Identification of potential biological biomarkers for acute ischemic stroke based on integrated bioinformatics analyses

Abstract

Ischemic stroke is a major cause of disability and mortality worldwide, accounting for approximately 66% of all stroke cases. Stroke has been reported to cause approximately 6.5 million deaths annually, and the global disability-adjusted life years attributable to stroke are projected to exceed 200 million by 2030. Therefore, identifying reliable diagnostic biomarkers for acute ischemic stroke (AIS) and exploring their underlying molecular mechanisms are of great importance for early disease recognition and clinical intervention. Transglutaminases (TGases) are widely distributed in the central nervous system and play important roles in inflammatory responses, neural repair, and vascular regulation. In this study, multiple bioinformatics approaches were used to identify TGM2 as a candidate gene associated with AIS. Single-cell RNA sequencing (scRNA-seq) data were further integrated to analyze the expression pattern of TGM2 across different cell types and to explore its potential intercellular communication context. Finally, TGM2 expression was validated in peripheral blood samples from patients with AIS. This study provides new evidence supporting the potential involvement of TGM2 in AIS and its value as a candidate biomarker.

Introduction

Ischemic stroke is a major cerebrovascular disease that poses a serious threat to global health. It is characterized by high incidence, high disability rate, high mortality, high recurrence rate, and substantial economic burden [1, 2]. Its incidence has continued to increase, and the age of onset is shifting toward younger populations [3–5]. The underlying pathophysiological mechanisms of ischemic stroke are highly complex and mainly involve multiple pathological processes, including energy metabolism disorders, glutamate-mediated excitotoxicity, neuroinflammation, oxidative stress, apoptosis, and disruption of the blood–brain barrier. These processes collectively contribute to the formation of irreversible brain injury [6, 7]. Therefore, early diagnosis, monitoring of disease progression, and prognostic evaluation of acute ischemic stroke are of critical importance; however, specific and reliable biomarkers are still lacking.

Transglutaminases are a family of enzymes that catalyze post-translational modifications of proteins [8]. Among them, transglutaminase 2 (TGM2) is the most abundant and multifunctional transglutaminase in the central nervous system. It is expressed in various glial cell types, with microglia serving as the primary source of TGM2 in the brain. TGM2 plays important roles in inflammatory responses, neural repair, and vascular regulation [9–13]. Nevertheless, its exact role in ischemic stroke remains unclear.

In this study, we used an integrated bioinformatics approach to identify TGM2 as a potential key gene associated with acute ischemic stroke. Single-cell RNA sequencing data were further used to analyze the expression pattern of TGM2 across different cell types and to explore its potential involvement in intercellular communication, providing clues to its cellular context in ischemic stroke. In addition, TGM2 expression was validated in peripheral blood samples from patients with ischemic stroke, and the results were consistent with the bioinformatics findings. Overall, this study provides AIS-specific evidence for the potential biomarker value of TGM2 and offers new insights into the molecular and cellular features of ischemic stroke.

Materials and methods

Acquisition and processing of GEO datasets

Transcriptomic datasets related to acute ischemic stroke were retrieved from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) [14, 15]. The datasets used for the identification of differentially expressed genes (DEGs) included GSE58720 (mouse brain tissue; 3 control samples and 3 stroke samples), GSE30655 (mouse brain tissue; 3 control samples and 7 stroke samples), and GSE28731 (mouse brain tissue; 2 control samples and 3 stroke samples). These discovery datasets were all derived from mouse brain tissue, which helped reduce the potential influence of tissue heterogeneity. The three discovery datasets were not directly merged into a single cross-platform expression matrix. Instead, each dataset was analyzed independently using the same analytical workflow. For each dataset, gene expression data were normalized separately using the normalizeBetweenArrays function in the limma package. When applicable, probe IDs were mapped to official gene symbols according to the corresponding platform annotation files, and duplicated probes mapped to the same gene were summarized before differential expression analysis. Differential expression analysis was then performed independently for each dataset using the limma package in R software (v4.0.3) [15, 16], with the screening criteria set as |logFC| ≥ 1 and adjusted P value ≤ 0.05. Overlapping DEGs among the independent datasets were identified using Venn diagrams generated with the ggvenn package [17]. DEGs identified from mouse datasets were subsequently mapped to their corresponding human orthologs to ensure consistency in cross-species analysis.

Inflammation-related gene sets were obtained from the GeneCards database (https://www.genecards.org/) using “Inflammation” as the search keyword, and genes with a relevance score ≥ 5.0 were selected as the inflammatory gene set to enhance the specificity of the screening results. Candidate genes were identified by intersecting the overlapping DEGs with the inflammation-related gene set.

After candidate gene screening, the GSE122709 dataset was used for external validation of the target gene. This dataset consisted of peripheral blood samples from patients with ischemic stroke and controls, including 5 control samples and 5 stroke samples.

Protein–protein interaction network and enrichment analysis

Candidate differentially expressed genes were submitted to the STRING database (https://cn.string-db.org/) to construct a protein–protein interaction (PPI) network [18]. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed using the clusterProfiler package in R [19]. Statistical significance was defined as P < 0.05. These analyses were used to explore the biological functions and signaling pathways associated with the candidate genes.

Protein–protein interaction network and enrichment analysis

Single-cell RNA sequencing (scRNA-seq) data related to stroke were obtained from the GEO dataset GSE174574, which included brain tissue samples from 4 patients with ischemic stroke and 3 healthy controls. The Seurat R package was used to perform quality control, normalization, dimensionality reduction, and clustering of the raw scRNA-seq data [20, 21]. The cell filtering criteria were set as follows: nFeature_RNA > 200, nFeature_RNA < 4,000, and the proportion of mitochondrial genes (percent.mt) < 15%. After quality control, cell clustering analysis was conducted using the Seurat package, with the resolution parameter set to 0.1 based on data quality. The FindAllMarkers function was applied to identify cluster-specific marker genes, and manual cell type annotation was performed in conjunction with the CellMarker 2.0 database. Differences in the distribution of cell subpopulations between the control and disease groups were visualized using the scRNAtoolVis package. Based on human genome annotation information, the circlize package was used in combination with chromosome length data for GRCh38 obtained from UCSC to visualize the chromosomal localization of the target gene.

Cell–cell communication analysis

Based on the selected target cell subpopulations, intercellular communication analysis was performed using the CellChat R package [22]. The normalized expression matrix was obtained using the GetAssayData function and combined with cell-type annotation information to construct a CellChat object. The mouse ligand–receptor database, CellChatDB.mouse, was then loaded [23]. After calculating intercellular communication probabilities and filtering out low-confidence interactions, the overall cell–cell communication network was aggregated. Inflammatory signaling pathways, including SPP1, CCL, CXCL, and MIF, were further evaluated to characterize their communication patterns, as well as sender and receiver roles, across different cell subpopulations. Combined with the expression distribution of TGM2 among cell subpopulations, these analyses were used to explore the potential intercellular communication context associated with TGM2.

Patient blood sample collection and processing

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of the First People’s Hospital Affiliated with Yangzhou University (Approval No. 2025-YKL03-(K02)). Written informed consent was obtained from all participants. Peripheral blood samples from patients with acute ischemic stroke and healthy controls were collected at the First People’s Hospital Affiliated with Yangzhou University.

The inclusion criteria for the acute ischemic stroke group were as follows: patients who met the diagnostic criteria for acute ischemic stroke, experienced a first-ever stroke, and had not received stroke-related treatment before blood sampling. The exclusion criteria were as follows: presence of a bleeding tendency or hematological or coagulation disorders; presence of severe cardiac diseases, such as myocardial infarction or heart failure; and presence of severe hepatic or renal dysfunction, infection, or malignant tumors. A total of 140 patients with acute ischemic stroke admitted between September 1, 2024, and September 1, 2025, were enrolled as the disease group, and 100 healthy individuals undergoing physical examinations during the same period were selected as the control group. Six milliliters of peripheral venous blood was collected from the antecubital vein of each participant, gently inverted to mix, and allowed to stand, followed by centrifugation at 3,000 rpm for 10 min to separate the serum. The isolated serum was aliquoted into sterile centrifuge tubes and immediately stored at − 80 °C for subsequent analysis.

Enzyme-linked immunosorbent assay (ELISA) for TGM2 detection

Serum TGM2 levels were quantified using a commercially available ELISA kit (Wuhan Huamei Biotech, China; Cat. No. CSB-E11797h) according to the manufacturer’s instructions. The washing buffer was prepared by diluting the concentrated solution 1:20 with deionized water. Standard solutions were prepared by serial dilution as specified in the kit protocol. Prior to use, the 100× concentrated horseradish peroxidase (HRP)–conjugated detection antibody was centrifuged at 1,000 × g for 1 min and subsequently diluted 1:99 with the universal diluent.

Serum samples were thawed on ice and diluted 10-fold with the universal sample diluent. A volume of 100 µL of diluted serum sample or standard solution was added to each well, while blank wells received 100 µL of universal diluent. All serum samples were tested under the same dilution conditions. After sealing, the plate was incubated at 37 °C for 60 min. The liquid was then discarded, and each well was washed with 300 µL of 1× washing buffer, allowed to stand for 1 min, and the buffer was removed. This washing procedure was repeated three times. Subsequently, 100 µL of HRP-conjugated detection antibody working solution was added to each well, followed by incubation at 37 °C for 60 min. After discarding the solution, the plate was washed five times as described above. Next, 90 µL of TMB substrate was added to each well, and the plate was incubated at 37 °C in the dark for 15 min. The reaction was terminated by adding 50 µL of stop solution to each well, and the optical density (OD) was immediately measured at 450 nm using a microplate reader (Tecan Spark, Switzerland; asset ID: 0303092304)). Serum TGM2 concentrations were calculated based on the standard curve and expressed as serum concentration. No normalization to total protein amount was performed.

Propensity-score matching sensitivity analysis

To reduce the potential influence of baseline imbalance between the AIS and control groups, propensity-score matching (PSM) was performed as a sensitivity analysis. The propensity score was estimated using a logistic regression model, in which group status was used as the dependent variable and age, sex, hypertension, diabetes, and atrial fibrillation were included as covariates. Patients with AIS and healthy controls were then matched using a 1:1 nearest-neighbor matching method with a caliper width of 0.2. After matching, the balance of baseline characteristics between the two groups was re-evaluated.

In the matched cohort, serum TGM2 levels were compared again between AIS patients and controls. ROC curve analysis was also repeated to assess the discriminatory performance of serum TGM2 after reducing the influence of baseline confounding. The AUC with 95% confidence interval, optimal cutoff value, sensitivity, and specificity were calculated in the matched cohort.

Statistical analysis

Statistical analyses were performed using SPSS software version 26.0. Continuous variables were first tested for normality. Data with a normal distribution are presented as mean ± standard deviation (x ± s), and comparisons between groups were conducted using the independent-samples t test. Data that did not conform to a normal distribution are presented as median (interquartile range) [M (Q1, Q3)], and comparisons between groups were performed using the Mann–Whitney U test. Categorical variables are expressed as counts and percentages [n (%)], and comparisons between groups were conducted using the χ² test. Correlation analyses were performed using Pearson or Spearman correlation coefficients, as appropriate based on data distribution, to evaluate the association between serum TGM2 levels and the occurrence of acute ischemic stroke. Binary logistic regression analysis was applied to identify independent factors associated with the occurrence of acute ischemic stroke. Receiver operating characteristic (ROC) curves were generated, and the area under the curve (AUC) was calculated to assess the predictive performance of TGM2. A p < 0.05 was considered statistically significant.

Results

Identification of key genes

The overall workflow of this study is illustrated in Fig. 1. First, differential expression analyses were performed on three mouse brain transcriptomic datasets, and the results are presented as volcano plots (Fig. 2A–C). A total of 1,064 DEGs were identified in GSE58720, including 885 upregulated and 179 downregulated genes; 511 DEGs were identified in GSE30655, comprising 385 upregulated and 126 downregulated genes; and 223 DEGs were identified in GSE28731, including 209 upregulated and 14 downregulated genes. After intersecting the DEGs from the three datasets, 142 common differentially expressed genes were obtained (Fig. 2D). In parallel, 729 inflammation-related genes were retrieved from the GeneCards database. The intersection of the inflammation-related genes with the 142 common DEGs yielded 29 candidate key genes (Fig. 2E).

Fig. 1.

Schematic illustration of the overall workflow of this study

Fig. 2.

Analysis of differentially expressed genes and inflammation-related hub genes in the GSE28731, GSE30655, and GSE58720 datasets. A–C Volcano plots showing the differentially expressed genes in the three datasets, respectively. D Intersection of DEGs from the three datasets, yielding 142 common differentially expressed genes. E Venn diagram analysis of the common DEGs and the inflammation-related gene set, identifying 29 candidate genes

PPI network and enrichment analysis

GO and KEGG enrichment analyses were performed on the 29 candidate genes (Fig. 3A–B). In the Gene Ontology biological process (BP) category, the genes were mainly enriched in leukocyte migration, myeloid leukocyte migration, positive regulation of tumor necrosis factor superfamily cytokine production, positive regulation of tumor necrosis factor production, and monocyte chemotaxis. In the cellular component (CC) category, the genes were primarily enriched in collagen-containing extracellular matrix, receptor complex, plasma membrane–anchored component, immunological synapse, and glial cell projection. In the molecular function (MF) category, the genes were mainly enriched in cytokine receptor binding, cytokine activity, G protein–coupled receptor binding, chemokine receptor binding, and chemokine activity. KEGG pathway analysis indicated that these genes were predominantly enriched in lipid and atherosclerosis, the tumor necrosis factor (TNF) signaling pathway, the Toll-like receptor signaling pathway, viral protein interaction with cytokine and cytokine receptors, and the interleukin-17 (IL-17) signaling pathway. In addition, a protein–protein interaction (PPI) network of the 29 candidate genes was constructed based on the STRING database (Fig. 3C). To further evaluate the consistency of the candidate genes in peripheral blood, validation was performed using the human peripheral blood transcriptomic dataset GSE122709. The results showed that seven genes remained significantly differentially expressed between the stroke group and the control group, namely CD14, CYBA, CXCL10, TGM2, LCN2, VWF, and SERPINE1 (Fig. 3D).

Fig. 3.

Functional enrichment analysis and validation of differentially expressed genes. A, B GO and KEGG enrichment analysis results of the 29 candidate genes; C PPI network constructed using STRING; D Validation results in the GSE122709 dataset (differential expression of seven genes)

Single-cell analysis and chromosomal localization

Based on the single-cell transcriptomic data from GSE174574, cells were subjected to quality control, normalization, dimensionality reduction, clustering, and cell type annotation. Ultimately, 12 distinct cell subpopulations were identified and annotated, including endothelial cells, CD14⁺ monocytes, homeostatic microglial cells, antigen-presenting cells, mural cells, radial glial progenitor cells, choroid plexus cells, oligodendrocytes, monocytes, ependymal cells, lymphocytes, and oligodendrocyte precursor cells (Fig. 4A). Further analysis of TGM2 expression across different cell subpopulations revealed that TGM2 exhibited relatively high expression levels and/or significant intergroup differences in CD14⁺ monocytes, antigen-presenting cells, monocytes, and oligodendrocyte precursor cells (OPCs) (Fig. 4B–C). In addition, chromosomal localization analysis demonstrated that TGM2 is located on human chromosome 20 (Fig. 4D).

Fig. 4.

Single-cell RNA sequencing analysis and chromosomal localization of TGM2. A Clustering and annotation of 12 cell subpopulations; B–C Expression distribution and differences of TGM2 across different cell subpopulations; D Chromosomal localization of TGM2 (Chr 20)

Intercellular communication analysis

Based on the single-cell analysis results, we performed cell–cell communication analysis on the 12 selected cell subtypes. Compared with the control group, myeloid cells in the stroke group, including CD14⁺ monocytes, monocytes, and antigen-presenting cells, appeared to have broader communication connections with multiple cell subpopulations (Fig. 5A–B). Further analysis of outgoing and incoming signaling patterns showed that these myeloid cells displayed relatively strong signal-sending activity in inflammation-related pathways, including SPP1, CCL, and CXCL (Fig. 5C). They also showed marked signal-receiving features in these pathways (Fig. 5D), suggesting that myeloid cells may be important participants in the inflammatory communication network after stroke.

Fig. 5.

Results of cell–cell communication analysis. A–B Visualization of the cell–cell communication networks in the MCAO and Sham groups; C Heatmap of outgoing signaling roles, showing the relative strength of different cell subpopulations as signal senders; D Heatmap of incoming signaling roles, showing the signal-receiving strength of different cell subpopulations in major inflammation-related pathways, such as SPP1, CCL, and CXCL

Clinical validation of serum TGM2

Comparison of general clinical characteristics and serum TGM2 levels

A total of 240 participants were enrolled in this study, including 100 healthy controls and 140 patients with acute ischemic stroke (AIS). Fasting serum samples were collected from the control group in the early morning, and fasting serum samples were collected from the AIS group on the morning of the day following hospital admission for the measurement of serum TGM2 levels. Normality of continuous variables was assessed using the Shapiro–Wilk test. Clinical variables with a normal distribution were compared using the independent-samples t test (b), whereas variables with a non-normal distribution were analyzed using the Mann–Whitney U nonparametric test (c). Comparisons of categorical variables were performed using the chi-square test (a).Comparison of general clinical characteristics between the two groups showed no statistically significant differences in sex, body mass index (BMI), hyperlipidemia, alcohol consumption history, or several laboratory parameters, including hemoglobin (Hb), triglycerides (TG), total cholesterol (TC), uric acid (UA), and alanine aminotransferase (ALT) (all p > 0.05), indicating good comparability between the two groups with respect to most baseline characteristics. In terms of medical history, the proportions of hypertension, diabetes mellitus, coronary heart disease, and atrial fibrillation were significantly higher in the AIS group than in the control group (all p < 0.05), suggesting that these cardiovascular and cerebrovascular risk factors are closely associated with the occurrence of AIS and may represent independent risk factors. In addition, compared with the control group, plasma TGM2 levels were significantly elevated in patients with AIS [median: 44.49 ng/L vs. 126.12 ng/L; p < 0.001], with the difference reaching statistical significance. These findings suggest that TGM2 may be involved in the occurrence and progression of AIS and may have potential clinical value (Table 1).

Table 1.

Comparison of clinical characteristics between the control group and AIS group

Variable	Control group (n = 100)	AIS group (n = 140)	Test statistic	P value
General clinical characteristics
Male, n(%)	59(59)	89(86.3)	0.516a	0.473
Age, years	50 ± 14	69 ± 11	-12.07b	<0.001
BMI, kg/m2	23.70(21.38,26.45)	24.22(22.31,26.06)	-1.064c	0.287
Hypertension, n (%)	23(23)	106(75.7)	65.205a	< 0.001
Diabetes mellitus, n (%)	10(10)	41(29.3)	12.965a	< 0.001
Coronary heart disease, n (%)	2(2)	17(12.1)	8.232a	0.004
Atrial fibrillation, n (%)	2(2)	30(21.4)	19.055a	< 0.001
Hyperlipidemia, n (%)	23(23)	19(13.6)	3.592a	0.058
Smoking history, n (%)	20(20)	47(33.6)	5.339a	0.021
History of alcohol consumption, n (%)	21(21)	44(31.4)	3.212a	0.073
WBC, ×109/L	6.70(5.16, 7.05)	6.82(5.67, 8.83)	-3.441c	<0.001
Hb, g/L	143(130.25, 152.75)	138.5(123, 149.75)	-1.744c	0.081
PLT, ×109/L	209.5(177.25, 257)	169(132.5, 214)	-5.321c	<0.001
TG, mmol/L	1.39(0.99, 2.15)	1.36(0.94, 2.11)	-0.421c	0.674
TC, mmol/L	4.72 ± 1.17	4.46 ± 1.10	1.74b	0.083
LDL-C, mmol/L	2.72(2.04, 3.21)	2.73(2.09, 3.25)	-0.268c	0.789
HDL-C, mmol/L	1.18(1.01, 1.33)	1.09(0.94, 1.32)	-1.98c	0.048
BG, mmol/L	5.24(4.83, 6.02)	6.77(5.64, 8.63)	-6.365c	<0.001
Alb, g/L	44.15(42.65,45.7)	41.25(38.9,44.25)	-5.298c	<0.001
Scr, umol/L	71.15(59.9, 82.83)	64.95(53.63, 80.80)	-2.102c	0.037
UA, umol/L	316.4(259.48, 395.35)	296.4(245.55, 369.58)	-1.618c	0.106
ALT, U/L	20.65(14.25,33.28)	21(16,30)	-0.111c	0.911
TGM2, pg/ml	44.49(28.21, 59.92)	126.12(79.55, 194.33)	-1.39c	<0.001

Data are presented as mean ± SD or median (interquartile range), as appropriate

^aχ² test

^bt test

^cMann–Whitney U rank-sum test

Correlation between serum TGM2 Levels and the occurrence of AIS

To further clarify the relationship between serum TGM2 levels and AIS, Spearman’s rank correlation analysis was performed, with the occurrence of ischemic stroke treated as one variable and serum TGM2 levels in the control and AIS groups treated as the other variable. The results showed a significant positive correlation between serum TGM2 levels and the occurrence of AIS (r = 0.672, P < 0.001), suggesting that TGM2 may be involved in the pathogenesis and progression of AIS and may serve as a potential biomarker. Detailed results are presented in Table 2.

Table 2.

Spearman correlation analysis between serum TGM2 Levels and AIS occurrence

Variable	TGM2
Spearman correlation coefficient (r)	0.672
P value	< 0.001

Binary logistic regression analysis

To exclude the potential influence of confounding factors on the occurrence of AIS, a binary logistic regression analysis was performed. The occurrence of ischemic stroke was set as the dependent variable, while variables that were statistically significant (P < 0.05) in univariate analysis—including TGM2, age, hypertension, diabetes mellitus, atrial fibrillation, coronary heart disease, smoking history, white blood cell count, platelet count, high-density lipoprotein cholesterol, blood glucose, and albumin—were included as independent variables in the model. The results showed that TGM2, hypertension, diabetes mellitus, and white blood cell count were significantly and positively associated with the occurrence of ischemic stroke. Notably, elevated TGM2 concentration was identified as an independent risk factor for AIS. In contrast, albumin acted as a protective factor, with higher levels being significantly associated with a reduced risk of ischemic stroke. Other variables, including atrial fibrillation, coronary heart disease, smoking history, platelet count, high-density lipoprotein cholesterol, blood glucose, and serum creatinine, did not show statistically significant associations (P > 0.05). Overall, increased TGM2 concentration exhibited a stable and independent association with an elevated risk of ischemic stroke. Detailed results are presented in Table 3.

Table 3.

Binary logistic regression analysis of factors associated with acute ischemic stroke

Variable	B	SE	Waldχ2	P value	OR	95%CI
TGM2	0.044	0.010	21.189	<0.001	1.045	1.025–1.064
Age	0.87	0.027	10.245	0.001	1.091	1.034–1.151
Hypertension	2.288	0.651	12.364	<0.001	9.859	2.753–35.299
Diabetes mellitus	1.729	0.775	4.978	0.026	5.636	1.234–25.742
Atrial fibrillation	-0.016	1.037	0.000	0.987	0.984	0.129–7.505
Coronary heart disease	-0.358	1.118	0.103	0.749	0.699	0.078–6.251
Smoking history	0.267	0.672	0.158	0.691	0.765	0.205–2.857
WBC	0.505	0.170	8.810	0.003	1.657	1.187–2.314
PLT	-0.005	0.006	0.848	0.357	0.995	0.984–1.006
HDL-C	-0.681	0.863	0.621	0.431	0.506	0.093–2.751
BG	0.163	0.114	2.025	0.155	1.177	0.940–1.473
Alb	-0.200	0.075	7.207	0.007	0.818	0.707–0.947

ROC curve analysis of serum TGM2 for AIS prediction

To further evaluate the discriminatory performance of serum TGM2 levels for AIS, ROC curve analysis was performed. The results showed that TGM2 concentration had good diagnostic value for AIS, with an AUC of 0.894 (95% CI: 0.854–0.933, P < 0.001). Based on the Youden index, the optimal cutoff value was determined to be 60.73, with a sensitivity of 82.9% and a specificity of 77.0% (Fig. 6). These findings suggest that serum TGM2 levels have good discriminatory ability for AIS.

Fig. 6.

ROC curve of serum TGM2 for predicting the occurrence of AIS

Propensity-score matching analysis of serum TGM2

To further reduce the potential influence of baseline confounding, PSM was performed using age, sex, hypertension, diabetes, and atrial fibrillation as matching variables. After matching, 39 matched pairs of AIS patients and controls were retained. Serum TGM2 levels remained significantly higher in the AIS group than in the control group [170.45 (104.27–263.28) vs. 45.66 (30.54–62.04), P < 0.001]. ROC analysis in the matched cohort showed that serum TGM2 retained good discriminatory ability for AIS, with an AUC of 0.957 (95% CI: 0.919–0.996). The optimal cutoff value was 90.00, with a sensitivity of 84.6% and a specificity of 94.9% (Fig. 7).

Fig. 7.

ROC curve analysis of serum TGM2 after propensity-score matching

Discussion

Ischemic stroke is caused by interruption of cerebral blood supply, leading to focal tissue ischemia, hypoxia, and disturbances in energy metabolism [3, 6, 7], which subsequently induce a series of pathological cascades, including glutamate-mediated excitotoxicity, oxidative stress, apoptosis, neuroinflammatory responses, and disruption of the blood–brain barrier, ultimately resulting in irreversible brain injury [24, 25] .Transglutaminase 2 (TGM2) participates in processes such as cell survival, inflammatory regulation, and tissue remodeling in the nervous system, and previous studies have suggested that it plays important biological roles in various neurological diseases [12] However, systematic investigations of TGM2 in acute ischemic stroke (AIS) remain limited. In this study, through integrated multi-omics analyses, single-cell transcriptomic profiling, and clinical sample validation, we systematically evaluated the expression characteristics of TGM2 in AIS and explored its potential biological significance.

At the transcriptomic level, 29 candidate genes were identified through integration of three independent brain tissue datasets combined with an inflammation-related gene set. Functional enrichment analyses revealed that these genes were mainly involved in leukocyte migration, chemokine signaling, and atherosclerosis-related pathways, which are highly consistent with immune activation and vascular pathology observed in ischemic stroke. Subsequent validation using a human peripheral blood dataset showed that only seven genes—CD14, CYBA, CXCL10, TGM2, LCN2, VWF, and SERPINE1—remained consistently differentially expressed, suggesting that they may represent a core peripheral blood molecular signature associated with stroke occurrence. Among these genes, TGM2 demonstrated the greatest consistency and stability across multiple analyses. Previous studies have shown that TGM2 regulates several signaling pathways involved in neurogenesis and inflammatory responses, including the PI3K–Akt, MAPK, and Notch pathways [9, 26]. Loss of TGM2 markedly suppresses the proliferation and neuronal differentiation of neural stem/progenitor cells (NSPCs), whereas its overexpression promotes cell migration and enhances neurogenesis following injury, indicating a critical role for TGM2 in endogenous repair processes in damaged brain tissue [10, 27]. These mechanistic findings provide a plausible biological basis for the abnormal TGM2 expression observed in peripheral blood in the present study. In addition to TGM2, the other candidate genes also showed notable contributions in differential expression and network analyses, suggesting their involvement in the initiation and progression of ischemic stroke. CD14 is a key mediator of innate immune responses and inflammation [28, 29]. CYBA participates in reactive oxygen species (ROS) production and has been linked to increased risk of cardiovascular and cerebrovascular diseases [30, 31]. CXCL10 is involved in inflammatory signaling and inhibition of angiogenesis [32, 33] and plays an important role in atherosclerosis and ischemic cerebrovascular disorders [34, 35]. LCN2 is an inflammation-associated lipid-binding protein predominantly expressed by astrocytes after cerebral ischemia, and its upregulation can amplify neuroinflammation and aggravate brain injury [36, 37]. VWF is essential for platelet adhesion and thrombus formation [38], while SERPINE1 regulates fibrinolysis and contributes to cell adhesion and vascular homeostasis [39–41]. Collectively, these genes may play important roles in the pathogenesis and progression of ischemic stroke, although their precise mechanisms warrant further investigation.

Further single-cell analysis showed that TGM2 was highly expressed in four cell subpopulations, including CD14⁺ monocytes, antigen-presenting cells (APCs), monocytes, and oligodendrocyte precursor cells (OPCs). These findings suggest that TGM2 may be associated with both inflammatory responses and repair-related processes after ischemic stroke. Previous studies have reported that TGM2 is highly expressed in myeloid cells and is involved in macrophage polarization, inflammatory mediator secretion, and tissue repair [42, 43]. In addition, microglia, as important antigen-presenting cells in the brain, have been reported to be a major source of TGM2, and TGM2 may influence neural circuit stability by modulating microglial phagocytic activity [9]. Together with our single-cell analysis, these findings suggest that TGM2 may be linked to myeloid-cell-associated phagocytic programs and inflammatory signaling during the injury and repair responses after ischemic stroke. However, these results should be interpreted as association-based evidence and require further experimental validation.

Cell–cell communication analysis showed that CD14⁺ monocytes, monocytes, and APCs were located at central positions in the stroke-related inflammatory communication network and exhibited extensive signaling interactions with multiple cell subpopulations. Ligand–receptor analysis suggested that these myeloid-related cells displayed both signal-sending and signal-receiving patterns in classical inflammatory axes, including SPP1–CD44, CCL–CCR, and CXCL–CXCR. These findings indicate that myeloid-related cells may be important participants in inflammatory communication after ischemic stroke. Consistent with previous studies, microglia in the central nervous system are rapidly activated after AIS and may acquire a pro-inflammatory phenotype, releasing neurotoxic mediators and contributing to neuronal injury through excessive phagocytosis [44, 45]. Peripheral myeloid cells, such as neutrophils and monocytes, can infiltrate the brain parenchyma through the disrupted blood–brain barrier, and their adhesion and migration may further enhance inflammatory responses and impair blood–brain barrier integrity [46]. Interactions between resident microglia and infiltrating peripheral myeloid cells may aggravate neuroinflammation and neurological dysfunction [47]. As the disease progresses, myeloid cells may also exhibit anti-inflammatory, immunosuppressive, phagocytic, angiogenic, and tissue-remodeling features [48, 49]. Thus, myeloid cells are not only associated with post-stroke inflammation but may also participate in repair-related processes. In this context, the differential expression pattern of TGM2 in these myeloid-related cell populations suggests that TGM2 may be involved in the inflammatory–repair response mediated by myeloid cells after ischemic stroke.

At the clinical level, our findings suggest the potential relevance of TGM2 as a serum biomarker in AIS. Serum TGM2 levels were significantly higher in patients with AIS than in healthy controls. After adjustment for major vascular risk factors and inflammation-related indicators, TGM2 remained independently associated with AIS. ROC curve analysis further showed that serum TGM2 had good discriminatory ability for distinguishing patients with AIS from healthy controls, supporting its potential value as a candidate serum biomarker. Importantly, the association between serum TGM2 levels and AIS remained evident after propensity-score matching, suggesting that the observed difference was not solely explained by baseline differences in age, sex, or major vascular comorbidities. Previous studies have reported various candidate biomarkers for AIS, including inflammation-related factors, vascular injury-related molecules, and miRNAs. Markers such as MMP-9, CRP, Lp-PLA2, and VEGF mainly reflect blood–brain barrier injury, atherosclerotic inflammation, inflammatory cascade activation, and vascular repair [50, 51], whereas miRNAs are involved in vascular inflammation, carotid atherosclerosis, plaque instability, and post-transcriptional regulation [52]. Unlike these biomarkers, TGM2 is a multifunctional enzyme with transamidase activity, GTP-binding activity, and GTPase activity, and it has been implicated in extracellular matrix remodeling and inflammatory responses [8]. Moreover, TGM2 is expressed in multiple cell types, including neurons, astrocytes, microglia, and oligodendrocyte precursor cells [53], suggesting that it may reflect cell-type-specific inflammatory and repair-related changes in AIS.

The novelty of this study lies in identifying and validating TGM2 as a potential serum biomarker in AIS. Single-cell and cell–cell communication analyses suggest its involvement in myeloid cell–mediated inflammatory responses, and clinical validation further supports its potential biomarker value. Several limitations should be acknowledged. First, the clinical validation was performed in a single center with a limited sample size, which may introduce selection bias and requires further validation in larger multicenter cohorts. Second, only pretreatment serum TGM2 levels were measured, and dynamic changes during disease progression or treatment were not evaluated. Finally, this study was mainly based on association analyses, and further in vivo and in vitro experiments are needed to clarify the mechanistic role and clinical utility of TGM2 in AIS.

Conclusions

This study identified TGM2 as a potential serum biomarker for AIS. Integrated bioinformatics, single-cell transcriptomic, and clinical validation analyses suggest that TGM2 may be associated with AIS-related inflammatory responses and may provide additional value for the clinical evaluation of AIS.

Authors’ contributions

Linling Ji: Designed the research, performed the experiments, analyzed the scRNA-seq data, data curation, and wrote the original draft. Lin Zhu: Interpreted experimental results, and contributed to writing-review & editing. Yiran Zhu: Data curation, and contributed to writing-review & editing. Kaiqi Ren: Analyzed the scRNA-seq data , and contributed to writing-review & editing. Jingyan Liang: Funding acquisition, and contributed to writing-review & editing. Yingge Wang: Funding acquisition, supervision, resources, writing-review & editing, and approved the final manuscript.

Funding

This work was supported by the Social Development Project Fund of Yangzhou City (Grant No. YZ2023067), the National Natural Science Foundation of China (NSFC, Grant No. 82370517).

Data availability

The publicly available datasets analyzed in this study are available in online databases, and the corresponding accession numbers are provided in the article . The clinical data were collected from our hospital and contain sensitive patient information; therefore, they are not publicly available. Reasonable requests for access to the clinical data may be directed to the corresponding author, subject to approval by the institutional ethics committee.

Declarations

Ethics approval and consent to participate

The study design and procedures were conducted in accordance with the Declaration of Helsinki and were approved by the Ethics Committee of the Affiliated First People’s Hospital of Yangzhou University (Approval No. 2025-YKL03-(K02)). Written informed consent to participate was obtained from all participants prior to their inclusion in the study.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.