Reconciling Hierarchical Taxonomy with Molecular Phylogenies

Taxonomy has a history dating back to Aristotle (350BC) and has facilitated a wide range of developments in the biological sciences. Linnaeus' Systema Naturae assumed that organisms were static creations of God and formulated the hierarchical framework of classification that we currently use. Today we know that organisms continuously evolve and it is generally accepted that these hierarchies are arbitrary constructs (Coyne and Orr 2004). The arbitrary nature of these higher taxonomic ranks does not prevent their practical use with regards to cataloguing, and communicating about, biological diversity, provided such arbitrary decisions are made on a consistent basis. However, current taxonomic hierarchies are suggested to be inconsistent, both between and within major clades (Avise and Liu 2011), reflecting the work of systematists with diverging views regarding the constitution of higher taxonomic ranks. In this study, we present a comprehensive analysis of the overall levels of consistency within current higher taxonomic ranks using dated phylogenies for all bird and mammal species. Building on work by Hennig (1966), and proposed by Avise and Johns (1999), “temporal banding” provides an opportunity to assess the consistency of ancestral relationships within and among higher taxonomic ranks, as well as a practical solution to these temporal inconsistencies. Temporal banding standardizes taxonomic ranks by “cutting” dated phylogenies at specified points in time and applying this concept requires comprehensive dated phylogenies, which has so far limited the application of the approach. Here, we apply temporal banding to two vertebrate classes (birds and mammals) and produce standardized error metrics to compare the consistency of existing taxonomic ranks within and between the two classes. We discuss the implications of error within the phylogenies used, as well as the practicalities of the temporal banding approach. Taxonomic inconsistencies have implications for our view on the uniqueness of organisms. Studies of ecology, evolution, and conservation often attempt to cover a particular hierarchical rank (for some examples, see Jønsson et al. 2010; Derryberry et al. 2011; Condamine et al. 2012; Holt et al. 2013). These approaches all implicitly assume that hierarchical ranks are to some extent comparable across taxa and some studies even make the assumption explicit by using these categories as units of study (e.g., Lagomarcino and Miller 2012; Rueda et al. 2013). Although the extensive use of current taxonomic rankings demonstrates both their practical benefits and scientific importance, the fact that no attempt has been made to delimit higher rankings consistently is cause for concern (Hennig 1966; de Queiroz and Gauthier 1992; Avise and Liu 2011). We intuitively assume that hierarchical ranks reflect similar temporal evolutionary histories across different taxa, but this is commonly not the case. For example, the age of orders may vary by more than 400 myr between vertebrates and invertebrates (Avise and Liu 2011). Heated debates have proved that it is not a trivial task to produce taxonomies that convey information about the evolutionary history of organisms on Earth (Hennig 1966; Mayr 1974; de Queiroz and Cantino 2001). These debates have centered on the evolutionary theory that should underpin the assignment of taxonomic ranks, with the two main competing views focusing on either the identification of lineages entering new adaptive zones or the identification of clades (Mayr 1974). Although both of these views represent a valid approach to rank taxonomic delimitation in theory, the practical application of these concepts requires relevant data for the focal organisms, as well as an analytical approach that can use such data to delimit taxonomies in a systematic manner. In the absence of such an objective analytical approach, higher taxonomic ranks cannot be expected to represent comparable biological units. Taxonomic inconsistencies have the potential to bias the results of scientific studies, as well as adversely influencing scientific prioritization and comparisons across studies. Consequently, new opportunities to quantify the full extent of such taxonomic inconsistencies need to be explored. The recent emergence of complete dated species-level DNA phylogenies for large well-studied animal groups such as birds (Jetz et al. 2012) and mammals (Bininda-Emonds et al. 2007; Fritz et al. 2009) has greatly benefited the study of the natural world. However, these new hypotheses have not yet been used to systematically review the classifications that ornithologists and mammalogists use on a daily basis. Analyzing levels of temporal divergence within these large phylogenies allows temporal inconsistencies (i.e., inconsistent levels of phylogenetic divergence among taxonomic groups) to be fully quantified within major taxonomic ranks. Temporal divergence is often an important consideration with modern taxonomic classifications and subjective interpretations of morphological and anatomical data to determine taxonomic ranks have clearly led to major temporal inconsistencies within ranks. This traditional focus on morphology has been suggested to inhibit taxonomic objectivity due to the lack of clear homologous features among major groups (Avise and Johns 1999). The goal of this study is to evaluate the consistency of current taxonomic rankings. We take advantage of these large complete, dated phylogenies and examine the extent of temporal inconsistency across bird and mammal genera, families, and orders. Specifically, we (i) produce the first full temporal banding classifications for these groups, (ii) explore the extent of current temporal inconsistencies in the ages of hierarchical clades, and finally (iii) test for phylogenetic bias in these inconsistencies, that is, are specific types of temporal inconsistency more commonly found toward the base or the tips of the trees? For birds, we used the taxonomy and phylogeny from Jetz et al. (2012). These data consisted of 9993 species, divided into 40 orders, 194 families, and 2091 genera. The bird phylogeny is based on molecular data from 6670 species, with the remaining species added according to taxonomy. In order to investigate the potential for bias created by the taxonomic addition of these species without data, we repeated our analysis using only the 6670 species for which molecular data existed. In addition, because the bird taxonomy used by Jetz et al. (2012) was not based on a single existing taxonomy but an amalgamation of several, we also ran our analysis using the International Ornithological Committee (IOC) taxonomy (Gill and Donsker 2013). For mammals, we used the taxonomy of Wilson and Reeder (2005) and the phylogeny of Bininda-Emonds et al. (2007) with updates according to Fritz et al. (2009). These mammal data consisted of 5020 species, divided into 29 orders, 148 families, and 1188 genera. The level of divergence (referred to as “clade age”) within taxonomic groups was calculated as the age, in millions of years ago (ma), of the oldest phylogenetic node connecting the immediately subordinate taxonomic groups within a clade (i.e., species within genera, genera within families, and families within orders). Monotypic taxonomic groups (with only a single immediately subordinate taxonomic group) were not assigned a clade age and are therefore not included within this specific aspect of the study. Note that this approach differs subtly, from the approach known as “crown age,” which is calculated as the age of the node representing the most recent common ancestor of all species within a clade. We rejected this approach because it artificially exaggerates taxonomic inconsistencies by unfairly assigning young ages to monotypic families and orders. For example, for a family that contains only one genus, but multiple species, both the family and the genus would have the same crown age despite the fact that these species would only be expected to show genus-level divergence. For analytical purposes, new hierarchical taxonomic rankings (orders, families, and genera) were produced by cutting the phylogenies at the age that returned the same number of groups as found in the original rankings. These arbitrary cut-off ages allow for direct comparisons with the original classifications and we are neither supporting nor rejecting the wider use of these cut-off ages to produce taxonomic hierarchies for these clades. Since the phylogenies are not fully resolved (the bird phylogeny contains some polytomies, in the form of zero length phylogenetic branches), it was not always possible to return the exact number of groups within ranks as found in the original classification. In these cases, we used the cut-off age that returned the closest possible number of groups within ranks to the original number in the taxonomy. We calculated error scores for each hierarchical taxonomic rank (Fig. 1). This method measures whether taxonomic groups have been either (i) split from their sister clade when they should not have been (i.e., the split is more recent than the standardized cut-off age) or (ii) lumped with their sister clade when they should not have been (i.e., the split with the sister clade is older than the standardized cut-off age). In both cases “temporal error” is measured in millions of years (myr) from the erroneous taxonomic split and the standardized cut-off age, with taxonomic clades that are split too recently giving negative values and taxonomic clades that contain splits older than the cut-off age giving positive values. The mean amount of absolute error was calculated across all groups. Graphical representation of temporal “error scores.” In all cases, the phylogeny is cut to produce four clades (the same as the number of genera according to the existing taxonomy). a) The empirical error is calculated (Genera: Platysmurus, Temnurus Crypsirina, and Dendrocitta). According to the “cut-off age” (dotted green line), the genera Platysmurus and Temnurus should be lumped (indicated by the red color). Dendrocitta formosae should be split from its congeners (indicated in blue). Crypsirina is consistent (indicated in gray) and consequently produces no error. The total error score is calculated as the sum of the absolute distances (in myr) between the node that causes temporal inconsistency and the cut-off age All possible random splits are shown as (b to e), including the one that matches the empirical case (d). For the main analysis, random splits were generated 1000 times and the empirical result was compared with the mean random result to produce standardized error scores. These raw scores are not directly comparable because expected levels of error differ among study groups and among taxonomic ranks. “Standardized error scores” were therefore produced for each taxonomic rank within each taxonomic group. These scores compare the total amount of error to random expectations, resulting in a metric for which zero represents a perfect score (a completely consistent taxonomy), and one represents a score that is equal to the mean random expectation. Random expectations were generated by randomly dividing phylogenies into the required number of monophyletic groups (Fig. 1). For example, for random bird orders, all bird species were randomly divided into 40 monophyletic groups in accordance with the number of orders in the bird taxonomy. Species were divided into random clades by randomly splitting the phylogenetic nodes, starting with the root, until the required number of clades was produced. Nodes were randomly selected for splitting in a weighted manner that ensured that all possible random results were equally possible. This process was repeated 1000 times and the mean random score calculated. Standardized error scores were then calculated by dividing the total observed error by the mean total random error. All analyses were performed using the R statistical analysis software (R Core Team 2013), using the packages ape, foreach and picante (Paradis et al. 2004; Kembel et al. 2010; Analytics and Weston 2013). We investigated whether particular parts of phylogenies are more prone to inconsistent taxonomic treatment. Specifically, we ask whether temporal error is randomly distributed across phylogenetic trees or if it is clustered toward the base or the tips. We conducted Spearman correlation tests to consider potential correlations between taxonomic clade phylogenetic position and clade temporal error score, for all three hierarchical ranks considered and for both birds and mammals. Taxonomic clade phylogenetic position was quantified as root node distance, that is, the number of nodes between the root and ancestral node of the clade, with clades having low root node distances referred to as being more basal and those with higher scores referred to as being more distal. Clade temporal error scores were calculated as explained in the “Consistency among taxonomic schemes” section. For birds, the new phylogenetically delineated orders, families, and genera were generated by cutting the phylogeny at 65, 37.5, and 11.405 ma, respectively (Supplementary Material S1, (available from http://dx.doi.org/10.5061/dryad.qd3pd). Standardized error scores for the three bird taxonomic ranks show that the taxonomy is most consistent at the order level and least consistent at the genus level (Fig. 2). Overall 695 of the 2325 taxonomic clades were completely consistent between the original and revised classifications, with roughly half of consistent clades being monotypic. Traditional hierarchical taxonomic clade age distributions overlapped considerably (Fig. 3a) and clades varied substantially in their amount of temporal error (Fig. 4a–c; Supplementary Material S2). The most extreme comparison was between the order Casuariiformes, “emus and cassowaries” (clade age = 22.31 ma), and the genera Caprimulgus and Eurostopodus, both nightjars, with clade ages of 55.13 ma. Our two repeated versions of the bird analysis, one using the data-only phylogeny and the other using the IOC taxonomic classification, both produced highly similar results to the main analysis (Supplementary Material S3). Consistency of current higher taxonomic classifications for birds and mammals, for genera, families, and orders. As measured by percentage of groups that are identical to a group produced by a temporal banding approach (bars) or by standardized error scores (points), for which zero = no error and one = mean random error. Standardized error scores are the most appropriate metric for comparing study groups or hierarchical ranks. See the for analytical for standardized error scores. for current hierarchical taxonomic clade ages of a) birds and mammals traditional taxonomy and the new temporal banding approach, in which time phylogenies are cut at specific points in time to consistent hierarchical taxonomic ranks. bias of taxonomic for bird and mammal orders, families, and genera. represent the cut-off ages required to split the phylogenies into the existing number of groups for each taxonomic a clade is split too recently (i.e., be the phylogenetic is a taxonomic clade is split too ago (i.e., be the phylogenetic is In both cases the of the temporal from the cut-off that are consistent with the temporal banding approach are For birds and mammals, a negative correlation between the root node age of clades and clade temporal error for families and genera (birds and and both For orders no correlation between these For mammals, the new phylogenetically based orders, families, and genera were generated by cutting the phylogeny at and ma, respectively (Supplementary Material to polytomies, it was not possible to cut-off ages the same number of clades for families and genera as suggested by Wilson and Reeder The selected cut-off ages returned the closest possible number of clades, with more genera and more families standardized error scores show that the taxonomy is most consistent at the order level and least consistent at the genus level (Fig. 2). hierarchical taxonomic ranks, the mammal taxonomy was more consistent than the bird taxonomy (Fig. 2). of the original mammal clades were completely consistent with a clade in the temporal banding with more than half of consistent clades being monotypic. The traditional mammal taxonomic clade age distributions overlapped across taxonomic ranks (Fig. and clades varied substantially in their amount of temporal error (Fig. Supplementary Material S2). The most extreme comparison was between the order (clade age = ma), and the genus (clade age = ma), which is within the For both birds and mammals, families and genera a negative correlation between root node distances and clade error scores (birds and and both This result that more basal taxonomic clades are more to be split temporal banding and more clades are more to be not show correlation for either birds or mammals is as Supplementary Material analyses that the current bird taxonomy is more inconsistent than that of mammals within the three taxonomic ranks hierarchical taxonomic ranks, mammal groups are consistently older than bird groups. For both birds and mammals, the amount of temporal inconsistency among clades as taxonomic hierarchical ranks genera are inconsistent and only than a randomly delineated taxonomy. only the of the bird phylogeny for which molecular data provides this that the temporal banding approach may even be to phylogenies and to the of current families and genera, basal taxonomic groups are more to be split a temporal banding approach, with negative correlations between phylogenetic position and temporal error within these taxonomic ranks, for both study groups. Our results have important implications for bird and mammal studies based on higher taxonomic ranks our results are consistent across the two groups, is also the potential that similar studies based on other groups be in a similar manner. for orders compare well with expectations and are for current This hierarchical rank error and no of phylogenetic bias for splitting or This result that currently observed in the number of species within orders may well be a of similar in ancestral For example, the bird order completely temporal with only of the species being For mammals, two orders currently and Although is split into new orders temporal one of these new orders for a of mammal species. fully and for a of mammal species. show a level of consistency that is between that shown by orders and temporal banding at this hierarchical rank more species than at the other two ranks. The for this is that revised families to contain more species than families that As a result of bird species and of mammal species have their families revised temporal The nature of temporal banding family to be by the position of families within the For birds and mammals, most of the existing families that were split temporal banding represent basal some for The most basal bird (the family within is also the most extreme of temporal banding to splitting into two orders to the currently of and with four new families and two of the most basal with split into new families temporal banding and split into two families by myr to the cut-off age for orders). of the most families were split temporal for example, only of the most of the families of these families were For birds, of the most of families are split temporal banding but more of these families are lumped The most extreme case of for birds (in of number of families is by the existing which of families, of which are lumped into one single family by temporal with the remaining family being the single species family In the of extreme temporal current families are lumped into a single family that is currently as the it that the of birds and mammals into orders have been in a consistent manner by the more into families and genera have led to a wide of temporal The total number of groups may an important with regards to consistency across taxonomic For example, it is that a can consider all 29 existing groups when current orders but an of a genera would be more without analytical Although we on two complete phylogenies for birds and mammals, it is important to be that the phylogenetic data are not of the temporal inconsistency shown may be due to phylogenetic error. Although these large phylogenies represent important they also represent the first at complete phylogenetic hypotheses at the of classes and new phylogenies some of the levels of in these phylogenies, we would not full taxonomic based on our new taxonomies without This is relevant at genus level the of species within the phylogenies without data is to be it should be that these phylogenies represent the of of systematic relationships within these groups and the results of our temporal banding specific lineages that may taxonomic The results for our bird analysis, which only species within the phylogeny based on molecular data species of 9993 are similar to the main which that the temporal banding approach may be to As more data are and phylogenetic we systematic within birds and mammals. because hierarchical ranks are is not a single number of groups at particular For analytical purposes, we delineated taxonomic hierarchical ranks using cut-off ages that returned the same number of groups shown by current is that cut-off ages may be on other such as morphology and to the temporal cut-off it is only too to the levels of among on such cut-off ages the arbitrary nature of these ranks, the approach used for our analysis is more and existing or objective analytical may an approach that cut-off ages and The hierarchical taxonomic has and to have a on modern taxonomy. the of phylogenetic data, and some for the of higher taxonomic ranks Queiroz and Gauthier the use of this classification as as both within and across the wider However, current taxonomic hierarchical ranks have not yet been to contain comparable groups for aspect of or Linnaeus' the of the of Linnaeus' work and that to by to the creations of the The of the of by the of a new in biological than organisms being static in need of they were considered continuously and is assumed to to a taxonomic from traditional taxonomic and toward an on evolutionary (e.g., Hennig 1966; de Queiroz temporal for a hierarchical a of species within a taxonomic group be more than of species that are in different taxonomic groups. Our analysis demonstrates that this is not to being in current bird and mammal taxonomic for families and genera. This temporal inconsistency the of evolutionary relationships among species, as well as biological analyses based on taxonomic hierarchical ranks. The new hierarchical ranks produced in this study convey comparable information about the temporal evolutionary histories of birds and mammals. based on taxonomic such as classifications based on ecology, or evolution, be considered to be a approach to some but would be highly if not For example, as by (Mayr requires decisions regarding or not it has a new adaptive and to extent it has a major to the lack of homologous features across the of it is to such decisions can be made on a consistent basis. The temporal banding approach is not completely from practical the of molecular data, be assigned to lineages their morphological this process be assigned to higher groups by temporal banding ranks in the same manner as they are Our study demonstrates that the temporal banding approach can produce consistently taxonomic groups. These groups therefore represent more units of comparison when other of such as divergence. Temporal banding hierarchies also more consistent units of study for analysis, in cases it may be to such analysis using the phylogenies directly than a taxonomic classification. Our consistent taxonomies should a of specific current taxonomic groups for birds and mammals, and this study that a consistent of large ranks is both and until a consistent of taxonomic ranks is to the complete of taxonomies to reflect the continuously nature of on from the the of in and the This work was by from the of the The and for their on this study and for their on the