Are polyploids really evolutionary dead-ends (again)? A critical reappraisal of Mayrose etal. ().

Abstract

Throughout the past century, hybridization and polyploidization have variously been viewed as drivers of biodiversity (e.g. Arnold, 1997) or evolutionary noise, unimportant to the main processes of evolution (e.g. Stebbins, 1950; Wagner, 1970). Wagner (1970) argued that while polyploids have always existed, they have never diversified or played a major role in the evolution of plants, and that the study of polyploidy (as well as inbreeding, apomixis, and hybridization) has led researchers to be ‘carried away with side branches and blind alleys that go nowhere’. However, the use of molecular tools revolutionized the study of polyploidy, revealing that a given polyploid species often forms multiple times (reviewed in Soltis & Soltis, 1993, 1999, 2000). The realization that recurrent polyploidization from genetically differentiated parents is the rule that shattered the earlier perceptions of polyploids as genetically depauperate (Stebbins, 1950; Wagner, 1970). Because of multiple origins, polyploid species can maintain high levels of segregating genetic variation through the incorporation of genetic diversity from multiple populations of their diploid progenitors (e.g. Soltis & Soltis, 1993, 1999, 2000; Tate et al., 2005). Numerous studies have also shown that polyploid genomes are highly dynamic, with enormous potential for generating novel genetic variation (e.g. Gaeta et al., 2007; Doyle et al., 2008; Leitch & Leitch, 2008; Flagel & Wendel, 2009; Hawkins et al., 2009; Chester et al., 2012; Hao et al., 2013; Roulin et al., 2013). Furthermore, genomic studies have also revealed numerous ancient polyploidy events across the angiosperms (e.g. Vision et al., 2000; Bowers et al., 2003; Blanc & Wolfe, 2004; Paterson et al., 2004; Schlueter et al., 2004; Van de Peer & Meyer, 2005; Cannon et al., 2006; Cui et al., 2006; Tuskan et al., 2006; Jaillon et al., 2007; Barker et al., 2008, 2009; Lyons et al., 2008; Ming et al., 2008; Shi et al., 2010; Van de Peer, 2011; Jiao et al., 2012; McKain et al., 2012; Tayale & Parisod, 2013; reviewed in Soltis et al., 2009); all angiosperms have undergone at least one round of polyploidy (e.g. Jiao et al., 2011; Amborella Genome Consortium, 2013). Polyploidy is now viewed not as a mere side branch of evolution, but as a major mechanism of evolution and diversification. Based on analyses of ferns and angiosperms, Mayrose et al. (2011) and Arrigo & Barker (2012) revived the concept of polyploids as ‘blind alleys’. Arrigo & Barker (2012, p. 140) refer to ‘rarely successful polyploids’ and state that ‘despite leaving a substantial legacy in plant genomes, only rare polyploids survive over the long term and most are evolutionary dead-ends’. Mayrose et al. (2011) also refer to polyploids as ‘dead-ends’. However, these authors use ‘evolutionary dead-end’ in a sense that differs from the traditional view of Stebbins (1950) and Wagner (1970). Whereas Stebbins (1950) and Wagner (1970) were concerned that polyploids did not contribute significantly to evolution, Arrigo & Barker (2012) argue that the analyses of Mayrose et al. (2011) indicate that polyploids are more likely to go extinct than are diploids and in this sense are typically ‘dead-ends’. We agree that most new polyploids likely go extinct early, probably at the population level before they are even detected; this has long been espoused (e.g. Levin, 1975; Ramsey & Schemske, 1998; reviewed in Rieseberg & Willis, 2007; Soltis et al., 2010). Ramsey & Schemske (1998) estimated that the rate of autotetraploid formation is high – comparable to the genic mutation rate – but most do not survive. However, these very early, extinction-prone stages of polyploids are much younger than the timeframe considered by Mayrose et al. (2011) and Arrigo & Barker (2012), who concluded that established polyploid species are more likely to go extinct than diploid congeners. We revisit the analyses and conclusions of Mayrose et al. (2011) and Arrigo & Barker (2012) and ask: Is there convincing evidence to support the hypothesis that extinction rates in established polyploids are higher than those of diploids? Perhaps this is true, but we do not feel that these authors have made a compelling case. While we have focused our review and examples on angiosperms, most of our criticisms are equally applicable to the fern dataset included in their studies. We emphasize philosophical, statistical, and analytical arguments against the study of Mayrose et al. (2011), as well as problems with sampling and their data-mining methods. We support our views by drawing on numerous examples from the recent and classic literature on polyploid complexes (most not included in Mayrose et al., 2011). We stress that our goals are to correct errors and stimulate discussion, not attack the authors of Mayrose et al. (2011) and Arrigo & Barker (2012). Increased diversification rates are more likely to arise in large clades than in small ones, simply because there are more opportunities for increased speciation or retarded extinction rates. Thus, large, old clades will, on average, yield more clades with increased diversification rates than young, small ones, and this probabilistic statement has relevance for patterns of diversification in polyploids relative to their diploid relatives (including their parents). By definition, a polyploid species is younger than its parents. The parental species are part of a clade that is more numerous and older than the single polyploid derivative – thus, through chance, greater diversification could occur at the diploid than the polyploid level. This relationship between clade age, clade size, and diversification – and its particular relevance to diploid–polyploid comparisons – was not considered by Mayrose et al. (2011). Because a polyploid can never be the same age as its parents, analyses within genera with both diploids and polyploids are biased in favor of greater diversification at the diploid level. A more appropriate comparison would be between related diploid and polyploid clades of the same age. Although evolutionary history is represented using bifurcating phylogenetic trees, it is well known that complex sequences of evolutionary events are poorly represented with a branching tree structure (Doolittle, 1999; Huson & Bryant, 2006). Hybridization, hybrid speciation, and allopolyploidy are not described well by bifurcating trees (e.g. Wagner, 1983; Linder & Rieseberg, 2004; McBreen & Lockhart, 2006; Brysting et al., 2007; Soltis & Soltis, 2009; Huson & Scornavacca, 2010). Thus, the evolutionary history of many plant groups is not tree-like, but a network with numerous reticulation events (e.g. Guggisberg et al., 2009; Marcussen et al., 2012; Pellicer et al., 2012; Abbott et al., 2013; Garcia et al., 2014). To reconstruct and visualize complex reticulate evolutionary scenarios, various methods may be employed, including splits graphs, other networks, and analysis of incongruence among gene trees (McBreen & Lockhart, 2006). Comparison of numerous nuclear gene trees is perhaps the best approach to tease apart the complexity of reticulation events (e.g. Linder & Rieseberg, 2004), but to date few studies have used this approach. Mayrose et al. (2011), however, relied entirely on phylogenetic and diversification analyses that have as an underlying assumption that evolution is strictly bifurcating. While we agree that diversification methods have not been developed for nonbifurcating evolutionary histories, we urge caution in interpreting results derived from analyses that violate many of the underlying assumptions of the algorithms used. How polyploids affect phylogeny reconstruction is poorly understood – even at the level of how the topology may be influenced (Soltis et al., 2008; Lott et al., 2009; Marcussen et al., 2012). How those effects cascade onto the accuracy of branch length estimation is unexplored. Finally, how these effects on topology and branch lengths influence diversification analyses is also untested. Again, improved methods and understanding of these nonbifurcating evolutionary influences will help address the question. However, we conclude that current methods are poorly suited when it comes to the complex evolution of polyploids. The conclusions of Mayrose et al. (2011) hinge upon accurate estimation of speciation and extinction rates from molecular phylogenies. Here, we focus on the problems with estimating speciation and extinction rates, as well as the statistical interpretation of the results. Estimating extinction rates from molecular phylogenies based solely upon extant taxa in the absence of a fossil record, as in Mayrose et al. (2011), is problematic (Rabosky, 2009, 2010; Quental & Marshall, 2010). Rabosky (2010) showed that when diversification rates vary among lineages, ‘simple estimators based on the birth–death process are unable to recover true extinction rates’ (p. 1816). Because variation in diversification rate is ‘ubiquitous,’ Rabosky (2010) cautioned that extinction rates should not be estimated in the absence of fossils. Similarly, Quental & Marshall (2010) stress that ‘a wide range of processes can give similarly shaped molecular phylogenies’. Mayrose et al. (2011) found relatively little difference in speciation rates between diploids and polyploids; their conclusion that polyploids diversify at a lower rate is based on higher inferred extinction rates in polyploids – if that rate is unreliably estimated, as others have cautioned, their conclusions cannot be supported. Speciation and diversification rates were estimated with BiSSE (Maddison et al., 2007; FitzJohn et al., 2009). However, Davis et al. (2013) examined the power of the BiSSE method and showed that phylogenies with fewer than 300 taxa should be treated with extreme caution. However, only three datasets in Mayrose et al. (2011) have over 100 terminals. Additionally, Davis et al. (2013) caution against using datasets where the frequency of one of the two states is under 10% – 11 of the 63 genera studied have frequencies outside this range (nine with polyploidy frequencies under 10% and two with polyploidy frequencies over 90%, another is reported as 10% polyploid species). Mayrose et al. (2011) cannot be criticized based on conclusions reached in 2013. However, even when BiSSE was published (Maddison et al., 2007), the authors stressed that large phylogenetic trees would be needed, suggesting c. 500 terminals. ‘It remains to be studied how the probability of rejecting a false null hypothesis (power) varies with number of species and with the degree of difference in rates. From our initial exploration, however, we suspect that large phylogenetic trees will generally be needed to have sufficient power’ (p. 708). They also note (p. 706), ‘This shows at least some power to reject the null hypothesis with trees of 500 species when there is a twofold difference in speciation rates’. Mayrose et al. (2011) show the results of the significance tests for each study but do not draw the reader's attention to the meaning of the information. The last three columns of supplementary material Table S2 from Mayrose et al. (2011) show the percent of the MCMC chains where the difference in diversification, speciation, and extinction estimates of diploids minus polyploids is positive (higher in diploids than polyploids). The test of significance is whether zero (no difference in rates) is outside of the 95% credible interval – or outside of the range seen in 95% of the MCMC steps. Thus, under equal rates, the percent is expected to be 50%, while if 95% of the MCMC steps show higher parameter estimates in diploids than polyploids, this would be a significant result (similarly, if only 5% show higher rates, this would be a significant result in favor of polyploids having higher rates). For diversification rate, only six of 63 genera support significantly higher diversification in diploids, and three of the 63 support significantly lower diversification in diploids relative to polyploids – hardly results that support a conclusion of dramatic differences in diversification rates. For speciation, none of the differences support significantly higher speciation in diploids, while 13 of 63 support higher speciation rates in polyploids – in fact, when examined within each MCMC step (as these data do), speciation is higher in polyploids on average, with only an average of 22% of the MCMC steps showing higher speciation in diploids than polyploids vs 50% for equal rates. While the estimated average speciation rates are higher in diploids (columns 1 and 2 of Table S2 in the supplementary material of Mayrose et al. (2011)), when association with MCMC steps is accounted for, the trend is the opposite. For the extinction results, we assume that the table is mislabeled (while the file has been corrected once since publication, the latest version was updated September 1, 2011, and remains labeled %(μD > μP), that is, that extinction in diploids is higher than polyploids), and the authors intended the label to be ‘%(μD < μP)’, as this would be consistent with their assertion that extinction is higher in polyploids than diploids. With that assumption, 32 of the 63 genera indicate significantly higher extinction in the polyploids. We again point to the Rabosky (2010) caution regarding estimation of extinction rates, as well as the following section as cautionary examples of how even this result may be questionable. In summary, while on average, speciation rates are higher in diploids, the only significant results are in the opposite direction, where polyploids have significantly higher speciation rates in 20% of the genera studied. Additionally, when the estimates are considered within MCMC steps, accounting for correlation among parameters, diploid species only have a higher rate of speciation 22% of the time – indicating that 78% of the time, polyploids have the same or higher speciation rates. The significant extinction rate differences (and thus diversification) seem compromised by the difficulties in estimating extinction rates. These findings alone call into serious question the validity of the conclusions of Mayrose et al. (2011). While further study and better techniques may well support the results, at this point, we argue that the conclusion that polyploids have lower rates of diversification is unjustified. Many polyploids are not recognized taxonomically. Many autopolyploids in particular are unnamed, and these may be especially numerous. J. Ramsey and B. C. Husband (Soltis et al., 2007) found that of 2647 species from 346 genera in 62 angiosperm families in the California Flora, 334 species (13%) have multiple cytotypes (3x, 4x, or higher multiples of the base chromosome number for the genus). Because some of these chromosomally polymorphic taxonomic species actually comprise more than two cytotypes, if each cytotype represented a distinct species, the total number of unrecognized species would be 483, or c. 15% of the flora. Although we, and many others, might consider each cytotype to represent a distinct species, others might advocate a more conservative species concept. If only one quarter of these cytotypes represent distinct species, the impact is considerable (121 species, c. 4% of the flora). This tremendous source of biodiversity has not been captured in the phylogenetic analyses from which Mayrose et al. (2011) drew their data. These enormous gaps in knowledge make it difficult to conduct meaningful analyses and reach firm conclusions on the extinction rate of diploid vs polyploid species. Mayrose et al. (2011) rely on the dataset of Wood et al. (2009) and assume that the angiosperm and fern genera analyzed therein are representative of these groups as a whole. Focusing on angiosperms, this assumption is incorrect for several reasons, and its violation has major implications. First, the Wood et al. (2009) sampling is very small. They sampled only 49 of c. 16 000 genera (0.31%) and 1984 of c. 300 000 species (0.66%) of angiosperms (www.theplantlist.org). In addition, phylogenetic analyses conducted to date represent the initial phase of phylogenetic inquiry. Investigators have not yet analyzed many of the really problematic angiosperm genera (of which there are many) in a thorough manner. Several genera with a large proportion of polyploid species have been analyzed phylogenetically, but were not included in Mayrose et al. (2011) (Supporting Information Table S1). Furthermore, because polyploidization violates the model of bifurcating evolution, systematists have often avoided polyploid complexes in phylogenetic analyses and have removed polyploid species from analyses of genera with high proportions of polyploids. As reviewed later, both factors suggest bias in the dataset of Wood et al. (2009). Using the data from Mayrose et al. (2011), for each angiosperm genus sampled, we plotted the % estimated polyploids vs % total sampled species for that genus. The more complete the sampling, the higher the percentage of polyploids (Fig. 1). These analyses support the idea that polyploid species are the ones excluded in the analyses and that the overall sampling approach is not representative and will lead to biased results. We provide a list of genera (Table S1) with a large proportion of polyploids, all of which have been analyzed phylogenetically. Many of these datasets and topologies (e.g. Aegilops, Draba, Elymus, Festuca, Hordeum, Hedera, Gossypium, Nicotiana, Paeonia, Rubus, and Triticum) could have been included in Mayrose et al. (2011). The percentage of polyploids in most of these genera is high (e.g. Draba, 78%; Festuca, 70%; Hedera, 60%; Hordeum, 50%; Nicotiana, 40%; Triticum, 70%). The frequency of polyploidy is only 10% in Gossypium (Grover et al., 2012) and 25% in Paeonia (Sang et al., 1997); in Gossypium, the polyploids form a clade – from a single origin. Other examples of polyploid complexes have only recently been published (e.g. Opuntia, Majure et al., 2012a; Viola, Marcussen et al., 2012; Fragaria, Njuguna et al., 2013) and hence could not have been included in Mayrose et al. (2011). In most of these recent studies, close to half or more than half of the species are polyploid (e.g. Fragaria, 45%; Opuntia, 59%). A high frequency of polyploids does not necessarily equate to diversification at the polyploid level. However, inclusion of a more representative set of taxa having a high frequency of polyploidy (see later) could greatly alter the results of Mayrose et al. (2011), if many of the polyploid species arose via cladogenesis from other polyploid species rather than from separate polyploidization events from diploid parents. Nicotiana (Solanaceae) comprises 75 species (40 diploids, 35 allopolyploids), with the polyploids ranging in age from 200 000 yr to c. 10 million yr (My) old (Clarkson et al., 2004; Leitch et al., 2008; Kelly et al., 2012). The older polyploid clades generally have more species (Leitch et al., 2008); for example, sect. Suaveolentes contains 26 species, all of which are polyploid (Chase et al., 2003; Marks et al., 2011). Opuntia (Cactaceae) is a young clade (5.6 ± 1.9 million yr old; Arakaki et al., 2011) of c. 200 species with a base chromosome number of x = 11 (Pinkava, 2002). Most species of Opuntia are polyploid, ranging from triploids (2n = 33) to nonaploids (2n = 99). Of 150 species with reported chromosome numbers, 59% were polyploid, 12% had both diploid and polyploid counts, and only 29% were diploid (Majure et al., 2012b). Most species of Triticum/Aegilops (Poaceae) are polyploid. Given that the parentage of polyploid species of Triticum includes several species of Aegilops, neither Triticum nor Aegilops is monophyletic (Petersen et al., 2006; Bordbar et al., 2011), and we consider them together here. Aegilops comprises 22–29 species. Using the higher species estimate, 11 are diploid (2n = 14), 13 are tetraploid (2n = 28), and five are hexaploid (2n = 42). Triticum comprises 11 species: three diploids (2n = 14), six tetraploids (2n = 28), and two hexaploids (2n = 42), including T. aestivum (common wheat). Viola (Violaceae) contains 500–600 species with numerous hybrid and polyploid complexes. From a putative base number of x = 6 or x = 7, extant chromosome numbers range from 2n = 4 (V. modesta) to at least 20-ploid, 2n = c. 160 (V. arborescens). The high polyploids (with > six sets of chromosomes) are monophyletic (Marcussen et al., 2012) and resulted from allodecaploidization 9–14 million yr ago (Mya), involving diploid and two paleotetraploid ancestors. Two of the high-polyploid lineages remained decaploid; recurrent polyploidization with tetraploids within the last five million yr has resulted in two 14-ploid lineages and one 18-ploid lineage (Marcussen et al., 2012). Polyploid speciation within polyploid clades has been a major contributor to diversification in Viola (Marcussen et al., 2012). Many angiosperm genera contain numerous polyploids but lack phylogenetic data. Indeed, classic polyploid complexes remain understudied simply because they are so complex. Thus, the examples analyzed by Mayrose et al. (2011; based on Wood et al., 2009) are not representative of angiosperm diversity, and their conclusions (and Arrigo & Barker, 2012) are premature. Systematists continue to avoid some of the most problematic polyploid complexes (Table S2) in favor of clades for which models of cladogenesis are appropriate and clear results may be obtained. Classic examples of such ‘messy’ genera (Stebbins, 1950, 1971; Grant, 1981) are Claytonia, Crepis, and Crataegus. Claytonia virginica itself comprises over 50 cytotypes in this single recognized species, with numbers ranging from 2n = 12 to c. 191 (Lewis, 1970; reviewed in Doyle, 1983). Salix (Salicaceae) comprises perhaps 400 species, approximately half of which are polyploid (Brunsfeld et al., 1991; Mabberley, 1997). Furthermore, genetic data and the genome sequence of Populus trichocarpa confirmed the prediction based on high chromosome number (Stebbins, 1950, 1971) that Salix and Populus are ancient polyploids (Soltis & Soltis, 1990; Tuskan et al., 2006). Other well-known examples of ‘messy’ groups (due in part to polyploidy) include Castilleja (Tank & Olmstead, 2008), Saxifraga, Micranthes (Webb & Gornall, 1989), Sedum and several other genera of Crassulaceae (Mort et al., 2010), as well as multiple genera of Cactaceae. Some examples are confounded by hybridization and apomixis, but such cases need to be factored into the polyploid equation if we are to ascertain whether polyploids have higher extinction rates than diploids. One can argue that we are preferentially seeking out polyploid complexes to counter Mayrose et al. (2011) and Arrigo & Barker (2012) and that these complexes are also not representative of angiosperms. However, our point is simple – many complex areas of the angiosperm tree of life are highly complicated due to polyploidy. Until more of those complexes are included in comparative studies, any broad assertions of diploid vs polyploid success and extinction are premature. Throughout their paper, Mayrose et al. (2011) adhere to the methods that they describe. However, as we show later, in so doing they ultimately misrepresent some of the originally published trees, yielding unsupported conclusions regarding the impact of polyploidy. We review the methods used by Mayrose et al. (2011), highlighting areas of concern. The bulk of the dataset is based on Wood et al. (2009), which collected data for 143 groups. Chromosome counts were taken from the phylogenetic study, or when not included, obtained from other sources such as the Index to Plant Chromosome Numbers (IPCN) and the Plant DNA C-values Database. Where there were discrepancies, the data reported in the phylogenetic study were favored. Unstated in the methods, but apparent in the supplementary data provided, for species with multiple chromosome numbers, the lowest value was used, generating a bias against polyploids within genera. In many cases, known chromosome counts were omitted with no explanation. Inconsistencies between the dataset of Mayrose et al. (2011) and the original sources were noted. For example, for Physalis (Solanaceae), Mayrose et al. (2011) listed only two polyploids: P. longifolia and P. minima (2n = 48). However, the ICPN lists values for several additional Physalis species, including three with polyploid counts of 2n = 48 (P. angulata (all reports), P. hederaefolia (also 2n = 12, 24), P. peruviana (2n = 48, 72)). In Achillea, A. asiatica, A. biebersteinii, A. crithmifolia, and A. holoserica are all listed in the original study as both 2x and 4x (Guo et al., 2004), but all were considered only 2x in Mayrose et al. (2011). Additional inconsistencies occur for Betula, Erodium, and Centaurium. This treatment of chromosome counts uniformly biased results toward having fewer polyploid counts in the trees, falsely creating a tree that is more diploid than reality. For each study, Mayrose et al. (2011) gathered molecular data and re-analyzed the published datasets. For studies with multiple loci, only those taxa included in the combined analysis were included. In some cases, this led to excluding whole clades of polyploids that had only been sequenced for some loci. While Mayrose et al. (2011) were clear about how the datasets were selected, and why some taxa were removed, the effects of their choices are not explored, either in terms of how many studies were affected or in how the exclusions may have biased the results. In some cases, the dataset chosen by Mayrose et al. (2011) resulted in what seems to be an arbitrary use of 35 to 98% of taxa from the original studies instead of the entire published dataset (Table 1). The removal of taxa can have a major impact. Specifically, in Cerastium, 21 taxa were omitted from the 57 species used in Scheen et al. (2004) – because they were not all sequenced for all genes – with major consequences (see later and Brysting et al., 2007); in Graptopetalum and allies, 15 of 43 taxa were removed (Acevedo-Rosas et al., 2004); in Penstemon, 31 of 163 species were omitted (Wolfe et al., 2006). Table 1 lists 21 examples where Mayrose et al. (2011) excluded species from their analyses – one-third of the studies analyzed. Yet there is no discussion of how this may have affected the results, or whether the removal was biased in terms of ploidy. In some cases, the dataset chosen by Mayrose et al. (2011) resulted in the deletion of polyploid taxa, making the trees ‘more diploid’ than they really are. Tiquilia (Boraginaceae), comprising c. 27 species, is an example in which several polyploids were excluded (Moore et al., 2006) (Fig. 2). Mayrose et al. (2011) only used those taxa that were included in the Moore et al. (2006) 5-gene combined analysis. However, six additional polyploid species were included in Moore et al. (2006). Including these polyploids would have increased the sampled polyploids from three out of 18 species (17%) to nine out of 28 (32%) (Fig. 2). We repeated the analyses of Mayrose et al. (2011) using a combined ITS and rps16 dataset from Moore et al. (2006), but including all 28 taxa for which data were available. With most of the polyploids removed, Mayrose et al. (2011) report that diversification is higher in diploid Tiquilias in 38% of the MCMC steps; with the full dataset, this is only the case in 24% of the steps. Notably, speciation is estimated to be higher in polyploids in both datasets. Extinction was estimated to be higher in polyploids in 82% of the MCMC steps by Mayrose et al. (2011), but in the larger dataset, only 23% of the MCMC steps show this (Fig. 2). This is consistent with Moore et al. (2006), who found that the large polyploid clade in Tiquilia shows high diversification in a short period of time (c. six million yr). Moore et al. (2006) demonstrated that this rapid polyploid diversification is associated with migration into a new geographic area and the evolution of novel morphologies. The deletion of taxa from Cerastium, as well as missing data, by Mayrose et al. (2011) also has a major impact. Cerastium comprises c. 100 species, nearly all of which are polyploid. Only a few species have the lowest (diploid) number of 2n = 18 (C. semidecandrum has 2n = 18, 36, 37; C. lithospermifolium has 2n = 18); many species have 2n = 36. The arctic high-polyploid species (2n = 72 and above) form a clade with relationships best represented as a polytomy. The lack of genetic variation and resolution among the arctic species indicates a recent origin and rapid radiation (Scheen et al., 2004). However, Mayrose et al. (2011) left the chromosome count for C. lithospermifolium as unknown when they ran ChromEvol to infer ploidy. Thus, 2n = 36 was considered the lowest number in Cerastium, and taxa with 2n = 36 were scored as diploid, resulting in a percentage of polyploids in the genus of 42% (Fig. 3). However, C. lithospermifolium is listed in Scheen et al. (2004) as 2n = 18. It is unclear why the known value of 2n = 18 was left as unknown in Mayrose et al. (2011). When ploidy is reconstructed with ChromEvol including the count for C. lithospermifolium, the actual percentage of polyploidy in Cerastium is c. 97% (Fig. 3). When BiSSE is run with the corrected data, instead of diversification being higher in diploids in 88% of the MCMC steps as reported by Mayrose et al. (2011), it is only higher in 76% of the MCMC steps, an odd result, given the distribution of ploidy, that clearly demonstrates the limits of BiSSE for these datasets. The results for Cerastium are a powerful demonstration of the lack of significance in the results reported by Mayrose et al. (2011). Superficially, 76% of the steps indicating higher diversification in diploids sounds like a strong conclusion in favor of the Mayrose et al. (2011) position that diversification is higher in diplo