Mechanochemical model for Adaptive Motility in E.coli via Peptidoglycan network training

How do single-celled organisms like E.coli exhibit adaptive motility without the need for a nervous system? We propose a novel mechanochemical learning mechanism that leverages the mechanical properties of the peptidoglycan (PG) wall—the cell’s own elastic, load-bearing mesh—as a computational substrate. Stresses across this mesh modulate stator occupancy, the number of torque-generating units engaged at each flagellar motor, effectively translating environmental forces into motor control to allow the cell to combat collision with the environment. We model the PG layer as a trainable spring network with overdamped langevin dynamics, which is coupled to an agent based model of E. coli chemotaxis with motor switching dynamics.

The core of our model is the feedback loop between environment and motor functions, which are anchored in PG with catch-slip bonds that are sensitive to mechanical load. By allowing environment constraints to create a stress pattern across the PG network, and using a local learning rule where stresses update the network stiffness, we alter the spatial pattern of stator occupancy, allowing the cell to reshape its net torque and tumble angle distribution to escape collisions faster.

By learning to bias away from frequently encountered walls, the cell adapts its stochastic tumbles through the use of PG mechanical memory. This work presents a framework for exploring how mechanical memory, embodied in the material properties of the cell itself, can drive adaptive behavior.