Evolution by natural selection in age-structured populations in fluctuating environments

Summary of thesis: In age-structured populations, subject to environmental and demographic stochasticity, the response to selection on fitness-related traits will be a complex result of selection at different life stages. Understanding how selection at different stages of the life history interacts to determine the total selection differential is important to enable predictions of evolutionary change. This thesis contributes to this end by showing how current methods for estimating selection can be extended using a demographic framework. Then reproductive values and the stable age distribution can be used to account for the effects of age-structure and estimate selection. The individual reproductive value is the relevant measure of fitness in age-structured populations with no density regulation. This measure of fitness is defined as an individual’s contribution to the total reproductive value of the population next year. Then the total selection on a trait was shown to be a weighted sum of age-specific selection gradients with weights equal to the stable age distribution. This enabled the estimation of temporal mean selection and fluctuation selection using maximum likelihood methods. The R package lmf was developed to implement these methods in statistical analyses. In the Robertson-Price equation, the total change in a mean trait over a time step is separated into two additive components. The first is the covariance of trait and relative fitness (i.e. the selection differential). The second, an expectation which describe how offspring differ from their parents (a transmission term). A generalization of the Robertson-Price equation for a weighted mean was derived. The correct selection differential was obtained by using reproductive values as weights. For any other choice of weights, estimated selection differentials contained transient quasi-selection due to fluctuations in the age-distribution and variation in the mean trait between age classes. Harvesting has repeatedly been shown to be non-random with respect to age in wild populations. For instance, due to a preference for old individuals which also are large. In a harvested moose population, an extension of the generalized Robertson-Price equation for multiple traits was applied to investigate harvest-induced selection. Hunters were shown to induce selection for later birth dates and smaller calf body mass. This may have detrimental effects on the population in the long run, as early birth date and high calf body mass are generally associated with high fecundity in moose. An artificial selection experiment successfully perturbed body size in a population of house sparrows from their natural mean. This induced large age-specific variation in mean phenotypes and fluctuations in the age distribution. Using the R package lmf, natural selection was shown to oppose the artificial selection, indicating the presence of an optimal phenotype. While offspring of artificially selected parents produced less recruits than individuals with unselected parents, there was also some indications that this effect was most pronounced in the first age class. The results in this thesis emphasize the advantages of using methods which allows for handling fluctuations in age distribution and variation in mean phenotypes between age classes when analysing selection in wild populations. Ignoring age-structure may seriously affect inferences of natural selection