Specialized specialists and the narrow niche fallacy: a tale of scale-feeding fishes

Although rare within the context of 30 000 species of extant fishes, scale-feeding as an ecological strategy has evolved repeatedly across the teleost tree of life. Scale-feeding (lepidophagous) fishes are diverse in terms of their ecology, behaviour, and specialized morphologies for grazing on scales and mucus of sympatric species. Despite this diversity, the underlying ontogenetic changes in functional and biomechanical properties of associated feeding morphologies in lepidophagous fishes are less understood. We examined the ontogeny of feeding mechanics in two evolutionary lineages of scale-feeding fishes: Roeboides, a characin, and Catoprion, a piranha. We compare these two scale-feeding taxa with their nearest, non-lepidophagous taxa to identify traits held in common among scale-feeding fishes. We use a combination of micro-computed tomography scanning and iodine staining to measure biomechanical predictors of feeding behaviour such as tooth shape, jaw lever mechanics and jaw musculature. We recover a stark contrast between the feeding morphology of scale-feeding and non-scale-feeding taxa, with lepidophagous fishes displaying some paedomorphic characters through to adulthood. Few traits are shared between lepidophagous characins and piranhas, except for their highly-modified, stout dentition. Given such variability in development, morphology and behaviour, ecological diversity within lepidophagous fishes has been underestimated.

Although rare within the context of 30 000 species of extant fishes, scale-feeding as an ecological strategy has evolved repeatedly across the teleost tree of life. Scalefeeding (lepidophagous) fishes are diverse in terms of their ecology, behaviour, and specialized morphologies for grazing on scales and mucus of sympatric species. Despite this diversity, the underlying ontogenetic changes in functional and biomechanical properties of associated feeding morphologies in lepidophagous fishes are less understood. We examined the ontogeny of feeding mechanics in two evolutionary lineages of scale-feeding fishes: Roeboides, a characin, and Catoprion, a piranha. We compare these two scale-feeding taxa with their nearest, non-lepidophagous taxa to identify traits held in common among scale-feeding fishes. We use a combination of micro-computed tomography scanning and iodine staining to measure biomechanical predictors of feeding behaviour such as tooth shape, jaw lever mechanics and jaw musculature. We recover a stark contrast between the feeding morphology of scale-feeding and non-scale-feeding taxa, with lepidophagous fishes displaying some paedomorphic characters through to adulthood. Few traits are shared between lepidophagous characins and piranhas, except for their highly-modified, stout dentition. Given such variability in development, morphology and behaviour, ecological diversity within lepidophagous fishes has been underestimated.

Introduction
Better minds than ours have pointed out the link between feeding morphology and diet, and many of the archetypes in  Gignac & Kley [22], to visualize muscle tissue. All specimens were soaked in an aqueous solution of 3% Lugol's iodine (i.e. 0.75% I 2 and 1.5% of KI) for 12 h or until specimens were completely stained. Specimens were then patted dry and prepared for CT scanning as described above.

Functional morphology of the feeding apparatus
Reconstructed scans were converted to .dcm format and exported into the CT segmentation program, HOROS (The Horos Project, 2015 http://www.horosproject.org/). We used the 3D-MPR mode in HOROS to measure functional aspects of the cranial morphology of these fishes, including: (i) tooth aspect-ratio for comparing tooth shape and robustness; (ii) occlusional offset, an indicator of slicing or crushing jaw action; (iii) anterior (AMA) and posterior jaw mechanical advantage (PMA), measurements of jaw leverage or force transmittance to prey (electronic supplementary material, figure S1); (iv) jaw adductor muscle cross-sectional area (CSA), a proxy for muscle force generation; and (v) second moment of area of the jaws, a proxy for jaw stiffness in either vertical or lateral bending. Tooth aspect ratio was calculated as the maximum tooth height divided by its perpendicular maximum tooth width. We measured three undamaged, mature teeth that were representative of overall tooth shape. To measure occlusional offset, we first drew a line from the tip of the anterior most tooth to the posterior most tooth in the lower jaw [29]. Then we measured the orthogonal distance from that axis to the jaw joint. We measured mechanical advantage (a ratio which evaluates trade-offs between jaw leverage and jaw-closing speed), in piranhas from the jaw joint to the insertion point of the adductor mandibulae tendon [30]. However, for the characids we approximated the centre of the concave area in the mandible and used that as the insertion point as muscle-scarring was not immediately obvious. Anterior and posterior out-lever were measured from the jaw joint to the middle of the anterior and posterior tooth, respectively. After iodine-staining, we could clearly observe and segment the primary jaw adductor muscles in HOROS. We looked at the primary jaw closing muscle, the adductor mandibulae (only the alpha division), and measured the cross-sectional area of those muscles [31]. For measurements of muscle CSA, we first determined myofibre directionality, made a digital slice through the muscle perpendicular to fibre direction at the estimated centre of muscle mass, and finally measured this area with the polygon tool in HOROS. Jaw height and width were measured at two different regions (0%, just adjacent to symphysis, and 90%, at jaw joint) along the central axis of the lower jaw. For calculations of second moment of area, jaw height was considered the major axis while jaw width was considered the where a is jaw height (major axis length) and b is jaw width (minor axis length). We used size-corrected jaw height as a proxy for jaw stiffness in the PCA.

Statistical analysis
We used analysis of variance (ANOVA) to contrast gross trends in skull skeletal architecture and muscle CSA among lepidophagous and non-lepidophagous taxa over their ontogeny. We size-corrected these data by regressing each morphological trait against each fish's standard length (SL) for ANOVA, calculated the residuals, and used these values as our size-corrected morphometric measures. Since ratios and angular measurements are proportions, and therefore naturally size-corrected, we did not transform mechanical advantage or aspect ratios [32]. We also examined the scaling relationships of how the above, measured morphometric traits changed over ontogeny using reduced major axis regression (RMA) [33] using the lmodel2 package [34]. We used the smatr 3 package [35] to confirm significant differences between these ontogenetic slopes. Scaling data were log-transformed prior to scaling analyses, excluding ratios as above [32]. We also visualized the functional feeding morphospace for all species using a principal components analysis. All statistical tests were analysed using R (www.r-project.org)

Intraspecific ontogenetic change in feeding morphology
Tooth aspect ratio scaled with positive allometry in both piranha species over ontogeny (Catoprion: slope = 0.15, Pygopristis: slope = 0.21; table 1, figure 3). However, the manner in which the teeth occlude did not deviate across ontogeny; occlusional offset showed isometric growth for both piranha species (Catoprion: slope = 1.05, Pygopristis: slope = 1.02; table 1, figure 3). In Catoprion, jaw length scaled Reduced-major axis regressions of feeding morphology traits and standard length in Catoprion mento, and Pygopristis denticulata over ontogeny. (a) Tooth aspect ratio, (b) anterior mechanical advantage, (c) posterior mechanical advantage, (d) occlusional offset, (e) jaw length, (f ) jaw adductor cross-sectional area, (g) second moment of area at the jaw joint (0%), (h) second moment of area at the jaw symphysis (90%). The dotted lines indicate the predicted isometric curve. Boxes represent the scaling pattern displayed by each species: N = negative allometry, P = positive allometry, I = isometric growth. Scale-feeders are outlined in red, their non-lepidophagous relatives outlined in blue.

Similarities between lepidophagous taxa
Of the ontogenetic jaw mechanics traits shared between species pairs, both serrasalmids and characids, only isometric growth of the lower jaw and CSA of the adductor muscle were similar between scalefeeding fishes. Between C. mento and R. affinis, a similar trait found in common was tooth aspect ratio, which had values that were similar to each other (0.85 ± 0.02 s.e. and 0.86 ± 0.01 s.e., respectively), as well as these teeth being significantly stouter than their non-scale-eating counterparts. Additionally, the CSAs of the scale-feeding fishes were significantly smaller than their respective sister taxa. Characid and serrasalmid species pairs showed elevational changes between their respective regression lines, i.e. where the ontogenetic trajectory of one species was higher than its sister taxon. Pygopristis and Charax had significantly more cuspidate teeth throughout their ontogeny relative to Catoprion and Roeboides (respectively), although the slopes of these relationships were indistinguishable   Figure 5. A principal component analysis of functional feeding traits from all characiform species in this study. Convex hulls are drawn around each species, points are individual specimens. Scale-feeder hulls are outlined in red, their non-lepidophagous relatives outlined in blue.
between sister taxa (tables 1 and 2). A similar pattern was evident for muscle CSA, where nonlepidophagous taxa had conspicuously larger muscle masses consistently over ontogeny than their scale-feeding counterparts (tables 1 and 2). No other morphological similarities were apparent among the two lepidophagous taxa. Scale-feeding taxa functionally resemble their sister taxon more closely than other scale-feeders. Roeboides and Charax largely overlap in their trait values, having similar mechanical advantages and jaw morphologies. The results of the PCA (electronic supplementary material, table S1) show that while individuals of Catoprion and Pygopristis generally do not overlap functionally, Roeboides and Charax have largely similar mechanical configurations (figure 5). The PCA loadings showed trends corroborated by ANOVA and regression results, e.g. Charax generally had narrower, more pointed teeth than Roeboides, which had broader, more robust teeth. Both characins had narrower teeth relative to the serrasalmid taxa. The average AMA of Pygopristis was greater than that of Catoprion; however, both serrasalmids had greater mechanical advantage than either characin taxa. Pygopristis had jaw occlusion suitable for slicing action, while Catoprion had jaws built for gripping or crushing. Both scale-feeders however had more robust jaws (more material distributed around a neutral axis in the Y-plane) than their non-lepidophage cousins.

Is scale-feeding an ecomorphological monolith?
Mechanically and functionally, the lepidophagous fishes we examined have little in common. There is no archetypal morphology for scale-feeding as seen for some other dietary strategists like hard prey crushers (robust teeth and jaws, large jaw muscles), piscivores (large epaxial muscles, protrusible or tube-like mouths) and herbivores (multicuspid teeth, grinding dentition, long gut tract). Instead we find three different morphologies, one specialized for bodily ramming into prey to dislodge small scales (Roeboides), another which feed orally on small scales as a juvenile (Pygopristis) but move on to other prey as an adult, and a third which appears to feed on large scales consistently throughout its ontogeny (Catoprion) [23,36]. Roeboides' morphospace largely overlaps with its sister taxon Charax, with only a opisthognathous jaw covered by stout teeth distinguishing the two. This shared morphospace is not the same as that of the piranhas, which have completely non-overlapping morphospaces ( figure 5). The many-to-one mapping scheme can be used to explain morphological diversity in ecological guilds [5],  resulting from functional equivalency in phylogenetically-conserved systems, yet here we find different functional outcomes mismatched to an overly-broad ecological category.
We believe these examples of lepidophagy are in fact very different niches from a functional and dietary point of view. The guts of Catoprion and Roeboides are packed with scales, but scales of very different size and number (figure 6). Catoprion eat large scales throughout their ontogeny [28,36]; in one specimen with 19 scales in its stomach, the average scale size was 80% the estimated maximum gape (assuming a gape angle of 120° [36]), and 14% larger than the length of the lower jaw. Larger scales are often found on larger fishes, which may be attractive to scale-feeders which retain a lepidophagous ecology as large adults [19]. Given the sort of niche partitioning diversity evident in 'narrow-niche' fishes like wood-eating catfishes [37], why should we not expect similar patterns to be replicated in the numerous scale-feeding fishes inhabiting similar habitats? Scale-feeding fishes typically inhabit an ecological niche-continuum spanning mucophages, and presumably pterygophages (perhaps even ectoparasite feeders [19,38]), facilitating specialization on any of these nuances of vertebrate ectoparasitism.
Catoprion is one of the few fish we know of that specialize in scales that are very large relative to their gape throughout ontogeny (figure 6, see Terapon jarbua [19]). As juveniles, wimple piranhas are one of many neotropical lepidophagous freshwater fishes, but as adults they alone among scale-feeders exploit the scales of comparably large prey [19,36,38]. This stands in contrast to lepidophagous characins, which consume small scales at small sizes and even when these characins reach larger sizes, continue to eat greater quantities of small scales rather than larger ones [19,39]. Adult Catoprion are also distinguished from most other scale-feeders (and other piranhas) in being a solitary predator, while even large Roeboides and Terapon are found interspersed among their prey or in shoals with conspecifics [38,40]. Juvenile lepidophagous characins occasionally mimic their prey, and are almost always found in schools. In contrast, adult Catoprion control specific territories [27,41], reflecting the need for access to large, mobile prey fishes that pass through these territories. The very traits that make piranhas excellent predators on fishes and fruits [42,43], wide gape, fast jaw closure, strong jaws, are exapted in Catoprion to lever scales from large fishes.

Development of scale-feeding morphologies
Ontogenetic slopes between lepidophage and non-lepidophage relatives show distinct differences in slope elevation while exhibiting few differences in actual slope. This pattern is consistent with the hypothesis that static allometries between species are difficult to evolve while allometric slope elevations

Catoprion mento
Pygopristis denticulata juvenile adult Figure 7. Comparison of tooth shape in Catoprion mento and Pygopristis denticulata as juveniles and as adults. Tooth outlines in lower right corner of each species' panel.

A behavioural hypothesis for tooth form and function in lepidophagous fishes
Our data agree with prior studies, all demonstrating that shared morphological adaptations for scalefeeding (at least among Characiformes) involve specialized dentition [41,52]. Catoprion and Roeboides