Data from: From gene trees to a dated allopolyploid network: insights from the angiosperm genus Viola (Violaceae)

Allopolyploidisation acounts for a significant fraction of speciation events in many eukaryotic lineages. However, existing phylogenetic and dating methods require tree-like topologies and are unable to handle the network-like phylogenetic relationships of lineages containing allopolyploids. No explicit framework has so far been established for evaluating competing network topologies, and few attempts have been made to date phylogenetic networks. We used a four-step approach to generate a dated polyploid species network for the cosmopolitan angiosperm genus Viola L. (Violaceae Batch.). The genus contains ca 600 species and both recent (neo-) and more ancient (meso-) polyploid lineages distributed over 16 sections. First, we obtained DNA sequences of three low-copy nuclear genes and one chloroplast region, from 42 species representing all 16 sections. Second, we obtained fossil-calibrated chronograms for each nuclear gene marker. Third, we determined the most parsimonious multilabelled genome tree and its corresponding network, resolved at the section (not the species) level. Reconstructing the ‘correct’ network for a set of polyploids depends on recovering all homoeologs, i.e. all subgenomes, in these polyploids. Assuming the presence of Viola subgenome lineages that were not detected by the nuclear gene phylogenies (‘ghost subgenome lineages’), significantly reduced the number of inferred polyploidisation events. We identified the most parsimonious network topology from a set of five competing scenarios differing in the interpretation of homoeolog extinctions and lineage sorting, based on (1) fewest possible ghost subgenome lineages, (2) fewest possible polyploidisation events, and (3) least possible deviation from expected ploidy as inferred from available chromosome counts of the involved polyploid taxa. Finally, we estimated the homoploid and polyploid speciation times of the most parsimonious network. Homoploid speciation times were estimated by coalescent analysis of gene tree node ages. Polyploid speciation times were estimated by comparing branch lengths and speciation rates of lineages with and without ploidy shifts. Our analyses recognise Viola as an old genus (crown age 31 Ma) whose evolutionary history has been profoundly affected by allopolyploidy. Between 16 and 21 allopolyploidisations are necessary to explain the diversification of the 16 major lineages (sections) of Viola, suggesting that allopolyploidy has accounted for a high percentage – between 67% and 88% – of the speciation events at this level. The theoretical and methodological approaches presented here for (1) constructing networks and (2) dating speciation events within a network, have general applicability for phylogenetic studies of groups where allopolyploidisation has occurred. They make explicit use of a hitherto underexplored source of ploidy information from chromosome counts to help resolve phylogenetic cases where incomplete sequence data hampers network inference. Importantly, the coalescent-based method used herein circumvents the assumption of tree-like evolution required by most techniques for dating speciation events.

异源多倍化（allopolyploidisation）在许多真核生物类群中占物种形成事件的重要比例。然而，现有系统发育与定年方法均依赖树状拓扑结构，无法处理包含异源多倍体类群的网状系统发育关系。目前尚未建立用于评估竞争性网络拓扑结构的明确框架，且针对系统发育网络的定年研究极少。我们采用四步流程，为世界性被子植物属堇菜属（Viola L.，堇菜科Violaceae Batch.）构建了定年的多倍体物种网络。该属包含约600个物种，涵盖新近形成的（新多倍体）与更为古老的（中古多倍体）多倍体类群，分布于16个组中。首先，我们从代表全部16个组的42个物种中获取了3个低拷贝核基因与1个叶绿体区域的DNA序列。其次，我们为每个核基因标记构建了经化石校准的年代树。第三，我们确定了以组（而非物种）水平解析的最简约多标记基因组树及其对应网络。重建多倍体类群的"正确"网络，有赖于在这些多倍体中回收所有同祖基因（homoeologs），即所有亚基因组。假设存在未被核基因系统发育检测到的堇菜属亚基因组类群（"幽灵亚基因组类群"ghost subgenome lineages），可显著减少推断出的多倍化事件数量。我们基于5种针对同祖基因灭绝与谱系分选的不同解释场景，筛选得到最简约的网络拓扑结构：(1) 幽灵亚基因组类群数量最少；(2) 多倍化事件数量最少；(3) 与所涉多倍类群现有染色体计数推断的预期倍性偏差最小。最后，我们对最简约网络的同倍体物种形成时间与多倍体物种形成时间进行了估算。同倍体物种形成时间通过基因树节点年龄的溯祖分析（coalescent analysis）估算；多倍体物种形成时间则通过比较存在与不存在倍性转变的类群的分支长度与物种形成速率完成。我们的分析确认堇菜属为一个古老类群（冠群年龄为31 Ma），其演化历史深受异源多倍化影响。解释该属16个主要类群（组）的分化，需要16至21次异源多倍化事件，这表明异源多倍化在该分类水平的物种形成事件中占比极高——介于67%至88%之间。本文提出的理论与方法学策略，(1) 构建网状系统发育网络，(2) 对网络内的物种形成事件进行定年，可广泛应用于发生过异源多倍化的类群的系统发育研究。这些策略明确利用了此前未被充分挖掘的染色体计数倍性信息，以解决因序列数据不完整阻碍网络推断的系统发育问题。值得注意的是，本文所用的基于溯祖分析的方法，规避了大多数物种形成事件定年技术所需的树状演化假设。