Prognostic value of myocardial redox signalling: a machine learning multi-omics approach

Abstract Background Oxidative stress has been proposed to be implicated in cardiovascular pathology. Studies have linked myocardial oxidative stress with the development of atrial fibrillation, but its role in the prediction of long-term outcomes and mortality is yet to be elucidated. Purpose To apply discovery network transcriptomics and genomics to atrial tissue (AT) obtained from patients undergoing cardiac surgery, to phenotype myocardial redox-state and identify potential therapeutic targets. Methods We included 1,032 patients undergoing cardiac surgery. Arm 1 included 258 patients in whom atrial tissue (right atrial appendage) was used for ex-vivo quantification of NOX-derived superoxide by lucigenin-enhanced chemiluminescence. Arm 2 included 414 patients in whom exploratory, unsupervised gene network analysis in human atrial tissue RNA sequencing data was performed to identify atrial tissue redox-sensitive signatures (blue signature). Arm 3 included 881 patients genotyped with the genome-wide SNP array UK Biobank Axiom Array and used for machine learning construction of GenRedOx, a SNP model built to predict the blue transcriptomic signature. Associations with future incidence of major adverse myocardial events (MAME: cardiovascular death and/or heart failure) was assessed across all arms using cox regression models (adjusted for age, sex, hypertension, dyslipidaemia, diabetes mellitus, body mass index, and plasma TNFa). Results Over a median follow-up of 5.27 years [IQR: 4.29-6.83] high atrial tissue (AT) NOX-derived superoxide was independently associated with MAME risk (Adj. HR[95%CI]: 11.7 [2.18 , 62.82], p=0.004, a). Unsupervised transcriptomic analysis in Arm 2 revealed 15 coexpressed gene ‘modules’ or 'signatures' (b). The blue signature was significantly associated with AT superoxide production with "immune response-regulating and activating signalling" being amongst the top enriched pathways. Patients with high eigengene values for the blue signature showed a 5.5-fold higher adjusted risk of MAME (c). In Arm 3, extreme gradient boosting of SNP data, trained using the blue transcriptomic signature as ground truth, revealed GenRedOx (d), a genomic artificial intelligence score, which was found to be prognostic of MAME in the validation stage (e). Homozygosity for the minor allele of SNP rs723897 in the Chemokine-Like Factor Super Family 7 (CMTM7) gene had the highest gain in the model. Conclusion We present for the first time two novel human myocardial transcriptomic and genomic signatures that reflect changes in redox state and identify long-term cardiovascular risk. Targeting pathways in the myocardium related with immune response signalling may lead to the development of novel therapeutics in cardiovascular disease.