Bacterial memory and history-dependent behavior

Lyon (2015) reviews Koshland's account of bacterial chemotaxis. In his 1977 explication, Koshland described the duration of protein phosphorylation as a bacterium's memory, because the period of modification holds information about the state of the environment at a particular time.⁴ Lyon reports that Koshland later located this memory in the enzymes that form and remove the modification. On this view the duration is not fixed. It reflects the difference between a fast excitation process and a slow adaptation process, and depends on the strength and salience of the stimulus.⁴ Lyon adds that response memory varies among chemotactic proteins from seconds to more than 5 minutes, the observational limit in the study he cites.⁵

Wolf, Fontaine-Bodin, Bischofs, Price, Keasling and Arkin (2008) argued that history-dependent behavior may be regulated by the cell and may confer fitness, so that it is more than a by-product of acclimation or metabolic stores.¹ They built an information-theory framework to measure how long memory persists and how much information about the past the dynamics encode. The measure is phenomenological and makes no reference to the specific mechanisms involved.² They applied it to Bacillus subtilis, chosen for its sensitivity to environmental conditions, its bistable and hysteretic regulatory switches, and its sporulation decision.³

Wolf and colleagues set mechanism aside by design.² Lyon (2015) treats mechanism as the unfinished task. He states that bacterial memory has yet to be properly investigated and its mechanisms characterized in different species.⁶ He also calls associative learning in prokaryotes unexplored. He suggests Myxococcus xanthus as a candidate, because it carries 97 genes for serine/threonine protein kinases, and he proposes checking for sequence similarity to CaMKII, a kinase implicated in animal memory.⁶

Katz and Fontana (2022) note that work on probabilistic inference has centered on animal nervous systems, while single-celled organisms also show complex, history-dependent behaviors in changing environments.⁷ They list the cognitive terms applied to this behavior: learning, memory and decision-making.⁷ Biswas, Clawson and Levin (2022) tested memory in computational transcriptional network models. Weaker training stimuli of 10X and 50X baseline, against 100X for normal training, left 52% of models with the same number of memories. They concluded that memory properties are robust across stimulus strengths.⁸