Identifying potential signatures for atherosclerosis in the context of predictive, preventive, and personalized medicine using integrative bioinformatics approaches and machine-learning strategies

Abstract

Background

Atherosclerosis is a major contributor to morbidity and mortality worldwide. Although several molecular markers associated with atherosclerosis have been developed in recent years, the lack of robust evidence hinders their clinical applications. For these reasons, identification of novel and robust biomarkers will directly contribute to atherosclerosis management in the context of predictive, preventive, and personalized medicine (PPPM). This integrative analysis aimed to identify critical genetic markers of atherosclerosis and further explore the underlying molecular immune mechanism attributing to the altered biomarkers.

Methods

Gene Expression Omnibus (GEO) series datasets were downloaded from GEO. Firstly, differential expression analysis and functional analysis were conducted. Multiple machine-learning strategies were then employed to screen and determine key genetic markers, and receiver operating characteristic (ROC) analysis was used to assess diagnostic value. Subsequently, cell-type identification by estimating relative subsets of RNA transcript (CIBERSORT) and a single-cell RNA sequencing (scRNA-seq) data were performed to explore relationships between signatures and immune cells. Lastly, we validated the biomarkers’ expression in human and mice experiments.

Results

A total of 611 overlapping differentially expressed genes (DEGs) included 361 upregulated and 250 downregulated genes. Based on the enrichment analysis, DEGs were mapped in terms related to immune cell involvements, immune activating process, and inflaming signals. After using multiple machine-learning strategies, dehydrogenase/reductase 9 (DHRS9) and protein tyrosine phosphatase receptor type J (PTPRJ) were identified as critical biomarkers and presented their high diagnostic accuracy for atherosclerosis. From CIBERSORT analysis, both DHRS9 and PTPRJ were significantly related to diverse immune cells, such as macrophages and mast cells. Further scRNA-seq analysis indicated DHRS9 was specifically upregulated in macrophages of atherosclerotic lesions, which was confirmed in atherosclerotic patients and mice.

Conclusions

Our findings are the first to report the involvement of DHRS9 in the atherogenesis, and the proatherogenic effect of DHRS9 is mediated by immune mechanism. In addition, we confirm that DHRS9 is localized in macrophages within atherosclerotic plaques. Therefore, upregulated DHRS9 could be a novel potential target for the future predictive diagnostics, targeted prevention, patient stratification, and personalization of medical services in atherosclerosis.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13167-022-00289-y.

Introduction

Atherosclerosis is the most common underlying pathology of atherosclerotic cardiovascular diseases (ASCVD) [1, 2], which has remained the leading cause of death, disability, and high healthcare costs worldwide [3, 4]. Many of the risk factors for atherosclerosis are modifiable, and thus, atherosclerosis can be prevented in many cases. However, it is usually not intervened until symptoms occur, even the onset of ASCVD. Actually, from viewpoint of predictive, preventive, personalized medicine (PPPM), the intervention after the occurrence of symptoms or the onset of ASCVD is considered as an obviously delayed response. Over the past years, PPPM is recommended to apply in the comprehensive management of atherosclerosis, because it adopts a holistic strategy to predict individual predisposition, provide targeted prevention, and create personalization of medical services [5–7]. Considerable effort has been directed towards biomarkers associated with atherosclerosis under the PPPM paradigm. For example, C-reactive protein and interleukin-6 have been used as the powerful predictors and targeted therapies for atherosclerosis [8–10]. The elevated concentrations of plasma homocysteine (Hcy) increase the risk of atherosclerosis and ASCVD, and Hcy metabolic pathways are regarded as the powerful predictive and prognostic targets, as well as the specific targets for cost-effective preventive treatments and optimized interventions tailored to the individualized patient [11]. In addition, angiopoietin-2 [12], angiopoietin-4 [13], matrix metalloproteinase-12 [12], C-X-C motif chemokine receptor 4 (CXCR4)/C-X-C motif chemokine ligand 12 (CXCL12) [14, 15], trimethylamine-N-oxide [16], and other predictive genes[17] may provide the opportunities of early diagnosis, risk stratification, and personalized intervention with atherosclerosis. Unfortunately, due to the divergent results in the PPPM practice of atherosclerosis, the molecular biomarkers have not been fully used to guide the clinical management of atherosclerosis [18–20]. Therefore, finding novel and robust biomarkers is a feasible option to improve PPPM in atherosclerosis.

Recent burgeoning evidence has demonstrated that immune and inflammation mechanisms play the crucial roles in the pathogenesis of atherosclerosis [21, 22]. Although innate and adaptive immunity cooperates to promote occurrence and progression of atherogenesis [23, 24], the molecular system permitting different immune cells to impact atherosclerosis has not been fully elucidated [25]. Hence, assessment of immune cells’ infiltration involving in atherosclerosis is essential for clarifying its molecular pathogenetic mechanisms from immune perspective. Studies have shown that the abundance of most immune cells have significant differences between atherosclerosis and controls, such as M0 and M1 macrophages, monocytes, and T cells [26, 27]. Moreover, single-cell RNA sequencing (scRNA-seq) data have revealed the characteristics of immune cell subsets in atherosclerotic arteries [28–30]. Thus, elucidating the molecular immune mechanism of genetic markers underlying atherosclerosis onset will contribute to the biomarkers’ application of PPPM. Accordingly, this encourages us to investigate the relationships between biomarkers and immune cells based on the omics data.

Working hypothesis

In the present study, we hypothesized that key biomarker genes of atherosclerosis were identified through integrative bioinformatics approaches and machine-learning strategies, and the immune mechanisms of biomarkers underlying atherosclerosis are identified by multi-omics analysis. The novel signatures were the potential targets for predictive/diagnostic tool, targeted prevention, and personalized therapy in atherosclerosis from the perspective of PPPM.

Study design

Firstly, we obtained atherosclerosis microarray datasets from Gene Expression Omnibus (GEO) repository to analyze for identifying differentially expressed genes (DEGs). We then combined bioinformatics analysis with multiple machine-learning strategies to screen and determine critical signatures of atherosclerosis in depth. Receiver operating characteristic (ROC) analysis was used to assess their diagnostic values. Since the investigation of signatures-related molecular mechanism may help to their further clinical practice in the context of PPPM in atherosclerosis, we sought to explore the relationships between biomarker genes and immunity. Next, we applied cell-type identification by estimating relative subsets of RNA transcript (CIBERSORT) to clarify the differences in immune infiltrates of atherosclerotic plaques between atherosclerosis and control tissues. To gain more insights into the molecular immune mechanism, we conducted correlation analysis to elucidate the relationships between key biomarkers and immune cells in atherosclerosis. In addition, we interrogated a scRNA-seq dataset to more precisely elaborate the expression of signatures in immune cells. Lastly, we validated the expression of biomarkers in atherosclerotic vessel tissues from patients and mice. Collectively, this study demonstrated a robust and feasible genetic marker of atherosclerosis and provided a potential target for the diagnosis, prevention, and treatment of atherosclerosis in framework of PPPM.

Materials and methods

Dataset collection

Data analysis procedures of our study are shown in Fig. 1. The original microarray datasets of GEO series (GSE)100,927 and GSE43292 were downloaded from National Center of Biotechnology Information-GEO (NCBI-GEO), which is an international public repository for microarray data. GSE100927 [31] dataset includes 29 atheromatous carotid plaques and 12 control carotid arteries without atherosclerotic lesions, based on GPL17077 platform. GSE43292 [32] contains 32 carotid atherosclerotic samples and 32 carotid non-atherosclerotic samples, based on GPL6244 platform. The annotation files for GPL570 and GPL6244 were downloaded from GEO. For a gene with multiple probe IDs measured, we calculated average expression value of all those probes to represent the single gene expression level. In addition, GSE28829 [33], composed of 13 early and 16 advanced atherosclerotic plaque samples from the human carotid artery, was used as a validation set to verify the expression of signatures. Furthermore, scRNA-seq data GSE159677 was performed to delineate the cell populations expressing key biomarkers in atherosclerosis. The samples of GSE159677 [28] are carotid atherosclerotic plaques and patient-matched proximal adjacent portions collected from three patients. The clinical information of these datasets is summarized in Table S1. Quality control analysis and microarray data pre-processing, including background correction and normalization, were performed in R using the Bioconductor package [34] before formal analysis.

Fig. 1.

Flowchart of the study

Identification of DEGs

Here, we conducted differential expression analysis to identify DEGs by R Bioconductor package limma (V3.46.0) [35]. P-values were adjust by Benjamini–Hochberg’s false discovery rate (FDR), and genes with the adjusted P-value < 0.05 and |Log2 fold-change (log2FC)|≥ 0.585 were defined as differentially expressed. Volcano plots were generated by R software ggplot2 V3.3.5 package [36], and heatmaps for top 50 DEGs from each dataset were plotted by R software Pheatmap V1.0.12 package [37], and the VennDiagram was drawn with R software VennDiagram V1.6.20 package[38].

Functional enrichment analysis of DEGs

To explore the function and pathway of the overlapping DEGs, the functional enrichment analysis was conducted using the R software clusterProfiler V3.18.1 package [39] and the Goplot V1.0.2 package [40] (significant as a P < 0.05 and a q-value < 0.05). For all overlapping DEGs, gene ontology (GO) terms (BP, biological process; CC, cellular component; and MF, molecular function) as well as kyoto encyclopedia of genes and genomes (KEGG) pathways enrichment analysis were conducted and visualized. In addition, we performed GO enrichment analysis for the upregulated and downregulated overlapping DEGs, respectively.

Screening and validation of critical gene signatures

Three algorithms were used to screen novel and key signatures for atherosclerosis in the present study, which were random forests (RF) [41], least absolute shrinkage and selection operator (LASSO) logistic regression [42], and weighted gene co-expression network analysis (WGCNA) [43]. RF model was implemented via the R software randomForest V4.7–1 package. LASSO logistic regression analysis was performed using R software glmnet V4.1–4 package. WGCNA was conducted using R software WGCNA V1.71 package. Then, overlapping genes were identified among the three classification models for further analysis. To evaluate the diagnostic value of signatures for atherosclerosis, ROC curves and area under the curve (AUC) were calculated using R software pROC V1.18.0 package [44]. A two-sided P < 0.05 defined statistical significance. For the in-depth validation of key biomarkers’ accuracy, the validation dataset GSE28829 combined with GSE100927 and GSE43292 was used to evaluate the expression of gene signatures between atherosclerosis and control.

Construction of a multifactor regulatory network

Long non-coding RNA (lncRNA), microRNA (miRNA), or transcription factor (TF) is the regulators of gene expression at transcriptional or posttranscriptional levels. To elucidate the factors contributing to the regulation of key gene signatures, we performed starbase [45], hTFtarget [46], and mirDIP [47] to predict lncRNAs, miRNAs, and TFs corresponding to DEGs, and a multifactor regulatory network was then visualized using Cytoscape V3.8.2 software [48].

Determination, evaluation, and correlation analysis of infiltrated immune cells

Before performing CIBERSORT analysis, we first visualized and plotted the principal component analysis (PCA) of gene expression profile across all samples in GSE100927 with R software factoextra V1.0.7 package. CIBERSORT was then performed to analyze the infiltration of 22 kinds of the immune cells. We obtained the relative abundance of infiltrated immune cell according to P < 0.05, and then drew the correlation heatmap for visualizing the correlation of infiltrated immune cells through R software corrplot V0.92 package, and next explored the differential infiltration in immune cells between atherosclerosis and control groups using Wilcoxon rank sum test, and subsequently analyzed the Spearman relationship between biomarkers and infiltrating immune cells. The results was visualized via R software ggstatsplot V0.9.1 package.

Single-cell RNA sequencing data analysis

The barcodes data, gene features data, and gene count matrix data of GSE159677 preprocessed by Cellranger (10X Genomics) were downloaded from GEO database. These data were imported in R, and analyzed using Seurat V4.1.0 package [49]. Firstly, quality control was conducted through filtering out cells satisfying the following criteria: a gene count per cell > 200 and < 2500, and a percentage of mitochondrial genes < 5%. Next, the data were normalized by NormalizeData function. For the downstream analysis, top-ranked 2000 variably expressed genes were selected using “vst” method in FindVariableFeatures function. Before the PCA, the data were scaled using ScaleData function. The data were then subjected to PCA, cluster analysis, and Uniform Manifold Approximation and Projection (UMAP) dimensional reduction with RunPCA, FindClusters, and RunUMAP functions. Subsequently, the cell clusters were visualized using the UMAP plots displayed by the DimPlot function. Different expressions of signatures were determined with the FindAllMarkers function. Violin plots were drawn using the VlnPlot function. Furthermore, we applied R software SingleR V1.4.1 package [50] to annotate cell types, and we employed the celldex V1.0.0 package to download the HumanPrimaryCellAtlasData reference.

Human atherosclerotic samples

To validate the expression of the critical biomarker in human atherosclerotic samples, human lower extremity arterial tissue samples were harvested from amputees during surgical operations, who underwent lower extremity amputations for suffering a traffic accident or other accident. The arteries were prepared for western blot analysis and histopathological examination. We detected hematoxylin and eosin staining to categorize atherosclerosis and normal control. Three lower extremity arterial samples with atherosclerosis and four normal controls without atherosclerosis were incorporated in the study. The protocol for collecting human tissue samples was approved by Ethics Committee of the General Hospital of Central Theater Command (Wuhan, China). Written informed consent was provided by all participants before enrollment.

Mice atherosclerotic samples

In the animal experiments, twelve male ApoE^–/– mice aged 6 weeks were obtained from Peking Weitong Lihua Experimental Animal Technology (Beijing, China). All mice were fed adaptively for 2 weeks prior to the experiment. To induce atherosclerosis, six mice were switched from a normal chow diet (NCD) to a western diet (WD) at 8 weeks of age and were fed WD (40% kcal fat, 43% kcal carbohydrates, and 17% kcal protein; Beijing Hfk Bioscience Co. Ltd., Beijing, China) for 12 weeks. Another six mice were fed with NCD for 12 weeks and used as the control group. At the terminal of the study, mice were anesthetized by intraperitoneal injection of pentobarbital sodium (60 mg/kg) and euthanized for aortic samples. The animal experimental protocols were permitted by the Animal Ethics Committee of the General Hospital of Central Theater Command (Wuhan, China).

Immunofluorescence staining

Tissue slices of human and mouse arterial tissues were obtained from samples set in paraffin. For immunofluorescence staining, the samples were incubated overnight at 4 °C, and stained with DHRS9 polyclonal antibody (1:300; Cat# 14,560–1-AP; Proteintech), and a Cy3-labeled polyclonal secondary antibody(1:500; Cat# G1223; Servicebio). We costained the sections with a CD68 antibody (1:5000; Cat# 66,231–2-Ig for human; 1:5000; Cat# 28,058–1-AP for mouse; Proteintech), and an FITC-labeled polyclonal secondary antibody (1:500; Cat# G1222; Servicebio). Slides were counterstained with 4′,6-diamidino-2-phenylindole (DAPI). Finally, images were acquired using a fluorescence microscope (Nikon Eclipse C1, Nikon, Japan) and were quantified using Image-Pro Plus analysis software V8.0 (Media Cybernetics).

Western blot

The protein expression of DHRS9 was measured by a standard western blot assay from human and mouse arterial tissues. Primary antibodies used were anti-DHRS9 (1:600; Cat# 14,560–1-AP; Proteintech) and anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (1:20,000; Cat# 60,004–1-Ig; Proteintech). HRP-linked secondary antibodies (anti-mouse IgG, Cat# 15,014; anti-rabbit IgG, Cat# 15,015; Proteintech) were applied to detect bound primary antibodies at 1:10,000 dilution. The western blots were quantified by densitometric analysis with ImageJ software V1.53a.

Statistical analyses

The human and mice western blot data were expressed as mean ± standard deviation. The comparisons were conducted using the unpaired t-test. A two-tailed P-value < 0.05 was judged statistically significant. Statistical analyses and the graphs were performed using GraphPad Prism V8.4.3 software (Prism 8 for macOS).

Results

Identification of DEGs

The results of differential expression analysis showed that a total of 2437 genes, consisted of 1401 upregulated genes and 1072 downregulated genes, were identified as DEGs in GSE100927 (Fig. 2a and d). Meanwhile, a total of 944 genes, consisted of 567 upregulated genes and 377 downregulated genes, were obtained from GSE43292 (Fig. 2b and e). The distribution of these DEGs was presented in the volcano plots (Fig. 2a and b). The top 50 DEGs, including 29 upregulated genes and 21 downregulated genes in GSE100927, and 18 upregulated genes and 32 downregulated genes in GSE43292, were exhibited in the heatmaps (Fig. 2c and f). To identify the overlapping DEGs between the two datasets, we then conducted the Venn analysis. Venn diagrams displayed 361 overlapping upregulated genes and 250 overlapping downregulated genes associated with atherosclerosis (Fig. 2g, Table S2).

Fig. 2.

Identification of differentially expressed genes (DEGs). a, b Volcano plots of DEGs distribution in GSE100927 (a) and GSE43292 (b). Nodes in red represent upregulated genes, nodes in blue represent downregulated genes, and gray dots represent no significantly changed genes. c, f Heatmaps of DEGs in GSE100927 (c) and GSE43292 (f). Legend on the top right indicates the log fold change of the genes. Horizontal axis represents each sample, and the vertical axis represents each gene. Blue and red colors represent low and high expression values, respectively. d, e Number of DEGs in GSE100927 (d) and GSE43292 (e). g Venn diagram of DEGs from the two datasets

GO and KEGG pathway analysis of DEGs

Next, we conducted GO and KEGG pathway functional enrichment analysis of the overlapping DEGs. The results showed that the significantly enriched BP included neutrophil activation, neutrophil degranulation, neutrophil activation involved in immune response, neutrophil mediated immunity, and T cell activation (Fig. 3a). In the CC category, secretory granule membrane, membrane raft, membrane microdomain, external side of plasma membrane, and endocytic vesicle were the top 5 enriched items (Fig. 3b). As for MF, the most enriched terms were actin binding, endopeptidase activity, phosphoric ester hydrolase activity, actin filament binding, and carbohydrate binding (Fig. 3c). We also performed the GO functional enrichment analysis in upregulated and downregulated DEGs, respectively. The remarkably enriched BP, CC, and MF terms were presented in the Fig. 3d and e. Subsequently, the DEGs were subjected to KEGG pathway functional enrichment analysis. Cytokine-cytokine receptor interaction, chemokine signaling pathway, regulation of actin cytoskeleton, tuberculosis, and lipid and atherosclerosis were considered to be the most highly enriched pathways (Fig. 3f).

Fig. 3.

Gene ontology (GO) and kyoto encyclopedia of genes and genomes (KEGG) pathway enrichment analysis of differentially expressed genes (DEGs). a–c Bubble charts show GO-enriched items of DEGs in three functional groups: biological processes (BP, a), cell composition (CC, b), and molecular function (MF, c). The x-axis labels represent gene ratios, and y-axis labels represent GO terms. The size of circle represents gene count. Different colors of circles represent different adjusted P-values. d, e Chord plots show GO-enriched items of upregulated DEGs (d) and downregulated DEGs (e). Symbols of DEGs are presented on the left side of the graph with their fold change values mapped by color scale. Gene involvement in the GO terms is determined by colored connecting lines. f Circle plot shows KEGG-enriched items of DEGs. The red dots in the graph mean upregulated genes, and the blue dots mean downregulated genes. The height of the bar in the inner ring indicates the significance of the term, and color corresponds to the z-score

Screening of critical signatures

Because the differential expressions in GSE100927 were more pronounced than that in GSE 43292, we utilized GSE100927 for all subsequent analyses. Here, we carried out RF algorithm, LASSO logistic regression algorithm, and WGCNA analysis to screen the critical marker genes. The results showed that 140 genes were identified with RF algorithm (Fig. 4a–c, Table S3), and 12 genes were determined by LASSO logistic regression algorithm (Fig. 4d and e, Table S3). Moreover, using WGCNA analysis with the default-recommended parameters (Fig. 4f and g), we identified 12 remarkable co-expression modules (Fig. 4h). As indicated from the investigations of module-trait correlations, multiple modules were related to atherosclerosis (Fig. 4i). Given that the association of turquoise module and atherosclerosis was the most significant, genes in the turquoise module were screened for the subsequent exploration, and 608 genes were successfully identified (Fig. 4j, Table S3). Subsequently, Venn diagrams indicated that dehydrogenase/reductase 9 (DHRS9) and protein tyrosine phosphatase receptor type J (PTPRJ) were overlapping genes by the three algorithms, which were both upregulated genes (Fig. 4k, Table 1).

Fig. 4.

Screening of critical signatures via multiple machine-learning algorithms. a–c Identification of signatures by random forests (RF). Distribution of out-of-band (OOB) error rate at various values of mtry (a) and trees (b). Variable importance, as measured by the mean decrease in accuracy (left panel) or the Gini coefficient (right panel), is computed using the OOB error (c). Genes are shown in descending order of importance. d, e Establishment of signatures by least absolute shrinkage and selection operator (LASSO) logistic regression analysis. LASSO coefficient profile of the 12 genes (d), and different colors represent different genes. Selection of the optimal parameter (lambda) in the LASSO model, and generation of a coefficient profile plot (e). f–j Process of weighted gene co-expression network analysis (WGCNA). Analysis of network topology for various soft-thresholding powers (f, g). The x-axis reflects the soft-thresholding power. The y-axis reflects the scale-free topology model fit index (f) and the mean connectivity (g). Clustering dendrogram of differentially expressed genes related to atherosclerosis, with dissimilarity based on topological overlap, together with assigned module colors (h). Module–trait associations (i). Each row corresponds to a module, and each column corresponds to a trait. Each cell contains the corresponding correlation and P-value. The table is color-coded by correlation according to the color legend. The gene significance for atherosclerosis in the turquoise module, and one dot represents one gene in the turquoise module (j). k Venn diagram shows the intersection of critical signatures obtained by the three strategies

Table 1.

Information of RNA-binding proteins of DHRS9 and PTPRJ

Gene symbol	Log2FC	P-value/Adj.P.Val	Gene title	Involved function
DHRS9	1.887	2.74E-15/5.44E-13	dehydrogenase/reductase 9	Converting 3-alpha-tetrahydroprogesterone and 3-alpha-androstanediol to dihydroxyprogesterone. Involving in the biosynthesis of retinoic acid from retinaldehyde. Utilizing both NADH and NADPH
PTPRJ	0.823	1.83E-17/1.83E-14	protein tyrosine phosphatase receptor type J	Dephosphorylation. Involving in vascular development, cell adhesion, migration, proliferation and differentiation. Regulator of macrophage adhesion and spreading. Positive regulator of platelet activation and thrombosis, endothelial cell survival, as well as of VEGF-induced SRC and AKT activation. Negative regulator of cell proliferation, PDGF-stimulated cell migration, EGFR signaling pathway, and T-cell receptor signaling. Affecting cell–matrix adhesion. Enhancing the barrier function of epithelial junctions during reassembly

DHRS9, dehydrogenase/reductase 9; PTPRJ, protein tyrosine phosphatase receptor type J

Verification of DHRS9 and PTPRJ in atherosclerosis

In order to assess the potential predictive value of key gene markers in atherosclerosis, we generated ROC curves. The AUCs for DHRS9, PTPRJ, and combined were all 1 (Fig. 5a), suggesting that the two crucial genes had the high accuracy of predictive value. Next, we verified the expression of DHRS9 and PTPRJ in atherosclerotic arterial tissues. The results showed that both DHRS9 and PTPRJ expressions were upregulated in atherosclerosis when compared with controls in the GSE100927 and GSE43292 datasets (all P < 0.01, Fig. 5b and c). In the validation dataset of GSE28829, the expression of DHRS9 in advanced lesions was significantly higher than that in early lesions (P = 0.0048, Fig. 5d). Similarly, the expression of PTPRJ in advanced lesions was also higher than that in early lesions, but there was no statistically significance (P = 0.78, Fig. 5d).

Fig. 5.

a The diagnostic power of DHRS9, PTPRJ, and combined in atherosclerosis by ROC curve. b–d The expressions of DHRS9 and PTPRJ in GSE100927 (b), GSE43292 (c), and GSE28829 (d). DHRS9, dehydrogenase/reductase 9; PTPRJ, protein tyrosine phosphatase receptor type J

Construction of a multifactor regulatory network based on key signatures

To further explore their potential regulatory mechanism, we conducted and visualized an integrated analysis of lncRNA/miRNA/TF-gene regulatory network. Here, we first extracted interaction pairs of miRNAs, lncRNAs, and TFs with the 2 marker genes, and then constructed the multifactor regulatory network. From the network, DHRS9 was regulated by 3 miRNAs, including hsa-miR-1264, hsa-miR-499b-3p, and hsa-miR-6890-5p, and the potential TFs for DHRS9 were ATF3, BRD4, CEBPA, EP300, FLI1, FOSL1, FOXA2, IRF1, and JUN. Simultaneously, the PTPRJ network contained 5 lncRNAs, 77 miRNAs, and 63 TFs. Strikingly, our regulatory network showed that TFs IRF1 and JUN modulated both DHRS9 and PTPRJ (Fig. 6).

Fig. 6.

The multifactor regulatory network based on DHRS9 and PTPRJ. DHRS9, dehydrogenase /reductase 9; PTPRJ protein tyrosine phosphatase receptor type J

Analysis of immune cell infiltration

There is now substantial experimental and clinical evidence that immune mechanisms can accelerate atherosclerosis [24]. This promotes us to explore the relationship between the key signatures and the immune infiltration in atherosclerosis. We first applied PCA of the samples in GSE100927. The result showed that different group samples were well separated (Fig. 7a). Next, we performed CIBERSORT algorithm to analyze the 22 immune cell phenotypes in GSE100927. Five types of immune cells with undetectable abundance were excluded, and 17 types of immune cells were utilized to further analysis. As indicated from the correlation heatmap of the 17 types of immune cells (Fig. 7c), B cell memory and plasma cells, T cells CD4 memory resting and macrophages M0, monocytes and macrophages M0, macrophages M0 and macrophages M1, macrophages M0 and macrophages M2, macrophages M0 and mast cells resting, and mast cells resting and mast cell activated displayed significant negative correlations, respectively. T cells CD4 memory resting and macrophages M1, T cells CD4 memory resting and mast cells resting, mast cells resting and macrophages M1, and mast cells resting and macrophages M2 exhibited significant positive correlations, respectively. In comparison with control samples, atherosclerotic samples had a higher proportion of B cell memory, macrophages M0, mast cells activated, and T cell gamma delta (all P < 0.01). However, the proportions of macrophages M1, macrophages M2, mast cells resting, monocytes, plasma cells, T cells CD4 memory activated, T cells CD4 memory resting, and T cell CD8 in atherosclerosis were relatively lower than that in control (all P < 0.05) (Fig. 7b).

Fig. 7.

Immune cell infiltration analysis, and relationships between key signatures and immune cells in atherosclerosis. a Principal component analysis (PCA) cluster plot of gene expression profile between atherosclerotic samples and control samples in GSE100927. b Box-plot of the proportion of 17 types of immune cells. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns P ≥ 0.05. c Heatmap of correlation in 17 types of immune cells. The size of the colored squares represents the strength of the correlation; red represents a positive correlation, and blue represents a negative correlation. Darker color implies stronger association. d Correlations between DHRS9, PTPRJ, and infiltrating immune cells. e UMAP visualization of clustering revealing 16 cell clusters. f Violin plots show expression distribution of DHRS9 and PTPRJ mRNA in different cell clusters in atherosclerosis. Cluster identities: 0, T cells; 1, T cells; 2, endothelial cells; 3, smooth muscle cells; 4, macrophage; 5, monocyte; 6, chondrocytes; 7, monocyte; 8, B cell; 9, endothelial cells; 10, tissue stem cells; 11, smooth muscle cells; 12, B cell; 13, monocyte; 14, NK cell; 15, NA. DHRS9, dehydrogenase/reductase 9; PTPRJ, protein tyrosine phosphatase receptor type J; UMAP, Uniform Manifold Approximation and Projection

Correlation analysis between key signatures and infiltration-related immune cells

Here we sought to explore the relationships between key signatures and infiltrated immune cells in atherosclerosis. Based on the results of correlation analysis, both DHRS9 and PTPRJ displayed positive correlations with macrophages M0 (r = 0.61 and 0.53, both P < 0.05) and activated mast cells (r = 0.51 and 0.66, both P < 0.05) and showed negative correlations with mast cells resting (r =  − 0.69 and -0.71, both P < 0.05), T cells CD4 memory resting (r =  − 0.68 and -0.55, both P < 0.05), and macrophages M1 (r =  − 0.62 and − 0.54, both P < 0.05). Also, DHRS9 showed a negative correlation with plasma cells (r =  − 0.51, P < 0.05). (Fig. 7d).

Expression of DHRS9 and PTPRJ in single cells by scRNA-seq analysis

To more precisely delineate the expression of key signatures in immune cells, we interrogated a scRNA-seq database to identify the cell populations expressing DHRS9 and PTPRJ mRNA in atherosclerosis. After implementing quality control and filtering cells, we applied unbiased clustering of the cells based on their gene expression profiles. Clustering identified 16 subpopulations as shown in UMAP plot (Fig. 7e). DHRS9 mRNA expression was significantly upregulated in macrophages (clusters #4) and NK-cells (clusters #14) in atherosclerotic arterial tissue (both P < 0.001) (Fig. 7f). Although PTPRJ mRNA was detected in several subpopulations, there were no significantly difference between atherosclerosis and control (Fig. 7f). These data suggest that DHRS9 might be specifically upregulated in atherosclerotic lesions that are rich in macrophages.

Expression of DHRS9 in atherosclerotic patients and mice

For further support, we measured DHRS9 protein expression in human and mice atherosclerotic tissues. Immunostaining of human and mice atherosclerotic tissues revealed that DHRS9 was colocalized with CD68 positive macrophages (Fig. 8a). Immunoblotting showed that the expression of DHRS9 was significantly higher in atherosclerosis patients than that in control subjects (P = 0.044, Fig. 8b and c). Similarly, the expression of DHRS9 was significantly increased in WD-fed ApoE^–/– mice when compared to NCD-fed ApoE^–/– mice (P < 0.001, Fig. 8d and e). Taken together, the results further support that DHRS9, as a potential diagnostic marker, is localized in macrophages in atherosclerosis.

Fig. 8.

Expression of DHRS9 in human and mice atherosclerotic tissues. a Immunostaining for DHRS9 and CD68 in 3 atherosclerotic tissues from human lower extremity arterial (the upper row), and in 6 aortic roots from WD-fed ApoE.^–/– mice (the lower row). b–e Protein levels of DHRS9 in human atherosclerosis (b, c) and mice atherosclerosis (d, e) were assessed by Western blot analysis. DHRS9, dehydrogenase/reductase 9

Discussion

Early detection, prevention, and intervention of atherosclerosis are key to reducing the incidence and fatality rates, and decreasing enormous socio-economic burden of the ASCVD [51–53]. Screening the potential susceptibility genetic markers and revealing their underlying mechanism in atherosclerosis are considered to be the effective strategies for predictive diagnosis, targeted prevention, and personalized treatment of the disease. In the present study, our findings have established that DHRS9 and PTPRJ are the atherosclerosis’s critical signatures, and both of them exhibited good diagnostic performances for atherosclerosis. Furthermore, our results have suggested that DHRS9 might be specifically upregulated in atherosclerotic lesions that are rich in macrophages, which has been confirmed in atherosclerotic patients and mice. Collectively, this is the first study to report in vivo evidence that DHRS9 localized in macrophages can be considered as a novel genetic marker for atherosclerosis in the context of PPPM.

The advent of multi-omics technology has advanced our understanding of the molecular mechanisms of atherosclerosis in a profound manner, resulting in progressions in clinical diagnosis and treatment strategies adopted for atherosclerosis. In this study, we conducted integrative bioinformatics analysis to identify 611 DEGs between atherosclerosis and controls, covering 361 upregulated and 250 downregulated genes. Among the overlapping DEGs, functional enrichment analysis revealed that the enrichment of terms related to immune cell involvements, immune activating process, and inflaming signals. It is suggested that atherosclerosis displays tight associations with immunity and inflammation, which is in line with the current view [17, 54].

Importantly, we used three machine-learning strategies, WGCNA, RF coupled with LASSO, to screen and determine critical signatures of atherosclerosis. The advantage of the integrative procedures is to fit a model with consensus performance on the atherosclerosis based on multiple machine-learning algorithms and their combinations, as well as taking the intersection of markers from three algorithms combinations can further reduce the number of markers, thus enhancing the specificity and sensitivity of signatures. Ultimately, DHRS9 and PTPRJ were screened as the gene signatures, and furthermore, ROC analysis on diagnostic performance demonstrated that the elevated DHRS9 and PTPRJ could accurately distinguish atherosclerosis from non-atheroma individuals. Equally important, the expression of DHRS9 and PTPRJ maintained the stable performance in different validation datasets. These results indicated that DHRS9 and PTPRJ may be causative factors for atherosclerosis, suggesting their great potential for clinical applications in disease prediction, targeted prevention, and personalized therapeutics. Interestingly, in the multifactor regulatory network, both DHRS9 and PTPRJ were regulated by TFs IRF1 and JUN, which have been demonstrated to be related to atherosclerosis [55, 56]. Therefore, further research regarding DHRS9 and PTPRJ transcriptionally modulated by IRF1 and JUN is needed.

Given the important role of immunity in atherosclerosis, we sought to explore the relationship between gene signatures and immune cells in atherosclerosis. The results showed that both DHRS9 and PTPRJ displayed varying degrees of correlation with immune cells, such as macrophages and mast cells. Very importantly, DHRS9 mRNA expression was significantly upregulated in macrophages in atherosclerotic plaques through analyzing single-cell omics data; nevertheless, there was no significantly difference in PTPRJ expression between atherosclerosis and control. The results were further verified in atherosclerotic patients and mice. Herein, for the first time, we reported DHRS9 as a novel signature for atherosclerosis, and further elucidated that DHRS9 localized in macrophages in atherosclerosis. These data thus provide novel insights and immune mechanism of DHRS9 in the process of atherosclerosis, which will establish a foundation for targeted prevention, progression monitoring, prognostic assessment, and personalized medicine, and drive the progression of PPPM practice based on DHRS9-targeting strategy.

DHRS9 has been identified as a moonlighting protein in the short-chain dehydrogenases/reductases family. Current studies have shown that DHRS9 is associated with cicatricial alopecia [57], polycystic ovary syndrome [58], rheumatoid arthritis [59], and various human tumors including colorectal, pancreatic, and oral squamous cell carcinoma [60–62]. To the best of our knowledge, however, no previous study has reported the relationship between DHRS9 and atherosclerosis. Of note, DHRS9 might have the ability to metabolize oxylipins [63]. Furthermore, DHRS9 has been identified as a specific and stable marker of human regulatory macrophage [64], which represents a unique state of macrophage polarization [65]. On the basis of these publications, it is speculated that DHRS9 might relate to oxylipins and immunity, which are increasingly recognized as the crucial pathogenic mechanisms of atherosclerosis [21, 22, 54, 66–69]. Therefore, there is a certain level of rationale for our findings.

Early detection and diagnosis of atherosclerosis are particularly important for the prediction of this disease. Carotid artery intima medial thickness(c-IMT), ankle brachial index (ABI), and pulse wave velocity (PWV) are usually performed to predict early atherosclerotic changes [70, 71], but there are no satisfactory biomarkers suitable for screening and early diagnosis of atherosclerosis. According to our results, detecting the DHRS9 level with or without c-IMT, ABI, and PWV may assist clinicians to identify individuals at high risk of developing atherosclerosis and diagnose the early-stage atherosclerosis.

Primary prevention against atherosclerosis, which is crucial for the management of ASCVD, is dependent on the effective reduction of the cardiovascular risk factors, including tobacco control, regular physical exercise, a healthier diet, loss weight, prevention, and treatment of hypercholesterolemia, hypertension, and diabetes [72]. Currently, changes in the intestinal microbiome play an important role in both protection and development of atherosclerosis [73]. From the perspective of PPPM, pre and probiotics treatments can inhibit inflammation and have a great clinical potential for effective prevention of atherosclerosis [74]. As for the preventive medicine in PPPM, for normal population with elevated DHRS9, we should assess the comprehensive cardiovascular risk factors and aggressively control them to prevent or delay atherosclerosis onset and progression; for patients with high DHRS9 in atherosclerosis, timely initiation of the secondary and tertiary prevention is recommended to prevent the deteriorating condition.

Patient risk stratification and identification are key to personalized medicine of atherosclerosis patients. In the stratification of patients with atherosclerosis, obesity is one of the established stratification factors for cardiovascular diseases. Nevertheless, the underweight population has a greater risk of cardiovascular diseases than the normal-weight controls [75]. Consequently, the experts recommend that stratification factors related to low body weight need more attention [76]. In our study, we found that the expression of DHRS9 in advanced atherosclerosis was significantly higher than that in early atherosclerosis, implying higher DHRS9 might be associated with greater disease severity. We speculated that downregulation of DHRS9 may inhibit the initiation and progression of atherosclerosis, which would be the promising application of DHRS9 in risk stratification and disease monitoring. In the context of PPPM, quantification of DHRS9 expression may help match established and latent atherosclerosis patients to different personalized therapies. Additionally, DHRS9 is also a potential therapeutic target for atherosclerosis. Therefore, the elevated DHRS9 in atherosclerosis not only presents the onset of the disease that can be served as a genetic marker to guide predictive diagnostics and targeted prevention but also indicates the disease severity that can be recommended as a special target for patient stratification and personalized therapy. Overall, the results support that determinations of DHRS9 expression in atherosclerosis have important and profound implications for the application of PPPM to atherosclerosis.

Our study has several weaknesses that are worth mentioning. Firstly, the data used in our study were obtained from the public dataset, and the relevant clinical data were missing, so we could not make prognostic analysis. Secondly, the involved studies were the limited sample size for atherosclerosis, and the analysis failed to cover the impact of the cardiovascular risk factors on the overall data analysis and results, which might affect the gene expression in patients with atherosclerosis. In order to solve these shortages, we will perform further clinical studies with more detailed clinical information and larger sample sizes to confirm the research results. Lastly, this study preliminarily screened the key marker genes for atherosclerosis, and we need to further confirm these findings in vitro and in vivo studies, as well as conduct further research to clarify the pathological mechanism of DHRS9 underlying atherosclerosis onset, which may help DHRS9 to apply in the field of PPPM for atherosclerosis.

Conclusions and expert recommendations

In conclusion, our study is the first to reveal the involvement of DHRS9 in the atherogenesis, and the proatherogenic effect of DHRS9 is mediated by immune mechanism. Furthermore, we demonstrate that DHRS9 is localized in macrophages within atherosclerotic plaques. Therefore, the upregulation of DHRS9 could be a novel potential target for the future predictive diagnostics, patient stratification, targeted prevention, and personalization of medical services in atherosclerosis, which would promote the development of PPPM of atherosclerosis.

Expert recommendations and outlook within the framework of PPPM

Herein, from the viewpoint of PPPM, we highly recommend elevated DHRS9 as a genetic marker for predictive diagnostics and targeted prevention, as well as a special target for patient stratification and personalized therapy for atherosclerosis. Firstly, this study focused on the identifying of key biomarker genes of atherosclerosis based on the integrative bioinformatic approach and multiple machine-learning strategies. The results provide the scientific data to screen critical signatures and elucidate the related molecular mechanisms in atherosclerosis, which will be helpful for PPPM practice of atherosclerosis. Secondly, DHRS9 could be served as a potential predicative/therapeutic target for the initiation and progression of atherosclerosis. Actually, DHRS9 has been found to be involved in several diseases, but this is the first time to reveal that DHRS9 is associated with atherosclerosis. In order to evaluate the association of DHRS9 with the risk and development of atherosclerosis, we recommend more laboratory experiments and clinical trials to assess the dose–response relationship between the alteration of DHRS9 level and the occurrence/progression of atherosclerosis, which has great clinical application value for early diagnosis, risk stratification, and even prognostic assessment in atherosclerosis. In addition, future clinical observational research is also warranted to identify normal reference values of DHRS9 for health population and diagnostic threshold for atherosclerosis that can be utilized to optimize the strategy of PPPM for atherosclerosis. Thirdly, the pathogenesis of atherosclerosis caused by high DHRS9 is possibly mediated by diverse immune cells, particularly macrophages. However, the immune mechanism of DHRS9 remains unclear in atherosclerosis. The potential underlying mechanisms may be helpful for the targeted prevention and personalized medical services of atherosclerosis. Lastly, DHRS9-based therapeutic target and drug for atherosclerosis are a very promising approach, which might develop a new precise and effective therapeutic target/drug for PPPM practice in atherosclerosis. Moreover, since the patients with high DHRS9 are susceptible to the immunity dysregulation, we also recommend that further researches should investigate the customized immunotherapy and gene therapy to regulate or decrease the expression of DHRS9 in patients with atherosclerosis, providing greater possibilities for personalized therapy in the atherosclerosis PPPM context. Expert recommendations and outlook associated with DHRS9 within the framework of PPPM are summarized in Supplementary Fig. 1.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ABI

Ankle brachial index

ASCVD

Atherosclerotic cardiovascular diseases

AUC

Area under the curve

BP

Biological process

CC

Cellular component

CIBERSORT

Cell-type identification by estimating relative subsets of RNA transcript

c-IMT

Carotid artery intima medial thickness

DAPI

4’,6-Diamidino-2-phenylindole

DEGs

Differentially expressed genes

DHRS9

Dehydrogenase/reductase 9

GAPDH

Anti-glyceraldehyde-3-phosphate dehydrogenase

GEO

Gene Expression Omnibus

GO

Gene ontology

GSE

Gene Expression Omnibus series

Hcy

Homocysteine

KEGG

Kyoto encyclopedia of genes and genomes

LASSO

Least absolute shrinkage and selection operator

lncRNA

Long non-coding RNA

MF

Molecular function

miRNA

MicroRNA

NCBI

National Center of Biotechnology Information-GEO

NCD

Normal chow diet

PCA

Principal component analysis

PPPM

Predictive, preventive, and personalized medicine

PTPRJ

Protein tyrosine phosphatase receptor type J

PWV

Pulse wave velocity

RF

Random forests

ROC

Receiver operating characteristic

scRNA-seq

Single-cell RNA sequencing

TF

Transcription factor

UMAP

Uniform Manifold Approximation and Projection

WD

Western diet

WGCNA

Weighted gene co-expression network analysis

Author contribution

JX and HZ were responsible for data acquisition and analysis. YC conducted the human and mice experiments. JX drafted the manuscript. GX designed the study and revised the article. All the authors read and approved the final manuscript for submission.

Funding

This work was supported by the National Natural Science Foundation of China (Grant no: NSFC 82170857).

Data availability

The data involved in this study had been described in detail in the “Materials and methods.” The data analyzed during the human and mice experiments are available from the corresponding author on reasonable request.

Code availability

All software applications used are included in this article.

Declarations

Ethics approval

The protocol for collecting human tissue samples was approved by Ethics Committee of the General Hospital of Central Theater Command (approval no. [2021]004–02), and all procedures were performed in accordance with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Written informed consent was provided by all participants before their inclusion in the study. The animal experimental protocols were permitted by the Animal Ethics Committee of the General Hospital of Central Theater Command (approval no.2021013).

Consent to participate

All individuals were informed about the purposes of the study and have signed their consent for publishing the data.

Consent for publication

After reviewing the manuscript, all authors agreed with its publication in the current form.

Competing interests

The authors declare no competing interests.