Evidence for high-performance suction feeding in the Pennsylvanian stem-group holocephalan Iniopera

Significance Suction is an especially effective way of feeding underwater, and adaptations to enhance it have evolved numerous times in jawed vertebrates. The only major living jawed vertebrate group including no specialist suction feeders is chimaeras, a handful of anatomically conservative fish species that feed on hard-shelled prey. Contrastingly, in the Carboniferous (359 to 299 Ma), diverse chimaeras formed a prominent part of aquatic ecosystems. Here, we use three-dimensional–preserved fossils of one of these Carboniferous chimaeras to reconstruct its cranial muscles and argue that it had the forward-facing mouth and expandable pharynx characteristic of high-performance suction feeders. This suggests that in the Carboniferous, some chimaeras were suction feeders in the water column, an ecological niche since monopolized by neopterygian fishes.


Details of skeletal 3D model
We built a model to approximate the range of movement of the pharyngeal and pectoral skeleton of Iniopera, using the 3D models described in the Methods (Figs 1, S1-S2). For detailed anatomical descriptions of components see [1][2][3][4] . The models were those used for the reconstruction of Pradel et al. 4 but with the following changes in positioning made compared to their reconstruction. Two specimens were used, KUVP 22060 and KUVP 158289. Of these, KUVP 158289 has in previous publications been referred to as "21894" [1][2][3][4] ; the specimen number has subsequently been revised due to a specimen number duplication in the KUVP collections.
The neurocranium, mandible, and upper tooth whorls are from KUNHM 22060 1 (figs 1-13, 31-32). Lower tooth whorls are missing from the specimen but are assumed to have had a similar morphology based on the mandible's shape and flattened Iniopera specimens 5 . Upper tooth whorls positions were estimated based on comparison to other Iniopera specimens 1,5 . A break in the mandible that affected its alignment was fixed by cutting along the line of the break, rotating the two parts to fit and reattaching them in Blender; it was then remeshed (Fig. S1j-g).
The mandible was fitted into articulation with the neurocranium; its resting position was approximated by copying models of the upper tooth whorls onto the lower jaw and closing the jaw until they made contact (Fig. S3a). The neurocranium and mandible were both remeshed in Blender to provide smaller more manageable models.
The intercoracoid, scapulocoracoid, suprascapular elements, and fin elements are from KUNHM 158289 2 (figs 1+2). Paired elements (scapulocoracoid, fin skeleton, and suprascapular) are from the right-hand side of the fossil and are mirrored across the centre of our model. Like with the mandible, a break in the scapulocoracoid was that affected its alignment was fixed by cutting along the line of the break, rotating the two parts to fit and reattaching them in Blender; it was then remeshed (Fig. S1k-n).
Additional visceral cartilages are taken from both KUNHM 22060basihyal, basibranchial, ceratohyal 4 ( fig. 1)-and KUNHM 158289-ceratobranchials 4 (fig. 2). Ceratohyals were articulated with corresponding facets on the basibranchial, their exact angle and position is unknown as is the presence of an epihyal/hyomandibula (Figs 1, S1). Unlike extant holocephalans the ceratohyals seem to have lain out of sequence of the branchial arches, and it is unknown whether the epihyal articulated with the braincase, although the lack of an obvious articulation surface makes it unlikely 1 . Ceratobranchials were used to inform the approximate shape and size of the orobranchial cavity (Figs 4, S6) but their exact placement is conjectural, and the presence or absence of hypo-, epi-, and pharyngobranchials is unknown 4 .
The exact placement of neurocranial, visceral, and pectoral elements relative to one another is necessarily subjective, but preserved positions and the articulations between elements were used as a guide to their relative positions.

Muscle placement justifications
Detailed justification for reconstructions of muscles related to high performance suction feeding in Iniopera follows. Muscles in Iniopera are reconstructed using the extant phylogenetic bracket 6 with particular reference to holocephalans and elasmobranchs.

M. adductor mandibulae.
This muscle is present in both holocephalans and elasmobranchs, and is innervated by the maxillary branch of the trigeminal (V) nerve 7,8 . In holocephalans the muscle has a broad preorbital origin on the neurocranium, and inserts on a sheet of connective tissue slung around the bottom of Meckel's cartilage (Figs 1b, S2f) 8,9 . In elasmobranchs the muscle has its origin on the palatoquadrate and inserts directly on the posterior part of the lateral face of Meckel's cartilage 9,10 .
Iniopera has a holostylic neurocranium and so this muscle must have had its origin on the neurocranium as in living holocephalans. Iniopera lacks the high-walled lamina orbitonasalis present in crown-group holocephalans but there is a large preorbital fossa, ventral to the nasal capsules, which we consider to a probable position for the mandibular adductor's antorbital origin (Figs 1a, S1d, S2a). A foramen in the wall of this fossa is interpreted by Pradel et al. 1 as providing passage for the maxillary branch of the trigeminal nerve 1 (fig. 6B, fVmx), further supporting this interpretation. There also seems likely to have been an insertion in the bottom of the orbit, as suggested by Dearden et al. 9 . We have reconstructed the muscle with these two origins, analogous to the suborbital and antorbital mandibular adductor muscles in living holocephalans 9 . Pradel et al 1 suggest instead that the origin of the mandibular adductor was on the ventral part of the postorbital wall: this seems unlikely as the muscle would have occluded the orbit.
The insertion of the muscle was on the lateral side of Meckel's cartilage (Figs 1ª, S1d, S2a,c). Notably the cartilage is demarcated by a ventral rim meaning that the muscle must have inserted on the cartilage itself as in crown-group elasmobranchs, rather than on a ventral sling as in crown-group holocephalans 8,9,11 .

Mm. anguli oris, M. labialis anterior, M. intermandibularis
In crown-group holocephalans, these muscles insert on the labial cartilages. The m. anguli oris anterior and the m. anguli oris posterior have preorbital origins and attach on the labial cartilages, the m. labialis anterior muscle connects certain labial cartilages, and the m. intermandibularis comprises two sections which link certain labial cartilages to the Meckelian cartilage; all are innervated by the trigeminal (V) nerve 8,9 . None of these muscles are present in crown-group elasmobranchs 9 .
In Iniopera there is no evidence for labial cartilages having been present 1 and no evidence for the presence of these muscles.

M. superficialis
This superficial muscle, innervated by the trigeminal (V) nerve, has only been reported in Callorhinchus among crown-holocephalans 8,9 and its presence in Iniopera is unknowable from the preserved remains.

M. mandibulohyoideus
In crown-group holocephalans the m. mandibulohyoideus has an origin on the posterior of the Meckelian symphysis and inserts on the ventral angle of the ceratohyal (Fig. 1f, S2g) 8,9,12 . It is innervated by the facial (VII) nerve as well as the glossopharyngeal (IX) nerve in Hydrolagus 8,9,12 . This muscle is absent in elasmobranchs and is likely apomorphic for total-group holocephalans 12 .
Iniopera lacks the insertion area on the postero-ventral part of the mandibular symphysis onto which this muscle has its origin in Callorhinchus (Fig. S2h) 9 . However, the ceratohyal does possess a large ventral angle similar to that which the muscle inserts on in living holocephalans (Fig. 1a, S1b) and which is absent in elasmobranch ceratohyals 4,9 . This muscle is the only candidate from the muscles of living chondrichthyans for attachment in this position, and seems likely to have been present in Iniopera (Fig. 1a,e). We consider it likely that it was present but it seems unlikely that it would have attached to the mandibular symphysis as in living holocephalans: its path is obstructed by the basihyal, instead it may have attached on the medial face of the Meckelian cartilage.

M. epaxialis
In living holocephalans the epaxial muscles attach over a large area of the dorsal surface of the neurocranium, on either side of the endolymphatic foramen (Figs 1b, S2f). In elasmobranchs these muscles attach in a similar position 8,9 . They are innervated by spinal nerves 7 .
In Iniopera fossae are present on either side of the endolymphatic foramen and occipital crest which where the epaxial muscles would have inserted (Figs 1a, S1e,f, S2a-d) 1 (fig. 6, pf). Compared to Callorhinchus the insertion area for these muscles on the neurocranium is limited 9 . Foraminae are present in the occipital region of Iniopera which may have carried spino-occipital nerves to innervate the epaxial musculature 1 ( fig. 21, so, bso). The lateral extent of the epaxial muscles would have been limited by the shoulder girdle.

M. protractor dorsalis pectoralis
In crown-group holocephalans the m. protractor dorsalis pectoralis has its origin posterior to the orbit, ventral to the lateral otic ridge and inserts on the anterior edge and medial side of the upper scapular process 9 where it meets the m. retractor dorsalis pectoralis (Figs 1a, s!b, S2a,c,d,e). The m. protractor dorsalis pectoralis is probably a trunk muscle, innervated by the nerves IX and X 13,14 . This muscle is absent in elasmobranchs.  fig. 7, ld). Pradel et al suggest that these probably accommodated the "jaw and visceral musculature (e.g. the cucullaris profundus, subspinalis, and levator hyoideus muscles) and gills. This is probably the case given their position relative to the branchial skeleton in our construction (see below). However, these muscles in living holocephalans are collectively fairly small. Their size and the large fossae that face them on the scapulocoracoid (Fig. S2e) suggest that they mainly served as origins for muscles linking the neurocranium and shoulder girdle. The closest analogue of this in living chondrichthyans is the m. protractor dorsalis pectoralis and it seems likely that a similar muscle joined the scapulocoracoid and neurocranium in Iniopera. In all figures we have termed this muscle m. protractor dorsalis pectoralis, but the exact homology of the muscle in this position is unclear.

M. retractor dorsalis pectoralis
In living holocephalans the m. retractor dorsalis pectoralis has its origin in the trunk musculature and inserts on the medial face of the upper scapulocoracoid (Figs 1b, S2g,h,i) 9 . As with the m. protractor dorsalis pectoralis it is probably a trunk muscle 13,14 and is absent in elasmobranchs. It is innervated by the spinal nerves.
The large areas on the posterior face of the scapulocoracoid of Iniopera suggest a large area for insertion of posteriorly oriented musculature (Figs 1a, S1f,e, S2e), which may have been homologous to the m. retractor dorsalis pectoralis.

M. retractor mesio-ventralis pectoralis
In living holocephalans m. retractor mesio-ventralis pectoralis is a part of the trunk musculature that inserts on the ventral part of the scapulocoracoid (Figs 1b,f, S2g) 8,9 . In elasmobranchs hypaxial muscles insert posteriorly on the scapulocoracoid 15 . It is innervated by the spinal nerves.
Iniopera has large, posteriorly facing surfaces on the rear of the scapulocoracoids, which likely served as insertions surfaces for some kind of ventral musculature (Fig. 1a,e, S1f, S2b). Whether this was a distinct muscle or not is unclear, but for ease of comparison with Callorhinchus we have labelled these as m. retractor mesio-ventralis pectoralis in our figures.

M. cucullaris superficialis
In living holocephalans the m. cucullaris superficialis overlies the m. protractor dorsalis pectoralis with an origin on the postorbital crest (Figs 1b, S2f,i) 8,9 , and inserts on the lateral face of the scapulocoracoid. It is innervated by nerves IX and X. It is absent in elasmobranchs.
In Iniopera the postorbital wall 1 (fig. 6B, pow) is divided into an upper and a lower part by a ridge 1 ( fig. 6B, imz). Pradel et al. 1 suggest that the upper part may have been the origin of a m. cucullaris superficialis: this seems likely (Figs 1a, S1d,e, S2a). If so it would have inserted on the lateral part of the scapulocoracoid (Figs 1a, S2a,e). It seems likely that a separate m. cucullaris profundus and m. cucullaris superficialis like living holocephalans were present due to the subcranial pharynx and this candidate origin site.

M. cucullaris profundus/m. subspinalis
In living holocephalans the m. cucullaris profundus has its origin on the neurocranial floor and inserts on the posterior pharyngobranchial complex (Fig. S2h) and is innervated by nerves IX and X 8,9 . The m. subspinalis has its origin medial to this and inserts on the first two pharyngobranchials (Fig. S2h) and is innervated by spinal nerves and cranial nerve X. In elasmobranchs both of these muscles are also present. The m. cucullaris profundus has its origin posteriorly to the neurocranium and inserts on the posterior pharyngobranchial complex and the scapulocoracoid. The m. subspinalis has its origin on the back of the neurocranium and inserts on pharyngobranchial I.
These muscles are present in living elasmobranchs and holocephalans and so were likely present in Iniopera. For the reasons outlined above, we consider it likely that Iniopera had a separate m. cucullaris profundus and m. cucullaris superficialis like living holocephalans. However, no muscle attachments are visible, and the pharyngobranchials are not preserved. As previously suggested 1 these muscles presumably both inserted somewhere in the large postorbital fossae with the protractor dorsalis pectoralis equivalent as the only possible insertion point place dorsal to the branchial skeleton (see above).

M. levator hyoideus
In living holocephalans the m. levator hyoideus has its origin on the basicranium and inserts on the epihyal 8,9 . It is innervated by cranial nerve VII. In elasmobranchs this muscle is present but incorporated into the dorsal hyoid constrictor muscle to varying degrees 10 .
Posterior to the orbits in Iniopera there is a small scar ( fig. 10, mi of 1 ) which could feasibly have been the origin point for this muscle. Another possibility is that it as the articulation of the hyoid arch, although the hyomandibula and whether it articulated with the neurocranium is unknown in Iniopera and

M. coracomandibularis
In living holocephalans the m. coracomandibularis inserts on the posterior edge of the mandible and has its origin on the coracoid and via two arms, on the bases of the scapular processes (Figs 1b, 1f, S2f-i) 8,9 . It is innervated by the spinal nerves. It is also present in elasmobranchs in which it is a proportionally thinner muscle which also extends from the mandible to the coracoid.
In Iniopera this muscle presumably inserted on or along the bottom of the mandible, although it lacks the attachment area present in living holocephalans (Figs 1a,e, S1d, S2a,d,e). Iniopera also lacks the fused coracoid of living holocephalans. The origin of the m. coracomandibularis may have been on some combination of the bottom of the intercoracoid element, and may also have lain on the intercoracoid element, where there is a plausible attachment surface.

M. coracohyoideus
In living holocephalans the origin of the m. coracohyoideus is on top of the m. coracomandibularis (Figs 1b,f, S2h). It then inserts on the posterior face of the basihyal 8,9 . It is innervated by the spinal nerves. It is also present in elasmobranchs, with the same insertion and origin points.
In Iniopera they probably inserted on the much larger attachment surface along the ventral posterior edge of the basihyal (Figs 1a,e, S1d, S2d,e). The origin was possibly on the anterior face of the scapulocoracoid, or on the dorsal side of the m. coracomandibularis. Their relative size appears likely to have been much larger in Iniopera than in living holocephalans and probably played a more major role relative to the smaller m. coracomandibularis.

Mm. coracobranchiales
In living holocephalans the mm. coracobranchiales have their origins along the medial margin of the ventral scapular shaft and insert between the hypobranchials (Fig. S2f,i) 8,9 . It is innervated by the spinal nerves. In elasmobranchs they have their origin along the edges of the coracoid and insert on the hypobranchials. It seems likely in Iniopera the mm. coracobranchiales had their origins along the side of the lower scapulocoracoid (Fig. S2a,e), as the broad basibranchial covered the intercoracoid. Insertions are not known, but presumably lay on the hypobranchials.

M. constrictor operculi dorsalis and ventralis
Living holocephalans have a single m. constrictor operculi dorsalis and ventralis (Fig. S2h) 8,9 , both of which are innervated by cranial nerve VII. The m. constrictor operculi dorsalis has its origin at the base of the scapular process and on the notochord, and inserts along the rim of the operculum as well as along the bottom edge of the orbit. The m. constrictor operculi ventralis has its origin on the opercular cover and inserts ventrally. Elasmobranchs instead have a m. constrictor hyoideus dorsalis, which has its origin across the otic part of the neurocranium and epaxials and inserts on its ventral counterpart and the hyomandibula, and a m. constrictor hyoideus ventralis, which has its origin along the median aponeurosis and inserts on its dorsal counterpart 9 .
No mineralised opercular cartilage is present in Iniopera, but one may have been carried by the large ceratohyal 4 ; the subcranial pharynx suggests that it probably had a single branchial opening like living holocephalans and that these muscles were more like those of living holocephalans than elasmobranchs. It seems likely that the m. constrictor operculi dorsalis had a suborbital origin on the coracoid and on the lamina ventral to the ridge in the postorbital wall, identified by Pradel et al. 1 as a possible mandibular adductor site.

Details of the 3D pharyngeal expansion animation
The model was animated in Blender using a single armature with a different bone for each component, placed down the centre of the model (Fig. S6). Joints between bones were defined as below and were translated in the X axis to provide the locations for each armature joint. 3D models of the individual skeletal elements were linked to armature bones using the "child of" object constraint. Paired elements (e.g. the scapulocoracoid) were mirrored around a central point. A posterior bone was added to the armature which was linked to an empty using the "inverse kinematics" bone constraint. This empty then moved posteriorly to simulate the contraction of the ventral trunk muscles. An anterior bone was added to the armature to simulate the basihyal being pulled posteroventrally. This was linked to an empty using the "inverse kinematics" bone constraint. The empty was then moved postero-ventrally along a curve.
To approximate pharyngeal expansion a polyhedron mesh was created which was shaped to fit one half of the interior of the buccal cavity at a closed position (Fig. S6d). Each vertex of this mesh was assigned to a different vertex group, each of which was then "attached" with a hook modifier to an empty in the same position as each vertex. These empties were then parented to parts of the skeleton of Iniopera so that the polyhedron moved with the animated model.

Neurocranium-suprascapular joint
There is no joint between the suprascapular and the neurocranium, but based on its preserved position it may articulate with a process on the back of the braincase 2 and may have been attached by a ligament (Figs S1,S6). This places the element at the lateral edge of the epaxial muscles 'attachment to the neurocranium, and by comparison with living holocephalan musculature it seems likely that the element was embedded between the epaxial musculature and the musculature linking the braincase to the scapulocoracoid (see below), perhaps to provide firm dorsal anchorage for the shoulder girdles movements. In our model we have enabled rotation between the neurocranium and suprascapular but have stiffened the joint quite substantially to simulate its likely embedding in muscle.

Suprascapular-scapulocoracoid
Again, there is no obvious joint between the suprascapular and scapulocoracoid, but based on its preserved position it seems to have overlain the top of the scapulocoracoid (Figs S1, S6) 2 . Again, it may have been attached via a ligament. In our model we have enabled rotation between the suprascapular and the top of the scapulocoracoid but have stiffened the joint quite substantially to simulate its likely embedding in muscle.

Scapulocoracoid-Intercoracoid
The intercoracoid and scapulocoracoids have a joint formed by paired rounded bulges on the posterior edge of the intercoracoid that fit into sockets in the upper face of the scapulocoracoids (Figs S1, S6) 2 . It seems likely these would have permitted quite free movement both dorsoventrally and laterally. In our model this is represented by free rotation and no stiffness modifier.

Intercoracoid-Basibranchial
A laterally elongated facet on the front of the intercoracoid meets a rounded ridge on the ventroposterior face of the basibranchial (Figs S1,S6). This seems likely to have permitted some limited rotation around the transverse axis. This is represented in our model by a stiff joint that can rotate around this axis

Basibranchial-basihyal
The basibranchial has two anterior processes which match facets on the posterior side of the basihyal (Figs S1, S6). This would have permitted some rotation around the transverse axis. This is represented in our model by a stiff joint that can rotate around this axis

Basihyal-Mandible
There is no articulation between the basihyal and the mandible (Figs S1,S6). However, the shape of the basihyal closely fits the shape of the mandible and its large size suggests that it probably acted somewhat like the 'tongue 'formed by the basihyal in living elasmobranchs and would have had limited movement relative to the mandible being embedded in the pharyngeal musculature. This is represented in our model by its movement following a curve posterior to the Meckelian cartilage.

Supplementary figures
Movie S2. Animation of the gape opening in Iniopera with adductor muscles modelled as cylinders, following method of Lautenschlager (29), starting at resting position of 6 degrees.
Movie S3. Animation of the gape opening in Iniopera with adductor muscles modelled as cylinders, following method of Lautenschlager (29), starting at resting position of 9 degrees.
Movie S4. Animation of reconstructed pharyngeal expansion in Iniopera, with the mouth opening to the optimal extension of mandibular adductor muscles.
Movie S5. Animation of reconstructed pharyngeal expansion in Iniopera, with the mouth opening to the maximum extension of mandibular adductor muscles.