A multi-calibrated mitochondrial phylogeny of extant Bovidae (Artiodactyla, Ruminantia) and the importance of the fossil record to systematics

Background Molecular phylogenetics has provided unprecedented resolution in the ruminant evolutionary tree. However, molecular age estimates using only one or a few (often misapplied) fossil calibration points have produced a diversity of conflicting ages for important evolutionary events within this clade. I here identify 16 fossil calibration points of relevance to the phylogeny of Bovidae and Ruminantia and use these, individually and together, to construct a dated molecular phylogeny through a reanalysis of the full mitochondrial genome of over 100 ruminant species. Results The new multi-calibrated tree provides ages that are younger overall than found in previous studies. Among these are young ages for the origin of crown Ruminantia (39.3–28.8 Ma), and crown Bovidae (17.3–15.1 Ma). These are argued to be reasonable hypotheses given that many basal fossils assigned to these taxa may in fact lie on the stem groups leading to the crown clades, thus inflating previous age estimates. Areas of conflict between molecular and fossil dates do persist, however, especially with regard to the base of the rapid Pecoran radiation and the sister relationship of Moschidae to Bovidae. Results of the single-calibrated analyses also show that a very wide range of molecular age estimates are obtainable using different calibration points, and that the choice of calibration point can influence the topology of the resulting tree. Compared to the single-calibrated trees, the multi-calibrated tree exhibits smaller variance in estimated ages and better reflects the fossil record. Conclusions The use of a large number of vetted fossil calibration points with soft bounds is promoted as a better approach than using just one or a few calibrations, or relying on internal-congruency metrics to discard good fossil data. This study also highlights the importance of considering morphological and ecological characteristics of clades when delimiting higher taxa. I also illustrate how phylogeographic and paleoenvironmental hypotheses inferred from a tree containing only extant taxa can be problematic without consideration of the fossil record. Incorporating the fossil record of Ruminantia is a necessary step for future analyses aiming to reconstruct the evolutionary history of this clade.

The mitochondrial matrix and all maximum clade credibility trees resulting from all analyses are available online through TreeBase at the following link: http://purl.org/phylo/treebase/ phylows/study/TB2:S14132  Table 2. Percent difference between the molecular estimate and the median fossil age for the root and 16 calibrated nodes. These differences are then summed at far right for each tree, and below for each node. Not surprisingly, the tree using all 16 calibrations also provides estimates with the lowest difference from the expected median age. The nodes with the highest summed difference scores (i.e. nodes that exhibited the highest proportion of molecular age error) are the Alcelaphus buselaphus crown and the Kobus ellipsiprymnus + K. leche crown nodes.

Fossil Calibration Data
Calibrated nodes are defined and identified with respect to living taxa (crown clades) as molecular phylogenies normally consider only extant or recently-extinct taxa. The phylogenetic relationships of fossil bovid taxa have typically been proposed descriptively, or through phylogenetic analyses that are limited in taxonomic representation and/or number of character. A few notable examples of larger phylogenetic analyses that include fossil taxa are those of Gentry S2 [1], Geraads [2], Vrba [3], Vrba and Gatesy [4], and Bibi [5]. In selecting the fossil specimens or species representing the earliest known appearances of a clade, I have focused on the most reliable information available through the literature or on my own personal studies of fossil specimens, or both. Poorly documented single occurrences, especially any greatly stretching the first appearance datum of a taxon, were not relied upon. In cases where a first appearance datum is taxonomically sound but based on uncertain provenience (e.g. Taurotragus from the Chemeron Formation), I relied on the next-youngest occurrence in addition.
All clade calibration ages given here are minimum or approximate ages and probability distributions extrapolated from these come from my own assessment of the fossil record and the paleontological literature. It will be clear from the data that some calibrations are much better substantiated than others, and the details provided here should help others evaluate (and potentially improve) each calibration point themselves. Some may be used directly for dating, others for comparisons with clade origination ages derived from molecular clocks. All calibration data may be expected to change or be refined with further discovery and improvement of the fossil record and our understanding of it. This list is not exhaustive and many more candidate calibrations can still be added.
All references to Pleistocene refer to the recently redefined Pliocene-Pleistocene boundary at 2.588 Ma (previously at 1.8 Ma) [6]. The term crown clade refers to a node-based clade originating with the last common ancestor of two or more extant species or organisms (e.g. Bovidae) [7]. The stem group, or stem lineage, is the ancestral lineage leading to a crown group (e.g. stem Bovidae). The branch-based clade that comprises a crown clade plus the stem leading to it is the 'total' or 'pan-' clade (e.g. Pan-Bovidae) [7]. Recognition of extant species follows S3 common accounts [8][9][10][11], rather than classifications that promote the elevation of most subspecies to the level of species [e.g. 12].
Fossil calibration data are presented, following the recommendations of Parham et al [13], in the following format:

Clade Name
Definition: a phylogenetic definition of the clade [e.g. 7] Calibration point: the name and placement of the respective calibration point Age: Numerical age of the oldest known fossil Probability distribution type, 2.5-97.5% probability range in Ma (mean (x), standard deviation (sd or log(sd)), and median (M) used to create the required probability distribution 95% range in Beauti v.1.7.4 [14]) 1. Identification of species and, when feasible, the best-preserved and oldest specimens on which apomorphies relevant to the diagnosis of the node being dated are visible. In some cases, evidence for cladogenesis relies on contemporaneous occurrences of multiple taxa rather on any individual specimen or collection of specimens (e.g. Pan-Cervidae).
2. Identification of the apomorphies diagnosing the specimen or taxon as belonging within the clade in question. These may be in the form of references to in-depth discussions or treatments (phylogenetic analyses) in other studies.
3. Evidence for clade monophyly and identification of any major issues that may affect reconciliation of morphological and molecular phylogenetic analyses of the clade in question. S4 4. Identification of locality and stratigraphic information from which the relevant specimen, specimens, or taxa are known to have been collected. 5. Numerical age and dating information for the relevant specimen, specimens, or taxa, from which the clade age above is derived. 6. Justification for the calibration prior probability distribution parameters. A lognormal prior distribution with the upper 2.5% limit set to the fossil's age is used when the oldest fossil belongs to the clade in question. This means that there is a 97.5% chance that the node being calibrated will be older than the fossil. A normal distribution with the mean set to the fossil's age is used in cases where it is not certain whether the fossil belongs inside the clade in question or on its stem lineage. This means that there is a 50-50 chance that the node age will be either older or younger than the fossil. While fossil evidence helps provide a good minimum age for a divergence, maximal ages are much more difficult to assess. The main criterion considered here is the completeness of the fossil record of the lineage in question just prior to the origination of the node in question [13,15]. I conservatively choose wide intervals.

Pan-Cervidae
Definition: the clade consisting of all extant Cervidae and all species (living or extinct) that share a more recent common ancestor with Cervidae than with any other living ruminant. 3. Cervidae is by all accounts monophyletic though questions have in the past surrounded the relationship of Hydropotes to remaining cervids [17]. A revised taxonomy of cervids based on molecular phylogenies is given by Gilbert et al. [18] and Hassanin et al. [19]. 6. The fact that three stem cervids appear within MN 3 means that the line leading to Cervidae diverged from the remaining Pecora during or before this time. Given a general lack of understanding of the phylogenetic relationships of late Oligocene ruminants relative to the living clades, a lognormal distribution is chosen with its 95% range covering the maximal age range of MN 3 and extending back to the beginning of the Late Oligocene (28.4 Ma). 1. The oldest fossil moschid is Dremotherium feignouxi [16,17,21].

Pan-Moschidae
2. The affinities of Dermotherium with Moschidae appear to be based largely on the presence of a laterally enclosed subcentral tympanohyal on the auditory bulla [17]. This and other characters are described and discussed by Webb and Taylor [22] and Janis and Scott [17,23], with numerous indications that Dremotherium is a stem taxon.
3. Living moschids are represented by a single genus with five or more species [21]. The position of Moschidae in Pecora has a controversial history, with moschids treated as stem cervids, as the sister clade to Cervidae, or as sister to Bovidae, among other configurations [17,21]. An early molecular analysis found some support for moschids being the sister group of cervids [24], as did a supertree approach [25], but more recent molecular work has placed the clade as the sister taxon to Bovidae [19,26]

Pan-Bovini
Definition: the clade consisting of Bovini and all species (living or extinct) that share a more recent common ancestor with Bovini than with Pseudoryx nghetinhensis. In the current analysis recognizing extant taxa only, the node from which stem Bovini originates is equivalent to that defining Bovini + Pseudoryx. Therefore I chose a lognormal distribution with its 2.5% upper limit set to 10.2 Ma and a 97.5% lower limit spanning back to 16Ma, being shortly after the age of appearance of Eotragus.

Crown Bovini
Definition: the clade originating from the last common ancestor of Bos taurus and Syncerus caffer. 1. Selenoportax giganteus [5,45,54]. S12 2. Selenoportax giganteus is derived on S. vexillarius and S. falconeri in: larger size, strong molar ribs, more convoluted enamel, larger m3 hypoconulid with more anteroposterior orientation and with distal flange present; perhaps greater horn core divergence, greater hypsodonty, and the presence of a small horn core pedicel sinus [5,54]. While still relatively conserved in many features, S. giganteus is the oldest fossil bovid that exhibits a basic suite of characters synapomorphic for Bovini. It therefore probably lies close to the last common ancestor of Bovini, but may itself be either on the stem lineage or within the crown clade. 6. Whether these earliest tragelaphins were actually crown or stem taxa is not clear, but their morphology suggests they must be close, either way, to the common ancestor of the crown S14 clade. A few rare and isolated dental specimens from Mpesida and the Lower Nawata (both > 6.5 Ma) have been referred to Tragelaphini [56,62], but these are not conclusive. A normal distribution with a mean at 5.7 Ma and 95% range covering ± 1 Ma leaves opportunity for a latest Miocene to earliest Pliocene origin of the crown clade.
Pheraios chryssomalos was described as a stem tragelaphin from the late Miocene of Greece [57]. It lacks the synapomorphic characters listed above for Tragelaphini. A wider analysis with a larger number of taxa would help further investigate the intriguing hypothesis that the ancestor of Tragelaphini migrated to Africa from Eurasia during the late Miocene.

Pan-Tragelaphus euryceros
Definition: the clade consisting of Tragelaphus euryceros and all species that share a more recent common ancestor with it than with any other living bovid species. 2. Arambourg [64], Gentry [65], and Harris [66] had noted resemblances of the living bongo (Tragelaphus eurycerus) to Tragelaphus nakuae from the African Plio-Pleistocene. Bibi [63] established Tragelaphus rastafari to accommodate older specimens previously assigned to T. S15 nakuae, and discussed the cranial characters relating these two consecutive fossil species to the bongo. Tragelaphus saraitu is slightly older and may itself be ancestral to T. rastafari [67]. 3. No good assessment of the relationships among living tragelaphin species exists based on the fossil record. Mitochondrial and nuclear DNA phylogenies indicate the sister taxon to T. euryceros is the sitatunga, T. spekii [19,59]. There is no fossil record known for T. spekii.

Pan-Tragelaphus strepsiceros
Definition: the clade consisting of Tragelaphus strepsiceros and all species that share a more recent common ancestor with it than with any other living bovid species.

Calibration point: stem Tragelaphus strepsiceros
Age: 3.40 Ma minimum Lognormal, 3.4-4.5 Ma (x = 3.9 Ma, log(sd) = 0.07, M = 3.9 Ma) S16 1. Tragelaphus lockwoodi [69], from Hadar, Ethiopia. [69] argued that numerous apomorphies (e.g. the mediolateral compression of the horn cores along with the loose but high degree of torsion and developed anterior keel, combined with large size) makeT. lockwoodi a viable candidate for the ancestry of the living greater kudu (T. strepsiceros).  [72] analyses of cytochrome b, however, did not support this arrangement, presenting the possibility that the clade uniting the waterbuck and lechwes might have to include the kob and puku. S19 4. Though a single specimen of K. ellipsiprymnus is reported from lower Member G (unit G1), three specimens are recorded from upper Member G (above Unit G13) [65]. 5. Shungura Formation Member G Unit 13 is dated to between 2.04 and 2.00 Ma using magnetostratigraphy calibrated with radiometric ages [74]. 6. Since the fossil age here is a minimum for the actual node age, and since both extant species are also hypothesized to descend from the late Pliocene Kobus oricornus [70], a lognormal distribution is chosen with it's older bound extending back to 3 Ma.

Pan-Hippotragini
Definition: the clade consisting of Hippotragus equinus and all species (living or extinct) that share a more recent common ancestor with it than with Alcelaphus buselaphus. In the current analysis recognizing extant taxa only, the node from which Pan-Hippotragini originates is equivalent to that defining Hippotragini + Alcelaphini.  5. The anthracotheriid unit has been biochronologically correlated to between 7 and 6 Ma [76].
2. Apomorphies establishing the associations of these fossil specimens to Hippotragini and to Hippotragus and Oryx are presented by Gentry [81] and Vrba and Gatesy [4]. Though postdating the divergence of Hippotragus from Oryx, these fossil species probably predate the origins of crown Hippotragus and crown Oryx clades.
3. Monophyly of Hippotragini is supported on both morphological and molecular grounds. 6. Given the general rarity and poor resolution of the early Pliocene hippotragin fossil record, the actual age of origin of the crown group could be significantly older than that of the Upper Laetolil fossils. Two specimens from the late Miocene (<6.5 Ma) of Kenya were tentatively attributed to 'Praedamlis?' (i.e. an Oryx relative) by Harris [62]. A wide lognormal age distribution is chosen, with a 95% range covering 3.6 to 6.5 Ma.

Crown Alcelaphini
Definition: The clade originating with the last common ancestor of Alcelaphus buselaphus, Connochaetes gnou, Damaliscus pygargus, and Beatragus hunteri. 1. The oldest alcelaphin to be attributed to the crown group is Damalacra neanica from the earliest Pliocene of South Africa [83].
2. In describing the species, Gentry [83] justified affinity to Alcelaphini by reference to numerous cranial and dental characters. A morphological phylogenetic analysis by Vrba [3] established the position of this fossil taxon (along with D. acalla) within the crown clade and as sister to Beatragus spp. A slightly modified version of Vrba's analysis was run by Faith et al. [84], supporting the position of this species (but not that of D. acalla) as the oldest known crown alcelaphin.

The monophyly of Alcelaphini is firmly established by both morphological and molecular
analyses. Vrba's [3] analysis indicates that all known Pliocene and Pleistocene fossil alcelaphins belong within the crown clade. Faith et al.'s [84] re-analysis suggests instead that Pliocene and Pleistocene Damalacra acalla and Parmularius spp. are stem taxa. Harris [62] described specimens from the late Miocene that might also be stem alcelaphins. 2. For apomorphies of Alcelaphus buselaphus and the description of the BOD-VP-1/20 skull, see Vrba [3]. This skull is most similar to the living tora, swaynei, cokii subspecies and so most likely lies inside the crown clade or else very close to the last common ancestor [3]. 3. Alcelaphus buselaphus is a widespread and highly variable species. Classifications recognize several subspecies, and some taxonomists have advocated species and even genus level distinction for A. b. lichtensteini (as Sigmoceros lichtensteini), but molecular phylogenies support subspecific designation [86,87]. S24 4. BOD-VP-1/20 derives from Bodo locality 1, from sedimentary unit 'u-t' of the Middle Pleistocene deposits of the Middle Awash, Ethiopia [88]. 5. On the basis of sedimentological, structural, faunal, and archaeological evidence, unit 'u-t' at Bodo correlates to just above unit 'u' at nearby Dawaitoli and Hargufia, where a maximum age of 0.064 ± 0.03 Ma is indicated by 40Ar/39Ar dating of an underlying tuff [88]. 6. The Bodo skull represents the oldest known Alcelaphus buselaphus, and the oldest Alcelaphus at all (crown or stem) from a middle and early Pleistocene African record rich in alcelaphins.
The ancestry of Alcelaphus has been linked to Numidocapra (=Rabaticerus) arambourgi [3,89] which is only slightly older if at all [58], and Numidocapra crassicornis [3] which is about 1.5-1.0 Ma in age [58]. The genus is therefore unlikely to be older than 1 Ma. Gentry [90] described specimens that appear attributable to Alcelaphus buselaphus lichtensteini from the middle Pleistocene of Zaire, or at about 0.5-0.3 Ma [3], potentially supporting a ≥ 0.6 Ma age for the origin of the species.

Crown Connochaetes spp.
Definition: The clade originating with the last common ancestor of Connochaetes taurinus and Connochaetes gnou.  [58,89]. The extinct Connochaetes africanus is also known from Olduvai Bed II [58,89] and is also believed to belong within the crown clade [3].
2. See Harris [66,91] and Gentry and Gentry [89] for descriptions of early fossil Connochaetes taurinus. See Harris [66], Vrba [3], and Gentry [58] for description and analysis of characters diagnosing Connochaetes and proposing C. gentryi as a stem taxon and possible ancestor for both living wildebeest species.

3.
Connochaetes is by all accounts monophyletic [3,86]. 4. The oldest specimens of Connochaetes taurinus are known from Olduvai Bed II [58,89]. 5. Bed II at Olduvai ranges in age from 1.79 to 1.15 Ma [92,93]. 6. The early Pleistocene Connochaetes gentryi [66], a stem taxon and a likely ancestor for the crown clade [3], is first known from Olduvai Bed I, the Kaitio Member, and the Upper Burgi Member [58,66,91], with a maximum age of around 2 Ma [92]. (Vrba [3] also includes two partial horn cores from the Upper Lomekwi, ca. 2.5 Ma, described by Harris [91] as Connochaetes sp., in this species.) The maximum age of origin of crown Connochaetes is therefore not likely to be older than 2 Ma.

Pan-Caprini
Definition the clade consisting of Capra ibex and all species that share a more recent common ancestor with it than with Pantholops hodgsonii. 2. Alcala et al. [94] diagnose Aragoral mudejar as a member of Caprini (= Caprinae) on the basis of large size, moderate hypsodonty, reduced premolar rows, horn cores with relatively large bases, frontal sinuses reaching the horn core base, and a greatly shortened and wide metacarpal. The presence of simple frontal sinuses in the pedicel and horn core base is probably a synapomoprhy of the Caprini + Alcelaphini + Hippotragini clade, with further developments in lineages such as Capra and Ovis. A greatly shortened and wide metacarpal is a synapomorphy of Caprini [e.g. 38], and therefore helps place Aragoral on the line to Caprini, or possibly within the crown clade.
3. Caprini is by all account monophyletic, though its internal subclades have received numerous systematic shakeups [95][96][97][98]. Pantholops is an enigmatic taxon that has never adequately fit into any of the main bovid tribes. Recent molecular and morphological analyses have placed it as the sister taxon to Caprini, from which it differs in many important behavioral and morphological characteristics [38]. Among these are shortened metacarpals, a character present in Aragoral mudejar and Caprini, but not in Pantholops. 4. Aragoral mudejar is from the site of La Roma 2, Spain, which is biostratigraphically correlated to European Neogene Mammal Zone MN 10, and, on the basis of high resolution local biostratigraphy and magnetostratigraphy, to an age of 8.9 Ma [99]. S27 5. Belonging to either crown or stem Caprini, Aragoral mudejar provides a minimum age for the origin of the stem group. Given a lack of phylogenetic resolution surrounding much of the late Miocene record, the maximum limit is liberally extended back to allow for a much earlier origin of the stem Caprini. Aragoral mudejar could alternately have been used to provide a calibration for crown Caprini, using a normal probability distribution (as for crown Bovidae, crown Bovini, crown Tragelaphini above) with a range of 8.9 ± 2 Ma or thereabouts.