Blood metabolome shows signatures of metabolic dysregulation in obese and overweight subjects that can be predicted by machine learning applied to heart rate variability

TL;DR

This study shows that heart rate variability can predict metabolic changes in overweight and obese individuals, suggesting a non-invasive way to monitor metabolic health.

Contribution

The novel use of machine learning to predict key metabolites from HRV in different BMI groups is a key contribution.

Findings

Sparse partial least square discriminant analysis revealed distinct metabolic profiles in normal-weight, overweight, and obese subjects.

Time- and frequency-domain HRV features correlated well with specific metabolites linked to metabolic dysregulation.

Machine learning successfully predicted key metabolite levels in overweight and obese individuals using HRV metrics.

Abstract

Obesity and overweight are linked to metabolic disturbances, which contribute to the onset of diseases like type 2 diabetes (T2D) and cardiovascular disorders. Metabolic health is also closely linked to autonomic function, as measured by heart rate variability (HRV), making HRV a potential non-invasive indicator of metabolic status. While studies have examined metabolic changes with body mass index (BMI), the link between HRV and specific metabolic profiles in normal-weight (NW), overweight (OW), and obese (OB) individuals is less understood. Additionally, whether HRV can reliably predict key metabolites associated with metabolic dysregulation remains largely unexplored.

This study uses targeted metabolomics to profile amino acids and acylcarnitines in a group of academic employees across BMI categories (NW, OW, and OB) and investigates correlations between HRV variables and these metabolites. Finally, a machine learning approach was employed to predict relevant metabolite levels based on HRV features, aiming to validate HRV as a non-invasive predictor of metabolic health.

NW, OW, and OB subjects showed different metabolic profiles, as demonstrated by sparse partial least square discriminant analysis (sPLS-DA). The main upregulated metabolites differentiating NW from OB were C6DC and C8:1, while C6DC and C10:2 were higher in OW than NW. Time- and frequency-domain HRV features show a good correlation with the regulated metabolites. Finally, our machine learning approach allowed us to predict the most regulated metabolites in OB and OW subjects using HRV metrics.

Our study advances our understanding of the metabolic and autonomic changes associated with obesity and suggests that HRV could serve as a practical tool for non-invasively monitoring metabolic health, potentially facilitating early intervention in individuals with elevated BMI.

Figures