Integrated Single-Cell Atlas Reveals Cross-Subtype Heterogeneity in Human Pulmonary Hypertension

BACKGROUND: Pulmonary hypertension (PH) is a severe vascular disorder characterized by pathological pulmonary vascular remodeling. Although single-cell RNA sequencing has provided insights into PH, patient studies remain fragmented across clinical subtypes and are processed with heterogeneous pipelines, limiting cross-study comparability and integrative interpretation of disease-relevant cellular states in human PH. METHODS: We integrated single-cell RNA sequencing data sets to construct a core PH-specific cellular atlas using a standardized quantification and batch-correction pipeline, followed by hierarchical cell type annotation to define major lineages and finer cell states. We then performed comparative analyses to reassess reported cellular alterations within a unified reference. Building on this atlas, we developed PH-Map, a multitask learning tool pretrained on this atlas for rapid hierarchical cell annotation and validated it in independent external cohorts. RESULTS: The core PH-specific atlas comprises 235 621 high-quality cells from 64 human samples spanning idiopathic pulmonary arterial hypertension, systemic sclerosis-associated pulmonary arterial hypertension, chronic thromboembolic PH, and healthy controls, organized into 6 major lineages and 58 distinct cell states. Cross-subtype comparison revealed shared and subtype-specific remodeling across vascular and immune compartments, including redefined smooth muscle cell states, fibroblast-associated complement activation in idiopathic pulmonary arterial hypertension, and heterogeneous immune transcriptional programs. We additionally observed system-dependent differences, including variable major histocompatibility complex class II activation in capillary endothelial cells across animal models versus human PH and convergent phenotypic shifts in primary cultured endothelial cells relative to native lung endothelial cells. PH-Map showed stable hierarchical annotation performance in external cohorts across multiple levels of annotation granularity with improved computational efficiency. CONCLUSIONS: This core human single-cell PH atlas enables systematic cross-subtype comparison and delineates shared and subtype-specific cellular programs across vascular and immune compartments. Together with PH-Map, it supports reproducible hierarchical annotation and prioritization of candidate cell states and pathways for downstream studies.