Interpreting Pulmonary Hypertension Beyond Single Cells

Pulmonary hypertension (PH) is a progressive cardiopulmonary disease characterized by pulmonary vascular remodeling, increased pulmonary vascular resistance, and eventual right ventricular failure^1,2. Key features include endothelial dysfunction, smooth muscle cell (SMC) phenotypic switching, fibrosis, inflammation, thrombosis, and immune remodeling, although their relative contributions differ across PH subtypes and models. As single-cell RNA sequencing (scRNA-seq) datasets continue to expand, a major challenge is no longer data generation but consistent interpretation of disease-associated cell states across studies, species, and tissue sources.

Feng and colleagues address this challenge by creating an integrated human PH single-cell atlas and developing PH-Map, a hierarchical annotation framework for PH single-cell datasets³. The value of PH-Map is its disease-specific reference structure. In PH, cell annotation is not simply technical: diseased endothelial cells (ECs), smooth muscle cells (SMCs), fibroblasts, and immune cells can acquire inflammatory, proliferative, metabolic, or remodeling programs that blur normal cellular identities. A PH-specific reference may therefore improve cross-study comparability, reduce inconsistent labeling of related disease states, and accelerate analysis of new datasets. In large atlases, harmonization can also reveal rare or transitional populations that are not easily detected in smaller studies. However, PH-Map should be viewed as a reference-mapping tool, not a substitute for biological validation by marker analysis, spatial localization, histology, and functional studies.

Several findings illustrate this value. Feng and colleagues identify CXCL12+ ECs as a recurrent disease-associated state across PAH subtypes, enriched in digital spatial profiling-defined plexiform lesion regions³. They also show that MHC-II activation in capillary ECs, previously observed in rodent PH models, is not consistently observed across human PH datasets or animal models^{3, 4}. Animal models can reveal meaningful biology, but model-specific programs should not be assumed to represent human disease without systematic comparison. In the stromal compartment, the atlas confirms loss of contractile CNN1+ SMCs in IPAH and SSc-PAH and refines several dedifferentiated SMC states³, consistent with prior evidence of smooth muscle heterogeneity and altered vascular-immune communication⁵. The fibroblast and immune findings also argue against a simple remodeling program. Although complement activation in fibroblasts has been proposed in PH⁶, C3 expression is concentrated in adventitial fibroblasts and appears most evident in IPAH³. Cytotoxic lymphocyte, NK-cell, and macrophage programs also vary by subtype, consistent with the distinct mechanisms underlying IPAH, SSc-PAH, and CTEPH.

A key limitation is that the atlas primarily reflects advanced human PH, as most datasets derive from transplantation or end-stage surgical specimens with incomplete clinical metadata on hemodynamics, treatment exposure, disease duration, genetic background, and precise sampling sites³. Thus, atlas-defined cellular states should not be automatically interpreted as early causal events, disease-stage markers, anatomical distributions, or direct evidence of lineage transitions. This distinction is particularly important for cell fate plasticity. Although recent studies suggest that capillary ECs may acquire arterial features through HIF-2α/Notch4 signaling in PH⁷, the current atlas does not directly demonstrate capillary-to-arterial transition, endothelial-to-mesenchymal transition, or smooth muscle lineage conversion. Rather, it provides a framework for testing these hypotheses. While datasets of this scale can support trajectory inference, regulon analysis, and ligand–receptor modeling^{8, 9}, trajectories derived from cross-sectional end-stage tissues should be interpreted as cell-state continua not direct temporal evidence of disease progression.

These cautions reflect broader limitations of single-cell RNA in cardiovascular diseases. Precise anatomical information is often limited, despite substantial regional variation in cellular composition and gene expression. Disease severity also varies between patients and across regions within the same specimen. Fibrosis, necrosis, dissociation efficiency, and cell loss can strongly influence recovered cell populations. Integration of normal and diseased tissues during clustering analyses can further complicate interpretation, because clustering may reflect altered cellular composition or tissue remodeling rather than authentic transcriptional state changes (Figure).

Figure: Challenges and future directions for single-cell RNA-seq analysis in CVD.

(1) Key limitations of scRNA-seq in CVD include regional sampling bias, heterogeneous tissue remodeling, loss of normal cellular architecture, and difficulty integrating samples across disease stages. Unlike cancer, CVD tissues undergo extensive fibrosis, cell loss, and structural remodeling, requiring analytical frameworks that account for altered tissue composition and spatial context.

(2) Future approaches should integrate spatial multi-omics, including histology, spatial transcriptomics, proteomics, and metabolomics, with clinical data such as hemodynamics, treatment history, and longitudinal outcomes to generate spatially and clinically anchored PH/CVD digital twins. This framework may improve understanding of disease-associated cellular states, cell-cell interactions, tissue remodeling, and therapeutic responses. Created with assistance from Biorender.

Analytical strategies developed for oncology may not be directly applied to non-neoplastic cardiovascular diseases. Cancer studies often focus on the emergence and expansion of malignant cell populations driven by oncogenic transformation, resulting in the appearance of novel cellular states with distinct transcriptional programs. Whereas cardiovascular diseases commonly show cell loss, extracellular matrix expansion, fibrosis, scar formation, and disrupted architecture. (Figure). Therefore, analytical frameworks for cardiovascular diseases must explicitly account for damage-induced changes in cell composition and spatial structure rather than treating all cluster shifts as new pathogenic cell states.

Future studies should integrate scRNA-seq with spatial multi-omics, histology, imaging, hemodynamics, treatment history, and longitudinal outcomes. Spatial transcriptomics, proteomics, and metabolomics can preserve tissue context and help distinguish true cell-state alterations from sampling bias or structural remodeling(Figure). large-scale spatial datasets, combined with machine learning, may improve standardization, reduce analytic bias, decrease costs, and identify cellular neighborhoods relevant to disease progression and related molecular mechanisms. (Figure).

These studies can support the development of patient-specific tissue digital twins. Here, a tissue digital twin refers to a patient-specific computational model that integrates spatial multi-omics, histopathology, imaging, and clinical data to represent tissue structure and function over time^{10, 11}. Such models could move the field from descriptive atlases toward dynamic biological systems in which cells, extracellular matrix, inflammatory signals, and metabolic programs interact during disease progression. Agent-based and reinforcement learning approaches, may be particularly valuable in remodeling processes such as necrosis, apoptosis, fibroblast activation, fibrosis, and regional vascular remodeling^{12, 13}. In PH and other cardiovascular diseases, spatially and clinically anchored digital twins could help identify disease-associated cellular niches (cellular interactions, inflammatory signaling, metabolic adaptation), predict vulnerable regions of vascular remodeling, and prioritize therapeutic strategies before clinical testing. Such approaches may improve precision medicine, accelerate drug development, and enhance patient stratification in clinical trials. Achieving this goal will require advances in high-resolution imaging, artificial intelligence, scalable computational infrastructure, and standardized multi-omic analytical frameworks across institutions (Figure).

Feng and colleagues is therefore an important step: it provides a valuable reference map for PH single-cell interpretation while highlighting the need for spatial, longitudinal, and clinically grounded models of disease, such as digital twins, which integrate with longitudinal patient histories, wearable health monitoring systems, radiographic imaging, and real-world clinical outcomes comprehensive tissue mapping to transform personalized medicine by enabling predictive, preventive, and highly individualized healthcare strategies.