Single-cell RNA sequencing (scRNA-seq) has significantly advanced the understanding of T cell heterogeneity in NMOSD. Studies have identified distinct T cell populations based on their gene expression profiles, including Naïve CD4+ and CD8+ T cells, Memory CD4+ and CD8+ T cells, Effector Memory CD8+ T cells, Cytotoxic CD8+ T cells, Exhausted CD8+ T cells, CD4-CD8- double-negative T cells, and Mucosal-associated invariant T (MAIT) cells (47, 99).

In NMOSD patients, the proportion of total T cells in peripheral blood mononuclear cells (PBMCs) is decreased compared to healthy controls. This reduction is primarily attributed to a decrease in CD4+ T cells, whereas the proportion of CD8+ T cells within the T cell population is notably increased. This shift suggests a specific immune dysregulation rather than a general T cell deficiency. MAIT cells, which possess cytotoxic and pro-inflammatory functions, show a significant proportional increase in NMOSD patients following steroid therapy (47).

Regulatory T cells (Tregs), characterized by the expression of the transcription factor FOXP3, are essential “brakes” of the immune system, maintaining self-tolerance by suppressing the activation and proliferation of autoreactive lymphocytes (100). A consistent finding in NMOSD is that Tregs are both numerically deficient and functionally impaired, particularly during disease relapses (101, 102). Evidence from mouse models highlights their importance; Treg depletion exacerbates astrocyte loss and demyelination, while adoptive transfer of Tregs attenuates brain damage (102). This Treg dysfunction, potentially driven by epigenetic modifications at the FOXP3 locus or a pro-inflammatory cytokine milieu that inhibits their suppressive capacity, is considered a key checkpoint failure that permits the development of autoimmunity (103).

T helper (Th) cells are master regulators of the adaptive immune response. In NMOSD, the balance between different Th subsets is skewed towards a pro-inflammatory phenotype. Th17 cells, characterized by the production of IL-17A, IL-17F, and IL-22, are potent inducers of tissue inflammation and are consistently found at elevated levels in the blood and CSF of NMOSD patients (43, 48, 104). IL-17 acts on endothelial cells of the BBB to disrupt tight junctions and promotes the recruitment of other inflammatory cells, such as neutrophils, into the CNS (105). Multi-omics studies have elucidated the signaling pathways driving Th17 differentiation in NMOSD, highlighting a critical role for IL-6, which is produced by B cells and innate immune cells in response to stimuli like type I interferons (43, 106). This establishes a pathogenic feedback loop where B cells fuel the differentiation of Th17 cells, which in turn promote CNS inflammation (106). Th1 cells, which produce IFN-γ, also contribute to the inflammatory environment, and a particularly pathogenic subset known as Th17.1 (or ex-Th17) cells, which co-express markers of both Th1 and Th17 lineages (e.g., produce both IFN-γ and IL-17), has been identified as being highly enriched in NMOSD (107).

T follicular helper (Tfh) cells, found in secondary lymphoid organs, are specialized providers of help to B cells. Their interaction with B cells, mediated through molecules like CD40L, ICOS, and the cytokine IL-21, is essential for germinal center formation, affinity maturation, and the generation of long-lived plasma cells and memory B cells (108). Tfh cells are expanded in the circulation of NMOSD patients, and their numbers correlate with AQP4-IgG titers, underscoring their critical role in driving the pathogenic humoral response (109).

Each T cell expresses a unique T-cell receptor (TCR) that recognizes a specific peptide-MHC complex. The collective diversity of all TCRs in an individual constitutes the TCR repertoire. High-throughput TCR sequencing (TCR-seq) has provided profound insights into the T-cell response in NMOSD. Compared to healthy individuals, NMOSD patients exhibit a significantly contracted and less diverse TCR$\beta$ repertoire, characterized by prominent oligoclonal expansions (110). This indicates that a limited number of T-cell clones are undergoing massive antigen-driven proliferation in response to specific epitopes. These expanded clones are found within pathogenic effector memory and cytotoxic T-cell subsets, and their frequency can decrease following effective therapy, suggesting they could serve as a dynamic biomarker of disease activity (47).

A leading hypothesis for the initiation of autoimmunity is molecular mimicry, where a foreign peptide from an infectious agent shares sufficient structural similarity with a self-peptide to trigger a cross-reactive T-cell response. Integrated TCR-seq and functional studies have provided compelling evidence for this mechanism in NMOSD. Evidence linking Cytomegalovirus (CMV) infection to AQP4-IgG+ NMOSD has been derived from integrated TCR/BCR repertoire analysis and functional validation (110). This work suggests a specific mechanism for disease initiation. A T-cell antigenic epitope of CMV was found to be identical to a sequence within AQP4, and its corresponding CDR3 sequence closely resembled an NMOSD-TCR sequence (110). Shared core peptides that partially overlap with Clostridium perfringens epitopes, previously reported to cross-react with AQP4, have also been identified, suggesting a broader mechanism of microbial molecular mimicry (44, 110).

Transcriptomic analyses reinforce this link by showing upregulated genes related to viral infection and innate immune pathways in NMOSD patients. Gene Set Enrichment Analysis (GSEA) further demonstrates significant activation of interferon-related and T-cell receptor signaling pathways. Functionally, in vitro experiments show that CD4+ T cells from untreated AQP4-IgG+ NMOSD patients are significantly activated upon stimulation with CMV peptide pools, a response not observed in healthy controls. Genetic factors, such as specific HLA genes (HLA-DPB1 05:01* and HLA-DRB1 03:01*), appear to influence susceptibility, suggesting that CMV infection may be a necessary but not sufficient trigger for the disease. These findings highlight a plausible pathway for disease initiation where environmental triggers (microbial infections) and genetic susceptibility (HLA type) converge to activate autoreactive T cells that provide help to AQP4-specific B cells (110).

The B-cell lineage is a continuum of developmental stages, from naïve B cells to highly specialized antibody-secreting cells (ASCs). scRNA-seq has mapped this landscape in NMOSD, revealing significant shifts in the B-cell compartment (78, 111). Compared to healthy controls, NMOSD patients exhibit an increased proportion of IgG+ plasma cells, IgA+ plasma cells, total plasma cells, and memory B cells, indicative of a robust humoral immune response (47, 112). The frequency of these plasmablasts often correlates with disease activity and AQP4-IgG titers, and they are considered a major source of the pathogenic antibodies during relapses (113).

Beyond ASCs, multi-omics has highlighted the importance of other B-cell subsets. Memory B cells, which persist after an initial immune response and can rapidly differentiate into ASCs upon re-exposure to antigen, are expanded and transcriptionally primed for activation in NMOSD patients (111). These cells likely represent a persistent reservoir of autoimmunity that contributes to relapses. Furthermore, studies have identified an expansion of so-called “atypical” B cells, including double-negative B cells and age-associated B cells (ABCs) (114, 115). These subsets have been implicated in other systemic autoimmune diseases and are characterized by the expression of transcription factors like T-bet, a pro-inflammatory phenotype, and a lower threshold for activation, potentially contributing to the cycle of inflammation in NMOSD.

The remarkable efficacy of therapies targeting the B-cell surface protein CD20 (rituximab, ocrelizumab) or CD19 (inebilizumab) has revolutionized NMOSD management (36, 116, 117). Multi-omics studies are helping to elucidate their precise mechanisms of action. By depleting circulating B cells, these therapies remove not only the precursors of AQP4-IgG-secreting plasmablasts but also a critical population of antigen-presenting cells required to sustain the autoreactive T-cell response, and a key source of pro-inflammatory cytokines like IL-6 (43, 118). This multifaceted impact likely explains their high efficacy.

Single-cell BCR sequencing (scBCR-seq) has revealed critical insights into antibody production in NMOSD. Oligoclonal expansions of BCRs are consistently observed, particularly after therapy, indicating a focused and persistent antigen-driven response. scBCR-seq data show increased proportions of immunoglobulin heavy chain gamma (IGHG) and alpha (IGHA) in NMOSD patients, signifying a class switch towards IgG and IgA production. Clonal BCRs observed after steroid treatment consist primarily of IgA and IgG subtypes, with widespread clonal expansion observed across naïve B, plasma, and memory B cell compartments (77).

The pathogenic autoantibody in NMOSD, AQP4-IgG, exhibits consistently greater binding affinity to the M23 isoform of AQP4 compared to the M1 isoform. This differential binding is attributed to the assembly of the M23 isoform into orthogonal arrays of particles (OAPs). Experiments with varying M1:M23 ratios and OAP-disrupting mutants of M23 have confirmed this conclusion. Furthermore, purified Fab fragments of NMO-IgG showed similar binding patterns, indicating that structural changes in the AQP4 epitope upon array assembly, not bivalent cross-linking, are responsible for the greater binding affinity (119). Analysis of these BCR sequences shows high rates of somatic hypermutation, particularly in the complementarity-determining regions (CDRs) that form the antigen-binding site. This is the molecular signature of an affinity-matured, T-cell-dependent immune response, consistent with the high-affinity binding of AQP4-IgG to its target (48, 120). These studies also confirm a strong bias towards the use of IgG1 and IgA isotypes within the expanded clones, the very isotypes known to be pathogenic or enriched in NMOSD (88).

B cells also function as professional antigen-presenting cells (APCs), further stimulating T cell activation. The interaction between B cells and T cells is a critical aspect of the adaptive immune response in NMOSD. For instance, IFN-I stimulates B cells to produce IL-6, which then drives pathogenic Th17 differentiation. The efficacy of B cell-depleting therapies, such as anti-CD20 (rituximab) and anti-CD19 (inebilizumab) monoclonal antibodies, in reducing relapses underscores the pathogenic role of B cells (111). The observation that IL-6 and IL-17 levels are lower in patients on anti-CD20 therapy suggests a link between B cell depletion and the reduction of pro-inflammatory cytokines, providing a molecular basis for the therapeutic benefit.

Computational tools designed for multi-omics integration are beginning to link findings across different data types. For example, by integrating genomic data (HLA risk alleles) with TCR-seq data, researchers can predict which specific self- or microbial peptides are likely to be presented by those risk alleles to drive the expansion of pathogenic T-cell clones (65). Similarly, integrating transcriptomic data from T cells with metabolomic data from the same patients can reveal how the pro-inflammatory gene expression programs of Th17 cells are fueled by specific metabolic pathways, such as aerobic glycolysis (92, 121). Identifying and targeting these metabolic dependencies offers a novel therapeutic strategy.

A key application of single-cell transcriptomics is the inference of intercellular communication networks. By analyzing the expression of ligand-receptor pairs across all cell types in a sample, computational tools like CellChat or NicheNet can construct a map of the signaling interactions that shape the immune response (122). In NMOSD, this approach has been used to model the critical cross-talk between Tfh cells and B cells, identifying IL-21 and CD40L as key signals driving B-cell differentiation (109). It can also reveal how astrocytes and microglia, the resident cells of the CNS, respond to infiltrating immune cells and, in turn, produce chemokines and cytokines that amplify the inflammatory cascade (83, 123). These communication maps provide a rich source of potential therapeutic targets aimed at disrupting these pathogenic cellular conversations.

From Discovery to Clinical Utility. The ultimate goal of multi-omics research is to improve patient care. This requires translating complex datasets into clinically actionable tools. A major focus is on biomarker development. The high dimensionality of omics data provides fertile ground for discovering novel diagnostic, prognostic, and predictive biomarkers (124). For example, a specific signature of clonally expanded TCRs in the blood could serve as a highly sensitive biomarker of impending relapse. A proteomic or metabolomic signature could predict which patients are likely to respond to a specific therapy, enabling the development of companion diagnostics for personalized treatment selection (58, 90). While many candidate markers have been identified, the path to clinical validation is long and requires large, prospective, longitudinal patient cohorts. Currently, fluid biomarkers like serum GFAP and neurofilament light chain (NfL), which reflect astrocyte and neuronal injury, respectively, are the most advanced and are being incorporated into clinical trials as measures of disease activity and treatment response (125, 126).

Despite technological advances, the paraneoplastic nature of some NMOSD cases presents fundamental clinical challenges that multi-omics has only just begun to address.

The Overarching Diagnostic Dilemma: Idiopathic vs. Paraneoplastic: The most critical shortcoming is the inability to reliably distinguish idiopathic from paraneoplastic NMOSD at disease onset. While AQP4-IgG is a superb diagnostic marker for NMOSD, its presence reveals nothing about the trigger (127). Clinicians are left to rely on demographic risk factors and extensive screening to search for an occult malignancy. Multi-omics has uncovered thousands of molecular changes, but has not yet consolidated these findings into a clinically validated biomarker panel that can calculate a “paraneoplastic risk score,” leaving a crucial diagnostic and prognostic gap (46). For example, the concept of using high-throughput sequencing of T-cell receptors (TCR) and B-cell receptors (BCR) to diagnose pNMOSD is immunologically elegant. The technique allows for the identification of clonally expanded populations of lymphocytes from peripheral blood or CSF (110). In a paraneoplastic context, such an expansion could theoretically represent the specific lymphocyte clones that are reacting to the tumor antigen (e.g., AQP4), thereby providing a molecular fingerprint of the paraneoplastic response (110). However, the time required to perform TCR/BCR sequencing and bioinformatic analysis is substantial. Commercial and academic laboratories typically report turnaround times of 14 days or longer. At this time a clinical report stating “TCRβ chain clonal expansion of unknown significance” provides no immediately actionable information for the treating neurologist, we still have a long way to go.

Limitations of Current Therapies for a Tumor-Driven Disease: Existing NMOSD treatments—B-cell depletion, complement inhibition, or IL-6 receptor blockade—are designed to suppress the autoimmune response but are fundamentally inadequate for paraneoplastic cases as they do not address the root cause: the tumor (110). This approach is akin to managing the smoke while ignoring the fire. Multi-omics reveals the immense heterogeneity of the anti-AQP4 response, but the challenge is no longer just suppressing it; it’s understanding how to stop the tumor from continuously stimulating it. The ultimate goal for these patients is not merely to re-establish immune tolerance, but to achieve this by finding and eliminating the malignancy that first broke it (42).

Bridging the Gap Between Data and Pathophysiology: While we can now generate vast multi-omics datasets, translating these into a coherent understanding of disease triggers remains a major hurdle. For instance, though studies have proposed a link between CMV infection and NMOSD via molecular mimicry, the precise events that initiate loss of tolerance in only a fraction of infected individuals are unknown (110). The lack of integrated, longitudinal data—tracking patients from the emergence of the tumor, through the onset of neurological symptoms, to post-cancer treatment—makes it exceedingly difficult to pinpoint the exact immunological factors that conspire to initiate paraneoplastic NMOSD.

While significant questions remain, multi-omics research is paving the way for clinical applications that could revolutionize the management of patients with suspected or confirmed paraneoplastic NMO.

Precision Biomarkers for Cancer Detection: The detailed cellular and molecular signatures of NMOSD can be repurposed as a “liquid biopsy” for an underlying malignancy. Specific TCR or BCR clonotypes that expand in response to tumor-expressed AQP4 could be developed into highly sensitive assays not just for monitoring NMOSD activity, but for detecting an occult cancer and monitoring its eradication following therapy. Proteomic or metabolomic profiles that differ between idiopathic and paraneoplastic cases could form the basis of companion diagnostics to trigger immediate and targeted cancer screening, personalizing the diagnostic workup (48). Although the clinical value of these methods is still limited, we can try to explore a combination of methods as a paraneoplastic risk score which could include:

Identification of Novel Therapeutic Targets: Multi-omics analyses have moved beyond implicating entire cell populations to pinpointing specific molecules and pathways as potential drug targets. The identification of the IFN-I → B-cell → IL-6 → pathogenic Th17 axis provides several nodes for targeted intervention beyond IL-6 itself (106).

The management of pNMOSD is fundamentally different from that of iNMOSD and requires tight collaboration between neurologists and oncologists. The central therapeutic principle must be that oncologic therapy is the definitive neurologic therapy. Successful treatment of the underlying malignancy—whether by surgical resection, chemotherapy, or radiation—frequently leads to stabilization of the neurological disease, a reduction in relapse rates, and, in some documented cases, a decrease in or complete seroreversion of AQP4-IgG titers (13). Conversely, relying on immunosuppression alone while the tumor remains untreated is often insufficient, with patients continuing to experience relapses and neurological decline (16). The identification of an AQP4-expressing tumor in a patient with NMOSD is a critical finding that should be treated as a therapeutic emergency. The immediate priority is to control the acute CNS inflammation and prevent further irreversible damage. This is achieved with standard acute NMOSD therapies. As soon as the patient is medically stable, definitive tumor-directed therapy should be pursued with urgency. In the future, understanding how tumor cells present AQP4 and interact with immune cells could enable the development of next-generation therapies that block the initial priming of autoreactive lymphocytes, potentially preventing the neurological syndrome entirely without requiring broad immunosuppression.

Realizing the full potential of multi-omics in paraneoplastic NMOSD will require a concerted effort to investigate both sides of the disease—the tumor and the brain.

Integration of Spatial Data from both Tumor and CNS: The future lies in comparing the spatial dynamics of the immune response in two distinct locations: the primary tumor and the CNS lesions. Spatial transcriptomics can map the interactions between AQP4-expressing cancer cells and immune cells, and compare that “immune synapse” to the one causing astrocyte damage in the brain. Combining this with longitudinal profiling of blood and CSF from large patient cohorts will be essential for building a complete, dynamic model of the disease (128).

Functional Validation in Paraneoplastic-Specific Models: Computational findings must be validated in advanced preclinical models that recapitulate the entire disease process. This requires the development of humanized mouse models that both bear an AQP4-expressing human tumor and are susceptible to developing the subsequent neurological autoimmune disease. Such models are crucial for testing therapies aimed at the tumor and observing the downstream effects on neuroinflammation, thereby de-risking new therapeutic strategies (129).

AI-Powered Clinical Decision Support for Dual-Disease Management: The complexity of paraneoplastic NMO necessitates the use of AI for true clinical integration. Future decision support tools should integrate a patient’s multi-omic profile, their HLA type, their clinical data, and tumor genomics to provide a real-time, data-driven “paraneoplastic probability score.” Such a tool could guide clinicians on when to initiate cancer screening, which organs to focus on, and how to select therapies that address both the oncologic and neurologic aspects of the disease. The primary obstacle to developing such a model is the acquisition of a suitable dataset. Given the rarity and clinical heterogeneity of pNMOSD, this would necessitate a large-scale, international collaboration to assemble a curated, multi-modal database.54 The required data modalities would include Structured Clinical Data, Imaging Data, Serology Data and Omics Data (130). The low prevalence of pNMOSD means that any single-center dataset would be too small and prone to overfitting (131). A necessary strategy to overcome this would be federated learning, a technique where the model is trained across multiple institutions on local data without the need to share the raw, sensitive patient information, thus preserving privacy while building a more robust and generalizable model.

In conclusion, the continued application of multi-omics to NMOSD provides a powerful roadmap for solving the unique clinical challenges posed by its paraneoplastic variant. By focusing on the tumor as the origin of the autoimmune cascade, we can translate discoveries into validated clinical applications that bridge the gap between oncology and neurology, moving closer to an era of preventative and personalized medicine for patients with this devastating condition.