Systematic Botany and Mycology, Department of Biology, University of Munich, D-80638 Munich, Germany.
Syst Biol. 2010 Jul;59(4):458-64. doi: 10.1093/sysbio/syq032. Epub 2010 Jun 3.
Studies of diversification patterns often find a slowing in lineage accumulation toward the present. This seemingly pervasive pattern of rate downturns has been taken as evidence for adaptive radiations, density-dependent regulation, and metacommunity species interactions. The significance of rate downturns is evaluated with statistical tests (the gamma statistic and Monte Carlo constant rates (MCCR) test; birth-death likelihood models and Akaike Information Criterion [AIC] scores) that rely on null distributions, which assume that the included species are a random sample of the entire clade. Sampling in real phylogenies, however, often is nonrandom because systematists try to include early-diverging species or representatives of previous intrataxon classifications. We studied the effects of biased sampling, structured sampling, and random sampling by experimentally pruning simulated trees (60 and 150 species) as well as a completely sampled empirical tree (58 species) and then applying the gamma statistic/MCCR test and birth-death likelihood models/AIC scores to assess rate changes. For trees with random species sampling, the true model (i.e., the one fitting the complete phylogenies) could be inferred in most cases. Oversampling deep nodes, however, strongly biases inferences toward downturns, with simulations of structured and biased sampling suggesting that this occurs when sampling percentages drop below 80%. The magnitude of the effect and the sensitivity of diversification rate models is such that a useful rule of thumb may be not to infer rate downturns from real trees unless they have >80% species sampling.
研究多样化模式的人经常发现，谱系积累在接近现在时会放缓。这种明显普遍的速度下降模式被视为适应辐射、密度依赖调节和元群落物种相互作用的证据。速度下降的意义是通过统计测试（伽马统计和蒙特卡罗常数速率（MCCR）测试；出生-死亡似然模型和赤池信息量准则（AIC）得分）来评估的，这些测试依赖于零分布，零分布假设包含的物种是整个进化枝的随机样本。然而，在实际系统发育中采样通常是非随机的，因为系统发育学家试图包括早期分化的物种或先前分类群的代表。我们通过实验修剪模拟树（60 和 150 个物种）以及一个完全采样的经验树（58 个物种）来研究有偏差采样、结构化采样和随机采样的影响，然后应用伽马统计/MCCR 测试和出生-死亡似然模型/AIC 得分来评估速率变化。对于具有随机物种采样的树，在大多数情况下可以推断出真实模型（即拟合完整系统发育的模型）。然而，过度采样深节点会强烈偏向下降的推断，结构化和有偏差采样的模拟表明，当采样百分比下降到 80%以下时，就会发生这种情况。这种影响的幅度和多样化率模型的敏感性使得一个有用的经验法则可能是，除非真实树木的物种采样率超过 80%，否则不要从真实树木中推断出速率下降。
