Proteomic and metabolomic landscapes of HFsnEF and HFnEF Identify molecular subtypes and prognostic markers in HFpEF

Background/Introduction Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous syndrome with limited therapeutic options. Recent guidelines suggest further stratification based on left ventricular ejection fraction (LVEF) thresholds, yet the underlying molecular and clinical differences remain poorly understood. Purpose This study aimed to characterize the proteomic and metabolomic profiles of HFpEF patients stratified by LVEF, identify molecular subtypes with distinct clinical features and outcomes, and develop a protein-based diagnostic model. Methods A total of 203 hospitalized HFpEF patients were enrolled and stratified into HFnEF (LVEF <65%, n=117) and HFsnEF (LVEF ≥65%, n=86). Clinical characteristics, including echocardiographic parameters, laboratory tests, and comorbidities, were collected. Plasma proteomic and metabolomic profiling was performed. Differential expression, correlation with clinical parameters, and pathway enrichment analyses were conducted. Unsupervised clustering was used to identify molecular subtypes. Machine learning models were applied to select key protein markers for subtype classification. Results HFsnEF and HFnEF groups exhibited distinct clinical and molecular signatures. Compared to HFnEF, HFsnEF patients had higher LVEF (65.2% vs. 56.5%, p<0.001), higher LVFS (35.9% vs. 29.6%, p<0.001), and smaller left ventricular end-systolic diameter (LVDs, 30.9 mm vs. 34.6 mm, p<0.01), suggesting a phenotype of smaller ventricular cavity. HFnEF showed upregulation of motor proteins and cytokine pathways, while HFsnEF enriched in oxidative phosphorylation and reactive oxygen species pathways. LVEF was significantly correlated with 413 proteins (|r|>0.3, p<0.05), including negative correlations with myocardial structure proteins (TNNI2, MYH3, CKMT2). Unsupervised clustering revealed three molecular subtypes (C1–C3) with divergent clinical outcomes: C3 had worse prognosis in HF hospitalization and composite endpoints (p<0.01). A 14-protein panel identified by random forest achieved high diagnostic accuracy (AUC=0.967 in test set). Protein–metabolite interaction networks revealed subtype-specific regulatory patterns. Conclusion Integrative multi-omics and clinical profiling reveals molecular heterogeneity in HFpEF beyond LVEF-based stratification. HFsnEF is characterized by smaller ventricular cavity and distinct metabolic pathways. The identified protein markers and subtypes provide new insights into HFpEF pathophysiology and may support personalized risk stratification and therapy.