The authors present a minimal model of an autonomous biological agent. Most sensorimotor models are stimulus-response models, and so there is no coupling between the goals of the agent and the function of the sensorimotor system. Here they investigate numerically a very simple model of bacterial chemotaxis, which has a run-and-tumble behavior. The tumbling rate is modulated by the product M of a simple metabolic reaction, which is essentially the transformation from a resource A to that product (they consider a slightly more complex autocatalytic reaction). When the resource is metabolized, the tumbling rate increases and so the bacterium tends to remain in place (in other words, this is a random walk with diffusion coefficient inversely proportional to the metabolic product). This results in chemotaxis, although there are no sensory receptors, because the bacterium slows down in places where it feeds.

This becomes especially interesting when variations are introduced. If the bacterium can also metabolize another product B into the same metabolic product M, then it also performs chemotaxis to that product, but the interactions between A and B becomes quite complex. If for example there is a homogeneous distribution of B in the environment, then chemotaxis to A is inhibited. This makes sense because the bacterium can just eat B and stay in place. If we think in terms of stimulus-response behavior, we can see that this is a complex nonlinear interaction between the two pathways (A->response and B->response) that appears to be goal-directed, and it does not involve explicit arbitration between the two signals (as in eg Bayesian multisensory integration).

Similarly, a compound that interferes with the metabolism (eg by reacting with A) becomes automatically a repellent. The authors also show that meaningful history-dependent effects can be introduced (e.g. if the metabolic reaction depends on two compounds, then chemotaxis to the first compound depends on whether the bacterium has previously encountered the second). I quote: “the behavioral significance of chemical compounds becomes a relational property that depends on the metabolic dynamics of the cell”, and the authors point out the “integrative role of metabolism”. I found this study original and insightful.