A new large squalodelphinid (Cetacea, Odontoceti) from Peru sheds light on the Early Miocene platanistoid disparity and ecology

The South Asian river dolphin (Platanista gangetica) is the only extant survivor of the large clade Platanistoidea, having a well-diversified fossil record from the Late Oligocene to the Middle Miocene. Based on a partial skeleton collected from the Chilcatay Formation (Chilcatay Fm; southern coast of Peru), we report here a new squalodelphinid genus and species, Macrosqualodelphis ukupachai. A volcanic ash layer, sampled near the fossil, yielded the 40Ar/39Ar age of 18.78 ± 0.08 Ma (Burdigalian, Early Miocene). The phylogenetic analysis places Macrosqualodelphis as the earliest branching squalodelphinid. Combined with several cranial and dental features, the large body size (estimated body length of 3.5 m) of this odontocete suggests that it consumed larger prey than the other members of its family. Together with Huaridelphis raimondii and Notocetus vanbenedeni, both also found in the Chilcatay Fm, this new squalodelphinid further demonstrates the peculiar local diversity of the family along the southeastern Pacific coast, possibly related to their partition into different dietary niches. At a wider geographical scale, the morphological and ecological diversity of squalodelphinids confirms the major role played by platanistoids during the Early Miocene radiation of crown odontocetes.

The South Asian river dolphin (Platanista gangetica) is the only extant survivor of the large clade Platanistoidea, having a well-diversified fossil record from the Late Oligocene to the Middle Miocene. Based on a partial skeleton collected from the Chilcatay Formation (Chilcatay Fm; southern coast of Peru), we report here a new squalodelphinid genus and species, Macrosqualodelphis ukupachai. A volcanic ash layer, sampled near the fossil, yielded the 40 Ar/ 39 Ar age of 18.78 ± 0.08 Ma (Burdigalian, Early Miocene). The phylogenetic analysis places Macrosqualodelphis as the earliest branching squalodelphinid. Combined with several cranial and dental features, the large body size (estimated body length of 3.5 m) of this odontocete suggests that it consumed larger prey than the other members of its family. Together with Huaridelphis raimondii and Notocetus vanbenedeni, both also found in the Chilcatay Fm, this new squalodelphinid further demonstrates the peculiar local diversity of the family along the southeastern Pacific coast, possibly related to their partition into different 2018 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited.

Body size
To estimate the TBL of MUSM 2545 and of all other fossil platanistoids with single-rooted posterior teeth included in our phylogenetic analysis, we used the equation provided by Pyenson & Sponberg [30] for stem Platanistoidea, based on the BZW: log(TBL) = 0.92 × (log(BZW) − 1.51) + 2.49 Based on the estimated TBL and the 50% majority-rule consensus tree obtained in our phylogenetic analysis as a backbone, we investigated changes in body size among squalodelphinids and related platanistoids using MESQUITE v. 2.74 [31]. For this analysis, the TBL was considered as an ordered character, including three distinct states that we defined by placing the division in correspondence with the largest gaps of our sample.

40 Ar/ 39 Ar isotopic analysis
To obtain an absolute dating through the 40 Ar/ 39 Ar method, a tephra layer CHILC-AT1 was sampled in the Chilcatay Fm outcropping 1.7 km southeast (SE) to the locality of the M. ukupachai holotype. The sample locality (geographical coordinates: 14°23 49.85 S, 75°53 27.35 W) is 150 m SE to an uncollected skeleton that most likely belongs to the same species as the M. ukupachai holotype.
After sieving, biotite crystals larger than 250 µm were separated by hand-picking under a stereoscopic microscope and a sample of 17.87 mg was selected for dating. Electron Probe Micro-Analyses (EPMAs) were performed using a JEOL 8200 Super Probe at the University of Milan to characterize the chemical composition of the tephra glass and to check for the lack of alteration of the biotite phenocrysts. For 40 Ar/ 39 Ar dating, the biotite sample was irradiated in the nuclear reactor at

General morphology
With a CBL greater than 770 mm and a BZW of 370 mm (table 1), the cranium of M. ukupachai is larger than in all other known squalodelphinids (H. raimondii CBL = 494 mm, BZW = 207 mm; N. vanbenedeni CBL = 582-634 mm, BZW = 235-254 mm; S. fabianii: CBL = 640 mm, BZW = 263 mm). The skull of the holotype of Medocinia tetragorhina Delfortrie, 1875 [40] is too fragmentary to provide estimates of these measurements, but other skull measurements, for example the width at rostrum base, are smaller than in Macrosqualodelphis. The original CBL of Macrosqualodelphis can be tentatively estimated as a percentage of the CBL within the ranges of the other squalodelphinids (63-70%) [13]. Using these percentages, we obtain an estimated rostrum length for Macrosqualodelphis varying between 488 and 644 mm, with the lower value smaller than the preserved rostrum length (490 mm). Using the higher value, the estimated missing anterior portion is 154 mm and the estimated CBL is 924 mm.
The rostrum is less abruptly tapering from its base to its anterior end than in Notocetus, Squalodelphis and particularly Huaridelphis, all of them having a narrow anterior half and a wide, triangular posterior half of the rostrum. To better quantify this feature, also on skulls lacking the anterior portion of the rostrum, we measured the width of the rostrum at a distance from the rostrum base twice the width across the antorbital notches. The ratio between this measurement and the width at the antorbital notch is 0.37 in Macrosqualodelphis and <0. 35 in Huaridelphis, Notocetus and Squalodelphis. A value close to the one of Macrosqualodelphis is observed in the aff. H. raimondii MUSM 603, also from Chilcatay Fm, as described by Lambert et al. [13].
As in the other squalodelphinids, the antorbital notches are distinctly asymmetrical, having (i) the right antorbital notch more posteriorly located than the left and (ii) the lateral and medial borders of the antorbital notch drawing a more open angle on the right side (86°) than on the left side (60°); the ratio between the two angles is approximately 1.4, intermediate between Notocetus (1.7) and Huaridelphis (1.2). Moreover, the left antorbital notch is more U-shaped than the V-shaped right antorbital notch. By contrast, in all other squalodelphinids, both antorbital notches are V-shaped.
As in all other squalodelphinids and in platanistids, the vertex and the bony nares are distinctly shifted on the left side, as clearly evidenced by the oblique orientation of the main transverse axis of the nasals and, perpendicular to this axis, of the nasal septum made of the presphenoid (see below).
The temporal fossa is dorsoventrally higher and more anteroposteriorly elongated than in all other squalodelphinids, extending posteriorly beyond the occipital condyles due to a salient temporal crest. The temporal fossa also exhibits a significant transverse widening, as can be seen in posterior view. Among squalodelphinids, a similar widening is only observed in Squalodelphis, related to a more lateral position of the zygomatic process of the squamosal. It is interesting to note that, although squalodontids have a dorsoventrally and anteroposteriorly large temporal fossa large as in Macrosqualodelphis, their fossa is significantly transversely narrower than in Macrosqualodelphis (figure 2).

Premaxilla
Owing to the poor preservation of the anteroventral portion of the rostrum, the extent of the anterior premaxillary portion of the rostrum and the presence of dental alveoli in this apical premaxillary portion cannot be assessed.
In dorsal view, the medial margins of the right and left premaxillae contact each other for approximately 150 mm from the preserved anterior end of the rostrum; then the premaxillae gradually diverge towards the V-shaped bony nares (figure 3). The dorsal opening of the mesorostral groove remains narrow for all its anteroposterior extent, reaching a transverse width of 15 mm near the anterior end of the bony nares. This condition is intermediate between Huaridelphis, whose premaxillae contacting medially for about half the length of the rostrum, and Notocetus, whose mesorostral groove is open until or in close proximity of the apex of the rostrum. As in Macrosqualodelphis, both Huaridelphis and Notocetus retain a narrow opening of the mesorostral groove near the rostrum base, whereas Medocinia and Squalodelphis display a wide opening.
From the anterior end of the anteromedial sulcus, between 220 and 330 mm anterior to the right antorbital notch, the right premaxilla is clearly narrower than the left, a condition shared with all other squalodelphinids and most platanistids [13].  Both premaxillae exhibit their maximum transverse width approximately 60 mm anterior to the right antorbital notch, a condition also observed, with some degree of intraspecific variation, in the other squalodelphinids.
The posterior rostral portion of the premaxillae is also featured by a marked medial slope, forming a prenarial depression having its maximum depth at the level of the right antorbital notch. Here, the vertical distance between the lateral and the medial margins of the premaxilla reaches 18 mm. A similar prenarial depression is observed in all other squalodelphinids and, more or less marked, in most of the other homodont platanistoids [13].
A single premaxillary foramen is clearly visible on both premaxillae, at the level of the right anterior notch. Among the other squalodelphinids, the premaxillary foramen is anterior to the right antorbital notch in Huaridelphis and Squalodelphis, whereas its position varies from anterior to weakly posterior to the right antorbital notch in the three known skulls of Notocetus. The elongated anteromedial sulcus (approx. 15 mm) and the posteromedial sulcus are both weakly excavated, whereas the posterolateral sulcus is deep and clearly discernible, reaching posteriorly the anterior limit of the bony nares. The premaxillary sac fossa is narrow, weakly concave and slopes medially. The maximum transverse widths of the right and left premaxillary sac fossae are roughly identical. Both ascending processes of the premaxillae are deeply incised by a longitudinal groove laterally margined by a thick ridge. This groove might be homologous to the premaxillary cleft described in Waipatia [41] and also observed in Papahu [42]. It is also visible in all other squalodelphinids having this region well preserved and in most platanistids [13]. Right and left ascending processes of the premaxillae display a short posteromedial angle contacting the corresponding frontal at the vertex.

Maxilla
In dorsal view, the transverse width of the maxilla is roughly constant on most of the length of the rostrum and consistently decreases at the level of the maximum widening of the premaxilla (at approx. 60 mm anterior to the right antorbital notch) (figure 3). At this level, the ratio between the transverse width of maxillae and the transverse width of the premaxillae reaches the minimum value of 0.30. More posteriorly, this ratio increases, reaching a value of 0.68 that is closer to Huaridelphis (0.60-0.61) and Notocetus (0.56-0.68), but significantly smaller than in Medocinia and Squalodelphis (0.82), both having the premaxilla nearly reaching the lateral margin of the rostrum.
In the posterior portion of the rostrum, the maxilla becomes dorsoventrally thinner laterally, with a slender, blade-like lateral margin of the rostrum, whereas this margin is significantly thicker in other squalodelphinids. This feature is clearly visible in lateral view, together with the steep ascent of this lateral margin towards the antorbital notch.
As in all other squalodelphinids with the exception of Medocinia (whose holotype skull only preserves the posterior part of the rostrum), the unfused lateral maxilla-premaxilla suture is not excavated by a deep groove, contrasting with most platanistids (except Araeodelphis [15], allodelphinids, eurhinodelphinids and eoplatanistids).
A single infraorbital foramen pierces the maxilla 50 mm posterior to the right antorbital notch, whereas two foramina are present 20 and 50 mm posterior to the left notch. All these foramina are located near the medial margin of the maxilla, in the area of the greatest concavity of the lateral margin of the premaxilla.
From the base of the rostrum, the maxilla extends posterolaterally, forming the posterior wall of the antorbital notch, but not covering the anterolateral portion of the preorbital process of the frontal and the antorbital process of the lacrimal. The antorbital process of the maxilla is elevated (the left more than the right) in relation to the dorsoventral thickening of both the maxilla and the frontal in this  area. A similar thickening is also observed in other squalodelphinids: it is much more pronounced in Squalodelphis, almost absent in Medocinia, and similarly developed in Notocetus. An extreme condition is observed in platanistids, in which it forms an elevated maxillary (or frontomaxillary) crest. Posterior to the postorbital process, the maxilla and the underlying frontal become very thin and, consequently, they are broken and partially missing on the left side. On the right side, however, the lateral edge of the maxilla and underlying frontal is apparently preserved, although a large breakage is observed in the middle of the dorsal aspect of the maxilla. The posterior dorsal infraorbital foramina are apparently absent, but they could have been originally located on the missing part of the maxillae.
The left maxilla descends more abruptly ventrolaterally from the vertex than the right maxilla, forming a deep fossa posterolateral to the left nasal. Linked to the shift of the vertex towards the left side, this feature is also observed in other squalodelphinids, even if is less marked in Huaridelphis. Owing to recent erosion, most of the palatal surface of the maxilla is missing (figure 4a,b). However, on the preserved portion of the rostrum, 10 eroded alveoli are visible near the lateral margin of each maxilla. Most of these alveoli still hold partly broken single-rooted teeth. The transverse diameter of the alveoli ranges from 10 mm anteriorly to 13 mm posteriorly. The apparent smaller transverse diameter of the anterior alveoli is due to the fact that the cross-sections of the alveoli and associated dental roots are closer to their narrower deeper portions. Although the spacing of the alveoli varies irregularly along the alveolar row, it is smaller posteriorly (0-5 mm) than anteriorly (up to 20 mm). The posteriormost right alveolus is 106 mm anterior to the right antorbital notch (115 mm on the left side). Posterior to the posteriormost alveolus, the maxilla rises abruptly posterodorsolaterally, generating a deep excavation of

Presphenoid
The ossified portion of the presphenoid ( = mesethmoid from previous works; see [43]) exhibits a narrow and elongated nasal septum separating the bony nares ( figure 3). This septum draws an angle of 14°w ith the main axis of the skull. The posterodorsal margin of the cribriform plate reaches the anterodorsal margin of the nasals.

Nasal
The nasals are nodular, with an inflated and subhorizontal dorsal surface, reaching a level higher than the frontals (figures 3 and 5). For all these features, the nasals of Macrosqualodelphis are similar to those of Notocetus, whereas they differ from those of the other squalodelphinids, all characterized by a flat and anteriorly sloping dorsal surface of the nasals. Nevertheless, the nasals of Macrosqualodelphis differ from those of Notocetus in their smaller general size and in being proportionally anteroposteriorly shorter.
There is no significant difference in size and shape between the right and left nasals. The longitudinal axis of the nasals is slightly obliquely oriented, drawing an angle of 6°with the main axis of the skull, and the anterolateral corner of the right nasal is 10 mm anterior to the anterolateral corner of the left nasal. A similar oblique orientation of the longitudinal axis of the nasals is present in Huaridelphis, Notocetus and Squalodelphis. The anterior margin of both nasals is weakly anteriorly convex, with the anterior edge of the joined nasals forming an anteromedial angle of 155°. The nasal-frontal suture is on the whole straight, with only a small anteromedial process of the left frontal wedged between the nasals. The condition of Macrosqualodelphis is intermediate between Notocetus, whose anterior and posterior margins of the nasals are anteriorly convex, and Huaridelphis, whose both margins are straight.

Frontal
The dorsal exposure of both frontals at the vertex slopes anteroventrally and has a minimal transverse width slightly smaller than the transverse width of nasals (figure 3). The right and left frontals are subequal in length, differing in this respect from Huaridelphis, Notocetus and Squalodelphis, but resembling Medocinia. By contrast, the left frontal is significantly transversely wider than the right as observed in the other squalodelphinids (but the medial suture between the frontals is not visible in Squalodelphis). The medial suture between the frontals is straight and the frontal-occipital suture is transversely oriented with a small anteromedial process of the supraoccipital wedged between the frontals anteriorly.
The preorbital process of the frontal is dorsoventrally thickened (more than in Huaridelphis and less than in Medocinia), the orbit is anteroposteriorly short, and the postorbital process is robust and dorsoventrally elongated (figure 5). On the medial portion of the ventral surface of the orbit roof, a fossa is partially filled by sediment (figure 4a,b). A similar fossa, but slightly larger and deeper, has been observed in Huaridelphis and Notocetus, and interpreted as corresponding to an extension of the pterygoid sinus in the orbit region [13,44].

Supraoccipital
The nuchal crest is prominent and, unlike in all other squalodelphinids, higher than the frontals and the nasals at the vertex (figures 3a-c and 5). This crest is markedly transversely wide, as in all other squalodelphinids with the exception of Squalodelphis, the latter having a narrower nuchal crest (approximately as wide as the greatest width of the premaxillae). The nuchal crest of Macrosqualodelphis is roughly straight in dorsal view, whereas the outline of the supraoccipital shield, formed medially by the nuchal crests and laterally by the two prominent temporal crests, is half-circle shaped in posterodorsal view.
In dorsal view, the temporal crest extends far posterolaterally, increasing the length of the temporal fossa. The complete right temporal crest extends far beyond the occipital condyles.
The posterodorsal surface of the supraoccipital shield is transversely concave and there is no external occipital crest (sensu [25]) (figure 4c,d).

Palatine
The palatines cannot be identified on the ventral surface of the skull, either due to their complete fusion with the maxillae or because they are fully covered by the pterygoids (figure 4a,b).

Pterygoid
The pterygoid is long and narrow, as in other squalodelphinids and platanistids. Its pointed anterior apex extends 60 mm beyond the level of the right antorbital notch ( figure 4a,b).
The pterygoid sinus fossa reaches the level of the right antorbital notch, whereas it extends beyond the antorbital notch in other squalodelphinids.
The lateral lamina of the pterygoid is a robust, plane plate that contacts posteriorly the falciform process of the squamosal as is observed in all the other squalodelphinids.

Jugal-lacrimal
In ventral view, the lacrimal and the fused preserved anterior portion of the jugal are longitudinally short as in the other squalodelphinids (figure 4a,b). They form the anteriormost portion of the ventral surface of the antorbital process and, more medially, the posterior margin of the antorbital notch.
A narrow ventral projection of the lacrimojugal complex is preserved on the left side of the skull (figure 5c,d). A similar peculiar structure was also observed in the holotype of N. vanbenedeni and seems to be analogous to the ventroposterior projection of the jugal described in the holotype of the eurhinodelphinid Eurhinodelphis cocheteuxi du Bus, 1867 [45] by Lambert ([46], fig. 3).

Squamosal
In lateral view, the zygomatic process of the squamosal shares with the other squalodelphinids the same strongly swollen aspect ( figure 5). This process is more anteriorly elongated than in Huaridelphis. Its posterodorsal margin is markedly convex and its anteroventral margin is slightly convex. The ratio between the maximum distance from the anteroventral margin of the zygomatic process to its posterodorsal margin, in lateral view, and the vertical distance from the lower margin of the occipital condyles to the vertex of the skull is 0.43, a value indicating a robustness of the zygomatic process similar to other squalodelphinids.
As mentioned above, the postglenoid process is lacking in both squamosals, due to recent erosion of the skull.
In ventral view, a deep, narrow, 50 mm long depression on the anterolateral margin of the zygomatic process represents the suture for the missing posterior portion of the jugal. The mandibular fossa is wide, occupying most of the ventromedial surface of the zygomatic process and being laterally defined by the thin ventral margin of the process. The tympanosquamosal recess is transversely narrow. More medially, the falciform process is a wide plate contacting anteriorly the lateral lamina of the pterygoid.

Exoccipital
The occipital condyles are posteriorly prominent with a conspicuous neck in ventral view (figure 4c,d).
The dorsal condyloid fossae are visible in posterior view, dorsolateral to the occipital condyles. The exoccipital extends far laterally and its dorsal margin contributes significantly to the transversely wide posteroventral margin of the temporal fossa. Together with all parts of the exoccipital ventral to the condyles, the paroccipital processes are missing.

Basioccipital
The ventral surface of the basioccipital is not well preserved and the basioccipital crests are broken and almost completely missing (figure 4a,b).

Vomer
In ventral view, the vomer is visible between the choanae, in a region partly filled with sediment (figure 4a,b).

Teeth
Based on the abraded ventral surface of the rostrum showing broken dental roots and alveoli, each maxilla carried more than 10 single-rooted teeth ( figure 4a,b). Moreover, considering the estimated missing portion of the rostrum (154 mm), it is probable that the original upper tooth count per quadrant reached a value close to Squalodelphis (15) and lower than other squalodelphinids for which the upper tooth count is known (Huaridelphis, 28-30; Notocetus, [22][23]. The posterior maxillary alveoli have a transverse diameter of approximately 13 mm, that is 3.5% of the BZW, a value higher than in other squalodelphinids (all with values lower than 3.0%).
Only one complete and two fragmentary detached single-rooted teeth are preserved ( figure 6). The complete tooth, the only one having the crown preserved, is 55 mm in length. It is curved and crescentiform in labial and lingual views, and straight and fusiform in mesial and distal views. The crown is small, having a diameter at its base of only 8 mm contra 48 mm of the maximum diameter of the root. The root is transversely flattened (ratio between the maximum mesiodistal and labiolingual diameters of 0.33). The maximum transverse diameter of the root is 18    diameter of the posterior maxillary alveoli reaching roughly 13 mm, suggesting that anterior teeth were significantly larger than the posterior teeth. The two other teeth only preserve the root, with a length of 58 and 55 mm corresponding to teeth even bigger than the complete tooth described above. Considering their shape and large size, with a diameter significantly larger than the posterior maxillary alveoli, it is probable that these teeth originate from the missing anterior portions of the rostrum and of the mandibles. Similar large anterior teeth are also present in Squalodelphis, and several platanistids including the extant Platanista, which displays anterior teeth considerably larger than posterior ones [47,48].

Atlas
The well-preserved, anteroposteriorly thick atlas of Macrosqualodelphis is not fused to the missing axis (figure 7a-e and   The other preserved thoracic vertebra of Macrosqualodelphis exhibits a more anteroposteriorly elongated centrum (ratio between length and height of the centrum = 1.2), shorter transverse processes, a dorsoventrally compressed, heart-shaped outline of the centrum in anterior and posterior views, pedicles vertically rather than obliquely oriented, a narrower neural arch and a higher neural spine (figure 7i-k). This vertebra is very similar to the 'Thoracic D' of Huaridelphis, also sharing similarities with T4-T5 of Phocageneus venustus. These latter differ from the vertebrae of Macrosqualodelphis and 'Thoracic D' of Huaridelphis in having a more rounded ventral margin of the centrum in anterior and posterior views and a neural spine vertical rather than posteriorly inclined.

Lumbar vertebrae
Two large vertebrae are interpreted as the two last lumbars, because they are the anteriormost and the only vertebrae without facets for the chevrons (haemal arches) of a sequence of six vertebrae found in articulation (figure 8e,f,k,l,q,r). These two vertebrae have a cylindrical, elongated centrum (ratio between length and height of the centrum = 1.26-1.27), bearing a marked medial keel on the ventral surface. A pair of wide and deep sulci runs obliquely from the centre of the ventral surface forward with the posterolateral margins. Similar grooves are also present on the following Ca1-Ca4 of MUSM 2545 (figure 8m-p) and have been named 'hypovertebral grooves' by Aguirre-Fernández et al. ([52], fig. 9) on two isolated lumbar vertebrae from the Miocene of Venezuela. According to these authors, together with the proportionally very elongated centrum, the presence of hypovertebral grooves supports the assignation of the two vertebrae from Venezuela to cf. Zarhachis flagellator Cope, 1868 [53], because similar grooves have been first observed on the four vertebrae described as the type material of Z. flagellator [53,54]. However, similar grooves have been described in several cetaceans and interpreted as related to the passage of the arteries departing from the caudal portion of the abdominal aorta [55]. We observed the same grooves, although generally less excavated than in Macrosqualodelphis, on the lumbar and caudal  vertebrae of most of the extant and many fossil odontocetes. In some cases, on the caudal vertebrae, we note that each groove is laterally connected to the vertebrarterial canal, suggesting, as already pointed out by Slijper [55], that the artery runs along the groove and crosses the transverse process of the vertebra to reach the dorsal tissues. We therefore rather use the term 'vertebrarterial groove' instead of 'hypovertebral groove' as proposed by Aguirre-Fernandez et al. [52]. Furthermore, by denying any systematic relevance to this character, we suggest that the referral to the family Platanistidae of the lumbar vertebrae from the Miocene of Venezuela should remain tentative. The elongated transverse processes of the two lumbar vertebrae of Macrosqualodelphis start from the lateral borders of the centrum, and are dorsoventrally flattened, weakly widened distally and ventrally and posteriorly directed. The neural canal is narrow and high. Partly preserved only on the

Caudal vertebrae
Four of the eight preserved caudal vertebrae are presumably the anteriormost ones (Ca1-Ca4) (figure 8ae,g-j,m-p). They are close in size and shape to the posteriormost lumbars, the main difference with the lumbars being the presence of facets for the chevrons. Other differences are the lesser elongation of the centrum (slightly decreasing from Ca1 to Ca4), the smaller size of the neural arch and the transverse processes being more posteriorly directed, but perpendicular to the longitudinal axis of the centrum in dorsal and ventral views. Moreover, the transverse processes of Ca3 and Ca4 do not widen distally, being instead anteroposteriorly pointed. The other four preserved caudal vertebrae are probably the last ones, corresponding to the fluke region ( figure 9). They are considerably smaller compared to Ca1-Ca4, they lack transverse processes and neural arch, and are anteroposteriorly and dorsoventrally compressed. Their surface is damaged by erosion and the vertebrarterial canals are only partly visible on the dorsal and ventral surface. The smaller last caudal has an irregular nodular shape.

Forelimb
The humerus, radius and ulna of the left forelimb have been maintained in anatomical connection after preparation (figure 10), as found in the field, whereas the two manus bones were found scattered in the sediment.

Humerus
The humerus is robust and transversely flattened, stockier than in allodelphinids and waipatiids [56][57][58]. It is somewhat longer than the radius (ratio between their respective lengths = 1.24). The humerus is similarly longer than the radius in only a few extant odontocetes, including monodontids, physeteroids, Inia and Platanista [59,60], whereas this feature is commonly observed in extinct platanistoids and related taxa (e.g. allodelphinids, eurhinodelphinids, the early platanistoid Otekaikea huata, the squalodontid Kelloggia (probably synonymous to Squalodon) barbara Mchedlidze, 1976 [56] and the probable waipatiid Sulakocetus [5,57]). The humeral head is hemispherical and protrudes posterolaterally. Medially to the head, the lesser tubercle is robust, higher than the head and the smaller, anteriorly located greater tubercle. The greater tubercle lies on the anteromedial margin of the humerus, extending distally in a salient and elongated deltopectoral crest. This crest reaches a level closer to the distal epiphysis than observed in allodelphinids, Otekaikea huata Tanaka & Fordyce, 2015 [5], and waipatiids, whereas Platanista lacks any defined crest [47,59]. On the lateral surface of the diaphysis, posterior to the deltopectoral crest, there is a large and deep fossa for insertion of M. infraspinatus. The posterior margin of the humerus is concave, due to the slight anteroposterior widening of the distal epiphysis (to a much lesser extent than in Platanista).

Radius
The radius lacks the posterior portion of its distal epiphysis. It is a transversely flattened trapezoidal bone that slightly widens distally. It is proportionally longer than in Platanista, but considerably shorter, stockier than in allodelphinids, O. huata and waipatiids, and more similar to some eurhinodelphinids (e.g. Schizodelphis sp. USNM 244413) and squalodontids (e.g. Kelloggia barbara [56] and Squalodon bellunensis Dal Piaz, 1901 [61] MGP 26092). The radius is proximally articulated with the humerus and, for a small tract of its posterior margin, with the ulna.

Ulna
The ulna lacks almost its whole distal half. Like the radius, it is transversely flattened; it is strongly anteriorly articulated with the latter bone and proximally with the humerus. The olecranon is roughly half-circle shaped in lateral and medial view, forming an open notch with the posterior margin of the diaphysis. The olecranon is less developed anteroposteriorly than in allodelphinids and waipatiids, with proportions more similar to eurhinodelphinids and squalodontids, whereas Platanista lacks such a process. Distal to the olecranon, the ulna is significantly anteroposteriorly narrower than the radius.

Manus bones
The preserved bones of the manus are two transversely flattened and straight small bones that differ significantly one from the other in the size and shape ( figure 11). The largest has the mesial and distal epiphyses wider than the diaphysis. Owing to its large size and convex proximal margin in lateral and medial view, this bone occupied a more proximal position along the corresponding digit, probably as a metacarpal.   The smaller bone narrows significantly distally and is interpreted as a phalanx, located in a distal position along the corresponding digit.
These two bones do not differ significantly from the manus bones of Platanista and Z. flagellator [47,54,59].

Phylogeny
The cladistic analysis produced 120 equally parsimonious trees, with tree length = 80, consistency index (CI) = 0.60 and retention index (RI) = 0.82. The strict consensus tree and the 50% majority-rule consensus tree are presented in figure 12. The strict consensus tree obtained here shows the same relationships within the homodont platanistoids as the tree of Godfrey et al. [15], summarized in the basalmost position of Allodelphinidae and the sister group relationship between Platanistidae and Squalodelphinidae, both families resulting as monophyletic groups. This analysis also confirms the position of Dilophodelphis within the platanistids, as already proposed by Boersma et al. [44] using a matrix modified from Godfrey et al. [15].
Our consensus tree supports the referral of Macrosqualodelphis to the family Squalodelphinidae, of which it is the earliest diverging lineage. The relationships between other squalodelphinids remain unresolved, as in previous analyses [13,15,44]. The 50% majority-rule consensus tree provides a more satisfactory result, with the specimen MUSM 603 branching before the two other South American genera Huaridelphis and Notocetus (unresolved relationships), and a clade including all the squalodelphinids from the North Atlantic realm.
The referral of Macrosqualodelphis to the family Squalodelphinidae is also supported when this taxon and Huaridelphis are included in the taxonomically broader matrix of Tanaka & Fordyce [6], as modified by Lambert et al. [20] and with the few further changes and additions reported in the Material and methods section and in table 4 of appendix B. Analysis 1 (equally weighted characters and no molecular constraint) produced 3919 equally parsimonious trees, with tree length = 1839, CI = 0.24 and RI = 0.65; analysis 2 (down-weighted homoplastic characters and no molecular constraint) produced 189 equally parsimonious trees, with tree length = 1888, CI = 0.23 and RI = 0.64; analysis 3 (equally weighted homoplastic characters and molecular constraint) and analysis 4 (weighted homoplastic characters and molecular constraint) both produced 272 equally parsimonious trees with tree length = 1925, CI = 0.23 and RI = 0.63 (consensus trees in figure 17 of appendix B).
Although this second set of analyses also supports the monophyly of Squalodelphinidae and the sister group relationship between Platanistidae and Squalodelphinidae, it does not resolve the relationships within the squalodelphinids.

Biostratigraphic and 40 Ar/ 39 Ar age constraint for
Macrosqualodelphis ukupachai The age of the Chilcatay Fm has been described in the past literature as spanning from the latest Oligocene to the earliest Middle Miocene based on diatoms, foraminifers and molluscs [34][35][36][37]. In the Western Ica Valley area, our biostratigraphic and 40 Ar/ 39 Ar datings converge and constrain the age of this formation to the Early Miocene. In Roca Negra, the type locality of the heterodont dolphin Inticetus vertizi Lambert et al. [20], the base of the Chilcatay Fm was assigned through silicoflagellate biostratigraphy to the Naviculopsis ponticula zone of Bukry [62], dated by Bukry [63] between 19 and 18 Ma by correlation with the coccolith Sphenolithus belemnos zone at DSDP Site 495 offshore Guatemala [20]. We identified the same biozone at the locality of Ullujaya, a rich fossil marine vertebrate-bearing locality [13,19,21]. In the latter locality, the presence of N. ponticula subsp. spinosa indicates, following the species dominance described by Bukry [64], a slightly younger age within the same biozone.
The top of the Chilcatay Fm is constrained at 18.02 ± 0.07 Ma, through 40 Ar/ 39 Ar age dating of a tephra layer collected by us 1 m below the erosional contact with the overlying Pisco Formation at Cerro Submarino. At the same locality, within diatomaceous sediments, the presence of Corbisema triacantha, Distephanopsis crux subsp. parva and subsp. scutulata, Stephanocha speculum cf. triommata and the absence of Naviculopsis allows us to assign these samples to the Cannopilus schulzii subzone within the C. triacantha zone, dated between 18 and 13.5 Ma [63] ( figure 13).
The tephra layer CHILC-AT1, sampled 1.7 km SE of the holotype of M. ukupachai, near an uncollected squalodelphinid skeleton most likely belonging to the same taxon, is composed of 90% glass shards and 10% juvenile crystals, mainly biotite, as estimated by optical microscopy. EPMA analyses on volcanic glasses show a rhyolitic composition, whereas the biotite crystals suggest a calc-alkaline origin. Biotite analyses reveal a slight loss of K in the interlayer occupancy, but the petrological composition, the chemical homogeneity of the biotite population and the lack of sedimentary evidence suggest that this tephra layer is a primary air-fall. The level of post-eruptive marine alteration was low.
Considering the 'isochemical steps' [66] as the heating steps most closely reflecting the degassing of biotite crystals, we calculated the 40 Ar/ 39 Ar age from steps 4-9 with the lowest Ca/K and Cl/K ratios, obtaining a weighted average of 18.80 ± 0.06 Ma (2σ ), with a mean square weighted deviation (MSWD) value of 12, and an isochron age of 18.70 ± 0.13 Ma (2σ ), with an MSWD value of 7.3 ( figure 14 and appendix C). However, both these dispersion values are too high, which points to a systematic bias, such    as suggested by the substoichiometric K concentration of 6.8% calculated from the total 39 Ar release. For this reason, we can consider only the age given by step 9, which is the most gas-rich step (greater than 30%) and the one with the lowest Ca/K: the age calculated is 18.72 ± 0.02 Ma (2σ ), which overlaps with the weighted average, as shown in figure 14.
The most conservative age estimate covers the entire 2-sigma confidence interval between 18.70 and 18.86 Ma. If this confidence interval was symmetrical and Gaussian, it would correspond to an age of 18.78 ± 0.08 Ma, which can be considered as the age of this tephra layer. The Ar results are available in appendix C.   [39]. Note that with its estimated TBL of 3.5 m, Macrosqualodelphis is markedly larger than any other homodont platanistoid analysed here.
with an estimated TBL of 3.5 m Macrosqualodelphis is by far the largest homodont platanistoid ( figure 15). In fact, more than half of the platanistoids of the sample have a TBL smaller than 2.3 m, and the others do not exceed 2.9 m in length except Macrosqualodelphis. A significant point is that squalodelphinids are the clade of homodont platanistoids displaying the widest range in TBL, varying from 2.0 m in Huaridelphis to 3.5 m in Macrosqualodelphis, with at least two evolutionary shifts to a smaller size and one to a larger size. This wide range in size is most likely related to the greater diversity of squalodelphinids included in the sample. High diversity in an evolving clade also generates an increase in the maximum body size [67]. Contrasted sizes, combined with other cranial and dental features, are examined below for Macrosqualodelphis and the other squalodelphinids from the Chilcatay Fm, in order to understand the ecological significance of the diversity of this fossil platanistoid clade.

Ecological segregation for squalodelphinids of the Chilcatay Fm
The description of a third squalodelphinid species from the same lithological unit (Chilcatay Fm) and the same geographical region (Pisco Basin) raises the question of how these three related species shared food resources along the western coast of South America during the early Burdigalian. Constituting a key parameter for local diversification, ecological niche segregation among closely related, sympatric species has been investigated in a number of extant cetaceans, including Delphinidae (true dolphins) and Ziphiidae (beaked whales) (e.g. [68][69][70]). These studies demonstrated that resource partitioning may result from several ecological traits, or combinations of these traits: different foraging habitat (depth, distance to the coast), different behaviour (for example, diel variations in foraging activities) and different feeding ecology (different prey types/position along the local trophic chains). Considering that specimens of M. ukupachai were, up to now, not found in the localities of H. raimondii and N. vanbenedeni (Ullujaya; [13,21], it is tempting to hypothesize a different foraging habitat) ( figure 16). In other respects, because they were not discovered in the same locality, there is no evidence that M. ukupachai on the one hand and H. raimondii and N. vanbenedeni on the other hand proceed from exactly the same level in the Chilcatay Fm, and therefore were coeval. If M. ukupachai and H. raimondii + N. vanbenedeni were not contemporaneous, the two groups may very well have had similar foraging habitats without competing. Therefore, our limited sample size, the lack of comparative data about the palaeoenvironmental conditions in various localities of the Chilcatay Fm and the need for an even more refined chronostratigraphic framework should urge for caution when dealing with such considerations. Obviously, diel variations in foraging activities cannot be tested for extinct taxa. Based on comparative skull morphology and dimensions, we thus only assess potential differences in feeding ecology among these three squalodelphinid species. Marked anatomical differences are noted at different levels: Extending the comparison to the whole family Squalodelphinidae, similar differences are observed, both considering the body size range (see above) and the disparity in cranial and dental morphology (even if the only other squalodelphinid whose rostrum is known is Squalodelphis, having the rostrum significantly tapered and 15 teeth per row). Therefore, at a wider geographical scale, the observed squalodelphinid morphological and ecological diversity further illustrates the broad diversification of homodont platanistoids during the Early Miocene crown toothed whale radiation. Altogether, these anatomical differences between squalodelphinids are reminiscent of differences between several morphotypes of extant delphinids, for example the smaller common dolphin Delphinus delphis Linnaeus, 1758 [71] with a more tapered rostrum, higher tooth count, and smaller teeth, the intermediate bottlenose dolphin Tursiops truncatus (Montagu, 1821) [72], and the larger false killer whale Pseudorca crassidens Owen, 1846 [73] with a much less tapered rostrum, lower tooth count, more robust dentition, larger temporal fossa and higher cranial crests [74][75][76]. The significantly greater body size of M. ukupachai combined with more powerful bites revealed by its cranial and dental features (e.g. [77]) suggest that it was capable of preying upon larger prey items, positioned higher along the local trophic chains, as observed for P. crassidens [78,79]. Such an ability to broaden the range of prey sizes constitutes a key parameter for ecological segregation among sympatric modern odontocetes [68][69][70]; it could be tested in fossil species via stable isotope analyses (e.g. [80,81]).

Conclusion
Macrosqualodelphis ukupachai is a new species of the extinct platanistoid family Squalodelphinidae based on a well-preserved partial skeleton collected from the Early Miocene (ca 19-18 Ma) fossiliferous beds of the Chilcatay Fm outcropping in the Western Ica Valley (southern coast of Peru). The age of this skeleton is further constrained via 40 Ar/ 39 Ar dating of a local volcanic ash layer to 18.78 ± 0.08 Ma (early Burdigalian).
Our phylogenetic analysis supports the referral of M. ukupachai to the monophyletic family Squalodelphinidae, of which it constitutes the earliest diverging lineage.
The main distinctive character of M. ukupachai is its large size: its estimated TBL is approximately 3.5 m, significantly larger than all other known squalodelphinids, including N. vanbenedeni (2.5 m) and H. raimondii (2.0 m), both also found in the Chilcatay Fm. Combined with cranial and dental features (robust rostrum less tapered than in other squalodelphinids, large temporal fossa, prominent nuchal and temporal crests, and more robust teeth), the large body size of M. ukupachai suggests that this squalodelphinid was able to prey upon larger prey items. Consequently, M. ukupachai would have been positioned higher along the local trophic chain than the roughly contemporaneous N. vanbenedeni and H. raimondii. Therefore, it is suggested that the squalodelphinid diversity, both locally and worldwide, could be related to their partition into different dietary niches, as is observed in the extant delphinids.
Step 9 is the one with the lowest Ca/K ratio and the youngest age. Other steps with a high Ca/K ratio have been considered as a result of alteration and not taken into account in the age calculation. All uncertainties are given as 2σ . (b) Isochron of the CHILC-AT1 biotite crystals, considering only the isochemical heating steps 4-9. The age obtained is 18.70 ± 0.13 Ma, with an MSWD of 7.3. The sizes of the ellipses representing points are the 2σ uncertainties.