Strong selection can cause rapid evolutionary change, but temporal fluctuations in the form, direction and intensity of selection can limit net evolutionary change over longer time periods. Fluctuating selection could affect molecular diversity levels and the evolution of plasticity and ecological specialization. Nonetheless, this phenomenon remains understudied, in part because of analytical limitations and the general difficulty of detecting selection that does not occur in a consistent manner. Herein, I fill this analytical gap by presenting an approximate Bayesian computation (ABC) method to detect and quantify fluctuating selection on polygenic traits from population-genomic time-series data. I propose a model for environment-dependent phenotypic selection. The evolutionary genetic consequences of selection are then modeled based on a genotype-phenotype map. Using simulations, I show that the proposed method generates accurate and precise estimates of selection when the generative model for the data is similar to the model assumed by the method. Performance of the method when applied to an evolve-and-resequence study of host adaptation in the cowpea seed beetle (Callosobruchus maculatus) was more idiosyncratic and depended on specific analytical choices. Despite some limitations, these results suggest the proposed method provides a powerful approach to connect causes of (variable) selection to traits and genome-wide patterns of evolution. Documentation and open-source computer software (fsabc) implementing this method are available from GitHub (https://github.com/zgompert/fsabc.git).

This dataset includes mostly output from simulations of evolution, along with associated scripts. An illustrative dataset from an experimental evolution study with seed beetles that was analyzed as part of this methods study is also included.

The following files or sets of files are included here.

ABCoutput.tar.gz. This compressed directory contains the output from simulations used to fit the ABC model. Subdirectories within the main folder contain the subset of output for each set of simulation conditions, along with R scripts used to process the simulation output.

cmac_L14.tar.gz. This compressed directory contains the input files, ABC simulations, and associated processing and analysis scripts for the analysis of the seed beetle (Callosobruchus maculatus) data set on which the method was tested.

methods_comparison.tar.gz. This compressed directory contains the input files, ABC simulations, and associated processing and analysis scripts for comparative analysis assessing the relative performance of the proposed method and the Berg & Coop method for detecting selection that varies in time.

simDat.R. This R script simulates the input files that were used as observed data under baseline conditions.

simDat1k.R. This is similar to simDat.R, but with 1000 loci.

simDat5x5.R. This is similar to simDat.R, but with 5 populations sampled for each of 5 generations.

simGaus.R. This is similar to simDat.R, but for stabilizing selection.