The oldest ceratosaurian (Dinosauria: Theropoda), from the Lower Jurassic of Italy, sheds light on the evolution of the three-fingered hand of birds

The new find, in the context of Early Jurassic neotheropods

Skeletal remains of theropod dinosaurs are extremely rare in the Lower Jurassic and most reports are of only fragmentary remains (Benton, Martill & Taylor, 1995; Owen, 1863; Woodward, 1908; Andrews, 1921; Cuny & Galton, 1993; Delsate & Ezcurra, 2014). Moreover, ceratosaurian-grade taxa are absent until Middle Jurassic times (Maganuco et al., 2007; Pol & Rauhut, 2012), with one exception from the Pliensbachian–Toarcian of Northern Africa (Allain et al., 2007). This paucity of skeletal remains results in a considerable gap in our knowledge of these animals at a time when theropods were diversifying rapidly in the aftermath of the Triassic–Jurassic mass extinction event, as it is proven by the rich and worldwide distributed ichnofossil record (Delsate & Ezcurra, 2014, and references therein).

In Europe, theropod remains are reported from Hettangian times and are mostly non-diagnostic at generic level: Scotland (Benton, Martill & Taylor, 1995), England (Owen, 1863; Woodward, 1908; Andrews, 1921), France (Cuny & Galton, 1993), and Luxembourg (Delsate & Ezcurra, 2014). Two species of the genus Sarcosaurus have been reported from the Hettangian of England, S. woodi from Barrow upon Soar, Leicestershire, based on an isolated pelvis, vertebra, and proximal femur (BMNH 4840/1), and S. andrewsi (Huene, 1932), based on a partial tibia (NHMUK R3542) (see also Woodward, 1908). The neotheropod Dracoraptor hanigani, from the Hettangian of Wales, has been recently described by Martill et al. (2016) on the basis of a 40% complete skeleton including cranial and postcranial material.

In the rest of the world, the most famous Early Jurassic theropod is certainly Dilophosaurus wetherilli from the Hettangian of Arizona (Welles, 1954, 1984), which is known from several specimens. Other relevant taxa are Sinosaurus (=“Dilophosaurus” sinensis) from the Hettangian–Sinemurian of China (Hu, 1993), Coelophysis rhodesiensis from the Hettangian–Pliensbachian of South Africa and Zimbabwe (Raath, 1990), Dracovenator from the Hettangian of South Africa (Yates, 2005), Cryolophosaurus from the Early Jurassic (?Sinemurian–Pliensbachian) of Antarctica (Hammer & Hickerson, 1994), Podokesaurus from the Pliensbachian to Toarcian of Massachussetts (Talbot, 1911), Segisaurus from the Pliensbachian to Toarcian of Arizona (Carrano, Hutchinson & Sampson, 2005), “Syntarsus” kayentakatae from the Hettangian of Arizona (Rowe, 1989), and Berberosaurus from the Toarcian of Morocco (Allain et al., 2007). We do not take into consideration the enigmatic genus Eshanosaurus from the Lower Jurassic of China, tentatively dated as Hettangian (Xu, Zhao & Clark, 2001), pending correct identification and reliably dating, as this purported therizinosaurian coelurosaur might be a sauropodomorph as well.

In this context, the discovery of a new specimen from the Sinemurian of Italy is extremely relevant as it is among the oldest Jurassic theropods, it is larger than all other pre-Aalenian theropods (see Skeletal reconstruction and body size section, below) and it improves our knowledge on some of the macroevolutionary patterns that would have characterized the evolution of Theropoda during the Jurassic. It also represents the first dinosaur skeleton from the Italian Alps, the first of Jurassic age, and the second theropod skeleton found in Italy after Scipionyx samniticus (Dal Sasso & Signore, 1998; Dal Sasso & Maganuco, 2011).

The discovery of the specimen here described was accidental (for a more detailed account, see Dal Sasso, 2004). In the summer of 1996, Angelo Zanella, fossil amateur and collaborator of the Museo di Storia Naturale di Milano (MSNM), spotted some bones emerging from large blocks of rock in a huge quarry located in the Alpine foothills, at the Swiss–Italian border near Saltrio, less than 80 km north of Milan (Varese Province, Lombardy). Mr. Zanella reported the bones to the MSNM, which arranged a rapid prospection and recovered more remains. The research was difficult because the explosives used for industrial quarrying had blown up the fossil-bearing layer and had broken it into hundreds of pieces. In fact, the Saltrio quarry is active since the 15th century as one of the finest sites of marble production, and the “Saltrio Stone” provided high quality matter during the building of famous Italian monuments, such as the Scala Theatre in Milan, and the Mole Antonelliana in Turin.

In 1999, after 1,800 h of chemical preparation in the Laboratory of the MSNM, 132 remains were extracted from three main blocks. Although fragmentary, jaw fragments, one tooth, rib remains, pectoral and limb bones were resulted to be part of a large theropod dinosaur. The Saltrio theropod (MSNM V3664) became popular by the name “saltriosauro” and so it was reported (Dal Sasso, 2001a) and preliminarily described (Dal Sasso, 2001b, 2004). Actually, even though sometimes latinized (Dalla Vecchia, 2001), any pseudo-scientific name given to the specimen in the past is a nomen nudum, not valid because its erection did not follow the International Commission on Zoological Nomenclature (ICZN) rules (i.e., no diagnosis, neither accession number were provided in the publication erecting that name): that is one of the aims of the present contribution.

Fossil preparation

Removal of the fossil bones from the hard dolomitic matrix took more than 1 year (Dal Sasso, 2001b, 2004; Dal Sasso, Magnoni & Fogliazza, 2001). The methods used were a combination of mechanical preparation and controlled chemical preparation. Once the largest portions of matrix devoid of bones were cut away, the fossiliferous blocks were repeatedly immersed in a water solution of formic acid (5%) previously saturated with calcium triphosphate, then washed under abundant water current, then dried up, and the gradually surfacing bone was protected with an ethyl methacrylate co-polymer (Paraloid B72). This cycle involved about half a ton of limestone and took about 1,800 h.

Material

A total number of 132 bone pieces were recovered in close association, all clearly belonging to a single individual (except for one tooth and one jaw fragment, pertaining to a bony fish). The material consists of: 35 determinable bones, representing the holotypic material and belonging to the right lower jaw, pectoral girdle, rib cage and forelimbs, right manus, right ankle, and metapodium; 29 partially determinable bone pieces (five cranio-mandibular fragments, four rib fragments; five coracoidal, five scapular, and three sternal fragments; four appendicular skeletal fragments, including three possibly ungual fragments); 68 totally indeterminate bone pieces, including 16 small fragments surfaced in situ and 52 very small fragments recovered during preparation.

Methods

Measurements of the bones were taken with a digital caliper and a goniometer. In the present paper, if not differently specified, length of a given fragmentary element indicates its maximum length, and its height or width or diameter were taken perpendicular to the maximum length.

Thin sections of the embedding sediment were made, in order to observe microfossils and study the sedimentology and the depositional environment; microfossils were also collected by sieving the residual fraction of the acid preparation process.

Two bone samples were obtained from selected skeletal elements, for paleohistological analysis. The samples were mounted on glass slides, polished down to obtain thin sections with a thickness of ∼50 μm, and analyzed under a Nikon Eclipse E600 POL mineralogical microscope. Photographs were taken with the gypsum plate inserted. Definition and terminology of lines of arrested growth (LAGs), external fundamental system (EFS), and vascular categorization follow Chinsamy (2005), Erickson (2005) and Francillon-Vieillot et al. (1990).

X-ray computed tomography (CT) of selected appendicular elements was performed at the Radiology Department of the Fondazione Ospedale Maggiore di Milano, with a Siemens Somatom Definition Dual Source CT Scanner. The best CT imaging was obtained with a bone algorithm on transverse (axial) slices, with scan parameters 120 kV, 120 mA, and slice thickness of 0.3 mm. Data was exported in DICOM format using eFilm (v. 1.5.3; Merge eFilm, Toronto, Canada). Analysis and post-processing were performed at Siemens Milano, Italy, with SyngoVia post-processing system using Region Growing Algorithm to segment volumes and see internal anatomical structures and vacuities.

We used photogrammetry to better show and study the mobility of the manus. 3D models of the bones were obtained with Agisoft PhotoScan, by processing 60 shots for each bone element. The photos were taken with a Nikon D90 camera, using a light box. The models were animated and rendered with Maxon Cinema 4D.

For the anatomical nomenclature, following Weishampel, Dodson & Osmólska (2004) we adopted the terminology of the Nomina Anatomica Veterinaria (World Association of Veterinary Anatomist (WAVA), 2005) and the Nomina Anatomica Avium (Baumel et al., 1993). Concerning the dental nomenclature, we followed the standardization established by Hendrickx, Mateus & Araújo (2015a).

Phylogenetic taxonomy

In this study, we adopted the following clade name definitions. Dinosauria: the least inclusive clade containing Megalosaurus bucklandii, Hylaeosaurus armatus, Plateosaurus engelhardti, and Iguanodon bernissartensis (emended). Saurischia: the most inclusive clade containing Allosaurus fragilis and Diplodocus longus but not I. bernissartensis. Theropoda: the most inclusive clade containing Allosaurus fragilis but not Plateosaurus engelhardti or Heterodontosaurus tuckii (Naish et al., in press). Neotheropoda: the least inclusive clade containing Allosaurus fragilis, Ceratosaurus nasicornis and Coelophysis bauri (emended). Coelophysoidea: the most inclusive clade containing Coelophysis bauri but not Allosaurus fragilis or Ceratosaurus nasicornis. Dilophosauridae: the most inclusive clade containing Dilophosaurus wetherilli but not Allosaurus fragilis, Coelophysis bauri, or Ceratosaurus nasicornis (new definition). Averostra: the least inclusive clade containing Vultur gryphus and Ceratosaurus nasicornis but not Coelophysis bauri (emended). Tetanurae: the most inclusive clade containing Vultur gryphus but not Ceratosaurus nasicornis (emended). Ceratosauria: the most inclusive clade containing Ceratosaurus nasicornis but not Vultur gryphus (emended). Neoceratosauria: the least inclusive clade containing Ceratosaurus nasicornis and Abelisaurus comahuensis (emended). Ceratosauridae: the most inclusive clade containing Ceratosaurus nasicornis but not Abelisaurus comahuensis or Noasaurus leali. Abelisauroidea: the least inclusive clade containing Abelisaurus comahuensis and Noasaurus leali.

Following Bristowe & Raath (2004), the binomial “Syntarsus rhodesiensis” is considered a junior synonym of Coelophysis rhodesiensis. The binomial “Syntarsus kayentakatae” is provisionally used for the Kayenta Formation coelophysid (Rowe, 1989), pending the formal definition of a genus name for the latter (see Bristowe & Raath, 2004; Tykoski & Rowe, 2004).

Phylogenetic analysis

The phylogenetic data set used for investigating the affinities of the new Italian theropod includes 87 operational taxonomic units scored for 1,781 morphological character statements (Data S1). Character statement definitions are based on Cau (2018). The data set was analyzed using maximum parsimony as tree search strategy. Parsimony analyses were performed using TNT version 1.5 (Goloboff, Farris & Nixon, 2008). Given the large size of the data set, the search strategy involved 100 “New Technology” search analyses using the default setting, followed by a series of “New Technology” search analyses exploring the tree islands found during the first round. Then, the analysis explored the tree islands recovered during the “New Technology” analysis rounds, using “Traditional Search” analysis and saving up to 99.999 shortest trees (default maximum storage in TNT). Nodal support was calculated saving all trees up to 10 steps longer than the shortest topologies found and using the “Bremer Supports” function of TNT.

Nomenclatural acts

The electronic version of this article in portable document format will represent a published work according to the ICZN, and hence the new names contained in the electronic version are effectively published under that Code from the electronic edition alone. This published work and the nomenclatural acts it contains have been registered in ZooBank, the online registration system for the ICZN. The ZooBank LSIDs (Life Science Identifiers) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix http://zoobank.org/. The LSID for this publication is:

LSID urn:lsid:zoobank.org:pub:DBF732EB-6D24-48D2-294 8E5E-1C83EB380FD2.

The online version of this work is archived and available from the following digital repositories: PeerJ, PubMed Central, and CLOCKSS.

Lithology and sedimentology

Kalin & Trumpy (1977) recognized four lithofacies in the Saltrio Fm., all mostly consisting of litho-bioclastic calcarenites rich in crinoid remains, gray-brown and sometimes greenish in color, with grainstone–packstone microfacies embedding ooliths, peloids, and bioclasts. Extraclasts, consisting of reworked penecontemporaneous shallow-water dolomitic and phosphatic grains eroded from the Triassic substratum, are also present.

Age

The Simenurian age of the Saltrio Fm. is well-supported by a hundred species of marine invertebrates, among which 19 ammonites are index fossils of that time (Sacchi Vialli, 1964; Jadoul et al., 2005). The stratigraphic position of specimen MSNM V3664 was confirmed in situ by the co-occurrence, in the bank embedding the bones, of the ammonite Paracoroniceras cf. gmuendense (Fig. 1F) and the nautiloid Cenoceras striatum (Fig. 1G) (V. Pieroni, 2017, personal communication), whose association is typical of the layers S3 and S5 (sensu Sacchi Vialli, 1964) of the Saltrio Fm., that is, of the bucklandi and semicostatum Zone. Of the two layers, according to the authors who investigated the Saltrio Fm. in the past decades (Sacchi Vialli, 1964; F. Jadoul, 2004, personal communication; Croce, 2005), the S3 is the only one containing glauconite as accessory mineral, therefore there is no doubt that the theropod bones were embebbed in the bucklandi Zone, which is then referable, more precisely, to the earliest portion of the early Sinemurian substage (199.3–197.5 Mya) (Ogg & Hinnov, 2012).

Depositional environment

In the Saltrio quarry, the sedimentary succession shows a deepening-upward trend, but it lacks frankly shallow-marine sedimentary structures (e.g., shoreface facies) at the base. In facts, the stratigraphic transition is from the unconformity on Upper Triassic deposits to dolomitic breccias with green marly matrix, which represent debris flow deposits, thus already subtidal conditions (Croce, 2005). In other words, the depositional environment of the Saltrio Fm. was a likely tectonic slope that connected differently subsiding areas. After long subaerial exposure, these areas became subject to intense rifting and sunk. Due to these tectonics, the shore facies were bypassed and a subtidal environment was established directly, with debris flow deposits supplied by active tectonic slopes (M. Croce, 2018, personal communication). The texture and irregular thickness of the Saltrio Fm., the sedimentological data, and the presence of normal-salinity marine biofacies in the bone-bearing layer, with abundant crinoids, outer-shelf lagenids, and benthic foraminifera, indicate that the depositional environment of the Saltrio theropod was a proximal slope or ramp, that is, an open subtidal zone reached by the effects of storm waves and with constant bottom currents, where re-sedimentation phenomena were frequent (Jadoul et al., 2005; Croce, 2005). A depth of some dozen of meters can be reasonably estimated (F. Berra, 2018, personal communication). The parautochthonous glauconite (sensu Amorosi, 1997) indicates intervals of reduced sedimentation, in sectors adjacent to the seafloor where the dinosaur carcass deposited (F. Berra, 2018, personal communication).

Encasing sediment

Specimen MSNM V3664 comes from the lower banks of the Saltrio Fm., which are characterized by abundant inclusions of glauconite (Fig. 1H), a green-colored iron potassium phyllosilicate which is considered a bathymetric indicator, as it originates typically in shallow marine depositional environments, during periods of slow rates of accumulation (Amorosi, 1997). Thin sections of the layer embedding the bones (Fig. 1J) show bioclastic packstone and grainstone, with abundant and sometimes large fragments of crinoids, echinoids, ostracods, brachiopods, bivalves, gastropods, and benthic foraminifers (F. Berra, 2018, personal observation). The skeletal grains are often rounded and sometimes micritized, which indicates the presence of continuous reworking bottom currents.

Taphonomical description

The blocks embedding the dinosaur, photographed during the acid preparation stages (Figs. 2 and 3), provided important taphonomical data, showing that the bones were: (1) laying in a single bedding plane and all disarticulated, albeit close one another; (2) not oriented but randomly scattered; (3) mostly broken into small pieces, but very rarely deformed by diagenesis. Of a hundred of specimens, only half a dozen of small and delicate bones have been compressed (two phalanges, rib ends, indeterminate bone laminae). Even the numerous scapular fragments, once reconnected, rendered a gentle continuous curvature, which is consistent with the shape of the left scapula naturally embracing the rib cage and with only the acromion taking a counter curve.

Long bones from both fore- and hindlimb show “coherent” anatomical proportions, which are consistent with the skeletal composition of one single individual, without any homologous overlapping element. No other vertebrate remains were found associated to this bone assemblage, except for one tooth and one jaw fragment pertaining to a bony fish. Likely, the Saltrio theropod fossilized almost in its entirety, but with some dispersal of body chunks in different clusters. This might explain why no vertebrae were found in blocks A and B, which contained mostly appendicular elements (about 85% of the total bone content). The complete lack of gastralia in the block that trapped the furcula, the pectoral girdle and the dorsal ribs, raises the suspect that the ventral dermal bones abandoned the carcass before it reached the sea bottom, when the decay gases caused the “explosion” of the abdomen and eventually its separation.

In block A—the largest and quantitatively most informative cluster of bones—the flattest bone (mostly scapular) fragments appear to be sorted as to floor the front side, and come to light contemporarily, during early preparation stages (Figs. 2A and 2B). In all likelihood, this apparent bedding plane represents the water-sediment paleosurface, thus the depositional succession must be imagined as upside down with respect to Figs. 2 and 3. In facts, the largest and most irregularly shaped bone (right humerus) covers the flooring fragments on the “back” side.

Furcula excluded, the bones of the pectoral girdle have suffered intense longitudinal and mosaic cracking (sensu Behrensmeyer, 1978) and moderate to intense abrasion (stage 1–2 of Boessenecker, Perry & Schmitt, 2014). Such a high degree of fragmentation, coupled with the taphonomical observations above listed, and the paleontological content and the sedimentology of the Saltrio calcarenites, suggests that the dinosaur carcass floated, entered a marine basin and sunk to the bottom not far from the shoreline, then decayed in shallow waters, remaining on the sea bottom for quite a long time before being completely buried. This hypothesis is further supported by a relevant taphonomical evidence: several bones of the Saltrio theropod suffered bioerosion, mostly by marine invertebrates.

Taphonomical interpretation

A map of the macroborings on the bones in situ (Figs. 2 and 3) shows that 27 of 30 marks faced the back (n = 15) and side (n = 12) directions, and only three marks faced the front of blocks A + B. This distribution confirms that the bedding layer was upside down with respect to Figs. 2 and 3, and that the bones of the Saltrio theropod remained exposed for most of their surface to bottom currents and scavengers, which easily rolled the elements with rounded cross-sections, in this case favoring the marks on multiple sides (e.g., humeri and metatarsals). The evaluation of the exposure time depends also on the estimated grazing and colonizing speed of the bone tissue, thus on the scavenging fauna (Boessenecker, Perry & Schmitt, 2014).

The identification of the tracemakers is beyond the aims of this study; however, it is worth to note that this is likely the first record of marine bioerosions on dinosaur bones. In turn, it is well-documented that whale falls at the sea floor can nourish subsequent communities of scavengers for several years (Smith & Baco, 2003), and there is evidence of the same processes in the fossil record of cetaceans (Dominici, Danise & Benvenuti, 2018), plesiosaurs (Kaim et al., 2008), and ichthyosaurs (Danise, Twitchett & Matts, 2014).

Similarly, necrophagy on the bones of the Saltrio theropod by a variety of taxa indicates that the dinosaur carcass remained exposed to the water-sediment interface for months, maybe years, long enough to being first defleshed by mobile scavengers, then colonized by a microbial community that spanned the bone–water interface, which in turn attracted slow-moving grazers and epibionts. The bones of the dinosaur were locally bioeroded by these opportunistic macroinvertebrates, furthermore fragmented, and partially abraded by the bottom currents and the sandblasting action of the calcarenites, which eventually covered them.

The fact that the main scavengers of the Saltrio theropod were benthic marine invertebrates is a further confirmation that the dinosaur carcass deposited on a well-oxygenated and well-illuminated sea bottom, in any case comprised within the photic zone, where the biotic activity was intense but, at the same time, the sedimentation rate was high enough to cover skeletal material before its complete destruction (Dominici, Danise & Benvenuti, 2018). In our material, this sequence of events (i.e., partial scavenging followed by burial and diagenesis) is best documented by a deep semilunate “bite,” produced along the shaft of metacarpal II (Fig. 3A). The gnawing action trenched a perfect semicircle; much later, the edge of the bite was splitted in two by subsequent collapse of the bone wall onto the hollow central cavity, caused by diagenetic pressure of the sediment that accumulated on top.

Skull

From a cranial element possibly comes a fragmentary bone with a very peculiar texture and high degree of vascularization (Figs. 5A and 5B). This bone is broken at any end, showing a T-shaped cross-section that at first glance recalls a vertebral transverse process with a deep and robust centrodiapophyseal lamina. However, the top of the T is perfectly flat and the two other bone surfaces are textured with fine ridges and pits, suggesting tight soft tissue attachments. This texture clearly differs from the parallel striations (i.e., muscle and ligament scars) seen on the vertebral processes (C. Dal Sasso & S. Maganuco, 2017, personal observation on Allosaurus fragilis MSNM V435). The internal structure also differs in being highly spongy rather than fibrolamellar, indicating a delicate, not robust structure. In addition, the purported centrodiapophyseal lamina widens toward its broken edge, suggesting a V-shaped branching or prosecution toward a wider portion of bone. One can hypothesize that this fragment was part of a cranial fenestra, but to relocate its anatomical position remains impossible.

Lower jaw

Three fragments that can be referred to the lower jaw have been recovered closely associated from block B. Besides their thin bony wall and finely parallel ridged texture, oriented rostrocaudally, the three fragments share complex grooved surfaces, reminiscent of the vascular grooves that usually run along the medial and internal sides of the lower jaw bones.

Splenial. The largest bone piece (Figs. 5C and 5E) preserves two other small fragments in tight sutural contact, respectively, with its dorsal and ventral margin; both sutures run restrocaudally, paralleling the finely ridged texture. The main fragment has a possibly medial surface missing the cortex and exposing the internal bone structure, a ventral sharp margin, oriented at 90°, and a flat ?lateral side that houses a longitudinal groove near its dorsal end. We think that this laminar element may be part of the middle portion of a right splenial, just caudal to the Meckelian foramen (absent in our fragment), where the splenial is clasped dorsally and ventrally by the caudal ends of the bifurcating dentary.

The second jaw fragment is much narrower dorsoventrally but preserves a sharp ?ventral margin with an angle of 90°, just like the previous fragment, which suggests it might be the rostral continuation of the same bone. In facts, the splenial of coelophysoid-grade theropods (including Dilophosaurus) is more elongate and rostrally tapering than that of tetanurans like Allosaurus. Interestingly, the splenial of Ceratosaurus nasicornis (C. Dal Sasso, 2017, personal observation on AMNH FR 27631- cast of the right lower jaw of USNM 4735) at mid-length displays a labioventral margin which is sharp-squared, highly similar to the margin of our fragments.

Prearticular. The third jaw fragment (Figs. 5F–5G) is here interpreted as a piece of the right prearticular, thanks to its very peculiar cross-section. The medial side is slightly convex and the lateral side is slightly concave, with the same curvature; the narrow ventrolateral and dorsal margins house a shallow groove each, whereas dorsolaterally a deep narrow groove enters the bone until the middle, giving its dorsal section a Y-shaped aspect. Such complex profile was used as a fingerprint to relocate the anatomical position of this bone fragment on complete theropod skulls and lower jaws. The best match occurred with the lower jaw of Allosaurus MOR 693 (C. Dal Sasso, 2004, personal observation). Carefully examining its disarticulated bones, we found an almost identical arrangement of grooves and processes at mid-caudal length of the right prearticular (Fig. 5H). That diagnostic cross-section, inferred by manual drawing, was later confirmed by unpublished CT data of the same specimen (E. Rayfield, 2016, personal communication; Fig. 5I). The prearticular of Ceratosaurus, “in so far as one may judge from the parts preserved it is very similar to that of Antrodemus” (Gilmore, 1920). In facts, the Saltrio fragment matches the prearticular of MOR 693 even in size (both are 35 mm tall), thus it is consistent with a lower jaw about 80 cm long (Fig. 5J), and a body length of a subadult Allosaurus fragilis (see below).

Yates (2005: fig. 5C, D) illustrates and describes a fragment “from near the posterior end” of the right prearticular of Dracovenator regenti that further confirms our interpretation: “the lateral surface bears two tall sharp-edged ridges, which extend across the length of the fragment, although their height decreases toward the posterior end. At the anterior end these ridges are closely spaced creating a deep, V-shaped sulcus between them. Toward the posterior end they diverge creating a broad, triangular fossa,” just like the dorsolateral groove in the Saltriovenator fragment (Fig. 5F). Moreover, “a thin, ventrally directed crest arises from the ventromedial margin. This creates a ventrolaterally facing, elongate fossa for the reception of the angular”: this is the ventrolateral groove seen in Fig. 5G.

Tooth

A single tooth (MSNM V3659) was found isolated within a small limestone block near block A. Considering the uniqueness of the find, we confidently refer this tooth to the same taxon represented by the assemblage of bones. The specimen, missing the root and the apex, is 43 mm long and 18 mm wide (thus the tooth crown height is 2.4 times the base length). The crown is typically ziphodont: elongate, pointed, distally recurved and laterally compressed, without basal constriction, and with denticulate carinae (Figs. 5L–5N).

Following Hendrickx, Mateus & Araújo (2015a), with a crown height ratio of 2.39 and a crown base ratio of 0.48 (Table 1), the tooth referred to Saltriovenator can be considered moderately elongated (category range 1.5–2.5) and moderately narrow (category range 0.5–0.6). At closer examination, the apicobasal curvature of the distal margin of the crown in labial/lingual view can be defined as marked, because the apex of the tooth is placed distally to the distal margin of the crown base, the mesial margin is clearly convex and the distal margin is concave.

The transverse cross-section of the crown is intermediate between lenticular and D-shaped types (sensu Hendrickx, Mateus & Araújo, 2015a), being moderately compressed but asymmetrical: both mesial and distal carinae face linguomesially and linguodistally, respectively, but the distal edge is sharper than the mesial one, and the labial side of the crown is more convex that the lingual one. Approaching the carinae, the crown edges remain convex either on the labial or on the lingual side, different from the condition seen in salinon-shaped and parlinon-shaped teeth (sensu Hendrickx, Mateus & Araújo, 2015a): the concave areas seen in Fig. 5L near the carinae are due to diagenetic crushing.

As in most basal theropods, the enamel surface texture is smooth without any wrinkles, also adjacent to the carinae, even at higher magnification (Figs. 5M–5N), and any ornamentation—such as flutes, longitudinal grooves or ridges, transverse, or marginal undulations—is absent.

The denticles are completely lost along the mesial carina, which is deformed, crushed, and eroded; small denticles (12 per 5 mm, i.e., 2.5 per mm) are preserved in a short medio-apical tract (7.3 mm long) along the less damaged distal carina (Figs. 5K, 5M and 5N). Following the morphological terms standardized by Hendrickx, Mateus & Araújo (2015a), the preserved denticles are chisel-shaped, apicobasally subrectangular, perpendicular to the carina, and symmetrically convex in the outline of the external margin; the interdenticular space is deep and narrow, the interdenticular slit—when not altered by erosion—seems shallow and triangular, without a lamina joining two neighboring denticles, and there are no interdenticular sulci (blood grooves).

The moderately compressed D-shaped cross-section and the lingually-sided carinae suggest a mesiolateral position for this tooth. In other words, it might be one of the first maxillary teeth from the upper right arcade, or one of the transitional dentary teeth from the lower left arcade. Comparison with the dentition of Early-Middle Jurassic theropod taxa allows to exclude affinity of the Saltrio tooth to known coelophysoids, which so far possess much smaller crowns (CH <15 mm) with minute denticles on the distal carina (>30 denticles per five mm; Buckley, 2009; Hendrickx & Mateus, 2014). Dilophosaurus is definitely more similar in denticle density (13 per 5 mm—C. Dal Sasso, 2004, personal observation on UCMP 37303), which in its turn is reported to be similar in Sinosaurus and Cryolophosaurus (Xing, 2012). On the other hand, the teeth of abelisaurids are usually low and weakly recurved, have a slightly concave, straight or convex distal profile, and irregular non-oriented enamel texture, and megalosaurid teeth are characterized by centrally-positioned carinae on both mesial and lateral crowns (Hendrickx, Mateus & Araujo, 2015b).

Affinities with the Ceratosauridae cannot be excluded, as the eroded lingual side in our specimen does not allow to verify the presence of the “diagnostic longitudinal grooves” described by Madsen & Welles (2000); however, in Saltriovenator it is absent “a wide concave area centrally positioned on the labial side of the crown,” mentioned as typical of this clade by Hendrickx, Mateus & Araujo (2015b). Similarity to allosaurid and metriacanthosaurid crowns is in the crown proportions, as well as denticle count (12 per mm—C. Dal Sasso, 2004, personal observation on Allosaurus MOR 693), but they differ in having apparent transverse undulations.

Axial skeleton

The axial skeleton of Saltriovenator zanellai is totally lost, except for a dozen of rib fragments, all coming from block A (Fig. 2). Small pieces of vertebral processes might be present among the indeterminate material.

Ribs. Based on the literature (Allain, 2005; Madsen, 1976; Madsen & Welles, 2000) and on mounted skeletons of Allosaurus fragilis (MSNM V435) and Tyrannosaurus rex (MSNM V3902), we tentatively refer four fragments to left dorsal ribs, and five fragments to right dorsal ribs (Figs. 5O–5Q). Our interpretation is based on the curvature of the preserved fragments, taking the keeled margin and the (usually laterodorsal) most flattened face as reference sides to orient the rib pieces, and assuming that the thicker cross-sections are proximal and the thinner-flatter ones are distal.

These fragments range from 28 to 18 mm in maximum diameter, and from 15 to 8 in minimum diameter, which is consistent with mid-distal shaft rib size in a theropod about 25% larger than the 6-m-long Allosaurus fragilis MSNM V435. The bulkiest rib fragment (30 mm in diameter) has a subtriangular cross-section, a concavo-convex caudal side, a cranial ridge and a robust tapering keel projected medially. By comparison with the cross-section of a Ceratosaurus rib figured by Madsen & Welles (2000: plate 19) and by direct comparison with MSNM V435 we refer this fragment to the proximal portion of a left dorsal rib.

A tiny fragment with similar cross-section, less than 10 mm in diameter and preserving a very sharp medial keel, emphasized by a deep groove running caudomedially along its base, likely belongs to the midshaft of a left cervical rib.

The preserved ribs do not show any pneumatic recess. Four very fragmentary rib pieces remain indeterminate.

Scapular girdle and forelimbs

This is the most represented portion of the appendicular skeleton of Saltriovenator zanellai, including the best preserved and most complete elements (right humerus, right manus). The bones of the scapular girdle and the two humeri come all from block A; the bones of the right manus and part of the left humerus have been extracted from block B.

Scapula. A total of 15 fragments of the left scapula have been recovered from block A (Fig. 2), and patiently reconnected into three main portions (Figs. 6A–6D). Although the broken edges of the three portions are not complementary, they can be referred to adjacent parts of the same bone thanks to similar size and craniocaudal diameter, flattened structure with continuous longitudinal ridged texture and continuous mediolateral curvature, lenticular cross-section with same cortical bone lamination and thickness, and macro-vacuolar aspect of the inner spongy bone. In addition, the presence of a longitudinal keel along a tapering thinner cranial edge, and of a thicker crest along a robust caudal edge, in all the three portions, allowed to orient them correctly (e.g., see the elongate drop-like cross-section in Madsen & Welles, 2000: p. 20). Reconstructed this way, the scapula of Saltriovenator results approximately two times longer than the humerus.

The distalmost portion of the left scapular blade is distinguished by its thinner cross-section, dorsally tapering in cranial and caudal view (Figs. 6B and 6D), dorsally diverging margins (Figs. 6A and 6C), and equally diverging surface texture. Possibly, five other small fragments showing similar flattening and texture (Fig. 2, sc label) are part of the same bone, or of the counterlateral element.

The costal (medial) surface of the scapular blade of Saltriovenator is flat. Only vascular pits and tracks, running on the medial surface of the acromion, are present. Below the neck, the scapula becomes much thicker (50 mm) and stouter along the caudal margin, whereas in cranial direction it tapers into the acromion. Only part of it is preserved in our specimen, with an axe-shaped fragment, that is, concave medially and convex laterally. Due to breakage, it is impossible to know how much was the acromion prominent, and if the scapulocoracoid was notched between acromion and coracoid (e.g., as it is in Dilophosaurus, unlike Ceratosaurus). On the lateral side of the scapula, a wide fossa proximal to the acromion and opposite to its medial concavity marks a powerful muscle attachment site, likely for the M. supracoracoideus (Burch, 2017). The maximum mediolateral diameter of the scapula (70 mm) is reached in the fragment that bears the glenoid face. The latter is intact, with an elongate D-shaped profile and a perfect line of contact (scapulocoracoid suture) with the glenoid of the right coracoid (Figs. 6E–6H), which fossilized close to it and to the right humerus (Fig. 2C). Given this, we refer the preserved scapular glenoid to the right scapula, albeit the fragments of the left scapula are by far most abundant. The well-preserved scapulocoracoid suture allows to restore the glenoid cavity, which appears directed mainly caudoventrally, without lateral exposition, as in basal neotheropods (Rauhut & Pol, 2017). The resulting glenoid angle, seen in lateral view, is a broad arc that measures about 110° (Fig. 6G). The scapular glenoid is wider (mediolaterally) than long (dorsoventrally), measuring 68 × 58 mm; its participation to the glenoid cavity is approximately equal to that of the coracoid. In medial and lateral view, the scapular glenoid shows a distinct outer lip that points caudally.

In the type specimen of Dilophosaurus wetherilli (C. Dal Sasso, 2004, personal observation on UCMP 37302) the scapular glenoid is squared rather than D-shaped, the angle formed by the glenoid with the articular surface for the coracoid is identical (140°), the glenoid angle is more open (125°), and the supraglenoid lip in lateral view is slightly more pronounced, almost hook-like, as in Ceratosaurus (Madsen & Welles, 2000) and Majungaurus (Carrano, 2007). In a subadult specimen of Allosaurus fragilis (C. Dal Sasso, 2004, personal observation on MOR 693) the scapular glenoid is subrectangular and much smaller than in Saltriovenator (49 mm long × 38 mm wide), as it is the coracoid glenoid (39 mm long × 40 mm wide), and they form a glenoid angle of 105°. In A. fragilis the scapula is dramatically narrower and more slender than in Saltriovenator, bladelike, with a dramatic proportional reduction of the coracoid. On the other hand, the scapula of Dilophosaurus wetherilli (Welles, 1984: fig. 25) has a subrectangular distal expansion, and a shaft with concave cranial and caudal edges.

Using the best preserved holotypic right scapula of Dilophosaurus to track a scaled reference silhouette in a tentative recomposition of the scapula of Saltriovenator, the latter fits a narrower, feebly cranially curved profile, without remarkable distal expansion: three important differences that make the scapula of Saltriovenator definitely more similar to those of Ceratosaurus dentisulcatus (Madsen & Welles, 2000: p. 20) and, secondarily, Eoabelisaurus (Pol & Rauhut, 2012).

Coracoid. A large thick, concavo-convex bone fragment (Figs. 6E–6J) is identified as the caudodorsal portion of the right coracoid, thanks to the preservation of the supracoracoid nerve foramen, the bicipital (also named lateral or coracoid) tubercle, the infraglenoid buttress, and the characteristic fossa than runs between these two prominent processes. As in several theropods, the bicipital tubercle is developed as a boss-like prominence: in some tetanurans, including Allosaurus (Madsen, 1976), this tubercle is extended along the lateral surface of the bone, forming a distinct ridge, but in Saltriovenator it is more prominent and forms a very elongate triangle, which is remiscent of the condition seen in several basal neotheropods, such as Coelophysis rhodesiensis (Raath, 1977), Zupaysaurus (Ezcurra & Cuny, 2007), and Dilophosaurus (see below), and different from the low rigde seen in Ceratosaurus (Madsen & Welles, 2000). The infraglenoid buttress seems taller and more pointed than the bicipital tubercle, but the latter is eroded, and the similar basal transverse diameter suggests that they were subequal in size, like in Dilophosaurus wetherilli (C. Dal Sasso, 2004, personal observation on UCMP 37302). In addition, in Saltriovenator the fossa is asymmetrical in the same way, with the bicipital side, which is subvertical, and the infraglenoid side oblique. In Allosaurus the infraglenoid-bicipital complex is much less pronounced, either in juvenile or adult specimens (C. Dal Sasso, 2004, personal observation on a growth series on loan to MOR from UUVP).

The coracoid of Saltriovenator lacks a lipped margin of the glenoid: the infraglenoid buttress forms a lip but it is directed laterally, not invading the glenoid margin. The supracoracoid nerve foramen continues in a groove, which is directed craniodorsally (dorsally in Sinosaurus—Hu, 1993; Ceratosaurus dentisulcatus—Madsen & Welles, 2000; Majungasaurus—Carrano, 2007), and still wide open at the broken end of the fragment. On the other hand, in Dilophosaurus wetherilli the supracoracoid nerve foramen widens in cranial direction, and with a more open angle, and does not show any groove or fossa (C. Dal Sasso, 2004, personal observation on UCMP 37302).

On the caudodorsal side of the bone, the coracoid glenoid is preserved as a smooth concave area, about 65 mm long (dorsoventrally) and 62 mm wide (mediolaterally). Laterally the glenoid is bordered by a rim, which extends in cranial direction from the infraglenoid buttress, and medially it becomes unclear because the bone cortex is missing. In facts, the nutrient foramen of the glenoid is widened by this lack of bone. The scapular face is deep and robust, remarkably similar to the “extremely thick contact with the scapula” described in Sinosaurus (Hu, 1993), also present in D. wetherilli (C. Dal Sasso, 2004, personal observation on UCMP 37302) and Segisaurus halli (Carrano, Hutchinson & Sampson, 2005: fig. 5).

The main coracoid fragment of Saltriovenator represents about one-third of the whole bone and preserves a good portion of the caudoventral margin, as shown by the ridge that borders the medial concavity. A second ridge marks the dorsomedial edge of the scapulocoracoid suture.

Four smaller bone pieces are referred to the flattened, fan-like portion of the coracoid as they show similar texture (fine parallel ridges), cross-section (concavo-convex bone, with one rounded margin), thickness (10–15 mm), and structure (thin-walled and finely spongy bone). Two of these fragments (rco2 and rco3 in Figs. 5E–5G) are likely the ventral continuation of the largest portion the right coracoid, as they were found overlapped onto it (Fig. 2B) and almost match each other along their fracture lines; the other ones, being thinner, are tentatively positioned more cranially (and might also belong to the left coracoid).

Based on preserved parts, the reconstructed coracoid appears proportionally smaller than expected from the size of the scapula, if compared to Dilophosaurus; the disproportion is minor in Ceratosaurus and Eoabelisaurus, and is the opposite in Allosaurus, due to its quite elongated scapula. Moreover, the coracoid of Saltriovenator is much longer parallel to the scapular suture than perpendicularly to it, and deep dorsoventrally (see depth of scapular facet in Table S1).

The caudoventral margin of the coracoid in Saltriovenator is gently rounded and lacks either a long pointed (e.g., Allosaurus) or distinctly hooked (e.g., Limusaurus) process, usually bound proximally by the infraglenoid buttress, which is present in Elaphrosaurus (Rauhut & Carrano, 2016), abelisaurids, and many averostrans, but not in Ceratosaurus (Rauhut, 2003). In facts, in Ceratosaurus the caudoventral margin is similarly curved, “short and bluntly rounded” (Tykoski & Rowe, 2004). In this aspect, the highest affinity is with the type specimens of Dilophosaurus wetherilli (C. Dal Sasso, 2004, personal observation on UCMP 37302) and Cryolophosaurus ellioti (Smith, Hammer & Makovicky, 2017; P. Makovicky, 2017, personal communication on FMNH PR 1821), both having the caudal margin of the coracoid regularly rounded with the same arch span. This suggests that also Saltriovenator had subelliptical rather than suboval coracoids, that is, it retained a rather plesiomorphic morphology, shared among basal saurischians.

Furcula. The furcula (Figs. 6K–6O) was extracted during acid treatment of block A, in close association to all other elements of the pectoral girdle (Fig. 2A). This bone cannot be misinterpreted as a gastral basket element because the two rami are stout, lack any longitudinal groove, and are medioventrally united in a clearly defined hypocleideum; furthermore, the complete right ramus terminates with a flat epicleideal facet (or epicleideum), which is typically spatulate and sulcated by ligamental scars, for articulation with the scapular acromion (Chure & Madsen, 1996; Carrano, Hutchinson & Sampson, 2005).

In the last two decades, furculae have been documented in nearly all but the most basal theropods (such as Herrerasaurus). The discovery of furculae in coelophysoids (Tykoski et al., 2002) has ruled out previous hypotheses on the phylogenetic position of Saltriovenator (Dal Sasso, 2001b), which were based on the idea that the fusion of the two clavicles occurred only in the Tetanurae. At present, the oldest known furculae belong to Coelophysis bauri and date back to the Late Triassic (Rinehart, Lucas & Hunt, 2007).

The furcula of Saltriovenator is V-shaped in ventral view (Fig. 6N) and U-shaped in cranial view (Fig. 6K) because, toward the symphysis, the dorsal margins of the two rami are concave, rather than straight. The preserved epicleideum is definitely twisted craniolaterally and expanded dorsoventrally at midlength of the facet, then it tapers to a pointed end. In cross-section (Fig. 6N), the two rami of the furcula are D-shaped in proximity to the symphysis, with the flat side facing dorsally; distally, the convex side develops a longitudinal ridge that eventually becomes the ventral edge of the epicleideum. The cross-section of the epicleideum is like a compressed D, with the flat side facing cranially. The hypocleideum of Saltriovenator projects caudoventrally seven to eight mm from the base of the clavicular rami, pointing to the left with a slight asymmetry. Interestingly, basal neotheropods such as Segisaurus and Dracoraptor lack (Carrano, Hutchinson & Sampson, 2005; Martill et al., 2016) or do not show (Chure & Madsen, 1996; Makovicky & Currie, 1998; Tykoski et al., 2002) prominent hypocleidea.

As commonly observed (Carrano, Hutchinson & Sampson, 2005), there is no trace of interclavicular suture between the two rami, which indicates a complete fusion. This was confirmed by CT analysis, which also excluded the presence of pneumatic openings and internal pneumatisation (Sereno et al., 2008), not to be confused with the wide medullary cavities visible especially inside the two rami.

With the method of measurement used by Nesbitt et al. (2009) we estimate an interclavicular angle of 140° for Saltriovenator zanellai. In coelophysoids, the furcula is variably U- or V-shaped and has an angle of 115–140°; the furcula is V-shaped also in allosauroids and ranges from 120° to 135° (Nesbitt et al., 2009). A “widely arched” furcula is present in Limusaurus (Xu et al., 2009).

Sternum. Remains of sternal plates were present in block A, partially mixed with other flat bone fragments of scapula and coracoid (Fig. 2B). In particular, we reconnected two fragments into a platelike, weakly curved bone margin (Figs. 5P–5R), which at first sight we hypothesized to be the distal end of the scapular blade, but eventually could not fit that position. This element shows a carinate (keeled) margin which is thinner than the thinnest preserved margin of the scapular blade, a similar spongy interior, but a different surface texture, which in facts is randomly oriented and finely pitted, well-vascularized, and it lacks the fine parallel striae that run all along the scapula. A couple of smaller fragments were recovered piled up on the former (Fig. 2A, st label) and share very similar shape and ornamentation. These features are also visible in the sternal plates MPG-KPC1 and 2, described by Sánchez-Hernández & Benton (2014: fig. 10) in Camarillasaurus cirugedae, a Cretaceous ceratosaurian from Spain. By comparison with the latter specimens, which are by far more complete, we suggest that our fragment may represent the caudolateral corner of the right sternal plate, and approximately one-eighth of the whole bone (Fig. 10B).

This would be the fourth time that sternal plates have been described in a ceratosaurian theropod, after Carnotaurus (Bonaparte, Novas & Coria, 1990), Limusaurus (Xu et al., 2009), and Camarillasaurus (Sánchez-Hernández & Benton, 2014).

Humerus. The humeri are the largest bones and the only paired elements known from both sides of Saltriovenator zanellai (excluding the clavicles that are fused into a furcula). The right humerus (Figs. 7A–7F) is by far more complete as it lacks only part of its head, and the adductor crest (=internal tuberosity of Madsen, 1976); the left humerus (Figs. 7G–7L) lacks not only the adductor crest, the extensor crest and part of the proximal diaphysis, but also the whole distal half. In both humeri, mainly on the fossae for the M. coracobrachialis, apparent subcircular marks are present; as written above (taphonomical section), these marks represent post-mortem damage (macroborings produced by marine invertebrates). The midshaft cut of the left humerus shows a wide open internal hollow, which occupies more than half of the diameter of the bone; CT analyses of the right humerus and right metatarsal II show that this relationship between cortex and medulla is present in the whole bone, even more marked towards and inside the epiphyses, as expected in the long bones of a theropod dinosaur.

The shaft torsion of the humerus of Saltriovenator, measured as the angle between the trasverse axes of proximal and distal ends when viewed proximally/distally, is about 74°. The main axis of the head is oriented transversally and collinear with the plane of the proximal expansion of the humerus, thus differing from some tetanurans (Benson & Xu, 2008) in which it forms a distinct acute angle with the main transversal plane of the proximal end. In proximal view, the head of the left humerus is more complete than the right one and appears ellipsoidal, expanded more lateromedially than proximodistally, that is, not inflated or dome-shaped, far from the globular shape seen in noasaurids and abelisaurids: thus remarkably plesiomorphic, as in Eoabelisaurus (Pol & Rauhut, 2012), and contrary to Limusaurus (Xu et al., 2009). The lateral tuberosity of the humerus is placed laterodistally to the head, at the level of the proximal end of the deltopectoral crest. It is well-developed, giving the lateral margin a straight profile in cranial and caudal view; in late-diverging ceratosaurians it is reduced, giving the humerus a broadly convex margin. The right humerus of Saltriovenator appears almost straight also in lateral and medial view, being just slightly bent in its distal third. Likely, the missing adductor crest was as slightly curved as the distal epiphysis, giving the whole bone a only moderately sigmoid shape, as in most Neoceratosauria (Tykoski & Rowe, 2004), and in some large-sized basal tetanuran taxa, such as Poekilopleuron (Allain & Chure, 2002; C. Dal Sasso & S. Maganuco, 2004, personal observation on plastotype MNHN 1897-2), Acrocanthosaurus, Szechuanosaurus (Gao, 1993), and Xuanhanosaurus (Novas, Aranciaga Rolando & Agnolín, 2016). It is also similar to Dilophosaurus (Welles, 1984), although in the latter the diaphysis is more slender and a little more bowed, with an arch which is continuous from the dorsal lamina to the entepicondylar crest (C. Dal Sasso, 2004, personal observation on UCMP 37302). In coelophysoids, the humerus shows a clearly sigmoid curvature, as well as torsion (Tykoski & Rowe, 2004). In Allosaurus (Madsen, 1976; C. Dal Sasso, 2004, personal observation on MOR 693) the humerus is markedly sigmoid, the diaphysis in craniocaudal view is narrow and bowed medially, and there is an increased torsion of the epiphyses, which are proportionally more enlarged.

The deltopectoral crest is the largest process of the humerus of Saltriovenator: proximally, it is not confluent with the humeral head, being separated from it by a shallow concavity that houses a thin extensor crest, like in Allosaurus (Madsen, 1976) and unlike Dilophosaurus (Welles, 1984). Distally, the deltopectoral crest becomes transversely inflated, and—remarkably and uniquely—it protrudes craniomedially for more than twice the midshaft diameter size, finally meeting the distal lamina abruptly, with an angle of 90°. The deltopectoral crest of Saltriovenator forms an angle of 50° with the plane of the distal condyles and it extends for more than 2/5 the humeral length, as in Dilophosaurus wetherilli (C. Dal Sasso, 2004, personal observation on UCMP 37303) and contrary to most tetanuran theropods, in which it extends in cranial direction. On the left humerus, the protruding end of the deltopectoral crest is much more pointed than in the right humerus, nearly hooked, being grown over the distal lamina—in this feature, it recalls Acrocanthosaurus (Currie & Carpenter, 2000). This condition, as well as the right-angled distal end, differs from the more gentle transition between the crest and the shaft seen in most theropods, including Ceratosaurus (Madsen & Welles, 2000) and Dilophosaurus (C. Dal Sasso, 2004, personal observation on UCMP 37303), and other taxa that possess a similarly protruding deltopectoral crest, such as Poekilopleuron (Allain & Chure, 2002), Szechuanosaurus (Gao, 1993), Torvosaurus (Galton & Jensen, 1979), and Australovenator (Novas, Aranciaga Rolando & Agnolín, 2016). A nearly perpendicular distal lamina of the deltopectoral crest can be seen only in Segisaurus (Carrano, Hutchinson & Sampson, 2005).

The proximodistal length of the remaining humeral shaft, between the deltopectoral crest and the distal condyles, is about five times the minimal shaft diameter. In this portion, the shaft does not bear any distinct tuber along the craniolateral surface, whereas on the caudolateral margin, at level of the distal lamina, an elongate scar for the M. latissimus dorsi is present, like in Majungasaurus (Carrano, 2007: fig. 3).

In cranial view, the humerus of Saltriovenator appears non-sigmoid, almost straight, similar to the holotype of Ceratosaurus dentisulcatus (Madsen & Welles, 2000: fig. B, D) but a little less bulky, with less pronounced, gently enlarged epiphyses; therefore it markedly differs from the midshaft-constricted holotype of C. magnicornis (Madsen & Welles, 2000: fig. A, C). In facts, in Saltriovenator the distally placed distal condyles are slightly less than twice larger than the diaphysis at its minimum transverse diameter.

In distal view, the partially eroded (or not completely ossified) condyles are only weakly convex (nor hemispherical, neither totally flattened) and subequal in size, the ectocondyle being slightly shorter in mediolateral direction, but deeper craniocaudally. The same condition is observed in Dilophosaurus wetherilli (C. Dal Sasso, 2004, personal observation on UCMP 37303) and Cryolophosaurus ellioti (Smith et al., 2007: fig. 14 C-D). The intercondylar sulcus is preserved only at its ends, it is shallower than in Dilophosaurus, mediolaterally narrow and slit-like in shape. The distal fossa is moderately developed. Although not hypertrophied, the ectepicondylar crest seems more developed than in Dilophosaurus and than the entepicondylar crest, but this may be an artifact of preservation, because the medial wall of the entocondyle is missing. In Allosaurus (Madsen, 1976; C. Dal Sasso, 2004, personal observation on MOR 693) the disproportions between the ecto- and the entocondyle increase, the latter becoming almost twice than the former in mediolateral length and much more compressed craniocaudally; the intercondylar groove markedly divides the two condyles and the ectepicondylar crest appears as robust as in Saltriovenator.

Manus: carpus and metacarpus

The preserved manual elements of Saltriovenator are one carpal, the right second metacarpal, the first phalanx and part of the second phalanx of the same finger, four phalanges that perfectly articulate each other when connected and are referred to the third finger, and the tip on an indeterminate ungual phalanx (Figs. 8–10 and 12). These bones were closely associated, although not in articulation, in the same limestone block, together with the right second metatarsal (Figs. 3H–3J).

The manual morphology of Saltriovenator demonstrates to be more similar to that of basal ceratosaurians (Ceratosaurus, Eoabelisaurus), and secondarily to that of dilophosaurids (Dilophosaurus) and allosauroids (Allosaurus, Sinraptor), than to the condition present in either late-diverging ceratosaurians (Limusaurus, abelisaurids), or in coelurosaurians. Interestingly, when rearticulated in neutral (straight) pose, the manual elements of both the second and third digit of Saltriovenator result in slightly side-bended fingers (Fig. 10C).

Carpal bone. At first sight this element (Figs. 13M–13R) appears as a little portion of a much larger bone, with a convex side slightly constricted medially, which recalls the articular condyles of a long bone. Actually, on the very eroded opposite side a small area of finished bone cortex is preserved, with a surface which is flat and parallels the other side, thus indicating that this bone was small and disk-like. Based on such shape, this element likely pertains to the carpus of Saltriovenator. No carpals have been found in the articulated hands of Ceratosaurus, Limusaurus, Majungasaurus, and Aucasaurus. Moreover, in Ceratosaurus USNM 4735 the whole arm is preserved in articulation, showing an apparent gap between the forearm and the metacarpals that indicates a non-ossified area (Carrano & Choiniere, 2016), as it is the case in Limusaurus. However, a large carpal 1 + 2 has been reported in a yet undescribed noasaurid from Niger (Sereno, Wilson & Conrad, 2004), a flattened discoidal bone was found associated with the proximal ends of metacarpals III and IV in Dilophosaurus (Welles, 1984), and only a single carpal is preserved in Eoabelisaurus (Pol & Rauhut, 2012), supporting the idea that Saltriovenator might have had a single carpal too. As it articulates to the metacarpal II, we consider the Saltriovenator’s carpal homologous to carpal 2 (Fig. 10B).

In sum, none of the known ceratosaurians has more than one ossified carpal bone: this leads us to suspect that most ceratosaurians had non-ossified, cartilaginous carpals. The presence of a single ossified carpal in Saltriovenator and Eoabelisaurus would represent an intermediate condition, before the complete loss of ossified carpals that can be observed in Limusaurus and in the Cretaceous ceratosaurians.

Metacarpal II. This bone from the right manus is almost complete and remarkably well-preserved (Figs. 8A–8F). Although metacarpal III was not recovered, its relationship with metacarpal II can be inferred from the anatomy of the latter: the distal end of metacarpal II unequivocally shows no evidence of a distal fusion with metacarpal III, and the contact for metacarpal III is limited to the lateral margin of the proximal portion of metacarpal lI. The same condition can be inferred for metacarpal I, whose contact projects proximomedially as a subrectangular facet at the apex of a low buttress, unlike tetanurans, and comparable to Ceratosaurus (Carrano & Choiniere, 2016: fig. 7), Dilophosaurus (Welles, 1984), and Eoabelisaurus (Pol & Rauhut, 2012). This condition was likely present—albeit not surely—also in Berberosaurus (Allain et al., 2007), pending confirmation (Carrano & Choiniere, 2016) that the preserved bone of Berberosaurus might instead represent a left metacarpal III (by affinity with the third metacarpal of Ceratosaurus: note that this interpretation is followed in the character score of the phylogenetic analysis used here).

The metacarpal II of Saltriovenator is peculiar in having dramatically enlarged and robust articulations, which give the bone an hourglass shape, especially in dorsal and palmar views, where the midshaft diameter reaches half the proximal transverse diameter; in lateral and medial views the shaft shows parallel margins, and its palmar side is flat, like in Ceratosaurus (Carrano & Choiniere, 2016), so that in cross-section it results subrectangular. As in Berberosaurus and Ceratosaurus, but not in Limusaurus (Xu et al., 2009), the proximal end is wider than the distal, and greatly expanded from the shaft. In this respect, as well as in size, the bone of Saltriovenator is more similar to that of an adult Allosaurus (Madsen, 1976) or Sinraptor (Currie & Zhao, 1993), than Dilophosaurus and coelophysoids (Welles, 1984), in which it is by far much more slender, with gently concave diaphyseal margins and moderately expanded epiphyses. The overall morphology of this bone suggests the presence of powerful manual muscles in Saltriovenator.

Although the dorsomedial process of the proximal articulation is missing, the original subtrapezoidal shape of the bone is still evident in proximal view, where the dorsal, palmar, and lateral sides are concave and three prominent bony processes, representing the contact limits for the adjacent metacarpals, mark the preserved articular apices. These processes continue in form of robust longitudinal ridges along the diaphysis, up to the midshaft. As in Ceratosaurus (Carrano & Choiniere, 2016), but less than in Berberosaurus (Allain et al., 2007), the most pronounced proximal concavity of Saltriovenator is the broad and triangular palmar flexor fossa, which occupies almost the entire width of the bone until the midshaft, and the second wide concavity is the dorsal extensor fossa. The similarly trapezoidal proximal articulation of Allosaurus (C. Dal Sasso, 2004, personal observation on MOR 693) differs in having a more excavated and more ventrally-facing contact for metacarpal III, a well-delimited articular facet for a carpal bone on the medioventral margin, and a depression for a second carpal on the proximolateral corner (both absent in Saltriovenator). In Dilophosaurus (Welles, 1984) the proximal end is also trapezoidal but less bulky, being compressed in dorsopalmar direction, having less pronounced dorsal and palmar concavities, and proximal ridges shorter and less prominent. In Saltriovenator the lateral margin of the proximal end meets the dorsal margin at an angle of about 80°, just like in Ceratosaurus (Carrano & Choiniere, 2016: fig.7E), forming a tab that, with metacarpal III in articulation, overlaps its proximomedial portion (Fig. 10C). A similar condition is also present in Dilophosaurus (Xu et al., 2009: fig. 2c; C. Dal Sasso, 2004, personal observation on UCMP 37302 and UCMP 37303).

In Saltriovenator the stout distal epiphysis terminates in a ginglymoid articulation which is asymmetrically and obliquely partitioned by the intercondylar sulcus into two condyles: a medial condyle, placed more proximally but deeper palmodorsally, centrally concave and with sharp medial edge; and a lateral condyle, more extended distally than the medial condyle, with convex (hemispherical) articulation and with rounded lateral edge. The same asymmetry and rotation of the distal condyles relative to the long axis of the bone (around 40°) is present in Berberosaurus (Allain et al., 2007) and Ceratosaurus (Carrano & Choiniere, 2016), and with minor degree (about 30°) in Dilophosaurus (Welles, 1984), Eoabelisaurus (Pol & Rauhut, 2012), Limusaurus (Xu et al., 2009), and some specimens of Coelophysis (Galton, 1971). It is absent in late-diverging abelisauroids as well as in most other theropods, including the tetanurans Allosaurus (Madsen, 1976) and Sinraptor (Currie & Zhao, 1993), which possess subequal, subparallel, and subvertical distal condyles, divided by a deeper intercondylar sulcus. In Saltriovenator both condyles are side-marked by well-developed fossae and pits for the collateral ligaments and extend in dorsal aspect, where the lateral condyle occupies almost two-thirds of the distal articulation. In lateral and medial view, the condylar surfaces further extend to the palmar side, tracing a semicircular excursion (as in Ceratosaurus) rather than three quarters of it (e.g., Dilophosaurus, Allosaurus); the medial condyle terminates in a pronounced “lip-like” projection directed proximopalmarly, a feature found only in Berberosaurus (Allain et al., 2007). In palmar view, the two condyles are divided by a shallow fossa for the flexor ligament. Continuous with the dorsal end of the intercondylar sulcus, a deep extensor ligament pit opens, bordered by an enlarged semicircular lip of bone that protrudes dorsolaterally over the condylar level. This lip delimits the dorsal excursion of the distal articular complex, functionally acting as a stop for the maximum extension/supination of the first phalanx of digit II (see below). The pit-and-lip complex is dramatically developed in Saltriovenator: more than in any other theropod, including Berberosaurus (Allain et al., 2007), Ceratosaurus (C. Dal Sasso & S. Maganuco, 2014, personal observation on USNM 4735), Eoabelisaurus (Pol & Rauhut, 2012), Sinraptor (Currie & Zhao, 1993), and Dilophosaurus. In the latter, the extensor ligament pit is not well-figured (Welles, 1984: fig. 37) and is described as six mm deep. Under direct observation (C. Dal Sasso, 2004, personal observation on UCMP 37302 and 37303), this pit results proportionally shallower and mediolaterally narrower than in Saltriovenator, and the lip reaches but not oversizes the dorsal limit of the condyles. On the other hand, in Sinraptor (Currie & Zhao, 1993: fig. 20) and Allosaurus (Madsen, 1976: pp. 43–45) the pit is wide and subcircular, extended towards the bone midshaft, and the lip is lower that the dorsal end of the distal condyles (Sinraptor) or absent (Allosaurus). Interestingly, the tetanurans Acrocanthosaurus (Currie & Carpenter, 2000), Szechuanosaurus zigongensis (Gao, 1993), Xuanhanosaurus (Dong, 1984), and Torvosaurus (Galton & Jensen, 1979) lack a protruding lip but retain a deep extensor ligament pit and asymmetrical distal condyles.

Hind limbs

Tarsus

Distal tarsals are not often recovered with specimens because of delayed ossification and small size (Currie & Zhao, 1993). Remarkably, the right distal tarsals III and IV are among the few complete bones of Saltriovenator, and are beautifully preserved in three dimensions (Fig. 11), unfused to the metatarsals (Figs. 11 and 12). They articulate pretty well with each other, and with the proximal epiphyses of the preserved metatarsals. The whole complex belongs to the right ankle of Saltriovenator, which is reconstructed in Fig. 14.

Distal tarsals are well-preserved in the type specimen of Dilophosaurus wetherilli, but inadequately described and not figured in the formal description (Welles, 1984). Also Dracoraptor nicely preserves these bones (Martill et al., 2016: fig. 27), showing remarkable—but not discussed—affinity with the former taxon. The distal tarsals III and IV of Dilophosaurus are similar to the ones of Saltriovenator, both in general shape and relative size, but are about one fourth smaller; in addition, they are proportionately less developed in craniocaudal direction and the contacts with proximal tarsals and metatarsals appear less marked (C. Dal Sasso, 2004, personal observation on UCMP 37302). On the other hand, either in absolute and relative size, the distal tarsals of Saltriovenator are not significantly different from those of the medium-large Jurassic tetanurans Allosaurus and Sinraptor. In shape, distal tarsal III of Saltriovenator is somewhat similar to the compact distal tarsal III of Allosaurus, whereas distal tarsal IV is more angled than in these tetanuran theropods, showing a more complex morphology.

Distal tarsal III. This is a rhomboid and nearly flattened element, with two sides tapering cranially into a thinner side with a sharp margin, and one side which is definitely thicker and bulky, displaced caudally; consequently, in lateral and medial view this bone has a markedly cuneiform profile (Figs. 11A–11F).

The distal tarsal III of Dilophosaurus (C. Dal Sasso, 2004, personal observation on UCMP 37302) possesses a more inflated mediocaudal portion and a cranial edge which is thicker, round-finished and not at all sharpened, as in more basal coelophysoids (e.g., Coelophysis rhodesiensis—Raath, 1990). In Allosaurus (C. Dal Sasso, 2004, personal observation on MOR 693) this bone is more flattened along all sides, the caudal process is pointed rather than rounded, and the proximal face is straight. Nevertheless, similar to Allosaurus, the lateral side of distal tarsal III of Saltriovenator bears two pointed apices, divided by a concavity, which fit exactly the wavy medial side of distal tarsal IV; in particular, the craniolateral pointed articulation inserts in the large medial notch of distal tarsal IV (Figs. 11M and 11N).

In distal view, the distal tarsal III of Saltriovenator is crossed at mid-length by a ridged articulation for metatarsal III, directed mediolaterally, which is absent in Allosaurus (MOR 693) and reduced to a feeble bump in Dilophosaurus (UCMP 37302); the articulation for metatarsal II is an elongate depression that runs along the distomedial margin; two adjacent notches mark the contact with metatarsal IV along the distolateral side; the largest notch bears two pits at its bottom. All these reference points, coupled with the undeformed condition of this bone, allow to place perfectly distal tarsal III on top center of metatarsal III, with marginal mediolateral lapping over (Fig. 14A).

Besides the features listed above, the distal tarsal III of Saltriovenator differs from Dilophosaurus in the distal side: in Dilophosaurus it is depressed in the middle, in Saltriovenator it is convex and ridged in the middle; in Dilophosaurus the contact for metatarsal II is limited to the craniomedial corner, in Saltriovenator the slit for metatarsal II develops along the bone edge in caudal direction for most of its length (C. Dal Sasso, 2004, personal observation on UCMP 37302).

Interestingly, the convex cranial margin shown by the distal tarsal III of Saltriovenator has been recently described in Powellvenator (Ezcurra, 2017), contra the concave margin of Coelophysis rhodesiensis (Raath, 1977), and Ceratosaurus (USNM 4735), or the straight margin of Dilophosaurus (UCMP 37302) and Dracoraptor (NMW 2015.10G.1a/b). However, Saltriovenator differs from Powellvenator in other aspects: the concave contact for the astragalus is limited to a mid-medial portion of the proximal surface, rather than to the whole medial third, and there is not any caudal depressed surface, absent also in most other basal neotheropods (Ezcurra, 2017).

Distal tarsal IV. This is a blocky element, almost twice thicker than distal tarsal III, with a complex subtrapezoid shape and with proximodistal thickness varying considerably (Figs. 11G–11L): in medial view, the distal surface appears convex with a central concavity that likely matched the proximal articular surface of metatarsal IV (not preserved in our specimen); the proximal surface appears weakly sigmoid, with a caudal convexity and a cranial concavity for the astragalus, like in Dilophosaurus (UCMP 37302) and unlike the uniformly flattened aspect described in Powellvenator (Ezcurra, 2017). The latter taxon is also very different in having a strongly convex, almost subcircular cranial margin, whereas the tarsal IV of Saltriovenator has a wing-like margin expanded craniolaterally. An almost equally developed wing-like convex margin is present in Dilophosaurus, Dracoraptor, and other basal neotheropods (Ezcurra, 2017), whereas in Ceratosaurus the craniolateral expansion is reduced by squared margins (Madsen & Welles, 2000: fig. 10).

In proximal and distal views, emphasized by the wing-like expansion, the tarsal IV of Saltriovenator narrows into a subrectangular caudomedial portion and displays a broad, equally subrectangular caudolateral notch, which was considered an unambiguous apomorphy of the Ceratosauria (sensu Tykoski & Rowe, 2004, a clade including coelophysoids and neoceratosaurians: note that our preferred phylogeny does not support coelophysoids in Ceratosauria). In facts, in Allosaurus (MOR 693) the caudal portion of the bone is less constricted and the notch is not a squared corner but a gentle concavity, and in Sinraptor (Currie & Zhao, 1993) there is not even a concavity. In proximal view, the mid-lateral portion of the wing-like expansion bears a facet, delimited caudally by a transverse shallow groove. This is likely the articular contact with the calcaneum (Fig. 14B) and cannot be compared to the marked lateral spur seen in some maniraptorans. In distal view, the pointed craniomedial end of the wing-like expansion bears a flat triangular facet that, with the ankle and foot bones re-articulated, contacts the proximal craniolateral end of metatarsal III (Figs. 14A and 14E). A similar process is present in several basal neotheropods, such as the “Padian’s Coelophysis” (UCMP 129618), Dracoraptor (NMW 2015.10G.1a/b), Dilophosaurus wetherilli (UCMP 37302), but also in Allosaurus (C. Dal Sasso, 2004, personal observation on MOR 693); it is absent in Powellvenator (Ezcurra, 2017), Coelophysis rhodesiensis (Raath, 1977: fig. 19), Segisaurus (UCMP 32101), and “Syntarsus” kayentakatae (Rowe, 1989).

Four pits open on the bone edges, three on the caudolateral side and one on the medial. The latter, and the two largest lateral pits, open at the bottom of notches that are homologous to the three pitted notches of Allosaurus described and figured by Madsen (1976: fig. 25); the medial pitted notch is for the craniolateral pointed articulation of distal tarsal III; the large caudolateral pitted notch is the articulation for the proximal head of metatarsal V. This notch is associated to the subrectangular “ceratosaurian” bone margin described above. In facts, it is remarkably squared in Ceratosaurus (Madsen & Welles, 2000: fig. 10); slightly less excavated in Dilophosaurus wetherilli (UCMP 37302), Dracoraptor (NMW 2015.10G.1a/b), Coelophysis rhodesiensis (Raath, 1977, 1990), Powellvenator (Ezcurra, 2017), feeble in Allosaurus (Madsen, 1976) and almost absent in Pandoravenator (Rauhut & Pol, 2017) and Sinraptor (Currie & Zhao, 1993).

Phylogenetic affinities of Saltriovenator

Our phylogenetic analysis (see also Data S1) found 624 shortest trees of 4,281 steps each (CI: 0.3069, RI: 0.5113). Exploration of the shortest topologies found shows that a large unresolved polytomy among the main dinosauriform lineages is due to the unstable position of Lewisuchus and Teleocrater, acting as “wildcards.” Once the two “wildcard” taxa have been pruned from the resulted topologies, the reduced strict consensus of the shortest trees found is well-resolved (Fig. 15) and places Saltriovenator as sister group of Berberosaurus along the basalmost ceratosaurian branch. Among theropods, the monophyly of Tetanurae, Averostra (the ceratosaurian-tetanuran clade excluding coelophysoids) and Neotheropoda is supported. Coelophysoid-grade theropods form a paraphyletic series leading to Averostra.

Unambiguous synapomorphies of the basal node of Ceratosauria (i.e., the least inclusive node containing Saltriovenator and other ceratosaurians) present in Saltriovenator zanellai are: a stout metacarpal II not longer than 5/2 its distal width; metacarpal II with distinctly shelf-like margin of collateral ligament fossa overhanging the fossae (particularly prominent along medial fossa); wide and prominent proximoventral processes in the manual phalanges, and reduction of the collateral ligament pits of the manual phalanges to shallow fossae not bordered by distinct lips. This node is also supported by the following synapomorphies present in Berberosaurus but not preserved in the Italian ceratosaurian: extensive pneumatization of the anterior presacral neural arches, and additional pneumatic foramen penetrating the posterior half of cervical centra.

A single unambiguous synapomorphy supports the sister-group relationship between Berberosaurus and Saltriovenator: the pronounced “lip-like” projection of the medial condyles of metacarpals II and III, which is directed proximopalmarly. This relatively weak support is explained by the limited anatomical overlap between the only two known specimens of Berberosaurus and Saltriovenator, restricted to the metacarpals.

Saltriovenator is excluded from Neoceratosauria (the ceratosaurid-abelisauroid clade) because it lacks: tooth crowns with flat or concave surfaces adjacent to the carinae; a relatively symmetrical and more equal development of the condyles of metacarpal II; a complete fusion of the tibiotarsus early in ontogeny; and a marked mediolateral expansion of the plantar margin of the proximal end of metatarsal III (“antarctometatarsalian” condition).

Enforcing Saltriovenator in Tetanurae (as originally suggested by Dal Sasso, 2001b), the shortest trees found are five steps longer than the unconstrained shortest trees: under this constraint, the Saltrio theropod is found in various alternative positions: as the basalmost tetanuran, as sister taxon of Szechuanosaurus zigongensis, as a coelurosaurian or within Megalosauridae. The latter two suboptimal scenarios are provisionally rejected also on stratigraphic ground, as they would imply several tetanuran branches currently unknown in the Early Jurassic (i.e., Megalosauridae, Spinosauridae, Allosauroidea, and Coelurosauria) to be extended back by at least 25 My (Carrano, Benson & Sampson, 2012).

Based on the phylogenetic framework of Wang et al. (2017), Delcourt (2018) proposed a radically alternative topology for Ceratosauria, where the noasaurid-elaphrosaurine grade taxa form the basalmost ceratosaurian clade which is sister group of Ceratosauridae + Abelisauridae. We re-run our phylogenetic analysis, enforcing the topology of Wang et al. (2017) and Delcourt (2018), setting Saltriovenator (not included in the analysis of Wang et al., 2017) as a floating taxon (i.e., its placement in the tree was not constrained by the enforced topology). The resulted shortest trees reconstructed under such constraint are 15 steps longer than our preferred scenario, and thus are rejected as not parsimonious interpretations of the data. Note that our data set is based on a larger character sample than the one of Wang et al. (2017)—1,781 characters vs 744—and includes several appendicular characters relevant in the placement of Saltriovenator. It is noteworthy that even under that alternative topological constraint, Saltriovenator resulted the basalmost ceratosaurian (i.e., sister group of Ceratosauroidea sensu Delcourt, 2018): thus, we conclude that our main evolutionary result (i.e., Saltriovenator representing the plesiomorphic condition of the ceratosaurian hand) is not biased by the data set used.

Histology and ontogenetic status

The age and maturity of the holotypic specimen of Saltriovenator zanellai can be inferred by suture fusion and by analysis of histological thin-sections. In MSNM V3664, the scapulocoracoid is unfused and unsutured; the tarsal bones are fully ossified but not fused to metatarsals; the metatarsals do not show any sign of proximal co-ossification. These features might indicate that MSNM V3664 was a subadult, showing some but not all of the skeletal transformations that mark cessation of growth (Tykoski & Rowe, 2004). However, these and other common characters used to assess the ontogenetic stage of fossil tetrapods, such as surface texture of bones, and obliteration of the sutures in skulls and vertebrae, have been found to be ambiguous (Brochu, 1996; Werning, 2012; Bailleul et al., 2016). Moreover, skull and vertebrae are not preserved in our specimen. Therefore, histological analysis is the most reliable method for ontogenetic assessment and absolute estimation of the age of an individual (Chinsamy, 2005; Erickson et al., 2004; Erickson, 2005).

We sampled the broken diaphysis of the left humerus, as well as a fragment of the shaft of a right dorsal rib (Figs. 16A and 16B). The type of microstructure, the density and type of vascular canals, the amount of remodelling, the number of LAGs, and the presence or absence of an EFS were the proxies used in this study to evaluate ontogenetic stage. Waskow & Mateus (2017) recently demonstrated that dorsal ribs record cyclical growth marks: sampling dorsal ribs 1–3 within the proximal third of the rib, but distal to the capitulum and tuberculum, recorded the most intact and complete history of LAGs. Our sample comes from the mid of the shaft and the LAGs record is therefore underestimated. Nonetheless, the data collected allowed to reliably infer the ontogenetic stage of the individual and if somatic and reproductive maturity were reached before death.

The sectioned bones do not differ in bone architecture to those of the other land-dwelling theropods (Erickson et al., 2004; Waskow & Mateus, 2017). In both sections, primary bone was observed in the outermost part of the compacta, towards the bone surface. It is woven fibrolamellar bone. In the rib section, most blood vessels are longitudinally oriented, and only a few ones are directed radially throughout the compacta. In the humerus the vascularization is more regularly organized, that is, radially directed in the inner portion and longitudinally directed towards the outermost cortex, suggesting a slowing down of the growth of the bone tissue and, as a consequence, of the animal. In the rib, the medulla is spongier in the middle, and the hollow medullary cavities are partly filled by matrix. In thin-sectioning the humerus, the medullary cavity was not sampled but the broken diaphysis shows that it is broad, open, and the passage between the medullary cavity and the compacta is abrupt.

Haversian systems are present and abundant in the inner cortex in both humerus and rib sections, and secondary osteons are visible closer to the outer surface. The latter elements are more abundant in the rib section. The amount of remodelling of the primary cortex clearly indicates that the animal was not juvenile. In the rib section, some LAGs are sometimes interrupted by secondary osteons but can be traced circumferentially. They are regularly spaced, except for the last (external, toward the outer cortex) LAGs, which are more closely spaced than the others. Due to the resorption of primary bone tissue caused by the expansion of the medullary cavity, only five LAGs are preserved in the cortical bone of the rib. As the rib section is not from the proximal third, this number of LAGs cannot be used to estimate its age at the time of death. Nonetheless, we performed a retrocalculation of the number of zones obscured from remodeling. For a consistent count of LAGs we followed Ibrahim et al. (2014) and employed three recognition criteria: the broadest zone, taken as representative of each missing band; the ultimate or penultimate zone; the mean interval between the three innermost zones. Using each criterion, respectively, we calculated 11, 27, and 19 missing LAGs, with a resulting mean of 19 years missing.

Thus, we estimate that the age of the holotype of Saltriovenator zanellai at the time of death was at least 24 years. The preservation of the humerus section does not allow to count LAGs. The combination of lamellar bone, reduced vascularization, and more closely spaced LAGs is here interpreted as an incipient EFS, more marked in the rib section. The presence of this incipient EFS and the remodelling observed in the compacta suggest that MSNM V3664 was a subadult approaching somatic maturity.

Skeletal reconstruction and body size

Despite the incompleteness of our material, we attempted to reconstruct the skeleton of Saltriovenator zanellai n. gen. n. sp. using comparative anatomy and character state inference to predict a plausible range for the size and the proportions of the missing elements. Alternative methods, discussed in literature, were used, and their results compared.

Comparison with more complete specimens

The scapula and humerus of Saltriovenator zanellai are, respectively, subequal and 10% longer than the corresponding elements of the articulated skeleton of a subadult Allosaurus fragilis (MOR 693), which is about 8 m long. In the hindlimb, the distal tarsals and the proximal end of metatarsals II and III of these two specimens are subequal in size. Compared to MOR 693, the forelimb elements in Saltriovenator are proportionally stockier and more robust (Figs. 16C and 17). We thus conservatively conclude that MSNM V3664 was at least seven to eight m long at the time of death.

Prior to the discovery of Saltriovenator, the coelophysoid-grade Cryolophosaurus was considered the largest Early Jurassic theropod (Smith et al., 2007). Although not enough comparable material with Saltriovenator is available (only the coracoids overlap), Cryolophosaurus is stated to be comparable in size to UMNH 5278, the largest specimen of Ceratosaurus (Smith et al., 2007; Madsen & Welles, 2000): based on comparable material with the latter, Saltriovenator is at least 25% larger.

Macroevolutionary implications of Saltriovenator

The holotypic skeleton of Saltriovenator zanellai belongs to a subadult individual approaching somatic maturity and shows a combination of coelophysoid-grade symplesiomorphies coupled with derived features shared with ceratosaurians and basal tetanurans (Figs. 4 and 16C). The strap-like scapular blade is comparable to basal ceratosaurians (e.g., Ceratosaurus; Madsen & Welles, 2000), differing from the broader fan-shape of coelophysoids (e.g., Dilophosaurus; Welles, 1984). The coracoid is similar to dilophosaurids (e.g., Cryolophosaurus; Smith et al., 2007) in the lack of a hooked caudoventral process and in the subtriangular prominent bicipital tubercle. The broad, V-shaped furcula bears a prominent hypocleideum. The humerus is robust and straight in lateral view, as in ceratosaurians and some basal tetanurans (Gao, 1993; Madsen & Welles, 2000), and the quadrangular hypertrophied deltopectoral crest is extended for half of bone length; in later ceratosaurians, the deltopectoral crest is variably reduced (Carrano & Sampson, 2008; Xu et al., 2009). The humeral condyles are stout but flattened distally as in most ceratosaurians (Carrano & Sampson, 2008). The partially-preserved hand combines a unique mosaic of tetanuran and ceratosaurian features. The second metacarpal is stout and robust, as in all ceratosaurians (Xu et al., 2009; Burch & Carrano, 2012; Carrano & Choiniere, 2016), whereas the high metacarpus/humerus ratio fits those of non-ceratosaurian theropods (Fig. 17). The distal end of the second metacarpal is abruptly expanded and twisted, bearing asymmetrically-developed condyles, as in Berberosaurus (Allain et al., 2007). The extensor ligament pit is very pronounced and bound proximally by a prominent lip, a combination of features recalling the longest metacarpal of basalmost tetanurans (Madsen, 1976; Dong, 1984; Gao, 1993; Senter & Robins, 2005): in most ceratosaurians, the pit is shallower and the lip poorly marked (Xu et al., 2009; Carrano & Choiniere, 2016). The diaphysis of the first phalanx of the second finger is very short and stout, as in ceratosaurians, and differing from the more slender and elongate phalanx seen in coelophysoid-grade theropods and basal tetanurans (Welles, 1984; Dong, 1984; Gao, 1993). As in many ceratosaurians and basal coelurosaurs, the proximal flexor processes are prominent (Novas, 1998; Carrano & Choiniere, 2016). Although the third metacarpal is lost, the third finger is completely preserved. It recalls the third finger of coelophysoid-grade theropods and most tetanurans in bearing four functional phalanges (Madsen, 1976; Welles, 1984), including an elongate penultimate phalanx and the ungual with a distinct articular surface and prominent flexor tubercle. This mix of plesiomorphic conditions is absent in other ceratosaurians, in which the third finger has variably-reduced phalangeal formulae (Xu et al., 2009), short distal phalanges and poorly-developed articular surfaces (Burch & Carrano, 2012). In the foot, the distal tarsals are not co-ossified neither fused to the metatarsals, a plesiomorphic condition that we do not consider biased by the relatively mature ontogenetic stage of the specimen. The fourth distal tarsal bears a distinct subrectangular notch for metatarsal V, as reported in some coelophysoids (Ezcurra, 2017). The proximal surface of the third metatarsal lacks both the mediolateral plantar expansion shared by coelophysoids and some ceratosaurians (Tykoski & Rowe, 2004), and the middle constriction present in basal tetanurans (Madsen, 1976). The fourth metatarsal is curved laterodistally, suggesting graviportal adaptations in the foot.

The evolution of the hand in Ceratosauria and Tetanurae

Using the strict consensus of the shortest trees found, we infer the ancestral state of the manual phalangeal formula along the internodes of the avian stem lineage leading to Tetanurae (Fig. 15): Dinosauria, 2-3-4-3-2; Saurischia, 2-3-4-1-0 (loss of the fifth finger and reduction of the fourth finger to a single phalanx); Theropoda, 2-3-4-1-0 (retention of the saurischian ancestral condition); Neotheropoda, 2-3-4-1-0 (retention of the saurischian ancestral condition); Averostra, 2-3-4-1-X (loss of the fifth metacarpal); Tetanurae, 2-3-4-0-X (loss of the fourth finger); several lineages among Tetanurae (i.e., Allosauria, advanced tyrannosauroids, and maniraptoromorphs, including birds), 2-3-4-X-X (loss of the fourth metacarpal).

The analysis supports a step-wise lateral reduction trend along the avian stem, leading to a tetrametacarpal and tridactyl condition at the root of Tetanurae. The presence of a complete formula for finger I in Eoabelisaurus, for finger II in Limusaurus, and for finger III in Saltriovenator (and the retention of a phalanx in finger IV in Limusaurus and Majungasaurus) implies that the ancestral ceratosaurian formula must be 2-3-4-1-X. Relevant for the discussion on the homology of the manual elements in tetanurans, Saltriovenator demonstrates the morphological similarity between the second metacarpal of basalmost ceratosaurians and the longest metacarpal of basal tetanurans (e.g., compare Saltriovenator with Acrocanthosaurus, Szechuanosaurus zigongensis, and Xuanhanosaurus; Dong, 1984; Gao, 1993; Senter & Robins, 2005), thus filling the gap between the latter taxa and the other ceratosaurians (e.g., Ceratosaurus, Carrano & Choiniere, 2016). This result confirms that the longest metacarpal in tetanurans is homologous to metacarpal II of other theropods, and not to metacarpal III (contra Xu et al., 2009). Furthermore, the relatively gracile but fully-functional third finger of Saltriovenator (which bears three pre-ungual phalanges and a claw-like ungual phalanx) closely fits both the third finger of non-averostran theropods and the lateral finger of tetanurans, strongly supporting the homology between these elements (Fig. 15, yellow finger). In their review of the alternative homology patterns for the theropod hand, Xu et al. (2014a) discuss four alternative models (i.e., “frame-shift,” “lateral-shift,” “axis-shift,” and “central loss,” Xu et al., 2014a, fig. 4A): the morphological consistence of both metacarpal and phalangeal patterns between earliest ceratosaurians and tetanurans dismisses all the homology frameworks alternative to the axis-shift model, which results the most robust scenario for the evolution of the hand along the avian stem.

The phalangeal formula of Limusaurus (0-3-3-1-X) is thus markedly modified compared to the ancestral ceratosaurian (and averostran) formula and cannot be considered an ancestral stage for that of the three-fingered tetanurans, contra Xu et al. (2009).

Phylogenetic analysis places Saltriovenator as sister taxon of the other Early Jurassic averostran Berberosaurus: this lineage results the basalmost ceratosaurian branch and the oldest averostran radiation (Carrano & Sampson, 2008; Carrano, Benson & Sampson, 2012). The combination of pronounced extensor pits and hypetrophied dorsal lips, deeply gynglimoid articular surfaces and prominent extensor/flexor processes and fossae in the metacarpus and phalanges of Saltriovenator (Figs. 4, 8 and 9; Supplemental Movie 1) shows that the hand of the basalmost ceratosaurians was well-adapted to struggle and grasp and to resist digital dislocation during violent movements by manually ensnared prey (see below), as in allosauroids (Senter & Robins, 2005). This is interpreted as the symplesiomorphic condition of all averostran forelimbs. The shorter and atrophied hand in Limusaurus (and abelisaurids) is thus a secondary condition restricted to late-diverging ceratosaurians, and is not directly related to the evolution of the tridactyl hand of tetanurans. In this scenario, the basalmost tetanurans (Dong, 1984; Gao, 1993) bear metacarpals I–IV and a robust metacarpal II sharing an enlarged asymmetrical distal end with a deep extensor pit and a robust lip, as in Saltriovenator. A vestigial metacarpal IV is retained in several tetanuran lineages, supporting 2-3-4-0-X as the ancestral phalangeal formula for that clade (Bever, Gauthier & Wagner, 2011). The persistence of a robust metacarpal IV eventually bearing one phalanx, even in the late-diverging ceratosaurians with atrophied hands (Xu et al., 2009; Burch & Carrano, 2012) suggests that a developmental constraint kept the primary axis of the hand in digit 4 position in all non-tetanuran theropods (Vargas et al., 2008; Tamura et al., 2011; Xu et al., 2014a). On the contrary, the independent reduction to only three metacarpals in allosaurians, tyrannosauroids, and maniraptoromorphs may indicate that a medial shift of the primary axis (from digit position 4 to digit 3) had occurred along the basal branch of Tetanurae after the complete loss of the fourth finger (Bever, Gauthier & Wagner, 2011), which then allowed multiple losses of the vestigial metacarpal IV in tetanuran subclades (Fig. 15). Accordingly, the evolution of the tridactyl hand of birds is more parsimoniously explained by lateral loss of elements among non-tetanuran dinosaurs, followed by a single medial shift of the primary axis at tetanuran root once the fourth finger was lost, and the retention of the ancestral fingers I–II–III along the whole avian stem.

The sister-taxon relationships between Saltriovenator and the other Early Jurassic ceratosaurian, Berberosaurus, indicates a previously unknown early radiation of averostrans along the western margin of the Tethys. We estimate Saltriovenator length at ∼7.5 m, thus resulting the largest Early Jurassic theropod based on skeletal remains (Smith et al., 2007; Carrano & Sampson, 2008). With a body size comparable to many Middle and Late Jurassic tetanurans (Carrano, Benson & Sampson, 2012), Saltriovenator pre-dates the occurrence of large theropods (body mass approaching 1,000 kg) by over 25 My (Benson, 2010; Carrano, Benson & Sampson, 2012), reinforcing a scenario recently suggested on the basis of ichnological evidence (Sciscio et al., 2017). The radiation of larger and relatively stockier averostran theropods earlier than previously known may represent one of the factors that ignited the trend toward gigantism in Early Jurassic sauropods (Sander et al., 2011; McPhee et al., 2018).

Remarks on the functional morphology of the manus

The phalangeal formula inferred for Saltriovenator gives solid ground to the prediction (Carrano & Choiniere, 2016) that the basal ceratosaurian manus retained a morphology similar to that of Dilophosaurus, with reduction in length of the hand initially occurring through shortening of each phalanx, while a full complement of phalanges (including unguals) was still present. As stated above, this also argues against identifying the basal node of Averostra as the phylogenetic location for a major shift in digit identity or homology (contra Xu et al., 2009; Bever, Gauthier & Wagner, 2011). The second metacarpal of Saltriovenator shows—and somewhat emphasizes—functionally-related similarities with those of Ceratosaurus and Dilophosaurus: the distal end exhibits well-developed articular surfaces of comparably wide extent, showing that the proximal phalanx was capable of similar ranges of flexion and extension; the same occurs in the deeply gynglimoid articulations of the preserved distal phalanges.

According to Welles (1984), the pit on metacarpal II of Dilophosaurus allowed a 90° hyperextension of the proximal phalanx. The pit-and-lip complex of Saltriovenator allowed a 65° hyperextension, and contemporarily a firm hold-in-place (Supplemental Movie 1). In other words, similar to the basal tetanuran Acrocanthosaurus, Saltriovenator was adapted to struggle and grasp and to resist digital dislocation during violent movements by manually ensnared prey (Senter & Robins, 2005). Although a relatively shorter manus may imply a reduction in the size of objects that could have been grasped, the proportionally stouter manual elements in Saltriovenator may represent an adaptation to sustain during predation mechanical loads more intense than those sustained by the more gracile-limbed coelophysoid-grade theropods.

A second important similarity with some medium and large-bodied neotheropods is the asymmetry and rotation of the metacarpal condyles. In Dilophosaurus the rotation of 30° with respect to the vertical, which is comparable to that of Saltriovenator (Supplemental Movie 1), causes the proximal phalanx to project about 15° medially at full extension (Welles, 1984). In Ceratosaurus, this flexion would turn the three first digits of the hand medially, instead of extending straightforward (Gilmore, 1920). This feature is widespread in different neotheropod taxa, and may represent a symplesiomorphy of this predatory clade, evidently giving functional advantage. For example, in Acrocanthosaurus, the asymmetrical joint allows the first phalanx of the second digit to hyperextend as much as it flexes (about 40°), and turns the digit so that the tip of the claw would have rotated medially during flexion and laterally during hyperextension (Currie & Carpenter, 2000). A comparable adaptation is also reported in megaraptorids (see White et al., 2015).