Unraveling metabolic reprogramming in Δhnox Paracoccus denitrificans: a time-resolved metabolomics and AI-Powered proteome modeling approach

TL;DR

This study explores how deleting a specific protein in a bacteria affects its metabolism and biofilm behavior using metabolomics and AI.

Contribution

The paper introduces an AI-powered approach to link metabolomic and proteomic data in a non-model bacterial species.

Findings

Deleting hnox disrupts central carbon metabolism in Paracoccus denitrificans.

Δhnox strains show altered glycolytic and pentose phosphate pathway metabolites.

AI models show promise in predicting proteomic changes from metabolomic data.

Abstract

Heme-nitric oxide/oxygen binding (H-NOX) proteins function as critical sensors for nitric oxide in many bacterial species. However, their physiological functions are surprisingly diverse, and most have yet to be thoroughly investigated. Here, we investigate the impact of hnox deletion in Paracoccus denitrificans, a species known for its metabolic versatility and the formation of unusually thin biofilm structures. Time-resolved targeted metabolomics across three growth phases (OD600 = 0.6, 2.0, and 4.0) indicates that the deletion of hnox is consistently associated with disruptions in central carbon metabolism. At early stages, the Δhnox strain exhibits increased abundance of glycolytic and pentose phosphate pathway metabolites accompanied by decreases in amino acids, suggesting dysregulation in late glycolysis or promotion of fermentative metabolism. Higher cell densities are characterized by increased quorum sensing, which is shown to promote biofilm dispersal in the WT but had little effect on the Δhnox strain. Metabolomics changes at these stages continue to highlight the pentose phosphate and glycolytic metabolites along with redox cofactors, implicating changes in energy metabolism or oxidative stress response. Total proteomics at OD600 = 2.0 were collected to explore connections between metabolism and proteome dynamics, and to provide an opportunity to test several machine learning (ML) models for predicting proteomic changes from metabolomic profiles. While constrained by limited sample size, these exploratory models showed biologically meaningful concordance with experimentally observed proteome shifts, highlighting both the promise and the current limitations of artificial intelligence (AI)-based methods in non-model microbial systems.

Figures