Divergent ontogenies of trophic morphology in two closely related haplochromine cichlids

Fish develop morphological specializations in their trophic and locomotor systems as a result of varying functional demands in response to environmental pressures at different life stages. These specializations should maximize particular performances in specialists, adapting them to their trophic and habitat niches at each ontogenetic stage. Because differential growth rates of the structural components comprised in the head are likely to be linked to the diet of a fish throughout its development, we investigated the ontogenetic development of two haplochromine cichlid species belonging to different trophic guilds. We employed geometric morphometric techniques to evaluate whether starting from morphologically similar fry they diverge into phenotypes that characterize trophic guilds and locomotor types. Our examination of overall body shape shows that certain specialized morphological features are already present in fry, whereas other traits diverge through ontogeny due to differences in species‐specific allometric variation. Allometric shape variation was found to be more relevant for the biter specialist than for the sucker morphotype. Our results confirm that phenotypic changes during ontogeny can be linked to dietary and habitat shifts in these fish. Furthermore, evidence for an integrated development of trophic and locomotor specializations in morphology was observed. J. Morphol. 276:860–871, 2015. © 2015 Wiley Periodicals, Inc.


INTRODUCTION
Cichlids are an excellent multidisciplinary model to investigate morphological evolution considering functional morphology, ecological speciation, phenotypic plasticity, and convergent morphotypes. In this context, East African cichlids exhibit a large array of ecotypes in relation to selective pressures on foraging performance and/or behavior, occupying a large range of habitats and trophic niches (Fryer and Iles, 1972;Liem and Osse, 1975;Van Oijen et al., 1981;Witte, 1981;Hoogerhoud et al., 1983;Witte and Van Oijen, 1990). In part, the characterization of these ecotypes is based upon the functional demands on their internal and external anatomy, which interrelate with environmental factors that stimulate the expression of genetic and plastic responses in their morphology (Sage and Selander, 1975). These functional demands change ontogenetically (Osse, 1990;Zengeya et al., 2007), parallel to dietary and niche shifts that many of these species experience (Galis and De Jong, 1988;Goldshmidt et al., 1990;Galis, 1993). This results in a progressive modification of the locomotor and trophic apparatus' morphology, making them more efficient towards a species-specific diet and habitat during ontogeny (Adriaens et al., 2001;Holzman et al., 2008).
Trophic specialization is reflected in an array of internal and external morphologies that can be situated along a biting/sucking functional continuum (Albertson and Kocher, 2006). This has led to the description of numerous trophic guilds (Greenwood, 1974). In the constructional sense, cichlid morphology can be divided into different functional apparatuses that are integrated spatially. It has been documented that certain morphological specializations in locomotor anatomical structures reiteratively correspond to specific trophic guilds, advocating a connection between the development of locomotor and trophic specializations in cichlid fish (Barel, 1983).
Of the functional systems known in cichlids, their oral apparatus is one of the best documented. It generally reflects a trade-off between two mechanically different functions involved in food acquisition: sucking and biting. Mechanically speaking, a fish jaw consists of two opposing lever systems, one for jaw opening and the other for jaw closing (Albertson and Kocher, 2006). The magnitude of how the lever system transmits force or speed is calculated by two ratios that are determined from the insertions of the interopercular mandibular ligament and adductor mandibulae muscle, respectively, at the mandibular articulation. The first characterizes jaw opening, and is the ratio of the retroarticular process (opening in-lever) and the length to the rostral (tooth) tip of the lower jaw (out-lever). The second ratio is calculated as the ratio between the length from the tip of the coronoid process (closing in-lever) and the length of the outlever, and characterizes jaw closing. These ratios reflect the mechanical advantage of the system. A low mechanical advantage predicts rapid jaw rotation, characteristic of sucking species, while a high mechanical advantage predicts powerful jaw rotation, characteristic of biting species.
Feeding performance is influenced by locomotor ability in labrid fish (Higham, 2007a;Collar et al., 2008). Integration of locomotor behavior and feeding kinematics in centrarchid fish (Higham, 2007b) and cichlids (Higham et al., 2007) has led to the prediction that physiological, behavioral, and morphological aspects implicated in these functions coevolve in fish. More recently, certain locomotor morphotypes have been associated to substrate type (Hulsey et al., 2007;Hulsey et al., 2013;Takeda et al., 2013), which is known to be correlated with diet (Winemiller et al., 1995;Genner et al., 1999;Kassam et al., 2004;Arbour and L opez-Fern andez, 2013). In general, four locomotor types have been related to body shape for fish (Webb, 1982): 1) fast steady swimming, 2) unsteady time-dependent swimming, 3) unsteady acceleration plus turning swimming, and 4) placebound maneuverability. Following this classification, zooplanktivores would require steady swimming, which is characterized by an efficient anterior streamline provided by a relatively narrow head and high postcranial body; and benthic oral-shelling molluscivores would require placebound maneuverability, which is characterized by round dorsal head profiles and a relatively deep body at the height of the paired fins (Barel, 1983).
Whether this integration of trophic and locomotor specializations is already present in fry morphology or develops later during ontogeny has not been documented yet. Here, we survey the morphological variation throughout the ontogeny of two haplochromine cichlids belonging to different trophic guilds to observe at what developmental moment species develop morphological specializations belonging to their respective trophic and locomotor ecological niches. Furthermore, we will discuss the functional implications of morphological specialization at different stages in ontogeny as predicted from literature.
The haplochromine species flock of Lake Victoria, the youngest of the African rift lakes, has led to the appearance of 300 endemic species in the last 200,000 years (Fryer and Iles, 1972;Elmer et al., 2009). Species have occupied basically every available niche and food resource, taking on a wide variety of morphotypes specific to the functional demands imposed by their particular environments (Rainey and Travisano, 1998). Within these, Haplochromis piceatus and Haplochromis fischeri are two syntopic endemic species from Lake Victoria (i.e., Mwanza Gulf). These two species are specialized feeders located on opposite sides of the sucking/biting functional axis, with adult head and body shape features representative of their respective trophic guilds: H. piceatus is a pelagic zooplanktivore (fast and steady swimmer) and specialized in suction feeding (Barel, 1983;, and H. fischeri is a benthivorous, oral-shelling molluscivore (place bound maneuverer) specialized in forceful biting (Greenwood, 1981in: Katunzi, 1983). As such, they form an excellent case for comparing shape differentiation reflecting morphological specializations during the different stages of ontogeny since they belong to different trophic guilds along the sucking/biting functional axis (Albertson and Kocher, 2001). However, the amount of shape variation that corresponds to genetic factors or to plastic factors cannot be accounted for because genetic relationships within the endemic Lake Victoria superflock are still under discussion (Meyer, 1993;Verheyen et al., 2003;Wagner et al., 2012).
To analyze shape variation reflecting morphological specializations, morphological features implicated in feeding and locomotion must be identified and quantified, taking into account the homology of structures in both the head and body (Kerschbaumer and Sturmbauer, 2011). Because Lake Victoria cichlids are known to exhibit low morphological variation albeit with important consequences for their ecology (Van Oijen et al., 1981), we expect that morphological specializations will become more pronounced in later stages of ontogeny given that functional requirements during larval stages are more similar (Moser, 1981).

MATERIALS AND METHODS Specimens
The H. piceatus (Greenwood and Gee, 1969) and H. fischeri [Seegers, 2008;formerly H. sauvagei (Pfeffer, 1896)] specimens that founded the aquarium population stock used in this study were collected at the Mwanza Gulf in southern Lake Victoria and shipped to the Haplochromis Ecology Survey Team (HEST) (Van Oijen et al., 1981) laboratory at the University of Leiden during the 1980s. Since then they have been tank bred and reared for 29 generations. In the aquarium facility of the Royal Belgium Institute of Natural Sciences specimens were fed "ad libitum" with commercial fish food (JBL Novostick and Hikari Cichlid Excel pellets) and a weekly complement of frozen Tubifex and Daphnia. Carcasses were fixated in 80% nondenaturalized ethanol after an overdose of MS-222. A total of 34 specimens of H. piceatus and 37 specimens of H. fischeri were used. The samples for each species comprised an ontogenetic series with individuals that had already absorbed their yolk sac, spanning from 1 to11.5 cm standard length (SL) ( Table 1). In order to account for the influence of domestication on shape, three type specimens from Lake Victoria, Tanzania were included for H. piceatus (RMNH 62769) and two for H. fischeri (formerly H. sauvagei; RMNH 70426), provided by the NCB Naturalis (the Netherlands Centre for Biodiversity, National Museum of Natural history and Research Center on Biodiversity in Leiden, The Netherlands).
Specimens were photographed with a Nikon D70 digital reflex camera using a Sigma 105 mm macro lens at five megapixels resolution. Fish were placed on a 20 3 15cm dissection board with a white paper background equipped with a scale bar. Specimens were centered to avoid optical distortion of the images at the lens borders (Arnqvist and Martensson, 1998). When needed, pins were placed in the tail and/or pectoral fin region to minimize unnatural bending of certain structures due to the fixation process.
To match the observed ontogenetic morphological changes to ecological data found in literature, specimens of both species were pooled into three size classes (1-4 cm SL, 4-8 cm SL, and 8 cm SL). We use size as a proxy for age, which has its pros and cons (Godfrey and Sutherland, 1995), but whose use has been justified before in ontogenetic studies (Zelditch et al., 2000). These size limits were established based on earlier work on ontogenetic development in African cichlids (Van Oijen et al., 1981;Witte, 1981;Hoogerhoud et al., 1983;. Maternal mouthbrooding care stops when fry reach an approximate length of 1 cm SL, and these experience an increase of approximately 4 cm SL during their first year (Witte, 1981).  noted changes in habitat, diet, and morphology at an approximate length of 7 cm SL and observed an increased growth (1 cm) of tank-bred specimens relative to wild individuals of the same species (H. piceatus). Adjusting our data to these observations, size ranges have been defined as "I" (1-4 cm SL), "II" (4-8 cm SL), and "III" (>8 cm SL).

Morphological Data Acquisition
To analyze shape variation in head and body morphology, 32 homologous landmarks (LMs; Fig. 1) were digitized: 13 in the head region, 11 outlining the exterior and denoting the base of the fins, two for the pectoral fin, and six indicating the lateral line and central longitudinal axis. The landmarks denoting the longitudinal axis were not included in the shape analysis, but were used as reference to apply the unbending procedure in TPS Util v1.38 (Rohlf, 2006a), in this way circumventing shape variation caused by unnatural bending during the fixation process.
In order to incorporate variation in head width in the analysis (as biter morphotypes tend to have wider heads), two measurements were taken on the head using an electronic caliper (0.1 mm accuracy). "Snout width" was measured at the height of the posterior extremity of the gape (LM7) and "head width" was measured at the level of the preopercular bone (LM1). In addition, standard length and interlandmark distances (calculated in Past v1.81 [Hammer et al., 2001]) were included as variables in the regression analysis.

Analysis of Shape
Shape data was analyzed statistically by means of Geometric Morphometrics (Zelditch et al., 2004). The correlation between Procrustes and tangent distances between specimens was tested using TPS Small v1.2 (Rohlf, 2003). A Principal Component Analysis (PCA) was performed on shape variables in Mor-phoJ v1.05b (Klingenberg, 2011) to search for the axes that maximize shape variation within the ontogenetic sample. Multivariate analysis of variance was performed on shape variables in IBM SPSS Statistics v19 (SPSS) to test for significant differences between species' ontogenetic trajectories. Ontogenetic growth vectors were calculated and their directions and lengths compared. To estimate the range of angles between growth vectors, the residuals from the regression of shape on size (CS [centroid size] and lnCS) were paired with predicted shape values and bootstrapped (2500 iterations) with replacement in IMP-VecCompare8 (Sheets, 2003(Sheets, -2014 to obtain significance values under the null hypothesis of parallel vectors. Ontogenetic growth vector lengths were calculated as Procrustes distances in IMP-Regress8 (Sheets, 2003(Sheets, -2014 using as a reference the consensus shape from the twelve smallest specimens (i.e., the six smallest specimens of each species). These Size classes are designated based on intervals of standard length (SL): "I" (1-4 cm SL), "II" (4-8 cm SL), and "III" (>8 cm SL).
distances were then regressed on size (lnCS) for each sample, and the slopes' mean, confidence intervals (95%), and P-values calculated through a Monte Carlo resampling procedure (1,000 iterations) with replacement (Zelditch et al., 2004) using the PopTools v3.2 (Hood, 2011) plugin in Microsoft Excel 2010. Next, the common allometric trajectory was calculated for both species and a novel PCA performed on the residuals to extract species-specific allometric shape variation. Shape changes are visualized by means of the deformation-based thin-plate spline interpolating function (Bookstein, 1991;Bookstein et al., 1996) and illustrated as wireframe grids.
To discern what structures were developing divergently at each ontogenetic stage and whether or not their development was correlated with size (lnCS), an analysis of variance (ANOVA) was performed to test for differences between group means in log-transformed biometric variables (SL, snout width, head width, and interlandmark distances). Afterwards, variables were corrected for size (lnCS) to eliminate ontogenetic size variation using General Linear Models (GLM) in IBM SPSS Statistics v19 (SPSS). To explore the differences between factor levels in GLM models with two categorical variables (i.e., SPE-CIES and SIZE_CLASS), in the absence of post hoc significance tests when the homogeneity of slopes assumption is violated, we compared the estimated marginal means plots. The level of statistical significance was set at a P-value <0.05. The Bonferroni and Tamhane's T 2 (when variables present unequal error variances across groups) adjustment for multiple comparisons were applied where necessary.

Ontogenetic Shape Trajectories
A multivariate analysis of covariance (MAN-COVA) was performed on shape variables using "size" (lnCS) as the covariate to test the null hypothesis of isometric growth and remove the effect of size differences between individuals within the ontogenetic series (Table 2). Wilk's k resulted significantly greater than expected by chance, indicating that species differ in their ontogenetic shape trajectories irrespective of differences in size. The multivariate distribution parameter was also significant for lnCS, leading us to reject the null hypothesis of isometric growth. This means that shape is allometric, so that it changes as a function of size. The interaction effect "species*lnCS" also resulted significant, which violates the homogeneity of slopes assump-tion in the MANCOVA. However, in biological terms this implies that each species has a different allometric trajectory in the shared ontogenetic shape space. "Size" explains a larger proportion of the variance (10%) in the model than "species" in view of the partial eta-squared (h 2 p) values (h 2 5 0.994 [size] vs 0.864 [species]). In units of Procrustes distance (d 2 ) this corresponds to 0.072 vs. 0.057 of 0.155, respectively.
The magnitude of the difference between species' ontogenetic shape trajectories was tested under the null hypothesis of parallel directions in the shared morphospace. The angle between species' ontogenetic vectors is of 34.4 , and the 95th percentile of the ranges of the within-species angles are 30.7 for H. piceatus and 24.2 for H. fischeri. The interspecific angle exceeds both within-species ranges, so we can conclude that the two species differ significantly in the direction of their ontogenies of shape.
To test for differences in the ontogenetic rate of amount of shape variation relative to increase in size between species, we calculated the Procrustes distance from each specimen to a consensus configuration calculated using the six smallest specimens of each species (1-2 cm SL). The Procrustes distances were plotted on size (CS) and the slope of the regression bootstrapped (1,000 iterations) to obtain the confidence intervals for each species (H. piceatus: 0.0024-0.0042; H. fischeri: 0.0026-0.0039). No significant differences were observed between species in the length of their ontogenetic vectors.

Ontogenetic Shape Variation
The PCA-analysis maximized between individual shape differences, revealing two trends in the shared ontogenetic morphospace: PC1 (37%) shape variation reflects similar shape changes for both species in relation with size increase, while PC2 (15%) reflects a component of shape variation that discriminates species (Fig. 2). Because PC1 shape variation is frequently considered a size axis in geometric morphometric studies, we calculated how much of PC1 and PC2 shape variation are correlated with size in our sample by regression. We observed that 64% (P < 0.0001) of PC1 shape variation is predicted by size, while 17% (P < 0.0001) is predicted for PC2.
Shape changes associated with the PC1 axis from smaller to larger individuals (positive to negative values;) (Fig. 2) involve i) a relatively shorter head, snout and oral jaws, ii) a dorsally shifted and reduced orbit, iii) a relatively longer ascending arm of the preopercular and larger opercular area, iv) a relatively deeper body and straightening of the dorsal outline, v) a rostral displacement and inclination of the pectoral fin, and vi) a steeply angled transition from the caudal peduncle towards the anal fin.
Shape changes associated with the PC2 axis from positive (H. fischeri) to negative (H. piceatus) values (Fig. 2) reflect i) a proportionally deeper head and cheek depth, ii) relatively longer snout, oral jaws and ascending arm of the preopercular, iii) a steeper angled transition from the neurocranium towards the dorsal fin, iv) a relatively deeper anterior body with a steeper angled transition towards the caudal peduncle, and v) a relatively shorter caudal peduncle.

Allometric Shape Variation
The multivariate regression of shape on size revealed that 28% (P < 0.0001) of ontogenetic shape variation is explained by size (Fig. 3). This allometric shape variation from positive to negative values is similar to PC1 shape variation, but differs in that i) there is no relative shortening of the head, ii) the leading edge of the dorsal fin shifts more dorso-rostrally, iii) there is no relative change in the inclination of the dorsal outline of the caudal peduncle, and iv) the bases of the leading edges of the anal and pelvic fins display a less marked ventral shift.
Deriving from the significant interaction effect between species and size in the MANCOVA that indicated different allometries of shape between species, we regressed species' allometries separately, but within the same Procrustes superimposition. For H. piceatus 28% of shape variation could be predicted by size and 42% for H. fischeri. The interspecific angle between them was of 35 (P < 0.0001). To test for allometric shape variation discriminating species, we performed a new PCA on the residuals from the shared allometric regression to maximize shape differences between individuals. Species were clearly discriminated (Wilk's k 5 0.031; F 5 14.333; P < 0.001; h 2 p 5 0.969) along residPC1 (30%; Fig. 4); H. piceatus individuals have positive residPC1 scores, while H. fischeri individuals have negative ones (with two exceptions). residPC1 axis shape variation predicted 34% of PC1 shape variation (with vectors at an angle of 40 ) and 48% of PC2 (with vectors at an angle of 55 ).
Shape changes described by the residPC1 axis ( Fig. 5) from H. fischeri to H. piceatus comprise i) a significant increase in head length, oral jaw length, snout height, and body height, ii) a more terminal positioned mouth, iii) a dorso-rostral shift of the origin of the first soft and hard dorsal fin rays, creating a steep transition towards the dorsal caudal peduncle, iv) a caudal shift of the pectoral fin, v) and a dorso-rostral shift of the origin of the first soft and hard anal fin rays, resulting in a steep transition towards the ventral caudal peduncle.

Biometric Variables
Interlandmark distances were chosen from the landmark configuration considering that they covered anatomical structures known to be implicated in sucking/biting performance and/or in other functions (Fig. 6). The linear measurements employed are defined in Table 3. All variables were transformed to their natural logarithm to linearize allometric relationships for regression analysis (Mascaro et al., 2014). A preliminary GLM was performed using lnCS as covariable to test what variables were correlated with an increase in size (Table 4). Variables not correlated with an increase in size were OpW, GH, LJ, PDA, BH, AF2, and PcF-PvF. It is noteworthy to mention that CS did not show significant differences Fig. 3. A plot of the regression scores of ontogenetic shape on size (lnCS). Confidence ellipses denote 90% mean value intervals for species' size classes.  between species for the corresponding age classes indicating a similar growth rate (as quantity of shape change per increase in size; Supporting Information Fig. S1) To observe what variables differed between species, a distinct ANOVA with SPECIES as the categorical variable was performed for each variable to avoid correlation interactions between variables in a multivariate GLM model (Table 4). Species had significantly different means for the variables BL, OpW, GH, HL, HH, and AF2. Because, species samples consist of an ontogenetic series, an analysis of covariance (ANCOVA) was performed to correct for size (lnCS). Additionally, the variables SL, HW, LJ, ChD, SnL, PDA, and PcF-PvF resulted significant, however violating the homogeneity of slopes assumption (except for SL). This indicates that the relationship between these variables and the covariate differ between species, suggesting different ontogenetic trends of these variables for each species.
To observe differences in our biometric variables between size classes through ontogeny, ANOVA was performed as before with SIZE_CLASS as the categorical variable (Table 4). Size classes presented significantly different means for the variables CS, SL, SW, HW, BL, HL, ChD, SnL, and NL. However, after correcting for differences in size (lnCS) through ANCOVA, only the variables HW, HH, BH, and PcF-PvF resulted significantly different between size classes. Of these, only BH violated the homogeneity of slopes assumption, suggesting a change in the ontogenetic trend of this variable at a determined size range for both species.
To further elucidate differences between species' ontogenetic series in biometric variables, a GLM was performed including both SPECIES and SIZE_CLASS as categorical variables in the model (Table 4). Once again, size correction was executed. Size classes had significantly different Fig. 6. Illustration of the wireframe used to describe body shape. Interlandmark distances calculated from the landmark configuration (in grey) constitute the variables used in the regression analysis that are described in Table 3.  (Osse, 1990) Interlandmark distances were calculated in Past v1.81 (Hammer et al., 2001). means for the variables SL, SW, HW, HL, LJ, ChD, SnL, PDA, and PcF-PvF. All of them violated the homogeneity of slopes assumption indicating differences in variable values between species, depending on the size range of individuals during ontogeny. Estimated marginal means plots were generated to estimate the timing of these ontogenetic shifts in variable values between species' size categories (Fig. 7).

Evolution of Morphological Allometry
Shape variation associated to a common allometric trajectory and that from species-specific allometry were examined separately to observe what shape changes were correlated solely to a common allometric trajectory from those that involved species-specific development (Fig. 5). Speciesspecific allometric shape variation accounted for a larger percentage of the shape variation within the ontogenetic sample (30%) than the common allometric component (28%). Both allometric components contribute to the shape differences associated to our PC1 and PC2 axes that maximize individual differences (Fig. 2), and an interaction between them in ontogenetic shape space is patent. Together they predict 98% (64% and 34%) of PC1 shape variation and 65% (17% and 48%) for PC2.
Allometric changes discriminating our species coincide with shape variation associated to their respective locomotor and trophic specializations (see below), similar to what has been observed in other Lake Victoria specialists (Bouton et al., 1999). Since species did not display differences in relative growth rates, allometric differences in biometric variables between species must improve some species-specific function at a certain moment in ontogeny (Pelabon et al., 2014). In a constructional context, this may be achieved by different spatial arrangements of the respective apparatuses between species and/or size classes (Strauss, 1984;Barel et al., 1989;Liem, 1991;Barel, 1993) originated by the reallocation of resources to meet functional demands at different ontogenetic moments (Von Bertalanffy, 1957;Ruehl and DeWitt, 2005;Taborsky, 2006). This seems to be the case with the oral jaws and the interpectoralpelvic fin length (Table 4). In relation to the recent literature on cichlid shape divergence along the benthic-limnetic axis (Hulsey et al., 2013;Takeda et al., 2013), the sucker morphotype apparently may be allocating more resources to increase in body length along the anterior-posterior axis during ontogeny, while the biter morphotype to increase lengths along the dorso-ventral axis and head width.
We expected that species' shape differences would become more pronounced through ontogeny starting from morphologically similar fry. We found that even though larvae were morphologically very similar, they already displayed differences in morphological characters uncorrelated with size that are implicated in trophic/respiratory (gill Fig. 7. Estimated marginal means of variables with a significant SPECIES*SIZE_CLASS interaction effect, and, that violate the homogeneity of slopes assumption. The horizontal axes denote size classes and individual lines represent each species (black: H. fischeri, grey: H. piceatus). Line segments that are parallel indicate that there is no interaction between the categorical variables at that ontogenetic interval. Estimated marginal means were calculated at the covariable value of lnCS 5 1.97. height and opercular width), a larger size of the gill arches enlarges the volume of the buccal cavity during suction feeding (Osse, 1990), and locomotor functions (soft anal fin region length). The former variables had larger values in the sucker morphotype, whereas the latter was larger in the biter morphotype. Hence, functionally relevant morphological differentiation between species is already present at the beginning of ontogeny for these characters, but is later magnified due to species-specific allometries that arise at specific moments in ontogeny (size classes). This implies that the developmental program of morphological specializations is decoupled in modular genetic programs throughout ontogeny, which may allow for phenotypic plastic adjustments at each ontogenetic stage (Atchley, 1984). In view of the morphologic (Barel et al., 1977) and genetic irresolution (Elmer et al., 2009;Wagner et al., 2012) of the Lake Victoria Haplochromis genus (including the species studied), we lack the phylogenetic framework to draw any conclusions on the divergence in the evolutionary direction of species' allometric trajectories. Nonetheless, the most recent common ancestor of the entire Lake Victoria Region haplochromine species flock was estimated to have existed at 4.5 million years ago (Elmer et al., 2009).

Trophic and Locomotor Functional Significance of Shape Variation
Species-specific allometric shape variation discriminating species (Fig. 5) agrees with similar comparisons relating to convergent sucker and biter morphotypes in all three East African Lake cichlid assemblages (Young et al., 2009): elongate bodies are typical of planktivorous suction feeders, whereas deep bodies with short down-turned heads are associated with diets comprised of harder prey items.
The functional implications of morphological specializations that facilitate more powerful biting have been evaluated in cichlids before (Barel, 1983;Van Leeuwen and Spoor, 1987;Galis, 1992;Bouton et al., 1998). It is agreed that in molluscivores, the jaw apparatus is more adapted to forceful biting. To this, we have to add the intraspecific differences in muscle recruitment and possible patterns of jaw movement (Liem, 1978;Galis, 1992). However, intraspecific shape variation due to phenotypic plastic adaptations to diet items (Bouton et al., 1999) can be ignored in our results because species were fed the same food regime. The pattern of morphological variation observed in H. fischeri in overall body shape predicts certain internal anatomical variation (Sanderson, 1990). In the head, the ample dorso-caudal shift of the eye and the substantial increase in length of the ascending arm of the preopercular bone and in height of the suspensorium, enlarges the space in this region, providing a larger insertion area and available volume for the adductor mandibulae muscle implicated in forceful biting (Barel, 1983). In the oral jaw lever system, we observe a relative increase in length of the coronoid process (closing in-lever) relative to the lower jaw (out-lever), which grants a higher mechanical advantage to the system (Albertson and Kocher, 2006). Both these changes mechanically lead to a progressively larger biting force (Bouton et al., 2002), which can thus be expected in H. fischeri. In the constructional context (Barel et al., 1989), the development of structures implicated in the trophic core functions of biting and sucking (Barel, 1983) is also constrained by that of adjacent apparatuses. All these apparatuses (oral jaw apparatus, expansion apparatus, gill apparatus, and locomotor apparatus) share spatial demands, resulting in morphological constraints reflected in functional tradeoffs. The different arrangements between apparatuses determine the range of form-features allowed architectonically. Based on these arrangements, Barel (1983) identified associated morphologies between the oral jaw apparatus and remaining apparatuses that either optimize one core trophic function or the other.
The head shape of H. fischeri has a more rounded profile resulting from the rostral-ventral shift of the anterior edge of the dorsal fin. The rostral-ventral shift of the leading edges of the anal and pelvic fins create a flat ventral margin, which is complementary to this shape of the head profile in providing rotation maneuverability characteristic of benthic feeders (Aleyev, 1977). In more recent investigations (Drucker and Lauder, 2001;Chadwell and Ashley-Ross, 2012), functional studies of locomotor specialization have revealed certain aspects of fin development that were also apparent in our results. Differences in the soft anal fin region length affect the generation/resistance of hydrodynamic forces during swimming. This is because the posterior region of the anal fin is functionally decoupled from the anterior region and provides roll and/or yaw stability, while generating additional thrust forces during slow turning maneuvers (Chadwell and Ashley-Ross, 2012). The development of these locomotor specializations associated with the biter trophic morphotype advocate a certain integration of feeding and locomotor functions Franchini et al., 2014), although it may just be a species-specific pattern.
The relative elongation of the lower jaw (outlever) in our H. piceatus sample results in a smaller mechanical advantage and consequently in an improved kinematic efficiency. The dorsal shift at the ventral intersection point between opercular and interopercular bone alters the inclination of the head occasioning an upturned mouth characteristic of pelagic feeders. The increase in size of the snout and increasing horizontal, dorsoventral orientation of the ventral head profile provide a more rectangular lateral head profile that when expanded results in a larger and more cylindrical buccal cavity with an increased buccal volume characteristic of suction feeders (Barel, 1983;Muller and Osse, 1984).
Associated changes in the locomotor apparatus are an efficient anterior streamline and a minimum body area reflected in relatively small widths and depths in outer head shape. In the constructional context, the increase in body height and the caudal displacement of the pectoral fin in H. piceatus leave more space adjacent to the head for the epaxial and hypaxial musculature, which coincides with the necessity of an increased need of power for head expansion in slow-swimming suckers (Barel, 1983;Wainwright et al., 2001;Carroll et al., 2004). In accordance with H. fischeri, the development of these locomotor specializations associated to the sucker morphotype advocate a certain integration of feeding and locomotor functions in Lake Victoria haplochromines.

Ecomorphological Implications of Morphological Specialization
The existence of differences between species in biometric variables correlated with size that are implicated in trophic and locomotor function advocate a benefit of increased growth considering that fish mortality is usually an inverse function of size (Galis and De Jong, 1988). Growths of characters in the head are especially important for food uptake. In the biter morphotype, the increase in head width allows individuals to feed upon larger prey items through ontogeny. Such a functional ontogenetic shift has been put forward for H. fischeri (Katunzi, 1983), and our observations corroborate that morphological specializations produced by its species-specific allometry facilitate a behavioral food-partitioning between individuals of different ontogenetic stages based on prey size in this species (Katunzi, 1983;Ferry-Graham et al., 2002). However, this is not the case for the sucker morphotype since Galis and De Jong (1988) observed during its ontogeny by means of optimal foraging models equal Chaoborus prey uptake and decreasing uptake of Daphnia prey with increasing size. We observed that variables in the head implicated in trophic specialization in this species do not begin to increase significantly in length until size class II, suggesting that a relatively larger buccal volume is not a constraint in food uptake until size class III, which coincides with the optimal foraging model of Galis and De Jong (1988). And on the contrary, oral jaw length increases in value through all of ontogeny, continuously potentiating suction feeding (kinematic effi-ciency and jaw protrusion) as H. piceatus individuals get bigger. The benefits of increased growth in size class II are less obvious in view of biometric variables implicated in locomotor performance. Both body height and interpectoralpelvic fin length display a similar increase in value at this size class for both species (H. piceatus displaying higher absolute values), but neither were correlated with size. The increase in interpectoral-pelvic fin length for the size class II biter morphotype may result in enhanced maneuvering capacities and force generation at the pectoral girdle (Drucker and Lauder, 2002). This morphological specialization can be linked to an ontogenetic habitat shift towards deeper waters  where benthic locomotion is more important. Similarly, the dorsal head profile at size class II becomes higher and more rounded which in addition to the flat ventral outline provided by the increase in interpectoralpelvic fin length, provides an adaptation to pitch over the bottom more effectively (Aleyev, 1977). These observations in our biter morphotype advocate an integrated development of the trophic and locomotor apparatus through ontogeny due to changing functional demands (Higham, 2007).
The development of locomotor specializations described by an efficient streamline in the sucker morphotype due to increased values in their body height and interpectoral-pelvic fin length is also more pronounced at size class II. However, body length displayed a significantly increased growth rate at size class I for this species. These observations support that morphogenetic programs are decoupled at different ontogenetic stages (Atchley, 1984), and coincide with the ontogenetic niche shift this species undergoes from shallow littoral nurseries to deeper waters when reaching size class III since predator avoidance and prey capture depend more on speed in pelagic waters (Witte, 1981;.
In the context of the adaptive radiation of East African cichlids, more ecological studies dealing with the biomechanics of the ontogenetic dietary and niche shifts that the two species studied undergo are necessary to evaluate whether the here observed morphological differentiation corresponds directly to differences in performance that can influence their survival at different moments in ontogeny. Although the species are syntopic in Lake Victoria, they should not compete with one another since they have different depth distributions (Van Oijen et al., 1981;, and differences in breeding strategies concerning timing, spawning, and brooding sites that are likely to contribute to the partitioning of resources . Thus, that the ontogenetic patterns of morphological specialization observed should be more the product of independent selective pressures for each species. The integration during ontogeny of shape variation patterns involving morphological features implicated in trophic and locomotor specializations does not agree with a three stage model of adaptive radiation in which habitat and trophic niche adaptation are considered independent of one another (Streelman and Danley, 2003), but puts forward an integration of these two stages in the adaptive radiation process.