# Assessing personalized molecular portraits underlying endothelial-to-mesenchymal transition within pulmonary arterial hypertension Researcher: Ge Zhang Chinese name: 张格 ORCID identity anchor: https://orcid.org/0000-0002-3116-3246 Canonical researcher: https://drgezhang.com/#person Journal: Molecular Medicine Year: 2024 Type: Article DOI: 10.1186/s10020-024-00963-z Canonical page: https://drgezhang.com/papers/doi-10-1186-s10020-024-00963-z.html Markdown record: https://drgezhang.com/papers/doi-10-1186-s10020-024-00963-z.md ## Full Authors 1. Ruhao Wu 2. Ge Zhang 3. Mingzhou Guo 4. Yue Li 5. Lu Qin 6. Tianci Jiang 7. Pengfei Li 8. Yu Wang 9. Ke Wang 10. Yize Liu 11. Zhiqiu He 12. Zhe Cheng ## Official Abstract Pulmonary arterial hypertension (PAH) is a progressive and rapidly fatal disease with an intricate etiology. Identifying biomarkers for early PAH lesions based on the exploration of subtle biological processes is significant for timely diagnosis and treatment. In the present study, nine distinct cell populations identified based on gene expression profiles revealed high heterogeneity in cell composition ratio, biological function, distribution preference, and communication patterns in PAH. Notably, compared to other cells, endothelial cells (ECs) showed prominent variation in multiple perspectives. Further analysis demonstrated the endothelial-to-mesenchymal transition (EndMT) in ECs and identified a subgroup exhibiting a contrasting phenotype. Based on these findings, a machine-learning integrated program consisting of nine learners was developed to create a PAH Endothelial-to-mesenchymal transition Signature (PETS). This study identified cell populations underlying EndMT and furnished a potential tool that might be valuable for PAH diagnosis and new precise therapies. Text reproduced verbatim from PMC ([source](https://pmc.ncbi.nlm.nih.gov/articles/PMC11513636/)); [DOI](https://doi.org/10.1186/s10020-024-00963-z); [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/). ## Research & review context ### Research fields - Pulmonary vascular biology - Computational biomarker research ### Review topics - Endothelial expression states in PAH - EndMT-related candidate classification ### Scientific role Exploratory integration of single-cell and bulk tissue expression to nominate endothelial states and an internally tested two-gene classifier. ### Evidence position Retrospective omics association with limited expression experiments; no direct lineage tracing, causal perturbation or independently validated clinical assay. ### Related concepts - Pseudotime interpretation - Tissue-to-blood translation boundary - Internal versus external validation ### Possible citation contexts - Background: endothelial heterogeneity and EndMT-associated transcriptional programs in PAH. - Methods: candidate-gene selection and internal classifier evaluation, with clear limits on causal and clinical interpretation. Possible literature-review contexts, not evidence of existing citations. Read the findings and Evidence Scope before citing the original article. ## Evidence Scale **Endothelial transcriptional states and the PETS candidate signature in pulmonary arterial hypertension** A retrospective analysis of public single-cell and bulk lung transcriptomes identifies endothelial expression patterns consistent with EndMT in pulmonary arterial hypertension and develops a two-gene candidate classifier, PETS, with internal test-set evaluation and limited expression-level experimental support. | Evidence | Value | | --- | --- | | Single-cell donor samples | 3 PAH samples and 6 control samples in GSE169471 | | Cells before and after quality control | 31,444 initially analyzed; 28,906 retained | | Endothelial trajectory subset | 3,165 cells; EC1, EC2 and EC3 computational states | | Candidate expression genes | 6 EndMT pattern genes; the final PETS score uses RARRES2 and TC2N | | Model evaluation | 9 algorithms; GSE117261 internal discovery/testing split 7:3; tenfold cross-validation repeated 10 times in discovery | ### Counting note No unique participant total is manufactured across cell-level data, donor samples, bulk expression datasets and experiments. The 3,165 endothelial cells are a subset of the retained single-cell population, not additional donors. The 7:3 GSE117261 split is internal evaluation, not an independent external cohort; exact analytical split sizes and experimental replicate counts are not inferred from percentages or plots. ## Research Question Which endothelial transcriptional states are associated with pulmonary arterial hypertension, and can genes selected from an EndMT-related state support a candidate molecular classifier of PAH versus control tissue? ## Author Evidence Summary This research combined public single-cell and bulk lung transcriptome analyses with targeted endothelial expression experiments. The single-cell analysis began with 31,444 cells from three PAH and six control samples and retained 28,906 cells after quality control. Trajectory analyses focused on 3,165 endothelial cells and identified an EC3 state with expression patterns consistent with mesenchymal transition and altered metabolism. Co-expression and marker-based selection nominated six EndMT pattern genes. Nine machine-learning approaches were compared using an internal split of GSE117261, and a glmNet-derived two-gene score, PETS, combined RARRES2 and TC2N. The reported internal testing-subset accuracy was 0.8333 and C-index was 0.8908; because that subset also guided algorithm comparison and model selection, these estimates may be optimistic. Targeted hypoxia experiments supported expression differences for selected genes, but the work did not demonstrate that the genes drive EndMT, establish a clinically validated serum assay, or show that a therapy directed at the score improves PAH. A secondary marker-count discrepancy between the article text and its supplementary table is retained as an explicit limitation. ## Key Findings ### KF1: Single-cell profiling showed PAH-associated heterogeneity across nine annotated lung cell types. Context: The observations are at cell and donor-sample levels in a small retrospective human lung series; cell counts are not participant counts. | Evidence | Value | | --- | --- | | Donor samples | 3 PAH and 6 control samples | | Cells initially analyzed | 31,444 | | Cells retained after QC | 28,906 | | Annotated cell types | 9, arising from 21 computational clusters | Source locator: PMC11513636: Materials and methods — scRNA-seq data collecting, processing and analysis; Results — scRNA-seq analysis revealed cell landscape and characteristics in PAH; Figure 2A; supplementary Figures S1–S2. ### KF2: Endothelial trajectory analyses identified expression states consistent with an EndMT-related program, including an EC3 state. Context: Slingshot and Monocle2 infer ordering from cross-sectional gene-expression data. They do not directly observe an individual endothelial cell converting in vivo or prove causality. | Evidence | Value | | --- | --- | | Endothelial cells analyzed | 3,165 | | Endothelial states | EC1, EC2 and EC3 | | Patterns along inferred trajectory | 3 gene-expression clusters (C1–C3) | | Associated biological programs | Mesenchymal differentiation, extracellular matrix organization and oxidative phosphorylation | Source locator: PMC11513636: Materials and methods — Pseudotime trajectory inference; Results — Trajectory analyses revealed EndMT in endothelial populations; Figure 4A–G; Figure 5A. ### KF3: Network and expression-based selection nominated six EndMT pattern genes from the EC3-related analysis. Context: This is computational prioritization of disease-associated expression features. A secondary discrepancy in the EC3 marker count is disclosed rather than resolved by assumption. | Evidence | Value | | --- | --- | | Candidate genes | 6: RARRES2, CBY1, MSRB3, TC2N, CKS1B and SMAGP | | Co-expression modules | 13 | | Reported greenyellow module genes | 66 | Source locator: PMC11513636: Results — Identification of EndMT-related modules and genes; Figure 6B–F; main-text description of Table S6. Table S4 has 604 unique gene rows while the main text reports 602 EC3 markers; the discrepant marker count is not used as an accepted quantitative claim here. ### KF4: A two-gene glmNet score showed PAH-versus-control discrimination in an internal held-out test set. Context: The testing subset came from the same GSE117261 dataset as discovery and was also used to compare the nine algorithms and select the preferred model. The reported performance may therefore be optimistic, rather than an untouched final validation estimate. The C-index is reported as such, not relabeled as ROC AUC, and the supplementary residual column is not relabeled as RMSR. | Evidence | Value | | --- | --- | | Candidate learners | 9 | | Discovery/testing split | 7:3 within GSE117261 | | Discovery resampling | 10-fold cross-validation repeated 10 times | | PETS formula | 2.6926723 × RARRES2 + 0.7410611 × TC2N | | glmNet internal test accuracy | 0.833333333333333 | | glmNet internal test C-index | 0.890756302521008 | | glmNet internal test F1-score | 0.888888888888889 | | glmNet internal test precision and recall | 0.842105263157895 and 0.941176470588235 | | Supplementary \|Residuals\| column | 0.253593 | Source locator: PMC11513636: Materials and methods — PETS generated from integrative machine-learning program; Results — Machine learning-based integrative program generates PETS; Figure 7A–D; Additional file 13, Table S7, glmNet row (worksheet row 3, columns B–G). ### KF5: Targeted experiments supported expression changes in selected candidate genes under hypoxic conditions. Context: Expression-level concordance does not establish that RARRES2 or TC2N causes EndMT or that inhibiting either gene improves PAH. No numerical fold changes or replicate counts are inferred from figures. | Evidence | Value | | --- | --- | | Human PAEC observations | Higher expression reported for RARRES2, TC2N, CKS1B and SMAGP under hypoxia | | Mouse PAEC observations | Higher Rarres2 and Tc2n mRNA under hypoxia | | Tissue observation | Reported chemerin signal in CD31-marked endothelial regions of hypoxic PAH mouse lung | | Human/mouse cell hypoxia condition | 1% oxygen stated in cell-culture methods | Source locator: PMC11513636: Materials and methods — Cell lines and cultures; RNA extraction and real-time quantitative PCR; Results — Identification of EndMT-related modules and genes, and Machine learning-based integrative program generates PETS; Figure 7E and supplementary Figure S6 caption. Experimental n and numerical fold changes are not supplied by these text/caption locators. ## Study Design & Model Development ### Study profile - Study Design: Retrospective public single-cell and bulk transcriptome analysis with computational trajectories, co-expression feature selection and internal machine-learning evaluation; targeted hypoxic endothelial-cell and tissue expression experiments - Evidence Type: Human tissue transcriptomic association and classifier development, supported by limited gene-expression experiments rather than causal gene perturbation or a prospective clinical trial - Population: GSE169471 lung single-cell samples: three PAH and six controls; GSE117261 used for internal classifier development/testing. Cells, donor samples, expression records and experimental systems are distinct units. - Primary Endpoint: PAH-associated endothelial expression states and EndMT-related gene prioritization in retrospective tissue datasets - Secondary Endpoint: Internal discrimination of PAH versus controls by the candidate PETS expression score, with selected gene-expression observations under hypoxia - External dataset evaluation: No - Data modalities: Human lung single-cell RNA sequencing, Public bulk lung transcriptomes, Pseudotime trajectory inference and co-expression networks, Internal machine-learning classification, RT-qPCR in human and mouse pulmonary artery endothelial cells, Reported chemerin/CD31 tissue immunofluorescence ### Model development - Model Name: PAH Endothelial-to-mesenchymal Transition Signature (PETS) - Model Family: glmNet candidate classifier - Features: RARRES2 and TC2N expression - Evaluation: Internal 7:3 discovery/testing split of GSE117261; tenfold cross-validation with ten repetitions within discovery. The testing subset also compared algorithms and selected the preferred model, so performance may be optimistic. - Clinical Boundary: Retrospective tissue-expression classification, not demonstrated early clinical diagnosis, serum-test validity, treatment response or clinical benefit ## What This Study Adds - It relates PAH-associated endothelial heterogeneity to an explicit EndMT-related transcriptional state using complementary computational trajectory methods. - It connects candidate-gene prioritization to an interpretable two-gene tissue-expression classifier and reports its internal model comparison. - It places limited expression experiments alongside retrospective computational evidence, creating testable hypotheses for prospective biomarker and mechanistic work. ## Evidence Scope ### Supports - PAH-associated endothelial transcriptional heterogeneity and computationally inferred EndMT-related expression programs. - A candidate two-gene PETS classifier with internal test-set performance under the reported dataset and analysis procedure. - Selected expression changes in hypoxic endothelial experimental systems. ### Does Not Establish - Direct lineage-tracing proof that the inferred trajectory is an in vivo cell-fate conversion. - A causal molecular mechanism from RARRES2 or TC2N to PAH or EndMT. - An independently externally validated or prospective diagnostic test, or demonstrated detection of early disease. - A validated serum PETS assay: the score was derived from transcriptomic expression, so discussion of serum translation remains a hypothesis. - Clinical efficacy or safety of chemerin/CMKLR1-targeted therapy, or any PETS-guided treatment benefit. ### Limitations - Human datasets are retrospective and the single-cell analysis has few donor samples; thousands of cells do not imply thousands of independent patients. - Trajectory inference, pathway scoring, ligand-receptor inference and co-expression are computational associations, not direct causal mechanisms. - The machine-learning testing subset is an internal split of GSE117261 and was also used to compare the nine algorithms and select the preferred model. Reported performance may therefore be optimistic. No untouched final test set, independent external clinical validation or prospective validation was established. - PETS is based on transcript expression. Discussion of serum-marker applications should not be read as a validated blood test. - Main text reports 602 EC3 marker genes, but the retrieved original Table S4 contains 604 distinct gene rows. This unresolved ancillary count difference is not silently harmonized. - A Results sentence describes lower incoming and higher outgoing communications but labels the role change in the opposite direction; no sender/receiver direction claim is carried into this evidence page. - A main-text sentence refers to MSRB1 while the candidate list and primer list use MSRB3. The page avoids drawing a gene-specific conclusion from that inconsistent sentence. - Targeted expression experiments do not demonstrate a beneficial intervention. Experimental sample sizes and fold-change values are not invented from plots. ## When This Study Is Useful to Cite ### CU1: What transcriptomic evidence links endothelial states with EndMT-related programs in PAH? Retrospective cell-state, pseudotime and expression observations, without claiming direct lineage tracing or causal mediation. Evidence: [KF1](#kf1), [KF2](#kf2), [KF3](#kf3) ### CU2: What is PETS and how was its classification performance evaluated? An exploratory RARRES2/TC2N tissue-expression score with internal test-set comparison; no established serum assay or clinical benefit. Evidence: [KF4](#kf4), [KF5](#kf5) ## What This Study Should Not Be Cited to Claim - Proven causal mediation of EndMT by the candidate genes. - A clinically established serum diagnostic test, prospective early detection, or effective chemerin-directed treatment. ## Evidence Matrix ### Human cell atlas - Context: The observations are at cell and donor-sample levels in a small retrospective human lung series; cell counts are not participant counts. - Finding: Single-cell profiling showed PAH-associated heterogeneity across nine annotated lung cell types. - Evidence Level: Retrospective descriptive human transcriptomics - Scope: Few donor samples; cells are not independent patients - Source Locator: PMC11513636: Materials and methods — scRNA-seq data collecting, processing and analysis; Results — scRNA-seq analysis revealed cell landscape and characteristics in PAH; Figure 2A; supplementary Figures S1–S2. - Evidence refs: [KF1](#kf1) ### Endothelial trajectory - Context: Slingshot and Monocle2 infer ordering from cross-sectional gene-expression data. They do not directly observe an individual endothelial cell converting in vivo or prove causality. - Finding: Endothelial trajectory analyses identified expression states consistent with an EndMT-related program, including an EC3 state. - Evidence Level: Computational state and pseudotime inference - Scope: Associated expression program, not lineage-tracing proof - Source Locator: PMC11513636: Materials and methods — Pseudotime trajectory inference; Results — Trajectory analyses revealed EndMT in endothelial populations; Figure 4A–G; Figure 5A. - Evidence refs: [KF2](#kf2) ### Candidate-gene prioritization - Context: This is computational prioritization of disease-associated expression features. A secondary discrepancy in the EC3 marker count is disclosed rather than resolved by assumption. - Finding: Network and expression-based selection nominated six EndMT pattern genes from the EC3-related analysis. - Evidence Level: Co-expression and marker-based feature selection - Scope: Candidate association; ancillary marker-count discrepancy retained - Source Locator: PMC11513636: Results — Identification of EndMT-related modules and genes; Figure 6B–F; main-text description of Table S6. Table S4 has 604 unique gene rows while the main text reports 602 EC3 markers; the discrepant marker count is not used as an accepted quantitative claim here. - Evidence refs: [KF3](#kf3) ### PETS classifier - Context: The testing subset came from the same GSE117261 dataset as discovery and was also used to compare the nine algorithms and select the preferred model. The reported performance may therefore be optimistic, rather than an untouched final validation estimate. The C-index is reported as such, not relabeled as ROC AUC, and the supplementary residual column is not relabeled as RMSR. - Finding: A two-gene glmNet score showed PAH-versus-control discrimination in an internal held-out test set. - Evidence Level: Internal held-out retrospective classification - Scope: Same-dataset split; no independent external clinical validation - Source Locator: PMC11513636: Materials and methods — PETS generated from integrative machine-learning program; Results — Machine learning-based integrative program generates PETS; Figure 7A–D; Additional file 13, Table S7, glmNet row (worksheet row 3, columns B–G). - Evidence refs: [KF4](#kf4) ### Expression experiments - Context: Expression-level concordance does not establish that RARRES2 or TC2N causes EndMT or that inhibiting either gene improves PAH. No numerical fold changes or replicate counts are inferred from figures. - Finding: Targeted experiments supported expression changes in selected candidate genes under hypoxic conditions. - Evidence Level: Targeted hypoxia-associated expression evidence - Scope: No causal perturbation or demonstrated therapeutic benefit - Source Locator: PMC11513636: Materials and methods — Cell lines and cultures; RNA extraction and real-time quantitative PCR; Results — Identification of EndMT-related modules and genes, and Machine learning-based integrative program generates PETS; Figure 7E and supplementary Figure S6 caption. Experimental n and numerical fold changes are not supplied by these text/caption locators. - Evidence refs: [KF5](#kf5) ## Q&A ### How many people and cells were represented in the single-cell analysis? The article reports three PAH and six control donor samples, initially containing 31,444 cells, with 28,906 retained after quality control. These are sample and cell counts, not interchangeable participant totals. Evidence: KF1 ### Does pseudotime prove that endothelial cells became mesenchymal cells? No. The ordering of 3,165 endothelial cells is inferred from cross-sectional expression. EndMT-related gene programs support a biological hypothesis, but direct lineage tracing or causal perturbation would be needed to establish cell conversion. Evidence: KF2 ### What genes form the final PETS score? The reported score is 2.6926723 × RARRES2 + 0.7410611 × TC2N expression. It was selected from a broader computationally nominated candidate set. Evidence: KF3, KF4 ### What does the reported model performance mean? In the GSE117261 internal testing subset, Table S7 reports glmNet accuracy 0.8333 and C-index 0.8908. The same subset was used to compare the nine algorithms and select the preferred model, so the reported performance may be optimistic. These are retrospective classification estimates, not prospective clinical accuracy or treatment benefit. Evidence: KF4 ### Was PETS externally validated in an independent clinical cohort? No independent external clinical validation is established by the reported 7:3 split. Discovery and testing came from GSE117261. The targeted expression experiments provide a different kind of evidence and do not substitute for clinical validation. Evidence: KF4, KF5 ### Do the experiments demonstrate a therapeutic mechanism? No. The reported experiments assess selected expression responses to hypoxia. They do not demonstrate that perturbing a candidate gene reverses EndMT or improves PAH outcomes. Evidence: KF5 ### Are all ancillary gene counts internally consistent? No. The main text reports 602 EC3 marker genes, whereas the retrieved Table S4 contains 604 unique gene rows. This page preserves the discrepancy and does not use that marker count as a confirmed scale metric. Evidence: KF3 ## Concepts & Entities ### Disease And Cell Context Pulmonary arterial hypertension, Pulmonary artery endothelial cells, Endothelial-to-mesenchymal transition ### Candidate Signatures PETS, RARRES2/chemerin, TC2N ### Methods Single-cell transcriptomics, Pseudotime inference, Co-expression feature selection, Internal test-set validation ## Related Research - [Smooth muscle cell fate decisions decipher a high-resolution heterogeneity within atherosclerosis molecular subtypes](https://drgezhang.com/papers/smc-fate.html) — Related vascular cell-state and molecular-subtype research in atherosclerosis; a distinct disease context rather than replication of PETS. (DOI: 10.1186/s12967-022-03795-9) - [Atherosclerotic plaque vulnerability quantification system for clinical and biological interpretability](https://drgezhang.com/papers/apvs.html) — Related work on interpretable vascular molecular signatures and classifier evaluation; does not provide external PAH validation. (DOI: 10.1016/j.isci.2023.107587) - [From Disease-Specific Models to Broad Clinical Utility: A Perspective on AI Hybrid Ensemble Frameworks](https://drgezhang.com/papers/doi-10-1002-ggn2-202500053.html) — Perspective relevant to separating disease-specific modeling performance from broadly established clinical utility. (DOI: 10.1002/ggn2.202500053) ## Publication & Provenance - DOI: https://doi.org/10.1186/s10020-024-00963-z - Source Of Truth: Version of record, Molecular Medicine 30:189, published 26 October 2024; PMID 39462326; PMC11513636. Complete PMC XML and retrieved original supplementary Tables S4 and S7 retained privately with SHA-256 hashes on 2026-10-14. - Original Article Url: https://doi.org/10.1186/s10020-024-00963-z - Full Text Url: https://pmc.ncbi.nlm.nih.gov/articles/PMC11513636/ - Publisher: https://link.springer.com/article/10.1186/s10020-024-00963-z - PubMed: https://pubmed.ncbi.nlm.nih.gov/39462326/ - License: Article-specific CC BY 4.0 notice in publisher Rights and permissions and PMC permissions. This is an original summary, not a reproduction of the full article or figures. - License Url: https://creativecommons.org/licenses/by/4.0/ - Identity Verification: Ge Zhang is listed with the Department of Cardiology, First Affiliated Hospital of Zhengzhou University, Henan cardiovascular research-center and cardiac injury/repair laboratory affiliations. The version of record identifies Ruhao Wu, Ge Zhang and Mingzhou Guo as equal contributors. The article does not supply a Ge Zhang-specific ORCID identifier; name/affiliation and the existing ORCID-derived master record are concordant. - Version Relationship: Published journal research article; no linked preprint relation was established from checked article metadata. - Numeric Verification: Cell/donor counts, endothelial subset and PETS formula checked against main text and figure captions; internal test metrics checked directly against the downloaded original Table S7 glmNet row. No total patient n, experimental replicate count, serum concentration or treatment effect was inferred. - Reporting Discrepancies: Ancillary count: main text 602 EC3 markers versus 604 distinct rows in original Table S4. Communication-role wording and MSRB1/MSRB3 inconsistency also retained in limitations. These passages are not used to establish directional cell communication, a specific MSRB expression result or a confirmed marker total. ## Evidence-page notice This page is an original evidence-focused explanation of the published article. PETS is an exploratory transcriptomic classifier, not an established diagnostic test or a demonstrated treatment strategy. Cite and inspect the original article and supplements. ## Links - DOI: https://doi.org/10.1186/s10020-024-00963-z - HTML page: https://drgezhang.com/papers/doi-10-1186-s10020-024-00963-z.html - Google Scholar query: https://scholar.google.com/scholar?q=%22Assessing%20personalized%20molecular%20portraits%20underlying%20endothelial-to-mesenchymal%20transition%20within%20pulmonary%20arterial%20hypertension%22 - ORCID: https://orcid.org/0000-0002-3116-3246