Ecological variation in South American geophagine cichlids arose during an early burst of adaptive morphological and functional evolution

Abstract

Diversity and disparity are unequally distributed both phylogenetically and geographically. This uneven distribution may be owing to differences in diversification rates between clades resulting from processes such as adaptive radiation. We examined the rate and distribution of evolution in feeding biomechanics in the extremely diverse and continentally distributed South American geophagine cichlids. Evolutionary patterns in multivariate functional morphospace were examined using a phylomorphospace approach, disparity-through-time analyses and by comparing Brownian motion (BM) and adaptive peak evolutionary models using maximum likelihood. The most species-rich and functionally disparate clade (CAS) expanded more efficiently in morphospace and evolved more rapidly compared with both BM expectations and its sister clade (GGD). Members of the CAS clade also exhibited an early burst in functional evolution that corresponds to the development of modern ecological roles and may have been related to the colonization of a novel adaptive peak characterized by fast oral jaw mechanics. Furthermore, reduced ecological opportunity following this early burst may have restricted functional evolution in the GGD clade, which is less species-rich and more ecologically specialized. Patterns of evolution in ecologically important functional traits are consistent with a pattern of adaptive radiation within the most diverse clade of Geophagini.

1. Introduction

Clades may differ in the tempo (rate) and mode (distribution or direction) of morphological evolution and these differences may contribute to the uneven phylogenetic and geographical distribution of morphological, ecological and taxonomic diversity. Understanding what processes affect the tempo and mode of evolution, as well as how the effects of these processes accumulate within groups over evolutionary time, is necessary to understand the unequal distribution of morphological and ecological diversity [1–3]. Adaptive radiation, in which lineages diversify rapidly while adapting to different ecological niches, suggests that both diversity and disparity are tightly linked to ecological opportunity [4]. Thus, a major expectation of adaptive radiation is a concomitant decrease in rates of lineage diversification and rates of evolution in ecologically relevant traits. An early burst of evolution in ecologically relevant phenotypic traits may be of particular importance in identifying adaptive radiation, although evidence from comparative datasets may yet be rare [5].

Adaptive radiation has been proposed to drive diversification patterns on a broad geographical and taxonomic scale [6,7]. However, many studies of adaptive radiation have focused on recent diversification events in isolated habitats such as islands or lakes [8], including Darwin's finches, Anolis lizards, African rift lake cichlids and Hawaiian silverswords [4,9–11], whereas radiations on a continental scale have been relatively poorly studied [12]. It is possible that the conditions underlying radiations in such isolated systems do not adequately scale to those of continental radiations. If adaptive radiation is a significant contributor to overall diversification patterns, it should be diagnosable on broad phylogenetic and geographical scales. The following study expands on the comparative analyses of morphological evolution and ecological diversity of the potential continental-scale, ancient adaptive radiation in the Geophagini tribe of Neotropical cichlids.

Neotropical cichlids are an ecologically and morphologically diverse group of riverine fishes [13] sister to the African clade that includes the extensively studied but relatively recent African Rift Lake cichlid radiations [8,10,14]. Geophagini is the most species-rich and morphologically diverse tribe of Neotropical cichlids, with 240 described species (species counts from ‘fishbase.org’) as well as numerous undescribed species [15,16]. Originating during the Cretaceous, this tribe has become widely distributed across South America and encompasses important ecological roles within many Neotropical aquatic communities [15–18]. Geophagini is composed of two sister clades with unequal species richness and ecological diversity. One clade (henceforth, clade GGD, for the three most species-rich genera Geophagus, Gymnogeophagus and Dicrossus) consists of 47 described species of which most are specialized, substrate-sifting benthivores in six genera: Geophagus, ‘Geophagus’ brasiliensis, ‘Geophagus’ crassilabris, Gymnogeophagus, Biotodoma and Mikrogeophagus [19,20]. Clade CAS (for the three most species-rich genera Crenicichla, Apistogramma and Satanoperca) consists of 193 described species ([15,16] and from fishbase.org), and similar to clade GGD includes substrate-sifting benthivores in the genera Satanoperca and Acarichthys [20], as well as numerous (greater than 90) dwarf species, such as those within the genus Apistogramma, living in leaf litter habitats and probably picking benthic invertebrates from the substrate [19,20]. Clade CAS however, also includes Crenicichla (which includes within it the genus Teleocichla, see López-Fernández et al. [17]; henceforth, the ‘Crenicichla clade’), which comprises 90 described and numerous undescribed species of elongate-bodied predators feeding on fishes and invertebrates [21,22].

Previous analyses have shown support for declining rates of lineage and body shape diversification over the evolution of Geophagini [18,20] consistent with a hypothesized adaptive radiation [23–25]. Geophagini thus provides an excellent opportunity to examine the relationship between morphological/ecological diversity and adaptive evolution. However, previous analyses have relied on traditional linear morphometrics [18,20] and quantifying ‘trait-utility’ [23] from such metrics is not straightforward. Functional morphological traits, on the other hand, are more closely associated with performance capability and therefore changes in these traits are more likely to reflect ecological characteristics [26]. In this study, we examined the tempo and mode of evolution in feeding functional morphology in the two major clades of Geophagini, and contrasted the observed patterns with those expected under a neutral, random-walk evolutionary process using Brownian motion (BM) simulations. From this we can determine whether (i) functional evolution in Geophagini exhibited an early burst consistent with adaptive radiation, and (ii) patterns of functional evolution are congruent with patterns of ecological diversity.

2. Material and methods

(a). Measurement of feeding biomechanics

We examined oral jaw, hyoid and pharyngeal jaw characteristics in 84 specimens representing 27 species of Neotropical cichlids (see the electronic supplementary material, table S1 and figure S1), of which 23 species are part of the Geophagini tribe, including at least one species from all known genera and four outgroup taxa. We focused on genus-level functional variation as previous studies have shown that morphological variation is much greater between than within genera of Geophagini [20]. Abbreviations for all taxonomic names used in the figures can be found in the electronic supplementary material, table S1 and figure S1. We focused on functional attributes that have been correlated with diet composition or feeding behaviours in cichlids or similar groups. Ten well-established morphological variables and biomechanical indices characterizing the production, transmission and distribution of force in the feeding apparatus of cichlids were measured from each specimen. The masses of the major jaw opening and closing muscles (sternohyoideus (ST) and adductor mandibulae (AM), respectively) were considered indicators of force production ability, as force production tends to scale with (muscle mass)^2/3. However, it is important to note that other properties of the muscles (e.g. muscle fibre length and angle) can impact force production and are likely to vary across geophagine cichlids [27]. Other measurements included pharyngeal jaw bone (CB5) mass indicating crushing potential, lever and linkage mechanical coefficients describing the transmission of force and velocity (lower jaw opening and closing mechanical advantage (MA), oral jaw and hyoid apparatus kinematic transmission (KT) coefficients), quadrate offset (QO, also see the electronic supplementary material, figure S2) describing bite occlusion and maximum jaw protrusion (JP) relating to attack velocity as well as suction index (SI) describing suction ability potential (see the electronic supplementary material, table S1 for a more detailed description of biomechanical indices and associated references and electronic supplementary material, table S2 for species mean values).

Comparative analyses within this study were based on a time-calibrated molecular phylogeny of the subfamily Cichlinae described by López-Fernández et al. [17,18], with all summarized results for all analyses presented as the mean ± s.d. calculated across 1000 posterior distribution trees (see the electronic supplementary material, figure S1 for more information). All size-dependent variables, including AM, ST and CB5 mass and JP distance (see the electronic supplementary material, table S1) were log-transformed and adjusted using phylogenetic size correction (‘phyl.resid’, phytools R package v. 0.1-8) by body mass or standard length [28]. The major axes of variation in functional morphology of the 27 taxa examined were characterized using phylogenetic principal component analysis (‘phyl.pca’, phytools R package v. 0.1-8) of size-corrected (mass and length residuals) and uncorrected, size-independent variables, including lower jaw opening and closing MA, oral jaw and hyoid KT, QO and SI. The eigenvalues of the observed principal component (PC) axes were contrasted with those generated by randomizing species trait values prior to phylogenetic PC analysis (PCA) to determine which axes represented critical variation [29,30]. Scores from the four outgroup taxa were used to help establish the ancestral trait values for all Geophagini but were excluded from further analysis.

(b). Phylomorphospace distribution

We examined whether the rate and distribution of functional evolution differed between the two clades of geophagine cichlids. The functional disparity of each clade was calculated using the sum of the variance of the scores from each PC axis scaled by the proportion of variation explained for each critical axis [31]. Morphometric branch lengths, the Euclidean distance in functional morphospace between the two nodes of a branch in the phylogenetic tree [3], were calculated after reconstructing ancestral character values using the maximum-likelihood function ‘ace’ from the ‘ape’ R package. Morphometric branch lengths were calculated across 1000 posterior distribution chronograms for both clade GGD and clade CAS and scaled by the length of each corresponding branch of the chronogram to determine the average evolutionary rate of each clade [3]. We compared the observed rates to those produced under a null model of BM evolution. The function ‘sim.char’ from the R package ‘geiger’ was used to simulate PC scores using eigenvalues from each of the PC axes as the variance of each axis.

We examined lineage density (LD) to determine the efficiency of each clade at expanding through functional morphospace [3]. The total morphometric branch length for each clade was compared with the volume of morphospace it occupied (LD = total distance/volume). Clades with a higher LD are more ‘packed’ within morphospace, whereas those with low LD have undergone greater relative expansion. If the evolutionary rates or LDs between the clades differed significantly, this would indicate that one clade was more rapidly or efficiently expanding through functional morphospace [3]. Furthermore, if the rates or LD ratios differed between the observed and simulated data, it would indicate that the differences observed between clades could not have occurred under strictly BM evolution. Functional disparity, evolutionary rate and LD were calculated over 1000 chronograms in the posterior distribution, and the observed and expected values were contrasted using two-sample t-tests. We calculated LD using the volume of three-dimensional convex hulls surrounding all PC scores within each clade (function ‘convhulln’, R package ‘geometry’ [32]).

(c). Patterns of disparity-through-time

Subclade disparity-through-time (DTT) analyses were used to examine how the evolution of functional disparity has been distributed during the evolution of the major lineages of geophagine cichlids. The average disparity of all subclades present at each node in the chronogram was contrasted with the modern disparity of the clade as a whole [2]. These observed values of relative subclade disparity were compared with those generated under BM simulations (1000 simulations for each of 1000 posterior distribution chronograms). Relative subclade disparity shows a fairly linear decline towards the present under BM evolution, whereas consistently low observed values of relative subclade disparity are indicative of early bursts of morphological evolution, as most variation corresponds to early divergence events and more recent subclades are restricted in morphospace [2]. Comparatively, high values of relative subclade disparity towards the present indicate that modern functional disparity originated during recent divergence events, a pattern which can be consistent with evolution around an adaptive optimum [18].

Departures from BM evolution were summarized using the morphological disparity index (MDI), which is the area between the observed and average simulated DTT curves, and has an expectation of 0 (under BM). We also examined the likelihood of MDI values occurring under a BM model of evolution by comparing the observed MDI value and the MDI value of each simulated character history [12,33]. If the observed MDI values occurred as a result of BM evolution, then the frequency of simulated values more extreme than those observed (i.e. MDI_BM> MDI_obs if MDI_obs is positive or MDI_BM < MDI_obs if MDI_obs is negative) should be approximately 50 per cent. All DTT analyses were carried out using the function ‘dtt.full’ from the R package ‘geiger’. We carried out all DTT analyses (calculation of MDI and % of extreme BM simulations) over the basal two-thirds of each chronogram to account for incomplete taxonomic sampling [2].

(d). Comparison of models for morphological evolution

We examined whether evolution in geophagine cichlid functional morphology is constrained towards adaptive optima by comparing BM and Ornstein–Uhlenbeck (OU) models for morphological evolution using a maximum-likelihood approach. While BM evolution is based on a random-walk process, OU models incorporate selection towards one or more adaptive optima [34–36]. We compared the likelihood of several models of functional evolution, including a BM model (random-walk with a constant rate of morphological evolution) as well as an OU model (a constant rate, random-walk around a selective optimum) for all Geophagini (‘single peak’ model), which would represent a general selective constraint on the feeding apparatus of all geophagine cichlids. We also examined whether the presence of different selective optima in each clade (i.e. higher or lower constraints in each clade) could have contributed to any observed differences in the functional evolution of the two clades by fitting an OU model with two adaptive peaks (one for clade GGD and one for clade CAS; ‘GGD–CAS peaks’ model). We also examined a two-peak OU model in which the Crenicichla clade evolved towards an adaptive peak different from all other Geophagine cichlids (‘Crenicichla’ model). As the Crenicichla clade is species-rich, includes ecological roles unique to clade CAS (e.g. piscivores) and exhibit unique ecomorphological characteristics [18,20], it is possible that they have reduced competitive pressure from having adapted to a different functional optimum. We did not test for different optima based on other ecological roles or other factors (e.g. benthivory, substrate sifting, body size, habitat type, etc.) as these occur in multiple clades (e.g. substrate sifting evolved independently in Geophagus in clade GGD and Satanoperca in clade CAS) and would require a detailed reconstruction of these traits that is beyond the scope of these analyses. All models for morphological evolution were fit using the ‘brown’ and ‘hansen’ functions from the ‘ouch’ package in R. The five models were compared using sample size corrected Akaike information criteria (AIC) values (AICc) and Akaike weights [37].

3. Results

(a). Functional morphospace

The first three axes of a phylogenetic PCA of the functional morphology of the 27 cichlid taxa included in this study explained 29.2 per cent, 18.8 per cent and 15.7 per cent of the variation in the data, respectively (see the electronic supplementary material, table S3). Species on the positive extreme of PC1 had lower jaws optimized for a fast, evenly distributed bite and poor suction ability, whereas those on the negative extreme had efficient lower jaw force transmission with less even force distribution and good suction ability (figure 1). Species with higher PC1 scores were generally more elongate in body form (e.g. Crenicichla, Dicrossus and Retroculus), whereas those with lower PC1 scores were relatively deep bodied (e.g. Guianacara and Biotodoma; figure 1). Fishes on the positive extreme of PC2 (figure 1a) had proportionately smaller biting muscles and larger jaw opening muscles, low oral JP and lower pharyngeal crushing potential. Species on the negative extreme of PC2 had protrusible jaws with proportionately larger biting muscles and smaller jaw opening muscles, as well as higher pharyngeal crushing potential. Species on the negative extreme of PC3 (figure 2, lower) had larger jaw opening muscles and efficient force transmission in the expansion of the oral jaws, whereas those on the positive extreme had rapid but inefficient (in force transmission) expansion of the oral jaws, smaller jaw opening muscles and subsequently poor suction ability. The high PC3 scores appear to be associated with taxa possessing comparatively short snouts and heads (e.g. Taeniacara candidi compared with ‘Geophagus’ brasiliensis in figure 1).

Figure 1.

Scores of the first three principal component (PC) axes of feeding functional morphology in 23 species of Geophagini and four outgroup Neotropical cichlid taxa (black circles), based on a phylogenetic PCA using the MCC chronogram. Blue diamonds and red squares indicate individual species scores for clade GGD and CAS, respectively, and dashed lines show the convex hulls of each clade. Text indicates the variables most strongly loading each axis (see the electronic supplementary material, table S3 for PCA results). Photographs are credited to J. Arbour, H. López-Fernández, C.G. Montaña and A. Lamboj. (Online version in colour.)

Figure 2.

A three-dimensional functional phylomorphospace of Geophagini. Points indicate the PC scores across the three critical axes of variation, and lines illustrate phylogenetic relationships between geophagine taxa (CAS, filled circles; GGB, open circles). Point size and colour intensity illustrate depth along the horizontal plane of the image along which PC1 and PC2 vary (e.g. lighter objects are further into the background), whereas PC3 is oriented with the vertical plane of the image. Blue (solid) rings indicates substrate-sifting taxa, red (horizontally hatched) rings indicates small-bodied taxa (less than 10 cm SL maximum body size) and purple (vertically hatched) rings indicates elongate-bodied, pursuit-predators/piscivores. (Online version in colour.)

Clades GGD and CAS overlapped in morphospace in the negative range of PC1 and the mid to positive range of PC2 (figure 1). However, the Crenicichla clade in clade CAS was relatively isolated in a unique portion of morphospace (primarily in the positive extreme of PC1 and somewhat on the negative extreme of PC2), united in possessing fast and highly protrusible oral jaws with gripping bites and poor suction ability (see the electronic supplementary material, table S3; figures 1 and 2). Substrate-sifting taxa from both clades GGD and CAS (Geophagus, Gymnogeophagus, Biotodoma, Mikrogeophagus, Satanoperca and Acarichthys) share similarly low PC1, high PC2 and low PC3 scores and this region of morphospace is maximized for suction feeding ability, i.e. high force transmission efficiency in both lower jaw MA and oral jaw KT, high ST mass and high SI (see the electronic supplementary material, table S3; figure 2, blue solid).

Small-bodied and large-bodied taxa (maximum body size less than or greater than 100 mm SL, respectively, J. Arbour (2011, personal observation) and S. Steele (2012, personal observation); species max SL from Reis et al. [15]) appear to occupy distinct regions of functional morphospace. In figure 2 (red, horizontally-hatched rings), the so-called ‘dwarf’ species (Apistogramma, Taeniacara, Dicrossus, Biotoecus, Mikrogeophagus and Crenicara) and other small-bodied species examined (e.g. Crenicichla sp. ‘Orinoco wallacii’ and Mazarunia mazarunii) were characterized by fast oral jaw movement (resulting in particular from velocity-optimized four bar kinematics) as well as poor crushing ability, JP and AM muscle mass. This ‘small-bodied morphospace’ shows low overlap with larger bodied species, and ‘dwarf’ taxa from both clades GGD and CAS have independently converged in these functional characteristics.

(b). Phylomorphospace characteristics

Clade CAS exhibited much higher multivariate functional disparity than clade GGD (table 1). Under BM simulations, clade GGD exhibited similar, although slightly lower, functional disparity than clade CAS (t = −2.30, p = 0.032; table 1). Comparatively, clade GGD actually exhibited only 35 per cent of the observed functional disparity of clade CAS (t = −20.0, p < 0.001; table 1). Interestingly, clade GGD also showed lower functional disparity (FD_obs = 84.2 versus FD_BM = 117; t = −6.33, p < 0.001) than expected under BM evolution, whereas clade CAS showed higher disparity than expected (FD_obs = 241 versus FD_BM = 129; t = 14.2, p < 0.001). It is therefore unlikely that functional disparity in Geophagini arose through a process of BM evolution.

Table 1.

Observed functional disparity and phylomorphospace metrics contrasted with values generated by Brownian motion simulations. (Mean ± standard deviations for each clade (as well as the ratio between the clade values) were calculated across 1000 posterior distribution chronograms.)

	observed			simulated
	GGD	CAS	GGD/CAS	GGD	CAS	GGD/CAS
functional disparity	84.2 ± 12.3	241 ± 24	0.349 ± 0.033	117 ± 12	129 ± 13	0.909 ± 0.074
morphometric branch length	9.46 ± 0.90	11.7 ± 1.7	0.821 ± 0.100	10.28 + 0.53	10.6 + 0.56	0.970 + 0.037
rate of morphological evolution	0.296 ± 0.034	0.369 ± 0.047	0.820 ± 0.161	0.313 + 0.021	0.306 + 0.021	1.02 + 0.065
lineage density	0.0155 ± 0.0027	0.0118 ± 0.0023	1.33 ± 0.16	0.0176 + 0.0019	0.0152 + 0.0016	1.16 + 0.01

The average rates of functional evolution were 18 per cent lower in clade GGD than clade CAS, despite an expectation under BM evolution of equal rates of morphological evolution between the two clades (t = 0.799, p = 0.433; table 1). Rates of evolution in clade GGD were not significantly different from those expected under BM evolution (t = −1.41, p = 0.18; table 1), but in clade CAS they were greater than expected under BM evolution (t = 4.23, p < 0.001; table 1). Lineage density in clade GGD was higher than in clade CAS (t = 3.52, p = 0.002; table 1), indicating clade GGD is more ‘packed’ in morphospace (see the electronic supplementary material, figure S3, A versus B). Clade GGD was expected to exhibit higher LD even under a process of BM evolution (t = 3.26, p = 0.004; table 1). Interestingly, both clades, GGD and CAS showed lower LD than expected under BM evolution (t_GGD = −2.11, p_GGD = 0.049; t_CAS = −4.20, p_CAS = 0.004), indicating that both clades were more efficient at colonizing morphospace than expected under a neutral process, particularly in the case of clade CAS.

(c). Disparity-through-time analyses

Clade CAS shows an early drop in relative subclade disparity (figure 3b) and an average MDI value of −0.129 ± 0.038 (s.d.). This negative MDI value suggests that this group underwent an early burst of evolution after which subclades remained relatively stable in average functional disparity. By approximately 95 Ma (time estimate based on MCC tree), subclades within clade CAS represented (on average) only 29 per cent of functional disparity in clade CAS as a whole (figure 3b), and subclade disparity remained below BM expectations until approximately 75 Ma. Simulated MDI values were as negative as observed values in only 6.41 per cent of BM simulations (compared with an expected frequency of 50%), suggesting that the observed MDI values were unlikely to have occurred under a BM process. Conversely, clade GGD followed a relatively linear decrease in subclade disparity similar to BM expectations up to approximately 60 Ma, after which it increased (figure 3a). This is consistent with the positive average MDI (0.194 ± 0.067), indicating that relatively young subclades within clade GGD exhibited a higher proportion of modern disparity than expected. Few BM simulations produced relative subclade disparity values as high as those observed during approximately 50–20 Myr ago (figure 3a), although overall a more moderate frequency of these simulations (12.4%) produced MDI values as high as those observed (expected frequency of 50%) compared with clade CAS. Modern functional disparity in clade GGD may have continued to arise during more recent divergence events compared with a strict model of BM evolution.

Figure 3.

Disparity-through-time analysis results for clades GGD (blue, a) and CAS (red, b), with the dark, solid lines indicating the median observed average subclade disparity (relative to modern disparity) and the dashed lines indicating the median simulated (BM) subclade disparity across all 1000 simulations for each of 1000 posterior distribution chronograms. The light, solid lines indicate 100 sampled simulations within the 95 per cent range of BM simulated values. The chronogram (c) illustrates the genus-level divergence times for all geophagine taxa examined in these analyses based on the MCC tree. (Online version in colour.)

(d). Comparison of models for morphological evolution

We examined whether the functional morphological data were best fit by a BM (model BM) or one of several OU models for morphological evolution. We found very strong support for the ‘Crenicichla’ model (see §2) on the first PC axis, with an Akaike weight of 0.999 (table 2) and all other ΔAIC values greater than 10 [37]. We also found weak support for the ‘Crenicichla’ model (wAIC = 0.560 and 0.473) over a BM model (wAIC = 0.299 and 0.329) for PC2 and PC3 (respectively), but strong support for the ‘Crenicichla’ model over the ‘single peak’ model or the ‘GGD–CAS’ model (table 2).

Table 2.

Model fitting results (log likelihood, ΔAIC and Akaike weight) for the comparison of Brownian motion (BM) and Ornstein–Uhlenbeck (OU) models for morphological evolution of each of the three critical PC axes. (Values display the average calculated across all 1000 posterior distribution chronograms; PC2 and PC3 do not have a model with a ΔAIC of 0 as no one model was supported across all 1000 trees. See §2 for description of each of the models.)

axis	model fitting variable	BM	‘single peak’	‘GGD–CAS’	‘Crenicichla’
PC1	logL	−95.8	−97.2	−97.0	−86.2
	ΔAIC	13.6	19.0	21.6	0.0
	wAIC	0.001	0	0	0.999
PC2	logL	−90.7	−90.3	−90.2	−87.2
	ΔAIC	2.5	4.4	7.2	1.2
	wAIC	0.299	0.113	0.028	0.560
PC3	logL	−87.5	−87.1	−86.2	−84.3
	ΔAIC	2.60	4.48	5.70	1.87
	wAIC	0.329	0.128	0.070	0.473

4. Discussion

(a). Ecology, ecomorphology and functional morphology of Geophagini

Variation in feeding biomechanics within Geophagini is congruent with the trophic ecology of the group [20,38], and appears to be associated with both patterns in body shape and size. Our analyses have demonstrated strong patterns of functional convergence between taxa sharing similar ecological roles (e.g. the substrate-sifting genera Geophagus from clade GGD and Satanoperca from clade CAS). In addition, variation in body shape along PC1 of feeding biomechanics (i.e. broad to elongate body morphologies) mirrors the first axes of variation in ecomorphology observed in two previous studies of geophagines based on morphometrics [18,20]. This suggests that previous analyses of ecomorphology in Geophagini have described functionally relevant axes of variation.

Geophagine cichlids exhibit a gradient between predators with fast, gripping bites, using ram-feeding behaviours (e.g. Crenicichla), and species feeding on or sifting through the benthos and relying on forceful suction ability (e.g. Geophagus; figures 1 and 2). This is consistent with the apparent trade-off in diet (between benthic invertebrates and fishes) observed in previous studies of geophagine diet composition [20,38]. This trade-off between fishes with deep bodies and subterminal or ventral mouths feeding primarily from the benthic community (e.g. substrate sifters) versus elongate fishes with terminal mouths feeding epibenthically or in the water column (e.g. piscivores and planktivores) has been observed in cases such as the African rift lake cichlids [39], fishes colonizing post-glacial lakes [40,41] and other examples of radiations in fishes [42]. Thus, the primary axis of biomechanical variation within Geophagini appears associated with benthic versus epibenthic/pelagic foraging, whereas other axes characterize different methods of specializing within these habitat zones, e.g. large-bodied taxa with long snouts and comparably large buccal volumes capable of sifting mouthfuls of substrate such as Geophagus and Satanoperca versus small-bodied taxa with short heads and snouts more capable of picking individual benthic invertebrates from the substrate such as Apistogramma and Biotoecus.

Evolution towards small body size (from a larger bodied lineage) has probably occurred repeatedly within Geophagini (e.g. independently in Mikrogeophagus in clade GGD versus Apistogramma in clade CAS) and other Neotropical cichlid tribes (e.g. Nannacara and Laetacara in Cichlasomatini, Cryptoheros in Heroini). Reduced body size in Geophagini appears to be associated with convergent feeding adaptations relating to fast but weak oral jaw four bar mechanics (figures 1 and 2). The long morphometric branch lengths observed among small-bodied lineages compared with larger bodied lineages (figure 2) and high variability of feeding biomechanics within these taxa suggests that reduced body size may be associated with accelerated morphological diversification rates. More detailed analyses are, however, necessary to confirm this interpretation.

(b). Timing and rates of morphological evolution

Processes of morphological evolution in traits directly affecting feeding performance and ecology have not followed a pattern of neutral, random-walk evolution in Geophagini. Rather, adaptive evolution of feeding biomechanics has been pervasive throughout geophagine evolutionary history. Clade CAS is more functionally disparate than expected under BM as well as compared with clade GGD. This high disparity appears to have been driven by an early burst of morphological evolution during which lineages became well dispersed in functional space [2,5]. By the end of this early burst, after which patterns increasingly correspond to BM expectations, lineages representing the majority of feeding-ecology roles within this group had already diverged (e.g. substrate sifters in Satanoperca and Acarichthys, piscivores in Crenicichla, small-bodied invertebrate pickers in Biotoecus and Apistogramma). Similarly, López-Fernández et al. [18] observed an early burst in body shape evolution between elongate and deep body morphologies within Geophagini. The association between body shape, feeding biomechanics and ecology suggests that patterns of morphological evolution on both these axes were driven by the ecological niches available early in Neotropical cichlid history.

Functional disparity was far lower in clade GGD than CAS as well as low compared with expectations under a random-walk process (table 1), which is consistent with the lower ecological variability in clade GGD. Interestingly, the vast majority of morphological evolution corresponding to modern functional disparity in clade CAS (prior to approx. 95 Ma; figure 3) may have occurred prior to significant lineage diversification within clade GGD (first divergence event within GGD occurs at approx. 90 Ma; López-Fernández et al. [18]). Functional diversification in clade GGD occurred more slowly than in clade CAS and was largely restricted to the region of morphospace occupied by substrate-sifting genera (i.e. Geophagus, ‘Geophagus’, Gymnogeophagus, Biotodoma and Mikrogeophagus). This restricted pattern of evolution is consistent with the observation that approximately 50 Ma †Gymnogeophagus eocenicus fossils are indistinguishable from modern substrate-sifting Gymnogeophagus species [20,43,44]. Based on the timing of the evolution of functional disparity, it is possible that the early burst of evolution and extensive morphospace occupation of clade CAS precluded a similar burst within clade GGD which underwent diversification later, thus limiting the potential ecological variation achievable by clade GGD.

(c). Adaptive peaks and key innovations

Our results indicate that clade CAS inhabits at least one adaptive peak not explored by clade GGD, and that it is characterized by poor suction ability and fast, protrusible, gripping oral jaws consistent with adaptation to predatory ram-feeding [45–47]. These traits are also consistent with the elongate body shape common among all Crenicichla and Teleocichla species, which are biomechanically suited to rapid acceleration but also poor critical swimming speeds and manoeuverability [48,49]. These functional attributes are consistent with those characteristic of adaptive peaks related to piscivory in other families of fishes [50], and species within this clade consume predominantly fishes and similarly evasive aquatic insects [20,22]. All geophagine species not within the Crenicichla clade appear to have been constrained around an adaptive peak characterized by good suction ability but comparably slower jaw mechanics. Most of these suction feeding taxa consume primarily benthic invertebrates or detritus, either by sifting or picking from the surface of the substrate [19,20,38]. This supports an adaptive basis for a trade-off between those geophagine taxa feeding primarily from the substrate and those feeding above it.

The Crenicichla clade encompasses 44 per cent of the species diversity of clade CAS and 35 per cent of the diversity of all geophagine cichlids, exhibits ecomorphological characteristics nearly unique among South American Neotropical cichlids and possesses as much body size and morphological disparity as many other Neotropical cichlid genera combined [18,20,21]. Evolution of the unique combination of feeding biomechanics probably permitted the evolution of a novel feeding strategy (fast-burst predators) among the Geophagini, allowing opportunities for rapid diversification of ecomorphological traits within this niche [51]. The Crenicichla clade may have colonized the predatory, ram-feeding adaptive peak early in Neotropical cichlid history, and this may have excluded some other Neotropical cichlids lineages from filling similar roles; for example, relatively elongate-bodied piscivores in the tribe Heroini are only found in Central America and not in South America [18]. The feeding biomechanics observed in Crenicichla may represent a key innovation [52] and if so it may have contributed to cichlid diversity and to shape the structure of fish communities in South America [38].

(d). Ancient adaptive radiations

Our results indicate that the adaptive underpinnings of ecological divergence in geophagine cichlids are ancient. Considering the important roles these species play in Neotropical ecosystems, aquatic communities in South America probably reflect ecological conditions that have persisted for tens of millions of years. This is consistent with fossil evidence showing that lineages of the most common modern families of Neotropical fishes were present as early as the Early Paleocene and Eocene [43,53–55]. Previous work on geophagine cichlids has supported high initial rates of lineage diversification [17,18], an early burst of body shape evolution [18] and correlations between ecomorphology and feeding ecology [20], all of which are evidence of an adaptive radiation. Functional morphology in Geophagine cichlids revealed an early burst of evolution consistent with these previous analyses. Therefore, patterns of functional evolution within Geophagini are consistent with a continental-scale adaptive radiation, at least within the species-rich clade CAS. Our results reflect the statement by Glor [25] that adaptive radiation must be diagnosed on a sufficiently narrow phylogenetic group as to remain ecologically and evolutionarily relevant; even closely related sister clades (i.e. GGD and CAS) in Geophagini display very different patterns of morphological evolution and should be examined independently.

Adaptive radiation has been implicated as playing a major role in the origin of modern biodiversity [6,7,23,25,56]. However, many studies of adaptive radiation are restricted to relatively recent and isolated systems [8] and some evidence points towards early bursts of morphological evolution being relatively rare from a broad, comparative perspective [5]. Our analyses, as well as that of López-Fernández et al. [18] indicate that Geophagini underwent a mid to Late Cretaceous burst of functional evolution that was associated with the advent of modern ecological groups. In addition to geophagine cichlids, early bursts of morphological evolution have been observed in continental-scale radiations of furnariid birds [12] and phyllostomid bats [57] within the Neotropics alone, as well as within the globally distributed Cetacea [33]. Together, these patterns suggest that it is possible that models of evolution consistent with adaptive radiation are applicable to patterns of diversity and disparity on a broad geographical scale and encompassing long evolutionary times.