Key genes and pathways of pulmonary hypertension identified by integrating bulk and single-cell RNA sequencing Authors: Peipei Wu, Jingru Huan, Chengliang Liu, Haiyang Jin, Ran Wang, Yong Wang, Cheng Cao, Binbin Zhang Published: 2026 ABSTRACT SummaryPulmonary hypertension (PH) is a progressive vascular disease, yet the underlying cell-type-specific molecular networks and diagnostic biomarkers remain incompletely defined. Integrating single-cell and bulk transcriptomics, we identified 14 pulmonary cell types in PH, with smooth muscle cells (SMCs) as a central communication hub interacting with monocytes and macrophages. High-dimensional weighted gene co-expression network analysis (hdWGCNA) revealed 4 SMC modules enriched in extracellular matrix organization, TGF-β, and HIF-1 pathways. A five-gene diagnostic signature distinguished PH from controls with an area under the curve of 0.895. Immune infiltration and enrichment analysis showed that activated CD4 memory T cells, activated dendritic cells, and resting mast cells increased in the PH group, while angiogenesis and epithelial-mesenchymal transition pathway were upregulated. In hypoxic human pulmonary artery SMCs, milk fat globule-epidermal growth factor-8 (MFGE8), secreted frizzled-related protein 2 (SFRP2), S100 calcium-binding protein A11 (S100A11), and procollagen C-endopeptidase enhancer 2 (PCOLCE2) were upregulated, and their knockdown suppressed proliferation, highlighting them as potential biomarkers and drivers of vascular remodeling, revealing a new molecular mechanism for PH and providing insights for potential therapeutic targets.Keywords: pulmonary hypertension, single-cell RNA sequencing, bioinformatics analysis, immune infiltration Introduction Pulmonary hypertension (PH) is a severe disease characterized by a progressive increase in pulmonary vascular resistance.1 It is hemodynamically defined as a mean pulmonary arterial pressure >20 mmHg at rest measured by right heart catheterization.2,3 Epidemiological data indicate that the annual incidence of PH is approximately 5 cases per million, and it affects individuals of all ages.4 The clinical manifestations are progressive, starting with exertional dyspnea, fatigue, and palpitations and eventually advancing to right heart failure, edema, syncope, and even sudden death.3 Despite the clinical application of drugs, such as prostacyclin analogs, endothelin receptor antagonists, and phosphodiesterase-5 inhibitors, which have improved symptoms to some extent, the long-term prognosis remains poor.5 Therefore, elucidating the molecular mechanisms underlying PH and identifying novel therapeutic targets are of great clinical significance. The pathogenesis of PH is exceedingly complex, involving endothelial dysfunction, aberrant proliferation of smooth muscle cells (SMCs), fibroblast activation, inflammatory responses, metabolic reprogramming, and genetic susceptibility.6 These pathological processes converge on pulmonary vascular remodeling, leading to lumen obliteration, increased right ventricular afterload, and right heart failure.7 However, the key molecules driving vascular remodeling and their cellular origins remain insufficiently understood, and single-target interventions have failed to reverse disease progression, highlighting the need to integrate multidimensional information at the systems biology level to dissect core regulatory networks. In recent years, the rapid advancement of high-throughput transcriptomics has enabled genome-wide dissection of disease mechanisms at the transcriptional level. Conventional bulk RNA sequencing (RNA-seq) reflects the average gene expression of mixed tissue samples and has successfully identified a series of differentially expressed genes (DEGs) and pathways in PH.8 However, because pulmonary vascular tissue comprises multiple cell types, including endothelial cells, SMCs, and infiltrating immune cells, bulk RNA-seq cannot resolve cell-type-specific transcriptional signatures. Single-cell RNA-seq (scRNA-seq) can delineate cellular heterogeneity at single-cell resolution and reveal dynamic changes and functional states of distinct cell subpopulations in PH, but it is often constrained by small sample sizes and limited sequencing depth.9 Therefore, integrating bulk and scRNA-seq data can not only select robust disease-associated genes from the global transcriptomic landscape but also anchor them to specific cell types, thereby identifying key genes and pathways of PH with greater precision. Based on this rationale, in the present study, we integrated PH-related bulk RNA-seq and scRNA-seq data from GEO datasets. By combining differential expression analysis, high-dimensional weighted gene co-expression network analysis (hdWGCNA), and machine learning-based feature selection algorithms, we identified key genes involved in the pathogenesis and progression of PH. We further performed functional enrichment analysis, immune infiltration analysis, and in vitro experimental validation, aiming to provide novel theoretical and experimental foundations for molecular diagnosis and targeted therapy of PH. Results Single-cell RNA data processing and cell annotation The single-cell sequencing data of GEO: GSE210248 were analyzed by R software. After excluding cells that do not meet the inclusion requirements, Figure 1A shows the features, counts, and percentage of mitochondrial genes of each sample. A total of 19,582 genes were identified, and the top 1,500 highly variable genes were identified (Figure 1B). Subsequently, the first 20 principal components (PCs) were identified by principal component analysis (PCA) on the expression matrix of highly variable genes (Figure 1C). The unsupervised clustering method based on uniform manifold approximation and projection (UMAP) dimensionality reduction technology was used to divide the cells into 14 cell clusters. Further cell annotation labeled these 14 cell clusters as 14 cell types, including SMCs, CD8+ T cells, macrophages, monocytes, natural killer (NK) cells, adipocytes, T cells, endothelial cells, dendritic cells, fibroblasts, epithelial cells, hematopoietic stem cells, B cells, and CD4+ T cells. Single-cell RNA sequencing analysis (A) The features, counts, and mitochondrial gene percentage of each sample. (B) Identification of highly variable genes. (C) PC selection using JackStraw function. (D) UMAP visualization of 14 cell clusters. Cell-cell communication analysis To investigate intercellular interactions, we performed cell-cell communication analysis using the CellChat tool. Figures 2A and 2B display the number of interactions and interaction strength, respectively, suggesting that SMCs play a prominent role. Figures 2C–2E show the communications of endothelial cells, epithelial cells, and SMCs with other cell types, where endothelial cells exhibited the strongest interactions with T cells, epithelial cells with macrophages, and SMCs with dendritic cells. Bubble plots further illustrate the ligand-receptor pairs mediating the interplay between SMCs and distinct cell populations, among which Annexin A1 (ANXA1)-formyl peptide receptor 1 (FPR1) mediates SMC-monocyte communication and complement component 3 (C3)-complement C3a receptor 1 (C3AR1) mediates SMC-macrophage communication (Figure 2F). Results of cell-cell communication analysis (A) Number of interactions in cell communication. (B) Interaction strength of cell communication. (C–E) Communication between endothelial cells, epithelial cells, and smooth muscle cells and other types of cells. (F) Bubble plot of interaction between smooth muscle cells and different cells. hdWGCNA and enrichment analysis Given that pulmonary artery SMCs (PASMCs) play a critical role in PH, we performed hdWGCNA on SMCs to delineate the gene co-expression networks underlying their biology. The soft threshold was optimized to 5 to ensure a scale-free network (Figure 3A). Hierarchical clustering of genes based on the topological overlap matrix (TOM) partitioned the transcriptome into eight distinct co-expression modules (Figure 3B). The top 10 key genes of each module are shown in Figure 3C. Among them, modules M2, M6, M7, and M8 displayed significant and specific expression enrichment in the SMC population and were therefore identified as SMC-associated modules (Figures 3D and 4A). Subsequently, the interaction networks of key genes in these 4 modules were constructed (Figures 4B–4E), the interactions among the top 3 key genes across all eight modules were examined (Figure 4F), and the spatial relationships of the top 10 key genes in each module were visualized on UMAP (Figure 4G). S100A10, ANXA2, RPL13A, and RPL13 were identified as key nodes within these modules. Gene Ontology (GO) analysis of the 4 specific module gene sets revealed enrichment in key biological process terms such as extracellular matrix (ECM) organization, extracellular structure organization, and external encapsulating structure organization. Significantly enriched cellular component terms included interstitial matrix, cytoplasmic vesicle lumen, and secretory granule lumen, while molecular function terms predominantly comprised ECM structural constituent, peptidase regulator activity, and glycosaminoglycan binding (Figure 4H). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis uncovered significant enrichment in the TGF-β signaling pathway, amino sugar and nucleotide sugar metabolism, and the HIF-1 signaling pathway (Figure 4I). These findings underscore a pivotal role of SMCs in the pathogenesis of PH. Results of hdWGCNA (A) hdWGCNA soft threshold selection plots. (B) Cluster dendrogram for hdWGCNA. (C) Gene enrichment levels of different modules and the top 10 genes were listed according to kME values in each module. (D) The enrichment atlas of each module in UMAP plots. Results of enrichment analysis (A) Correlation between smooth muscle cells and different modules. (B–E) Interaction level of key genes in related modules of smooth muscle cells. (F) Interaction between the Top 3 core genes in 8 modules. (G) Spatial relationship of the top 10 core genes in each module in UMAP plots. (H) GO enrichment analysis of 4 specific module gene sets in PH. (I) KEGG enrichment analysis of 4 specific module gene sets in PH. Identification of DEGs and their diagnostic value After batch effect correction of GEO: GSE113439 (15 PH vs. 11 controls), GEO: GSE53408 (12 PH vs. 11 controls), and GEO: GSE117261 (58 PH vs. 25 controls), DEGs were identified using |logFC| > 0.585 and adjusted p value < 0.05. The heatmap in Figure 5A displays the top 40 DEGs, and the accompanying boxplots present their expression levels in the PH and control groups (Figure 5B). The area under the curve (AUC) was calculated for these DEGs, and an AUC > 0.7 was considered indicative of diagnostic value for PH. Figures 5C–5G present representative receiver operating characteristic (ROC) curves, among which MFGE8 achieved an AUC of 0.81, demonstrating favorable diagnostic performance. A total of 19 genes with AUC >0.7 were obtained. To further ascertain their diagnostic utility, multiple machine learning models were constructed. The partial least squares (PLS) model yielded the highest AUC of 0.895 (Figure 5H) and was therefore selected for downstream analyses. Figure 5I shows the disease-characteristic key genes identified by the PLS model, with complement component 1 (C1R), milk fat globule-epidermal growth factor-8 (MFGE8), S100 calcium-binding protein A11 (S100A11), procollagen C-endopeptidase enhancer 2 (PCOLCE2), and secreted frizzled-related protein 2 (SFRP2) ranking as the top five, and Figure 5J maps their chromosomal locations. In the external validation cohorts (GSE15197 and GSE24988), the AUC of the PLS model was 0.756 (95% confidence interval [CI]: 0.583–0.930) and 0.712 (95% CI: 0.612–0.812), respectively, which confirmed the diagnostic utility of these genes in different groups (Figures 5K and 5L). Identification of DEGs and their diagnostic value (A) Heatmap displaying the top 40 DEGs. (B) Difference of DEG expression between PH group (n = 85, n represents the number of human participants) and control group (n = 47, n represents the number of human participants). Data are presented as box-and-whisker plots. The center line indicates the median, the box indicates the interquartile range (IQR; 25th–75th percentiles), and the whiskers extend to the smallest and largest values within 1.5×IQR. p values are based on Mann-Whitney U test. (C–G) ROC curve of different genes on PH. (H and I) ROC curves and hub genes of different machine learning modes. (J) Chromosome position information map of 5 hub genes. (K and L) ROC curves of the external validation cohorts. ∗p < 0.05, as compared with control group; ∗∗p < 0.01, as compared with control group; ∗∗∗p < 0.001, as compared with control group. Immune infiltration analysis The immune microenvironment plays a critical role in the pathogenesis and progression of PH. Boxplots and heatmaps were generated to visualize immune cell infiltration patterns. In the control group, the levels of resting NK cells, monocytes, and neutrophils were elevated, whereas resting CD4 memory T cells, activated CD4 memory T cells, activated dendritic cells, and resting mast cells were highly enriched in PH samples (Figures 6A and 6B). Furthermore, S100A11 and PCOLCE2 showed significant positive correlations with macrophages M0. SFRP2 was positively correlated with activated CD4 memory T cells and plasma cells. MFGE8 demonstrated significant correlations with monocytes, aligning with its reported function as a bridging molecule in phagocytosis and immune regulation. C1R showed relatively weaker but notable correlations with several immune subsets (Figure 6C). Immune infiltration analysis and ssGSEA (A) Differential distribution of immune cell types between PH group (n = 85, n represents the number of human participants) and control group (n = 47, n represents the number of human participants). Data are presented as box-and-whisker plots. The center line indicates the median, the box indicates the IQR (25th–75th percentiles), and the whiskers extend to the smallest and largest values within 1.5×IQR. p values are based on Mann-Whitney U test. (B) Correlation analysis of immune cells in PH cohort. (C) The associations between the 5 hub genes and immune infiltration levels. (D) Difference of HALLMARK pathway between PH group (n = 85, n represents the number of human participants) and control group (n = 47, n represents the number of human participants). Data are presented as box-and-whisker plots. The center line indicates the median, the box indicates the IQR (25th–75th percentiles), and the whiskers extend to the smallest and largest values within 1.5×IQR. p values are based on Mann-Whitney U test. (E) Correlation between 5 hub genes and HALLMARK pathway. ∗p < 0.05, as compared with control group; ∗∗p < 0.01, as compared with control group; ∗∗∗p < 0.001, as compared with control group. Single-sample gene set enrichment analysis To dissect the core signaling pathways associated with PH, we systematically analyzed correlations between the 5 hub genes and pathways by integrating the MSigDB database. Analysis of the HALLMARK pathway collection revealed that angiogenesis and epithelial-mesenchymal transition pathways were significantly upregulated in the PH group (Figure 6D). Furthermore, C1R was predominantly associated with xenobiotic metabolism and glycolysis pathways; SFRP2 with epithelial-mesenchymal transition and coagulation; MFGE8 with myogenesis and pancreas beta cells; S100A11 with the reactive oxygen species (ROS) pathway and IL-6/JAK-STAT3 signaling; and PCOLCE2 with oxidative phosphorylation (Figure 6E). Expression level of 5 hub genes in human PASMCs To validate our findings, we cultured human PASMCs (hPASMCs) under normoxic (control) or hypoxic (PH) conditions. Reverse transcription quantitative real-time polymerase chain reaction (RT-qPCR) analysis showed that, compared with the control group, the expression of MFGE8, SFRP2, S100A11, and PCOLCE2 was upregulated in the PH group, whereas C1R exhibited no statistically significant difference between the two groups (Figure 7A), a result partially consistent with the bioinformatics analysis. These observations were subsequently confirmed by western blotting (WB) (Figure 7B). Immunofluorescence staining further revealed that MFGE8, SFRP2, S100A11, and PCOLCE2 were upregulated in PH and predominantly localized to the cytoplasm (Figure 7C). RT-qPCR data showed the mRNA expression levels of MFGE8, SFRP2, S100A11, and PCOLCE2 after small interfering RNA (siRNA) transfection under hypoxic conditions, which demonstrated substantial reduction for each target gene (Figure 7D). The cell counting kit-8 (CCK-8) proliferation assay indicated that knockdown of MFGE8, SFRP2, S100A11, or PCOLCE2 inhibited the proliferation of hPASMCs (Figure 7E). The 5-ethynyl-2′-deoxyuridine (EdU) assay results confirmed that knockdown of MFGE8, SFRP2, S100A11, and PCOLCE2 significantly reduced the percentage of EdU-positive proliferating hPASMCs under hypoxic conditions, consistent with our CCK-8 findings (Figures 7F and 7G). To explore the correlation between our hub genes and the hypoxia-inducible factor 1-alpha (HIF-1α) pathway, we analyzed the correlation between the expression levels of the four validated hub genes (MFGE8, SFRP2, S100A11, and PCOLCE2) and HIF-1α target gene signatures in our bulk transcriptomic datasets. The results showed that SFRP2 and S100A11 were positively correlated with HIF-1α (Figure 7H). WB analysis suggested that knockdown of S100A11 and SFRP2 partially reduced HIF-1α protein expression (Figures 7I–7K), and overexpression of HIF-1α could partially reverse the proliferation inhibition caused by knockdown of S100A11 and SFRP2 (Figures 7L and 7M). Collectively, these results suggest that MFGE8, SFRP2, S100A11, and PCOLCE2 may play important roles in the development and progression of PH. Expression level of 5 hub genes in hPASMCs (A) mRNA expression of hub genes in each group. (B) Protein expression of hub genes in each group. (C) Immunofluorescence staining of hub genes in each group. Scale bars: 100 μm. (D) mRNA expression of hub genes after transfection with siRNA. (E) CCK-8 assays for hub genes in each group. (F and G) EdU analysis of hPASMCs in each group. Scale bars: 100 μm. (H) Correlation between hub genes and HIF-1α. (I) Protein level of HIF-1α in each group. (J and K) mRNA level of HIF-1α in each group. (L and M) CCK-8 assays of hPASMCs in each group. n = 3, n represents biological replicates. Data are represented as mean ± SEM. p values are based on Mann-Whitney U test. ∗p < 0.05, as compared with control group or hypoxia+si-NC group or hypoxia group. #p < 0.05, as compared with hypoxia+si-SFRP2 group or hypoxia+si-S100A11 group. Discussion In the present study, we integrated scRNA-seq with bulk RNA-seq datasets to comprehensively characterize the cellular and molecular landscapes of PH. Our single-cell profiling identified 14 cell types within the pulmonary microenvironment, among which SMCs emerged as a central hub of intercellular communication. Cell-cell interaction analysis revealed that SMCs engage in robust crosstalk with dendritic cells, monocytes, and macrophages, predominantly through ligand-receptor pairs such as ANXA1-FPR1 and C3-C3AR1. These findings align with growing evidence that immune-stromal paracrine loops drive pulmonary vascular remodeling. The increase of plasma ANXA1 has been proven to be related to the decrease of PH survival rate, and it has been identified as a potential prognostic marker of PH.10 At the same time, in the development of metabolic dysfunction-associated steatohepatitis, ANXA1 derived from neutrophils was found to increase the expression of S100A11 in neutrophils and macrophages by activating FPR1 signaling.11 C3, situated at the intersection of the classical, lectin, and alternative complement pathways, integrates systemic immune signals with microenvironmental stimuli to coordinate endothelial activation, smooth muscle phenotype switching, immune cell recruitment, platelet activation, and fibroinflammatory remodeling.12 In macrophages, C3a binds C3AR to activate NF-κB signaling, ROS generation, and calcium influx, resulting in assembly of the NLR family pyrin domain-containing 3 inflammasome and caspase-1-dependent maturation of IL-1β and IL-18.13,14 These cytokines enhance sterile perivascular inflammation and promote PASMC proliferation by inducing PCNA and cyclin D expression.15 Our data therefore highlight potential immune signals that promote PASMC dysfunction in PH. Using hdWGCNA, we delineated gene co-expression networks specific to SMCs and identified four modules (M2, M6, M7, and M8) that are significantly enriched in ECM organization, TGF-β signaling, and HIF-1 signaling. These pathways are well-established drivers of pulmonary vascular pathology. Overactive TGF-β/BMP imbalance promotes the proliferation of PASMCs.16 Concurrently, stabilization of HIF-1α under hypoxia transcriptionally upregulates numerous genes involved in proliferation, glycolysis, and ECM remodeling.17,18 The enrichment of terms such as “extracellular matrix structural constituent” and “interstitial matrix” further underscores the importance of matrix-driven signaling in pathological remodeling. Among the identified hub genes, pharmacological inhibition or genetic deletion of ANXA2 has been shown to confer effective protection against PH in rodents, indicating a pathogenic role for ANXA2 in the development of PH.19 Furthermore, S100A11 was found to mediate hypoxia-induced mitogenic factor-induced SMC migration, vesicular exocytosis, and nuclear activation.20 RPL13A and RPL13, as ribosomal proteins, have been implicated in translational control of specific mRNA subsets, potentially providing an additional layer of regulation in the hyperproliferative SMC phenotype, although their precise functions in PH warrant further investigation. By integrating three independent transcriptomic cohorts and applying a PLS-based machine learning model, we established a five-gene diagnostic signature (C1R, SFRP2, MFGE8, S100A11, and PCOLCE2) that discriminated PH from controls with an AUC of 0.895 in the training set. In vitro, hypoxia significantly upregulated MFGE8, SFRP2, S100A11, and PCOLCE2 in hPASMCs, and siRNA-mediated knockdown of each of these genes attenuated cell proliferation. MFGE8 is a secreted glycoprotein that binds integrins and has been reported to promote vascular SMC proliferation and migration in an integrin-dependent manner, supporting its potential role in medial thickening.21 SFRP2, often viewed as a modulator of the Wnt pathway, can either inhibit or potentiate Wnt signaling in a context-specific manner; elevated SFRP2 has been documented in pulmonary arteries from PH patients and may contribute to aberrant angiogenic and proliferative responses.22,23 S100A11 is a calcium-binding protein that activates RAGE and triggers ROS generation and IL-6/JAK-STAT3 pathways,24 a finding that matches our single-sample gene set enrichment analysis (ssGSEA) results showing S100A11 linked to ROS and IL-6/JAK-STAT3 signaling. PCOLCE2, a procollagen C-endopeptidase enhancer, facilitates collagen fibril maturation and could thus promote matrix stiffening, a hallmark of advanced vascular remodeling.25 C1R, although identified as a top-weighted gene in the diagnostic model, did not show differential expression under acute hypoxia in cultured hPASMCs. This may be caused by the following reasons: (1) C1R may be predominantly expressed by other cell types within the pulmonary vascular niche, such as immune cells or endothelial cells, rather than SMCs. Our scRNA-seq data from GSE210248 indicate that C1R expression is not strictly confined to SMCs, and its transcriptional regulation may be more prominent in immune compartments; (2) the 24-h acute hypoxia exposure in our in vitro model may not recapitulate the chronic, progressive nature of PH pathology. C1R might be involved in later stages of vascular remodeling or in response to sustained inflammatory stimuli that are not captured by short-term hypoxic stress; (3) given that C1R encodes a complement component, its expression may be more sensitive to inflammatory signals rather than direct hypoxic stimulation.26 The monoculture system lacks the complex immune-stromal crosstalk present in vivo, which could be essential for C1R upregulation. Nevertheless, its contribution to the multivariate model might capture aspects of systemic immune dysregulation that are not replicated in monocultures. The immune infiltration analysis revealed distinct immune landscape shifts in PH, characterized by enrichment of activated CD4 memory T cells, activated dendritic cells, and resting mast cells, alongside a relative reduction in NK cells, monocytes, and neutrophils. This pattern indicates an adaptive immune activation coupled with chronic inflammation, which is a recognized feature of PH pathobiology.27 Further ssGSEA confirmed the upregulation of angiogenesis and epithelial-mesenchymal transition signatures in PH samples, both of which are closely linked to endothelial dysfunction and smooth muscle hypertrophy. The observed associations of the five key genes with pathways such as glycolysis, oxidative phosphorylation, and ROS emphasize the metabolic reprogramming that sustains the hyperproliferative phenotype of PASMCs.28 Our integrative transcriptomic analysis identifies SMCs as key orchestrators of intercellular communication in PH and reveals MFGE8, SFRP2, S100A11, and PCOLCE2 as novel proliferative factors and potential diagnostic biomarkers. These findings reinforce the concept of smooth muscle-driven vascular remodeling and may inform the development of targeted therapeutic interventions in PH. Limitations of the study This study has several limitations. The analyses are based on publicly available datasets and, despite batch correction, require validation in independent prospective cohorts. Subclustering of SMCs was not pursued due to the limited number of cells within the SMC subset, which precluded statistically reliable subtype delineation. The in vitro experiments were limited to hPASMCs under acute hypoxia, and the functional significance of these genes in complex in vivo models, including their impact on hemodynamics and right ventricular function, remains to be determined. The mechanistic link between C1R and PH pathology and the precise signaling pathways through which MFGE8, SFRP2, S100A11, and PCOLCE2 regulate PASMC proliferation require more detailed exploration. Resource availability Lead contact Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Binbin Zhang (binbinzhang2025@fy.ahmu.edu.cn). Materials availability No new materials were generated by this work. Data and code availability •Data from publicly archived datasets are available from the GEO database (GSE210248, GSE113439, GSE53408, and GSE117261): https://www.ncbi.nlm.nih.gov/geo/. All data reported in this study will be shared by the lead contact upon request. Data from publicly archived datasets are available from the GEO database (GSE210248, GSE113439, GSE53408, and GSE117261): https://www.ncbi.nlm.nih.gov/geo/. All data reported in this study will be shared by the lead contact upon request. •This paper does not report original code. This paper does not report original code. •Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request. Acknowledgments This research was supported by the fund for Anhui Province Clinical Medical Research Translation Special Program (202304295107020045), Health Research Program of Anhui (AHWJ2024Bac30073), Excellence Young and Middle-aged Physician Training Program, the Scientific Research Cultivation Fund of the North District of the First Affiliated Hospital of Anhui Medical University (2024YKJ04), and the Medical Backbone Talent Project of the North District of the First Affiliated Hospital of Anhui Medical University. Author contributions Conceptualization, P.W., J.H., and C.L.; methodology, P.W., J.H., and C.L.; investigation, H.J. and R.W.; writing – original draft, P.W., J.H., and C.L.; writing – review and editing, P.W., J.H., Y.W., C.C., and B.Z.; funding acquisition, C.C.; resources, C.C. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesC1RBioworldBS8518; RRID: AB_2720761MFGE8BioworldBS8327; RRID: AB_2855366S100A11BioworldBS78280; RRID: AB_2852178PCOLCE2AbmartPHL8632; RRID: AB_2867378SFRP2BioworldBS7759; RRID: AB_2766193HIF-1αBioworldBS3514; RRID: AB_2927791GAPDHServicebioGB11002; RRID: AB_2769572Chemicals, peptides, and recombinant proteins4% paraformaldehydeServicebioG11013% H2O2Beyotime BiotechnologyP0100ACritical commercial assaysRNA Quick Purification KitGenuine400–105Reverse Transcription KitYeasen11150ES60CCK-8 Assay KitServicebioG4103Click-iT EdU-555 Cell Proliferation Assay KitServicebioG1602Deposited dataSingle-cell RNA DataGEOGSE210248Microarray DataGEOGSE113439Microarray DataGEOGSE53408Microarray DataGEOGSE117261Experimental models: Cell linesHuman Pulmonary Artery Smooth Muscle CellsATCCPCS-100–023OligonucleotidesC1R primer: forward 5′- TTCCCCAAGCCTTACCCCAA-3′ and reverse 5′-GCTGGAAGACGAGCTTCACC-3′This studyN.A.MFGE8 primer: forward 5′- TCTGTGCGTGTGACCTTCTTG-3′ and reverse 5′-CCAGGGGTTATCGTCATTGCT-3′This studyN.A.S100A11 primer: forward 5′-CTGAGCGGTGCATCGAGTC-3′ and reverse 5′-TGTGAAGGCAGCTAGTTCTGTA-3′This studyN.A.PCOLCE2 primer: forward 5′-TGCCGCTATGACTTTGTGGAT-3′ and reverse 5′-GGCATCAGAAATCATCTGCACC-3′This studyN.A.SFRP2 primer: forward 5′-ACGGCATCGAATACCAGAACA-3′ and reverse 5′-CTCGTCTAGGTCATCGAGGCA-3′This studyN.A.HIF-1α primer: forward 5′- GAACGTCGAAAAGAAAAGTCTCG-3′ and reverse 5′-CCTTATCAAGATGCGAACTCACA-3′This studyN.A.GAPDH primer: forward 5′-AGGTCGGTGTGAACGGATTTG-3′ and reverse 5′-TGTAGACCATGTAGTTGAGGTCA-3′This studyN.A.Software and algorithmsRThe R FoundationN.A.ImageJNIHN.A. Experimental model and study participant details hPASMCs were obtained commercially from the American Type Culture Collection (ATCC, Manassas, VA, USA). hPASMCs were authenticated as a smooth muscle cell line by immunofluorescence staining for α-smooth muscle actin. The cell line was tested for mycoplasma contamination, and no contamination was detected. Cells were maintained in Smooth Muscle Cell Medium (SMCM, ScienCell, CA, USA) supplemented with 10% fetal bovine serum, 1% smooth muscle cell growth supplement, and 1% penicillin-streptomycin antibiotic solution. Normoxic cultures were incubated at 37°C in a humidified atmosphere containing 21% O2 and 5% CO2. For the PH group, cells were exposed to 1% O2, 5% CO2, and 94% N2 for 24 hours. According to the manufacturer’s instructions, the specific siRNA (GenePharma, China) for C1R, MFGE8, S100A11, PCOLCE2 and SFRP2 was used to knock down the target genes in the cells. The plasmid for HIF-1α (GenePharma, Shanghai, China) were used to overexpress target gene in cells. Sex and gender were not considered to influence the study results. Method details Single-cell RNA data processing and cell annotation Single cell sequencing data of PH patients (3 cases) and healthy controls (3 cases) were obtained from GEO dataset GSE210248. After removing genes detected in 3 or fewer cells and cells expressing 200 or fewer genes, cells with less than 15% mitochondrial genes were retained for downstream analysis. Seurat package (Version 5.5.0) was used for quality control.29 Gene expression data were normalized using the “LogNormalize” method and subsequently scaled. The top 1500 highly variable genes were identified with the FindVariableFeatures function and the variance-stabilizing transformation method. PCA was performed to determine significant principal components, and the JackStraw and ScoreJackStraw functions were used to visualize the corresponding p-values. Dimensionality reduction was carried out via UMAP based on 20 PCs, followed by clustering. The SingleR package (Version 2.13.5) was employed for cell annotation. Cell-cell communication analysis Cell-cell communication networks were inferred, analyzed, and visualized using the CellChat package (Version 1.6.1).30 During initial preprocessing, ligand-receptor mapping was established via the “identifyOverExpressedGenes”, “identifyOverExpressedInteractions”, and “projectData” functions. Communication probabilities were then calculated with “computeCommunProb”, and interactions involving fewer than 10 cells were filtered out. Pathway-level interactions were integrated using “computeCommunProbPathway”. For network construction, “aggregateNet” was applied to generate the global network. Cell-cell communication was subsequently examined by iterating through individual cells to observe their interactions with others, and the ligand-receptor results were presented as bubble plots. hdWGCNA Given the critical role of PASMCs in the pathogenesis of PH, we applied the hdWGCNA package (Version 0.4.11) to perform weighted gene co-expression network analysis on smooth muscle cells.31 Genes detected in at least 5% of cells were selected using the fraction parameter. Metacells were constructed with the MetacellsByGroups function, and the data were subsequently normalized. A soft-thresholding power of 5 was chosen as optimal, and gene co-expression networks were built using the constructNetwork function, which generated the topological overlap matrix (TOM). A hierarchical clustering dendrogram was then plotted to visualize the module structure. The ModuleConnectivity function was used to calculate the module characteristic genes and analyze the module correlation degree (kME). The associations between the identified modules and cell populations were visualized with DotPlot. Furthermore, hub gene networks for all modules were plotted via HubGeneNetworkPlot, and the distribution of module eigengenes on the UMAP embedding was displayed. Enrichment analysis Genes within the four modules (M2, M6, M7, and M8) most strongly associated with smooth muscle cells were used for downstream functional enrichment analysis. GO and KEGG enrichment analyses were conducted using the clusterProfiler package (Version 4.19.8),32 and the results were visualized with the enrichplot package (Version 1.31.5). A p-value less than 0.05 was considered statistically significant. Identification of DEGs and their diagnostic value Three microarray datasets (GSE113439, GSE53408, and GSE117261) were selected, and the raw expression data from each dataset were normalized using the quantile normalization method implemented in the limma package (Version 3.67.3) to minimize potential batch effects.33 We applied the “removeBatchEffect” algorithm implemented in the limma package (Version 3.67.3) in R to correct for batch effects across the three GEO datasets. Following batch correction, PCA was performed to confirm the effective removal of batch effects and the proper integration of the datasets. The expression of the genes most strongly associated with smooth muscle cells was examined across the three GEO datasets, and the DEGs were visualized. Then, the ROC curve of DEGs was drawn by the pROC package (Version 1.19.0), and the genes with AUC>0.7 were screened out. A p-value <0.05 was considered statistically significant. Identification characteristic genes of PH by machine learning method Multiple machine learning models were constructed using the 19 genes with an AUC >0.7, with the data split into training and test sets at a 7:3 ratio. ROC curves were plotted for the 19 genes using the pROC package (Version 1.19.0). The top 5 genes (hub genes) from the machine learning method that achieved the highest AUC were visualized using the circlize package. Two microarray datasets (GSE15197 and GSE24988) were used to verify the accuracy of the machine learning method. Immune infiltration analysis The abundance of 22 immune cell types in PH and control samples was estimated using the CIBERSORT algorithm. The comparison of immune cell distribution between PH patients and controls was visualized with box plots. Correlations among immune cell populations were examined, and the potential associations between the 5 hub genes and immune infiltration levels were explored. ssGSEA ssGSEA evaluates the enrichment of gene sets in individual samples. The Hallmark gene set collection comprises 50 independent gene sets, each associated with a distinct biological process or pathway. Differences in Hallmark gene set enrichment between the PH and control groups were assessed using the GSVA package (Version 2.5.40),34 and the potential associations between the 5 hub genes and the Hallmark gene sets were explored. RT-qPCR Total RNA was isolated from treated cells with an RNA Quick Purification Kit and reverse transcribed to cDNA using a specific reverse transcription kit following the manufacturer’s protocol.35 GAPDH was used as the internal control. WB Total protein was isolated from cells, and the protein concentration was measured with a BCA protein assay kit (Beyotime, Shanghai, China). Equivalent amounts of protein were resolved by SDS-PAGE and then transferred to polyvinylidene difluoride (PVDF) membranes (Millipore, USA). The membranes were incubated at 4°C overnight with the following primary antibodies: anti-C1R, anti-MFGE8, anti-S100A11, anti-PCOLCE2, anti-SFRP2, anti-HIF-1α and anti-GAPDH. Following incubation with the appropriate secondary antibodies (Servicebio, Wuhan, China) for 1 h at room temperature, immunoreactive bands were detected using a chemiluminescence imaging system (Tanon, Shanghai, China). Band intensities were quantified with ImageJ software. Immunofluorescence staining Cells were fixed with 4% paraformaldehyde, and endogenous peroxidase activity was quenched by treatment with 3% H2O2. After blocking with goat serum, the cells were incubated overnight at 4°C with one of the following primary antibodies: anti-MFGE8, anti-S100A11, anti-PCOLCE2, or anti-SFRP2. The cells were then incubated with Cy3-conjugated goat anti-rabbit secondary antibody (Servicebio, Wuhan, China) for 1 h, and nuclei were counterstained with DAPI. Fluorescence images were acquired using an Olympus fluorescence microscope (Olympus, Tokyo, Japan). CCK-8 assay PASMCs were plated in 96-well plates at a density of 5,000 cells per well and treated as indicated. After treatment, 10 μL of CCK-8 working solution was added to each well, and the plates were incubated in the dark for 2 h. Cell viability was assessed by measuring the absorbance at 450 nm with a microplate reader (BioTek, VT, USA). EdU assay After different treatments, cell proliferation was assessed using the Click-iT EdU-555 Cell Proliferation Assay Kit according to the manufacturer’s instructions. Cells were incubated with EdU working solution (20 μM) for 2 hours, then fixed, washed, and permeabilized. Subsequently, cells were incubated with Click-iT reaction solution for 30 minutes for EdU detection, and nuclei were counterstained with Hoechst 33342. Representative images were captured using a fluorescence microscope (Olympus, Tokyo, Japan). Quantification and statistical analysis Statistical analyses were performed using R (Version 4.6.0). The non-parametric Kruskal-Wallis test was applied to compare multiple independent groups, and the Mann-Whitney U test was used for comparisons between two groups. n represents the number of human participants or biological replicates. Data were presented as mean ± SEM unless otherwise indicated. For box-and-whisker plots, the center line represents the median, the box represents the IQR (25th–75th percentiles), and the whiskers represent the smallest and largest values within 1.5×IQR. A p-value <0.05 was taken to indicate statistical significance for all tests. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Exact n values, statistical tests, and p values are reported in the figure legends. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.117846. Contributor Information Yong Wang, Email: wangyong1@ahmu.edu.cn. Cheng Cao, Email: chengcao2015@126.com. Binbin Zhang, Email: binbinzhang2025@fy.ahmu.edu.cn. Supplemental information Data S1. The raw data of Figure 7 References Associated Data Supplementary Materials Data S1. The raw data of Figure 7 Data Availability Statement •Data from publicly archived datasets are available from the GEO database (GSE210248, GSE113439, GSE53408, and GSE117261): https://www.ncbi.nlm.nih.gov/geo/. All data reported in this study will be shared by the lead contact upon request. Data from publicly archived datasets are available from the GEO database (GSE210248, GSE113439, GSE53408, and GSE117261): https://www.ncbi.nlm.nih.gov/geo/. All data reported in this study will be shared by the lead contact upon request. •This paper does not report original code. This paper does not report original code. •Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request.