Convergence is Only Skin Deep: Craniofacial Evolution in Electric Fishes from South America and Africa (Apteronotidae and Mormyridae)

Synopsis Apteronotidae and Mormyridae are species-rich clades of weakly electric fishes from Neotropical and Afrotropical freshwaters, respectively, known for their high morphological disparity and often regarded as a classic example of convergent evolution. Here, we use CT-imaging and 3D geometric morphometrics to quantify disparity in craniofacial morphologies, and to test the hypothesis of convergent skull-shape evolution in a phylogenetic context. For this study, we examined 391 specimens representing 78 species of Apteronotidae and Mormyridae including 30 of 37 (81%) of all valid genera with the goal to sample most of the craniofacial disparity known in these clades. We found no overlap between Apteronotidae and Mormyridae in the skull-shape morphospace using PCA and a common landmark scheme, and therefore no instances of complete phenotypic convergence. Instead, we found multiple potential instances of incomplete convergence, and at least one parallel shift among electric fish clades. The greatest components of shape variance in both families are the same as observed for most vertebrate clades: heterocephaly (i.e., opposite changes in relative sizes of the snout and braincase regions of the skull), and heterorhynchy (i.e., dorsoventral changes in relative snout flexion and mouth position). Mormyrid species examined here exhibit less craniofacial disparity than do apteronotids, potentially due to constraints associated with a larger brain size, ecological constraints related to food-type availability. Patterns of craniofacial evolution in these two clades depict a complex story of phenotypic divergence and convergence in which certain superficial similarities of external morphology obscure deeper osteological and presumably developmental differences of skull form and function. Among apteronotid and mormyrid electric fishes, craniofacial convergence is only skin deep.

Synopsis Apteronotidae and Mormyridae are species-rich clades of weakly electric fishes from Neotropical and Afrotropical freshwaters, respectively, known for their high morphological disparity and often regarded as a classic example of convergent evolution. Here, we use CT-imaging and 3D geometric morphometrics to quantify disparity in craniofacial morphologies, and to test the hypothesis of convergent skull-shape evolution in a phylogenetic context. For this study, we examined 391 specimens representing 78 species of Apteronotidae and Mormyridae including 30 of 37 (81%) of all valid genera with the goal to sample most of the craniofacial disparity known in these clades. We found no overlap between Apteronotidae and Mormyridae in the skull-shape morphospace using PCA and a common landmark scheme, and therefore no instances of complete phenotypic convergence. Instead, we found multiple potential instances of incomplete convergence, and at least one parallel shift among electric fish clades. The greatest components of shape variance in both families are the same as observed for most vertebrate clades: heterocephaly (i.e., opposite changes in relative sizes of the snout and braincase regions of the skull), and heterorhynchy (i.e., dorsoventral changes in relative snout flexion and mouth position). Mormyrid species examined here exhibit less craniofacial disparity than do apteronotids, potentially due to constraints associated with a larger brain size, ecological constraints related to food-type availability. Patterns of craniofacial evolution in these two clades depict a complex story of phenotypic divergence and convergence in which certain superficial similarities of external morphology obscure deeper osteological and presumably developmental differences of skull form and function. Among apteronotid and mormyrid electric fishes, craniofacial convergence is only skin deep.
Spanish Resumen Apteronotidae y Mormyridae son clados ricos en especies de peces débilmente eléct ricos de aguas dulces Neotropicales y Afrotropicales, respectivamente, conocidos por su alta disparidad morfológica y, a menudo, considerados como un ejemplo clásico de evolución convergente. Aquí usamos imágenes de tomografía computarizada y morfometría geométrica 3D para cuantificar la disparidad en las morfologías craneofaciales y para probar la hipótesis de la evolución convergente de la forma del cráneo en un contexto filogenético. Para este estudio, examinamos 391 especímenes que representan 78 especies de Apteronotidae y Mormyridae, incluidas 30 de 37 (81%) de todos los géneros válidos, con el objetivo de muestrear la mayor parte de la disparidad craneofacial conocida en estos clados. No encontramos superposición entre Apteronotidae y Mormyridae en el morfoespacio en forma de cráneo utilizando PCA y un esquema de referencia común y, por lo tanto, no hay instancias de Introduction Convergent evolution, referring to the independent origins of similar traits in distantly related species, is widely considered resulting from selection for phenotypes that solve similar functional, physiological, or ecological problems ( Revell et al. 2007 ;Losos 2011 ;Stayton 2015 ;Ord and Summers 2015 ;Sackton and Clark 2019 ;Grossnickle et al. 2020 ). Under the umbrella idea of convergence, traits can vary in the degree of structural or functional similarities, and also in the covariances of similarities among levels in the hierarchy of organismal design ( Striedter and Northcutt 1991 ). These levels, including genetics, development, morphology, and function, can appear as convergent or divergent across species, and there is not always a oneto-one connection between them (different morphologies may result in the same behavior, such as flight in birds and bats) ( Striedter and Northcutt 1991 ). Even at the morphological level, we cannot assume that internal morphologies are completely homologous across taxa, even if external morphologies appear similar ( Shubin et al. 2009 ). Morphologists have used morphotypes (qualitative external shape categories) as heuristics for understanding the ecology and evolution of organismal trait evolution ( Cresko and Baker 1996 ;Berrebi and Valiushok 1998 ;Dimmick et al. 2001 ;Meier et al. 2017 ;Meier et al. 2019 ;McGee et al. 2020 ). They assign species to distinct phenotypic categories to make inferences about ecology and habitat use ( Cresko and Baker 1996 ;Berrebi and Valiushok 1998 ;Dimmick et al. 2001 ;Meier et al. 2017 ;Meier et al. 2019 ;McGee et al. 2020 ). While morphotypes are useful for identifying possible instances of convergence, new phylogenetic comparative methods and ways to quantify shape are better suited for analyzing convergent evolution. In this paper, we utilize these cutting-edge methods to study convergent evolution in apteronotid and mormyrid weakly electric fishes ( Fig. 1 ).
In phylogenetic comparative methods, researchers have put forward definitions to identify different types  of convergence and quantify shape differences among taxa. Complete convergence can be said to occur when species or higher taxa from phylogenetically distant clades exhibit considerable or total overlap in their phenotypes ( Losos 2011 ;Meachen-Samuels 2012 ), as represented, for example, in a multivariate shape-space where taxa are connected by lines depicting their phylogenetic interrelationships; i.e., a phylomorphospace ( Sidlauskas 2008 ). By the same token, incomplete convergence occurs when taxa in a phylomorphospace are closer together than were their ancestors, although still occupying distinct portions of the phylomorphospace (see Fig 3 . in Stayton 2006 ). Another possibility is parallel evolution (coined "parallel shifts"), in which different lineages undergo similar morphological changes represented as parallel lines in a phylomorphospace, often thought to arise from changes in similar underlying genetic or developmental factors affecting the production of phenotypic variation ( Simpson 1961 ;Schluter et al. 2004 ;Arendt and Reznick 2008 ;Bolnick et al. 2018 ). The development of new phylogenetic comparative methods has increased the reliability and analytical tractability of quantifying these different modes of phenotypic evolution using statistical models ( Stayton 2015 ;Grossnickle et al. 2020 ). These methods also permit us to differentiate between different types of convergence using morphological and morphometric data, and multivariate statistical and comparative phylogenetic analyses.
Our study aims to examine internal morphological characters of two distantly-related groups of electric fishes, Gymnotiformes and Mormyridae, for patterns of convergent evolution. Gymnotiformes are a clade of weakly electric fishes from the humid Neotropics containing approximately 262 valid species ( Albert 2001 ;Albert and Crampton 2006 ;Evans et al. 2017 ;Bernt et al. 2018 ;2019 ;Ivanyisky and Albert 2014 ). In this group, the family Apteronotidae (with 99 valid species) includes a high proportion of total morphological disparity observed in Gymnotiformes as a whole, particularly in terms of head shape and craniofacial phenotypes ( Albert 2001 ;Evans et al. 2017 ;Ford et al. 2022 ). Salient diagnostic characters of Apteronotidae include: presence of a small caudal fin with segmented fin rays, an elongate fleshy electroreceptor organ located on the dorsal  Table 1 for landmark definitions. Each scale bar is 10 mm.
body margin, and a neurogenic electric organ that generates a high-frequency wave-type electric signal ( Bennett 1965 ;Albert 2001 ;Albert and Crampton 2005 ;Bernt et al. 2019 ). Ford additional diagnostic osteological traits see Albert (2001) . Apteronotid species inhabit most aquatic habitats of lowland tropical South America, and are most diverse in deep ( > 5 m) and swiftly flowing river channels of large Amazonian rivers. The high disparity in head, snout, and mouth shape in deep channel apteronotids is accompanied by high trophic diversity, and these traits are hypothesized to represent adaptations to utilize different habitats and trophic resources ( Marrero and Winemiller 1993 ;Winemiller and Adite 1997 ;Albert and Crampton 2005 ;Albert and Reis 2011 ;Evans et al. 2019 ).
Mormyrid fishes are another clade of weakly electric fishes from the Afrotropics with approximately 227 valid species ( van der Bank and Kramer 1996 ; Lavoué et al. 2004 ;Kramer 2013 ;Lamanna et al. 2016 ). As with apteronotids, mormyrids exhibit high disparity of head shape, craniofacial phenotypes, and electric signal waveforms ( Ford et al. in review ;van der Bank and Kramer 1996 ;Lavoué et al. 2004 ;Kramer 2013 ;Lamanna et al. 2016 ). Mormyrids have a relatively large brain as compared to their body size, hypothesized to be connected to cognitive functions such as environmental mapping and signal recognition ( Nilsson 1996 ;von der Emde and Bleckmann 1998 ;Butler and Hodos 2005 ;Striedter 2005 ;Sukhum et al. 2016 ). Salient diagnostic characters of Mormyridae include: non-protrusible mouths; cycloid scales covering the body; small pores in the skin covering the body and head; posterior dorsal and anal fins; and a deeply forked caudal fin with rounded lobes ( Kramer 1994 ;Sullivan and Hopkins 2005 ;Kramer 2013 ;Hilton and Lavoue 2018 ;Amen et al. 2020 ;Mulelenu et al. 2020 ). Additional diagnostic osteological traits are provided by Hilton (2003) . Mormyrids inhabit many habitats across continental Africa, including small streams, fast-moving rivers, and swamps ( Chapman et al. 2002 ;Montchowui et al. 2007 ;Lavoué et al. 2012 ;Jackson et al. 2013 ;Adjibade et al. 2020 ). Mormyrids exhibit moderate levels of dietary diversity, hypothesized to have contributed to the disparity of  head and mouth phenotypes observed in this family ( Okedi 1971 ;Fawole 2002 ;Arnegard and Carlson 2005 ;N' da et al. 2014 ). Gymnotiformes and Mormyridae have long been viewed as a case of convergent evolution, including genetic, physiological, morphological, and behavioral traits associated with active electroreception (e.g., electrosensory receptor organs and central neural pathways, electrocytes and electromotor neural pathways), but also craniofacial phenotypes associated with trophic behaviors ( Figs. 2 and 3 ) ( Bullock and Heiligenberg 1986 ;Marrero and Winemiller 1993 ;Winemiller and Adite 1997 ;Zakon et al. 2006 ;Gallant et al. 2014 ). A large portion of the work on electric fish convergence has focused on similarities in electric signal (both electroreception and electrogeneration) and the genetic basis of signal diversity, and have found evidence for convergence in these areas ( Bullock et al. 1983 ;Zakon et al. 2006 ;Lavoué et al. 2012 ;Wang and Yang 2021 ). In a few species of gymnotiform and mormyrid, there is evidence of convergence in external morphologies Table 2 Proposed groups of convergent apteronotid and mormyrid species for convevol analysis. Proposed convergence groups of apteronotid and mormyrid species based on qualitative proximity on the phylomorphospace. These groups were used in an analysis of convergence based on the distance between the hypothesized convergent taxa divided by the maximum distance between any two species in trait space. The results from the convevol analysis are included to show significant levels of convergence across some of the h ypothesized g roups  ( Winemiller and Adite 1997 ), and qualitatively in osteological characters ( Marrero and Winemiller 1993 ). In our study, we quantitatively assessed similarities in craniofacial morphology across a diverse sampling of species in Apteronotidae and Mormyridae ( Fig. 3 ), including 78 total species (40 of 227 mormyrid species, 36 of 97 apteronotid species, and two outgroups). We used 3D geometric morphometrics and phylogenetic comparative methods to determine if there is significant craniofacial similarity among species between these families. The aims of this study were to: (1) obtain a diverse open-source CT dataset for both families; (2) quantify craniofacial shape using 3D geometric morphometrics; and (3) statistically assess convergence across species of apteronotids and mormyrids hypothesized to be convergent using phylogenetic and morphological data.

Specimen preparation
We scanned 391 specimens from two families of weakly electric fishes, the Apteronotidae ( n = 162) and Taible 3 Proposed groups of convergent apteronotid and mormyrid species for windex analysis. Proposed convergence groups of apteronotid and mormyrid species based on qualitative proximity on the phylomorphospace. These groups were used in an analysis of convergence comparing the mean phenotypic distances between all species and the distances between our species of interest. The results from the windex analysis are included to show non-significant levels of convergence across all the hypothesized groups  scans were isolated using CT-Vox and DataViewer , and individual fish segmented using 3D-Slicer ( Fedorov et al. 2012 ). We generated surface renderings and volumes for geometric morphometrics ( Fig. 3 ). We used one scan of the skull of outgroup species for each ingroup clade (Hiodontidae and Pimelodidae) deposited at MorphoSource.org (ark:/87602/m4/M51250; ark:/87602/m4/M53402). We deposited all our scan data at MorphoSource.org.

Geometr ic mor phometr ics
We performed three-dimensional geometric morphometrics in 3D-Slicer using a 22-point homologous landmark scheme across the entire skull (neurocranium, suspensorium, and lower jaw) ( Fig. 4 , Table 1 ) (Federov et al. 2012). We used the R package geomorph to complete a generalized Procrustes superimposition to remove the effects of size, rotation, and relative location from the shape analysis ( Collyer and Adams 2018 ;Adams et al. 2021 ;RStudio Team).

Phylogenetic tree
A combined phylogenetic tree was generated using pruned versions of the Bernt et al. (2019) apteronotid phylogeny and the Peterson et al. (2022) mormyrid phylogeny (only species sampled in this study were kept in the new phylogeny). Bayesian and maximumlikelihood methods were used to infer each phylogeny based on multiple nuclear and mitochondrial genes ( Bernt et al. 2019, Peterson et al. 2022 ). These two phylogenies were then combined in R (v.4.0.3) using the command bindtree in the R package ape (v. 5.4-1, Paradis and Schliep 2019 ). Next, the chronos function Tree ( ape) and six fossil and secondary calibrations from Peterson et al. (2022) and Arcila and James 2017 were used to time-calibrate the combined topology ( Paradis and Schliep 2019 ).

Shape analysis
We completed a principal component analysis (PCA) in geomorph using all specimens to identify the major axes of shape change and visualize shape differences within and between families ( Collyer and Adams 2018 ;Adams et al. 2021 ). A phylomorphospace was generated in MorphoJ using species shape averages to visualize phylogenetic relationships and identify potential instances of convergence ( Klingenberg 2011 ). We completed an analysis of morphological disparity by genera which showed significant distances between multiple genera in our dataset (Supplemental Table 2).

Assessment of convergence
We used two methods to test for convergence: the R packages convevol and windex ( Arbuckle et al. 2014 ;Arbuckle 2015 ;Stayton 2015 ). The package convevol uses a priori hypotheses of convergence based on similarities in external phenotypes ( Table 2 ) and assesses the phenotypic distance between taxa and the most recent common ancestor ( Stayton 2015 ). Values of C1 describe the strength of convergence (distance between proposed tips divided by the maximum distance between any pair of taxa in those lineages) and C2 is the absolute amount of morphological evolution during convergence (subtract the maximum distance between two species from the distance between tips of proposed taxa) ( Stayton 2015 ). We tested several groups of potentially convergent species based on proximity and phylogenetic distance in the phylomorphospace (based on species averages) and used groupings that provided the highest degrees of convergence. We also calculated the Wheatsheaf index ( Arbuckle 2015 ) to test for convergence and assess the degree to which incomplete convergence best describes our data. This method also uses a priori hypotheses ( Table 3 ) and calculated the ratio of the mean distances between all species to the distances between focal species (using species averages).

Mor pholog ical diversity
There are high levels of diversity in craniofacial morphology across the two groups, and the PCA shows broad coverage of the morphospace by both families, with no overlap between them when PC1 and PC2 are the axes of shape change ( Fig. 5 ). The first three PC axes represent 62.71% of the morphological variation for both Apteronotidae and Mormyridae (with outgroups), although PC3 (12.0%) does not show a singular phenotypic trend. PC1 (33.79%) represents the shape change trend heterocephaly ( Evans et al. 2017 ), while PC2 (16.92%) visualizes the shape change trend heterorhynchy ( Ford et al. 2022 ). Heterocephaly is the inverse relationship between the size of the snout (the pre-orbital portion of the skull) and braincase. Extreme PC1 phenotypes include Sternarchorhynchus and Petro-cephalus . Heterorhynchy is the relative dorso-ventral flexion of the snout. These trends are not only seen when visualizing both families together but also in independent studies ( Evans et al. 2017 ;Ford et al. 2022 ). Extreme morphologies of PC2 include Sternarchella and Sternarchorhynchus . There were significant differences in morphological disparity (based on the Procrustes absolute distances across genera) both within each family and across each family (Supplemental Table 2).

Craniofacial evolution
When both families are visualized in a phylomorphospace, there is no overlap between species averages of phenotypes ( Fig. 6 ). There are instances of convergence and divergence within each family (Apteronotidae and Mormyridae), but no complete convergence between families. Instead, we see multiple instances of what we identify as incomplete convergence towards certain morphologies in each family and an additional parallel shift ( Fig. 7 ( Table 3 ). These results confirm our earlier result that there is no complete convergence between Apteronotidae and Mormyridae.

Discussion
Many aspects of the phenotype in the electric fishes of the Afrotropics (Mormyroidea) and Neotropics (Gymnotiformes) have been interpreted as convergent, including: genes, cells and tissues of the electrosensory and electromotor systems, foraging and sexual behaviors, and foraging ecology ( Bullock et al. 1983 ;Zakon 1986 ;Marrero and Winemiller 1993 ;Winemiller and Adite 1997 ;Zakon et al. 2006 ;Gallant et al. 2014 ). Researchers recognized similarities in external head and body morphology among distantly related taxa, hypothesizing that these aspects of external morphology reflect similar internal structures and functions; for example, grasp-suction feeding ( Marrero andWinemiller 1993 , Winemiller andAdite 1997 ). In our study of craniofacial evolution, however, we find a more complicated story of convergence, divergence, and independent trait evolution at different scales. The two electric fish clades (Mormyridae and Apteronotidae) are completely separated on the PCA and phylomorphospace, with no instances of overlap in craniofacial morphology ( Figs. 5 and 6 ). This is not what we expect if there were "complete" morphological convergence ( sensu Losos 2011 ), and this interpretation is confirmed with the Wheatsheaf indices of potentially convergent taxa. In this sense convergence is superficial in that there have been multiple instances of incomplete morphological convergence, and one instance of a parallel shift ( Fig. 7 ) ( Stayton 2015 ;Grossnickle et al. 2020 ). This lack of complete convergence in head shape between mormyrid and apteronotid electric fishes is presumably based on developmental canalization ( Evans et al. 2017 ), such that the convergent phenotypes exhibit a mosaic pattern of diversification and may be said to be only skin deep. Furthermore, the instances of incomplete convergence may be closely related to ecological factors such as diet preference and habitat occupancy. The extremely dolichocephalic mormyrid species (e.g., Campylomormyrus ) has substrate preferences based on morphology (Amen 2020), with longer snouts allowing for moving substrate for foraging. Although it has not been examined in the same laboratory setting, the same is hypothesized regarding dolichocephalic apteronotid species.
Mormyrids and apteronotids occupy nonoverlapping areas of the common skull morphospace, and mormyrids exhibit less total disparity ( Figs. 5 and 6 ). Both families include species with foreshortened and elongate skulls, but some apteronotids (i.e., Sternarchorhynchus ) exhibit the most extreme dolichocephalic phenotypes in the morphospace ( Fig. 5 ). Mormyrids have shorter skulls in general, and Petrocephalus occupies the most extreme brachycephalic portion of the morphospace. The mormyrids sampled do not occupy the extreme dolichocephalic (long snouts with a small braincase) end of the continuum in part because they have a relatively larger brain and braincase than do apteronotids ( Fig. 5 ;Carlson et al. 2011 ;Carlson and Gallant 2013 ;Stevens et al. 2013 ).
Mormyrids also occupy the exact range of PC2 values left unoccupied by apteronotids in the PCA, indicating stark differences in mouth position and dorsoventral flexion of apteronotid and mormyrid snouts ( Fig. 5 ). Mormyrids may be morphologically constrained by sensory related tissues that extend into the nasal region, leading to sub-terminal mouths and rounded foreheads ( Carlson et al. 2011 ;Carlson and Gallant 2013 ;Stevens et al. 2013 ). An extreme example is Petrocephalus with very large bony canals at the anteriormost region of the skull ( Fig. 8 ). In contrast, many apteronotids have terminal mouths, perhaps because some species have a habit of male combat which involves biting the tail of competitors ( Lundberg et al. 1996 ;Albert and Crampton 2009 ). A terminal mouth in the apteronotid taxa that have it may therefore allow the ethmoid region of the skull more dorso-ventral freedom to flex forming concave-down or convex-up snout morphologies ( Fig. 8 ).
Although weakly electric fishes occupy a large portion of the morphospace, there is a large, empty area with no representatives of either group (low PC1 values coupled with high PC2 values). This empty region of the morphospace is where we would see species with dolichocephalic, upturned snouts, a phenotype not observed in apteronotids or mormyrids ( Ford et al. 2022, Ford et al. 2022 ), but which is observed in some other teleost fishes (Aulostomidae, Syngnathidae, etc.) ( de Lussanet and Muller 2007 ;Lees et al. 2012 ). We interpret this empty region of the morphospace as a constraint in apteronotids and mormyrids, although not among teleost fishes in general. The relatively distant evolutionary relationship of apteronotids and mormyrids, with a most recent common ancestor approximately 150 million years ago, suggests that shared history is a poor explanation for the "missing phenotypes" in these clades. The role of genetic or developmental mechanisms underlying these phenotypic constraints could be explored using genetic editing methods and ontogenetic studies across multiple taxa.

Conclusion
The story of craniofacial evolution between Afrotropical mormyrid and Neotropical apteronotid electric fishes is complex, illustrating several common themes in comparative biology. The morphological similarities observed within each clade exhibit a mosaic pattern of occurrence among species, with many examples of phylogenetic convergence, divergence and stasis (conservatism). Individual traits may be convergent at one or more levels in the hierarchy of biological organization (e.g., cellular, tissue, organ, whole body) and not necessarily at other levels. Certain phenotypes of external anatomy (e.g., body size, head and mouth shape, eye size, fin configuration, etc.) may be similar despite different underlying structures (e.g., skeletal, musculature, nervous innervation, etc.). Although these external phenotypes evolved to perform certain functions and behaviors, they are not necessarily built by ontogenies in the same way. In this sense, morphological convergence between these groups may be viewed as superficial.
Supplemental Table 1. Catalog Information. Catalog and collection information for apteronotid and mormyrid species sampled in this analysis. Specimen numbers, museum ID numbers, and n -numbers are included.

Supplementary data
Supplementary Data available at IOB online.