Accounting for relatedness and spatial structure to improve plant phenotypic selection in the wild

Abstract

Identifying natural selection in wild plant populations is a challenging task, as the reliability of selection coefficients depends, among other factors, on the critical assumption of data independence. While rarely examined, selection coefficients may be influenced by the spatial and genetic dependence among plants, which violates the independence criterion, leading to biased selection estimates. In this study, we examine the extent to which frugivore-mediated selection coefficients are influenced by spatial and genetic information. We used Generalized Additive Models to deal with spatial and relatedness issues. We compared the fit of the Lande and Arnold multivariate model with models including spatial, genetic relatedness, and spatial + genetic relatedness corrections. Our results indicate that fit in standard models was substantially increased after including the spatial structure. Likewise, the model including the genetic relatedness accounted for a variance fraction not explained by spatial structure, which permitted the identification of significant selection acting upon fruit size, a trait not detected under selection otherwise, and dealt better with autocorrelation that any other model. The model including spatial and genetic effects altogether accounted for 65% of the variance, compared to 13% of the standard model. The spatial structure and genetic relatedness played an important role in this system. As genetic effects revealed significant selection upon fruit traits otherwise hidden under standard selection estimates, field studies that control for plant dependency may provide more realistic selection estimates in natural plant populations.