The Middle Cambrian lophotrochozoans Odontogriphus omalus and Wiwaxia corrugata have been interpreted as stem-group members of either the Mollusca, the Annelida, or a group containing Mollusca + Annelida. The case for each classification rests on the organisms' unusual mouthparts, whose two to three tooth-rows resemble both the molluscan radula and the jaws of certain annelid worms. Despite their potential significance, these mouthparts have not previously been described in detail. This study examined the feeding apparatuses of over 300 specimens from the 505-million-year-old Burgess Shale, many of which were studied for the first time. Rather than denticulate plates, each tooth row comprises a single axial tooth that is flanked on each side by eight to 16 separate shoehorn-shaped teeth. Tooth rows sat on a grooved basal tongue, and two large lobes flanked the apparatus. New observations—the shape, distribution and articulation of the individual teeth, and the mouthparts' mode of growth—are incompatible with an annelid interpretation, instead supporting a classification in Mollusca. The ancestral molluscan radula is best reconstructed as unipartite with a symmetrical medial tooth, and Odontogriphus and Wiwaxia as grazing deposit-feeders.
Although the molluscs represent the second-largest animal phylum and have a rich shelly fossil record, their early evolutionary history remains contentious. The soft-bodied organisms Odontogriphus omalus and Wiwaxia corrugata from the Middle Cambrian Burgess Shale (Series 3—505 Ma) seem to represent early members of this group [1,2]; if so, they provide our best constraint on the ancestral mollusc. However, it has been suggested that Wiwaxia is more closely related to the annelid worms [3–5], and that Odontogriphus could have diverged from a lineage ancestral to both the molluscs and the annelids [4,5]. The fossils' mouthparts, which have been interpreted as a radula (the molluscan feeding apparatus) or as an annelid jaw [1,2,5–8], are pivotal to the classification—but have not been described in detail.
The mouthparts of Wiwaxia are understood to represent a series of flexible denticulate plates, with either one [1,9] or two [3,6] plates per row (depending on interpretation). They closely resemble the better-understood Odontogriphus feeding apparatus. According to Caron et al. [2,7], this consists of two (occasionally three) bilaterally symmetrical tooth rows, with the posterior row usually wider than the anterior. The teeth, which have a consistent relative position even when isolated from the body, are held to have been embedded in or on a non-fossilized radular membrane. Caron et al. deduce that this membrane sat upon a supporting apparatus or ‘tongue’; teeth occupied its dorsal surface and sometimes passed round to the underside, inverting as they did so. Caron et al. go on to interpret the mouthparts as bipartite and distichous: that is, comprising just two denticulate teeth in each symmetrical row, with the radular membrane divided along its centre. The posterior genesis of new tooth rows is inferred from the faintness of the third tooth row, where present. It is not clear whether adjacent rows were the same size; nor whether denticle shape varied between rows.
The present study supplements previously examined material with new specimens collected by the Royal Ontario Museum. One hundred and seventy Odontogriphus specimens were examined, of which 155 bore mouthparts; 10 further Odontogriphus mouthparts lacked accompanying body tissue. Four hundred and seventy-six Wiwaxia specimens were examined, 140 with clear mouthparts, and one isolated Wiwaxia apparatus was identified. Backscatter electron micrographs were acquired under environmental pressure , complementing traditional light microscopy and digital interference of images obtained under plane-polarized and cross-polarized light . This high-resolution imaging allows a fundamentally new reconstruction of the mouthparts, invalidating their interpretation as an annelid jaw and instead revealing striking similarities with the radula—bringing a new perspective to the origin of this defining molluscan synapomorphy.
The mouthparts are primarily preserved as a non-mineralized film of carbon silhouetted by trace amounts of Ca and P (figure 1). The regions corresponding to the tooth rows are depleted in Si and show raised concentrations of Al and K: an elemental signature that is commonly associated with voids [12,13]. Along with the thickness of the carbon film, this suggests that the teeth were robust enough to resist compaction—accounting for their heightened relief and their tendency to interrupt the plane of splitting.
Each of the mouthparts' ‘denticulate plates’ actually comprises an array of individual articulating teeth (figure 2). Except for a bilaterally symmetrical central tooth in each row, teeth within the same taxon have a consistent shape. Each tooth is shoehorn-shaped, tapering from a pointed root to a broad, flattened scoop. The root sits at the front of each tooth, pointing towards the apex of the chevron (figure 2b). The scoops lie either parallel or sub-perpendicular to the bedding plane (figure 2a,b), indicating that they originally sat at an oblique angle to the tongue. Marginal thickenings (figure 2b,c) emphasize the slight transverse curvature of each tooth. The ratio of scoop width to root length is around 1 : 1 in Odontogriphus but closer to 2 : 1 in Wiwaxia.
(c) Tooth row disposition
All the examined Odontogriphus and Wiwaxia specimens bear two tooth rows; some possess a third. This additional posterior row resembles the others, although it is often fainter: probably reflecting its weaker carbonization, its preservation at a different level within the rock, or a combination thereof (figure 3a). The total number of tooth rows (two or three) is unrelated to size (binary logistic regression against log(apparatus width): adjusted R² = 0.004, p = 0.28, n = 44). In Odontogriphus the rows are separated by 1.7 ± 0.7 times the width of the widest tooth (n = 18), whereas in Wiwaxia, tooth rows are immediately adjacent (separation = 0.18 ± 0.16 times maximum tooth width, n = 14).
In each Odontogriphus tooth row (figures 2 and 3), the central tooth is flanked by nine to 11 medial teeth—of which the abaxial six or seven are substantially larger—and sometimes a further one or two diminutive lateral teeth (figure 2b,c). When the mouthparts are in their resting position, teeth form anteriad-directed chevrons with a basal angle of 34 ± 6° (irrespective of specimen size; n = 40). In their active position, where chevrons have passed around the end of the supporting apparatus and onto its dorsal surface (figure 2c), their angle is steeper (53 ± 6°, n = 8; two-tailed t-test p < 10−4). This reconfiguration would not occur if the supporting apparatus was flat: a medial groove, representing about 20 per cent of the supporting apparatus's depth, must have been present. This caused each tooth to rotate relative to its neighbours as it rounded the end of the supporting apparatus (figure 3b). The lateral teeth, like a racer in an outside lane, had further to travel; they would fall back as the chevron rounded the support. This produces the anteriad decrease in total chevron width measured by Caron et al. , but does not affect the along-chevron distance from the central tooth to the most distal; this measurement, being independent of chevron angle, is not significantly different between the two posterior rows (paired t-test, p = 0.40, d.f. = 14; data log-transformed). Measured this way, the size of adjacent rows is identical (slope not significantly different from 1, Pearson correlation p = 0.22; constrained to pass through origin, d.f. = 12, adjusted R² = 0.9996).
Wiwaxia's tooth rows (figure 4) differ from those of Odontogriphus in a number of respects. Each row consists of a central tooth, flanked on each side by six to 11 medial teeth; larger tooth-rows bear more teeth (adjusted R² = 0.76; p ≪ 0.001). Three to five diminutive lateral teeth are sometimes preserved on each side. Rather than chevrons, teeth form an open-based isosceles trapezium. In the resting position, the base angle measures 33° to 61° (mean = 44 ± 9°, n = 26; angle significantly smaller than Odontogriphus, with two-tailed Mann–Whitney p < 10−4). Smaller specimens are more steeply angled (exponent = 0.13 ± 0.05, p = 0.02, adjusted R² = 0.17), indicating a loose ontogenetic trend towards straighter tooth rows. Only three Wiwaxia specimens preserve tooth rows in the active position; once round the end of the supporting apparatus, the top of the trapezoid remains straight as the legs become more steeply angled—denoting a broad medial groove that represents 30 per cent of the support's thickness. As in two specimens of Odontogriphus , partial tooth rows (recognizable by their morphology, size and composition) are present in the guts of two Wiwaxia specimens (figure 5). This indicates that mouthparts were sloughed, swallowed and—judging by their posterior position in the gut—indigestible.
(d) Mouthpart lobes
Some Odontogriphus specimens preserve an internal lobe of resistant tissue on each side of the tooth rows (figure 6a–c), usually as aluminosilicate films that silhouette the original carbon. In some cases, original carbon is still present and elicits a backscatter SEM response, suggesting that the structure's original constitution was similar to that of the teeth. The outer margins of the lobes, which are darker and more pronounced, were presumably their toughest or thickest parts; these become progressively more labile towards the rear. The lobes emanate from the front of the support and arch dorsally towards its posterior; their attachment was sufficiently robust to survive post-mortem displacement of the feeding apparatus. The lobes do not invert when tooth rows pass round the end of the supporting apparatus, indicating a fixed position relative to mobile tooth rows. The lobes were originally inclined at a small angle to the vertical: in most cases, they are compressed into a butterfly-shape, but sometimes both are flattened in the same direction. Posteriad of the mouthparts, the margins of the lobes come together and connect to a chevron-shaped feature that is similar in angle and width to the tooth rows, but always lacks teeth or other projections (figure 6a,b). Given their form, ventro-lateral position and detachment from the tooth rows, the lobes are interpreted as attachment rather than as masticatory structures.
This study confirms that the Odontogriphus and Wiwaxia feeding apparatuses (figure 7) comprise two to three equally sized, self-similar rows. Significant extracellular secretion is indicated by the residual three-dimensionality of teeth . As iterated by Caron et al. [2,7], the mouthparts comprise symmetrical transverse rows of solid teeth that are distinct from, but embedded in, a radular membrane; the rows pass round the end of a supporting apparatus, are sloughed anteriorly and replaced from the posterior, and contain more teeth when they are larger. But where Caron et al. identified two denticulate teeth per row [2,7], this study recognizes each ‘denticle’ as a separate articulating tooth. The robust lobes preserved in Odontogriphus attach to the radula, have thickened margins, become increasingly labile towards the rear and are constructed from the same material (presumably chitin) as the radular teeth; they presumably represent a muscle attachment structure. The supporting apparatus bore a medial groove, causing chevron angle to increase and teeth to rotate as they round its cusp. Chevrons are uniformly sized and directed anteriad while the supporting apparatus is in its resting position; all teeth except the symmetrical central tooth have a uniform shape (even though their silhouette on bedding planes varies with burial angle).
This new reconstruction is inconsistent with previous interpretations of the Odontogriphus and Wiwaxia mouthparts as annelid jaws. A comparison with the dorvilleid polychaetes was founded on the difference in morphology between adjacent tooth rows , which this study shows to be a taphonomic feature. Dorvilleids have a single planar pincer-like jaw  with teeth in a fixed position ; Odontogriphus and Wiwaxia have multiple, transverse rows containing teeth that rotate relative to one another. In dorvilleids, jaw exuviae are shed laterally and are fainter than functional rows ; in Odontogriphus and Wiwaxia, moulted rows are shed from the front of the apparatus and remain strongly carbonized. Dorvilleids replace moulted jaws with new ones that are substantially (1.4 times) larger , but each subsequent tooth row in Odontogriphus or Wiwaxia is the same size. Tooth shape changes as the dorvilleid jaw grows , but never changes in Odontogriphus or Wiwaxia. Odontogriphus and Wiwaxia have no equivalent to the permanent (non-moulted) dorvilleid mandible.
Given these barriers to a dorvilleid interpretation, might the single chevron-shaped tooth row in the ampharetid annelids be a better proxy for the Odontogriphus and Wiwaxia mouthparts ? Butterfield  no longer considers these jaws to be comparable with multi-rowed apparatuses, and indeed their construction from bilateral series of teeth on the posterior edge of a ventral bulb [18,19] is at odds with the central tooth and anterior placement of the tooth rows of Odontogriphus and Wiwaxia. Furthermore, I am aware of no ampharetid that moults its teeth.
The similarities between the fossils' mouthparts and annelid jaws are only superficial, whereas the molluscan radula bears a number of specific features  that are also present in Odontogriphus and Wiwaxia:
— a basal membrane: implied in Odontogriphus and Wiwaxia because isolated feeding apparatuses remain articulated;
— a supporting apparatus with a median groove: indicated by the flexure of rows and the rotation of teeth;
— adjacent tooth rows of similar sizes;
— deciduous teeth: tooth rows are occasionally present in the guts of both Odontogriphus and Wiwaxia, and the number of tooth rows varies from two to three with no relation to size;
— addition of new tooth rows at the posterior: evidenced by fainter, weakly carbonized tooth rows intermittently present in this position;
— significant extracellular secretion: indicated by the residual three-dimensionality of teeth ;
— a symmetrical central tooth: establishing each tooth row as a single (not bipartite) entity;
— a robust structure (alary process/hyaline shield) attached to the radula, with thickened margins, increasingly labile towards the rear, and constructed from the same material (chitin) as the radular teeth [20,21]: observed in Odontogriphus and inferred in Wiwaxia; and
— more teeth per row in larger specimens.
Although the small number of tooth rows in Odontogriphus and Wiwaxia is unusual among modern molluscs, some juvenile Polyplacophora have three (or perhaps fewer) tooth rows when they begin grazing . The strong morphological overlap between the molluscan radula and the mouthparts of Odontogriphus and Wiwaxia clearly eclipses their similarities with dorvilleid and eunicid jaws, whose convergent origin is exposed by fundamental differences in construction.
(c) Comparison with microfossils
Harvey et al.  recovered tooth-like elements from acid macerates of the early Middle Cambrian Kaili biota, some of which resemble the teeth of Odontogriphus and Wiwaxia in outline and in having thickened margins, but the lack of articulated apparatuses impedes a detailed comparison.
A resemblance can also be found with certain boot-shaped and fibrous elements from Early Cambrian radulae , which also have a fibrous (microvillar?) construction and imbricate to form a bilaterally symmetrical array. However—notwithstanding tooth morphology—these rows lack a symmetrical central tooth, are very closely spaced and have many (up to 29) teeth; so they do not merit close comparison with Odontogriphus or Wiwaxia.
Scoop-shaped teeth from the Mount Cap formation display pointed roots and thickened margins, and take a strikingly similar formation when articulated . These teeth, which have an internal fibrous microstructure and are found in association with Wiwaxia sclerites, may well have belonged to an Odontogriphus- or Wiwaxia-like organism.
(d) Radula and ecology
The diet of many molluscs is constrained by radular morphology , but radulae can also adapt to available food sources—either by plasticity  or natural selection . Accordingly, the radula provides a rich ecological signal.
Although the match between radula and foodstuff is not always exact, some patterns persist throughout the Mollusca and constrain the Odontogriphus or Wiwaxia diet. All molluscan macroalgivores have a pair of pronounced gouging teeth with which to incise their food ; if their foodstuff is calcified, these teeth are inevitably mineralized (although the converse relationship does not hold). Long, thin, raking teeth are necessary to sweep up filamentous algae, and all gastropods that feed on calcified or leathery algae, or on meat, have serrated teeth. To rasp hard substrates (such as calcifying algae) requires a file-like radula with teeth in a fixed relative position. When teeth rotate relative to their neighbours, as in Odontogriphus and Wiwaxia, the radular function is generally limited to sweeping food from a surface, abrading soft tissue or excavating sediment [28,29].
Radulae that comprise morphologically uniform, unornamented, shoehorn-shaped teeth (figure 8) are today found in particle-feeding gastropods, some extant monoplacophorans (Neopilina) and chitons—although in the latter case the marginal and central teeth are reduced to plates. Neopilina and chitons are a promising ecological analogue to Wiwaxia and Odontogriphus. Some chitons feed on sponges , but sponge spicules (although indigestible) are never preserved in the Odontogriphus or Wiwaxia digestive tract. Most chitons, like Neopilina, are deposit feeders [32,33]—reflected by the substantial radular musculature, a limited number of teeth (less than 20) per row and teeth that rotate relative to their neighbours . These traits are shared by Odontogriphus and Wiwaxia (the musculature, at least in Odontogriphus, indicated by the large alary process): this substantiates previous speculation  that the organisms were deposit feeders.
(e) Ancestral molluscan radula
Even relatively minor changes to a species' radula can have dramatic ecological consequences , and the rise of radula-driven herbivory likely contributed to the Cambrian substrate revolution. Unfortunately, the organ's origin is ill-constrained. The first fossil evidence of a radula-like organ comes from scrape-marks left by the putative Ediacaran mollusc Kimberella [35,36], but radulae themselves are rarely fossilized. The oldest fossil radulae (dating to the Early Cambrian) already share derived characters associated with aplacophorans and gastropods . Like those of Odontogriphus and Wiwaxia, these resemble flexoglossate radulae with independently rotating teeth—contrasting with the parallel grooves scratched by Kimberella , which denote stereoglossy .
The unsettled status of molluscan phylogeny means that extant taxa do little to constrain the ancestral radula. Wingstrand, who interpreted the radulae of chitons and extant monoplacophorans to represent the ancestral format , postulated that paired fluid-filled vessels created a groove in the radular support (resulting in flexoglossy). He imagined a small number of teeth on each row: a small medial tooth flanked by larger hooked teeth, then a comb-shaped tooth and diminutive marginal teeth . But if the Aplacophora are interpreted as basal, a bipartite and distichous reconstruction results, with no median tooth or tooth rotation . Recent analyses generally recover a derived position for aplacophorans [38–43], and molecular phylogenies imply that the ancestral radula bore multiple teeth per row . Taking Odontogriphus and Wiwaxia to be more basal than the Aplacophora, the ancestral mollusc bore a functionally flexoglossate, unipartite radula with multiple regions of distinct teeth and a symmetrical central tooth.
I thank Jean-Bernard Caron for discussion and guidance, and Nicholas Butterfield for reading an earlier version of this manuscript. Parks Canada provided research and collection permits to Royal Ontario Museum teams led by Desmond Collins, and Douglas Erwin facilitated access to specimens at the Smithsonian Institution, National Museum of Natural History (USNM). Peter Fenton, Rachel Thorpe, Mark Florence and Gene Hunt assisted with collections; Scott Whittaker and Sharon Lackie with scanning electron microscopy; and Tomohiko Kawamura provided photographs of modern taxa. This research was undertaken as part of a PhD thesis in the Department of Ecology and Evolutionary Biology, University of Toronto, and was funded through a Geological Society of America research grant and University of Toronto fellowships to M.R.S. and a Natural Sciences and Engineering Research Council of Canada Discovery Grant through Jean-Bernard Caron. This is Royal Ontario Museum Burgess Shale Research Project 37.
- Received July 11, 2012.
- Accepted July 31, 2012.
- This journal is © 2012 The Royal Society