first. Overlap with general reactive gliosis would bear on how sharply the state can be defined, but not on whether the function matters.These considerations suggest one way of reading intervention studies. Astrocyte-directed is used here in a strict sense: it is a claim about dependence rather than delivery, narrower than cell-typerestricted exposure alone, and narrower than state-restricted action. A treatment effect justifies the term astrocyte-directed only when the four requirements set out above are met. Where a compound lowers disease activity without any such demonstration, the result supports the pathway, and would be more accurately reported as pathway-level evidence than as validation of the cell state. If a sufficiently selective intervention leaves pathology unchanged, that argues against developing astrocytic antigen presentation as a target. The same alternatives would have to be excluded. The absence of a usable readout, or of a selective agent, argues only for sequencing, and for investing in methods rather than abandoning the question. Longitudinal human association and selective perturbation would justify further translational testing, but would not establish clinical utility (Figure 1).What would have to be shown, and what each result would licence: a proposed validation framework. Four claims are set out in order of the evidential burden each imposes, with the evidence each would require, the conclusion that evidence would permit, and the decision that would follow, although decisions in practice also depend on tractability and unmet need. A phenotypically similar astrocyte population exists in human lesions, licensed by correlative human tissue alone. The state is clinically relevant, licensed by a longitudinal association with relapse-independent progression. The state contributes causally to pathology, licensed by cell-specific functional intervention. Treatment benefit is attributable to astrocytedirected action, licensed by the selectivity requirements. A result at one level does not licence the conclusion of another. The figure also marks whether current evidence reaches each claim in mouse, human tissue, or neither. The band carries the highest tier reached: blue, human correlative; green, mouse cell-specific; grey, neither. Each row also states the tier in words, so the marking does not depend on colour alone. The mouse tier for the causal claim rests on promoterrestricted, cell-specific rather than astrocyte-specific perturbation, in one model and one species. The framework is proposed by the authors rather than derived from the study under discussion, and the human evidence tier remains correlative throughout.Conflict of interest. The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.Funding. The authors received no financial support for the research, authorship, and/or publication of this article.Contributors and acknowledgments. ZC and ML conceived the article. ZC performed the literature search and drafted the manuscript. ML critically revised it for important intellectual content and supervised the work. Both authors approved the final version and agree to be accountable for all aspects of the work. No third-party human writing assistance was used; generative AI use is declared in the following statement.The author(s) declared that generative AI was used in the creation of this manuscript. During the preparation and revision of this manuscript, the authors used WorkBuddy, version 5.5.6 (powered by DeepSeek v4.1 flash; source: DeepSeek; accessed September 2026) to assist with drafting, language and structure editing, and compilation of reference details. The tool was used solely to assist with drafting and with language and structure editing of text written by the authors, and to assist with the compilation of reference details. After using this tool, the authors thoroughly reviewed, edited, and verified all content, and take full responsibility for the integrity and originality of the published work. No AI tools were used to generate data, to select or verify the scientific argument, or to produce the figure, which is original vector artwork prepared by the authors.Ethical approval and data availability. Not applicable: no primary, human or animal data were generated for this article, which is based entirely on previously published literature.
Source: Frontiers in Immunology
Authors: Zhufeng Chen, Ming Luo
Institutions: Wenzhou Medical University, Lishui Central Hospital
Language: English
Subfield: Clinical Neurology
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7225661875