Stochasticity in Metabolism: Oxidative Stress Response Phenotypes As Emergent Consequences of Network Topology

Abstract

Reactive oxygen species (ROS) are ubiquitous yet elusive due to their reactivity and short half-lives, making simulation essential for mechanistic insight. Traditional ROS modeling has been confined to small ODE frameworks limited by kinetic-parameter scarcity. Here, we recast ROS dynamics as a flux-balance analysis (FBA) optimization problem using COBRA and CPLEX, augmenting the Escherichia coli iJO1366 genome-scale model with 84 spontaneous oxidation, 14 redox, and 9 detoxification reactions, targeting three ROS species: hydroxyl radicals, superoxide, and hydrogen peroxide (H2O2). We also formalize five fundamental rules of oxidation biology in the form of mixed-integer linear constraints, allowing us to dynamically incorporate reaction rate constants and let complex behavior emerge from first principles. Among these phenomena is the bimodal killing pattern under H2O2 stress, first announced over 40 years ago and never computationally recapitulated until now. We observe it by stochastically sampling our models with GapSplit, simulating bet-hedging and partitioning solutions into growers, persisters, and non-survivors. In our simulations, this killing effect arises solely from the metabolic network topology during random exploration of alternative metabolic configurations. The conservation of steady state creates a stochastic bias for detoxification when central carbon metabolism reaches its metabolic capacity to deal with ROS, causing the famous survival recovery phase. The recovery phase is thus part of Mode 1 metabolism, characterized by spare detoxification capacity and fulfilled by the Ahp reaction in E. coli. Mode 2 has no spare detoxification. We anticipate this to be an overlooked universal principle.