Body shape differences in a pair of closely related Malawi cichlids and their hybrids: Effects of genetic variation, phenotypic plasticity, and transgressive segregation

Abstract Phenotypic differences may have genetic and plastic components. Here, we investigated the contributions of both for differences in body shape in two species of Lake Malawi cichlids using wild‐caught specimens and a common garden experiment. We further hybridized the two species to investigate the mode of gene action influencing body shape differences and to examine the potential for transgressive segregation. We found that body shape differences between the two species observed in the field are maintained after more than 10 generations in a standardized environment. Nonetheless, both species experienced similar changes in the laboratory environment. Our hybrid cross experiment confirmed that substantial variation in body shape appears to be genetically determined. The data further suggest that the underlying mode of gene action is complex and cannot be explained by simple additive or additive‐dominance models. Transgressive phenotypes were found in the hybrid generations, as hybrids occupied significantly more morphospace than both parentals combined. Further, the body shapes of transgressive individuals resemble the body shapes observed in other Lake Malawi rock‐dwelling genera. Our findings indicate that body shape can respond to selection immediately, through plasticity, and over longer timescales through adaptation. In addition, our results suggest that hybridization may have played an important role in the diversification of Lake Malawi cichlids through creating new phenotypic variation.


| 4337
HUSEMANN Et Al. of phenotypic variation for selection to act upon. However, additional mechanisms have been recognized as generators of phenotypic variation.
Therefore, recently separated species driven apart by consistent divergent selection are not expected to exhibit transgressive segregation (Albertson & Kocher, 2005). However, due to their potential to generate phenotypic variation, both phenotypic plasticity and transgressive segregation have been suggested to play crucial roles in adaptive radiations (Genner & Turner, 2012;Seehausen, 2004;Selz et al., 2014).
One of the most diverse and best-known adaptive radiation is represented by the East African cichlid fish. More than 2000 species of haplochomine cichlids in the three East African Great Lakes (Tanganyika, Victoria, and Malawi) exhibit an extraordinary amount of phenotypic diversity allowing cichlids to occupy all major ecological niches within the lakes (Seehausen, 2006;Sturmbauer, Husemann, & Danley, 2011). Of the three lakes, Lake Malawi harbors the most species-rich radiation with more than 700 species (Danley et al., 2012). Selection is thought to be the main driver of diversification in the different stages of the cichlid radiations (Danley & Kocher, 2001;Muschick et al., 2014). In the early stages, natural selection may lead to macrohabitat divergence and the differentiation of genera with different trophic traits, respectively. The divergence in trophic traits can be accompanied by the divergence in body shapes helping to exploit a variety of resources and microhabitats (Hulsey, Roberts, Loh, Rupp, & Streelman, 2013;Husemann, Tobler, McCauley, Ding, & Danley, 2014) leading to highly complex communities (Ding, Daugherty et al., 2014). During the most recent stage of diversification, natural and sexual selection drove the divergence of signaling phenotypes, microhabitat preferences, and body morphologies Kerschbaumer, Mitteroecker, & Sturmbauer, 2013;Streelman, Albertson, & Kocher, 2007;Sturmbauer, 1998). Hence, body shape is a key trait at multiple levels of the cichlid radiation.
Although body shape in fish can evolve in response to a variety of evolutionary forces, including predation (Langerhans, 2009), abiotic environmental factors (Neves & Monteiro, 2003), and competition (Scott & Johnson, 2010), it often evolves in response to ecological selection and can be used as an ecological marker when studying differentiation in natural populations (Tobler et al., 2008). A classic axis of body shape divergence in fishes is the divergence into deep-bodied and slender-bodied morphs associated with adaptation to benthic and limnetic macrohabitats. This pattern has been documented in a variety of species (e.g., Schluter, 1993;Willacker, Hippel, Wilton, & Walton, 2010) including Lake Malawi cichlids (Hulsey et al., 2013).
However, the underlying mechanisms which drove this divergence remain unknown. While several studies have addressed the genetic makeup, plasticity, and transgression of body shape in other cichlid radiations (e.g., Franchini et al., 2014;Kerschbaumer, Postl, Koch, Wiedl, & Sturmbauer, 2011;Kerschbaumer et al., 2013;Klingenberg, Barluenga, & Meyer, 2003;Selz et al., 2014;Stelkens et al., 2009), we have relatively little information on Lake Malawi cichlids. To gain a better understanding of the forces driving body shape evolution in Malawi cichlids, we performed a common garden experiment and generated hybrid crosses between two closely related species of the genus Maylandia. We expected the main differences between species remaining stable in fish bred under standardized conditions, yet also predicted a plastic response to the new environment. Our experimental design further allowed us to investigate the mode of gene action underlying the differences in body shape and to determine the potential for transgressive segregation. As body shape is a complex, modular phenotype, we expected that the differences would not be explained by a simple additive model, but rather involve epistatic interactions, as it has been shown for example in sticklebacks (Schluter, Clifford, Nemethy, & McKinnon, 2004). Complex gene interactions are thought to promote transgressive segregation, generating new body shape phenotypes as a result of hybridization (Rieseberg, Widmer, Arntz, & Burke, 2003;Rieseberg et al., 1999;Tobler & Carson, 2010). However, we expected transgression to be limited in this pair of closely related species, as the amount of transgression exhibited in a cross is often correlated with the genetic distance between the parentals (Stelkens & Seehausen, 2009;Stelkens et al., 2009). Overall, we anticipated to gain an insight into the roles of plasticity and transgression for body shape divergence in the adaptive radiation of Lake Malawi cichlids.

| Study species
We used two closely related species of the rock-dwelling cichlid genus Maylandia to study the environmental and genetic components of body shape differentiation and to test for transgressive segregation.
Maylandia benetos is a microendemic only occurring at a single location in the lake, Mazinzi Reef, where it is sympatric with three other Maylandia species, including Maylandia zebra. Maylandia zebra is one of the few rock-dwelling cichlids (mbuna) found at most rocky habitats throughout the lake (Ribbink, Marsh, Marsh, Ribbink, & Sharp, 1983).
Both species can be readily distinguished by their body coloration, distinct behavior, and microhabitat choice (Danley, 2011;Ding, Curole, Husemann, & Danley, 2014;Husemann et al., 2014). The species do not hybridize in nature, but can produce viable and fertile offspring if artificially fertilized (Ding, Daugherty et al., 2014). A previous study has shown that sympatric barred and nonbarred Maylandia species, including M. benetos (nonbarred) and M. zebra (barred) from Mazinzi Reef, are differentiated in their body shape in a predictable manner . To further understand the repeated, parallel divergence of body shape differentiating barred and nonbarred species, we used M. benetos and M. zebra as a model to study the transmission of body shape variation and to understand potential mechanisms creating phenotypic divergence in Lake Malawi cichlids. Specifically, (1) populations of both species were raised in identical aquaculture environments, resembling neither of the natural niches of either species, to quantify the degree of plasticity influencing this phenotype, and (2) these species were artificially hybridized in the laboratory to investigate the underlying mode of gene action and the amount of transgressive segregation observed in the body shape phenotype.

| Field sampling and laboratory breeding conditions
We collected adult specimens of M. zebra (N = 38) and M. benetos generations. Fish were kept in 110 cm × 28 cm × 30 cm tanks at water temperatures between 26 and 28°C. Light was kept at a 12-hr day/night cycle using timer-controlled fluorescent lights. Fish were fed a mixture of food flakes twice daily. Bidirectional F 1 (N = 96) were produced from five independent broods, and these F 1 were used to produce F 2 (N = 326) from three independent crosses. We then generated one backcross with each parental species (F 1 × M. benetos N = 20, F 1 × M. zebra N = 42). A total of 702 individuals were used in this study (Table 1). Pictures were taken with a Canon Eos 540d under standardized conditions including a ruler as length standard. All specimens were sexed, and the standard length was measured.

| Morphometric analyses
We quantified body shape variation in the two species and hybrid generations using geometric morphometric analyses. Lateral pictures of individual fish were imported into tpsDig v.2.16 (Rohlf, 2006), and we digitized 16 landmarks (Figure 1, see figure caption for a description of the landmarks). To address our question regarding the genetic and environmental components of body shape, we used data collected from wild-caught and laboratory-reared M. benetos and M. zebra. In a second analysis, we tried to determine the mode of gene action and test for transgressive segregation in the hybrid generations. For this, we included all laboratory-reared parental generations and the F 1 , F 2 , and backcrosses. For each data set, landmark coordinates were aligned using least-square superimposition as implemented in the program tpsRelw (Rohlf, 2007)

| Estimating genetic and plastic components
To distinguish genetic and plastic components of shape differen-  (Klingenberg & Spence, 1993). This allows for the visualization of body shape variation in response to a particular factor, while correcting for all other effects in the model.
Individual divergence vector scores were then used as independent variables in tpsRegression (Rohlf, 2005) to generate thin-plate spline deformation grids highlighting shape differences among groups.

| Analyses of body shape in laboratory crosses
To investigate the mode of gene action and test for transgressive segregation in body shape, we examined the laboratory-reared parental species and their F 1 , F 2 , and backcross hybrids for a total of 620 individuals. We calculated divergence vector scores from the species term of a MANCOVA described above and plotted the means and variances for each generation. If the body shape variation between the parental species is heritable and the genetic basis for these differences is additive, all hybrid generations' mean phenotypes are expected to be intermediate to the parentals. However, the variance in the F 2 generation is expected to be higher in comparison with the parentals and the F 1 hybrid generation (Lynch & Walsh, 1998). To test for additive effects, we compared the additive model of gene action to the additive-dominance model using the joint-scaling test to determine which model better fits our data (Lynch & Walsh, 1998). The joint-scaling method can also be used to detect the action of epistasis; however, testing for epistasis would require more than the six lines available from our cross (P 1 , P 2 , F 1 , F 2 , BC 1 , and BC 2 ). Therefore, we used a t-test based on P 1 , P 2 , F 1 , and F 2 data as suggested by Lynch and Walsh (1998) to evaluate epistasis.
To estimate and quantify the amount of transgressive segregation found in our crosses, we used two separate approaches. First, we em- As sample sizes differed across generations, we performed a second analysis testing for transgression while adjusting for different sample sizes. We generated estimates of convex hull volumes for each parental species, the combined parentals, and the F 2 generation. The convex hull of a set of points is a geometric measure describing the volume of the smallest convex set of points that contains all points in that data set. We removed variation due to sex and size by using the residuals from a preparatory MANCOVA and performed a PC analysis using the first nine axes (all axes with an Eigenvalue larger than the mean Eigenvalue) to calculate a convex hull for each group using the Quickhull algorithm (Barber, Dobkin, & Huhdanpaa, 1996). Due to different sample sizes among groups, we used a randomization procedure to calculate morphospace as described in Tobler and Carson (2010). Random distributions of morphospace were generated using 1,000 iterations of randomly selected specimens with replacement from the respective pool of individuals. A convex hull was calculated for each sample. Means and confidence intervals were calculated for each group through the examination of 1,000 iterations of this process. If body size exhibits transgressive segregation, the F 2 is expected to occupy significantly more morphospace than that of the combined parental generations.
Finally, we also visualized body shape variation of parentals and hybrids in order to identify the nature of shape differences among groups. To do so, we generated two distinct divergence vectors: (1) We calculated a divergence vector describing the morphological gradient between the two parental species.

| Genetic and plastic components of body shape
The analysis of wild-caught and laboratory-reared fish revealed significant plastic and genetic components that influence body shape.
The greatest differences between the species, independent of the environment, were found in the shape and slope of the head and body depth (Figure 2, upper panel). When examining the differences between the treatments, wild-caught individuals of both species had much deeper bodies with higher caudal peduncles, whereas body shape was more elongated and fusiform in the laboratory-reared individuals ( Figure 2, lower panel). The variance in the species score was higher in the wild-caught samples compared to the laboratory-raised fish. The MANCOVA showed that all main effects and most interaction terms were significant (Table 2); however, the effects of standard length and sex and the interaction terms involving sex were generally weak (η 2 p < 0.15). The rearing environment (laboratory vs. field, plastic component) had the strongest effect on body shape (η 2 p = 0.585), followed by the species identity (genetic component, η 2 p = 0.400). The interaction of species by environment was significant as well, yet had a weaker effect (η 2 p = 0.356).

| Body shape in laboratory crosses
In the laboratory-bred stocks, the means of the divergence vector  Table 2 T A B L E 2 Results of the multivariate analysis of covariance (MANCOVA) of body shape in the laboratory-field comparison of Maylandia benetos and M. zebra. F-ratios were approximated using Wilks' lambda; effect sizes were estimated with partial Eta squared (η 2 p ). Significant values are printed in bold  benetos. In addition, the variance of the F 1 generation was lower than that observed in M. benetos. The variance of the F 2 (N = 326, μ = 2.28, σ 2 = 1164.61) was only slightly higher than that in M. benetos ( Figure 3). The additive-dominance model was rejected as well (p.AD = 1.01 × e −14 ); yet, when comparing both models, the additivedominant model explains the data slightly better (p.A.AD = 4.68 × e −08 ).
As we did not have sufficient hybrid lines, we could not use the jointscaling test for epistasis. Instead we used the test based on the variances of the parental lines, F 1 and F 2 as proposed by Lynch and Walsh (1998).
The t-test could not reject epistatic effects (test statistic: −0.22; epistasis can only be rejected if this value is above 1.96; Lynch & Walsh, 1998).
Using the approach provided by Stelkens et al. (2009), our data show clear signs of transgressive segregation (Figure 4, Table 3). The range of the F 2 phenotypes clearly exceeds the ranges of shape space on each PC of the combined parentals (Figure 4). The amount of transgression found in the combined hybrid generations differed between 7% (PC4) and 52.1% (PC3) ( Table 3). The total amount of transgression across all axes and adjusted for the variance explained by each axis was 23.8% (Table 3).

| DISCUSSION
In this study, we performed a common garden experiment and hybrid crosses to estimate the potential of plasticity and transgressive segregation contributing to body shape variation in a sympatric pair of F I G U R E 3 Mean vs. variance of the species vector plotted for the parental and hybrid generations F I G U R E 4 Transgressive segregation in body shape. PC1 and PC2 for body shape were plotted for the Maylandia benetos (green triangles), M. zebra (blue kites), and F 2 hybrids (red triangles). Note that only the first two PC axes are shown, and therefore, the total amount of transgressive segregation is not displayed here T A B L E 3 Amount of transgressive segregation found at each axis calculated according to Stelkens et al. (2009) closely related rock-dwelling cichlids. We found that the body shape differences between the two species were largely maintained despite clear differences between wild-caught and common garden-raised fish. The mode of gene action underlying variation in body shape is complex, and the additive and additive-dominance models were both rejected. Furthermore, the analysis of body shape variation in hybrids revealed a high potential for transgressive segregation despite low genetic divergence between the species suggesting that transgression as well as plasticity may generate considerable amount of phenotypic variation. Transgressive phenotypes resemble some other taxa within the Lake Malawi cichlid radiation suggesting that some of the radiation may have been seeded by variation resulting from hybridization.

| Genetic and plastic components of body shape
Our data clearly show that species differences in body shape have a genetic basis. Despite significant differences between wild-caught and laboratory-reared fish, body shape differences observed between the two species in the wild are maintained after ~12 generations in a laboratory environment (Figure 2). In addition, the data from hybrids support a genetic component of body shape with the F 1 and F 2 generation being roughly intermediate between the two parental species (Figures 3 and 6). This strong genetic component to variation body shape suggests that the species are adapted to different microniches in nature and that this adaptation is the result of habitat-or community-specific selective pressures .
Nonetheless, there is also a strong plastic component to body shape: Both species consistently changed body shape under standardized laboratory conditions ( Figure 2). The changes followed a similar path, as both species became more slender with shallower bodies and thinner caudal peduncles. This might be a plastic response to smaller spaces, less activity, a structurally less complex habitat, and/ or reduced predation pressure in the laboratory (Kerschbaumer et al., 2011). The observed plasticity in body shape may play an adaptive role in the diversification of Lake Malawi cichlids; this plasticity may allow species to rapidly respond to novel environmental challenges.
This in turn may prevent competitive exclusion in species-rich and F I G U R E 5 Comparison of morphospace occupation (Convex hull volume) in each of the parental species, both parentals combined, and the F 2 hybrid generation adjusted to different sample size using the Quickhull algorithm; 95% confidence intervals are too narrow to be visible (see online supplement Table for actual values) F I G U R E 6 Visualization of the nature of transgressive phenotypes: species and generational differences along the species divergence vector (top) and along the parental/hybrid divergence vector (bottom).
Presented are estimated marginal means (and standard errors) of divergence vector scores highly competitive communities like those observed in rock-dwelling cichlids (Ding, Daugherty et al., 2014;Ghalambor et al., 2007;Olsson & Eklöv, 2005). Furthermore, a plastic response can lead to heritable adaptive changes, if selection favors a specific character state within the reaction norm across multiple generations (Via et al., 1995, see below).
Plasticity in body shape is not surprising as cichlids exhibit plasticity in a variety of morphological traits, such as jaw morphology and dentition (Muschick, Barluenga, Salzburger, & Meyer, 2011), body shape (Wimberger, 1992), mouth orientation, the size and orientation of fins, and the thickness of the caudal peduncle (Kerschbaumer et al., 2011). Interestingly, we find many similar traits to be plastic in our hybrid crosses: Body depth, orientation of the mouth, and the thickness of the caudal peduncle show strong variation between environments ( Figure 2). These traits are important for feeding and swimming performance, and therefore, they can be assumed to be under strong ecological selection (Langerhans & Reznick, 2009). However, even in traits under strong selection, high genetic variability can be maintained if strong species-by-environment interactions occur (Greenfield, Danka, Gleason, Harris, & Zhou, 2012). Such interactions, similar to genotype-by-environment interactions, may provide the genetic variation to quickly react to environmental change via plastic responses (Rodriguez, 2012). The resulting phenotypic plasticity may have contributed to the large phenotypic diversity observed in the East African cichlid radiation.

| Analyses of hybrid crosses
In a second step, we hybridized our two model species to get some insights into the genetic basis of body shape in these species. Little is known about the genetic basis of fish body shape. In sticklebacks, divergence in body shape between marine and freshwater lines appears to be determined by many genes (Schluter et al., 2004). In turn, a QTL analysis of the same stickleback species indicated that body shape is determined by few genes with large effects in addition to multiple genes with smaller effects (Albert et al., 2007). In cichlids, the genetic basis of body shape has been investigated in a pair of crater lake cichlids (Franchini et al., 2014). The study used genomewide SNPs and geometrics morphometrics to identify QTL and found that few genetic regions of large effects contribute to the divergence along the benthic-limnetic axis (Franchini et al., 2014). However, several other traits, including tropic morphology, coloration, and mate choice, have been studied (Albertson & Kocher, 2005Albertson, Streelman, & Kocher, 2003a,b;Ding, Curole et al., 2014;O'Quin, Drilea, Conte, & Kocher, 2013;O'Quin, Drilea, Roberts, & Kocher, 2012): These studies, for example, suggested that one to eleven genetic factors underlie shape differences of individual elements of trophic structures, and pleiotropic effects appear to be a common feature in the genetic architecture of these traits (Albertson et al., 2003a). Our data similarly indicate a complex genetic basis of body shape. The additive and dominant modes of gene action were rejected and epistatic interactions seem likely. Despite our limited knowledge, our data and previously published studies suggest that body shape represents a composite trait (see also Selz et al., 2014)

| Transgressive segregation
Another source of phenotypic variation is transgressive segregation resulting from hybridization of distinct genetic lineages. Hybridization is often considered a force leading to a decline of biodiversity, because it disrupts species boundaries (e.g., Perry, Lodge, & Feder, 2002).
However, the elevated genetic and phenotypic variance resulting from hybridization provides new variation that selection can act upon.
A variety of cichlid phenotypes are known to exhibit transgressive segregation. Albertson and Kocher (2005) et al., 2012). This high potential for transgressive segregation in different phenotypes suggests that hybridization can promote evolvability in East African cichlids and might be an important mechanism in generating new variation for selection to act on (Parsons et al., 2011;Seehausen, 2004;Stelkens et al., 2009). This has led some to suggest that transgressive segregation resulting from hybridizations at a variety of taxonomic scales has contributed to the origin of new species (Smith, Konings, & Kornfield, 2003), genera (Albertson & Kocher, 2005), and even whole clades (Genner & Turner, 2012) of Lake Malawi cichlids.
Our findings suggest that body shape is yet another cichlid phenotype exhibiting transgressive segregation: Hybrids occupy body shape morphospace beyond what is found in the parentals indicating significant transgression. F 1 and F 2 hybrids have more fusiform bodies in comparison with both parentals, a phenotype resembling the "aggressive" and "elongata" species groups within the genus Pseudotropheus (Figure 7; Konings, 2007). The amount of transgression found in the hybrid crosses was 23.8 %, which is remarkably high for a cross of two species so closely related (Stelkens & Seehausen, 2009;Stelkens et al., 2009). It has to be noted, however, that the approach for quantification is dependent on the coordinate system used. New less biased approaches need to be developed to make results more comparable across studies. The genetic distance between these two Maylandia species is estimated at 0.002 (based on mitochondrial D-Loop sequences), which is lower than any of the distances separating other cichlid hybrids reporting transgressive segregation (Albertson & Kocher, 2005;Genner & Turner, 2012;Stelkens et al., 2009). Furthermore, the amount of transgression observed in our study exceeded estimates of transgression for F 2 generations of taxa with an order of magnitude higher genetic divergence.
This runs counter to the typical pattern of transgressive segregation in which the degree of phenotypic novelty increases with genetic distance and suggests that hybridization can lead to high amounts of transgression, even in very closely related species such as those commonly found in Lake Malawi. In addition, the observation of transgressive segregation in two such closely related species suggests that, even within very recently diverged Lake Malawi cichlids, stabilizing selection on body shape can play an important role in shaping the observed phenotypic diversity in the lake (Rieseberg et al., 1999).
Overall, our results and those of previously published studies suggest that transgressive segregation seem to be the rule rather than the exception in cichlids, and as such, transgressive segregation may be an important mechanism in generating phenotypic variation in cichlids and may have played an important role in the evolution of this adaptive radiation.

| CONCLUSIONS
Our analysis of body shape in wild-caught and laboratory-reared specimens of two closely related species of cichlids revealed that speciesspecific differences have a genetic basis. In addition, body shape has a plastic component providing the potential to promote and maintain diversity. The mode of gene action of the species differences is complex, likely polygenic, and involves dominant and epistatic interactions. The potential for transgressive segregation is high, supporting the possibility of an important role of hybridization of closely related species in cichlid diversification. Therefore, our study supports the idea that plasticity and transgressive segregation as result of occasional hybridization may have been important factors in the evolution of the Lake Malawi cichlid radiation and potentially in other rapidly diverging systems.

ACKNOWLEDGMENTS
We would like to thank Richard Zatha, Aimee Howe, and Jason Curole for help with fish catching and processing in the field. We are grate-