Unsupervised Learning Based on Proteomic Signatures Identifies Distinct Subgroups of Heart Failure With Mildly Reduced Ejection Fraction

Heart failure with mildly-reduced ejection fraction (HFmrEF) lacks therapeutic strategies due to heterogeneity and dynamic transitions between HFrEF/HFpEF. Proteins constitute predominant drug targets and primary mediators of signaling pathways in HF. We measured 92 plasma proteins (Olink CardiovascularIII) in 230 HF patients from BIOMS-HF registry. Fifteen, eighteen, and fifteen baseline proteins discriminated MACEs were determined in HFmrEF, HFpEF, and HFrEF, respectively. Pathway enrichment revealed shared signaling in HFmrEF/HFpEF (apoptosis, etc.), HFmrEF/HFrEF (vascular regulation, etc.), and HFmrEF/HFrEF/HFpEF (inflammatory/hormonal signaling). Four patient phenotypes were identified according to proteomic signatures using unsupervised learning: Cluster1 (younger, smokers, lowest MACEs [29.5%]); Cluster2 (elderly, higher comorbidity, diastolic dysfunction); Cluster3 (systolic dysfunction, elevated heart rates, responsive to HFrEF therapies); Cluster4 (high inflammation, cardiometabolic disturbances, highest MACEs [74.4%]). Cross-referenced with druggable genome database, TNF-R1 was revealed as an appealing druggable target for cluster2/4, while OPN and MMP-2 for cluster3/4. Unsupervised learning based on proteomics identified four HFmrEF phenotypes, each providing druggable targets according to distinct pathophysiological pathways.