Exploring potential biomarkers of NETosis-Related genes in spinal cord injury through machine learning and multi-omics analysis

Abstract

Spinal cord injury (SCI) is a serious condition typically caused by mechanical trauma, often resulting in significant motor, sensory, and autonomic dysfunction. It places a heavy burden on individuals, families, and society; however, effective treatment options are still limited because of the complex pathophysiology behind primary and secondary injury mechanisms. Neutrophil extracellular traps (NETs) created by neutrophils play a crucial role in exacerbating secondary injury following spinal cord injury by promoting inflammation and hindering neural repair. This study aims to clarify the molecular basis of neutrophil extracellular trap-related genes (NRGs) in SCI through an integrated bioinformatics approach. We utilized the GSE151371 dataset from the GEO database, which includes gene expression profiles from 38 SCI patients and 10 healthy controls, and we identified differentially expressed genes (DEGs) using the limma package in R. We identified 4878 DEGs, and we performed functional analysis of these genes using GO and KEGG. Immune cell infiltration analysis conducted with CIBERSORT showed significant differences in immune cell populations between the SCI group and the control group, with notable differences in the infiltration of Neutrophils, B cells memory and Macrophages M0. Weighted gene co-expression network analysis (WGCNA) identified a module highly associated with SCI, which resulted in the selection of 12 candidate genes. We built a predictive model using machine learning algorithms, identifying NLRP3, LRG1, and TLR8 as key genes with high diagnostic potential (AUC >0.9). Subsequently, through multi-omics analysis, including gene set enrichment analysis (GESA), protein interaction analysis, and correlation analysis between key genes and immune cells, we explored the relationship between key NRGs and the pathological processes in SCI patients. Finally, these findings were validated through molecular biology experiments in a rat SCI model and clinical samples, confirming the clinical relevance of our findings regarding these biomarkers. In summary, this study provides a comprehensive analysis of NRGs in SCI, highlighting their diagnostic and therapeutic potential. Future research could focus on developing interventions targeting NETs formation, providing new opportunities to enhance treatment outcomes for SCI.

Highlights

• •First systematic analysis links neutrophil extracellular traps (NETs) to spinal cord injury pathology.
• •Machine learning identifies NLRP3, LRG1, TLR8 as diagnostic biomarkers (AUC >0.9).
• •NET-related genes drive immune dysregulation via neutrophil/macrophage infiltration.
• •Multi-omics validation confirms NRGs' role in neural repair inhibition.
• •Targeting NET formation offers novel therapeutic strategy for SCI.

1. Introduction

Spinal cord injury (SCI) is a destructive neuropathological condition characterized by motor, sensory, and autonomic dysfunction. Its pathological progression involves two distinct phases: primary mechanical injury and secondary cascading reactions [[1], [2], [3], [4]]. Given the difficulty in preventing primary injury, regulating secondary damage has become a critical therapeutic approach for improving SCI prognosis [4,5]. Neutrophils, as the earliest infiltrating immune effector cells at the injury site (comprising 50–70 % of circulating leukocytes), participate in the pathological process through phagocytosis, degranulation, and the formation of Neutrophil Extracellular Traps (NETs) [[6], [7], [8], [9]]. NETs are three-dimensional reticular structures composed of chromatin DNA, histones, and granule proteins. While physiologically responsible for pathogen clearance, their pathological accumulation impedes neural repair by exacerbating inflammation, disrupting the blood-spinal cord barrier, and inhibiting axonal regeneration [[10], [11], [12], [13], [14]]. Previous studies have demonstrated that targeting NETs (e.g., with CD47-blocking antibodies) reduces tissue damage and neuronal death following SCI [15]. Furthermore, NETs play a pathologically promoting role in multiple neurological disorders [16], making targeted regulation of NET formation a crucial therapeutic strategy across diseases. However, systematic research on NET-associated genes (NRGs) in spinal cord injury remains insufficient.

With the rapid advancement of bioinformatics, screening differentially expressed genes (DEGs) based on gene microarray and high-throughput sequencing data has become a crucial strategy for elucidating disease molecular mechanisms. In the field of spinal cord injury (SCI) research, systematic analysis of SCI-related DEGs can profoundly reveal its pathological processes and uncover potential therapeutic targets. This study aims to comprehensively analyze the association between NET-related genes (NRGs) and DEGs to effectively identify SCI patients. SCI patient DEGs will be sourced from databases. We employed various bioinformatics methods to screen for differentially expressed NRGs (DE-NRGs) in SCI, ultimately focusing on three core genes playing pivotal roles in immune and inflammatory regulation: LRG1, NLRP3, and TLR8. LRG1, as a crucial immunoregulatory factor, exhibits significant expression correlations with the activation and infiltration of immune cells such as neutrophils, and participates in regulating NET formation [17]; NLRP3, as a core component of inflammasomes, engages in key pathways of neuronal necrotic apoptosis and neuroinflammation [18,19]; TLR8, a member of the Toll-like receptor family, plays a vital role in recognizing damage-associated molecules and initiating innate immune responses, closely linked to the remodeling of the immune microenvironment [20,21]. These characteristic genes may participate in regulating the inflammatory process of SCI by modulating NETs, demonstrating significant diagnostic potential. We further conducted pathway enrichment analysis and immune cell infiltration analysis. To validate the reliability of bioinformatics predictions, core genes will undergo molecular biological validation using rat SCI models and clinical blood samples. This cross-validation strategy holds promise for providing translational medical evidence for precise diagnosis and targeted therapy of SCI.

2. Methods

2.1. Data sources

By employing "SCI" as the search keyword, we successfully retrieved the gene expression dataset GSE151371 from the GEO database (https://www.ncbi.nlm.nih.gov/geo/).The criteria for the Homo sapiens species and GPL20301 platform were met. The dataset of microarrays includes gene expression profiles derived from 38 patients who have experienced spinal cord injury as well as 10 healthy individuals. Drawing from these examples, the dataset was divided into two distinct categories: one comprising individuals with SCI and the other consisting of healthy controls (HC). After conducting an extensive literature review, we successfully identified the NRGs. Furthermore, we accessed the GeneCards database (https://www.genecards.org/),a public resource dedicated to biomarkers and regulatory elements associated with various cell death manifestations and ailments [22]. In order to identify particular NETosis markers, the process of NETosis is utilized as the key term for screening. These genes have been extensively validated for their crucial involvement in the process of NETosis formation.

2.2. Differentially expressed gene processing

To analyze the differential expression between SCI samples and healthy controls in the GSE151371 dataset, the limma package in R was employed. This analysis involved identifying DEGs based on thresholds of | log2 fold change (FC) |> 1 and an adjusted P-value <0.05.

2.3. Functional and pathway enrichment analysis

We performed functional enrichment analysis on DEGs using the clusterProfiler R package within the Gene Ontology (GO) [23] and Kyoto Encyclopedia of Genes and Genomes (KEGG) [24] frameworks. The GO analysis covered three distinct categories: Biological Processes (BP), Cellular Components (CC), and Molecular Functions (MF). To gain insights into the signaling pathways linked to DEGs, we utilized KEGG analysis.

2.4. Immune cell infiltration assessment

The IOBR package in R was utilized to perform comprehensive immune cell infiltration profiling on GSE151371 samples, with the CIBERSORT algorithm [25] integrated into the process. CIBERSORT deconvolutes gene expression data to estimate relative abundances of 22 immune cell subtypes and reveals compositional differences across samples.

2.5. Weighted gene Co-expression network construction

The WGCNA R package was utilized to deploy the Weighted Gene Co-expression Network Analysis (WGCNA) and construct co-expression modules for the GSE151371 cohort, employing the pick Soft Threshold function [26]. Clusters of highly correlated genes (modules) were identified by a topological overlap matrix (TOM) and dynamic tree-cutting algorithm. The values of module membership (MM) were used to determine the modules associated with SCI. The limma package was utilized to intersect genes in key modules with NETosis markers, resulting in NRGs that can be used for further analysis.

2.6. Model construction

Eleven different methods for selecting variables and building models, such as Lasso, Ridge, Elastic Net (Enet), Stepglm, SVM, glmBoost, LDA, plsRglm, and RandomForest, are available. XGBoost and NaiveBayes were implemented in R through a combinatorial framework. In the process of cross-validation, different algorithms were employed: one for variable selection and another for constructing predictive models. The area under the curve (AUC) of the receiver operating characteristic (ROC) was computed for all 101 algorithm combinations using external datasets. All model combinations that resulted in fewer than two variables were excluded, and the remaining results were visualized using heatmaps. Through meticulous planning and coordination, a total of 56 distinct training models were created by combining various pairs. These models were further validated by internal test sets. The best model was identified as the one that attained the highest average AUC in both training and testing groups.

2.7. Protein-protein interaction (PPI) network analysis

Protein and protein interaction networks for LRG1, NLRP3, and TLR8 were constructed using the GeneMANIA [27] data resource (http://www.genemania.org) and the STRING online tool, along with predictions of their potential target proteins. (In the STRING online tool, input LRG1, NLRP3, and TLR8, set the species to Homo sapiens, and adjust parameters (interaction score threshold ≥0.4).)

2.8. Prognostic analysis

The pROC R package was used to generate ROC curves, which were employed to assess the diagnostic potential of key DEGs. In the GSE151371 dataset, candidate biomarkers were validated and a high diagnostic discriminative power was indicated by an AUC greater than 0.7. To predict outcome probabilities based on multiple variables, a nomogram was constructed using the “rms” R package. Decision curve analysis (DCA) evaluated clinical utility by measuring net benefits across threshold probabilities compared to "treat-all" or "treat-none" strategies [28]. The clinical impact curves (CIC) were used to complement the evaluation of the model [29].

2.9. Gene set enrichment analysis (GSEA)

GSEA [30] v3.0 (https://www.gsea-msigdb.org/gsea/index.jsp) categorized the samples into two distinct groups: one with high expression (>50 %) and another with low expression (<50 %), based on the levels of hub genes. The c2.cp.kegg.Hs.symbols.gmt gene set was downloaded from the molecular signature database for analyzing biological pathways and molecular mechanisms. The enrichment analysis employed gene set sizes ranging from a minimum of 5 to a maximum of 5000, utilizing 1000 permutations. Pathways that met the criteria of P < 0.05 and had normalized enrichment scores (NES) surpassing thresholds were grouped and represented visually.

2.10. Correlation analysis of hub NRGs with immune cells

A Pearson correlation analysis [31] was conducted to investigate the relationships between Hub NRGs and immune cell subsets, such as neutrophils, B cells, and T cells, in the dataset GSE151371. Using R's LinkNet package, interaction networks were visualized.

2.10.1. Clinical blood samples

This study was approved by the Ethics Committee of the First Affiliated Hospital of Guangxi Medical University and conducted in accordance with the Declaration of Helsinki. Venous blood samples were collected from 3 SCI patients and 3 healthy individuals.

2.10.2. Animal Cultivation

Ten female SD rats aged 8–10 weeks and weighing 180–200 g were purchased from the Animal Experiment Center of Guangxi Medical University (Guangxi, China). Rats were housed under standard light-dark cycles and in a specific pathogen-free environment with free access to food and water. Due to the loss of sensory and motor function in the lower body following injury, animals were unable to urinate autonomously. Female rats were selected due to their shorter urethra and relatively weaker urinary control, facilitating manual voiding. Male rats possess a longer urethra and stronger sphincter control, significantly increasing the difficulty of manual voiding procedures and potentially compromising animal welfare and model consistency. All animal experiments were approved by the Animal Ethics Committee of the First Affiliated Hospital of Guangxi Medical University (Guangxi, China) and strictly conducted in accordance with the guidelines of the National Health Commission of the People's Republic of China.

2.10.3. Establishment of SCI SD rat model and sample preparation

Ten rats were randomly divided into two groups of five rats each: the control group (Sham) and the SCI model group. Rats in the Sham group underwent laminectomy only, while rats in the SCI group underwent spinal cord injury modeling under sterile conditions using the modified Allen's method [32]. Briefly, anesthesia was induced with a 5 % isoflurane +70 % N₂O + 30 % O₂ mixture, maintained at 1.5–2 % isoflurane (delivered continuously via nasal prongs) [33], and the T9–T11 spinal cord level was exposed via laminectomy. Injury was then inflicted by freely dropping a 2 mm diameter, 10 g impact rod from a height of 5 cm onto the dorsal surface of the spinal cord [32]. The incision site was subsequently closed in layers and disinfected with topical iodine solution. Postoperatively, SCI rats underwent manual bladder emptying twice daily until bladder reflex recovery. Tissue collection occurred on day 7 post-surgery. Each rat underwent cardiac perfusion under anesthesia, followed by extraction of spinal cord tissue extending 0.5 cm (approximately 1 cm) cephalad and caudal from the lesion center (T10) for subsequent experiments.

2.10.4. Quantitative real-time PCR (qRT-PCR)

The NucleoZOL RNA Kit (MACHEREY-NAGEL, Düren, Germany) was utilized to extract total RNA from both human blood and rat spinal tissues. The process of reverse transcription to create complementary DNA (cDNA) was carried out using the PrimeScript RT Master Mix manufactured by Servicebio in Wuhan, China. The qPCR assays were conducted using PerfectStart Green qPCR SuperMix (manufactured by TransGen Biotech in Beijing, China) on a Gentier 96E/96R system (produced by TIANLONG in Shaanxi, China).Relative gene mRNA expression was calculated using the ΔΔCT method [34] and normalized against β-actin and GAPDH. Sangon Biotech in Shanghai, China synthesized the primers (refer Table 1, Table 2).

Table 1.

Gene primer sequences for qRT-PCR (human).

Primer Name	Sequence(5′to3′)	Primer Length
β-actin(F)	CCTGGCACCCAGCACAAT	18
β-actin(R)	GGGCCGGACTCGTCATAC	18
NLRP3(F)	GGCAACACTCTCGGAGACAA	20
NLRP3(R)	GGAAAGATCCCAGCAGCAGT	20
TLR8(F)	CCAAACTGCCAAGCTCCCTA	20
TLR8(R)	TGAAGCACCTCGGACAGTTC	20
LRG1(F)	AGCTGAGGGTGCTGGATCTA	20
LRG1(R)	CAGATGCCCCAGAGCTTTCA	20

Table 2.

Gene primer sequences qRT-PCR (rat).

Primer Name	Sequence(5′to3′)	Primer Length
GAPDH(F)	GGCACAGTCAAGGCTGAGAATG	22
GAPDH(R)	ATGGTGGTGAAGACGCCAGTA	21
NLRP3(F)	CCAGGAGTTCTTTGCGGCTA	20
NLRP3(R)	GCCTTTTTCGAACTTGCCGT	20
TLR8(F)	GCCTCCCAAACTACCAAGCT	20
TLR8(R)	AACACCTTGGGCAGTTTCCA	20
LRG1(F)	ATGACAACCAGCTAACCGCA	20
LRG1(R)	CAGGGGTTGTGGGAGACATC	20

2.11. Statistical analysis

The research analyses were conducted utilizing the R programming language (version 4.5.0) within the RStudio integrated development environment (version 2025.05.0–496), complemented by Perl (version 5.40.2) for file processing tasks. The analysis of the data was conducted using GraphPad Prism software, specifically version 10.1.2.Continuous variables are presented as mean ± standard deviation (SD), and intergroup comparisons were carried out through independent samples t-tests. The flow chart depicting the study is displayed in Fig. 1.

Fig. 1.

Flow chart of this study. A comprehensive visual representation of the research methodology employed in this study is presented here. Utilizing the GSE151371 dataset from the GEO database, we employed the limma package in R to identify differentially expressed genes (DEGs). Following this, Functional and pathway enrichment analyses were conducted using GO and KEGG for these genes. Subsequently, employing weighted gene co-expression network analysis (WGCNA), a module was identified that exhibited a strong correlation with SCI, which overlapped with NETosis. This process resulted in the identification of 12 potential candidate genes. Using machine learning algorithms, we constructed a predictive model that identified NLRP3, LRG1, and TLR8 as crucial genes with significant diagnostic potential (AUC >0.9). Additionally, comprehensive multi-omics investigation encompasses gene set enrichment analysis (GSEA), protein interaction examination, and correlation assessment of crucial genes with immune cells. A study was undertaken to investigate the connection between the three hub NRGs and the pathological processes occurring in individuals with SCI. In conclusion, a series of molecular biology experiments were conducted to confirm the results obtained from rat SCI models and clinical samples.

3. Results

3.1. Identification of DEGs between SCI and Control Groups

The GSE151371 dataset, comprising 38 patients with SCI and 10 healthy controls, was subjected to differential analysis(Fig. 2A). The analysis utilizing the "limma" package in R revealed a total of 4878 differentially expressed genes (DEGs), with 2409 being upregulated and 2469 downregulated (Fig. 2B). The heatmap presents the top 50 upregulated and top 50 downregulated DEGs categorized by groups (Fig. 2C).

Fig. 2.

Identification of differentially expressed genes between SCI and Control Groups. (A) Cohort Composition of the GSE151371 Dataset. (B–C) Volcano plots and heatmap depicting the differentially expressed genes between the aforementioned groups.

3.2. Functional and pathway enrichment analysis

The analysis of GO includes Biological Processes (BP), Cellular Components (CC), and Molecular Functions (MF) (Fig. 3A and B) demonstrated notable enrichments of DEGs: The terms encompassed in BP include lymphocyte differentiation, T cell differentiation, α-β T cell activation, as well as leukocyte/cell activation associated with immune response. The activation of CD4-positive α-β T cells, the signaling pathways involved in immune response regulation and activation of cell surface receptors, the activation of myeloid leukocytes, and the differentiation of α-β T cells are all important processes in the immune system. The CC analysis revealed an abundance of secretory vesicle components, including specific/tertiary granules and the external side of plasma membrane. MF categories highlighted immune/G protein-coupled/chemokine/cytokine/C–C chemokine receptor activities, structural constituent of ribosomes, chemokine/cytokine binding, and pattern recognition receptor activity. KEGG pathway analysis further showed that there was enrichment of DEGs in pathways related to COVID-19 infection, Th1/Th2 cell differentiation, and PD-L1 expression/PD-1 checkpoint pathways (Fig. 3C and D).

Fig. 3.

Differential expression analysis in SCI. (A–B) Top 10 enriched functional terms in biological processes (BP), cellular components (CC), and molecular functions (MF). (C–D) Pathway enrichment analysis using KEGG for DEGs.

3.3. Weighted gene Co-expression network construction

The scale-free topology fit index reached its peak at a soft thresholding power of 10 (as shown in Fig. 4A), and the mean connectivity remained stable near 1, which made it easier to compare networks. Hierarchical clustering produced a dendrogram (Fig. 4B) that was based on topological overlap, and color-coded modules were assigned through dynamic tree-cutting. The analysis of the module-trait relationship demonstrated that the light coral module exhibited the most robust positive correlation with SCI, as depicted in Fig. 4C-D. The analysis using a Venn diagram revealed that there are 12 genes that overlap between the light coral module and NRGs (Fig. 4E).

Fig. 4.

Network construction and module detection in GSE151371 using WGCNA. (A) Selection of the soft-thresholding power in WGCNA and Analysis of network connectivity within WGCNA. (B) The WGCNA generated a cluster dendrogram. Identification of candidate hub genes. (C) Cluster dendrogram of gene modules from WGCNA. (D) Distribution of gene significance across modules. (E) The Venn diagram illustrating the overlap between the lightcoral module and NRGs.

3.4. Machine learning-based hub gene selection

Using a combinatorial framework, 11 machine learning algorithms were employed to generate 101 models aimed at identifying robust diagnostic signatures among 12 candidate NRGs. The training and validation of models were conducted on a dataset that was split in an 8:2 ratio. The evaluation of performance was conducted using the mean AUC metric across both training and internal test sets, as depicted in Fig. 5A. Integrated analysis revealed that NLRP3, LRG1, and TLR8 are hub genes exhibiting exceptional classification performance.

Fig. 5.

Screening of optimal signature genes and evaluation of diagnostic performance. (A) Multiple machine learning algorithms achieved the highest AUC values on the internal test set (1). (B) ROC curve analysis for the signature genes NLRP3, LRG1, and LYZ. (C) Diagnostic prediction map constructed based on the three signature genes. (D) Calibration curve evaluating the predictive capability of the prediction map. (E) Decision curve analysis (DCA) of the prediction map's predictions.

3.5. Diagnostic evaluation and nomogram construction

The analysis using the ROC curve demonstrated AUC values of 0.904 for NLRP3, 0.966 for LRG1, and 0.987 for TLR8, which suggests outstanding diagnostic capability (Fig. 5B). A nomogram that incorporates these genes was created utilizing the rms package, as depicted in Fig. 5C. The calibration curves exhibited remarkable predictive accuracy, as shown in Fig. 5D. The analysis using DCA curves demonstrated that the combined model or individual genes outperformed the "treat-all" and "treat-none" strategies in terms of net benefits, as shown in Fig. 5E across various threshold probabilities.

3.6. Expression differences and protein interactions of characteristic NRGs between normal and disease groups

To investigate the expression differences of these three SCI-specific NETosis-related genes (NRGs) between healthy controls and SCI groups, we analyzed the expression levels of the three characteristic NRGs in both groups using the ggplot package in R. As shown in the box plots (Fig. 6A), NLRP3, LRG1, and TLR8 exhibited significant expression differences between healthy controls and SCI groups (P < 0.05). Both box plots and volcano plots revealed significantly higher expression levels of these three characteristic genes in the SCI group compared to healthy controls (Fig. 6B). The protein-protein interaction network generated by GeneMANIA (Fig. 6C) revealed interactions among LRG1, NLRP3, and TLR8 proteins, with 20 potential target proteins identified: TXNIP, PYCARD, SUGT1, UNC93B1, MEFV, P2RX7, HSP90AB1, NEK7, MAVS, CTSB, CNPY3, TLR7, CASP1, CTSK, TLR9, CARD8, HSP90B1, CYCS, ENG, IRAK4) interact with LRG1, NLRP3, and TLR8. Additionally, we employed the online tool STRING (https://string-db.org/) to construct a protein-protein interaction network for comparative analysis. The resulting protein interaction network comprised 23 nodes and 99 edges. And results revealed that only four genes—CARD8, NEK7, MEFV, and CNPY3—were predicted by both tools (Fig. 6D).

Fig. 6.

Expression differences and protein-protein interaction analysis of characteristic NETosis-related genes (NRGs). (A) Box plot (R/ggplot): Expression comparison of NLRP3, LRG1, and TLR8 between healthy controls and SCI group (P < 0.05). (B) Volcano plot: Validation that NLRP3, LRG1, and TLR8 expression is significantly higher in the SCI group than in healthy controls. (C) GeneMANIA-constructed PPI network: Interactions exist between LRG1/NLRP3/TLR8, predicting 20 potential target proteins. (D) Intersection of STRING and GeneMANIA predictions: Only CARD8, NEK7, MEFV, and CNPY3 are common core target proteins in the interaction network.

3.7. GSEA enrichment analysis of hub NRGs

The Gene Set Enrichment Analysis (GSEA) of three key NRGs demonstrated unique pathway connections that were categorized based on their expression levels. In the case of TLR8, notable enriched pathways in the low-expression group included hematopoietic cell lineage, MAPK signaling, primary immunodeficiency, ribosome, and T cell receptor signaling (Fig. 7A). NLRP3 and LRG1 showed increased presence in both high-expression and low-expression groups (Fig. 7B–E). In the high-expression group, both genes exhibited immune-related enrichment in carbohydrate metabolism, lysosomal pathways, neuroactive ligand-receptor interactions, and PPAR signaling.On the other hand, the low-expression group showed increased enrichment in allograft rejection, ribosome, and T cell receptor signaling.

Fig. 7.

Gene set enrichment analysis (GSEA) of signature genes in SCI. Analysis of immune cell infiltration. (A) Results of GASE enrichment analysis for TLR8. (B) GASE enrichment analysis of NLRP3 in high-expression subgroup. (C) GASE enrichment analysis of NLRP3 in low-expression subgroup. (D) GASE enrichment analysis of LRG1 in high-expression subgroup. (E) GASE enrichment analysis of LRG1 in low-expression subgroup. (F) Distribution of 19 types of immune cells in tissues of healthy controls and SCI patients. (G) Exhibited statistically distinct immune infiltration patterns between SCI patients and healthy controls. (H) Heatmap of co-expression of the 19 types of immune cells.

3.8. Immune cell infiltration

The CIBERSORT analysis quantified 19 distinct immune cell types in the comparison between the SCI group and the control group (Fig. 7F). Subsets that underwent substantial changes (P < 0.05) encompassed B cell memory, plasma cells, T cells CD8, T cells CD4 naïve, T cells CD4 memory resting, T cells gamma delta, macrophages M0, and neutrophils (Fig. 7G). Patients with SCI exhibited a notable rise in the infiltration of plasma cells, T cells gamma delta, macrophages M0, and neutrophils. In contrast, the memory of B cells, as well as the CD8 T cells, naïve and memory resting CD4 T cells, were observed to be reduced. The correlation analysis uncovered the most robust negative correlations between Macrophages M0 and T cells CD4 memory resting, as well as between Monocytes and Neutrophils. The strongest positive associations were found between the resting state of NK cells and the naïve CD4 T cells. Between eosinophils and T cell regulatory cells (Tregs) (Fig. 7H).

3.9. Correlation analysis of core genes with immune cells

Correlation assessments showed varied immune cell associations for the three hub genes, with NLRP3 showing no significant correlations with any immune cells (P > 0.05, Fig. 8A); TLR8 showed associations with seven immune subsets (P < 0.05, Fig. 8B). Including positive correlations with CD4 memory activated T cells, resting Mast cells, Plasma cells, γ-δ T cells, and Neutrophils, alongside negative correlations with Monocytes and T cells CD8. Correlations (|r|> 0.5) were observed for Neutrophils, T cells γ-δ, and Monocytes (Fig. 8D–F). LRG1 was found to be associated with five different immune cell types, as evidenced by statistical significance (P < 0.05) and illustrated in Fig. 8C, indicating a positive correlation with plasma cells, T cells γ-δ, along with Macrophages M0, exhibit negative associations with T cells CD4 memory resting and Dendritic cells activated. In the study, macrophages M0, T cells γ-δ, and T cells CD4 memory resting were found to have r values greater than 0.5 (Fig. 8G–I). Further investigation using the LinkNet package revealed the interaction networks formed between these NRGs and the involved immune cells (Fig. 8J).

Fig. 8.

Correlation analysis between hub genes and immune cells. (A–C) Visualization of Spearman correlations between immune cells and NLRP3, TLR8, and LRG1, respectively. (D–F) Correlation analysis of immune-related biomarkers (TLR8) with neutrophils, γδ T cells, and monocytes in patients with spinal cord injury. (G–I) Correlation analysis of immune-related biomarkers (LRG1) with M0 macrophages, γδ T cells, and CD4 memory quiescent T cells in patients with spinal cord injury. (J) Correlation analysis between immune cells and hub genes.

3.10. Experimental validation of hub NRGs

3.10.1. Validation in human SCI peripheral blood

Quantitative real-time polymerase chain reaction (qRT-PCR) analysis confirmed significant upregulation of NLRP3 (3.188-fold, P < 0.0001) and TLR8 (2.381-fold, P < 0.0001) mRNA in SCI patient blood versus healthy controls (Fig. 9B and C). LRG1 expression increased 1.34-fold in the SCI group but did not reach statistical significance (P = 0.0531, Fig. 9A). These results align with transcriptomic data trends.

Fig. 9.

The expression of hub NRGs was validated through qRT-PCR. (A–C) show the mRNA expression levels of LRG1, NLRP3, and TLR8 in human blood samples from spinal cord injury patients and healthy controls, respectively. (D–F) The mRNA expression levels of LRG1, NLRP3, and TLR8 were detected in spinal cord tissue from rat models of spinal cord injury. (∗∗:p < 0.01; ∗∗∗∗: p < 0.0001.)

3.10.2. Validation in rat spinal cord tissues

qRT-PCR analysis of rat spinal cords demonstrated significantly increased mRNA levels in the SCI group compared to the sham-operated control group:LRG1 exhibited a significant increase of 4.022-fold (P < 0.01), NLRP3 showed a substantial rise of 6.723-fold (P < 0.0001), and TLR8 demonstrated an even greater elevation of 10.88-fold (P < 0.0001) (Fig. 8D–F). This confirms alignment with transcriptomic profiling.

4. Discussion

Spinal cord injury (SCI) is a severe neuropathological condition that leads to significant motor, sensory, and autonomic dysfunction. The secondary neuroinflammatory cascade triggered by acute injury represents a core pathological mechanism underlying the progression of neurological deficits [35]. Neutrophils, as innate immune phagocytes, play a central role in inflammation and immune defense [36,37]. In recent years, researchers have made significant advances in understanding the roles of neutrophils in pathogen clearance, immune regulation, and disease pathology [[38], [39], [40]]. Reticulated chromatin structures known as neutrophil extracellular traps (NETs) have remained at the forefront of neutrophil biology research [41]. The inflammatory regulatory role of NETs has been demonstrated in bacterial sepsis, septic shock [42], ischemia-reperfusion injury [43], atherosclerosis [44], and autoimmune diseases [45]. However, studies on neutrophils and their extracellular traps within the central nervous system remain limited. In vitro studies demonstrate that co-culturing migrating neutrophils with neurons significantly reduces neuronal viability, while inhibiting NET-associated neutrophil-derived extracellular proteases markedly attenuates neutrophil-mediated neurotoxicity [46]. This may be related to the active components within NETs. Additionally, studies have shown that increased NET production and reduced clearance impair hemorheological remodeling and vascular remodeling after brain injury [47], thereby exacerbating neurological deficits [48]. Research indicates infiltrating neutrophils generate NETs at injury sites, exacerbating secondary damage by promoting post-SCI neuroinflammation and disrupting the blood-spinal cord barrier. Limiting NETs effectively reduces SCI-induced neuronal cell death [14]. However, the precise mechanisms of NETs in SCI remain unclear.

This study integrates bioinformatics analysis with experimental validation to reveal the diagnostic value and therapeutic target potential of NETosis-related genes (NRGs) in SCI. Based on the GSE151371 dataset, we identified 4878 differentially expressed genes. KEGG pathway enrichment analysis revealed that these genes were enriched in pathways related to COVID-19 infection, Th1 and Th2 cell differentiation, PD-L1 expression, and PD-1 checkpoint signaling. COVID-19 infection-related pathways involve ERK1/2, JNK, p38, PI3K/AKT, and NF-κB signaling pathways [49]. These molecules and pathways mediate the onset and progression of inflammation, closely related to the secondary inflammatory pathomechanisms in spinal cord injury. COVID-19 infection activates various immune cells (including neutrophils, macrophages/monocytes, etc.) and cytokines (including interleukins (IL), interferons (IFN), tumor necrosis factor (TNF), and chemokines) [50,51]. Simultaneously, it activates related pathways to release neuroinflammatory cytokines such as interleukin-6, IL-15, and tumor necrosis factor-alpha (TNF-α) [52]. These cytokines can compromise the blood-brain barrier (BBB), enabling viral entry into the brain and ultimately inducing CNS disease-related symptoms [[53], [54], [55], [56]]. Under normal physiological conditions, the dynamic equilibrium between Th1 and Th2 cells and their associated pathways play a critical role in allergic and autoimmune diseases [57,58]. When interacting with antigen-presenting cells, Th1 cells are characterized by producing IFN-γ and IL-2; Th2 cells produce IL-4, IL-5, and IL-13 [59,60]. The PD-L1/PD-1 signaling pathway has been demonstrated to participate in the pathological processes of intracerebral hemorrhage (ICH), ischemic stroke, and multiple sclerosis (MS) [61]. Studies indicate that various signals in the stroke environment, such as inflammatory signals and hypoxia, can upregulate the expression of PD-L1 and PD-1 pathways. Conversely, PD-L1 mitigates acute stroke progression by suppressing inflammatory responses [62]. These findings suggest that DEGs may interact with the aforementioned signaling pathways during the immune-inflammatory processes occurring in SCI, offering new directions for subsequent immune-targeted therapeutic research in SCI.

The intersection of core modules identified by WGCNA and NRGs yielded 12 candidate genes. Machine learning models further identified NLRP3, LRG1, and TLR8 as key biomarkers with high diagnostic performance (AUC >0.9). Box plots and volcano plots revealed significantly higher expression of these three characteristic genes in the SCI group compared to the normal group (Fig. 4a). LRG1 is a secreted glycoprotein that binds to endorphins, a key component of transforming growth factor-β (TGF-β). Previous studies indicate that targeted inhibition of LRG1 expression promotes vascular normalization in tumor tissues (e.g., lung cancer, prostate cancer), thereby enhancing potential for drug delivery within tumors and lymphocyte infiltration to boost immunotherapy efficacy [63]. NLRP3 is a sensor component of the NLRP3 inflammasome, mediating inflammasome activation in response to membrane integrity defects, leading to the secretion of inflammatory cytokines IL-1β and IL-18 and pyroptosis [64]. TLR8 is an endosomal receptor playing a critical role in both innate and adaptive immunity [65,66]. It regulates the host immune response to pathogens by recognizing microbial-specific RNA degradation products [67]. Upon binding to agonists, TLR8 dimerizes, leading to activation of downstream transcription factors NF-kappa-B and IRF7, which induce proinflammatory cytokines and interferons, respectively [68]. The aforementioned studies indicate that NLRP3, LRG1, and TLR8 genes are all closely associated with immune and inflammatory responses, playing important roles in cancer and various acute and chronic inflammatory diseases. Our research findings also reached similar conclusions. This may provide potential targets for future SCI interventions.

This study integrated the complementary bioinformatics platforms GeneMANIA and STRING to construct a protein interaction network involving LRG1, NLRP3, and TLR8. The aim was to identify potential interaction targets and elucidate their possible synergistic mechanisms in SCI. We identified four core target proteins jointly predicted by both platforms: CARD8, MEFV, NEK7, and CNPY3. Among these, NEK7 is an essential kinase for NLRP3 inflammasome activation [69,70], while CARD8 and MEFV (Pyrin) are core components of other inflammasomes [71,72]. This finding strongly suggests that LRG1, NLRP3, and TLR8 may form a complex inflammatory signaling hub by interacting with these core inflammatory components, jointly regulating the maturation and release of key inflammatory factors in SCI. Additionally, platform-specific predictions revealed broader pathway connections. The GeneMANIA predicted network is rich in classical immune signaling molecules (e.g., MAVS, IRAK4, CASP1), outlining a broad innate immune and inflammatory response landscape. The STRING prediction significantly enriched other members of the NOD-like receptor (NLR) family (e.g., NLRP1, NLRP6, NLRC4), suggesting that the signaling axis under investigation may be part of a larger NLR family regulatory network. The evolutionary homology among these proteins likely contributes to their functional interconnectivity. Notably, LRG1—a protein whose full function remains unclear—was predicted to interact with multiple key inflammatory mediators [17]. This suggests LRG1 acts as a novel regulator actively involved in inflammation control, potentially functioning alongside NLRP3 and TLR8 to modulate NETosis-related pathways in SCI. This offers a novel perspective for understanding SCI pathogenesis. Of course, this conclusion stems from bioinformatics predictions, whose limitations necessitate subsequent experimental validation. Future work will focus on verifying direct binding between core genes and downstream target genes via immunoprecipitation (Co-IP) and pull-down techniques. Furthermore, using gene knockdown or inhibitor technologies in cellular and animal models, we aim to elucidate the specific functional roles of NETosis-related core genes in regulating SCI inflammation. In summary, our analysis not only predicts key interaction targets but also provides a robust theoretical foundation and clear research direction for subsequent in-depth exploration of SCI inflammatory mechanisms.

This study further examined the expression levels of NETosis-related genes via RT-qPCR. Results showed that NLRP3 and TLR8 mRNA expression levels were significantly upregulated in blood samples from SCI patients compared to healthy controls, while LRG1 mRNA expression was slightly elevated in the SCI group relative to healthy controls, though the difference was not statistically significant (p = 0.531). In rat SCI group spinal cord tissue, mRNA expression levels of LRG1, NLRP3, and TLR8 were significantly higher than in the sham-operated group, with statistically significant differences. These genes are closely associated with inflammatory responses and the NETosis process. NLRP3, a key component of inflammasomes, is typically upregulated during pyroptosis or activation of inflammatory signaling pathways [73], potentially driving neutrophil conversion to NETosis. As a Toll-like receptor, TLR8 participates in recognizing pathogen-associated molecular patterns and initiating innate immune responses. Its elevated expression may enhance inflammatory signaling [20,74], promoting NET release. LRG1 plays a role in inflammation and fibrosis [75], and its expression changes may reflect tissue repair or injury responses. In SCI patients' blood, LRG1 expression showed no significant change. This may result from blood's role as part of the systemic circulatory system, where gene expression is regulated by multiple physiological and pathological factors, potentially diluting or counteracting local injury signals. Conversely, rat spinal cord tissue, being the direct site of injury, exhibits a more concentrated inflammatory response, leading to significant upregulation of LRG1 expression. This suggests that NETosis-related gene responses may be more pronounced in local tissues, correlating with the immediate injury microenvironment. This aligns with neutrophils primarily recruiting, activating, and executing NETosis functions at the site of inflammatory injury [76]. It may also result from species differences, as human SCI patients and rat SCI models may exhibit variations in disease progression, genetic background, and immune responses, potentially leading to differing degrees of gene expression changes. Rat models offer controlled experimental conditions, whereas human samples reflect clinical complexity, underscoring the need for caution when interpreting cross-species results. Differences may also stem from smaller sample sizes, greater inter-individual variability, or limitations in detection methods. In contrast, the consistent significant upregulation of all genes in rat experiments may benefit from the model's homogeneity.

The co-upregulation of these genes collectively suggests an exacerbated inflammatory response following SCI. This may promote neutrophil-derived NET release by activating NETosis-related genes and pathways, thereby worsening tissue injury and hindering neural repair. The significant alterations in NLRP3 and TLR8 particularly imply the pivotal role of inflammasome and TLR signaling pathways in SCI pathology. The upregulation of LRG1 in spinal cord tissue may further contribute to inflammation-mediated fibrosis or apoptotic processes.

4.1. Strengths and limitations

By integrating multi-omics data with animal experiments and clinical samples, this study systematically validated the reliability of NLRP3, LRG1, and TLR8 as SCI-related clinical biomarkers through gene enrichment, protein interaction analysis, and molecular biology approaches. Previous multi-omics strategies were often limited to single-dimensional validation, struggling to establish causal relationships across data layers. Through machine learning-based multi-omics analysis and experimental validation, we clarified the expression dynamics of these biomarkers in SCI models and their direct associations with NETosis, inflammation, and barrier damage, fully addressing previous research gaps in integrating mechanistic pathways. Through protein interaction networks and in vivo experiments, this study further revealed the critical roles of these molecules in regulating NETosis, cytokine release, and barrier integrity, advancing SCI secondary pathology research from single-point observation to systemic functional validation. This study comprehensively analyzed SCI-related genes in NETs, revealing their significance in SCI disease mechanisms and providing novel insights for developing NET-targeted SCI interventions. These findings may enhance SCI treatment efficacy and offer valuable references for related disease research. Subsequent studies will further explore the clinical application potential of these biomarkers to improve outcomes for SCI patients.

However, it must be acknowledged that this study does have certain limitations. First, the relatively limited sample size (including both clinical blood samples and animal tissue samples) may restrict the generalizability and statistical power of the results; future studies should incorporate larger sample sizes to further validate the current findings. Second, gene expression changes were verified solely at the transcriptional level, without involving validation of protein activity or function. Future studies could confirm the causal roles of these genes in NETosis and SCI progression by expanding sample sizes and integrating protein assays with functional experiments (e.g., gene knockout or inhibitor interventions), thereby exploring their potential as therapeutic targets. Additionally, observed heterogeneity across datasets constitutes a potential source of bias that requires consideration and sensitivity analysis in subsequent investigations. Finally, while we identified key marker genes associated with SCI, the specific molecular mechanisms by which these genes influence SCI through regulation of neutrophil extracellular traps (NETs) remain unelucidated and require in-depth functional studies. Nevertheless, this study's focus on NETs-related genes provides valuable new insights for discovering diagnostic biomarkers and exploring therapeutic targets in spinal cord injury (SCI).

5. Conclusions

In summary, this study conducted a comprehensive bioinformatics analysis to explore the relationship between NETosis-associated genes and SCI, ultimately identifying three characteristic genes: LRG1, NLRP3, and TLR8. These hallmark genes were highly correlated with immune cell infiltration and immune responses, suggesting they may promote SCI by regulating inflammatory processes. Finally, experimental validation revealed that LRG1, NLRP3, and TLR8 likely participate in SCI-related pathological processes through NETosis-associated pathways, thereby driving SCI progression. This research will contribute to the development of novel biomarkers and potential therapeutic targets for SCI.

Ethical approval and consent to participate

This human study was approved by the Medical Ethics Committee of The First Affiliated Hospital of Guangxi Medical University (Approval No.: 2025-E0489). Informed consent was waived as this study utilized archived biological specimens obtained during prior clinical management. All procedures followed the principles of the Declaration of Helsinki.

The animal experiment was approved by the Medical Ethics Committee of The First Affiliated Hospital of Guangxi Medical University (Guangxi, China; Approval No.: 2025-E0489) and strictly conducted in accordance with the guidelines of the National Health Commission of China.

Availability of data and materials

The dataset GSE151371 generated in this study is publicly available in the GEO repository and requires no additional ethical review. Female Sprague Dawley rats (8–10 weeks old, 180–200g) were obtained from the Animal Experiment Center of Guangxi Medical University (Guangxi, China). Animals were housed under standard light-dark cycles in specific pathogen-free conditions. Clinical samples were collected from the Department of Rehabilitation Medicine at The First Affiliated Hospital of Guangxi Medical University, including blood specimens from 3 spinal cord injury patients and 3 healthy controls.

Funding

This work was supported by the Department of Science and Technology of Guangxi Zhuang Autonomous Region (Grant No.: Guike AB25069038).

CRediT authorship contribution statement

Xinliao Sun: Conceptualization, Data curation, Investigation, Validation, Visualization, Writing – original draft, Writing – review & editing. Yuchang Gui: Investigation, Validation, Writing – original draft, Writing – review & editing. Yuting Lu: Investigation, Validation, Writing – original draft, Writing – review & editing. Kewen Wang: Formal analysis, Writing – original draft, Writing – review & editing. Jingzhi Yao: Software, Validation, Writing – original draft, Writing – review & editing. Zi Li: Methodology, Writing – original draft, Writing – review & editing. Dandan Lu: Validation, Writing – original draft, Writing – review & editing. Qian Guo: Conceptualization, Data curation, Formal analysis, Methodology, Supervision, Writing – review & editing. Ruixue Liu: Conceptualization, Data curation, Formal analysis, Methodology, Supervision, Writing – review & editing. Jianwen Xu: Conceptualization, Funding acquisition, Methodology, Project administration, Supervision, Writing – review & editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

Data will be made available on request.