https://doi.org/10.1007/s00392-025-02625-4
1Universitätsmedizin der Johannes Gutenberg-Universität Mainz Präventive Kardiologie und Medizinische Prävention Mainz, Deutschland; 2University of Milan Department of Computer Science Milan, Italien; 3Universitätsmedizin der Johannes Gutenberg-Universität Mainz Mainz, Deutschland
Introduction: Heart failure is a complex and heterogeneous clinical syndrome characterized by an impaired capacity of the heart to pump sufficient blood, thereby compromising oxygen supply to the body. Emerging evidence suggests a critical molecular link between human metabolism and heart failure, presenting opportunities for patient stratification and novel therapeutic targets. In this context, unbiased subgroup identification based on circulating protein and lipid profiles is a promising strategy to refine risk assessment and reveal new biomarkers for the onset and progression of heart failure.
Methods: In this study, we analyzed plasma samples from 1,678 individuals with stage C/D heart failure, derived from the MyoVasc cohort, using targeted proteomics (Olink panels: Cardiometabolic, CVD2, CVD3, Inflammation, Neurology, Organ Damage) and lipidomics (4D-LC/TIMS-IMS mass spectrometry). Integrative data fusion approaches were employed to combine proteomics and lipidomics data (305 lipid species and 529 proteins after preprocessing). Using similarity network fusion (SNF), subject-subject networks were generated, and clusters were derived via spectral clustering. The optimal cluster number was determined by rank test for heart failure worsening outcomes, and the specific contributions of proteins and lipids were assessed through multinomial Lasso regression.
Results: This integrative approach distinctly identified eight clusters of patients each characterised by unique clinical profiles. Among the identified patient subgroups, four subgroups had statistically different outcomes for heart failure worsening. Notably, clinical parameters not only emphasised differences in metabolic phenotypes among patient clusters, such as diabetes and obesity, but also underscored critical markers related to kidney and liver function, atherosclerosis, and atrial fibrillation. Comparative analysis of protein and lipid abundance profiles across subgroups illuminated new clinically undefined subtypes, with a notable contribution from immune-modulating proteins of interleukin (IL) and tumour necrosis factor (TNF) superfamilies, driving the unbiased subtyping framework.
Conclusion: Our findings demonstrate that proteins and lipids can define patient subgroups at higher risk for worsening heart failure, underscoring these biomarkers' potential for refined patient stratification and targeted therapeutic intervention. The network-based approach effectively overcomes data heterogeneity, leveraging the unique contributions of each biological platform to enhance clustering resolution.