Species are a fundamental unit of biodiversity that are delimited via genetic data and coalescent-based methods with increasing frequency. Despite the widespread use of coalescent-based species delimitation, we do not fully understand the sensitivity of these methods to potential sources of bias and violations of their underlying assumptions. One implicit assumption of coalescent-based species delimitation is that geographic sampling is adequate and representative of genetic variation among populations within the lineage of interest. Yet exhaustive geographic sampling is logistically difficult, if not impossible, for many taxa that span large geographic expanses or occupy remote regions. Here, we examine the impact of geographic sampling on the output of Bayes-factor delimitation with SNAPP, a popular coalescent-based species delimitation pipeline. First, we demonstrate the problematic nature of sparse geographic sampling and isolation by distance for species delimitation using simulated data sets of populations connected by different levels of gene flow. We then examine whether similar trends are present in an empirical dataset of Andesiops mayflies (Ephemeroptera: Baetidae) from a high elevation transect in the Ecuadorian Andes. In both the simulated and empirical analyses, we systematically exclude geographically intermediate sites to quantify the impact of geographic sampling and isolation by distance on coalescent-based species delimitation. We find that removing intermediate sites with genetically admixed individuals incorrectly favors multi-species delimitation scenarios. Oversplitting is especially pronounced when isolation by distance is strong, but exists even when gene flow among neighboring populations is relatively high. These findings highlight the importance of adequate geographic sampling in species delimitation and urge caution in interpreting the output of such methods when species’ distributions are sparsely sampled and in systems characterized by strong patterns of isolation by distance.

中文翻译：

---

基于聚结的物种定界对地理采样和距离隔离很敏感

物种是生物多样性的基本单位，通过遗传数据和基于聚结的方法越来越频繁地进行界定。尽管广泛使用基于聚结的物种定界，但我们并不完全了解这些方法对潜在的偏见来源和违反其基本假设的敏感性。基于聚结的物种定界的一个隐含假设是地理采样是足够的并且代表了感兴趣谱系内种群之间的遗传变异。然而，对于跨越大面积地理区域或占据偏远地区的许多分类群来说，详尽的地理抽样在逻辑上是困难的，如果不是不可能的话。在这里，我们使用 SNAPP（一种流行的基于聚结的物种定界管道）检查地理采样对贝叶斯因子定界输出的影响。首先，我们使用由不同基因流水平连接的种群的模拟数据集证明了稀疏地理采样和物种定界距离隔离的问题性质。然后，我们检查来自厄瓜多尔安第斯山脉高海拔横断面的 Andesiops 蜉蝣（蜉蝣科：Baetidae）的经验数据集中是否存在类似的趋势。在模拟和实证分析中，我们系统地排除了地理中间站点，以量化地理采样和距离隔离对基于聚结的物种定界的影响。我们发现删除具有基因混合个体的中间位点不正确地有利于多物种定界方案。当距离隔离很强时，过度分裂尤其明显，但即使在相邻种群之间的基因流动相对较高时也存在。这些发现强调了充分地理采样在物种划界中的重要性，并敦促在物种分布稀疏采样和以距离隔离的强烈模式为特征的系统中解释此类方法的输出时要谨慎。