Wing shape evolution is not constrained by ancestral genetic covariances in the invasive Drosophila suzukii

  1. Antoine Fraimout
  2. Stéphane Chantepie
  3. Nicolas Navarro
  4. Céline Teplitsky
  5. Vincent Debat

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

The extent to which phenotypic evolution can be constrained by genetic correlations is an important question in evolutionary biology. To address this question, biological invasions are opportune models where derived, invasive populations can be compared to their extant ancestors, allowing to track the evolution of genetic correlations from the ancestor, throughout the invasion process. In this paper, we focused on the worldwide invasion of Drosophila suzukii (Matsumara, 1931), and investigated the evolution of the genetic covariance matrix G of wing shape between ancestral native, and derived invasive populations. Leveraging demographic history resolved by population genetics approaches, we tested whether G remained stable during the invasion. Using a multivariate Q ST -F ST approach, we further tested whether or not the observed phenotypic divergence in wing shape aligned with a neutral scenario of evolution. Our results show moderate yet significant quantitative genetic differentiation of wing shape among D. suzukii populations and a relative stability in the structure of G , presenting a roughly spherical shape but slightly different volumes. These characteristics likely reflect the demographic history of populations and suggest a low level of genetic constraint on wing shape evolution. The divergence between populations was greater than expected under a purely neutral model of evolution, compatible with an effect of divergent selection among them. Overall, our study suggests that selection and drift, but not ancestral genetic constraints, affected the early stages of wing shape evolution during D. suzukii invasion.

Article activity feed

  1. Peer Community in Evolutionary Biology

    Dear Antoine Fraimout,

    I apologize for taking long to review your preprint, “Evolution of genetic (co)variances during the worldwide invasion of Drosophila suzukii.” I have received two reviews—one more enthusiastic than the other. Both reviewers and I agree that the work is of interest to invasion biology and quantitative genetics. However, there are critical issues with the analyses and interpretation that need to be resolved before I can consider recommending the preprint. Please address the following points in detail, as well as all those raised by the reviewers, if you wish to resubmit a revision for possible PCI recommendation (in no specific order of importance):

    1.   In the introduction, you state that your objectives are to determine whether the G matrix is stable despite a complex evolutionary history and to ascertain whether selection and drift can be implicated in phenotypic divergence. However, these two objectives are confounded in the analysis and interpretation of the results. Specifically, you need to be more explicit about your expectations for selection. In your first test for selection (line 253), you suggest that a difference between the coefficient of proportionality between G and D and predicted neutral Fst is indicative of selection. However, I am not entirely convinced that you can dismiss the possibility that demographic history could produce the same difference. Your second test for selection (line 281), using the breeder’s equation, is not really a test for selection once you conclude that the G matrices do not constrain evolution. I would suggest eliminating the analysis in Figure 5. In your third test (line 352), you ask whether there is less genetic variance in the main divergence dimension than by chance, but you do not consider whether drift alone under infinitesimal assumptions would also lead to less genetic variance. See also point 5 below.

    2.   The morphometric analysis of divergence between populations is hard to follow. I could not understand the difference between the PC (line 153, supplementary figure 2) and the divergence analysis (line 171, figure 2). For example, you mention two MANOVAs, one for each analysis (lines 163 and 174), but it seems only one was performed. Additionally, I could not understand what the pairwise permutational MANOVAs were testing for or if they were represented somewhere in the figures or a table. At one point, you mention that the angles between divergence eigenvectors are different and refer to table S2. However, table S2 shows the angles between divergence and gmax. Could you show the eigen decomposition for divergence as well as the null angle expectations for divergence in Figure 2? Additionally, please better explain how the wing outlines were obtained.

    3.   You do not seem to have the power to estimate a large G matrix for each population with your sample sizes. Critically, you need to show the null expectations of the expected genetic variance after eigen decomposition of the G matrices in a main figure. As suggested by one of the reviewers, most phenotypic variation in wing shape can likely be summarized in just a few dimensions (3-5). I recommend deciding on this number from the start, after the PCA, and then using these dimensions for the remaining analyses. If you do not have genetic variances for some or all of the populations, then you should continue with the comparisons you are conducting (G differences, divergence, selection). This is because evidence of a signal in your comparisons (or its absence) is not evidence of significance, to paraphrase a common trope. For example, the US-WAT population has more genetic variance than the other populations (Table 1), but what if you cannot show that some or all of them have more genetic variance than their respective null expectations for the first few eigenvectors? Similarly, with the eigentensor analysis, you attempt to interpret differences between populations despite them not being significant (line 358). Once the eigentensors are not different from chance, no further decomposition should be presented (Figures 3B and 3C). Your conclusion that the G matrices are similar and spherical is also not supported, as they could have different eccentricities that you cannot detect.

    4.   One possible solution to the lack of power could be to merge the data from populations of similar geographic origin and then estimate the G matrices with a location covariate (or perhaps using the Fst neutral markers as a kin matrix covariate). After all, you do not explicitly test for selection with the latitude cline. However, you would still face the problem of having only one Japanese ancestor, which might limit your comparisons to US and France differentiation. However, I am not entirely sure how legitimate such an analysis would be.

    5.   I found it difficult to interpret Figure 4, specifically what is being projected onto the main d eigenvector. Perhaps you could consider separating the A projections from the POP projections. The first projections seem to repeat the results from Figure 3, showing that the G matrices do not differ among themselves. Shouldn't the null be represented for each population? There are two steps that do not seem appropriate: first, testing for more variance along d for the populations, and then suggesting that there is selection because there is more variance than when the randomized ancestral G matrix is projected. Maybe I misunderstood what is being represented. In any case, besides the comparison with Fst, this is the figure that suggest selection during divergence and so it needs to be clear. What is the argument that larger than expected genetic variances in the main direction of divergence are not consistent with complicated evolutionary demographic histories?

    6.   I could not fully understand your experimental design. If the male wings were measured in brother- sister inbred lines, why was there a need to model a pedigree when estimating the G matrices? Additionally, please clarify the potential of non-additivity inflating your genetic variance estimates, which could confound the role of selection.

    7.   You need to present all the data and analyses. All variance-covariance matrices and their eigen decomposition estimates (P, G, D, S), including credible intervals and nulls, must be made available either as supplementary files in the preprint or on the GitHub site you mentioned (I accessed it, but it was empty/not publicly available). Likewise, the code for analysis should be made available. Please clearly state whether the divergence data and analysis have been published before. And finally, please be more attentive in referencing all the studies you mention in the text.

    I hope that these comments and those of the reviewers are helpful.

    Best regards,

    Henrique Teotónio

  2. Peer Community in Evolutionary Biology

    Fraimout et al. present an interesting study on the genetic architecture of Drosophila wings following a worldwide invasion of D. suzukii. They sampled two populations from the native range (Japan) and four populations from the USA and France, where the species has established relatively recently. Using isofemale lines, the authors estimated genetic variance in wing size and shape to understand whether phenotypic divergence has occurred following a global invasion, and whether divergence is associated with patterns of genetic variance. 

    They found evidence of moderate differences in phenotype and some differences in genetic variance in wing shape. Their main result is that G-matrices for wing shape are spherical and have not constrained phenotypic divergence following the recent invasion of D. suzukii to Europe and the US. Overall, I enjoyed reading the work, and I appreciate that this is a difficult subject to tackle. I did, however, struggle with the analyses and their biological interpretation. I have four major comments, I hope the authors find them helpful.

     

    Major comments:

    1) The description of the statistical tests could be simpler to make it easier to understand what was done and why. I found that the introduction was fairly light on hypotheses, and they were only touched upon briefly in the methods and results. To be more focused, it could help to make the hypotheses clearer in the introduction and then describe how they are tested in the methods and results. Otherwise, there is a disconnect between the hypotheses, methods and results. Sections I struggled with include the description of the geometric analysis (paragraph starting L.154), which is is a list of different methods, often without justification as to why they were done (e.g., L.166). The similarity of G and D (L.253 onwards) is similarly hard to follow, especially the rationale and decription of how selection is differentiated from drift. It would help if these areas stepped the reader through the methods a bit more carefully. The comparison between proportionality of G and the neutral expectation (using Martin et al. 2008) was difficult to grasp in the methods and results – this is a hypothesis test that the authors could consider including background in the introduction. Some clarity here would help the reader understand how you are discerning selection from drift in the alignment between G and D.

    2) In general, the analyses are comprehensive and robust. However, I don’t think that a G-matrix of 26 principal components is a good idea. Is there a justification for using 26 PCs rather than particular traits that encompass wing shape? While all PCs together capture shape, it is difficult to understand what each individual PC represents biologically – especially when decomposing G means that we are assessing eigenanalyses of principal components. This aside, my main concern is that while there is good replication at the line level (c.30 isofemale lines per population), a G-matrix created using 26 PCs is estimating 351 unique parameters (one diagonal + variances) so the number of observations (30 lines) is far lower than the number of parameters being estimated (and this doesn’t even include estimation of the residual covariance). This could be the reason that the authors found spherical G-matrices. Covariance matrices will become non-positive definite if too many parameters are estimated for the amount of data (an older paper that discusses this: https://doi.org/10.2307/2530605), which becomes problematic because MCMCglmm constrains matrices to be positive definite, and so will always estimate a matrix (even if the estimates are very imprecise), whereas other approaches, such as REML, will not allow such a model to run. It is therefore considered best practice to estimate less parameters than the number of observations at the necessary level of replication (e.g., a 6×6 matrix = 21 parameters, which would be more suitable for 30 isofemale lines). Given that MCMCglmm constrains variances to be positive, have the authors tested whether their estimates of genetic variance exceed random sampling (i.e., are significant)? Having a quick look at the model output, it seems as though many estimated parameters have posterior distributions centered on zero, but with very large ranges, often -500 to 500. This could be due to the nature of the variables analyzed, but I suspect means that the model parameters are not well-estimated. My suggestion would be to choose ~5 traits that encompass wing shape (the first 5 principal components, and/or the traits in Fraimout et al. 2018), scale them and test whether they produce the same results.

    3) I struggled to see the value of the selection analysis on flight velocity. The justification is a paper (Frazier 2008) that isn’t included in the reference list. Including a longer justification on why velocity should be used as a performance metric is important given that the breeders equation is based on evolutionary change using relative fitness. In the results, it seems that spherical G-matrices mean that evolution hasn’t been constrained – calculating delta z would then seem to be a little redundant as you already know that there is little constraint in G.

    4) The biological interpretation of the results is missing. Describing how different the populations are from each other and from they hypothetical answer is great, but with this information it is difficult to understand how wing shape has changed between the populations (e.g., rounder, more slender etc…). This is also the case, for example, on L.361-363 and L.378 (where there is reference to the wing figure, but the text only talks about MCMC samples). Furthremore, as the G-matrices (or their decomposition) are not available, it is difficult to understand what the G-matrices look like, or what they represent biologically. In any case, it would be nice to link the results with their biological significance, as has been done in Fraimout et al. 2018.

     

    Minor comments:

    - As a suggestion, the title could be clearer and more results driven – this paper has not explored everything about their invasion, but has used it for specific tests. Perhaps focus on the patterns of genetic variation following invasion.

    - It would be good to be clear on what the estimates of genetic variance represents given that isofemale lines have been used, which include non-additive genetic variation. The cited papers David et al. 2005 and Berger et al. 2013 provide good background for this.

    - L.50-51 the second paragraph could benefit from a more gentle introduction to the topic, in particular, phrases like ‘the orientation of genetic (co)variance between traits relatively to the direction of selection applied’ might be quite abstract for a broader readership.

    - L.71 The statement, “G cannot be taken for granted and requires additional empirical evaluation” could be more specific and feels a bit conversational. What about G needs to be explored empirically?

    - L.81-82 This is the premise of the study, so it could help to describe the differences in wing shape and size, where the populations were sampled and how the current study builds from this previous work.

    - L.85 Should “allows to investigate” be something like “provides the opportunity to investigate”?

    - L.96-104 Here it isn’t clear to me how comparing G and D can distinguish the difference between neutral versus selective effects on divergence. Under both conditions, G is being compared with D. In the second there is reference to ‘the axis of greatest genetic variation’, but isn’t this the same as gmax in the 1st condition? Overall, I found this section a little confusing, and it might be useful to describe it in more detail as a paragraph on comparing G and D before discussing the study system. 

    - L.145 It would be helpful to provide replication per population at the level of isofemale lines and replicates within.

    - L.173 The references are missing for the citations to Stuart et a. 2017 and Fraimout et al. 2022b.

    - L.189 ‘synthetize’ should be synthesize.

    - L.342 I understand what you mean, but I think how 1.5 reflects sphericity could be explained in a bit more detail. Also, the biological meaning (that genetic variance is in more than one direction) could be important to highlight.

    - L.368-370 This is quite difficult to follow and could benefit from a simplified explanation.

     

    Title and abstract
    Does the title clearly reflect the content of the article? 
    [] Yes, [X] No (please explain), [ ] I don't know
    Good, but could be more results driven to emphasise the main findings of the paper. Currently too broad to reflect the goal of the article. 

    Does the abstract present the main findings of the study? 
    [X] Yes, [ ] No (please explain), [ ] I don’t know

    Introduction
    Are the research questions/hypotheses/predictions clearly presented? 
    [ ] Yes, [X] No (please explain), [ ] I don’t know
    The hypotheses could be more clearly articulated and supported with appropriate background. They often appear in the methods/results without a clear justification or background in the introduction.

    Does the introduction build on relevant research in the field? 
    [X] Yes, [ ] No (please explain), [ ] I don’t know

    Materials and methods
    Are the methods and analyses sufficiently detailed to allow replication by other researchers? 
    [X] Yes, [ ] No (please explain), [ ] I don’t know

    Are the methods and statistical analyses appropriate and well described? 
    [ ] Yes, [X] No (please explain), [ ] I don’t know
    See major comments. Analyses could be better justified and described in some more detail. There is also an issue with estimating a 26x26 matrix with only 30 isofemale lines. 

    Results
    In the case of negative results, is there a statistical power analysis (or an adequate Bayesian analysis or equivalence testing)? 
    [ ] Yes, [X] No (please explain), [ ] I don’t know
    See major comments: The spherical G-matrices are a kind of negative result that could be due to low sample size (for the number of traits) - this should be verified and supported by additional analyses.

    Are the results described and interpreted correctly? 
    [ ] Yes, [X] No (please explain), [ ] I don’t know
    Many of the analyses are described and interpreted well, but the connection to their biological interpretation is often missing. From the results it is difficult to understand how wing shape has changed, and what the patterns of genetic variation are (except that they are spherical, a result that should be verified).

    Discussion
    Have the authors appropriately emphasized the strengths and limitations of their study/theory/methods/argument? 
    [X] Yes, [ ] No (please explain), [ ] I don’t know

    Are the conclusions adequately supported by the results (without overstating the implications of the findings)? 
    [X] Yes, [ ] No (please explain), [ ] I don’t know
     

     

  4. Peer Community in Evolutionary Biology

    The manuscript reads much better now, and you did a great job addressing my concerns and those of the reviewers. I’m not sending this version back to the reviewers and I'm happy to recommend the preprint. Before I do, however, there are a few small issues that could be taken care of if you would like to go straight for publication.

    The first one is that some studies are still cited in the text but missing from the reference list. This is a problem, particularly because the ones I noticed are important for understanding the Qst-Fst analysis (Martin et al. 2008, Chapuis et al. 2008, Rogers and Harpending 1983). There are other instances, e.g. Sztepanacz & Blows 2015.

    Regarding the Qst-Fst comparison, I still find it difficult to follow, in particular how the rho_ST relates to Qst. Can you give a more extensive explanation?

    The two angles that you compute have the same letter name. Probably better to distinguish them with a superscript perhaps? When reporting these angles, it is not entirely clear what null is expected (45° or 90°?).

    Throughout the methods section, I believe you should refer to the figures or tables that report a particular approach. For example, in line 211, you can refer to the fact that the analysis is presented in Figure 2. The methods are quite technical and numerous, and most readers probably forgot about them by the time they get to the results. I would also detail these methods more in the figure legends, even if it is redundant with the methods. For example, explain how the null expectations were obtained in Figure 3 or Figure 5.

    Figure 3 is nice, but it is not entirely obvious that the G matrices are spherical. I understand that the conclusion comes from the ratio of the first with the second eigenvalue, but still they are more eccentric than expected. Further, you don't mention that after the 3rd dimension nothing can be concluded because the order of the PCs is not the same in the observed G matrices and the randomized ones (or did you rotate the random matrices?). A reader will have the impression that there is actually less variance in them. I think you should mention these issues in the main text.

    The new Figure 5 is also nice, but to help the reader, wouldn't it make sense to have the colored lines in the derived populations shorter than the dashed lines in the ancestor, if the hypothesis is that selection will decrease genetic variances?

    In Figure 6, maybe color some of the vectors to illustrate that at the tip of the wings in the populations they have different sizes and directions than expected from Lande's equation?

    In equation 2, what are the fixed effects (beta vector)?

    Typo in line 143: adaption should be adaptation.

    I wonder if part of the justification for the wings to be under selection couldn't be because of male courtship performance. Is there any work on this? I think it is worth mentioning because you measure male wings and there are probably differences in wing shape between sexes. How sexual dimorphism limits evolvability has been much discussed, and it would be worthwhile to refer to Sztepanacz and Houle 2019 (10.1111/evo.13788).

    Despite the revisions, I’m still confused about how the animal model is being used to estimate the G matrix. In your reply you mention that lines were measured before and after brother sister mating for 5 generations. Or did I misunderstand this? In the text you mention measurements just after the 5 generations, so how do you have differences in inbreeding? I think you need to detail more how the pedigree was obtained.

    I think it is important to mention in the discussion how the use of isofemale lines represents the standing genetic variation found in nature. In Drosophila there is certainly inbreeding depression that can bias G matrix estimates, due to non-additivity, which in turn questions the role of drift/selection or admixture in decreasing or increasing genetic variances.

    Opedal et al. 2023 (10.1073/pnas.2203228120) seems to be an important study to cite, as it is a meta analysis of plant data that suggests that standing genetic variance predicts phenotypic differences between populations. And to "feather my own nest", and that of PCI, it also seems relevant to cite Mallard et al. 2023 (10.24072/pci.evolbiol.100627) where in nematode C. elegans evolution experiments it was shown that differentiation of the G matrix by genetic drift occurs despite phenotypic stasis due to stabilizing selection. You do refer to Sikkink et al. 2015 experiments, with the nematode C. remanei, but here the G matrix as a “constraint” is inferred from correlated responses to selection.

     

     

     

     

     

  5. Peer Community in Evolutionary Biology

    Adaptation to novel environments, the maintenance of standing genetic variation within populations, and the eventual divergence of populations and species, requires the coordinated evolution of multiple traits whose genetic and selective relationships vary and are typically unknown (Arnold 2023). Since the seminal work of R. Lande, the degree and direction of phenotypic evolution has been understood to depend on the genetic variances and covariances among traits, summarized in the G matrix, together with the directional selection gradients acting on each of them (Lande 1979; Lande and Arnold 1983). However, Lande's formulation applies only over short timespans because the G matrix itself evolves, even when populations face constant and/or homogeneous environments (Lande 1980; Lynch and Hill 1986).

    Despite many theoretical and empirical studies during the next two decades, only at the turn of the century did Steppan and colleagues clearly ask the key questions of comparative quantitative genetics (Steppan et al. 2002): why and how quickly does the G matrix evolve, and is the evolution of G decoupled from short- and long-term phenotypic divergence among populations and species? Since then, comparative and experimental studies have shown that the G matrix can evolve more slowly than trait means, but also, puzzlingly, that phenotypic stasis occurs alongside extensive G matrix instability (Arnold 2023). This has shifted attention toward identifying which aspects of genetic covariances are relatively stable, potentially due to the input of pleiotropic mutations, and how these covariances can bias evolutionary trajectories, particularly when traits experience correlated selection (Svensson et al. 2021).

    The Drosophila wing has long been a favorite model for studying phenotypic evolution. In the 1990s, K. Weber showed that arbitrary dimensions of wing shape, independently of wing size, contain genetic variance and respond to artificial selection (Weber 1990; Weber 1992). Ten years later, P. Phillips and colleagues demonstrated that a single generation of brother-sister mating unpredictably alters genetic covariances among such arbitrarily defined wing shape traits, and that these changes are sufficient to impact phenotypic divergence 20 generations later (Phillips et al. 2001; Whitlock et al. 2002). More recently, it has been shown that patterns of mutational covariances influence the rate and direction of wing shape divergence across 40 million years of Drosophila evolution (Houle et al. 2017), and that the distribution of standing genetic variation across wing shape traits may reflect the combined effects of mutational bias introducing variation unevenly across phenotypic space and stabilizing selection filtering out that variation (Dugand et al. 2021).

    The recent global invasion of the spotted-wing Drosophila suzukii, a fruit pest, from its native East Asian range into North America and Europe over the past two decades provides a unique natural "replicated experiment" to study the evolution of the G matrix and phenotypic divergence during range expansion. In the preprint "Evolution of genetic (co)variances during the worldwide invasion of Drosophila suzukii", A. Fraimout and colleagues compared the G matrix of wing shape between ancestral native and derived invasive populations sampled from two localities each at a northern and southern latitude in the USA and France (Fraimout et al. 2026). Leveraging the species' demographic history, their main result is that there was moderate but significant genetic differentiation in wing shape among populations, although the overall structure of the G matrix remained relatively stable. The study is particularly original in applying a multivariate extension of the classic Qst-Fst comparison, following (Martin et al. 2008), to test whether wing shape divergence is explained by genetic drift alone. Under neutrality, the expected relationship between the among-population divergence D and the G matrix is D=2Fst/(1-Fst)G, where population differentiation (Fst) is estimated from neutral DNA markers (Leinonen et al. 2013). Although by not much, the observed divergence exceeded this neutral expectation, indicating that ancestral standing genetic variation did not strongly constrain early wing-shape evolution and that selection likely played a role in phenotypic divergence despite founder effects and hybridization.

    Using the multivariate Qst-Fst comparison to infer selection on wing shape is nevertheless challenging because information about individual traits is lost, see discussion in (Martin et al. 2008), and, as one reviewer of the preprint astutely noted, the proportionality coefficient between standing genetic variation and divergence "does not drill down into the specific combinations of traits". A recent preprint specifically measured the relationship between wing shape and fitness in D. melanogaster, finding that stabilizing selection on wing shape, presumably for flight performance, constrained phenotypic divergence and was mostly driven by males (Brändén and De Lisle 2026). These results join those of J. Sztepanacz and D. Houle indicating that cross-sex covariances may considerably bias wing-shape evolution, constraining responses to antagonistic selection between sexes while enhancing responses whenever selection is concordant between them (Sztepanacz and Houle 2019).

    Fraimout et al.'s study suggests that selection and drift influence wing-shape evolution during the D. suzukii invasions, and it would be valuable to further investigate sex-specific selection on wing shape, particularly because males have spotted wings that may play a role in sexual signaling or courtship. More broadly, this system offers the opportunity to study the interplay between genetic architecture and adaptive evolution in real time across independently founded invasive pest populations (Camus et al. 2025).

    References

    Arnold, S. J. 2023. Evolutionary Quantitative Genetics. Oxford University Press, Oxford.

    Brändén, A. and S. P. De Lisle. 2026. Apparent stabilizing selection explains micro and macroevolution of Drosophila wings. bioRxiv. https://doi.org/10.64898/2026.01.08.698402

    Camus, L., N. O. Rode, S. Serga, A. Loiseau, X. Chen, C. Iampietro, M. Kenis, W. Marande, J. Mensch, H. Parinello, M. Savić Veselinović, S. Valière, J. Zhang, A. Estoup, S. Boitard, and M. Gautier. 2025. Adaptive Challenges of Past and Future Invasion of Drosophila suzukii: Insights From Novel Genomic Resources and Statistical Methods Combining Individual and Pool Sequencing Data. Molecular ecology 34:e70192. https://doi.org/10.1111/mec.70192

    Dugand, R. J., J. D. Aguirre, E. Hine, M. W. Blows, and K. McGuigan. 2021. The contribution of mutation and selection to multivariate quantitative genetic variance in an outbred population of Drosophila serrata. Proceedings of the National Academy of Sciences 118:e2026217118. https://doi.org/10.1073/pnas.2026217118

    Fraimout, A., S. Chantepie, N. Navarro, C. Téplitsky, V. Debat (2026) Wing shape evolution is not constrained by ancestral genetic covariances in the invasive Drosophila suzukii. bioRxiv, ver.3 peer-reviewed and recommended by PCI Evolutionary Biology. https://doi.org/10.1101/2024.01.02.573869

    Houle, D., G. H. Bolstad, K. van der Linde, and T. F. Hansen. 2017. Mutation predicts 40 million years of fly wing evolution. Nature 548:447-450. https://doi.org/10.1038/nature23473

    Lande, R. 1979. Quantitative genetic analysis of multivariate evolution, applied to brain: body size allometry. Evolution 33:402–416. https://doi.org/10.1111/j.1558-5646.1979.tb04694.x

    Lande, R. 1980. The genetic covariance between characters maintained by pleiotropic mutations. Genetics 94:203-215. https://doi.org/10.1093/genetics/94.1.203

    Lande, R. and S. J. Arnold. 1983. On the measurement of selection on correlated characters. Evolution 37:1210-1226. https://doi.org/10.1111/j.1558-5646.1983.tb00236.x

    Leinonen, T., R. J. McCairns, R. B. O'Hara, and J. Merila. 2013. Q(ST)-F(ST) comparisons: evolutionary and ecological insights from genomic heterogeneity. Nature reviews. Genetics 14:179-190. https://doi.org/10.1038/nrg3395

    Lynch, M. and W. G. Hill. 1986. Phenotypic Evolution by Neutral Mutation. Evolution 40:915-935. https://doi.org/10.1111/j.1558-5646.1986.tb00561.x

    Martin, G., E. Chapuis, and J. Goudet. 2008. Multivariate QST-FST comparisons: a neutrality test for the evolution of the g matrix in structured populations. Genetics 180:2135-2149. https://doi.org/10.1534/genetics.107.080820

    Phillips, P., M. C. Whitlock, and G. R. Fowler. 2001. Inbreeding changes the shape of the genetic covariance matrix in Drosophila melanogaster. Genetics 158:1137-1145. https://doi.org/10.1093/genetics/158.3.1137

    Steppan, S. J., P. C. Phillips, and D. Houle. 2002. Comparative quantitative genetics: evolution of the G matrix. TREE 17:320–327. https://doi.org/10.1016/S0169-5347(02)02505-3

    Svensson, E. I., S. J. Arnold, R. Burger, K. Csillery, J. Draghi, J. M. Henshaw, A. G. Jones, S. De Lisle, D. A. Marques, K. McGuigan, M. N. Simon, and A. Runemark. 2021. Correlational selection in the age of genomics. Nature ecology & evolution 5:562-573. https://doi.org/10.1038/s41559-021-01413-3

    Sztepanacz, J. L. and D. Houle. 2019. Cross-sex genetic covariances limit the evolvability of wing-shape within and among species of Drosophila. Evolution 73:1617-1633. https://doi.org/10.1111/evo.13788

    Weber, K. 1990. Selection on wing allometry in Drosophila melanogaster. Genetics 126:975-989. https://doi.org/10.1093/genetics/126.4.975

    Weber, K. 1992. How small are the smallest selected domains of form? Genetics 130:345-353. https://doi.org/10.1093/genetics/130.2.345

    Whitlock, M. C., P. C. Phillips, and K. Fowler. 2002. Persistence of changes in the genetic covariance matrix after a bottleneck. Evolution 56:1968-1975. https://doi.org/10.1111/j.0014-3820.2002.tb00122.x

  6. Version published to 10.1101/2024.01.02.573869 on bioRxiv