The fossil material object of this contribution has been collected along successive fieldtrips led by one of us (GWR) in collaboration with the Museo Paleontológico Egidio Feruglio to Mesozoic outcrops in northern central Chubut Province (Fig. 1). Fossils are relatively well preserved but show weathering and polished breakage surfaces, particularly in articulation facets, likely due to hydraulic transport prior to deposition (Harper et al., 2018 and Varela and Parras, 2013). Measurements were mainly those defined by LaDuke (1991) and Rage (2001) and subsequently used by several authors (Rio and Mannion, 2017 and Vasile et al., 2013) and were either taken digitally from photographs or directly on vertebrae with a manual digital caliper (0.01 mm error). Drawings were made with the aid of a Zeiss Stemi SV11 binocular microscope equipped with a camera lucida and photographs were taken with a Nikon D3200 digital camera equipped with a macro lens. The osteological terminology follows that of Rage, 1984, Rage, 1998 and Rage, 2001 and LaDuke (1991), with a few additional terms from Scanferla and Canale (2007) and Garberoglio et al. (2019a). The collected fossils are housed at the Museo Paleontológico Egidio Feruglio.

The position of the new taxon from La Colonia Formation as well as the monophyly and internal relationships of Madtsoiidae were tested through a quantitative phylogenetic analysis. For this we used a modified version of the data set of Garberoglio et al. (2019a), which in turn stemmed from that of Caldwell et al. (2015). For the purpose of this study, we added several additional madtsoiid taxa known mostly or only from vertebrae and vertebral characters based on the data set of Vasile et al. (2013), as later modified by Rio and Mannion (2017). Also, we included the presumably basal snake Seismophis from the Upper Cretaceous of Brazil (Hsiou et al., 2014), which is known from Najash-like vertebrae (Garberoglio et al., 2019a), and the Cretaceous marine snake Simoliophis, also known only from isolated vertebrae (Rage et al., 2016). In total, 18 extant and 36 extinct taxa were scored for 264 osteological characters, of which 52 are of the axial skeleton (Supplementary data).

A maximum parsimony analysis of the complete data set was conducted in TNT Version 1.5-beta (Goloboff et al., 2008). Heuristic searches under equal weights consist of 500 tree replicates obtained by random addition sequence (RAS), searching for new tree topologies with tree bisection and reconnection (TBR), and saving 100 trees per replication and collapsing branches of zero length after tree search. The resulting trees were subjected to a final round of TBR branch swapping and optimal topologies were rooted with Varanus. During tree searches, multistate characters based on continuous underlying data or depicting clear morphoclines (Wiens, 2001) were treated as ordered (Supplementary data). Node support was estimated by calculation of the Bremer support (BS) and absolute frequencies under jackknifing, performing 10,000 pseudo-replicates of 10 random sequence additions each followed by TBR swapping, keeping up to 10 trees, with a probability of alteration P = 0.36.

We explored trends in body size evolution among madtsoiids by mapping a proxy of absolute size onto a stratocladogram. As a proxy of size we measured the maximum width of cotyle of middle precloacal vertebrae (Supplementary data). Measurements were either taken digitally from photographs or directly on vertebrae with a manual digital caliper (0.01 mm error). The stratocladogram was derived from one of the optimal trees obtained in the parsimony analysis, including madtsoiids plus a few outgroup taxa. Numerical ages of each extinct taxon were derived from the youngest possible age of the oldest fossil-bearing unit following the current chronostratigraphic framework (Cohen et al., 2013, updated) and internodes were set to a minimum time interval (1 Myr herein), following previous approaches (Gómez and Pérez-Ben, 2019, Marjanović and Laurin, 2008 and Marjanović and Laurin, 2014). Stratocladogram assemblage and character optimization under squared change parsimony were performed in Mesquite Version 3.31 (Maddison and Maddison, 2011).

MPEF-PV: Vertebrate Paleontological collection, Museo Paleontológico Egidio Feruglio, Trelew, Chubut Province, Argentina; MML-PV: Vertebrate Paleontological collection, Museo Municipal de Lamarque, Lamarque, Río Negro Province, Argentina.

ar, Arqual ridge; cd, condyle; cot, cotyle; hk, haemal keel; izc, interzygapophyseal constriction; izr, interzygapophyseal ridge; lf, lateral foramen; lr, lateral ridge; na, neural arch; naf, neural arch foramen; nc, neural canal; ns, neural spine; pcf, paracotylar foramen; pe, posterior embayment; prz, prezygapophysis; pzf, parazygantral foramen; scf, subcentral fossa; scr, subcentral ridge; sy, synapophyses; zp, zygosphene; zt, zygantrum.

Large madtsoiid with wide and short, massively built mid-precloacal vertebrae showing a single parazygantral foramen located in an elliptical shallow depression on each side of the zygantrum and lacking prezygapophyseal processes. Eomadtsoia is distinct from all other madtsoiids by the following combination of features (autapomorphies indicated by character and character state numbers): moderately vaulted neural arch shallowly indented posteriorly (215:0) lacking dorsolateral (= parasagittal) ridges (226:0), pierced by paired small foramina, and bearing a low neural spine (249:2); thick zygosphene nearly as wide as the cotyle; very short and wide triangular centrum in ventral view having clearly laterally convex subcentral ridges; narrow haemal keel clearly demarcated from the centrum by subcentral fossae pierced by small subcentral foramina; large lateral foramina associated with short lateral ridges; at least a large paracotylar foramen on each side of the cotyle; and finally, synapophyses broad and medial to prezygapophyseal lateral edge (221:1).

Eomadtsoia further differs from most other madtsoiids from Patagonia by its markedly larger size. Additionally, it differs from the mid-sized Rionegrophis (Albino, 1986 and Albino, 2007; ROG pers. observ. based on MML-PV91–115) by the lower neural spine and the less laterally extended synapophyses. It is easily distinguished from Madtsoia by the well-demarcated, narrow and sharp haemal keel and the lack of parasagittal ridges.

All the vertebrae identified as belonging to Eomadtsoia are from the precloacal region. Among the well-preserved vertebrae, MPEF-PV 2383 is one of the largest with a cotylar width of 10.7 mm, but a specimen tentatively referred to E. ragei that consists of an isolated zygosphene and portion of the neural arch (MPEF-PV 2388) is noticeably larger than the remaining material, indicating an even larger maximum size for this species.

Anterior view. The cotyle is big and subcircular, slightly wider than high (Fig. 2A). The robust zygosphene is as wide as the cotyle and relatively thick (low width/height ratio). The zygosphene has moderately inclined articular facets (angle between 25–35 degrees from vertical), and an almost straight, horizontal, dorsal margin. The neural canal is relatively small, about half the area of the cotyle, and subtrifoliate in cross section. The prezygapophyses are stout, well projected laterally and moderately inclined, up to 15–20° from horizontal. There is no prezygapophyseal process, though a subtle dorsoventral crest on the prezygapophyseal buttress is present, which extends to the anterodorsal edge of the synapophyses. The synapophyses are heavily weathered and, thus, diapophysis and parapophysis do not differentiate well. It is clear though that the lateral edge of the diapophyses remain clearly medial to the lateral limit of the prezygapophyses, whereas the parapophyses surpass the cotylar margin ventrally. Deep paracotylar foramina, at least one on each side of the cotyle, are located in shallow depressions that are ventrally delimited by blunt short buttresses (Fig. 2 and Fig. 3).

Lateral view. The neural spine is nearly as long as the neural arch, very low along its length, and only slightly slanted posteriorly (Fig. 3B and E). The zygosphenal articular facets are oval in shape. The condyle is eroded in all specimens, but it is clearly posterodorsally oriented and well demarcated from the centrum by a slight precondylar constriction. The synapophyses are broad and dorsoventrally elongated. A thick and oblique lateral ridge is more or less evident among specimens (Fig. 2 and Fig. 3). The subcentral ridge is almost straight all along its length. The interzygapophyseal ridge is distinct and sigmoid-shaped (Fig. 3E). There is a big lateral foramen of central position, below the above-mentioned lateral ridge. A small lateral ridge or bulge can be present on the lateral wall, below the lateral foramen.

Posterior view. The neural arch is only moderately vaulted (Fig. 2C). The condyle is subcircular. The parazygantral foramina are wide, with a distinct elliptical shape and obliquely oriented, one on each side of the zygantrum, located in small fossae. The zygantral roof is slightly convex ventrally. A pair of zygantral foramina is located on separate deep rounded fossae. The synapophyses are posterolaterally oriented, showing nearly all their articular surface in this view.

Dorsal view. The vertebrae are wider than long with a neural arch only slightly constricted (Fig. 2D). The interzygapophyseal width (minimal neural arch width) is almost the same as the prezygapophyseal width. The prezygapophyseal facets are anterolaterally oriented (angle between 60–80° from the anteroposterior axis) and have a nearly oval outline, although slight posterolateral and anteromedial corners exist. The zygosphenal anterior margin is almost straight, with only a very shallow medial notch. The neural spine is narrow, extending from the back of the anterior margin of the zygosphene, and widens as it extends posteriorly. The neural spine likely has a posterior tubercle based on its terminal widening, but its surface is not well preserved enough to confirm this feature. To both sides of the neural spine, posteriorly to the zygosphenal base, several foramina are present on the neural arch (at least one to each side in all available vertebrae). Posteriorly, the neural arch shows a shallow embayment. The parasagittal ridges are not developed on the neural arch, which in turn is not markedly faceted. Arqual ridges are present on the posterior dorsal margin of the neural arch, anteromedial to the postzygapophyseal lappet (Fig. 3F).

Ventral view. The centrum is very short and subtriangular (Fig. 2E). The subcentral ridges are straight to slightly convex laterally. Much of the synapophyseal articular surface is visible in ventral view. The haemal keel is well developed and demarcated, low and narrow, laterally is delimited by shallow depressions, where small subcentral foramina are located in anterior position. The haemal keel contacts the condylar constriction posteriorly, not reaching the condyle. Posterior processes on the haemal keel are absent.

Among the four snakes that have been previously reported from La Colonia Formation, only one of these, namely the small madtsoiid Alamitophis argentinus, was unambiguously referred to family level (Albino, 2000). A single vertebra (MPEF-PV 642) originally reported as a ‘Serpentes incertae sedis’ (Albino, 2000 and Albino, 2007) is more likely to represent a dolichosaur rather than a snake (Albino, 2011 and Scanlon and Hocknull, 2008). The remaining snakes from this unit, formerly reported as a possible madtsoiid and tentatively as a boid or madtsoiid (Albino, 2000), were based on poorly preserved material that precluded a better taxonomic assignment at the time. Now, additional and better preserved snake vertebrae from the La Colonia Formation that show a unique combination of characters justifying the erection of a new taxon, Eomadtsoia ragei gen et sp. nov., allow referral of previously known material of uncertain madtsoiids or boids to this same taxon. Therefore, the snake assemblage from La Colonia Formation is restricted to madtsoiids, with E. ragei increasing the known diversity of this basal snake radiation.

Among snakes, the overall vertebral configuration of E. ragei is only present in the Madtsoiidae and boids, but it can be set apart from extinct and extant boids by the presence of distinct parazygantral foramina along with paracotylar foramina, arqual ridges on the neural arch, and the absence of prezygapophyseal processes (Gómez and Báez, 2005, Rage, 1998 and Scanlon, 2005). It is readily distinguishable from small-to-mid-sized madtsoiids from the Late Cretaceous of Patagonia and Europe and Paleogene of Australia, namely Alamitophis, Herensugea, Nanowana, Nidophis, Patagoniophis, Rionegrophis, and Wonambi barriei (Albino, 1986, Albino, 1994, Albino, 2007, Rage, 1996, Scanlon, 1997, Scanlon, 2005, Scanlon and Lee, 2000 and Vasile et al., 2013), by its conspicuous larger size, along with the wide and short triangular centrum and the distinctly medial position of the synapophyses relative to the lateral edge of the prezygapophyses. Eomadtsoia clearly contrasts with the much older Dinilysia (Caldwell and Albino, 2002 and Scanferla and Canale, 2007), which is the only other Late Cretaceous large snake from Patagonia, in having parazygantral foramina, lower neural spine, and poorly constricted neural arch, among other features.

Noteworthy, the narrow and sharp haemal keel of Eomadtsoia contrasts with that of large-to-gigantic madtsoiids, such as species of Gigantophis, Madtsoia, Menarana, and Platispondylophis, in which the haemal keel is a somewhat broad and flat process more or less demarcated from the rest of the centrum (Andrews, 1901, LaDuke et al., 2010, Rage, 1996, Rage, 1998, Rage et al., 2014, Rio and Mannion, 2017 and Smith et al., 2016). In addition, an interzygapophyseal constriction as shallow as that of E. ragei is rarely found among large madtsoiids, which generally show a more marked constriction. The lack of parasagittal ridges of Eomadtsoia is also uncommon, only resembling Herensugea (Rage, 1996) and Adinophis (Pritchard et al., 2014) among the surveyed madtsoiids; although some intracolumnar variation cannot be ruled out taken into account what is seen in Gigantophis (Rio and Mannion, 2017)

The parsimony analysis yielded 42 most parsimonious trees (MPTs) of 803 steps (consistency index [CI] = 0.400; retention index [RI] = 0.697), the strict consensus of which is relatively well resolved and depicts all madtsoiids in a clade also including the basal snakes Najash, Dinilysia, and Seismophis (Fig. 4). In agreement with recent hypotheses, these snakes lie outside crown-group Serpentes (Caldwell et al., 2015, Garberoglio et al., 2019a, Garberoglio et al., 2019b, Harrington and Reeder, 2017, Martill et al., 2015 and Vasile et al., 2013). The monophyly of madtsoiids is here rigorously tested, since we use as outgroup taxa all major groups of extant snakes plus several Mesozoic snakes and the lizard Varanus for rooting (Fig. S1), instead of solely the basal snakes Najash and Dinilysia as in previous similar studies (Rio and Mannion, 2017).

Our results depict Najash and Dinilysia forming a clade crown-ward to Seismophis and all nested within taxa usually considered part of Madtsoiidae, since the poorly known Herensugea from the Maastrichtian of Iberia (Rage, 1996) emerges as an early diverging taxon sister to all other snakes in this clade; a position that contrasts with that recovered in previous studies (Rio and Mannion, 2017 and Vasile et al., 2013). The isolated vertebrae of Seismophis recall in many aspects the vertebrae of Najash (Garberoglio et al., 2019a and Hsiou et al., 2014), both from the Cenomanian of South America and, hence, the oldest members of this clade. Interestingly, the vertebral morphology of Herensugea is quite unique among madtsoiids, having depressed neural arch with low parasagittal keels, and prezygapophyses and main axis of zygosphenal facets close to horizontal (Rage, 1996), and it would not be surprising if it represented an end-Cretaceous relict of a much earlier snake radiation along with a still unnamed putative madtsoiid from the Cenomanian of North Africa (Rage and Dutheil, 2008).

All these basal snakes from Gondwana and neighboring terrains share the presence of conspicuous parazygantral foramina, which has been considered the clearest synapomorphy of Madtsoiidae by most authors (Albino, 2000, Rage, 1984, Rage, 1998, Rio and Mannion, 2017, Scanlon, 2005 and Vasile et al., 2013), although they have also been reported as irregularly present in a few other snakes (Holman and Case, 1992, Rage, 1998 and Rage et al., 2004). In view of our results depicting Seismophis, Najash, and Dinilysia basal to all madtsoiids other than Herensugea, and that the anatomy and relationships of the enigmatic Herensugea needs further scrutiny, we opt to not strictly consider the former taxa as part of Madtsoiidae and instead regard the parazygantral foramen as synapomorphic of a more inclusive clade than traditionally acknowledged.

Our results recover all large-bodied madtsoiids as a clade sister to mostly small-sized madtsoiids (Fig. 4), mirroring to some extent previous hypotheses derived from comprehensive analyses of Madtsoiidae (Rio and Mannion, 2017 and Vasile et al., 2013). However, our tree topology showing two clades reflecting overall size has not been recovered previously (Vasile et al., 2013: fig. 5A). Large madtsoiids are not members of a single clade either in the analysis of Rio and Mannion (2017: fig. 10B). These size-related groups are here recovered despite the fact that ‘body size’, in contrast with the previous studies, was not included as a discrete character in our analysis.

Eomadtsoia ragei is here recovered as an early diverging lineage of the clade of large madtsoiids, crown-ward to Sanajeh and Rionegrophis and sister to a poorly resolved clade encompassing snakes in the genera Gigantophis, Madtsoia, Menarana, and Platyspondylophis from the Maastrichtian–Eocene interval, as well as the younger Wonambi and Yurlunggur from Australia (Fig. 4). Interestingly, deeply nested within this clade also appears the small Adinophis fisaka (Pritchard et al., 2014), which was previously recovered as more closely related to other small madtsoiids (Rio and Mannion, 2017). None of the madtsoiid genera represented by more than one species have been recovered as monophyletic neither here nor in previous analyses (Rio and Mannion, 2017 and Vasile et al., 2013), calling for further revision of the taxonomy of these snakes.

Gigantic Madtsoiidae are some of the largest snakes that have ever existed, reaching in some cases 10 m of total length (Head et al., 2009 and Rio and Mannion, 2017), but according to our results they evolved from small-to-mid-sized forms at some point during the Cretaceous (Fig. 5), likely in the Late Cretaceous of Gondwana. Among the closest relatives of true madtsoiids are the relatively small snakes Seismophis and Najash, but also the much larger Dinilysia that might have exceeded 1.8 m in length (Yi and Norell, 2015), suggesting that this early snake radiation was not constrained in size, which is also echoed by the initial split of madtsoiids into sister clades of disparate body size.

Different body-size evolutionary trends characterize these madtsoiid clades. That of small forms, which appear to have radiated to reach a widespread geographic distribution during the Maastrichtian (Vasile et al., 2013), shows relative stasis in body size over time. Into the Cenozoic they remained diverse apparently only in Australia (Scanlon, 1997 and Scanlon, 2005), although Paleogene putative madtsoiids scattered across Gondwanan landmasses (Rage, 1991 and Rage et al., 2008) might also belong to this group. Conversely, its sister clade shows a general trend of increased body size since the Latest Cretaceous with several independent transitions towards gigantism prior to the K/Pg boundary (Fig. 5). In addition, this subclade exhibits a higher diversity of body sizes with some reversal shifts to a smaller size in Menarana nosymena and, more markedly, Adinophis fisaka, both from Madagascar (LaDuke et al., 2010 and Pritchard et al., 2014).

In the current phylogenetic framework, Eomadtsoia, together with Sanajeh from the Maastrichtian of India and Rionegrophis from the slightly older Los Alamitos and Allen formations of Patagonia (Albino, 1986, Albino, 2007 and Gómez, 2006), emerge as early diverging mid-to-large madtsoiids with vertebral morphologies and body sizes that could be interpreted as foretelling of the size increase characteristic of the more nested clades. Eomadstsoia fits in a morphocline, linking the ancestral madtsoiid conditions with those of large-to-gigantic taxa such as Madtsoia bai from the Eocene of Patagonia (Simpson, 1933).