Plasma proteomic analysis identified potential therapeutic targets for pulmonary arterial hypertension

Summary

Pulmonary arterial hypertension (PAH) remains a progressive disease with high mortality. By integrating the UK Biobank Pharma Proteomics Project (49,272 participants, 337 incident cases) with drug-target Mendelian randomization, we identified 398 plasma proteins associated with incident PAH and 19 with potential causal relevance, 15 of which are druggable. Then we prioritized 5 candidates for expression validation in the Sugen5416 and hypoxia rat model and confirmed significant upregulation of TNFRSF10A and downregulation of UMOD in lung tissue, consistent with human plasma proteomic trends. Subsequent in vitro functional assays demonstrated that UMOD treatment attenuates PAH-related cellular phenotypes, whereas TNFRSF10A knockdown exacerbates pro-pathogenic progression. These results suggest UMOD as a promising protective therapeutic candidate and indicate the inherent complexity of the TNFRSF10A signaling. This integrative study provides a robust framework for translational research, future studies are warranted to elucidate downstream molecular mechanisms and further evaluate the therapeutic efficacy of these candidates.

Graphical abstract

Highlights

• •Plasma proteomic signature for incident PAH was established in a general cohort
• •19 proteins were causally associated with PAH, with 15 identified as druggable
• •Dysregulation of TNFRSF10A and UMOD were validated in PH rat lungs
• •Showed UMOD protects against PAH and highlighted TNFRSF10A signaling complexity

Complication; Pharmaceutical science; Proteomics

Introduction

Pulmonary arterial hypertension (PAH) is a progressive vascular disease characterized by proliferative remodeling, elevated pulmonary pressures, and vascular resistance. Initially, the right ventricle compensates for heightened afterload through hypertrophy and enhanced contractility. However, persistent exposure to elevated afterload leads to its eventual failure, resulting in right heart failure and even death if left untreated.¹ Although the current treatment strategies have significantly improved the survival of the patients, PAH remains a devastating disease that cannot be cured.² Moreover, the unspecific symptoms and signs of PAH often delay diagnosis, potentially worsening the clinical outcomes of patients.¹ Therefore, identifying promising biomarkers and therapeutic targets in PAH is a pressing concern.

The plasma proteome, which encompasses the full spectrum of proteins actively secreted or leaked from tissues and cells, serves as a reflection of the body’s health status. An imbalance in the plasma proteome can act as an early marker of disease.³ Recent studies have leveraged high-throughput methods to identify proteins associated with PAH and those that may inform patient prognosis.^4,5 However, most previous efforts have relied on case-control designs, focusing on proteins linked to established PAH, and there is a lack of a study to explore proteins associated with incident PAH in the general population. Furthermore, while correlating protein levels with disease status provides valuable insights, it does not indicate causality. Integrating genome-wide association studies (GWAS) with molecular quantitative trait locus (molQTL) data, such as gene expression quantitative trait locus (eQTL) or protein quantitative trait locus (pQTL), offers the potential for causal inference and may reveal promising drug targets for PAH.^6,7

In this study, we employed a multi-stage analytical framework that integrated large-scale proteomic profiling with causal inference and experimental validation to identify therapeutic targets for PAH. Initially, we defined a proteomic signature for incident idiopathic PAH, the predominant subtype in Western countries, using the prospective UK Biobank (UKB) cohort. Subsequently, we leveraged drug-target Mendelian randomization (MR) to evaluate the putative causal relationships between these proteins and PAH risk. To corroborate these findings in a disease-relevant context, we confirmed the prioritized targets of candidate proteins in the Sugen5416 and hypoxia (SuHx) rat model. Finally, the phenotypic effects of the identified candidates were validated in vitro using human pulmonary artery endothelial cells (hPAECs) and human pulmonary artery smooth muscle cells (hPASMCs), establishing a robust link between genetic-proteomic associations and functional biology.

Results

Baseline characteristics of the cohort

To identify proteomic signatures associated with PAH risk, we analyzed a cohort of 49,272 participants from the UKB. During a median follow-up of 13.64 (interquartile range: 12.92–14.39) years, 337 incident PAH cases were identified. The baseline characteristics of individuals who later developed PAH were shown in Table 1. Compared to non-PAH participants, those in the PAH group were generally older (62.88 ± 6.06 vs. 57.61 ± 8.12 years), exhibited a higher body mass index (29.65 ± 5.82 vs. 27.42 ± 4.76 kg/m²) and a higher proportion of male (51.63% vs. 45.98%). However, non-PAH participants was more likely to meet the criteria for ideal physical activity (54.27% vs. 45.36%). Moreover, participants who had PAH demonstrated a significantly higher prevalence of cardiovascular comorbidities at baseline, such as hypertension (57.27% vs 30.36%), diabetes mellitus (17.51% vs. 5.25%), and dyslipidemia (43.62% vs 18.87%). These findings were consistent with higher blood pressure, glucose, and lipid levels measured at recruitment in the PAH group.

Table 1.

The baseline characteristics of participants in UKB-PPP cohort

Variables	Total (n=49,272)	Non-PAH (n=48,935)	PAH (n=337)	p value
Age, years	57.64 ± 8.12	57.61 ± 8.12	62.88 ± 6.06	<0.001
Male, n (%)	22,675 (46.02)	22,501 (45.98)	174 (51.63)	0.044
BMI, kg/m2	27.43 ± 4.78	27.42 ± 4.76	29.65 ± 5.82	<0.001
Smoking status, n (%)				<0.001
Never	26,222 (53.39)	26,080 (53.47)	142 (42.64)	–
Previous	17,728 (36.10)	17,585 (36.05)	143 (42.94)	–
Current	5,160 (10.51)	5,112 (10.48)	48 (14.41)	–
Alcohol drinking status, n (%)				0.003
Never	1,572 (3.19)	1,553 (3.18)	19 (5.65)	–
Previous	1,827 (3.71)	1,807 (3.70)	20 (5.95)	–
Current	45,822 (93.09)	45,525 (93.13)	297 (88.39)	–
Education level, n (%)				<0.001
Low	17,185 (35.37)	17,015 (35.26)	170 (51.99)	–
Intermediate	15,874 (32.67)	15,789 (32.72)	85 (25.99)	–
High	15,529 (31.96)	15,457 (32.03)	72 (22.02)	–
Systolic blood pressure, mmHg	137.94 ± 18.70	137.91 ± 18.68	142.00 ± 21.57	0.001
Diastolic blood pressure, mmHg	82.06 ± 10.29	82.07 ± 10.29	80.67 ± 10.93	0.017
Triglycerides, mmol/L	1.49 [1.06, 2.15]	1.49 [1.05, 2.15]	1.50 [1.13, 2.10]	0.542
HDL cholesterol, mmol/L	1.45 ± 0.38	1.45 ± 0.38	1.35 ± 0.43	<0.001
LDL cholesterol, mmol/L	3.54 ± 0.88	3.54 ± 0.88	3.19 ± 0.90	<0.001
Glucose, mmol/L	4.94 [4.61, 5.33]	4.94 [4.61, 5.33]	5.08 [4.66, 5.77]	<0.001
Comorbidity
Hypertension, n (%)	15,051 (30.55)	14,858 (30.36)	193 (57.27)	<0.001
Diabetes mellitus, n (%)	2,626 (5.33)	2,567 (5.25)	59 (17.51)	<0.001
Dyslipidemia, n (%)	9,380 (19.04)	9,233 (18.87)	147 (43.62)	<0.001
Ideal physical activity, n (%)	24,946 (54.22)	24,809 (54.27)	137 (45.36)	0.002

Data were presented as mean ± standard deviation or median [Q1, Q3] according to their distributions. BMI, body mass index; HDL, high-density lipoprotein; LDL, low density lipoprotein; PAH, pulmonary arterial hypertension; UKB-PPP, UK Biobank Pharma Proteomics Project.

Proteomics signature of incident PAH

As shown in Figure S1, this study included 49,272 White European participants, with 34,492 randomly assigned to the discovery cohort and the remaining 14,780 to the replication cohort, with comparable baseline characteristics (Table S1). In the discovery cohort, 230 new-onset PAH cases were recorded during a median follow-up of 13.65 (interquartile range: 12.93–14.40) years, and 107 new-onset cases were documented in the replication cohort during a median follow-up of 13.53 (interquartile range: 12.90–14.35) years. Multivariate Cox regression analysis revealed 847 and 459 proteins were significantly associated with incident PAH in the discovery (Figure 1A) and replication (Figure 1B) cohort, respectively, all with a false discovery rate (FDR)-adjusted p (FDR-p) < 0.05. Notably, 398 proteins emerged as consistent candidates, all exhibiting uniform effect directions across both cohorts (Table S2). Of these, 376 proteins (94.47%) were positively associated with incident PAH, while 22 (5.53%) showed an inverse association.

Figure 1.

Volcano plots and functional enrichment of PAH-associated proteins

(A and B) Volcano plots illustrating the associations between protein levels and PAH risk in the discovery (A) and replication (B) cohorts. The x axis represents the effect size (β coefficients) from multivariable Cox proportional hazards models, and the y axis shows the −log10(FDR-p). Red and blue dots denote proteins significantly associated with increased and decreased PAH risk, respectively (FDR-p < 0.05), while gray dots represent non-significant proteins. The 19 causal targets are explicitly labeled.

(C) GO enrichment analysis was performed on 398 proteins associated with PAH, categorized by Biological Process (BP), Cellular Component (CC), and Molecular Function (MF).

(D) KEGG pathway analysis revealing the top enriched signaling pathways for 398 proteins associated with PAH.

For (C) and (D), the horizontal axis indicates the count of enriched proteins. Significance for enrichment was defined as an adjusted p value (Benjamini-Hochberg procedure) and q value < 0.05. FDR-p, false discovery rate-adjusted p value; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; PAH, pulmonary arterial hypertension.

Functional enrichment analysis

To investigate the biological significance of the 398 proteins that were associated with PAH, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed (Figures 1C and 1D). The Biological Process (BP) category within GO analysis highlighted a predominant enrichment in leukocyte migration, chemotaxis, and the positive regulation of the MAPK cascade, suggesting robust immune cell recruitment and active signaling favoring cell survival.^8,9 These findings were complemented by Cellular Component (CC) results, where proteins were significantly localized to the collagen-containing extracellular matrix, basement membrane, and focal adhesions. Notably, their enrichment in organelles involved in protein processing suggests that a substantial portion of these candidates function as secreted proteins. Molecular Function (MF) categories further centered on glycosaminoglycan binding and cytokine activity.

Consistent with these findings, KEGG analysis identified cytokine-cytokine receptor and extracellular matrix-receptor interactions as the most prominent pathways. The activation of critical axes, such as PI3K-Akt, TNF, NF-kappa B, MAPK, Ras, and HIF-1 signaling, underscores a complex network driving the hyperproliferation and metabolic reprogramming characteristic of PAH.^10,11

Sensitivity analyses

To ensure the reliability of our findings, the association of the 398 proteins with PAH was further replicated using both the competing risk model in the original datasets (Table S3) and the multivariate Cox proportional hazards model in the post-imputation datasets (Table S4). As shown in Figure S2, the majority of these proteins were successfully replicated. Specifically, 366 (91.96%) of the 398 proteins maintained significance in the competing risk model, and 342 (85.93%) remained significant in the multivariate Cox model after data imputation. The effect sizes also showed strong agreement in both discovery and replication cohorts, suggesting these proteins as reliable markers of incident PAH.

Causal inference identifies 19 potential drug targets

To investigate the causal roles and identify potential therapeutic targets, we performed a two-sample MR analysis. Among the 398 proteins associated with PAH, 325 proteins possessed at least three independent cis-pQTLs suitable for use as instrumental variables (IVs) (Table S5). In the pQTL MR analysis, a total of 21 proteins were found to be causally associated with PAH by the inverse variance weighting (IVW) method (FDR-p < 0.05, Table S6), and all of which exhibited effect directions consistent with observational Cox models.

To ensure the robustness of these findings, we prioritized proteins that maintained directional consistency across all complementary MR methods. Excluding for ITGBL1 and TNFRSF6B, 19 proteins emerged as high-confidence causal targets (Figure 2). For example, per standard deviation (SD) unit increased in AOC3 level was causally associated with 24% increased risk of PAH (odds ratio [OR]: 1.24, 95% confidence interval [CI]: 1.15–1.35, p = 2.26×10^-7), while with each SD increase in Uromodulin (UMOD) level was causally associated with 14% decreased risk of PAH (OR: 0.86, 95% CI: 0.80–0.92, p = 6.77×10^-6). Moreover, sensitivity analyses confirmed the reliability of these targets, with 18 and 16 proteins successfully replicated in competing risk and post-imputation models, respectively.

Figure 2.

Association and causal effects of 19 candidate proteins

(A and B) Forest plots illustrating the associations between the 19 candidate proteins and PAH risk in the discovery (A) and replication (B) cohorts.

(C) Causal effects of the 19 proteins on PAH risk estimated via drug-target MR analysis.

MR, Mendelian randomization; PAH, pulmonary arterial hypertension; pQTL, protein quantitative trait locus.

We further assessed the horizontal pleiotropy using the MR-Egger intercept test, and the results showed there is no horizontal pleiotropy present (p > 0.05, Table S7). Although the Q statistics indicated that the potential heterogeneity for TNFRSF10A (p < 0.05, Table S7), the causal estimates calculated by IVW with the multiplicative random effects method still remained significant.

To bridge the gap between protein levels and gene expression, we performed summary-data-based MR (SMR) using cis-eQTL data from the eQTLGen and GTEx Consortiums (Table S8). We identified 19 significant associations between gene expression (mRNA level) and PAH, with seven targets receiving robust support at both the proteomic and transcriptomic levels (Table 2). For example, in 11 different tissues, the increased expression of TNFRSF10A was associated with a higher risk of PAH, such as whole blood (OR: 1.39, 95% CI: 1.05–1.85, p = 0.022) and lung (OR: 1.35, 95% CI: 1.04–1.75, p = 0.024). Moreover, the heterogeneity in dependent instruments (HEIDI) test suggested these associations were not due to genetic linkage (p > 0.01, Tables 2 and S8).

Table 2.

Results of SMR analysis

Gene name	OR (95% CI)	p value	p value of HEIDI test	Tissue source
FAM3C	1.47 (1.11-1.95)	0.007	0.597	Skeletal muscle
FBLN2	1.36 (1.01-1.81)	0.040	0.209	Brain
QSOX1	1.20 (1.00-1.43)	0.049	0.535	Artery coronary
QSOX1	1.15 (1.00-1.31)	0.045	0.377	Artery aorta
QSOX1	1.38 (1.01-1.91)	0.046	0.512	Skeletal muscle
RNF149	2.31 (1.20-4.43)	0.012	0.191	Whole blood
SLAMF8	1.29 (1.01-1.65)	0.039	0.909	Whole blood
SORT1	2.58 (1.06-6.26)	0.036	0.940	Whole blood
TNFRSF10A	1.39 (1.05-1.85)	0.022	0.431	Whole blood
TNFRSF10A	1.35 (1.04-1.75)	0.024	0.458	Lung
TNFRSF10A	1.26 (1.02-1.55)	0.030	0.931	Heart_Atrial appendage
TNFRSF10A	1.47 (1.04-2.09)	0.029	0.137	Heart_Left ventricle
TNFRSF10A	1.15 (1.02-1.30)	0.023	0.478	Artery aorta
TNFRSF10A	1.37 (1.04-1.79)	0.024	0.216	Adipose subcutaneous
TNFRSF10A	1.11 (1.02-1.22)	0.022	0.274	Adrenal gland
TNFRSF10A	1.73 (1.06-2.81)	0.028	0.193	Transverse colon
TNFRSF10A	1.42 (1.04-1.93)	0.026	0.248	Skeletal muscle
TNFRSF10A	1.32 (1.03-1.69)	0.027	0.974	Spleen
TNFRSF10A	1.45 (1.04-2.02)	0.029	0.138	Stomach

CI, confidence intervals; HEIDI, heterogeneity in dependent instruments; OR, odds ratio; SMR, summary-data-based Mendelian randomization.

Druggability investigation

To evaluate the therapeutic potential of the 19 identified causal genes, we performed a comprehensive druggability investigation using the druggable genome list¹² and the Drug-Gene Interaction Database (DGIdb) (https://dgidb.org/) (Table 3). We found that 15 of the 19 target genes were included in the druggable genome, excluding MAD1L1, NRP2, RNF149, and SORT1. While FAM3C, MAMDC2, NPPB, QSOX1, and SLAMF8 have the potential to be druggable, no existing drug-gene interactions were identified in DGIdb for these targets. Ultimately, 10 druggable genes with potential drug-gene interactions were identified and shown in Table S9.

Table 3.

A summary of the druggability for 19 target proteins

Classificationa	Target	UniPort ID	Function	Subcellular location	The possible role in PAH	Druggable gene
Strong evidence	NPPB	P16860	Vasoactive; Vasodilator	Secreted	Mediating cardio-renal homeostasis	Yes
	EDN1	P05305	Vasoactive; Vasoconstrictor	Secreted	Vasoconstrction, vascular remodeling	Yes
	NOTCH3	Q9UM47	Activator; Developmental protein; Receptor; Differentiation; Notch signaling pathway; Transcription	Cell membrane; Membrane; Nucleus	Phenotype switch of VSMCs	Yes
Moderate evidence	TNFRSF10A	O00220	Receptor; Apoptosis	Cell membrane; Cytoplasm; Membrane	Anti-apoptosis	Yes
	SORT1	Q99523	Developmental protein; Receptor; Differentiation; Endocytosis; Osteogenesis; Transport	Cell membrane; Endoplasmic reticulum; Endosome; Golgi apparatus; Lysosome	Inflammation and vascular calcification	No
	UMOD	P07911	Innate immunity; Cytokine receptor	Cell membrane; Cell projection; Cilium; Membrane; Secreted	Inhibiting vascular calcification	Yes
	HGF	P14210	Growth factor; Serine protease homolog	Secreted; Membrane; Platelet alpha granule lumen	Cardiac protection	Yes
	SPINT1	O43278	Inhibitor of HGFAC; Serine protease inhibitor	Cell membrane; Cytoplasm; Membrane; Secreted	Inhibiting HGF activator	Yes
Limited evidence	AOC3	Q16853	Oxidoreductase; Cell adhesion	Cell membrane; Membrane	Amplifying inflammatory response	Yes
	CD38	P28907	Hydrolase; Receptor; Transferase	Membrane	Intracellular calcium release, vasoconstriction	Yes
	FBLN2	P98095	Calcium-dependent; Adapter linking fibrillin-1 and tropoelastin	Extracellular matrix; Secreted	Vascular remodeling	Yes
	IL6ST	P40189	Receptor; Host-virus interaction	Cell membrane; Membrane; Secreted	Inflammation promotion	Yes
	NRP2	O60462	Developmental protein; Heparin-binding; Receptor; Differentiation; Host-virus interaction; Neurogenesis	Membrane; Secreted	Proliferation of endothelial cells and VSMCs	No
	QSOX1	O00391	Oxidoreductase; Fibroblasts adhesion and migration	Golgi apparatus; Membrane; Secreted	Proliferation of VSMCs	Yes
	RNF149	Q8NC42	Transferase; Ubl conjugation pathway	Membrane	Cellular growth, survival, differentiation, and transformation	No
	SLAMF8	Q9P0V8	B-lineage commitment and B cell receptor signaling	Membrane	Fibroblasts proliferation, migration and inflammatory response	Yes
	FAM3C	Q92520	Developmental protein; Promotes epithelial-mesenchymal transition	Cytoplasmic vesicle; Secreted	Endothelial cell mesenchymal transition	Yes
	MAD1L1	Q9Y6D9	Cell cycle; Cell division; Mitosis	Centromere; Chromosome; Cytoplasm; Cytoskeleton; Kinetochore; Nucleus	Tumorigenesis	No
	MAMDC2	Q7Z304	Glycosaminoglycan binding; Nervous system development	Extracellular matrix; Secreted	Tumorigenesis	Yes

^aBased on the available evidence, we classified the 19 proteins into three categories based on their potential: (1) well-established targets with strong evidence, (2) potential targets with moderate evidence, and (3) emerging targets with limited evidence. Function and subcellular location of proteins can be tracked at UniProt (https://www.uniprot.org/). PAH, pulmonary arterial hypertension; VSMC, vascular smooth muscle cell.

Furthermore, we stratification the 19 candidate proteins into three levels based on the strength of existing evidence linking them to PAH: well-established targets with strong evidence (3 proteins), potential targets with moderate evidence (5 proteins), and emerging targets with limited evidence (11 proteins) (Table 3). This evidence-based stratification facilitates the prioritization of candidates for further functional validation and translational research.

Validation of prioritized targets expression in SuHx rat lungs and plasma

To experimentally validate the proteomic findings, we first examined the expression levels of proteins categorized as “potential targets with moderate evidence” in the lung tissues and plasma of SuHx-induced pulmonary hypertension (PH) rats.

Compared with the control group, TNFRSF10A exhibited significantly upregulated mRNA (Figure 3A) and protein levels (Figures 3B and 3C) in lung tissue of PH rats. Conversely, UMOD expression was significantly downregulated at both the mRNA (Figure 3A) and protein levels (Figures 3B and 3C). Other candidates, including SORT1, HGF, and SPINT1, showed no significant differences between groups. TSA-enhanced immunofluorescence was employed to characterize the spatial distribution of these targets, revealing a marked upregulation of TNFRSF10A (Figure 3D) alongside a significant downregulation of UMOD within the remodeled pulmonary vasculature of SuHx rats (Figure 3E). Additionally, systemic levels via ELISA showed that plasma UMOD levels were significantly lower in PH rats than in controls, a trend that directly mirrors the protective proteomic profile observed in our human cohort (Figure 3F). In contrast, circulating levels of TNFRSF10A showed no significant difference between the groups, despite its marked upregulation observed in lung tissues (Figure 3G).

Figure 3.

Validation of prioritized targets in the lung tissues and plasma of PH rat models

(A) Relative mRNA expression levels of Tnfrsf10a, Umod, Sort1, Hgf, and Spint1 in lung tissues from control and SuHx-induced PH rats, as determined by quantitative real-time PCR (real-time qPCR).

(B and C) Protein expression of TNFRSF10A, UMOD, SORT1, HGF, and SPINT1 validated by western blotting. Representative blots (B) and densitometric quantification (C) demonstrate a significant upregulation of TNFRSF10A and downregulation of UMOD in PH lungs compared with controls.

(D and E) Representative immunofluorescence images of TNFRSF10A (D) and UMOD (E) (red) in lung tissues from control and SuHx rats. Tissues were co-stained with CD31 (green, endothelial marker) and α-smooth muscle actin (α-SMA, cyan, smooth muscle marker). Nuclei were counterstained with DAPI (blue). Scale bars: 50 μm.

(F and G) Plasma levels of TNFRSF10A (F) and UMOD (G) measured by ELISA in control and SuHx rats.

Data are expressed as mean ± SEM of the fold-change relative to the control group (n = 5–10 per group). Significance was determined by unpaired Student’s t-test: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001 and ∗∗∗ p < 0.0001. Con, control; ns, non-significant; ELISA, enzyme-linked immunosorbent assay; PH, pulmonary hypertension; SuHx, Sugen5416 and hypoxia.

Notwithstanding the discrepancy in systemic TNFRSF10A levels, the strong associations observed in our human proteomic analysis and the localized dysregulation of both proteins in PH lung tissues justified their selection for further functional investigation. We therefore prioritized TNFRSF10A and UMOD to explore their roles in hPAECs and hPASMCs, the primary effectors of vascular remodeling.

The impact of UMOD on vascular cell phenotypes

Given that UMOD is a secreted protein, we investigated its functional impact on vascular cells using recombinant human UMOD (rhUMOD) treatment. We initially focused on core pathological phenotypes: angiogenic capacity in hPAECs (via tube formation assays) and migratory potential in hPASMCs (via wound-healing assays).

Intriguingly, rhUMOD treatment (100 and 250 ng/mL) did not significantly alter the total tube length or tube numbers of hPAECs, suggesting a limited role in modulating endothelial angiogenesis (Figure S3). However, rhUMOD treatment markedly impaired the migratory capacity of hPASMCs (Figure 4A). Based on this selective effect, we further explored the impact of UMOD on hPASMCs proliferation. Compared with the control group, rhUMOD treatment significantly reduced hPASMC viability, as assessed by CCK-8 assays (Figure 4B). Consistently, the proportion of EdU-positive proliferating cells was significantly lower in the rhUMOD-treated groups (Figure 4C). Collectively, these findings indicate that UMOD exerts a protective role by specifically restraining the hyperproliferative and migratory phenotypes of smooth muscle cells.

Figure 4.

Impact of rhUMOD treatment on hPASMCs pathological phenotypes

(A) Wound-healing assays demonstrate that rhUMOD treatment (100 or 250 ng/mL) significantly impairs the migratory capacity of hPASMCs compared with the control group (0 ng/mL). Scale bars: 1,000 μm.

(B) CCK-8 assays showing that rhUMOD treatment reduces the cell viability of hPASMCs (n = 5).

(C) Representative images and quantification of EdU incorporation assays. rhUMOD treatment significantly inhibits hPASMC proliferation (n = 5). Nuclei were counterstained with DAPI (blue), and EdU-positive cells (green) indicate actively proliferating cells. Scale bars: 100 μm.

Data are expressed as mean ± SEM. Significance was determined by one-way ANOVA analysis: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001 and ∗∗∗∗ p < 0.0001. hPASMCs, human pulmonary artery smooth muscle cells; ns, non-significant; rhUMOD, recombinant human UMOD protein.

The impact of TNFRSF10A knockdown on vascular cell phenotypes

To explore the role of the cell membrane receptor TNFRSF10A, we performed small interfering RNA (siRNA)-mediated knockdown in both hPAECs and hPASMCs. Among four candidate siRNAs designed, siTNFRSF10A-3 and siTNFRSF10A-4 were the most effective, and siTNFRSF10A-3 also showed excellent silencing efficiency (Figure S4). Nevertheless, because the target sites of siTNFRSF10A-3 and siTNFRSF10A-4 are too closely located, siTNFRSF10A-2 and siTNFRSF10A-3 were chosen for further experiments. Then, we also examined the key phenotype that was involved in PAH pathogenesis, including tube formation assays for hPAECs and wound-healing assays for hPASMCs. Similarly, knock down of TNFRSF10A still did not significantly affect the tube formation capacity of hPAECs (Figure S3) while promote the migratory ability of hPASMCs (Figure 5A). Furthermore, siRNA-mediated knockdown of TNFRSF10A appeared to promote, rather than inhibit, the PAH-like phenotype. Specifically, TNFRSF10A silencing led to a significant increase in cell viability (Figure 5B) and the proportion of EdU-positive cells in hPASMCs (Figure 5C).

Figure 5.

Impact of TNFRSF10A knockdown on hPASMCs pathological phenotypes

(A) Wound-healing assays show that TNFRSF10A knockdown significantly accelerates the migratory capacity of hPASMCs compared with the negative control (siNC) group. Scale bars = 1,000 μm.

(B) CCK-8 assays demonstrating that silencing of TNFRSF10A significantly enhances hPASMCs viability (n = 5).

(C) Representative images and quantification of EdU incorporation assays. TNFRSF10A knockdown (siTNFRSF10A-3) significantly promotes DNA synthesis in hPASMCs (n = 5). Nuclei were counterstained with DAPI (blue), and EdU-positive cells (green) indicate actively proliferating cells. Scale bars: 100 μm.

Data are expressed as mean ± SEM. Significance was determined by one-way ANOVA analysis: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001 and ∗∗∗∗ p < 0.0001. hPASMCs, human pulmonary artery smooth muscle cells; ns, non-significant; siNC, negative control siRNA.

Discussion

Through integrating large-scale plasma proteomics with genomic causal inference, we identified and validated potential therapeutic targets for PAH. To the best of our knowledge, this is the first study to define a proteomic signature for incident PAH within a general population, linking 398 out of 2,923 screened proteins to disease risk. MR analysis implicated 19 proteins as causal, 15 of which are currently categorized as druggable. Preclinical validation of prioritized targets in SuHx-induced PH model demonstrated the significant upregulation of TNFRSF10A and downregulation of UMOD in lung tissues, consistent with the trends observed in human plasma proteomics. Subsequent in vitro functional assays further supported that UMOD treatment attenuates PAH-related cellular phenotypes, while TNFRSF10A knockdown appears to drive pro-pathogenic progression. Collectively, these results provide a multi-level evidence base for the involvement of these candidates in pulmonary vascular remodeling.

The identification of well-established targets, specifically B-type natriuretic peptide (BNP and NPPB) and endothelin-1 (EDN1), further reinforces the validity of our analytical approach. Currently, BNP and its precursor fragment NT-proBNP remain the most widely utilized biomarkers for early detection and risk stratification in clinical PAH management.¹ Similarly, ET-1, a potent vasoconstrictor implicated in vascular remodeling, inflammation, and fibrosis, represents a cornerstone of PAH therapeutics, as evidenced by the clinical success of endothelin receptor antagonists, such as ambrisentan, bosentan, and macitentan.^1,13

Furthermore, our study also supports the emerging role of NOTCH3 signaling, a pathway central to pulmonary vascular remodeling in PAH pathogenesis. Unlike conventional therapies that primarily target vasoconstriction, interventions targeting the vascular remodeling address a key driver of disease progression, marked by excessive vascular smooth muscle cells (VSMCs) proliferation and inflammatory activation.¹⁴ NOTCH3, a signaling receptor involved in cell proliferation and differentiation, is almost exclusively expressed in VSMCs.¹⁵ Previous studies have shown that NOTCH3 is overexpressed in the VSMCs of lung tissue from PAH patients, correlating with disease severity.¹⁶ Experimental evidence further indicates that the activation of NOTCH3 and its downstream effector Hes5 promotes a proliferation and undifferentiated phenotype transition of VSMC, contributing to vascular remodeling in small precapillary pulmonary arteries.¹⁵ Importantly, targeting NOTCH signaling, such as γ-secretase inhibitors and monoclonal anti-NOTCH3 antibodies, have shown promise in reversing experimental PH in murine models.^16,17,18 Of note, the FDA-approved γ-secretase inhibitor Nirogacestat (Table S9), currently used for desmoid tumors, provides a compelling opportunity for drug repurposing in PAH. In future, prospective clinical trials are required to evaluate its safety and efficacy in this population.

UMOD was identified as the sole candidate in our study to exhibit a putative protective effect against PAH. In support of this role, we observed a significant reduction in UMOD levels in both the plasma and pulmonary arteries of SuHx rats, a finding consistent with the loss of a protective mediator during disease pathogenesis. Although UMOD is a renal-specific protein synthesized by the kidneys, it also undergoes basolateral release into the systemic circulation.¹⁹ Elevated circulating UMOD level has been associated with reduced risks of chronic kidney disease, systemic infection, and cardiovascular events.^19,20 While most evidence linking UMOD to cardiovascular diseases derives from observational studies, emerging mechanistic data indicate that UMOD exerts potent immunomodulatory and anti-calcification effects, including vasoprotective actions in human VSMCs through the inhibition of pro-inflammatory cytokine signaling, such as TNF-α and IL-1β.²¹ Vascular calcification, driven by the phenotypic transition of VSMCs in response to pathological stimuli,^22,23 is increasingly recognized as a hallmark of PAH pathogenesis that frequently manifests as calcified lesions in the distal pulmonary arteries, even at the earliest stages of the disease.²⁴ Our in vitro data demonstrated that UMOD treatment attenuates pathological changes primarily in hPASMCs, further supporting its protective role. Collectively, these findings suggest that UMOD is a potential protective factor, though the precise molecular pathways regulating its action in PAH warrant further investigation.

Our findings regarding TNFRSF10A underscore the complex role of the TRAIL receptor system in PAH. Paradoxically, while our human proteomics and expression validation in PH rats established elevated TNFRSF10A as a risk factor, its siRNA-mediated knockdown in hPASMCs exacerbated pro-pathogenic phenotypes. Although TNFRSF10A (also known as DR4 or TRAIL-R1) is classically characterized as one of the pro-apoptotic receptor for TRAIL in tumor cells, the TRAIL system exerts divergent, pleiotropic effects within the cardiovascular field.^25,26,27 In human vascular endothelial cells and smooth muscle cells, TRAIL does not induce apoptosis but instead exerts anti-apoptotic and mitogenic effects, promoting the survival, proliferation, and migration of endothelial cells and VSMCs through the activation of NF-κB, Akt, and ERK pathways.^28,29,30,31 Furthermore, TRAIL has been demonstrated as a potent stimulus for pulmonary vascular remodeling, and its pharmacological blockade effectively prevents PAH development in preclinical animal models.³² The exacerbated PAH-like phenotype observed upon TNFRSF10A knockdown may stem from a “receptor switching” mechanism. The loss of TNFRSF10A likely redirects endogenous TRAIL to alternative receptors (e.g., DR5 or specific decoy receptors, such as DcR1 and DcR2) that possess stronger signal amplification capabilities, thereby triggering an intensified, compensatory pro-proliferative cascade.^33,34,35 Beyond this specific signaling switch, the observed upregulation of TNFRSF10A in our PH model may represent a broader endogenous compensatory response aimed at mitigating pulmonary vascular remodeling during disease pathogenesis. Taken together, these findings underscore the complexity of the TRAIL/TRAIL-receptor network, suggesting that although ligand blockade remains a promising therapeutic strategy in PAH, targeting a single receptor may be insufficient to fully interrupt its pathogenic effects.

For the remaining “moderate evidence” targets that did not exhibit significant expression changes in our rat validation, this discrepancy may reflect a common translational challenge, as human-derived associations do not always replicate in animal systems due to interspecies differences or the involvement of systemic rather than purely pulmonary mechanisms. Although these targets may warrant lower priority for validation in this model context, their potential contribution to human PAH pathogenesis cannot be entirely excluded. Moreover, proteins categorized as limited evidence remain largely uncharacterized in pulmonary vascular research, yet available data suggest their possible involvement in vascular remodeling-related processes. Of note, PAH shares several hallmark characteristics with cancer biology, including enhanced cellular plasticity, migration, proliferation, and inflammation.³⁶ This biological overlap raises the possibility of repurposing established oncogenic drivers as PAH targets. For example, FAM3C promotes epithelial-mesenchymal transition in cancer, analogous to endothelial-mesenchymal transition in PAH.³⁷ Similarly, MAD1L1 and MAMDC2 are implicated in tumorigenesis but remain largely unexplored in cardiovascular contexts.

In summary, we established a distinct molecular signature for incident PAH by integrating large-scale plasma proteomics with genomic causal inference. Validation of prioritized targets confirmed the dysregulation of TNFRSF10A and UMOD in the SuHx rat model, the subsequent in vitro functional investigations identified UMOD as a promising protective therapeutic candidate and revealed the inherent complexity of the TNFRSF10A signaling network. Collectively, our findings provide robust evidence for translational research. Future studies are warranted to fully elucidate the downstream molecular mechanisms of these candidates and to evaluate their therapeutic efficacy in additional in vivo models and prospective clinical trials, ultimately aiming to improve clinical outcomes for patients with PAH.

Limitations of the study

This study presents a large-scale proteomic profiling of incident PAH within a general population, providing robust observational, causal, and experimental evidence to prioritize targets for future translational research. However, several limitations warrant consideration. First, PAH identification in the UKB relies on International Classification of Diseases, 10th Revision (ICD-10) coding rather than hemodynamic measurements, which may capture individuals with group 2 or 3 PH and introduce diagnostic bias, a recognized challenge in biobank-scale research.³⁸ To rigorously address this, we performed a sensitivity analysis entirely excluding participants with left-heart or parenchymal lung diseases (detailed ICD-10 codes are provided in part A of Table S10) prior to PAH diagnosis; notably, the associations of our targets with PAH remained robust (part B, Table S10). We also validated the expression and spatial distribution of prioritized targets in lung tissues from the SuHx rat model (a system that recapitulates severe human-like vascular remodeling³⁹) and subsequently confirming their phenotypic effects in human pulmonary vascular cells. Second, the higher proportion of males in our baseline PAH cohort (51.63% vs. 45.98%, p=0.044) contrasts with the female predominance typically seen in clinical registries. This discrepancy likely reflects the older age of the UKB cohort (recruitment age ≥40 years). Young females with early onset, severe disease may be underrepresented; moreover, epidemiological evidence demonstrates that the pronounced female-to-male PAH incidence ratio observed in young adults approaches parity with advancing age.^40,41 Third, the reliance on GWAS and cohort data primarily from individuals of European descent may limit the generalizability of these findings to other ethnic groups. Fourth, a minor degree of sample overlap between the UKB and the PAH GWAS cannot be entirely ruled out given the use of summary-level data, but the consistently high F-statistics indicate that our causal estimates remain robust against potential bias. Finally, although our multi-stage approach identifies candidates with multi-level support, in vivo pharmacological intervention studies and prospective clinical cohorts are necessary to define the therapeutic windows, safety profiles, and long-term potential of these targets.

Resource availability

Lead contact

Requests for further information should be directed to and will be fulfilled by the lead contact, Dr. Lu Hua (ethannan@126.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •The UKB-PPP GWAS summary statistics can be found at https://metabolomips.org/ukbbpgwas/; eQTL summary data from the eQTLGen (https://www.eqtlgen.org/) and the GTEx Consortium V8 (https://gtexportal.org/), and the GWAS summary statistics related to PAH can be downloaded at GWAS Catalog (https://www.ebi.ac.uk/gwas/). These data are publicly available as of the date of publication.
• •This study was conducted using data from the UKB-PPP cohorts under application number 103409.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request

Acknowledgments

This work was supported by grants from the National Clinical Research Center of Cardiovascular Diseases, Fuwai Hospital Chinese Academy of Medical Sciences, Shenzhen (NCRCSZ-2025-009 and NCRCSZ-2023-015), Shenzhen Medical Research Fund (C2501036), Joint Funds for the innovation of science and Technology, Fujian province (2024Y96010029), National Natural Science Foundation of China (8247023090 and 82500078), High-Level Hospital Clinical Research Funding (2025-GSP-QN-27), and Capital’s Funds for Health Improvement and Research (2024-1-4031).

Author contributions

S.H., Y.X., J.-S.T, T.-T.G., S.Z., and Z.L., conceptualization, data curation, formal analysis, investigation, methodology, software, and writing – original draft; S.H., Y.X., and Z.L., validation and visualization; Z.L., funding acquisition; L.H. and J.Z., conceptualization, methodology, funding acquisition, supervision, and writing – review and editing.

Declaration of interests

The authors have no relevant financial or non-financial interests to disclose.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-TNFRSF10A	Invitrogen	Cat#PA5-87131; RRID: AB_2803881
Anti-UMOD	Bioss	Cat#BS-2189R
Anti-SPINT1	Abcam	Cat#ab272721
Anti-HGF	Invitrogen	Cat#PA5-115354; RRID: AB_2899990
Anti-SORT1	Proteintech	Cat#12369-1-AP; RRID: AB_2302242
Anti-β-actin	Proteintech	Cat#20536-1-AP; RRID: AB_10700003
Anti-CD31	Abcam	Cat#ab182981
Anti-α-SMA	Abcam	Cat#ab7817
HRP Conjugated AffiniPure Goat Anti-mouse IgG (H+L)	BOSTER	Cat#BA1050
Anti-rabbit IgG, HRP-linked Antibody	Cell Signaling Technology	Cat#7074
HRP-labeled goat anti-rabbit secondary antibody	Servicebio	Cat#GB23303
Biological sample
Sprague-Dawley rats	BEIJING HFKBIOSCIENCE CO., LTD	N/A
Chemicals, peptides, and recombinant proteins
rhUMOD	Proteintech	Cat#Ag27750
Critical commercial assays
Reverse transcription kit	Transgen	Cat#AT341-01
SYBR Green Master Mix	Transgen	Cat#AQ601-01-V2
BCA protein assay kit	Thermo Fisher	Cat#23225
Tanon™ High-sig ECL Western BlottingTMSubstrate	Tanon	Cat#180-5001
Rat TNFRSF10A ELISA Kit	Elabscience	Cat#ELK9377
Rat UMOD ELISA Kit	Elabscience	Cat#ELK9377
Cell Counting Kit-8	Beyotime	Cat#C0039
BeyoClick™ EdU Cell Proliferation Kit with Alexa Fluor 488	Beyotime	Cat#C0071L
TSA PLus Fluorescence Triple Staining Kit	Servicebio	Cat#G1236-50T
Incucyte® Cell Migration Kit	Sartorius	Cat#BA-04858
Invitrogen™ Lipofectamine™ RNAiMAX	Invitrogen	Cat#13778150
Deposited data
UKB-PPP GWAS summary statistics	Plasma proteomic associations with genetics and health in the UK Biobank	https://metabolomips.org/ukbbpgwas/
eQTL summary data	eQTLGen	https://www.eqtlgen.org/
GTEx Consortium V8	GTEx Consortium	https://gtexportal.org/
PAH GWAS summary statistics	GWAS Catalog	https://www.ebi.ac.uk/gwas/
UKB-PPP cohorts data	UK Biobank	Available upon request
Deposited data
N/A	N/A	N/A
Experimental models: Cell lines
Human pulmonary artery endothelial cells	iCell Bioscience Inc	HUM-iCell-a008
Human pulmonary arterial smooth muscle cells	iCell Bioscience Inc	HUM-iCell-a009
Oligonucleotides
Primer for qPCR, see Methods S1	This paper	N/A
Sequence for siRNA, see Methods S1	This paper	N/A
Software and algorithms
R software (4.0.5)	R	https://www.r-project.org/
SMR software tool (1.3.1)	SMR	https://yanglab.westlake.edu.cn/software/smr/
ImageJ software (1.52)	ImageJ	https://imagej.net/ij/
GraphPad Software (8.0)	GraphPad Software	https://www.graphpad.com/
Incucyte® SX5 Live-Cell Analysis System	Sartorius	https://www.sartorius.com.cn/

Experimental model and study participant details

Human participants

In this study, we utilized data from the UKB Pharma Proteomics Project (UKB-PPP), which characterized the plasma proteomic profiles of 54,219 UKB participants using the Olink Explore 3072 panel. To align with the GWAS summary data used for downstream causal inferences (which were derived solely from European populations), our study specifically included White European individuals without PAH at baseline. This cohort was randomly split into discovery and replication groups in a 7:3 ratio (n=34,492 for the discovery cohort; n=14,780 for the replication cohort). The UKB received approval from the North West Multi-Centre Research Ethics Committee (11/NW/0382). All participants provided written informed consent for participation and follow-up, and the study was conducted according to the guidelines of the Declaration of Helsinki.

Animal models

To biologically validate the targets identified from the human proteomic analysis, a SuHx-induced PH rat model was established. Eight-week-old male Sprague-Dawley rats were utilized for the experiments. All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC #0109-7-90-ZX(X)-25) and were performed in strict accordance with the ARRIVE guidelines. Rats were randomized into control or PH groups (n=5–10 per group). PH was induced via a single subcutaneous injection of Sugen5416 (20 mg/kg) followed by three weeks of chronic hypoxia (10% O₂). The animals were then transferred to a normoxic environment (21% O₂) for an additional two weeks. The control group received a subcutaneous injection of an equal volume of dimethyl sulfoxide and was maintained in a normoxic environment for the entire five-week period. Rats were housed in standard pathogen-free laboratory conditions with a 12-h light/dark cycle and ad libitum access to food and water.

Primary cell cultures

hPAECs and hPASMCs were commercially acquired from iCell Bioscience Inc. (Shanghai, China). The cells were authenticated and characteristically certified by the manufacturer upon purchase. hPASMCs were authenticated by immunofluorescence staining for α-smooth muscle actin (α-SMA) using a rabbit polyclonal anti-α-SMA antibody (55135-1-AP; Proteintech; 1:100). The percentage of α-SMA-positive cells was >95%. Similarly, hPAECs were authenticated by immunofluorescence staining for vWF using a mouse anti-vWF antibody (66682-1-Ig; Proteintech;100), with a positive cell percentage >95%. To ensure biosafety, mycoplasma contamination of all cell lines was screened by PCR assay in four independent cell batches, and all batches were negative. Cells between passages 3 and 7 were utilized for all in vitro experiments. hPAECs were maintained in Endothelial Cell Medium (ECM; Thermo Fisher Scientific, US) supplemented with 5% fetal bovine serum (FBS), 1% endothelial cell growth supplement (ECGS), and 1% penicillin/streptomycin. hPASMCs were cultured in DMEM/F12 (Thermo Fisher Scientific, US) supplemented with 10% FBS and 1% penicillin/streptomycin. All cell cultures were incubated at 37°C in a humidified atmosphere containing 5% CO₂.

Method details

Study design

Between 2006 and 2010, the UKB recruited over 500,000 participants aged 40–69 years from 22 assessment centers throughout the UK. The goal was to establish a large prospective cohort with extensive phenotypic and genotypic data. Each participant provided electronic signed consent, and a large number of data were collected at baseline, including self-reported information on medications and health, physical measurements and health-related factors that were collected by answered touchscreen and verbal interviews.⁴² The ongoing follow-up of participant’s health was performed via linkage to a range of national datasets.⁴² Moreover, the UKB has launched the UKB-PPP. It is a collaborative effort involving 13 biopharmaceutical companies to characterize the plasma proteomics profiles of 54,219 UKB participants using the antibody-based Olink Explore 3072 panel, measuring 2,923 unique proteins.⁴³

The study design is illustrated in Figure S1. Given that the GWAS data used to explore the causal associations of proteins with PAH were based solely on European participants, we included White European individuals without PAH at baseline from the UKB-PPP cohort. The cohort was randomly split into discovery and replication groups in a 7:3 ratio, with 34,492 assigned to the discovery cohort and the remaining 14,780 to the replication cohort. We then assessed the association between 2,923 proteins and incident PAH. Proteins showing significant associations in both the discovery and replication cohorts were selected for further causal evaluation. To explore potential causal effects of these proteins on PAH, we performed a two-sample MR analysis using QTL (pQTL and eQTL) GWAS summary data. We selected significant proteins that exhibited directional consistency between MR and Cox proportional hazards estimates as candidate protein targets. This conservative filtering strategy was employed to identify robust disease drivers and potential therapeutic targets while minimizing the confounding effects of complex feedback loops or secondary compensatory responses. Subsequently, the “druggability” of the candidate proteins was evaluated. Finally, to corroborate our cohort and genomic findings, we validated the prioritized targets in PH rats and human vascular cells.

Study variables and outcomes

The detailed definitions of study variables are listed in Table S11. Briefly described, some were defined as follows: (1) Body mass index: calculated as weight (kg) divided by height squared (m²). (2) The diagnosis of hypertension, diabetes mellitus, and dyslipidemia were acquired by ICD-10 codes, and individuals with their diagnosed date before or at their first assessment were considered to have these diseases at baseline. Additionally, responses to the questionaries regarding “Vascular/heart problem diagnosed by doctor”, “Diabetes diagnosed by doctor”, and “Medication for cholesterol, blood pressure, diabetes” were also extracted to replenish their medical history. (3) Smoking and alcohol drinking status were categorized as never, previous and current smoking or drinking. (4) Education level was classified into three categories: high (college or university degree), intermediate (A/AS levels or equivalent, O levels/CSEs or equivalent), and low (none of the aforementioned). (5) Ideal physical activity: at least 150 minutes of moderate-intensity, or 75 minutes of high-intensity physical activity per week, as WHO suggested.

The primary outcome was incident PAH, identified using inpatient ICD-10 codes (I27.0 for idiopathic PAH). In UKB, ICD-10 diagnosis codes are extracted from hospital inpatient records, and each code is recorded only after standard diagnostic workflow in participating hospitals. Participants were followed up until PAH was diagnosed, occurrence of death, or the last day of follow-up (31/Oct/2022). The follow-up period was calculated as the difference between the censoring date and the date of the participant’s initial assessment at the recruitment center.

Functional enrichment analysis

To elucidate the biological functions and potential therapeutic pathways associated with the proteins identified, GO and KEGG enrichment analyses were conducted using the R package “clusterProfiler”.⁴⁴ GO analysis was performed across three categories: BP, CC, and MF. KEGG pathway enrichment analysis was simultaneously carried out to identify potential signaling cascades. Statistical significance was stringently defined as an adjusted p-value < 0.05 and a Q-value < 0.05, with multiple testing correction performed using the Benjamini–Hochberg method.

Drug-target mendelian randomization analysis

Drug-target MR simulates the genetic variation of pharmacological inhibition of drug-gene targets to investigate the role of specific proteins in disease. By using cis-instruments, this method enables large-scale analysis of thousands of proteins across diseases, advancing more effective drug development when the target protein is druggable. The causal effects of proteins on PAH were evaluated using drug-target MR analysis, with QTLs (pQTLs and eQTLs) serving as IVs. Specifically, cis-acting single nucleotide polymorphisms (SNPs), which are located in or near a gene and can influence the activity or expression of its encoded protein, were used to infer the potential impact of pharmacological interventions on PAH.

In the pQTLs MR analysis, cis-pQTLs were defined as variants within ± 500 kb from the transcription start site with a genome-wide significant association (p < 5 × 10^-8) with plasma protein levels.⁴⁵ To ensure the independence of the selected IVs, we performed linkage disequilibrium clumping with a threshold of r² < 0.2 within a 10,000 kb window, based on the European reference panel from the 1000 Genomes Project. Moreover, SNPs with an F-statistic > 10 [F-statistic=(β/SE)²] were retained to mitigate weak instrument bias. Only proteins with at least three independent IVs after clumping were included in the primary analysis to ensure the stability and robustness of the causal estimates. The pQTL summary statistics for 2,923 plasma proteins were obtained from the UKB-PPP study.⁴³ Moreover, PAH summary statistics were derived from the largest available GWAS, comprising 11,744 individuals of European ancestry (2,085 PAH cases, 88% cases were diagnosed with idiopathic PAH).⁴⁶ The PAH summary statistics were derived from several specialized clinical cohorts.⁴⁶ While these cohorts included participants from the UK, the extent of potential sample overlap with the UKB remains unconfirmed due to the use of summary-level data. Nonetheless, potential bias is minimized as these GWAS individuals were recruited through specialized clinical registries rather than general population frameworks of the UKB. To mitigate potential bias arising from such overlap, we restricted our MR analyses to IVs with an F-statistic > 10. The distribution of F-statistics is provided in Table S5; the consistently high F-statistics (mean: 76.4, range: 30.0–13425.4) observed across our prioritized targets indicate that the causal estimates are robust and less likely to be driven by weak instrument bias or sample overlap.

For each IV, a variant-specific Wald ratio method was used to estimate the causal effect. The IVW method was then employed to meta-analyze these estimates for proteins with multiple variants, with results expressed as ORs and 95% CIs per SD increase in genetically predicted protein levels. Other MR methods, including MR egger regression, weighted median, weighted mode and simple mode, were performed as additional sensitivity analyses. Causal associations were considered robust if the IVW estimate showed an FDR-p < 0.05 and the results from the other MR methods were directionally consistent with IVW. Additionally, heterogeneity of the IVs was assessed using Q statistics, and potential horizontal pleiotropy was tested with MR Egger regression. p < 0.05 indicated possible heterogeneity or pleiotropy. In cases of detected heterogeneity, the IVW method with a multiplicative random effect was used.

To further support the causal associations identified by the pQTL MR analysis, we also performed an eQTL MR analysis. The SMR method was used to assess whether gene expression (mRNA levels) was causally associated with PAH. The causal estimates were expressed as OR and 95% CI for one SD in unit, and p < 0.05 were considered significant. In this analysis, only available cis-eQTLs were used as proxies for candidate proteins. Similarly, cis-eQTLs were defined as those located within 1 Mb on either side of the encoding gene. The common [minor allele frequency (MAF) >1%] eQTLs SNPs that were significantly (p < 5.0×10^-8) associated with the expression of candidate protein genes (mRNA level) in whole blood or other available tissues (except reproductive system) were obtained from the eQTLGen and the GTEx Consortium V8. The PAH GWAS summary statistics were the same as those used in the pQTL MR analysis. Additionally, the HEIDI test was performed to examine if the association between the mRNA level of proteins and PAH was due to linkage, which is indicated by a HEIDI test of p < 0.01.⁴⁷

Sensitivity analysis

We performed two sensitivity analyses for the associations between proteins and incident PAH. First, since competing risk should be considered given that mortality may preclude the diagnosis of PAH, we applied the Fine & Gray competing risk model, an approach to resolve the competing risk of mortality.⁴⁸ Sub-distribution hazard ratios (SHRs) was calculated in a multivariate manner, to examine the associations between proteins and incident PAH in the entire cohort. Second, missing data for covariates were imputed using a random forest imputation method, after which the associations between proteins and incident PAH were re-evaluated.

Classification of target proteins

To prioritize the identified proteins according to the robustness of existing evidence, we stratified them into three levels:

• (1)Well-established targets with strong evidence: supported by strong clinical or experimental data, often serving as recognized mature biomarkers or therapeutic targets, or previously validated in PH models;
• (2)Potential targets with moderate evidence: supported by moderate evidence from cardiovascular research but lacking validation in PAH;
• (3)Emerging targets with limited evidence: for which evidence remains limited and their roles in cardiovascular and pulmonary vascular diseases are largely unexplored.

Validation of prioritized targets

To bridge our proteomic findings with biological relevance, candidate proteins categorized as “potential targets with moderate evidence” were selected for experimental validation. The SuHx rat model was employed in this study to induce PAH, because it reliably produces severe and irreversible PH accompanied by the development of occlusive vascular remodeling and plexiform-like lesions, which are critical pathological hallmarks of human PAH that are not consistently observed in other models.³⁹ Protein and mRNA expression levels were quantified in rat lung tissues via Western blot and qPCR, while cellular localization and plasma levels were assessed through TSA-enhanced immunofluorescence and ELISA, respectively. Furthermore, the functional roles of TNFRSF10A and UMOD were characterized in hPAECs and hPASMCs using siRNA-mediated silencing and recombinant protein treatment, respectively. Phenotypic alterations, including cell viability, proliferation, migration, and angiogenic tube formation, were evaluated via CCK-8, EdU incorporation, wound healing assay and in vitro tube formation assay. All procedures were approved by the Institutional Animal Care and Use Committee (IACUC #0109-7-90-ZX(X)-25), adhering to ARRIVE guidelines. Further method details kregarding the STAR Methods can be found in the supplemental information (Methods S1).

Quantification and statistical analysis

Cohort analysis

Continuous variables were presented as mean ± SD or median (interquartile range) according to their distributions. The differences between groups were compared by t-test for normalized distributed variables, otherwise a Wilcoxon rank-sum test was used. Categorical variables were shown as frequency (proportion, %), and the differences were compared by Chi-square test. Multivariate Cox proportional hazards model was used to explore the associations between the plasma protein level and incident PAH in both the discovery and replication groups, with HRs and 95% CIs estimated. To minimize the impact of potential confounders, we adjusted for key demographics and cardiometabolic risk factors in the multivariate model, including age, sex, body mass index, hypertension, diabetes mellitus, dyslipidemia, smoking status, alcohol consumption, education level, and physical activity. The proteins that showed significant associations with incident PAH (FDP-p < 0.05) in both the discovery and replication cohorts were selected for further causal evaluation.

All analysis were performed by using the R software (R version 4.0.5; R Foundation for Statistical Computing, Vienna, Austria). The R packages “TwoSampleMR” were used for pQTLs MR analysis, and the SMR software (version 1.3.1) was used for SMR analysis (https://yanglab.westlake.edu.cn/software/smr/). All p-values are two-sided, and as described above, p value or FDR-p < 0.05 demonstrated statistical significance.

Experimental analysis

Quantitative analysis was performed using ImageJ software (version 1.52, NIH, USA). Representative figures reflected the average level of each experiment. For Western blot and qPCR, the relative expression levels were normalized to β-actin. For immunofluorescence, the relative fluorescence intensity was quantified from at least three random high-power fields per section. Wound healing was assessed by calculating the percentage of wound closure area, while tube formation was quantified by determining the total tube length.

All statistical analyses were performed using GraphPad Prism (version 8.0; GraphPad Software Inc., La Jolla, CA). Data are presented as mean ± standard error (SEM) from at least five independent replicates (n = 5) for both animal and cell experiments. Comparisons between two groups were analyzed using an unpaired, two-tailed Student’s t-test. For multiple comparisons, one-way ANOVA was employed. A p-value < 0.05 was considered statistically significant. Significance levels are indicated in the figures as follows: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001 and ns indicated no statistical difference.