Genetic estimates of population age in the water flea, Daphnia magna

Abstract

Genetic datasets can be used to date evolutionary events, even on recent time scales if sufficient data are available. We used statistics calculated from multilocus microsatellite datasets to estimate population ages in data generated through coalescent simulations and in samples from populations of known age in a metapopulation of Daphnia magna in Finland. Our simulation results show that age estimates improve with additional loci and define a time frame over which these statistics are most useful. On the most recent time scales, assumptions regarding the model of mutation (infinite sites vs. stepwise mutation) have little influence on estimated ages. In older populations, size homoplasy among microsatellite alleles results in a downwards bias for estimates based on the infinite sites model (ISM). In the Finnish D. magna metapopulation, our genetically derived estimated ages were biased upwards. Potential sources of this bias include the underlying model of mutation, gene flow, founder size, and the possibility of persistent source populations in the system. Our simulated data show that genetic age estimation is possible, even for very young populations, but our empirical data highlight the importance of factors such as migration when these statistics are applied in natural populations.

Figure 1.

Boxplots representing the accuracy of age estimation methods: (a) P_K distribution (t) (B) variance in repeat number (s ² _AS). Means are plotted as open and filled circles for datasets of 5 and 20 loci, respectively. Lines extend from the upper and lower quartiles to the maximum and minimum values (excluding outliers), respectively. Outliers are represented as asterisks. True ages are those specified during simulations. The diagonal line indicates perfect correspondence between simulated and estimated ages. Plotted data are from simulations with N _e = 1000. Large Xs on the x axis indicate parameter combinations that frequently resulted in monomorphic datasets. In these instances, both the minimum and the lower quartile values are zero, the log of which is undefined.

Figure 2.

Precision of the age estimation methods employed in this study. CVs are plotted against the log of genome-wide θ, defined here as the sum of the locus-specific θ values (θ _i = 4N _e µ _i). Separate plots are provided for each of the five simulated ages. Within each plot, parameter combinations sampling 5, 10, and 20 loci are indicated with squares, circles, and triangles, respectively. The relationship between CV and the informational content of the dataset [log(genome-wide θ)] is shown for each method. This relationship was assessed using a generalized linear model and trend lines for each individual method are plotted.

Figure 3.

Boxplots of estimated ages for each of 200 replicate datasets simulated under identical conditions. Each individual simulation is composed of 20 loci sampled from a contemporary population of one million alleles. The magnitude of the population expansion was varied from the most severe case (a single founding allele) to a situation wherein 500 diploid individuals (1000 allele copies) colonize a new population. Dotted lines represent the simulated population age, 100 generations.

Figure 4.

Comparison of known (survey data) and estimated (genetic data) ages in generations for the 14 sampled Finnish D. magna populations. Correlations were tested using Pearson’s product moment correlation coefficients; when significant, the correlation coefficient is reported. Estimated ages are from (a) P_K distribution (t) and (B) the variance in allele size (in units of repeats, s ² _AS).