---
title: "Single-cell study signals neuroimmune remodeling and flags SPP1, LGALS3 in neuropathic pain"
id: "frontiers-in-immunology-11-single-cell-analyses-identify-persistent-neuroimmune-remodeling-and-prioritize"
canonical_url: "https://medichelpline.com/clinical-feed/frontiers-in-immunology-11-single-cell-analyses-identify-persistent-neuroimmune-remodeling-and-prioritize"
content_type: "clinical_feed_article"
specialty: "Infectious Disease"
source_name: "Frontiers in Immunology"
source_url: "https://www.frontiersin.org/articles/10.3389/fimmu.2026.1981038"
published_at: "2026-11-04T00:00:00.000Z"
evidence_level: "Journal Feed"
license: "CC-BY-NC-4.0 / Informational Use"
---
# Single-cell study signals neuroimmune remodeling and flags SPP1, LGALS3 in neuropathic pain
## Provenance & Clinical Metadata
- **Canonical URL:** https://medichelpline.com/clinical-feed/frontiers-in-immunology-11-single-cell-analyses-identify-persistent-neuroimmune-remodeling-and-prioritize
- **Specialty:** [Infectious Disease](https://medichelpline.com/clinical-feed/infectious-disease.md)
- **Primary Source:** Frontiers in Immunology
- **Source URL:** [Original Journal Publication](https://www.frontiersin.org/articles/10.3389/fimmu.2026.1981038)
- **Published At:** 2026-11-04T00:00:00.000Z
- **Evidence Rating:** Journal Feed
## Executive GIST (TL;DR)
- The source article title indicates a **single-cell** analytic approach applied to neuropathic pain, with persistent neuroimmune remodeling reported and two molecules prioritized: **SPP1** and **LGALS3**. - The published record visible to this task is incomplete; the source document body was not provided beyond header metadata, so investigators' methods, cohort composition, sequencing platform, and statistical results were not available for extraction. - The title implies identification of cell-type–specific immune changes that remain after injury or disease, but the article text containing cell cluster definitions, timepoints, or persistence criteria was not present in the supplied source. - The authors appear to prioritize **SPP1** and **LGALS3** as candidate mediators or biomarkers, yet the supplied source did not report experimental validation approaches, expression levels, or functional assays supporting prioritization. - No details were provided about sample origin (human, animal, tissue type), single-cell modalities used (RNA, multiome, spatial), or downstream validation (immunohistochemistry, knockout/knockdown experiments), so those elements remain unspecified. - Because primary data and numeric outcomes were not available in the provided content, exact effect sizes, significance testing, and cohort demographics cannot be summarized or cited here. - Key outstanding questions left unanswered by the supplied material include: which immune cell subsets remodel persistently; how persistence was defined; whether **SPP1** and **LGALS3** were elevated or suppressed; and what therapeutic hypotheses the study supports. - The absence of the full article text prevents confirmation of claims, replication details, and clinical-translational milestones such as suggested biomarkers or drug targets that might follow from the reported findings.
## Clinical Analysis & Structured Key Points
Frontiers | Single-cell analyses identify persistent neuroimmune remodeling and prioritize SPP1 and LGALS3 in neuropathic pain ORIGINAL RESEARCH article Front. Immunol. , 04 November 2026 Sec. Multiple Sclerosis and Neuroimmunology Volume 17 - 2026 | https://doi.org/10.3389/fimmu.2026.1981038 Published in Frontiers in Immunology Multiple Sclerosis and Neuroimmunology 7 impact factor 11.3 citescore Part of a Research Topic Molecular Mechanisms and Therapeutic Targets in Neuropathic Pain Submission open 2200 views 3 articles Editor & Reviewers Edited by M Z Mingzhu Zhai Reviewed by G X Gangcai Xie M M Mahdi Mehmandoost Outline Figures and Tables Figure 1 View in article Figure 2 View in article Figure 3 View in article Figure 4 View in article Figure 5 View in article Figure 6 View in article Figure 7 View in article Figure 8 View in article Figure 9 View in article ORIGINAL RESEARCH article Front. Immunol. , 04 November 2026 Sec. Multiple Sclerosis and Neuroimmunology Volume 17 - 2026 | https://doi.org/10.3389/fimmu.2026.1981038 Single-cell analyses identify persistent neuroimmune remodeling and prioritize SPP1 and LGALS3 in neuropathic pain J L Jianhui Li † S M Shanghai Ma † K Y Kaisheng Ye † B Z Bin Zhang * Department of Anesthesiology, The First Affiliated Hospital of Jinan University, Guangzhou, Guangdong, China Article metrics View details Abstract Introduction: Persistent neuropathic pain involves lasting sensory and immune changes. We examined which expression programs persist across injury settings and whether prioritized genes are associated with pain severity. Methods: We analyzed five public single-cell or single-nucleus transcriptomic resources and two independent bulk RNA-sequencing resources, then examined tissue expression and behavior in spared nerve injury experiments. Results: Microglial expression showed a distinct early response and persistent glial-program activity. The glial signature remained elevated at day 14 and month 5 after exclusion of Spp1 and Lgals3 (false-discovery-adjusted q=0.0473 and 0.00128). Independent bulk data supported neuronal injury-related and glial programs, although candidate-gene effects varied across populations and stages. Spp1 and Lgals3 retained high priority in ranking sensitivity analyses, and their dorsal root ganglion expression correlated with mechanical hypersensitivity after adjustment for collection endpoint. Post-injury Lumogallion treatment increased the day-14 paw-withdrawal threshold (treated/vehicle geometric-mean ratio 4.08, 95% confidence interval 2.53–6.55) and reduced selected dorsal root ganglion and spinal transcripts. Structural modeling proposed an LGALS3–Lumogallion interaction for experimental testing. Discussion: These findings link persistent neuroimmune expression programs to tissue-level Spp1 and Lgals3 associations and show partial behavioral improvement with Lumogallion. Direct target-engagement studies are planned to determine whether LGALS3 contributes to this response. Graphical Abstract Single-cell analyses and independent bulk comparisons support neuronal injury-related and persistent glial expression programs. SPP1 and LGALS3 were prioritized as candidates; higher dorsal root ganglion (DRG) expression was associated with greater mechanical hypersensitivity in the local spared nerve injury (SNI) cohorts. Lumogallion treatment on days 3–13 increased the day-14 paw withdrawal threshold relative to SNI vehicle (geometric-mean ratio 4.08, 95% confidence interval 2.53–6.55; n=6 animals per group) and reduced selected DRG and spinal transcripts. The illustration summarizes expression associations and treatment effects; it does not establish a causal SPP1 pathway or LGALS3-dependent drug action. 1 Introduction Neuropathic pain can persist long after the initial nerve injury, but the molecular features that accompany its maintenance are difficult to distinguish from the acute injury response ( 1 ). Early increases in a transcript may reflect tissue damage, repair or inflammation without contributing to sustained hypersensitivity. Identifying expression programs that persist in defined tissues and injury phases can help select candidates for mechanistic investigation. Single-cell and single-nucleus studies have described injury responses in sensory neurons, peripheral glia and spinal immune populations. Peripheral nerve injury alters dorsal root ganglion (DRG) cell populations and spinal macrophages ( 2 , 3 ), and spinal microglia show distinct early and late responses after spared nerve injury (SNI) ( 4 ). The iPain atlas links nociceptor senescence to chronic pain ( 5 ); chronic constriction injury (CCI) profiling describes loss of neuronal identity and the emergence of injury-associated states ( 6 ). Comparing these resources can identify recurring expression programs while accounting for differences in tissue, platform and sampling design. SPP1 and LGALS3 are extracellular candidates of interest in these injury responses. SPP1, also known as osteopontin, has been studied in sensory and peripheral nerve injury and has functions involving integrins and the extracellular matrix ( 7 , 8 ). LGALS3, or galectin-3, has been implicated in spinal microglial activation and neuropathic pain ( 9 – 11 ). Structures of its carbohydrate-recognition domain bound to inhibitors also permit examination of candidate ligand interactions ( 12 , 13 ). These observations raise two questions: where the genes recur in injury-related expression programs and how their tissue expression relates to pain severity. We asked whether neuroimmune expression programs persist across injury settings and whether selected genes relate to mechanical hypersensitivity. Fixed signatures allowed comparison of five public single-cell or single-nucleus resources, with two independent bulk RNA-sequencing resources used to assess recurring programs and candidate expression. Expression-state analysis and injury-group classification addressed how well the profiles distinguished injury phases. An SNI tissue series examined the relationship between prioritized DRG transcripts and pain severity. A separate post-injury experiment tested the behavioral and transcriptional effects of Lumogallion, whose proposed interaction with LGALS3 was examined by structural modeling. 2 Methods 2.1 Analytical design We combined dataset-specific transcriptomic analysis, candidate prioritization, tissue quantitative PCR (qPCR) and behavioral measurements. Five single-cell or single-nucleus resources provided discovery data, and two bulk RNA-sequencing resources provided independent expression comparisons. We applied the same gene signatures within each study, then compared the resulting injury contrasts across tissues and sampling designs. Expression-state analysis and classification were retrospective. Chromatin-accessibility and trigeminal profiles from the parent iPain atlas were outside the analysis. 2.2 Data sources GSE162807 provided mouse spinal microglial libraries from naive, sham and SNI 3-day, 14-day and 5-month groups, plus three human spinal microglial libraries ( 4 ). GSE253345 contributed four pooled DRG gene-expression libraries from the iPain resource: one CCI day-14 pool, one mixed-sex control pool, one female control pool and one male control pool ( 5 ). Each pool contained three mice according to the deposited series metadata; pooling prevents animal-level inference. GSE216039 contributed four neuron-enriched DRG libraries, representing male/female sham and CCI day-7 groups ( 6 ). GSE155622 contributed Control and SNI day-14 cells from deposited matrix parts 1 and 3; parts 2 and 4 were not included ( 3 ). GSE134003 provided spinal cells collected 14 days after SNI or superficial injury (SI) in adult male mice ( 2 ). The GEO series specifies 20 mice, one SNI exclusion for DRG contamination, and 19 submitted samples (10 SI, 9 SNI). SI denotes the superficial-injury comparator used in the source study; it is not an uninjured control. Library, sex, condition and grouping metadata accompany the source tables. 2.3 Preprocessing and quality control Preprocessing used Scanpy, with version 1.11.5 specified in the reproducibility environment ( 14 ). For GSE162807, GSE253345 and GSE216039, respective minimum counts/detected genes were 250/120, 300/150 and 400/180. Mitochondrial fractions were calculated but were not used as exclusion thresholds for these three resources; no additional doublet filter or upper gene-count cutoff was applied. The GSE155622 count objects included cells with 300–12,000 detected genes and mitochondrial fractions at most 30%; GSE134003 used 200–6,500 genes and at most 25% mitochondrial reads. After cell filtering, the working subsets were capped at 1,100 cells per mouse GSE162807 library, 1,500 per human library, 5,500 per GSE253345 library and 3,800 per GSE216039 library. GSE155622 and GSE134003 were sampled by condition, cell class and source label with a nominal cap of 8,500, yielding 7,714 and 8,417 cells, respectively. Random seeds and selection logic are retained in the code. Genes detected in fewer than three sampled cells were removed before library-size normalization. Each retained cell was normalized to a fixed total of 10,000 and transformed with natural log(1+x); the normalization target was identical across libraries. Raw input barcodes, post-filter cells and sampled analysis cells are reported separately; raw barcodes in unfiltered matrices are not all called cells. Gene-name cleanup removed the GRCh38:___ and mm10:___ prefixes and made duplicate symbols unique. Human–mouse correspondence was established using Ensembl Compara release 115 from the September 2025 archive, retrieved on 24 September 2026 ( 15 ). Source stable gene identifiers were used where available; otherwise, a symbol had to resolve to a unique mouse identifier. Cross-species expression summaries and the human-supported ranking required a high-confidence one-to-one orthologue represented unambiguously in the human matrix. Other mapping classes and absent features were recorded separately rather than assigned zero expression. Mouse-only analyses retained their original gene sets. Working embeddings used Seurat-flavor highly variable genes, up to 30 principal components and the archived uniform manifold approximation and projection (UMAP) coordinates. Broad iPain labels were assigned from the maximum mean expression among predefined marker panels (threshold 0.02 log-normalized units), merging nociceptor labels into DRG neurons. The 38 cells initially assigned an astrocyte label were retained as unresolved glia because their sparse marker expression did not establish that identity. We inspected all four astrocyte-panel genes, retaining Gfap in the display despite its absence, and repeated candidate ranking after omitting these cells. Because the same markers informed the initial assignments, their expression provides descriptive support rather than independent validation of cell identity. 2.4 Expression signatures and injury-stage definitions Senescence-like, nociceptor-associated and chronic-glial scores were arithmetic means of log1p-normalized expression across available genes in fixed panels of 16, 11 and 12 genes, respectively. The panels were specified independently of fitting to iPain or the microglial time series. Projection applies the same gene list and scoring rule to each target dataset. Gene lists, coverage and missing-symbol handling are provided in Supplementary Methods S1 . Senescence-like expression does not by itself establish cellular senescence. Naive, sham and deposited control labels defined baseline comparators; SNI day 3 and CCI day 7 were early injury. SNI/CCI day 14, SNI month 5 and GSE134003 SNI day 14 defined the operational chronic-like class used in classification. These operational labels do not imply identical disease phases across tissues, models or sexes. The original chronification index combined expression with known-stage terms. Supplementary Methods S1 and Supplementary Figure S1 provide its formula and decomposition; temporal interpretation relies on the expression-only analyses below. 2.5 Expression-state analysis in spinal microglia For each of 25 conservative GSE162807 source groups, we calculated mean natural-log-transformed counts per 10,000 (log1p CP10K). We selected 2,000 genes with the highest group-level variance among genes with mean expression above 0.02, excluding mitochondrial, ribosomal and specified low-information symbol patterns. Principal component analysis (PCA) used standardized group-level expression. Nonnegative matrix factorization (NMF) used four components, nonnegative double singular-value initialization, gene-standard-deviation scaling with a 0.05 floor, a 2,000-iteration limit and seed 20260922. Loadings were normalized to unit sum and group program weights to proportions. Stage labels were excluded from both feature selection and factorization; all four programs are reported. A complementary cell-level diffusion graph excluded the curated signature and marker genes, including Spp1 and Lgals3. Two thousand highly variable genes were selected using source-library batches, followed by sparse scaling, 30 principal components, 30-neighbor graph construction and 15 diffusion components. Diffusion pseudotime used 10 components and a naive cell nearest the naive median in PCA space as its root. The baseline label therefore orients the graph but does not enter its expression features. Group medians summarize diffusion coordinates and pseudotime. Time association was assessed separately within the 10 injured groups. These cross-sectional analyses describe expression-state relationships, not cell lineage or longitudinal transitions in individual animals. At each injury stage, we compared four signature scores (glial, target-excluded glial, senescence-like and nociceptor) and three genes (Spp1, Lgals3 and Apoe) with pooled naive/sham groups. Each ordinary least-squares model included a sex covariate. Heteroskedasticity-consistent type-3 (HC3) standard errors and residual-degree-of-freedom t intervals were used; Benjamini–Hochberg (BH) adjustment covered all 21 comparisons. The target-excluded glial panel removes both Spp1 and Lgals3. Group counts were 4 naive, 11 sham, 3 day-3, 4 day-14 and 3 month-5 groups. Grouping accounts for related libraries; animal identities could not be independently verified for all groups. 2.6 Signature and integration sensitivity Sensitivity analyses varied normalization targets across 1,000, 10,000 and 100,000 on the same retained gene/cell space, gave source groups equal weight, omitted individual source groups where both conditions remained represented, and removed the tested target from glial correlation scores. We assessed sensitivity to quality filtering by applying uniform thresholds to all six species-specific datasets: at least 400 counts, 300–6,500 detected genes and at most 25% mitochondrial counts. Reconstruction from the available raw inputs preserved source conditions and group definitions. Within each library or source label, originally sampled cells that passed the new thresholds were retained first. Additional eligible cells were selected in a fixed barcode order, independently of outcomes, up to the original sampled count. The gene sets retained after filtering and the normalization procedure were held fixed. We compared cell retention, cell-type composition, expression contrasts with equal weighting of source groups, candidate ranks and DRG-neuron integration between QC schemes. These thresholds provide a common sensitivity test across technologies; they do not establish an optimum or recover cells removed by the source studies before deposition. Supplementary Methods S8 describes the reconstruction. An alternative integration analysis was restricted to DRG-neuron annotations from GSE253345, GSE216039 and GSE155622. At most 1,200 cells per conservative source group were selected; 2,000 study-aware highly variable genes from the shared gene space and 30 principal components were used for Harmony (theta=2, maximum 20 iterations, seed 20260922) ( 16 ). Scores were transferred from each query cell’s 50 nearest cells in other studies before and after correction. Direct scores on the identical subset provided a comparator. Condition labels were excluded from integration and neighbor selection. This analysis tests sensitivity to the expression representation. When study and condition are confounded, correction may remove biological variation as well as batch effects. 2.7 Expression contrasts and candidate communication Expression contrasts were calculated as case-minus-comparator differences in mean log1p expression. The original cell-weighted summaries are retained alongside summaries giving equal weight to conservative source groups. The source-table column logfc contains this mean log-expression difference. It is not an estimated log2 fold change from a pseudobulk differential-expression model. We evaluated 12 manually specified ligand–receptor pairs in the context of published communication resources ( 17 – 19 ). For each pair, ligand evidence L was the mean expression difference across three DRG-neuronal contrasts and receptor evidence R the mean across four glial/spinal contrasts. The raw score, , was min–max scaled across the 12 pairs. Of 84 requested role-by-contrast entries, 83 were observed; the missing Ccl3 value in iPain neurons was assigned zero and is identified in the coverage table. This descriptive score uses individual receptor subunits, including separate Spp1–Itgav and Spp1–Itgb1 entries. It does not reconstruct receptor complexes or assign inferential significance, and no versioned LIANA/OmniPath inference was run. 2.8 Target prioritization The candidate pool combined fixed panels, druggability priors, the 150 genes with the largest positive effects per contrast and the 150 genes most correlated with sampling stage. The original weighted score allocated 25, 20, 15, 15, 10, 10 and 5 units to progression, senescence consistency, crosstalk, cross-model support, cell-type concentration, model interpretation and human-expression/druggability context. A multiplier of , where A is min–max-scaled actionability, modified the total. Three comparison genes from an earlier opioid-induced hyperalgesia analysis (Tyrobp, P2rx4 and P2ry1) originally received a separate 0.65 multiplier. Candidate scores excluded the ten-unit model component and used the same actionability rule for every gene. The human-supported ranking included 717 of 913 candidate features that met the one-to-one mapping and human-feature requirements. All six entries in three duplicate-alias pairs were excluded, without selecting or merging a representative. Cell-type concentration reflected the revised unresolved-glia label. The comparison of name lookup with standardized mapping used this same eligible set and common scaling limits. A sensitivity ranking of all 913 candidates excluded human expression from both the weighted score and the actionability multiplier while retaining the druggability prior. Supplementary Methods S2 and S8 give the formulas, and Supplementary Figure S7 shows the components. These scores are heuristic; SHAP values from independently evaluated classifiers do not contribute to them. 2.9 Source-group classification and SHAP interpretation Classification used one mean-expression row per conservative source group. The primary task comprised 25 GSE162807 mouse microglial groups, including seven chronic-like groups; baseline and early injury formed the negative class. A pooled task comprised 54 groups, including 18 positives, from those microglia, the three DRG-neuronal resources and GSE134003 microglia. Repeated/split microglial title stems 103, 106 and 307 were combined, and GSE155622 records were grouped by condition where animal identity could not be resolved. Source groups may represent libraries
## Related Clinical Research

- [Herpesvirus–autoantigen antibody networks linked to chronic disease in large biobank profiling](https://medichelpline.com/clinical-feed/nature-immunology-2-large-scale-antibody-reactome-profiling-identifies-herpesvirus-autoantigen.md)
- [Herpesvirus Antibody Networks Linked to Autoimmunity and Chronic Disease](https://medichelpline.com/clinical-feed/nature-immunology-0-herpesvirus-exposures-map-to-self-reactivity-networks-in-chronic-disease.md)
- [Strain-specific and HIV-driven humoral responses to Epstein-Barr virus mapped by PhIP-Seq](https://medichelpline.com/clinical-feed/pubmed-42841646.md) (DOI: 10.1128/jvi.01043-26)
- [Cell Therapy's Role in Humoral Alloimmunity Post-Transplantation](https://medichelpline.com/clinical-feed/frontiers-in-immunology-11-cell-therapy-and-humoral-alloimmunity-post-transplantation-a-review-of.md)
- [Age-specific contact matrices across 28 EU/EEA countries during COVID-19: linking NPIs to social m](https://medichelpline.com/clinical-feed/medrxiv-21-reconstructing-age-specific-social-contact-patterns-across-28-eu-eea-countries.md)

## Navigation
- [← Back to Infectious Disease Feed](https://medichelpline.com/clinical-feed/infectious-disease.md)
- [← All Clinical Specialties](https://medichelpline.com/clinical-feed.md)
## Medical & Regulatory Disclaimer

> [!CAUTION]
> MedicHelpline content is structured for research, educational, and professional discovery purposes. It does not constitute individual medical advice, clinical diagnosis, or treatment recommendations.
> Always verify dosing, contraindications, and regulatory alerts against official product labeling and primary regulatory sources before clinical decision-making.