A simple regulatory architecture allows learning the statistical structure of a changing environment.

Bacteria live in environments that are continuously fluctuating and changing. Exploiting any predictability of such fluctuations can lead to an increased fitness. On longer timescales, bacteria can 'learn' the structure of these fluctuations through evolution. However, on shorter timescales, inferring the statistics of the environment and acting upon this information would need to be accomplished by physiological mechanisms. Here, we use a model of metabolism to show that a simple generalization of a common regulatory motif (end-product inhibition) is sufficient both for learning continuous-valued features of the statistical structure of the environment and for translating this information into predictive behavior; moreover, it accomplishes these tasks near-optimally. We discuss plausible genetic circuits that could instantiate the mechanism we describe, including one similar to the architecture of two-component signaling, and argue that the key ingredients required for such predictive behavior are readily accessible to bacteria. Associations inferred from previous experience can help an organism predict what might happen the next time it faces a similar situation. For example, it could anticipate the presence of certain resources based on a correlated environmental cue. The complex neural circuitry of the brain allows such associations to be learned and unlearned quickly, certainly within the lifetime of an animal. In contrast, the sub-cellular regulatory circuits of bacteria are only capable of very simple information processing. Thus, in bacteria, the ‘learning’ of environmental patterns is believed to mostly occur by evolutionary mechanisms, over many generations. Landmann et al. used computer simulations and a simple theoretical model to show that bacteria need not be limited by the slow speed of evolutionary trial and error. A basic regulatory circuit could, theoretically, allow a bacterium to learn subtle relationships between environmental factors within its lifetime. The essential components for this simulation can all be found in bacteria – including a large number of ‘regulators’, the molecules that control the rate of biochemical processes. And indeed, some organisms often have more of these biological actors than appears to be necessary. The results of Landmann et al. provide new hypothesis for how such seemingly ‘superfluous’ elements might actually be useful. Knowing that a learning process is theoretically possible, experimental biologists could now test if it appears in nature. Placing bacteria in more realistic, fluctuating conditions instead of a typical stable laboratory environment could demonstrate the role of the extra regulators in helping the microorganisms to adapt by ‘learning’.

Informations de copyright

© 2021, Landmann et al.

Déclaration de conflit d'intérêts

SL, CH, MT No competing interests declared