Machine Learning‐supported Characterization of the Oxidative Stress‐Sensitive Posttranslational Modifications in ISO‐induced Murine Cardiac Hypertrophy

IntroductionOxidative stress disrupts cellular redox homeostasis and chemically alters proteins at the proteome scale in the form of oxidative sensitive post‐translational modifications (O‐PTMs). In stressed heart, these O‐PTMs often lead to protein dysfunction and proteotoxicity, which have been implicated as pathological drivers for various cardiovascular diseases. However, little is known about the dynamics of myocardial O‐PTMs during disease progression, or the O‐PTM molecular signatures pertinent to pathophysiological phenotypes of the heart.MethodsWe characterized a large‐scale longitudinal proteomics dataset previously collected from mice under isoproterenol (ISO)‐induced cardiac hypertrophy. Six genetic mouse strains exhibiting distinct susceptibilities to ISO challenge were employed to recapitulate genetic diversity in humans. Left ventricle tissue samples were analyzed at 7 time points following ISO treatment (0, 1, 3, 5, 7, 10, 14 days); cardiac function was measured concomitantly via heart weight/body weight ratios and ejection fraction. Molecular features (e.g., modification site & occupancy) of 7 prevalent types of O‐PTMs were extracted using the Integrated Proteomics Pipeline. Stringent filtering criteria were applied to identify O‐PTMs of high confidence in both Control and ISO‐treatment samples, as well as to enable data‐imputation to recover missing data. A customized analytic pipeline empowered by Machine Learning (ML) approaches was employed to pinpoint a panel of O‐PTMs as molecular signatures. Subsequently, we constructed a multi‐layer Bayesian network to unearth pivotal relationships among molecular signatures and phenotypic features of heart.ResultsWe conducted the first large‐scale O‐PTM analysis in murine cardiac remodeling, which elucidates the impact of ISO‐induced oxidative stress on 7 irreversible O‐PTM subtypes. After data‐imputation, a completed data matrix containing temporal occupancy values of 2,877 O‐PTMs maximally defined the O‐PTM landscapes in healthy and stressed mouse hearts. The ML‐supported feature selection and validation extracted 9 O‐PTMs that are highly indicative of pathophysiological conditions. The Bayesian network modeling unveiled hierarchical relationships among O‐PTM signatures and cardiac function, offering mechanistic insights underlying oxidative stress and cardiac dysfunction. These O‐PTM signatures are enriched in biological pathways supporting energy production, ROS regulation, Ca2+signaling, and ECM remodeling, which are all benchmark processes of cardiac remodeling.ConclusionProteomics captures in remarkable detail the molecular interplays underlying cardiovascular pathophysiology. ML methods support in‐depth analyses to tease out key molecular features unique to disease phenotypes, providing a novel strategy to identify potential diagnostic and therapeutic targets. Application of these cutting‐edge computational approaches on O‐PTM investigations will enhance our capability to substantiate precision medicine.Support or Funding InformationNIH R01HL146739 to D. Wang & D. Ng; NIH R35HL135772 to P. Ping.