Individual symmetry is believed to be advantageous and reflecting developmental stability, but frequency-dependent selection can also maintain polymorphisms of asymmetric phenotypes. There are many examples of so-called antisymmetry, where mirror image morphs occur at equal frequencies. With very few exceptions, these are caused by nongenetic variation. One notable exception is handedness and mouth bending variation in the scale-eating cichlid Perissodus microlepis, which has been suggested to be an example of antisymmetry determined by a single genetic locus of large effect. Here, we report that this handedness and mouth bending asymmetry are not jointly and exclusively determined by a single major locus. We found no evidence of a major locus for asymmetry and some support for a major handedness locus. Also, asymmetry is plastic in this species: it can change in adults. We suggest that behavioral handedness in this system precedes and guides morphological asymmetry.
In the scale-eating cichlid fish Perissodus microlepis, two phenotypic morphs are believed to coexist. Individuals having a mouth opening asymmetrically to the left are called left-handed, when the mouth opens to the right, the fish is called right-handed (Hori 1993). The morphs seem to differ in “flank specialization.” When attacking from behind, left-handed (or right-handed) fish preferentially rip scales from the right (or left) flank of their prey. To avoid confusion, we will call mouth bend “asymmetry” and behavioral preference “handedness” in this article. Some results support the idea that prey guard one flank better when one morph becomes more common, thereby automatically giving the rare other morph an advantage (Hori 1993). Other factors have been suggested to contribute to the maintenance of the polymorphism, such as lethality of homozygote right-bending asymmetry (Hori et al. 2007), interspecific coupling of asymmetries and handedness, and disassortative mating for asymmetry (Takahashi and Hori 2008).
While the evidence for ecological effects of asymmetry and handedness is convincing, our understanding of the determination of the traits is limited. Evolutionary theory states that several different modes of determination can evolve as a response to frequency-dependent selection (Rueffler et al. 2006). Phenotypic polymorphisms can be determined by genotypic differences, but also by nongenetic environmental input, such that one should not restrict attention to a specific mode too soon. Hori (1993) suggested that mouth bend in P. microlepis is heritable in a Mendelian fashion with a dominant gene for right-bending (left-bending recessive). In a recent review Palmer (2005) has pointed out that P. microlepis would be an unusual antisymmetry (Van Valen 1962) in that case, because most antisymmetries are environmentally determined. The data on asymmetry are contradictory. Two of the five families with left-left parents had about 25% right-bending individuals among the brood, contradicting the inferred Mendelian inheritance. In a more recent study, Hori et al. (2007) conclude that the gene for right-bending is homozygous lethal, whereas the 1993 data suggest that at least three parents were of that genotype. P. microlepis farm out their fry. Hori (1993) excluded differently sized fry from the data for that reason, but extra-pair young might still be present. One can conclude that the suggested inheritance system is incorrect, or that there still is a reasonable probability that family groups are heterogeneous, or both. In published photographs, the mouth asymmetry seems conspicuous, whereas an older histogram of asymmetry values (Hori 1991) suggests that there are near-symmetric individuals. Asymmetry determination is therefore likely not exclusively Mendelian, and both polygenic and environmental effects may contribute to asymmetry variation.
No study so far takes effects of measurement error into account to determine whether the distribution of asymmetry values is truly antisymmetric, or maybe just another case of fluctuating asymmetry (Polak 2003). The coupling between morphological asymmetry and behavioral handedness has only been investigated to a limited extent (Hori 1993), and not in controlled conditions. We also do not know whether morphology and behavioral preference would be determined separately, or sequentially.
We have therefore collected estimates of individual asymmetry and handedness in laboratory conditions to investigate their joint and marginal trait distributions. In addition, we have tested whether these traits are partly determined by environmental variation, by forcing individuals to forage from the flank of our preference.
Material and Methods
We have collected repeated measurements of asymmetry on 47 individuals and handedness observations on 39 of these. The fish were imported from Burundi in December 2005. The collection site was in the Northern part of the lake, on the opposite side from Uvira (Hori 1993) and far from Mpulungu in the Southern part of the lake (Hori et al. 2007). In the laboratory, the fish were housed individually and fed daily with live chironomid larvae and loose scales.
In our opinion, mouth bend asymmetry is difficult to determine by visual inspection alone (as in Hori 1993; Hori et al. 2007). It can suffer from bias or exaggeration when occurring on individuals fixed in ethanol or formalin. Because visual inspection tends to require some manipulation of the mouth, it is not well suited for repeated measurements on living individuals. We have estimated asymmetry in a quantitative manner using Procrustes shape analysis (Dryden and Mardia 1998) of head landmarks on digital photographs (Fig. 1A). We used a fixed setup to photograph simultaneous lateral and dorsal views in a single photograph. With a transparent blunt stick we gently pushed the fish for each photograph into a standardized upright position as much as possible. Per session, three photographs were taken. There were four sessions in February and April 2006 before the experiment described below started on June 20, 2006, two sessions at end of the experiment (December 2006), and one session two months after the end of the experiment. We decided to use the simultaneous lateral views only for putting the dorsal landmarks at the correct coordinates and not for analysis. The resolution of the lateral view was worse than the dorsal view due to the fact that the image was transferred through an aquarium wall and a mirror, and because extra measurement error in asymmetry was generated on the lateral views by the usually unequal distances of a fish to the two mirrors that reflected the images toward the camera. We measured landmarks on the dorsal view of minimum 12 photographs per individual, from at least the first four sessions to assess head shape variation and construct individual asymmetry values (Fig. 1). We carried out the Procrustes analysis on all seven sessions combined, and assessed the variation caused by each principal shape component using the shape library written by Mardia for R (Ihaka and Gentleman 1996; Dryden and Mardia 1998). We inspected components in order of magnitude of their relative contribution to shape variation, until we found a component that had clear effects on symmetry, from left-bending to symmetric to right-bending. The score for the shape component was calculated per photo. Other shape components described head shape variation from blunt to sharp, or effects caused by the fish not standing exactly vertical under the camera. We estimated individual size by measuring distances on each dorsal photograph between landmarks along the body until the caudal peduncle.
Individual preference for attacking a certain prey flank was determined by observing minimum 20 attacks per individual on goldfish. Goldfish were added one at a time to a tank with a single cichlid. There were never more than seven attacks per goldfish, and all regenerated the lost scales within three weeks. Eight fish did not attack goldfish.
DATA PROBABILITY DISTRIBUTIONS
Even though our data are used to assess variances of traits of individuals in a population, only fixed effects were used to extract estimates of individual trait values. This avoids the distributional assumptions of most random effect models. The handedness data were first analyzed using a binomial generalized linear model, with fish and date effects. There were no significant date effects, so we further used the proportion of attacks directed to the left flank as a handedness measure. The asymmetry scores from photographs taken before the experiment were treated as asymmetry measurements and analyzed using a linear mixed model with different error variances per individual, fixed individual and random session effects (function lmer, Pinheiro and Bates 2000). These last effects correct for random differences in the position of the camera on different days. Gape size did not affect asymmetry significantly and was therefore immediately removed from the model.
Antisymmetries (Van Valen 1962) can be called “weak” if the data distribution of a continuous trait is not bimodal but still more platykurtic than the normal distribution. Antisymmetries are “strong” when the data are clearly bimodal. A mixture analysis can give significant support to the presence of two components in a trait distribution, for both weak and strong antisymmetries.
We first tested whether the null hypothesis could be rejected that asymmetry and handedness were each samples from a unimodal distribution, using the dip test (Hartigan and Hartigan 1985). This is then a test for the presence of a strong antisymmetry. Additionally, we fitted mixtures of normal distributions (asymmetry) and binomial distributions (handedness) to the asymmetry trait values or attack data, and inspected the fit of mixtures with one to four components (FlexMix package, Leisch 2004). We used a parametric bootstrap test to investigate whether trait distributions were composed of one or two component distributions (McLachlan and Peel 2000), if the main comparison to be made was between these two numbers of components. This test can detect both strong and weak antisymmetries. As suggested in other studies (Palmer and Strobeck 2003), we conducted an Anscombe-Glynn test for platykurtosis relative to the normal distribution (Anscombe and Glynn 1983). We view this as a test for the presence of at least a weak antisymmetry. In the case of handedness, we used proportions of attacks toward the left flank as trait values. We did not apply nonlinear transforms to the data that might introduce new clustering or separation in the distribution.
The tests on trait distributions were complemented with a post hoc power analysis of the dip test, the mixture analysis, and the Anscombe-Glynn test (Supporting Information). We estimated the power of the three tests if morphological asymmetry or handedness would be determined by a large genetic effect (from a locus with a dominant and a recessive allele), plus measurement error in the case of asymmetry. For the purpose of the power analysis, we estimated the measurement error of individual trait values using the repeated photographs per individual.
To test whether asymmetry and handedness can change during adult life, we selected 10 individuals with a strong preference for the left flank. From these, five randomly chosen individuals were allowed to forage scales on the flank of their preference, whereas the five others were forced to feed from their nonpreferred flank. This was achieved by presenting them soft-bait dummy fish wrapped in trout skin and with spikes preventing foraging from the “forbidden” flank. The fish ate exclusively from these dummies for six months. The asymmetry data on these individuals were analyzed using a linear mixed model (lmer, Pinheiro and Bates 2000). We were primarily interested in the fixed treatment effects and treated fish effects as random. The number of fish in this analysis was small and a specific subsample of our population, such that it is unlikely that their trait values significantly differ from a sample drawn from a normal distribution. We scored the handedness of these fish at the end of the experiment, as above.
At the time of the first session, the average fish length was 8.18 cm (SD 1.44). The asymmetry values explain 5.3% of shape variation in our data (see Supporting Information for an overview of shape components). From Figure 1A, it is clear that asymmetry variation is not very conspicuous and not easy to observe in a dorsal view. Overall, 62.5% of attacks were directed to the left flank. The number of individuals that attacked the left flank more than the right was nearly significantly different from 50% (25 vs. 14, χ2= 3.103, P= 0.08).
A linear mixed model with fixed individual effects, individual differences in error variance, and a random set effect fitted the data better (P < 0.001) than a model with a single error variance, indicating that individual differences in measurement error indeed occurred. The fixed individual effects were used as asymmetry trait values. A dip test applied to them did not demonstrate a significant deviation from a unimodal distribution (dip statistic 0.056, n= 47, P= 0.3). The Anscombe-Glynn test for a deviation from the normal demonstrated significantly platykurtosis (P= 0.025). However, the parametric bootstrap test did not reject the null hypothesis that the asymmetry traits are a sample from a single normal distribution (P= 0.11). When mixtures with two to four components are fitted to the data, these hardly differ in terms of log likelihood. We conclude that asymmetry shows at most a weak antisymmetry. The “single-component” normal distribution that fits the trait values best has mean 0.06 and standard deviation 0.71, which is about four times larger than the average measurement error SD, which is 0.17 (see Supporting Information). We plotted on purpose a histogram with 15 bins for asymmetry values (Fig. 1B), to show that the data do not show perfect normality for all choices of histogram categories, but never undisputable bimodality either (see Supporting Information for histograms with different numbers of bins).
The distribution of individual handedness values is not different from unimodal (dip statistic 0.058, n= 39, P= 0.35), but the mode is not at an even probability of attacking left and right. When mixtures of binomials are fitted to the attack data, distributions with several components always fit the data far better than a single binomial (difference in log-likelihood always more than 80). When comparing mixtures, the conclusions are less unequivocal and differ depending on the information criterion considered. The Akaike information criterion (Akaike 1973) and the Bayesian information criterion (Schwarz 1978) prefer a mixture with four components. According the integrated classification likelihood criterion (Biernacki et al. 2000), mixtures with two to four components are all nearly equally preferred. All these mixtures contain a binomial component with preference parameter within the range [0.3, 0.7], that is, there are clearly a substantial group of individuals with at most a weak handedness. The estimated binomial parameters Pi are, for two components (P1= 0.87, P2= 0.35), three components ((P1= 0.87, P2= 0.46, P3= 0.17), four components (P1= 0.99, P2= 0.82, P3= 0.45, P4= 0.17). Based on the attack data, a model in which handedness is exclusively determined by a major locus cannot be ruled out. However, there are no two clearly distinct groups with either a very strong preference for the left or the right, as predicted by the original Mendelian model (Hori 1993), but the antisymmetry for handed behavior certainly seems stronger than for morphological asymmetry.
The scatter plot (Fig. 1D) of morphological asymmetry and handedness does not show two clearly distinct clusters, in accordance with the analysis of the marginal distributions. Morphological asymmetry correlates significantly with handedness (rS= 0.50, P= 0.001, Fig. 1), and is therefore functionally significant.
INFLUENTIAL OBSERVATIONS—POWER ANALYSIS
Whether individual differences in measurement error affected conclusions on morphological asymmetry was investigated as explained in the Supporting Information. Individuals with weak asymmetry were not the ones for which the measurement error was relatively large (see also Fig. 1B). From the power analysis, we conclude that if asymmetry were determined by a single Mendelian locus and measurement error, and with differences between genotypes similar to the estimated difference between means in the two-component mixture, that our tests would likely reject the null hypothesis of a simple normal trait value distribution, and that the dip test would have been significant. Our power analysis also indicated that the Anscombe-Glynn test for platykurtosis has largest power when there was no underlying antisymmetry and only individual variation in measurement error, but that power was very weak (15–20%). The significant platykurtosis in our data could be caused by additive genetic variation or by environmental variation between individuals. For the handedness data, the power analysis indicated that the dip test had undesirable properties. Its power was 73% when the data were simulated from a single binomial distribution and hence when there would be no antisymmetry in handedness. The power decreased for weak antisymmetry, and increased again when that became stronger. The test for kurtosis never had good power in any of the handedness parameter settings simulated. Mixture analysis was therefore most appropriate for the handedness analysis.
Most of the 10 fish we selected for our controlled foraging experiment had a mouth oriented toward the right (Fig. 2). We measured asymmetry at the end of the experiment, and detected a significant treatment effect (P < 0.001, Fig. 2). The fish that could forage as they preferred increased their asymmetry further toward the right (asymmetry value increase 0.645). The fish foraging against their preference did not change significantly (asymmetry value change −0.138). This shows that asymmetry is developmentally plastic, even in mature fish, and that experimental treatments can create differences between individuals. The effect of session was not significant in this analysis. The fish foraging against their preference responded in a heterogeneous manner and not all as we expected. The significant treatment difference we found was thus mainly due to increased asymmetry in the control group. The heterogeneous response in the other group might be caused by new heterogeneity in attack behavior that we observed in that group. For example, fish started attacking from the front instead of the back. One individual learned to roll across the body of the prey dummy to reach the preferred side. When we included the data collected two months after the end of the experiment, the treatment effect was still significant (likelihood ratio test P < 0.001). To assess whether we could detect any significant change in an analysis with fixed effects at the individual level, we fitted a generalized linear model (gls, Venables and Ripley 2002) with different error variances between individuals, individual effects, and also estimates of individual changes at the end of the experiment. In this analysis, we found that we managed to make one asymmetric fish as good as symmetric (t-test P < 0.001, asymmetry value before the experiment 1.60 (SE 0.22), and at the end 0.22 (SE 0.25)). We could not detect a significant treatment effect in a handedness test carried out at the end of the experiment, but observed that overall, attack preference had become slightly weaker (from over 90% to 75% of attacks directed toward the left flank). This argues against a strong dependence of handedness on environmental conditions experienced by adults.
Neither morphological asymmetry nor handed behavior in P. microlepis appeared to exhibit a strong antisymmetry (Palmer 2005; Schilthuizen et al. 2007). The distribution of morphological asymmetry was not strictly bimodal and many fish did not show a strong attack preference for a specific flank. Exclusive determination of combined morphological asymmetry and behavioral handedness by a single major gene with dominance recessivity is therefore excluded because this would imply the absence of near-symmetric individuals or without a clear behavioral preference.
We propose that morphological asymmetry in scale-eating cichlids is more likely a case in which developmental plasticity (Polak 2003) contributes to trait variation. The platykurtosis we found for mouth bending suggests a weak antisymmetry. Our power analysis shows that just individual variation in measurement might cause that result, but that is not very likely because the power is very low. Because we do find much more variation in asymmetry than our estimate of measurement error predicts, we conclude that the platykurtosis can be caused by other sources of variation between individuals, for example additive genetic variation. We cannot reject the possibility that some genetic variation in morphological asymmetry is present or that genetic determination may be coupled with environmental variation.
In humans, handed behavior shows a mode of determination with genetic and environmental or developmental effects (Vuoksimaa et al. 2009). On the other hand, handed behavior in P. microlepis is multimodal, which still allows for exclusive determination by one or a few major genes. For handedness, the mixture analysis supports the conclusion that the handedness trait distribution consists of a few discrete groups with significant differences in handedness. There is no evidence of an environmental contribution to handedness differences over the time scale of the plasticity experiment. Overall, handedness seems a stronger antisymmetry than mouth bending.
Because behavioral handedness in so-called laterality traits is common in fish (Sovrano et al. 2005; Bisazza et al. 2007), and given that we can reinforce or limit asymmetry variation over time by affecting foraging, we suggest that behavioral handedness variation will precede and guide variation in mouth asymmetry. If there would be a major gene involved in foraging and asymmetry in this cichlid, we expect that it primarily affects behavioral laterality traits rather than morphology, such that any correlated effects on mouth asymmetry could be masked by effects of variation at other loci, making morphological asymmetry more apparently polygenic and with an environmental component. The same pattern of distributions would be found if juveniles can develop handed behavior through learning, and the reinforcement involved in learning makes the distribution of handedness in adults a mixture of discrete strategies. The fact that we could influence morphological asymmetry in an experimental treatment is in line with the general plasticity and versatility of the jaw apparatus in cichlids (Galis and Metz 2000).
Another issue to pursue is whether asymmetry as measured or determined on surface measures on intact fish (Hori 1991, 1993; Hori et al. 2007) strongly depends on underlying skeletal asymmetries (Liem and Stewart 1976) or not. In any case, the presence of near-symmetric individuals warrants a reassessment of the advantages and disadvantages of asymmetry in these cichlids. They could uncover an evolutionary constraint or pleiotropic fitness effects affecting the maintenance of the polymorphism.
Associate Editor: H. Hoekstra
TVD was supported by a Dutch NWO VENI post-doc grant. Rich Palmer and Angus Davison gave helpful comments on the manuscript.