Donor-Level Public-Data Analysis of Proteostasis–Translation–Metabolism Coupling in Cardiomyocytes From Patients With Heart Failure With Preserved Ejection Fraction

Abstract Heart failure with preserved ejection fraction (HFpEF) is biologically heterogeneous, and the relationship between myocardial metabolic stress, proteostasis, and immune composition remains incompletely defined. We established a prespecified proteostasis–translation–metabolism coupling-failure (PTM-CF) framework and evaluated it using public human and mouse datasets. The primary analysis used the complete authorized H5AD CellBender raw-count layer from the human SCP3342 single-nucleus dataset and treated donors rather than nuclei as biological replicates. The discovery set comprised 14,644 cardiomyocyte nuclei from 43 donors, including 19 HFpEF donors and 24 non-failing controls. Donor-level pseudobulk models were adjusted for sex and biosample/pool. Translation-load index (TLI), proteostasis-capacity index (PCI), metabolic-supply index (MSI), and a prespecified PTM-CF composite were calculated from frozen gene modules. TLI was lower in HFpEF cardiomyocytes (β = −0.260, 95% CI −0.419 to −0.100, FDR = 0.00460), as was MSI (β = −0.233, 95% CI −0.320 to −0.147, FDR = 1.65 × 10−5). PCI was directionally lower but not significant (β = −0.0635, 95% CI −0.190 to 0.0628, FDR = 0.314). The PTM-CF composite was higher in the discovery model (β = 0.953, 95% CI 0.123 to 1.783, FDR = 0.0342), requiring independent replication. A prespecified immune-composition analysis found a higher lymphocyte fraction (β = 0.01351, FDR = 0.00382) and lower cycling-macrophage fraction (β = −0.00579, FDR = 0.000816), whereas the overall immune-labeled fraction was not different (FDR = 0.4529). A frozen 18-gene cardiomyocyte inflammatory score was not significantly higher in HFpEF (FDR = 0.6623) and was not significantly associated with PTM-CF (FDR = 0.2395). In human myocardial proteomics, 25 frozen-module genes were matched and 17/25 showed concordant RNA–protein direction, although global correlation was not significant (Spearman ρ = 0.158, P = 0.452). ACADVL and HADHA were the most convergent computational candidates across human cardiomyocyte RNA, human myocardial protein, and mouse HFpEF protein evidence. These findings support an auditable model of HFpEF metabolic, proteostasis, and immune-composition remodeling, but do not establish generalized immune activation, cytokine release, HFpEF specificity, causality, therapeutic efficacy, or target validation.