Mendozasaurus, GONZALEZ RIGA, 2003

MENDOZASAURUS GONZÁLEZ RIGA, 2003

Type species: Mendozasaurus neguyelap González Riga, 2003 .

Holotype: Twenty-two mostly articulated caudal vertebrae (IANIGLA-PV 065/1–22), three anterior chevrons (IANIGLA-PV 065/23–25) and fragments of posterior chevrons (IANIGLA-PV 065/26–30).

Referred material: The following disarticulated bones were found associated with the holotype, including remains not mentioned in previous publications on Mendozasaurus : five cervical vertebrae (IANIGLA-PV 076/1–3; 076/5, 084/1); two dorsal vertebrae (IANIGLA-PV 76/4, 066); one thoracic rib (IANIGLA-PV 084/2); a right scapula (IANIGLA-PV 068); a right sternal plate (IANIGLA-PV 067); a left and right humerus (IANIGLA-PV 069/1–2); a right radius (IANIGLA-PV 070/2); a right ulna (IANIGLA-PV 070/1); six metacarpals (IANIGLA-PV 071/1–5, 154); a fragment of pubis (IANIGLA-PV 072); the proximal half of a right femur (IANIGLA-PV 073/1) and a left femur (IANIGLA-PV 073/4); one left (IANIGLA-PV 074/2) and three right tibiae (IANIGLA-PV 073/2–3, 074/1); a left and right fibula (IANIGLA-PV 074/3 and 074/4, respectively); a right astragalus (IANIGLA-PV 155); 12 metatarsals (IANIGLA-PV 077/1–5, 100/1–6, 153); ten pedal phalanges (IANIGLA-PV 077/6–12, 078/1–2, 079) and four osteoderms (IANIGLA-PV 080/1–2, 81/1–2).

Revised diagnosis: Mendozasaurus neguyelap can be diagnosed by seven autapomorphies (marked with an asterisk), as well as two local autapomorphies: (1) middle–posterior cervical vertebrae with tall and transversely expanded neural spines that are wider than the centra, with the lateral expansion formed by spinodiapophyseal laminae, without contribution from the pre- or postzygapophyses*; (2) anteriormost caudal neural spine dorsoventral height divided by centrum height> 1.2; (3) anterior caudal vertebrae (excluding anteriormost) with ventrolateral thickening of prezygapophyses*; (4) middle caudal centra with greatly reduced posterior condyles displaced dorsally*; (5) laterally compressed and anteroposteriorly elongated middle caudal neural spines, with the horizontal dorsal margin forming a 90° angle with the dorsal portion of the anterior margin in lateral view*; (6) humerus with divided lateral distal condyle on anterior surface; (7) second ridge on posterior surface of distal third of radius, parallel to main interosseous ridge*; (8) metacarpal I with ridge or tubercle on the dorsolateral margin at approximately two thirds of length from proximal end*; (9) large subconical to subspherical osteoderms, lacking a cingulum*.

Locality and horizon: Arroyo Seco, south of Cerro Guillermo , Malargüe Department, Mendoza Province, Argentina ( González Riga, 2003); upper levels of the Sierra Barrosa Formation, Río Neuquén Subgroup, Neuquén Group; middle– upper Coniacian, early Late

Cretaceous ( Garrido, 2010; González Riga & Ortiz David, 2014) ( Figs 1 View Figure 1 , 2 View Figure 2 ).

DESCRIPTION AND COMPARISONS

Nomenclature for vertebral laminae and fossae follows the standardized terminology of Wilson (1999) and Wilson et al. (2011b), and serial variation in caudal vertebrae is demarcated using the scheme proposed by Mannion et al. (2013).

CERVICAL VERTEBRAE

Five cervical vertebrae are preserved ( González Riga, 2005), including two that are described here for the first time (see Table 1 for measurements). A middle– posterior cervical vertebra (IANIGLA-PV 076/5) is relatively complete ( Fig. 4 View Figure 4 ), but the posterior end has been eroded away, and the element is still in its field jacket. It is probably the most anterior cervical vertebra preserved. IANIGLA-PV 076/3 preserves the centrum and lower part of the neural arch, including the right diapophysis, of a middle–posterior cervical vertebra, but it has been strongly compressed dorsoventrally ( Fig. 5 View Figure 5 ). IANIGLA-PV 076/2 preserves only a fragmentary neural spine and partial postzygapophysis. IANIGLA-PV 076/1 is a posterior cervical vertebra that is generally complete, but poorly preserved in places, and has undergone anteroposterior compression ( Fig. 6 View Figure 6 ). The fifth element (IANIGLA-PV 84/1) is probably one of the posteriormost cervical vertebrae ( Fig. 7 View Figure 7 ) and is less distorted than IANIGLA-PV 076/1. Apart from its eroded posterior surface, IANIGLA-PV 84/1 is largely complete. This latter cervical vertebra, as well as IANIGLA-PV 076/ 5 in particular,

All measurements are in millimetres.

*Measurements have been heavily affected by crushing.

† A measurement that is extrapolated based on only one half being complete.

demonstrates that the cervical (and dorsal) centra of Mendozasaurus were not especially short anteroposteriorly, and that the apparently short length of the centrum of IANIGLA-PV 076/1 is best regarded as a taphonomic artefact. Below we describe all of the cervical vertebrae together, rather than individually, noting where there is morphological variation between elements.

All of the cervical centra are opisthocoelous and dorsoventrally compressed, with the height:width ratio varying between 0.86 in IANIGLA-PV 076/5 and 0.49 in IANIGLA-PV 84/1. The ventral surface of the centrum is transversely concave in between the parapophyses, flattening and becoming convex posteriorly. Whereas there is no ventral ridge on IANIGLA-PV 076/3 ( Fig. 5E View Figure 5 ), there is a low, rounded midline ridge along the anterior half of the non-condylar centrum in IANIGLA-PV 076/5 and 076/1 ( Fig. 6E View Figure 6 ). Although generally absent in most macronarians, a small number of somphospondylans also preserve a ventral ridge in at least some cervical vertebrae, for example Rapetosaurus ( Curry Rogers, 2009) and Savannasaurus ( Poropat et al., 2016) . All of the cervical centra lack ventral fossae and ventrolateral ridges. The lateral surface of the centrum is excavated by a fairly deep lateral pneumatic fossa, but this does not open into a foramen, and there are also no dividing ridges within the fossa. This ‘simple’ fossa is comparable to many somphospondylans, but it tends to be much shallower in those taxa ( Upchurch, 1998; Curry Rogers, 2005), including Futalognkosaurus ( Calvo et al., 2008b) . A distinct ridge forms the dorsal margin of the lateral fossa. Parapophyses project laterally and, with the exception of the most posterior cervical vertebra preserved (IANIGLA-PV 84/1), quite strongly ventrally ( Figs 4A View Figure 4 , 6A View Figure 6 ), with those of IANIGLA-PV 076/1 similar to the condition in some euhelopodids ( D’Emic, 2012) and diplodocoids ( Mannion et al., 2013), as well as at least some other titanosaurs, for example Isisaurus ( Jain & Bandyopadhyay, 1997) , Overosaurus ( Coria et al., 2013) , Patagotitan ( Carballido et al., 2017) and Puertasaurus ( Novas et al., 2005) . Unlike some saltasaurids ( D’Emic, 2012), the parapophyses do not extend as far as the midlength of the centrum. As is the case in most derived titanosaurs ( Upchurch, 1998; Curry Rogers, 2005), the dorsal surfaces of the parapophyses are unexcavated.

Relative to the height of the centrum, the neural arch of IANIGLA-PV 84/1 is dorsoventrally low in anterior view ( Fig. 7C View Figure 7 ), consistent with the posterior cervical vertebrae of most derived somphospondylans ( Bonaparte, González Riga & Apesteguía, 2006; Mannion et al., 2013). Both the anterior and posterior neural canal openings are subcircular. Centroprezygapophyseal laminae (CPRLs) are flat, mediolaterally wide sheets of bone that are not excavated or divided. The prezygapophyses are well separated from one another by a transversely elongate intraprezygapophyseal lamina (TPRL) that is mainly horizontal in anterior view ( Fig. 7C View Figure 7 ), dipping only very gently towards the midline; in dorsal view, it is U shaped ( Fig. 7B View Figure 7 ). Only in the posteriormost cervical vertebra (IANIGLA-PV 84/1) does the TPRL form the dorsal margin of the neural canal ( Fig. 7C View Figure 7 ); in more anterior cervical vertebrae (see IANIGLA-PV 076/1) the anterior surface of the arch surrounding the neural canal is flat and featureless ( Fig. 6A View Figure 6 ). The prezygapophyses are widely separated along their midline. Each prezygapophyseal articular surface is flat and faces mainly dorsally, but also medially and slightly anteriorly. They increase in anteroposterior length laterally and extend a very short distance beyond the anterior margin of the condyle. In posterior view, the intrapostzygapophyseal lamina (TPOL) has a shallow, transversely wide U shape, with no midline ventral ridge extending between it and the dorsal margin of the neural canal ( Fig. 6C View Figure 6 ). The postzygapophyseal articular surfaces are flat and face mainly ventrally, but also laterally and very slightly posteriorly. There are no pre-epipophyses or epipophyses.

The diapophysis is supported from below by prominent anterior centrodiapophyseal (ACDL) and posterior centrodiapophyseal laminae (PCDL). The ACDL and PCDL form the margins of the centrodiapophyseal fossa (CDF), with the ventral margin of this fossa formed by the sharp ridge that delimits the dorsal margin of the lateral fossa of the centrum. A prezygodiapophyseal lamina (PRDL) and postzygodiapophyseal lamina (PODL) also contribute to the sheet-like diapophysis, with the dorsal surface of the latter tilted to face posterodorsally. The convex anterior border of the PRDL gives the diapophysis a ‘wing’ shape in dorsal/ ventral and anterior/posterior views ( González Riga, 2005). The diapophyses project mainly laterally, but also curve slightly ventrally (e.g. Fig. 6A View Figure 6 ). An accessory lamina runs along the posterior surface of the neural arch, emanating from the posterior margin of the PCDL ( Fig. 6A, E View Figure 6 ). A broken portion of the diapophysis of IANIGLA 076/5 reveals a camellate internal tissue structure, as characterizes the cervical and anterior dorsal vertebrae of Galveosaurus + Titanosauriformes ( Wilson & Sereno, 1998; Mannion et al., 2013).

Although accentuated by crushing, the neural spine is anteroposteriorly short along its length, lacking bifurcation ( Fig. 6F View Figure 6 ). It projects mainly dorsally, although we cannot be certain whether the anterior deflection of IANIGLA-PV 076/1 ( Fig. 6B, D View Figure 6 ) is a genuine feature. The neural spine is a dorsoventrally tall structure, exceeding twice the height of the centrum in IANIGLA-PV 076/1 ( Fig. 6 View Figure 6 ). Despite poor preservation, a midline prespinal ridge extends along most, or all, of the anterior surface of the neural spine of IANIGLA-PV 076/1 ( Fig. 6A View Figure 6 ; see also González Riga, 2005), as is the case in the posterior cervical vertebrae of most somphospondylans ( Salgado, Coria & Calvo, 1997; D’Emic, 2012), but is absent from the other preserved cervical neural spines, including IANIGLA-PV 84/1 ( Fig. 7C View Figure 7 ). Robust lateral spinoprezygapophyseal laminae (lSPRLs) extend dorsomedially from the posterolateral corners of the prezygapophyses and extend along most of the spine ( Fig. 6A View Figure 6 ). Although broken, there are also remnants of medial SPRLs (mSPRLs) at the base of the neural spine ( Fig. 6A View Figure 6 ), which presumably must have merged with the prespinal lamina dorsally. The presence of paired SPRLs might represent an autapomorphy of Mendozasaurus , but we exclude it from our diagnosis because of the poor preservation in this region. Spinopostzygapophyseal laminae (SPOLs) form the posterolateral margins of a mediolaterally wide spinopostzygapophyseal (=postspinal) fossa (SPOF). The SPOLs are directly strongly dorsally, as well as being slightly medially deflected ( Fig. 6C View Figure 6 ). There does not appear to be a postspinal ridge, but this region is poorly preserved.

On the lateral surface, at the base of the neural spine, there is a deep (but not sharp-lipped) spinodiapophyseal fossa (SDF) that is floored by the diapophysis, bounded anteriorly by the lSPRL and posteriorly by the postzygapophysis ( Figs 4B View Figure 4 , 6D View Figure 6 , 7A View Figure 7 ). Although many taxa have an SDF in their cervical vertebrae ( González Riga, 2005; Wilson et al., 2011b), the depth noted in Mendozasaurus otherwise appears to be restricted to the posterior cervical vertebrae of Futalognkosaurus ( Calvo et al., 2008b) , Alamosaurus ( Tykoski & Fiorillo, 2017) and possibly Isisaurus ( Jain & Bandyopadhyay, 1997) . Within this fossa, a spinodiapophyseal lamina (SPDL) starts at the base of the spine and continues dorsally, where it is the sole contributor to the lateral expansion of the upper portion of the neural spine ( Figs 4 View Figure 4 , 6 View Figure 6 ). This lateral expansion means that the neural spine extends further laterally than the margins of the centrum, although it does not extend as far as the lateral margins of the prezygapophyses ( González Riga, 2005). It also gives the neural spine a strongly convex dorsal margin in anterior view. Although this ‘paddle’-shaped morphology has been described in the posteriormost cervical and anteriormost dorsal vertebrae of a number of somphospondylans ( Bonaparte et al., 2006; Calvo et al., 2008b; D’Emic, 2012), the laminar contribution to the lateral expansion differs between taxa ( González Riga, 2010; Gallina, 2011; Gallina & Apesteguía, 2015). Only in Futalognkosaurus , Mendozasaurus , Quetecsaurus (Gallina, 2011; González Riga & Ortiz David, 2014) and Alamosaurus ( Tykoski & Fiorillo, 2017) is this known to be formed entirely by the SPDL, and Mendozasaurus is distinct in that this lateral expansion results in the neural spine being wider than the centrum.

DORSAL VERTEBRAE

Two dorsal vertebrae are preserved (IANIGLA-PV 076/4 and 066; see Table 1 for measurements). IANIGLA-PV 076/4 is interpreted as one of the anteriormost dorsal vertebrae ( González Riga, 2005). It preserves an incomplete neural arch, most of the neural spine, and the right diapophysis, although the posterior surface is largely incomplete. IANIGLA-PV 066 is complete, but it is poorly preserved in places, and has undergone some anteroposterior compression ( Fig. 8 View Figure 8 ). It is also from the anterior region of the dorsal series (suggested to be Dv3 by González Riga, 2005), evidenced by the position of the parapophysis on the dorsal half of the centrum and lower portion of the neural arch.

Although the anterior condyle of the centrum of IANIGLA-PV 066 has been worn ( Fig. 8A View Figure 8 ), it was clearly strongly convex, as evidenced by the fairly deep posterior cotyle. The centrum is dorsoventrally compressed (width to height ratio = 1.4), comparable to the anterior dorsal centra of several titanosaurs, including Malawisaurus , Notocolossus , Opisthocoelicaudia and Rapetosaurus ( Mannion et al., 2013; González Riga et al., 2016). Ventrally, the centrum is transversely convex, lacking ridges or fossae ( Fig. 8E View Figure 8 ). A pneumatic foramen excavates the lateral surface of the dorsal half and anterior two thirds of the centrum ( Fig. 8B, D View Figure 8 ), although it does not ramify deeply. This foramen is posteriorly acute, and seems to be set within a fossa, as in many somphospondylans ( Upchurch, Barrett & Dodson, 2004; Mannion et al., 2013). There are no ridges within either the fossa or foramen. The erosion of the anterior surface of the centrum reveals that the internal tissue structure is camellate.

On the lateral surface of the neural arch, the paradiapophyseal lamina (PPDL) and PCDL define a narrow, subtriangular CDF ( Fig. 8B View Figure 8 ). The CPRLs are not bifid, but there are shallow, paired fossae dorsolateral to the neural canal opening on IANIGLA-PV 066. Similar excavations are present in the dorsal vertebrae of several other titanosaurs, including Pitekunsaurus (MAU- Pv-AG-446: PDM pers. observ. 2014) and Rinconsaurus (MAU-PV-CRS-05: PDM pers. observ. 2014). There is a large, deep, semicircular shaped prezygapophyseal centrodiapophyseal fossa (PRCDF) limited by the CPRL, PRDL and PPDL ( Fig. 8B View Figure 8 ). In anterior view, the TPRL is gently convex ( Fig. 8A View Figure 8 ). Whereas the flat prezygapophyseal articular surfaces mainly face dorsally in IANIGLA-PV 076/4, they face dorsomedially and slightly anteriorly in IANIGLA-PV 066, tilted at ~40° to the horizontal, as is the case in most titanosaurs ( Carballido et al., 2012).

Each CPOL has a free anterior margin as a result of excavation of its lateral surface, whereas the remainder of the posterior surface of the neural arch lacks ridges or fossae ( Fig. 8C View Figure 8 ). The area above the posterior neural canal opening, in between the CPOLs, is shallowly excavated, and is divided by a ridge ( González Riga, 2003) that has been broken and distorted. This was presumably a midline ridge that extended from the ventral midpoint of the V-shaped TPOL: a comparable feature is present in the anterior dorsal vertebrae of several other sauropods ( Poropat et al., 2016), including the titanosaurs Bonitasaura , Malawisaurus , Muyelensaurus and Rapetosaurus ( Curry Rogers, 2005, 2009; Gallina & Apesteguía, 2011). The postzygapophyseal articular surfaces are flat and strongly tilted to face ventrolaterally. They have almost certainly been crushed anteriorly in IANIGLA-PV 066, so that they are extremely well separated along the midline; this has also resulted in the PODL being extremely short and barely discernible in this vertebra ( Fig. 8D View Figure 8 ). Although the absence of a hyposphene characterizes the middle–posterior dorsal vertebrae of Lithostrotia ( Salgado et al., 1997; Upchurch, 1998), the two preserved dorsal vertebrae of Mendozasaurus are probably too anterior in the sequence to determine whether this structure was genuinely absent in this taxon.

Diapophyses project laterally and very slightly dorsally, and they are longer along their dorsal than ventral margins ( Fig. 8A, C View Figure 8 ). They also expand slightly dorsoventrally towards their distal tips. Each diapophysis has a subtriangular cross section, with the apex of this triangle formed by the PCDL, and the other corners formed by the PODL and PRDL. As is the case in the cervical vertebrae, the PRDL extends laterally in IANIGLA-PV 076/4, whereas it is a reduced structure in IANIGLA-PV 066. It is not possible to determine whether an ACDL is genuinely absent. The posterior surface of the diapophysis lacks excavations.

The neural spine is anteroposteriorly compressed and projects primarily dorsally, and very slightly posteriorly ( Fig. 8B, D View Figure 8 ). Dorsal to the postzygapophyses, the neural spine rapidly decreases in transverse width, before forming subparallel margins along the dorsal half, with a gently convex dorsal margin. In IANIGLA-PV 076/4, the neural spine appears to have small triangular aliform processes, which seem to be formed entirely by the SPOLs ( Fig. 8C View Figure 8 ). Distinct midline prespinal and postspinal laminae begin at the base of the neural spine and extend until at least close to the spine apex ( Fig. 8A, C View Figure 8 ), as is the case in most somphospondylans ( Salgado et al., 1997; Curry Rogers, 2005; D’Emic, 2012; Mannion et al., 2013). Short, but distinct, spinoprezygapophyseal laminae (SPRLs) are present at the base of the neural spine ( Fig. 8A View Figure 8 ). In IANIGLA-PV 076/4, these merge at spine midheight with the undivided SPDLs, resulting in a prominent fossa on the dorsal surface of the diapophysis, posterior to the prezygapophysis. In contrast, the SPRLs in IANIGLA-PV 066 are less robust and merge with the prespinal ridge a short distance up the spine. SPDLs form the anterolateral margins of the neural spine and remain separate from the prespinal lamina. The presence of SPDLs in anteriormost dorsal vertebrae is generally a feature restricted to lithostrotians and some diplodocoids ( Salgado et al., 1997; D’Emic, 2012; Poropat et al., 2016). A well-developed postzygapophyseal spinodiapophyseal fossa (POSDF) is present between the SPDL and SPOL ( González Riga, 2003). SPOLs are undivided and form the posterolateral margins of the neural spine; there also seems to be evidence for weakly developed epipophyses.

The shaft of a large thoracic rib (IANIGLA-PV 084/2) is preserved, demonstrating its plank-like cross-sectional shape. However, its incomplete nature means that we cannot determine its position in the dorsal sequence.

CAUDAL VERTEBRAE

A total of 22 caudal vertebrae are preserved ( Fig. 9; see Table 2 for measurements). These comprise: (1) four anterior caudal vertebrae (IANIGLA-PV 065/1–4) that were found disarticulated ( Fig. 9A–M); (2) a series of nine partially articulated anterior–middle caudal vertebrae (IANIGLA-PV 065/5–13) ( Fig. 9N–T, AC–AD); (3) a series of six articulated middle–posterior caudal vertebrae (IANIGLA-PV 065/14–19) ( Fig. 9U –AD); and (4) three disarticulated posterior caudal vertebrae (IANIGLA-PV 065/20–22) ( González Riga, 2003). Although they do not form a continuous series, these 22 caudal vertebrae are described as Cd I–XXII below. Rather than providing a complete description of each caudal vertebra, we record anatomical features the first time that they can be observed and then document how they change along the sequence.

Cd I is fairly complete, but has undergone anteroposterior crushing, with most of the anterior surface eroded ( Fig. 9A–D). It is from the anteriormost region of the tail, although it is not the first caudal vertebra. The centrum is strongly procoelous, as is the case in the anterior caudal vertebrae of nearly all lithostrotian titanosaurs, as well as a number of other eusauropod taxa ( Salgado et al., 1997; Upchurch, 1998; Whitlock,

Centrum heights do not include chevron facets. All measurements are in millimetres. Abbreviations: LIC, length of centrum including condyle; LEC, length of centrum excluding condyle; ACH, anterior centrum height; ACW, anterior centrum width; PCH, posterior centrum height; PCW, posterior centrum width; NAH, neural arch height; NSH, neural spine height; NSW, neural spine width at base; MNSW, maximum neural spine width; NSL, neural spine length at base; TWD, transverse width across diapophyses.

*A measurement that is extrapolated based on only one half being complete.

D’Emic & Wilson, 2011; Mannion et al., 2013). A small central depression excavates the posterior condylar surface ( González Riga, 2003). Although the ventral surface is not well preserved, it is clearly transversely convex, and seems to lack ridges or fossae. No fossa or foramen excavates the lateral surface of the centrum, in contrast to the anterior caudal vertebrae of many diplodocoids ( Upchurch, 1998), and a small number of titanosauriforms ( D’Emic, 2012; Mannion et al., 2013). Caudal ribs are situated on the upper part of the lateral surface of the centrum and extend onto the neural arch. They project posterolaterally, as is the case in most titanosauriforms ( Mannion & Calvo, 2011), although their extension beyond the posterior margin of the non-condylar centrum might be an artefact of anteroposterior crushing (see Cd II). In anterior view, the caudal ribs are triangular, tapering distally ( Fig. 9A), in contrast to the wing-like caudal ribs that characterize the anteriormost caudal vertebrae of most diplodocoids ( Upchurch, 1998; Whitlock et al., 2011). The ventral margins of the caudal ribs are dorsally deflected along their medial sections in anterior view, but become close to horizontal laterally. The posterior surface of the neural arch, ventrolateral to each postzygapophysis, forms a large, sharp-lipped postzygapophyseal centrodiapophyseal fossa (POCDF; Fig. 9D). Postzygapophyses have flat to very mildly concave articular surfaces that face ventrolaterally. Each postzygapophysis is supported ventrally by a dorsomedially directed CPOL that also forms the lateral and dorsal margins of the posterior neural canal opening. There is no hyposphene, an absence that characterizes most somphospondylans ( Upchurch, 1998; Mannion et al., 2013). A sharp PRDL is present, and there are remnants of a PODL on the right side. A sharp-lipped SDF is present above the PODL on the lateral surface of the neural spine ( González Riga, 2003), a feature that also characterizes several other titanosaurs, including Alamosaurus , Dongyangosaurus , Futalognkosaurus and Malawisaurus ( Wilson, 2002; Whitlock et al., 2011). The neural spine projects mainly dorsally, and slightly posteriorly. It is tall relative to the centrum height (ratio of 1.25), contrasting with most other titanosauriforms, although similar ratios are present in Futalognkosaurus ( Calvo et al., 2008b: fig. 16) and Saltasaurus ( Powell, 2003: pl. 33). There is no SPDL, but the SPRL and SPOL merge along the dorsal half of the neural spine to form a laterally thickened apex. A similar morphology is present in the anterior caudal vertebrae of several other titanosaurs, including Futalognkosaurus and Malawisaurus (González Riga et al., 2009) . The presence of the SPRL on the lateral surface of the neural spine, ventral to this lateral thickening, appears to be a result of the anteroposterior crushing of the caudal vertebra: this lateral extent is otherwise known only in diplodocoids ( Wilson, 2002). It could be argued that the SPRL is actually an SPDL, as interpreted by Carballido et al. (2017), but there seems to be a clear change in orientation at the prezygapophyseal level between the lower part of this lamina (our PRDL) and the upper part (our SPRL), even accounting for crushing. As such, we regard both Mendozasaurus and Futalognkosaurus (MUCPv-323: P.D.M., pers. observ. 2009; contra Carballido et al., 2017) as lacking a SPDL in anterior caudal vertebrae. In contrast, Patagotitan seems to have a continuous lamina extending from the caudal rib up to the neural spine, and we agree with Carballido et al. (2017) in identifying this as a SPDL. The SPOLs of Cd I of Mendozasaurus form the posterolateral margins of the neural spine, with a postspinal fossa in between. A ridge-like midline postspinal lamina is present and seems to extend for most of the spine length, although dorsally it becomes transversely widened and more rugose. The dorsal margin of the neural spine lacks the trifid morphology that characterizes the anterior caudal vertebrae of several somphospondylans ( D’Emic et al., 2013), including Futalognkosaurus ( Calvo et al., 2008b) .

Cd II preserves most of the centrum and the base of the arch, including the prezygapophyses, but has been transversely compressed ( Fig. 9E–G). It reveals that the internal tissue structure of the vertebra is noncamellate, contrasting with the camellae that pneumatize the anteriormost caudal vertebrae of several lithostrotians ( Wilson, 2002; Mannion et al., 2013). Caudal ribs project posterolaterally, but do not extend beyond the posterior margin of the non-condylar centrum, contrasting with the condition in the titanosaurs Andesaurus ( Mannion & Calvo, 2011) and Notocolossus ( González Riga et al., 2016) , as well several basal titanosauriforms ( Mannion & Calvo, 2011; D’Emic, 2012; Mannion et al., 2013). There is a tubercle on the dorsal surface of the caudal rib ( D’Emic et al., 2013), approximately at the midpoint between the prezygapophysis and the distal end of the rib ( Fig. 9E, F). A comparable feature has previously been noted in several other somphospondylan taxa, including Baurutitan , Epachthosaurus and Huabeisaurus ( Martínez et al., 2004; Kellner, Campos & Trotta, 2005; D’Emic et al., 2013), but is also present in a wider array of eusauropods ( Poropat et al., 2016). The prezygapophyses project anterodorsally and extend well beyond the anterior margin of the centrum.

Cd III lacks the posterior surface of the neural arch and spine, and the latter is also incomplete dorsally ( Fig. 9H–J). It demonstrates the presence of a midline prespinal lamina, in addition to well-developed SPRLs. Only the right half of the vertebra of Cd IV is preserved, although the neural spine is largely complete ( Fig. 9K–M). A sharp-lipped POCDF remains present ( González Riga, 2003). There is a prominent tubercle (‘SPRL-process’ sensu D’Emic et al., 2013) on the dorsal margin of the SPRL, close to the prezygapophysis ( González Riga, 2003), that is best observed in lateral view ( Fig. 9L; note that this feature might also be subtly present on Cd III). A comparable, although often less prominent, tubercle is present in several other titanosauriforms, including Alamosaurus , Giraffatitan and Saltasaurus ( González Riga, 2003; D’Emic & Wilson, 2011; D’Emic, 2012). The neural spine of Cd IV projects posterodorsally. In this regard, Mendozasaurus differs from Epachthosaurus , in which there is a reverse shift, from posterodorsally to vertically oriented neural spines in the anteriormost caudal vertebrae ( Martínez et al., 2004). In anterior view, the neural spine of Cd IV of Mendozasaurus is transversely expanded dorsally.

Cd V is complete apart from the neural spine, but is heavily distorted, such that the vertebra is strongly sheared anteriorly. The centrum is procoelous, the caudal ribs curve posterolaterally, and a POCDF is present, as in preceding caudal vertebrae. The ventrolateral surfaces of the prezygapophyses are thickened, a feature that continues into at least the next few caudal vertebrae (Cd VI–VIII), in which this thickening becomes a prominent swelling ( Fig. 9H, I, N). A similar feature is present in the anterior caudal vertebrae of Diplodocus , although in that taxon the expansion forms a distinct ridge ( Tschopp, Mateus & Benson, 2015). We consider this as a convergent feature and regard the swelling as an autapomorphy of Mendozasaurus . Cd VI ( Fig. 9N–Q) and Cd VII ( Fig. 9R) are less deformed than Cd V, although Cd VII is missing the right side of the centrum, the right prezygapophysis, and the posterior condyle. Both vertebrae demonstrate the retention of a small, rugose tubercle on the dorsal surface of the SPRL. In Cd VII, the neural spine is reduced to a simple, transversely compressed plate that projects dorsally and posteriorly.

The centrum of Cd VIII is the first caudal vertebra that lacks strong procoely, with a concave anterior articular surface ( Fig.9S), and an irregularly concave posterior surface that is slightly convex dorsally ( Fig. 9T). In the middle and posterior centra, the posterior articular surfaces are practically planar, with the exception of the reduced condyles that are dorsally displaced ( González Riga, 2003). This morphology is different to most lithostrotians [e.g. Narambuenatitan ( Filippi, García & Garrido, 2011) ; Rinconsaurus ( Calvo & González Riga, 2003) ; Overosaurus ( Coria et al., 2013) ], in which a prominent condyle is retained ( Fig. 10 View Figure 10 ). There are no excavations or ridges on the lateral surface of the centrum of Cd VIII of Mendozasaurus , and its ventral surface is transversely convex. As such, the caudal centra of Mendozasaurus lack the ventrolateral ridges and midline hollow that characterize many somphospondylans and diplodocoids ( Upchurch, 1998; Wilson, 2002; Mannion et al., 2013). The caudal ribs (= transverse processes) curve posterolaterally, although they do not extend as far as the posterior margin of the centrum. In contrast, Notocolossus exhibits well-developed and ventrally curved transverse processes ( González Riga et al., 2016: fig. 3). The prezygapophyses of Cd VIII of Mendozasaurus project anteriorly and slightly dorsally and extend well beyond the anterior margin of the centrum. Cd IX is very incomplete and poorly preserved, and the centrum of Cd X has a small posterior convexity that is dorsally restricted.

Cd XI and Cd XII are fairly complete and articulated, although the former is poorly preserved and more distorted than the latter. The posterior articular surface of the centrum of Cd XII is irregular, and caudal ribs are absent on both vertebrae, with the exception of a very small bulge-like process around the arch-centrum junction. As such, we regard these as some of the first middle caudal vertebrae ( Fig. 9 AD). Anterodorsally projecting prezygapophyses extend well beyond the anterior margin of the centrum. There is a shallow fossa on both lateral surfaces at the base of arch, ventrolateral to the postzygapophyses. The anterior margin of the neural spine is subvertical; unfortunately, the posterior margin is incomplete. Cd XIII is missing most of the anterior half of the centrum and the left prezygapophysis. The dorsal half of the posterior articular surface of the centrum of Cd XIII is gently convex, whereas the ventral half is concave. As in the middle caudal vertebrae of other titanosauriforms ( Calvo & Salgado, 1995), the neural arch is restricted to the anterior half of the centrum.

The ventrolateral surfaces of the anteriorly projecting prezygapophyses lack a raised ridge-like expansion in Cd XIV and subsequent caudal vertebrae. In lateral view, the anterior margin of the neural spine is vertical, the dorsal margin is horizontal and the posterior margin slants such that it faces posteroventrally. This morphology was described by González Riga (2003: fig. 5) as an autapomorphic character of Mendozasaurus and is considered valid herein. Some titanosaurs exhibit laterally compressed and anteroposteriorly elongated middle caudal neural spines, including Andesaurus , Dreadnoughtus , Epachthosaurus , Malawisaurus and Narambuenatitan ( Fig. 10 View Figure 10 ). However, the shape of this neural spine is different to that of Mendozasaurus . The neural spines of Epachthosaurus form an obtuse angle at their anterodorsal corner ( Fig. 10D View Figure 10 ) ( Martínez et al., 2004: fig. 8). In Andesaurus ( Fig. 10E View Figure 10 ) and Malawisaurus ( Fig. 10F View Figure 10 ), the anterodorsal corner is rounded and the dorsal margin is slightly convex ( Jacobs et al., 1993: fig. 2; Mannion & Calvo, 2011: fig. 6). In Narambuenatitan ( Fig. 10G View Figure 10 ), the dorsal border is not horizontal: it is higher posteriorly ( Filippi et al., 2011: fig. 9). Finally, in Dreadnoughtus ( Fig. 10B View Figure 10 ), the shape of the neural spine is different from that of Mendozasaurus and changes along the caudal series ( Lacovara et al., 2014: supp. fig. 1). In the anteriormost middle caudal vertebrae (Cd XI–XVII) of Dreadnoughtus , the anterodorsal border is sharply pointed and anteriorly projected, extending, in some cases, beyond the anterior margin of the centrum. In contrast, from Cd XVIII to XXII, the neural spines of Mendozasaurus are posteriorly orientated.

Cd XVI of Mendozasaurus ( Fig. 9U–X) has a shallow prespinal fossa at the base of the neural spine ( Fig. 9U). Chevron facets are incomplete in Cd XVI, but the posterior ones are prominent structures that are widely separated from one another along the midline. Cd XVIII ( Fig. 9Y –AB) and subsequent caudal vertebrae (Cd XIX–XXII) are from the posterior region of the tail. In these vertebrae, a prominent, anteroposteriorly elongate ridge is retained at the arch-centrum junction, and the posterior articular surface of the centrum still has a dorsally restricted convexity.

CHEVRONS

Three fairly complete chevrons (IANIGLA-PV 065/23–25) from the proximal part of the tail are preserved ( Fig. 11 View Figure 11 ; see Table 3 for measurements), as well as four fragments (IANIGLA-PV 065/26–30). Although it is not possible to directly match them to particular caudal vertebrae, these probably belong to the series of nine partially articulated anterior–middle caudal vertebrae (Cd V–XIII). All three chevrons are proximally unbridged, as characterizes most macronarians ( Upchurch, 1995), and their proximal articular surfaces have a midline furrow, such that they are separated into distinct anterior and posterior portions ( González Riga, 2003). This proximal morphology also characterizes the titanosaurs Aeolosaurus ( Powell, 1987; Santucci & Arruda-Campos, 2011), Epachthosaurus (UNPSJB-PV 920: P.D.M., S.F.P., pers. observ. 2013) and Maxakalisaurus ( Kellner et al., 2006) , as well as some euhelopodids ( D’Emic, 2012). Haemal canal depth is between 40 and 45% of total chevron proximodistal height, as is the case in many other titanosauriforms ( Wilson, 2002; Mannion et al., 2013).

The lateral surfaces of the proximal rami lack ridges, such as those seen in some titanosaurs (e.g. Alamosaurus , Epachthosaurus , Saltasaurus ; Poropat et al., 2016), although these ridges are not always

All measurements are in millimetres.

present in the anteriormost chevrons. There are also no ridges on the lateral surfaces of the transversely thin distal blades of the chevrons of Mendozasaurus , but sharp anterior and posterior midline ridges are present. In lateral view, the distal blade curves posteriorly, with only a subtle anteroposterior expansion occurring distally, and it has a convex distal margin.

SCAPULA

The right scapula (IANIGLA-PV 068; see Table 4 for measurements) is here described with the long axis of the scapular blade oriented horizontally ( Fig. 12A, B View Figure 12 ). The acromion (proximal plate) is damaged dorsally and

All measurements are in millimetres.

*Measurements have been affected by crushing.

anteroventrally, and the dorsal margin of the blade is missing at its very distal end ( Fig. 12A View Figure 12 ). Although incomplete, the coracoid articular surface does not seem to have been strongly tilted posteriorly, relative to the long axis of the scapular blade, contrasting with the condition observed in many diplodocoids ( Tschopp et al., 2015), several derived titanosaurs ( Wilson, 2002), and a number of taxa close to the titanosaurian radiation, for example Ligabuesaurus ( Bonaparte et al., 2006) . As in other somphospondylans ( Wilson, 2002), the glenoid is bevelled medially. The acromial ridge is an anteroposteriorly wide, rounded ridge that posteriorly bounds the excavated lateral surface of the acromion ( Fig. 12A View Figure 12 ). There is no fossa or excavation posterior to the acromial ridge, such as that seen in a number of neosauropods ( Upchurch et al., 2004). As preserved, the posterodorsal margin of the acromion is straight. Although damaged, there is evidence for a ventral process at the posterior end of the acromion ( D’Emic, Wilson & Williamson, 2011). A comparable feature is present in several other titanosauriforms ( Fig. 13 View Figure 13 ), for example Alamosaurus (D’Emic et al., 2011) , Chubutisaurus ( Carballido et al., 2011a) , Dreadnoughtus ( Ullmann & Lacovara, 2016) , Ligabuesaurus ( Bonaparte et al., 2006) and Paralititan ( Smith et al., 2001) .

The lateral surface of the scapular blade is dorsoventrally convex, forming a rounded ridge that is ventrally biased ( Fig. 12A View Figure 12 ). This ridge fades out at approximately midlength of the scapular blade, from which point the lateral surface is relatively flat. In contrast, the medial surface of the scapular blade is gently concave dorsoventrally. As such, the cross section at the base of the scapular blade is closest to the ‘D’ shape of most eusauropods, differing from the subrectangular shape that characterizes many somphospondylans ( Wilson, 2002), although taxa such as Diamantinasaurus ( Poropat et al., 2015b) , Ligabuesaurus ( Bonaparte et al., 2006) , Opisthocoelicaudia ( Borsuk-Białynicka, 1977) and Patagotitan ( Carballido et al., 2017) , also have this D-shaped cross section. The scapular blade is transversely thicker ventrally than dorsally at its base, a morphology that has also been reported in other titanosaurs (e.g. Alamosaurus ; D’Emic et al., 2011). There are no ridges on the medial surface of the proximal portion of the distal blade ( Fig. 12B View Figure 12 ), differing from those seen in a small number of derived titanosaurs ( Sanz et al., 1999). There is also no ventral process towards the anterior end of the scapular blade, in contrast with some somphospondylans, for example Alamosaurus and Euhelopus (D’Emic et al., 2011; D’Emic, 2012; Mannion et al., 2013). The scapular blade expands dorsoventrally towards its distal end, and its ventral margin is concave in lateral view.

STERNAL PLATE

The preserved sternal plate (IANIGLA-PV 067; Fig. 12C View Figure 12 ) is a right, rather than left element, as originally identified ( González Riga, 2003). Its lateral margin is incomplete, meaning that it was probably more strongly concave than it presently appears (see Table 4 for measurements). The medial margin is convex, and there is a prominent ridge at the anterior end of the ventral surface, situated on the lateral margin. A similar ridge is present in a wide range of eusauropods ( Sanz et al., 1999; Upchurch et al., 2004; Poropat et al., 2016). As noted by González Riga (2003), the posterior margin is straight in dorsal view, lacking the convexity that characterizes the sternal plates of most sauropods. Mendozasaurus shares this posterior morphology with a small number of other titanosaurs, including Alamosaurus and Malawisaurus ( González Riga, 2003) . Assuming that the sternal plate is from the same individual as the two humeri (see below), the ratio between its maximum length and the proximodistal length of the humerus is between 0.75 and 0.78, similar to other derived titanosaurs, for example Alamosaurus and Opisthocoelicaudia ( Upchurch, 1998) .

HUMERUS

A right (IANIGLA-PV 069/1) and a left (IANIGLA-PV 069/2) humerus, probably from the same individual,

are preserved ( Fig. 14 View Figure 14 ; see Table 4 for measurements). Both are relatively complete: the right humerus is better preserved ( Fig. 14A–F View Figure 14 ), whereas the left humerus is slightly less distorted ( Fig. 14G–J View Figure 14 ).

The humerus is relatively slender throughout its length, with low ratios for the mediolateral widths of the proximal (<0.35), midshaft (<0.15) and distal ends (<0.30), relative to that of the proximodistal length of the humerus. As such, the humerus of Mendozasaurus is closer in morphology to the humeri of taxa like Ligabuesaurus and Rapetosaurus , rather than the robust forelimb elements that characterize many saltasaurids ( Curry Rogers, 2005), as well as titanosaurs such as Diamantinasaurus ( Poropat et al., 2015b) and Dreadnoughtus ( Ullmann & Lacovara, 2016) . If the femur (IANIGLA-PV 073/4) is from the same individual as the two humeri, then Mendozasaurus has a low humerus to femur length ratio of between 0.72 and 0.75, similar to basal eusauropods and a small number of derived titanosaurs, such as Jainosaurus ( Wilson, Barrett & Carrano, 2011a) and Opisthocoelicaudia ( Borsuk-Białynicka, 1977) . However, other titanosaurian taxa have higher ratios, for example Rapetosaurus [0.80 ( Curry Rogers, 2009)], Dreadnoughtus [0.84 ( Lacovara et al., 2014)] and Epachthosaurus [0.85 ( Martínez et al., 2004)], which might suggest that the humeri of Mendozasaurus are from a smaller individual than the femora.

There is no strong degree of torsion between the proximal and distal ends, and the proximal margin is gently convex in anterior view ( Fig. 14B, G View Figure 14 ), lacking the well-developed process for M. supracoracoideus that creates a sinuous outline in some titanosaurian taxa ( Upchurch, 1998; González Riga, 2003; González Riga et al., 2009; González Riga & Ortiz David, 2014). As is typical in most somphospondylans ( Wilson, 2002; Mannion et al., 2013), the proximolateral corner is square shaped. The proximomedial corner forms an acute, triangular projection, similar to that illustrated in Paralititan ( Smith et al., 2001: fig. 2a), and previously considered an autapomorphy of Angolatitan by Mateus et al. (2011). The humerus of Mendozasaurus lacks the extreme proximomedial expansion that characterizes that of Notocolossus ( González Riga et al., 2016) . As is the case in other titanosauriforms ( Poropat et al., 2016), the proximal end is asymmetrical, with no notable expansion of the lateral margin relative to the shaft.

The humeral head extends onto the posterior surface as a prominent projection in Mendozasaurus , although crushing has obscured its morphology. There is no ridge along the lateral margin of the proximal third of the posterior surface, such as that recognised in several titanosaurs [e.g. Epachthosaurus , Notocolossus and Saltasaurus ( González Riga et al., 2016; Poropat et al., 2016)]. Although there is some evidence for such a ridge on the right humerus, this has been caused by crushing; the left humerus confirms its absence. As in many titanosauriforms ( Upchurch, Mannion & Taylor, 2015), a prominent bulge for M. scapulohumeralis anterior is present, but there is no equivalent site for M. latissimus dorsi , the presence of which seems to characterize saltasaurids ( Otero, 2010; D’Emic, 2012). Although the latter muscle scar was considered present in Patagotitan ( Carballido et al., 2017) , we interpret this as more likely to represent the site for M. scapulohumeralis anterior based on comparisons with other titanosauriforms.

The deltopectoral crest of the right humerus projects strongly medially, but this has been almost certainly accentuated by crushing. That of the left humerus projects anteromedially and is likely to be closer to the genuine orientation of the deltopectoral crest. A medially deflected deltopectoral crest characterizes many titanosauriforms ( Mannion et al., 2013). Distally, the deltopectoral crest doubles in mediolateral thickness, a feature previously recognised in some saltasaurids ( Wilson, 2002), but that is also present in several more basal titanosaurs ( Fig. 15 View Figure 15 ), including the Argentinean taxa Muyelensaurus (MAU-PV-LL-70: P.D.M., pers. observ. 2014), Narambuenatitan (MAU-PV-N-425: P.D.M., pers. observ. 2014) and Rinconsaurus (MAU- PV-CRS-47: P.D.M., pers. observ. 2014). A tubercle for attachment of the M. coracobrachialis is present on the anterior surface of the proximal third. At midshaft, the humerus has an elliptical cross section, with a mediolateral to anteroposterior width ratio of ~2.0. The lateral margin of the midshaft is straight in anterior view.

The lateral half of the anterior surface of the distal end has a clearly divided condyle ( Fig. 14C, G, H View Figure 14 ). In this regard, Mendozasaurus differs from nearly all other titanosaurs in which this condyle is undivided ( D’Emic, 2012; Mannion et al., 2013), with the exception of Diamantinasaurus ( Poropat et al., 2015b) , with which it shares this reversal to the plesiomorphic sauropod state. We regard this feature as a local autapomorphy of Mendozasaurus . A well-developed supracondylar fossa is bound by medial and lateral ridges on the posterior surface of the distal end ( Fig. 14E, J View Figure 14 ), as is the case in most somphospondylans ( Mannion & Calvo, 2011). The undivided distal articular surface does not expand strongly onto the anterior surface of the humerus, contrasting with the condition in some saltasaurids ( Wilson, 2002). There is some bevelling of the distal end, with the medial condyle extending further distally than the lateral one. Comparable bevelling is present in the humeri of at least some other titanosaurs, including Saltasaurus (PVL 4017–63: P.D.M., S.F.P., pers. observ. 2013) and Neuquensaurus (MLP CS 1050: P.D.M., pers. observ. 2013).

RADIUS

The right radius (IANIGLA-PV 070/2; Fig. 16A–D View Figure 16 ) is complete aside from a portion of the medial half of the shaft, but is broken into two pieces, and has undergone some distortion, particularly at the proximal end, as well as anteroposterior compression (see Table 4 for measurements).

In anterior view ( Fig. 16B View Figure 16 ), the lateral margin is concave, whereas the medial margin is gently sinuous. Although distorted, the proximal end clearly becomes anteroposteriorly narrow medially, forming a distinct medial projection. The element is too crushed and damaged to determine whether a ridge for attachment of M. biceps brachii and M. brachialis inferior (see Upchurch et al., 2015) was present on the medial surface of the proximal end.

A posterolateral ridge extends along most of the radius length ( Fig. 16D View Figure 16 ), beginning a short distance from the proximal end, as is the case in many titanosaurs, as well as a few more basal taxa ( Curry Rogers, 2005; Mannion et al., 2013). There is evidence for a second, parallel ridge along the distal third of the posterior surface, medial to the posterolateral ridge, with a shallow groove separating the two ridges ( Fig. 16D View Figure 16 ). Although some other titanosauriform taxa also possess two ridges, in those taxa the second ridge is restricted either to the anterolateral margin (e.g. Diamantinasaurus ; Poropat et al., 2015b) or to the distal quarter of the radius (e.g. Muyelensaurus ; MAU- PV-LL-71: P.D.M., pers. observ. 2014). As such, the presence of a second ridge along the distal third of the radius is tentatively regarded as an autapomorphy of Mendozasaurus .

Although distorted, the distal end was clearly bevelled, with this bevelling restricted to the lateral two thirds of the distal surface; however, poor preservation means that it is not possible to determine the angle of bevelling. A gentle concavity is situated on the posterior margin of the distal end ( Fig. 16D View Figure 16 ), approximately equidistant from the medial and lateral margins. The distal end is also mediolaterally wider than the proximal end, a feature that Mendozasaurus shares with several macronarians ( Curry Rogers, 2005; Mannion et al., 2013), including the titanosaurs Patagotitan ( Carballido et al., 2017) and Rapetosaurus ( Curry Rogers, 2009) .

ULNA

The right ulna (IANIGLA-PV 070/1; Fig. 16E–H View Figure 16 ) is missing the proximal and distal articular surfaces, as well as a large amount of the posterior surface of the proximal third (with no posterior process preserved) (see Table 4 for measurements). As such, it is not possible to determine the nature of the olecranon process, or whether the articular surface of the anteromedial process was concave. The anteromedial and anterolateral processes form a right angle to one another in proximal view, with a well-developed radial fossa. All three proximal processes continue distally as rounded ridges.

Distally, the ulna is posteriorly expanded, as is the case in most sauropods, but contrasting with several titanosauriforms with unexpanded distal ulnae, for example Alamosaurus , Giraffatitan and Saltasaurus ( D’Emic, 2012; Mannion et al., 2013). There is a very gentle fossa on the anteromedial surface of the distal end, for reception of the radius. In distal view, the ulna has a subtriangular or semicircular outline, with a flat anteromedial margin. A similar morphology has been documented in several other titanosaurs, including Diamantinasaurus , Epachthosaurus and Saltasaurus ( Upchurch et al., 2015) .

METACARPUS

A total of six metacarpals are preserved ( Figs 17 View Figure 17 , 18; see Table 5 for measurements). IANIGLA-PV 071/1–5 potentially represent metacarpals I–V of manüs of one individual ( Figs 17 View Figure 17 , 18A –AC), although all of them have undergone crushing. There is also a metacarpal of a smaller individual (IANIGLA-PV 154; Fig. 18 AD– AI). The position of several elements within the manus is also revised, with three of the four metacarpals

All measurements are in millimetres.

originally described by González Riga (2003) re-identified. All metacarpals are described as if held horizontally, with the long axis of the distal end oriented transversely. No carpal or manual phalangeal elements are preserved. Although we cannot rule out that their absence might be preservational, the titanosaurian affinities of Mendozasaurus mean that it probably genuinely lacked these elements in vivo.

If we assume that IANIGLA-PV 071/1–5 are from a single individual, then metacarpal IV would be the longest metacarpal in the manus, with metacarpals I–III subequal in length, and metacarpal V the shortest. This would be potentially autapomorphic, as in other sauropods one of metacarpals I–III is the longest in the manus ( Curry Rogers, 2005; Poropat et al., 2015a); however, because of uncertainty in the number of individuals, we regard this as only a tentative autapomorphy of Mendozasaurus , pending the discovery of an articulated manus. Although the metacarpals are distorted, the metacarpus clearly would have formed a semi-tubular, ‘U’-shaped outline in proximal view ( Fig. 17 View Figure 17 ).

There are two elements that can definitely be identified as metacarpal I from the left manus [IANIGLA-PV 071/4 ( Fig.18A–E) and 154 ( Fig. 18 AD–AI)]. Metacarpal I has a D-shaped outline in proximal view ( Fig. 18D, AE), with a mildly concave ventrolateral margin and a dorsolateral projection. The proximal and distal ends are twisted relative to one another. There is no lateral bowing of the metacarpal in dorsal view, contrasting with the morphology observed in the titanosaurs Andesaurus and Argyrosaurus ( Apesteguía, 2005) . At approximately two thirds of the length from the proximal end, the dorsolateral margin forms a proximodistally short, sharp ridge or tubercle ( Fig. 18E) that we consider an autapomorphy of Mendozasaurus . In distal view, the metacarpal is dorsoventrally taller along its lateral half, and the ventral margin is gently concave ( Fig. 18C, AI). The distal end is not bevelled relative to the long axis of shaft, and there are no distinct distal condyles. As in nearly all titanosauriforms ( Salgado et al., 1997; D’Emic, 2012), the distal articular surface does not extend onto the dorsal surface in any of the metacarpals.

Only one metacarpal II is preserved (IANIGLA-PV 071/3; Fig. 18F–K), and it is from a right manus. The proximal end is incomplete dorsally, and the element has undergone more deformation than the other metacarpals. The metacarpal decreases in dorsoventral height distally, and its ventral margin is dorsally bowed, although the latter might be a preservational artefact. A rounded, ventral ridge extends distally from the proximal end and is deflected medially; it disappears a short distance before the distal end. There is also evidence for a tubercle on the dorsomedial margin of the distal end, but this area is poorly preserved. In distal view, the metacarpal is slightly dorsoventrally taller along its medial margin and overall has a transversely elongate, trapezoidal outline. None of the metacarpals possess the ventral ‘channel’-like concavities described in Argyrosaurus ( Mannion & Otero, 2012) .

IANIGLA-PV 071/1 ( Fig. 18L–Q) is a right metacarpal III [illustrated by González Riga & Astini (2007) as metacarpal IV?]. In proximal view, the metacarpal is wedge shaped ( Fig. 18O). The apex of this triangular shape continues distally as a laterally deflected ventral ridge, situated on the ventrolateral margin at the distal end. In distal view, the metacarpal has a trapezoidal outline, with medially slanted margins, and a very mildly concave ventral margin ( Fig. 18M).

In proximal view,the left metacarpal IV(IANIGLA-PV 071/2; Fig. 18R–W) seems to have a dorsoventrally tall ‘T’ shape ( Fig. 18V), lacking the ‘chevron’ shape that characterizes this element in brachiosaurids ( D’Emic, 2012) and some other sauropods ( Mannion et al., 2013). The ventral part of this ‘T’ shape continues as a ridge along the proximal two thirds of the ventral surface. Although poorly preserved and probably slightly incomplete distally, there is a dorsomedial flange along the distal third. In distal view ( Fig. 18U), the metacarpal has a transversely elongate trapezoidal outline, similar to Epachthosaurus ( Poropat et al., 2016) with a mildly concave lateral margin.

The left metacarpal V (IANIGLA-PV 071/5; Fig. 18X – AC) has a compressed ‘D’ shape in proximal view, with the flat margin of this ‘D’ facing ventromedially ( Fig. 18X). A ventral ridge extends distally from the proximal end and is deflected medially along its length; it becomes increasingly low and rounded distally and is present until at least close to the distal end (the very distal end of the ventral surface is incomplete). Although this does not form a medially biased flange-like swelling, such as that observed in Andesaurus and Epachthosaurus ( Apesteguía, 2005; Mannion & Calvo, 2011; Poropat et al., 2016), it might have been affected by crushing. A dorsomedial flange is present along the distal half ( Fig. 18 AB), similar to that present in Muyelensaurus (MAU-PV-LL-152: P.D.M., pers. observ. 2014) and Petrobrasaurus (MAU-Pv-PH-449: P.D.M., pers. observ. 2014); although this feature is less prominent in Mendozasaurus , this might just be a result of dorsoventral crushing. In distal view, the metacarpal decreases in dorsoventral height towards its medial margin, with this reduction entirely restricted to the ventral margin ( Fig. 18 AB).

FEMUR

Two femora are preserved (IANIGLA-PV 073/1 and 073/4; Fig. 19 View Figure 19 ; see Table 6 for measurements). Only the proximal half of the right femur (IANIGLA-PV

073/1) is preserved ( Fig. 19G View Figure 19 ) and has undergone anteroposterior compression. The left femur (IANIGLA-PV 073/4) is complete, although it is slightly crushed and a little poorly preserved in places ( Fig. 19A–F View Figure 19 ).

The femoral head projects mainly medially, lacking the dorsal deflection that characterizes some sauropods ( Upchurch et al., 2004; Curry Rogers, 2005). There is no evidence for a longitudinal ridge (linea intermuscularis cranialis) on the anterior surface of the shaft, contrasting with several derived titanosaurs ( Otero, 2010; D’Emic, 2012; Poropat et al., 2015b). In contrast to brachiosaurids and several additional taxa ( Mannion et al., 2013), the well-developed fourth trochanter is not visible in anterior view. As in nearly all eusauropods ( Upchurch, 1998), it is restricted to the medial margin of the posterior surface. Its distal tip is situated at approximately midlength of the femur. The lateral margin of the proximal end is deflected medially relative to the lateral margin of the shaft ( González Riga, 2003). This is the condition in most basal macronarians, but several derived titanosaurs lack this medial deflection ( Mannion et al., 2013). A trochanteric shelf appears to be present, a feature Mendozasaurus shares with most titanosaurs, as well as some taxa outside of Titanosauria ( Otero, 2010; Mannion et al., 2013).

At midshaft, the femur has an anteroposteriorly compressed elliptical cross section. Although anteroposteriorly longer than the fibular distal condyle, the tibial distal condyle is mediolaterally narrower, as is the case in many other titanosauriforms ( Wilson, 2002; Poropat et al., 2016). The fibular condyle is divided posteriorly into two well-developed condyles, but poor preservation means that we cannot be certain whether a ridge is present within this division, such as that seen in Diamantinasaurus and Magyarosaurus ( Poropat et al., 2015b) . As in many derived titanosaurs ( Wilson, 2002), the fibular condyle extends further distally than the tibial condyle. The distal articular surface is anteroposteriorly convex, but it is not possible to determine whether the distal condyles extended onto the anterior surface.

TIBIA

One left tibia (IANIGLA-PV 074/2) and three right (IANIGLA-PV 073/2, 073/3, 074/1) tibiae are preserved

( Fig. 20A–N View Figure 20 ). The sole left tibia belongs to a larger individual than the remaining tibiae (see Table 6 for measurements). Although IANIGLA-PV 073/2 is complete, it has undergone extreme anteroposterior compression, whereas IANIGLA-PV 074/1 is much less deformed, but is missing most of its distal end. The proximal end of the third tibia is missing, and the distal end is slightly incomplete. As such, anatomical information on the tibia of Mendozasaurus is limited.

The proximal end is anteroposteriorly longer than mediolaterally wide, and the prominent cnemial crest projects primarily anteriorly, curving slightly laterally. There is no tuberculum fibularis, but a small ‘second cnemial crest’ is present. Posterior to this, there is an anterolateral expansion of the proximal end, although this is not as pronounced as it is in Uberabatitan ( Salgado & Carvalho, 2008) .

FIBULA

A left (IANIGLA-PV 074/3) and a right (IANIGLA-PV 074/4) fibula, probably from the same individual, are preserved ( Fig. 20O–R View Figure 20 ; see Table 6 for measurements). The left element is largely complete, with the exception of small pieces missing in places, including part of the anterior margin of the proximal end. The right element is distally incomplete, and much of the medial surface is not preserved.

In lateral view ( Fig. 20O View Figure 20 ), the fibula is sinuous, a morphology that Mendozasaurus shares with a wide array of somphospondylans ( Canudo, Royo-Torres & Cuenca-Bescós, 2008; D’Emic, 2012). The medial surface is flat for most of the length of the fibula ( Fig. 20R View Figure 20 ), whereas the lateral surface is anteroposteriorly convex. Proximally, the medial surface is striated, with a weak ridge delimiting the ventral

All measurements are in millimetres.

*Measurements have been affected by crushing.

margin of this striated region, directed anteroventrally from the posterodorsal corner of the proximal end. As in most somphospondylans ( D’Emic, 2012; Mannion et al., 2013), an anteromedial crest is present at the proximal end. An anterolateral trochanter is also present ( González Riga, 2003): there is a ridge on the lateral surface, a short distance from the anterior margin, situated above the level of the lateral trochanter, with an associated groove anterior to this ridge. A comparable feature has also been noted in several other titanosaurs, including Jainosaurus ( Wilson et al., 2011a) , Laplatasaurus ( González Riga, 2003) and Uberabatitan ( Salgado & Carvalho, 2008: fig. 19g).

The lateral trochanter consists of a rugose area, comprising two parallel ridges, as is the case in many somphospondylans ( Upchurch, 1998; Powell, 2003; Mannion et al., 2013). It is restricted to the proximal half of the fibula. Although fairly well developed, it does not project beyond the lateral margin of the remainder of the fibula, in contrast with the hypertrophied lateral trochanters that characterize the fibulae of some derived titanosaurs, such as Laplatasaurus ( Powell, 2003) and Uberabatitan ( Salgado & Carvalho, 2008) .

The anterior margin of the distal third forms a ridge. Distally, the fibula expands strongly laterally, as well as medially and a little anteroposteriorly. There is no concavity on the medial surface of the distal end. In distal view, the fibula has a rounded, subtriangular outline, with the apex of this triangle pointing anteriorly, as characterizes many titanosaurs ( Upchurch et al., 2015).

ASTRAGALUS

The right astragalus (IANIGLA-PV 155; Fig. 21A–F View Figure 21 ) has undergone extreme dorsoventral compression (see Table 6 for measurements). It clearly decreases in dorsoventral height and anteroposterior length towards its medial margin, as in all derived eusauropods ( Upchurch, 1998), and differs from derived titanosaurs ( Wilson, 2002) in that it is not pyramidal. Little more anatomical information can be provided with confidence. No calcaneum is preserved, but we cannot be certain that its absence is genuine.

PES

A total of 12 metatarsals and ten pedal phalanges are preserved ( Figs 22–24 View Figure 22 ; see Tables 7 and 8 for measurements). Although we cannot be certain, it is possible that these are the pedal elements of just two individuals. It seems likely that IANIGLA-PV 077/1–5 represent metatarsals I–V of a right pes of one individual [with a left metatarsal V (IANIGLA-PV 153) also preserved], along with a complete set of right pedal phalanges (IANIGLA-PV 077/6–10, 078/1–2 and 079), and two left phalanges (IANIGLA-PV 077/11 and 077/12). IANIGLA-PV 077/12 is extremely proximodistally compressed ( Fig. 22 View Figure 22 AH–AM). The remaining pedal remains are from a larger individual and comprise right metatarsalsI(IANIGLA-PV100/1),III [IANIGLA-PV 100/3; interpreted by González Riga & Astini (2007) as a titanosaur metacarpal and numbered as IANIGLA-PV 100] and the proximal end of V (IANIGLA-PV 100/5), and left metatarsals I (IANIGLA-PV 100/2; very poorly preserved and incomplete), IV (IANIGLA-PV 100/4) and V (IANIGLA-PV 100/6).

If our interpretation of the number of individuals represented by pedal remains is correct, then metatarsal IV is the longest element in the metatarsus, followed by metatarsals III, V, II and I. Although metatarsal IV is often the longest element in the pes (see González Riga et al., 2016: table 2), there is variation between taxa, for example metatarsal III is the longest in Opisthocoelicaudia ( Borsuk-Białynicka, 1977) . Furthermore, the order of decreasing size appears to be highly variable. Following our reconstruction of the pes, the phalangeal formula of Mendozasaurus is 2-2- 2-2-0 ( González Riga et al., 2016). The same formula has been described in several other titanosaurs, including the ‘Invernada titanosaur’ ( González Riga, Calvo & Porfiri, 2008) and Notocolossus ( González Riga et al., 2016) , and fits the general trend of increased pedal phalangeal loss in titanosaurs compared with other sauropods ( Upchurch, 1998; Wilson & Sereno, 1998; Bonnan, 2005; Nair & Salisbury, 2012; González Riga et al., 2016).

The proximal end of metatarsal I is ‘D’ shaped, with an approximately flat lateral margin, and a pointed dorsolateral projection ( Fig. 23D, AN). Whereas the proximal end is not bevelled relative to the long axis of the shaft, the distal end is, as a result of the lateral distal condyle extending further distally than the medial one, which is the condition in most eusauropods ( Wilson, 2002). There are no foramina on the dorsal surface, nor is there a rugosity on the dorsolateral margin of the distal end of the metatarsal (or any subsequent metatarsals), such as that seen in many diplodocoids ( Upchurch, 1998). The distal end lacks a distinct ventrolateral process and does not extend further laterally than the proximal end. In this regard, the first metatarsal of Mendozasaurus differs from those of most diplodocoids ( Upchurch, 1998) and several other eusauropods ( Mannion et al., 2013), including a number of titanosauriforms (D’Emic et al., 2011). In distal view, the metatarsal has a semicircular outline ( Fig. 23B, AL).

The proximal end of metatarsal II is a little deformed, but its long axis is oriented dorsoventrally, with a mildly concave lateral margin ( Fig. 23J). The distal end is bevelled, such that the lateral distal condyle extends further distally than the medial one. In distal view, the metatarsal has an almost semicircular outline, with a mildly concave ventral margin and a convex dorsal margin ( Fig. 23H).

In proximal view, metatarsal III has a dorsoventrally compressed trapezoidal outline ( Fig. 23P, AU), tapering dorsoventrally along the ventromedial margin to a small point. The proximal articular surface is gently ‘domed’. In dorsal view, the medial margin of the metatarsal is strongly concave ( Fig. 23M, AR). As in metatarsals I and II, the distal end of metatarsal III is bevelled. In distal view, the metatarsal is slightly taller dorsoventrally at its medial than lateral margin ( Fig. 23N, AS), and the distal articular surface extends onto the dorsal surface, with a medial bias. There is no midline foramen on the ventral surface, close to the distal end, such as that observed in Epachthosaurus (UNPSJB-PV 920: P.D.M., S.F.P., pers. observ. 2013).

Metatarsal IV is trapezoidal shaped in proximal view ( Fig. 23V, BA), with the ventral margin the shortest, and a ventrolaterally facing lateral margin. In contrast with several titanosauriforms (D’Emic et al., 2011; D’Emic, 2012), there is no medial embayment for reception of metatarsal III. In dorsal view ( Fig. 23S, AX), the lateral margin of the metatarsal is concave, with a mildly concave medial margin. In distal view ( Fig. 23T, AY), the metatarsal has a transversely elongate, elliptical outline, with a flattened medial margin, and very subtly concave ventral margin. The distal articular surface is dorsoventrally convex, especially towards the ventral margin.

In dorsal view, metatarsal V is funnel shaped ( Fig. 23Y, AE, BD, BI), with only a slight transverse expansion of the distal end relative to the midshaft. The proximal end is dorsally expanded relative to the shaft ( Fig. 23 AB, AH, BF, BL), contrasting with the dorsoventrally compressed proximal ends seen in the fifth metatarsals of some derived titanosaurs, for example Alamosaurus and Saltasaurus ( Poropat et al., 2016) . At the medial margin, the proximal end thins dorsoventrally to form a small flange, presumably for articulation with metatarsal IV. The lateral and medial margins of the proximal third are concave in dorsal view, with this concavity more pronounced along the lateral side. The ventral surface is transversely concave proximally, and mainly flat distally, and there is a rugose tubercle at approximately midlength, situated close to the lateral margin. A comparable ventral tubercle (or ridge) is present on the fifth metatarsals of the titanosaurs Epachthosaurus (UNPSJB-PV 920: P.D.M., S.F.P., pers. observ. 2013), Neuquensaurus (MLP CS 1180: P.D.M., pers. observ. 2013) and Saltasaurus (PVL 4017–121: P.D.M., S.F.P., pers. observ. 2013) (see also Poropat et al., 2016). The distal end of metatarsal V of Mendozasaurus thickens dorsoventrally towards its lateral margin, where there is also a slight lateral expansion.

Phalanx I-1 (IANIGLA-PV 077/6) is complete and only slightly distorted ( Fig. 24A–F). In proximal view ( Fig. 24D), it has an approximate ‘D’ shape, with a gently convex dorsal margin. The proximal articular surface is irregularly flat. Both the dorsal ( Fig. 24A) and ventral ( Fig. 24E) surfaces are anteroposteriorly concave, but whereas the dorsal surface is transversely convex, the ventral surface is fairly flat transversely. The distal end is very slightly bevelled, as a result of the medial margin being very slightly longer proximodistally than the lateral margin. In distal view ( Fig. 24B), the phalanx is dorsoventrally tallest along its medial margin, and the distal articular surface is dorsoventrally convex. A left phalanx I-1 (IANIGLA-PV 077/12; Fig. 24 AH–AM) has been strongly crushed anteroposteriorly.

Specimens 077/1–4 have all undergone significant distortion. All measurements are in millimetres.

*Measurements have been affected by crushing.

† Measurements are incomplete.

The proximal end of phalanx II-1 (IANIGLA-PV 077/7) has a transversely wide semicircular outline ( Fig. 24G–L), with a fairly flat ventral margin ( Fig. 24J). Its proximal articular surface is flat, and the phalanx is slightly taller dorsoventrally along its medial margin compared to its lateral margin. Whereas the dorsal surface of the phalanx is transversely convex and gently concave proximodistally, the ventral surface is very mildly concave in both directions. In dorsal view ( Fig. 24G), the medial margin is more strongly concave than the lateral margin, and the medial distal condyle extends very slightly further distally than the lateral condyle. The distal end has a similar morphology to that of the proximal end, but is dorsoventrally shorter ( Fig. 24H). The distal articular surface is convex dorsoventrally, especially along its medial half, where it extends prominently onto the dorsal and ventral surfaces. A left phalanx II-1 ( Fig. 24 AN– AS) is also preserved.

Phalanx III-1 (IANIGLA-PV 077/8) is complete, but has undergone some deformation, such that the distal end curves upwards ( Fig. 24A–F). It decreases in dorsoventral height towards its lateral margin. In proximal view ( Fig. 24D), the phalanx has a transversely elongate D shape, with a flat ventral margin (note that the apparent ventral concavity is a result of breakage). The proximal articular surface is fairly flat, and there is no proximoventral projection. The ventral surface is gently concave in both directions and lacks foramina ( Fig. 24E). There is no well-defined separation of the distal end into distinct condyles, and no distal bevelling ( Fig. 24B).

Phalanx IV-1 (IANIGLA-PV 077/9) has a transversely elongate, semicircular proximal outline, with a flat ventral margin ( Fig. 24 AB–AG). The proximal articular surface is fairly flat ( Fig. 24 AE). Both the medial and lateral margins of the phalanx are concave in dorsal view ( Fig. 24 AB), and the medial distal condyle extends further distally than the lateral condyle. The distal end is much shorter dorsoventrally than the proximal end and decreases in dorsoventral height laterally. The distal articular surface is gently convex dorsoventrally ( Fig. 24 AC).

Ungual claws are present on digits I –III (IANIGLA-PV 078/1, 078/2, 079; Figs 22A View Figure 22 , 24S – AA). These are strongly compressed mediolaterally ( Fig. 22A View Figure 22 ). In lateral view, they have a convex dorsal margin and concave ventral margin ( Fig. 24U, X, AA). A ridge-like tubercle is present along the ventral surface of the distal two fifths of each ungual claw. Similar ventral ridges or tubercles are present in a wide array of titanosauriforms ( Canudo et al., 2008; Mannion et al., 2013), including the titanosaurs Dreadnoughtus ( Ullmann & Lacovara, 2016) , Epachthosaurus ( Martínez et al., 2004: fig. 13; UNPSJB-PV 920: P.D.M., S.F.P., pers. observ. 2013), Malawisaurus ( Gomani, 2005) and Muyelensaurus (MAU-PV-LL 58, 59, 144–146: P.D.M., pers. observ. 2014). Phalanx IV-2 (IANIGLA-PV 077/10) is a reduced, proximodistally short ungual, with subcircular proximal and distal ends ( Figs 22A View Figure 22 , 24 AT, AU).

OSTEODERMS

Four osteoderms (IANIGLA-PV 080/1–2, 081/1–2) were found ( Figs 25 View Figure 25 , 26 View Figure 26 ) associated with the anterior caudal vertebrae ( González Riga, 2003). Two of them are large and have a subspherical shape (IANIGLA-PV 080/1–2; Fig. 25 View Figure 25 ). Their internal side is slightly convex, whereas this convexity is more pronounced on their external side. They correspond to the morphotype 1 (ellipsoid shape) described by D’Emic, Wilson & Chatterjee (2009). IANIGLA-PV 080/2 ( Fig. 25G–L View Figure 25 ) is slightly crushed. It has a subconical shape, with its dorsal surface dominated by an apex at which fibres and grooves converge ( Gonzalez Riga, 2003: fig. 7). IANIGLA-PV 080/1 ( Fig. 25A–F View Figure 25 ) is better preserved than IANIGLA-PV 080/1. It has a subspherical shape with a less pronounced dorsal apex. Neither osteoderm appears to be hollow. These two osteoderms lack the cingulum present in the osteoderms of Ampelosaurus ( Le Loeuff, 1995) and are different in shape and size to the osteoderms of any other titanosaur. Consequently, we regard their shape as an autapomorphy of Mendozasaurus . The other two osteoderms (IANIGLA-PV 081/1–2) are small and bulbous ( Fig. 26 View Figure 26 ) ( Table 9).