Estimating heritability of pest resistance in forest trees: exploring potential biases from methodological and ecological factors
University of New Brunswick
Heritability describes the proportion of phenotypic trait variance attributed to genetic effects, and is used to identify traits likely to respond to selection. Selectively breeding for increased pest resistance is often appealing to plant breeders. Because resistance is a multifactorial trait (influenced by many host attributes) and cannot be measured directly, to estimate its heritability researchers must select a proxy variable that captures variance in resistance among individuals. Frequently, this proxy is quantified on a non-Gaussian (i.e., non-normal) scale, which can be problematic because non-normal data do not strictly meet the assumptions of the linear models traditionally used to estimate heritability. As a result, non-normal data are either transformed to fit the model or a generalized linear mixed effect model (GLMM) – which can accommodate non-normal data – is used instead. In this thesis, I identify and compare the common methods used to estimate heritability of pest resistance in forest trees. From the literature, no clear evidence suggests that the scale of the proxy or choice of statistical method has a strong influence on estimates. However, my analysis of field data suggests that choice of statistical model for a percent trait can influence estimates of heritability.With field data, the true value of heritability is unknown, making it impossible to determine which model produces a more accurate estimate. As such, to determine how choice of methods (scale and modeling technique) bias estimates of heritability I used data simulations and found that GLMMs can dramatically underestimate heritability. Heritability of pest resistance has an additional level of complexity, as its variance is affected by factors influencing expression in the host and factors influencing the pest population (e.g., fluctuations in pest density or distribution). Using data simulations, I examine how changes in pest densities or predictable environmental patterns in pest distribution, such as edge effects, influence estimates of heritability. I find that pest density and heritability have a parabolic relationship and heritability estimates are strongly reduced when pest damage is influenced by edge effects. Taken together, this thesis contributes toward a better understanding of the factors that contribute to variation in estimates of heritability of non-normal traits. Overall, the data presented within, will help breeders and ecologists better estimate and interpret estimates of heritability, allowing them to make more accurate predictions about how traits will respond to selection.