Phylogenetic tree of Litopterna and Perissodactyla indicates a complex early history of hoofed mammals

The Litopterna is an extinct clade of endemic South American ungulates that range from Paleocene up to late Pleistocene times. Because of their unique anatomy, litopterns are of uncertain phylogenetic affinities. However, some nineteenth century authors, considered litopterns as related to perissodactyl ungulates, a hypothesis recently sustained by molecular data. The aim of the present contribution is to include litopterns and other South American related taxa in a comprehensive phylogenetic analysis together with several extant and extinct basal perissodactyl ungulates. The analysis resulted in the nesting of litopterns and kin as successive stem-clades of crown Perissodactyla. Further, litopterns are not phylogenetically grouped with any North American basal ungulate, in agreement with some previous proposals. Presence of pan-perissodactyls in South America and India indicates that southern continents probably played an important role in the early evolution of hoofed mammals.

Dental similarities between South American condylarths and litopterns were previously noted by several authors, whom indicate that they may form a monophyletic clade 4,8,16,18,20 , for which the name Panameriungulata is available. Present results partially agree with such proposal, being congruent in that South American condylarths and litopterns constitute successive stem-taxa of Perissodactyla.
South American condylarths have been variously allied to the North American families Arctocyonidae, Hyopsodontidae, Phenacodontidae, Periptychidae, and Mioclaenidae 8,15,17,[19][20][21][22][23][24] . Cifelli 15 suggested that North American mioclaenines could serve as structural ancestors for the South American Didolodontidae, and numerous workers sustained a close relationship between North American Mioclaenidae and South American ungulates 18,[20][21][22][23][24][25] . However, it has recently been remarked that there is no support of close phylogenetic relationships between North American Mioclaenidae and South American condylarths and native ungulates. Even detailed morphological analysis did not find any derived character shared between Mioclaenidae and South American or African taxa 26 . www.nature.com/scientificreports/ In sum, present analysis indicates that South American condylarths are probably not closely allied to Northern Hemisphere taxa. As indicated above, South American forms share a number of derived features with perissodactyls that are absent in basal North American ungulate taxa.
The monophyly of Kollpaniidae resulted unresolved, with Pucanodus, Molinodus, Simoclaenus and Tiuclaenus, conforming a basal polytomy to remaining Pan-Perissodactyla. Because it is not the aim of the present analysis to resolve the internal relationships among kollpaniids, we do not discuss the monophyly of this grouping in length.
In addition to the above mentioned synapomorphies, some other key-traits shared by didolodontids, litopterns and perissodactyls include a fused mandibular symphysis, twinned lower molar metaconids, and a well-defined third lobe on the last lower molar, a combination of traits previously considered as unique to perissodactyls [32][33][34][35] (Fig. 2). Didolodontidae shares with basal perissodactyls as cambaytheriids and anthracobunids many plesiomorphic features including bunodont cheek-teeth with well-developed conules on upper molars, and the lack of any hint of lophodonty 35 . In fact, very prominent conules are usually considered to be diagnostic of didolodontids 16,27 , but are present also in cambaytheriids and anthracobunids [34][35][36] , sustaining close relationships between these clades.
In addition, litopterns share a large number of postcranial traits previously regarded as typical of Perissodactyla, including mesaxonic foot symmetry with reduced metapodials I and V, and hoof-like terminal phalanges, femur with large third trochanter and prominent greater trochanter, and very expanded greater trochanter on humerus (much more expanded than in basal condylarths as Phenacodus, Arctocyon or Tetraclaenodon 40,41 ) (Fig. 3), the distal humeral articulation is strikingly narrow and high, proximally delimited by a large foramen, and the radius is anteriorly located to the ulna. These features are correlated with an increased stride length and joints with reduced rotation 1,35,42 , a combination of characters typical of perissodactyls 35 .
In litopterns, as occurs in perissodactyls, the deltopectoral crest of humerus is not protrudent, and is restriced to the proximal half of the bone, whereas in phenacodontids and cambaytheriids the crest is distinct and plesiomorphically extends towards the distal end of the bone 35 . Further, the entepicondyles and the lateral supinator crest are reduced, contrasting with condylarths and basal perissodactyls as cambaytheriids 35,40,41 . In addition, the posterior astragalar facet of the calcaneum is angular and interlocks with the atragalus, whereas in cambaytheriids and condylarths this facet is rounded 35 .
One surprising result of present analysis was the nesting of the South American condylarth Escribania among Paleogene Indian Cambaytheriidae and Anthracobunidae. These taxa share some unambiguous synapomorphies, including absence of lower molar metaconid buttress (character 188, state 0), individualized protostyle on upper molars (character 212, state 1), and distinct entoconulid on lower molars (character 207, state 0) (Fig. 5). We here interpret the large and well-developed cusp in the lower molars of Escribania, and described as the "accesory cusp 2" by Gelfo et al. 10 as the entoconulid, because it is located anteromedially to the entoconid cusp.
Escribania shares with didolodontids, litopterns and perissodactyls several features (e.g., m3 with entoconid similar in size to hypoconulid, entoconid and hypoconulid separate, absence of entocristid, and presence of additional cusp mesial to entoconid). However, it differs from didolodontids in several dental traits: m3 with inflated metaconid that invades the talonid basin, relatively narrower talonid, and large trigonid with well-developed paraconid 10,24 . Further, Escribania shows a large parastyle as large as the mesostyle 10 . These features are clearly present in cambaytheriids, such as Cambaytherium 35

Discussion
Recently, on the basis of protein analysis, Welker et al. 12 suggested that notoungulates and litopterns may belong to Perissodactyla. Regarding notoungulates, many authors indicate that they are probably not phylogenetically close to litopterns 18 , and that notoungulates share features with afrotherians 44,45 . This last proposal resulted in a hot debate about notoungulate origins 46,47 . In this way, present discussion will focus on the biogeographic implications of perissodactyl affinities for litopterns.

Litoptern affinities and the splendid isolation of South America. Seminal studies by Florentino
Ameghino on fossil mammals from Patagonia resulted in a number of biogeographical relationships for the entire mammalian clade. This paleontologist 14 , proposed that most mammals originate in the Southern Cone and from there dispersed trough the entire world, a point of view known as "Extreme Australism" 48 . This was  www.nature.com/scientificreports/ refuted by Albert Gaudry 49 who considered that most characters linking Argentinean fossils with those of other landmasses are the result of convergences through a long time of isolated and parallel evolution, a "Splendid Isolation" as coined by Simpson 50 .
In spite that most authors (with exception of Muizon and Cifelli 18 ) were not able to find special similarities between North American and South American basal ungulates, it was clear to them that South American Condylarths undoubtedly arrived from North America 8,11 . Present work failed to find a clade encompassing South American and North American Condylarths, suggesting the possibility that South American litopterns may not be necessarily related to Northern Hemisphere taxa, in agreement with some previous authors 26 .
In this sense, the model of South America isolation may be too biotically simplistic 51 , as demonstrated by several studies which indicate that several animal and plant lineages reached South America from Africa by Late Cretaceous and Tertiary (e.g., legumes, lauraceans, and several others 52 ). On this basis, authors indicate that Africa and South America may have been united by Walvis Ridge-Río Grande Rise, and Sierra Leone-Ceará Rises during the Early Tertiary 53-55 . This is sustained by a large number of taxa shared between Africa and South America, but also with other landmasses and especially India, including hystricognath rodents, anthropoid monkeys, afrotherian mammals, pipid frogs, freshwater fishes (cichlids and aplocheiloids), birds (parrots, hoatzins, phororhacoids), and lizards (geckos), and Malpighiaceae, Asteraceae, and Bromeliaceae among plants 52,[54][55][56][57][58][59] . Further support for this interchange includes the finding of several lineages of metatherians, anthropoid monkeys and hystricognath rodents in South America, indicating multiple dispersals between South America and Africa and vice-versa during the Paleogene 60-64 . As enumerated above, the evidence indicating a fluid interchange between South America and other Southern Hemisphere landmasses and India has been greatly increasing during the last years (see below). This is in agreement with the seminal idea of Lavocat 65 whom suggested that the fossil record indicates closer biogeographical ties between South America and Africa than between North and South America.
As summarized above, strong biotic connection between South America and former Gondwanan landmasses appears to come to light. This point is crucial for understanding early biogeographical relationships of mammals, and more efforts are urgently need in order to analyze and criticize in detail different biogeographical scenarios.
India-South America biogeographical relationships. As indicated above, there are striking similarities between the Latest Cretaceous and Paleogene faunas and floras of former Gondwanan continents, including South America, Africa, and India. Bonaparte 66 noted that Mesozoic faunas from India were undoubtely Gondwanan in origin. In contrast, authors agree that the collision of India with Asia during the latest Cretaceous or Paleogene resulted in an important faunistic exchange 67,68 , and conclude that Paleogene faunas from India were entirely composed by Laurasian taxa 69,70 .
However, some recent workers sustained an important influence of Gondwanan biogeographical ties on India up to the early Tertiary. New findings suggest that by Eocene times Indian faunas were "mixed", having both European and Gondwanan lineages. Typically Gondwanan taxa include madtsoiid snakes, dyrosaurid crocodiles and pelomedusoid turtles 43 . More recently, adapisoriculid mammals with strong Gondwanan ties were reported for the first time in the early Eocene of India 71,72 .
Present analysis resulted in the shared presence of basal perissodactyls in both India and South America (Fig. 1). Further, the genus Escribania was included as the sister group of the Indian clade Cambaytheriidae + Anthracobunidae. In this way, perissodactyls constitute another clade that adds to the list of taxa shared by India and South America. It is possible that as soon as the fossil record of Paleogene faunas of India becomes improved, the list of taxa shared by both landmasses might increase.
Smith et al. 43 summarized two main hypotheses explaining occurrence of Gondwanan faunas on India. The first hypothesis proposed that these Gondwanan taxa may be the descendants of taxa already present by Cretaceous times that survived the K/T boundary. The second hypothesis sustain that a dispersal of Gondwanan taxa occurred from North Africa along the margins of the Neotethys to India. In this regard, an island arch (Oman-Kohistan-Dras) has been the route of migration proposed between Africa and India, during the Latest Cretaceous [71][72][73][74] . Because of the meagre fossil record, both hypotheses still lack important empiric support. However, because perissodactyls lack Cretaceous records, the shared presence of these taxa in both South America and India (and possibly Africa) may indicate Early Tertiary dispersal of Gondwanan taxa between India and North Africa. origin and early radiation of perissodactyla. The first works that deal with the origin of hoofed mammals indicate an Holarctic craddle for the Perissodactyla, particularly North American 29,75,76 or Asiatic origins 36 .
However, in the last decades many authors proposed that perissodactyls may have originated on India prior to its collision with Asia. Under this hypothesis the Indian plate may have acted as a "Noah´s Ark" during the Cretaceous and Paleocene 73 . Then, India carried Gondwanan forms to Asia after the break-up of the Gondwana super continent. This "Out of India" model was followed with modifications by some authors whom sustained that Indo-Pakistan area was most likely the center of origin for the Perissodactyls 35,38,74 . Further, Rose et al. 35 suggested that stem-Perissodactyla could have dispersed to India from Africa, by early Paleocene, and then, given rise to Perissodactyla before contact of India with Asia. In partial agreement with these contributions, present phylogenetic analysis indicates that pan-perissodactyls were widespread on southern continents, particularly in India and South America (and possibly in Africa) by early Tertiary times. This suggests that the southern continents may have played an important role in the early evolution and radiation of hoofed mammals.

Materials and methods
We follow the general concept and nomenclature of Litopterna and Didolodontidae of Simpson 16 with modifications by more recent authors 10,19,27 , and the Kollpaniinae of recent authorities 18,20,24 . In the later case, with the aim to emphasize the distinctiveness of the South American "mioclaenines" we opt to use Kollpaniidae rather than Kollpaniinae.
With the aim to analyze the phylogenetic relationships of Litopterna and kin we run a phylogenetic analysis following the comprehensive data matrix confected and employed by Rose et al. 35 . As in Rose et al. article, current study is limited to ungulates and does not address the possibility of a close relationship of litopterns to several disparate placental mammals (e.g., Glires, Primates, Carnivora). This data matrix was originally composed by 208 characters and 53 taxa. To this matrix we added 26 taxa (mostly "condylarths", litopterns, and didolodontids) and 6 characters that were employed by previous authors and that are key to dilucidate the affinities of litopterns and basal ungulatomorphs. This resulted in a matrix of 214 characters and 79 taxa ( Supporting Information 1-3). In addition, the character-taxon matrix is available on MorphoBank (Project 3768).
The phylogenetic analysis was performed using TNT 1.5 77 . All characters were equally weighted and treated as unordered. The data matrices were analysed under equally weighted parsimony. A total of 1,800,000 trees was set to be retained in memory, which is the maximum number of trees possible that could be saved on the computer used for these analyses. A first search using the algorithms Sectorial Searches, Ratchet (perturbation phase stopped after 20 substitutions), and Tree Fusing (5 rounds) was conducted, performing 1,000 replications in order to find all tree islands (each replication starts from a new Wagner tree). The best tree or trees obtained at the end of the replicates were subjected to a final round of TBR (tree-branch-swapping) algorithm.
The phylogenetic analysis resulted in the recovery of 8 Most Parsimonious Trees (MPTs), of 1518 steps, with a consistency index of 0.220, and a retention index of 0.625 which are summarized using a strict consensus ( Fig. 1; Supporting Information 4).
As a branch support measure, Bremer support was calculated, and as a measure of branch stability, a bootstrap resampling analysis 78 was conducted, performing 10,000 pseudoreplicates. Bremer support was calculated after searching for suboptimal trees and not with the script that accompanies the program 79 . Both absolute and GC 80 bootstrap frequencies are reported (Supporting Information 5).