Measuring viability selection from prospective cohort mortality studies: A case study in maritime pine

Abstract

By changing the genetic background available for selection at subsequent life stages, stage-specific selection can define adaptive potential across the life cycle. We propose and evaluate here a neutrality test and a Bayesian method to infer stage-specific viability selection coefficients using sequential random genotypic samples drawn from a longitudinal cohort mortality study, within a generation. The approach is suitable for investigating selective mortality in large natural or experimental cohorts of any organism in which individual tagging and tracking are unfeasible. Numerical simulation results indicate that the method can discriminate loci under strong viability selection, and provided samples are large, yield accurate estimates of the corresponding selection coefficients. Genotypic frequency changes are largely driven by sampling noise under weak selection, however, compromising inference in that case. We apply the proposed methods to analyze viability selection operating at early recruitment stages in a natural maritime pine (Pinus pinaster Ait.) population. We measured temporal genotypic frequency changes at 384 candidate-gene SNP loci among seedlings sampled from the time of emergence in autumn until the summer of the following year, a period with high elimination rates. We detected five loci undergoing allele frequency changes larger than expected from stochastic mortality and sampling, with putative functions that could influence survival at early seedling stages. Our results illustrate how new statistical and sampling schemes can be used to conduct genomic scans of contemporary selection on specific life stages.

Keywords: Pinus pinaster; adaptive conservation management; contemporary adaptation; longitudinal study; neutrality test; selection coefficient; selective mortality.

Figure 1

Effect of sample size (n) and minor allele frequency on selection coefficient estimates for a neutral locus (s = 0). RMSE is the root mean square error, and NCR the noncoverage rate of 95% credible intervals (the dotted line shows the nominal 5% value). Bias and RMSE were measured with respect to the mean (white bars) or with respect to the median (gray bars). Based on 1,000 Monte Carlo replicates per scenario, assuming a cohort of size N = 10,000 and a biallelic locus

Figure 2

Effect of sample size (n) and minor allele frequency on selection coefficient estimates for a locus under strong selection (s = −0.1). RMSE is the root mean square error, and NCR the noncoverage rate of 95% credible intervals (the dotted line shows the nominal 5% value). Bias and RMSE were measured with respect to the mean (white bars) or with respect to the median (gray bars). Based on 1,000 Monte Carlo replicates per scenario, assuming a cohort of size N = 10,000 and a locus with two codominant alleles

Figure 3

Effect of sample size (n) and minor allele frequency on selection coefficient estimates for a locus under weak selection (s = −0.01). RMSE is the root mean square error, and NCR the noncoverage rate of 95% credible intervals (the dotted line shows the nominal 5% value). Bias and RMSE were measured with respect to the mean (white bars) or with respect to the median (gray bars). Based on 1,000 Monte Carlo replicates per scenario, assuming a cohort of size N = 10,000 and a locus with two codominant alleles

Figure 4

Examples of posterior probability distributions of selection coefficient estimates in a simulated cohort mortality study under different selective scenarios. Each gray point and line indicates the posterior median and 95% credibility interval for the selection coefficient of one locus in an independent Monte Carlo replicate. Horizontal black lines indicate the assumed value of the selection coefficient. Obtained assuming a large simulated cohort (N = 10,000), two temporal samples of size n = 1,000, and biallelic loci with minor allele frequency of 0.3

Figure 5

Effect of sample size and minor allele frequency on the false positive (Type I error) rate and the power of neutrality tests. The Type I error was calculated assuming a neutral locus (s = 0; top panel), and the power assuming a strongly (s = −0.1; middle panel) or a weakly (s = −0.01; bottom panel) selected locus. The tests corresponded to either the quasi‐exact neutrality test in Equation (2) (“QE” white symbols) or to the proportion of times the 95% CI of Bayesian s estimates did not include zero (“By” black symbols). Assumed temporal sample sizes were n = 100 (circles), n = 500 (triangles), n = 1,000 (diamonds) and n = 10,000 (squares). Based on 1,000 Monte Carlo replicates per scenario, assuming a cohort of size N = 10,000 and a locus with two codominant alleles

Figure 6

Distribution of standardized temporal allele frequency changes observed at 237 polymorphic SNP loci in a Pinus pinaster seedling cohort mortality study in Fuencaliente (Spain). The two sampling periods were November 2010–March 2011 (white bars) and March–July 2011 (gray bars). Changes larger than expected (p < 0.05 after FDR correction) from random mortality and sampling are marked with asterisks