Maternal effects, stabilizing selection and information fidelity

Background/Question/Methods

Although non-genetic (e.g. maternal) inheritance plays a key role in phenotypic evolution, evolution, its role in organismal adaptation is currently poorly understood. Lande and Kirkpatrick used an extended multivariate breeder’s equation to show that maternal effects cause momentum in phenotypic change. Without non-genetic inheritance, the relationships among multivariate genetic traits are embodied in the genetic variance-covariance matrix G. G encapsulates the response to selection via the multivariate breeders' equation. Under stabilizing selection and in the absence of maternal effects, when phenotypic optima of different time-fluctuating traits are “out of phase”, then there is interference in the signal tracking, phenotypes do not match their optima and fitness is sub-optimal. So what is the impact of maternal effects when traits are under temporally fluctuating, stabilizing selection and the fidelity of the traits is compromised? To answer this question we set up a nonlinear optimization problem, with the maternal effects matrix M as a parameter, to minimize the “square error” between the required and actual phenotypes. Using this optimization formulation, we compute the optimum M in a number of scenarios.

Results/Conclusions

We focus on the illustrative case of two traits under temporally fluctuating, stabilizing selection: (i) when both traits are nearly in phase, so that trait 1 in generation t is a good predictor of trait 2 at generation t+1, the optimization procedure predicts that the off-diagonal elements in the optimum M (covariances between traits) are large. In contrast, when both traits are out of phase, so that the phenotype of one trait is not a good one-generation ahead predictor of the other, then off-diagonal elements in the optimum M are small; (ii) when information in one trait (say, trait 2) is corrupted, for example due to noise, then we find that the optimum M has strong cross talk from trait 1 to trait 2 so as to compensate for the low fidelity in trait 2. More generally, we find that non-genetic inheritance compensates for information loss and reduces information corruption.