Whence the beardogs? Reappraisal of the Middle to Late Eocene ‘Miacis’ from Texas, USA, and the origin of Amphicyonidae (Mammalia, Carnivora)

The Middle to Late Eocene sediments of Texas have yielded a wealth of fossil material that offers a rare window on a diverse and highly endemic mammalian fauna from that time in the southern part of North America. These faunal data are particularly significant because the narrative of mammalian evolution in the Paleogene of North America has traditionally been dominated by taxa that are known from higher latitudes, primarily in the Rocky Mountain and northern Great Plains regions. Here we report on the affinities of two peculiar carnivoraforms from the Chambers Tuff of Trans-Pecos, Texas, that were first described 30 years ago as Miacis cognitus and M. australis. Re-examination of previously described specimens and their inclusion in a cladistic analysis revealed the two taxa to be diminutive basal amphicyonids; as such, they are assigned to new genera Gustafsonia and Angelarctocyon, respectively. These two taxa fill in some of the morphological gaps between the earliest-known amphicyonid genus, Daphoenus, and other Middle-Eocene carnivoraforms, and lend additional support for a basal caniform position of the beardogs outside the Canoidea. The amphicyonid lineage had evidently given rise to at least five rather distinct forms by the end of the Middle Eocene. Their precise geographical origin remains uncertain, but it is plausible that southern North America served as an important stage for a very early phase of amphicyonid radiation.

ST, 0000-0001-6038-899X; ZJT, 0000-0001-5335-4230 The Middle to Late Eocene sediments of Texas have yielded a wealth of fossil material that offers a rare window on a diverse and highly endemic mammalian fauna from that time in the southern part of North America. These faunal data are particularly significant because the narrative of mammalian evolution in the Paleogene of North America has traditionally been dominated by taxa that are known from higher latitudes, primarily in the Rocky Mountain and northern Great Plains regions. Here we report on the affinities of two peculiar carnivoraforms from the Chambers Tuff of Trans-Pecos, Texas, that were first described 30 years ago as Miacis cognitus and M. australis. Re-examination of previously described specimens and their inclusion in a cladistic analysis revealed the two taxa to be diminutive basal amphicyonids; as such, they are assigned to new genera Gustafsonia and Angelarctocyon, respectively. These two taxa fill in some of the morphological gaps between the earliest-known amphicyonid genus, Daphoenus, and other Middle-Eocene carnivoraforms, and lend additional support for a basal caniform position of the beardogs outside the Canoidea. The amphicyonid lineage had evidently given rise to at least five rather distinct forms by the end of the Middle Eocene. Their precise geographical origin remains uncertain, but it is plausible that southern North America served as an important stage for a very early phase of amphicyonid radiation.

Introduction
The extinct family Amphicyonidae, commonly called the 'beardogs', represents a major group of caniform carnivorans with a rich evolutionary history of over 30 Myr extending from the Middle Eocene to the Late Miocene. During the course of their evolution, amphicyonids spread to all northern continents as well as Africa, and attained formidable ecomorphological diversity that included not only the iconic bear-like forms but also small fox-sized animals and wolf-like pursuit predators [1][2][3][4][5][6]. While amphicyonid systematics has a long history of study, the origin and initial evolution of the group during the Eocene have long remained nebulous. Examination of the holotype of Miacis australis Gustafson, 1986, from the Middle-Eocene portion of the Chambers Tuff of Texas, USA (see [7] for geologic context), prompted the first author to reassess its affinity with other early carnivoraforms, including Miacis cognitus Gustafson, 1986, from a higher stratigraphic level in the same formation. The phylogenetic affinities of these two taxa turn out to be eminently relevant to the question of amphicyonid origin as they exhibit combinations of primitive carnivoraform characters-which led to their original assignment to the genus Miacis [8]-with some derived amphicyonid conditions. Since 1986, several cladistic analyses of early carnivoraforms have recovered M. cognitus as a close relative of the earliestknown amphicyonid, Daphoenus (e.g. [9,10]), but the systematic position of the former has not been revised or discussed in detail. Based on a new cladistic analysis that included a larger sample of early amphicyonids than in previous studies, we propose here reassignment of M. australis and M. cognitus to new genera within the Amphicyonidae. We discuss the results of our analyses and their implications on the origin of beardogs-'animals of extraordinary structural characteristics and of remarkable interest' in the words of Scott & Jepsen ( [11], p. 76)-with the emphasis on the group's Middle-Eocene fossil record.     a Taxonomic composition of the amphicyonid clade (as recovered in the consensus tree from the unconstrained search; figure 1)-but not its internal topology-was held constant across all constrained searches.
Distribution-As for type species. Remarks-Determination of the subfamilial affiliation of Gustafsonia and, indeed, testing for the reality of currently recognized amphicyonid subfamilies [2] would require broader taxonomic sampling than was attempted here. Essential to such effort would be additional work on the taxonomy of species currently assigned to the genus Cynodictis, most of which are based on mandibular dental morphology alone [33] and have not been cladistically analysed.
Distribution-Known only from the holotype locality; upper portion of the Chambers Tuff, Presidio County, Texas, corresponding to the early Chadronian NALMA [52]. Part of the Little Egypt l.f. [7], which has been assigned to the upper part of Chron 17n [53] [52].
Remarks-The specific name becomes cognita (Latin feminine participle) when combined with the new genus name, Gustafsonia, to match the gender of the latter. Gustafsonia cognita is known only from a single cranium. Gustafson [8] figured and described the holotype in detail, and assigned the new species to the genus Miacis, emphasizing its generally primitive character (compared with fossil representatives of extant families) and overall similarity to M. parvivorus. While its resemblance in dental morphology to the early amphicyonid Daphoenus was noted in the original description, an amphicyonid affinity of 'M.' cognitus was rejected on the basis of perceived absence of deep basioccipital embayment. Wang  Tedford [9] reported additional observations on its basicranial anatomy, and noted more derived features of 'M.' cognitus compared with M. parvivorus, such as more extensive bony coverage of the facial nerve canal and likely medial passage of the internal carotid artery (inferred from the absence of a groove on the petrosal promontorium that would be left by a promontory artery). In a departure from Gustafson's description [8], Wang & Tedford [9] considered the degree of basioccipital embayment in the holotype of 'M.' cognitus indeterminate. Nevertheless, a cladistic analysis by the latter authors suggested a close relationship of 'M.' cognitus to Daphoenus. This hypothesis was further supported by subsequent studies that analysed additional taxa and morphological characters (e.g. [10,14,46]). In its original description, G. cognita was described as possessing a 'narrow' ( [8], p. 42) inferior petrosal sinus, presumably because the approximately 3 mm long contact between the lateral edge of basioccipital ventral floor and the petrosal promontorium leaves little externally visible gap between them. However, high-resolution X-ray computed tomographic (HRXCT) images of TMM 40209-200 [54] . reveal substantial embayment of basioccipital above its ventral floor along the line of contact with the promontorium and precisely where it cannot be observed by external inspection (figure 2c-f ). Tracing the embayment from its posterior end forward, it is rather constricted in size immediately anterior to the posterior lacerate foramen but rapidly expands thence to attain a width of 2.0 mm and a height of 1.5 mm at the anteroposterior level of mastoid tubercle; these dimensions are largely maintained anteriorly along the length of the rest of promontorium (figure 2d). Thus, at its widest point, the embayment corresponds to as much as approximately 40% of the width of the ventral floor of the basioccipital. This degree of embayment is much greater than seen in the early stem canids Prohesperocyon wilsoni and H. gregarius, as well as a specimen referred to the possible basal ursidan Am. shackelfordi [9,24]. Further, the basicranial structure surrounding the inferior petrosal sinus differs fundamentally from those in more primitive carnivoraforms such as Miacis parvivorus, in which the lateral edge of basioccipital floor does not bifurcate into ventral and dorsal lips (contrast [9], fig. 2 with figure 2f ).
It should be noted that rigorous quantitative scoring of this character (Character 31) as State 1 (deep excavation) or State 2 ('extremely deep' excavation as defined by Wesley-Hunt & Flynn [10]) is not possible at present as it would require comparative CT data from additional taxa and allometric consideration. Nevertheless, the result of cladistic analysis reported here is robust to this uncertainty because combining States 1 and 2 into a single state representing deep basioccipital excavation did not alter the topology of consensus tree (electronic supplementary material, S2). The presence of looped internal carotid artery within the inferior petrosal sinus (a feature unique to ursids among extant carnivorans and thought to serve as part of a countercurrent heat exchange system for the blood entering the brain; [55]) cannot be inferred with certainty for G. cognita based solely on the deep basioccipital embayment. It is, however, intriguing that the enlargement of inferior petrosal sinus apparently started in the earliest, very small amphicyonids, long before they gave rise to taxa with bear-like proportions.
Gustafsonia cognita represents one of the smallest amphicyonids known to date, with a condylobasal length (79.5 mm) approximately 10% shorter than that of the Early-Oligocene Pa. minimus from the central Great Plains, which was previously recognized as the smallest North American beardog [3]. A body mass of 2.3 kg (±1 s.e. = [1.4, 3.8]) is estimated from the cranial length using the allometric equation of Van Valkenburgh [56] for mammalian carnivores. Recognition of the genet-sized G. cognitawith dentition that lacks obvious signs of dietary specialization-as one of the earliest beardogs accentuates the picture of rapid ecological, if not taxic, diversification of amphicyonids during their earliest-known phase of evolution: the clade gave rise to coyote-sized Daphoenus and apparently hypercarnivorous Daphoenictis by the Late Eocene (cf. [47]).
Distribution-As for type species. Remarks-As is the case with Gustafsonia, additional systematic work is needed to determine the subfamilial affiliation of Angelarctocyon.    Emended Diagnosis (modified from Gustafson ([8], p. 44))-Differs from: (i) non-amphicyonid carnivoraforms in M1 lingual cingulum lingually elongate with a sharp lingual margin (Character AC1, State 1); (ii) other amphicyonids except G. cognita in more gracile dentition; (iii) G. cognita in more labial (rather than anterior) direction of M1 parastylar region, more robust paracone and metacone of M1, shallow and expansive lingual slope of M1 protocone; and M2 with labiolingually narrower trigon basin (apex of protocone is located approximately halfway between the level of paracone-metacone apices and lingual margin of tooth; figure 3c). Further differs from 'daphoenine' amphicyonids in greater anteroposterior constriction of lingual portion of M1, more labial direction of M1 parastylar region and larger M2 (length across paracone and metacone is comparable with that on M1). Further differs from: (i) Miacis parvivorus in shorter M1 stylar shelf labial to base of metacone (Character AC2, State 1), less pronounced height difference between M1 paracone and metacone, greater swelling of M1 lingual cingulum (Character 50, State 2), more elongate p4 main cuspid across its base (including the portion that bears the posterior accessory cuspulid, this length is approximately 120% of the height of the main cuspid (measured from its apex to the junction of anterior and posterior roots) compared with rsos.royalsocietypublishing.org R. Soc. open sci.   approximately 73% in USNM 214706), more dorsal position of p4 posterior accessory cuspulid, more open m1 trigonid, more robust m1 entoconid and more elongate m2 (m2 L/W > 1.51 compared with 1.36 for AMNH FM 5019); (ii) basal arctoids (possibly ursidans) Amphicynodon, Pachycynodon and early stem ursids such as cephalogalines (cf. [21,27,28]) in more acute and anteroposteriorly constricted P4 protocone, relatively taller M1 paracone and metacone, greater anteroposterior constriction of M1 lingual portion, labiolingually wider M1 stylar shelf (not reduced to a mere rim), more triangular outline of M2, more closed m1 trigonid and less reduced m2 paraconid; (iii) early stem ursids in greater lingual elongation and lingual orientation of M1-2 lingual cingula (figure 5d).
Distribution-Lower portion of the Chambers Tuff, Presidio County, Texas, corresponding to the late Duchesnean NALMA [52]. Part of the Porvenir l.f. [7], which has been assigned to the lower part of Chron 17n [53] [52].
Description-The following observations on the holotype FMNH PM 423 supplement those of Gustafson [8].
The maxillae are embedded in matrix of hard, tuffaceous sandstone (figure 3a). Most of the preserved teeth are extensively damaged. In addition, the dentaries and lower teeth are heavily coated with consolidant, obscuring some of the morphological details. The infraorbital foramen is located directly above the posterior alveolus for P3 (Character 4, State 1; figure 3b). Its precise shape is unclear because of the matrix infill, but it appears oblong and taller than wide (Character 3, State 0).
Of the anterior upper dentition, only the root of left canine is preserved. On the left maxilla, diastemata are present between the canine and the single-rooted P1 (approx. 2.2 mm), between P1 and P2 (approx. 3.4 mm) and between P2 and P3 (approx. 1.6 mm), and the spacing of these teeth appears to have been comparable with that in G. cognita as well as 'Miacis' cf. 'M.' sylvestris (FMNH PM 55953 = cast of AMNH FM 129284). The alveoli for P3 suggest posterior widening and triangular outline of the tooth in occlusal view.  The right P4 (inverted in figure 3e) is missing its posterolingual base, posterior end of metastylar blade and a small piece of the parastylar area on the anterolabial cingulum; the paracone is broken into pieces and its apical portion is slightly dislodged. The left P4 is heavily damaged except for the labial portion of paracone and the metastylar blade (figure 3c,d). The parastyle, which is not preserved, is likely to have been absent or very small judging from the surrounding areas. The cingulum is clearly defined along the lingual wall of tooth and anterior to the protocone, but is poorly developed on the labial side of tooth. The preparacrista on the right P4 is a weak ridge that becomes very faint as it approaches the parastylar area. Posterior to the paracone, a deep carnassial notch is followed by a moderately long metastylar blade. The protocone is roughly conical in shape and has a little over a third of the height of paracone.
The right M1 (inverted in figure 3e) is extensively damaged except for its anterolingual portion bearing a clearly delineated anterior cingulum (contra absence of such a cingulum reported in the original description ( [8], p. 46)) and the lingual portion of trigon basin formed by the labial base of protocone. The left M1 (figure 3c,d) likewise shows some major breakages, but its basic morphology is discernible. A small piece of the posterolabial wall ('rlw' in figure 3c) was apparently broken off, rotated lingually and lies on the labial side of metacone; the posterolabial outline of the tooth is thus obscured, but labial projection of that area appears to have been minimal as in G. cognita or absent as in Cynodictis. The lingual base of paracone and the middle part of the lingual slope of metacone are both missing, but a small portion of the very base of the lingual slope of metacone (immediately labial to the metaconule) is preserved. Also missing are most of the labial slope of protocone, the surface of lingual cingulum immediately lingual to the protocone and the cingulum anterior to the paracone. The preserved portion of protocone is slightly dislodged from the rest of tooth.
A notable feature of the M1 is its prominent lingual cingulum: it extends from the anterior base of the protocone to at least the base of metaconule, from where it seems to merge with the border of tooth (however, it may be partially covered by the matrix). The lingual cingulum ('lc' in figure 3c) is elongate in lingual direction, and connects smoothly (i.e. without a sharp border) with the lingual to posterolingual base of protocone, forming a concave slope (figure 3d). The very apex of protocone is missing, but appears to have been located at about the same anteroposterior level as that of the paracone. The preserved portion of protocone reaches approximately the same height as the paracone. The stylar shelf is very limited in labiolingual width (though not as much as in stem ursids) and is abutted by the steep labial slopes of the paracone and metacone. The low and not particularly sharp parastylar ridge ('psr' in figure 3c) runs from near the labial base of preparacrista to the labial border of tooth in nearly labiolingual direction. The centrocrista is rather sharp, while the postmetacrista is weakly defined. The paracone (taking into account its missing tip) and the metacone have subequal heights and comparable anteroposterior lengths; both of these cusps are more robust than in G. cognita. The paraconule, while completely missing from the left M1, appears to have been well defined, judging from its still fairly prominent remnant on the right M1 ('pcl' in figure 3e). The metaconule ('mcl' in figure 3c) is low in height but clearly recognizable, having a somewhat angular appearance. The posterolabial edge of the metaconule gives rise to a narrow cingulum-like band immediately posterior to the metacone. The trigon basin as seen on the right M1 is moderately deep.
The right M2 (figure 3e; inverted) is mostly broken except for the trigon basin and the apices of metacone and protocone; the paracone and the lingual cingulum are largely broken off. The left M2 (figure 3c,d) is missing the labial margin, posterolingual portion of the paracone, and most of the metacone; the lingual portion of the trigon basin is intact, including the paraconule and much of the protocone. The lingual cingulum together with the very shallow lingual slope of the protocone makes the lingual portion of M2 notably elongate. As in the M1, no sharp boundary exists between the lingual base of protocone and the surrounding cingulum, at least in preserved portions; this condition is frequently seen in amphicyonids but not in other caniforms [57]. The protocone is low in height and bears an essentially flat, shallow (and consequently rather long) labial slope that forms a large part of the gently concave trigon basin. Although unclear from the illustration in Gustafson ([8], fig. 29a), the apex of paracone is preserved ('apa' in figure 3c), revealing the paracone to be low and not much taller than the M1 metaconule. The preparacrista is not particularly sharp. The paraconule ('pcl' in figure 3c) is small with a somewhat angular appearance; it is separated from the protocone by a small notch. An anterior cingulum is present anterior to paracone. The metacone on the right M2 (figure 3e) has roughly the same height as the protocone. A very small metaconule (originally described as 'indistinct' ( [8], p. 45)) is present on the right M2 ('mcl' in figure 3e); it is recognizable as a slight bulge on the posterior border of the trigon basin lingual to the base of metacone, from which it is separated by a superficial wrinkle.
No M3 is preserved in the holotype, and no M3 alveolus is visible such that its presence was originally considered indeterminate [8]. However, the presence of M3 is almost certain based on the minimum posterior extent of the right maxilla behind (and lingual to) the M2 (white arrow in figure 3e).
The dentary (figure 4a) is slender (depth below right m1 = 12.0 mm; measured on the lingual side below the paraconid-metaconid junction) with a relatively straight ventral border. The anterior and posterior mental foramina are located below p1 and below the anterior border of p3, respectively. The left c1 measures 4.2 mm in length and 2.8 mm in width at the base of crown; much of the crown above that level is broken off.
The p3 (figure 4a,c-e) is missing its anterolabial portion, and the posterolingual portion is broken into at least two pieces that are held together by adhesive. The broken piece on the lingual side ('pmc' in figure 4e) appears to be somewhat dislodged, presumably giving it a false appearance of a 'very small metaconid on the lingual side of the posterior slope of the protoconid' reported by Gustafson ([8], p. 45). The lingual position of this apparent 'metaconid' is, in fact, inconsistent with where an accessory cuspulid is normally located on a carnivoraform p3 (and where it is located on the p4 of FMNH PM 423), which is on a ridge that runs on the labial side of the posterior slope of the main cuspid; thus, there is no plausible sign of a posterior accessory cuspulid or a 'metaconid' on the p3 of A. australis. The anterior and posterior cingulids do not appear to form any appreciable cuspulids, either, similar to early daphoenines and arctoids but in contrast with other early carnivoraforms such as Lycophocyon hutchisoni and H. gregarius.
The p4 (figure 4) is essentially complete except for chipped pieces of enamel on its lingual base. Mirroring the p3, the p4 lacks anterior and posterior cingular cuspulids, again, similar to early daphoenines and arctoids but in contrast with other early carnivoraforms such as L. hutchisoni and Hesperocyon gregarius (figure 6). The anterior and posterior cingulids themselves are for the most part only weakly demarcated and do not form a shelf or basin of appreciable size; instead, there is at the anterior end of tooth a short, posterolabially inclined slope whose anterolingual margin forms a very short ridge (figure 4c). Unlike in Hesperocyon, the central cuspid is positioned more centrally along the length of tooth, has a very weakly sigmoidal anterior border (in contrast with the simply bowing border in H. gregarius) and is anteroposteriorly more symmetrical in profile. Further, the notch between the main cuspid and the posterior accessory cuspulid is very shallow, and the posterior accessory cuspulid ('pac' in figure 4d; 'hypoconid' of [8]) is short, in contrast with the elongate, blade-like one in H. gregarius (figure 6). The step-like appearance in profile of the posterior cuspulid is characteristic of many amphicyonids (although exceptions are seen in e.g. D. tedfordi) and early stem ursids such as Amphicynodon and some species of Parictis; the p3 and p4 of A. australis, however, differ from those of typical 'amphicynodonts' in being narrower and lacking thick, clearly defined labial and lingual cingulids. The p4 is narrower and more gracile in appearance than that of H. gregarius. Hesperocyon gregarius Angelarctocyon australis Daphoenus hartshornianus The right m2 (figure 4) is complete except for the broken posterolingual corner. The paraconid is more distinct than in Hesperocyon. The size of m2 relative to m1 is considerably larger than in H. gregarius (m2LxW/m1LxW = 0.58 compared with 0.41 for H. gregarius calculated from species mean dimensions reported in Wang [41]). The trigonid cuspids are separated by three small notches. As in the m1, the cristid obliqua and the hypoconid are labially positioned, although they are very low in height. The labial portion of talonid is rather horizontal and more so than in L. hutchisoni. The entoconid and its surrounding areas are broken off. Separate and weakly defined cingulids are present anterolabial and posterolabial to the protoconid. The posterior slope of metaconid that descends onto the talonid is shallow.
The presence of a single-rooted m3 is evident from its alveolus on the left dentary (figure 4b). It seems probable that the tooth was less than half the size of m2.
Remarks-The original diagnosis for 'Miacis' australis was confined to comparisons with other species of Miacis and Hesperocyon ( [8], p. 44). Re-examination of the holotype and its inclusion in cladistic analysis for the first time support an amphicyonid affinity of this taxon. We consider the precise phylogenetic relationship between A. australis and G. cognita to be uncertain: their relative positions within the Amphicyonidae as shown in the strict-consensus tree (figure 1) should be viewed with caution because they are separated on the strict-consensus tree only by subtle differences in the position and the shape of infraorbital foramen-characters that are prone to subjective interpretation as well as taphonomic distortion-and because only the P4-M2 can be directly compared between the two taxa, which are known from a total of three specimens. The preserved upper teeth of the two species are nevertheless quite distinct, and we do not see a compelling reason to assign them to the same genus. The two taxa are known from the same geologic formation in the same general area of Trans-Pecos, Texas, but occur at different stratigraphic levels and have been assigned to separate local faunas that, in turn, are biostratigraphically distinguished as late Duchesnean (the Porvenir l.f. with A. australis) and early Chadronian (the Little Egypt l.f. with G. cognita) in age.

Discussion
Originally discussed as 'relicts of the considerable radiation of miacine miacids' ( [8], p. 46), A. australis and G. cognita are here formally recognized as basal members of the Amphicyonidae that, in fact, provide important insights into the little-known earliest evolutionary phase of a major carnivoran clade. This reinterpretation follows the progress in understanding of early carnivoraform basicranial anatomy and phylogeny that has been achieved over the last 30 years since the time of their first description (e.g. [3,9,10,24,36,[44][45][46]).
With the exception of Cynodictis, lingual rather than posterolingual elongation of the M1 is one of the most diagnostic synapomorphies of early amphicyonids that differentiate them from other early caniforms, including stem canids and stem ursids. This feature typically manifests as some combination of well-developed lingual cingulum (directly lingual to the apex of protocone) and extension of the lingual slope of the protocone, which frequently produces a smooth transition from the protocone to the lingual cingulum [57]. In Cynodictis, the orientation of elongate M1 lingual cingulum is decidedly posterolingual as in 'Miacis' gracilis (although the latter has a considerably shorter cingulum), canids and several early arctoids including stem ursids; this condition may, therefore, be ancestral to amphicyonids given their phylogenetic position and internal topology. An isolated caniform upper molar (SMNH P661.1492) from the Chadronian Calf Creek l.f. of the Cypress Hills Formation essentially shows this condition but was questionably reported as an M1 of the 'amphicynodont' Subparictis parvus ( [58], fig. 3.3; 'Parictis' parvus in that publication) and later described as being 'much more primitive than those of other species of Subparictis' ( [22], p. 490). In fact, no unquestionable upper tooth of Su. parvus is known, and SMNH P661.1492 more closely resembles the M2 s of A. australis (FMNH PM 423) and Daphoenus cf. D. lambei (OMNH 2953) than the M1 s of other species of Subparictis, in which the extension of M1 lingual cingulum is less pronounced and is more posteriorly directed (cf. [20]).
The removal of G. cognita from the genus Miacis leaves the latter with no specifically determinate or generically secure occurrence in the Late Eocene. Here, it is worth noting that Miacis represents perhaps one of the most confused genera in the classification of early carnivoraforms, with at least 19 species having been assigned to the genus (cf. [59][60][61]). None of them other than the type species M. parvivorus from the Bridgerian of North America has received cladistic support for its generic assignment [10,36,46]; as such, the genus as currently recognized defies apomorphic diagnosis. We suspect that the historical tendency of taxonomic workers to place primitive carnivoraforms in Miacis has obscured more than it clarified the patterns of their diversification, and await a comprehensive revision of the genus.

Phylogenetic position of the Amphicyonidae
In assessing the affinity of what is still the earliest-known amphicyonid (then considered a canid) genus, Daphoenus, Scott and Jepsen remarked in their 1936 paper: 'This very difficult problem is full of paradoxes and seeming contradictions' ( [11], p. 74). At the heart of the problem was the amalgam of plesiomorphic and homoplasic traits seen in the White River carnivorans, which the authors were well aware of but lacked an analytical framework to dissect. Homoplasy in skeletal morphology is well known among carnivorans (cf. [62][63][64]), and can become increasingly problematic in phylogenetic inference as the taxic and spatio-temporal scope of analysis expands. Comparison of early members of major clades is thus expected to be more informative for phylogenetic reconstruction than that of more derived members, which tend to be better represented in the fossil record but are separated by longer branches; the latter approach (e.g. [65,66]) may have contributed to the long history of systematic instability surrounding the beardogs, much of it predating widespread use of the cladistic method [2,[67][68][69][70][71] (see [5,69] for historical reviews).
To date, computer-based cladistic analyses of early carnivoran phylogeny have included only one to three OTUs representing amphicyonids (e.g. [10,14,29,46]). The results of previous studies variously suggested amphicyonids to be the sister group of the Ursidae [29] or the Canidae [46], or a basal clade outside the Canoidea [10,14]. Though still limited in taxic coverage, this study expands the previous efforts by substantially increasing the sampling of amphicyonids from the earliest phase of their radiation, which is key to deciphering their relationship to other carnivoran clades, and by incorporating hitherto-neglected morphological characters to help resolve their phylogenetic positions.
The identification of amphicyonids as an early diverging group of caniforms outside the Canoidea is in agreement with, and provides additional support for, previous findings by Wesley-Hunt & Flynn [10] and Tomiya [14], and was already considered a possibility 20 years ago [47]. proportionately small number of traits that represent synapomorphies and already a high degree of homoplasy. Nevertheless, alternative placement of amphicyonids as the sister group to canids, arctoids, or any subgroup of arctoids is clearly less acceptable based on the parsimony criterion (table 1).
The hypothesis of amphicyonid-ursid relationship based on the shared presences of deep basioccipital embayment and postscapular fossa [2,29,72,73] merits additional discussion. The main difficulty with this hypothesis, which is not supported by the present parsimony analysis, is that the earliest-known arctoids including some stem ursids are considerably more derived than the earliest amphicyonids. For example, Amphicynodon, Cephalogale (a constituent of the cephalogaline OTU in figure 1) and their close relatives from the Oligocene of Eurasia almost always lack the M3 and, where known, possess a more flat petrosal promontorium as well as horizontally expanded auditory (primarily ectotympanic) bullae that differ from the simple crescentic ones in the earliest amphicyonids (cf. [21,24,27,28]). Similarly non-crescentic and extensively ossified bullae are suspectedalbeit from poorly preserved material-for an even older amphicynodont, Campylocynodon personi from the Late Eocene (Chadronian) of North America [20]. The appearance of sister-taxon relationship may be reinforced when comparing more derived members of the two families, but many of their most conspicuous evolutionary trends (e.g. development of the flask-shaped Type A auditory bulla of Hunt [55] as opposed to its precursors seen in e.g. Daphoenus (cf. [73], p. 830, [47], p. 502), enlargement of posterior molars and pronounced body size increase) are known to represent parallelism or convergence [2,47,68,70,73].
An exclusive amphicyonid-ursid relationship, therefore, cannot be supported unless (i) more primitive stem ursids are discovered that reveal additional synapomorphies with the earliest amphicyonids or (ii) the two potential synapomorphies (basioccipital embayment and postscapular fossa) are more heavily weighted in cladistic analysis. However, for carnivoramorphans, the traditional expectation that basicranial anatomy is less prone to homoplasy through ecological adaptations than other craniodental traits (e.g. [69]) is not supported by analysis of empirical data [10]. Indeed, while the degree of basioccipital embayment that characterizes amphicyonids and ursids is not seen in other carnivorans, it should be noted that enlargement of the inferior petrosal sinus also occurred-presumably independently-in mustelidans [10,43]. A quantitative survey of the basioccipital embayment in early carnivorans and greater knowledge of their postcranial and brain morphology (cf. [46,74]) would enable additional testing of the various phylogenetic hypotheses.

Time and place of amphicyonid origin
Before their geographical range extension into Africa and Panama in the Early Miocene, amphicyonids were confined to the Holarctic (cf. [1,75,76]). Their first occurrences in different regions of the Holarctic are reviewed here.
Simamphicyon helveticus, known from the late-Middle Eocene of Europe (first occurrence in the mammal reference level MP 16, ca 40-37 Ma; [71]; see [77,78] for timescale) has generally been treated as an amphicyonid since the time of its original description ( [79]; see also [80] and [67], p. 81), and was regarded as the oldest amphicyonid by Springhorn [71]. However, several authors have expressed doubts as to its amphicyonid affinity [71,[81][82][83][84] [89] for the ages of assemblages), but we have not been able to verify either identification in the absence of unique identifiers for the pertinent specimens. A revision of the genus Cynodictis [33], which restricted its constituents to European species, and frequent misassignment of non-amphicyonid carnivorans to Cynodictis (rectified by e.g. [45,57,90]) cast further doubt on the generic identification as well as the amphicyonid affinity of the purported Middle-Eocene 'Cynodictis'. The only other possible occurrences of beardogs in the Middle Eocene of Asia consist of fragmentary teeth (a canine and a possible m1 talonid) from the late-Middle Eocene of Myanmar, and are questionably referred to the family [91]. The earliest definitive amphicyonid occurrences in Asia, then, are from the Late Eocene of China and Mongolia [57,92]. Despite the limited material available for comparison (Guangxicyon sinoamericanus is represented by a dentary and two limb elements [92], and the unnamed taxon from Mongolia by an incomplete M2 [57]), these taxa appear to be considerably more derived than the older Angelarctocyon and Gustafsonia from North America: for example, Gu. sinoamericanus is approximately 270% the size of A. australis (based on m1 length) and possesses low-crowned lower cheek teeth unlike those of A. australis but similar to some European amphicyonine amphicyonids (cf. [92]).
Thus, at present, the only verifiable Middle-Eocene occurrences of the family are from the Duchesnean NALMA (roughly comparable with the Bartonian age) of North America. Chronologically, they can be summarized as follows (see [47,52,93] [47,97]. Other authors, who have proposed a latest-Duchesnean age for the Little Egypt l.f. from the Chambers Tuff [98], would add to this list G. cognita and B. dodgei (cf. [8]). Curiously, amphicyonids are absent from the latest-Uintan to early-Duchesnean assemblages west of the Rocky Mountain region (e.g. Hancock Quarry l.f. (Clarno Formation) of Oregon, Laguna Riviera and Pearson Ranch l.f. (Santiago and Sespe Formations) of southern California), which include some of the oldest North American occurrences of immigrant carnivore lineages from Eurasia (represented by hyaenodontids Hemipsalodon and Hyaenodon and a nimravid [99][100][101]), and of an early non-canoid carnivoraform, Lycophocyon [14,102].
The topology of consensus tree, the morphological diversity of Duchesnean taxa and the temporal distribution of fossil occurrences together make it clear that the beardogs had not only originated but also given rise to at least five distinct genera or their precursors (Cynodictis, Gustafsonia, Angelarctocyon, Daphoenictis and Daphoenus) by the end of Middle Eocene, approximately 37 Ma. The geographical origin of the family, on the other hand, is difficult to infer from the available data. Arguments for and against three alternative scenarios can be summarized as follows: (1) Europe. The European species C. lacustris is more plesiomorphic than the earliest North American amphicyonids in some respects (e.g. greater labial extent of M1 parastylar region, small M2/m2, retention of vestigial anterior cingular cuspulids on lower premolars). However, the complete absence of beardogs from the fairly substantial Middle-Eocene faunal records for Europe (cf. [103,104]) makes a European origin less likely. Detailed investigation of the phylogenetic relationships among various species of Cynodictis and Eurasian amphicyonines, which has not been attempted since the work of Teilhard de Chardin [51], may help clarify the geographical context of early amphicyonid evolution. (2) Asia. Few carnivoraforms were reported-and even fewer were confidently identified-from the Middle Eocene of Asia in the faunal compendium of Russell & Zhai [88], and the poor state of knowledge has not fundamentally changed since then despite some important findings [105]. As a result, there is little empirical evidence for an Asian origin of amphicyonids. Still, recent discoveries of Late-Eocene taxa in China and Mongolia that appear to be more derived toward later amphicyonines than the roughly contemporaneous Cynodictis in Europe suggest eastern Eurasia as a potentially important stage for the early diversification of amphicyonids. Eurasia. A better fossil record compared with Asia and perhaps also Europe may explain the first observation, but the second point is puzzling under the hypothesis of Eurasian origin. However, the cladistic position of European Cynodictis in the consensus tree makes a North American origin less than certain.
Noteworthy from a broader biogeographic perspective are the endemic occurrences of some of the earliest (evolutionarily, if not always chronologically) members of emerging mammalian groups in the late Duchesnean to early Chadronian of Trans-Pecos, Texas. They include the most primitive stem canid Prohesperocyon wilsoni [41], the 'anchitheriine' equid Mesohippus texanus [106,107] and the advanced stem primate or basal crown primate Rooneyia viejaensis [108,109] in addition to the early amphicyonids listed above. Together with a recent discovery of the first Chadronian fauna in southern Mexico that contained an early jimomyid rodent and a derived chalicothere along with two undetermined caniform species (from published data, we think they might represent amphicyonids given their relatively large sizes) [110], these biogeographic patterns highlight the low-latitude regions within North America as a potential centre of mammalian diversification during the Middle to Late Eocene.   Appendix B. Additional notes on character matrix data Changes to data from previous studies [10,14] Character 25: The definition of State 2 was slightly modified such that it included all conditions in which there was no evidence for transpromontorial passage of internal carotid artery, without implying a specific alternative position (e.g. inside or outside the auditory bulla) for that artery. Accordingly, the character state for Daphoenus spp. (originally scored as unknown [10]) was scored as State 2. While Hunt ([117], p. 1032) listed probable transpromontorial course of internal carotid artery as one of the diagnostic features of the subfamily Daphoeninae, a groove on the promontorium that would indicate such a condition has only been documented for F:AM 76205 of D. tedfordi [117]. for the presence of rostral entotympanic. The problem is reflected in the apparently conflicting states that have been reported for the same taxa in different studies: e.g. viverravid Protictis schaffi either has a rugose area on the promontorium ([120], p. 112; interpreted by Wang & Tedford ([9], p. 5) as a probable site of attachment for the rostral entotympanic) or does not ( [10]; Character 30 scored as State 0), and the basal carnivoraform Vu. profectus either has ( [10]; Character 30 scored as State 1) or lacks ( [9], p. 5) such an area. The character state for Daphoenus was originally scored as State 1 [10], but neither the rostral entotympanic itself nor its attachment site on the promontorium or basisphenoid has been clearly identified in amphicyonids [2]. While the rostral entotympanic was present in all of the extant carnivorans examined by Hunt [55], further anatomical investigation is warranted to determine-if at all possible-its presence among basal carnivoramorphans.
Character 59: 'Miacis' gracilis (originally scored as unknown [10]) was scored as State 1 (m2 with short talonid and oval outline) because the holotype CM 11900 and a referred specimen CM 12063 clearly show this condition.
Note Daphoenus spp. was scored as State 2, but one of the seven specimens (FMNH UM 741, tentatively identified as D. cf. D. hartshornianus) examined for this character showed a small cuspulid (State 0); the latter is here considered to be an aberrant condition, but polymorphic scoring may be warranted if such a condition is found to be more prevalent.