The Anatomy and Phylogenetic Relationships of “Pelorosaurus“ becklesii (Neosauropoda, Macronaria) from the Early Cretaceous of England

The sauropod dinosaur “Pelorosaurus” becklesii was named in 1852 on the basis of an associated left humerus, ulna, radius and skin impression from the Early Cretaceous (Berriasian-Valanginian) Hastings Beds Group, near Hastings, East Sussex, southeast England, United Kingdom. The taxonomy and nomenclature of this specimen have a complex history, but most recent workers have agreed that “P.” becklesii represents a distinct somphospondylan (or at least a titanosauriform) and is potentially the earliest titanosaur body fossil from Europe or even globally. The Hastings specimen is distinct from the approximately contemporaneous Pelorosaurus conybeari from Tilgate Forest, West Sussex. “P.” becklesii can be diagnosed on the basis of five autapomorphies, such as: a prominent anteriorly directed process projecting from the anteromedial corner of the distal humerus; the proximal end of the radius is widest anteroposteriorly along its lateral margin; and the unique combination of a robust ulna and slender radius. The new generic name Haestasaurus is therefore erected for “P.” becklesii. Three revised and six new fore limb characters (e.g. the presence/absence of condyle-like projections on the posterodistal margin of the radius) are discussed and added to three cladistic data sets for Sauropoda. Phylogenetic analysis confirms that Haestasaurus becklesii is a macronarian, but different data sets place this species either as a non-titanosauriform macronarian, or within a derived clade of titanosaurs that includes Malawisaurus and Saltasauridae. This uncertainty is probably caused by several factors, including the incompleteness of the Haestasaurus holotype and rampant homoplasy in fore limb characters. Haestasaurus most probably represents a basal macronarian that independently acquired the robust ulna, enlarged olecranon, and other states that have previously been regarded as synapomorphies of clades within Titanosauria. There is growing evidence that basal macronarian taxa survived into the Early Cretaceous of Europe and North America.


Introduction
Sauropod dinosaurs were globally distributed mega-herbivores that dominated many Mesozoic terrestrial ecosystems [1][2][3]. Several aspects of the evolution of sauropods remain poorly understood, ranging from taxonomic and nomenclatural issues (e.g. [1]), to the causes of largescale fluctuations in their diversity [2] and the impact of gigantic body size on their growth, physiology and biomechanics [3]. Nevertheless, in recent years, significant progress in our understanding of sauropod evolution has been driven by several factors, including an influx of information on new taxa (e.g., see [4]: fig 4), phylogenetic analysis [1,[5][6][7][8][9][10][11][12][13][14][15][16][17][18][19], and new technology such as CT-scanning and Finite Element Analysis (e.g., [20,21]). These advances provide fresh opportunities to revisit material that was first discovered in the 19th or early 20th centuries, in order to address previously intractable taxonomic, phylogenetic or other problems. The current study represents one such reappraisal, focussed on the Early Cretaceous taxon "Pelorosaurus" becklesii Mantell 1852 [22]. This taxon is significant regionally and globally for several reasons: it potentially represents the earliest known European titanosaur [6,18]; it provides a glimpse of sauropod evolution during the otherwise poorly represented Berriasian-Valanginian, a period that documents the initial recovery of sauropods from a significant extinction event around the Jurassic-Cretaceous boundary [2,[23][24][25]; and it includes the first dinosaurian skin impression recognised by science [26]. Here, we survey the complex and convoluted taxonomic history of "Pelorosaurus" becklesii, describe and illustrate the specimen in detail, and identify a suite of character states that diagnose this animal as a new taxon. As a by-product of this investigation, we note a number of new or revised characters in the sauropod fore limb. These characters are discussed and incorporated into phylogenetic analyses in order to assess the relationships of"Pelorosaurus" becklesii. Finally, we consider the new information on "P." becklesii in the wider context of sauropod evolutionary history.
Universidad Nacional de la Patagonia San Juan Bosco, Comodoro Rivadavia, Argentina; USNM, United States National Museum of Natural History, The Smithsonian Institute, Washington D.C., USA; WDC, Wyoming Dinosaur Center, Thermopolis, USA: YPM, Yale Peabody Museum, New Haven, USA; ZDM, Zigong Dinosaur Museum, Zigong, People's Republic of China; ZPAL, Institute of Paleobiology, Polish Academy of Sciences, Warsaw, Poland (N.B. the holotype of Opisthocoelicaudia was held at this institution when it was described in 1977 [27], but it has subsequently been returned to Ulan Bator, Mongolia).
Historical Background: Discovery and Taxonomy of "Pelorosaurus" becklesii [22] The first sauropod specimens to be described scientifically were discovered and studied in the United Kingdom during the early and mid-19 th Century [28][29][30]. This 'head start' has had some unfortunate consequences because of the combined deleterious effects of fragmentary material, a tendency for early workers to refer non-overlapping specimens to the same taxon, the absence of opportunities to make comparisons with well-preserved and complete skeletons, over-enthusiastic naming of new taxa, and the phenomenon of historically obsolete diagnostic characters [31]. As a result, the taxonomy and nomenclature of British sauropods is notorious for its complexity and confused nature, and it is only in the past two decades that it has been possible to begin to resolve these problems using carefully evaluated synapomorphies and autapomorphies (e.g. [1,6,17,18,28,[32][33][34][35][36][37][38][39]. "Pelorosaurus" becklesii became thoroughly enmeshed in this taxonomic tangle during the 1880s, and numerous claims concerning its status and affinities have been sporadically proposed since then. Here, therefore, we briefly summarise the history of this species in order to present background information relevant to the revised diagnosis and other conclusions resulting from this reassessment of "P." becklesii. In 1841, Richard Owen [40] named Cetiosaurus (without a species name) on the basis of fragmentary sauropod remains from various localities in England [28,35]. At this time, and for many years subsequently, Owen maintained that these and other sauropod specimens belonged to gigantic carnivorous marine reptiles, and he later went on to exclude them from Dinosauria [41,42]. Owen [41] named four species of Cetiosaurus, one of which was C. brevis based on several vertebrae and chevrons, such as NHMUK R10390 from Sandown Bay and Culver Cliff, Isle of Wight, and NHMUK R2133, R2115, R2544-2550 from the Hastings Beds in what is now West Sussex (Fig 1). Most of these specimens were re-identified as belonging to Iguanodon by Melville [43], except for four anterior caudal vertebrae (NHMUK R2544-2547) and three chevrons (NHMUK R2548-2550) from Tilgate Forest, West Sussex, which he proposed as the type material of a new species of Cetiosaurus, C. conybeari ( [43]: p.297). A large right humerus (NHMUK 28626) was discovered in 1850 from '. . .a few yards. . .' away from the Tilgate Forest site that had yielded the C. conybeari tail elements [44]. On this basis, Mantell [44] erected the new generic name Pelorosaurus and created the combination of P. conybeari for the humerus, caudals and chevrons. Mantell also noted the robust and straight morphology of the humerus, and the presence of a medullary cavity in its shaft, and proposed that Pelorosaurus therefore represented a large terrestrial dinosaur rather than a marine reptile. Owen [42] disagreed with both Melville and Mantell and therefore retained the name Cetiosaurus brevis for NHMUK R2544-2550 and used Pelorosaurus conybeari (N.B. misspelt by Owen as 'conybearii') for the right humerus alone.
In 1852, Samuel H. Beckles recovered a block of 'Wealden Sandstone' exposed at low tide on the Sussex Coast ( [22]: p. 143). This block contained the associated left humerus, ulna and radius (all three numbered NHMUK R1870), as well as a skin impression (NHMUK R1868), from what we now recognise to be a sauropod. The report of this discovery has always doi: 10.1371/journal.pone.0125819.g001 Anatomy and Relationships of "Pelorosaurus" becklesii previously been referenced as 'Mantell 1852' [22], although in fact it is clearly an anonymous account of a talk given by Mantell. There are few descriptive details (and no plates or illustrations), except for some measurements of the limb bones and a mention that the scales of the skin impression are hexagonal. Mantell noted the general similarity of the NHMUK R1870 humerus to that of Pelorosaurus conybeari, but also recognised that they represented distinct species because the new humerus was somewhat shorter and more robust than the latter. On this basis, Mantell [22] erected a new species of Pelorosaurus, which he named P. becklesii. Little more was written about the latter taxon during much of the next three decades (i.e., from 1853 to 1888). For example, Owen [42] completely ignored "P." becklesii, despite the fact that he paid considerable attention to the Cetiosaurus brevis and Pelorosaurus conybeari specimens from Tilgate Forest. This long neglect might reflect the fact that the "P." becklesii material remained in the private Beckles collection until it was purchased by The Natural History Museum, London, in 1891 [45].
Significant advances in the understanding of sauropod dinosaurs were made as a result of discoveries in the 1870s and 1880s. In particular, more complete material of Cetiosaurus from Oxfordshire was described by Phillips in 1871 [46], enabling him to recognise that this animal was a gigantic terrestrial herbivorous dinosaur. This view was reinforced by the numerous new discoveries made in the Western USA by O. C. Marsh, E. D. Cope and associated colleagues, resulting in key advances such as: the recognition of the group Sauropoda [47]; the naming of iconic taxa, including Camarasaurus [48] and Apatosaurus [49]; description of the first wellpreserved skulls and cervical series (e.g., [50]); and the first publication of a skeletal reconstruction of a sauropod [51]. A series of papers by R. Lydekker and others during the late 1880s and 1890s aimed to revise and clarify the taxonomy, nomenclature and relationships of British sauropods in the light of these new discoveries. Unfortunately, these studies tended to have the opposite effect, at least when viewed with the benefits of hindsight and modern taxonomic and phylogenetic practices. Lydekker ([52]: p.55) mentioned the humerus of Pelorosaurus conybeari (N.B. he cited the specimen number '28266', but this should be '28626') and suggested that it was potentially synonymous with Ornithopsis. In the same paper (p.58) Lydekker discussed the material of "P." becklesii as follows: 'I have already mentioned Cetiosaurus brevis under the head of Pelorosaurus, but I may here bring to notice an associated humerus, radius, and ulna, from the Wealden of the Isle of Wight, in the collection of Mr. S. H. Beckles, of which the British Museum possesses casts (no. 28701). The length of the humerus is 0.620 (24.5 inches); its shaft is much shorter than that of the corresponding bone of C. [Cetiosaurus] oxoniensis; but it approximates to that piece in its widely expanded head, and there appears a probability that these bones may belong to C. [Cetiosaurus] brevis, in which case that form would differ widely from the type species, and would likewise be certainly distinct from Pelorosaurus. On the other hand, these limb-bones may perhaps be referable to Titanosaurus, or possibly even to a new genus. ' Lydekker was clearly referring to "P." becklesii, as demonstrated by the facts that, at the time, this was the only British sauropod from the Wealden that had produced an associated humerus, ulna and radius, that the original material was in the Beckles collection, and that the stated length of the humerus agrees closely with that of NHMUK R1870 (Table 1). The claim that these specimens came from the Isle of Wight, rather than East Sussex, is therefore an error.
Marsh [53] erected the genus "Morosaurus" for sauropod remains from the Late Jurassic Morrison Formation of North America, but this has subsequently been shown to be a junior synonym of Camarasaurus [54]. When Marsh visited England in 1888, he compared British dinosaur material with that from North America and published some of his conclusions in 1889 [55]. Marsh ([55]: p.325) suggested that "P." becklesii was referable to "Morosaurus" based on overall similarity and limb proportions, and created the new combination 'Morosaurus" becklesii. Nicholson and Lydekker [56] and Lydekker [57,58] synthesized the original views of Melville and Mantell with the more recent proposal of Marsh, recognising that "Cetiosaurus brevis", the caudals of Pelorosaurus conybeari, and 'M. becklesii' were congeneric and so created the new combination "M." brevis (see also [59,60]). "M." brevis was assigned to the Cetiosauridae, and its diagnosis was expanded to include dental and vertebral characters based on the referred specimens (e.g. [57]). As noted by several previous authors (e.g. [36,37]), the referral of "Pelorosaurus" becklesii to 'Morosaurus' cannot be supported because the latter genus is not a valid taxon and the diagnostic characters cited by Marsh, Lydekker and others are vague and/ or known to be widespread among many sauropod species.
Huene ([61]: p122-123) regarded Pelorosaurus as synonymous with Ornithopsis Seeley 1870 [62], with the former name having priority because it was erected earlier. He did not discuss the different species of Pelorosaurus, but listed it as including material from the Kimmeridgian of England and France, and Wealden of England: presumably this distribution reflects the inclusion of Jurassic Ornithopsis specimens. Huene placed Pelorosaurus in the 'Subfamily Brachiosauridae' (sic). A more detailed indication of Huene's opinion on these issues was presented in 1932 [63]. Huene ( [63]: pp.286-288, figs 34, 35) described and illustrated the limb material of "P." becklesii. He argued that the caudal vertebrae assigned to Cetiosaurus brevis (NHMUK R2544-2547) belonged to a member of the theropod family Megalosauridae: thus he rejected previous claims by Lydekker that these specimens and "P." becklesii should be considered to be conspecific. Huene also argued that the humerus of "P." becklesii differed from that of Pelorosaurus conybeari in both its proportions and the morphology of key features such as the 'processus lateralis' (= the deltopectoral crest). He therefore considered "P." becklesii to be distinct from Pelorosaurus conybeari, and referred to the former as 'Gen. (?) becklesii' (N.B. not 'Camarasaurus becklesii' as claimed by McIntosh [64]). Although Huene did not follow Marsh or Lydekker in regarding "P." becklesii as referable to "Morosaurus", he does seem to have believed that the two taxa were closely related because he placed 'Gen. (?) becklesii' and Camarasaurus in the Camarasaurinae, within the family Brachiosauridae ( [63]: p.251). "Pelorosaurus" becklesii then received very little attention during the rest of the 20 th Century, apart from occasional passing references or inclusion in lists of sauropod taxa. Swinton ([60]: p.211) mentioned "P." becklesii in passing, misspelling the species name as 'becklesi'. He noted the presence of "Morosaurus brevis" on the Isle of Wight, but did not suggest that "P." becklesii should be considered congeneric or conspecific with this taxon. Romer [65] and Steel [66] accepted the validity of Pelorosaurus and regarded it as synonymous with several other poorly known European taxa, including: "Chondrosteosaurus" Owen 1876 [67], "Dinodocus" Owen 1884 [68], Eucamerotus Hulke 1872 [69], "Gigantosaurus" Seeley 1869 [70], "Hoplosaurus" Lydekker 1890 [57], "Ischyrosaurus" Hulke 1874 [71], 'Morinosaurus" Sauvage 1874 [72], "Neosodon" Moussaye 1885 [73], Oplosaurus Gervais 1852 [74], and Ornithopsis Seeley 1870 [62]. All of these referrals made by Lydekker, Romer and Steel have been rejected recently [1,32,36] on the basis that either there is no anatomical overlap between the Pelorosaurus conybeari holotype and the referred taxon (e.g., Oplosaurus armatus is based on an isolated tooth, NHMUK R964), or because there are no autapomorphies uniting specimens where comparisons can be made. Romer [65] only listed genera, so there is no information on his views concerning Pelorosaurus conybeari and "P." becklesii. However Steel ([66]: p.68) proposed a detailed revision of sauropod taxonomy and nomenclature at the species level. Steel regarded Pelorosaurus as a member of the subfamily Brachiosaurinae, and P. conybeari and "P." becklesii as separate valid species of Pelorosaurus.
McIntosh [64] regarded Pelorosaurus as a valid brachiosaurid that included several species (e.g., P. conybeari and "P. mackesoni"). However, he excluded "P." becklesii from Pelorosaurus, instead considering it to be Sauropoda incertae sedis. This opinion was based largely on limb proportions: 'The ulna: humerus ratio is 0.71. The latter character and its robustness immediately excludes the animal from the genus Pelorosaurus and any other brachiosaurid such as Pleurocoelus.' ( [64]: p.398). More recently, most workers have regarded "P." becklesii as a titanosaur [1,6,18,36,[75][76][77]. For example, Upchurch [6] noted the robust nature of the forearm elements and the concave profile of the articular surface of the proximal anteromedial process of the ulna, features which were then believed to be synapomorphies of Titanosauria (see also [1]). Tidwell and Carpenter [75] also regarded "P." becklesii as a probable titanosaur and suggested that it shared similarities with an unnamed 'titanosaur' from the Cloverly Formation of Wyoming. However, because Tidwell and Carpenter [75] is only a published abstract, there are no further details concerning the character states used to support this claim. The first phylogenetic analysis to include "P." becklesii was that of Mannion et al. [18]. This indicated that "P." becklesii is a somphospondylan, but only a subset of the analyses supported its inclusion within the Titanosauria. If "P." becklesii does indeed represent a titanosaur, then it would be the earliest body fossil material pertaining to a member of that clade from Europe [1,6,36], although trackways from the Middle Jurassic of England are the earliest putative record globally [78,79] (but see [17] for a contrary opinion). In contrast, D'Emic [17] rejected the identification of "P." becklesii as a titanosaur, although he did accept it as a titanosauriform. More recently, Poropat et al. [80,81] updated the character scores for the Australian sauropods Diamantinasaurus and Wintonotitan for the 'Lusotitan Standard Discrete Matrix' (LSDM) data set presented by Mannion et al. [18]. However, "Pelorosaurus" becklesii was one of eight taxa that were pruned, a posteriori, from the resulting 5334 most parsimonious trees in order to generate an agreement subtree. Thus, Poropat et al. [80,81] did not evaluate the phylogenetic relationships of "P." becklesii, and no analysis has examined the impact of the new Australian data on titanosauriform relationships based on the 'Lusotitan continuous and discrete matrix' (LCDM) data set of Mannion et al. [18].
Most previous workers have acknowledged that "P." becklesii represents a distinct taxon (e.g. [1,63,64,66]) but have been reluctant to erect a new generic name because the holotype is somewhat incomplete, and clear autapomorphies have proved to be elusive. Thus, Naish and Martill ( [29]: p.499) wrote: '. . .whether the material is diagnostic is arguable.' Upchurch et al. [36], however, presented a preliminary reassessment of "P." becklesii and concluded that there were at least two potential autapomorphies of the humerus (see below).
In summary, the recent consensus among sauropod workers is that "P." becklesii is a distinct taxon that merits a new generic name provided that sufficiently strong autapomorphies can be identified. None of the other sauropod taxa from the Late Jurassic or Cretaceous of Britain can be justifiably referred to, or combined with, "P." becklesii. The latter taxon is generally regarded as a member of the Titanosauriformes and is potentially a somphospondylan or even a titanosaur.

Methods
No permits were required for the described study, which complied with all relevant regulations. The specimens studied for this work are housed at The Natural History Museum, London, UK (institutional abbreviation NHMUK). The specimen catalogue numbers are NHMUK R1868, R1869 and R1870. Other specimens examined in order to make comparisons with Haestasaurus are cited in the text as required, with full institutional catalogue numbers and references where appropriate. Cladistic data sets and analytical techniques are outlined in 'Phylogenetic Analyses' below.

Nomenclatural Acts
The electronic edition of this article conforms to the requirements of the amended International Code of Zoological Nomenclature, and hence the new names contained herein are available under that Code from the electronic edition of this article. This published work and the nomenclatural acts it contains have been registered in ZooBank, the online registration system for the ICZN. The ZooBank LSIDs (Life Science Identifiers) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix "http://zoobank.org/". The LSID for this publication is: urn:lsid:zoobank.org:pub: 9D2E9827-D6D5-444A-A01C-69CAE4FFCA22. The electronic edition of this work was published in a journal with an ISSN, and has been archived and is available from the following digital repositories: PubMed Central, LOCKSS, https://iris.ucl.ac.uk/iris/browse/profile?upi= PUPCH49 of the Beckles collection [45] mentions the proximal end of a metacarpal that was apparently regarded as part of "P." becklesii. It is probable that this catalogue entry refers to the proximal portion of NHMUK R1869. This is actually a large, robust and nearly complete sauropod metacarpal, which is broken into three pieces that fit together. Its relative size (e.g. maximum width of proximal end = 152 mm, total length = 330 mm) means that it cannot belong to the same individual as NHMUK R1870 (see Table 1). None of the literature dealing with "P." becklesii prior to, or since, 1891 mentions this metacarpal. We suspect that the metacarpal was not found with the humerus, ulna, radius and skin impression, but became 'associated' with them when the Beckles collection was purchased and catalogued by the NHMUK. This would explain why several early papers (e.g. [22]) specifically mention the association of the humerus, ulna, radius and skin impression, but do not note the presence of the metacarpal. Thus, NHMUK R1869 cannot be considered part of the holotype of Haestasaurus, or be referred to that taxon, and will not be discussed further here.
Etymology of new generic name-From 'Haesta', the name of the putative pre-Roman chieftain whose people apparently settled the area of Hastings and gave the town its name [83], and 'sauros', Greek for 'reptile'.
Diagnosis-Haestasaurus becklesii is diagnosed on the basis of the following autapomorphies: (1) the anteromedial corner of the distal end of the humerus projects to form an 'anterior entepicondylar process'; (2) there are two small vertical ridges situated between the lateral and medial anterodistal processes of the humerus; (3) the proximal articular surface of the radius is widest anteroposteriorly along its lateral margin, and this margin is nearly straight rather than strongly convex; (4) at the distal end of the shaft, the anterior surface of the radius is shallowly concave, between anterolateral and anteromedial ridges; and (5) the combination of a robust ulna (maximum proximal width:proximodistal length ratio > 0.4) and a slender radius (transverse proximal width:proximodistal length ratio < 0.3).
The proximal articular surface is strongly rugose (Fig 3C). The proximolateral corner lies only slightly below the level of the humeral head, whereas the medial half of the proximal articular surface is convex and curves strongly medially and distally in anterior view (Fig 2). Thus, . Abbreviations: aep, anterior entepicondylar process; af, anconeal fossa; alf, anterolateral fossa; dp, deltopectoral crest; hh, humeral head; lad, lateral anterodistal process; mad, medial anterodistal process. All parts are at the same scale.  Haestasaurus possesses the 'square' proximolateral corner that occurs in Tehuelchesaurus and most somphospondylans, rather than the plesiomorphic rounded condition observed in other taxa [1,10,17,18,85]. The profile of the proximal end, in anterior view, lacks the strongly sigmoid curvature that occurs in several titanosaurs [7,86] (Fig 2A). In proximal end view the lateral half of the articular surface curves strongly anterolaterally towards the proximolateral corner and top of the deltopectoral crest, whereas the medial half is nearly straight and projects medially. There are no large fossae or major processes along the junction of the proximal and anterior faces, although some small rugosities do occur (especially on the medial half). The anteroposteriorly widest part of the proximal end is located at about mid-width (slightly nearer the medial than the lateral margin), where the humeral head expands prominently backwards. Here, the proximal humeral head forms a prominent process that overhangs the posterior surface of the shaft (Fig 3A and 3B): a similarly distinct process is also present in some titanosauriform taxa, such as Giraffatitan . This has a rugose articular surface that curves posterodistally in lateral view. Distally, this process gives rise to a stout vertical ridge that fades out very rapidly into the posterior surface, but still extends downwards to divide this part of the proximal end into lateral and medial fossae. These fossae ( Fig 2B) probably represent the origins of the lateral and medial heads of the triceps muscle [27]. On the posterior surface, approximately level with the most prominent point of the deltopectoral crest, there is a low, rounded and vertically elongated bulge situated a short distance laterodistal to the distal end of the ridge that separates the triceps fossae. This bulge is subtle but can be seen in lateral view ( Fig 3A). Based on the muscle reconstructions of Borsuk-Bialynicka [27], this is probably the insertion for the M. latissimus dorsi. A second striated muscle scar is located a short distance proximolateral to the one just described, and probably marks the insertion of the M. there is a prominent striated projection in this region that has usually been identified as the insertion for the M. latissimus dorsi [90], or for 'brachial musculature' [91]. In these titanosaurs, the attachment for the M. scapulohumeralis anterior forms a prominent lateral projection that is visible in anterior view (see 'New and Revised Characters' below), but this projection does not occur in Haestasaurus.
The anterior surface of the proximal half of the humerus is mediolaterally concave. This surface does not display the low rounded bulge or rugosity that marks the site of insertion of the M. coracobrachialis, but this might be because of the relatively small size of Haestasaurus and/ or the presence of a museum label that potentially covers this structure. This muscle scar occurs in most neosauropods, and seems to be particularly well developed in titanosaurs such as Diamantinasaurus (AAOD 603 [80]), Elaltitan (PVL 4628 [77]: fig 6), Neuquensaurus (MLP-CS 1050 [91]) and Opisthocoelicaudia (Z.PAL MgD-I/48 [27]: fig 7). In Haestasaurus, the deltopectoral crest is most prominent at approximately 0.35 of humerus length from the proximal end. Although the deltopectoral crest does not extend as far distally as in most sauropods, we do not regard this as a diagnostic character state because it might be related to the small size of Haestasaurus, and a few other taxa (e.g. Barapasaurus, Epachthosaurus, Nigersaurus, Omeisaurus) have similar values ( Table 2). The crest is a transversely thin plate in its most prominent section, projecting anteriorly and slightly laterally (Fig 2A). Thus the deltopectoral crest of Haestasaurus is restricted to the lateral margin (i.e., the plesiomorphic state present in most sauropods and some titanosauriforms) rather than displaying the derived medial deflection of the distal part observed in certain titanosauriforms such as Alamosaurus (USNM 15560 [92]:  7) (see [1,7,10,18]). In Haestasaurus, the posterolateral surface of the humerus (posterior to the deltopectoral crest) forms a broad rounded ridge that extends distally. Between this ridge and the lateral surface of the deltopectoral crest there is therefore a shallow vertical fossa.
The humerus displays little torsion along the shaft, resulting in the long-axes across the proximal and distal articular faces being approximately parallel to each other. In medial view, the posterior margin of the humerus is strongly concave. Both the lateral and medial margins of the diaphysis are concave in anterior view (Fig 2A). Possession of a straight lateral margin is a derived state that occurs in some titanosauriforms, such as Alamosaurus (USNM 15560 [92]:  [11,18]). The midshaft is wider transversely than anteroposteriorly, and has a rounded 'D'-shaped cross-section with a flattened, but still mildly convex, anterior face.
The posterior surface of the distal shaft bears a moderately deep anconeal (= 'cubital' or 'supracondylar') fossa bounded by two vertical ridges, the medial one being more prominent and acute, and the lateral one more rounded. This asymmetry in the ridges bounding the anconeal fossa occurs in several titanosaurs, including Argyrosaurus (MLP 77-V-29-1 [77]: p.616) and Neuquensaurus (MLP-CS 1050 [90]: fig 3, PU and PDM pers. obs. 2013). An enlarged anconeal fossa occurs in many somphospondylans [1,36], but see 'New and Revised Characters' below, for quantification and further evaluation of this character. In anterior view, the distal end projects nearly as far medially as the medial process of the proximal end. This feature was proposed as an autapomorphy of Haestasaurus by Upchurch et al. [36], but many sauropods display some degree of medial flaring of the distal humerus (e.g. "Astrodon johnstoni", USNM 2263 [97]: fig 3.10), and the comparative measurements in Table 2 demonstrate that the distal end of the Haestasaurus humerus is not noticeably wider relative to the proximal end than in other sauropods. We therefore propose abandonment of this character state as an autapomorphy of Haestasaurus (contra [36]). The anterior face of the distal end of the humerus bears two prominent processes located just medial and lateral to the midline. These processes (here termed the lateral and medial anterodistal processes, Fig 3D, [14,17] (see Fig 4). In Haestasaurus, between these prominent processes, there are two smaller vertical ridges located virtually on the midline ( Fig  3D). These two ridges have not been observed in any other sauropod, except possibly one such ridge in Apatosaurus excelsus (YPM1980 [100]: pl. 48, fig 4) and Diamantinasaurus (AAOD 603 [80]: fig 13B, PU and PDM pers. obs. 2012), and this feature is therefore provisionally regarded as an autapomorphy of Haestasaurus. The anteromedial corner of the distal end is also drawn out into an anterior process (here termed the anterior entepicondylar process: Fig 3D). The latter process projects almost as far anteriorly as the lateral and medial anterodistal , it is provisionally regarded as an autapomorphy. The distalmost part of the medial surface is flat and faces medially. This surface is particularly wide anteroposteriorly because it is extended by the anterior entepicondylar process. The lateral surface of the distal shaft forms a vertical ridge that is rounded anteroposteriorly. This ridge projects laterally, defining the posterior wall of a broad and deep anterolateral fossa (Fig 3A). Such a ridge and fossa are frequently present in sauropod humeri (e.g. Diamantinasaurus (PU and PDM pers. obs. 2012) and Saltasaurus [PVL 4017-63, PU and PDM pers. obs. 2013]). The depth of this fossa is variable, reflecting the relative prominence of the lateral anterodistal process and lateral ridge (and perhaps also post-mortem crushing in some cases). The distal articular surface of the humerus is rugose and mildly convex anteroposteriorly: thus, it does not curl strongly up onto the anterior and posterior surfaces of the shaft unlike the derived condition observed in many titanosaurs, especially saltasaurids such as Saltasaurus and Neuquensaurus [103]. In anterior view, the distal articular surface is nearly flat (Fig 2A): thus Haestasaurus lacks the division of the distal humerus into distinct rounded ulnar and radial condyles that occurs in several derived titanosaurs such as Opisthocoelicaudia (Z.PAL MgD-I/ 48 [27]: fig 7B and 7D) and Saltasaurus (PVL 4017-63, PU and PDM pers. obs. 2013) (see character no. 164 in [10]).
Ulna (Figs 5 and 6, Table 1). The ulna is complete but was broken and repaired at approximately midlength. As noted by Upchurch [6], the Haestasaurus ulna is a relatively robust element. For example, the maximum width of the proximal end of the ulna is 0.44 of the proximodistal length of the element. According to Wilson ([10]: character no. 168), values of this ratio above 0.4 represent a synapomorphy of Isisaurus+Saltasauridae, although he also noted that this occurs independently in some non-neosauropod taxa such as Mamenchisaurus. Robust ulnae also occur in Apatosaurus louisae (CM 3018 [104]), Bellusaurus ([105]: fig 9) and Hudiesaurus (IVPP V11121-1 [106]): thus, some caution is required when interpreting the robust ulna of Haestasaurus as evidence for titanosaurian affinities.
The proximal articular surface is generally rugose. In proximal end view (Fig 6C), the ulna is 'V'-shaped, with an anteromedial process that is 1.29 times as long as the anterolateral process (Table 2). D'Emic [17] suggested that Haestasaurus possesses a derived state that is also seen in other titanosauriforms, in which the anteromedial process is longer than the anterolateral one: however, the values for this ratio shown in Table 2 cast some doubt on the validity of this character as an indicator of titanosauriform affinities (see 'New and Revised Characters' below for further consideration of this issue).
The olecranon of the Haestasaurus ulna is well developed as a rounded region that is higher than the articular surfaces of the anterior processes. The articular surface of the anteromedial process is concave along its length ( Fig 5A) and flat transversely. Both the prominent olecranon and concave anteromedial process are derived states that are moderately well-developed in some basal titanosauriforms, including the brachiosaurid Giraffatitan and the somphospondylan Sauroposeidon (= Paluxysaurus: see [17,18]), and more strongly marked in advanced titanosaurs such as Alamosaurus (USNM 15560 [92] [17] argued that Haestasaurus is likely to be a member of the Titanosauriformes: however, it should be noted that a distinct olecranon and concave anteromedial process also occur in some probable non-titanosauriforms, such as Hudiesaurus (IVPP V.11121-1 [106], PU and PDM pers. obs. 2007) and Janenschia (MfN MB. R.2095.11 [87]: pl. 17, fig 7a; see also [6,17,18]). In Haestasaurus, the surface of the anteromedial process does not slope strongly distally, unlike the condition observed in some basal macronarians such as Tehuelchesaurus and Lusotitan (character no. 235 in [18]). The anterolateral process has an articular surface that is flat both transversely and longitudinally, but slopes quite strongly downwards towards its tip, as occurs in other sauropods such as Epachthosaurus (UNPSJB-PV 920 [89] 5) and Rapetosaurus (FMNH PR 2209 [91]: 37D). In Haestasaurus, both processes are transversely narrower near their bases and then widen slightly towards their tips, before tapering to points. The third, posterior, process of the proximal end is relatively small, unlike the large processes that occur in some titanosaurs, such as Diamantinasaurus (AAOD 603 [80]: fig 15F) and Saltasaurus (PVL 4017-74, PU and PDM pers. obs. 2013) (see 'New and Revised Characters' below). In Haestasaurus, the posterior process gives rise to a vertical ridge that is wide mediolaterally and has an almost flat surface at its proximal end, becoming less prominent and more rounded distally. The anterolateral and anteromedial processes also produce prominent vertical ridges that extend distally and define the fossa for reception of the proximal end of the radius. The anterolateral ridge is rounded throughout its length, whereas the anteromedial ridge is sharper. A weak rugosity is present in the strongly concave radial fossa. Close to the proximal end, the lateral surface of the anterolateral process is mildly concave, whereas the medial surface of the anteromedial process is more strongly concave in a region that lies further distally.
The subtriangular cross-section of the proximal end persists along the shaft up to approximately midlength. Here a sharp interosseous ridge develops on the anterior face and so transforms the cross-section into a square with rounded corners. Posteromedial to the interosseous ridge, the anteromedial surface is shallowly concave. This concavity extends distally and merges into the mildly concave area that faces anteromedially and lies immediately above the distal end itself. This anteromedial concavity articulates with the distal part of the posterolateral surface of the radius. There is no vertical ridge and associated groove on the posterolateral surface of the ulna, immediately above the distal end, unlike the ulnae of Losillasaurus and Turiasaurus [109,110]. Towards the distal end, the ulna widens markedly anteroposteriorly, mainly as a result of posterior expansion (Fig 6A and 6B): thus, Haestasaurus lacks the derived unexpanded distal end that occurs in some titanosaurs such as Diamantinasaurus (AAOD 603 [80]: fig 15) and Epachthosaurus (UNPSJB-PV 920 [89]: fig 9B and 9C, PU and PDM pers. obs. 2013) (see [17]). In distal view (Fig 6D), the articular surface of the Haestasaurus ulna has a comma-shaped outline, with a strongly rounded posterior margin and tapering anterior process that bears the anteromedial fossa for the radius (this character might be phylogenetically informative-see 'New and Revised Characters' below). The distal end surface is mildly convex both anteroposteriorly and transversely, and is strongly rugose, as is typical for most sauropods.
Radius (Figs 7 and 8, Table 1). The radius is virtually complete, lacking only the tip of the medial process of the proximal end (Figs 7 and 8). This element was originally recovered intact, but was broken into three portions that fit together at the breaks. These breaks have been repaired recently.
The radius is a relatively slender element, with a proximal transverse width to shaft length ratio of approximately 0.25 (Table 1). Such slender radii represent the plesiomorphic state, in contrast to the robust elements (proximal width to length ratio equals 0.30 or higher) observed in titanosaurian sauropods [6] and some probable non-titanosaurs (e.g. Aragosaurus, I.G. 484 [111] fig 8). Typically, sauropods with robust ulnae (see above) also have robust radii. However, Haestasaurus appears to be unique in possessing a robust ulna but a relatively slender radius, and this combination of character states is provisionally regarded as autapomorphic. As in most sauropods, the transverse width of the proximal articular surface is greater than its maximum anteroposterior width in Haestasaurus (Fig 8C, Table 1). Also, Haestasaurus lacks the derived condition seen in Turiasaurus and Zby in which the anteroposterior width of the proximal end of the radius is less than 0.5 that of the distal end [112]. There is a small degree of twisting between the proximal and distal ends of the radius, such that the long-axes of the two surfaces are not quite in the same plane: however, Haestasaurus does not display the strong twisting (greater than 45°) that occurs in the radii of taxa such as Epachthosaurus (UNPSJB-PV 920, PU and PDM pers. obs. 2013), Huabeisaurus (HBV-20001 [113]: fig 19), Malawisaurus [95] and Rapetosaurus (FMNH PR 2209 [91]: fig 36C,D) (see [18]). The proximal articular surface is rugose (especially towards its margins) and mildly concave centrally. It has a subtriangular outline, created by an anteroposteriorly wide and nearly straight lateral margin and a tapering medial process (Fig 8C). In other sauropods such as Diamantinasaurus (AAOD 603 [80]: fig 12D), Epachthosaurus (UNPSJB-PV 920, PU and PDM pers. obs. 2013), Patagosaurus (MACN 932, PU and PDM pers. obs. 2013) and Tehuelchesaurus (MPEF-PV 1125, PU and PDM pers. obs. 2013), the greatest anteroposterior width of the proximal radius occurs close to the midline of the element and the proximal end has an elliptical or oval profile (Fig 9; see 'New and Revised Characters' below). Therefore, we regard the nearly straight and anteroposteriorly widened lateral margin as a potential autapomorphy of Haestasaurus. Curry Rogers ([11]: character no. 282, state 2) (see also character no. 46 in [18]) noted that the transverse width of the proximal end of the radius is less than that of the distal end in some titanosaurs, such as Alamosaurus and Rapetosaurus. The ratio of proximal end:distal end transverse width in Haestasaurus is > 0.89 (Table 1): the '>' reflects the fact that this is a minimum value because a small part of the proximal end is missing. When complete, it seems probable that this ratio was still less than 1.0 (estimated at 0.92), suggesting that Haestasaurus shares the derived state with several titanosaurs. However, this condition also occurs in some non-titanosaurs, including Mamenchisaurus youngi (ZDM 0083 [107) and Tehuelchesaurus (MPEF-PV 1125, PU and PDM pers. obs. 2013), so caution is required when interpreting the phylogenetic significance of this character.
The radius bows anteriorly in lateral and medial views (Fig 8A and 8B). In anterior view (Fig 7A), the shaft displays the typically sigmoid medial and gently concave lateral margins seen in other sauropod radii (Fig 10). In Haestasaurus, this profile is largely produced by expansion of the distal end, which projects somewhat further laterally than the proximal end. In some sauropods (e.g. Mamenchisaurus and Camarasaurus), the proximal end projects as far laterally as the distal end, but many taxa (e.g. Ferganasaurus, Epachthosaurus, Diamantinasaurus) possess a milder version of the condition seen in Haestasaurus (Fig 10). The medial process of the proximal end merges into a vertical edge that extends distally and defines the margin where the anterior and posterior faces of the shaft meet each other. At approximately 20% of radius length from the proximal end, this margin bears a vertical ridge that extends   9) and several titanosaurs (such as Rapetosaurus, FMNH PR 2209 [91]: fig 36C).
At its midlength, the radial shaft has a rounded trapezoidal cross-sectional profile, with strongly convex lateral and medial margins and flattened anterior and posterior faces (the anterior one being narrower transversely than the posterior one). The flat anterior face extends to the distal end, widening transversely and becoming concave between two low vertical ridges. This means that the anterior margin of the distal articular surface is shallowly concave ( Fig  8D). This distally located fossa on the anterior surface is absent in other sauropods, including , and is therefore regarded as a potential autapomorphy of Haestasaurus (Fig 9I-9P). The flattened posterior surface of the shaft at midlength extends distally and bears a faint but distinct vertical interosseous ridge for attachment to the ulna. Distally, the posteromedial part of the radius produces a prominent bulge, and a smaller rounded projection also occurs posterolaterally ( Fig 8D): thus, the radius terminates distally in two posteriorly placed 'condyles'. Such distal radial condyles occur in several other sauropods and their distribution is evaluated in more detail in 'New and Revised Characters' below. In Haestasaurus, these condyles create a shallowly concave area on the posterior face, immediately above the distal end, and below the distal termination of the interosseous ridge described above. Another shallow concavity is situated posteromedial to the ridge that defines the medial margin of the anterior fossa, and anterior to the posteromedial condyle just described.
The transverse width of the distal end of the radius is 1.85 times the width of the midshaft (Table 2). Thus, Haestasaurus possesses the plesiomorphic state (i.e. values of this ratio less than 2.0) seen in taxa such as Shunosaurus, Omeisaurus, and Diplodocus, rather than the derived condition that occurs in a clade of titanosaurs comprising Nemegtosauridae, Isisaurus and Saltasauridae according to Wilson ([10]: character no. 170) (but note that Mannion et al. ([18]: character no. 47) found that values of 2.0 or higher are more widespread in Neosauropoda, occurring in Apatosaurus, Camarasaurus, Giraffatitan and Tehuelchesaurus, and the derived state is also seen in Zby, ML 368 [112], Table 2).
The distal articular surface of the radius is mildly convex and strongly rugose (Figs 7 and  8D). If the long-axis of the shaft is oriented vertically, the lateral portion of this articular surface slants strongly proximolaterally in anterior view (Fig 7A), at approximately 21°to the horizontal. This proximolateral bevelling of the distal radius was regarded as a synapomorphy of Saltasauridae by Wilson ([10]: character no. 171) (see also [114]): however, the definition and distribution of the derived state requires some clarification. Mannion et al. ( [18]: character no. 49) noted that many sauropods have radii in which the distal articular surface has a medial portion that is approximately perpendicular to the shaft long-axis, and a lateral portion that is bevelled. Therefore, for the purposes of measurement and comparison, the bevelling angle is estimated using only the lateral half of the distal articular surface. This angle varies from 0°in basal forms (e.g. Shunosaurus, ZDM T5402 [115]; Mamenchisaurus youngi, ZDM 0083 [107]: pl. xvii, fig 7), through values of around 20°in Apatosaurus louisae (CM 3018 [104]: fig 12A) and Tehuelchesaurus (MPEF-PV 1125 [15]: fig 17A, PU and PDM pers. obs. 2013), to more than 25°in many derived titanosaurs such as Alamosaurus (USNM 15560 [92]: fig 9A) and Opisthocoelicaudia (Z.PAL. MgD-I/48 [27]: fig 8B). Thus, the angle of 21°in Haestasaurus is consistent with its placement within basal Neosauropoda or Somphospondyli, rather than Titanosauria (see also [112] for discussion of strong distal radial bevelling in Turiasaurus and Zby).
Skin impression (Fig 11). The skin impression (NHMUK R1868) associated with the Haestasaurus fore limb was the first specimen of dinosaurian integument to be discovered [22,26]. Upchurch et al. [1] described the Haestasaurus skin impression as being composed of 'ossicles'. If correct, this would represent the stratigraphically oldest evidence for osteoderms in the sauropod fossil record and would also strengthen the suggestion (e.g. [6]) that this taxon belongs to the Titanosauria or even the Lithostrotia. However, as noted by D'Emic et al. [116], the specimen actually represents an impression of the integument, not ossifications.
The Haestasaurus skin impression comprises numerous small hexagonal scales that range in size from approximately 10-25 mm in diameter (Fig 11) (see also [117] and [118]: fig 2). These scales tessellate and cover an area of approximately 195 mm x 215 mm. They might have become smaller towards the elbow joint as also occurs in iguanodonts and hadrosaurs according to Steel [66]. If correct, this pattern presumably reflects greater flexibility in the region of the joint [36]. Steel [66] argued that the convex surfaces of these scales (the side that is exposed) actually faced inwards towards the limb bones, based on the observation that their flat surfaces are covered by matrix. However, the dermal remains found with Saltasaurus [119], suggest that the exposed convex surfaces might represent the exterior of the skin [36].
The Haestasaurus skin impression generally resembles several others assigned to sauropod taxa. For example, a small patch of skin (MWC 6718, a thin carbonaceous film), from the Late Jurassic Mygatt-Moore Quarry, comprises seven hexagonal tubercles and was found in close association with remains of Apatosaurus excelsus [26]: fig 2A]. There are several other reports of similar integumentary structures found in association with Morrison Formation sauropods (see review in [26]), and also the basal macronarian Tehuelchesaurus (MPEF-PV 1125 [120]: figs 2 and 3). The tubercles of these impressions or films resemble those of Haestasaurus in both size and shape. Thus current evidence suggests that asymmetrical hexagonal scales or tubercles that tessellate rather than overlap were probably widespread among at least Neosauropoda.
Czerkas [121] and Foster and Hunt-Foster [26] noted that some sauropod skin impressions or carbonaceous films demonstrate the presence of small (1-2 mm diameter) tubercles on top of the larger hexagonal scales. However, these smaller tubercles cannot be observed in Haestasaurus, perhaps because the latter is represented by an impression in relatively coarse grained sediment, or because they were destroyed during preparation of the specimen. Anatomy and Relationships of "Pelorosaurus" becklesii

Data Sets
In order to determine the phylogenetic relationships of Haestasaurus, we have scored this taxon for the data matrix of Carballido and Sander [19] (the 'CSM' hereafter), and have revised the scores for "Pelorosaurus" becklesii in the 'Lusotitan Standard Discrete Matrix' (LSDM) and 'Lusotitan Continuous and Discrete Matrix' (LCDM) of Mannion et al. [18]. The CSM has the advantage of a larger and broader taxon sample (71 versus 63 taxa) and more characters (341  Table 2). Note that X excludes the anterodistal processes if present: this is because these processes are often absent or highly reduced in titanosaurs, and their inclusion in X would mean that the fossa depth ratio would be estimated in an inconsistent manner across Sauropoda. Abbreviation: af, anconeal fossa.  . Abbreviations: alp, anterolateral process; amp, anteromedial process; pp, posterior process; raf, fossa for reception of the proximal end of the radius. The blue lines in A mark the long-axes of the anteromedial and anterolateral processes; X, Y and Z mark the tip of the anteromedial process, the tip of the anterolateral process and the intersection of the process long-axes respectively. The anteromedial:anterolateral process length ratio ('Uppl' in Table 2) can thus be defined as X-Z/Y-Z. versus 279) than the Mannion et al. [18] matrices. However, the latter study might be better placed to assess the relationships of a putative basal titanosauriform, such as Haestasaurus, because its taxon and character sampling strategy focussed on this region of the sauropod tree. In addition, whereas the CSM and LSDM treat all character states as discrete, the LCDM allows exploration of the impact of treating quantitative characters as continuous data [18].
"Pelorosaurus" becklesii was not included in the original CSM: we have therefore scored Haestasaurus for this matrix based on the anatomical information presented above. The scores for Haestasaurus and the modified CSM (with six new characters, see below) are presented in S1 and S2 Files.
The LSDM and LCDM have been modified in several ways since publication. First, we have incorporated the revised scores for Diamantinasaurus, Wintonotitan, Malawisaurus and Rapetosaurus proposed by Poropat et al. [80,81]. Second, most of the scores for "Pelorosaurus" becklesii employed by Mannion et al. [18] are accepted here. However, we have made the following five changes based on our more detailed examination of Haestasaurus: 1. C46. Radius, mediolateral width of proximal to distal end ratio: 1.0 or greater (0); less than 1.0 (1) ( [11], modified, quantified, and polarity reversed in [18]. Note that in taxa with a twisted radius, the dimension of the long axis of the distal end is used). In the LCDM this was scored with a value of 0.92 in Mannion et al. [18]. However, after new measurements were taken, the actual value of this ratio is 0.89 (probably slightly higher because of the small portion missing from the medial process of the proximal end). Here, therefore, we have adjusted this score for Haestasaurus to be 0.89-0.92 in the LCDM, in order to reflect this slight uncertainty.
2. C47. Radius, distal end mediolateral width to midshaft mediolateral width ratio: less than 2.0 (0); 2.0 or greater (1) ( [10,122], modified by [18]. Note that in taxa with a twisted radius, the dimension of the long axis of the distal end is used). The new measurements for the Haestasaurus radius mean that its score in the LCDM has been adjusted from 1.71 in [18] to 1.852 here.
3. C48. Radius, distal end mediolateral to anteroposterior width ratio: 1.5 or greater (0); less than 1.5 (1) ( [9], quantified and polarity reversed by [18]). The new measurements for the Haestasaurus radius mean that its score in the LCDM has been adjusted from 1.405 in [18] to 1.527 here. This also means that the score for Haestasaurus in the LSDM has been changed from 1 to 0.
4. C228. Humerus, distal-most part of the posterior surface (supracondylar fossa) is: flat or shallowly concave (0); deeply concave between prominent lateral and medial vertical condylar ridges [1,18]. Haestasaurus was scored with state 1 by Mannion et al. [18]. However, although the ridges that define the anconeal (= supracondylar) fossa are fairly prominent in this taxon, the actual depth of the fossa is shallow (see Table 2 and 'New and Revised Characters' below'). We have therefore changed the score for this character to state 0 in both the LSDM and LCDM.
The revised character scores for the LSDM and LCDM and the complete data matrices (with six new characters, see below) are presented in S1, S3 and S4 Files.

New and Revised Characters
The detailed description and comparison of Haestasaurus has highlighted some problems with existing fore limb characters, and has also identified several new and potentially phylogenetically informative features. Below, we briefly discuss three revised characters ("RC") and six new characters ('NC'). The latter have been added to the CSM, LSDM and LCDM. Complete scores for all taxa for NC1-6 are presented in Tables B and C in S1 File. These new characters have been added at the end of each character set, so that NC1-6 form characters 342-347 in the CSM and characters C280-285 in the LSDM and LCDM, respectively (this has been done so that the original character numbers used by Carballido and Sander [19] and Mannion et al. [18] remain unaltered, and therefore correspond with those cited in the text here).
Character revisions and comments. RC1. 256. Humerus, RI (sensu [31]): gracile (less than 0.27) (0); medium (0.28-0.32) (1); robust (more than 0.33) (2) [123]. Table 2 indicates that the RI can reach particularly low values (0.21) in brachiosaurids such as Cedarosaurus and Giraffatitan, reflecting the relative elongation of the humerus in such taxa. At present, this potential synapomorphy of Brachiosauridae, or a clade within this family, is not captured by the state definitions. In future analyses, it might be appropriate to introduce an additional state, or simply treat the RI values as continuous data.
RC2. C228. Humerus, distalmost part of the posterior surface (supracondylar fossa) is: flat or shallowly concave (0); deeply concave between prominent lateral and medial vertical condylar ridges (1) [1,18]. This fossa varies in depth among sauropod taxa (see Fig 4), and a deep fossa, bounded by acute lateral and medial ridges, has generally been regarded as a derived state that characterises most somphospondylans (e.g. [1,18]): however, the depth of this fossa has not been quantified. Here, we have devised a simple ratio to capture fossa depth (Fig 12) and use this to redefine this character in quantitative terms as follows: 'Humerus-anconeal fossa /depth: shallow (depth of fossa divided by anteroposterior width of humeral distal end) is less than 0.4 (0); deep, fossa depth ratio is 0.4 or higher (1) (N.B. the anteroposterior width of the distal humerus ['X' in Fig 12] excludes the depth of the anconeal fossa itself and the contribution made by the lateral and medial anterodistal processes).' The boundary between states is based on the data in Table 2, and it appears that most titanosauriforms have a fossa depth ratio of at least 0.4 (e.g. Elaltitan, Epachthosaurus, Giraffatitan, Malawisaurus and Rapetosaurus), whereas non-neosauropod eusauropods, diplodocoids and basal macronarians typically have values of 0.2 or lower. Based on this criterion, Haestasaurus has state 0, and its score has therefore been adjusted here (from its previous score of state 1) in the LSDM and LCDM. One other aspect of this feature requires further investigation. Haestasaurus and many titanosauriforms appear to have more prominent lateral and medial ridges that define the margins of the anconeal fossa. The medial ridge is more prominent than the lateral one, and rather than being transversely rounded (as in more basal taxa) those of titanosauriforms have more acute apices. It seems that ridge shape and fossa depth are not completely correlated: future phylogenetic analyses might need to treat these two features as independent characters. However, some caution is required because the apparent independence of ridge shape/prominence and anconeal fossa depth might be an artefact produced by post-mortem crushing of some humeri. Furthermore, it is difficult to quantify the difference between a low rounded ridge and a more prominent and acute one.
RC3. C51. Ulna, ratio of maximum mediolateral width of proximal end (equivalent to anteromedial arm) to maximum anteroposterior width of proximal end (equivalent to anterolateral arm): less than 2.0 (0); 2.0 or greater (1) ( [10], modified and quantified by [18]; see also [17]). There are two problems associated with the character state definitions used by Mannion et al. [18]. First, the use of maximum anteroposterior width of the proximal ulna to represent the length of the anterolateral process means that the posterior process usually contributes to this measurement. However, as noted in NC2 below, the posterior process itself varies in relative prominence among sauropods. Second, the process length ratio can alter markedly because of changes in the orientations of the anterolateral and anteromedial processes, rather than because of changes in their relative lengths. In particular, anteromedial processes can project more medially or more anteriorly; such shifts in orientation affect the mediolateral width of the proximal ulna, even if the relative length of the process has not changed. In order to avoid possible inconsistencies in measurement and comparison, we propose a standardised method for defining the lengths of these processes (as illustrated in Fig 13) that excludes the size of the posterior process. Essentially, the long-axes of the anteromedial and anterolateral processes are extrapolated posteriorly so that they intersect close to the position of the olecranon. Process length is then measured from this intersection to the tip of each process. The resulting process length ratios are shown in Table 2, where they can be compared with those obtained by Mannion et al. [18]. We have utilised these revised values here. As a result, the "Cloverly titanosauriform' has been scored with state 0 (rather than 1) for C51 in the LSDM. Moreover, many of the scores for C51 in the LCDM have been modified (see Table A in S1 File for details).
We noted above (see 'Description') that D'Emic [17] supported inclusion of Haestasaurus within Titanosauriformes on the basis that the anteromedial process of the ulna is longer than the anterolateral one. Here, the data in Table 2 do not support this view, irrespective of whether the process length ratio is based on the state definition used by Mannion et al. [18] or the revised version proposed here. Most sauropods have an anteromedial process that is longer than the anterolateral one, and especially long anteromedial processes (i.e. process length ratio of 2.0 or higher) are restricted to a minority of titanosauriforms (e.g. Cedarosaurus, Epachthosaurus and Venenosaurus). NC2. Ulna-posterior process of proximal end: is weakly developed, so that the proximal profile of the ulna is 'V'-shaped (formed by the anteromedial and anterolateral processes) (0); is strongly developed, so that the proximal profile of the ulna is 'T'or 'Y'-shaped, and there is a deep fossa between the anteromedial and posterior processes rivalling the radial fossa in depth (1) (Fig 13). The plesiomorphic state (Fig 13A) 16B). In contrast, the derived state is seen in several somphospondylans, such as Diamantinasaurus (Fig 13B), Magyarosaurus (NHMUK 3859, PU and PDM pers. obs. 2014), Rapetosaurus (FMNH PR 2209 [91]: fig 37D), Saltasaurus (PVL 4017-74, PU and PDM pers. obs. 2013) and Wintonotitan (QM 7292 [81]: fig 10E). In these somphospondylans, it can sometimes be difficult to correctly determine whether an ulna is from the right or left side because the deep fossa created between the anteromedial and posterior processes can be mistaken for the radial fossa, and the relatively short anterolateral process can be confused with the enlarged posterior process (especially when the proximal profile is more 'T'-shaped, as in Rapetosaurus). However, the anterior surface of the ulna can be identified by the well-developed interosseous ridge for attachment of the radius, which lies distal to the true radial fossa.
NC3. Ulna-shape of the distal end: comma-shaped, with tapering curved anterior process associated with an anteromedial fossa for reception of the radius (0); elliptical or subtriangular in outline, with the anteromedial fossa reduced or absent (1) (Fig 14). The comma-shaped distal end, and associated anteromedial fossa for reception of the distal radius, is widespread among neosauropods, occurring in taxa such as Apatosaurus excelsus (YPM4633 [100]: pl. 52, fig 4),  15H), Epachthosaurus (UNPSJB-PV 920, PU and PDM pers. obs. 2013) and Saltasaurus (PVL 4017-74, PU and PDM pers. obs. 2013) also lack the 'comma'-shaped distal ulna, and the reduced fossa for the reception of the distal radius might have faced more anteriorly rather than anteromedially. In at least some titanosaurs, such as Diamantinasaurus (Fig 14B), the distal ulna is heart-shaped/subtriangular rather than elliptical in outline. At present, this condition is scored as state 1 because it is different from the comma-shape scored as state 0. In future analyses, it might be preferable to score the heart-shaped distal end as a separate state from the elliptical one: however, we have not done this here because the distribution of these morphologies requires further investigation via firsthand observation of additional titanosaurian taxa.

Analytical Methods
The CSM, LSDM and LCDM were analysed using TNT vs. 1.1 [127]. The multistate characters treated as ordered by Carballido and Sander [19] and Mannion et al. [18] were also treated as such in all analyses here. For all data sets, the New Technology search was first applied (including Drift, Sectorial Searches and Tree Fusing) with the consensus being stabilised 10 times. The most parsimonious trees (MPTs) produced by these searches were then used as the starting trees for two consecutive rounds of Traditional (heuristic) search using TBR. The LCDM was analysed in the same way, but with application of Implied Weights (with a k value of 3) [18,128].
The robustness of the resulting relationships was evaluated by generating relative bootstrap frequencies (known as GC values) using symmetric resampling in TNT [127,129]. In each case, 5000 bootstrap replicates were generated using the Traditional Search with TBR.
The character states that support the relationships of Haestasaurus in the various MPTs have been examined using character mapping in Mesquite vs. 2.75 [130].

Results
Carballido and Sander [19]-CSM. Analysis of the CSM in TNT yielded 99,999 MPTs of length 1126 steps, before the available memory became full. The 'pruned trees' option was then applied to these MPTs. This indicated that Tendaguria was the most unstable taxon. Tendaguria was therefore excluded a priori, and the analyses were run again. This produced 28 MPTs of length 1123 steps. The strict consensus tree of these 28 MPTs is shown in Fig 15, along with Mannion et al. [18]-LCDM. Analysis of the LCDM in TNT yielded 17 MPTs of length 105.53919 steps (N.B. the non-integer tree length, and a tree length that is shorter than the total number of characters, result from the use of continuous data and the application of implied weights [18]). The strict consensus tree of the 17 MPTs is shown in Fig 17, along with GC values. In this strict consensus tree, Haestasaurus is placed in a clade with Janenschia and Dongbeititan. This clade is more closely related to Saltasaurus than is Andesaurus, so technically Haestasaurus is placed within Titanosauria. However, this analysis has produced an unusual result in which Andesaurus, and several other taxa that are normally considered to be somphospondylans (e.g. Chubutisaurus), form a clade that is the sister-taxon to a clade that comprises traditional brachiosaurids.

Haestasaurus and Pelorosaurus
As explained in the 'Historical Background' section above, Haestasaurus has been known as "Pelorosaurus" becklesii since its discovery and announcement [22]. In order to justify the new generic name, it is not only necessary to identify autapomorphies of H. becklesii, but also to demonstrate that this taxon is distinct from Pelorosaurus conybeari. The latter taxon is known from the syntype specimens comprising a right humerus (NHMUK 28626), four anterior caudal vertebrae (NHMUK R2544-2547) and three chevrons (NHMUK R2548-2550), as well as some referred middle and distal caudals [36]. Thus, at present, only the humerus allows comparisons between Pelorosaurus conybeari and Haestasaurus becklesii.
Descriptions, figures and photographs of Pelorosaurus conybeari have been presented in several recent studies ( [28,34] and [36] : fig 7), and these will not be repeated here. However, we have provided a new figure of the right humerus of Pelorosaurus conybeari (Fig 18) in order to facilitate comparisons with Haestasaurus. The 1320 mm long Pelorosaurus conybeari humerus is incomplete, lacking both the lateral and medial margins of the proximal end (Fig 18). Nevertheless, it is clear that the humeri of Pelorosaurus and Haestasaurus are very different in terms of their robustness. The RI [sensu 31] for Haestasaurus is 0.33, whereas that for P. conybeari is estimated to be approximately 0.23 (Table 2). Thus, whereas Haestasaurus would be scored with state 2 for Carballido and Sander's [19] character no. 256 dealing with humeral RI values, Pelorosaurus would be scored with state 0. Although the proximal end of the Pelorosaurus humerus is badly damaged, enough remains to indicate that the proximal articular surface was convex transversely and curved distally towards the proximolateral corner ( Fig 18B): this suggests that the proximolateral corner was rounded rather than 'square', and that the lateral portion of the proximal articular surface lay below the level of the humeral head. If correct, then Pelorosaurus retains the plesiomorphic state, whereas Haestasaurus possesses the derived state that occurs in most somphospondylans (see 'Description'; above). The lateral margin of the Pelorosaurus humerus is straight proximodistally, whereas Haestasaurus retains the plesiomorphic concave profile. Finally, although both genera would be scored with state 0 for the character pertaining to anconeal fossa depth (see 'RC2' above), this fossa is more prominent and bounded by well-developed lateral and medial ridges in Haestasaurus, whereas in Pelorosaurus it is very shallow (    Haestasaurus differ in at least four humeral characters that have been employed in recent phylogenetic analyses (e.g. [17][18][19]).
The two humeral autapomorphies of Haestasaurus both pertain to the distal end. Unfortunately, this region is damaged and worn in Pelorosaurus. Thus, it is not clear whether the latter taxon possessed one or two small ridges between the lateral and medial anterodistal processes (indeed, these processes are themselves not well preserved). However, enough of the distal end is preserved to suggest that Pelorosaurus lacked the prominent anterior entepicondylar process that characterises Haestasaurus.
In short, the limited data available at present indicate that Haestasaurus becklesii and Pelorosaurus conybeari differ in at least five character states pertaining to the humerus alone. This indicates that they are distinct taxa and should be treated as separate genera.

Phylogenetic Affinities of Haestasaurus
As noted in the 'Historical Background' section above, most recent studies have concluded that Haestasaurus becklesii is a titanosauriform, but there has been disagreement over whether it is a basal member of this clade or a more derived titanosaur (e.g. [1,6,17,18,36]). The phylogenetic analyses presented here all confirm that Haestasaurus is at least a macronarian. This is supported by the presence of derived states such as the square proximolateral corner of the humerus and the associated straight lateral portion of the proximal end that lies level with the humeral head (character nos. 260 and C223 in the CSM and LSDM/LCDM respectively). Moreover, all analyses, both here and by other workers, agree that Haestasaurus is not a brachiosaurid. None of the putative brachiosaurid synapomorphies discussed by D'Emic [17] and Mannion et al. [18] can be observed in Haestasaurus, and the humerus of the latter taxon clearly lacks the rather slender and elongated morphology that characterises most brachiosaurids (e.g. see [131]).
The current phylogenetic analyses tend to support one of two placements of Haestasaurus: (1) closely related to taxa such as Camarasaurus, Janenschia and Tehuelchesaurus at the base of Macronaria (CSM and LSDM); or (2) positioned as a basal member of Titanosauria (LCDM). It seems likely, therefore, that some of the fore limb characters tend to exclude Haestasaurus from Titanosauriformes, whereas others represent synapomorphies that unite this English taxon with titanosaurs. However, given the large amounts of missing data and homoplasy, most of these characters form weak synapomorphies with low individual consistency and retention indices. In particular, Haestasaurus is very incomplete: it can be scored for just 21 (6.1%) and 31 (10.9%) of the characters for the modified CSM and LSDM/LCDM respectively. Moreover, homoplasy seems to be rife in fore limb characters for sauropods. For example, consider character nos. 264 and C50 in the CSM and LSDM/LCDM respectively, in which the derived state is a robust ulna. Based on the results of Mannion et al. [18], it appears that robust ulnae have evolved at least four times independently in the following taxa: (1) a potentially monophyletic assemblage of Jurassic Chinese sauropods that includes Bellusaurus, Hudiesaurus and some species of Mamenchisaurus; (2) the diplodocoid Apatosaurus; (3) basal macronarians such as Janenschia and Tehuelchesaurus; and (4) derived titanosaurs such as Opisthocoelicaudia and Saltasaurus. Finally, many of the available fore limb characters relate to the proportions of elements. The phylogenetic significance of such quantitative characters is much more strongly influenced by the methodological strategy adopted by a systematist than would be the case for a simple qualitative character (e.g. an absence/presence character). Further insights into these issues, and the evidence supporting the alternative relationships of Haestasaurus, can be obtained by detailed character mapping.
In order to determine how the support for a given relationship of Haestasaurus varies with data set and tree topology, character mapping was carried out in Mesquite 2.75 [130]. tends to exclude Haestasaurus from Titanosauria or a slightly more inclusive clade (but there is considerable homoplasy in this character).
12. The transverse width of the distal end of the radius is less than twice the width of the shaft at midlength (character nos. 266 and C47 in the CSM and LSDMLCDM respectively). This character state distribution tends to exclude Haestasaurus from Titanosauria or Chubuti-saurus+Titanosauria (but this is highly homoplastic, with taxa such as Apatosaurus, Camarasaurus and Giraffatitan also possessing the derived state [see 'Description']).
13. Relatively little anteroposterior compression of the distal end of the radius (mediolateral: anteroposterior width ratio is less than 1.9) (character no. C48 in the LCDM). This character state distribution tends to exclude Haestasaurus from Saltasauridae (though note that Opisthocoelicaudia lacks the high ratio seen in other saltasaurids).
14. The distal radius possesses posterolateral and posteromedial condyles (character nos. 347 and C285 in the CSM and LSDM/LCDM respectively). This character state distribution tends to exclude Haestasaurus from Titanosauria (or possibly Titanosauriformes because the reversal that produces loss of the distal radial condyles in titanosaurs also occurs in taxa such as Cedarosaurus, Lusotitan and Venenosaurus). However, there is considerable homoplasy in this character, with the condyles occurring in some titanosauriforms such as Giraffatitan and Rapetosaurus.
In contrast, the following eight characters support inclusion of Haestasaurus within Titanosauriformes, or a less inclusive clade such as Somphospondyli, Titanosauria or Lithostrotia:

2.
A prominent olecranon process (character nos. 263 and C233 in the CSM and LSDM/ LCDM respectively). This character state distribution tends to support inclusion of Haestasaurus within Somphospondyli or Titanosauria (depending on the optimisation of missing data). However, as noted in the 'Description', several non-titanosaurs also possess relatively large olecranon processes.
3. Robust ulna (character nos. 264 and C50 in the CSM and LSDM/LCDM respectively). This character state distribution tends to support inclusion of Haestasaurus within Saltasauridae (but with several convergences elsewhere among sauropods).
4. The proximal anteromedial process of the ulna has a strongly concave articular surface in anterior or posterior view (character no. C234 in the LSDM/LCDM). This character state distribution tends to include Haestasaurus within Titanosauriformes, especially Titanosauria.
5. The radius:humerus length ratio is greater than 0.6 (character no. C44 in the LSDM/ LCDM). It appears that basal sauropods and eusauropods have values greater than 0.6, but this ratio drops below this value in many titanosauriforms. The presence of values greater than 0.6 in Diamantinasaurus, Haestasaurus, Opisthocoelicaudia and Rapetosaurus could therefore be interpreted as a derived reversal to the state seen in basal sauropods.
6. The proximal end of the radius is subtriangular in outline, tapering towards its medial process (character nos. 345 and C283 in the CSM and LSDM/LCDM respectively). This character state distribution tends to support inclusion of Haestasaurus within Titanosauriformes (although the derived state also occurs in a few taxa, such as Aragosaurus and Lusotitan, whose status as titanosauriforms is uncertain or contradicted by some phylogenetic analyses [18,111] Thus the evidence available from character mapping is equivocal, as is expected given the bimodal placement of Haestasaurus in the phylogenetic analyses presented above. Moreover, the fact that two separate data sets (CSM and LSDM) produce similar results when conventional parsimony analysis is applied, whereas a radically different topology is generated by the LCDM, suggests that the selection of different methodological approaches is key to the lack of consensus regarding the relationships of Haestasaurus. It could be argued that more weight should be given to the LCDM result because this data set avoids some of the problems associated with the discretisation of quantitative characters (e.g. see [132]). Moreover, Goloboff et al. [133] found that the application of implied weighting produced MPTs that are more stable over research time (i.e. topologies change less markedly as new taxa and characters are added) than those produced by conventional parsimony approaches. Despite these theoretical points in favour of the LCDM result, we are sceptical about the suggestion that Haestasaurus is a titanosaur. This is partly because the current LCDM MPTs contain a very unexpected topology in which a clade containing Andesaurus, Chubutisaurus and other taxa, is the sister-taxon to a cluster of traditional brachiosaurids (Fig 17). Such a result has not been found by any previous analysis, and the fact that this topology differs so markedly from the LCDM MPTs found by Mannion et al. [18] somewhat undermines the suggestion that the application of implied weighting produces more stable trees through research time. Finally, the majority of fore limb characters suggest that it is unlikely that Haestasaurus is a member of Titanosauria, and most of the few remaining characters support placement within Titanosauriformes rather than less inclusive clades such as Somphospondyli or Titanosauria. Here, therefore, we conclude that Haestasaurus is most plausibly interpreted as a non-titanosauriform macronarian, rather than a titanosauriform, somphospondylan or titanosaur, contrary to most recent studies [1,6,17,18,36,[75][76][77]. If correct, then Haestasaurus convergently acquired certain features of the fore limb seen in derived titanosaurs. These features include increased robustness of the humerus and ulna, the enlarged olecranon, a concave profile to the anteromedial process of the ulna, and strong proximolateral bevelling of the distal end of the radius. Many of these derived character states were initially considered to be synapomorphies of the Titanosauria, or a less inclusive titanosaurian clade (e.g. [1,6,7,10]), but have subsequently been shown to be more widespread among Sauropoda [17][18][19]. Presumably, the repeated occurrence of these features in different regions of the sauropod tree reflects convergence in fore limb function, but at present the biomechanical and palaeoecological significance of this phenomenon is unclear.

Sauropod Evolution in the Early Cretaceous
Jurassic sauropod faunas were dominated by various non-neosauropod eusauropod clades (e.g. mamenchisaurids, turiasaurians) and, in the Late Jurassic, the specialised flagellicaudatans [1,2]. At, or close to, the Jurassic-Cretaceous (J-K) boundary, sauropods underwent a major extinction, in which 60-80% of taxa disappeared [2,23]. Cretaceous sauropod faunas were rather different from those of the Jurassic, being dominated by narrow-tooth-crowned clades (notably rebbachisaurids and titanosaurs with dental slenderness indices [sensu 7] of 3.0 or higher) from the late Early Cretaceous onwards [2,25,134,135]. The earliest Cretaceous (Berriasian-Barremian) therefore represents an interesting 'transitional' period in which a few broadtoothed sauropods persisted (but dwindled to extinction) and titanosauriforms and rebbachisaurids radiated. Haestasaurus thus potentially represents one of the few non-titanosauriform macronarians that survived into the earliest Cretaceous. Table 3 summarises those putative Early Cretaceous sauropods that have either been proposed to be non-titanosauriform macronarians, or potentially represent broad-toothed nonmacronarians. There are 21 such taxa, but the phylogenetic relationships of most of these are uncertain. Many of the putative non-titanosauriform macronarians are more plausibly interpreted as titanosauriforms, somphospondylans or even titanosaurs. For example, Xianshanosaurus was identified as a basal neosauropod by Lü et al. [136], but this was not based on a phylogenetic analysis. The only formal cladistic analyses to examine the relationships of Xianshanosaurus are those of Mannion et al. [18] (and subsequent studies such as [80] and here, based on the LSDM and LCDM data sets), which include first-hand observations of this Asian taxon (PDM pers. obs. 2012). These analyses all agree that Xianshanosaurus is a titanosaur. When doubtful non-titanosauriform macronarians are excluded, there are only four genera that can be provisionally regarded as true basal macronarians: Aragosaurus, Galveosaurus, Haestasaurus, and the unnamed Dalton Wells sauropod. There are also four other broadcrowned (or at least potentially broad-crowned) sauropods, two of which are probably not macronarians (Losillasaurus and Turiasaurus) and two that have uncertain affinities (Oplosaurus and the unnamed Hell Canyon sauropod). However, it should be noted that the four Spanish taxa (Aragosaurus, Galveosaurus, Losillasaurus and Turiasaurus) might actually be Late Jurassic in age (Table 3): if so, then Early Cretaceous non-rebbachisaurid and non-titanosauriform sauropods are very rare indeed.
Current evidence suggests that basal macronarians and other broad-crowned sauropods died out in Gondwana at the end of the Jurassic, and only persisted into the earliest Cretaceous in Europe and North America. This pattern might indicate some regional variations in the J-K boundary diversity decline among sauropods, perhaps reflecting differences in environmental conditions. However, the record of earliest Cretaceous sauropods has probably been strongly influenced by sedimentary rock availability and sampling. In particular, sauropod-bearing deposits of Berriasian-Valanginian age are rare globally, especially in Gondwanan continents. The Paleobiology Database (www.paleobiodb.org: accessed 15 th August 2014) and Fossilworks (www.fossilworks.org: accessed 15 th August 2014) list only 13 collections of sauropod body fossils from the Berriasian-Barremian of Gondwana: (1) the Bajada Colorada Formation (late Berriasian-Valanginian), Argentina, which has yielded the diplodocid Leinkupal [137]; (2) an unnamed unit (late Hauterivian-early Barremian), Croatia, which has produced the rebbachisaurid Histriasaurus and indeterminate titanosauriforms [138,139]; (3) an unnamed unit (Berriasian-Hauterivian), Lebanon, which has yielded two collections of titanosauriform teeth (possibly a brachiosaurid) [140]; (4) the Cabao Formation (Hauterivian-Barremian), Libya, which has yielded an incomplete sauropod tooth [141]; (5) the Irhazer Shale Formation (Berriasian-Valanginian), Niger, which has produced six collections of sauropod remains [142]  Identified as a titanosauriform by Canudo et al. [152] and D'Emic [17], but these opinions were not based on inclusion of Aragosaurus into a phylogenetic analysis. Recovered as MN in all analyses to date (e.g. [18,80,109,111]; LSDM and LCDM here).
(but see below); and (6) two collections from the Kirkwood Formation (Berriasian-Valanginian), South Africa, which include "Algoasaurus" and other indeterminate sauropods [143,144]. It could be argued that the existence of these Gondwanan deposits means that we do have opportunities to observe Southern Hemisphere basal macronarians and/or broadcrowned taxa from the Early Cretaceous if they were present. Note, however, that the quality and quantity of the specimens from five of these formations is low, and 10 of the localities are limited to only a small part of Gondwana (i.e. the northern part of the Afro-Arabian plate). Moreover, recent studies [145,146] have argued that the age of the Irhazer Formation of Niger is probably Middle Jurassic, rather than Early Cretaceous. Thus, sampling of Early Cretaceous sauropod-bearing deposits in Gondwana would need to improve substantially before we can conclude, with any confidence, that basal macronarians were genuinely absent. Considerations of fossil record quality aside, we tentatively suggest that the J-K boundary extinction reduced the geographic range of broad-crowned sauropods by exterminating them in most regions apart from Europe and North America. Broad-crowned sauropods, especially non-titanosauriform macronarians, were still relatively diverse and abundant in Europe during the earliest Cretaceous (N.B. but this relative diversity strongly depends on the precise age of the four Spanish genera mentioned earlier). Their presence in the late Berriasian-Valanginian Lakota Formation and Barremian Dalton Wells Quarry implies the persistence of endemic North American lineages from the Late Jurassic into the earliest Cretaceous. Although poor sampling obscures the true geographic distribution of broad-crowned sauropods in the earliest Cretaceous, it seems likely that such taxa were restricted to Laurasia from approximately the Hauterivian onwards, and even here they declined in diversity and eventually disappeared by the end of the Early Cretaceous.

Conclusion
The English sauropod "Pelorosaurus" becklesii is based on fore limb elements and a skin impression. This taxon can be diagnosed on the basis of five autapomorphies, and is distinct from Pelorosaurus conybeari; therefore the new generic name Haestasaurus is warranted. The phylogenetic relationships of Haestasaurus remain uncertain, although it can be regarded as a nonbrachiosaurid macronarian based on several phylogenetic analyses. Moreover it is likely that Haestasaurus is a basal macronarian, perhaps closely related to Janenschia, Tehuelchesaurus and/or Camarasaurus. The evidence for any particular phylogenetic relationship is weak, mainly because of the incompleteness of the holotypic material of Haestasaurus and prevalent homoplasy in fore limb characters. Several new characters might make a useful contribution to future phylogenetic analyses, and here they generally reinforce the view that Haestasaurus is basal within Macronaria. If Haestasaurus is a basal macronarian, then this adds to the evidence that such taxa (also including Aragosaurus, Galveosaurus and the unnamed Dalton Wells sauropod) survived the Jurassic-Cretaceous boundary decline of diplodocid and broad-toothed sauropods and did not finally die out until some point in the late Early Cretaceous. However, it is not clear whether the occurrence of several Early Cretaceous non-titanosauriform macronarians solely in Europe and North America indicates differential regional patterns in extinction/ survival across the J-K boundary, or merely reflects poor sampling of earliest Cretaceous deposits elsewhere.