Sensorimotor feedback and active perception in cells

Sorek, Balaban and Loewenstein (2013) tied memory to positive feedback. They wrote that associative learning needs memories, that memories need a long time-scale, and that bistability emerges when positive feedback and saturation are both present¹. Their model for associative learning in gene networks assumes a positive feedback loop in which a protein binds the promoter of its own gene and raises its own synthesis. They state that the deterministic approximation of the model's kinetics is equivalent to positive feedback loops such as the lac system². The same two conditions, they state, characterise many gene regulatory networks¹.

Pietak and Levin (2016) used a bioelectric tissue simulation to ask whether feedback can build voltage patterns across a group of cells. In their model, tight junctions let voltage develop in the extracellular space. A positively charged gating ligand collects around more depolarized cells and opens a Na+ channel, which depolarizes them further. Gap junctions close as voltage gradients grow, which sharpens the two regions³. They concluded that positive feedback can generate strong Vmem gradients in a cell collective and may be one way bioelectric pre-patterns are generated and manipulated⁴.

Zakirov and colleagues (2021) studied how endothelial cells settle into tip and stalk roles during angiogenesis, in silico and in vivo. They report that filopodia extend and retract, which moves receptors and changes their activation, and that this amplifies differences between cells in a positive feedback cycle⁵. In their simulations the feedback gave the tip–stalk selection machinery bistable, rapid and robust switch-like behaviour with hysteresis⁶. They argue that filopodia keep the decision timely and may be a basal form of active perception⁵. Information theory measures from computational neuroscience met a necessary but not sufficient criterion for filopodia enriching sensory input⁶.

Other authors place feedback in wider settings and note its costs. Reid (2023), reviewing slime mould cognition, describes a comparison in which ants, which rely more on positive feedback, are more likely to converge quickly on the optimal maze solution but are less adaptable to dynamic changes than Physarum, which likely relies less on it⁸. Fields and Levin (2022) argue that increasing information processing is inevitably a positive feedback loop, an arms race with selection pressures from the environment⁷. Iorio and colleagues (2024) report that filopodia and nanotube machinery do not simply coincide: Eps8, which reduces filopodia formation, positively regulates TNT biogenesis⁹.