Identification of blood-brain barrier injury-related biomarkers in cerebral infarction using transcriptomic analysis

Abstract

Background

Cerebral infarction leads to blood-brain barrier (BBB) disruption, exacerbating brain injury through edema, inflammation, and neuronal death. Although BBB damage is a critical event in stroke pathology, the underlying molecular mechanisms and reliable biomarkers remain poorly understood. This study aimed to identify key biomarkers associated with BBB injury following cerebral infarction using an in vitro model and transcriptomic approaches. Human cerebral microvascular endothelial cells (hCMEC/D3) were subjected to oxygen-glucose deprivation (OGD) and OGD/reoxygenation (OGD/R) to simulate ischemic and reperfusion injury. Cell viability, inflammatory cytokines, LDH release, and angiogenesis were assessed. Transcriptomic sequencing, weighted gene co-expression network analysis (WGCNA), and random forest algorithms were employed to identify differentially expressed genes and key biomarkers. OGD treatment significantly increased IL-1β, IL-6, TNF-α, and LDH levels, which were partially reversed by OGD/R. Transcriptomic analysis identified 1229 and 800 differentially expressed genes respectively in OGD and OGD/R comparisons. Enrichment analysis highlighted ribosome, endoplasmic reticulum, and mitochondrial pathways. Six core genes were screened, including RPS7, RPL36A, RPS9, RSL24D1, RPL41, and OSTC, all of which were upregulated under OGD and normalized after reoxygenation. We identified ribosome-related genes as potential biomarkers of BBB injury in cerebral infarction. These findings contribute to our understanding of BBB pathophysiology and suggest possible targets for future diagnostic and therapeutic development in ischemic stroke.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-39763-2.

Introduction

Cerebral infarction is an important form of ischemic cerebrovascular disease, often resulting from the interruption of cerebral blood supply, leading to localized hypoxia and insufficient nutrient supply to the brain tissue¹. After ischemia, the function of the BBB is compromised, which subsequently causes further damage to the brain tissue and loss of neurological function^2,3. The blood-brain barrier is a selective barrier composed of endothelial cells, the basement membrane, and tight junction proteins. Under normal conditions, it protects the brain tissue from harmful substances and maintains the homeostasis of the central nervous system⁴. However, after cerebral infarction, the integrity of the blood-brain barrier is affected, and its permeability significantly increases, resulting in exacerbated brain tissue damage².

The blood-brain barrier is a highly selective barrier formed by the endothelial cells of cerebral blood vessels, the basement membrane, astrocytes, and perivascular cells. The primary function of the blood-brain barrier is to prevent harmful substances in the blood from entering brain tissue, while allowing essential substances such as oxygen, glucose, and amino acids to pass through the endothelial cells into the brain tissue^5,6. Endothelial cells maintain the integrity of the blood-brain barrier through tight junctions, preventing large molecules and cells in the blood from entering the brain parenchyma.

Studies have found that the core mechanism of cerebral infarction is the interruption of cerebral blood flow, which leads to hypoxia and insufficient energy supply to brain tissue⁷. Under ischemic conditions, endothelial cells are stimulated by hypoxia, activating a series of signaling pathways that result in the degradation of tight junction proteins and disruption of endothelial cell function. As ischemia persists, the permeability of the blood-brain barrier gradually increases, allowing water, proteins, and other large molecules from the blood to infiltrate brain tissue, leading to cerebral edema and cellular damage. Moreover, studies have also shown that after reperfusion, the rapid restoration of oxygen results in the generation of large amounts of reactive oxygen species (ROS) and nitric oxide (NO), which further exacerbate the damage to the blood-brain barrier⁸. Oxygen free radicals activate intracellular signaling pathways, damaging the structure of endothelial cells and promoting the degradation of the basement membrane and tight junction proteins⁹. Additionally, it has been found that after cerebral infarction, local hypoxia and cell death lead to the release of inflammatory factors. Cytokines (such as IL-1β, TNF-α) and chemokines (such as MCP-1) are secreted in large quantities, recruiting neutrophils, monocytes, and other immune cells into the infarct area¹⁰. The ROS, enzymes (such as matrix metalloproteinases, MMPs), and inflammatory factors released by immune cells further damage the integrity of the blood-brain barrier, significantly increasing its permeability.

The damage to the blood-brain barrier following cerebral infarction is a major cause of severe complications such as cerebral edema, neurological dysfunction, and hemorrhagic transformation. To better understand the mechanisms of cerebral infarction-induced blood-brain barrier damage and its clinical implications, it is crucial to identify biomarkers related to blood-brain barrier damage. Biomarkers are molecules detectable in body fluids (such as blood or cerebrospinal fluid) that can indicate disease status or physiological changes. By detecting these biomarkers, it is possible to early assess the pathological state of cerebral infarction patients, predict prognosis, and evaluate treatment efficacy. Some enzymes or protein activations have been found to be closely related to blood-brain barrier damage, such as MMPs, which play a significant role in blood-brain barrier damage¹¹. After cerebral infarction, MMPs are activated and begin degrading collagen and other extracellular matrix components in the basement membrane, leading to structural damage to the blood-brain barrier¹². MMPs can also break down tight junction proteins, further increasing the permeability of the blood-brain barrier. Additionally, studies have shown that aquaporin-4 (AQP4) plays an important role in the formation of cerebral edema¹³. After cerebral infarction, AQP4 expression is upregulated, causing a large accumulation of water in the brain tissue, which induces cytotoxic cerebral edema. The activation of AQP4 not only worsens the edema but may also further disrupt the blood-brain barrier by regulating the flow of water¹⁴. However, research on biomarkers related to blood-brain barrier damage in human cerebral infarction remains limited, and many potential biomarkers are still unclear. This study aims to identify blood-brain barrier damage-related biomarkers by combining in vitro cell modeling with transcriptomics.

Materials and methods

Cell culture and treatment

The hCMEC/D3 cells (human Cortical Microvascular Endothelial Cell Line/D3 which was obtained from Fuheng Biology, Shanghai, China) were cultured in DMEM medium (containing 10% FBS and 2% antibiotics) and incubated at 37 °C with 5% CO₂¹⁵. The medium was changed every day, and when the cell density reached 80%, the cells were passaged at a 1:3 ratio. Cells in the logarithmic growth phase with good growth status were used for subsequent experiments. The cells were then divided into three groups: one group served as the control group, and the other two groups underwent OGD treatment and OGD/R treatment, respectively. In the OGD treatment, the hCMEC/D3 cells were cultured in normal DMEM medium to 80% confluence, then switched to glucose-free medium and incubated in an anaerobic chamber (95% N₂ and 5% CO₂) for 4 h. For the OGD/R treatment, the cells from the OGD condition were transferred to a normal incubator (21% O₂ and 5% CO₂) and treated with normal medium for 16 h to induce reoxygenation. Following OGD and OGD/R treatments, the culture supernatant containing detached cells was carefully removed. The adherent cells were then gently washed twice with phosphate-buffered saline (PBS) to remove any non-viable or loosely attached cells. Subsequently, adherent, viable cells from all groups (Control, OGD, and OGD/R) were collected by standard trypsinization, centrifuged, and resuspended in fresh complete medium. For downstream assays requiring reseeding (e.g., CCK-8 assay), the cell concentration was determined using a trypan blue exclusion method with an automated cell counter (or hemocytometer), and cells were seeded at equal densities across all groups to ensure standardized input. This protocol ensured that subsequent analyses were performed on the population of cells that survived the treatments and remained adherent.

CCK-8 cell viability assay

Following treatment, cells in each group were trypsinized, resuspended in complete medium supplemented with 10% fetal bovine serum (FBS), and seeded into 96-well plates at a density of 1 × 10⁴ cells per well¹⁶. Throughout the entire incubation period (24, 48, and 72 h), both the control and treatment groups were maintained in complete medium supplemented with 10% FBS. The plates were then incubated in a humidified atmosphere of 37 °C and 5% CO₂, 10 µL of CCK-8 reagent (Shanghai Maokang Biotechnology Co., Ltd. China) was added to each well. Prior to the initiation of Oxygen-Glucose Deprivation (OGD), microscopic examination confirmed that both the control and the groups designated for OGD treatment had reached approximately 80% confluence. The plates were further incubated for 2 h, and the absorbance was measured at 450 nm with a reference wavelength of 600–650 nm using a microplate reader¹⁷. Each experimental group was assayed in triplicate wells, and three independent experiments were performed. Cell viability was expressed as a percentage relative to the control group.

Cytokine measurement (IL-1β, IL-6, and TNF-α)

The levels of interleukin-1β (IL-1β), interleukin-6 (IL-6), and tumor necrosis factor-α (TNF-α) in the cell culture supernatant were quantified using commercial enzyme-linked immunosorbent assay kits, according to the manufacturer’s instructions. Briefly, the supernatant was collected and centrifuged at 1000 × g for 10 min at 4 °C to remove debris. The standards were reconstituted and serially diluted to generate the calibration curves with the concentrations specified in Table 1. For each assay, 100 µL of sample or standard was added to the pre-coated wells and incubated (Table 1). After incubation, the plates were washed five times with the provided wash buffer. Subsequently, 100 µL of the respective biotinylated detection antibody was added to each well and incubated for 60 min at room temperature. Following another wash cycle, 100 µL of horseradish peroxidase (HRP)-conjugated streptavidin was added and incubated for 20 min at room temperature in the dark. After a final wash, 100 µL of tetramethylbenzidine (TMB) substrate solution was added and incubated in the dark for 20 min for color development. The reaction was stopped by adding 50 µL of stop solution, and the absorbance was immediately measured at 450 nm using a microplate reader. All samples and standards were assayed in duplicate. The cytokine concentration in each sample was determined by interpolating from the respective standard curve.

Table 1.

Key parameters for the cytokine measurements.

Cytokine	Standard curve concentrations (pg/mL)	Sample/Standard incubation
IL-1β	15.62, 31.25, 62.5, 125, 250, 500, 100018,19	120 min, Room Temperature
IL-6	3.9, 7.81, 15.62, 31.25, 62.5, 125, 25020	120 min, Room Temperature
TNF-α	15.62, 31.25, 62.5, 125, 250, 500, 100021	90 min, 37 °C

LDH assay

The cells were collected into a centrifuge tube and centrifuged, discarding the supernatant. For cell extraction, 1 mL of extraction buffer was added per 5 million cells, and the mixture was sonicated (ice bath, 200 W, 3 s on, 10 s off, repeated for 30 cycles). The solution was centrifuged at 8000 × g for 10 min at 4 °C, and the supernatant was collected and kept on ice for further testing. The spectrophotometer was preheated for over 30 min, and the wavelength was adjusted to 450 nm, with distilled water used to zero the instrument. Standard solutions of 20 µmol/mL were diluted to concentrations of 2, 1, 0.5, 0.25, 0.125, and 0 µmol/mL to create a standard curve²². The samples were mixed thoroughly and incubated at room temperature for 3 min, and absorbance was measured at 450 nm.

Angiogenesis assay

Matrigel was thawed in a 4 °C refrigerator, and 250 µL per well was added to a 24-well plate, avoiding air bubbles²³. The plate was incubated in the cell culture incubator for 60 min to allow the gel to solidify. After starving the cells for 24 h, they were digested and counted, then resuspended in medium to adjust the cell density to 2–3 × 10⁵ cells/mL. Once the Matrigel had solidified, 500 µL of cell suspension was added to each well, and the plate was placed in the incubator for routine culture for 3–6 h, with images taken at the end of the incubation period.

Angiogenic activity was quantified based on the capillary-like structures formed on Matrigel. Image analysis was performed using the “Angiogenesis Analyzer” plugin for ImageJ (Fiji, National Institutes of Health, USA). For each well, three randomly selected, non-overlapping fields of view were captured using an inverted phase-contrast microscope. The captured images were thresholded and skeletonized using the plugin to extract the capillary network. The following parameters were automatically calculated. First, the combined length (in pixels, converted to micrometers using a scale bar) of all connected capillary-like structures. A “tube” was defined as a continuous structure formed by connected cells. Then, number of branch points: The count of points where three or more capillary segments intersect. The mean value from the three fields per well was calculated to represent a single biological replicate (n = 1). Experiments were performed with six independent biological replicates (n = 6) per experimental group. Data from all replicates were pooled for statistical analysis.

RNA sequencing and analysis

After total RNA was extracted from the cell sample (Health normal group: H, OGD group: OGD; OGD/R group: OGD/R), mRNA was enriched with magnetic beads with Oligo (dT). The mRNA fragments were divided into fragments by fragmentation buffer. The first cDNA strand was synthesized by six-base random primers using the fragment mRNA as template. The second strand of cDNA was synthesized by adding buffer solution, dNTPs, RNase H and DNA polymerase I, purified by QiaQuick PCR kit and eluted by EB buffer solution. After terminal repair, base A and sequencing joint were added, the target fragment was recovered by agarose gel electrophoresis, and then PCR amplification was performed to complete the preparation of the whole library. The constructed library was sequenced by Illumina HiSeqTM.

The reads comparison tool bowtie2(2.2.8)²⁴ was used to compare High Quality Clean Reads to the ribosomes of Homo sapiens, and the reads of ribosomal RNA were removed. The retained data were compared to the reference genome (https://ftp.ncbi.nlm.nih.gov/genomes/all/GCF/000/001/405/GCF_000001405.40_GRCh38.p14/) of Homo sapiens using comparison software Tophat2(2.1.1)²⁵. Then, we applied Expectation-Maximization algorithm for Allele Specific Expression (EMASE)²⁶ to quantify gene expression from the individual aligned RNA-seq data. Count data were normalized using DESeq2²⁷ variance stabilizing transformation. Tissue levels of mercury and physiological parameters were statistically evaluated using GraphPad Prism^® 7.05 using one-way analysis of variance (ANOVA) followed by Tukey’s multiple comparison test. Qlucore Omics Explorer 3.5 was used for statistical analysis of transcriptomic data. Prior to statistical analysis, pre-processed RNA seq data were log2 transformed. Data were analyzed using ANOVA comparing the 3 groups followed by planned contrasts, to investigate the specific comparisons between groups. In the omics analyses, P-values were used at a threshold of P < 0.05 for statistical significance. The data were further examined by principal component analysis (PCA).

The differentially expressed genes were analyzed for enrichment using the Gene Ontology (GO) database and the KEGG pathway which were conducted using the DAVID database²⁸. The known and predicted protein-protein interactions (PPIs) from STRING database version 11.0, http://www.string-db.org/) were downloaded. The interactions included direct (physical) and indirect (functional), derived from experimental repositories and computational prediction methods. The subnetwork was identified with highest degree by MCODE in Cytoscape v3.9.1²⁹ . In the PPI network, each edge denotes a score to quantify the interaction confidence, that is, the likelihood of an interaction.

WGCNA analysis

As an efficient and accurate bioinformatics data mining method, weighted gene co-expression network algorithm aims to find synergistically expressed gene modules³⁰. And explore the relationship between gene networks and the phenotypes that researchers are interested in. Microarray data differential expression analysis and Weighted Gene Co-Expression Network Analysis (WGCNA) analysis based on the clusterProfiler package for Rtudio (version 4.2.1) were performed with multiple comparison corrections (adjP < 0.05 as cut-off value)³⁰.

Candidate key gene screening

The randomForest R software package was used to further screen the intersection of WGCNA and significantly different genes by random forest algorithm. The randomForest algorithm determines the variables by MeanDecreaseGini value > 0, and selects the TOP30 variables based on Gini > 0.

Results

Cell culture and detection

In this study, OGD and OGD/R treatments were applied to the hCMEC/D3 cell line. The cell activity of health normal control group (H) and OGD group were higher than that of OGD group (Fig. 1A). The results also showed that, compared to the health normal control group, the OGD group exhibited significantly elevated levels of IL-1β, IL-6, TNF-α inflammatory cytokines, and LDH. However, after reoxygenation (OGD/R treatment), the levels of IL-1β, IL-6, TNF-α, and LDH were significantly reduced compared to the OGD-alone group (Fig. 1B–E). The angiogenesis assay results indicated that the number of vascular tubes formed by cells in the OGD/R group was significantly higher compared to the OGD group (Fig. 1F–H).

Fig. 1.

Construction of cell model of blood-brain barrier injury caused by cerebral infarction. (A) cell activity detected by CCK-8; (B) IL-1β detected; (C) IL-6 detected; (D) NF-α detected; (E) LDH detected; (F) Blood vessel formation test results; (G) Branch points statistics of blood vessel formation test; (H) Capillary length of blood vessel formation test.

Transcriptome data analysis and identification of differentially expressed genes

PCA analysis of quantitative data showed that the sample distribution was consistent with the cell modeling (Fig. 2A), and the sample repeatability in the group was also well, which could better conduct subsequent difference analysis.

Fig. 2.

Transcriptome data analysis. (A) PCA analysis; (B) Statistically significant differentially expressed genes; (C) Heat map of significantly differentially expressed genes of OGD-vs-H; (D) Heat map of significantly differentially expressed genes of OGD_R-vs-OGD.

Comparing the OGD group and OGD/R group with the H group respectively revealed 1229 differentially expressed genes (DEGs) (679 up/550 down) and 800 DEGs (590 up/210 down) (Fig. 2B). The top30 genes were shown in the heat map, and it was found that NBPF19, NOVA2, PPAN-P2RY11, PRAP1 and RAB4B-EGLN2 were significantly different in both groups (Fig. 2C and D).

Enrichment analysis

In the results of GO enrichment analysis, a significant number of terms were found to be associated with the cytoplasm, endoplasmic reticulum (ER), mitochondria and ribosome, such as ribosomal small subunit biogenesis (GO:0042274), mitochondrial respiratory chain complex I assembly (GO:0032981) and ribosomal large subunit assembly (GO:0000027) in OGD-vs-H or OGD_R-vs-OGD (Fig. 3A and C). Interestingly, in the KEGG enrichment results, more pathways were found to be enriched in ribosome and endoplasmic reticulum related pathways, such as: ribosome (hsa03010) and protein processing in endoplasmic reticulum (hsa04141) in both OGD-vs-H and OGD_R-vs-OGD. There are also some diseases associated with nerve damage were enriched, such as: Parkinson disease (hsa05012) and pathways of neurodegeneration - multiple diseases (hsa05022) in OGD_R-vs-OGD (Fig. 3B and D).

Fig. 3.

Enrichment analysis. (A) GO analysis of OGD-vs-H; (B) KEGG analysis of OGD-vs-H; (C) GO analysis of OGD_R-vs-OGD; (D) KEGG analysis of OGD_R-vs-OGD.

WGCNA analysis

The original data totaled 15,570 genes and 12 samples, and the genes with low expression fluctuation (standard deviation ≤ 0.5) were filtered, and the final remaining 2133 genes were 12 samples. In order to satisfy the premise of scale-free network distribution as far as possible, it is necessary to select the value of the adjacency matrix weight parameter power. When power was 30, the correlation coefficient was higher, and the average connectivity of the network was also higher, so the Power value used in the subsequent module construction was 30 (Fig. 4A). Set the power value from 1 to 30, and calculate the corresponding correlation coefficient and the average connectivity of the network respectively. A total of five modules (1663 magenta, 332 lightcyan, 56 grey, 46 tan and 36 cyan) were obtained by merging dynamic modules with DissThres set to 0.2 (Fig. 4B and C). The pearson correlation algorithm was used to calculate the correlation coefficient and p-value of the characteristic genes and traits of the modules. The absolute value of the correlation coefficient was greater than or equal to 0.3 and p-value was less than 0.05 as the threshold value, and the modules related to each trait were screened. For each trait related module, the correlation between the module gene expression and the corresponding trait (Gene Significance) was calculated, and the correlation between the module gene expression and the Eigengene was calculated. The results showed that magenta (1663) and lightcyan (332) were closely related to the characters (Fig. 4D).

Fig. 4.

Gene co-expression network analysis. (A) Left: illustrates the correlation coefficients corresponding to different power values; Right: depicts the average connectivity of networks constructed with varying power values; (B) Presents the Cluster Dendrogram, where the upper section displays the gene clustering tree constructed from the dissTOM matrix derived from weighted correlation coefficients, and the lower section shows the distribution of genes within each module, with the same color indicating genes belonging to the same module; (C) Displays the heatmap of all gene clusters; (D) Shows the heatmap of trait-module associations, ‘***’P-value < 0.001, ‘**’0.01 < P-value < 0.001, ‘*’0.01 < P-value < 0.05, ‘.’0.05 < P-value < 0.1.

Core candidate gene screening

In order to obtain the significant differential genes associated with OGD, the significant differential genes of the two difference comparison groups were intersected with the significant gene sets of lightcyan and magenta in WGCNA, and a total of 497 communicative candidate genes were obtained (Fig. 5A).

Fig. 5.

Machine learning screens for candidate genes. (A) Venn among WGCNA, OGD-vs-H and OGD_R-vs-OGD; (B) Random forest error rate graph, which shows how error varies with the number of trees. The solid black line represents the overall error rate, which represents the average error rate of the model for all data classifications. The line gradually stabilizes as the number of trees increases, indicating that the error rate gradually decreases and levels off as more trees are added to the model. The red and green dashed lines represent the OOB error rate for each classification category; (C) The figure shows the importance ranking of feature variables in the random forest model and the importance evaluation of each feature variable in the random forest model. Mean Decrease Gini was used to measure the contribution of each feature; (D) PPI of candidate genes; (E, F) The expression box pattern of candidate genes.

The R software package “randomforest” was used to further screen the candidate genes by random forest algorithm. (PMID: 37671131) and filtered TOP30 variables based on Gini > 0. The error rate of a random forest varies with the number of trees. The solid black line represents the Overall Error Rate, which represents the average error rate of the model for all data classifications. The line gradually stabilizes as the number of trees increases, indicating that the error rate gradually decreases and levels off as more trees are added to the model (Fig. 5B). Random forest screened the top 30 genes for subsequent analysis (Fig. 5C). PPI network analysis was performed on 30 candidate genes from random forest screening, and 6 key genes (RPS7, RPL36A, RPS9, RPL41, RSL24D1 and OSTC) with interaction relationship were obtained (Fig. 5D). In order to clarify the expression levels of key genes in OGD samples, OGD_R samples and normal samples, the expression levels of key genes were displayed based on FPKM. ggpubr package violin diagram was used for visualization and wilcox test was performed. The results showed that 5 key genes were significantly up-regulated in the OGD group compared with the H group, while the expression levels of 5 key genes were significantly down-regulated in the OGD_R group compared with the OGD group, which was consistent with the differences in sample processing methods (Fig. 5E and F).

Discussion

In this study, we utilized transcriptomic analyses to uncover biomarkers associated with BBB injury caused by cerebral infarction. Our findings provide novel insights into the molecular mechanisms underlying BBB disruption and highlight potential therapeutic targets for mitigating secondary brain injury following ischemic stroke.

Our transcriptome analysis extends the findings of early research by providing a comprehensive gene expression profile targeting BBB damage. The heatmap displays the top 30 genes, among which NBPF19, NOVA2, PPAN-P2RY11, PRAP1, and RAB4B-EGLN2 showed significant differences between the two groups (Fig. 2C and D). Some study has found neuro-oncological ventral antigen 2 regulates splicing of vascular endothelial growth factor receptor 1 and is required for endothelial function which explained NOVA2 associated with the blood-brain barrier³¹. Previous study also found PART1 destabilized by NOVA2 regulates blood-brain barrier permeability in endothelial cells via STAU1-mediated mRNA degradation³². Proline-rich acidic protein 1 (PRAP1) was found to protect the gastrointestinal epithelium from irradiation-induced apoptosis.

In the enrichment analysis of differentially expressed genes, it was found that genes were significantly enriched in pathways related to ribosomes, endoplasmic reticulum, and mitochondria, which have indeed been found to be closely associated with BBB damage in previous studies. For instance, studies have shown that during the injury process, mitochondrial dysfunction leads to a decrease in ATP supply, affecting cell viability and accelerating the functional decline of the blood-brain barrier³³. And BBB damage is often accompanied by local brain tissue hypoxia or ischemia, which may exacerbate ER stress². Additionally, ribosomes on the rough ER synthesize many secretory and membrane-associated proteins that are critical for maintaining the structure and function of the BBB. Although previous studies have emphasized the role of individual molecules such as MMP9 and vascular endothelial growth factor (VEGF) in blood-brain barrier disruption³⁴, this study identified six new potential targets through transcriptome analysis. Four genes (RPS7, RPL36A, RPS9 and RSL24D1) were closely related to ribosome function, demonstrating their potential as biomarkers^35–38. Although these molecules have been less studied in the context of blood-brain barrier injury, they may represent unexplored therapeutic targets.

Furthermore, our study highlights the dynamic changes in inflammatory cytokines and LDH levels under OGD and OGD/R conditions, reinforcing the link between inflammatory response and BBB integrity. The reduction in IL-1β, IL-6, and TNF-α following reoxygenation suggests that inflammatory pathways are reversible to some extent, which may inform timing for anti-inflammatory interventions.

The identification of BBB-specific biomarkers has significant implications for clinical practice. First, these biomarkers could serve as diagnostic tools for assessing the extent of BBB injury in real-time, potentially aiding in the stratification of stroke severity and prognosis. Second, targeting these molecules may pave the way for developing therapies aimed at preserving BBB integrity, thereby reducing secondary neuronal damage. For instance, pharmacological inhibitors of potential target or modulators of tight junction proteins could be explored as therapeutic interventions³⁹. Moreover, the ribosome-related genes identified here may open new avenues for neuroprotective strategies focusing on cellular stress responses and protein synthesis regulation.

Despite these promising findings, several limitations warrant consideration. First, the transcriptomic data were derived from cell models, and their relevance to human pathophysiology requires validation through clinical studies. Second, while transcriptomics provides a snapshot of gene expression changes, integrating proteomic and metabolomic data would offer a more comprehensive understanding of BBB injury. Finally, longitudinal studies are needed to delineate the temporal dynamics of gene expression changes during the acute and chronic phases of cerebral infarction.

Future research should focus on validating the identified biomarkers in human cohorts and exploring their mechanistic roles in BBB pathology. Additionally, the development of non-invasive imaging techniques or liquid biopsies to monitor these biomarkers in vivo would significantly enhance their translational potential.

In summary, our study identifies key biomarkers and molecular pathways associated with BBB injury in cerebral infarction. These findings enhance our understanding of the molecular mechanisms underlying BBB disruption and suggest potential therapeutic targets. Further investigation of these biomarkers could help guide future strategies aimed at preserving BBB integrity and improving outcomes after ischemic stroke.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Chuyu LIU designed the research. Xiaoyan Liu was involved in the composition of the manuscript. Yulan cai carried out cell experiments. Xiaoyan Liu conducted data analyses, generated figures, and drafted the manuscript. Xiaoyan Liu contributed to technical support and the creation of the figures. Chuyu LIU supervised the study and reviewed the whole text. All authors reviewed with all co-authors.

Funding

This study was supported by the: 1.National Natural Science Foundation of China (82260167;82460167). 2.Joint Fund of Zunyi Science and Technology Bureau, No.132, Zunshi Kehe HZ (2025). 3.Zunyi Medical University Future Clinical Excellence Physician Training Program (20231009) 4.High-Level Talent Fund of the Second Affiliated Hospital of Zunyi Medical University(GRC-2025-005)5.Guizhou Science and Technology Association Youth Science and Technology Talent Promotion Project (NO.GASTYESS202434).

Data availability

The transcriptome sequencing data analyzed in this study were generated from an in vitro model using hCMEC/D3 cells under three experimental conditions: Control (H group), Oxygen-Glucose Deprivation (OGD group), and OGD/Reperfusion (OGD/R group). The normalized gene expression matrix, which is the key data supporting the findings and conclusions of this manuscript, has been provided as Supplementary Information files accompanying this submission. The datasets generated and/or analysed during the current study are available in the GEO repository, [https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1367797 Accession: PRJNA1367797]

Declarations

Competing interests

The authors declare no competing interests.