Proteomic profiling of atrial fibrillation in heart failure predicts clinical outcome and uncovers key pathways

Abstract Introduction Despite advances in understanding the pathophysiology of atrial fibrillation (AF), data on the interaction between cardiovascular risk factors (CVRFs) and molecular alterations are limited. Further, biomarkers for AF in heart failure (HF) are lacking. Combining proteomic profiling with bioinformatics, we aimed to i) characterize the protein profile of AF, ii) identify signaling pathways linked to AF and iii) estimate the diagnostic and prognostic utility of a new AF protein score in HF. Methods Data from the MyoVasc Study, a prospective cohort of HF (n=3,289), were analyzed. Subjects with active cancer (n=262) or missing protein data (n=89) were excluded. Baseline and biannual follow-up comprised clinical phenotyping and biobanking. All subjects with prior or primary diagnosis of AF at baseline were identified, individuals without diagnosis of AF served as controls. The AF polygenic risk score (AF-PRS) was calculated. 536 proteins were analyzed by proximity extension assay technology. Differentially expressed proteins were identified and a novel protein signature for AF was derived by supervised machine-learning. The association of the resulting AF protein score with prevalent AF was validated by multivariable logistic regression analyses adjusting for age, sex, CVRFs, humoral markers, CHA2DS2-VA-score and HF status. The prognostic utility of the AF protein score was analyzed via multivariable Cox regression models. We performed external validation using a comparable subcohort of the Gutenberg Health Study. Bioinformatics were applied for pathway analysis. Results Prevalence of AF in the study sample was 28% (n=812; 69.9% men; mean age 68.9±9.3 years). Subjects with AF had a higher burden of CVRFs. Between the AF subsample and controls, the proportion of HF (78 vs 46%) and the AF-PRS were different (both p<0.0001). 145 proteins associated with AF were identified. Fibroblast growth factor 23, tartrate-resistant acid phosphatase type 5 and Notch 3 were the most important predictors for prevalent AF. Proteins with the most pronounced expression changes (Cohen’s d) showed a notable gradient of effect size from group paroxysmal to group permanent AF. The AF protein score predicted AF independent of the clinical profile (OR 3.09, 95%CI 2.56–3.73, p<0.0001; AUC 0.81). This effect persisted across all phenotypes and AHA stages of HF. Worsening of HF (HR 1.27; 1.09–1.47), cardiac death (HR 1.47; 1.22–1.78) and all cause death (HR 1.36; 1.22–1.53) were predicted by the AF protein score independent of the clinical profile. Bioinformatics revealed clustering in immune- and endothelial function-related pathways. Conclusion We identified novel protein markers and signaling pathways associated with AF. Combining proteomic profiling with bioinformatics emerges as a promising approach to advance our understanding of the molecular mechanisms of AF in HF, providing a basis for the development of biomarkers and novel antiarrhythmic therapies.