Translate this page into:
Integrative multi-omics analysis of plasma pQTL prioritizes CELSR2 as a novel therapeutic target for atherosclerosis
*Corresponding author: E-mail address: coolzhangbin22@163.com (B. Zhang); liujunwen@csu.edu.cn (J. Liu)
-
Received: ,
Accepted: ,
Abstract
Atherosclerosis (AS) is a multifactorial disease and a leading cause of cardiovascular diseases, imposing a significant burden on global health. While there is considerable progress in elucidating its pathogenesis, the understanding of AS remains incomplete. Consequently, the identification of novel therapeutic targets is paramount to developing more comprehensive therapeutic interventions. We applied integrative approaches, including Summary-data-based Mendelian randomization (SMR), colocalization, and machine learning to prioritize causal genes. Experimental validation encompassed reverse transcription-real-time polymerase chain reaction (RT-qPCR), immunofluorescence, and single-cell RNA-seq in human plaques, complemented by ox-LDL-induced cellular models. Drug prediction and molecular docking were performed to assess therapeutic potential. A total of 15 relevant positive genes were identified by SMR, and the gene cadherin EGF LAG seven-pass G-type receptor 2(CELSR2) was verified as a key gene. Subsequent RT-qPCR and immunofluorescence of human carotid atherosclerotic plaques confirmed that CELSR2 showed different expression in the pathological state of AS. RT-qPCR combined with single-cell sequencing analysis led us to conclude that CELSR2 is a key target in the pathogenesis of AS. CELSR2 is an experimentally validated AS-related gene, which is important for unraveling the pathogenesis of AS and developing new therapeutic options.
Keywords
Atherosclerosis
CELSR2
Colocalization
Drug prediction
Machine learning
pQTL
1. Introduction
Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide (Herrington et al., 2016). Atherosclerosis (AS), the primary pathological basis of CVDs, is characterized by the deposition of lipids and cholesterol within the coronary arteries, leading to the formation of plaques. The progression of these plaques can severely restrict blood flow to the heart, potentially triggering angina or myocardial infarction (Libby et al., 2019). Although the incidence of AS has decreased in some countries, its incidence is increasing globally, especially in the elderly population.
Despite advances in lipid-lowering therapies like statins, current treatments fail to reverse advanced plaque progression and carry risks of adverse effects. This underscores the urgent need to identify novel therapeutic targets through innovative genetic and multi-omics approaches (Ferri et al., 2023).
To support the data acquisition and analytical workflows in this study, we utilized the following key public databases and tools:
IEU OpenGWAS (https://gwas.mrcieu.ac.uk): A repository of GWAS summary data, essential for genetic analyses.
SMR Portal (https://yanglab.westlake.edu.cn/smr-portal/): A tool for integrating GWAS and QTL data to identify genes with putative causal effects on traits.
NCBI (https://www.ncbi.nlm.nih.gov/): A central hub for biomedical data, including literature, sequences, and genetic variants, used for data retrieval and annotation.
Immune Single Cell (https://immunesinglecell.org/): A curated database for immune cell single-cell RNA-seq data, enabling the exploration of cell-type-specific gene expression.
Ma’ayan Lab Cloud (https://maayanlab.cloud): A platform for systems biology tools, such as Enrichr, used for pathway and gene set enrichment analysis.
AlphaFold (https://alphafold.com): An AI system providing highly accurate protein structure predictions, crucial for structural bioinformatics and functional inference.
CB-Dock2 (https://cadd.labshare.cn/cb-dock2/php/index.php): A server for automated protein-ligand docking and virtual screening, aiding in the prediction of binding interactions.
The integrated use of these resources was fundamental to the design and execution of our study.
Summary-data-based Mendelian randomization (SMR) extends this framework by integrating protein quantitative trait loci (pQTLs) to prioritize genes with putative causal links to disease. In this study, we systematically integrated SMR, colocalization analysis, and machine learning to identify AS-associated genes, and subsequently validated prioritized candidates through experimental assays in human plaques and functional studies. Our integrative approach identifies cadherin EGF LAG seven-pass G-type receptor 2 (CELSR2) as a key driver of AS pathogenesis, providing a compelling new target for therapeutic development.
2. Materials and Methods
2.1 Exposure data
The plasma protein quantitative trait locus (pQTL) data were obtained from the UK Biobank Pharma Proteomics Project (UKB-PPP; https://registry.opendata.aws/ukbppp/) (Sun et al., 2023). This collaboration between the UK Biobank and several biopharmaceutical companies performed comprehensive plasma proteomic profiling on 54,219 participants. Through high-resolution mapping, the study identified 14,287 significant genetic associations spanning 2,923 distinct proteins.
2.2 Outcome data
The outcome data were sourced from the IEU Open GWAS platform (https://gwas.mrcieu.ac.uk) (Lyon et al., 2021), (Elsworth et al., 2020), (Hemani et al., 2018), utilizing two distinct datasets. The first, identified as ukb-d-I9_CORATHER (coronary atherosclerosis), comprised 14,334 cases and 346,860 controls. The second, ieu-a-7 (coronary heart disease), included 60,801 cases alongside 184,305 control subjects.
2.3 SMR analysis
The SMR analysis was conducted using the SMR Portal (Zhu et al., 2016). To pinpoint genes associated with complex traits and diseases, the SMR and HEIDI methods were applied to summary-level data, with the HEIDI test specifically evaluating heterogeneity due to linkage disequilibrium. It aims to facilitate the discovery of genes for complex traits, elucidate the regulatory mechanisms behind GWAS signaling, and improve the research community’s access to these gene‒trait associations.
2.4 Machine learning
The gene expression profile was from the GEO dataset GSE104140, normalized by FPKM with duplicate genes deleted. The samples with only endometrial thickening were set into the CON group, and the other samples with significant lesions, such as fibrotic plaques, were set into the AS group. The gene list was the 15 genes that were significant in the SMR analysis with the AS outcome. Lasso, Elasticnet, and Ridge algorithms were selected, and biowinford (http://biowinford.site:3838/trial/) was used for feature selection and classifier analysis, respectively.
2.5 Colocalization analysis
For the 15 genes identified in the SMR analysis, we used biowinford (http://biowinford.site:3838/trial/) for colocalization analysis. With pQTLs as the exposure and ukb-d-19_CORATHER as the outcome, 15 genes were input into the sequence for analysis, and PP.H4 was compared. If PP.H4 > 0.75, colocalization between this gene and the outcome was considered.
2.6 Protein interaction network construction
15 significant genes obtained from SMR were imported into the String database, and the default parameters were used (Szklarczyk et al., 2023). Following visualization of the PPI results in Cytoscape (V3.10.2) (Shannon et al., 2003), the top 10 hub genes were selected using Cytohubba to map the core interaction network (Chin et al., 2014). Additionally, the GeneMANIA (https://genemania.org/) tool was employed to conduct a supplementary PPI analysis (Warde-Farley et al., 2010).
2.7 Single-cell RNA-seq analysis of the plagues
Data on CELSR2 expression in the vasculature were obtained from the Human Protein Atlas (http://www.proteinatlas.org/). The single-cell sequencing dataset GSE260657 was derived from the GEO database. We used R 4.4.1 to convert the expression matrix for each case into the RDS format, followed by the integration of asymptomatic patients (n=7) and symptomatic patients (n=8) via DISCO (https://immunesinglecell.org/) (Li et al., 2022), analysis and visualization via ezSinglecell (https://www.immunesinglecell.org/ezsc/) (Sethi et al., 2024).
2.8 Cell culture
Human coronary smooth muscle cells (HCASMCs; Sigma‒Aldrich, USA) were donated by Nanjing Medical University and cultured in complete culture medium (Pricella, Wuhan, China). Based on published research, we selected appropriate concentrations of ox-LDL and induction times, and appropriately expanded the scope to ensure the reliability of the results (Z. Wang et al., 2025; Zhu and Yan, 2025; Li et al., 2015). When cells reached 70–80% confluence, they were first incubated in DMEM (Gibco, Massachusetts, USA) for 10 h. The medium was then replaced with complete culture medium containing 100 μg/mL ox-LDL (Yiyuanbiotech, Guangzhou, China) for 24, 48, or 72 h.
When culturing THP-1 cells (Zhong Qiao Xin Zhou Biotechnology, Shanghai, China), we used 1640 medium (Gibco, Massachusetts, USA) containing 15% fetal bovine serum (cell mixture, Changsha, China) and 1% penicillin‒streptomycin. THP-1 cells were stimulated with 50 nM phorbol myristate acetate (PMA) for 24 h to induce macrophages (M. Wang et al., 2025). After THP-1 cells were differentiated into macrophages, the culture medium was substituted for RPMI-1640 containing 10% fetal bovine serum. Subsequently, the cells were exposed to 80 μg/mL ox-LDL for varying durations (24, 48, or 72 h).
2.9 RT‒qPCR
Total RNA was isolated from both cellular and tissue samples using TRIzol reagent (Genview, Florida, USA). cDNA synthesis was performed with a reverse transcription kit (Promega, Wisconsin, USA), followed by quantitative polymerase chain reaction (qPCR) executed using a corresponding kit (Promega, Wisconsin, USA). The sequences of all gene-specific primers have been listed in Table S1.
2.10 H&E staining
Human carotid atherosclerotic plaques were cut into sections after paraffin embedding, dewaxed, and hydrated with xylene, ethanol, and ultrapure water. Paraffin sections of atherosclerotic plaques were subjected to hematoxylin and eosin (H&E) staining for visualization via an inverted microscope (Leica DM IL LED, Wetzlar, Germany).
2.11 Immunofluorescence
The treated sections were treated with sodium citrate antigen repair solution (Proteintech, Wuhan, China) for microwave antigen repair. After repair, the sections were sealed with goat serum (Bioss, Beijing, China) and incubated overnight with 15 μg/mL primary antibody (R&D Systems, AF6739, Minnesota, USA). After the primary antibody incubation at 4°C overnight, the samples underwent three PBS washes. Subsequent steps included a 1-h incubation with fluorescent secondary antibodies and a 5-min nuclear staining with DAPI. Following a final PBS wash, fluorescence images were acquired using a fluorescence microscope (Zeiss Axio Imager Z2, Oberkochen, Germany).
2.12 Candidate drug prediction
The Enrichr platform (Chen et al., 2013), (Kuleshov et al., 2016), (Xie et al., 2021) is used to predict potential drugs or molecular compounds that target CELSR2, an accessible resource that collects and integrates vast amounts of information about gene function. The discovery of possible targeted drugs based on the Drug Characterization Database (DSigDB) (Yoo et al., 2015) can be achieved.
2.13 Molecular docking
The structure of CELSR2 was retrieved from the AlphaFold database (Jumper et al., 2021; Varadi et al., 2023), while the ligand structures were obtained from PubChem (https://pubchem.ncbi.nlm.nih.gov). Molecular docking was carried out on the CB-Dock 2 server (Y. Liu et al., 2022; Yang et al., 2022).
2.14 Statistical analysis
Statistical significance between two groups was determined using a two-tailed Student’s t-test on normally distributed data with GraphPad Prism (v10.3.0), defining a p-value < 0.05 as significant.
3. Results
3.1 Identification of AS-associated genes via SMR analysis
Using pQTL data from the UK Biobank (UKB-PPP) and AS summary statistics from the IEU Open GWAS, SMR analysis identified 15 genes significantly associated with AS (adjusted p < 0.05), including CELSR2, apolipoprotein E (ApoE), Proprotein convertase subtilisin/kexin type 9 (PCSK9), and so on (Fig. 1a). PECAM1 was excluded because its mr keep is false (Table S2).
- Identification and Analysis of AS-Related Genes. (a) Manhattan plot of SMR analysis in the discovery phase. (b) GO analysis of 15 significant genes. The main components of GO analysis are immunity, receptors, etc. (c) KEGG analysis of 15 significant genes. The most notable pathways were the pathways related to cancer and amoebiasis, both of which are related to the immune response and fit with the results of the GO analysis. (d) PPI network of top 10 genes. CELSR2 interacts with PCSK9 and APOE, both of which are well-characterized genes with established links to AS.
The enrichment analysis revealed a significant involvement of these genes in several biological processes, including those related to blood microparticles and cancer pathways (Figs. 1b and c).
CELSR2 interacts with PCSK9 and APOE (Fig. 1d), both of which are well-characterized genes with established links to AS (Wang et al., 2023; Ilyas et al., 2022).
3.2 Colocalization analysis
Colocalization analysis can elucidate whether exposures and outcomes share the same causal SNPs (Zhang et al., 2025), especially when SNPs are explicitly associated with both the exposures and outcomes. Therefore, we conducted pQTL-based colocalization analysis on the 15 genes identified in our preceding SMR analysis. As detailed in Table 1, eight of these proteins demonstrated robust evidence of colocalization with AS (Table S3), with a posterior probability of the presence of horizontal pleiotropy (PP.H4) exceeding 0.75.
| Table of colocalization analysis | |
|---|---|
| Gene | PP.H4 |
| CELSR2 | 1.000 |
| APOE | 1.000 |
| LPA | 1.000 |
| PCSK9 | 1.000 |
| FES | 1.000 |
| COL4A1 | 0.910 |
| FGF5 | 0.932 |
| VAMPS | 0.822 |
3.3 Machine learning prioritizes CELSR2 as a key candidate
We employed three machine learning algorithms, Lasso, ElasticNet, and Ridge, to refine gene selection. This rigorous feature selection process pinpointed CELSR2 as the most consistently significant gene across all models (Fig. 2). Classifier analysis further confirmed its high importance, with ROC curves demonstrating robust predictive accuracy (AUC = 1 for Lasso/Elasticnet, 0.938 for Ridge) (Fig. 3).
- Feature selection of candidate genes for machine learning. (a) When lambda was set at 2.01 in the Elasticnet algorithm, the model demonstrated the highest goodness of fit. (b) Six genes showed strong correlations with the disease. (c) When lambda was set at 2.11 in the Lasso algorithm, the model demonstrated the highest goodness of fit. (d) Four genes showed strong correlations with the disease. (e) When lambda was set at 0.7 in the Ridge algorithm, the model showed the highest goodness of fit (f) Lambda=0.7, 15 genes showed strong correlations with the disease.
- Classifier of candidate genes for machine learning. (a) Lasso confusion (b) Lasso ROC (c) Lasso importance calculation (d) Elasticnet confusion (e) Elasticnet ROC (f) Elasticnet importance calculation (g) Ridge confusion (h) Ridge ROC (i) Ridge importance calculation.
3.4 External verification
A total of 11 genes demonstrated significance in the SMR analysis with coronary heart disease (CHD) as the outcome (Fig. 4a, Table S4). Of these, seven genes exhibited a consistent effect in the analysis with AS as the outcome (Fig. 4b). Subsequent enrichment analysis revealed that these seven overlapping genes were strongly implicated in lipid and lipoprotein metabolism (Figs. 4c-e).
- External validation of AS-related genes. (a) External verification. (b) A total of seven genes were found to be significantly associated with both AS and CHD outcomes simultaneously. (c) GO analysis of seven co-significant genes. (d) The most significant GO components (e) Reatcome enrichment analysis of seven co-significant genes.
3.5 Experimental validation in human plaques and cellular models
Fibrous caps, foam cells and other structures often exist in atherosclerotic plaques (Fig. 5a). Reverse transcription-real-time polymerase chain reaction (RT-qPCR) and immunofluorescence revealed significantly elevated CELSR2 expression in human carotid atherosclerotic plaques compared to adjacent normal tissue (Figs. 5b and c). Single-cell RNA-seq analysis of vascular cells highlighted CELSR2 enrichment in vascular smooth muscle cells (VSMCs) and macrophages (Figs. 6a-c). In vitro, ox-LDL treatment induced dynamic CELSR2 expression changes: downregulation in VSMCs and a biphasic response (initial suppression followed by upregulation) in macrophages (Figs. 6d-i).
- Experimental validation of atherosclerotic plaques. (a) HE staining of atherosclerotic plaques. Fibrous caps are observed around the periphery (bottom), and foam cells are observed in the core of the plaque (top). (b) RT‒qPCR results of CELSR2 in plaques. CELSR2 was significantly upregulated in plaque tissues. (c) Immunofluorescence in paraffin sections of atherosclerotic plaques. Nuclei were labelled with DAPI (blue), and CELSR2 (red) expression was lower in peripheral areas with mild lesions and significantly increased in areas with more severe lesions. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001
- Expression analysis of CELSR2 at the cellular level. (a) CELSR2 expression in normal vascular tissues obtained by the Human Protein Atlas. Mainly concentrated on VSMCs and macrophages. (b) In the plaques of asymptomatic patients, various cell types were identified, including macrophages, VSMCs, endothelial cells, and monocytes. Among these, CELSR2 was found to be highly expressed in macrophages and VSMCs. (c) In the plaques of symptomatic patients, various cell types were identified, including macrophages, VSMCs, endothelial cells, and pericytes. Among these, CELSR2 was found to be highly expressed in macrophages, VSMCs, and pericytes. (d-f) The expression of CELSR2 in VSMCs after induction with 100 μg/mL ox-LDL for 24, 48, and 72 h. (g-i) The expression of CELSR2 in macrophages after induction with 80 μg/ml ox-LDL for 24, 48, and 72 h.*p<0.05, **p<0.01, ***p<0.001, ****p<0.0001
3.6 Functional insights and drug prediction
PPI networks linked CELSR2 to Wnt signaling and lipid metabolism pathways (Fig. 7a).
- Prediction of the Mechanism of Action and Drug Prediction for CELSR2. (a) Construction and Function Prediction of PPI Network Centered on CELSR2. The canonical and non-canonical Wnt pathways, as well as cell and tissue polarization, play a dominant role. These functions are closely related to VSMCs and macrophages. (b) Molecular docking of verteporfin with CELSR2 (c) Molecular docking of emetine with CELSR2 (d) Molecular docking of mehp with CELSR2.
We interrogated the DSigDB database to identify potential therapeutic compounds. Following filtration based on adjusted p-values (p < 0.05), a total of 8 statistically significant candidate drugs were identified (Table 2). To ensure safety, agents with documented substantial toxicity or carcinogenicity were excluded. Based on this screening, mehp, verteporfin, and emetine were selected for subsequent evaluation.
| Drug prediction results table | |||
|---|---|---|---|
| Term | P value | Adjusted P value | Gene |
| lndeno[l,2,3-cd]pyrene CTD 00001895 | 0.002 | 0.022 | CELSR2 |
| Mehp CTD 00000849 | 0.003 | 0.022 | CELSR2 |
| Benz[a]anthracene CTD 00001470 | 0.003 | 0.022 | CELSR2 |
| Benzo[k]fluoranthene CTD 00001069 | 0.004 | 0.022 | CELSR2 |
| cephaeline MCF7 DOWN | 0.010 | 0.039 | CELSR2 |
| verteporfin MCF7 DOWN | 0.012 | 0.039 | CELSR2 |
| emetine MCF7 DOWN | 0.012 | 0.039 | CELSR2 |
| piperlongumine MCF7 DOWN | 0.013 | 0.039 | CELSR2 |
Through molecular docking, we identified emetine as the most potent binder, with a Vina score of -7.4, suggesting its potential as a novel drug candidate for the treatment of AS (Figs. 7b-d).
4. Discussion
Leveraging pQTLs, this investigation pinpointed 15 potential AS targets, including CELSR2, FES, APOE, and LPA. The application of the SMR method effectively ruled out measurement confounding, and the colocalization results offered additional validation. Furthermore, the combination of diverse machine learning algorithms not only clarified the associations between the candidate genes and AS but also precisely ranked their importance. Impressively, CELSR2 exhibited a significant association with AS in every algorithmic model, emerging as the most stable and critical gene.
To substantiate the clinical relevance of the identified genes, we performed external validation using an independent CHD dataset. We then employed enrichment analyses and PPI networks to explore the biological functions of CELSR2. Experimental studies in human plaque samples and cellular models further corroborated these bioinformatic findings. Finally, we performed molecular docking to identify potential drugs targeting CELSR2, underscoring its therapeutic potential.
In recent studies, it has been proposed that AS is a tumor-like lesion driven by VSMCs; thus, we hypothesized that CELSR2 may be involved in some of the mechanisms shared by AS and tumors (Pan et al., 2024). CELSR2 interacts with PCSK9 and APOE, both of which are well-characterized genes with established links to AS. PCSK9 is known to regulate lipid metabolism by promoting the degradation of the LDL receptor on hepatocytes, and recent studies suggest that it may also function as an inflammatory modulator in AS (Wang et al., 2023). The role of APOE in AS has been more extensively documented, with APOE knockout mice being a prevalent model for AS research globally. APOE, encoding an apolipoprotein, is intricately associated with lipid metabolism (Ilyas et al., 2022). Meanwhile, we found that CELSR2 has a high functional correlation with the FZD family, which is a member of both the canonical and non-canonical Wnt pathways (Hayat et al., 2022). These pathways are closely related to both tumors and AS (J. Liu et al., 2022; Akoumianakis et al., 2022). This once again indicates that CELSR2 may affect the tumor-like transformation of VSMCs and thereby influence AS through this way.
In human atherosclerotic plaques, both RT‒qPCR and immunofluorescence analyses revealed significantly high expression levels of CELSR2. However, a contrasting phenomenon was observed in VSMCs, which presented the highest expression of CELSR2 in the vascular system. Upon induction with ox-LDL, CELSR2 expression was significantly downregulated in these cells. This paradox has sparked our interest.
We hypothesize that the diverse cell types in AS plaques interact within distinct microenvironments, potentially leading to the high expression of CELSR2 in these plaques being associated with the activation status of specific cell subsets. ox-LDL may induce different signaling pathways in VSMCs, resulting in altered CELSR2 expression. This discrepancy may reflect the biological differences between the complex microenvironments in vivo and single-cell models in vitro. Additionally, ox-LDL can activate inflammation-related signaling pathways, such as the NF-κB and MAPK pathways (Fang et al., 2021), which may exhibit varying activation states in AS plaques and VSMCs models, thereby influencing CELSR2 expression. Furthermore, gene expression is intricately regulated through a network involving transcription factors and epigenetic modifications. The observed dysregulation of CELSR2 may be attributed to context-dependent differences in these regulatory mechanisms when comparing complex AS plaques to cultured VSMCs.
The expression dynamics of CELSR2 in macrophages exhibited a more intricate pattern, characterized by an initial decrease, followed by an increase, and ultimately a decline. Macrophage polarization yields two main subtypes: pro-inflammatory M1 macrophages with roles in inflammatory responses, and anti-inflammatory M2 macrophages responsible for tissue repair and anti-inflammatory functions. The fluctuating expression of CELSR2 may mirror the response of macrophages to ox-LDL stimulation at various stages, as well as their transition between the M1 and M2 phenotypes (Eshghjoo et al., 2022). This observation aligns with our previous gene function prediction analysis, which suggested that CELSR2 influences cell polarization.
CELSR2 deficiency can cause intracellular lipid accumulation, inhibit cell proliferation, and promote cell apoptosis (Tan et al., 2021). We suspect that the significant downregulation of CELSR2 in VSMCs may cause a similar effect. First, lipid accumulation in VSMCs accelerates the process of AS. Second, as an important component of the fibrous cap of atherosclerotic plaques, the apoptosis of VSMCs may lead to instability of the fibrous cap and further lead to serious consequences of plaque rupture (Childs et al., 2021). Based on the machine learning prediction that CELSR2 promotes AS progression, its observed overexpression in plaques, and the insights from enrichment and PPI analyses, we hypothesize that CELSR2 exacerbates AS by modulating macrophage polarization and lipid endocytosis via the canonical and non-canonical Wnt signaling pathways. The paradoxical CELSR2 expression dynamics in VSMCs (downregulation) versus macrophages (biphasic response) may reflect distinct roles in plaque stability and inflammation. In VSMCs, reduced CELSR2 could exacerbate lipid accumulation and apoptosis, destabilizing fibrous caps. In macrophages, temporal regulation may modulate polarization states (M1/M2) or lipid uptake.
Our study provides the first experimental evidence confirming the differential expression of CELSR2 under pathological conditions, thereby offering more direct and compelling evidence. The integration of single-cell sequencing analysis and cellular experiments has laid a solid foundation for future research directions. To ensure that the observed cellular responses in VSMCs and THP-1-derived macrophages are comparable and reflect coordinated disease mechanisms, we normalized all data to baseline controls and applied standardized assays. Our pathway analysis further substantiates this approach, revealing that both cell types, despite their distinct CELSR2 expression dynamics, engage common pathways critically involved in CAD pathophysiology, such as Wnt signaling and lipid metabolism.
The final drug prediction phase underscored the therapeutic potential of the genes under investigation, and the high binding affinity observed in molecular docking studies indicates the promising potential of these genes as drug targets.
5. Conclusions
By integrating multi-omics data with experimental validation, we pinpoint CELSR2 as a central player in AS pathogenesis. Its dysregulation in plaques, cell-type-specific functions, and confirmed druggability converge to nominate CELSR2 as a promising therapeutic target. This integrative framework not only elucidates a key AS driver but also provides a blueprint for translating genetic findings into tangible clinical applications.
Acknowledgement
We are most grateful to all those who contributed to this study, including all the public data collectors, the creators of the database, the inventors of the analytical methods, the patients who contributed samples to scientific research, all funders who supported this research and all the researchers around the world who are working to combat cardiovascular diseases.
CRediT authorship contribution statement
Haotian Tang: Conceptualization and study design; data collection; experimental research and analysis; writing of the original draft. Zhenrui Liu: Study design; data collection; analysis and visualization; equal contribution to the work. Bin Zhang: Study conceptualization and design; guidance and supervision throughout the study. Junwen Liu: Study conceptualization and design; overall guidance and inspection. Ruofan Zhao: Data collection; analysis and visualization. Bingxin Qi: Collection of human tissue samples. Yanqing Huang: Collection of human tissue samples. Xunliang Cheng: Collection of human tissue samples. Didi Yuan: Experimental research and analysis. All authors read and approved the final manuscript.
Declaration of competing interest
The authors declare that they have no competing financial interests or personal relationships that could have influenced the work presented in this paper.
Data availability
All the datasets mentioned in the text can be downloaded from public databases. The relevant database and dataset numbers have been indicated. If you have any other data query requirements, please contact the corresponding authors.
Declaration of generative AI and AI-assisted technologies in the writing process
The authors confirm that there was no use of artificial intelligence (AI)-assisted technology for assisting in the writing or editing of the manuscript and no images were manipulated using AI.
Funding
This research was funded by grants from the Hunan Province Natural Science Foundation, No. 2022JJ30780. The Education Reform Project of Hunan Province, No. HNJG-2022-04591 Postgraduate Research and Innovation Project of Central South University, 1053320242086.
Supplementary data
Supplementary material to this article can be found online at https://dx.doi.org/10.25259/JKSUS_1072_2025.
References
- Non-canonical WNT signalling in cardiovascular disease: Mechanisms and therapeutic implications. Nat Rev Cardiol. 2022;19:783-797. https://doi.org/10.1038/s41569-022-00718-5
- [Google Scholar]
- Enrichr: Interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinformatics. 2013;14:128. https://doi.org/10.1186/1471-2105-14-128
- [Google Scholar]
- Senescent cells suppress innate smooth muscle cell repair functions in atherosclerosis. Nat Aging. 2021;1:698-714. https://doi.org/10.1038/s43587-021-00089-5
- [Google Scholar]
- cytoHubba: Identifying hub objects and sub-networks from complex interactome. BMC Syst Biol 8 Suppl. 2014;4:S11. https://doi.org/10.1186/1752-0509-8-S4-S11
- [Google Scholar]
- The MRC IEU OpenGWAS data infrastructure. 2020 https://doi.org/10.1101/2020.08.10.244293
- [Google Scholar]
- Macrophage polarization in atherosclerosis. Genes (Basel). 2022;13:756. https://doi.org/10.3390/genes13050756
- [Google Scholar]
- IRGM/Irgm1 facilitates macrophage apoptosis through ROS generation and MAPK signal transduction: Irgm1+/- mice display increases atherosclerotic plaque stability. Theranostics. 2021;11:9358-9375. https://doi.org/10.7150/thno.62797
- [Google Scholar]
- Fixed combination for the treatment of dyslipidaemia. Curr Atheroscler Rep. 2023;25:691-699. https://doi.org/10.1007/s11883-023-01142-x
- [Google Scholar]
- AlphaFold protein structure database and 3D-Beacons: New data and capabilities. J Mol Biol. 2025;437:168967. https://doi.org/10.1016/j.jmb.2025.168967
- [Google Scholar]
- Wnt signaling pathway: A comprehensive review. Cell Biol Int. 2022;46:863-877. https://doi.org/10.1002/cbin.11797
- [Google Scholar]
- The MR-Base platform supports systematic causal inference across the human phenome. Elife. 2018;7:e34408. https://doi.org/10.7554/eLife.34408
- [Google Scholar]
- Epidemiology of atherosclerosis and the potential to reduce the global burden of atherothrombotic disease. Circ Res. 2016;118:535-546. https://doi.org/10.1161/CIRCRESAHA.115.307611
- [Google Scholar]
- Mouse models of atherosclerosis in translational research. Trends Pharmacol Sci. 2022;43:920-939. https://doi.org/10.1016/j.tips.2022.06.009
- [Google Scholar]
- Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596:583-589. https://doi.org/10.1038/s41586-021-03819-2
- [Google Scholar]
- Enrichr: A comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res. 2016;44:W90-W97. https://doi.org/10.1093/nar/gkw377
- [Google Scholar]
- Oxidized low-density lipoprotein inhibits THP-1-derived macrophage autophagy via TET2 down-regulation. Lipids. 2015;50:177-183. https://doi.org/10.1007/s11745-014-3977-5
- [Google Scholar]
- DISCO: A database of deeply Integrated human single-cell omics data. Nucleic Acids Res. 2022;50:D596-D602. https://doi.org/10.1093/nar/gkab1020
- [Google Scholar]
- Atherosclerosis. Nat Rev Dis Primers. 2019;5:56. https://doi.org/10.1038/s41572-019-0106-z
- [Google Scholar]
- Wnt/β-catenin signalling: Function, biological mechanisms, and therapeutic opportunities. Sig Transduct Target Ther. 2022;7 https://doi.org/10.1038/s41392-021-00762-6
- [Google Scholar]
- CB-Dock2: Improved protein–ligand blind docking by integrating cavity detection, docking and homologous template fitting. Nucleic Acids Res. 2022;50:W159-W164. https://doi.org/10.1093/nar/gkac394
- [Google Scholar]
- The variant call format provides efficient and robust storage of GWAS summary statistics. Genome Biol. 2021;22:32. https://doi.org/10.1186/s13059-020-02248-0
- [Google Scholar]
- Atherosclerosis is a smooth muscle cell-driven tumor-like disease. Circulation. 2024;149:1885-1898. https://doi.org/10.1161/CIRCULATIONAHA.123.067587
- [Google Scholar]
- ezSingleCell: An integrated one-stop single-cell and spatial omics analysis platform for bench scientists. Nat Commun. 2024;15:5600. https://doi.org/10.1038/s41467-024-48188-2
- [Google Scholar]
- Cytoscape: A software environment for integrated models of biomolecular interaction networks. Genome Res. 2003;13:2498-2504. https://doi.org/10.1101/gr.1239303
- [Google Scholar]
- Plasma proteomic associations with genetics and health in the UK Biobank. Nature. 2023;622:329-338. https://doi.org/10.1038/s41586-023-06592-6
- [Google Scholar]
- The STRING database in 2023: Protein-protein association networks and functional enrichment analyses for any sequenced genome of interest. Nucleic Acids Res. 2023;51:D638-D646. https://doi.org/10.1093/nar/gkac1000
- [Google Scholar]
- CELSR2 deficiency suppresses lipid accumulation in hepatocyte by impairing the UPR and elevating ROS level. FASEB J. 2021;35:e21908. https://doi.org/10.1096/fj.202100786RR
- [Google Scholar]
- AlphaFold protein structure database in 2024: providing structure coverage for over 214 million protein sequences. Nucleic Acids Res. 2024;52:D368-D375. https://doi.org/10.1093/nar/gkad1011
- [Google Scholar]
- METTL14 knockdown augmented the polarization of M2-like macrophages to promote acute myeloid leukemia progression. Hum Cell. 2025;38:168. https://doi.org/10.1007/s13577-025-01300-0
- [Google Scholar]
- Interactions between PCSK9 and NLRP3 inflammasome signaling in atherosclerosis. Front Immunol. 2023;14:1126823. https://doi.org/10.3389/fimmu.2023.1126823
- [Google Scholar]
- The clinical significance of miR-139-5p in acute coronary syndrome and its potential effect on the progression of acute coronary syndrome. J Cardiovasc Transl Res. 2025;18:983-992. https://doi.org/10.1007/s12265-025-10625-y
- [Google Scholar]
- The GeneMANIA prediction server: Biological network integration for gene prioritization and predicting gene function. Nucleic Acids Res. 2010;38:W214-W220. https://doi.org/10.1093/nar/gkq537
- [Google Scholar]
- Gene set knowledge discovery with Enrichr. Curr Protoc. 2021;1:e90. https://doi.org/10.1002/cpz1.90
- [Google Scholar]
- FitDock: Protein–ligand docking by template fitting. Brief Bioinform. 2022;23 https://doi.org/10.1093/bib/bbac087
- [Google Scholar]
- DSigDB: Drug signatures database for gene set analysis. Bioinformatics. 2015;31:3069-3071. https://doi.org/10.1093/bioinformatics/btv313
- [Google Scholar]
- Exploring the role of B-cell biotherapy in myasthenia gravis from a genetic perspective: A mendelian randomization study. Mol. Neurobiol.. 2025;63:115. https://doi.org/10.1007/s12035-025-05440-0
- [Google Scholar]
- LncRNA MAGI2-AS3 promotes the progression of atherosclerosis by sponging miR-525-5p. J Cardiothorac Surg. 2025;20:206. https://doi.org/10.1186/s13019-025-03451-5
- [Google Scholar]
- Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat Genet. 2016;48:481-487. https://doi.org/10.1038/ng.3538
- [Google Scholar]