Semicircular canals in Anolis lizards: ecomorphological convergence and ecomorph affinities of fossil species

Anolis lizards are a model system for the study of adaptive radiation and convergent evolution. Greater Antillean anoles have repeatedly evolved six similar forms or ecomorphs: crown-giant, grass-bush, twig, trunk, trunk-crown and trunk-ground. Members of each ecomorph category possess a specific set of morphological, ecological and behavioural characteristics which have been acquired convergently. Here we test whether the semicircular canal system—the organ of balance during movement—is also convergent among ecomorphs, reflecting the shared sensory requirements of their ecological niches. As semicircular canal shape has been shown to reflect different locomotor strategies, we hypothesized that each Anolis ecomorph would have a unique canal morphology. Using three-dimensional semilandmarks and geometric morphometrics, semicircular canal shape was characterized in 41 Anolis species from the Greater Antilles and the relationship between canal shape and ecomorph grouping, phylogenetic history, size, head dimensions, and perch characteristics was assessed. Further, canal morphology of modern species was used to predict the ecomorph affinity of five fossil anoles from the Miocene of the Dominican Republic. Of the covariates tested, our study recovered ecomorph as the single-most important covariate of canal morphology in modern taxa; although phylogenetic history, size, and head dimensions also showed a small, yet significant correlation with shape. Surprisingly, perch characteristics were not found to be significant covariates of canal shape, even though they are important habitat variables. Using posterior probabilities, we found that the fossil anoles have different semicircular canals shapes to modern ecomorph groupings implying extinct anoles may have been interacting with their Miocene environment in different ways to modern Anolis species.

Anolis lizards are a model system for the study of adaptive radiation and convergent evolution. Greater Antillean anoles have repeatedly evolved six similar forms or ecomorphs: crown-giant, grass-bush, twig, trunk, trunk-crown and trunkground. Members of each ecomorph category possess a specific set of morphological, ecological and behavioural characteristics which have been acquired convergently. Here we test whether the semicircular canal system-the organ of balance during movement-is also convergent among ecomorphs, reflecting the shared sensory requirements of their ecological niches. As semicircular canal shape has been shown to reflect different locomotor strategies, we hypothesized that each Anolis ecomorph would have a unique canal morphology. Using three-dimensional semilandmarks and geometric morphometrics, semicircular canal shape was characterized in 41 Anolis species from the Greater Antilles and the relationship between canal shape and ecomorph grouping, phylogenetic history, size, head dimensions, and perch characteristics was assessed. Further, canal morphology of modern species was used to predict the ecomorph affinity of five fossil anoles from the Miocene of the Dominican Republic. Of the covariates tested, our study recovered ecomorph as the single-most important covariate of canal morphology in modern taxa; although phylogenetic history, size, and head dimensions also showed a small, yet significant correlation with shape. Surprisingly, perch characteristics were not found to be significant covariates of canal shape, even though they are important habitat variables. Using 2017 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited.
The vestibular system has, however, been little explored outside the Mammalia. While the semicircular canals are physiologically and anatomically homologous in all vertebrates, we cannot assume that the relationship between form and function in mammals will hold true for other taxa, especially given considerable morphological differences between mammals and other amniote groups. Extrapolation from mammals is particularly problematic for studies wishing to reconstruct the ecology of non-mammalian fossil species [21,[23][24][25][26]. Recent studies looking at the morphology of the semicircular canals with respect to ecology in amphibians [27], squamates [28,29] and birds [30,31] have begun to expand our knowledge of the vestibular system beyond mammals, though much work is still needed to fully understand the system from a greater evolutionary and ecological spectrum.
Anolis lizards (Dactyloidae) represent a unique opportunity for furthering such research. Greater Antillean anoles originated ca. 65 Ma, diversifying throughout the Caribbean and neotropical mainland [32]. Within Anolis, six ecomorph types have evolved independently on each of the four Greater Antilles islands (with several exceptions), with each ecomorph encompassing a specific suite of anatomical (e.g. short versus long limbs), ecological (e.g. tree trunk versus branches), and behavioural (e.g. locomotion, territoriality) characteristics (reviewed in [33]): trunk-ground (TG) ecomorphs inhabit lower tree trunks, using infrequent, rapid descents to the ground where they capture prey; trunk-crown (TC) ecomorphs occupy the upper reaches of the tree, navigating the complex 3D canopy with a high rate of movement; trunk (Tr) ecomorphs occupy the trunk area between TG and TC ecomorphs with some overlaps, and are fairly active locomotors; crown-giant (CG) ecomorphs occupy similar habitats to TC ecomorphs, yet are substantially larger, and move more slowly; twig (Tw) ecomorphs occupy

Data acquisition and landmarks
Various methods for measuring the complex structure of the three semicircular canals have been used to date. Traditional morphometric approaches, in the form of linear and angular measures of size and orthogonality, have been used extensively in the past [2,6,7,11] with the benefit of being easily comparable across studies. However, these measurements struggle to completely capture the full shape variation in the canals owing to their complex curvature. Geometric morphometrics (GM; [41,42]) has been used increasingly to overcome this shortcoming, with landmark [9,16,28,35] and semilandmark [19,43,44] approaches, particularly the latter, capable of capturing far more morphological variation than standard morphometrics. GM methods do, however, vary between studies and are thus less easily compared across studies and broader taxonomic groups. For morphometric approaches, digital thresholding and segmenting of micro computed tomography (µCT) data can introduce significant variation in canal lumen thickness [45], making measuring and digitizing the canal surface error prone. Instead, using a centreline through the lumen overcomes this potential error as it is not affected by threshold values [43,45]. This is the approach that we took.
Specimens were µCT scanned using a variety of imaging systems and settings (electronic supplementary material, table S1) and the semicircular canals manually segmented and 3D rendered using MATERIALISE MIMICS ® software. 3D landmarks were derived from the centreline of the semicircular canals, calculated using the 'medcad' module in MIMICS ® , and then manually adjusted to optimize the position of the centreline through the lumen. This centreline was then split into four segments-the three canals (anterior, posterior and lateral) and the crus commune (figure 3a)-and exported as 3D coordinates describing a curve. The centreline curves were then resampled using the RESAMPLE executable [46] so that the three canals were each described by 28 equally-spaced semilandmarks, and the crus commune by three semilandmarks. These semilandmarks were anchored by five landmarks positioned at the junction of the anterior and lateral canals, the posterior and lateral canals with the vestibule, and the junction of the crus commune, resulting in 92 landmarks in total (figure 3b).

Data analysis
Landmark coordinate data were aligned by Procrustes superimposition, allowing semilandmarks to slide along their tangent directions in order to minimize bending energy [47,48] using the R statistical environment v. 3.2.3 [49] and the package geomorph v. 2.1.4 [50]. The resulting Procrustes residuals were used as shape variables in the subsequent analyses. Alignment was done on both the full shape variable dataset (n = 131) and a phylogenetic subset (n = 41). Since the full dataset represented ontogenetic series of each species, the phylogenetic subset was represented by the largest adult male from each species paired with the phylogeny of Gamble et al. [51] (figure 3c). As a previous study found significant sexual dimorphism in some Anolis species [52], comparing all males avoids a potential source of bias. Partitioning the data was necessary to investigate the role phylogeny might play in determining canal shape as no current methods are available that would account for ontogenetic variation in phylogenetic comparative analyses.
Principal    centroid size, performed on PC scores using the 'phyl.resid' function of phytools v. 0.5-10 [55,56]. Centroid size is a measure of size calculated as the square root of the sum of squared distances of a set of landmarks from their centroid [42]. To visualize shape changes throughout the PCA morphospace, partial warps were used to generate maximum and minimum shape warps along each principal component (PC) axis by back-transformation through the eigenvectors. To determine whether ecomorphs occupy different regions of morphospace (and thus have significantly different canal morphologies), and to test the effect of size and head proportions on canal shape, analysis of covariance (ANCOVA) and phylogenetic generalized least squares (PGLS) were performed on the phylogenetic dataset using the 'procD.lm' and 'procD.pgls' functions respectively [57] of geomorph, with pairwise comparisons tested using 'advanced.procD.lm' [50]. These functions perform statistical assessment of the terms in the model using Procrustes distances among specimens, rather than explained covariance matrices among variables, and are thus suitable for multivariate datasets [57,58]. Three log-transformed size metrics were used: semicircular canal centroid size, skull length (measured between the premaxilla and the dorsal margin of the foramen magnum) and skull width (measured between the paraoccipital processes). The relationship between the size metrics was also explored using linear regressions. Head proportions were determined by taking the ratio of skull length : width. In addition, we investigated the relationship between canal shape and habitat use, represented by perch height and diameter ( [50] and J. B. Loses 1988-2005, unpublished)-the two most frequently reported habitat variables-by ANCOVA and PGLS.
Finally, a canonical variate analysis (CVA) with cross-validations was run using the 'CVA' function of the package Morpho v. 2.3.0 [59] to explore the morphological shape variables that maximize betweenecomorph-group variance relative to within-group variance, and to predict the potential ecology of the fossil anoles. Prior to running the CVA, a PCA was performed on the full extant dataset (excluding unique species) and the first 40 PC axes representing 99% of the variation were extracted; this reduction in dimensionality was done to ensure that the number of shape variables (n = 40) was less than the number of individual specimens (n = 99) [60] and to remove minor components of shape variance that might be attributable to error. In addition, the full specimen dataset was used to take into consideration both specific and ontogenetic variation to compare against the fossil specimens and to increase the power of the test by incorporating a larger sample size. Within the CVA morphospace, 95% confidence intervals (CI) were generated around each of the modern ecomorph groups. The unique and fossil specimens were then projected into this morphospace using the canonical variates. As CVA is not a rigid ordination and the resulting morphospace may deviate from a Euclidean space, we use Mahalanobis distances in subsequent analyses to correct for any distortions in shape-space [60]. The posterior (typicality) probability of ecomorph-group membership for fossil and unique specimens was assessed by calculating the Mahalanobis distances of each specimen to the mean of each ecomorph. This distance was then compared to within-ecomorph-group distances which had been resampled 10 000 times [61][62][63][64][65]. If the distance between a specimen and group mean was greater than 95% (p < 0.05) of within-group distances, we reject the null-hypothesis that it belongs to that ecomorph group [61,[65][66][67][68]. Further, loglikelihood estimations were also calculated to allow comparison with previous work [34]. To visualize shape changes throughout the CVA morphospace, partial warps were used to generate maximum and minimum shape warps along each canonical variate (CV) axis [60].

Patterns of shape variation
PCA (figure 4a,b) shows that PC1 (40.2% of variation) largely represents changes in anterior and lateral canal morphology. Moving from PC1 positive to PC1 negative, there is a trend for the canals to become more rounded and anterodorsally shortened. PC2 (11.0% of variation) represents moderate changes in all three canals, with a PC2 positive to PC2 negative shift showing rounding of the anterior-most section of the anterior canal, more torsion (out-of-plane curvature) of the lateral canal, and less torsion of the posterior canal. PC3 (10.5% of variation) represents changes in the anterior and posterior canals, with a transition from PC3 positive to PC3 negative showing increased curvature and deepening of the posterior canal and reduction of the lateral aspect of the anterior canal. MANOVA results show that ecomorphs at not distinct on PC1, but they do significantly separate over subsequent PCs (electronic supplementary material, table S2).  Visually, there is significant overlap between ecomorph groupings along PC1 and PC2, with GB anoles occupying most of PC1. PC3 separates the three 'trunk' ecomorphs from the Tw and CG ecomorphs. When the shape data were corrected for size and phylogeny, the PC morphospace is minimally altered. The unique species are widely distributed across morphospace, overlapping with most ecomorph grouping (PC2 versus PC1) and falling outside the variation enclosed by the ecomorphs (PC2 versus PC3). All fossil specimens fall along the positive end of PC1, in the GB area of morphospace, which represents flattening and anterodorsal elongation of the anterior canal. Furthermore, three fossil specimens (M-1153, M-3410, USNM580060) overlap with multiple ecomorph groupings along PC3. Fossil specimens M-525 and OAAAA appear to fall beyond the morphologies of all living taxa sampled.

Predictors of shape
Phylogeny was found to have only a weak influence on semicircular canal morphology (K = 0.58), though permutation found this influence to be greater than expected from random (p = 0.0034). Mapping of the phylogeny onto morphospace (figure 5) shows extensive overlapping of branches through morphospace, indicating convergence towards similar semicircular canal morphologies.
ANCOVA (table 1) found that ecomorph is a moderate and significant predictor (R 2 = 0.36, p < 0.001) of canal shape. Multivariate pairwise post hoc tests found all ecomorphs to be significantly different (p < 0.05) from one another except Tr and TG (p = 0.300, electronic supplementary material, table S3). A weak but significant relationship also exists between canal centroid size and canal shape (R 2 = 0.11, p = 0.001) and their interaction (R 2 = 0.11, p < 0.001), as well as head proportions and canal shape (R 2 = 0.03, p = 0.041) and their interaction (R 2 = 0.23, p < 0.001) (table 1). Further, PGLS (table 2) found ecomorph to be a significant predictor of canal shape, though the effect was less strong (R 2 = 0.20, p < 0.001). This reduction in the correlation coefficient indicates an interaction between phylogeny and ecomorph and that ecomorph groupings are not entirely independent of phylogeny. PGLS (table 2) also reveals a relationship between canal centroid size and shape (R 2 = 0.19, p < 0.001) and an interaction between ecomorph and centroid size (R 2 = 0.19, p < 0.001). Head proportions remains weak, yet significant (R 2 = 0.08, p = 0.006), though PGLS reveals a much stronger interaction between head proportions and ecomorph (R 2 = 0.3, p = 0.002). There was no significant relationship between either perch height or diameter and canal shape, with and without phylogenetic correction (tables 1 and 2).

Ecomorph differences
The CVA biplots (figure 6) and posterior probabilities (table 3)            These results indicate that unique species display greater variation in semicircular canal shape than what is encompassed by the ecomorph categories. Our log-likelihood calculations do, however, assign all unique species to the defined ecomorphs, although these assignments are also inconsistent within and between species (electronic supplementary material, table S5).
All the fossil specimens fall either outside or just on the margin of the 95% CI of the modern ecomorphs, much like the unique species ( figure 6). Our posterior probabilities support this: all fossil specimens are highly unlikely to belong to any modern ecomorph group (table 4). This contrasts with the log-likelihood tests which assign each fossil to the 'closest' ecomorph group regardless of actual morphological distance (table 4). Moreover, log-likelihood tests found only one instance of correspondence with the ecomorphs inferred by Sherratt et al. [34]: OAAAA, which is assigned to TC. Visually, all fossils are broadly distributed around the first three CV axes, with some being in extreme regions of morphospace ( figure 6). However, M-525 generally falls close to TG (CV1 vs CV2) and Tr (CV2 vs CV3), even though it statistically falls outside their 95% CIs (table 4).

Convergence of semicircular canal shape
Our results support the hypothesis that phylogenetically disparate Anolis species have convergent semicircular canal morphologies, allowing them to navigate similar ecological niches. Of the covariates we tested (ecomorph, size, head proportions), we found ecomorph grouping to be the best determinant of canal shape, even when phylogeny is accounted for (25-35%, tables 1 and 2). This is remarkable considering the taxa we examined constitute only a single genus which inhabits a relatively constrained geographical and ecological space. However, given the role semicircular canals play in coordinating fast and complex movements in 3D, and the differences in ecology and locomotor behaviour among the Anolis ecomorphs, this result aligns well with the body of work supporting a relationship between semicircular canal morphology and locomotion [5,7,9,22,27,28].
Though differences among the ecomorphs are subtle, CVA posterior probabilities found that all ecomorphs form significantly different groupings, though our ANCOVA post hoc pairwise comparisons did not find a statistical difference between TG and Tr ecomorphs (electronic supplementary material, table S3). While CV1 and 2 do not discriminate all groups, all axes of variation must be considered to properly establish group separation [67]. Why our CVA results and post hoc tests do not fully align is uncertain-both statistical tests use similar non-parametric methods, though use different distant measures (Procrustes [69] versus Mahalanobis [61]). Of all the groups, Tr and TG ecomorphs are certainly the most similar-demonstrated by their overlap in CVA morphospace (figure 6), which may reflect locomotor/behavioural similarities.
Both CG and Tw ecomorphs display more torsion of the anterior canal than those of the other ecomorphs, separating significantly along CV1 (figure 6a): CG and Tw ecomorphs are the two groups that generally run the least, but also have to negotiate extremely complicated 3D environments. In mammals, increasing out-of-plane torsion of a canal may increase sensitivity to rotations out of the canal's major plane of motion detection [64]; thus such a morphology in CG and Tw ecomorphs may potentially reflect the coordination required to negotiate their complex environment. Further, CV3 groups the three most arboreal ecomorphs: TCs alongside CG and Tw based on increased circularity and length of the lateral canal ( figure 6). Both increased canal circularity [5] and length [3] have been associated with greater canal sensitivity and agility in mammals. These potential increases in sensitivity of both anterior and lateral canals may represent adaptations to the specialized arboreal niches of these three ecomorphs-CG, Tw and TC-occupying the complex upper reaches of the canopy, requiring greater sensitivity to movements. For the remaining ecomorphs, out of plane sensitivity may not be as essential to locomotor performance. For Tr and TG ecomorphs the trunk provides a broad, uncomplicated surface on which locomotion is much easier [70][71][72], requiring less refined balance. The perch diameter for GB ecomorphs is indeed relatively much more narrow [33,73] and complex, though the consequences of falling from grass or a bush are far less severe than falling from the tree canopy as in the higher dwelling ecomorphs. Perhaps these relaxed locomotory constraints result in less drastic semicircular canal specialization among Tr, TG and GB ecomorphs.
Generally, however, the ecomorphological signal we found does not fully explain variation in semicircular canal shape. Analysis of canal shape and perch height and diameter returned non-significant results (tables 1 and 2), despite both being correlated with ecomorph [33]. This finding was unexpected given the importance of balance during locomotion on narrow perches, and the assumed consequences of falling from high perches. The perch data included here are from a different population than our morphological data; perhaps this limitation introduced sufficient error into our analysis to confound the relationship. Further work is needed. Other behavioural characteristics may also be associated with canal shape variation, such as locomotor performance over varied substrates and/or head rotational velocities [19], and we encourage collection of such data. The remaining variation in canal morphology may also be the result of morphological 'noise' introduced by the skull. As the morphology of the semicircular canals must be accommodated by the skull, there may be trade-offs with the other functional requirements of the skull-such as the brain, the feeding apparatus and other senses of sight and hearing. Reduced penalties for locomotory performance in less arboreal ecomorphs may release the skull to accommodate these other vital functions. Although our results do not indicate canal shape is strongly influenced by head proportions (tables 1 and 2), recent explorations of Anolis skull morphology using geometric morphometric techniques have established differences in skull shape between ecomorphs [52,74]. Perhaps the semicircular canals are being influenced by covariation of the skull, but our head proportion ratio was not sensitive enough to capture it. Comparisons between semicircular canal shape and multidimensional skull shape should be an interesting route of future inquiry that might reveal additional influences on vestibular anatomy.

Role of phylogeny and size
We found a small yet significant relationship between phylogenetic relatedness and semicircular canal morphology (figure 5; tables 1 and 2). This weak phylogenetic signal may be the result of repeated convergent evolution for which anoles are famous [33], though similarly weak yet significant phylogenetic signals are consistent across studies dealing with other taxonomic groups [28,[35][36][37]. It is also likely that we are simply dealing with limited divergences as we are working within a single genus-previous studies have generally compared broader taxonomic groups [28,[35][36][37]. Based on these significant phylogenetic signals, some authors have suggested using the inner ear as a source of phylogenetic characters [75,76]. However, the results of our study demonstrate that while semicircular canal morphology is related to phylogenetic history, size and ecology are more important factors (tables 1 and 2) and any phylogenetic analysis based on such characters would be unreliable. Billet et al. [77] concluded similarly in their phylogenetic analysis of litopternan petrosal and inner ear characters, finding a potentially confounding allometric signal.
Skull length and width (and canal size) is correlated with semicircular canal morphology in Anolis and our results also show that it covaries with ecomorph (electronic supplementary material, table S6). This interaction suggests differences among the allometric shape trajectories of the six ecomorph groups. Previous studies have found that canal size appears to scale with negative allometry, such that smaller animals have relatively larger canals [7,9,37,38]. Some have postulated that smaller animals experience relatively greater angular accelerations of the head than do large animals [7,9,78] and that canal sensitivity is tied to canal radius, suggesting that larger canals are more sensitive to rotation [3]. We found that all three size metrics were highly correlated with canal morphology (tables 1 and 2; electronic supplementary material, table S6), though negative (or positive) allometry cannot be determined when the response variable (canal shape) is multivariate. However, regression of log centroid size on skull length in our dataset found evidence of strong negative allometry (slope = 0.63; 95% CI = 0.54-0.71) which is in keeping with prior studies. Further analyses exploring the allometric variation in our data will be the subject of future publications.

Affinities of fossil anoles
Although further research is needed to determine other factors that may covary with semicircular canal morphology, the significant relationship between ecomorph and canal shape in extant Anolis species enabled us to explore the palaeoecology of fossil taxa. Using posterior (typicality) probabilities, we find that the semicircular canal shapes of all five fossils are significantly different from modern ecomorph groupings (table 4), and that all five also differ from each other (figures 2, 4 and 6). It is not unreasonable for the fossil taxa to differ from modern morphological patterns: Anoles probably first reached Hispaniola in the late Eocene approximately 40 Ma [34], so these 20 Ma Miocene fossils probably represent an intermediate period of diversification between Eocene and modern anoles. Further, the ecological context of Miocene anoles was probably different from the modern Antillean ecosystems. Though little is known about the forest ecosystem structure of the Antilles during the Miocene, Hymenaea protera, the amber forming tree in which the fossils are contained, is more closely related to the African Hymenaea verrucosa than the modern Antillean species [40,79]. Differences in floral composition in the Miocene may have influenced how extinct anoles were navigating their island environment, meaning semicircular canal shape may have been under different selective pressures.
Using a log-likelihood approach (table 4), the fossil anoles are sorted into modern ecomorph groupings, however, only OAAAA matches the predicted groupings of Sherratt et al. [34] who used external morphological features to define ecomorphs. It is possible that the discrepancy between the two studies is a result of quantifying different anatomical structures. Taphonomic distortion may also be an unavoidable factor in fossil specimens. Of the five fossil anoles, two fall into an extreme region of PCA morphospace (figure 4), which could imply taphonomic distortion. However, close inspection of the fossils finds that only M-525 has any noticeable deformation of the basicranium (lateral compression, figure 2). Therefore, we do not expect taphonomic distortion to be causing these differences between studies. Alternatively, the discrepancy between our study and Sherratt et al. [34] may be owing to mosaic evolution: the vestibular system may responded differently to selective pressures than other ecomorphological traits (e.g. limb length, digit length, subdigital lamellae) resulting in varying rates of morphological change [80].

Conclusion
Here we demonstrate that the classic ecomorph definitions of Anolis of the Greater Antilles are supported by inner ear morphology, with each ecomorph possessing a distinctive semicircular canal shape. We find that, of the covariates we tested, ecomorph is the single-most important covariate of canal morphology, although phylogenetic history, canal size and head proportions are also significantly correlated with canal shape. Surprisingly, we were unable to find any correlation between canal shape and perch variables; this result may suggest that canal shape is not influenced by where anoles live, but rather how they locomote. Still, much of the morphological variance seen in our sample remains unexplained and further work is required to tease out other ecological, behavioural, and/or anatomical characteristics that may covary with semicircular canal morphology. Using the more conservative metric of posterior (typicality) probabilities, we were unable to assign fossil anoles to modern ecomorph groups based upon semicircular canal shape. Our results indicate that the semicircular canals of these extinct anoles are morphologically different from modern Anolis ecomorphs, suggesting fossil taxa may have been interacting with their Miocene environment in different ways to modern Anolis species.
Data accessibility. All data and code for analyses are available in the Dryad data repository at the following link: http:// dx.doi.org/10.5061/dryad.8s586 [81].