A gigantic new dinosaur from Argentina and the evolution of the sauropod hind foot

Titanosauria is an exceptionally diverse, globally-distributed clade of sauropod dinosaurs that includes the largest known land animals. Knowledge of titanosaurian pedal structure is critical to understanding the stance and locomotion of these enormous herbivores and, by extension, gigantic terrestrial vertebrates as a whole. However, completely preserved pedes are extremely rare among Titanosauria, especially as regards the truly giant members of the group. Here we describe Notocolossus gonzalezparejasi gen. et sp. nov. from the Upper Cretaceous of Mendoza Province, Argentina. With a powerfully-constructed humerus 1.76 m in length, Notocolossus is one of the largest known dinosaurs. Furthermore, the complete pes of the new taxon exhibits a strikingly compact, homogeneous metatarsus—seemingly adapted for bearing extraordinary weight—and truncated unguals, morphologies that are otherwise unknown in Sauropoda. The pes underwent a near-progressive reduction in the number of phalanges along the line to derived titanosaurs, eventually resulting in the reduced hind foot of these sauropods.

The Notocolossus specimens come from the same area in the northern part of Cerro Guillermo. They were found 403 m apart in the same, basal-most stratigraphic horizon-a red mudstone facies-of the upper Coniacian-lower Santonian Plottier Formation. Based on the contacts of the strata that crop out in this area, the specimens were deposited simultaneously under the same sedimentological regime. Each specimen represents a single individual. Following the sizes of their respective anterior caudal vertebrae and the inferred lengths of their femora (see below), the holotype (UNCUYO-LD 301) pertains to a larger individual than the referred specimen (UNCUYO-LD 302).

III. Justification for referral of UNCUYO-LD 302 to Notocolossus gonzalezparejasi
The referral of UNCUYO-LD 302 to Notocolossus gonzalezparejasi is based on a unique combination of synapomorphic characters of the anterior caudal vertebrae that this specimen shares with the holotype of this species (UNCUYO-LD 301). To our knowledge, no other titanosaurian anterior caudal vertebrae exhibit the following character combination that is present in these two specimens: centra with (1) deeply concave anterior articular cotyles and strongly convex posterior articular condyles (i.e., strong procoely); (2) circular anterior articular surfaces and slightly quadrangular posterior articular surfaces; (3) anteroposteriorly concave lateral surfaces; (4) multiple vascular foramina on the lateral surfaces ventral to the transverse processes; and (5) anteroposteriorly narrow, slightly concave ventral surfaces; transverse processes that are (6) robust, elongate, and posteroventrally directed, nearly reaching the anteroposterior level of the posterior condyle of the centrum; (7) wide and rounded at their lateral ends; and (8) ornamented by longitudinal ridges on their anteroventral margins at the approximate midlength of the process; and (9) neural arches that are anteriorly placed. Qualitatively, the anterior caudal vertebrae of UNCUYO-LD 301 and 302 are nearly indistinguishable (see Fig. 3), with the few differences between them presumably attributable to the larger body size of the holotype and/or the more anterior position of the only preserved caudal vertebra of this individual (Fig. 1b). Furthermore, the two specimens are almost identical in stratigraphic and geographic provenance; as indicated above, they were recovered from the same bed of the Upper Cretaceous Plottier Formation at sites only 403 m apart.

IV. Supplemental description
The referred specimen of Notocolossus (UNCUYO-LD 302) preserves an articulated series of seven partial anterior caudal vertebrae and seven incomplete haemal arches (Fig. 2f, Table S2). Except for their smaller size (which is due in part to their slightly more posterior position in the series), the anterior-most vertebrae are nearly identical to the anterior caudal vertebra of UNCUYO-LD 301. The centra exhibit the vascular foramina evident in the latter, and the transverse processes are elongate and swept strongly posterolaterally; furthermore, the complete left transverse processes of the anterior-most three vertebrae exhibit a rugose anteroventral ridge, as in UNCUYO-LD 301 (Fig. 3). The centrum of the anterior-most and most complete vertebra of the referred specimen is quadrangular in posterior view, and its posterior articular condyle is offset dorsally (Fig.  2f,g,h,i). Its transverse process is approximately 60 percent as long as the posterior end of the centrum is wide, as is also the case in the anterior caudal of the holotype. The centra rapidly become less strongly procoelous posteriorly. Their ventral faces are slightly concave anteroposteriorly and relatively narrow (Fig. 3).
One of the haemal arches is nearly complete, but the remaining six are fragmentary. All are proximally open ('unbridged'), lack ridges on their lateral surfaces, and seem unusually elongate relative to the sizes of the centra. The most anteriorly-positioned of these haemal arches corresponds to the anterior part of the tail (Supplementary Fig. S2a,c,e). It includes both proximal rami, the proximal articular surfaces of which exhibit a central groove (which is better preserved on the right side) such that they are separated into distinct anterior and posterior portions, as in Mendozasaurus 1 . In lateral view, the proximal-most part of each ramus curves posteriorly. In more posteriorly-situated haemal arches, the proximal facets have a single articular surface that is subtriangular in proximal contour ( Supplementary Fig. S2b,d,f). In the nearly complete haemal arch (the third in the preserved sequence), the depth of the haemal canal is 40-45% the total proximodistal length of the bone, as is the case in many other titanosauriforms (Supplementary Table S2). The distal blades are straight.  Table S1). The iliac peduncle is subtriangular in proximal view, broader anteroposteriorly than mediolaterally. The obturator foramen is proximodistally elongate and slit-like in lateral view, though its morphology has probably been modified by crushing.
Although the complete right tarsus and pes of the referred specimen (UNCUYO-LD 302) were disarticulated during preparation, a cast was made prior to the initiation of this process ( Supplementary Fig. S4). This cast, as well as photos and field observations, demonstrates that all five metatarsals were preserved in contact at their proximal ends. As articulated, the orientations of the proximodistal axes of the metatarsals differed dramatically from one another. Metatarsal I was strongly inclined, such that its proximal end was situated much more dorsally and laterally than its distal end. Metatarsal II was oriented in generally the same fashion but tilted less steeply. In sharp contrast, metatarsals III-V were oriented essentially vertically in the mediolateral plane but angled strongly proximoventrally-distodorsally in the dorsoventral plane. This proximoventral-distodorsal angle was at least 60° in metatarsal III and approximately 45° in metatarsals IV and V. Thus, in the field, and probably also in life, the metatarsals of Notocolossus were arrayed in a semi-plantigrade conformation, as in other sauropods. The proximal phalanges were slightly displaced with respect to the distal ends of their respective metatarsals.
The articulated tarsus and pes were found 60 cm beneath the surface of the modern outcrop ( Supplementary Fig. S5). As such, we regard the peculiar morphology of the pedal unguals as authentic, either the 'normal' condition for Notocolossus or a pathology of this particular individual. The remainder of the pes, including the diminutive phalanx IV-2, was also well preserved. None of the other phalanges possess the truncated, highly irregular distal surfaces evident in the three unguals.

V. Comparison with other Plottier Formation titanosaurs
Although this situation is beginning to change, the fossil vertebrate record of the Plottier Formation is, at present, relatively poorer than those of most other geologic units of the Neuquén Group 13 . Nevertheless, Notocolossus gonzalezparejasi is at least the third titanosaurian species to be erected based on material from this formation; the others are 'Antarctosaurus' giganteus [20][21][22] and Petrobrasaurus puestohernandezi 23 . According to Garrido 19 , the titanosaur Muyelensaurus pecheni also comes from the Plottier Formation, although this taxon was assigned to the Portezuelo Formation by its describers 24 . Furthermore, at least two additional titanosaurian specimens have been reported from the Plottier Formation, although these currently remain unidentified at the generic level: four anterior caudal vertebrae of an indeterminate aeolosaurine (MAU-Pv-N-414 25 ) and UNCUYO-LD 313, a partial appendicular skeleton that includes, among other elements, the complete left pes 7 . Notocolossus is easily distinguished from Muyelensaurus and Petrobrasaurus in, for example, the morphology of the humerus, which is much more slender, especially proximally, in these two taxa than it is in the new form (see Calvo et al. 24 : fig.  12b; Filippi et al. 23 : fig. 6a). Unfortunately, however, comparisons with 'Antarctosaurus' giganteus are more difficult. The holotype of 'A.' giganteus (MLP 26-316) consists of rib fragments, two partial posterior caudal vertebrae, both incomplete pubes, both femora, a distal tibia, and other poorly preserved and indeterminate limb elements 20,22 ; some recent authors (e.g., Upchurch et al. 26 ) have regarded this taxon as a nomen dubium. Like Notocolossus, 'A.' giganteus was undoubtedly very large: for instance, at 2310 mm, the left femur of MLP 26-316 is the longest complete limb bone that has been described for any titanosaur (see Paul 27 :table 1; Lacovara et al. 28 :table 1). Nevertheless, the only skeletal element preserved in common to Notocolossus and 'A.' giganteus is the pubis, and it is highly incomplete in both taxa, precluding any meaningful comparisons between them.

VI. Comparison with Mendozasaurus neguyelap
Mendozasaurus neguyelap is the most completely known titanosaur from Mendoza Province 1,2 . Here, we compare skeletal elements of this taxon to overlapping bones in Notocolossus gonzalezparejasi to definitively differentiate these taxa. As noted above, although the localities that have produced fossils of Mendozasaurus and Notocolossus are close to one another-the former comes from south of Cerro Guillermo in southern-most Mendoza, the latter from Cerro Guillermo itself-these sites are stratigraphically separated: Mendozasaurus was recovered from the middle-upper Coniacian Sierra Barrosa Formation, whereas Notocolossus comes from the upper Coniacian-lower Santonian Plottier Formation.
Most elements preserved in Notocolossus are also represented in Mendozasaurus, facilitating comparisons between these taxa. The anterior dorsal vertebrae of the two forms differ in multiple respects. Although some of these differences could conceivably be due to positional variation, the single known anterior dorsal vertebra of Notocolossus and the best-preserved anterior dorsal of Mendozasaurus (IANIGLA-PV 066 1,2 ) are thought to be closely comparable in serial position (the Notocolossus dorsal is here regarded as the second or third; the Mendozasaurus dorsal is considered to be the third 2 ). The centrum of the anterior dorsal vertebra of Notocolossus has a proportionally more prominent, anteriorly-projecting articular condyle and relatively small lateral pneumatic fossae ('pleurocoels'). The centrum is much larger relative to the diameter of the neural canal than is the centrum of IANIGLA-PV 066. Overall, the Notocolossus neural arch is considerably lower and wider due to its longer transverse processes and shorter neural spine; moreover, the dorsal edge of the left transverse process meets the neural spine at an obtuse angle, whereas in Mendozasaurus the processes intersect the spine at nearly right angles. In IANIGLA-PV 066, the ventral edge of the intraprezygapophyseal lamina is linked to the dorsal margin of the neural canal by a stout 'medial pillar' 1 , whereas in Notocolossus the intraprezygapophyseal lamina forms the dorsal margin of the neural canal. In Mendozasaurus, the base of the prespinal lamina is connected to the prezygapophyses by short spinoprezygapophyseal laminae; these latter laminae do not occur in the UNCUYO-LD 301 dorsal vertebra. Furthermore, the two 'accessory' laminae within the parapophyseal centrodiapophyseal fossa of Notocolossus are not present in Mendozasaurus (or, to our knowledge, any other titanosaur). Unfortunately, the posterior surface of the UNCUYO-LD 301 vertebra is currently obscured by a protective 'cradle'. As such, we are unable to determine whether the two hypothesized autapomorphies evident on the posterior aspect of the anterior dorsal neural arch of Mendozasaurus 1 ([1] subtriangular centropostzygapophyseal [= 'infrapostzygapophyseal'] fossae and [2] dorsolaterally-ventromedially-oriented 'postzygapostspinal' laminae connecting the postzygapophyses to the base of the postspinal lamina) are also present in Notocolossus.
The anterior caudal vertebrae of Mendozasaurus and Notocolossus may be easily distinguished as well. As with the dorsal vertebrae, a few of their differences may be attributable to serial variation; however, because one of the best preserved anterior caudals of Mendozasaurus (IANIGLA-PV 065/1, regarded as the first or second in the holotypic series 1 ) is thought to be close in position to the most complete caudal of Notocolossus (that of the holotype, UNCUYO-LD 301, here considered the third or fourth), most of these distinctions are probably taxonomic in nature. The centrum of IANIGLA-PV 065/1 is much shorter anteroposteriorly than is that of the caudal of UNCUYO-LD 301; this could, however, be due to taphonomic deformation, especially since another anterior caudal vertebra of the Mendozasaurus holotype (IANIGLA-PV 065/4) has a much longer centrum 1 that is more comparable to those of Notocolossus (both specimens, UNCUYO-LD 301 and 302) in this regard. Perhaps more significantly, the transverse processes of IANIGLA-PV 065/1 are substantially shorter than are those of the anterior caudals of UNCUYO-LD 301 and 302, despite the fact that this Mendozasaurus vertebra is thought to have been situated more anteriorly in the series than are all known caudals of Notocolossus. Because, in titanosaurs represented by complete anterior-middle caudal series (e.g., Alamosaurus 29 , Baurutitan 30 , Dreadnoughtus 28 , Epachthosaurus 31 ), the transverse processes gradually decrease in length as one moves posteriorly through the series, we regard the considerably longer anterior caudal transverse processes of Notocolossus as a well-supported distinction between this taxon and Mendozasaurus.
Several differences are evident between the anterior caudal neural arches of IANIGLA-PV 065 and UNCUYO-LD 301 as well. In Notocolossus, there is no evidence of the dorsal prominences of the prezygapophyses or the deep 'interzygapophyseal fossa' (the confluent postzygapophyseal spinodiapophyseal/postzygapophyseal centrodiapophyseal fossa of Wilson et al. 32 ) that characterize Mendozasaurus (e.g., IANIGLA-PV 065/4). Also, whereas Mendozasaurus has elongate, well-developed spinoprezygapophyseal and prespinal laminae, in Notocolossus, there is no clear evidence of the prespinal lamina, and the spinoprezygapophyseal laminae are short. The anterolateral margins of the neural spine of the new taxon are instead comprised by seemingly novel laminae that converge ventrally and merge immediately dorsal to the prezygapophyses, forming a 'V-shape' in anterior view. The haemal arches of Notocolossus (UNCUYO-LD 302) are too incomplete for meaningful comparisons with those of Mendozasaurus.
The humeri of Notocolossus (UNCUYO-LD 301) and Mendozasaurus (IANIGLA-PV 069) differ dramatically, especially at their proximal ends. The mediolateral width of the proximal end of the Notocolossus humerus is 2.88 times the width at midshaft, substantially greater than in any other titanosaur; in Mendozasaurus, by contrast, this ratio is only 2.41 (Table 1). The medial half of the proximal margin of the Notocolossus humerus is strongly projected proximomedially, whereas in Mendozasaurus this same margin is horizontal. Notocolossus also bears a small proximolateral process (= 'supracoracoideus tuberosity') that is absent in Mendozasaurus.
Finally, the complete, articulated pes of the referred specimen of Notocolossus (UNCUYO-LD 302) differs from the known pedal material of Mendozasaurus as well. Compared to that of metatarsal I, the proximodistal length of metatarsal III is shorter in Notocolossus than in any other titanosaurian taxon, including Mendozasaurus, though it is perhaps notable that a probably associated metatarsus of the latter (IANIGLA-PV 077) comes the closest in this regard (Table 2). Notocolossus also exhibits a substantially lower metatarsal IV:metatarsal I length ratio (1.33) than does Mendozasaurus (1.46) ( Table 2). Metatarsal V is only minimally distally expanded in Mendozasaurus, as is the case in most other sauropods, but not in Notocolossus. The two preserved pedal unguals of Mendozasaurus differ from those of UNCUYO-LD 302 in being proximodistally elongate and dorsoventrally tapered distally, as is typical of sauropods, instead of short and distally truncated as they are in Notocolossus.

VII. Estimates of body dimensions
The incompleteness of known specimens of Notocolossus, coupled with the variability in body proportions observed within Titanosauria (e.g., long-necked taxa such as Futalognkosaurus 33 versus relatively short-necked forms such as Mendozasaurus 1,2 ), renders estimation of the body dimensions of the new taxon problematic. Nevertheless, we estimated the proximodistal lengths of the femora of the two known Notocolossus specimens (UNCUYO-LD 301 and 302) and the body mass of the holotype (UNCUYO-LD 301) using methods closely comparable to those employed by Smith et al. 34 and Lamanna 35 to estimate these same dimensions in the type specimen of the giant Egyptian titanosaur Paralititan (CGM 81119).
To estimate the femoral length of the Notocolossus holotype, we first compiled humeral and femoral lengths of articulated or definitively associated titanosaurian skeletons that preserve both of these elements in their entirety, then transformed these data into logarithms (base 10) (see Supplementary Table S4). We then plotted log humeral versus log femoral length and performed a linear regression to produce an allometric equation relating these dimensions, allowing one to be estimated from the other ( Supplementary Fig. S9). The humeral length of UNCUYO-LD 301 (1760 mm) was then 'plugged in' to this equation, generating an estimated femoral length of 2166 mm for this specimen (Supplementary Table S4). By comparison, the maximum femoral lengths of the type specimens of the giant titanosaurs Dreadnoughtus (MPM-PV 1156) 28 , Futalognkosaurus (MUCPv-323) 36 , and 'Antarctosaurus' giganteus (MLP   20 Table S4). Identical methods applied to the proximodistal length of the femur versus that of metatarsal III and used in concert with the metatarsal III length of UNCUYO-LD 302 (197 mm) yielded an estimated femoral length of 1283 mm for this referred specimen of Notocolossus (Supplementary Fig. S10; Supplementary Table S5). The length of metatarsal III of the ?Alamosaurus specimen NMMNH P-49967 (270 mm) produced an estimated femoral length of 1632 mm, lower than the 1.7-2.1 m length estimated by D'Emic et al. 37 .
Having demonstrated that, based on estimated femoral length, the holotype of Notocolossus represents a considerably larger individual than does the referred specimen, we then estimated the body mass of the former. Again, we used methods comparable to those employed by Lamanna 35 . We compiled minimum midshaft humeral and femoral circumferences of titanosaurian skeletons that preserve both of these dimensions, then transformed these data into logarithms (base 10) (Supplementary Table S6). We plotted log humeral versus log femoral midshaft circumference and performed a linear regression to produce an allometric equation relating these dimensions, allowing one to be estimated based on the other (Supplementary Fig.  S11). The humeral circumference of UNCUYO-LD 301 (770 mm) was then 'plugged in' to this equation, generating an estimated femoral circumference of 936 mm for this specimen (Supplementary Table S6 Table  S6). Although the applicability of this scaling equation to unusually large titanosaurs has recently been challenged 39 , it may still provide insight into the body masses of these dinosaurs relative to one another. As such, although (due to the incomplete nature of the specimen) this must be regarded as tentative, it appears that the holotype of Notocolossus may represent one of the most massive titanosaurian individuals-and terrestrial animals-that has been discovered to date.

VIII. Photogrammetric models
To better document the skeletal morphology of Notocolossus, we herein provide threedimensional digital models of the anterior dorsal vertebra and right humerus of the holotype (UNCUYO-LD 301) and a cast of the complete and articulated right tarsus and pes of the referred specimen (UNCUYO-LD 302). These models were kindly produced by Stephen Poropat of the Australian Age of Dinosaurs Natural History Museum (Winton, Queensland, Australia). Poropat photographed each of these specimens from all feasible perspectives using a Nikon D90 digital camera fitted with a VR 18-105 mm lens, then used Agisoft PhotoScan Professional Edition software to assemble the resulting photos into photogrammetric digital models. He then used this software to convert each model into a three-dimensional (3D) Adobe Portable Document Format (.pdf) file (see Supplementary Figs. S1, S3, and S4). (Viewing and navigating Adobe 3D .pdf files requires Adobe Acrobat or Acrobat Reader. The latter is freely available for download at http://get.adobe.com/reader/.) Users may download and rotate each model into whatever orientation they prefer, zoom in on particular osteological structures, etc. (Note, however, that the posterior surface of the dorsal vertebra and the posterodistal surface of the humerus are obscured by supportive 'cradles'; as such, these areas are not represented in the digital models.) Surface files (.obj, .stl) of all models are available to qualified researchers upon request to the senior author (B.J.G.R.).

IX. Phylogenetic character list
The following 350 morphological characters were employed in our phylogenetic analysis and are listed by general anatomical region. The vast majority (340) of these characters (numbers 1-130, 134-256, 259-330, 332, 333, and 335-347) were drawn directly from Carballido and Sander 40 and references cited therein. (Their respective character numbers in that analysis are as follows: 1-130, 132-254, 255-326, 327, 328, and 329-341.) Except for minor editing (e.g., correction of typographical errors), the descriptions of these 340 characters are as presented by Carballido and Sander 40 ; also, the literature attributions of a few characters (103, 135, 149, 213, 245, and 249) have been corrected. Furthermore, character 133 (character 131 of Carballido and Sander 40 ) was slightly modified from that study. A significant aspect of the current analysis is the addition of nine characters that are presented in italics below (numbers 131, 132, 257, 258, 331, 334, and 348-350). Of these, characters 131 and 132 were taken from González Riga and Ortiz David 4 and sources in that paper, character 257 was modified from Mannion et al. 41

XI. Supplementary Figures S1-S11
To keep the size of our Supplementary Information file to a minimum, we have included preview images of each of the three three-dimensional photogrammetric digital models of Notocolossus gonzalezparejasi bones (Supplementary Figures S1, S3, and S4). Readers may access and download the models themselves (as 3D .pdfs) from the data repository Figshare by following the hyperlinks provided in the caption of each figure.   63 . Specimens are listed by increasing femoral length. All specimens that preserve both the femur and metatarsal III in their entirety (i.e., MACN-PV RN 1061 ['Individual D'] through ZPAL MgD-I/48) were used to generate the regression equation presented in Supplementary Fig. S10; this equation was then used to estimate the femoral lengths of the referred specimen of Notocolossus gonzalezparejasi (UNCUYO-LD 302) and ?Alamosaurus sanjuanensis specimen NMMNH P-49967 at 1283 mm and 1632 mm, respectively. Abbreviations: L F , femoral length; L MTIII , metatarsal III length; log, logarithm (base 10). Institutional abbreviations see section I above. ** = measurement estimated.