The mould of the skeleton is incomplete and articulated (Fig. 1). It lies mostly on its dorsal side. The skull and the anterior region of the postcranial skeleton are well preserved. The bones have been dissolved in the ferro-calcareous nodule but a silicone cast, recently made, reveals important and new anatomical characters described below.

The posterior half of the right maxilla is partly visible in lateral and palatal views (Fig. 2). Its lateral surface is smooth without any foramina. The dorsal flange of the maxilla rapidly decreases in height posteriorly, then more steadily. In contrast, the ventral margin is subhorizontal. The posterior extremity of the maxilla is missing. The postorbital region was pushed in medially post-mortem, so the jugal and quadratojugal are now located medial to their anatomical position while the squamosal has rotated counterclockwise, revealing the tympanic fossa and the quadrate. The cheek region is otherwise well preserved and only visible in lateral view. The jugal is a rod-like bone with weakly dilated extremities. Its anterior end shows a laterally directed but poorly defined surface for the suture with the posterior extremity of the maxilla. In contrast, the shortness and the dilated ends of the jugal of Procolophon trigoniceps (Carroll and Lindsay, 1985) produce an hourglass shape. The quadratojugal is subcomplete. It is roughly quadrangular with a smooth lateral surface. Dorsally, the quadratojugal meets the jugal along a straight suture. These two perpendicular elements form a right-angled subtemporal emargination, a configuration also known in Coletta seca (Modesto et al., 2002) and Sauropareion anoplus (MacDougall et al., in press). Morphological variation occurring in the latter (MacDougall and Modesto, 2011, MacDougall et al., in press and Modesto and Damiani, 2007) still indicates that this condition should be regarded as a variant of the broadly excavated emargination seen in most procolophonids (Cisneros, 2008a). It is distinct from the acute emargination formed by the boomerang-like jugal in Nyctiphruretus (Säilä, 2010b) and Owenettidae (Meckert, 1995 and Reisz and Scott, 2002). In the postorbital region, the supratemporal is partly preserved. Exposed in ventral view, it is a wide blade overlying the dorsal part of the squamosal and forming the posterolateral corner of the skull. The exact shape of the supratemporal is not well defined on the silicone cast because a part of the silicone became stuck in the nodule in the course of freeing the mold from the specimen. Observation of the original print after the removal of the residual silicone suggests that the supratemporal terminates in a pointed corner, with an angle slightly lower than 90°. The squamosal and quadrate are visible in posterolateral occipital view. The squamosal presents, on its anterodorsal side, a thick tympanic flange. The temporal notch, where the tympanic membrane was attached, is represented by a shallow posterolateral depression. The quadrate is covered posteriorly by the vertical occipital flange of the squamosal. Fine parallel crests extend mediodorsally from the middle of this region toward the supratemporal–a structure unique so far in procolophonids (Fig. 2A–B). The squamosal is broken anteriorly, at the contact with the quadratojugal. The relatively high quadrate is massive and trapezoidal. The two condyles are well developed. The medial one remains in contact with the articular. The lateral condyle consists of an anteroposterior ridge located slightly dorsal to the level of the medial one. Although the quadrate is poorly preserved medially, the contact with the respective quadrate process of the pterygoid is discernible.

The palate, visible in ventral view, is articulated except for the left pterygoid. The anterior process of the pterygoid bears denticles (Dentition) along the anterolaterally directed ridge bordering the interpterygoid cavity. The basipterygoid recess of the pterygoid is short and ends in a concave surface for the basipterygoid process of the parabasisphenoid. The quadrate process forms an arcuate blade sutured extensively to the quadrate. A low, straight ridge runs anterolaterally from the centre of the pterygoid. This ridge is edentulous, in contrast with ‘Owenetta’ kitchingorum (Reisz and Scott, 2002: fig. 2B) and P. trigoniceps (Carroll and Lindsay, 1985: fig. 1B).

The right ramus, missing its anterior tip, is visible in ventral and lateral views. The massive dentary deepens steadily posteriorly and shows a smooth lateral surface except for a labial foramen, as in Coletta seca (Modesto et al., 2002). The blade-like splenial occupies the medial surface of the jaw, covering the dentary medially, and reaching the angular posteriorly. The subtriangular coronoid forms, in lateral view, the eponymous prominence and is sutured ventrally to the surangular and dentary. The subtriangular surangular contributes anteriorly to the coronoid prominence. The surface of the angular and surangular, in lateral view, appears rough but only reflects tiny bubbles in the silicone cast. The articular is preserved close to the quadrate condyles. The articular region is relatively straight in lateral view, in contrast with leptopleurines in which the whole region is downturned (Li, 1989 and Sues et al., 2000). The robust, subtriangular angular extends up to the level of the supralabial foramen mentioned above.

Four teeth are preserved on the maxilla. An empty space, distal to the fourth tooth, suggests an additional tooth. In apical view, the teeth are connected by subparallel ridges running along the dental row (Fig. 3B), a condition that suggests a strong tooth attachment or even tooth ankylosis (Discussion). Although this morphology might have been artifactual, a closer look at the mould confirms the presence of such ridges. The kind of tooth implantation is hard to identify (Discussion). The teeth are separated from each other at their bases by spaces of about the same length as the teeth, but there is no evidence for tooth replacement.

The first three mesial teeth are subconical, and this aspect becomes more pronounced distally. These teeth have a single cusp, a condition contrasting with the fourth, bicuspid tooth. In this molariform tooth, the distolabial cusp is higher and thinner than the mesiolingual one. In occlusal view, both cusps are connected by a sigmoid labiolingual crest known in some other procolophonids despite differences in the degree of curvature and sharpness (Carroll and Lindsay, 1985, fig. 14; Cisneros, 2008b). It is also broader, being as high as wide, and more bulbous than the mesial teeth. At least two tooth prints are observed on the dentary, directly on the nodule, but they are difficult to interpret. This area is unfortunately obscured by the maxilla on the silicone cast. Six large and at least three small denticles are respectively present on the anterior process of the left and right pterygoids. These denticles are scarce and widely spaced, a possible result of the poor preservation of the pterygoid area. Furthermore, both pterygoids are covered by a shagreen of denticles, close to the supposed symphyseal region, as in Owenetta rubidgei (Reisz and Scott, 2002).

The copula or corpus hyoideum is a thin, flat, and butterfly-shaped bone, which is wider than long. It shows two (anterior and posterior) paired processes of similar length, as in O. rubidgei and ‘Owenetta’ kitchingorum (Reisz and Scott, 2002) but not in P. trigoniceps (Carroll and Lindsay, 1985), which has a longer posterior process. The deep concavity of the posterior margin reaches a third of the length of the bone, resulting in a U-shaped arch. Similarly, the anterior margin is V-shaped in L. beltanae (Fig. 2A–B), whereas that of P. trigoniceps (Carroll and Lindsay, 1985: figs. 9A, 13C), O. rubidgei (Reisz and Scott, 2002: figs. 4D, 5D), and ‘Owenetta’ kitchingorum (Reisz and Scott, 2002: fig. 2B) is only gently concave. It differs even further from Sauropareion, in which the copula is bowed anteriorly (MacDougall et al., in press and Modesto and Damiani, 2007). The first ceratobranchial pair is preserved. The right one has been shifted between the pterygoids, while the left one is still connected with the copula and extends posterolaterally, overlying a part of the anterior cervical region. These ceratobranchials are slender, rod-like elements with a ventrally bowed shaft and terminating distally in a flat, spatula-shaped end.

The preserved presacral series is articulated, for a minimal vertebral count of 20 (Fig. 1C). Because the individual described here is exposed in ventral view, three to four vertebrae are obscured by the pectoral girdle. A minimal presacral count of 23 would suggest that few presacral vertebrae are missing. There are indeed 25 to 27 presacral vertebrae in procolophonoids (Broili and Schröder, 1936, MacDougall et al., in press, Meckert, 1995, Reisz and Scott, 2002 and Säilä, 2010a). For convenience, the vertebral series preserved anteriorly and posteriorly to the scapular girdle of L. beltanae will be referred to here as the cervical and dorsal series, respectively. Six cervical vertebrae are visible in lateral view, except the missing atlas and the penultimate cervical, which is also preserved in anterior view. The third cervical is partially covered by the left ceratobranchial. The axis shows a massive neural arch with a typical hatchet-like neural spine. Its corresponding pleurocentrum could not be identified. The neural spine and the pedicel of the neural arch are relatively high. Other cervicals have shorter and more slender neural spines. The neural arches and pleurocentra have a similar height, but the latter are about twice as wide as long, resulting in a compressed aspect. The fourth cervical pleurocentrum shows a concave anterior surface suggestive of the plesiomorphic amphicoelous condition seen in procolophonids (deBraga, 2003). The pleurocentra also have a pronounced ‘lip’ on their anterior and posterior articular surfaces, which are connected ventrally by a strong ridge. This configuration results in a deep lateroventral excavation on the pleurocentrum. Short diapophyses are also visible anteriorly, at the junction between the neural arch and pleurocentrum. There is no trace of parapophyses, which are present on the intercentra in other procolophonoids (deBraga, 2003 and Reisz and Scott, 2002).

Fourteen dorsal vertebrae are preserved in ventral view. As in the cervical series, the dorsal pleurocentra are wider than long and bear a low ventral ridge and two lateral shallow excavations. The ventral longitudinal sulcus seen in P. trigoniceps (deBraga, 2003) is absent. Intercentra are short, crescent-shaped, and ventrally arched. They are well developed, with a width sub-equal to that of the corresponding pleurocentra, and well ossified. This is also the case in P. trigoniceps (deBraga, 2003), Nyctiphruretus acudens (Efremov, 1940) and B. besairiei (Meckert, 1995). Conversely, vacant spaces between two vertebrae suggest cartilaginous intercentra in Hypsognathus fenneri (Colbert, 1946: pl. 27, figs. 5–6) and Soturnia caliodon (Cisneros and Schultz, 2003: fig. 4A–C).

Only the presacral ribs are preserved. Two long, straight cervical ribs are located to the right of the pleurocentra of the fourth and fifth preserved vertebrae. The first one is apparently holocephalic, but the second one bears two slender articular heads separated by a depression. This condition differs from the dichocephalic one of P. trigoniceps (deBraga, 2003) which has two depressions. Sixteen pairs of dorsal ribs are preserved. They are long, slender, and bowed, forming a wide rib cage. The left rib series has been slightly displaced compared to the right series, and the first ribs are broken. The ribs are flared proximally and terminate in a flat head, where the capitulum is indistinguishable from the tuberculum. The distal extremities of the ribs are also mediolaterally dilated, ending in a flat tip. From the thirteenth pair onward, they become progressively narrower and straighter, though this apparent straightening seems to be an artefact of compression.

Gastralia are numerous subparallel, bowed, needle-shaped elements, visible on the right side of the rib cage (Fig. 1B–C). They are thinner, shorter and slightly more curved than those of P. trigoniceps (deBraga, 2003) and SAM-PK-K7711 [a specimen assigned to P. trigoniceps by deBraga (2003), but considered later as an indeterminate procolophonid by Modesto and Damiani (2007), and as a possible Teratophon spinigenis by Cisneros (2008c)].

The pectoral girdle is articulated and nearly complete, missing only the left clavicle and cleithrum. The clavicle, still in its natural position, is a slender, crescentic element ending medially in a pointed head. The shape of the dorsal process is poorly preserved. The junction of the head and branch forms an angle of about 120°. A splint-like bone, interpreted here as the cleithrum, is preserved close to its natural position on the right scapula. The cleithrum has pointed extremities and its anterior half is wider and flatter than its posterior one. This bone is rarely preserved in procolophonoids (MacDougall et al., in press, Modesto and Damiani, 2007 and Reisz and Scott, 2002). The interclavicle has a typical ankyramorphan T-shape. The left lateral process of the interclavicle is missing its tip, but the right one is complete. The lateral processes are straight and the tip is recurved posteriorly, ending at the junction between the scapula and the anterior coracoid, as in P. trigoniceps (Colbert and Kitching, 1975: figs. 16–17). Both lateral extremities are connected by a high ridge along the anterior border of the interclavicle. A second ridge runs posteriorly along the median process. This ridge becomes gradually lower until becoming indistinguishable from the remainder of the process, at about two-thirds of the interclavicle length. Both the lateral and median ridges are approximately perpendicular and meet anteriorly on the interclavicle. The posterolateral margin of the interclavicular head is slightly concave as in P. trigoniceps and SAM-PK-K7711 (deBraga, 2003), whereas it forms a right angle in Pentaedrusaurus ordosianus (Cisneros, 2008a and Li, 1989). The anterior coracoids are flat and subcircular. A medial ridge separating the coracoid in two sub-equal parts runs along its dorsal face. A deep depression at the base of this coracoid corresponds to the opening of the coracoid foramen. The posterior coracoids, thicker and longer than the anterior ones, are flat and subcircular too. They also present a large glenoid articular surface dorsally. The scapulae are subrectangular blades, stretched dorsolaterally. Laterally, the scapular blade presents two shallow depressions separated by a low anteroposterior ridge at mid-length. The anterior scapular margin is thin in contrast to the thick, rounded posterior margin. The latter bears a subtriangular ventral supraglenoid buttress.

The forelimbs are well preserved and their bones are articulated. The humerus is very robust, with broad, flat heads that are twisted at about 90° from each other, in contrast with SAM-PK-K7711 (deBraga, 2003), in which it is no more than 45°. In L. beltanae, the deltopectoral crest is restricted to the proximal head of the humerus in lateral view. Both humeral heads are about the same size. The entepicondyle, more prominent than the ectepicondyle, forms an angle of about 45° with the diaphyseal axis. The entepicondylar foramen, visible on both humeri, is large and elongated. The presence of an ectepicondylar foramen or groove cannot be assessed. The distal head of the humerus shows the two small lenticular surfaces for the articulation with the zeugopod.

The left ulna of L. beltanae is only preserved proximally. This bone is larger than the radius and has a distinct triangular proximal head with a smooth convex articular surface. The radius is a slender element showing a flat, smooth triangular articular surface for the humerus. The distal part of both radii is missing (Fig. 1C).

The autopodium is incomplete, consisting here of seven carpals, two metacarpals and ten phalanges (Fig. 4). The carpals are more or less lined up in a proximal, a medial, and a distal row. The proximal row comprises three different carpals. The lateral one, the most massive, has a pentagonal shape, with blunt edges, a large central depression, and a semicircular medial notch. The central carpal, much smaller and oval, presents a short, shallow groove in its proximal part. The medial carpal is rectangular and exhibits many rugosities. These elements are tentatively identified as the ulnare (the semicircular notch would thus border the manual perforating foramen), the intermedium, and the radiale, respectively. The medial row includes two carpals. The carpal showing a medially oriented tip and a large foramen may represent the lateral centrale. Laterally, a larger, pentagonal or round carpal with two small foramina is possibly the weakly ossified medial centrale. The distal row comprises two carpals, which may be the distal carpals 2 and 3 because they are very close to the metacarpals.

The metacarpals are recognizable by their typical hourglass shape. The medial one presents on its medial side the typical lateral bowing of the metacarpal I seen in P. trigoniceps, in SAM-PK-K7711 (deBraga, 2003: figs. 3–5), and in B. besairiei (Meckert, 1995: figs. 24–26). Whereas its distal end is almost perpendicular to the long axis, its proximal end is deflected at 45° medially. Metacarpal II is as wide as metacarpal I, but longer. Its proximal extremity is less deflected but more developed than in the latter, however. It shows two articular surfaces, the medial one being particularly prominent. The distal end of metacarpal II bears two small and symmetrical condyles.

Four partial digits are preserved in articulation. The phalangeal formula is 2–3– > 2– > 2–?. Given the relative size, position, and articulation of the phalanges, digits III and IV may have had four phalanges each. The two elements proximal to ungual I belong to the same proximal phalanx I, artifactually separated by a break. Also, digit I overlaps digit III proximally. The print of other phalanges is possibly within the nodule. The phalanges are relatively robust, with bulbous extremities showing two distinct surfaces each. The ungual phalanges are recurved and bear a deep median lateral groove.

Clustered rounded pits are visible all over the rib cage, especially on the right side (Fig. 1B–C). They are considered either as sedimentary micronodules or as possible geogastroliths (sensu Wings, 2007) that would have been dissolved and/or removed by weathering. Though geogastroliths are so far unknown in procolophonids, quartz pebbles (5 mm > 1 cm) are frequently found in the left ventral region of specimens of Barasaurus (Smith, 2000; J.F., pers. obs. of MNHN.F.MAP 88, 161, 191).

The immaturity of MNHN.F.MAE 3039 at the time of death is revealed by the weak ossification and co-ossification of its postcrania. The former phenomenon is illustrated by: (1) the nearly circular shape of the coracoids; (2) the convex but smooth and weakly developed articular surfaces of limb elements (humerus, ulna, radius, carpus); and (3) the absence of a distinct olecranon. The lack of co-ossification is also notable, given that: (4) the axial neural arch is free; (5) post-axial cervical vertebrae show an open neuropleurocentral suture–though they may be actually free because the area was left undisturbed after death; (6) the scapula and coracoids are free. In fact, this is the case in all procolophonids in which this region is preserved, even in mature specimens (Colbert and Kitching, 1975, deBraga, 2003, Li, 1989, Säilä, 2010a and Sues and Reisz, 2008), suggesting that the pectoral girdle elements did not co-ossify during ontogeny in these reptiles.

Until recently, Barasaurus was represented by dozens of nodules collected from the Lower Sakamena Formation only, in the Morondava Basin (Meckert, 1995, Piveteau, 1955 and Smith, 2000). The description of additional material showed that this genus is also present in the Middle Sakamena Formation in the Diego Basin (Ketchum and Barrett, 2004). Thorough comparisons between MNHN.F.MAE 3039 and Barasaurus are therefore necessary to justify their separation. The following comparisons are based on specimens assigned to B. besairiei (casts of MNHN.F.MAP 88, 166, 186, 191, and an unnumbered MNHN specimen, the last two being respectively cited as ‘P 1’ and ‘P 6’ in Meckert, 1995).

In this case, the dentition is particularly informative. The most conspicuous feature is the presence of a bicuspid tooth in MNHN.F.MAE 3039, distally. Among procolophonians, bicuspidy is so far only known in procolophonids. The low, blunt conical shape of the more mesial teeth is also distinctive. The dentition of Barasaurus consists of slender conical teeth with a pointed apex. These teeth are furthermore tightly packed on the tooth row, where they are inserted in a distinct groove (MNHN.F.MAP, unnumbered). There is also an empty alveolus for each missing tooth. This is not the case in MNHN.F.MAE 3039. Its teeth are widely spaced, but there is neither a groove nor an indication of tooth loss or replacement explaining this pattern. The tooth row surface is covered instead by mesiodistal ridges that are connected to the teeth. These ridges might have strengthened the attachment of the teeth to the maxilla. They may even be related to tooth ankylosis, though this interpretation must remain speculative given the absence of actual bone and teeth. Interestingly, both MNHN.F.MAP (unnumbered) and MNHN.F.MAE 3039 bear the high lingual walls typical of acrodont implantation, although the presence of a groove and alveoli in the former suggest rather a protothecodont condition. This apparent inconsistency may be the first stage of the setting of a ‘pseudo-acrodont’ implantation–described in detail for the procolophonids Soturnia by Cabreira and Cisneros (2009) and Sauropareion by MacDougall and Modesto (2011)–characterized by the combined presence of high lingual walls, protothecodont implantation, tooth replacement (albeit very slow), and tooth ankylosis (MacDougall and Modesto, 2011). This configuration is likely that displayed by MNHN.F.MAE 3039 (except for the speculative ankylosis). However, because the mode of tooth replacement and implantation of procolophonids remain highly disputed (Cabreira and Cisneros, 2009, MacDougall and Modesto, 2011 and Säilä, 2009), it will not be discussed further.

The maxillary tooth count is also helpful to distinguish Lasasaurus from Barasaurus. That of the latter can be estimated using MNHN.F.MAP (unnumbered), on which the maxillae are adequately exposed. This specimen bears 12 teeth on the left maxilla and 13 on the right one, with respectively nine and seven empty alveoli. This suggests a total tooth count of 20–21, close to the reconstructed count of 19–20 of Meckert (1995) and to that of 20 for ‘Owenetta’ kitchingorum, (J.F., pers. obs. of BP/1/4195a). Such a count is much higher than in MNHN.F.MAE 3039 with its four teeth, considering the preserved length of the maxilla in the latter. Though its actual tooth count cannot be assessed, this likely reflects the trend to reduction of tooth count seen in procolophonids.

Other features supporting the separation of Lasasaurus from Barasaurus include the right angle (versus acute) temporal emargination, the presence (versus the absence) of dendritic crests on the squamosal, the presence (versus the absence) of an entepicondylar foramen, and the prominent (versus smooth) interclavicular medial ridge. These characters also justify the assignment of Lasasaurus to Procolophonidae.

A phylogenetic analysis was performed to determine the position of L. beltanae within Procolophonidae and to test the monophyly of the group. The data matrix was taken from Cisneros (2008a) and modified following Cisneros (2008b), Modesto et al. (2010), MacDougall and Modesto (2011), and MacDougall et al. (in press). The characters are numbered here from 1 to 59. The new Malagasy procolophonid was added to the data matrix in order to test its phylogenetic position. In a recent redescription, N. acudens was shown having an acute temporal emargination (Säilä, 2010b). This taxon was therefore rescored accordingly as ‘1’ for character 13. New material referred to Sauropareion (MacDougall et al., in press) indicates that Coletta should be scored as ‘2’ for the same character (Description). Twenty-six taxa and 59 characters (42 cranial and 17 postcranial) were considered (Appendix A for data matrix and Appendix B for character list). Following Cisneros (2008a), the parareptiles Nyctiphruretus and Owenettidae were selected as outgroups. Owenettidae were subdivided into ‘Owenetta’ kitchingorum (Reisz and Scott, 2002; J.F. pers. obs. of BP/1/4195a and b) and B. besairiei (Meckert, 1995; J.F., pers. obs. of MNHN.F.MAP 88, 161, 186, 191, and unnumbered). The scoring is consistent with that of Owenettidae except for three characters: character 23 was rescored as ‘1’ for both taxa (Meckert, 1995 and Reisz and Scott, 2002), character 38 as ‘1’ for Barasaurus because it has no pterygoid dentition (Meckert, 1995), and character 56 as ‘1’ for Barasaurus according to the femur/humerus length ratio given by Meckert (1995) (p. 75). The analysis was performed using the heuristic search of the software PAUP 4.0 beta10 (Swofford, 2002) and TNT 1.1 (Goloboff et al., 2008). Characters were considered unordered and equally weighted, and polymorphism was treated using the “uncertainty” option. Branches with a maximal length of zero were collapsed using Collapsing Rule 3.

The first analysis resulted in 255 equally optimal trees of 133 steps each (CI = 0.617; RI = 0.787) of which the strict consensus is 181 steps long (CI = 0.453; RI = 0.586). The topology of the latter (Fig. 5) is identical to that found by MacDougall and Modesto (2011) and MacDougall et al. (in press), except, of course, for the presence of Lasasaurus. Coletta and Sauropareion are successive sister taxa to clade #C, which contains the remaining procolophonids. This clade consists of a huge polytomy formed by clades #D, #E (=Procolophoninae), and #G (=Leptopleurinae) and 11 other terminal taxa including Lasasaurus. This lack of resolution is not only caused by the amount of missing data, but also by the erratic behaviour of some of the taxa included in the analysis. Using the comparative pruning option of TNT, the most erratic taxa were identified as Phonodus (three nodes gained) and Kitchingnathus (two nodes gained).

A second analysis was therefore performed without Phonodus and Kitchingnathus. It generated 15 optimal trees of 126 steps each (CI = 0.651; RI = 0.787) for a strict consensus of 137 steps (CI = 0.599; RI = 0.770). Procolophonid interrelationships were better resolved than in the first analysis. About half of the clades have a very low Bremer support (1), however. It is a bit higher in clades #c, #d, #f, #n, and #o (2) and much stronger in clades #a (=Procolophonidae) and #o (4). The strict consensus (Fig. 5) shows a monophyletic Procolophonidae as in previous analyses (Cisneros, 2008a, Modesto and Damiani, 2007, Modesto et al., 2002, Piñeiro et al., 2004 and Säilä, 2008). Unequivocal synapomorphies of procolophonids include the presence of roughly circular or dorsoventrally expanded external nares (3[1*]) bordered posteroventrally by a depression of the maxilla (6[1*]), a broadly excavated temporal ventral margin (13[2*]), four premaxillary teeth (26[1*]), and vomerine teeth anteriorly (35[1*]) but not along the posteromedial suture (36[1]) (changes followed by a star are non-homoplastic). Lasasaurus belongs to clade #e, where it forms a polytomy with Tichvinskia, Timanophon, #f (=Theledectinae), #g, #h (=Procolophoninae), and #j (=Leptopleurinae). Clade #e is supported univocally by the presence of external nares which are anterior to the first premaxillary tooth (2[1*]), a wide internarial bar (4[1*]), an orbitotemporal fenestra of which the posterior margin is beyond the posterior border of the pineal foramen (8[2*]), enlarged mesialmost premaxillary teeth (27[1*]), both mono- and bicuspid maxillary teeth (31[1*]), eight to six maxillary teeth (32[3*]), an inset of maxillary cheek teeth (33[1*]), and two widely separated cusps on dentary molariforms (41[2*]). Among these characters, Lasasaurus could only be scored for characters 31 and 33. Its inclusion in clade #e is founded on the presence of mono- and bicuspid maxillary teeth (31[1]) despite the fact that they are not inset from the lateral surface of the bone (33[0]) as are those of most members of the clade. The latter feature, interpreted here as a reversion, is shared with Theledectinae. It might represent a mere convergence, a potential synapomorphy uniting Lasasaurus to this clade, or, more likely, a symplesiomorphy. In the last two cases, it would mean that Lasasaurus is not as close as Tichvinskia is to Procolophoninae and Leptopleurinae.

In previous analyses of procolophonid interrelationships, Tichvinskia was found as either a Leptopleurinae (deBraga, 2003 and Modesto and Damiani, 2007) or the sister taxon of the clade (Procolophoninae + Leptopleurinae) (Cisneros, 2008a, Cisneros, 2008b, Modesto et al., 2001, Modesto et al., 2002, Modesto et al., 2010, Piñeiro et al., 2004 and Säilä, 2008). In addition, Theledectinae has been suggested as the sister taxon to the clade (Tichvinskia (Procolophoninae, Leptopleurinae)) (Cisneros, 2008a, Cisneros, 2008b and Modesto et al., 2010). The discovery of new material referable to Sauropareion sheds light on the relationships of this genus with Coletta and the clade including other procolophonids, but it results in the collapse of many branches in the latter. The phylogenetic resolution could be improved by removing Phonodus and Kitchingnathus, but the interrelationships of Theledectinae, Procolophoninae, Leptopleurinae, Tichvinskia, and, of course, Lasasaurus remain uncertain. This lack of resolution within clade #e is unfortunate because the interrelationships of its members are crucial to elucidate aspects of the evolution of procolophonids, such as their ecological adaptation, the timing of their diversification, and their biogeographical range.

Procolophonoids are the only parareptiles that survived the PT mass extinction (Cisneros, 2008a and Modesto et al., 2001). The Early Triassic radiation of procolophonids is likely the result of diverse dietary specializations related to deep modifications of the jaw apparatus and teeth (Cisneros, 2008a). The ecological success of these reptiles appears linked to their spread across Pangaea–including Madagascar–as early as the Early Triassic. Although in some areas procolophonids are the most common reptiles (Nicolas and Rubidge, 2010 and Thulborn, 1986), MNHN.F.MAE 3039 is the first procolophonid identified from the Middle Sakamena Formation, despite intensive collection of fossil vertebrates. Finally, a comprehensive palaeoclimatic study has shown promising results in investigating the geographical distribution and relative abundance of leptopleurines and traversodont therapsids in eastern North American rift basins during the Late Triassic (Whiteside et al., 2011). It showed that their respective ecological preferences have a major bearing on their biogeography. Such a study is needed for the Early Triassic, but, meanwhile, it is hard to interpret the occurrence of procolophonids in Madagascar at that time. In fact, their scarcity in the Middle Sakamena Formation can be hypothesized as the mere result of unfavourable palaeoenvironmental settings.

MNHN.F.MAE 3039 provides palaeoecological and taphonomical evidence to help reconstruct the depositional environment. This evidence includes several suboval structures bearing concentric striations, located right to the neck region of the holotype of L. beltanae. These remains are interpreted as carapaces of euestheriid conchostracans, identified as Palaeolimnadia sp. and Euestheria sp. or Magniestheria sp. by Heinz Kozur (pers. comm. to M.A., 2011). Their presence here may be explained by scavenging on L. beltanae, a frequent behaviour in these crustaceans (Martin and Christiansen, 1995). Euestheriids are abundant in the Middle Sakamena Formation of Madagascar (Besairie, 1972). They are represented by M. truempyi, which is considered euryhaline, living in brackish, estuarine, deltaic, or coastal palaeoenvironments (Shen et al., 2002). However, because most extant species live in temporary freshwater ponds, they could be also considered freshwater indicators (Vannier et al., 2003). The conchostracans preserved on MNHN.F.MAE 3039 are also associated with a wide range of facies (from freshwater ponds, shallow lakes, or floodplains to brackish estuarine deposits), but are at least indicative of a low energy depositional environment (Heinz Kozur, pers. comm. to. M.A., 2011). Here, in the Middle Sakamena Formation, euestheriids are typically associated with marine interbedding, suggesting that the palaeoenvironment was probably deltaic and under tidal influence. These conclusions are reinforced by the co-occurrence of (possibly) terrestrial, freshwater, brackish, and especially marine indicators (Maganuco et al., 2009 for a review).

Lasasaurus shows no osteological adaptation to an aquatic or even amphibious mode of life, in contrast to the temnospondyls (Lehman, 1966, Maganuco et al., 2009, Steyer, 2002 and Steyer, 2003) and diapsids (Ketchum and Barrett, 2004) found in the same formation. Yet, one may argue that the presence of (possible) geogastroliths in Lasasaurus may have been used to regulate buoyancy. This is the case, for instance, in the supposedly amphibious diapsid Hovasaurus boulei (Currie, 1981, Piveteau, 1926 and Smith, 2000) from the Lower and Middle Sakamena formations (Ketchum and Barrett, 2004). This species displays various morphological adaptations to swimming and would have used geogastroliths as ballast (Currie, 1981 and Smith, 2000). Geogastroliths are also common in Barasaurus (Smith, 2000; J.F. pers. obs. of MNHN.F.MAP 88, 161, 191), which shows no osteological evidence of aquatic adaptation. Surprisingly, lithological and taphonomical data suggest otherwise. According to Smith (2000), Barasaurus is indeed the dominant component of a fossil assemblage (comprising also actinopterygians and plant remains) that was deposited in a distal lacustrine environment. He also noted that, if the geogastroliths of Barasaurus and Hovasaurus had been used as a gastric mill, plants fragments should be expected in its pebble mass. Yet, despite the preservation of macroplant fossils next to several skeletons, Smith (2000) observed no plant remains in the pebble mass of any of the Barasaurus and Hovasaurus-bearing nodules he inspected, hence concluding that the concentration of geogastroliths should be rather interpreted as having a ballast function in these reptiles. From these data, an amphibious to mostly aquatic mode of life is inferred for Barasaurus. In contrast, procolophonids are considered as terrestrial. Their taphonomy and the various anatomical adaptations they exhibit even suggest that they were burrowers (MacDougall et al. (in press) for a discussion). There is currently no evidence for the need of buoyancy regulation in these reptiles. These geogastroliths would have been more likely used by L. beltanae to crush hard prey or to grind hard-fibre plants to help digestion, a hypothesis that is consistent with the supposed specialization toward a durophagous diet in procolophonids. Finally, the limited disarticulation of its remains indicates that it was deposited and buried shortly after death. This implies brief transport and the proximity of the terrestrial source, consistent with the supposed depositional environment.

The deltaic to coastal lagoonal environmental setting of the Diego Basin was likely the main influence on the preserved composition of the Madagascar tetrapod fauna during the Early Triassic. This depositional environment contrasts strongly with that of the contemporaneous strata of the Beaufort Group, in the main Karoo Basin, that were deposited in wide floodplains associated with braided and meandering river systems (Catuneanu et al., 1998). Additionally, the Diego Basin was closer to the Equator and had therefore a much warmer and a bit wetter climate than the main Karoo Basin (Péron et al., 2005). Although Madagascar and Africa were in contact at that time, it is therefore not surprising that the Early Triassic Beaufort Group and Middle Sakamena Formation tetrapod faunas are very different taxonomically and ecologically (Battail et al., 1987). Besides, the association of relictual Permian tetrapods (Ketchum and Barrett, 2004) alongside typical Triassic forms (Maganuco et al., 2009, Steyer, 2002 and Steyer, 2003; this work) in the Middle Sakamena Formation suggests that Madagascar acted as a refuge zone during the PT extinction events.