Eupatilin ameliorates spinal cord injury by inhibiting damage-associated microglia and optimizing the regenerative microenvironment

Summary

Microglia represent critical therapeutic targets in spinal cord injury (SCI), with damage-associated microglia (DAM) playing key roles in neuroinflammation and tissue repair. Through integrated in-silico analysis of single-cell RNA sequencing (scRNA-seq) and microarray datasets, we identified DAM subsets specific to acute SCI characterized by hub genes Fcer1g, Grn, and Gusb. Using a C57BL/6 mouse spinal cord contusion model, we validated increased DAM accumulation post-injury and demonstrated their propensity to transition toward homeostatic microglia (MG2). Eupatilin treatment promoted DAM-to-MG2 differentiation, as confirmed through bulk and scRNA-seq analyses, revealing supportive gene expression changes. These findings establish DAM as functionally distinct microglial populations in acute SCI and identify Eupatilin as a therapeutic agent that facilitates beneficial microglial polarization. This work provides mechanistic insights into microglial dynamics during SCI and suggests targeted modulation of DAM-to-MG2 transitions as a promising therapeutic strategy for promoting inflammation resolution and functional recovery.

Highlights

• •DAM subsets with hub genes Fcer1g, Grn, and Gusb are specific to acute SCI
• •Homeostatic microglia (MG2) naturally transition toward the DAM phenotype following SCI
• •Eupatilin promotes beneficial DAM-to-MG2 differentiation in SCI mouse model
• •Targeted DAM modulation represents a therapeutic strategy for SCI recovery

Neuroscience; Immunology

Introduction

Spinal cord injury (SCI) is one of the most common and severe conditions affecting the central nervous system (CNS), often leading to permanent motor dysfunction and prolonged disability.^1,2 SCI initiates a complex cascade of inflammatory responses at the injury site, involving the activation of various cell types and the release of inflammatory mediators.² Microglia, in particular, play a crucial role in shaping the inflammatory environment and influencing the recovery process after SCI. Traditionally, microglia have been thought to differentiate from a homeostatic state (M0) into pro-inflammatory (M1) and anti-inflammatory (M2) phenotypes with distinct spatiotemporal patterns following SCI.^3,4,5 However, prolonged neuroinflammation and inadequate anti-inflammatory responses result in a persistently inflammatory microenvironment that hinders tissue repair.⁶ As a result, targeting microglial polarization to shift the balance toward the reparative M2 phenotype has emerged as a promising therapeutic strategy. However, with deeper insights into microglial biology, it has become clear that activated microglia cannot be simply categorized into M1/M2 phenotypes.⁷

The development of single-cell technology has revealed that microglia usually co-express M1 and M2 markers rather than polarize to either of M1/M2 phenotypes.^8,9 Recent studies indicate that some microglia display an array of unique molecular markers and functional capacities that extend beyond the M1/M2 framework, suggesting hybrid or transitional states that reflect the complexity of their role in injury response.^10,11 Notably, a subset of microglia in neurodegenerative conditions, termed disease-associated microglia, exhibit a unique transcriptional and functional signature and play critical roles in the progression of diseases such as Alzheimer disease (AD), multiple sclerosis (MS), amyotrophic lateral sclerosis (ALS), and aging.^{12,13,14,15,16} These findings suggest that context-dependent microglia could potentially serve as therapeutic targets in these corresponding conditions. However, in the context of SCI, the existence and specific roles of damage-associated microglia (DAM) remain underexplored.

Eupatilin, an active ingredient isolated from Artemisia argyi, exhibits several pharmacological activities including anti-oxidative, anti-inflammatory, and anti-apoptosis effects in various diseases.^{17,18,19,20,21} Our previous study demonstrated that Eupatilin inhibits microglial activation following intracerebral hemorrhage.¹⁸ However, it remains unclear whether Eupatilin modulates the function of DAM in SCI.

In this study, we identified DAM subpopulations in SCI mice through in-silico analysis of single-cell RNA sequencing (scRNA-seq) and transcriptome microarray data, revealing distinct changes compared to homeostatic microglia. These findings were further validated in a thoracic SCI mouse model. Using molecular docking, we hypothesized that Eupatilin could modulate these DAM populations. To explore its therapeutic potential, we administered Eupatilin via intrathecal injection in the thoracic SCI model, confirming its inhibitory effect on DAM and its promising role in promoting spinal cord repair (Figure 1).

Figure 1.

Overview of workflow and the mechanism of Eupatilin in treating SCI

The upper panel delineates the design of the in-silico data analysis pathway, outlining the methodologies employed to investigate the potential therapeutic impacts of Eupatilin on SCI.

The left lower panel delineates the workflow of the animal experimentation conducted in this study. The right lower panel elucidates the mechanism underlying the therapeutic effects of Eupatilin on SCI, as confirmed by this study. Specifically, Eupatilin exerts its effects by inhibiting damage-associated microglia (DAM)-related key genes, thereby promoting the transition of DAM to a homeostatic microglial phenotype. This transition, in consequence, alters the expression profile of cytokines, enhances the regenerative microenvironment, and facilitates neural regeneration and functional recovery following SCI.

Our findings provide an insight into the critical role of DAM in SCI and highlight Eupatilin as a potential therapeutic agent for modulating microglial responses, offering a strategy to improve spinal cord regeneration and recovery.

Results

scRNA-seq reveals microglial subpopulation dynamics in SCI

To investigate changes in key cellular subpopulations within the microenvironment after SCI, we reanalyzed publicly available scRNA-seq data (GSE162610). This dataset includes spinal cord tissue from female C57BL/6J mice (8–10 weeks old) across three conditions: Sham (three replicates), 3 days post-injury (dpi) (two replicates), and 7dpi (two replicates). After rigorous quality control, we obtained a total of 45,168 high-quality cells, distributed as follows: Sham (12,488 cells), 3dpi (17,491 cells), and 7dpi (15,189 cells) (Figures 2A and S1). Using classical marker genes and the SingleR annotation tool, we identified 11 common cellular subpopulations (Figure 2B). Among these subpopulations, the SCI microenvironment was found to be predominantly composed of immune cells, including microglia and macrophages, along with astrocytes and endothelial cells (Figure 2C). As resident immune cells of the CNS, microglia exhibited strong communication probabilities with other cells both before and after injury, particularly with astrocytes, macrophages, and monocytes, as revealed by CellChat cell-cell communication analysis (Figures 2D and 2E).

Figure 2.

Single-cell transcriptomic analysis reveals microglial subpopulation dynamics in SCI

(A) UMAP visualization of 45,168 cells across Sham, 3dpi, and 7dpi conditions. Cells are color coded by annotated cell types.

(B) Volcano plot showing expression of representative marker genes used to annotate 11 cell subpopulations.

(C) Proportional composition of cell types across Sham, 3dpi, and 7dpi groups, indicating immune cells dominate the SCI microenvironment.

(D) CellChat analysis of cell-cell communication probabilities.

(E) CellChat analysis of internation weight showing microglial interaction networks after injury.

(F) UMAP of microglia subclusters (clusters 0–6) identified in SCI samples.

(G) Quantification of clusters 1 and 2 proportions across conditions, showing significant increases at 3dpi and 7dpi.

(H) Expression of damage-associated microglia (DAM) markers (Lyz2, Igf1, and Lpl) within microglia in SCI.

(I) UMAP of microglia subclusters in samples across Sham, 3dpi, and 7dpi conditions. Cells are color coded by annotated cell types.

(J) Proportional representation of DAM and other microglial clusters across groups, with marked expansion of DAM post-injury.

(K) Functional enrichment analyses (GO) and cluster heatmap of microglial clusters.

Focusing on microglia, we initially subdivided this population into seven clusters (cluster 0–cluster 6) using Seurat-based unsupervised clustering (Figures 2F and S2). Notably, cluster 1 and 2 showed a significant increase in proportion at 3dpi and 7dpi compared to the Sham group (Figure 2G). Interestingly, classical markers such as Lyz2, Igf1, and Lpl, known to define DAM in AD, were highly expressed in these clusters (Figure 2H). We also found that their numbers increased from 131 cells in the Sham group to 2,571 cells at 3dpi and 3,271 cells at 7dpi (Figures 2H–2J). Based on these characteristics, we functionally re-annotated clusters 1 and 2 as DAM subpopulations in SCI, while the remaining clusters were re-defined as MG0-MG3 according to their distinct biological properties (Figures 2I–2K). DAM, characterized by high expression of Spp1, Fabp5, and Lgals3, was implicated in lipid metabolism and energy regulation processes through gene ontology (GO) analysis (Figure 2K). Additionally, KEGG pathway analysis highlighted DAM involvement in apoptosis and ferroptosis, suggesting their role in promoting pathological progression following SCI (Figure S3). Simultaneously, we observed a marked reduction in MG2 subpopulations following injury. These cells were characterized by high expression of Jun, Fos, and Siglech and were functionally associated with the regulation of hematopoiesis and RNA stability (Figure 2K). These features suggest that MG2 represents homeostatic microglia. In summary, our analysis identifies DAM as a potentially critical regulatory subpopulation in SCI progression, offering an insight into their role in the injury microenvironment.

Dynamic plasticity and pro-inflammatory roles of DAM in SCI

To further investigate the heterogeneity of DAM and other microglial subpopulations in SCI, we performed pseudotime trajectory analysis using Monocle2 (Figures 3A–3C). The computational trajectory analysis suggested potential transcriptomic relationships between microglial subpopulations, with trajectories indicating possible transitions from DAM through intermediate states over time (Figures 3A and 3C). Specifically, the analysis revealed that DAM may transition through intermediate states (MG0, MG1, and MG3) before reaching transcriptomic profiles similar to the homeostatic microglial subpopulation MG2, suggesting potential plasticity and dynamic relationships following injury (Figures 3A and 3B). A pseudotime heatmap further highlighted the key genes involved in these potential differentiation processes (Figure S4). Validation using CytoTRACE confirmed DAM’s high differentiation potential and plasticity (Figures 3D–3F). Although MG2 cells also exhibit some differentiation potential, the highly undifferentiated and plastic state of DAM after SCI may enable them to adapt quickly to the injury environment and potentially transition into homeostatic microglia, characterized by lower gene expression diversity.

Figure 3.

Pseudotime trajectory, regulatory modules, and cell-cell communication analysis of DAM in SCI

(A–C) Pseudotime trajectory analysis of microglial subpopulations using Monocle2. (A) Cells are ordered along the pseudotime axis, transitioning from DAM (early state) to MG2 (homeostatic state). (B) Subcluster assignments across the trajectory. (C) Group-level comparisons showing progression from SCI to Sham microglial states.

(D) Predicted differentiation potential of microglial subtypes by CytoTRACE, highlighting DAM as the most undifferentiated and plastic state.

(E and F) CytoTRACE scoring shows spatial and trajectory-level differentiation across subclusters.

(G) HdWGCNA analysis of single-cell gene expression data with a soft power threshold of 8, identifying five distinct gene modules.

(H) Gene importance (MEEs) for each module.

(I) Module expression patterns across microglial subpopulations.

(J) GO enrichment analysis of blue module genes.

(K) Interaction weights for DAM with other cell types pre- and post-injury, highlighting distinct signaling changes.

(L) Upregulated pro-inflammatory pathways (e.g., TNF, CXCL, and TWEAK) and downregulated survival pathways (e.g., PDGF, TGF-β, and MIF) in the injury microenvironment.

(M) PCA of cells contributing to TNF and TWEAK signaling, showing DAM as the primary source.

(N) DAM-specific ligand-receptor interactions, such as Tnfsf12-Tnfrsf12a (inflammatory response amplification) and CXCL2-CXCR2 (granulocyte recruitment).

To identify key regulatory modules in DAM, we employed HdWGCNA. By applying a soft threshold of eight, we identified five key modules from the single-cell gene expression data (Figure 3G). Among these, the blue module, strongly associated with DAM, was enriched for genes such as Lpl, Spp1, and Ftl1 (Figure 3H). Functional analysis revealed that this module was linked to pathways involved in inflammatory responses, lipid metabolism, and neuronal death, all of which are detrimental to neural regeneration (Figures 3I and 3J). In contrast, MG2 homeostatic microglia were primarily associated with the turquoise module, containing genes such as Atf3, Gas5, and Mcl1, which were enriched in pathways related to mRNA metabolism and myeloid cell differentiation (Figures 3H, 3I, and S5).

Finally, we applied CellChatV2 software to compare DAM communication with other cells before and after injury (Figure 3K). Post-injury, DAM demonstrated upregulated pathways such as tumor necrosis factor (TNF), CXCL, and TWEAK, which promoted apoptosis and inflammatory responses in the microenvironment (Figure 3L). Conversely, pathways supporting cell survival, such as platelet-derived growth factor (PDGF), transforming growth factor β (TGF-β), and macrophage migration inhibitory factor (MIF), were significantly downregulated (Figure 3L). PCA of cells contributing to TNF and TWEAK signaling pathways identified DAM as the primary source of these pro-inflammatory signals (Figure 3M). Compared to other microglial subpopulations, DAM uniquely utilized the Tnfsf12-Tnfrsf12a ligand-receptor pair to amplify inflammatory responses in astrocytes, endothelial cells, fibroblasts, and macrophages (Figure 3N). Additionally, DAM recruited granulocytes through the CXCL2-CXCR2 axis, further exacerbating inflammation. In summary, our combined analyses of cell trajectory, HdWGCNA, and cell-cell communication reveal the significant plasticity of DAM and their critical role in driving inflammation, influencing surrounding cells, and impeding spinal cord repair.

Identification of hub genes regulating DAM differentiation and function in SCI

The identification and study of hub genes provides a powerful tool for understanding the complex regulatory mechanisms of cellular functions and offers potential avenues for developing therapeutic strategies. In this study, we aimed to identify hub genes regulating DAM by integrating previous microarray data (GSE47681 and GSE5296) with tissue- and cell-level analyses. Differentially expressed genes (DEGs) were identified based on criteria of Log₂|FC| > 2 and p < 0.05. At 3dpi, 352 upregulated genes and 10 downregulated genes were identified, while at 7dpi, 266 upregulated genes and 2 downregulated genes were detected (Figures 4A and 4B). By combining DEGs from 3dpi and 7dpi with genes from the blue module identified in HdWGCNA as most associated with DAM, we ultimately identified 32 key regulatory genes for DAM (Figure 4C).

Figure 4.

Identification and characterization of hub genes regulating damage-associated microglia (DAM) in spinal cord injury (SCI)

(A and B) Volcano plots of differentially expressed genes (DEGs) at 3dpi (A) and 7dpi (B) compared to the Sham group. Genes meeting the threshold (Log₂|FC| > 2, p < 0.05) are highlighted.

(C) Venn diagram showing the overlap of DEGs from 3dpi and 7dpi with genes from the HdWGCNA blue module most associated with DAM, identifying 32 hub genes.

(D) Metascape analysis of the 32 hub genes (p < 0.05).

(E) STRING protein-protein interaction (PPI) network of the 32 hub genes, highlighting five core regulatory genes (Ctsz, Folr2, Gusb, Grn, and Fcer1g) involved in inflammatory response regulation (p < 0.001).

(F) Single-cell RNA sequencing (scRNA-seq) analysis showing significantly higher expression of the five hub genes in DAM compared to Sham conditions.

(G) Pseudotime trajectory analysis demonstrating the progressive increase in the expression of hub genes (Ctsz, Fcer1g, Folr2, Grn, and Gusb) from MG2 homeostatic microglia to DAM.

Metascape analysis revealed that these hub genes are primarily involved in immune regulation, extracellular matrix degradation, inflammatory response regulation, and lipoprotein response (p < 0.05) (Figure 4D). Protein interaction analysis using STRING further identified five core regulatory genes—Ctsz, Folr2, Gusb, Grn, and Fcer1g—that are central to inflammatory response regulation (p < 0.001) (Figure 4E). scRNA-seq analysis confirmed that these five hub genes were significantly upregulated in DAM compared to the Sham group post-injury (Figure 4F). Further examination of these hub genes along the computational microglial trajectory suggested a potential progressive increase in their expression from MG2 homeostatic microglia to DAM (Figure 4G). In summary, through the integration of microarray and scRNA-seq data, we identified five key hub genes potentially involved in regulating DAM differentiation and their transition to MG2 homeostatic microglia.

Validation of DAM regulatory genes and the potential of Eupatilin regulating the DAM

Given the validation of Fcer1g, Grn, and Gusb as hub genes regulating DAM differentiation following SCI, we sought to identify a small molecule drug that could target these genes to modulate DAM function. Eupatilin, a flavonoid derived from Artemisia species, has demonstrated significant anti-inflammatory and neuroprotective effects in various studies. To investigate whether Eupatilin can interact with hub proteins to regulate the differentiation trajectory of DAM, molecular docking analyses were conducted (Figure 5A). Eupatilin interacted with FCER1G, forming hydrophobic interactions with ILE27 and a hydrogen bond with GLN15 via its hydroxyl group. Also, Eupatilin showed strong binding affinity with the GRN protein. It binds to a hydrophobic cavity formed by PHE203 and PRO202, with its hydroxyl group forming hydrogen bonds with SER204 and PRO306 of the GRN protein. Additionally, Eupatilin exhibited favorable binding with GUSB. It formed hydrophobic interactions with THR231, hydrogen bonds with ASP229 and GLU427 through two hydroxyl groups, and π-π stacking interactions between its benzene ring and TRP246. In summary, our findings suggest that Eupatilin is a potential compound capable of modulating DAM differentiation trajectories through its interactions with key regulatory proteins.

Figure 5.

The regulatory potential of Eupatilin on damage-associated microglia (DAM) hub genes

(A) Molecular docking of Eupatilin with hub proteins FCER1G, GRN, and GUSB. Molecular docking analysis illustrating the binding interactions between Eupatilin and key hub proteins. (Top) Eupatilin interacts with FCER1G (−5.084 kcal/mol), forming hydrophobic interactions with ILE27 and a hydrogen bond with GLN15 through its hydroxyl group. (Middle) Eupatilin binds strongly to GRN (−6.973 kcal/mol), occupying a hydrophobic pocket formed by PHE203 and PRO202. Its hydroxyl group forms hydrogen bonds with SER204 and PRO306, stabilizing the interaction. (Bottom) Eupatilin exhibits favorable binding with GUSB (−6.217 kcal/mol), engaging in hydrophobic interactions with THR231 and forming hydrogen bonds with ASP229 and GLU427. Additionally, its benzene ring interacts with TRP246 via π-π stacking.

(B) qPCR validation of the expression of hub genes in the control and Eupatilin group in BV-2 cell.

(C and D) Western blot validation of the expression of hub genes in the control and Eupatilin group in BV-2 cell.

(E) qPCR validation of the expression of hub genes in the Sham, control, and Eupatilin group.

(F and G) Western blot validation of the expression of hub genes in the control and Eupatilin group. Data are presented as mean ± SD; unpaired t test: ∗p < 0.05, ∗∗p < 0.01; n.s., non-significant; n = 5.

To investigate the effects of Eupatilin on microglia, we treated microglia (BV-2 cell) with Eupatilin and compared the mRNA and protein expression levels of these genes (Figures 5B–5D). Quantitative PCR analysis revealed that Eupatilin significantly downregulated the mRNA expression of Fcer1g (p = 0.043), Grn (p = 0.004), and Gusb (p = 0.001) in BV-2 cell, while Folr2 (p = 0.164) and Ctsz (p = 0.160) mRNA expression had no significant changes after Eupatilin treatment. Consistent with the transcriptional changes, western blot analysis demonstrated a significant reduction in the protein levels of FCER1G (p = 0.015), GRN (p = 0.031), and GUSB (p = 0.024), indicating that Eupatilin suppresses the expression of these pro-inflammatory or microglia-associated markers at both the mRNA and protein levels (Figures 5B–5D).

To validate the regulatory role of Eupatilin on the hub genes identified following SCI, we treated T9 spinal cord contusion mice with Eupatilin and compared the mRNA and protein expression levels of each group (Figures 5E–5G). Our findings revealed that Fcer1g (p = 0.004), Grn (p = 0.001), Gusb (p = 0.001), and Ctsz (p = 0.017) mRNA expression level increased significantly after injury, and Eupatilin treatment significantly reduced the expression of Fcer1g (p = 0.016), Grn (p = 0.006), and Gusb (p = 0.013). As shown in Figures 5F and 5G, FCER1G (p = 0.004), GRN (p = 0.026), and GUSB (p = 0.028) protein expression decreased after Eupatilin treatment in vivo. The experimental findings validated that the three hub genes (Fcer1g, Grn, and Gusb) are potentially crucial in regulating the behavior of DAM following SCI. Furthermore, Eupatilin emerges as a promising therapeutic candidate for modulating DAM function by inhibiting the expression of these three genes and fostering microglial homeostasis during SCI treatment.

Eupatilin inhibits the activity of DAM in the acute phase of SCI

To clarify the effect of Eupatilin on DAM activity during the acute phase of SCI, we administered intrathecal Eupatilin injections in a mouse model of thoracic contusion SCI and collected spinal cord tissue on day 7. Immunofluorescence staining was used to assess the impact of Eupatilin on DAM-related markers, i.e., FCER1G, GRN, and GUSB. As shown in Figure 6, Eupatilin altered the spatial distribution of the markers at the injury site. FCER1G expression was observed at the lesion and adjacent tissue (rostral and caudal), with higher levels in the adjacent tissue compared to the lesion. The Eupatilin-treated group showed significantly reduced expression in both the lesion and adjacent tissue compared to controls (rostral: t = 4.214, p = 0.003; lesion: t = 4.130, p = 0.003; caudal: t = 7.294, p = 0.0003) (Figures 6A and 6B). GRN followed a similar pattern (rostral: t = 10.22, p = 0.0001; lesion: t = 5.080, p = 0.0009; caudal: t = 5.884, p = 0.005) (Figures 6C and 6D). GUSB expression was markedly decreased in the adjacent tissue, even falling below lesion levels (rostral: t = 3.127, p = 0.009; lesion: t = 3.115, p = 0.009; caudal: t = 4.210, p = 0.005) (Figures 5E and 5F). These results indicate that Eupatilin suppressed the expression of DAM-related genes in both the lesion and adjacent tissues after SCI, suggesting an inhibitory effect on DAM activity post-SCI.

Figure 6.

Eupatilin inhibits DAM-associated gene expression 1 week after SCI

Representative immunofluorescence images illustrate the expression of FCERG1 (A, purple), GRN (C, purple), and GUSB (E, purple) in sagittal sections of the injured spinal cord. Nuclei were stained with DAPI (blue), and microglia were stained with IBA1 (yellow). The yellow dashed curve in the left panels delineate the approximate area of injury. The yellow rectangular areas in the left panels are enlarged in the corresponding panels on the right. Quantitative analysis of the fluorescence intensity of FCERG1 (B), GRN (D), and GUSB (F) in the rostral, lesion, and caudal sites. Data are presented as mean ± SD; unpaired t test: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; n = 5.

Eupatilin optimizes the regenerative microenvironment in the acute phase of SCI

Given that Eupatilin inhibits DAM activity during the acute phase of SCI, it may also impact the composition of the inflammatory microenvironment. To explore this possibility, we performed immunofluorescence staining to examine the distribution of two representative markers in microglia at the injury site: the pro-inflammatory TNF-α and the anti-inflammatory ARG1. The ratios of TNF-α- and ARG1-positive areas to the IBA1-positive (microglial) area were calculated to characterize changes in the inflammatory microenvironment. As shown in Figure 7, Eupatilin significantly reduced the proportion of TNF-α-positive area (t = 7.290, p = 0.007) and increased the ARG1-positive area (p = 0.0079). These results indicate that Eupatilin reshapes the inflammatory microenvironment at the SCI site during the acute phase, decreasing pro-inflammatory cytokine secretion and promoting anti-inflammatory microenvironment formation, suggesting its potential to optimize the regenerative microenvironment and promote post-SCI neural regeneration.

Figure 7.

Eupatilin alleviates microglia-mediated inhibitory microenvironment 1 week after SCI

(A and C) Representative immunofluorescence images illustrate the IBA1 (white), TNF-α (A, red), ARG1(C, green), and DAPI (blue) expression patterns in sagittal sections of the injured spinal cord. The yellow dashed curve in the left panels delineate the approximate area of injury. The yellow rectangular areas in the left panels are enlarged in the right panels.

(B and D) Quantitative analysis of the TNF-α (B) and ARG1 (D) expression (the ratio of TNF-α- or ARG1-positive area and IBA-1 positive area) in the lesion sites. Data are presented as mean ± SD; unpaired t test (B) and Mann-Whitney test (D): ∗∗p < 0.01; n = 5.

Eupatilin promotes motor functional recovery and tissue repair after SCI

To assess the impact of Eupatilin on spinal cord repair and hindlimb motor function, BMS scores were recorded weekly post-SCI (Figure 8A), and CatWalk footprint mapping was performed at 8 weeks (Figure 8B). In both groups, BMS scores dropped to 0 immediately after SCI. From the second week post-surgery, motor function gradually improved in both groups. Two-way repeated measures ANOVA revealed a significant group and time interaction (F = 4.165, p < 0.001). Bonferroni post hoc tests showed that the treatment group had significantly higher BMS scores than the control group from week 5 (p = 0.029) to week 8 (p < 0.001). Within-group comparisons indicated significant improvement from baseline in the treatment group starting at week 3 (p < 0.001), while the control group showed improvement from week 4 (p < 0.001). Footprint mapping at 8 weeks also showed occasional alternating hindlimb weight bearing in the Eupatilin group, a pattern rarely observed in the control group.

Figure 8.

Eupatilin promotes functional recovery and neural regeneration 8 weeks after SCI

(A) Time course of BMS scores in control and treatment groups (n = 10). Data are presented as mean ± standard error of the mean (SEM). Statistical significance: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001 (between-group comparisons, Bonferroni post hoc test); #p < 0.05 (within-group comparisons vs. week 1 baseline, Bonferroni post hoc test).

(B) The footprints shape depiction recorded with CatWalk analysis system.

(C) Representative immunofluorescence images illustrate the GFAP (green), TUJ1 (purple), and DAPI (blue) expression patterns in the injured spinal cord. The yellow dashed curve delineates the approximate area of injury.

(D) Quantitative analysis of the positive area of TUJ1 in the lesion sites. Data are presented as mean ± SD; unpaired t test: ∗∗p < 0.01; n = 5.

(E) Representative immunofluorescence images illustrate the NGF (green), 5HT (red), and DAPI (blue) expression patterns in the injured spinal cord. The yellow dashed curve delineates the approximate area of injury.

(F) Quantitative analysis of the positive area of NGF in the lesion sites. Data are presented as mean ± SD; unpaired t test: ∗p < 0.05; n = 5.

To further elucidate the impact of Eupatilin on neural tissue repair during the chronic phase following SCI, we conducted immunofluorescence staining of spinal cord tissue at 8 weeks post-injury to assess the status of neuronal regeneration. Specifically, we utilized GFAP and TUJ1 as markers for astrocytes and axons, respectively (Figures 8C and 8D). In both groups, a notable proliferation of GFAP-positive astrocytes was observed at the injury site. However, Eupatilin treatment attenuated the barrier function of the glial scar, leading to the enhanced infiltration of TUJ1-positive neuronal fibers into the lesion core (Figure 8C). The expression of TUJ1 was significantly higher in the Eupatilin-treated group compared to the control group at the lesion site (t = 2.469, p = 0.038) (Figures 8C and 8D). As shown in Figures 8E and 8F, we found the Eupatilin-treated group had a higher NGF secretion compared to the control group at lesion cite (t = 2.406, p = 0.042). Moreover, it was found that Eupatilin treatment facilitated neuron regeneration at lesion site, as evidenced by immunostaining of 5-hydroxytryptamine (5HT), neurofilament (NF), and growth-associated protein 43(GAP43) (Figures 8E and S8). These findings suggest that Eupatilin treatment alters the tissue recovery process after SCI, by inhibiting the expansion of glial scars while promoting the preservation and regeneration of neural axons in the lesion and adjacent sites.

Eupatilin promotes functional recovery by regulating DAM-to-MG2 transition

To investigate the potential mechanisms underlying Eupatilin’s promotion of functional recovery in SCI mice, we conducted bulk RNA-seq of the injury core region from Sham-operated mice (Sham), SCI mice at 7 days post-injury (SCI), and Eupatilin-treated SCI mice (Eupatilin). PCA revealed distinct transcriptomic profiles among the groups, indicating that Eupatilin altered the transcriptional features of the spinal cord in SCI mice (Figure 9A). Differential expression analysis identified 2,518 DEGs in the SCI vs. Sham comparison and 494 DEGs in the Eupatilin vs. SCI comparison (Figures 9B and 9C). Additionally, comparison between Eupatilin vs. Sham groups revealed 1,090 DEGs (Figure S7). Mfuzz clustering analysis of these 1,090 DEGs from the Eupatilin vs. Sham comparison grouped the genes into three distinct clusters (C1, C2, and C3), each associated with unique biological functions (Figure 9D). C1 genes, highly expressed in the Sham group and modestly upregulated in the Eupatilin group compared to SCI, were enriched in pathways related to neurotrophic factor secretion and synaptic signaling, suggesting Eupatilin’s ability to improve neuronal signaling post-injury. C2, a Eupatilin-specific cluster, included genes involved in tissue remodeling, wound healing, and lipid metabolism, highlighting Eupatilin’s role in promoting repair mechanisms. C3, highly expressed in the SCI group but downregulated in the Eupatilin group, was associated with inflammatory cytokine secretion, oxidative stress, and apoptosis pathways that exacerbate SCI progression. These findings indicate that Eupatilin alleviates inflammation and oxidative damage, contributing to spinal cord repair.

Figure 9.

Eupatilin promotes functional recovery by modulating microglial states and transcriptional profiles in SCI

(A) Principal-component analysis (PCA) of bulk RNA-seq data from Sham, SCI, and Eupatilin-treated mice at 7dpi.

(B and C) Volcano plots showing differentially expressed genes (DEGs) in SCI vs. Sham (B) and Eupatilin vs. SCI (C) groups, identified based on Log_2|FC| > 1 and p < 0.05.

(D) Heatmap and Mfuzz clustering analysis of Eupatilin vs. SCI DEGs, divided into three functional clusters.

(E) CIBERSORTx deconvolution of microglial subpopulations using bulk RNA-seq data, showing relative proportions of DAM and homeostatic MG2 microglia across groups.

(F–H) Bar plots (F) and boxplots (G, H) depicting the reduction of DAM and restoration of MG2 proportions in Eupatilin-treated mice compared to SCI controls.

(I) Conceptual illustration of the proposed transition of DAM to MG2, with Eupatilin promoting this shift, based on transcriptional and cellular analyses. Data are presented as mean ± SD: Wilcoxon rank-sum test: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

To explore Eupatilin’s effects on microglial subpopulation dynamics, we utilized CIBERSORTx to deconvolute bulk RNA-seq data and predict changes in cellular proportions (Figure 8E). SCI induced a marked increase in DAM, whereas Eupatilin treatment significantly reduced DAM proportions (Figures 9F and 9G). Conversely, the proportion of MG2 homeostatic microglia, which decreased following SCI, was restored by Eupatilin treatment (Figures 8F and 8H). In summary, our results demonstrate that Eupatilin promotes the transition of DAM to MG2 homeostatic microglia, reduces inflammation, and enhances neuronal function recovery in SCI mice (Figure 9I).

To further validate the bulk transcriptomic findings, we performed scRNA-seq on spinal cord tissues from Sham, SCI, and Eupatilin-treated groups, identifying a total of 10 distinct cell types (Figures 10A and 10B). Following SCI at 7dpi, the overall proportion of microglia was markedly decreased, while Eupatilin treatment did not significantly alter the total microglial proportion (Figure 10C). Subsequently, we classified microglia into subpopulations (MG0-3 and DAM) according to our previously established criteria, with DAM characterized by high expression of marker genes including Lyz2, Igf1, and Lpl (Figures 10D–10G). Regarding specific proportions, MG2 homeostatic microglia increased following injury, and Eupatilin treatment further enhanced MG2 microglial populations by approximately 2-fold (Figure 10H). Concurrently, Eupatilin treatment reduced the DAM subpopulation by 8.9% compared to the SCI group. Comparative analysis between Eupatilin-treated and SCI groups revealed that gene set enrichment analysis demonstrated Eupatilin’s capacity to ameliorate acute inflammatory response pathways in microglia within the injury region (Figure 10I). Pseudotime trajectory analysis using Monocle2 revealed a consistent differentiation trajectory from DAM toward MG2 microglia (Figures 10J and 10K). Direct comparison of microglial differentiation trajectories across different treatment groups further confirmed through scRNA-seq analysis that Eupatilin promotes the transition of DAM to MG2 homeostatic microglia, providing single-cell resolution validation of our bulk transcriptomic findings (Figures 10L and 10M).

Figure 10.

scRNA-seq analysis of Eupatilin’s role in modulating microglial dynamics in SCI

(A) UMAP plot displaying the 10 distinct cell types identified in spinal cord tissues from Sham, SCI, and Eupatilin-treated groups at 7 dpi.

(B) Volcano plot showing expression of representative marker genes used to annotate 10 cell subpopulations.

(C) Proportional composition of cell types across Sham, SCI, and Eupatilin-treated groups at 7 dpi.

(D) UMAP plot of microglia subpopulations (MG0-3 and DAM) identified based on established criteria.

(E–G) Feature plots showing the expression levels of marker genes Lyz2, Igf1, and Lpl, respectively, with high expression characterizing the DAM subpopulation.

(H) Bar plot depicting the relative proportions of MG2 homeostatic microglia and DAM subpopulations across Sham, SCI, and Eupatilin-treated groups.

(I) Gene set enrichment analysis (GSEA) of microglial pathways. GSEA plot highlighting the enrichment of acute inflammatory response pathways in microglia from the SCI group compared to the Eupatilin-treated group.

(J and K) Pseudotime trajectory plot generated using Monocle2, illustrating potential transcriptomic relationships among microglia subpopulations, with trajectories suggesting a differentiation trend from DAM through intermediate states (MG0, MG1, and MG3) toward MG2 homeostatic microglia.

(L) Violin plots or bar graphs validating the transition of DAM to MG2 homeostatic microglia in the Eupatilin-treated group.

(M) DDRTree overlay comparing microglial differentiation trajectories between SCI and Eupatilin-treated groups, highlighting the shift from DAM toward MG2 in the Eupatilin group.

Discussion

Context-dependent microglia are recognized for their distinct roles in driving disease progression through specialized functions, making them potential therapeutic targets in a variety of diseases. However, DAM specific to SCI has not been extensively studied. In this study, we used scRNA-seq to identify SCI-specific DAM subpopulations and explore therapeutic strategies targeting these cells. Our findings reveal a significant increase in DAM levels following SCI, with functional enrichment analyses indicating their critical role in promoting inflammatory responses within the injury microenvironment. Notably, post-SCI DAM exhibits a tendency to transition into homeostatic microglia (MG2), a shift associated with functional repair and tissue regeneration. To better understand this phenomenon, we combined scRNA-seq with microarray analyses to identify key regulatory genes within DAM, including Fcer1g, Grn, and Gusb. Using molecular docking techniques, we screened for potential compounds that could modulate the activity of these hub genes, leading to the identification of Eupatilin as a promising candidate. Eupatilin was found to promote the differentiation of DAM into MG2, thereby facilitating functional recovery. In vivo experiments further confirmed that Eupatilin alleviates local inflammation in SCI mouse models and enhances long-term recovery outcomes. Finally, bulk RNA-seq of injury tissues, coupled with microglial subpopulation deconvolution, validated that Eupatilin’s therapeutic effects stem from its ability to drive the DAM-to-MG2 transition. This study highlights the critical roles of DAM in SCI and positions Eupatilin as a promising therapeutic agent for modulating microglial plasticity and improving recovery outcomes in SCI.

Microglia play a central role in the pathophysiology of SCI by dynamically regulating inflammation and tissue repair, thus maintaining the homeostasis of the central nervous system. Given the diverse and context-dependent functions of microglia, accurately classifying these cells is crucial when selecting specific microglial subpopulations as therapeutic targets for SCI. Traditional classification systems, such as the dichotomous “M1/M2” paradigm derived from macrophages, have proven insufficient in capturing the full complexity of microglial behavior. Recent advances in scRNA-seq provide deeper insights into microglial heterogeneity, revealing a spectrum of transcriptional and functional states rather than rigid categories. Keren-Shaul et al. employed scRNA-seq in a transgenic mouse model (5XFAD) to define disease-associated microglia, a population linked to neurodegenerative diseases. Disease-associated microglia exhibit downregulation of homeostatic microglial markers, including P2ry12/P2ry13, Cx3cr1, and Tmem119, while genes involved in lipid metabolism, phagocytosis, and lysosomal function are upregulated, such as Apoe, Ctsd, Lpl, Tyrobp, and Trem2.¹² This gene-expression-pattern-based definition more precisely characterizes the functional roles of this microglial subset. Similarly, Hammond et al. identified injury-responsive microglia (IRM) in lysolecithin-induced brain injury using scRNA-seq, characterized by high expression of Fcrls, Apoe, and Ifi27L2a.¹⁶ Such context-dependent microglial states have been identified across various models, emphasizing the close relationship between gene signatures and microglial function. Krasemann et al. defined the microglial neurodegenerative phenotype (MGnD) in mouse models of ALS, AD, and MS.²⁸ This phenotype is characterized by the downregulation of 68 homeostatic microglial genes, including Tmem119, P2ry12, Gpr34, Cx3Cr1, and Tgfb1, and the upregulation of 28 genes, including Spp1, Itgax, Axl, Lilrb4, Clec7a, Ccl2, Csf1, and Apoe, associated with neurodegenerative processes.²⁸ Marschallinger et al. identified lipid-droplet-accumulating microglia (LDAM) in aged mice, marked by impaired phagocytosis, elevated reactive oxygen species production, and pro-inflammatory cytokine secretion, with a gene signature including CD63, Tuba1, Rab5b, and Rab7.²⁹ Li et al. described proliferative-region-associated microglia (PAM) in developing white matter, which shares gene signatures with DAM and exhibit impaired phagocytic activity, generate elevated levels of reactive oxygen species, and release pro-inflammatory cytokines.³⁰ These findings highlight how specific gene signatures underpin microglial functional diversity, enabling precise subpopulation classification and expanding our understanding of their roles in health and disease. In our study, analysis of scRNA-seq data from SCI mice revealed a subset of microglia associated with damage, which we define as DAM. The gene signature of these cells includes Lyz2, Lgf1, and Lpl, with partial overlap to previously reported disease-associated microglia, PAM, and MGnD signatures.^16,28,30 We observed a significant increase in the number of these cells early after SCI, suggesting they may serve as critical targets for SCI therapy (Figure 2). Consistent with previous studies, our findings confirm that microglial subsets with unique gene signatures exhibit specialized functions in specific pathological contexts.

Our further analysis, combining scRNA-seq and microarray data, identified 32 genes significantly upregulated in SCI-associated DAM, with functions primarily related to immune regulation, extracellular matrix degradation, inflammatory response modulation, and lipoprotein response (Figure 4). Using MetScape analysis and the PPI network, followed by subsequent validation in SCI mouse models, we confirmed the significant upregulation of three genes (Fcer1g, Grn, and Gusb) involved in DAM, indicating their role as hub genes (Figure 5). Fcer1g is widely expressed on the surface of the immune cells and mediates interactions between antibodies and the immune system.³¹ The expression changes in Fcer1g have been linked to several diseases, including leukemia,³² eczema,^33,34 osteoarthritis,³⁵ and acute myocardial infarction.³⁶ Zhang et al. reported a remarkable increase in FCER1G expression after SCI, which was correlated with the antigen-presenting cell infiltration, such as M1 macrophages.³⁷ These findings were consistent with the results of our scRNA-seq analysis and immunofluorescence (IF) staining (Figures 4 and 6). Moreover, we suggested that the upregulation of FCER1G in DAM can promote neuroinflammatory microenvironment and impair spinal cord repair. In addition, we observed the upregulation of the inflammatory gene Grn in DAM after SCI (Figure 5). Generally, GRN is associated with the lysosomal function, neuronal survival, astrogliosis, and neuroinflammation.³⁸ Marschallinger et al. discovered that Grn was one of the key genes causing autosomal-dominant forms of neurodegeneration and served as a genetic modulator in lipid droplet formation of microglia.²⁹ They used sgRNA targeting these genetic modulators to increase Grn expression, leading to significant defects in phagocytosis and reactive oxygen species production in microglia. These microglia showed functional impairments similar to those of DAM in SCI. Moreover, the knockout of Grn induced intracellular and intralysosomal long polyunsaturated triacylglycerol accumulation in the brain, which might be correlated with the disease pathology in frontotemporal dementia.³⁹ Therefore, Grn might be an alternative therapeutic target in several central system diseases, including SCI. Additionally, GUSB played a crucial role in cleaving β-linked glucuronides from diverse glycosaminoglycans, preventing the process of autosomal recessive disease mucopolysaccharidosis type VII (Sly syndrome).^40,41 The overexpression of GUSB was linked to several pathologies including inflammation and cancer metastasis.^42,43 A recent study revealed that GUSB is associated with juvenile myelination, suggesting its role in neural plasticity and function.⁴⁴ However, our results showed that the overexpression of GUSB in DAM was detrimental to nerve regeneration post-SCI. Overall, we found that the high expression of hub genes (Fcer1g, Grn, and Gusb) in DAM fosters an inflammatory microenvironment, preventing nerve regeneration and locomotor function recovery.

The transition between MG2 and DAM played a crucial role in SCI and recovery. Pseudotime trajectory analysis in current study revealed that MG2 progressively transformed into DAM post-injury (Figure 3). MG2 was involved in various biological processes, including inflammatory activation, lipid metabolism regulation, and neuron apoptosis, which resulted in the transformation to the DAM phenotype with high expression of Fcer1g, Grn, and Gusb. Meanwhile, DAM amplified inflammatory responses to various local immune cells through the Tnfsf12-Tnfrsf12a ligand-receptor interaction (Figure 3N). This phenomenon suggested that targeting the dynamic DAM-to-MG2 transition could provide a potential therapeutic strategy for SCI. Several bioactive substances such as neurotrophin-3, interleukin-10, and minocycline were widely investigated in spinal cord functional recovery research, while the high cost and unstable bioactive structure limited the clinical translational process. Eupatilin, an active ingredient isolated from Artemisia argyi, exhibits several pharmacological activities including anti-oxidative, anti-inflammatory, and anti-apoptosis effects across various disease contexts.^{17,18,19,20,21} Additionally, Eupatilin has an advantage in availability, biosafety, stability, and multifunction in spinal cord recovery, which makes it easier for clinical translation. Furthermore, the present study revealed that Eupatilin decreased the expression of FCER1G, GRN, and GUSB in DAM, thereby reducing the secretion of inflammatory cytokines (Figures 6 and 7). Then, we investigated the underlying mechanism with Eupatilin treatment. Local Eupatilin treatment remodeled the inflammatory microenvironment by reducing interleukin-1 and TNF secretion and inhibiting glial cell apoptosis. Meanwhile, we noticed that the injured area exhibited changes in biological processes, such as lipid storage, iron transport, tissue remodeling, and wound healing. Additionally, the proportion of DAM notably decreased while MG2 levels increased, indicating the DAM-to-MG2 transition after Eupatilin treatment (Figure 8). Therefore, we believe that Eupatilin optimizes the inflammatory microenvironment by downregulating DAM hub genes (Fcer1g, Grn, and Gusb) and promoting the DAM-to-MG2 transition, thus contributing to nerve regeneration and locomotor function recovery.

In conclusion, our study identifies DAM subpopulations as potential therapeutic targets in SCI and highlights Eupatilin’s ability to modulate the regenerative microenvironment by regulating DAM plasticity and facilitating their dynamic transition. These findings offer an insight into the mechanisms underlying neuronal recovery after SCI and lay a solid experimental foundation for future clinical approaches targeting DAM in SCI treatment.

Limitations of the study

Our findings highlight the heterogeneity and functional diversity of microglia in response to SCI, suggesting the potential for targeted therapies based on specific subpopulations. While our study introduces an approach to treating SCI by modulating DAM, several limitations remain to be addressed. First, while our pseudotime trajectory analysis provides valuable computational insights into potential microglial state transitions, it is important to acknowledge that these findings represent mathematical modeling of transcriptomic relationships rather than direct evidence of cellular lineage dynamics. Future studies incorporating lineage tracing methodologies and temporal single-cell profiling will be essential to definitively distinguish between DAM representing a plastic differentiation state versus a distinct injury-responsive subpopulation with enhanced proliferative capacity. Second, although Eupatilin was identified as a modulator promoting this transition, the precise molecular pathways underlying specific effects on DAM remain unclear and warrant further investigation. Furthermore, without targeted delivery systems, we cannot definitively attribute the observed effects solely to direct microglial modulation. Eupatilin may influence multiple cell types within the neuroinflammatory environment, including astrocytes and neurons, potentially affecting microglia through indirect mechanisms such as paracrine signaling or alterations in the local inflammatory milieu. While our transcriptomic analysis revealed significant modulation of microglia-associated pathways, these changes could represent downstream effects of primary actions on other cell types. Future studies utilizing microglia-specific targeting approaches such as specialized delivery systems would provide clearer evidence for direct versus indirect microglial effects of Eupatilin. Moreover, comprehensive characterization of Eupatilin’s biocompatibility and pharmacokinetic profile is essential for its potential clinical translation. Additionally, the choice of female subjects was based on practical considerations related to post-operative care, as the shorter female urethra facilitates catheter management and reduces urinary complications following SCI. However, this limits the generalizability of our findings, as sex-related differences in immune responses, neuroinflammation, and tissue repair mechanisms may influence the therapeutic efficacy of Eupatilin in SCI recovery. Future studies should include both male and female subjects to fully characterize any sex-specific effects of Eupatilin treatment and to enhance the translational relevance of these findings to diverse patient populations.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Weitao Man (mwta03607@btch.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •All data associated with this study are present in the paper or the supplemental information.
• •Raw sequencing data and processed expression matrices generated in this study have been deposited to the Gene Expression Omnibus (GEO) under the accession number GSE306745. The published mouse data are downloaded from GEO (GSE162610, GSE47681, and GSE5296). Source data are provided with this paper.
• •This paper does not report the original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was supported by Beijing Natural Science Foundation (grant no: 7242187), National Natural Science Foundation of China (grant no: 82201521 and 82301560), and Tsinghua University Initiative Scientific Research Program (grant no: 20257020014).

Author contributions

Z.W., conceptualization, methodology, investigation, original draft, review & editing, and funding acquisition; Z.M., conceptualization, methodology, investigation, original draft, and review & editing; Y.L., B.S., G.L., and P.Z., methodology, investigation, formal analysis, and review & editing; J.Y., K.Y., and G.W., investigation, validation, and formal analysis; X.W., supervision, methodology, and review & editing; B.Y., conceptualization, supervision, methodology, formal analysis, original draft, and review & editing; W.M., conceptualization, supervision, methodology, resources, project administration, and funding acquisition. All the authors take full responsibility for the work.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
rabbit anti- FCER1G	Affinity Biosciences	DF13263, 1:50
rabbit anti-GUSB	Affinity Biosciences	DF3843, 1:50
rabbit anti-GRN	Affinity Biosciences	DF7997, 1:50
rabbit anti-β-tubulin III	Abcam	ab18207, 1:500
mouse anti-IBA1	Sigma	MABN92-25UG, 1:250
mouse anti-GFAP	Santa Cruz Biotechnology	sc-65343, 1:50
mouse anti-NF 200	Sigma	N0142, 1:500
rabbit anti-GAP43	Abcam	ab16053, 1:500
rabbit anti-5-HT	Sigma	S5545, 1:1000
mouse anti- NGF	Biolegend	84004, 1:1000
goat anti-Mouse IgG (H + L) Secondary Antibody, Alexa Fluor 647	Thermo Fisher Scientific	A121235, 1:1000
goat anti-Rabbit IgG (H + L) Secondary Antibody, Alexa Fluor 594	Thermo Fisher Scientific	A11012, 1:1000
Chemicals, peptides, and recombinant proteins
Eupatilin	MedChemExpress	22368-21-4
Deposited data
scRNA-seq data	GEO	GSE306745, GSE162610, GSE47681 and GSE5296
Oligonucleotides
Ctsz, forward 5'- TATGCCAGCGTCACCAGGAAC -3'	Beijing Genomics institution	N/A
Ctsz, reverse 5'-CCTCTTGATGTTGATTCGGTCTGC-3'	Beijing Genomics institution	N/A
Fcer1g, forward 5'- CTCAAGATCCAGGTCCGAAAG-3'	Beijing Genomics institution	N/A
Fcer1g, reverse 5'- GGGAAAAGAATGCAGCCAAG-3'	Beijing Genomics institution	N/A
Folr2, forward 5'- GGCTGTGGACGAAGACTGTA-3'	Beijing Genomics institution	N/A
Folr2, reverse 5'- TCCAGGCCATGTCTTTCTCAA-3'	Beijing Genomics institution	N/A
Grn, forward 5'- GTCCTGGGAGCCAGTTTGAA-3'	Beijing Genomics institution	N/A
Grn, reverse 5'- CATCCCCACGAACCATCAAC-3'	Beijing Genomics institution	N/A
Gusb, forward 5'- TCATGACGAACCAGTCACCG-3'	Beijing Genomics institution	N/A
Gusb, reverse 5'- CGGTTTCGTTGGCAATCCTC-3'	Beijing Genomics institution	N/A
Gapdh, forward 5'- GCACCGTCAAGGCTGAGAAC-3'	Beijing Genomics institution	N/A
Gapdh, reverse 5'- TGGTGAAGACGCCAGTGGA-3'	Beijing Genomics institution	N/A
Software and algorithms
Prism	Graphpad	https://www.graphpad.com/scientific-software/prism
ImageJ	National Institutes of Health	https://imagej.net/ij/
STRING	European Molecular Biology Laboratory Peer Bork	https://cn.string-db.org
R software	R software R Core Team	https://www.R-project.org/
RCSB Protein Data Bank	Research Collaboratory for Structural Bioinformatics	https://pdb.org

Experimental model and study participant details

Cell lines

The mice microglial cell line BV-2 was obtained from Procell Life Science & Technology (China) and cultured in Dulbecco’s Modified Eagle Medium (DMEM; Procell Life Science & Technology, China) at 37°C in a humidified incubator with 5% CO₂. Cells were passaged every 2–3 days to maintain exponential growth and used for experiments between passages 3 and 10. For Eupatilin treatment, BV-2 cells were seeded in 6-well plates at a density of 5 × 10⁵ cells/well and allowed to adhere overnight. The next day, cells were treated with 100 μM Eupatilin diluted in DMSO and control groups received the same volume of DMSO. After 48 h or 72 h of incubation, cells were harvested for total RNA and protein extraction.

Animals

The research was conducted in compliance with the National Institutes of Health's Guide for the Care and Use of Laboratory Animals and received approval (No. 23-MWT1) from the Institutional Animal Care and Use Committee at Tsinghua University in Beijing, China. Female C57BL/6 mice (8-week-old, weighing 18–20 g) were utilized in this study. A total of 79 mice were used in this study. For bulk RNA sequencing, three mice per group were randomly assigned to the sham, SCI, and treatment groups. Quantitative PCR, Western blotting, and single-cell RNA sequencing were performed at 7 days post-SCI using five mice per group from the SCI and treatment groups. Immunofluorescence (IF) analysis was conducted at both 7 days and 8 weeks post-SCI in the SCI and treatment groups, with five mice per group at each time point. BMS score analysis was conducted at each week post-SCI in the SCI and treatment groups, with ten per group at each time point. Mice were housed in ventilated cages (max 5/cage) in a controlled environment with a 12/12-hour light/dark cycle (22–26°C) and had access to sterile food and water.

Method details

Microarray and scRNA-seq datasets

We sourced microarray data from the spinal cords of young female C57BL/6 mice (2–3 months old), specifically from sham-operated and subacute thoracic contusion spinal cord injury (SCI) groups. The SCI group included 14 samples (7 at 3 days post-injury [dpi] and 7 at 7 dpi), while the sham group comprised 8 samples. Data was obtained from two datasets, GSE47681 and GSE5296, sequenced on the GPL1261 platform. In addition, Single-cell RNA sequencing (scRNA-seq) data were acquired from female C57BL/6J mice aged 8–10 weeks. This dataset (GSE162610) included samples from sham and SCI groups at 3 dpi and 7 dpi.²² The data for our study were all retrieved from the GEO database, which is accessible through the National Center for Biotechnology Information (NCBI).²³ We extend our gratitude to the original authors for making these raw data sets available for our analysis.

Bulk RNA isolation and library preparation

Total RNA was extracted using TRIzol reagent (Invitrogen, USA) following the manufacturer’s protocol. RNA purity and concentration were assessed with a NanoDrop 2000 spectrophotometer (Thermo Scientific, USA), and RNA integrity was verified using the Agilent 2100 Bioanalyzer (Agilent Technologies, USA). RNA-seq libraries were prepared using the VAHTS Universal V6 RNA-seq Library Prep Kit according to the manufacturer’s instructions. Transcriptome sequencing and downstream analyses were conducted by OE Biotech Co., Ltd. (Shanghai, China).

Bulk RNA sequencing and differential gene analysis

RNA-seq libraries were sequenced on the Illumina NovaSeq 6000 platform, generating 150 bp paired-end reads. Raw FASTQ reads were processed using fastp, and low-quality reads were removed to yield clean reads. These were mapped to the mm10 mouse reference genome using HISAT2, and gene expression levels were calculated as FPKM values. Read counts were obtained using HTSeq-count. Principal Component Analysis (PCA) was performed in R to assess sample reproducibility. Differentially expressed genes (DEGs) were identified using the limma package (version 3.52.1) with a cutoff of |log₂(fold-change) | > 1.0 and p-value < 0.05. All analyses and visualizations were conducted in R version 4.2.1.

Microarray data processing and pathway enrichment analysis

To eliminate batch effects arising from multi-dataset integration, we employed the ComBat function from the sva package (version 3.44.0) for correction. ComBat utilizes an empirical Bayes framework to address batch effects through the following sequential steps: (1) establishment of a linear model incorporating biological covariates and batch variables; (2) estimation of location and scale parameters for each batch; (3) application of empirical Bayes shrinkage methods to stabilize parameter estimates; (4) standardization procedures to eliminate systematic differences between batches. The effectiveness of batch effect removal was validated through PCA conducted both before and after correction. Hub genes of DAM were identified with the limma package using stringent criteria: |log₂(fold-change) | > 2.0 and p-value < 0.05. Heatmaps were generated using the pheatmap package (version 1.0.12). Functional enrichment of DEGs was performed using the clusterProfiler package (version 4.9.3) for gene function analysis. We performed enrichment analysis of key pathways using the Metscape platform (https://metascape.org/), a valuable online resource for functional enrichment analysis.²⁴ The protein–protein interaction network analysis was carried out using STRING database (https://cn.string-db.org).

Single-cell RNA sequencing and data collection

Single-cell suspensions were prepared from spinal cord injury sites at 7 dpi. Briefly, mice were anesthetized and perfused with ice-cold PBS. Spinal cord segments encompassing approximately 1 mm of tissue centered on the injury epicenter were carefully dissected under a stereomicroscope. The tissue samples were immediately transferred to ice-cold Hibernate-E medium (Thermo Fisher Scientific) supplemented with B27 and GlutaMAX to maintain cell viability during processing. For single-cell dissociation, the collected spinal cord segments were enzymatically digested using the Neural Tissue Dissociation Kit (Miltenyi Biotec) according to the manufacturer's protocol.

Tissues were minced into small fragments and incubated with papain solution at 37°C for 15 minutes with gentle agitation. The digestion was stopped by adding ovomucoid protease inhibitor, and the tissue was mechanically dissociated using fire-polished Pasteur pipettes. The resulting cell suspension was filtered through a 40 μm cell strainer and centrifuged at 300 × g for 5 minutes at 4°C. Cell viability was assessed using trypan blue exclusion, and only samples with >80% viability were processed further. Single-cell suspensions were loaded onto the Singleron GEXSCOPE platform for droplet-based single-cell RNA sequencing. Libraries were prepared using the GEXSCOPE Single Cell RNA Library Kit following the manufacturer's instructions. Briefly, individual cells were encapsulated in droplets containing barcoded beads, and cell lysis, reverse transcription, and amplification were performed within the droplets. The resulting cDNA libraries were sequenced on an Illumina NovaSeq 6000 platform with a target depth of 50,000 reads per cell.

Single-cell RNA-seq data processing

Base call files were demultiplexed into FASTQ reads and aligned to the mm10 mouse genome using STAR aligner. Raw count matrices were processed with Seurat package (version 4.3.0), and doublets were removed using DropletUtils. Quality control retained cells with 300–10,000 gene counts and mitochondrial gene content <20%. Highly variable genes were identified using the FindVariableGenes function, followed by PCA for dimensionality reduction. The UMAP technique was used for visualization, and batch effects were corrected with the Harmony package. Differential gene expression analysis was conducted using the FindAllMarkers function (thresholds: |log₂(fold-change) | > 0.8, p-value < 0.05). Cell subgroups were annotated with the SingleR package using MouseRNAseqData as a reference. Data visualizations were performed with DimPlot, FeaturePlot, and SCP package. Cell-cell communication networks were analyzed using CellChatV2.

Pseudotime analysis

We performed pseudotime analysis using Monocle2 (version 2.24.1) and CytoTRACE. Monocle2 projected gene expression data into a lower-dimensional space using the DDRTree algorithm to infer cell state transitions. Differential expression across pseudotime states was analyzed to identify key genes. CytoTRACE (version 0.3.3) assessed cell potency and developmental trajectories by identifying gene sets associated with cell progression. Model performance was evaluated using weighted Kendall correlations, and robust hyperparameters were optimized through Bayesian cross-validation.

High dimensional weighted gene co-expression network analysis (hdWGCNA)

We applied hdWGCNA (version 0.2.1) to identify gene modules associated with microglia subtypes in SCI. After filtering genes expressed in >5% of cells, metacells were constructed and normalized using the MetacellsByGroups and NormalizeMetacells functions. A soft-thresholding power of 8 was selected to construct the co-expression network with the ConstructNetwork function. Module-trait relationships were explored using ModuleFeaturePlot and HubGeneNetworkPlot functions.

CIBERSORTx deconvolution analysis

We employed CIBERSORTx (https://cibersortx.stanford.edu/) to perform deconvolution analysis on bulk RNA-seq data, which allowed us to estimate the abundance of various cell types and infer cell-type-specific gene expression levels without the need for physical cell isolation. We initiated the process by extracting a single-cell expression matrix from a Seurat object, with genes as rows and cells as columns. Subsequently, we constructed a bulk matrix, aligning with the same gene-centric format but with samples as columns, to facilitate comparison with the single-cell data. We then utilized the CIBERSORTx online platform, which is user-friendly and does not require prior bioinformatics training or programming experience, to upload our prepared files. On the CIBERSORTx website, we first created a signature matrix from our single-cell data (Data S1). With the signature matrix in place, we proceeded to impute cell fractions from our bulk tissue transcriptome profiles.

Molecular docking analysis

The 2D structure of Eupatilin was retrieved from PubChem, while protein structures were obtained from the RCSB Protein Data Bank (https://pdb.org). AutoDock 4 was employed to perform flexible ligand docking simulations, enabling an analysis of protein-ligand interactions. Structural visualizations were created using PyMOL for presentation and interpretation.

Spinal cord injury model

Prior to surgery, mice were fasted for 6 hours received anesthesia with a single intraperitoneal injection of 270-330 mg/kg Avertin (Sigma-Aldrich, MO, USA, T48402). The T9 lamina was surgically exposed while preserving the integrity of the dura mater, followed by a 30-second aneurysm clip compression for all injured groups. The muscle and skin were then sutured closed. Following SCI, manual bladder expression was performed 3 times daily until bladder function recovered. For the first 3 days, intrathecal injections were administered: the treatment group received Eupatilin (30 μl, 1.5 mg/mL, Bide, Shanghai, China, BD298186), while the SCI control group received an equivalent volume of saline.

Functional behavior evaluation

Motor function was evaluated utilizing the Basso Mouse Scale (BMS), a standardized metric ranging from 0 (total paralysis) to 9 (normal function). Assessments were conducted on the day of injury and weekly thereafter for a duration of 8 weeks, with ten mice per group at each time point. Evaluations were conducted by three independent, blinded observers who observed each animal for 5 minutes to determine the BMS score through consensus. Furthermore, at the 8-week post-SCI, gait analysis was performed using the Catwalk XT 10.6 System (Noldus, Wageningen, Netherlands). Mice traversed a fluorescently illuminated glass walkway, and footprints were captured by a camera positioned below. The collected data were analyzed using CatWalk 10.6 software, which automatically identified and labeled paw prints for further analysis.

Tissue preparation

Following anesthesia, mice underwent a perfusion process, initially with saline and subsequently with 4% (w/v) paraformaldehyde, administered through the heart. The spinal cord was carefully dissected, and the dura mater was removed. Spinal cords were sectioned into 1-cm segments centered on the lesion site and subsequently cryosectioned into 10-μm-thick sagittal slices using a freezing microtome (CM1900, Leica Microsystems, Germany). Subsequently, these sections underwent processing for immunofluorescence (IF) staining.

Immunofluorescence staining

As described in our previous study, Tissue sections were rinsed in phosphate-buffered saline (PBS), blocked for 2 hours at room temperature in 10% goat serum and 0.3% (w/v) Triton X-100 in PBS, and incubated with primary antibodies overnight at 4°C.^25,26 After washing three times with PBS, sections were incubated with secondary antibodies for 1 hour at room temperature in the dark. Finally, sections were mounted using a medium containing DAPI (Abcam, UK). IF-stained sections were scanned with a Zeiss Axio Scan.Z1 scanner (Carl Zeiss, Germany) and analyzed using Zen 2.6 software (Carl Zeiss, Germany).

Quantitative PCR

Spinal cord injury tissues were harvested at 7 days post-SCI (5 samples per group. The regions surrounding the lesion site (approximately 1 mm on either side) were rapidly snap-frozen in liquid nitrogen. mRNA was isolated utilizing the mRNA Isolation Kit (Tiangen, Beijing, China, DP501), and cDNA was synthesized with the FastQuant RT Kit (Tiangen, Beijing, China, KR-106). As described in our previous study, quantitative real-time polymerase chain reaction (qRT-PCR) was conducted on the CFX96 Real-Time System (Bio-Rad Laboratories, Shenzhen, China), employing iTaq Universal SYBR Green Supermix (Bio-Rad Laboratories, Shenzhen, China, 172-5122).²⁷ Gene expression was normalized to the housekeeping gene Gapdh, and relative expression levels were calculated using the 2ˆ-ΔΔCT method. The primer sequences used are as follows: for Ctsz, forward 5'- TATGCCAGCGTCACCAGGAAC -3' and reverse 5'-CCTCTTGATGTTGATTCGGTCTGC-3'; for Fcer1g, forward 5'- CTCAAGATCCAGGTCCGAAAG-3' and reverse 5'- GGGAAAAGAATGCAGCCAAG-3'; for Folr2, forward 5'- GGCTGTGGACGAAGACTGTA-3' and reverse 5'- TCCAGGCCATGTCTTTCTCAA-3'; for Grn, forward 5'- GTCCTGGGAGCCAGTTTGAA-3' and reverse 5'- CATCCCCACGAACCATCAAC-3'; and for Gusb, forward 5'- TCATGACGAACCAGTCACCG-3' and reverse 5'- CGGTTTCGTTGGCAATCCTC-3'; for Gapdh, forward 5'- GCACCGTCAAGGCTGAGAAC-3' and reverse 5'- TGGTGAAGACGCCAGTGGA-3'.

Western blot

Total protein was extracted from BV-2 cells using RIPA lysis buffer (Beyotime, China) supplemented with protease and phosphatase inhibitors (Thermo Fisher Scientific, USA). Lysates were incubated on ice for 30 min and then centrifuged at 12,000 × g for 15 min at 4°C. The supernatant was collected. Equal amounts of protein (20–30 μg) were mixed with loading buffer, denatured at 95°C for 5 min, and separated by SDS-PAGE on 10% polyacrylamide gels. The proteins were then transferred to polyvinylidene difluoride (PVDF) membranes (Millipore, USA) using a semi-dry transfer system. Membranes were blocked with 5% non-fat milk in Tris-buffered saline containing 0.1% Tween-20 (TBST) for 1 h at room temperature and incubated overnight at 4°C with primary antibodies diluted in TBST with 5% BSA. After three washes with TBST, membranes were incubated with HRP-conjugated secondary antibodies (1:5,000; ZSGB-Bio, China) for 1 h at room temperature. Protein bands were visualized using enhanced chemiluminescence (ECL) reagents (Bio-Rad, USA), and images were captured using a ChemiDoc XRS+ imaging system (Bio-Rad, USA).

Quantification and statistical analysis

Statistical analyses were performed using SPSS 23.0 (IBM, Chicago, IL). Quantitative analysis of IF images was conducted with ImageJ 1.52 (National Institutes of Health, USA). Graphs and illustrations were created using GraphPad Prism 8.0 (GraphPad, San Diego, CA) and Adobe Illustrator 22.1 (Adobe, San Jose, CA). Data are presented as mean ± standard deviation (SD). The BMS scores were analyzed using a two-way repeated measures ANOVA with a Bonferroni post hoc test. Comparisons between two groups were analyzed using two-tailed Student's t-tests. For data that did not meet the assumption of normality, nonparametric rank-based tests were applied. All statistical tests were two-sided, and a p value < 0.05 was considered statistically significant.

Published: October 1, 2025