Data from: The effects of inference method, population sampling and gene sampling on species tree inferences: an empirical study in slender salamanders (Plethodontidae: Batrachoseps)

Species tree methods are now widely used to infer the relationships among species from multi-locus datasets. Many methods have been developed, which differ in whether gene and species trees are estimated simultaneously or sequentially, and in how gene trees are used to infer the species tree. While these methods perform well on simulated data, less is known about what impacts their performance on empirical data. We used a dataset including five nuclear genes and one mitochondrial gene for 22 species of Batrachoseps to compare the effects of method of analysis, within-species sampling and gene sampling on species tree inferences. For this dataset, the choice of inference method had the largest effect on the species tree topology. Exclusion of individual loci had large effects in *BEAST and STEM, but not in MP-EST. Different loci carried the greatest leverage in these different methods, showing that the causes of their disproportionate effects differ. Even though substantial information was present in the nuclear loci, the mitochondrial gene dominated the *BEAST species tree. This leverage is inherent to the mtDNA locus and results from its high variation and lower assumed ploidy. This mtDNA leverage may be problematic when mtDNA has undergone introgression, as is likely in this dataset. By contrast, the leverage of RAG1 in STEM analyses does not reflect properties inherent to the locus, but rather results from a gene tree that is strongly discordant with all others, and is best explained by introgression between distantly related species. Within-species sampling was also important, especially in *BEAST analyses, as shown by differences in tree topology across 100 subsampled datasets. Despite the sensitivity of the species tree methods to multiple factors, five species groups, the relationships among these, and some relationships within them, are generally consistently resolved for Batrachoseps.

当前，物种树推断方法已被广泛应用于基于多位点数据集推演物种种间亲缘关系。目前已开发出多种此类方法，其差异体现在基因树与物种树的估计是同步进行还是分步完成，以及利用基因树推演物种树的具体策略。尽管此类方法在模拟数据上表现优异，但学界对哪些因素会影响其在实证数据中的性能仍知之甚少。本研究采用涵盖22种无肺螈属（Batrachoseps）物种、包含5个核基因与1个线粒体基因的数据集，对比分析了分析方法、物种种内采样策略以及基因采样策略对物种树推断的影响。针对该数据集，推断方法的选择对物种树拓扑结构的影响最为显著。剔除单个基因位点对*BEAST与STEM的物种树推断结果存在显著影响，但对MP-EST并无此类效果。不同方法中对推断结果发挥最大调控作用的位点各不相同，这表明此类位点不成比例影响的成因存在差异。尽管核基因位点已蕴含丰富的系统发育信息，但线粒体基因主导了*BEAST方法构建的物种树。这种影响作用是线粒体DNA（mtDNA）位点固有的特性，源于其较高的进化变异率与较低的假定倍性。当线粒体DNA发生基因渐渗时，这种主导性影响可能会引发问题，而本数据集的情况大概率存在此类渐渗事件。与之相反，STEM分析中RAG1基因（Recombination Activating Gene 1）对推断结果的影响并非源于该位点自身的固有特性，而是源于其基因树与其他所有基因树存在强烈的拓扑冲突，这种冲突最合理的解释是远缘物种种间发生了基因渐渗。物种种内采样策略同样具有重要影响，尤其在*BEAST分析中表现明显——100个不同子采样数据集得到的物种树拓扑结构存在差异，也印证了这一点。尽管物种树推断方法对多种因素存在敏感性，但针对无肺螈属（Batrachoseps）的分析仍稳定解析出5个物种类群、类群间的亲缘关系以及部分类群内部的亲缘关系。