The effects of sampling on delimiting species from multi-locus sequence data

As a fundamental unit in biology, species are used in a wide variety of studies, and their delimitation impacts every subfield of the life sciences. Thus, it is of utmost importance that species are delimited in an accurate and biologically meaningful way. However, due to morphologically similar, cryptic species, and processes such as incomplete lineage sorting, this is far from a trivial task. Here, we examine the accuracy and sensitivity to sampling strategy of three recently developed methods that aim to delimit species from multi-locus DNA sequence data without a priori assignments of samples to putative species. Specifically, we simulate data at two species tree depths and a variety of sampling strategies ranging from five alleles per species and five loci to 20 alleles per species and 100 loci to test (1) Structurama, (2) Gaussian clustering, and (3) nonparametric delimitation. We find that Structurama accurately delimits even relatively recently diverged (greater than 1.5 N generations) species when sampling 10 or more loci. We also find that Gaussian clustering delimits more deeply divergent species (greater than 2.5 N generations) relatively well, but is not sufficiently sensitive to delimit more recently diverged species. Finally, we find that nonparametric delimitation performs well with 25 or more loci if gene trees are known without error, but performs poorly with estimated gene genealogies, frequently over-splitting species and mis-assigning samples. We thus suggest that Structurama represents a powerful tool for use in species delimitation. It should be noted, however, that intraspecific population structure may be delimited using this or any of the methods tested herein. We argue that other methods, such as other species delimitation methods requiring a priori putative species assignments (e.g. SpeDeSTEM, Bayesian species delimitation), and other types of data (e.g. morphological, ecological, behavioral) be incorporated in conjunction with these methods in studies attempting to delimit species.

Highlights► We test the sensitivity of three species delimitation methods to sampling strategy. ► Structurama accurately delimits recently diverged species with moderate sampling. ► Gaussian clustering only successfully delimits more divergent species. ► Nonparametric delimitation performs well only if gene trees are without error. ► We suggest combining Structurama with non-molecular data for species delimitation.