An empirical evaluation of two-stage species tree inference strategies using a multilocus dataset from North American pines

Abstract

BackgroundAs it becomes increasingly possible to obtain DNA sequences of orthologous genes from diverse sets of taxa, species trees are frequently being inferred from multilocus data. However, the behavior of many methods for performing this inference has remained largely unexplored. Some methods have been proven to be consistent given certain evolutionary models, whereas others rely on criteria that, although appropriate for many parameter values, have peculiar zones of the parameter space in which they fail to converge on the correct estimate as data sets increase in size.ResultsHere, using North American pines, we empirically evaluate the behavior of 24 strategies for species tree inference using three alternative outgroups (72 strategies total). The data consist of 120 individuals sampled in eight ingroup species from subsection Strobus and three outgroup species from subsection Gerardianae, spanning ∼47 kilobases of sequence at 121 loci. Each “strategy” for inferring species trees consists of three features: a species tree construction method, a gene tree inference method, and a choice of outgroup. We use multivariate analysis techniques such as principal components analysis and hierarchical clustering to identify tree characteristics that are robustly observed across strategies, as well as to identify groups of strategies that produce trees with similar features. We find that strategies that construct species trees using only topological information cluster together and that strategies that use additional non-topological information (e.g., branch lengths) also cluster together. Strategies that utilize more than one individual within a species to infer gene trees tend to produce estimates of species trees that contain clades present in trees estimated by other strategies. Strategies that use the minimize-deep-coalescences criterion to construct species trees tend to produce species tree estimates that contain clades that are not present in trees estimated by the Concatenation, RTC, SMRT, STAR, and STEAC methods, and that in general are more balanced than those inferred by these other strategies.ConclusionsWhen constructing a species tree from a multilocus set of sequences, our observations provide a basis for interpreting differences in species tree estimates obtained via different approaches that have a two-stage structure in common, one step for gene tree estimation and a second step for species tree estimation. The methods explored here employ a number of distinct features of the data, and our analysis suggests that recovery of the same results from multiple methods that tend to differ in their patterns of inference can be a valuable tool for obtaining reliable estimates.

Inferring rooted species trees from unrooted gene trees using approximate Bayesian computation

“…When the species tree had a pseudocaterpillar shape, the correct species tree was returned 85% and 100% of the time with branch lengths of 0.1 and 1.0, respectively. Strong shape biases have been known to exist for rooted MDC in previous research (Than and Rosenberg, 2011; DeGiorgio et al , 2014), and more work would be needed to understand its behavior in the unrooted setting.…”

Phylogenomic subsampling: a brief review

“…; DeGiorgio et al . ; reviewed in Edwards ). These methods are known to vary in their efficiency and also in their ability to handle large data sets, with Bayesian methods being the desired goal, specifying the MSC completely but unable to handle large data sets, and summary statistic methods being less efficient, but still statistically consistent and able to handle data sets befitting the title ‘phylogenomics’.…”

“…We use a model, the multispecies coalescent (MSC) model, that provides an elegant means of assessing the likelihood of an overarching species tree in the presence of gene tree variation that is dominated by incomplete lineage sorting (ILS; Rannala & Yang 2003, 2008Degnan & Rosenberg 2009;Liu et al 2009aLiu et al , 2015b. As reviewed extensively elsewhere, there is now a diverse ecosystem of species tree methods, ranging from Bayesian to likelihood to summary statistic methods, and which can handle a variety of data types, from multiple SNPs linked into single loci to multiple unlinked SNPs (Bryant et al 2012;DeGiorgio et al 2014;reviewed in Edwards 2016). These methods are known to vary in their efficiency and also in their ability to handle large data sets, with Bayesian methods being the desired goal, specifying the MSC completely but unable to handle large data sets, and summary statistic methods being less efficient, but still statistically consistent and able to handle data sets befitting the title 'phylogenomics'.…”

Mosaic genetic differentiation along environmental and geographic gradients indicate divergent selection in a white pine species complex

“…3c), despite the shared polymorphism observed. Shared polymorphism and contrasting genealogical signals among genes, have been observed for this species complex in a recent study by DeGiorgio et al (2014). Under a scenario of drift and gene flow (Knowles and Carstens 2007;Yeaman and Whitlock 2011), we would expect to see a similar pattern of differentiation across different loci between species, as gene flow would act on entire genomes (Shafer and Wolf 2013).…”

“…However, shared polymorphism is a common characteristic in white pines, and represents a challenge for multilocus genetic studies, as patterns are not clear cut (DeGiorgio et al 2014). Therefore, it becomes important to explore the role of divergent selection in genes reinforcing postzygotic reproductive barriers by reducing the fitness of hybrids, and thus promoting speciation even if the rest of the genome is not affected in the same fashion (Smadja and Butlin 2011).…”