Phenotypic memory and fitness optimization in fluctuating environments

Kashiwagi, Urabe, Kaneko and Yomo (2006) studied a gene network in cells grown in different media. Single-cell microscopy showed three states, called Attractors 1, 2 and W, each linked to a medium. Cells in Medium N expressed both reporters weakly. Cells in Medium T and Medium M showed strong expression of one reporter and repression of the other. Measured mRNA copy numbers agreed with this. The authors concluded that the cells chose the adaptive attractors reproducibly¹. They allowed that this stochastic route may be less efficient than ordinary signal transduction. They proposed it could be a primitive adaptation that needs no large set of specific sensors, and a preadaptation for later signal-specific pathways².

Lambert and Kussell (2014) asked which lac operon proteins hold cells in an induced state. They made cells express LacZ, LacY or LacA constitutively and timed the lag and recovery phases. With LacZ over-expressed, the phases lasted under 10 minutes. With LacY they lasted about 40 minutes. With LacA they were more variable and usually longer than 60 minutes, which the authors took to rule out a role for LacA in memory³. They then adapted an established metabolic model to fluctuating conditions. In it, protein levels keep rising after an inducer pulse ends, giving an overshoot, if enough mRNA and intracellular inducer remain⁴.

Katz and Springer (2016) place adaptation in a statistical frame. They recall Levins, who argued in early theory that the statistical relations between signals in a fluctuating environment are central to adaptation. They note that strains differ in cost, for example in yeast preparation times after a switch from glucose to galactose, yet the statistics stay informative⁵. Their own framework treats adaptation to changing environments as inference under uncertainty. It is built up toward a strategy for multi-nutrient settings that a simple cellular circuit could in principle carry out. They illustrate it with glucose and galactose in yeast⁶.

Schavemaker and Lynch (2025) review the energetic cost of cellular traits. They describe a cost-benefit experiment on the lac operon, cited as ref. 18, in which expression was altered artificially to different levels in the absence of lactose. Growth rate fell by more than a linear model predicted. The authors note that chaperone burden or cross-reactivity of lac proteins may have contributed⁸. Timsit and Grégoire (2021) state the general case. They hold that cells perceive and integrate stimuli and decide in ways that maintain homeostasis and select survival strategies in fluctuating environments⁷.