Environ Biol Fish (2015) 98:713–724 DOI 10.1007/s10641-014-0306-z

Morphology predicting ecology: incorporating new methodological and analytical approaches

Nathan R. Franssen & Christopher G. Goodchild & Donald B. Shepard

Received: 6 December 2013 /Accepted: 24 June 2014 /Published online: 9 July 2014 # Springer Science+Business Media Dordrecht 2014

Abstract Associations between the morphology of an- to characterize ecologically relevant morphological var- imals and their ecology have contributed to our under- iation. Furthermore, we compared morphometric ap- standing of phenotypic diversity by helping to relate proaches employing two analyses (partial Mantel test form and function. Most early studies on fishes used and Phylogenetic Canonical Correlation Analysis traditional measurements of linear distances on the body (PCCA)) that test for correlations among data sets while or fins to quantify morphological variation among taxa. explicitly accounting for phylogenetic relationships. More recently, geometric morphometric analyses have Traditional morphology and body shape showed similar gained popularity for assessing phenotypic shape varia- correlations with habitat use in all analyses. In contrast, tion. Along with new methodologies for quantifying only traditional morphology was correlated with diet; morphological variation, researchers have become in- however, this was only revealed by the PCCA. Our creasingly aware of the influence of phylogeny on mor- findings indicated the taxonomic span of species under phological and ecological traits. Our study, which study and the statistical treatment of data are important spanned seven cyprinid genera, assessed the abilities factors to consider when choosing between traditional or of traditional and geometric morphometric approaches geometric morphometric approaches. In addition, a bet- ter understanding of phylogenetic relationships will im- prove our ability to establish associations between mor- Electronic supplementary material The online version of this phology and ecology. article (doi:10.1007/s10641-014-0306-z) contains supplementary material, which is available to authorized users. . . * : Keywords Cyprinidae Ecomorphology Geometric N. R. Franssen ( ) C. G. Goodchild . Department of Biology, University of Oklahoma, morphometrics Phylogenetic canonical correlation Norman, OK 73019, USA analysis . phytools e-mail: [email protected]

N. R. Franssen Department of Biology and Museum of Southwestern Introduction Biology, University of New Mexico, Albuquerque, NM 87131, USA Relationships between morphology and ecology of an- C. G. Goodchild imals have been recognized for centuries, dating to early Department of Marine Sciences, University of New England, Sanskrit writings on the morphology of fishes from Biddeford, ME 04005, USA different environments and to Aristotle’s “Historia Animalia” (Matthews 1998). In modern ecology, Carl D. B. Shepard Department of Biology, University of Central Arkansas, Hubbs was one of the first to summarize relationships Conway, AR 72035, USA between morphology and ecology in a chapter on the 714 Environ Biol Fish (2015) 98:713–724 morphological correlates of hydrology in fishes (Hubbs In parallel with the advent of new methodologies for 1941). Application of multivariate techniques applied to assessing morphological variation, researchers have be- morphology of for taxonomic purposes in the come increasingly aware of the importance of accounting 1970s was soon followed by use of these approaches to for phylogenetic nonindependence in comparative stud- predict ecological relationships, giving rise to the disci- ies (Felsenstein 1985; Harvey and Pagel 1991). Species pline of “ecomorphology”. Noteworthy early contribu- sharing a common ancestor more recently tend to be tions to the field were those of Findley (1973, 1976)for moresimilartoeachotherthantheyaretomoredistantly bats, Gatz (1979a, 1979b) for fishes, Ricklefs and Cox related species and thus, interspecific data are not likely (1977) for birds, and Ricklefs et al. (1981) for lizards. to be independent. The goal of most ecomorphological After these seminal contributions, numerous studies studies is to causally link morphological variation to followed (e.g., Winemiller 1991; Douglas and Mat- organismal ecology (i.e., structure and function to envi- thews 1992; Winemiller 1992; Winemiller et al. 1995) ronmental context; Wainwright and Reilly 1994;Norton in which an ecomorphological approach was used to et al. 1995). However, two traits (e.g., morphology and predict or compare ecological relationships within fish microhabitat use) might be correlated across a group of communities (reviewed in Matthews 1998 pp 403–410). species due to phylogeny rather than ecological factors Essentially all of the ecomorphological studies of repeatedly shaping phenotypes of species (Felsenstein fishes cited above, and numerous others since (e.g., Piet 1985; Harvey and Pagel 1991; Freckleton et al. 2002). 1998; Pouilly et al. 2003;Herler2007;Ibañezetal. Therefore, assessing and accounting for variation due to 2007; Hoagstrom and Berry 2008;Reechtetal.2013), phylogeny is of key importance to distinguish pattern used traditional morphometric measurements of linear (correlation) and process (causation) in distances on the body or fins, more or less following ecomorphological relationships (Wainwright and Reilly Hubbs and Lagler (1941). Overall, these measurements 1994; Norton et al. 1995). To this end, several analytical appear to capture useful information about the ecology methods have been developed to quantify the influence of fish species in communities (e.g., microhabitat asso- of phylogeny on trait evolution, typically termed phylo- ciations and dietary preferences; Matthews 1998). genetic signal, and for analyzing different types of data in Nonetheless, multivariate analysis of these measures a phylogenetic comparative framework (Felsenstein presented several statistical issues, most notably uncer- 1985;Grafen1989; Maddison 1990; Freckleton et al. tainty of how to control for allometric relationships, 2002;Blombergetal.2003). In tandem, many software deflated statistical power for distinguishing shape, and programs and statistical packages have been developed to the inability to graphically represent shape because geo- perform these analyses (Martins 2004; Paradis et al. metric relationships among variables are not retained 2004; Revell and Harrison 2008;Maddisonand (Adams et al. 2004; McCoy et al. 2006). These concerns Maddison 2011; Revell 2012), including some specific led to the development of a novel approach to quantify to geometric morphometric data (Adams and Otárola- morphology termed geometric morphometrics (Rohlf Castillo 2013). and Marcus 1993a, 1993b), also commonly called anal- Given the diversity of available methodological and ysis of shape. This approach was first described by analytical approaches, appropriately incorporating tradi- Strauss and Bookstein (1982), in which form is summa- tional and novel morphometric analyses is of foremost rized by documenting as cartesian coordinates the posi- interest in ecomorphological endeavors (Norton et al. tion of a series of homologous, repeatable landmarks 1995). Here, we compared different ways to quantify (e.g., tip of snout, origin or insertion of dorsal fin) on the and analyze interrelationships among morphology, bodies of fishes (typically viewed laterally). Geometric ecology, and phylogeny. To do so, we combined morphometric analyses have been used extensively in previously published morphological and ecological ecology (Breno et al. 2011; Colborne et al. 2013; data from Douglas and Matthews (1992), newly gener- Franssen et al. 2013) as well as in systematics (Rohlf ated geometric morphometric data, and phylogenetic 1998) to characterize general shapes of fishes. Due to its analysis of mitochondrial DNA sequences. Douglas prevalence in such studies, the geometric morphometric and Matthews (1992) empirically assessed the ability approach has been touted for initiating a “revolution” in of morphological measures to predict ecology in a fish morphometric analysis (Rohlf and Marcus 1993a, assemblage. They used a series of traditional morpho- 1993b; Adams et al. 2004). logical measurements for body and fin structures to Environ Biol Fish (2015) 98:713–724 715 predict habitat use and diet for 17 native fishes from five Douglas and Matthews (1992). We also quantified geo- families in the Roanoke River, Virginia, USA. They metric morphometric variation using specimens collect- found that multivariate morphological analysis effec- ed from the Roanoke River in 1978, and inferred phy- tively sorted species by family, suggesting an influence logenetic relationships of species using mitochondrial of phylogeny, and that morphology was related to mi- DNA sequences. crohabitat use within a subset of eight minnow species (cyprinidae; Douglas and Matthews 1992). Because Traditional morphology Douglas and Matthews (1992) used traditional morpho- logical measures, we generated geometric morphomet- Thirty-five traditional morphological measurements on ric data for the same eight minnow species that demon- the body and fins were taken by William J. Matthews strated morphological-ecological correlations (Luxilus (Douglas and Matthews 1992) from ten individuals of albeolus, Campostoma anomalum, Luxilus cerasinus, each species (Fig. 1; Appendix S1). To size-correct Clinostomus funduloides, Nocomis leptocephalus, morphological measurements for all species, we

Lythrurus ardens, Chrosomus oreas, Rhinichthys Log10-transformed all measurements, regressed each atratulus) using specimens originally collected from variable against Log10-transformed Standard Length the Roanoke River in the 1970s. We then combined (SL), and retained the residuals (Collar et al. 2005; our new geometric morphometric data with the tradi- Collar and Wainwright 2006). For each species, we tional morphology, diet, and habitat use data (Surat et al. calculated the mean residual value for each morpholog- 1982; Douglas and Matthews 1992) to examine differ- ical measurement and used these size-corrected means ences between the two different morphometric ap- to construct a pairwise dissimilarity matrix using taxo- proaches to predict the ecology of fishes. Although nomic distances in NTSYS v.2.2 (Rohlf 2009). comparisons between traditional and geometric mor- phometrics have been made previously (e.g., Body shape Maderbacher et al. 2008), our study aimed to explore relationships between morphology and species’ ecology Body shape of each species was quantified using geo- generated by these disparate methodologies. Further- metric morphometric analyses in the tps software (Rohlf more, given that phylogeny can be a major influence 2004). We digitally photographed a total of 130 individ- on both morphology and ecology (Eggleton and Vane- uals (mean=16.25 per species, range: 3–20; Wright 1994; Freckleton et al. 2002;VittandPianka Campostoma anomalum had the smallest sample size) 2005; Losos 2011), we examined morphometric ap- on the left lateral side with a reference scale. We ran- proaches employing two analyses that test for correla- domized the order of the digitized photographs using tions among data sets while accounting for phylogenetic tpsUtil (Rohlf 2004) to reduce potential bias associated relationships. These two analytical approaches consid- with the sequence in which specimens were ered phylogeny in different ways and thus, we also photographed and subsequently landmarked. On each compared results between analyses to gauge any advan- specimen, 10 homologous landmarks were set on dis- tages or shortcomings with respect to their use in tinct anatomical points (Fig. 2). We resized landmark ecomorphological studies. coordinates using the reference scale and aligned all species together using least-squares superimposition to remove the effects of scale, translation, and rotation with Materials and methods the program tpsRelw (Rohlf 2004;AppendixS1). We calculated relative warps and uniform components (i.e., Study species and data sources weight matrix, hereafter referred to as shape variables, n=16) as well as centroid size (the square root of the We compiled traditional morphological, habitat use, and sum of the variances of the landmarks about that cen- diet data for eight species of cyprinids (Luxilus albeolus, troid in x- and y-directions, a measure of fish size) using Campostoma anomalum, Luxilus cerasinus, tpsRelw (Rohlf 2004). To control for allometric varia- Clinostomus funduloides, Nocomis leptocephalus, tion in shape data in a manner consistent with traditional Lythrurus ardens, Chrosomus oreas, Rhinichthys morphological methods, we regressed each shape vari- atratulus) from the Roanoke River, VA, USA, from able against Log10 (centroid size), saved the residuals, 716 Environ Biol Fish (2015) 98:713–724

Fig. 1 Traditional morphological 1 characters measured on all species 2 by Douglas and Matthews (1992): 3 (1) Standard Length, (2) pre- 4 12 dorsal fin length, (3) tip of snout to maximum depth of body, (4) 13 11 head length, (5) head depth, (6) 5 18 20 pre-pectoral fin length, (7) pre- 10 16 pelvic fin length, (8) pre-anal fin 9 14 length, (9) length of pectoral fin 6 17 15 base, (10) pectoral fin length, (11) 7 19 body depth, (12) dorsal fin height, 8 (13) length of dorsal fin base, (14) length of pelvic fin base, (15) pelvic fin length, (16) length of anal fin base, (17) anal fin height, (18) caudal peduncle depth, (19) 21 caudal peduncle length, (20) 22 vertical span of caudal fin, (21) 23 snout length, (22) snout to occiput, (23) eye diameter, (24) 24 27 upper jaw length, (25) lower jaw length, (26) suborbital distance, 25 26 (27) postorbital distance, (28) distance between nares, (29) pre- naris length, (30) interorbital distance, (31) head width, (32) maximum body width, (33) caudal peduncle width. 29 Characters of height of opened mouth and gape width not pictured 28 30 31 32 33

and calculated the mean value for each species. We used predominant substrate size, water depth, and position in these size-adjusted shape variable means to construct a the water column were collected from several seine pairwise dissimilarity matrix based on taxonomic hauls within mesohabitats (e.g., pool, riffle, run); sample distances. sizes ranged from 363 to 2,395 individuals per species. Each habitat variable was divided into categories based – – Habitat use on natural breaks: water velocity (0 25, 26 75, and ≥76 cm/s), substrate size (≤5, 6–18, 19–33 cm, and – – ≥ We obtained habitat use data for the eight species from solid), water depth (0 35, 36 75, and 76 cm), and Douglas and Matthews (1992). Data on water velocity, position in water column (benthic, non-benthic). Habitat use for each species was based on percent occurrence within each category (Appendix S1). We arcsin-square- root transformed the proportion of occurrences in each category and constructed a pairwise dissimilarity matrix using taxonomic distances.

Trophic ecology

We quantified the trophic ecology of each species using Fig. 2 The 10 homologous landmarks set on the eight species of cyprinids to assess geometric morphological variation (i.e., shape 11 diet categories from Douglas and Matthews (1992); variation) among taxa sample sizes ranged from 20 to 120 individuals per Environ Biol Fish (2015) 98:713–724 717 species. For each diet category, the percent occurrence runs using the program AWTY (Wilgenbusch et al. in the stomachs of individuals of each species was 2004;Nylanderetal.2008) and by viewing plots of— quantified (Appendix S1). We arcsin-square-root trans- lnL and parameter values in Tracer v.1.5 (Rambaut and formed the proportion of occurrences of each diet cate- Drummond 2007). We discarded trees sampled before gory and constructed a pairwise dissimilarity matrix burn in (20 %), combined post-burn in trees from the using taxonomic distances. two independent runs, and generated a Bayesian con- sensus phylogram. Most nodes in the consensus tree had Phylogenetic relationships posterior probabilities of 1.0 except one node was 0.98 and one was 0.93. A Bayesian posterior probability Because the ecomorphological analyses we performed ≥0.95 is typically considered strong nodal support (described below) used different types of data to account (Felsenstein 2004); therefore, we considered phyloge- for phylogeny, we quantified phylogenetic relationships netic uncertainty to be minimal and proceeded with of species in two ways. First, we calculated pairwise analyses using the consensus phylogram. We made this genetic distances between species using cytochrome b tree ultrametric using nonparametric rate smoothing in (cytb) mitochondrial DNA (mtDNA) sequences. We TreeEdit v.1.0 (Sanderson 1997;Rambautand downloaded cytb sequences of each species from Charleston 2001) and then pruned the outgroup taxon, GenBank (Appendix S1), aligned them using the MUS- leaving the eight focal minnow species (Fig. 3). CLE algorithm in Geneious Pro v.5.5.7 (Drummond et al. 2012), and checked them by eye. The final cytb Data analysis alignment was 1,140 base pairs with complete coverage for all eight species and no gaps or premature stop To visualize relationships among species in morpholo- codons. We then calculated a pairwise maximum likeli- gy, body shape, diet, habitat use, and genetic relatedness hood (GTR) genetic distance matrix in PAUP*v.4.0 (genetic distances), we constructed phenograms in (Swofford 2003). which species were clustered using the Unweighted Pair Second, we inferred the phylogeny of the eight focal Group Method with Arithmetic Mean (UPGMA) in species using Bayesian inference in MrBayes v.3.2 NTSYS (Rohlf 2009). (Ronquist et al. 2012). We downloaded partial cyto- In the first analysis, we tested for correlations be- chrome c oxidase 1 (cox1) mtDNA sequences for each tween morphology (traditional morphology and body species from GenBank (Appendix S1) and concatenated shape), ecology (habitat use and diet), and phylogeny them with their respective cytb sequences from above. using standard Mantel tests (Mantel 1967). To compare We also downloaded cytb and cox1 sequences for the results while accounting for phylogenetic relationships, Eurasian cyprinid, Tinca tinca, for use as an outgroup we also tested for correlations between matrices using (Bufalino and Mayden 2010). We aligned all sequences partial Mantel tests. The partial Mantel tests for correla- and checked them by eye as above. The concatenated tion between a dependent variable matrix and an inde- cytb and cox1 alignment was 1792 base pairs with pendent variable matrix while holding other indepen- complete coverage for all species and no gaps or pre- dent variable matrices constant (Smouse et al. 1986). mature stop codons. We conducted two independent For all partial Mantel tests, we used the pairwise species runs in MrBayes consisting of three heated and one cold distance matrices for morphology, body shape, habitat chain for one million generations retaining every 1000th use, and diet (i.e., morphology vs. habitat use, morphol- sample. We applied a separate substitution model to ogy vs. diet, body shape vs. habitat use, body shape vs. each codon position within each gene and unlinked all diet) and input the species genetic distance matrix as a parameters across partitions (i.e., a partitioned analysis). third constant matrix. We ran standard and partial Man- The best model for each partition was chosen using the tel tests in PASSaGE v.2 (Rosenberg and Anderson Akaike Information Criterion (AIC) in jModelTest 2011) using 999 permutations with two-tailed P-values (Posada 2008) and models were as follows: SYM+Γ to determine significance. for cytb first positions, GTR+Γ for cytb second and In the second analysis, we used Phylogenetic Canon- third positions and cox1 first and third positions, and ical Correlation Analysis (PCCA) to test for correlations F81 for cox1 2nd positions. We used default priors and between morphology and ecology while accounting for assessed burn in within runs and convergence between phylogeny (Revell and Harrison 2008). PCCA is a 718 Environ Biol Fish (2015) 98:713–724

Fig. 3 UPGMA trees depicting Traditional Morphology Body shape morphological (using traditional C. anomalum C. anomalum and geometric morphometrics), R. atratulus N. leptocephalus ecological (diet and habitat use) and phylogenetic and genetic C. oreas R. atratulus distance relationships among the N. leptocephalus C. oreas eight cyprinids examined C. funduloides C. funduloides L. albeolus L. albeolus L. cerasinus L. cerasinus L. ardens L. ardens 0.097 0.069 0.040 0.016 0.012 0.009 Distance Distance Phylogenetic relationships Genetic distance C. oreas C. oreas

C. funduloides R. atratulus

C. anomalum N. leptocephalus

N. leptocephalus C. funduloides

R. atratulus C. anomalum L. albeolus L. ardens L. cerasinus L. albeolus L. ardens L. cerasinus

0.4 0.2 0.0 0.225 0.161 0.096 Substitutions per site Distance Habitat use Diet C. anomalum C. anomalum N. leptocephalus C. oreas C. oreas N. leptocephalus R. atratulus C. funduloides

C. funduloides L. albeolus L. ardens L. cerasinus L. albeolus L. ardens

L. cerasinus R. atratulus 0.404 0.269 0.135 0.313 0.191 0.068 Distance Distance modification of Canonical Correlation Analysis (CCA) with many variables that can be divisible into two dis- that first corrects for phylogenetic nonindependence tinct groups, such as morphological and ecological var- using a phylogenetic generalized least squares transfor- iables. CCA computes two sets of derived variables mation (PGLS; Grafen 1989) based on Pagel’s λ (Pagel from two sets of original variables whereby the correla- 1999; Freckleton et al. 2002). Pagel’s λ is a measure of tions between the derived variables are maximized phylogenetic signal, which can range from 0 to 1; λ=0 (Miles and Ricklefs 1984; Revell and Harrison 2008). indicates that trait evolution is independent of phyloge- In the case of PCCA, running the analysis with an ny whereas λ=1 indicates that traits are evolving on the ultrametric tree under λ=0 is equivalent to CCA with phylogeny under constant-rate Brownian motion no phylogenetic transformation whereas running with (Freckleton et al. 2002). The use of CCA in λ=1 is equivalent to CCA performed on the phyloge- ecomorphology and phylogenetic comparative studies netically independent contrasts (Felsenstein 1985; has a long history (Miles and Ricklefs 1984; Losos Revell and Harrison 2008). 1990; Winemiller et al. 1995;Giannini2003;Vittand Because the number of species limits the number of Pianka 2005) and is especially well suited to data sets variables in a PCCA, we first conducted a Principal Environ Biol Fish (2015) 98:713–724 719

Components Analysis (PCA) on each data set (i.e., Similarly, the diet of fishes showed a deep bifurcation, morphology, body shape, habitat use, and diet) to reduce but contrary to habitat use, this phenogram grouped the number of variables. For the PCAs, we used the C. oreas with C. anomalum and N. leptocephalus,and correlation matrix, retained axes with eigenvalues >1.0, grouped R. atratulus with the other species. These gen- and saved the Principal Component scores for use in eral groupings reflected high frequency of invertebrates further analyses (PCAs were conducted in SPSS v.18.0). in the diet (e.g., L. ardens) and fishes that consumed PCA resulted in four axes retained for the morphology large amounts of algae and detritus (e.g., C. anomalum). data set (explaining 91.6 % of the variation), four axes The genetic distance phenogram and the phylogeny for body shape (89.0 %), three axes for habitat use show C. oreas as sister to all other species and grouped (86.7 %), and three axes for diet (86.8 %). L. ardens, L. albeolus,andL. cerasinus, but placement We used the Principal Component scores for each of other taxa varied. data set (morphology, body shape, habitat use, and diet) Standard Mantel tests revealed that both traditional and the ultrametric phylogeny for the eight minnow morphology and shape were correlated with habitat use, species to perform a series of PCCAs using the phytools but traditional morphology showed a stronger correla- package in R (phyl.cca; Revell 2012): morphology vs. tion (Table 1). Neither morphology nor body shape was habitat use, morphology vs. diet, body shape vs. habitat correlated with diet; however, diet was correlated with use, and body shape vs. diet. We ran each analysis twice, phylogeny. Using partial Mantel tests to account for under λ=1 and λ=0, to compare results with and with- evolutionary relationships resulted in both traditional out accounting for phylogeny, respectively. morphology and shape still being correlated with habitat use. Diet was still uncorrelated with traditional morphol- ogy and body shape after accounting for phylogeny. Results PCCA revealed significant correlations between tra- ditional morphology and habitat use as well as shape Both morphology and body shape phenograms grouped and habitat use on the first canonical axes (Table 2). This C. funduloides, L. cerasinus,andL. albeolus, and split was the case for both λ=0 and λ=1. Traditional mor- out L. ardens as the most divergent (Fig. 3). The primary phology was also correlated with diet on the first canon- difference between morphology and body shape clus- ical axis; however, the correlation was only significant tering was placement of C. oreas and N. leptocephalus. under λ=1. In contrast, shape showed no correlation In the morphology phenogram, C. oreas grouped with with diet in the PCCA (Table 2). C. anomalum and R. atratulus,butN. leptocephalus grouped with C. anomalum and R. atratulus in the body shape phenogram. Clustering of both morphology and Discussion body shape generally grouped fish by more fusiform or rounded bodies (e.g., C. anomalum, R. atratulus)and How to best characterize ecologically relevant morpho- fish that are more laterally compressed (e.g., L. albeolus, logical variation is a primary consideration for L. cerasinus) fish, and L. ardens as an outgroup in both ecomorphological studies (Norton et al. 1995). Here cases. It is worth noting C. anomalum had a limited we found that both traditional morphology and body sample size (n=3) for body shape analysis; however, the shape uncovered relationships between morphology and general relationships among species were similar to ecology of eight species of cyprinids. Both traditional those described in Douglas and Matthews (1992)and and geometric morphometric approaches quantify linear thus the small sample size for this species did not seem distances on a fish’s body, but each also captures some to influence relationships among species in this case. unique aspects of morphology (Adams et al. 2004). For The habitat use phenogram had two major clusters and example, traditional morphometrics includes fin mor- somewhat mirrored the morphology and body shape phology and considers some three-dimensional aspects phenograms by grouping C. oreas and R. atratulus with through measures such as gape width and interorbital C. anomalum and N. leptocephalus. Habitat use group- distance (Hubbs and Lagler 1941), which lateral-only ings primarily split fishes that occupied benthic, shallow shape quantification often ignores. In comparison, geo- habitats (e.g., C. anomalum) and fishes that occupied the metric morphometrics focuses on the “fuselage” and water column in deeper habitats (e.g., L. ardens). captures the geometry of shape (Adams et al. 2004). 720 Environ Biol Fish (2015) 98:713–724

Table 1 Mantel’s r (P-value) from standard and partial Mantel distances included to control for phylogenetic relationships. Sig- tests. The lower triangle is results from standard Mantel tests and nificant matrix correlations are in bold (P≤0.05) the upper triangle is results from partial Mantel tests with genetic

Traditional morphology Shape Habitat Diet

Traditional morphology – NA 0.76 (0.001) 0.25 (0.185) Shape NA – 0.45 (0.024) −0.08 (0.684) Habitat 0.74 (0.001) 0.44 (0.026) – NA Diet 0.19 (0.305) −0.04 (0.475) NA – Phylogeny −0.11 (0.653) −0.04 (0.861) 0.13 (0.504) 0.37 (0.019)

The strong correlation between traditional morphology subtle ways that are better revealed by geometric and ecology suggests that lateral body measures and fin morphometrics. In support of this, Maderbacher et al. lengths captured ecologically relevant morphological (2008) found that closely related were easier to variation among fish species. However, some species, differentiate using geometric morphometrics. In our particularly closely related ones, may differ in more study, which spanned seven cyprinid genera, traditional

Table 2 Results from phylogenetic canonical correlation analysis phylogeny, and λ=1, which assumes that traits are evolving under (PCCA) testing for correlations between data sets. Pagel’s λ is a Brownian motion. Significant canonical correlations (Can. Cor.) measure of phylogenetic signal; each analysis was performed areinbold(P≤0.05) under λ=0, which assumes that trait evolution is independent of

Comparison λ Can. Var. Can. Cor. χ2 df P

Traditional morphology 0 1 0.999 25.31 12 0.013 vs. Habitat 2 0.949 7.60 6 0.269 3 0.450 0.68 2 0.712 1 1 0.999 26.96 12 0.008 2 0.968 8.94 6 0.177 3 0.438 0.64 2 0.727 Traditional morphology 0 1 0.998 20.03 12 0.067 vs. Diet 2 0.829 4.12 6 0.660 3 0.438 0.64 2 0.726 1 1 0.999 24.15 12 0.019 2 0.886 6.18 6 0.404 3 0.639 1.57 2 0.456 Shape vs. Habitat 0 1 0.999 31.24 12 0.002 2 0.951 7.32 6 0.293 3 0.309 0.30 2 0.860 1 1 0.999 29.28 12 0.004 2 0.942 6.87 6 0.333 3 0.330 0.35 2 0.841 Shape vs. Diet 0 1 0.950 9.75 12 0.638 2 0.770 2.76 6 0.838 3 0.148 0.07 2 0.967 1 1 0.970 10.26 12 0.593 2 0.625 1.74 6 0.942 3 0.283 0.88 2 0.883 Environ Biol Fish (2015) 98:713–724 721 morphology and body shape had similar associations nonindependence is also an important consideration. with habitat use, but differed in their associations with This is necessary in order to determine the extent of diet. Douglas and Matthews (1992) found that tradition- morphological and ecological variation that is causally al morphology was significantly correlated with ecology linked independent of evolutionary relationships for the same eight species studied here, but their results (Wainwright and Reilly 1994; Norton et al. 1995). In depended on how morphology data were analyzed. An- our analyses, accounting for phylogenetic relationships alytical methods in traditional morphometrics have sta- did not substantially change results regarding habitat tistical issues, particularly in regards to controlling for use. This is likely due to a lack of correlation with allometric relationships, which can ultimately yield dif- genetic distance (Mantel test) and low phylogenetic ferent results (Adams et al. 2004; McCoy et al. 2006; signal in the data (PCCA). In contrast, diet was corre- Mitteroecker and Bookstein 2011). Taken together, the- lated with genetic distance and incorporating phylogeny se findings indicate that the taxonomic span of species improved correlations between traditional morphology under study and the statistical treatment of data are and diet. This correlation indicates that the trophic ecol- important factors to consider when choosing between ogies of these eight fish species are significantly influ- traditional and geometric morphometric approaches. enced by evolutionary history. Diets of lizards and Across all analyses, habitat use was correlated with snakes are also significantly influenced by phylogeny both traditional morphology and body shape. Habitat with major dietary shifts occurring early in their evolu- segregation among coexisting species is widespread in tionary history (Vitt and Pianka 2005; Colston et al. fishes (Werner et al. 1977; Gorman and Karr 1978; 2010). Douglas and Matthews (1992) found significant Willis et al. 2005;Kanoetal.2013), although the degree pairwise correlations between morphology, diet, and of segregation is likely dependent on habitat availability phylogeny when analyzing all 17 fish species in their (Jepsen et al. 1997). The observation that fishes occu- study, but the correlation between morphology and diet pying comparable habitats have similar morphologies decreased to non-significance when they accounted for suggests that certain suites of traits are better adapted to phylogeny. They did not consider phylogeny in their different microhabitats found within streams. Both mor- analyses on the eight cyprinids, likely because of limi- phology and body shape phenograms (Fig. 3)grouped tations imposed by using taxonomic ranks. In compar- C. funduloides, L. albeolus,andL. cerasinus: all are ison, DNA sequences used in our study provide a much “slab-sided” minnows with laterally compressed bodies better estimate of phylogeny than taxonomic ranks used and use open-water and pool habitats. Conversely, by Douglas and Matthews (1992). DNA sequences can L. ardens uses similar pool habitats but its body shape resolve relationships irrespective of and pro- is more elongate and fusiform compared to the three vide estimates of branch lengths, which are important to species above. The other grouping consisted of determine the amount of time or character change ex- C. anomalum, N. leptocephalus, C. oreas,and pected between branching events. In contrast, taxonom- R. atratulus; all are more “round-bodied” species mostly ic ranks are on an ordinal scale and thus, the number of associated with substrates and occur in both pools and available ranks limits phylogenetic resolution and riffles. In the habitat use phenogram (Fig. 2), the top branch lengths between adjacent ranks are not scaled four species are essentially substrate affinity species, to time and therefore considered equal, which is unlikely whereas the bottom four species are all open-water, both between ranks and across the tree. larger pool fishes. These patterns are generally consis- We tested for correlations between morphology and tent with previous investigations demonstrating that ecology data sets using two analyses (Mantel test and fishes from lotic habitats often have fusiform morphol- PCCA) that account for phylogeny in different ways. ogies that reduce drag and facilitate sustained swim- The Mantel test has a long history of use in systematics ming, whereas more laterally compressed body shapes and ecomorphology (Mantel 1967;Smouseetal.1986; in lentic waters can facilitate faster burst speeds and Douglas and Matthews 1992). Phylogeny is incorporat- increased maneuverability (Alexander 1967;Gosline ed by including a matrix of pairwise species distances 1971; Langerhans and DeWitt 2004; Langerhans based on genetic distance, patristic distance, taxonomic 2008). rank, or similar measure. Despite its wide use, the Man- Given that most ecomorphological studies involve tel test has been criticized for low power and inflated data on multiple species, accounting for phylogenetic Type I error in phylogenetic comparative analyses 722 Environ Biol Fish (2015) 98:713–724

(Raufaste and Rousset 2001; Harmon and Glor 2010). and indicate that previous phylogenies based on mor- In fact, Harmon and Glor (2010) recommended using phology were misled by convergence (Wiens et al. Mantel tests only when data cannot be expressed in any 2005, 2012). Future ecomorphological investigations other way than distances. Phylogenetic Canonical Cor- will no doubt benefit from incorporating new methods, relation Analysis (PCCA) is a relatively new analysis analyses, and phylogenetic information, but researchers and is among several recent phylogenetic comparative will need to consider which approaches may be best methods that rely on a measure of phylogenetic signal in given their questions and the specifics of their study the data (Revell and Harrison 2008;Revell2012). In system. PCCA, phylogeny is incorporated by including an ultrametric tree with branch lengths, which are then Acknowledgments We thank W.J. Matthews for spurring this transformed using Pagel’s λ (Pagel 1999; Revell and research, providing the history of traditional morphometrics, for Harrison 2008). Although λ can be estimated directly generously sharing his specimens and data, and for comments that substantially improved the manuscript. We also thank two anony- and the estimate used in PCCA, a power issue exists in mous reviewers for their helpful suggestions for improvement. data sets with low numbers of species, and this may be worse when jointly estimating λ from sets of variables such as in PCCA (Freckleton et al. 2002; Revell and References Harrison 2008). Given these issues with estimating λ, some authorities recommend running analyses under λ λ Adams DC, Otárola-Castillo E (2013) geomorph: an R package both =0 and =1 with the idea being that the actual for the collection and analysis of geometric morphometric value of λ does not matter if one gets similar results shape data. Methods Ecol Evol 4:393–399 (Schluter 2000; Freckleton et al. 2002; Revell and Adams DC, Rohlf FJ, Slice DE (2004) Geometric morphometrics: ten years of progress following the “revolution”. Ital J Zool Harrison 2008). Our PCCA results did not differ sub- – λ λ 71:5 16 stantially under =0 and =1 for correlations between Alexander RM (1967) Functional design in fishes. Hutchinson, both measures of morphology and habitat use, but the London correlation between traditional morphology and diet Blomberg SP, Garland T Jr, Ives AR (2003) Testing for phyloge- improved to significance under λ=1 (Table 2). Although netic signal in comparative data: behavioral traits are more labile. Evolution 57:717–745 Mantel and PCCA analyses mostly yielded similar re- Breno M, Leirs H, van Dongen S (2011) Traditional and geometric sults in our study, previous criticisms of the Mantel test morphometrics for studying skull morphology during the (Raufaste and Rousset 2001; Harmon and Glor 2010) growth in Mastomys natalensis (Rodentia: Muridae). J and the potential to incorporate an estimate of phyloge- Mammal 92:1395–1406 Bufalino AP, Mayden RL (2010) Phylogenetic relationships of netic signal in data advocate for the use of PCCA. North American phoxinins (: Cypriniformes: Furthermore, PCCA revealed a significant correlation Leuciscidae) as inferred from S7 nuclear DNA sequences. here that was not evident using traditional Mantel tests. Mol Phylogenet Evol 55:143–152 Because ecological traits can be shaped over both Colborne SF, Peres-Neto PR, Longstaffe FJ, Neff BD (2013) Effects of foraging and sexual selection on the evolutionary and contemporary timescales, multi- ecomorphology of a fish with alternative reproductive tactics. faceted approaches will likely be most effective at Behav Ecol 24:1339–1347 uncovering relationships between morphology and ecol- Collar DC, Wainwright PC (2006) Discordance between morpho- ogy (Norton et al. 1995). New methodological and logical and mechanical diversity in the feeding mechanism of centrarchid fishes. Evolution 60:2575–2584 analytical approaches will continue to improve our abil- Collar DC, Near TJ, Wainwright PC (2005) Comparative analysis ity to quantify morphological variation and predict the of morphological diversity: does disparity accumulate at the ecology of organisms, but more studies are needed to same rate in two lineages of centrarchid fishes? Evolution 59: connect morphological variation to performance and 1783–1794 fitness in order to reveal the underpinnings of how Colston TJ, Costa GC, Vitt LJ (2010) Snake diets and deep history hypothesis. Biolog J Linn Soc 101:476–486 natural selection drives morphological and ecological Douglas ME, Matthews WJ (1992) Does morphology predict diversification. In addition, a better understanding of ecology? Hypothesis testing within a freshwater stream as- phylogenetic relationships will improve our ability to semblage. Oikos 65:213–224 establish associations between morphology and ecolo- Drummond AJ, Ashton B, Buxton S, Cheung M, Cooper A, Duran C, Field M, Heled J, Kearse M, Markowitz S, Moir R, gy. In many groups, new molecular-based phylogenies Stones-Havas S, Sturrock S, Thierer T, Wilson A (2012) have changed our views of evolutionary relationships Geneious v5.6. http://www.geneious.com Environ Biol Fish (2015) 98:713–724 723

Eggleton P, Vane-Wright R (1994) Phylogenetics and Ecology. Langerhans RB, DeWitt TJ (2004) Shared and unique features of Acad Press, San Diego evolutionary diversification. Am Nat 164:335–349 Felsenstein J (1985) Phylogenies and the comparative method. Losos JB (1990) Ecomorphology, performance capability, and Am Nat 125:1–15 scaling of West Indian Anolis lizards: an evolutionary analy- Felsenstein J (2004) Inferring Phylogenies. Sinauer, Sunderland, sis. Ecol Monogr 60:369–388 MA Losos JB (2011) Lizards in an Evolutionary Tree: Ecology and Findley JS (1973) Phenetic packing as a measure of faunal diver- Adaptive Radiation of Anoles. University of California Press, sity. Am Nat 107:580–584 Berkeley Findley JS (1976) The structure of bat communities. Am Nat 110: Maddison WP (1990) A method for testing the correlated evolution 129–139 oftwobinarycharacters:Aregains or losses concentrated on Franssen NR, Harris J, Clark SR, Schaefer JF, Stewart LK (2013) certain branches of a phylogenetic tree? Evol 44:539–557 Shared and unique morphological responses of stream fishes Maddison WP, Maddison DR (2011) Mesquite: a modular system for to anthropogenic habitat alteration. Proc R Soc B 280: evolutionary analysis. Version 2.75. http://mesquiteproject.org 20122715 Maderbacher M, Bauer C, Herler J, Postl L, Makasa L, Freckleton RP, Harvey PH, Pagel M (2002) Phylogenetic analysis Sturmbauer C (2008) Assessment of traditional versus geo- and comparative data: a test and review of the evidence. Am metric morphometrics for discriminating populations of the Nat 160:712–726 moorii species complex (Teleostei: Cichlidae), a Gatz AJ Jr (1979a) Ecological morphology of freshwater stream model for allopatric speciation. J Zool Syst fishes. Tulane Stud Zool Bot 21:91–124 Evol Res 46:153–161 Gatz AJ Jr (1979b) Community organization in fishes as indicated Mantel NA (1967) The detection of disease clustering and a by morphological features. Ecol 60:711–718 generalized regression approach. Cancer Res 27:209–220 Giannini NP (2003) Canonical phylogenetic ordination. Syst Biol Martins EP (2004) COMPARE v.4.6b. Computer programs for the 52:684–695 statistical analysis of comparative data. http://compare.bio. Gorman OT, Karr JR (1978) Habitat structure and stream fish indiana.edu communities. Ecol 59:507–515 Matthews WJ (1998) Patterns in freshwater fish ecology. Kluwer Gosline WA (1971) Functional morphology and classification of Academic Press, New York teleostean fishes. University of Hawaii Press, Honolulu McCoy MW, Bolker BM, Osenberg CW, Miner BG, Vonesh JR Grafen A (1989) The phylogenetic regression. Philos Trans Royal (2006) Size correction: comparing morphological traits Soc Lond B 326:119–157 among populations and environments. Oecologia 148:547– Harmon LJ, Glor RE (2010) Poor statistical performance of the 554 Mantel test in phylogenetic comparative analyses. Evolution Miles DB, Ricklefs RE (1984) The correlation between ecology 64:2173–2178 and morphology in deciduous forest passerine birds. Ecol 65: Harvey PH, Pagel MD (1991) The Comparative Method in 1629–1640 Evolutionary Biology. Oxford University Press, Oxford Mitteroecker P, Bookstein F (2011) Linear discrimination, ordina- Herler J (2007) Microhabitats and ecomorphology of coral- and tion, and the visualization of selection gradients in modern coral rock-associated gobiid fish (Teleostei: Gobiidae) in the morphometrics. Evol Biol 38:100–114 northern Red Sea. Mar Eco 28:82–94 Norton SF, Luczkovich JJ, Motta PJ (1995) The role of Hoagstrom CW, Berry CR (2008) Morphological diversity among ecomorphological studies in the comparative biology of fish- fishes in a Great Plains river drainage. Hydrobiologia 596: es. Environ Biol Fishes 44:287–304 367–386 Nylander JAA, Wilgenbusch JC, Warren DL, Swofford DL (2008) Hubbs CL (1941) The relation of hydrological conditions to AWTY (are we there yet?): a system for graphical exploration speciation in fishes. In: Needham JG, Sears PB, Leopold A of MCMC convergence in Bayesian phylogenetics. (eds) A symposium on hydrobiology. University of Biogeosciences 24:581–583 Wisconsin Press, Madison, pp 182–195 Pagel M (1999) Inferring the historical patterns of biological Hubbs CL, Lagler KF (1941) Guide to fishes of the Great Lakes evolution. Nature 401:877–884 and tributary waters. The Cranbrook Institute of Science. Paradis E, Claude J, Strimmer K (2004) APE: analysis of phylo- Bloomfield Hills, MI genetics and evolution in R language. Biogeosciences 20: Ibañez C, Tedesco PA, Bigorne R, Hugueny B, Pouilly M, Zepita 289–290 C, Zubieta J, Oberdorff T (2007) Dietary-morphological Piet GJ (1998) Ecomorphology of a size-structured tropical fresh- relationships in fish assemblages of small forested streams water fish community. Environ Biol Fishes 51:67–86 in the Bolivian Amazon. Aquat Living Resour 20:131–142 Posada D (2008) jModelTest: phylogenetic model averaging. Mol Jepsen DB, Winemiller KO, Taphorn DC (1997) Temporal pat- Biol Evol 25:1253–1256 terns of resource partitioning among Cichla species in a Pouilly M, Lino F, Bretenoux JG, Rosales C (2003) Dietary- Venezuelan blackwater river. J Fish Biol 51:1085–1108 morphological relationships in a fish assemblage of the Kano Y, Miyazaki Y, Tomiyama Y, Mitsuyuki C, Nishida S, Bolivian Amazonian floodplain. J Fish Biol 62:1137–1158 Rashid ZA (2013) Linking mesohabitat selection and eco- Rambaut A, Charleston M (2001) TreeEdit: phylogenetic Tree logical traits of a fish assemblage in a small tropical stream Editor v.1.0a10. http://tree.bio.ed.ac.uk/software/treeedit/ (Tinggi River, Pahang Basin) of the Malay Peninsula. Zool Rambaut A, Drummond AJ (2007) Tracer v1.4. http://beast.bio. Sci 30:178–184 ed.ac.uk/Tracer Langerhans RB (2008) Predictability of phenotypic differentiation Raufaste N, Rousset F (2001) Are partial Mantel tests adequate? across flow regimes in fishes. Integr Comp Biol 48:750–768 Evolution 55:1703–1705 724 Environ Biol Fish (2015) 98:713–724

Reecht Y, Rochet MJ, Trenkel VM, Jennings S, Pinnegar JK Strauss RE, Bookstein FL (1982) The truss: body form reconstruc- (2013) Use of morphological characteristics to define func- tions in morphometrics. Syst Zool 31:113–135 tional groups of predatory fishes in the Celtic Sea. J Fish Biol Surat EM, Matthews WJ, Bek JR (1982) Comparative ecology of 83:355–377 Notropis albeolus, N. ardens,andN. cerasinus (Cyprinidae) Revell LJ (2012) phytools: an R package for phylogenetic com- in the upper Roanoke River drainage, Virginia. Am Midl Nat parative biology (and other things). Methods Ecol Evol 3: 107:13–24 217–223 Swofford DL (2003) PAUP*. Phylogenetic Analysis Using Revell LJ, Harrison AS (2008) PCCA: a program for phylogenetic Parsimony (*and Other Methods). Version 4. Sinauer canonical correlation analysis. Biogeosciences 24:1018– Associates, Sunderland, MA 1020 Vitt LJ, Pianka ER (2005) Deep history impacts present-day Ricklefs RE, Cox GW (1977) Morphological similarity and eco- ecology and biodiversity. Proc Natl Acad Sci U S A 102: logical overlap among passerine birds on St. Kitts, British 7877–7881 West Indies. Oikos 29:60–66 Wainwright PC, Reilly SM (1994) Ecological Morphology: Ricklefs RE, Cochran D, Pianka ER (1981) A morphological Integrative Organismal Biology. University of Chicago analysis of the structure of communities of lizards in desert Press, Chicago habitats. Ecol 62:1474–1483 Werner EE, Hall DJ, Laughlin DR, Wagner DJ, Wilsmann LA, Rohlf FJ (1998) On applications of geometric morphometrics to Funk FC (1977) Habitat partitioning in a freshwater fish studies of ontogeny and phylogeny. Syst Biol 47:147–158 community. J Fish Res Board Can 34:360–370 Rohlf FJ (2004) tps software. http://life.bio.sunysb.edu/morph/ Wiens JJ, Bonett RM, Chippindale PT (2005) Ontogeny discom- Rohlf FJ (2009) NTSYSpc: Numerical Taxonomy System, v.2.2. bobulates phylogeny: paedomorphosis and higher-level sal- Exeter publishing, Setauket, NY amander relationships. Syst Biol 54:91–110 Rohlf FJ, Marcus LF (1993) A revolution in morphometrics. Wiens JJ, Hutter CR, Mulcahy DG, Noonan BP, Townsend TM, Trends Ecol Evol 8:129–132 Sites JW Jr, Reeder TW (2012) Resolving the phylogeny of Ronquist F, Teslenko M, van der Mark P, Ayres DL, Darling A, lizards and snakes (Squamata) with extensive sampling of Höhna S, Larget B, Liu L, Suchard MA, Huelsenbeck JP genes and species. Biol Lett 8:1043–1046 (2012) MrBayes 3.2: Efficient Bayesian phylogenetic infer- Wilgenbusch JC, Warren DL, Swofford DL (2004) AWTY: a ence and model choice across a large model space. Syst Biol system for graphical exploration of MCMC convergence in 61:539–542 Bayesian phylogenetic inference. http://ceb.csit.fsu.edu/awty Rosenberg MS, Anderson CD (2011) PASSaGE: Pattern analysis, Willis SC, Winemiller KO, Lopez-Fernandez H (2005) Habitat spatial statistics and geographic exegesis. Version 2. Methods structural complexity and morphological diversity of fish Ecol Evol 2:229–232 assemblages in a Neotropical floodplain river. Oecologia Sanderson MJ (1997) A nonparametric approach to estimating 142:284–295 divergence times in the absence of rate constancy. Mol Biol Winemiller KO (1991) Ecomorphological diversification in low- Evol 14:1218–1231 land freshwater fish assemblages from five biotic regions. Schluter D (2000) The ecology of adaptive radiation. Oxford Ecol Monogr 61:343–365 Series in Ecology and Evolution. Oxford University Press, Winemiller KO (1992) Ecomorphology of freshwater fishes. Natl Oxford Geogr Res Explor 8:308–327 Smouse PE, Long JC, Sokal RR (1986) Multiple regression and Winemiller KO, Kelso-Winemiller LC, Brenkert AL (1995) correlation extensions of the Mantel test of matrix correspon- Ecomorphological diversification and convergence in fluvial dence. Syst Zool 35:627–632 fishes. Environ Biol Fishes 44:235–261