REPLYING TO J. van Gestel & F. J. Weissing Nature 555, (2018); F. Mallard, A. M. Jakšic´ & C. Schlötterer Nature 555, (2018)

The concerns raised in the accompanying Comments by Mallard et al.1 and van Gestel and Weissing2 that natural selection could not generate the transcriptomic results reported in our Letter3 overlook the larger context of previous work documenting rapid parallel evolutionary changes in guppies4,5,6,7. Here we also show compelling evidence that their alternative interpretations simply do not match our published datasets.

Mallard et al.1 argue that the negative correlation between plasticity and evolution reflects neutral processes rather than selection, but they only report simulations using a subset of criteria our conclusions were based on; when all criteria are applied, their simulations support our conclusions. We based our conclusions on four criteria with analyses that appropriately accounted for stochastic variation and non-independence of our data: (1) we found more concordantly differentially expressed (CDE) transcripts than in permuted datasets; (2) we did not find more genes that diverged in opposite directions than in permuted datasets; (3) the association between plasticity and divergence was more negative than in permuted datasets; and (4) the direction of association between plasticity and divergence was more extreme than in permuted datasets. Had our data not met both of the first two criteria, we would not have concluded that our CDE genes were enriched for genes evolving under selection and would not have proceeded to assess the relationship between ancestral plasticity and divergence. The results presented in figures 1 and 2 of Mallard et al.1 do not meet criterion (1) and thus provide no evidence that weakens our conclusions. The authors acknowledge this potential flaw and present simulations in their supplementary methods that meet criterion (1), yet they omit criterion (2) in those analyses. In fact, only 26 of their 400 parameter sets replicate all four criteria, and only a fraction of these yield datasets with distributions that might reasonably match our results (Fig. 1 and ref. 8). Therefore, the simulations of Mallard et al.1 are not only consistent with our interpretation, but also other studies concluding that rapid evolution of genes exhibiting non-adaptive plasticity is more likely due to selection and very unlikely to arise by chance3,9.

Figure 1: A small fraction of simulations performed by Mallard et al.1 meet all four criteria that support our conclusion that CDE genes diverged under selection, and that they diverged in the opposite direction as ancestral plasticity.
figure 1

Blue boxes indicate parameter sets that met the criteria, and red boxes indicate parameter sets that did not meet the criteria. a, Only a subset of parameters meet criterion (2), positing that differentially expressed genes that diverge in opposite directions should not be overrepresented compared to the permuted datasets. b, All four criteria merged together shows that only 26 parameter sets recreate the results found in Ghalambor et al.3 This contrasts with the much higher fraction claimed by Mallard et al.1, as they did not consider criterion (2) in their analyses.

PowerPoint slide

We agree with van Gestel and Weissing2 that individual transcripts are not independent, and recognize the robustness of gene expression networks to produce similar phenotypes via different mechanisms10. We explicitly incorporated non-independence among transcripts in designing our permutations to preserve entire transcriptional profiles. The model proposed by van Gestel and Weissing2 positing a single regulatory change underlying patterns in our set of CDE transcripts simply does not match our data, as such a model would generate a strongly correlated set of CDE transcripts. Correlational analyses show that few of our CDE genes are strongly correlated, and that numerous clusters of transcripts independently evolved CDE expression patterns and negative associations between plasticity and divergence (see ref. 8) to compare correlations in experimental data and simulated datasets with a single regulatory change). We thus refute the claim that variation in a single modulator can account for the rapid evolution of our CDE transcripts, just as it is unlikely to explain the rapid parallel evolution of other complex phenotypes in guppies.