Unraveling the neurotoxic mechanisms of tranexamic acid in epilepsy induction: a network toxicology and molecular docking approach

Abstract

Background

Tranexamic acid, an antifibrinolytic drug, is effective for surgical hemostasis and bleeding-related disorders but may induce neurotoxicity, such as epilepsy. The mechanisms driving TXA-induced epilepsy remain poorly understood, as traditional toxicology approaches fall short in capturing its complex toxicity profile. This study employs network toxicology and molecular docking to elucidate the multi-target molecular mechanisms of TXA-induced epilepsy, aiming to enhance its safe clinical application.

Methods

TXA toxicity was predicted using ADMETlab2.0, PROTOX3.0, toxCSM, and ADMET-AI, with SMILES sequences obtained from PubChem, and a TXA target library was constructed using databases such as ChEMBL. EP-related targets were screened from databases such as GeneCards, and target intersections were analyzed using R software. Networks were constructed using Cytoscape and STRING, with GO and KEGG analyses performed to investigate molecular pathways. Molecular docking was conducted using the CB-Dock2 platform to analyze TXA binding to core targets.

Results

TXA poses a neurotoxicity risk, with 51 intersecting targets identified between TXA and EP, among which GABRA1, GABBR2, GABRA5, GABRA2, and GAD1 exhibited the greatest relevance. Analysis revealed that TXA induces EP by disrupting GABA signaling and synaptic function. The PPI network confirmed five core targets, and GO and KEGG analyses elucidated their roles in neural suppression. Molecular docking demonstrated stable binding of TXA to these targets, with Vina scores ranging from − 5.2 to − 6.9.

Conclusion

This study elucidates the mechanisms of TXA-induced EP through network toxicology and molecular docking, confirming its interference with GABA-related targets and neural suppression functions. The study overcomes the limitations of traditional toxicology, providing a scientific basis for the safe use of TXA and prevention of neurotoxicity.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-025-04605-x.

Background

Tranexamic acid (TXA), with the chemical formula C₈H₁₅NO₂, is an antifibrinolytic drug widely used in surgical settings, renowned for its potent hemostatic properties [1, 2]. TXA effectively controls or reduces bleeding by inhibiting excessive fibrinolysis, holding a significant position in international medical practice. Its clinical applications span multiple fields, including the treatment of heavy menstrual bleeding, prevention of blood loss in major surgeries (e.g., cardiac, orthopedic, and postpartum hemorrhage), reduction of mortality risk in trauma patients, and management of hereditary angioedema [3–6]. In medicinal chemistry, TXA, as a lysine analog and its derivatives, serves as a critical structural template and tool molecule for studying antifibrinolytic mechanisms and developing novel hemostatic agents, anti-inflammatory drugs, and treatments for vascular angioedema [7]. Its exceptional hemostatic efficacy establishes it as a cornerstone drug in the medical field. In the Expert Consensus on the Application of TXA and Anticoagulants in the Perioperative Period of Enhanced Recovery After Orthopedic Surgery in China, the recommendation of TXA for topical use in spinal surgery has sparked widespread controversy [8]. Given TXA’s extensive use in clinical and research settings, elucidating its toxicological mechanisms is of paramount importance for optimizing its safe application and developing more effective therapeutic strategies.

To date, the most common toxicities associated with TXA include gastrointestinal reactions, thromboembolism, visual disturbances, and central nervous system effects [9–12]. Among reported TXA adverse reactions, literature on thromboembolism risk is abundant, whereas studies on epilepsy (EP) induction risk are relatively scarce. EP is a chronic condition caused by sudden abnormal neuronal discharges in the brain, often resulting in transient brain dysfunction [13]. Currently, most studies focus on TXA’s coagulant effects, while the potential mechanisms underlying EP induction remain insufficiently elucidated. Thus, whether and how TXA induces EP risk warrants further in-depth investigation. As a higher neural center, the brain’s abnormal neuronal discharges are triggered by complex and diverse factors, making the identification of underlying issues particularly challenging. Therefore, preventing the occurrence of EP is of paramount importance. Multiple studies have demonstrated TXA’s broad and excellent clinical value in managing severe trauma, surgical procedures, obstetric hemorrhage, coagulation disorders, and vascular angioedema, profoundly impacting the development of surgical medicine [14–16]. Consequently, thoroughly investigating TXA’s adverse effects holds significant practical importance. However, traditional toxicological testing methods struggle to comprehensively and systematically evaluate the interrelationships among TXA’s various toxicities. Future research should adopt more advanced methodologies to elucidate the underlying mechanisms of TXA’s toxic effects, providing a scientific basis for its safe application.

This situation is expected to change, as network toxicology and molecular docking, as novel network-based assessment methods, provide innovative perspectives for toxicology research. Network toxicology integrates multidisciplinary technologies such as bioinformatics, big data analysis, and genomics, focusing on the toxicological pathways of compounds and their disease-related molecular mechanisms [17]. Through systematic network analysis, this field aims to reveal the interactions and toxic effects of different molecules in biological systems, deeply exploring the complex relationships between chemicals, biological targets, and adverse reactions, thereby enhancing the systematic understanding of toxicological effects. Network toxicology breaks through the traditional “single target, single drug” research paradigm, proposing a new “multi-target, single drug” model, which contrasts sharply with traditional toxicology methods. Through a network perspective, it clearly demonstrates the dynamic interactions of genes, proteins, and metabolites in disease and drug action [18, 19]. Molecular docking technology enables researchers to simulate the spatial conformation and binding interactions of small molecule ligands with biological macromolecules (such as receptors or enzymes) at active sites, providing important insights into potential therapeutic effects or side effects. As a key computational tool in drug discovery, molecular docking has revolutionized the study of interactions between small molecules and biological target molecules. In the field of toxicology, this technology can be used to predict and analyze the mechanisms of interaction between toxins and biological molecules, revealing the toxic characteristics of toxins and their potential harm to organisms [20, 21].

The purpose of this study is to elucidate the complex molecular pathways and interaction mechanisms related to TXA-induced neurotoxicity and EP, while providing a scientific basis for formulating relevant safety measures and developing TXA-related products. The core advantage of this technology lies in its ability to efficiently analyze vast amounts of biological information and datasets, thereby predicting and identifying potential toxic effects and providing a comprehensive and systematic assessment of the potential risks of TXA exposure. Furthermore, an in-depth understanding of the toxicological mechanisms of TXA aids in evaluating its safety in medical applications and drug development, and provides a solid scientific foundation for the formulation of relevant safety standards and protective measures.

Methods

Network toxicology analysis

The toxicological profile of TXA was evaluated using the ADMETlab 2.0, ProTox 3.0, toxCSM, and ADMET-AI databases. Initially, the SMILES string for TXA was retrieved from the PubChem database [22–27]. This SMILES string was then inputted into each of the aforementioned databases to conduct toxicity predictions, and the resulting data were downloaded for further analysis [28–34].

Construction of tranexamic acid targets

In this study, we first searched for “Tranexamic Acid” in the ChEMBL database, confirmed the name and molecular formula of the best match, and then retrieved and obtained the target information for TXA, specifying the species as human [35]. Second, to verify the reliability of the targets, we used the STITCH database and searched based on the SMILES sequence of TXA (with the species specified as human and the minimum interaction score set to 0.400) to obtain predicted target information [36]. Finally, we utilized the SwissTargetPrediction database, also based on the SMILES sequence and specifying the species as human, to retrieve predicted targets. The result files from the above operations are detailed in (Additional file 1). By integrating the target data from the three databases, removing duplicate targets, and taking the union, we constructed the final TXA target library (Additional file 2) and drew a Venn diagram illustrating the relationship between the targets from the three databases [37–41].

Construction of disease targets

The EP disease target library was constructed by integrating data from GeneCards, OMIM, and TTD databases, using the keyword “Epilepsy” to retrieve relevant targets (Additional file 3). To ensure reliability, two filtering criteria were applied. First, GeneCards targets were filtered to include only those with a Relevance score > 10, which integrates literature, functional, and pathway evidence to prioritize genes strongly associated with EP while excluding weakly related targets. Second, a STITCH interaction score ≥ 0.400 was used to select protein–compound interactions with at least medium confidence, a threshold widely adopted in network pharmacology and toxicology studies to balance reliability and connectivity. The filtered GeneCards targets were then combined with those from OMIM and TTD, duplicates were removed, and a union set was created to form the final EP disease-related target collection (Additional file 4). A Venn diagram was generated to visualize the relationships among targets from the three databases [23, 42–52].

Intersection of drug and disease targets

The intersection analysis between drug targets (TXA target library) and disease targets (EP target library) was conducted using R software (version 4.4.1). By performing an intersection analysis on the TXA and EP target libraries in R, common target genes were identified (Additional file 5). A total of 51 core intersection genes were obtained, and the results were visually represented in a Venn diagram.

Construction of the drug-target-pathway network

The drug-target-pathway network was constructed to visually represent the relationships between the drug, its targets, and associated pathways, as well as to highlight the connections between key targets and the analysis results, thereby identifying core targets and key pathways. This involved integrating the core targets corresponding to the drug and all pathways related to each core target. Subsequently, the attribute table from the Network file was imported into Cytoscape software to load node positions, set the starting point, and adjust the image layout to generate a visual network diagram [53].

PPI network construction

The PPI network was constructed using the STRING database [54]. The 51 core intersection genes identified previously (Additional file 5) were uploaded to the STRING database, with the species specified as Homo sapiens, to generate the PPI network (Fig. 4). Additionally, the interaction data were downloaded in TSV format. Subsequently, Cytoscape was utilized to perform topological structure analysis on the downloaded PPI network, identifying key network attributes and hub nodes, and the results were visualized [55, 56].

Fig. 4.

Diagram of the intersection gene target pathway network of tranexamic acid and epilepsy

GO and KEGG pathway analysis

Gene ontology (GO) enrichment analysis, encompassing biological process (BP), cellular component (CC), and molecular function (MF), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis were conducted on the 51 core intersection genes (Additional file 5) using the clusterProfiler package in R. Gene identifiers were converted to Entrez IDs via the org.Hs.eg.db package. To comprehensively capture all potentially relevant biological terms and pathways and avoid premature exclusion of biologically meaningful results, both analyses were initially performed with pvalueCutoff = 1 and qvalueCutoff = 1 to retain the full set of enriched terms. Statistically significant results were subsequently defined as those with both raw p value < 0.05 and adjusted p value < 0.05. Significant GO terms were exported to Additional file 6, and significant KEGG pathways were converted from Entrez IDs to gene symbols and exported to Additional file 7. Visualization of GO enrichment included (1) faceted bar plots (grouped by ontology, top 10 terms each); (2) faceted bubble plots (ranked by gene ratio); (3) categorized bar plots (colored by ontology, top 10 terms each); and (4) circle plots (showing gene count, enrichment factor, and p value gradient) [39]. KEGG pathway visualization comprised (1) bar plots (top 30 or all significant pathways); (2) bubble plots (ranked by gene ratio); and (3) lollipop plots (with line lengths representing gene count and point sizes indicating significance) [57–60].

Molecular docking

The computational analyses employed (network toxicology and molecular docking) model ligand–target interactions, but they do not incorporate dose–response or route-dependent effects, which are known determinants of TXA neurotoxicity. This methodological limitation should be addressed in future in silico pipelines. The crystal structures of the human proteins GAD1(PDB ID:2okj), GABRA1(PDB ID:6cdu), GABBR2(PDB ID:4f11), GABRA2(PDB ID:9cx7), and GABRA5(PDB ID:8bej) were obtained from the Protein Data Bank (PDB) and subjected to molecular docking using the CB-Dock2 platform, which integrates automated binding pocket detection, docking, and visualization based on the AutoDock Vina algorithm [61–64]. Compared with standalone AutoDock Vina, CB-Dock2 offers automated cavity prediction via the CurPocket algorithm, streamlined input preparation, and cloud-based computation, enabling high-throughput blind docking without manual grid box definition, particularly suitable when detailed information on the target’s active site is limited [65]. The three-dimensional (3D) structure of TXA was uploaded as the ligand, followed by sequential docking with the PDB structures of the five target proteins as receptors. Binding affinities were evaluated using the Vina score generated by CB-Dock2, derived from the AutoDock Vina scoring function. The Vina score represents the predicted binding free energy (kcal/mol), with more negative values indicating stronger predicted affinity. In this study, the lowest Vina score for each ligand was used as the primary docking performance indicator. All dockings were performed in a structure-based blind docking mode, and the docking pose corresponding to the lowest score was considered the most stable binding mode. Three-dimensional visualizations of the resulting ligand–receptor complexes were generated [66]. PDB ID and related information are detailed in Table 3.

Table 3.

PDB ID and information

Gene	PDB ID	Resolution (Å)	Total structure weight (kDa)	Atom count	Modeled residue count	Deposited residue count	Unique protein chains
GAD1	2okj	2.3	115.31	8314	1005	1008	1
GABRA5	8bej	3.24	206.4	13,650	1680	1800	1
GABRA2	9cx7	3.30	329.72	14,573	1755	2851	6
Gabra1	6cdu	3.45	375.58	25,930	3140	3240	1
GABBR2	4f11	2.38	49.13	3478	417	433	1

Results

Toxicity analysis of tranexamic acid

Toxicity predictions from the ADMETlab2.0, PROTOX3.0, toxCSM, and ADMET-AI databases indicate that the primary toxicities of TXA include liver damage, respiratory toxicity, brain injury, and neurotoxicity (Table 1). Additionally, the four databases predicted potential eye irritation, respiratory toxicity, and skin sensitization for TXA, although these toxicities were only identified in one of the databases. Notably, neither database predicted an association between TXA and EP. Based on the database analysis and literature review, the following sections will further explore the predictive analysis of the molecular mechanisms linking TXA and EP.

Table 1.

Tranexamic acid toxicity prediction database comparison tip: 0.3–0.5 (−), 0.5–0.7 (+), 0.7–0.9 (+ +), 0.9–1.0 (+ + +). Larger probability represents a greater risk of toxicity. N/A indicates that there is no information in the database

TXA toxicity prediction database comparison
Toxicity type	ProTox3.0	ADMETlab2.0	toxCSM	ADMET-AI
Hepatotoxicity	−	−	+ +	−
Neurotoxicity	+	+	−	N/A
Nephrotoxicity	+	−	+	N/A
Respiratory Toxicity	+ + +	+ +	+	N/A
Cardiotoxicity	+	N/A	+	−
Carcinogenicity	+	−	+ + +	−
Brain Injury	+ + +	+ + +	−	+ +
Skin Sensitization	N/A	−	+	−
Eye Irritation	N/A	−	−	N/A
GABA receptor	+	N/A	N/A	N/A

Potential targets of tranexamic acid inducing epilepsy

Initially, 344 TXA-related targets were obtained through the SwissTargetPrediction, STITCH, and ChEMBL databases, and their intersection Venn plots are shown in Fig. 1. Subsequently, a total of 2051 EP-related targets were collected using GeneCards, OMIM, and TTD databases, and their intersection Venn plots are shown in Fig. 2. Finally, 51 crossover targets were identified as potential targets for TXA-induced EP (Fig. 3).

Fig. 1.

Venn of Targets for tranexamic acid in the ChEMBL Database, STITCH Database, and SwissTargetPrediction Database

Fig. 2.

Venn of epilepsy targets in GENECARD, OMIM, and TTD databases

Fig. 3.

Venn of the intersection gene targets of tranexamic acid and epilepsy

Network analysis of tranexamic acid-induced epilepsy

Using Cytoscape software, a drug-target network was constructed to represent the relationships between TXA and EP, resulting in a comprehensive visualization. As shown in Fig. 4, this network diagram illustrates the 51 intersecting targets identified between TXA and EP, including key proteins such as GABRA1, GABRA2, GABRA5, GABBR2, and GAD1. The network connects these targets to TXA and EP, providing a structural overview of their potential interactions. However, specific pathway relationships are not explicitly labeled in this figure and are further elucidated through subsequent GO and KEGG analyses.

PPI network and hub target identification

Figure 5A depicts the protein–protein interaction (PPI) network of differentially expressed genes associated with TXA-induced EP, constructed using the STRING database and visualized in Cytoscape. In this network, nodes represent proteins and edges denote known or predicted interactions, including experimentally validated associations, curated database annotations, gene proximity, co-expression patterns, and protein homology. The network exhibits a high degree of interconnectivity, suggesting that neurotransmitter receptors and their regulatory factors may be critically involved in TXA-induced alterations of neural activity. Figure 5B shows a concentric circle network diagram derived from degree analysis performed with the NetworkAnalyzer plugin in Cytoscape. Briefly, the TSV file containing network relationships was imported to Cytoscape to automatically generate the PPI network. NetworkAnalyzer was then used to calculate topological parameters, with node size and color mapped to degree values (i.e., the number of connected nodes). A continuous color scale was applied, with node colors transitioning from light blue to red to indicate increasing degree values, and nodes with higher connectivity positioned toward the network center. This analysis revealed that GABAergic signaling pathway proteins, including GAD1, GABRA1, GABRA2, GABRA5, and GABBR2, occupy central positions and display the deepest red coloration, indicating their prominent roles in network regulation. These hub proteins are likely key mediators of the pathological link between TXA and EP, offering potential targets for further mechanistic investigation and therapeutic intervention.

Fig. 5.

A Protein–protein interaction (PPI) network of TXA-induced EP-related genes constructed using the STRING database. B Concentric circle layout highlighting hub proteins by degree value, with deeper red indicating higher connectivity

GO analysis

GO enrichment analysis identified 79 BP terms, 57 CC terms, and 33 MF terms (Additional file 6). The most significantly enriched BP terms included the γ-aminobutyric acid (GABA) signaling pathway (GO:0007214), regulation of membrane potential (GO:0042391), chloride transmembrane transport (GO:1,902,476), and monovalent inorganic anion transport (GO:0006820) (Fig. 6A, B). In the CC category, the GABA-A receptor complex (GO:0099010), postsynaptic membrane (GO:0099068), chloride channel complex (GO:0099012), and synaptic membrane (GO:0097060) were highly enriched. MF terms were dominated by GABA receptor activity (GO:0016917), GABA-A receptor activity (GO:0004890), neurotransmitter receptor activity (GO:0030594), and chloride channel activity (GO:0005254). Figure 6B shows that GABA receptor activity exhibited the highest statistical significance (-log10(P) > 20), followed by postsynaptic membrane and ion channel-related terms (-log10(P) between 10 and 20). Figure 6C and D further indicates that within BP, “regulation of membrane potential”; within CC, “synaptic membrane”; and within MF, “neurotransmitter receptor activity” ranked highest in both enrichment counts and statistical significance.

Fig. 6.

A Shows the name of GO on the abscissa and the number of GO-enriched genes on the ordinate axis, showing the functional enrichment. B Circle diagram structure: the outermost circle is the GO ID, the second circle is the total number of genes in each GO, the third circle is the number of GO enrichment genes, and the fourth circle is the gene ratio. C The ordinate is the name of GO, the abscissa is the proportion of genes, the size of the circle indicates the number of genes, and the color reflects the significance (the darker the red, the higher the significance). D The ordinate is the GO name, the abscissa is the number of enriched genes, and the color reflects the significance (the darker the red, the higher the significance)

KEGG pathway analysis

KEGG pathway enrichment analysis revealed significant overrepresentation of pathways related to neurotransmission, addiction, and metabolic regulation (Additional file 7; Fig. 7A–C). Neuroactive ligand–receptor interaction showed the highest gene ratio and statistical significance, underscoring the central role of neurotransmitter signaling. Synapse-related pathways, including GABAergic synapse, glutamatergic synapse, and serotonergic synapse, were also enriched, highlighting the involvement of excitatory–inhibitory balance in TXA-induced EP. Addiction-related pathways—such as morphine addiction, nicotine addiction, and retrograde endocannabinoid signaling—were significantly enriched. Although their direct association with EP remains uncertain, these pathways may share overlapping neurobiological mechanisms with TXA-induced neural alterations. Several metabolic pathways, including sphingolipid metabolism, phenylalanine metabolism, tyrosine metabolism, and fatty acid metabolism, were also significantly enriched despite involving fewer genes, suggesting that metabolic dysregulation may act as a concomitant or contributory factor in the TXA–EP relationship. Functional categorization demonstrated that most significant pathways clustered into neurotransmission-, addiction-, and metabolism-related categories, offering key mechanistic insights into the potential effects of TXA on EP.

Fig. 7.

KEGG pathway enrichment analysis of TXA-related genes associated with epilepsy. A Bubble plot of the top enriched KEGG pathways. X-axis: gene ratio; Y-axis: pathway name. Bubble size: gene count; color: adjusted p-value. B Bar plot of the top enriched pathways ranked by gene count. Bar length: gene count; color: adjusted p value. C Functional classification of enriched pathways grouped by KEGG categories. Bubble size: gene count; color: category. Most pathways were related to neurotransmission, addiction, and metabolism

Molecular docking

To elucidate the binding mechanisms of TXA with its core targets, the crystal structures of GABRA1, GABBR2, GABRA5, GABRA2, and GAD1 were retrieved from the Protein Data Bank and subjected to structure-based blind docking (Fig. 8). In the docking visualizations, TXA is depicted in ball-and-stick format, while the receptor proteins are rendered in cartoon representation. From left to right, the panels illustrate (1) the protein surface view, highlighting the overall molecular surface of the target and potential ligand-binding regions; (2) the ribbon diagram of secondary structures, indicating approximate TXA binding sites; (3) a magnified view of the binding pocket, showing TXA insertion; and (4) detailed docking interactions with key amino acid residues. The docking results demonstrated that TXA forms stable interactions with multiple residues across different chains of the five targets, with Vina scores ranging from − 5.2 to − 6.9 kcal/mol (Table 2). Notably, TXA exhibited favorable binding affinities toward all five proteins—GAD1, GABRA1, GABBR2, GABRA5, and GABRA2—with the strongest predicted interaction observed for GABRA2 (Vina score: − 6.9 kcal/mol). Across these targets, TXA primarily engages in hydrogen bonding and hydrophobic interactions with residues in key functional domains, suggesting potential modulation of GABAergic receptor activity and glutamate decarboxylase function. These findings offer structural insights into the multi-target neuropharmacological potential of TXA and imply mechanisms underlying TXA-induced EP, although experimental validation is essential to corroborate these computational predictions.

Fig. 8.

Molecular docking of TXA with core targets (GAD1, GABRA1, GABBR2, GABRA2, GABRA5). From left to right: I Protein surface representation showing the overall molecular surface morphology of the target protein and potential ligand-binding regions; II Protein secondary structure and binding site presented as a ribbon diagram indicating the approximate location of tranexamic acid binding; III Enlarged view of the binding site illustrating how TXA is embedded within the protein pocket; IV Docking details depicting the interactions between TXA and surrounding key amino acid residues

Table 2.

Molecular docking scores and key interacting residues between TXA and its core protein targets (GABRA1, GABBR2, GABRA5, GABRA2, and GAD1). Vina scores reflect binding affinity, with more negative values indicating stronger predicted interactions

TXA interaction table
Protein	Interacting residues	Vina score
GAD1	Chain A: ASN189 GLN190 LEU191 SER192 GLY251 GLY252 ALA253 ASN256 HIS291 SER293 THR344 GLY346 THR347 THR348 ASP373 ALA374 ALA375 TRP376 GLY377 ASN402 PRO403 HIS404 MET406 MET407 GLN412 ARG567 Chain B: ASN212 PHE214 TYR434 LEU435 CYS455 GLY456	− 5.7
GABRA1	Chain A: ASN21 LYS22 TYR24 LYS34 VAL35 ASP 36 ASP86 ARG105 LEU107 GLY108 SER109 ILE152 ASP153 GLU155 GLU159 TYR225 Chain B: VAL26 ASN27 THR28 LEU29 GLU30 ILE79 ASN80 VAL81 VAL82 GLY83 SER84 SER111 ASN112 ASP113 GLU129 GLU159 TYR175 ARG190 ARG199 ASN275 SER276 LEU277 PRO278 LYS279	− 5.7
GABBR2	Chain A: PRO161 VAL162 LEU163 ALA164 ASP165 LYS166 ARG174 PRO177 SER178 ASP179 ASN180 ALA181 VAL182 ASN183 PRO184 GLU210 VAL211 ASN213 ASP214 GLY217 VAL218 ARG418 ASN419 GLU421 ARG422 MET423 GLY424 ALA444	− 5.2
GABRA5	Chain C:TYR49 ASN51 SER52 PHE53 GLY54 PRO55 VAL56 SER57 ASP58 THR59 PHE68 ALA185 ASP187 GLY188 SER189 ARG190 LEU191 ASN192 GLN193 TYR194 TYR228 GLN232 ALA276 ARG277 ASN278 SER279 Chain D:ASP58 THR59 GLU60 MET61 GLU62 HIS105 LYS108 GLU141 CYS142 PRO143 MET144 LYS159 GLU204 ILE206 ILE215 THR217 ILE274 ASN278 LEU280 PRO281 LYS282	− 5.3
GABRA2	Chain B:GLU59 PHE100 HIS102 GLY104 LYS105 LYS106 ARG136 GLU138 SER159 TYR160 VAL203 SER205 SER206 THR207 TYR210 Chain C:ASP56 SER61 ILE62 GLY63 PRO64 THR73 ILE74 ASP75 PHE77 ALA79 THR81 LYS118 ASP120 HIS122 MET130 TRP134 ARG138 LEU140 THR142 ARG144 THR146 ASP148 LYS184 ARG197	− 6.9

To evaluate the biological relevance of the predicted TXA binding sites, we compared the interacting residues with known active and allosteric sites of GAD1, GABRA1, GABBR2, GABRA5, and GABRA2, as reported in structural and functional studies. For GAD1, TXA binds near the active site, interacting with His291, Phe214, and Tyr434, in proximity to the pyridoxal 5′-phosphate (PLP) pocket (Lys396, His295), suggesting potential inhibition of GABA synthesis [67]. However, residues such as Asn189–Ser192 and Asp373–Arg567 are peripheral, indicating possible nonspecific interactions. In GABRA1, TXA interacts with Tyr225 and Glu159, located near the orthosteric GABA binding site (Tyr159, Arg144) and the benzodiazepine pocket, implying disruption of inhibitory signaling [68, 69]. The GABBR2 docking site (Pro161–Val218, Arg418–Ala444) aligns with the extracellular Venus flytrap domain and TM6, in proximity to the GABA binding site (Ser130, Arg202) and the EP-associated region (Ser695) [70]. For GABRA5, residues Tyr194, Arg190, and His105 overlap with orthosteric and benzodiazepine sites, while Cys142–Met144 (M1–M2) suggests potential channel regulation [69, 71]. GABRA2 showed the strongest alignment, with Arg144, Tyr160, and His102 matching orthosteric and allosteric sites, supporting a competitive antagonistic effect [71]. Although these overlaps suggest that TXA may disrupt GABAergic signaling, certain peripheral residues (e.g., Asn21–Asp36 in GABRA1) may reflect limitations of blind docking. Further studies are needed to validate these interactions and confirm their role in TXA-induced seizures (Table 3).

Discussion

This study systematically investigates the toxicological mechanism of TXA-induced EP through toxicology database analysis and literature review. TXA, a hemostatic drug widely utilized in clinical settings, requires thorough examination of its toxicological properties to prevent and mitigate potential side effects. This research adopts a multi-disease perspective to comprehensively evaluate TXA’s potential adverse reactions. Using Venn diagram analysis, we identified 51 potential targets associated with TXA-induced EP. Subsequently, a protein–protein interaction (PPI) network analysis pinpointed five core targets critical to this mechanism. Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses of these core targets uncovered significant associations between TXA and EP. Furthermore, the study suggests that TXA may trigger multiple diseases by acting on shared target genes. Molecular docking analysis revealed that the binding affinity of TXA to these core targets could be a pivotal factor in its adverse reactions.

Traditional toxicology relies heavily on animal experiments and cell-based tests to assess substance toxicity. For example, pre-market drug evaluation typically involves acute and chronic toxicity testing in mice and rats, with human trials commencing only after confirming minimal toxicity. Cell experiments are also frequently employed for toxicity assessment [72, 73]. However, these approaches are expensive, raise ethical concerns, and struggle with determining appropriate sample types and quantities, complicating the acquisition of rare toxicological data. Consequently, the application of network toxicology becomes increasingly vital [74].

The advent of network toxicology offers a significant advancement in addressing the limitations of traditional methods. As an integrative approach, it leverages vast bioinformatics datasets and big data analytics, enabling researchers to systematically dissect the complex interaction networks between toxins and biological systems. Unlike traditional toxicology’s focus on single targets, network toxicology highlights multiple targets and their interconnections, markedly improving the depth of understanding of toxin-induced biological effects. Moreover, it facilitates the prediction of potential toxic effects, providing a robust foundation for comprehensive toxicity assessments and the formulation of precise safety measures and risk management strategies. Additionally, network toxicology accelerates the investigation of toxicological mechanisms, paving innovative pathways for drug development and environmental toxicity evaluation. Public databases, serving as repositories for adverse drug reactions and drug-target interactions, are indispensable in drug safety research, risk assessment, and development [28]. By employing network toxicology, this study predicts TXA’s role in inducing EP, offering valuable insights into its clinical application.

Since its introduction in Japan in the 1960 s, TXA has been widely recognized for its effectiveness in perioperative bleeding control, with expanded use in nebulized inhalation for hemoptysis and in managing abnormal bleeding in gynecology and pediatrics [75–77]. Although oral TXA has low toxicity—minimum adverse reaction doses of 500 mg usually requiring only supportive care—its epileptogenic potential has gained increasing attention. Intrathecal misadministration (200–250 mg) can rapidly induce persistent seizures via competitive GABAA receptor antagonism, as evidenced by 48 cases reported to the California Poison Control System between 1997 and 2021 [78, 79]. Clinical and experimental data further confirm this risk. A 44-year-old male with hemoptysis developed altered mental status, myoclonus, and hyperthermia within 1 h after combined nebulized (500 mg thrice daily) and bronchoscopic TXA instillation (1000 mg), indicating potential neurotoxicity of inhaled administration [80]. In animals, cortical application of high-concentration TXA (≥ 0.5 mg/mL) induced focal EP, whereas chronic high-dose intravenous administration (500–600 mg/kg) triggered seizures lasting up to 12 h. Current research on the interaction between TXA and EP remains limited. Although interspecies differences in receptor composition and pharmacodynamics may limit the direct applicability of these findings to humans, animal studies nonetheless offer valuable insights into the underlying mechanisms [81]. Mechanistic studies show TXA suppresses hippocampal GABAA receptor–mediated inhibition, disrupting excitatory–inhibitory balance and triggering seizures [82]. In summary, TXA’s neurotoxicity is dose-, route-, and patient-dependent, necessitating individualized dosing strategies to optimize hemostatic efficacy while minimizing neurological risks.

The results of this study indicate that TXA can form stable interactions with GABRA1, GABBR2, GABRA5, GABRA2, and GAD1,thereby potentially inducing EP.

GABRA1 encodes the α1 subunit of the γ-aminobutyric acid type A (GABA_A) receptor, a core component of the brain’s principal inhibitory neurotransmitter system [83]. The α1 subunit is widely expressed throughout the brain, where it contributes to receptor assembly and localization, modulates GABA binding affinity, regulates ion channel gating properties, and plays a pivotal role in determining the efficacy of inhibitory synaptic transmission. Through these functions, the α1 subunit is essential for maintaining the excitatory–inhibitory balance that underlies normal brain activity [84]. Pathogenic variants in GABRA1 are strongly associated with EP [85]. Loss-of-function (LoF) mutations, often occurring in the extracellular domain, typically reduce receptor sensitivity to GABA or decrease receptor surface expression, thereby weakening inhibitory signaling, relatively enhancing excitability, and predisposing to milder phenotypes such as genetic generalized EP. In contrast, gain-of-function (GoF) mutations, particularly those within the transmembrane domain, frequently lead to severe developmental and epileptic encephalopathies (DEE). These variants can cause excessive activation of GABA_A receptors or abnormally heightened GABA sensitivity, which disrupts neuronal development, alters network homeostasis, and may impair processes such as neuronal migration and synapse formation. The phenotypic severity is determined by both the variant location—extracellular domain variants being associated with milder presentations and transmembrane domain variants with more severe phenotypes—and the extent of functional alteration in receptor activity [86–88].

GABRA5 encodes the α5 subunit of the γ-aminobutyric acid type A (GABA_A) receptor, which is predominantly expressed in the hippocampus and plays a critical role in inhibitory neurotransmission. By mediating GABA-induced chloride influx, the α5 subunit induces neuronal hyperpolarization and inhibition. Anchored to the postsynaptic membrane by proteins such as gephyrin, it regulates inhibitory synaptic strength and maintains the excitatory–inhibitory balance in neural networks, thereby preventing pathological hyperexcitability. De novo mutations in GABRA5, including p.V294F and p.S413F, have been implicated in early-onset epileptic encephalopathy (EOEE) through distinct mechanisms. The p.V294F mutation causes receptor retention in the endoplasmic reticulum, reducing surface expression and synaptic clustering, which in turn diminishes GABA-induced currents and inhibitory function. By contrast, p.S413F leaves surface expression intact but alters channel properties, leading to reduced inhibitory efficacy. Both mutations disrupt gephyrin clustering at synapses, impairing synaptic structure and GABAergic inhibition, thereby promoting network hyperexcitability and seizures [89]. Clinically, a 2020 case report described a 4-year-old Thai girl with a de novo p.V294F mutation, confirming its pathogenic role in DEE. Similarly, another study reported a 2-year-old male with a p.V294L variant who presented with severe early-onset EP; electrophysiological recordings in HEK293T cells expressing the α5(V294L)β2γ2s receptor revealed a marked increase in apparent GABA affinity [90, 91]. Together, these findings suggest that GABRA5 mutations impair inhibitory neurotransmission, particularly in extrasynaptic regions, thereby facilitating epileptogenesis.

GABBR2 encodes the γ-aminobutyric acid type B receptor subunit 2, a critical component of the GABA_B receptor that mediates inhibitory neurotransmission in the central nervous system. Pathogenic variants or abnormal expression of GABBR2 can impair receptor function, diminish inhibitory regulation, and increase neuronal excitability, which are closely linked to epileptic encephalopathy (EE) and related neurodevelopmental disorders, such as Rett-like syndrome [92–94]. Variants located within the sixth transmembrane domain (TM6) are particularly enriched in EE patients. Notably, the de novo mutations S695I and I705N, along with G693W, all cluster in this domain [95]. These TM6 variants reduce GABA_B receptor surface expression and signaling efficiency, resulting in neuronal hyperexcitability and seizure susceptibility. Among them, S695I causes the most severe impairment in receptor expression and function, potentially explaining its association with more severe EP phenotypes [96]. Clinically, GABBR2-related EE may present with EP, generalized hypotonia, and paroxysmal dystonic limb postures, as exemplified by an 11-year-old female patient harboring a de novo pathogenic GABBR2 variant [94].

GAD1 encodes glutamate decarboxylase 1, the key enzyme catalyzing the conversion of glutamate, the principal excitatory neurotransmitter, into γ-aminobutyric acid (GABA), the primary inhibitory neurotransmitter in the central nervous system. GAD1 is predominantly expressed in GABAergic neurons and plays a pivotal role in maintaining the excitatory–inhibitory balance by ensuring adequate GABA synthesis [97, 98]. Deficient GAD1 expression, as observed in schizophrenia, leads to impaired inhibitory function and parallels the pathophysiological mechanisms of EP [99]. Experimental knockout of Gad1 in rats via CRISPR/Cas9 significantly reduces GABA levels, attenuates responses to sensory stimuli, and increases absence seizure frequency, underscoring its role in sustaining inhibitory tone [100]. Clinically, biallelic GAD1 mutations—including homozygous missense, frameshift, nonsense, or compound heterozygous variants—result in loss or reduction of GAD67 enzymatic activity, leading to insufficient GABA synthesis [101]. This disruption of inhibitory–excitatory balance not only precipitates early-onset EP in the neonatal period but also impairs neurodevelopment, often causing severe developmental delay and abnormal muscle tone [102].

GABRA2 encodes the α2 subunit of the γ-aminobutyric acid type A (GABAA) receptor, an integral receptor component predominantly expressed in the hippocampus and cortex, and is critical during early development for maintaining the excitatory–inhibitory balance of neural circuits. Structural or functional perturbations of the α2 subunit can impair channel gating or reduce GABA affinity, thereby inducing neuronal hyperexcitability and increasing the risk of EP. In a 2018 study employing whole-genome sequencing and targeted resequencing of 279 individuals with EP, a de novo missense variant in GABRA2 (c.875C > A, p.T292K) was identified within the M2 transmembrane domain. This variant markedly reduced receptor surface expression and produced constitutive channel activity in the absence of GABA, indicating dysfunctional ion-channel behavior [91]. A subsequent 2019 study reported four additional GABRA2 missense variants (p.Val284Ala, p.Leu291Val, p.Met263Thr, p.Phe325Leu), all of which caused loss of function with significantly diminished GABA-evoked currents, with p.Leu291Val exerting the most pronounced effect. Variants located in the transmembrane domains (M1–M3) were typically associated with severe phenotypes, whereas extracellular variants correlated with comparatively milder clinical presentations, supporting a clear genotype–phenotype relationship [103]. In a 2025 multicenter series analyzing six individuals harboring novel GABRA2 variants, extracellular-domain changes (e.g., p.Phe230Leufs*3) were primarily linked to mild developmental encephalopathy accompanied by late-onset, medication-responsive EP [104].

Our findings suggest that TXA may induce seizures through its interactions with GABAergic targets, underscoring the importance of clinical risk mitigation. Specifically, TXA was identified to interact with key proteins involved in GABAergic neurotransmission, including GAD1, GABRA1, GABBR2, GABRA2, and GABRA5. Docking analysis further confirmed its potential to bind both functional and allosteric sites within these proteins, indicating a direct disruption of GABA synthesis and receptor activity. Given that GABA is the principal inhibitory neurotransmitter in the central nervous system, such interference may reduce inhibitory tone, disturb the excitatory–inhibitory balance, and predispose neurons to hyperexcitability. Consequently, TXA exposure could promote neuronal overexcitation and contribute to seizure induction, providing a mechanistic basis for its pro-epileptogenic effects. Clinical evidence from meta-analyses supports these mechanistic insights, showing a dose-dependent increase in seizure incidence, particularly at doses exceeding 2 g/day [105]. Even moderate doses (~ 24 mg/kg, approximately 1.8 g in a 75-kg patient) have been shown to double the risk of postoperative seizures compared to no TXA use [106]. Therefore, clinicians should employ the lowest effective dose, avoid regimens above 2 g/day, and preferentially consider continuous infusion or low-dose bolus strategies. Most critically, inadvertent intrathecal or spinal administration must be strictly avoided, as even trace amounts can precipitate catastrophic neurotoxicity, including refractory seizures and high mortality [107, 108]. By integrating low-dose regimens, careful route monitoring, and patient-specific considerations (e.g., renal function), clinicians can maximize TXA’s hemostatic efficacy while minimizing seizure risk.

This study, based on public data, systematically explored the potential mechanisms of TXA-induced EP using network toxicology and molecular docking methods. The results were derived from computational models and predictive analysis, thus requiring further experimental validation to confirm their reliability and applicability. Additionally, due to the complexity of biological systems, network toxicology and molecular docking methods have certain limitations in fully simulating complex biological interactions, which may affect the accuracy of the results. The databases employed in this study, such as ChEMBL, GeneCards, OMIM, TTD, STITCH, and SwissTargetPrediction, provide valuable resources for target prediction and disease association analysis; however, they also carry inherent limitations. For instance, these databases are subject to incomplete or uneven annotations, potential biases in literature-based relevance scoring, and species-specific differences that may reduce target identification accuracy. Consequently, some identified targets may not fully capture the true molecular mechanisms underlying TXA-induced EP. Nevertheless, long-term drug toxicity studies face significant challenges in the laboratory due to the involvement of multiple factors. By applying network toxicology and molecular docking models to predict the pathogenic mechanisms of compounds, this study provides important support for developing targeted prevention and treatment strategies. This research not only offers key therapeutic insights for TXA-induced clinical EP but also contributes valuable toxicological insights for its use in surgical procedures.

Conclusions

Following the “Expert Consensus on the Application of TXA and Anticoagulants in the Perioperative Period of Accelerated Rehabilitation in Orthopedic Surgery in China,” TXA’s safety in spinal surgery has gained significant attention. This study investigates TXA’s neurotoxic mechanisms, specifically EP induction, using network toxicology and molecular docking to optimize its clinical use. While TXA effectively reduces intraoperative bleeding by inhibiting fibrinolysis, its potential neurotoxicity raises safety concerns. The analysis identified 51 targets linked to TXA and EP, with GABRA1, GABBR2, GABRA5, GABRA2, and GAD1 as core targets critical for neuronal excitatory-inhibitory balance. GO and KEGG analyses suggest TXA disrupts GABA signaling and synaptic function, potentially triggering EP, with molecular docking confirming stable target binding. Limitations include the lack of in vivo/in vitro validation and unexplored effects of dose or administration variations. By innovatively applying network toxicology for multi-target analysis, this study transcends single-target limitations, indicating TXA may induce multiple diseases via shared targets. Though further validation is needed, it offers vital theoretical support for safe TXA use and neurotoxicity prevention, providing significant clinical and scientific value.

Abbreviations

TXA

Tranexamic acid

EP

Epilepsy

BP

Biological process

CC

Cellular component

DEE

Developmental and epileptic encephalopathy

EE

Epileptic encephalopathy

EOEE

Early-onset epileptic encephalopathy

GABA

Gamma-aminobutyric acid

GABRA1

Gamma-aminobutyric acid type a receptor subunit alpha 1

GABRA5

Gamma-aminobutyric acid type a receptor subunit alpha 5

GABBR2

Gamma-aminobutyric acid type b receptor subunit 2

GAD1

Glutamate decarboxylase 1

GO

Gene ontology

GoF

Gain-of-function

KEGG

Kyoto Encyclopedia of Genes and Genomes

LoF

Loss-of-function

MF

Molecular function

Authors’ contributions

TC: Data curation, Investigation, Software, Writing – original draft, Writing – review & editing. DL: Data curation, Investigation, Software, Writing – original draft, Writing – review & editing. YG: Visualization, Writing – original draft. TG: Visualization, Writing – original draft. XL1: Conceptualization, Writing – review & editing. XL2: Conceptualization, Methodology, Writing – review & editing. All authors read and approved the final manuscript.

Funding

This work was supported by the Major Science and Technology Project of Jilin Province (Grant No. 20210304001YY), the Regional Innovation and Development Joint Fund of the National Natural Science Foundation of China (Grant No. U22A20367), the Jilin Province Youth Science and Technology Talent Support Program (Grant No. QT202227), the Jilin Province Traditional Chinese Medicine Science and Technology Project (Grant No. 2023027), the Young Academic Backbone Talent Training Program (Grant No. 202317), the Jilin Province Clinical Research Center for Orthopedics of Traditional Chinese Medicine (Grant No. 20180623048TC), the Science and Technology Project of the Education Department of Jilin Province (Grant No. JJKH20230987KJ), and the “Qi-Huang Scholar” Project of the National Administration of Traditional Chinese Medicine.

Data availability

The datasets analyzed in this study were obtained from multiple publicly available databases and platforms, including ADMETlab 2.0, ProTox 3.0, toxCSM, ADMET-AI, ChEMBL, STITCH, SwissTargetPrediction, PubChem, GeneCards, OMIM, TTD, STRING, PDB, and the CB-Dock2 molecular docking platform. Detailed information and validated data can be accessed through their respective official websites. All data generated or analyzed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable. This study did not involve human participants, human data, human tissue, or animals. All analyses used publicly available databases and computational tools.

Consent for publication

Not applicable. This study did not involve human participants, personal data, or clinical images that require consent for publication. All authors consent to the publication of this article.

Competing interests

The authors declare no competing interests.