Multi-omics integration study of vascular smooth muscle cell phenotypic conversion identified novel biomarkers in idiopathic pulmonary arterial hypertension

Abstract

Background

Idiopathic pulmonary arterial hypertension (IPAH) is a progressive, high‑mortality disease driven in part by maladaptive vascular remodeling. Phenotypic conversion of vascular smooth muscle cells (VSMCs) is implicated in this process, but the cellular heterogeneity and regulatory drivers in IPAH remain incompletely defined.

Methods

We integrated scRNA‑seq datasets from IPAH and control lungs samples with bulk RNA‑seq from IPAH patients, MCT‑induced rat IPAH models, applying diffusion‑map dimensionality reduction, pseudotime inference, CellChat, and machine learning to nominate phenotypic conversion regulatory genes (PCRGs). Candidate genes were validated in independent cohorts and functionally tested in hypoxia‑induced human PASMCs.

Results

Single-cell analysis identified three VSMC clusters-contractile, proliferative, and synthetic-and reconstructed a dedifferentiation trajectory. 40 PCRGs correlated with the primary diffusion-map component, of which POSTN and CCDC80 emerged as hub genes up-regulated in rat and human IPAH cohorts. Hypoxia induced their expression in PASMCs, and siRNA knockdown attenuated hypoxia‑driven proliferation and migration. A multivariable VSMC phenotypic conversion signature (VPCS) integrating these biomarkers and clinical variables demonstrated robust diagnostic discrimination in validation datasets.

Conclusions

Our from high‑dimensional datasets.integrative VSMC‑centered approach delineates dynamic state transitions in IPAH, nominates CCDC80 and POSTN as candidate regulators and biomarkers, and offers a practical framework for prioritizing targets

Supplementary Information

The online version contains supplementary material available at 10.1186/s12931-026-03568-8.

Introduction

Idiopathic pulmonary arterial hypertension (IPAH) is a rare but devastating disease characterized by progressive elevation of pulmonary arterial pressure, leading to right ventricular dysfunction and eventual heart failure [1, 2]. Despite therapeutic advances, IPAH remains associated with high morbidity and mortality [3]. Its pathobiology is complex and multifactorial, involving an aberrant immune-inflammatory response, dysregulated vascular remodeling, and phenotypic modulation of vascular smooth muscle cells (VSMCs) [4, 5]. A detailed molecular understanding of these alterations is essential for developing targeted therapies and improving clinical outcomes.

The complex interplay of genetic and environmental factors in IPAH requires high-resolution approaches to dissect underlying mechanisms [6–8]. Although high-throughput sequencing has produced abundant RNA expression data, limited sequencing depth and pervasive cellular heterogeneity are obstacle to further exploration of IPAH [9]. Fortunately, single-cell RNA sequencing (scRNA‑seq) has emerged as a powerful tool to profile biological process at the single‑cell resolution. This approach delineates cellular heterogeneity, identifies discrete cell states and lineage trajectories, and uncovers candidate regulatory genes and rare pathogenic cell populations [10, 11]. Recent studies suggest that VSMCs in IPAH undergo marked transitions toward a synthetic phenotype [12, 13]. However, the detailed process and its regulatory factors remain largely unexplored at the single‑cell level.

Multi‑omics approaches are increasingly recognized for their ability to provide complementary perspectives on complex biology [14, 15]. Integrating scRNA‑seq with bulk RNA‑seq leverages the strengths of both modalities: scRNA‑seq reveals cell type specific mechanisms, while bulk RNA‑seq delivers greater quantitative robustness and larger clinically annotated cohorts that enhance statistical power [16, 17]. This synergy enables the translation of mechanistic insights at the cellular level into reproducible expression signatures at the population scale [18], which strengthens biomarker discovery, enhances the generalizability of models, and improves clinical applicability. Additionally, it further refines this integrative framework by allowing for efficient feature selection and dimensionality reduction from high-dimensional transcriptomic data [19]. Machine learning methods capture nonlinear relationships and complex interactions, integrate heterogeneous predictors, and support rigorous validation through resampling and ensemble techniques [20, 21]. With machine learning tools, single cell observations can be projected onto bulk RNA‑seq datasets-focusing attention on a small number of robust targets and thus greatly facilitate mechanistic understanding and clinical translation.

Moreover, by integrating trajectory inference, cell-cell communication analysis, and machine-learning based feature prioritization, such strategies closely parallel analytical pipelines increasingly employed in tumor microenvironment (TME) research to bridge mechanistic discovery with clinical translation. Recent TME-focused single-cell studies have demonstrated that fibroblasts, smooth muscle–like cells, and perivascular populations undergo coordinated and continuous state transitions driven by extracellular matrix remodeling, hypoxia, and bidirectional signaling with immune and endothelial cells [22, 23]. Importantly, intercellular communication networks have emerged as central regulators of these processes, enabling adaptive responses to chronic stress and sustaining pathological tissue remodeling [24]. Applying this conceptual framework to IPAH supports the view that VSMC phenotypic conversion represents a dynamic, microenvironment regulated continuum rather than a binary switch.

In this study, we integrate single‑cell and bulk transcriptomic data to investigate the cellular dynamics and regulatory pathways underlying VSMC phenotypic conversion in IPAH. We employ scRNA‑seq to identify key regulatory genes and cell‑state trajectories, and we apply machine‑learning methods to bulk RNA‑seq cohorts to select and validate clinically translatable biomarkers and to construct a nomogram prediction model. Our aim is to deepen understanding of VSMC‑driven vascular remodeling in IPAH and to advance biomarker guided strategies for diagnosis and personalized treatment.

Method

Single-cell RNA-seq data collection and processing

We retrieved scRNA-seq data from GSE210248 (three IPAH samples and three control samples) and GSE169471 (three IPAH samples and six control samples) in the Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/) database. Quality control pipeline was executed with the following criteria: > 200 genes/cell, < 4000 genes/cell, > 3 cells/gene, and < 10% mitochondrion genes. Then, we integrated two single cell datasets through canonical correlation analysis (CCA)-based workflow from Seurat [25]. Briefly, each sample was log-normalized and 3,000 highly variable genes (HVGs) were identified per dataset using the “vst” method. The datasets were split into a list, and “FindIntegrationAnchors” function was used to identify anchors-pairs of cells representing shared biological states across datasets. Finally, “IntegrateData” was applied to perform anchor-based batch correction and generate the integrated expression matrix.

The top 20 principal components (PCs) were selected based on HVGs’ expression. Using the first 20 PCs (as guided by the elbow plot, Figure S1A), we constructed a shared nearest-neighbor (SNN) graph and performed community detection with FindNeighbors and FindClusters (resolution = 0.5), which employs a Louvain algorithm. Subsequently, cell clusters were identified using the FindClusters function, which employs the Louvain algorithm. The t-distributed Stochastic Neighbor Embedding (t-SNE) approach was then used for further dimensionality reduction and to visualize cluster structure. Cell type assignment was achieved by examining marker genes highly expressed in each subcluster and cross‑referencing these candidates with canonical cell type markers from the CellMarker database [26].

Bulk data processing and analysis

The raw data for GSE117261 and GSE254617 were downloaded from the GEO database. Probe identifiers were mapped to gene symbols using the corresponding platform annotation files. Expression values were then transformed to log2(counts + 1) for downstream analyses.

Tissue distribution of distinct cell subtypes

Tissue preference of each cell type was estimated by the observed to predicted cell number ratio (Ro/e) [27]. Re/o denotes enrichment or depletion of a given cell type within a particular tissue.

Cell–cell communication analysis

Based on the CellchatDB, we applied CellChat to reconstruct intercellular communication networks from the scRNA-seq data [28]. Normalized expression matrices were provided to CellChat to detect and quantify signaling interactions among different cell types.

Assessing the contribution score of different cell types

To quantify specific cell type’s contribution to IPAH, we first used the “FindAllMarkers” function to obtain the top 100 differentially expressed genes (DEGs) between IPAH and control. Subsequently, we calculated the fold change in gene expression (FCexp) and the proportion of gene expression (FCprop) of PAH relative to the control group across different cell types. The contribution score for a cell type was defined as the square root of the product FCexp × FCprop.

Functional enrichment analysis at the single cell level

After further dimensionality reducing and clustering, the “FindAllMarkers” function was performed to identify VSMC subtypes’ markers to further explore biological functions. Genes with average expression of Fold Change (FC) > 1.5 and p-value < 0.05 were regarded as significant DEGs, which were subjected to Gene Ontology (GO) terms enrichment analyses to characterize functional differences among distinct VSMC subtypes.

Pseudotime trajectory inference

We first applied the Slingshot algorithm to reconstruct the differentiation trajectories of VSMCs [29]. The input for the Slingshot analysis and cell lineage trajectory formation comprised the Uniform Manifold Approximation and Projection (UMAP) embedding and subclustering assignments of VSMCs [30]. Additionally, we applied the Diffusion Map method [31] for dimensionality reduction: PCA was first performed, and Euclidean distances between cells in the resulting PCA space were used to construct the diffusion kernel. Thereafter, pseudotime and diffusion pseudotime (DPT) were calculated based on the the diffusion-map component (DC) embedding of VSMCs.

Identification of VSMC phenotypic conversion genes

During dimensionality reduction through Diffusion Map, we observed clear separation of VSMC subpopulations along the DC1. This discovery demonstrated that DC1 served as a critical signature for delineating heterogeneity among VSMC subtypes. Accordingly, we assessed the correlation between gene expression and DC1 to identify VSMC Phenotypic Conversion Regulatory Genes (PCRGs). Genes with an absolute correlation with DC1 > 0.5 and with DC2 < 0.2 were defined as PCRG members.

Development and validation of VSMC phenotypic conversion signature

To identify hub PCRGs in IPAH, we applied multiple machine-learning approaches: least absolute shrinkage and selection operator (LASSO) regression, random forest-recursive feature elimination (RF-RFE), and a variance-threshold filter. RF-RFE was performed with ten-fold cross-validation. Hub PCRGs were defined as genes selected by both LASSO and RF-RFE and having a variance greater than 0.5. We then used logistic regression to evaluate the associations of these hub PCRGs and clinical variables with disease status, and constructed a diagnostic nomogram based on the multivariable logistic regression coefficients, termed the VSMC phenotypic conversion signature (VPCS). Receiver operating characteristic (ROC) curves were generated to evaluate the diagnostic accuracy and robustness of the VPCS and hub PCRGs, with performance quantified by the area under the curve (AUC).

RNA sequencing and profiling

RNA sequencing was performed on lung samples from seven monocrotaline‑induced rats and six control rats. Total RNA was extracted with Trizol reagent (Thermo Fisher) and assessed for quantity and integrity on an Agilent Bioanalyzer 2100 using the RNA 6000 Nano kit (Agilent); samples with RIN > 7.0 were used for library construction. mRNA was purified from 5 µg total RNA using Dynabeads Oligo(dT) (Thermo Fisher) with two rounds of enrichment, then fragmented with the Magnesium RNA Fragmentation Module (NEB) at 94 °C for 5–7 min. First‑strand cDNA synthesis was performed with SuperScript™ II (Invitrogen), followed by second‑strand synthesis incorporating dUTP using E. coli DNA polymerase I and RNase H. After end repair and A‑tailing, dual‑index adapters were ligated and libraries were size‑selected with AMPure XP beads to an average insert size of ~ 300 ± 50 bp. U‑labeled second strands were treated with heat‑labile UDG (NEB), and adapter‑ligated products were PCR‑amplified (initial denaturation 95 °C for 3 min; 8 cycles of 98 °C 15 s, 60 °C 15 s, 72 °C 30 s; final extension 72 °C for 5 min). Final libraries were purified, evaluated on the Agilent Bioanalyzer 2100, and sequenced as 2 × 150 bp paired‑end reads on an Illumina NovaSeq 6000 (LC‑Bio).

Cell lines and cultures

Human pulmonary artery smooth muscle cells (HPASMCs) were obtained from Wuhan Procell Life Science & Technology Corporation. Cells were maintained in DMEM supplemented with 10% fetal bovine serum and incubated at 37 °C in a humidified atmosphere with 5% CO2. For hypoxia experiments, HPASMCs were cultured in a 1% O2 chamber (Galaxy R, RS Bitotech, Alloa, UK) continuously supplied with a gas mixture producing 1% O2 and 5% CO2 (balance N2).

Quantitative real-time PCR

Quantitative real-time PCR (qRT-PCR) was performed using the SYBR Green Premix Pro Taq HS qPCR Kit (Rox Plus; Accurate Biology, China) on a QuantStudio 5 real-time PCR system (Applied Biosystems, Thermo Fisher Scientific). The primer sequences were as follows: CCDC80 forward, 5′-AACGATGAAGACTGGGCCTATTC-3′, reverse, 5′-CAACTTCCTCTCCAACGCCTAAA-3′, POSTN forward, 5′-AGGATGTGACGGTGACAGTATAA-3′, reverse, 5′-CAGGTGCCAGCAAAGTGTATTC-3′. The cycling conditions were as follows: initial denaturation for 10 min at 95 °C followed by 40 cycles of denaturation (15 s at 95 °C), annealing and elongation (30 s at 60 °C). Relative RNA expression levels were calculated using the 2^−ΔΔCt method, with GAPDH as the internal reference gene for normalization.

Transient cell transfection

Small interfering RNA (siRNA) targeting CCDC80 and POSTN was purchased from OBiO Technology (OBiO, Shanghai, China) and transfected into HPASMCs using Lipofectamine 3000 transfection reagent according to the manufacturer’s instructions (Thermo Scientific, Shanghai, China).

Cell proliferation and transwell assay

Cell Counting Kit-8 (Servicebio, Hubei, China) assays were used to detect the proliferation of HPASMCs according to the manufacturer’s instructions. HPASMCs were plated in 96-well plates at a density of 3000 cells per well and incubated for 24 h. The medium was then replaced with various transfection mixtures. After a 6-hour incubation period, the transfection mixtures were substituted with medium containing 10% fetal bovine serum, and the cells were incubated for an additional 24 h. Subsequently, the cells were subjected to either normoxic or hypoxic conditions for another 24 h. Finally, 10µL of CCK-8 reagent (Dojindo, Japan) was added to each well, and the optical density was measured using an ELx800 Universal Microplate Reader (BioTek Instruments, Winooski, VT, USA).

After transfection, cell migration assays were performed using a 24-well plate (Corning, MA, USA) in accordance with the manufacturer’s protocol. Initially, 200µL of cell suspension (10,000 cells in total) containing 1% fetal bovine serum was added to the upper chamber, while 600µL of medium with 15% fetal bovine serum was placed in the lower chamber. The cells were incubated under normoxic or hypoxic conditions for 24 h. After incubation, cells that had adhered to the lower surface of the membrane were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet. Non-migrated cells were removed from the upper surface by gently scraping the membrane with a cotton swab, and the migrated cells were counted using an optical microscope (Zeiss, Oberkochen, Germany).

Statistical analysis

All data processing, plotting, and statistical analysis were conducted in R (version 4.3.1). The T test was applied to compare continuous variables. For all statistical tests, a two-sided p < 0.05 was considered to indicate statistical significance, with the following notation: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Result

scRNA-seq analysis revealed cell landscape and characteristics in IPAH

After quality control and integration, 51,610 cells from six IPAH samples and nine control samples were retained. t-SNE analysis identified 18 cell clusters (Figure S2B), which were annotated into nine cell types: B cells, endothelial cells (ECs), epithelial cells, fibroblasts, macrophages/monocytes, mast cells, natural killer (NK) cells, smooth muscle cells (SMCs), and T cells (Figure S2C, Fig. 1A). We then quantified changes in cell-type proportions in IPAH. As shown in Fig. 1B, several immune cell populations were increased in IPAH, including T cells, mast cells, and B cells, while fibroblasts, SMCs, and NK cells were more abundant in controls. Marker gene expression for each cell type was presented in Fig. 1C to validate the annotations.

Fig. 1.

scRNA-seq depicted the cell landscape in IPAH. A Cell landscapes of 51,610 cells from nine types (B cells, endothelial cells, epithelial cells, fibroblasts, macrophages/monocytes, mast cells, natural killer cells, smooth muscle cells, and T cells). B Cell proportion in IPAH and control groups. C Heatmap showing marker genes’ expression in distinct cell types. D Line chart showing tissue prevalence for each cell type estimated by Ro/e score. E Bar plot exhibiting the intercellular interactions number in the IPAH and control groups

To assess tissue-specific enrichment of cell populations, we performed Ro/e analysis. Consistent with the proportion changes, most immune cells were enriched in IPAH lung tissue, whereas fibroblasts, SMCs, and NK cells were preferentially located in the control group (Fig. 1D, Table S1).

Besides, using ligand-receptor pairs and co-factors identified by CellChat, we assembled a cell-cell communication network. Notably, the IPAH group displayed a higher number of intercellular interactions (Fig. 1E).

Identification and functional characterization of vascular smooth muscle cell subpopulations

We then quantified contribution scores across cell types. At single-cell resolution, IPAH-associated changes were mainly attributable to SMCs and fibroblasts, while mast and NK cells contributed little (Fig. 2A, Table S2). Accordingly, we focused on SMCs for dimensionality reduction and clustering using UMAP, which resolved three SMC clusters (Fig. 2B). Afterwards, SMC cluster 1 and cluster 2 were annotated as vascular SMCs (VSMCs) based on the high expression of NOTCH3 (Figure S2A). Moreover, compared with the control group, VSMCs in the IPAH group displayed markedly stronger interactions with nearly all other cell types (Figure S2B, C). Subsequently, UMAP-based analysis identified three VSMC subclusters (Fig. 2C).

Fig. 2.

Identification and functional characterization of vascular smooth muscle cell clusters. A Pie chart exhibiting contribution score of distinct cell types. B Cell landscapes of three smooth muscle cell clusters. C Cell landscapes of three vascular smooth muscle cell clusters. D Feature plot exhibiting makers’ expression among different vascular smooth muscle cell clusters. E Word cloud chart showing biological pathway enrichment results of each vascular smooth muscle cell cluster

We then investigated the functional characteristics of the VSMC subclusters by examining marker gene expression and performing biological pathway enrichment analyses based on significant DEGs among VSMC subclusters (Table S3). As described in Fig. 2D, from Cluster 1 to Cluster 3, the expression of markers for contractile VSMCs, such as ACTA2 and MYH11, gradually decreased, while expression of extracellular matrix (ECM) components, such as FN1 and DCN, progressively increased. Further biological pathway enrichment analysis revealed that Cluster 1 displayed features of contractile VSMCs, including enrichment for muscle system processes and muscle contraction; Cluster 2 was enriched for ECM-related pathways, notably extracellular matrix organization and extracellular structure organization; and Cluster 3 was associated with elevated proliferative activity, with activation of pathways governing smooth muscle cell proliferation (Fig. 2E, S3A). Therefore, Cluster 1 was classified as contractile VSMCs, Cluster 2 as proliferative VSMCs, and Cluster 3 as synthetic VSMCs.

Pseudotime trajectory inference

To further elucidate the dynamics of VSMCs’ differentiation, we performed the pseudotime trajectory analysis. Firstly, slingshot pseudotime analysis inferred a lineage placing VSMC Cluster1 at the origin and VSMC Cluster3 at the terminus (Fig. 3A). Similar findings were demonstrated in the diffusion-map pseudotime trajectory analysis. Both pseudotime and diffusion pseudotime (DPT) analyses indicated an ordering from VSMC Cluster 1 through Cluster 2 to Cluster 3 (Fig. 3B, C). At the single-cell level, we identified a phenotypic transition from contractile to synthetic VSMCs in IPAH.

Fig. 3.

VSMCs’ phenotypic conversion process and involved regulatory genes in IPAH. A The developmental trajectory of VSMCs identified by Slingshot. B Mountain plot exhibiting VSMC clusters’ distribution along pseudotime identified by Diffusion Map. C VSMC clusters’ diffusion-map component distribution, colored-coded by diffusion-map pseudotime (DPT). D VSMC clusters’ diffusion-map component distribution, colored-coded by cell type. E Correlations of diffusion-map component (DC) 1 with markers of contractile and synthetic VSMCs. F Absolute correlations of genes with DC1 and DC2

Interestingly, after dimensionality reduction using diffusion map, the different VSMC clusters were clearly separated along Diffusion-map Component (DC)1, while showing no discernible separation along DC2 (Fig. 3D). This implies that DC1 might be a key determinant for understanding differences among VSMC clusters and their phenotypic conversions. To enhance the interpretability of DC1 and facilitate understanding of VSMC phenotypic conversion for clinical translation, we computed the correlations between gene expression and DC1 in VSMCs. As shown in Fig. 3E, DC1 was strongly negatively correlated with contractile VSMC (e.g., ACTA2 and TAGLN) and strongly positively correlated with synthetic markers (e.g., LUM and COL1A2), consistent with the phenotypic conversion process of VSMCs in IPAH. Finally, we identified 40 genes that were strongly associated with DC1 (|r| > 0.5) but weakly associated with DC2 (|r| < 0.2), which we designated as VSMC Phenotypic Conversion Regulatory Genes (PCRGs) (Fig. 3E, Table S4).

Development and validation of VSMC phenotypic conversion signature

Machine learning algorithms were applied for hub PCRGs identification to distinguish IPAH and control group. Under ten-fold cross‑validation, LASSO selected the most parsimonious model comprising 11 genes (Fig. 4A, B); RF‑RFE retained seven genes when the root mean square error (RMSE) reached its minimum (Fig. 4C); and nine genes passed the variance‑threshold filter (Fig. 4D). POSTN and CCDC80, which were common to all three methods were designated as hub PCRGs (Fig. 4D). Then, we accomplished RNA-seq on lung samples from the MCT-induced rat IPAH model and identified 301 up-regulated genes and 108 down-regulated genes (Fig. 4E). Postn and Ccdc80 were significantly upregulated in IPAH rats compared with controls (Fig. 4F). Consistent results emerged in both GSE117261 and GSE254617 cohorts (Fig. 4G, H).

Fig. 4.

Hub PCRGs identification and validation. A-B In the LASSO regression model, the determination of the optimal λ and included genes. C The relationship between features and the root mean square error (RMSE) in RF-RFE process. D Venn plots exhibiting two identified genes from distinct machine learning algorithms. E Volcano plot displaying up-regulated genes and down-regulated genes in MCT-induced rat IPAH model. F-H Box plot showing elevated level of PSTN and CCDC80 in MCT-induced rat and IPAH patients in GSE117261 and GSE254617

We then applied multivariable logistic regression to evaluate associations between the hub PCRGs, clinical covariates, and IPAH status. The forest plot showed that POSTN and CCDC80 were risk factors for IPAH, whereas sex was protective (Fig. 5A). Based on the three significant predictors from the model, we then constructed a diagnostic nomogram, termed the VSMC phenotypic conversion signature (VPCS). As displayed in Fig. 5C, the IPAH group had higher VPCS level than controls. Moreover, in two independent RNA-seq cohorts, the VPCS outperformed individual predictors in diagnostic performance (Fig. 5D, E).

Fig. 5.

Development and accuracy of VSMC phenotypic conversion signature. A Forest plot showing odds ratios (ORs) and p-values for CCDC80, POSTN, sex, and age as risk factors for IPAH. B Nomogram model developed for IPAH risk prediction. C Dot plot exhibiting risk score of IPAH and control group in GSE117261. D-E Receiver operating characteristic curves of the VSMC phenotypic conversion signature, POSTN, and CCDC80 in GSE117261 and GSE254617

Meanwhile, in hypoxia-induced HPASMCs, elevated mRNA expression levels of CCDC80 and POSTN were detected, consistent with our above conclusions (Fig. 6A, B). We further selected CCDC80 and POSTN for knockdown and performed cell function experiments to decode the role of VPCS in HPASMCs’ function regulation. The CCK-8 assay was conducted to provide cell viability evaluation and the Transwell assay was carried out to assess cell migration. HPASMCs demonstrated increased viability and migration under hypoxia (Fig. 6C-F). To assess the role of CCDC80 and POSTN involved in the process in HPASMCs, siRNA against CCDC80 and POSTN were transfected into HPASMCs. As displayed in Fig. 6C-F, the knockdown of CCDC80 and POSTN partially abrogated the excessive cell proliferation and migration under hypoxia.

Fig. 6.

Verification of key biomarkers. In hypoxia-treated human pulmonary artery smooth muscle cells (HPASMCs), mRNA levels of CCDC80 and POSTN are upregulated (A, B). Hypoxia enhances HPASMC viability and migration as measured by CCK-8 and Transwell assays (C–F). siRNA-mediated knockdown of CCDC80 or POSTN partially attenuates the hypoxia-induced increases in proliferation and migration (C–F)

Discussion

Pulmonary arterial hypertension (PAH) is a progressive disease with high morbidity and mortality and currently lacks curative therapies [32]. Idiopathic PAH (IPAH) denotes cases without an identifiable cause. Increasing evidence implicates phenotypic conversion of vascular smooth muscle cells (VSMCs) as a central contributor to pulmonary vascular remodeling, with proposed mechanisms including oxidative stress [33, 34], autophagy [35, 36], endoplasmic reticulum stress [37] and mitochondrial dysfunction [38, 39]. However, the cellular heterogeneity and regulatory circuitry driving VSMC switching in human IPAH remain incompletely defined.

To address this gap, we applied an integrative multi‑omics strategy combining scRNA‑seq and bulk RNA‑seq with downstream machine‑learning prioritization to characterize VSMC phenotypic conversion and nominate candidate regulators with diagnostic and mechanistic relevance. Our single‑cell analyses profiled changes in cell proportions, low‑dimensional spatial distribution and altered cell–cell interactions in IPAH samples. By integrating contractile marker expression, extracellular matrix (ECM) signatures and pathway enrichment, we resolved three VSMC states-contractile, proliferative and synthetic-and reconstructed a pseudotime trajectory consistent with dedifferentiation from contractile toward synthetic phenotypes.

Functionally, this transition was marked by loss of contractile markers and upregulation of cell‑cycle and ECM‑related programs, together with increased proliferative and migratory potential, supporting a role for VSMC state changes in vascular remodeling. To move beyond descriptive signatures, we used diffusion‑map based dimensionality reduction to separate VSMC subtypes in low‑dimensional space and derived correlation‑based thresholds to nominate candidate regulators. Subsequent machine‑learning prioritization highlighted CCDC80 and POSTN as top candidates; both genes showed concordant upregulation in an MCT‑induced rat model and in two independent human bulk RNA‑seq cohorts.

We further validated these candidates experimentally: hypoxic conditions induced higher expression of POSTN and CCDC80 in VSMCs, and their upregulation enhanced VSMC proliferation and migration in vitro, linking these genes to the phenotypic features observed in single‑cell profiles. Integrating these biomarkers with clinical variables produced a multivariable diagnostic model that was validated in independent cohorts, indicating potential utility for diagnosis or risk stratification pending further testing.

POSTN (periostin) is a matricellular protein associated with the ECM. Induced by TGF‑β, mechanical stress and hypoxia, POSTN contributes to ECM remodeling and collagen organization [40, 41]. A previous study has shown that POSTN mediates the proliferative and migratory responses of PASMCs to cigarette smoke extract [42]. Moreover, POSTN has been shown to promote endothelial dysfunction in PAH, thereby contributing to vascular remodeling [43]. CCDC80 is a secreted/co-secreted protein with a coiled‑coil domain, which has been reported in ECM-associated fractions and implicated in extracellular signaling [44]. CCDC80 is critically involved in regulating vascular remodeling and homeostasis [22, 45]. Recent research has linked it to VSMC phenotypic switching in aortic dissection [46]. An increasing body of evidence links POSTN and CCDC80 to vascular diseases. Notably, as ECM‑associated proteins, POSTN and CCDC80 are readily detectable in patient’s blood and other body fluids. Thus, VSMC phenotypic conversion signature (VPCS), based on serum POSTN and CCDC80 levels, could serve as a potential early biomarker of IPAH risk, enabling earlier intervention and timely treatment. It may contribute to targeted therapies against vascular remodeling treatment option for IPAH patients, but only after rigorous preclinical studies. It may inform the development of targeted anti‑vascular‑remodeling therapies for IPAH patients, but only after rigorous preclinical validation.

Several limitations should be noted. The number of human scRNA‑seq samples is modest, and larger, multicenter datasets are required to confirm the generality of the identified VSMC states and trajectories. Tissue dissociation and cell capture can introduce compositional bias and erase spatial context; complementary spatial transcriptomics or in situ validation would strengthen the anatomical and microenvironmental interpretation. The MCT-induced rat model, while useful for concordance testing, does not fully recapitulate human IPAH pathophysiology; rigorous in vivo gain and loss of function studies across multiple models are needed to establish causality. Finally, although diffusion‑map selection and machine learning provide a structured prioritization workflow, translating high‑dimensional signals into actionable therapeutic targets demands additional orthogonal validation and mechanistic dissection.

Conclusion

In summary, our integrative VSMC‑centered approach delineates dynamic state transitions in IPAH, nominates CCDC80 and POSTN as candidate regulators and biomarkers, and offers a practical framework for prioritizing targets from high‑dimensional datasets.

Abbreviation

IPAH

Idiopathic pulmonary arterial hypertension

VSMC

Vascular smooth muscle cell

scRNA‑seq

Single-cell RNA sequencing

RNA‑seq

RNA sequencing

ML

Machine learning

GEO

Gene Expression Omnibus

CCA

Canonical correlation analysis

HVGs

Highly variable genes

PC

Principal component

UMAP

Uniform Manifold Approximation and Projection

SNN

Shared nearest-neighbor

DEG

Differentially expressed gene

FC

Fold Change

DPT

Diffusion pseudotime

DC

Diffusionmap component

LASSO

Least absolute shrinkage and selection operator

RF-RFE

Random forest-recursive feature elimination

ROC

Receiver operating characteristic

AUC

Area under the curve

VPCS

VSMC phenotypic conversion signature

ECM

Extracellular matrix

PCRGs

Phenotypic Conversion Regulatory Genes

RMSE

Root Mean Square Error

Authors’ contributions

1. C, and R.W. contributed study design and paper revising. R.J, Y.M., C.Y. and M.Y. contributed project oversight. M.G., T.J., Y.L. and R.W. contributed data analysis, visualization, and paper writing. K.W. and M.F. conducted experiments. Y.W., Y.L., P.L. and Y.G. contributed paper revisiting. All authors approved this manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (grant no. 82170037), Key Scientific Research Project of Science and Technology Department of Henan Province (grant no. 231111310800).

Data availability

This study did not generate new unique reagents.Data and code availability The code that supports the findings of this study is available from the corresponding author upon reasonable request. The datasets can be accessed via the following links: GSE210248 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE210248), GSE169471 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE169471), GSE117261 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE117261), GSE254617 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE254617).

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not required.

Competing interests

The authors declare no competing interests.