A unique Cretaceous–Paleogene lineage of piranha-jawed pycnodont fishes

The extinct group of the Pycnodontiformes is one of the most characteristic components of the Mesozoic and early Cenozoic fish faunas. These ray-finned fishes, which underwent an explosive morphological diversification during the Late Cretaceous, are generally regarded as typical shell-crushers. Here we report unusual cutting-type dentitions from the Paleogene of Morocco which are assigned to a new genus of highly specialized pycnodont fish. This peculiar taxon represents the last member of a new, previously undetected 40-million-year lineage (Serrasalmimidae fam. nov., including two other new genera and Polygyrodus White, 1927) ranging back to the early Late Cretaceous and leading to exclusively carnivorous predatory forms, unique and unexpected among pycnodonts. Our discovery indicates that latest Cretaceous–earliest Paleogene pycnodonts occupied more diverse trophic niches than previously thought, taking advantage of the apparition of new prey types in the changing marine ecosystems of this time interval. The evolutionary sequence of trophic specialization characterizing this new group of pycnodontiforms is strikingly similar to that observed within serrasalmid characiforms, from seed- and fruit-eating pacus to flesh-eating piranhas.

. Under special lighting conditions (i.e., strong, laterally oriented light beam), this polished section shows thin and straight tubules penetrating from the dentine into the acrodin (Supplementary Fig. S6). Several incremental growth lines can also be observed in the dentine layer ( Supplementary Fig. S6). The posterior half of this tooth ( Supplementary Fig. S5c) was etched a few minutes with HCl and subsequently observed under scanning electron microscope (SEM). The acrodin layer has an inner portion consisting of fibres more or less perpendicular to the dentine layer. The outer portion of the acrodin layer consists of woven fibre bundles which become densely arranged approaching to the crown surface ( Supplementary Fig. S5d-f). The crown surface shows strongly woven fibres (Supplementary preserved (all labiolingually compressed teeth). The coronoid process is relatively low, with a short neck and a dorsal border markedly rounded. In dorsal view, it is strongly mediolaterally compressed ( Supplementary Fig. S2h). The ventral margin is partly broken. The symphysial region is not preserved. Fig. S2k-n) is a nearly complete right prearticular (63 mm in preserved length) with 11 teeth preserved (all labiolingually compressed teeth). An additional tooth is broken at the crown/root boundary in the anteriormost region of the row. A small, subconical vestigial tooth is also present on the medial wall of the bone, 4 mm below the third anterior tooth (as preserved) of the row. This tooth shows a small vertical wear facet on its lingual face. The symphysial region and the coronoid process are not preserved. Fig. S2o) is a nearly complete left prearticular (66 mm in preserved length) with 13 teeth preserved (12 labiolingually compressed teeth and one anterior subconical tooth). This specimen is still included in matrix and seen in lingual view.

MHNM KHG 156 (Supplementary
The symphysial region and the coronoid process are not preserved. Fig. S2d, e) is a nearly complete left prearticular (40 mm in preserved length) with 10 teeth preserved (nine labiolingually compressed teeth and one anterior subconical tooth). The ventral margin is preserved. The symphysial region and the coronoid process are not preserved. The lateral face of the prearticular shows a low coronoid ridge and a shallow longitudinal depression for the insertion of the angular ventrolaterally.

MHNM KHG 155 (Supplementary
MHNM KHG 152 (Figs 1c, d, 2f and Supplementary Fig. S2i, j) is a nearly complete left prearticular (86 mm in preserved length) lacking its ventral region and most of the coronoid process. The tooth row as well as the symphysial region are well preserved. 13 teeth are preserved (12 labiolingually compressed teeth and one anterior subconical tooth), and at least two additional teeth are broken at the crown/root boundary in the anteriormost region of the row. The highest teeth are located in the middle part of the row and are 8 mm high.

Revised diagnosis
Large-sized serrasalmimid pycnodontiform fish characterized by a crushing-type dentition and distinguished by the following unique combination of characters (autapomorphies marked with an asterisk): 1) teeth elevated, more or less conical, rugose, and bearing a basal cingulum (mammiform teeth)*; 2) teeth with a relatively narrow size range; 3) teeth of main rows slightly longer than wide; 4) vomerine dentition with three main rows separated by smaller, irregularly arranged teeth; 5) lateral rows more extended posteriorly that the medial row*; 6) medial row with teeth decreasing in size posteriorly*; 7) prearticular dentition elongated and relatively narrow; 8) prearticular dentition showing four or five badly defined rows of irregularly arranged teeth*; 9) short symphysis, about half the prearticular length; 10) coronoid process low and relatively broad anteroposteriorly.

Remark.
As originally noted by Woodward [39], the dentary teeth seem to be preserved in NHMUK PV P 11157, a nearly complete lower dentition from the English Chalk still embedded in matrix ( Fig. 2b and Supplementary Fig. S3p). The supposed dentary teeth (four on the right side and three on the left side) are slightly larger than the adjoining teeth present in the anteriormost portion of the specimen. However, their morphology (i.e., occlusal contour rounded to oval, crown elevated, wide basal cingulum) is very similar to that of the prearticular teeth, making their identification as dentary teeth uncertain. Therefore, this tentative observation is not taken into account in our phylogenetic analysis (see part E).   [45][46][47][48]. Some definitions and state orderings have been altered in order to fit with the new set of taxa. Characters 105-111 were added in order to address the conditions in the new taxa. The data were analysed using PAUP* 4.0b10 [49]. The tree resulting from this analysis is shown in Supplementary Fig. S4, with uniquely derived character states given along the branches corresponding to the new taxa.

Tree description:
Unrooted tree(s) rooted using outgroup method Prefrontal bones:

Data matrix:
See Supplementary Table S2.

Part F. Carnivorous specialization of the Serrasalmimidae
The serrasalmimid dentitions show a spectacular carnivorous specialization, which is most There may be one or two wear facets (of slightly different inclination) in Serrasalmimus teeth. Several specimens show distinct (vertical and oblique) wear striae generated by opposed shearing crests that illustrate the jaw movements during tooth-to-tooth contact ( Supplementary Fig. S2n); -Presence of a linguobasal cingulum in prearticular teeth, which has the function to protect the gum from food or tooth contact during bite and occlusion, as seen in mammals (e.g., refs [50,51]). The functional role of this cingulum is indicated by the extension, in mature individuals, of the shearing wear facet onto it (cingulum truncated by the wear) ( Supplementary Fig. S2n); -Short symphysis of the prearticular (Fig. 4a).

Dalatiidae (Chondrichthyes: Squaliformes)
Among cartilaginous fishes, a similar foraging strategy is known among the Dalatiidae, especially in kitefin and cookiecutter sharks, Dalatias and Isistius, respectively. These dwarf to medium-sized specialized sharks have powerful jaws with a lower dentition consisting of a single functional row of large, labiolingually compressed, triangular teeth [58], roughly similar to that of piranhas and Serrasalmimus (Supplementary Fig. S8). Teeth are imbricated and show an overlapping arrangement [59], like in serrasalmimids. However, the dignathic heterodonty is strongly developed and the upper jaw of these sharks has a clutching function. Cookiecutter sharks prey mostly upon small deep-water cephalopods and fishes, but are also known to regularly cut out flesh plugs from the body or fins of large pelagic fishes and marine mammals [60,61].

Part H. Comments on the taxonomic assignment of Serrasalmimus
It worth noting that the dentition of Serrasalmimus misleadingly resembles to some extent to those found in some reptiles. Thus, Serrasalmimus could be erroneously misinterpreted as a squamate. However, the various features which could be regarded as squamate-like characters correspond to superficial convergence. It is true that some squamates, such as teiid scincomorph lizards, roughly show a similar tooth crown morphology (i.e., bicuspid teeth). However, these teeth are pleurodont, spaced, and relatively small (unlike in Serrasalmimus) [62]. In squamates with a cutting dentition (e.g., agamid lizards), the upper dentition is labial to the lower dentition when mouth closed (unlike in Serrasalmimus) [63,64], and the organization of the enamel layer (i.e., parallel rods running out towards the tooth surface in the agamid spiny-tailed lizard Uromastyx [63]) is distinct from that of the acrodin layer (i.e., woven fibre bundles densely arranged) observed in Serrasalmimus teeth. The tooth-bearing bone of the lower jaw of squamates, corresponding to the dentary, shows a typical subdental self on its lingual surface and some foramina of its labial surface (unlike in Serrasalmimus) [65,66]. The posterior margin of the dentary of squamates shows a typical surangular sinus located between two (i.e., surangular and angular) processes [66] (all absent in Serrasalmimus). The shape of the coronoid process of the dentary of squamates (including amphisbaenians) is relatively small and pointed [65][66][67] . 4b).
Most of the synapomorphic characters defining the Pycnodontiformes in previous studies [5,8,18] cannot be observed in the material of Serrasalmimus. However, it is worth noting that this is due to the lack of the corresponding skeletal elements in our material and not to a "true" absence of these characters. Since vomers and prearticulars are the only skeletal elements known in serrasalmimids, the only synapomorphic characters that can be observed are "teeth on vomer and prearticular arranged in more or less regular rows" and "long, stout In addition, the presence of a vestigial tooth in the prearticular MNHM KHG 157 is the remnant of one of the medial tooth rows lost during serrasalmimid evolution. This unique combination of characters makes Serrasalmimus an unusual but unambiguous pycnodont fish which is placed as the closest relative of the Late Cretaceous genus Eoserrasalmimus.
Like in durophagous pycnodontiforms, Serrasalmimus shows a typical woven pattern of acrodin bundles in the outer tooth layer (Supplementary Fig. S5g). This condition differs from crushing teeth of teleosts with an outer layer constituted of parallel fibre bundles [8].
Acrodin canals of type A (sensu Ørvig 1978 [73]) are present in Serrasalmimus ( Supplementary   Fig. S6). Ørvig [73] defined these canals as follows: "the acrodin canals of type A are thin and as a rule fairly straight. They penetrate for varying distances into the hard tissue from its basal boundary, sometimes reaching almost to its external surface, and they frequently lie in direct continuity with the dentinal tubules of the adjoining dentine". This type is mainly present in pycnodontiforms (and in basal actinopterygians such as palaeonisciforms and polypteriforms) (see ref.
[8], fig. 52 and ref. [73], fig. 63 for examples of pycnodont tooth sections showing a condition similar to that observed in the Serrasalmimus tooth section), but it seems to be absent in teleostean fishes [8,73]. In the latter, the acrodin layer is generally divided in two sublayers, and thus shows two co-occurring types of canals: 1) straight and parallel canals (acrodin canals of type C sensu Ørvig 1978 [73]), which are restricted to the external part of the layer, and 2) sinuous, irregularly arranged canals (acrodin canals of type B sensu Ørvig 1978 [73]), which are located in the internal part of the layer (see ref.
Serrasalmimus is described on the basis of isolated tooth-bearing elements that are identified as vomer and prearticular bones. The upper jaw bone cannot be interpreted as fused premaxillae because of the lack of anterior ascending processes. In lateral view, the vomer of Serrasalmimus shows a well-developed vertical oral border, with a slight anteroposterior ridge of the bone present above the tooth row ( Supplementary Fig. S9a, b). This is very similar to the condition observed in some pycnodonts (e.g., Gyrodus, Acrotemnus [10,81], R.V. pers. obs.; Supplementary Fig. S9c-f). In most teleosteans, the vomer is unpaired and bears tooth whose location show a wide array of patterns. In most of these fishes, however, the vomer is very shallow, without lateral vertical border and without a dorsal medial crest (e.g., the osteoglossiform Scleropages formosus [82]; Supplementary Fig. S9g, h). In anguilliforms, the vomer is fused with the ethmoid complex to form the ethmovomer (e.g., Moringua edwardsi [83]; Supplementary Fig. S9i, j). This ossification is stout and bears very different kind of teeth depending of the species. Although this massive ossification might recall the vomer of serrasamimids, its organization is very different from the vomer of Serrasalmimus because it forms the tip of the snout including its dorsal side, and consequently it never bears a medial dorsal crest that inserts within other bones as in pycnodonts. Among anguilliform fishes, some pike congers such as Muraenesox and Cynoponticus (Muraenesocidae) have a vomer bearing large, triangular, laterally compressed teeth [84]. However, these principal vomerine teeth are arranged on a single row located medially and do not occlude with any teeth of the lower jaw.
In addition to the main medial row, the lateral margins of the vomer bear a row of small teeth.
This condition is clearly distinct from that found in Serrasalmimus.
In most non-teleost actinopterygians, the vomer is a paired ossification. However, there are a few exceptions, such as in some pachycormiforms. Protosphyraena is a Cretaceous pachycormiform with laterally compressed teeth and with an unpaired toothed rostrodermethmoid [85][86][87]. Here again, the tooth attachment and the mode of occlusion of the jaw, as well as the very different shape of the unpaired toothed element, which consists of a vomer bearing one pair of large paramedial teeth fused to a cylindrical rostrodermethmoid in Protosphyraena ( Supplementary Fig. S9k, l), prevent any phylogenetic connections of the latter with Eoserrasalmimus and Serrasalmimus.
The lower jaw bone of Serrasalmimus cannot be interpreted as a dentary because of the features of the symphysis. The symphysial articulation is deep and rugose, like in the prearticular of most pycnodonts [7,8,72] (e.g., Phacodus; R.V. pers. obs.), but unlike in most teleostean fishes, in which it is often marked by a series of diagonal ridges [88,89]. The welldeveloped coronoid process, arising from the posterolateral side, is stout and shows a rounded shape in lateral view. This condition is found in many pycnodonts [7,8,18,72] whereas the coronoid process of teleostean dentaries is not as developed [88]. The lower jaw elements of Eoserrasalmimus and Serrasalmimus show superficial resemblance to jaw bones of saurodontids (Cretaceous ichthyodectiforms), such as pointed teeth with cutting edges, a deep symphysis and a rounded coronoid process posteriorly located. However, the ankylothecodontlike tooth attachment, the absence of a well-defined fossa on the medial side of the mandible near its anterior extremity and the mode of occlusion indicated by the wear facet on teeth in Eoserrasalmimus and Serrasalmimus are incompatible with saurodontid features [90,91].
Posteriorly, the lower jaw bone of Serrasalmimus shows no articulation fossa, unlike teleostean dentaries. The latter show a deep sinus for the articulation with the articular [88,92].
Interestingly, upper and lower jaw bones of Serrasalmimus are both devoid of teeth in their anteriormost (frontal) portion, a feature that is characteristic of the vomerine and prearticular dentitions of pycnodonts. In the anteriormost portion of the jaws, upper and lower prehensile teeth were borne by the premaxillary and dentary bones, respectively, which are here unpreserved. Lastly, the lateral surface of lower jaw bone of Serrasalmimus shows no pores or grooves, unlike in teleostean dentaries. In the latter, these openings are related to the mandibular sensory canal (e.g., [89,91,92]). In pycnodonts, the mandibular sensory canal seems to run along the entire length of the jaw, but it has never been specifically described in the prearticular bone