Cerebral stroke-related genes were retrieved from the GeneCards (https://www.genecards.org), DisGeNET (https://www.disgenet.org/), and DigSee (http://210.107.182.61/geneSearch/) databases. Duplicate entries were removed. Active compounds of LCH were obtained from the HIT (http://hit2.badd-cao.net), TCMSP (http://tcmspw.com/tcmsp.php), and TCMID (https://bidd.group/TCMID/) databases. Repetitive compounds and those lacking Pubchem IDs were excluded. According to Pubchem ID, targets of the active compounds in LCH were searched in the HIT database.

Cerebral stroke blood transcriptome datasets GSE16561 and GSE22255 for cerebral stroke were retrieved from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo). Batch effects between datasets obtained from different platforms were removed using the surrogate variable analysis (SVA) algorithm. Screen for Differentially Expressed Genes (DEGs) between the disease and control groups were identified using the limma program package (https://bioconductor.org/packages/release/bioc/html/limma.html) [13] with p 0.05.

The cerebral stroke-related genes obtained, the active compound-targeted genes of LCH, and the DEGs in cerebral stroke were intersected using a Venn diagram (version 1.7.3) in R 4.0 software (https://www.r-project.org/).

Interactions between the 55 intersecting genes and their reciprocal proteins were predicted using the STRING database (https://string-db.org/, version 11.0). The screening conditions included a score cutoff $>$0.4 and a size cutoff 10. The construction of Protein-Protein Interaction (PPI) networks were visualized using Cytoscape (version 3.8.0; http://www.cytoscape.org/). The MCC algorithm in CytoHubba was used to calculate the hub genes in the PPI network.

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses of genes in the PPI network were performed using clusterProfiler (http://www.bioconductor.org/packages/release/bioc/html/clusterProfiler.html, version 4.4.2), where p 0.05 was considered a significant enrichment result.

A diagnostic marker model for cerebral stroke was constructed using the machine learning algorithm lasso regression applied to the screened hub genes. To avoid overfitting the model and evaluate the stability of the model prediction, 60% of the samples were randomly assigned to the training set and the remaining 40% constituted the test set. Considering sex and age as relevant risk factors for cerebral stroke, the predictive performance of the model was further assessed in the training and test sets for sex and age groups. The difference in the mean values of the basic blood test items were compared and items exhibiting statistically significant differences were included in the lasso regression analysis. The regularization parameter $\lambda$, corresponding to one standard error from the minimum Mean-Squared-Error, was selected as the most appropriate $\lambda$ by cross-validation. Indicators with non-zero coefficients were selected for the receiver operating characteristic curve analysis.

Molecular docking was conducted to examine the interactions between marker genes in the model and the active ingredients of LCH. Proteins structures were downloaded from the RCSB protein database (http://www.rcsb.org), while structure files of the active ingredients in sdf format were obtained from PubChem. They were then converted to PDBQT format using Open Babel GUI (http://openbabel.org/docs/2.3.1/GUI/GUI.html). The protein was optimized by removing original ligands, dehydration, adding hydrogen bond and amino acids, and calculating charges using Pymol (https://pymol.org/2/) and Autodock 1.5.6 (http://mgltools.scripps.edu/). The molecular docking results were visualized using Autodock 1.5.6 and Pymol.

All data are presented as mean $\pm$ standard deviation. Group comparisons were analyzed using a one-way analysis of variance (ANOVA) followed by Turkey’s tests. All statistical analyses were conducted using GraphPad 8.0 software (GraphPad Software, Inc., San Diego, CA, USA). A p 0.05 was considered statistically significant.

A total of 2603 DEGs were identified. After removing batch effects in the combined GSE16561 and GSE22255 datasets, the distribution trend of gene expression levels was more consistent (Fig. 1A,B). The DEGs were visualized using a heatmap and a volcano plot (Fig. 1C,D). Red represents upregulated genes, while blue represents downregulated genes on the heatmap (Fig. 1C). The significance cut-off criterion for differential expression was set at p 0.05 on the volcano plot (Fig. 1D).

Fig. 1.

Differentially expressed genes (DEGs) analysis in cerebral stroke. (A) Expression level of sample screened in GEO database before removing batch. (B) Expression level of sample screened in GEO database after removing batch. (C) Heatmap of DEGs. (D) Volcano plot of DEGs. GEO, Gene Expression Omnibus.

We obtained a total of 9217 cerebral stroke-related genes and 191 active compounds of LCH. Among these, 43 active compounds were found to target 195 specific genes. By intersecting the cerebral stroke-related genes, LCH target genes, and DEGs, we identified a set of 55 genes that included 52 upregulated genes and 3 downregulated genes (Fig. 2A and Table 1). The 55 genes and active compounds of LCH were intersected using Cytoscape to construct a compound-target network (Fig. 2B). This network consisted of 85 nodes and 78 relationships. Oleic acid (CID: 445639) exhibited the highest degree of connectivity, interacting with 11 targets. Caffeic acid (CID: 689043) was the second most connected compound, interacting with 10 targets.

Fig. 2.

Targets of Ligusticum Chuanxiong Hort. (LCH) against cerebral stroke. (A) Venn diagram of cerebral stroke-related genes, DEGs, and LCH-target genes. (B) Compound-target network.

The reciprocal proteins of the 55 intersecting genes were predicted by the STRING database and a PPI network was constructed in Cytoscape. The PPI network consisted of 62 nodes and 393 edges in the network (Fig. 3A). To identify the core genes within the PPI network, the MCC algorithm in the CytoHubba plugin was applied. The top 15 hub genes, ranked by node degree, were IL1B, CCL2, MAPK3, PTGS2, JUN, MMP9, TLR4, HIF1A, PPARA, FOS, PTEN, NFE2L2, TLR2, IMP1, and SOD2 (Fig. 3B). These genes showed a high level of connectivity within the network, indicating their potential importance in the context of cerebral stroke.

Fig. 3.

Protein-protein interaction (PPI) network. (A) PPI network of targets in Ligusticum Chuanxiong Hort. (LCH) against cerebral stroke. (B) PPI network of top 15 hub genes.

A total of 743 results were enriched in the GO enrichment analysis. The top 15 results were filtered based on the p-value. In the KEGG pathway enrichment analysis, 93 signaling pathways were enriched and the top 15 results were displayed based on the p-value. The GO enrichment analysis results were primarily related to cellular response to chemical stress, response to oxidative stress, and response to extracellular stimulus, etc. (Fig. 4A). The KEGG pathways include the TNF signaling pathway, Lipid and atherosclerosis, and IL-17 signaling pathway, etc. (Fig. 4B).

Fig. 4.

Gene Ontology (GO) biological functional analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. (A) Bubble of GO enrichment analysis. (B) Bubble of KEGG enrichment analysis. The size of the dot indicates the number of genes annotated in the entry and the color indicates the range of the p-value, with lower values represented by darken shades of yellow.

To assess the diagnostic capability of the identified hub genes in cerebral stroke, the lasso logistic regression algorithm was used. The penalty parameter tuning process was performed using 10-fold-cross-validation (Fig. 5A). Based on the analysis, six genes (FOS, MMP9, PPARA, PTEN, TIMP1, and TLR2) were identified as diagnostic markers for cerebral stroke. The model parameters are shown in Table 2. A total of 60% of samples screened from GEO were randomly selected as the training set, and the rest of the samples were set as the test set. The area under curve (AUC) value of the model in the training set was 0.927 and 0.855 in the test set (Fig. 5B). Considering that age and sex are also important risk factors for cerebral stroke, the predictive performance of the model was further examined by sex and age within the training and test sets. The model had AUC values greater than 0.915 in the training set for sex and were greater than 0.86 for age (greater than 60 for the high age group and less than 60 for the low age group) (Fig. 5C,D). The values in the test set were more than 0.8 for sex and more than 0.78 for age (Fig. 5E,F).

Fig. 5.

Diagnostic model of cerebral stroke. (A) Screening processes for diagnostic biomarkers for cerebral stroke. (B) Receiver operating characteristic curve (ROC) of the training set and test set. (C) Sex-related ROC in the training set. (D) Age-related ROC in the training set. (E) Sex-related ROC in the test set. (F) Age-related ROC in the test set. CI, confidence interval; AUC, area under curve.

The target genes corresponding to each active compound in LCH were individually intersected with the 55 genes and the results are shown in Table 3. Among them, oleic acid (CID: 445639) and caffeic acid (CID: 689043) emerged as the core active compounds of LCH. The model marker genes FOS, MMP9, PPARA, PTEN, TIMP1, and TLR2 were selected for molecular docking with oleic acid and caffeic acid. The criteria were that the binding energy be less than –5.0 kcal/mol and that hydrogen bonds could be formed between the receptor ligands [14]. The docking results are summarized in Table 4. Caffeic acid was combined with MMP9, PPARP, PTEN, and TIMP1, with binding energies of –7.9, –6.4, –5.9, and –6.9 kcal/mol, respectively (Fig. 6A). Oleic acid bound well with MMP9 (–6.6 kcal/mol), PPARA (–6 kcal/mol), and TIMP1 (–5.5 kcal/mol) (Fig. 6B). Molecular docking simulations indicated a stable point docking structure for the binding of small-molecule ligands and protein receptors (Fig. 6A,B).

Fig. 6.

Docking model. (A) Caffeic acid with MMP9, PPARA, PTEN, and TIMP1. (B) Oleic acid with MMP9, PPARA, and TIMP1.

Cell validation experiments were performed to validate the results obtained from molecular docking. An OGD model was induced in SH-SY5Y cells, which were then treated with LCH extract. After treatment with 0, 12.5, 25, 50, and 100 µg/mL LCH, cell viability showed no significant difference (Fig. 7A), indicating the safety of LCH in SH-SY5Y cells. Subsequently, 25, 50, and 100 µg/mL LCH were chosen for further experiments. In comparison with control cells, the OGD-induced cells showed lower cell viability and higher LDH releasing (p 0.01, Fig. 7B,C). However, treatment with 50 and 100 µg/mL LCH dose-dependently elevated cell viability and reduced LDH level (p 0.01, Fig. 7B,C). These findings revealed the therapeutic potential of LCH for treating stroke.

Fig. 7.

LCH regulates the protein expression of identified targets. (A) Cell viability was determined by a Cell Counting Kit-8 (CCK-8) assay. SH-SY5Y cells were treated with different concentrations of LCH extract (0, 12.5, 25, 50, and 100 µg/mL). (B) Cell viability was determined using a CCK-8 assay. (C) LDH analysis. (D) Western blotting detected expression of MMP9, PPARA, PTEN, and TIMP1. Oxygen and glucose deprivation (OGD)-induced SH-SY5Y cells were treated with different concentrations of LCH extract (25, 50, and 100 µg/mL). ${}^{**}$p 0.01 compared to control group; ${}^{\mathrm{\#}\mathrm{\#}}$p 0.01 compared to OGD group; ${}^{\$\$}$p 0.01 compared to 25 µg/mL LCH group; ${}^{\mathrm{@}\mathrm{@}}$p 0.01 compared to 50 µg/mL LCH group.

Regarding the identified targets, the protein expression levels of MMP9 and PTEN were found to be increased, while those of PPARA and TIMP1 were decreased in the OGD-induced cells compared to control cells (p 0.01, Fig. 7D). However, treatment with 50 and 100 µg/mL LCH reversed the expression of these proteins in a dose-dependent manner (p 0.01, Fig. 7D). These results suggest that LCH treatment can modulate the expression levels of these target proteins associated with cerebral stroke.