Conceived and designed the experiments: CK CB. Performed the experiments: CK CB JP. Analyzed the data: CK CB. Wrote the paper: CK CB
Current address: Department of Biology, University of Antwerp, Antwerp, Belgium
The authors have declared that no competing interests exist.
The nature of developmental buffering processes has been debated extensively, based on both theoretical reasoning and empirical studies. In particular, controversy has focused on the question of whether distinct processes are responsible for canalization, the buffering against environmental or genetic variation, and for developmental stability, the buffering against random variation intrinsic in developmental processes. Here, we address this question for the size and shape of
Developmental buffering is an important factor in evolutionary processes, because it can maintain adaptive phenotypic traits in the presence of genetic and environmental variation and it can conceal genetic variation from selection
Empirical studies have tackled the question of whether canalization and developmental stability are distinct processes by comparing variation among individuals and the left-right asymmetries within individuals. Two main approaches have been used, which focus either on the amounts of variation or on covariance structures of multivariate features such as shape. Some studies have indicated that the amounts of individual variation and fluctuating asymmetry (FA) are correlated among genotypes
This study used both these approaches simultaneously in the context of an experimental design with complete control of genetic variation, replicated for 115 distinct genotypes from the Exelixis deficiency stocks of
We digitized 15 landmarks on the left and right wings of each fly (
Source | Sums of Squares | Degrees of freedom | Mean Squares ×106 |
Individuals | 0.1366 | 1846 | 74.0 *** |
Sides | 0.0015 | 26 | 57.0 *** |
Individuals×sides | 0.0307 | 1846 | 16.7 *** |
Measurement error 1: Imaging | 0.0015 | 3744 | 0.403 * |
Measurement error 2: Digitizing | 0.0028 | 7488 | 0.378 |
Procrustes analysis of variance
***
We used two different methods to quantify the amounts of shape variation among individuals and FA
The amounts of variation differed markedly among the different genotypes, although there was also a consistent effect of the vials in which the flies had been reared. For centroid size, the ANOVAs indicated that the variation among genotypes exceeded the variation among vials both for variation among individuals (
For both measures of shape variability, the amounts of shape variation among individuals and of shape FA were significantly correlated across genotypes (
In contrast, the correlation between individual variation and FA of centroid size was low and not statistically significant (
(A) Shape variation and FA quantified by Procrustes distance.
(B) Shape variation and FA quantified using Mahalanobis distance.
(C) Variation and FA of centroid size.
The solid lines are major axis regression lines, and the dashed lines in (A) and (C) are the theoretical limits for the situation when left and right sides are independent (FA variance is 4 times the variance among individuals; see text for details).
The correlations between the amounts of FA of shape and of centroid size were 0.46 and 0.36 for the shape measures using Procrustes and Mahalanobis distances, respectively (
To assess the possibility that this association was caused by a direct developmental link between size and shape, we tested for allometry within genotypes by multivariate regression of shape on centroid size
To examine whether among-individual variation and FA primarily concern the same or different features of shape, we quantified the degree of congruence between the respective patterns of covariation in landmark shifts. For this purpose, we computed matrix correlations between the respective covariance matrices for those 95 genotypes for which there were at least 50 specimens. Matrix correlations were computed both with the diagonal blocks included and excluded to examine whether the total patterns of landmark variation differ from the patterns of covariation among different landmarks
(A) Matrix correlations including the diagonal blocks (variances and covariances for
(B) Matrix correlations for covariance matrices without the diagonal blocks (only covariances among landmarks).
This study shows a significant genetic effect on the amounts of individual variation and FA as well as a clear association in both the amounts and patterns of variation between individual variation and FA for wing shape. This is consistent with the idea of a common genetic and developmental basis for buffering of wing patterning processes against variation from different sources. The correlations between size and shape in the amounts of FA and individual variation provide further evidence in favor of a common basis for buffering. In contrast, the lack of association between the amounts of individual variation and FA for centroid size indicates that these relationships depend on the specific traits under study and the processes involved in their development. Here we discuss these findings and their implications for interpreting the mixed results of published empirical studies on canalization and developmental stability.
Our data indicate a clear association of the amounts of individual variation and FA of wing shape across genotypes (
In stark contrast to the shape data, the association between individual variation and FA did not hold in the analysis for centroid size (
The discrepancy between these findings for size and shape highlights a methodological problem inherent in studies of developmental buffering: how can the effects of buffering be distinguished from differences in the initial input of developmental variation? Buffering is only observable if there is variation, and the resulting phenotypic variation is the joint expression of both the input of variation and the buffering of that variation by the developmental system. The original amount of variation, however, which is the input for the buffering processes in the developmental system, is unknown. The input of variation and buffering are therefore almost inextricably linked and cannot be separated without specifically designed experiments. Here we used samples of flies with controlled genotypes, so that genetic variation within samples can be ruled out. However, non-genetic effects cannot be controlled in this manner. The theoretical limit is a situation in which the environment is held constant so that the conditions under which the wings of two different flies develop are no more different than the conditions encountered by the two wings of the same fly. In this case, the left and right wings of individual flies would not be correlated, and the variance for individuals, var(0.5(right+left)), would be one-quarter of the variance for asymmetry, var(right−left). For shape, individual variation (quantified using Procrustes distance) exceeded this theoretical limit consistently, but only by relatively small amounts (
The correlation between the amounts of FA for centroid size and shape across the 115 genotypes exceeds the within-sample correlations of size and shape asymmetry for all but a few samples. Therefore, the direct developmental association of size and shape is not sufficient to account for the agreement of amounts of FA of size and shape. This is further evidence for a common genetic control of developmental variation of size and shape, although the data do not permit one to distinguish whether this control affects the origin of developmental noise or the developmental stability buffering against it. The association across many deficiency genotypes affecting different genomic regions may also be taken as evidence that a range of different genes contribute to the control of developmental stability, rather than just a few specialized genes
We not only compared the amounts of variation, but also the patterns of landmark shifts associated with individual variation and FA. There is a close and consistent correspondence between the patterns of individual variation and FA. A similar correspondence of patterns of shape integration for individual variation and FA has been found previously in the wings of
Because each of our samples was genetically uniform, we can rule out a contribution from allelic differences to the variation among individuals, which would produce effects that depend on the genetic composition of the sample and usually would differ from the within-individual effects. Imagine a population in which one locus with two alleles affects shape, so that the allelic differences will cause variation along a single line (with additive effect only) or in a plane (with additive and dominance effect). Unless the non-genetic components of variation also happen to be concentrated in the direction of this particular line or plane, the two components of variation will therefore be different. Even when more complex genetic models are used, the covariance structure among individuals will depend on the particular mix of genotypes, and may not reflect the inherent patterns of canalization. This reasoning can explain the closer resemblance of the patterns of FA to those of environmental rather than of genetic variation that has been found in empirical studies that specifically examined this effect
Overall, the results of this study clearly indicate that both the amounts and the patterns of individual variation and FA of shape are associated consistently across a broad spectrum of distinct genotypes. This suggests that canalization and developmental stability for wing shape share a common basis
The flies used here were offspring from crosses between the Exelixis deficiency stocks
A set of 15 landmarks was digitized on each image (
The shape information was extracted from the landmark coordinates with a generalized least-squares Procrustes fit
To quantify individual variation and FA of wing size, we used the within-sample variance of the centroid size
We used two different methods to quantify variation, which are based on different measures of morphological distance: Procrustes distance and Mahalanobis distance
To test whether the amounts of variation differed among genotypes, we used an extension of Levene's test
To examine the correspondence between the amounts of individual variation and FA, we computed the variances based on the two distance measures for data sets with either the mean shapes of both wings or the signed (right−left) differences of wing shape. Product-moment correlations were then computed across genotypes. The statistical significance of correlations was assessed with permutation tests
Allometry within genotypes was tested by multivariate regression of shape on centroid size
For the strains for which at least 50 specimens were available, we also compared the patterns of shape variation between individual variation and FA
Strains used in this study and various sample statistics.
(0.36 MB DOC)
We thank S. Harold, K. Stewart, and P. Tickle for technical assistance and an anonymous reviewer and the editor for helpful comments.