Hidden Heart Fibroblast State Found Across Human DCM Datasets

A single-nucleus RNA-sequencing study of human hearts has identified a fibroblast state that appears reproducibly expanded in dilated cardiomyopathy, one of the leading causes of heart failure and transplantation. The work, published in Frontiers in Genetics by Yimeng Li of Liyang People’s Hospital and Jiangnan University, reanalyzed two independent public datasets rather than generating new tissue samples, and applied unusually strict statistical controls to a question that has long frustrated the field: which of the many fibroblast states visible in single-cell data is most reliably tied to disease, rather than being an artifact of one particular cohort or clustering run.

Dilated cardiomyopathy, or DCM, is characterized by enlargement and weakening of the ventricles, and cardiac fibrosis — the excessive accumulation of extracellular matrix proteins such as collagens — is a major driver of that adverse remodeling. Fibrosis stiffens the heart wall, impairs its ability to fill with blood, and progressively worsens function. But fibrosis is not produced by a single, uniform population of cells. Cardiac fibroblasts occupy dynamic, injury-dependent transcriptional states, and single-cell and single-nucleus profiling has revealed marked stromal heterogeneity in both healthy and diseased myocardium. Activated fibroblasts can be sparse in non-failing hearts yet enriched in cardiomyopathy, and previous work has implicated the transcription factor MEOX1 as a regulator of fibroblast activation. What remained uncertain was which human DCM fibroblast state is most reproducibly enriched across cohorts, and whether its disease association is stronger than that of the conventional ACTA2-high myofibroblast state that has dominated the literature.

To answer that question, the study used a prespecified discovery-and-validation design built entirely on publicly available data. The discovery cohort, GSE183852, contained left-ventricular nuclei from 25 non-failing donor hearts and 13 DCM hearts. The external validation cohort, SCP1303, contained 16 normal and 11 DCM hearts, with hypertrophic cardiomyopathy samples excluded from the primary comparison. Fibroblasts from the discovery cohort — 49,124 nuclei in total — were independently reclustered in Seurat using log normalization, 2,500 variable features, 30 principal components, a shared-nearest-neighbor graph, and Louvain clustering at resolution 0.5, with a fixed random seed to ensure reproducibility. Critically, the author treated the donor, not the individual nucleus, as the statistical unit of inference, a decision that guards against the false discoveries that arise when thousands of nuclei from one patient are mistakenly counted as independent biological replicates.

The reclustering resolved multiple fibroblast states and revealed substantial compositional variation between donors. One cluster stood out: cluster two showed high expression of MEOX1 and FAP together with relatively low ACTA2, the gene encoding alpha-smooth muscle actin that marks the classical contractile myofibroblast. The study is careful to note that ACTA2-low is a relative property — ACTA2 was detected in 14.7 percent of cluster-two nuclei versus 57.3 percent in the discrete ACTA2-high cluster 16 — and does not imply absent expression or a lack of overall ACTA2 induction in disease. Cluster two nonetheless retained clear extracellular-matrix features, expressing FN1, POSTN, COL1A1 and COL3A1, and an unbiased marker analysis surfaced RFX2, CYR61, FOS, EGR1, DDAH1, CTGF, FAP and MEOX1 among its leading genes. The state was therefore defined by the combined three-marker pattern rather than any single gene.

The quantitative case for disease enrichment rests on donor-level proportions. Cluster two made up a median 5.81 percent of fibroblasts in DCM hearts versus 0.28 percent in donor hearts, a difference that was highly significant after false-discovery-rate correction. The mean DCM-minus-donor difference was 0.165, with a 95 percent confidence interval of 0.035 to 0.349 derived from 10,000 iterations of donor bootstrap resampling. Perhaps most tellingly, sequentially omitting each of the 38 donors one at a time never flipped the direction of the effect, which stayed within a range of 0.102 to 0.179. Sensitivity analyses across clustering resolutions showed that while the default state merged into broader compartments at lower resolutions, higher-resolution counterparts retained disease enrichment with p-values below 2 × 10⁻⁶. By contrast, the ACTA2-high cluster 16 showed only a smaller, non-significant abundance difference between groups.

Covariate modeling added nuance rather than contradiction. Heteroscedasticity-consistent HC3 linear models estimated a DCM coefficient of 0.165 unadjusted, 0.166 after adjusting for sex, and 0.184 after adjusting for both sex and ordinal age tertile — the finest age granularity the source data allowed. The effect magnitude was preserved in every model, but the confidence intervals marginally crossed zero, and the author explicitly declines to claim covariate-independent statistical significance. This kind of restraint is uncommon in reanalysis studies and reflects the modest number of donors available: with 38 hearts in the discovery cohort, even a robust effect can produce intervals that graze the null.

External validation in the independent SCP1303 cohort strengthened the finding while also defining its boundary. The author-annotated activated fibroblast population, FB-ActFB, was far more abundant in DCM than in normal hearts — a median of 24.11 percent versus 5.16 percent of fibroblasts — with a mean difference of 0.192 and a bootstrap confidence interval of 0.144 to 0.256. All 27 leave-one-donor-out estimates remained positive, and the sex-adjusted coefficient was 0.199 with a p-value of 6.61 × 10⁻⁷. The external state showed concordant MEOX1 and FAP expression, but ACTA2 was not available in the retained external marker summary, and validation relied on the source study’s annotations rather than frozen one-to-one label transfer. The author therefore concludes that the replication supports a shared activation program and abundance shift, not a complete reproduction of the three-marker phenotype or identical cellular identity.

Beyond the state itself, donor-pseudobulk differential expression — in which raw fibroblast counts are summed within each donor before testing — identified several genes elevated in DCM fibroblasts after false-discovery-rate correction. DDAH1 led the list with an adjusted log2 fold change of 2.01, followed by MEOX1 at 2.29 and FAP at 1.89; POSTN, TGFBI, CTGF and COL1A2 also passed correction. Within discovery cluster two, nuclei positive for DDAH1 showed higher extracellular-matrix scores than DDAH1-negative nuclei in paired donor summaries. DDAH1, an enzyme involved in nitric oxide metabolism, has previously been linked to cardiac hypertrophy and remodeling in animal models, which makes it a biologically plausible candidate. The study is emphatic, however, that DDAH1 is a prioritized state-associated candidate, not an established causal regulator, and it was deliberately excluded from the state-defining marker panel to avoid circularity.

The extracellular-matrix evidence itself proved partly method-dependent, a finding the author reports transparently. The donor-pseudobulk ECM score was significantly higher in DCM hearts, with a Mann–Whitney p-value of 0.00975, whereas the cell-derived donor mean showed the same positive direction without reaching significance. The two scoring approaches were nonetheless strongly correlated across donors, with a Spearman correlation of 0.891. Whole-fibroblast pseudobulk differences can reflect both shifts in state composition and expression changes within states, so they cannot by themselves establish cluster-specific regulation. ACTA2, notably, showed a modest overall increase in DCM fibroblasts, reinforcing that the ACTA2-low label describes a relative cluster phenotype rather than a claim that canonical myofibroblasts are absent from the diseased heart.

The study’s limitations are acknowledged candidly: it is a secondary analysis without spatial, protein-level or functional validation; age information was incomplete in both cohorts; reliable etiologic subdivisions of DCM were unavailable; and public-cohort processing differences precluded exact state transfer between datasets. Still, the work delivers something the field genuinely needs — a technically controlled, hypothesis-generating target for prospective experiments. The author proposes that multiplex tissue imaging should test the spatial coexistence of MEOX1, FAP, ACTA2 and DDAH1 in diseased myocardium, while fibroblast perturbation studies should determine whether DDAH1 alters collagen synthesis, matrix organization or state transitions. If those experiments pan out, a fibroblast state first glimpsed in reprocessed public data could point toward new anti-fibrotic strategies for a disease that currently has few options beyond transplantation.

Subject of Research: Identification of a disease-associated MEOX1+FAP+ACTA2-low fibroblast state in dilated cardiomyopathy using cross-cohort single-nucleus RNA sequencing

Article Title: Cross-cohort single-nucleus transcriptomic identification of a MEOX1+FAP+ACTA2-low extracellular matrix-remodeling fibroblast state in dilated cardiomyopathy

Article References: Li, Y. (2026). Cross-cohort single-nucleus transcriptomic identification of a MEOX1+FAP+ACTA2-low extracellular matrix-remodeling fibroblast state in dilated cardiomyopathy. Frontiers in Genetics, 17, Article 1987893. https://doi.org/10.3389/fgene.2026.1987893

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1987893

Keywords: dilated cardiomyopathy, cardiac fibrosis, fibroblasts, single-nucleus RNA sequencing, MEOX1, FAP, ACTA2, DDAH1, extracellular matrix, pseudobulk differential expression, biomarkers, Cross-cohort