Time series analysis reveals the effect of shear stress profile on endothelial gene expression

Abstract

Background

Endothelial cells (ECs) are central to vascular homeostasis, and their dysfunction is a key driver of atherosclerosis (AS). Hemodynamic forces, particularly laminar and disturbed flow (LF, laminar flow; DF, disturbed flow) patterns, differentially regulate ECs phenotypes via mechanosensing and epigenetic mechanisms. However, prior studies, largely reliant on single time-point analysis at isolated time points, fail to capture the dynamic gene expression responses of ECs to shear stress (SS). To address this, our study employs a time-series approach to dynamically delineate how distinct flow patterns shape ECs transcriptomes over time, aiming to identify pivotal regulatory genes and provide novel insights into AS pathogenesis.

Methods

In this study, samples were collected at multiple time points (0, 0.5, 2, 4, 6, 12, and 24 h) under LF or DF stimulation and subjected to RNA-seq. Time-series analysis was then performed to correlate data across these time points, thereby dynamically reflecting the effects of SS induced by different blood flow patterns on gene expression in endothelial cells. We identified two specific modules through weighted gene co-expression network analysis (WGCNA) to elucidate the effects of different blood flow patterns on ECs. Subsequently, a multi-database integration analysis was performed to screen for candidate genes.

Results

RNA-seq analysis of the multi-time-point transcriptomes of endothelial cells under different blood flow patterns revealed, through time-series analysis, that LF exerts a protective effect on homeostasis, whereas DF drives energy metabolism dysregulation, protein damage, and immunosuppression. WGCNA further identified that DF-specific modules were enriched in the endoplasmic reticulum stress pathway, while LF-specific modules were enriched in the oxidative phosphorylation pathway. Based on a multi-dataset cross-screening, the core genes HMOX1 and PTGIS were identified; HMOX1 was upregulated upon LF stimulation but downregulated in inflammatory and plaque tissues.

Conclusion

Through transcriptomic sequencing and RT-qPCR validation, this study’s further analysis of the HMOX1 gene reinforces the role of blood flow patterns in the progression of atherosclerosis and provides a basis for exploring hemodynamic-related diagnostic markers and therapeutic strategies. Further research on mechanosensitive genes and their signaling pathways may offer insights into the prevention and treatment of cardiovascular diseases.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12872-026-06233-6.

Background

ECs, as crucial components of the vascular system, dynamically regulate barrier function, maintain tissue microenvironment homeostasis, and actively participate in angiogenesis, tissue repair, and intercellular signal integration. The abnormality of these functions is closely associated with the onset and progression of various cardiovascular diseases, particularly pathological states such as AS [1]. Recent studies have revealed that distinct blood flow patterns—including LF and DF—profoundly influence the EC phenotype and function through mechanoreceptor activation and epigenetic reprogramming mechanisms [2]. LF, which predominantly occurs in straight vascular segments, produces stable shear stress known as laminar shear stress (LSS), exerting anti-inflammatory and antioxidant effects. It maintains endothelial homeostasis through epigenetic regulation, significantly inhibiting atherosclerotic plaque formation [3]. Conversely, DF, which is common at vascular bifurcations, bends, or stenotic regions, generates low and oscillatory shear stress termed oscillatory shear stress (OSS), readily inducing inflammatory responses, compromising the endothelial barrier, promoting lipid infiltration, and ultimately driving atherosclerotic plaque progression.

Although the effects of SS on ECs have received considerable attention [4], existing research has largely been confined to single time-point analysis at single or a few time points, making it difficult to comprehensively capture the dynamic changes in gene expression and molecular responses. This limitation hinders a deeper understanding of the regulatory mechanisms governing blood flow patterns and restricts the discovery of relevant biomarkers. To address this, the present study employs time series analysis (encompassing multiple time points: 0, 0.5, 2, 4, 6, 12, and 24 h) to dynamically elucidate the effects of SS imposed by different blood flow patterns on EC gene expression. By utilizing WGCNA and a multidatabase integration approach, we screen for key candidate genes to fill existing research gaps, thereby providing novel perspectives for mechanistic studies and clinical diagnosis of AS.

Method

Cell culture and processing

Human umbilical vein ECs (HUVECs, ScienCell, Catalog #8000) were cultured in Endothelial Cell Medium(ECM, ScienCell, Catalog #1001) supplemented with 10% fetal bovine serum (FBS, ScienCell), 1% endothelial cell growth factor (ECGS, ScienCell), 100 U/mL penicillin, and 100 µg/mL streptomycin (37 °C, 5% CO₂). The primary HUVECs were purchased from ScienCell Research Laboratories. Cells at passages 3 to 5 were used for the following experiments.

For shear stress (SS) treatment, 6–8 h before the experiment, observe and evaluate the growth status of the endothelial cells under a microscope. Prepare a suspension of 1.6 × 10⁶ cells/mL from well-grown endothelial cells (6th generation) and seed it onto µ-Slide I 0.4 Luer (channel slides). Seed 120–150 µL of cell suspension onto each slide to achieve a cell density of 1 × 10⁵ cells/cm². Place the lid on the Luer adapter and transfer the slides to a 37 °C cell culture incubator for later use. Cell samples collected from every 6 µ-Slide I 0.4 Luer (channel slides) are used as a single experimental sample.

For the inflammation assay, cells were seeded in 6 cm dishes and allowed to grow until approximately 80% confluence. They were then stimulated with either PBS (control) or TNF-α (10 ng/ml, Novoprotein, Catalog C008) for 6 h, followed by lysis and sample collection.

Fluid shear stress loading

Fluid shear stress loading was performed using the ibidi fluid shear stress system (version 2.4.1). As shown in Fig. 1A, the device consists of four fluid units; after connecting two slides via a serial connector, each unit can hold two ibidi slides to apply shear stress. For the laminar shear stress (LSS, 12 dyn/cm²) experiments, after applying LSS for 0.5 h, a total of 8 slides from the 4 units were collected. The two slides from the same unit were grouped together, resulting in 4 replicate groups. After collection, a new perfusion set and culture medium were replaced, and samples were collected at the other five time points (2, 4, 6, 12, and 24 h) following this procedure. Similarly, for oscillatory shear stress (OSS, 0.5 ± 4 dyn/cm²) experiments, samples were collected independently at the same six time points. The static control group (STA, 0 h) consisted of 8 slides without stimulation, randomly divided into 4 replicate groups, ultimately forming 13 independent sample groups, each containing at least 4 replicate samples. The flow medium was M199 culture medium supplemented with 10% FBS, and the system temperature was maintained at 37 °C (± 0.5 °C) with a CO₂ concentration of 5% V/V.

Fig. 1.

Sample quality control and data validation: A Schematic of SS loading in the ibidi fluid control system; B Pearson correlation heatmap between samples; C Principal component analysis (PCA) plot of samples; D Box plot of gene expression distribution, displaying gene expression levels across different sample groups on a log2 (FPKM + 1) scale

Transcriptome sequencing and preprocessing

Samples were collected at various time points (n = 4/group). Total RNA was extracted via the TRIzol method, and RNA integrity was assessed via Agilent 2100 (RIN ≥ 8.0). Beijing Novogene Co., Ltd. performed library preparation (NEBNext Ultra II Kit) and Illumina NovaSeq 6000 paired-end sequencing (150 bp, 6 Gb/sample). The raw data were subjected to FastQC quality control before alignment to the hg38 reference genome via HISAT2 (v2.2.1). Use the HISAT2 software to perform fast and accurate alignment of clean reads against the reference genome, thereby obtaining the reads’ mapping information on the reference genome. The raw fragment counts were normalized to FPKM (Fragments Per Kilobase of transcript per Million mapped fragments) values. This normalization accounts for both the sequencing depth (total mapped fragments) and the gene length, enabling comparative analysis of expression levels across different genes and samples. The resulting FPKM matrix was used for all subsequent differential expression and functional enrichment analyses.

Differential expression analysis

Based on the SS patterns, the samples were grouped into LF and DF categories. Time-gradient differential analysis was performed via DESeq2 (v1.38.3) for each time point versus the STA group. Significantly differentially expressed genes (DEGs) were identified via the following criteria: |log2FoldChange| > 1 and padj < 0.05.

Visualization implementation

Differentially expressed genes were visualized via volcano plots generated via the EnhancedVolcano package in R (v4.2.2). Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment bubble plots and receiver operating characteristic (ROC) curves were generated using Xiantao Academic (v4.0) (enrichment significance thresholds: P < 0.01, false discovery rate(FDR) < 0.05). Time-series analysis and clustering plots were generated via the ClusterGVis R package [5].

Atherosclerotic samples

Internal mammary artery (IMA) and atherosclerotic plaques were both harvested from the same patient during coronary artery bypass graft (CABG) surgery, and the patient provided informed consent. This study protocol was reviewed and approved by the Ethics Committee of the Affiliated Hospital of Southwest Medical University (NO.:KY2022179).

Real-time quantitative polymerase chain reaction (RT‒qPCR)

Total RNA was extracted from the cell samples via TRIzol (Invitrogen). Complementary DNA was synthesized via HiScript III RT SuperMix for qPCR (+ gDNA wiper) (Vazyme). RT–qPCR was performed with ChamQ Universal SYBR qPCR Master Mix (Vazyme). Human GAPDH served as an endogenous calibrator to normalize the expression levels of the target genes. Relative gene expression (normalized to that of the endogenous control GAPDH) was calculated via the comparative Ct method formula (2-ΔΔCt). The primers used for quantitative PCR were designed via Sangon Biotech.

Statistical analysis

Statistical analysis was performed with GraphPad Prism 10.1.2. The data were presented as the mean ± standard deviation (SD). An unpaired Student’s t-test was used to determine statistical significance. P < 0 0.05 was considered significant.

Results

Sampling quality control and data validation

Data quality control and preprocessing validation are paramount for ensuring the reliability of research findings. This study was conducted in strict accordance with standardized protocols, including sample collection, processing, and sequencing. To simulate the effects of human blood flow on the vascular endothelium, we employed German ibidi devices to subject ECs to LF or DF stimulation at durations of 0, 0.5, 2, 4, 6, 12, and 24 h (Fig. 1A).

To validate the reproducibility of the sequencing data, Pearson’s correlation analysis and principal component analysis (PCA) were employed. Pearson’s correlation results revealed strong correlations among the STA control, DF, and LF samples (Fig. 1B). PCA indicated that the samples within each group clustered closely on Principal Component 1 (PC1), whereas the groups remained distinct (Fig. 1C). Furthermore, box plots revealed a uniform overall gene expression distribution across all samples with no outliers, although some partial outliers suggested inter group differences (Fig. 1D).

Endothelial cell responses to different blood flow patterns

Differential expression analysis was performed via the limma R package for distinct time points (0.5 h, 2 h, 4 h, 6 h, 12 h, and 24 h). Within the LF and DF groups, each time point was compared against the STA control. Conversely, within the DF versus LF comparison group, time points were paired for comparison. Following external stimulation, cellular responses manifest in distinct, chronologically ordered phases. Within the initial seconds to minutes of stimulation, second messengers (e.g., cAMP and Ca²⁺) rapidly accumulate and activate protein kinase cascades. These rapid responses, independent of new protein synthesis, typically occur within 30 min [6]. Subsequently (approximately 1–2 h), transcription factors such as c-Fos and c-Jun are activated via phosphorylation, thereby initiating the transcription of immediate-early genes including FOS and JUN family genes. The mRNA levels of these genes typically peak at 1–3 h post-stimulation, with expression independent of new protein synthesis [7, 8]. Newly synthesized mRNAs must undergo translation to form functional proteins, a process that generally takes 3–6 h. The responses observed beyond 6 h often depend on the accumulation and functional realization of newly synthesized proteins (e.g., cyclin proteins, and inflammatory mediators) [9]. Accordingly, this study defines the 0–6 h post-stimulation period as the early cellular response phase, with 6–24 h designated the late response phase.

Multigroup volcano plots revealed that under LF stimulation, the number of DEGs gradually decreased and stabilized. In contrast, under DF stimulation, the number of DEGs fluctuated patterns even beyond the early response phase (Fig. 2A-B). Similarly, the number of DEGs fluctuated when the DF and LF control groups were compared. To investigate the impact of differential gene fluctuations on cells, enrichment analysis was performed on DEGs across groups. Considering 6 h as both the conclusion of the early response and the commencement of the late response, this intermediate point was selected as a reference to examine the effects of the two blood flow patterns on ECs (Fig. 2D-F). After 6 h of LF stimulation, compared with those in the static control group, the DEGs in ECs were significantly enriched in multiple biological processes related to vascular homeostasis and development. These included positive regulation of cell‒cell adhesion, positive regulation of vascular development, development of cardiac ventricles and chambers, the cellular response to vascular endothelial growth factor stimulation, and endothelial development, all of which presented positive correlations. Concurrently, although the positive regulation of the acute inflammatory response, NF-κB transcription factor activity, the NF-κB signaling pathway, and the tumor necrosis factor signaling pathway were also significantly enriched, their expression changes exhibited negative correlations. These findings suggest that LF may promote endothelial functional maturation and structural formation in the early stages by inhibiting inflammatory signaling pathways (Fig. 2D, G). In contrast, genes that were differentially expressed following 6 h of DF stimulation predominantly clustered in processes such as the response to hypoxia, the response to decreased oxygen levels, the response to oxygen levels, DNA replication initiation, and mitotic DNA replication. Among these pathways, hypoxia-related responses were positively correlated with the tumor necrosis factor signaling pathway, whereas the cell cycle, DNA replication, and cell adhesion molecule pathways were negatively correlated. These findings suggest that DF may initially induce EC stress responses by activating hypoxia and inflammatory signaling pathways, while simultaneously suppressing proliferation and adhesion-related functions (Fig. 2E, H). To distinguish between the two blood flow patterns more clearly, we performed paired analyses of DF (6 h) and LF (6 h) samples and identified enriched DEGs (Fig. 2C, F, I). Compared with those in LF-stimulated ECs, genes that were differentially expressed in ECs after 6 h of DF exposure were significantly enriched in functional categories related to circulatory system regulation and stress responses. Notably, positive correlations were observed for vascular processes in the circulatory system, the hypoxia response, the oxygen level response, the hypoxic level response, the regulation of anatomical structure size, fluid level regulation, and blood circulation regulation. Concurrently, the TNF signaling pathway, the PI3K‒Akt signaling pathway, and the NF‒κB signaling pathway were significantly activated. Notably, negative regulation of the transforming growth factor-β (TGF-β) receptor signaling pathway and responses to laminar fluid shear stress exhibited negative correlations. These findings suggest that DF may drive early endothelial phenotypic responses by inhibiting TGF-β signaling pathways and laminar-specific adaptive mechanisms, thereby activating inflammation- and stress-related signaling networks. These findings indicate that DF tends to induce stress and inflammatory responses, whereas LF promotes homeostasis and adaptive regulation, thereby clearly illustrating the distinct molecular-level effects of these two blood flow patterns.

Fig. 2.

Differential gene expression patterns: A–C Multigroup volcano plots of differentially expressed genes across time points for different sample groups, with red and green representing upregulated and downregulated genes respectively; D–F Volcano plots of differentially expressed genes(DEGs) between the LF 6 h and C groups, the DF 6 h and C groups, and the DF 6 h and LF 6 h groups; red and blue denote upregulated and downregulated genes respectively ; gray indicates no significant difference; G–I Combined GO-KEGG enrichment and logFC bubble plots for each group at 6 h; J Heatmap pattern illustrating the expression of key molecules at different time points under distinct blood flow patterns, linking phenotype-core pathway-key molecules. LF: laminar flow; DF: disturbed flow; C: control (static control)

To further explore the specific effects of different blood flow patterns on endothelial cells (ECs), we constructed a model that includes phenotype-core pathway-key molecule-key molecule expression heatmaps at different time points under two shear stress conditions. We selected several key phenotypes for study (Fig. 2J), aiming to infer phenotypic changes at the transcriptional level through changes in molecular expression. Gene expression heatmaps and phenotype-pathway correlation analyses indicate that DF and LF are involved in regulating various endothelial cell phenotypes. However, the color intensity in the heatmaps exhibits opposite patterns between DF and LF for genes such as CCND1, CYCS, Atg101, and KLF2; nevertheless, changes in individual gene expression at the transcriptomic level, without supporting functional evidence, are insufficient to reflect changes in the pathways.

Time-series analysis reveals dynamic changes in molecules

The phenotype-pathway-molecule model can only reflect changes in the transcriptional levels of individual molecules, but cannot account for pathway regulation. Therefore, are there molecules that exhibit similar trends and have significant expression levels, such that their synergistic effects can characterize the effects of laminar flow and DF on endothelial cells? To answer this question, we performed time-series analysis on data collected at seven time points under the two flow regimes.

We first determined the optimal number of clusters to be 8 via the getCluster function(Fig. 3A, C). Clustering was subsequently performed with the mfuzz R package, and expression trend lines were plotted for the clustered groups(Fig. 3B, D). GO/KEGG enrichment analyses were conducted on these 8 gene groups, with the top 5 enriched functions and pathways visualized (Figs. 4 and 5).

Fig. 3.

Time series analysis grouping: A Elbow method analysis diagram for the optimal number of clusters in DF group gene clustering. The horizontal axis represents the number of clusters k, whereas the vertical axis denotes the sum of squares within clusters, which is used to determine the optimal clustering number; B Normalized expression pattern diagram of the eight clustered groups within the DF group genes. Each sub diagram corresponds to one cluster, displaying the expression trends of genes within the cluster at different time points (C, DF 05, DF 2, DF 4, DF 6, DF 12, DF 24). The color represents the degree of gene membership within the cluster; C Elbow method analysis plot for the optimal number of clusters in the LF group gene clustering; D Standardized expression pattern diagram of the 8 clustering clusters for the DF group genes

Fig. 4.

Time-series analysis of disturbed flow: The left panel displays gene expression heatmaps for each cluster (C1-C8). The annotations above each heatmap indicate the number of genes within the cluster and their expression trend curves. The colors in the heatmaps denote the gene expression levels (red - high expression, green - low expression), with the x-axis representing the time points. The right panel lists representative genes for each cluster. The right panel presents functional enrichment analyses for each cluster, including GO terms (displayed as bar charts and scatter plots showing log10 (ratio) and -log10 (p-value)), and KEGG pathways (scatter plots showing -log10 (p-value)). These analyses reveal the functional characteristics of genes within each cluster from the perspectives of biological processes and pathways

Fig. 5.

Time-series analysis of laminar flow: The left panel displays gene expression heatmaps for each cluster (C1-C8). The annotations above each heatmap indicate the number of genes within the cluster and their expression trend curves. The colors in the heatmaps denote the gene expression levels (red, high expression, green, low expression), with the x-axis representing the time points. The right panel lists representative genes for each cluster. The right panel presents functional enrichment analyses for each cluster, including GO terms (displayed as bar charts and scatter plots showing log10 (ratio) and -log10 (p-value)), and KEGG pathways (scatter plots showing -log10 (p-value)). These analyses reveal the functional characteristics of genes within each cluster from the perspectives of biological processes and pathways

Through analysis of the temporal response patterns to fluid shear stress, we observed distinct sequential reprogramming in endothelial cell gene expression. Under DF stimulation, C1 and C3 exhibited gradual elevation, C2 and C7 showed fluctuating increases, C4 and C5 demonstrated fluctuating downregulation, C8 exhibited sustained downregulation, while C6 initially increased before decreasing. This reveals a multi-layered, dynamically coordinated stress adaptation program within the cells: Based on dynamic gene expression pattern analysis, endothelial cells under DF stimulation exhibit the following transcription remodeling characteristics associated with cellular damage: sustained upregulation of pathways related to energy metabolism (C1) and core protein degradation (C3), indicating persistent pressure from compensatory energy depletion and accumulating protein damage; Fluctuating upregulation of the autophagy pathway (C2) signaled organelle and protein aggregate damage; concurrently, fluctuating suppression of ribosomal biosynthesis (C4, C5) and sustained inhibition of innate immune defense responses (C8) respectively reflected depleted synthetic repair capacity and compromised immune surveillance. Collectively, these alterations indicate that sustained DF represents not merely physiological stimulation, but rather propels endothelial cells into a pathological state characterized by heightened metabolic expenditure, protein damage overload, diminished repair capacity, and weakened immune barriers. This may constitute the early molecular basis for vascular endothelial dysfunction arising from hemodynamic abnormalities(Fig. 4). Under LF stimulation, gene set expression exhibits highly dynamic and programmed cyclical variations. In the early phase (0.5–2 h), the NF-κB pathway (C6) undergoes transient activation, triggering an initial inflammatory stress response. Subsequently, the expression of protective transcription factors KLF2/4 is continuously induced and progressively amplified (C1). These factors powerfully suppress the early inflammatory program and dominate the establishment of a protective phenotype characterized by anti-inflammatory, antioxidant, and vasodilatory effects [10, 11]. This precise temporal transition from ‘brief alarm’ to ‘steady-state protection’ constitutes the core molecular mechanism by which endothelial cells adapt to laminar flow and maintain vascular health(Fig. 5). The periodic expression patterns of distinct gene sets directly reflect this multi-stage, adaptive reprogramming process at the genomic level. Moreover, sustained laminar flow stimulation may act as a potent ‘timing factor’, coupling or synchronizing with these intrinsic rhythms to induce regular periodic fluctuations in the expression of certain genes.

WGCNA identifies blood flow pattern-specific genomes

To identify key modules governing the effects of the two blood flow patterns on ECs, we analyzed gene expression changes across different time points for both flow patterns on the basis of time-series data. By employing WGCNA, combined with average connectivity hierarchical clustering and soft threshold selection, we ultimately identified 34 modules (Fig. 6A-C). Among these modules, the MEtan module was upregulated in DF and downregulated in LF, whereas the MEgreen module exhibited the opposite pattern (Fig. 6D-E). The MEtan module included 312 genes, whereas the MEgreen module included 734 genes. GO enrichment analysis revealed that the MEtan module, which represents DF specificity, was predominantly enriched in response to endoplasmic reticulum stress, positive regulation of the IκB kinase/NF-κB pathway, centrosome assembly, endoplasmic reticulum protein complexes, and ribosomal precursor functions (Fig. 6F). The MEgreen module, which included genes specific to LF, was enriched predominantly in functions related to cellular respiration, aerobic respiration, oxidative phosphorylation, and the mitochondrial inner membrane (Fig. 6G). Furthermore, KEGG enrichment analysis indicated that the MEtan module was enriched primarily in protein processing in the endoplasmic reticulum, whereas the MEgreen module was enriched predominantly in pathways associated with Huntington’s disease, Parkinson’s disease, and oxidative phosphorylation (Fig. 6F, G).

Fig. 6.

WGCNA screening of blood flow pattern-specific modules: A Left panel: scale independence analysis of soft threshold (power); right panel: average connectivity analysis of soft threshold, used to determine the optimal soft threshold for constructing co-expression networks; B cluster dendrogram displaying hierarchical clustering results, with colored bands below representing distinct co-expression modules; C module-sample relationship heatmap illustrating correlations between co-expression modules and samples, where color transitions from blue (negative correlation) to red (positive correlation) indicate varying strength of association; D MEtan module gene expression heatmap and histogram of module gene expression pattern distribution; E MEgreen module gene expression heatmap and histogram of module gene expression pattern distribution; F MEtan module gene GO‒KEGG enrichment analysis bubble plot; G MEgreen module gene GO‒KEGG enrichment analysis bubble plot

Screening and validation of core genes

To investigate the functional characteristics of the DF-specific and LF-specific modules identified by WGCNA within the protein‒protein interaction (PPI) network, this study conducted PPI analyses on both modules. Using the STRING database, an interaction network was constructed for genes within each module, setting a minimum interaction confidence threshold of 0.4 and hiding disconnected nodes. Visualization of the networks was performed via Cytoscape software (version 3.10.3), and the MCODE plugin was employed to identify core hub genes within each module on the basis of degree values. Within the MEtan module, the MCODE algorithm identified a highly interconnected core module comprising CPSF1, CPSF3, DIMT1, HEATR1, KRI1, MTREX, NMD3, NUP155, NUP188, NUP214, NUP62, PUS7, RRP1B, SYMPK, and TSR1 (Fig. 7A). The GO enrichment of this module was associated primarily with mRNA processing, ribosome biosynthesis, and nuclear‒cytoplasmic transport pathways. Specifically, further analysis of its expression patterns revealed that within DF, cells may synergistically increase gene expression by activating mRNA processing and nuclear export mechanisms, thereby promoting pathological responses such as inflammation and proliferation. Conversely, within protective LF, the suppression of this pathway may contribute to maintaining vascular homeostasis. Within the MEgreen module, the MCODE algorithm identified a highly synergistic core mitochondrial energy metabolism module comprising ATP synthases (ATP5F1A, ATP5F1E, ATP5MC1, ATP5MC3, ATP5MF, ATP5MG, ATP5PB, ATP5PF, ATP5PO), cytochrome c oxidase (COX6C, COX7B, COX8A), NADH dehydrogenase (NDUFA1, NDUFA11, NDUFA12, NDUFA4, NDUFA6, NDUFA9, NDUFAB1, NDUFB1, NDUFB2, NDUFB6, NDUFB8, NDUFB9, NDUFC1, NDUFS3, NDUFS4, NDUFS5, NDUFS6, NDUFV1), succinate dehydrogenase (SDHB, SDHD), and the ubiquitin-cytochrome c reductase complex (UQCR10, UQCR11) (Fig. 7B). This module comprehensively covers the four core complexes of the oxidative phosphorylation system. Its expression pattern indicates that under protective LF conditions, cells maintain efficient mitochondrial energy metabolism by fully activating the electron transport chain and ATP synthesis pathways. Conversely, under pathological DF conditions, widespread inhibition of this pathway may lead to an insufficient cellular energy supply, potentially promoting vascular endothelial dysfunction.

Fig. 7.

Screening and validation of core genes: A protein interaction network diagram of hub genes in the MEtan module; B protein interaction network diagram of hub genes in the MEgreen module; C Venn diagram of genes common to atherosclerosis, inflammation, and shear stress datasets in the GeneCard database, the DF C1 module in time-series analysis, and the MEtan module in WGCNA, identifying HMOX1, GFPT1, and P4HB; D ROC curves for the three genes in the public database GSE43292; E expression profiles of HMOX1 in the public databases GSE28829, GSE167024, and GSE213323, scaled as log2(fpkm + 1); F RT‒qPCR validation of HMOX1 expression across sample groups (IMA, Internal mammary artery; AS, Atherosclerosis; STA, static)

Further exploration of the WGCNA results via PPI analysis failed to definitively identify representative key molecules. Consequently, we sought to integrate public databases to identify potential core genes within our dataset that are correlated atherosclerosis, inflammation, and shear stress. Through temporal analysis, we selected the C6 dataset from the DF dataset, whose functional enrichment analysis confirmed that it was significantly associated with atherosclerosis pathways. To further elucidate its potential associations with shear stress and inflammatory responses, we integrated gene sets from the GeneCards database involving atherosclerosis, inflammation, and shear stress, alongside the light brown module from WGCNA. Through multidataset intersection analysis, we further identified core genes closely associated with shear stress and inflammatory processes, including HMOX1 and PTGIS (Fig. 7C). To evaluate the discriminatory capacity of these two genes for atherosclerosis, we employed receiver operating characteristic (ROC) curve analysis to assess their diagnostic efficacy. The GSE43292 dataset from the Gene Expression Omnibus (GEO) database, comprising tissue sections collected from 32 hypertensive patients who underwent carotid endarterectomy procedures, was selected. The samples comprised the media layer without the adventitia and the neointima layer, paired per patient: each patient provided one atherosclerotic plaque sample (Stary stage IV or higher, including the plaque core and shoulder regions) and one control sample of macroscopically intact distal tissue (corresponding to Stary stage I–II lesions). ROC analysis demonstrated moderate accuracy for HMOX1 (AUC = 0.847) and PTGIS (AUC = 0.783) (Fig. 7D). Furthermore, we examined the expression patterns of the HMOX1 in additional datasets. We observed that HMOX1 expression exhibited a decreasing trend in late-stage plaques compared to early-stage plaques. Following 48-hour and 72-hour exposure to laminar shear stress stimulation, HMOX1 expression significantly increased. (Fig. 7E). RT‒qPCR validation of these three genes revealed that HMOX1 expression was lower in atherosclerotic plaque tissue (n = 3) and inflammatory conditions (n = 12) than in control tissue but was significantly elevated under laminar shear stress (Fig. 7F).

Discussion

This study aimed to investigate the effects of different blood flow patterns on vascular ECs from a transcriptomic perspective. Through a combined approach of in vitro cell culture and bioinformatics analysis, we conducted an in-depth investigation into the effects of hemodynamics on EC gene expression. This comprehensive research methodology not only reveals the multifaceted impacts of varying blood flow patterns on ECs but also may provide novel biomarkers and therapeutic targets for the prevention and treatment of AS, thereby advancing the development of personalized medicine. This study aimed to investigate the effects of different blood flow patterns on ECs. By thoroughly analyzing the significance of hemodynamics in vascular biology, we seek to elucidate the specific impacts of blood flow patterns on EC function and pathological states. The strength of this study lies in its combination of multiple experimental techniques, ensuring both the reliability and comprehensiveness of the results. Key findings include significant alterations in EC gene expression, cellular phenotype, and function under varying blood flow conditions.

On the basis of high-throughput sequencing data, the use of bioinformatics to investigate the mechanisms of AS may facilitate the development of therapeutic agents [12]. This study employed multitime point expression data from ECs stimulated by different blood flow patterns. Through differential expression analysis, time series analysis, and WGCNA, key genes influencing AS progression under varying blood flow conditions were identified.

Through differential gene expression analysis, we observed that ECs presented significant alterations in gene expression under different blood flow patterns. Under LF, the number of DEGs gradually decreased toward a stable, lower level (Fig. 2A); conversely, under DF conditions, the reduction in DEGs was less pronounced and accompanied by fluctuations. Specifically under DF stimulation, multiple endothelial phenotypes exhibit pronounced temporal regulation, suggesting a complex vicious-cycle mechanism.

Under DF stimulation, ECs exhibit significant dynamic changes in gene expression and functional dysregulation, whereas under LF conditions, they maintain a relatively stable, quiescent state. Endothelial homeostasis-related genes show a dysregulated pattern in DF: KLF2 displays early high expression followed by a sharp decline, while NOS3 and NFKBIA maintain moderate expression, and SOD2 shows early high expression followed by attenuation. This collectively reflects disturbances in the KLF2/eNOS axis, antioxidant defense, and NF-κB pathways. In contrast, these genes remain low and stable under LF. In terms of inflammatory activation, DF drives upregulation of NLRP3 (moderately fluctuating high expression), IL1B (early high expression followed by decline), and CASP1 (sustained moderate expression) via the NLRP3 inflammasome pathway, whereas no significant activation is observed in LF. Endothelial-mesenchymal transition (EndMT) in DF is regulated by the TGF-β/Smad pathway, with SNAI2 (shifting from early low to moderate expression), VIM (late-phase high expression), and TWIST1 (early high expression followed by a sharp decrease) indicating the initiation and progression of EndMT. The barrier dysfunction gene CLDN5 remains persistently low in DF, while OCLN shows late-phase high expression in LF, suggesting differential regulation of tight junction pathways.

Analysis of cellular behavior further reveals a vicious cycle under DF: the proliferation-related gene PCNA increases continuously over time, reaching significantly elevated levels at 12 h, while MKI67 shows high expression in the early phase (e.g., 6 h) before declining around 12 h, indicating an initial increase followed by decreased cell cycle activation. Concurrently, the apoptosis-related gene CYCS exhibits low expression in the early phase (e.g., 0.5 h), gradually increases thereafter, and maintains moderate-to-high levels between 12 and 24 h, suggesting delayed activation of the apoptotic pathway. This pattern of early proliferation surge and late-phase enhanced apoptosis implies that DF may induce a vicious cycle of “malignant proliferation–apoptosis bursts,” where cells initially attempt to proliferate under stress but ultimately undergo programmed cell death. Additionally, late-phase activation of the autophagy gene Atg101 reflects cellular attempts to counteract stress via autophagy, while the transient early rise and subsequent sharp decline of the endothelial homeostasis gene KLF2, along with early activation and subsequent attenuation of inflammatory genes NLRP3 and IL1B, highlight disrupted homeostasis and an imbalanced inflammatory response. EndMT genes such as VIM show high expression in the late phase, indicating cellular reprogramming toward a mesenchymal phenotype, and the persistent low expression of the barrier function gene CLDN5 reveals tight junction dysfunction.

In contrast, under LF conditions, cellular proliferative capacity, though weak, remains relatively stable, consistent with the quiescent state of vascular ECs. Apoptosis and autophagy trends are steady, endothelial homeostasis is maintained, and inflammatory activation, EndMT, and barrier dysfunction are all suppressed. Only the late-phase high expression of OCLN suggests mild late-stage barrier dysfunction in LF. Overall, DF induces a sustained dysregulated state by perturbing endothelial homeostasis, driving inflammation and EndMT, and triggering an imbalance between proliferation and apoptosis. In comparison, LF maintains relative environmental stability, underscoring the critical role of hemodynamic differences in regulating endothelial cell function.

Overall, DF drives the temporal expression of genes through core pathways (such as the PI3K-AKT-mTOR pathway, mitochondrial apoptosis, and the TGF-β/Smad pathway), leading to the loss of normal barrier function in ECs and promoting inflammation and tissue fibrosis [13, 14]. Genes, as molecular markers, effectively reveal the dynamic interactions between phenotypes. These findings indicate that stable LF contributes to maintaining normal EC function, potentially benefiting cardiovascular health [15], whereas low and oscillatory DF compromises endothelial barrier integrity, thereby predisposing patients to subsequent disease progression [16].

In the time-series analysis, certain genes exhibited fluctuating patterns during early exposure to DF or LF stimuli, despite showing overall up- or downregulation trends. These findings indicate that negative feedback mechanisms within static ECs counteract these external stimuli to maintain endothelial homeostasis [17]. We observed that both DF and LF, when applied to quiescent ECs, induce inflammatory responses. The distinction lies in the fact that under DF conditions, the expression of inflammatory cytokines in ECs is significantly greater than that under LF conditions. However, once ECs have passed the initial response phase (0–6 h), DF ultimately disrupts endothelial homeostasis, whereas LF maintains endothelial homeostasis, with most gene expression levels approximating quiescence. In other words, owing to its stability, LF can be regarded as a form of quiescence once cells adapt to its stimulus. In contrast, DF is oscillatory, continuously delivering novel stimuli to ECs, disrupting their homeostasis and thereby contributing to subsequent disease progression [18]. Furthermore, we observed that the gene expression trends across all the groups under LF exhibited smoother, wave-like patterns with apparent periodicity. The C6 group also clearly enriched circadian rhythms. This raises the following question: does this periodic gene expression originate from the inherently static nature of the cells themselves, or does it emerge only after cells adapt to LF and enter a different form of ‘static’ state?

In WGCNA, we identified two modules exhibiting distinct characteristics: one specific to DF and the other to LF. Enrichment analysis of these modules revealed that DF induces a ‘stress-adaptation-dysfunction’ state in cells subjected to high metabolic stress. Ribosome synthesis and centrosome expansion jointly drive the high metabolic state of the cell, thereby supporting the protein translation required for the stress response. However, abnormal centrosome assembly can lead to chromosomal instability [19], causing ECs to enter a state of dysregulation with impaired cellular function. These findings provide a pathological basis for the onset and progression of various diseases. Conversely, LF promotes efficient energy generation by enhancing mitochondrial ATP synthesis, thereby supporting endothelial homeostasis. The downregulation of this gene cluster under DF suggests a shift from efficient oxidative metabolism to inefficient glycolysis, indicating diminished adaptability. As energy metabolism disruption precedes morphological changes (e.g., lipid deposition), it may serve as an early biomarker for blood flow-related diseases.

This study, based on a multi-database integrated analysis, ultimately identified two genes with potential diagnostic value: HMOX1 and PTGIS. Among these, HMOX1 demonstrated optimal diagnostic capability, with an area under the receiver operating characteristic curve (AUC) of 0.847. RT-qPCR experiments validated HMOX1 expression in samples subjected to atherosclerosis, inflammation, and laminar shear stress. Results revealed that HMOX1 was significantly downregulated in atherosclerotic plaques and inflammatory tissues compared to controls, whilst being highly expressed under laminar shear stress conditions, suggesting its potential as a reliable biomarker. These findings indicate that HMOX1 not only possesses strong diagnostic potential but may also serve as a target for subsequent mechanistic studies and therapeutic interventions. Indeed, existing literature indicates that HMOX1 plays a central role in regulating endothelial homeostasis under blood flow shear stress and in the progression of atherosclerosis. Hemoxygenase-1 (HO-1), encoded by the HMOX1 gene, is an important stress-induced enzyme. Its core function is known to involve the degradation of heme—a compound with pro-oxidant and pro-inflammatory properties—to produce carbon monoxide (CO), biliverdin/bilirubin, and free iron, thereby exerting broad cytoprotective, antioxidant, anti-inflammatory, and anti-apoptotic effects under various pathophysiological conditions [20–22].

Its expression and function are precisely modulated by distinct shear stress patterns, profoundly influencing atherosclerotic development. LF significantly upregulates HMOX1 expression in endothelial cells, with LF-induced HMOX1 upregulation exhibiting significant anti-inflammatory and antioxidant effects [23, 24]. Conversely, the atherogenic DF also induces HMOX1 expression [25], but this is accompanied by sustained oxidative stress. In contrast, protective laminar shear stress not only induces HMOX1 but also activates other antioxidant defense mechanisms, thereby restoring redox homeostasis. Although studies suggest that high HMOX1 expression may adversely affect plaque stability [26], atherosclerosis is a complex pathological process involving multiple stages and mechanisms. The function of HMOX1 may exhibit significant differences across distinct cell types. As the first line of defense within the vascular endothelium, endothelial cells possess unique response mechanisms to hemodynamic alterations and circulating harmful factors. We therefore hypothesize that HMOX1 may exert anti-atherogenic protective effects in endothelial cells by responding to laminar shear stress and suppressing inflammatory responses. This action is not contradictory to its influence during plaque progression; rather, it reflects the functional diversity of this molecule across different disease stages.

Although this study, through temporal analysis, WGCNA, and multidatabase integration, dynamically elucidates for the first time the mechanisms by which different blood flow patterns shape EC phenotypes via epigenetic reprogramming and inflammatory regulation, and identifies HMOX1 as a key responsive gene, offering new perspectives for the early diagnosis and targeted treatment of atherosclerosis, certain limitations remain. First, while the ibidi in vitro shear stress loading model simulates LF and DF, it cannot fully replicate the complex in vivo microenvironment (e.g., cell‒cell interactions and nutriregulation). Furthermore, the inherent differences between the HUVECs used and the arterial endothelium may limit the generalizability of the conclusions. Second, although the time series covers multiple points, the maximum 24-hour observation period struggles to capture long-term epigenetic adaptation events during the chronic progression of atherosclerosis. These limitations point to future research directions, including integrating multiomics data, utilizing primary cells or animal models for long-term validation, and conducting large-scale clinical studies to deepen the understanding of the mechanisms by which hemodynamics modulate endothelial function and to advance clinical translation.

Within contemporary medicine, atherosclerosis represents an increasingly severe disease, that has emerged as a significant contributor to cardiovascular events worldwide. It alters vascular wall structure and function by impairing EC performance, thereby increasing the risk of cardiovascular and cerebrovascular disorders. The pathological mechanisms of atherosclerosis are complex, and involve multiple factors, including genetic predispositions, environmental influences, and mechanical stress. Among these factors, alterations in blood flow patterns are recognized as critical determinants of EC function. Stable LF patterns help maintain the normal physiological state of ECs, whereas DF patterns may accelerate atherosclerotic progression by inducing inflammatory responses and apoptosis. Consequently, therapeutic strategies based on regulating blood flow patterns to improve vascular EC function are particularly crucial.

Conclusion

This study suggests that shear stress may influence endothelial cell phenotypes, potentially through mechanisms involving inflammatory regulation, as supported by transcriptomic sequencing and RT‑qPCR validation. The observed association with the HMOX1 gene highlights a potential link between blood flow patterns and atherosclerosis progression, offering a basis for further exploration of hemodynamics‑related markers and therapeutic strategies. Future investigation into mechanosensitive genes and their signaling pathways may provide additional insights for cardiovascular disease prevention and treatment.

Abbreviations

ECs

Endothelial cells

AS

Atherosclerosis

LF

Laminar flow

DF

Disturbed flow

SS

Shear stress

WGCNA

Weighted gene co-expression network analysis

HMOX1

Heme oxygenase-1

HUVECs

Human umbilical vein ECs

FBS

Fetal bovine serum

ECGS

Endothelial cell growth factor

LSS

Laminar shear stress

OSS

Oscillatory shear stress

STA

Static

DEGs

Differentially expressed genes

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

ROC

Receiver Operating Characteristic

FDR

False discovery rate

IMA

Internal mammary artery

CABG

Coronary artery bypass graft

PCA

Principal component analysis

PC1

Principal Component 1

EndMT

Endothelial-mesenchymal transition

PPI

Protein-protein interaction

GEO

Gene Expression Omnibus

Authors' contributions

Zhenyu Gao was responsible for the study design, data analysis, and manuscript drafting. Haozhong Huang undertook model construction, sample collection, and initial manuscript review. Linlu Liu and Yuhao Li contributed supplementary content. Yu Fengxu and Nie Yongmei conducted the final manuscript review and submission. All the authors approved the final manuscript.

Funding

Sichuan Provincial Department of Science and Technology Natural Science Foundation General Program (Number: 2025ZNSFSC0707), Key Laboratory of Medical Electrophysiology Open Fund research project (No.: KeyME- 2023-01). Supported by Health Commission of Sichuan Province Medical Science and Technology Program (No: 24LCYJPT19).

Data availability

The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA-Human: HRA014259) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human [27, 28]. Data available on request from the authors.

Declarations

Ethics approval and consent to participate

This study protocol was reviewed and approved by the Ethics Committee of the Affiliated Hospital of Southwest Medical University (NO.:KY2022179), and all participants provided written informed consent at the time of participation. This study complies with the requirements of the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.