Phenotypic variability can promote the evolution of adaptive plasticity by reducing the stringency of natural selection

Abstract

Adaptive phenotypic plasticity is a potent but not ubiquitous solution to environmental heterogeneity, driving interest in what factors promote and limit its evolution. Here, a novel computational model representing stochastic information flow in development is used to explore evolution from a constitutive phenotype to an adaptively plastic response. Results show that populations tend to evolve robustness to developmental stochasticity, but that this evolved robustness limits evolvability; specifically, robust genotypes have less ability to evolve adaptive plasticity when presented with a mix of both the ancestral environment and a new environment. Analytic calculations and computational experiments confirm that this constraint occurs when the initial mutational steps towards plasticity are pleiotropic, such that mutant fitnesses decline in the environment to which their parents are well‐adapted. Greater phenotypic variability improves evolvability in the model by lessening this decline as well as by improving the fitness of partial adaptations to the new environment. By making initial plastic mutations more palatable to natural selection, phenotypic variability can increase the evolvability of an innovative, plastic response without improving evolvability to simpler challenges such as a shifted optimum in a single environment. Populations that evolved robustness by negative feedback between the trait and its rate of change show a particularly strong constraining effect on the evolvability of plasticity, revealing another mechanism by which evolutionary history can limit later innovation. These results document a novel mechanism by which weakening selection could actually stimulate the evolution of a major innovation.

Phenotypic noise and the cost of complexity

“…The fitness function w(z) is assumed positive, three times differentiable, and has one nondegenerate optimum at z = 0 (Martin 2014). We extended the fitness function from Tenaillon (2014) by adding a minimal fitness (Zhang et al 2009;Draghi 2019),…”

“…When β = 0 and Q = 2, w(z) is Gaussian-shaped. Note that a nonnull β is sometimes used in numerical simulations (Zhang et al 2009;Draghi 2019). The absolute fitness of the genotype {μ, σ} (or genotypic fitness) is (Lande 1979…”

Phenotypic Noise and the Cost of Complexity

“…The fitness function w(z) is assumed positive, three times differentiable, and has one non-degenerate optimum at z = 0 (Martin, 2014). We extended the fitness function from Tenaillon (2014) by adding a minimal fitness (Zhang et al, 2009;Draghi, 2019),…”

“…When β = 0 and Q = 2, w(z) is Gaussian-shaped. Note that a non-null β is often used in numerical simulations (Zhang et al, 2009;Draghi, 2019).…”

“…Since phenotypic noise increases the genotypic fitness of an organism far from the fitness optimum, it seems logical that it should facilitate adaptive evolution by increasing the fitness of beneficial mutations (Zhang et al, 2009;Bódi et al, 2017;Draghi, 2019;Schmutzer and Wagner, 2020). However, this claim is in contradiction with classical theory that states that phenotypic noise always reduces the selection gradient (Lande, 1975;Gavrilets and Hastings, 1994;Wang and Zhang, 2011;Mineta et al, 2015), as confirmed experimentally (Mulder et al, 2016;Keren et al, 2016).…”