Construction of the data matrix.

Previous phylogenies of the genus Bothriolepis were done using selected synapomorphies to denote nodes, rather than composed from computer algorithms (Fig 12; see further). A new taxon*character data matrix was constructed in Mesquite v. 3.61 [22], based on the data published in [10]. The initial set of 20 characters was slightly modified and expanded to 59 characters (see Supplementary Information), and the original range of 12 Euramerican Bothriolepis species (+ Grossilepis as outgroup) was extended to 54 species, collectively of global distribution. Depending on their completeness or conflicts, some taxa or characters were removed in order to obtain a final data matrix composed of 44 taxa * 50 characters; 43 characters could be coded effectively for Bothriolepis dairbhrensis sp. nov. from the Valentia Slate Formation (completeness 73%; see S3 Table). Minor modifications in the coding and the labelling of characters and their states are included in the matrix file and in Supplementary Information). All characters are unordered, unpolarised, and to be of equal weight a priori. More characters would be necessary to resolve the relationships between the different coded species of Bothriolepis; we must hence keep in mind that polytomies and lower consistency indices (CI) are to be expected (addition of taxa decreases the CI more than the addition of characters in random generated matrices; [23]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 12.

Historical overview of the relationships within the genus Bothriolepis according to: A. Janvier and Pan ([24]:fig. 12); B. Long ([19]:text-fig. 13); C. Young ([25]:text-fig. 68B); D. Johanson and Young ([26]:fig. 10); and E. Lukševičs ([10]:fig. 83). Colours indicate the shape of the preorbital recess for species in which it is known (see Fig 13).

https://doi.org/10.1371/journal.pone.0280208.g012

All characters were re-arranged according to a morphological logical sorting in the data matrix. Similarly, all taxa were re-arranged by convenience using the following iterative rules: 1) their outgroup vs. ingroup status, 2) their palaeogeographic origin (Gondwana, China, then Euramerica), and 3) alphabetical order. It should be noted that this sorting is only intended for user convenience, and has no impact on the results (it should be however kept in mind that TNT always uses the first taxon as outgroup).

Once constructed, a survey of the matrix could reveal the completeness of the taxa and of the characters (i.e. the percentage of actually coded information vs. “missing data”; see S3 Table). The following characters and taxa were removed from later phylogenetic analyses because their completeness was lower than 25%:

• Character 5 (0%): ratio length anterior tip / width between lateral corners of the skull roof; this character was difficult to assess regarding the 3D vs. flattened preservation and available reconstructions in literature; besides, the range of values was quite consistent throughout all species, with ratios ranging between 0.6 < L/W <0.9 (see Supplementary Information).
• Character 7 (24%): bone thickness on orbital edges of premedian (PRM) and lateral plates (in [25]:text-fig. 68, character 27o; in [10]:char 8).
• Character 18 (24%): separate ventrally facing attachment surface for prelateral (PrL) on lateral plate (in [25]:text-fig. 68, character 27m).
• Character 25 (5%): shape of nuchal plate #3 (short or long posterolateral corners; in [25]:text-fig. 68, character 27p; split into three characters in the present analysis).
• Character 34 (15%): anterior submarginal (SM) plate attachment on lateral plate as a transverse ridge (in [25]:text-fig. 68, character 27j).
• Character 35 (7%): anterior portion of submarginal attachment covers spiracular groove (sp.g; in [26]:fig. 10, character 9).
• Character 57 (20%): pectoral pit-line traced on the central ventral plate 1 (Cv1) continuing on the central ventral plate 2 (Cv2; in [10]:character 5).
• Character 58 (2%): dermal ornamentation (in [25]:text-fig. 68, character 27a).
• Character 59 (0%): general size and ornamentation (in [25]:text-fig. 68, character 27q).
• Bothriolepis obesa (0%)
• B. paradoxa (7%)
• B. stevensoni (2%)
• B. coloradensis (17%)
• B. darbiensis (24%)
• B. nielseni (19%)
• B. alexi (5%)

A similar survey revealed the degree of polymorphism coded for each taxon and character; these values can also be seen in the S3 Table. 44 taxa include polymorphic coding (81%); 34 characters include polymorphic coding (58%).

The following taxa are removed from the final analysis because they are considered as junior synonyms of B. nitida (see discussion between [27–29]: B. virginiensis (63%), B. coloradensis (17%), B. nielseni (19%).

Bothriolepis sinensis, although initially considered, proved to be too poorly preserved and was removed at an early stage of the study (see [30]).

Ornamentation.

The ornamentation consists of rounded tubercles that may join up to form coalescent thick ridges in the larger individuals (e.g. ventral lamina of the posterior ventrolateral plate TCD19618-FB4033, Fig 8D). In many specimens the external ornament is worn or poorly expressed due to weathering and preservation (e.g. Fig 2A, 2C and 2D). The smaller specimens, and hence considered younger individuals, display a more nodose or reticulated ornament. This change of ornamentation through ontogeny is common in the genus Bothriolepis [9, 43–46].

Neurocranium.

The shape of the neurocranium in antiarchs is only known from impressions on the visceral surface of the head shield. A portion of the outline of the dorsal surface of the neurocranium is accessible from only one specimen (NHMP59677, Fig 4). It can be restored in its anterior half only with certainty anteriorly and laterally from the orbital fenestra, where it is overlaid by the premedian and lateral plates, and with a lesser degree of confidence in its posterolateral portion.

As described above/below, the infilling of the preorbital recess reveals unambiguously a trifid preorbital recess (“groenlandica type” of [20, 25, 43]). The median branch of the preorbital recess is triangular, flat, and is located below the posterior half of the premedian plate. The two lateral branches of the preorbital process are more pointed than the median one; they are flat at their tip but thicker posteromesially at the level of the orbital fenestra. The axis of the right lateral branch of the preorbital recess is directed towards the internal pit (p, Fig 4B and 4C). The angle between the axes of the median and the left branches is 65°, between the axes of the median and right branches is 69°. These measured values are considered accurate enough, considering the very different deformation for either lateral plate (inrolled for the left one, flattened for the right one) and very limited deformation for the premedian plate.

The depression for the postorbital process (pop, Fig 4B and 4D) is located under the pocket containing the right lateral branch of the preorbital recess. It is very slender and its tip extends anteriorly as far as the lateral branch of the preorbital recess, and thus more anteriorly than the orbital fenestra. Its axis is directed toward the internal pit.

The infilling of the otico-occipital depression (ood, Fig 4D) of the inner side of the right paranuchal plate reveals a square posterolateral corner. It is bounded laterally by the paramarginal crista (pm.cr, Fig 4C and 4D).

Skull roof.

The skull or skull elements are preserved in 9 specimens (NMING:F35202, NMING:F35205, NMING:F35206, NMING:F35207, NMING:F35204, NMING:F35201, NHM PV P 59677, NHM PV P 59678, NMING:F35203). It is restored from nearly all elements being preserved.

The premedian plate is trapezoidal, and is larger anteriorly than on its orbital edge (PRM, Figs 2A–2D, 3C and 4). It is widest between the lateral extremities of the sensory line groove. It is best seen on NHM V P 59677 (Fig 4) as it is incomplete and preserved in dorsal view. The ratio (Width max/Width orbital) is 2.0 in NHM V P 59677 (23.62mm*2/23.51mm). The orbital edge is straight in large individuals such as NHM V P 59677, but could be slightly concave in smaller individuals such as NMING:F35201 (in which it is difficult to assess because of the lateral compression of the specimen; Fig 2A–2C). A very shallow pentagonal depression in the midline of the posterior half of the plate corresponds to the underlying preorbital recess. The ethmoid commissure (ethm.c, Fig 2A and 2B) corresponds to a posteriorly directed “pinch” of the infraorbital sensory line groove on the external side, the posterior point of which corresponding to the median anterior apex of the underlying preorbital recess wall. The ventral side of the premedian plate is concave, and clearly reveals the nasal and facial laminae of the preorbital recess in its posterior half (n.l, f.l, prh, Fig 4). The facial lamina is part of the main component of the plate. The left lateral side shows the contact surface with the adjacent lateral plate better than the right side, both on the ornamented part and the nasal and facial laminae. The nasal lamina looks slightly sheared to the left in ventral view compared to the dermal part of the plate. As noted above, the premedian component of the preorbital recess is pentagonal and can be filled digitally in order to reveal its complete shape (see above/further; Fig 4). The preservation of the material and its scan cannot illustrate the presence or not of a rougher area just anterior to the preorbital recess in ventral view.

Only three specimens indicate size and shape of the orbital fenestra, although at different degrees of preservation, and compression orientation (NMING:F35201, NMING:F35202, NHMP59677; orb, Figs 2A–2D and 4). The orbital fenestra appears proportionally larger in the skull roof in younger specimens than older individuals. There is unfortunately no information relative to the size and shape of the suborbital fenestra internally. NMING:F35201 reveals anterolaterally a millimetre difference between the orbital fenestra and the suborbital fenestra (orb, s.orb, Fig 2C).

The rostropineal plate is observed on NMING:F35202 but does not present any obvious nor describable feature (RoPi, Fig 2D).

The postpineal plate is preserved as an incomplete part in specimen NMING:F35201 (PPi, Fig 2A–2C) but its lateral and posterior outline can be seen on NMING:F35202 (Fig 2D). It is slightly concave anteriorly and more strongly convex posteriorly.

The external surface of the lateral plate harbours the main and lateral portions of the infraorbital sensory line groove (ioc1, ioc2, Figs 2–4); a semi-circular groove (cir, Figs 2–4), the concavity of which is oriented anterolaterally, connects mesially to the intersection of the main and lateral branches of the infraorbital sensory line groove. A weakly defined notch on the anterolateral corner of the plate, at the tip of the lateral extension of the infraorbital sensory line groove, indicates the insertion area of the prelateral plate (n.PrL, Fig 4C). The central sensory line (csc, Figs 2–4) groove extends anterolaterally between the contact with the nuchal plate posteriorly and the semi-circular sensory line groove anteriorly but without reaching it. In the largest individual the central sensory line groove is located closer to the triple point between the lateral, nuchal and paranuchal plates (NHMP 59677, Fig 4), whereas in a smaller individual it is located closer to the orbital fenestra (NMING:F35201; Fig 2A–2C), unless the latter would be related with a compression of the specimen on its lateral side (i.e. median part closer to orbit shortened). These grooves are best visible on the right lateral plate of NHM P 59677 (La, Fig 4). A triangular area lower than the rest of the headshield on the lateral plate of NHMP 59677 indicates the position and shape of the underlying lateral branch of the preorbital recess (Figs 4A and 11A). The internal surface of the lateral plate reveals the flat lateral branches of the preorbital recess wrapped by the facial and nasal laminae. The postorbital process is situated laterally to the lateral side of the nasal lamina. An internal pit of moderate size is visible anteriorly to the tip of the lateral nasal lamina, as well as the oblique spiracular groove bordered by the anterior and posterior crests anterolaterally to the ensemble (sp.g, cr1, cr2, Fig 4C). The lateral edges of the lateral plates of NHMP59677 reveal in ventral view the two attachment areas for the submarginal plate (a1.SM, a2.SM, Fig 4C). The anterior insertion area is short, round and located just posteriorly to the notch for the prelateral plate (n.PrL, Fig 4C); the second one is located a bit posteriorly from the first one, and is elongated anteroposteriorly. Specimen NMING:F35203, although incomplete, shows contact surfaces for the adjacent nuchal and paranuchal plates (c.N, c.PaN, c.PM, Fig 2E and 2F). This specimen thus shows that the postmarginal plate contacts both the lateral and paranuchal plates. The internal side of the same specimen shows a paired attachment area located behind the notch for the prelateral plate, in a pattern similar to that observed in NHM P 596777.

The nuchal plate is observed on NMING:F35201 in lateral view and in NMING:F35202 for the anterior half in dorsal view (Fig 2A–2D). It produces a pair of anterolateral processes avoiding a contact between the lateral and postpineal plates. The posterior edge appears thicker because of the transverse nuchal crista of the internal side, laterally continued on the paranuchal plate. Its shows a well pronounced median process (mp, Fig 2A and 2B). Just anteriorly to the transverse nuchal crista thickening a transverse supraoccipital sensory line groove can be distinguished, also continued on the paranuchal plate (tn.cr, socc, Figs 2A–2C and 3C). An unidentified plate lying just posteriorly to the right lateral plate of NHMP 59677 could be a portion of the nuchal plate. A more complete specimen (NMING:F35205; Fig 3C) reveals a clear concave contact edge for the postpineal plate, a central sensory line, a posterior supraoccipital sensory line, and a thick transverse nuchal crest continued on the right paranuchal plate, although very weathered; the posterior median crest is damaged and not preserved.

The paranuchal plate is visible in lateral view on NMING:F35201 and in dorsal view on NMING:F35202 (Fig 2A–2D). Its posterior edge harbours mesially the continuation of the supraoccipital sensory line groove from the nuchal plate. It is crossed anteroposteriorly by the infraorbital sensory line groove, which shows a short and sharp posterolateral curve in the posterolateral corner of the plate. Two very damaged elements are identified as possible portions of the left and right paranuchal plates on NHMP 59677. The remains of the left would correspond to the lateral and posterolateral edges of the plate showing the dermal articulation area with the anterior dorsolateral plate. The right element is bigger, displaced under the premedian plate, and folded in a similar way that what occurred to the left lateral plate; the element can be identified as a paranuchal plate because of the otico-occipital depression and pocket for the posterolateral edge of the neurocranium (ood, Fig 4D). The supraoccipital thickening is not preserved. When repositioned next to the right lateral plate, the contact edges fit somehow in our model. The most anterior part of this element may belong to another plate, possibly the submarginal plate if those elements were really connected. A very poorly preserved element can be interpreted as a left paranuchal plate and can be observed in NHMP 59677; it was broken in two parts; the lateral edge of the plate remained connected to the lateral plate, while the more mesial part was folded and rotated next to the previous element. Reconstructed, the dorsal side shows a shallow and wide groove for a supraoccipital sensory line. The lateral edge shows a slender overlap area for the anterior dorsolateral plate. Ventrally, a transverse depression can be interpreted as the para-articular fossa (pa.f, Fig 4C), or as a simple break. NHMP 59677 also shows a right paranuchal plate, displaced under the premedian and left lateral plates and folded on itself, with its ventral otico-occipital lamina underlying the otico-occipital process, which is broken and displaced. Once digitally unfolded, it accommodates the lateral and submarginal plates. The posteromesial edge of the mesial lamina shows indentations to fit the nuchal plate.

The postmarginal plate is partly visible in NMING:F35201, still attached to the skull roof, and maybe as a detached plate in NHMP69677 (PM; Fig 2A–2C), but its outline remains uncertain. It seems to be a small quadrangular element (the latter corner is missing in NMING:F35201). In NMING:F35201 it is aligned with the orbital fenestra. In the larger individual NHMP 59677, it overlies the right paranuchal plate but is very compressed and appears triangular; it however contacts the lateral and paranuchal plates, and its ventral side accommodates the posterodorsal edge of the submarginal plate.

Elements of the cheek.

The prelateral plate (PrL, Figs 2A–2C and 5B–5F) is a small tear drop-shaped element, the top of which inserts in the notch of the lateral plate. It is known only in NMING:F35201 and NHM P 59677, in which only the right one shows a shallow groove for the second branch of the infraorbital sensory line.

The infraprelateral plate is known only in NHM P 59677 (IPrL, Fig 5B and 5D). Its external surface shows a very shallow and curved groove for the second branch of the infraorbital sensory line (ioc2, Fig 5B). Its outline is uncertain, but appears more similar to the rounder homologous element in a Bothriolepis sp. from Gogo [25, 42] than to the quadrangular one of B. canadensis [43, 47]. The posterodorsal edge of the plate is straight and accommodates the anterior edge of the submarginal plate. The posteroventral edge of the infrapreorbital plate is sigmoid and presents an overlap area to accommodate the anterior edge of the anterior ventrolateral plate. The anterior edge is straight, and the ventral one is rounded. A faint groove interpreted as the second infraorbital sensory line groove can be distinguished on the external surface of the plate (ioc2, Fig 5B); vertical dorsally, it bends anteriorly in the lower third of the plate. A dorsal process of the infraprelateral plate penetrates into a space between the prelateral and submarginal plate; this pattern is unknown in other members of the genus Bothriolepis, but we have to keep in mind that the cheek elements are seldom preserved.

The submarginal plate is found only in NHM P 569677 (SM, Fig 5A–5D). Although the ornamentation cannot be observed because of the preservation state, its outline, articular surfaces are clear and indicate that it is a right submarginal plate. It is oblong in shape, and slightly higher posteriorly than anteriorly. The dorsal edge of the plate is straight, at the exception of a small and rounded notch between the anterior and posterior articulation areas with the lateral plate, corresponding to the infraspiracular incisure (ad1, ad2, ispi, Fig 5A–5D). The anterior facet ad1 articulates on the lateral plate posteriorly to the spiracular groove, somewhat resembling the condition in B. mcphersoni in which the articulation area a1.SM is located in the spiracular groove rather than anteriorly to it as in most Bothriolepis species. Unlike B. macphersoni, the articulation area is oriented longitudinally along the edge of the lateral plate rather than transversally and more mesially (see [25]:text-fig. 66C). A longitudinal groove can be seen under the posterior lateral articular facet on the anterior part of the external side of the plate, shorter to what is observed in Young ([25]:text.-fig. 55). The posterodorsal edge is slightly curved and accommodates the postmarginal plate of the skull roof. The anterior edge of the plate shows an articular facet to accommodate the infraprelateral plate (a.IPrL, Fig 5A–5C). Mesial to this articular facet lies a groove extending ventrally and posteriorly on the internal side of the plate, and underlying a mesial vault (gr, m.v, Fig 5C and 5D). This vault tapers posteriorly into a low thickening.

Jaw apparatus.

The suborbital plates (SO, Fig 5G–5J) are subquadrangular elements with rounded corners, wider than high. The symphysial edge is much shorter than the lateral one. They are strongly tuberculated, and do not show any sensory line groove. Both left and right plates are preserved in NHMP 59677, although the ventral masticatory blades are not preserved. It is however noteworthy that the ventral edge of the internal side of the left suborbital plate is smooth, and could indicate a very low and abraded masticatory blade, advocating for a very old individual, comforted by the large size of this specimen.

The infragnathal plates are elongated (IG, Fig 5K–5R). The plates are naturally folded so that the masticatory median and articular lateral flanges are not in the same plane (d.IG, art.IG, Fig 5K–5N). The masticatory flange is tuberculated externally; the rest of the plate is smooth. Neither plate is complete.

Six elements are encountered in the material, next to the skull. They could be considered as hyomandibular and ceratohyals, but as their shape is undiagnostic they could be other elements as well. They can be seen on the 3D pdf available as supplementary material.

Sensory lines.

Their description is given above in the relevant dermal plates. Despite being considered non diagnostic within the genus Bothriolepis [9], we use some characters related to their course (see phylogenetic analysis sections), keeping in mind the possibility of extreme intraspecific or even intra-population variation (see for example [48]).

Trunkshield.

The anterior median dorsal plate is observed on two specimens (NMING:F35201, TCD 19681 FB1xxx; AMD, Figs 2A, 2B and 6A). It is visible in dorsal view as indicated by the posterior extension of the oblique sensory line groove as a vertical line, but the right lateral half is very compressed in NMING:F35201 (approximately 33% shorter). This is due to the lateral main axis of compression of the specimen during fossilisation. The ratio L/W must be slightly greater than 1.3 (width value calculated as the double of the measured left unaltered side of the plates in NMING:F35201 (ratio equals 1.35) and TCD 19681 FB1xxx (ratio equals 1.4)). The anterior edge of the anterior median dorsal plate that articulated to the skull roof is hidden under the posterior edge of the skull roof. The plate overlaps anteroventrally the anterior dorsolateral plate and is overlapped posteroventrally by the mixilateral plate and posteriorly by the posterior median dorsal plate. The posterior edge of the ornamented part of the plate is very clear as it displays a step pattern, and is followed by the overlap area for the posterior median dorsal plate (oa.PMD, Fig 6A). This posterior edge must have been a very wide angle open posteriorly, greater than the measured 145° value. The posterior oblique dorsal sensory line groove on the anterior median dorsal plate is an inverted V-shaped structure (dlg1, Figs 2A, 2B and 6A). Its anterior tip is located at the first third of the length of the plate, which may correspond to the tergal angle (dma, Fig 2A). The groove extends posterolaterally onto the mixilateral plate. A low median ridge is observed in the second third of the plate, between the tergal angle and the level of the articulation of the mixilateral plate (dmr, Fig 2A and 2B). The internal side of the anterior median dorsal plate is not available.

The posterior median dorsal plate is observed on one specimen (NMING:F35201, PMD, Fig 2A and 2B). In NMING:F35201, it is visible in left lateral view and although the right half is very compressed, a portion of the right mesial lamina is visible. It overlaps both the anterior median dorsal and the mixilateral plate. The measured angle of the anterior edge is of 89°, and obviously does not fit the value of the posterior edge of the anterior median dorsal plate; this discrepancy is related to its compression during the fossilisation process. Because the anterior edge is not very damaged, we are confident in its completeness. A calculated angle taking into account only a lateral compression gives a value of 117° ([145+89]/2). In the anterior median dorsal plate TCD 19681 FB1xxx, the value of the half angle between the midline and the posterior edge of the plate is 55°; the total value of the angle for the overlap area for the posterior median dorsal plate (and thus the angle of its anterior edge) is 110°. The posterior edge of the plate is broken, but the right lateral edge is as long as the anterior median dorsal plate. A low posterior median ridge extends along the second third of the plate (dmr, Fig 2A and 2B). A second oblique dorsal sensory line groove, inverted V shaped is observed and continues onto the mixilateral plate (dlg2, Fig 2A and 2B).

The anterior dorsolateral plate (ADL, Figs 2A, 2B and 6B–6E), is known in two specimens: the subarticulated NMING:F35201 and NMING:F35208. It is roughly quadrangular and longer than high. One specimen shows a clear lateral ridge (lr, Fig 6E). The plate is overlapped mesially by the anterior median dorsal plate (oa.AMD, Figs 2A, 2B and 6B–6E) and overlaps posteriorly the mixilateral plate (cf.MxL, Fig 6B). The overlap area for the anterior median dorsal plate is wider anteriorly, is located behind the low postnuchal ornamented corner of the plate and extends between the notch for the external postlevator process and the internal postlevator process of the anterior dorsolateral plate (pn.oa, n.prpl, pr.pn, Fig 6B). A low and smooth lip is visible at the boundary between the ornamented and unornamented areas. The dorsal lamina is much wider than the lateral one is high; this is more marked in adult specimens than in younger ones. The obstantic process is not preserved in its entirety as only a short buttress is visible on NMING:F35201 (p.obs, Fig 2A and 2B). However, the articulated specimen NMING:F35201 shows that the dermal articulation occurred with the postmarginal and the paranuchal plate. A very small anterior dorsolateral plate most likely belonging to a very young individual was discovered in the material (NMING:F35210, Fig 6D). Its ornamentation consists of straight tuberculated ridges, oblique against the sensory line groove.

The mixilateral plate (MxL), is visible in two specimens: in external view with the semi articulated NMING:F35201 (Fig 2A and 2B) and in internal view in NMING:F35212 (Fig 6F). It is broken in two parts: the posterolateral portion broke off and overlaps slightly the dorsal lamina posteriorly. The dorsal lamina is overlapped by the anterior dorsolateral plate anteriorly and by the posterior median dorsal plate dorsoposteriorly, and overlaps the anterior median dorsal plate anterodorsally (oa.ADL, oa.PMD, Figs 2A, 2B and 6F); this overlapping pattern is observed in all Bothriolepis species. The overlap area for the posterior median dorsal plate is slightly higher in its posterior end. The mesial edge of the ornamented dorsal lamina is straight at the exception of the posterior end indented by the posterior median dorsal plate. Similarly, the lateral edge of the ornamented area of the lateral lamina is straight except near the posterior end where the posterior ventrolateral plate indents the mixilateral plate dorsally. The posterior edge of the lateral lamina is concave. The dorsal lamina is crossed in its anterior half by the posterior oblique dorsal sensory line groove continued from the anterior median dorsal plate (dlg1, Figs 2A, 2B and 6A), and in its most posterolateral part by the second posterior oblique dorsal sensory line groove continued from the posterior median dorsal plate; this feature is present in all Bothriolepis species. The main lateral sensory line groove cannot be recognised due to the damage of the lateral lamina in its anterior part.

The anterior ventrolateral plate (AVL, Figs 6G–6I, 7A–7G, 8B and 8C) is known from specimens NMING:F35213, NMING:F35218, NMING:F35219, IS073, IS075, NHMP 596777 and NMING:F35216 (although only digitally for the two latter ones). It is known from both internal and external surfaces with both the ventral and lateral laminae preserved. The lateral lamina is highest above the pectoral fin attachment area. The ventral lamina of NMING:F35213 is wide and displays tubercles anteriorly and anastomosed ridges posteriorly. It overlaps posteriorly the posterior ventrolateral plate (oa.AVL, Fig 6K). The ventral lamina extends anteriorly under the head as a subcephalic division, overlapped by the submarginal plate, visible in NHMP59677 (c.al, Fig 6G and 6I). The prepectoral corner (pr.c, Fig 6G) is well marked and protects the brachial process. The overlap area for the semilunar and the other anterior ventrolateral plate are visible in NMING:F35213 (m1, oa.AVL, Fig 6G and 6K). There is a small unornamented area on the ventral lamina of the plate immediately posteriorly at the articular fossa, as observed in Antarctica forms (a.un, Fig 6G; see [25]:23). The diameter of the axillary foramen is estimated as wide as the length of the brachial process (Fig 7C). A wide and low crista transversalis interna anterior is observed on the internal side of NHMP 59677 (cit, Fig 6I). It looks more like a thickening than a crest, and can be related to either the poor preservation or the adulthood of the specimen, or both.

The brachial process is large (in some individuals it can be 24 mm in diameter as in modelled NMING:F35216; NHMP 59677; p.br, Figs 6G–6I and 7). It is rectangular rounded at its corners (although the rectangular aspect is largely due to compression during fossilisation) and pierced by a large funnel pit (f.p, Figs 6H and 7D–7E). The process faces posteriorly. The anterior dorsal, anterior ventral and posterior dorsal muscle insertions are located above and below the funnel pit respectively (a.sup, a.inf, p.sup, Fig 6H, 6D and 6E). The pars pedalis is short and extends horizontally (pe, Fig 6C–6E). The triangular protractor area is triangular and observed on the mesioventral part of the condyle on NMING:F35216 (f.mp, Fig 6D and 6E). The poorly preserved margo limitans is located anteriorly from the brachial process, and extends dorsally to form a supra-articular ridge (m.lim, sar, Fig 7C–7E). This ridge thickens slightly anteriorly and shows what could be interpreted as an auxiliary articular facet (a.fac, Fig 6D and 6E). The dorsal and ventral edges of the axillary foramen can be distinguished, though poorly preserved (f.ax, Fig 6C). If our reconstruction is correct, the foramen is as large as the whole brachial process.

The posterior ventrolateral plate is known from NMING:F35201, NMING:F35214, NMING:F35215 and TCD 19618 FB4033 (Figs 2A, 2B, 6J, 6K and 8D). NMING:F35214 and NMING:F35201 correspond to small and juvenile specimens, as witnessed by anastomosed ridged ornament in NMING:F35214. NMING:F35214 is preserved as bone posterolaterally and as internal mould in the rest of the plate. NMING:F35215 belongs to a much larger and adult individual, and harbours a tuberculated ornamentation. The lateral lamina shows a typical posterior buttress indenting the mixilateral plate. The ventral lamina is as wide as the lateral is high in the juvenile specimen.

No semilunar nor median ventral plate were identified in the material.

Pectoral fins.

The elements of the pectoral fins can be found dissociated, in chunks or still attached to the armour. The fins are long, and are reconstructed as extending far beyond the posterior edge of the armour (confirming the belonging to the family Bothriolepididae). On NMING:F35201, the left proximal segment (in anatomic connection with the armour) is almost as long as the armour itself (lateral corner of PVL); the right element is observed on the other side of the specimen (Fig 2A and 2B). NHM P 59677 corresponds to an assemblage of head, trunk armour and pectoral fin elements that belonged to one adult individual, indicating a [headshield length / proximal pectoral fin length] of 0.33.

The fins are better known from the proximal segments than the distal segments, but isolated bones of the latter occur in our material (Fig 10). The proximal segments appear gently bowed, convex laterally and concave medially in younger forms (see NMING:F35201, Fig 2A and 2B). In larger specimens, the lateral edge seems straighter, while the medial side keeps its concavity.

The first central dorsal and first central ventral are the most proximal elements (Cd1, Cv1, Figs 2A, 2B, 7, 8E-8K and 11), and they articulate with the brachial process of the anterior ventrolateral plate. When preserved, these articulation areas appear narrow (ar3d, ar3v, Fig 8E and 8G). Tentative reconstructions show that these plates did not enclose completely the brachial process; on the contrary two slits, the mesial one being larger, allowed for some proximolateral fin movements. The lateral marginal plate 2 (Ml2) contacts the central dorsal and central ventral with an acute angle.

The first mesial marginal plate (Mm1, Figs 7B, 7H, 7I, 8K, 11E and 11F) is seldom complete (but see NMING:F35230, Fig 9A). It separates the first central dorsal and central ventral plates approximately at the same level than the latter two connect. The proximal edge of the mesial marginal plate is rounded, located slightly behind the brachial process, and this associated with the axillary foramen of the anterior median ventral plate; this is particularly well visible in NMING:F35216 (Fig 7). Based on the shape and relationships of the other plates of the proximal segment of the fin, it is likely that that the mesial marginal plate 1 does not contact the central ventral 2, because of the median proximal process of the lateral marginal plate 2 (see other side of NMING:F35201; Fig 8K).

The lateral marginal plate 2 (Ml2, Figs 8O, 9D and 9E) is the most striking bone of the fin; it wraps the proximal segment of the fin laterally, and is gently curved; its curvature may increase in very large individuals (NMING:F35224, Fig 9D and 9E). Although ornamentation varies throughout the size and hence the age of each individual, so is the case of the lateral edge of the plate.

The mesial marginal plate 2 is another large element of the fin, visible on both the ventral and mesial side. Younger and older individuals show a single row of bigger protrusions on the angle ventromesial edge, with the median ones fused to each other and the most distal ones being separated (see NHMP 59683, Fig 9B). The mesial edge of the plate is gently concave. Central ventral 1 and median marginal 2 contact each other.

The second central dorsal plate is a small rounded element, elongated proximally, and terminated distally by the narrow articular area for the distal segment of the pectoral fin (Cd2, afd, Figs 8O, 9G–9J). The second central ventral plate is a much larger element, although not connecting anteriorly with the first mesial marginal plate (Cv2, Figs 7, 8H and 8I).

Few elements can be attributed with certainty to the distal segment of the fin. A central dorsal 3 and a central ventral 3 plates are recognized in respectively NMING:F35226 and NMING:F35227 (Cd3, Cv3; Fig 10A and 10B). A series of elements, possibly preserved in medial view, are recognised in NMING:F35228a. A large pointed element is tentatively assigned to a terminal element (Fig 10C and 10D).

The posterior part of the body was likely smooth and not covered with bony scales, as none is encountered in our material, even in the articulated ones (see NMING:F35201, Fig 2A).

Outgroup = Remigolepis + Grossilepis.

A first set of analysis includes the Asterolepididae Remigolepis sp. and the Bothriolepididae Grossilepis sp. as outroups (50 characters * 47 taxa). But as TNT can take into account as outgroup the first taxon only (i.e. Remigolepis), Grossilepis was resolved within the ingroup, as sister-group to B. cellulosa; this poor result, contradictory with the conception of a separate genus for Grossilepis is likely related to long-branch attraction phenomenon (see [49]:277; [32]) but also reminds Janvier and Pan’s hypothesis [24] (see also Fig 12A). The most basal species are Euramerican in the obtained topologies, highly contradicting previously considered scenarios (topologies and indices are given in S9-S12 Figs in S1 File).

Outgroup = Grossilepis.

Remigolepis is removed from the data matrix and Grossilepis is considered as single outgroup (50 characters * 46 taxa). The most basal species are again Euramerican in the obtained topologies, highly contradicting previously considered scenarios (topologies and indices are given in S13, S14 Figs in S1 File).

Outgroup = Bothriolepis niushoushanensis.

Non Bothriolepis taxa (i.e. Remigolepis and Grossilepis) are removed from the data matrix, and Bothriolepis niushoushanensis is taken as outgroup (50 characters * 45 taxa). B. niushoushanensis is considered as one of the most ancient species of the genus, as it occurs in the Eifelian of Ningxia (Northern China Palaeoblock). The topology of the consensus tree reveals a high number of polytomies and is poorly resolved (topologies and indices are given in S15, S16 Figs in S1 File). It is noteworthy that above the outgroup an ensemble of four Gondwanan species (B. karawaka, B. portalensis, B. kohni and B. macphersoni) is in polytomy with a cosmopolitan clade. The resolution is however considered non-satisfactory enough for further consideration.

Outgroup = Bothriolepis shaokuanensi.

B. shaokuanensis is also considered as one of the most ancient species of the genus, as it occurs in the Eifelian of Guangdong (Southern China Palaeoblock). The topology of the consensus tree reveals an irresolution above the outgroup (topologies and indices are given in S17, S18 Figs in S1 File).

Tree topology.

Bothriolepis askinae, from the Givetian of Antarctica, is taken as outgroup (50 characters * 45 taxa). This species is selected preferentially because it is resolved as sister group to all other Bothriolepis species since 1988 (misspelled B. askini in [25]:text-fig. 68B). Another reason to consider this taxon as outgroup is related to the presence of a “simple” preorbital recess (type 1), considered as the primitive condition since the first phylogenetic reconstructions attempts. The research led to 96 most parsimonious trees of 210 steps each (n_MPT = 96, L_MPT = 210 steps, CI_MPT = 0.333, RI_MPT = 0.509) then used for a Traditional Search (tree bisection–reconnection (TBR) algorithm), from which a strict consensus tree was calculated (topologies and indices are given in S23, S24 Figs in S1 File). The strict consensus tree is L_CS = 253 steps long, has a CI_CS = 0.277, and a RI_CS = 0.358; the majority rule consensus tree is L_C50% = 251 steps long, has a CI_C50% = 0.279, and a RI_C50% = 0.365 (Figs 13–15; S25, S26 Figs in S1 File). Robustness analysis was performed (1 ≤ Bremer Index ≤ 5; see further).

The majority rule consensus tree is reasonably well resolved, especially at its base. The Givetian Irish B. dairbhrensis n.sp. is sister group with the Frasnian Scottish B. gigantea (node 5, Fig 13). This clade (node 4) sits in a paraphyletic sequence composed at its base and respectively by the Gondwanan B. askinae, B. mawsoni and B. fergusoni (nodes 1, 2, 3) and above with a clade (node 6) divided into a Gondwanan clade containing B. longi, B. baretti and B. grenfellensis (node 7)and a more cosmopolitan clade (node 9) described below. From node 9, the species B. nitida is sister group of two clades (node 10), the first one being Euramerican (node 11) with a short Gondwanan grade (nodes 12, 13) in-between, and a second clade (node 17) showing B. hickingli resolved as sister group of a very poorly resolved clade containing Euramerican, Chinese and Gondwanan taxa (node 18).

Description of selected clades.

A complete list of character transformation generated from PAUP is provided in Supplementary Information. We focus below on major clades of the majority rule consensus in Fig 13A.

The clade(B. gigantea, B. dairbhrensis sp. nov.) at node 5 is characterised by a ratio of the (anterior edge/mesial constriction width) of the premedian plate < 1.0 (CI₁₂ = 0.222, non-ambiguous transformation), a flat anterior edge of the postpineal plate (CI₁₇ = 0.286, ambiguous transformation), a long supra-occipital sensory line groove (CI₄₂ = 0.200, non-ambiguous transformation).

The clade at node 6 is characterised by an axillary foramen longer than high (CI₃₈ = 0.5), a common contact point between the central ventral plates 1 and 2 and the mesial marginal plates 1 and 2 (Cv1, Cv2, Mm1 and Mm2) in the pectoral fin (CI₃₉ = 0.4, non-ambiguous transformation), and a curved central sensory line groove (CI₄₈ = 0.143, non-ambiguous transformation).

The clade (B. longi (B. barretti, B. grenfellensis) at node 7 is characterised by a position of lateral corner of skull roof at the same level as the junction between the premedian, lateral and paranuchal plates (CI₀₇ = 0.250, non-ambiguous transformation) and by large pits on the internal side of the lateral plate (CI₁₅ = 0.500, ambiguous transformation). This clade is reminiscent of the one suggested by Long [19].

The clade (B. hayi (B. yeungae (B. africana (B. alvesiensis (B. hydrophila (B. taylori, B. canadensis)))))) at node 11 is characterised by a round shape of the skull roof (CI₀₃ = 0.429, ambiguous transformation) and a light posterior indentation of the junction of the infraorbital sensory line groove on the premedian plate (CI₄₆ = 0.286, non-ambiguous transformation). The first character transforms again in B. africana (quadrangular shape of the skull roof, non-ambiguous transformation) and the second for the clade at node 15 ((B. hydrophila (B. taylori, B. canadensis))) (straight junction of the infraorbital sensory line groove on the premedian plate, ambiguous transformation).

The clade (B. niushoushanensis, B. karawaka, B. portalensis, (B. kohni, B. macphersoni)) at node 25 is characterised by a pentagonal preorbital recess (CI₀₁ = 0.333, non-ambiguous transformation), a posterior postorbital process extending anteriorly from the orbital fenestra (CI₀₂ = 0.286, non-ambiguous transformation), a lag between the mesial and ventral laminae of the lateral and paranuchal plates (CI₀₆ = 0.333, ambiguous transformation), a lateral corner of the skull roof anterior to the junction between the premedian, lateral and paranuchal plates junction point (CI₀₇ = 0.250, ambiguous transformation), a ratio of the [width of the interomesiolateral edge / width interoposterolateral edge) of the nuchal plate > 1 (CI₂₃ = 0.2502, ambiguous transformation), a quadrangular postmarginal shape (CI27 = 167, non-ambiguous transformation), a minor crest on the median dorsal plate (CI₃₂ = 0.250, ambiguous transformation), a straight contact between the anterior median dorsal and anteriordorsolateral plates (CI₃₃ = 0.111, ambiguous transformation), a contact between the central first ventral and the second mesial marginal plates in the pectoral fin (CI₃₉ = 0.400, non-ambiguous transformation). The mesial and ventral laminae of the lateral and paranuchal plates are at the same level (ambiguous transformation), ratio of the [width of the interomesiolateral edge / width interoposterolateral edge] of the nuchal plate = 1 (ambiguous transformation) in B. karawaka. The postmarginal plates becomes pentagonal in B. karawaka and the clade (B. kohni, B. macphersoni) (ambiguous transformations). The lateral corner of skull roof becomes posterior to the junction point between the premedian, lateral and paranuchal plates, there is no dorsal crest on the median dorsal plates and the contact between the anterior median dorsal and anterior dorsolateral plates becomes sinuous in B. niushoushanensis (ambiguous transformations). This grouping is reminiscent of the one suggested by Young [25] although this author envisaged a paraphyletic ensemble at the base of a “trifid” preorbital recess rather than a monophyletic group.

Character history.

Below is described and discussed the distribution of character states deemed important by previous authors.

The character “shape of the preorbital recess” (character 1) is divided into 5 states (character 1; state 0: simple; 1: trilobate; 2: pentagonal; 3: trifid; 4: trapezoid) and is a priori unordered and unpolarised (Fig 13B). The simple shape of the preorbital recess is considered plesiomorphic as it is also present outside the genus Bothriolepis (but poorly developed and restricted to the premedian plate), and de facto as it is the pattern of the chosen outgroup (B. askinae). The character has a consistency index of 0.333, and its retention index is of 0.5.

The previous character state transformation sequence for this character was thought as follows [10, 25]: 1←0→2→3 [trilobate←simple→pentagonal→trifid], with a strong probability according to which the trifid state is derived from the pentagonal one, rather than two independent states. It is however noteworthy that the group supported by a pentagonal preorbital recess appears monophyletic (see topology in Young, 1988 [25]:text-fig. 68B, node defined by character i). or possibly paraphyletic (see unresolved topology in Lukševičs, 2001 [10]:fig. 83). As for the fifth state “trapezoid” identified by Moloshnikov [37], it would be the most derived state (see topology in [10]:fig. 83, despite it was not coded as trapezoid). It is nevertheless noteworthy that in Young’s scheme, a clade based on the trifid preorbital recess appears one node more coronally in the tree at node r; Young thus considered the pentagonal state as intermediate, or he would have used it to characterise the sister clade of the one defined by r. Last, the shape of the preorbital recess is unknown in seven out of the eleven species in the “pentagonal–non-trifid” clade have a known preorbital (node defined by characters j and k in [25]).

The current phylogenetic analysis based on the strict or majority rule consensus trees reveals a completely different transformation sequence. The distribution of the states reveals only one unresolved clade supported by the pentagonal recess (node 24: B. karakawa, B. portalensis, B. niushoushanensis, B. macphersoni and B. kohni; Fig 13A). This clade is in polytomy with an assemblage of Gondwanan, Euramerican and Chinese forms, showing simple, trifid and trapezoid shaped preorbital recesses (node 18).

Another important result is that the trifid state is not resolved as the most apomorphic state, but rather seems to constitute an important grade between the simple shaped recess basally, and the simple, trilobate, pentagonal and trapezoid more coronally. The states are however not scattered randomly in the tree, as evidenced by the character consistency index (CI = 0.333; RI = 0.500).

Last, the distribution of the states in the tree reveals the a posteriori ordering and polarisation of this character: 0↔3, 3→1, 3→2, 3→5 (Fig 13C). The central position of the trifid preorbital recess questions its previously assigned apomorphic status, but rather pledges for a plesiomorphy instead, and hence echoes the original hypothesis by Janvier and Pan [24]. This fits with the stratigraphic record, as the earliest occurrence of Bothriolepis is B. shaokuanensis from the Eifelian of South China showing a trifid preorbital recess.

The presence of large pits on the internal side of the lateral plate (character 15) was proposed by Long ([19]:text-fig. 13) to support a clade of three Gondwana forms: B. cullodenensis, B. gippslandiensis and B. fergusoni). More taxa were added later to this clade based on this same character: Long & Werdelin [45] added B. tatongensis, Young ([25]:text-fig. 68B) added B. barretti, Johanson ([9]:fig. 12) added B. africana, and Johanson & Young ([26]:fig. 10) considered the possible inclusion of B. longi (among other positions). This character is not listed in Lukševičs’s matrix and cannot consequently become a synapomorphy ([10]:fig. 83, appendix). The clade of previous authors is not retained; the species previously listed are scattered into three clusters (B. cullodenensis + B. gippslandiensis; B. barretti; B. longi) or not being included in the study (B. africana, B. yeungae). Our study added the species B. grenfellensis [50]. We did not retrieve a clade based on large lateral pits either in our analysis. The consistency index for this character in our majority rule consensus tree is CI = 0.500, and the retention index is RI = 0.666. The derived state is distributed in two distant monophyletic clusters: B. longi, B. barretti and B. grenfellensis basally, and B. cullodenensis and B. gippslandiensis more apically.

Characters of the cheek have been considered. An anterior submarginal plate attachment supported by a separate ridge in the transverse lateral groove is suggested by Young ([25]:text-fig. 68) to support an unresolved and cosmopolitan clade, to which Johanson ([9]:fig. 12) proposed the possible addition of B. yeungae, and Johanson & Young ([26]:fig. 10) that of B. longi. These last authors also considered a smaller Gondwanan and South Chinese clade devoid of Euramerican and other Gondwanan species ([26]:fig. 11). This last smaller option is the one retained by Lukševičs ([10]:fig. 83; inclusion of B. macphersoni, B. karawaka, B. longi, B. kohni, B. portalensis, B. niushoushanensis; exclusion of B. vuwae, B. mawsoni, B. paradoxa, B. obesa). Originally coded in our data matrix, the character was finally removed based on a poor completeness (15%; character 34 in original matrix; see S3 Table). The shape of the submarginal plate is however taken into consideration (character 29, 31% completude).

Regarding the fin elements (character 39), Young [25] considered that the contact between the mesial marginal 1 and central ventral 2 plates, separating the central ventral 1 and mesial marginal 2 plates, supported a clade within the simple preorbital recess grade, consisting of the Gondwanan B. alexi, the Euramerican B. canadensis, B. cellulosa, B. panderi, B. nielseni, B. alvesiensis, B. hydrophila and the South Chinese B. tungseni). This clade was not retained in Lukševičs’ tree [10], the state being scattered in two different branches of his tree (fig. 83). This cannot be related to the slight difference of the chosen taxa (B. alexi and B. tungseni are not part of Lukševičs’s study), but rather to the assumption that B. nielseni is considered as belonging to a trifid preorbital recess group and is resolved very apical compared to the other forms which remain closer to the root. In our analysis, this character is modified from two to three states: 0: all plates contact in one point; 1: contact between the mesial marginal 1 and central ventral 2 plates; 2: contact between the mesial marginal 2 and central ventral 1 plates. This refining of the character results in a decrease in its consistency index (CI = 0.400; RI = 0.666). No clade nor grade can be identified immediately based on this character. However, three interesting observations can be made, but rely on the optimisation of missing data. 1) A contact between the mesial marginal 1 and central ventral 2 plates is distributed as a basal grade and present in the coronal polytomy. 2) All four plates contacting in one point are distributed among an intermediate grade between the base and the coronal polytomy. 3) The contact between the mesial marginal 1 and central ventral 2 plates is distributed in the apical polytomy only (Fig 14A–14D).

Best indexed characters (CI or RI ≥ 0.666).

Ten characters have a Consistency or Retention Index equal to or greater than 0.666. They are discussed below.

A shorter antorbital region (character 5) is primitive and present in the most basal forms of Bothriolepis (B. askinae, B. fergusoni; CI = RI = 1).

The character 15 involving the size of the lateral pits has a CI = 0.500 and RI = 0.666. See above for discussion.

A lateral corner of the skull roof posterior to the level of the contact between the postpineal plate and the orbital margin (character 16) is resolved as the primitive state. A lateral corner at the same level or anterior is observed only in the apical polytomy (node 18) and in B. grenfellensis; one occurrence of a posterior lateral corner is observed in this clade with B. shaokuanensis (CI = 0.250; RI = 0.684).

Anterior flanges of the postpineal plate (character 18) are observed only in the basal form B. fergusoni (CI = 1, autapomorphy).

A nuchal plate as long as wide (character 20) is observed only in B. tatongensis (CI = 1; autapomorphy).

A squarish Nu with convex anterior division of the lateral margin and short posterolateral corners (character 25, [25]) is observed in B. leptocheira-curonica (CI = 1; autapomorphy).

A mixilateral plate not broadest at its dorsal corner in adults (character 35) was observed in B. canadensis (CI = 1; autapomorphy)

The character 39 involving the fin elements has CI = 0.400; RI = 0.666. See above for discussion.

The derived condition of a branch of infraorbital sensory line diverging on prelateral plate parallel to the rostral margin of the head-shield (character 43) is displayed only in B. maxima(CI = 1; autapomorphy).

The derived condition of a long branch of the infraorbital sensory groove/canal diverging on prelateral plate (character 45) is observed only in B. heckeri (CI = 1; autapomorphy).

Iterative reweight of the character “shape of the preorbital recess”.

A series of analyses following the same parameters as the one before (OG = B. askinae; 50 characters * 45 taxa) was performed but by changing iteratively the weight of the first character “shape of the preorbital recess”. The weight was increased until a value of w = 8 (analyses with reweight values of 9 and greater were impossible to perform due to memory errors). We focus on the main indices of the trees: number, length, general CI and RI, CI and RI for the first character for the obtained trees and the majority rule consensus tree; these results are displayed in Table D in S1 File and figured in S33-S47 Figs in S1 File.

Evolution of indices.

The number of retained trees increases when the reweight of the character equals w = 2 (n = 268), but then decreases until a reweight of w = 4 (n = 30), from which the number of trees remains somehow constant. The length of obtained trees increases gently with the reweight of the character (between L = 210 at w = 1 to L = 247 at w = 8); the length of the majority rule consensus trees is more fluctuant but remains stable (L_50max = 251 at w1; L_50min = 232 at w = 3; L50 = 251 at w = 8). The consistency and retention indices of the trees and of the first character for the obtained trees and the majority rule consensus trees increase slowly with the reweighing of the character, but not at the same rate. A reweight of w = 5 is necessary to obtain a CI = RI = 1 for the original trees, but w = 6 is required to obtain the same values for the majority rule consensus. None of the reweighing options solves the discrepancy between stratigraphy and phylogeny. From w = 4 and above, L. zadonica is resolved with an ensemble of trifid preorbital recess, thus avoiding this latter state to characterise a clade; consequently, there is no clade defined by a trifid preorbital recess in any of the reweight options.

Evolution of topology relative to iterative reweighting.

At w = 2 (S34, S35 Figs in S1 File), the topology of the majority rule consensus tree differs greatly from the one previously described (w = 1). B. canadensis is resolved as intermediate within a basal Gondwanan grade; B. dairbhrensis remains sister group for B. gigantea, but this clade is in polytomy with Gondwanan and Euramerican species, and is not resolved as basally as in the previous analysis. The states of character are not consistently distributed, except for the pentagonal shape which characterises a clade in a trifid preorbital grade.

At w = 3 (S36, S37 Figs in S1 File), the character states are a better distributed, although the plesiomorphic state (simple preorbital process) characterises a basal grade and two more apical ensembles (one cosmopolitan grade and one Euramerican clade); the trilobate and trifid shapes are successive grades, and the pentagonal shape characterises one clade within the cosmopolitan trifid grade. B. canadensis is still basally resolved within a Gondwanan grade; B. dairbhrensis and B. gigantea are part of a partly resolved grade of Euramerican forms resolved above the basal Gondwanan grade.

At w = 4 (S38, S39 Figs in S1 File), B. canadensis is again resolved in a basal Gondwanan grade. B. dairbhrensis and B. gigantea are intergroup of B. hayi and B. yeungae, and this larger clade is at the base of a grade characterised by a trifid preorbital recess, located just apical to a trilobate grade. Better distributed, only the pentagonal shape characterises an apical clade resolved above a simple cosmopolitan shape grade, itself resolved above a cosmopolitan trifid shape, itself resolved above the trilobate grade.

At w = 5 (S40, S41 Figs in S1 File), the majority rule consensus tree is less well resolved than at w = 4. The clade ((B. dairbhrensis, B. gigantea), (B. hayi, B. yeungae)) is unresolved in polytomy with many other Euramerican trifid preorbital recess forms. A clade characterised by the trilobate shape is resolved within the simple shape grade. The pentagonal shape corresponds to a clade resolved with an unresolved coronal ensemble of simple shape.

At w = 6 (S42, S43 Figs in S1 File), the state of characters are well distributed. The trilobate shape forms constitute a clade in the basal simple-shape grade. The pentagonal shape forms a clade sister group with a more apical clade of simple shapes. The clade ((B. dairbhrensis, B. gigantea), (B. hayi, B. yeungae)) is unresolved in polytomy with many other Euramerican trifid preorbital recess forms. The Chinese taxa B. shaokuanensis and B. niushoushanensis belong to an unresolved clade characterised by the pentagonal shape of the preorbital recess.

At w = 7 (S44, S45 Figs in S1 File), and w = 8 (S46, S47 Figs in S1 File), the topology is similar to the one observed in w = 6.

Robustness of the tree: Bremer decay index.

We used Bremer Index (BI) to evaluate the robustness of the tree. We ran a series of searches for suboptimal trees corresponding to n = n_MPT +x, n_MPT being the number of steps in the most parsimonious trees, and x being the Bremer index value. The following code line was added to the original TNT search script:

    subopt x; bbreak;

We then computed the consensus trees and observed which branches collapsed from the MPT topology presented above.

For BI = 1, the number of maximal trees retained was expanded to 100 000, and the research led to 12 629 trees. The strict consensus is poorly resolved (S48 Fig in S1 File), B. askinae remains sister taxa to a polytomy containing unresolved species, except a few sister group species: B. gigantea and B. dairbhrensis, B. tatongensis and B. jarviki, and a trichotomy between B. hydrophila, B. taylori and B. canadensis. The majority rule consensus tree topology is preserved, and is even slight better resolved than that of the most parsimonious trees (S49 Fig in S1 File): B. karawaka is sister group to B. portalensis, and B. kohni to B. mcphersoni. These two latter groups are in polytomy with B. niushoushanensis.

For BI = 2, the research had to be limited to 1 million trees. The strict consensus tree topology is reduced to a complete polytomy with B. askinae as outgroup (S50 Fig in S1 File). The majority rule consensus tree is still surprisingly well resolved, and only a few branches collapse (S51 Fig in S1 File). B. yeungae, B. hayi are in polytomy with a resolved clade containing ((((B. canadensis, B. taylori) B. hydrophila) B. alvesiensis) B. africana). B. volongensis is resolved in polytomy one node more basally than in BI = 1.

For BI = 3, the number of maximal trees retained is kept at 100 000 000. The strict consensus tree topology remains a complete polytomy with B. askinae as outgroup (S50 Fig in S1 File). The majority rule consensus tree is slightly better resolved than at BI = 2, as a clade reappears containing B. niushoushanensis in polytomy with two clades consisting of (B. karawaka, B. portalensis) and (B. kohni, B. macphersoni) (S52 Fig in S1 File).

For BI = 4, the number of maximal trees retained is kept at 100 000 000. The strict consensus tree topology remains a complete polytomy with B. askinae as outgroup (S50 Fig in S1 File). The majority rule consensus tree is again better resolved than at BI = 3, as a clade reappears containing B. ornata as sister group to a clade containing B. ciercere and B. volongensis (S53 Fig in S1 File).

For BI = 5, the number of maximal trees retained is kept at 100 000 000. The strict consensus tree topology remains a complete polytomy with B. askinae as outgroup (S50 Fig in S1 File). The majority rule consensus tree is again better resolved than at BI = 4. Several clades appear in the apical polytomy sister group with B. hickingli: B. niushoushanensis becomes sister group with B. kohni and B. macphersoni (S54 Fig in S1 File), and the three of them is sister group with a clade uniting B. karawaka and B. portalensis; B. traudscholdi becomes sister group of the clade (B. leptocheira-curonica (B. evaldi, B. jani)); B. maxima becomes sister group of B. panderi. The clade sister group to the clade holding B. hickingli and the apical polytomy is also resolved: compared to BR = 4, B. yeungae and B. hayi are resolved as successive sister groups for the previously resolved (B. africana (B. alvesiensis (B. hydrophila (B. taylori, B. canadensis)))).

Summary of results

Several important pieces of information must be retained from the unweighted majority rule consensus tree. 1) The base of the tree is very well resolved and kept in the strict consensus tree, providing us a higher degree of confidence than previous analyses. 2) A short Euramerican clade containing the Givetian Irish species B. dairbhrensis n. sp. is quite basally resolved in a Gondwanan paraphyletic sequence. 3) Except at the base of the tree, there is no clear larger palaeogeographic clade or grade identifiable (i.e. which would have been totally Gondwanan, Euramerican or Chinese). 4) The character “shape of the preorbital recess” (character 1 of the current matrix) is not as consistently resolved as what it was assumed from previously published phylogenies (see distribution in dedicated column of Fig 12). 5) The most ancient forms are not the most basal in the tree (but this is pro parte related artificially to the choice of the outgroup, see previous attempts above); on the contrary they are resolved quite apically. These remarks invite for important palaeobiogeographic and taxonomic consideration undertaken in the discussion part. 6) The species Livnolepis zadonica is resolved within the genus Bothriolepis. 7) Even a slight reweight of a character modifies greatly the tree and characters indices and topologies.

Preorbital recess.

Different preorbital recesses were first noted by Stensiö [43] who distinguished a “canadensis” (i.e. “simple, type 1) from a “groenlandica” type (i.e. trifid, type 4). Young [25] added two more types, a trilobate one (after B. barretti, type 2), and a pentagonal one (after B. portalensis, type 3). The latest addition of a fifth pattern consists in a very thick and trapezoid preorbital recess, hitherto known only in Livnolepis (Bothriolepis) zadonica [37].

As mentioned above in the discussion, the shape of the preorbital recess has been used to build morphotype-based groups and phylogenies [41]. In other words, the shape of the preorbital recess was considered to record the evolution of the genus Bothriolepis in a consistent and strict parsimonious way, no reversion or convergence being deemed possible. This comfortable status-quo remained unchanged and rarely challenged until today. Johanson indeed mentioned some problems of keeping phylogenies based on preorbital recesses: keeping this in mind while hand-building a tree inevitably may lead to unconscious over-weighing of the character (when all characters should have the same weight a priori; [9]:338–339). This was reflected in her attempts to place B. yeungae in the clades identified by Young ([25]: 3 different position hypotheses; [9]:fig. 12). This problem persisted later, as she and G. Young tried to resolve B. longi: 4 positions and 2 different tree topologies were suggested ([26]:figs 10 and 11). As a matter of fact, the authors mentioned clearly that “none of the [major clades previously recognized] should be regarded as strongly supported” ([26]:68). This poses the problem of a possible inherent homoplasy within the character “shape of the preorbital recess”, but also all the other ones. This was tested in our own analysis, and it appears that the “preorbital recess shape” has some degree of homoplasy (CI = 0.333, RI = 0.500; see below/above).

In 1988, Young considered that the trifid recess unifies all northern hemisphere taxa [25]. This hypothesis is no longer supported in our cladogram, as it was already challenged by the occurrence of such characteristic in Gondwanan (B. yeungae) and Chinese forms (B. shaokuanensis).

Far from supporting a very apical clade as expected by Young, the trifid preorbital recess is resolved as potentially leading to other shapes (Fig 13C). Although the morphology-based cladistic analysis is disconnected from stratigraphic information, this result echoes with the fact that an Eifelian Chinese form (B. shaokuanensis) possesses a trifid preorbital recess. It is even worth remembering that the trifid recess was once considered plesiomorphic among Bothriolepidoidei because suggested possibly homologous with the preorbital depression of the Yunnanolepidoidei [24]. On the contrary, Long denied such homology on the basis that the yunnanolepidoideid preorbital depression was ornamented and thus could not hold the rhinocapsular ossification; consequently he considered the trifid process as apomorphic among Bothriolepididae [19]; Wang and Zhu confirmed the non-homology between the preorbital recess and the preorbital depression in antiarchs [80].

It is also noteworthy that the trifid shape is quite widespread, and that a common condition is often a priori considered as the primitive condition; this however relates to the knowledge we have of the group (a fortiori based on incomplete fossil data) and the sampling (see [81, 82]). Moreover, the simple (semicircular) preorbital recess is encountered outside of the genus Bothriolepis (i.e. Grossilepis and Wufengshania); despite these genera contain very few species, such distribution of a simple recess can be considered even wider than a trifid one.

By extended outgroup comparison, and considering that a simple preorbital recess occurs outside Bothriolepis (although restricted to the premedian plate), it is difficult to conceive a morphologic continuous transformation from the trifid to pentagonal shape. The state distribution observed in the main majority rule consensus as well as in those issued from the reweights of the preorbital recess shape character invite for an exploration of Emsian and Eifelian terranes and taxa in order to possibly discover new forms which will help answer this ordination and polarisation problem, and consequently the importance of the character in the Bothriolepis phylogeny.

Cheek plates.

Young considered the morphology of the anterior attachment area of the submarginal plate on the lateral plate, associated with the shape of the prelateral plate attachment area: a strong transverse anterior attachment for the submarginal plate, separate from the one for the prelateral plate would have occurred in the pentagonal group. Such pattern was also identified in the trifid form B. yeungae [9, 25]. Young also envisaged the modification of the shape of the anterior median dorsal plate throughout ontogeny as an relevant character [25]. As mentioned above, the poor completeness of these characters was not considered high enough to include the characters in the analyses (see S3 Table).

One character considering the shape of the submarginal plate is retained in our analysis (Character 29; completeness = 31%; CI = 0.500). A low and long submarginal plate is considered as the plesiomorphic state (present in B. askinae); only two species in polytomy in the most coronal unresolved clade (B. ciercere and B. ornata) display the apomorphic state (short and deep).

How many (valid) species of Bothriolepis?

Bothriolepis is undoubtedly a speciose genus (i.e. related to the high number of species rather than a visually pleasing aspect; see [83]). The increasing number of Bothriolepis species through time is plotted on a diagram, based on the taxonomic authority of each species, including synonyms; articles mentioning indeterminate species and redescribing previously named species are not taken into account (S2 Fig and Table A in S1 File). A different colour-code indicates whether a given species remained valid, became a junior synonym of another Bothriolepis species or was reassigned to another genus.

Obviously the publication of new species depends on many factors (e.g. access to localities, historical, anatomical and systematics comprehension of the material, etc.) and should be seen through the eyes of a natural science historian rather than through those of a data analyst. The discovery and publication of new species relies on many independent factors, such as access to fossil locality (sometimes extremely difficult, such as Antarctica), geopolitical context (some fossils could indicate terranes of economic importance considered non-disclosable and embargoed for a period of time), but also the technology available for excavation, preparation and study, as well as previously published works. However, a rapid overview is possible in order to get a general idea of the research dynamics associated with the different Bothriolepis species.

95 species are recorded, ranging from 1840 with the erection of Bothriolepis ornata [8], to B. dairbhrensis sp. nov. in 2021 Several articles published more than one new species at a time and suddenly increased the number of species and contributed to the then knowledge of the genus diversity. Chronologically, they are: 1) Agassiz, 1844 (3 species [84]), 2) Traquair, 1888 (2 species [52]), 3) Traquair, 1890–92 (1 species, but it is also associated with an other assemblage; i.e. B. leptocheira curonica [85]), 4) Gross, 1942 (2 species [86]), 5) Stensiö, 1948 (3 species [43]), 6) Miles, 1968 (6 species [20]), 7) Long, 1983 (4 species [19]), 8) Young, 1988 (10 species; 13 more indeterminate species are not taken into account because unnamed [25]) and Malinovskaya, 1988 (4 species, one of them was later reassigned to the genus Tubalepis [87]).

Many species are described on an incomplete set of dermal plates and endocranial structures (some taxa lack head or thoracic armour). The individual variability within a population and/or a species, associated with growth heterochronies and ontogenetic variations increase the taxonomic complexity of the task (e.g. plate shapes and sizes, ornamentation, sensory line system; e.g. [46, 48, 88]). Some species may have been erected based on stratigraphic and (palaeo-)geographic reasons rather than morphological, leading to the increase of taxonomic complexity of Bothriolepis (see for example B. nitida vs. B. minor, B. leidyi, B. coloradensis, B. darbiensis and B. virginiensis, although the latter is considered morphologically distinct and therefore a valid species by Weems, 2004; see [27–29]). As far as the current study is concerned, 77 species of Bothriolepis are considered valid, and 43 are retained for our final phylogenetic analysis (plus Livnolepis (Bothriolepis) zadonica).

Could the genus Bothriolepis be paraphyletic?

For decades, the monophyly of the genus Bothriolepis has remained essentially unquestioned. This is due to a combination of different factors: the number of species increasing with time (see above), the late access to cladistic methodology (Hennig’s work was translated in 1966 in English, and 50 species of Bothriolepis were then recorded; [89–91]), and computing facilities and calculation power for numerous taxa and characters data matrices (the first phylogenetic software was PHYLLIP by Felsenstein in 1980; 59 species recorded then; [75]). To these technical limitations it should be added that the difficulty of data treatment (polymorphism, missing data, stratigraphic considerations, etc.), together with a morphologically ungrateful material, has contributed to the procrastination of a complete Bothriolepis overview study. A review of previous phylogenetic analyses is given above, and all were handmade (see also Fig 12).

The non-monophyly of Bothriolepis (rather its paraphyly) was suggested by Janvier & Pan ([24]:fig. 12) when they considered the possible inclusion of Grossilepis within the genus Bothriolepis. These authors reminded that Grossilepis differs only slightly from the then known species of Bothriolepis. They even considered that the ‘simple’ preorbital recess shared by G. tuberculata, B. canadensis and B. cellulosa may be a synapomorphy for these forms.

The non-monophyly of Bothriolepis was later suggested in a two-step process: 1) the publication of Lukševičs’ tree including B. zadonica ([10]:fig. 83), and 2) the exclusion of this species from the genus Bothriolepis and the erection of the genus Livnolepis by Moloshnikov [47]. The main reason to assign B. zadonica to a distinct genus relies on the thickness of the preorbital recess and the angle between its nasal and facial laminae. For the same reason, Moloshnikov assigned B. heckeri to Livnolepis [92]. Surprisingly, these authors have never acknowledged the phylogenetic status consequence of this change of genus.

Interestingly, our phylogenetic analysis recovers the idea of Janvier & Pan because Grossilepis is resolved within the genus Bothriolepis when Remigolepis is taken as outgroup. This can be explained by either long branch attraction phenomenon or parallelism in Bothriolepis or a real misinterpretation of Grossilepis as a separate genus; these hypotheses advocate for a non-paraphyly of Bothriolepis.

However, the non-legitimacy of the separate genus Grossilepis appears a very weak argument, and is hardly believable considering the diagnosis of this taxon, relying on clear morphologic differences with Bothriolepis since its erection in 1948 by Stensiö. We consequently consider that analytic biases rather than real biological and evolutionary trends explain the position of Grossilepis within Bothriolepis, and that Grossilepis [43] should remain a valid genus.

Similarly, Livnolepis zadonica is resolved within the genus Bothriolepis in all of our analyses. Moloshnikov erected the genus several years after describing extensively B. zadonica [37, 47]. The most striking feature of the genus is the thick trapezoid preorbital recess, which was originally considered as a fifth morphologic possibility in the genus Bothriolepis [37]. It is however noteworthy that Lukševičs resolved B. zadonica as sister species of B. wilsoni, on the shared possession of a large median dorsal crest ([10]:fig. 83), and both species were included in a clade based on a trifid preorbital recess. The question hence revolves on the anatomical distinction and taxonomic distribution of this pattern.

Morphologically speaking, the distinctiveness of the preorbital process in Livnolepis (Bothriolepis) zadonica is unquestionable: the wide lateral flanges of the recess cannot be mistaken with longer lateral protrusion of the classic trifid recess. The apparent thickness of the preorbital recess in Livnolepis pledges for the recognition of a real distinct process rather than a variation of the trifid type.

Our phylogenetic analysis resolves Livnolepis within the genus Bothriolepis, but we acknowledge the lack of any other genus of the Bothriolepididae in the current data set. A wider sampling of Bothriolepididae genera and closely related forms, including a number of Bothriolepis species should be analysed in an attempt to resolve the validity of the genus Livnolepis (and possibly others); such analysis is beyond the scope of this article.

The question remains open regarding either the paraphyly of the genus Bothriolepis or a contrario the validity of the genera Grossilepis and Livnolepis (which should then be considered as junior synonyms of Bothriolepis). Arguing these questions is beyond the scope of the present article and will be the topic of another study (in progress).

Emsian.

The earliest record of Bothriolepis occurs in the upper part of the Emsian of South China with two indeterminate species from the Jiucheng and the Chuandong formations in Wuding and Qujing respectively (Yunnan, Assemblage V in [70]; S3 Fig in S1 File). It is noteworthy that other Bothriolepididae were identified and described in the Emsian of Taemas Wee Jasper in eastern Australia (Monarolepis verrucosa, previously assigned to the genus Bothriolepis; see [25, 51]) and of South China (Wufengshania magniforaminis, Houershanolepis zhangi, and Wudinolepis weni; see [60, 68–70]).

Eifelian.

The Eifelian record of the genus Bothriolepis is restricted to the Chinese palaeoblocks: B. niushoushanensis is encountered in the north in Ningxia [70, 93], while B. shaokuanensis, B. sinensis, B. tungseni, B. yunnanensis, and Bothriolepis sp. occur in the southern palaeoblock in Yunnan, Guangxi and Guangdong [30, 60, 70, 94, 95], as is the Bothriolepididae Dianolepis liui in Yunnan [61, 70]; the latter species is also encountered in the Givetian of Yunnan [96], although not listed as such by Zhu who considers the species present only in the Eifelian of the Haikou and Wuding Formations ([70]:378) (S4 Fig in S1 File).

Givetian.

The Givetian record of the genus Bothriolepis starts getting more cosmopolitan with evidence in the South China Palaeoblock, in Eastern Gondwana, in Euramerica and Kazakhstan (Figs 16 and 17, S5 Fig in S1 File). B. lochangensis and B. kwangtungensis are encountered in the Dahepo Formation in Lechang, Guangdong, and Anfu, Jiangxi [70, 97, 98], and Bothriolepis sp. is encountered in the Laohuao Group of Taishan, Guangdong [70, 98]. It is also present in eastern Gondwana, especially in the Antarctic Aztec siltstone with B. antarctica, B. askinae, B. kohni, B. portalensis, B. barretti, B. alexi, B. karawaka, B. macphersoni, B. mawsoni, B. vuwae and 13 indeterminate species [25, 99, 100], but also in eastern Australia with B. longi, B. bindarei, B. tatongensis, the latter possibly belonging to early Frasnian [19, 26, 45]. Bothriolepis babichevi and B. kassini occur in the Konyr Formation of central Kazakhstan, which corresponds to the end of Givetian or early Frasnian [62, 87, 101, 102].The earliest evidence of Bothriolepis in Euramerica comes from the mid-Givetian Valentia Slate Formation in Ireland, described in the present article. Other Bothriolepididae occur around the Givetian world: Dianolepis liui in the Haikou Formation of Yunnan ([96]; although not in [70]), Vietnamaspis trii from the Givetian–Frasnian of Vietnam [64], in eastern Australia with Briagalepis warreni (originally belonging to Bothriolepis and assigned to a new genus by Long, 1990 in [64]). Finally, Grossilepis spinosa is encountered in the Givetian Gauja Formation according to Sallan and Coates [96], but Lukševičs [10] considers this taxon as middle Frasnian (Pamūšis regional stage) (Figs 16 and 17, S5 Fig in S1 File).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 16. Distribution and composition of Givetian vertebrate faunas based on macroremains.

Pies correspond to best known localities and associated updated taxa formerly listed in Sallan and Coates [96], with the addition of Valentia Slate formation; they are plotted on a palaeomap of the Givetian (modified after [38]; oceans in dark grey, shallow and epicontinental seas in light grey, emerged continents in white). Early Givetian localities have their pies closer to the edge of the map; late Givetian are furthest. NB. 13 unnamed and indeterminate species of Bothriolepis in Antarctica mentioned by Young [25] are taken in consideration in this figure.

https://doi.org/10.1371/journal.pone.0280208.g016

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 17. Distribution and composition of Givetian Antiarcha.

Pies correspond to best known localities and associated updated taxa formerly listed in Sallan and Coates (96), with the addition of Valentia Slate formation; they are plotted on a palaeomap of the Givetian (modified after [38]; oceans in dark grey, shallow and epicontinental seas in light grey, emerged continents in white). Early Givetian localities have their pies closer to the edge of the map; late Givetian are furthest. NB. 13 unnamed and indeterminate species of Bothriolepis in Antarctica mentioned by Young [25] are taken in consideration in this figure.

https://doi.org/10.1371/journal.pone.0280208.g017

Frasnian.

The genus Bothriolepis becomes really cosmopolitan in the Frasnian with fossil evidence in South China, eastern, northern and western Gondwana, eastern and western Euramerica (on either side of the Appalachian—Caledonian mountains) and in Kazakhstan (S6 Fig in S1 File). Bothriolepis sp. is observed in the Shetianqiao Formation of Taojiang, Hunan [70, 98]. Its presence in the early Frasnian of Mount Howitt, Victoria, Australia can be evidenced with B. longi and B. tatongensis (but see comment on possible Givetian age in former paragraph; [19, 26, 45]). Bothriolepis sp. is also present in the late Frasnian Gogo Formation in Western Australia [96]. As for the northern margin of Gondwana, Bothriolepis sp. would occur in the lower Frasnian of Iran (Chahriseh and Kerman; [96]). As for the north-western part of Gondwana, B. perija is encountered in western Venezuela [103, 104], and Bothriolepis sp. in the Cuche Formation of Columbia [105, 106]. Eastward from the Caledonian Mountains in Euramerica, the genus is present in the mid-late Frasnian of the Alves Beds of Scotland with B. gigantea, B. alvesiensis, B. macrocephala, B. major, B. obesa, B. paradoxa, B. stevensoni, B. taylori and B. wilsoni [5, 107], in the Baltic states and western Russia (East European Platform) with B. leptocheira, B. leptocheira curonica, B. evaldi, B. favosa, B. cellulosa, B. maxima, B. obrutschewi, B. prima (e.g. [5, 10]; B. prima and B. obrutschewi would be the earliest forms in the upper part of the Amata formation), and in Armenia, eastern Russia, Siberia, Timan and/or Tuva, with B. dorakarasugensis, B. jeremijevi, B. siberica, B. panderi, B. markovski, B. prima, B. traudscholdi, B. maeandrina, B. volongensis [5, 10, 62, 108, 109]. Westward from the Caledonian mountains in Euramerica, the genus occurs in the Frasnian of Canada (B. rex, B. canadensis, B. traquairi; [5, 74, 96]), in the eastern (B. minor and B. virginiensis; [5, 27, 96]) and western parts of the USA (B. coloradensis and B. darbiensis; [5, 28, 96, 110]). In the Kazakhstan palaeoblock, five species are recorded: B. kassini, B. amakonyrica, B. nikitinae, B. tastenica, and B. sanzarensis [57, 62, 87, 111, 112]. Indeterminate Bothriolepididae are encountered in Tarim (Xinjiang Province, China; [70]). Briagalepis warreni occurs in the early Frasnian of Victoria, Australia (but could possibly be dated as Givetian; see above), Vietnamaspis trii occurs in Vietnam, while Grossilepis brandi occurs in Scotland, and Grossilepis spinosa and G. tuberculata occur in the Baltic States.

Famennian.

The distribution of the genus Bothriolepis remains cosmopolitan until the end of the Famennian, during which the genus becomes extinct as well as all other placoderms (S7, S8 Figs in S1 File). In the South China palaeoblock, Bothriolepis sp. is encountered in the Xikuangshan Formation of Lengshuijiang and Liuyang, Hunan [70, 98]. Occurrence in Gondwana is as scarce as in China, with few occurrences in the east (B. yeungae in Canowindra, New South Wales, Australia; [9]), and in South Africa (B. africana in Witpoort; [41]). The genus appears to be the most diversified in Euramerica, especially east from the Caledonian mountains. In England and Scotland, the following species are encountered B. alvesiensis, B. cristata, B.hayi, B. hickingli, B. hydrophila, B. laveroklochensis, B. leptocheira, B. macrocephala, B. obesa, B. paradoxa, B. stevensoni, B. wilsoni [5, 10, 20, 43, 96]. In Eastern Europe the following species are encountered: B. sosnensis, B. maeandrina, B. ciercere, B. jani, B. ornata, B. parvaniensis, B. heckeri, B. jazwicensis, B. jeremjevi and Bothriolepis sp. [5, 10, 96], but also the Bothriolepididae Livnolepis zadonica, Rossolepis brodensis and Grossilepis tuberculata [5, 47, 65, 96]; Tubalepis extensa may belong to a separate family Tubalepididae [56, 57]. In the western part of Euramerica, Bothriolepis nielseni, B. groenlandica and B. jarviki occur in Greenland [5, 43, 96], B. nitida and B. minor occur in the eastern part of the U.S.A. [5, 96] and B. coloradensis in the west [5, 96, 110]. At the very end of the Famennian, Bothriolepis subsists in South Africa (see above) and in Belgium with B. lohesti and Bothriolepis sp., together with Grossilepis rikiki [113–115].

Previous mentions of pre-Frasnian occurrence of Bothriolepis in Euramerica.

Below we review the occurrences of Bothriolepis in the pre-Frasnian terranes of Euramerica (i.e. Baltica and North America). This review is important as we advocate that the new Irish B. dairbhrensis sp. nov. is the most ancient occurrence of the genus in Euramerica, and that this taxon (with others) is used for important palaeogeographic implications.

Regarding Baltica, authors have moved the Givetian–Frasnian boundary between the Burtnieki, Gauja, Amata and Pļaviņas Regional Stages or their coevals in Estonia, Lithuania, Latvia, Belarus but also the western part of Russia [116–122]. Most recent works seem to consider a late Givetian age for the Gauja Formation and that the Givetian-Frasnian boundary would be situated at the base of the Amata formation [122–124].

It is also noteworthy that “the oldest remains of Bothriolepis in the Main Devonian Field, of probable Givetian age (Gauja Formation), were found in 1964 in borehole material from Latvia [125]. Unfortunately these remains are lost” ([10]:497). All the Bothriolepis material described in Lukševičs [10] come from the Main Devonian Field belonging to Upper Devonian deposits ([10]:499), with B. prima, B. obrutschewi, B. cellulosa, B. panderi, B. traudscholdi, B. maxima and B. evaldi coming from the Frasnian deposits, while B. leptocheira curonica, B. ornata, B. jani, B. heckeri and B. ciercere occur in the Famennian sequence [10].

Bothriolepis curonica, entered by Sallan and Coates [96] as occurring in the Givetian Gauja Formation actually refers to B. leptocheira curonica from the Famennian Eleja Formation (see [10]). Likewise, Grossilepis spinosa and Taenolepis speciosa rather belong to the Frasnian Pamūšis Regional Stage [10]. Only Asterolepis ornata from the Givetian Gauja Formation remains from Sallan and Coates’s original list ([96]; supplementary information: 27).

In north America, Bothriolepis sp. was recorded from the mid-Givetian Rockport formation (see [96]), but the material was later re-attributed to juvenile (smaller) forms of the arthrodire Protitanichthys rockportensis ([126]:14). In other words, there is no pre-Frasnian occurrence of Bothriolepis in North America (see S3-S8 Figs in S1 File).

Frasnian Western Gondwana.

The following forms could not be taken into account in our phylogenetic analysis because of the incompleteness of the taxa. However, their occurrence is acknowledged in our palaeogeographic discussion.

The Upper Devonian Cuche Formation of Columbia yielded some material attributed to Bothriolepis by Janvier & Villaroel described in 2000, who refer to an “elliptic shape” of the preorbital recess (thus “simple, type 1 preobital recess in our article). The inner side of the plate however shows a triangular vault which could indicate a trifid type (i.e. type 4 in our work; see [105]:pl. 1, fig. 7). A more recently described Bothriolepis sp. seems to have a simple preorbital recess ([106]:fig. 4H). This would indicate the presence of at last two species of Bothriolepis in the Cuche Formation.

The Frasnian Campo Chico Formation in Venezuela was studied notably by Young and Moody in in 2000 [103]. The authors erected and described the species Bothriolepis perija from Venezuela. Rests of this species are found in “locality 139, Caño Colorado Sur, in the upper member of the Campo Chico Formation”([103]:166), which corresponds to an early to mid Frasnian age ([103]:164). They also raised concerns about the processes leading to the known spatial and temporal distribution of the Venezuelan fauna, and especially Bothriolepis. Such processes can only be inferred from phylogenetic analyses ([127] in [103]). Bothriolepis is one of the most widespread and abundant vertebrate in the Devonian world, and its occurrences span a large variety of environments (e.g. freshwater, reef or open marine, even brackish waters). Young et al. [104] suggested that “Bothriolepis no doubt dispersed through the sea”. Interestingly, because of morphological affinities of B. perija with other Gondwanan rather than closer Euramerican forms, Young & Moody considered a dispersal scenario along the coasts rather than by crossing large areas of open water, and consequently concur with Rosen’s definition of a continental fish for Bothriolepis ([103]:163).

Notes on the stratigraphic uncertainties of some taxa.

Below we report the uncertainty of the stratigraphic information displayed in the cladograms (dashed lines in Figs 13–15). Although much larger than their real and likely more finite occurrences, these larger ranges do not infer the discussion regarding a northward dispersal of Gondwanan faunas in the Givetian.

• Regarding B. darbiensis, the most precise available information is Frasnian in the lower part of the Upper Devonian Darbty Fm. [5, 110].
• B. cellulosa occurs in the earliest Frasnian (Pļaviņas Baltic Regional stage, in [10]; Ust’ Yarega Fm. in Timan, in [128, 129]).
• B. volongensis occurs in the Upper Part of the Rassokha Fm., dated as Frasnian [130], without further precision.
• B. obesa occurs in the Upper Devonian of the Jedburgh beds [5], without further precision.
• B. niushoushanensis occurs in zone VI in [70] dated as Eifelian.
• B. longi occurs in the Middle-Upper Devonian (Givetian-Frasnian) according to Johanson and Young [26]. The authors note strong affinities with Aztec Antarctic fauna, but acknowledge that this does not mean synchronicity.
• B. hayi is recorded in the Upper Devonian [5, 131], without further information.
• B. hickingli occurs in the Famennian [131], without further information.
• B. virginiensis occurs in the Upper Devonian, without further information
• B. coloradensis is noted as Frasnian (Elbert Fm.) by Denison [5, 110], but is actually Famennian (Chaffee Fm.; see [96]).
• Regarding B. wilsoni, we do not possess more precise available information than Upper Devonian [5]; it could at best belong to the upper part of the Upper Old Red Sandstone Lithofacies in ([131]:49).
• B. gigantea could be mid-Late Frasnian in age according to Becker et al. [107] in Parfitt et al. [132].
• Lastly, B. alvesiensis occurs in the Alves beds considered as mid–late Frasnian according to Becker et al. [107] in Parfitt et al. [132]. Rosebrae beds are considered as late Famennian according to Marshall in Rogers [133] (see Stephenson and Gould, 1995 [134], at http://earthwise.bgs.ac.uk/index.php/Devonian,_Grampian_Highlands); Rosebrae beds are considered as early Famennian by Sallan and Coates [96]. NB to coauthors, reviewers and editors: KEEP URL IN TEXT UNTIL PUBLICATION (OTHERWISE IT DISAPPEARS FROM THE ENDNOTE GENERATED LIST)

Phylogeny and palaeobiogeographic implication: Proposal of an early northward dispersal scenario.

The Euramerican clade containing B. dairbhrensis n. sp. and B. gigantea resolved at the base of the tree in a Gondwanan paraphyletic sequence indicates a first northward dispersal episode from Gondwana to Euramerica of Bothriolepis, although it cannot be precised whether their common ancestor was Euramerican or Gondwanan (node 5, Fig 13). This first dispersion occurred in the middle Givetian, thus much earlier than what previously thought based on the sole stratigraphic occurrence of the most ancient Euramerica species of Bothriolepis (B. cellulosa in Latvia and Timan, B. obrutshewi in Baltic states, B. prima in Latvia and Armenia, for the species with the shortest and most secure stratigraphic range; see S. Fig 6 for other possible early Frasnian distribution in Euramerica).

It is noteworthy that the above hypothesis derived from the phylogenetic analysis of the sole genus Bothriolepis is consistent with stratigraphic data (except maybe the Chinese record), but is also supported by other vertebrate taxa (Figs 16 and 17). Preliminary observations indicate the presence of rhizodont sarcopterygians in the same Givetian Valentia Slate Formation fauna (i.e. different from the Frasnian Sauripterus sp. mentioned in [4]).

After this first Givetian northward dispersion, it is unsure whether some newly Euramerican forms swam back to Gondwana or whether a pre-existing pool remained. What is certain is that later Euramerican species are not directly related to the B. dairbhrensis–gigantea group, but instead presumably represent later dispersal waves.

The most apical half of the tree (node 18 and more apical) is mainly poorly resolved. This can be explained by several reasons. Taxonomically, the speciose aspect of the genus is problematic, and related to the validity of all species, together with the incomplete aspect of the fossil material. Biologically, this could correspond to interbreeding between closely related species (although this is obviously non-testable hypothesis). Evolutionarily, a rapid diversification can blur the relationships between species; this is likely to occur when ecological niches become newly available, intraspecies competition increases, and possibly with invasive species in a new environment devoid of predators (see [135]).

Methods.

In order to comprehend the palaeogeographic importance of the Valentia Slate vertebrate assemblage, it is important to compare its components with those of other Givetian localities (Figs 16 and 17).

Sallan & Coates [96] compiled a list of vertebrate taxa from several well-known macrofossil localities and compared their assemblage over several orders of Chondrichthyes, Acanthodii, Actinopterygii, Sarcopterygii and Placodermi. Their list is here updated (Figs 16 and 17; S5 and S6 Tables). The Bothriolepis material of Rockport Formation, MI, USA, was since re-attributed to juvenile smaller forms of the arthrodire Protitanichthys rockportensis ([126]:14). Palaeospondylus is considered as incertae sedis. The Bothriolepididae Briagalepis warreni is added to Mount Howitt fauna [19, 64]. For Gauja, Asterolepis ornata remains the only Givetian Antiarcha from their original list, as Grossilepis spinosa, Taenolepis speciose and Bothriolepis curonica are dated as middle Frasnian [10]. Vorobjevaia dolonodon Young et al., 1992 [136] is added to the Megalichthyiformes of the Givetian Antarctic Aztec Siltstone [136, 137]. Sarcopterygii are slightly reorganised from the original analysis, with Sallan and Coates’ ‘Tetrapoda’ being now divided into Elpistostegidae and Tetrapoda s.s. We have expanded this study at the family level for Antiarcha, and special attention to the genus Bothriolepis is emphasised at each level (Figs 16 and 17). Pie-diagrams corresponding to relative proportions for each group are then plotted on a palaeomap of the Givetian [38]. Relative ages of localities (early, middle or late Givetian) are indicated by the relative distance of the pies from the edge of the map.

A brief characterisation of Chinese, Gondwanan and Euramerican Givetian vertebrate assemblages.

A few immediate distinctions can be made between Gondwanan, Euramerican and Chinese faunal assemblages during the Givetian period. Gondwanan localities exhibit the most important proportion of Bothriolepis species compared to all other taxa (at least 20%; Mount Howitt, Victoria, Antarctica and Aztec, Antarctica; Fig 16; see S5 Table). Besides, until our current study, Bothriolepis was an exclusively Gondwanan and Chinese taxon for early and middle Givetian (Aztec, Mount Howitt, Haikou; Fig 16; see also other localities in S5 Fig in S1 File). The Aztec Siltstone of Antarctica is by far the richest formation in term of species number of Bothriolepis (9 species; Fig 17) and also contains the Rhizodontida Aztecia [138]; it should however be reminded that all species were not strictly contemporaneous nor from the same exact locality.

Valentia Slate macro-vertebrate assemblage.

Because of the important palaeogeographic significance of this taxon, an update of the Valentia Slate macrovertebrate assemblage originally given by Russell [4] should be given. Russell recognised the presence of two taxa: 1) Bothriolepis sp. (Placodermi, Antiarcha, diverse anatomy and the taxon yielding by far the most abundant specimens; [4]:figs 3–6, 11), and 2) one indeterminate Acanthodii (fin spine; [4]:fig. 9). Sauripterus (Sarcopterygii, dermal scales; [4]:figs 7–8) come from the overlying Frasnian St Finan’s Formation and belong in fact to Holoptychius; these are not taken further into consideration in this article.

A complete review of the material published by Russell reveals that the supposed acanthodian fin spine is actually a rhizodont coronoid fang (NHM P 59686; [4]:fig. 9), probably closely related to the Gondwanan genus Gooloogongia [138]. A dipnoan tooth plate is also identified in the matrix of the previous specimen (both identifications were made possible by CT-scanning the specimen at NHM). Both these taxa belong to the Valentia Slate formation (locality FB1 in [4]:fig. 2), and will be the subject of a separate publication.

The updated macro-vertebrate assemblage of the Givetian Valentia Slate is as follows: 1) Bothriolepis dairbhrensis sp. nov., 2) Rhizodontida indet., 3) Dipnoi indet., and 4) Tetrapoda indet. (ichnofossils only so far; Fig 16).

Comparison: Valentia slate formation fauna vs. the rest of the Givetian world.

It is noteworthy that the macrovertebrate faunal list of Valentia is very reduced compared to those of other localities listed by Sallan and Coates [96], and it would have certainly not been considered by these authors in their study (“Only macrofossil sites containing at least five named species in three groups in two divisions were eligible for inclusion in this smaller dataset”; see [96]:sup. inf. p. 9).

But despite this overall lower diversity, the presence of the antiarch Bothriolepis and of a rhizondont in the middle Givetian Valentia Slate is unique enough in Givetian Euramerica to invite for a comparison between those different localities.

The Valentia Slate Formation assemblage reminds the situation in the Mount Howitt and Aztec Siltstone faunas (these three localities hold the highest proportions of Bothriolepis compared to all other tatxa or only Antiarcha: Bothriolepis in Valentia corresponds to 25%_total, 100%_Antiarcha; Aztec Siltstones has 32.9%_total and 92%_Antiarcha; Bothriolepis in Mount Howit corresponds to 20%_total and 75%_Antiarcha; by comparison, Bothriolepis composes 9.1%_total and 25%A_ntiarcha in Haikou, and of course 0%_{total/Antiarcha} in Givetian Euramerican localities; Figs 16 and 17). It should be noted that the Aztec Siltstone covers six biozones, possibly ranging from the Eifelian to early Frasnian; comparisons should hence remain careful with this locality (J. A. L., pers. obs.). The presence of Bothriolepis in the Euramerican Valentia Slate of Ireland is thus very reminiscent of a Gondwanan fauna. This is emphasised with the presence of an indeterminate rhizodont in Valentia Slate, recalling the occurrence of Aztecia in the Givetian of Antarctica and of Gooloogongia in the Famennian of Canowindra, New South Wales, Australia; [138, 139]), while Rhizodontida were hitherto unknown in Euramerica.

Likewise, the faunal composition of the Valentia Slate Formation differs strikingly from that of sub-contemporaneous Euramerican localities. Bothriolepis is the only placoderm encountered in Valentia, and whereas this genus is absent in other localities. Beyond a possible collection bias, it should be noted that none of the hundreds of collected specimens could be attributed to other groups of placoderms. The contrast is even sharper with the localities on western side of the Caledonian–Appalachian Mountains, which are devoid of antiarchs and actinopterygians, but with over 50% of placoderms, a feature shared with Gerolstein in Germany, but also Haikou in the South China Palaeoblock and the Aztec Siltstones.

The fauna in the Valentia Slate Formation also differs from other well-known Givetian localities as it lacks (so far) both actinopterygians and acanthodians, a situation also encountered in Haikou and Cedar Valley, while Gerolstein and Rockport, although lacking actinopterygians, possess acanthodians.

The early Givetian Orcadian Lake in Scotland, U.K., and Harma in Estonia as well as the late Givetian Abava and Gauja in Latvia and Estonia share a likelihood in the faunal composition by possessing a small amount of actinopterygians and chondrichthyans, between 20% and 50% of acanthodians or sarcopterygians.

The strong Gondwanan faunal affinity of the Valentia slate contrasts with its Euramerican location as well as its early Givetian stratigraphic occurrence. We thus hypothesise that a cluster of Gondwanan vertebrates dispersed northward as soon as the early Givetian in Euramerica. It is noteworthy that not all taxa of this cluster remained in Euramerica and dispersed throughout the northern supercontinent. Indeed, our phylogenetic analysis seems to point toward a common ancestor of the Irish Bothriolepis with B. gigantea from Scotland, and further Euramerican occurrences of Bothriolepis are related to later dispersal waves.

Bothriolepis was likely a poor demersal swimmer, unable to cross large oceanic masses. It was likely restricted to freshwater environments and continental platform (following Young and Moody’s assumptions of dispersal processes of Bothriolepis; [103]:163). Besides, the Gondwanan terrane was likely already closer to Euramerica than previously suggested. Such early connections between the two super continents may not be the most ancient, as evidenced by other earlier vertebrate and invertebrate proxies (for a Lochkovian proximity between the two supercontinents and their related terranes, see [140–142]). Besides, it is worth mentioning that Scotese figured a land connection in the West between Euramerica and Gondwana (Figs 16 and 17; [38]).