Dataset: Results from a meta-analysis investigating covariance between genetic and environmental (CovGE) effects in phenotypic results in published literature

We searched the Web of Science database for experimental studies that evaluated differences in phenotypic responses across different genotypes and environments. We conducted the initial search on June 24, 2019. We used the search terms, ("cogradient variation" OR "countergradient variation" OR "cogradient selection" OR "countergradient selection" OR "co-gradient variation" OR "counter-gradient variation" OR "co-gradient selection" OR "counter-gradient selection") OR ("GxE" OR "genotype by environment" or "gene by environment") OR ("nonadaptive plast*" OR "non-adaptive plast*" OR "maladaptive plast*" OR "adaptive plast*") OR ("phenotypic plast*" AND "adapt*") AND ("common garden" OR "reciprocal transplant"). Initial searches returned approximately 5,900 hits. Results were further refined by including only those articles within Web of Science categories that related to ecology, evolution, or any ecological or evolutionary subdiscipline (e.g., papers categorized as engineering or biomedical were excluded). Refining reduced the search results to 4,458 studies. We also added studies that were included in previously published meta-analyses by Murren et al. (2015) and Hereford (2009) for screening.

Results were exported, compiled, and primed using package "metagear" (Lajeunesse 2016) in the R statistical environment (Team 2018). Studies were screened for inclusion by scanning titles and abstracts. We required that studies collect phenotypic data from at least two genotypes or populations across at least two different environments. We assumed that author-specified "populations" or "genotypes" are groups of interbreeding individuals experiencing different selection pressures and therefore are likely to be genetically divergent although we acknowledge that this is not always the case (Merilä and Hendry 2014). We excluded studies that only provided genomic data with no other phenotypic anchors, studies that did not provide information about the native environments of genotypes used in experiments, and studies that used genotypes produced by artificial selection. Additionally, a prerequisite for the estimation of CovGE is that phenotypic data from each genotype is required from the same environment in which the genotype evolved (i.e. its native environment). More simply, we cannot estimate CovGE if any genotype (‘G’) is missing its environment (‘E’). Because we use linear models to generate estimated marginal mean phenotypes (see methods below), if the experimental treatments did not align to the home (native) environments of each genotype, interpolation would be required to predict the mean phenotype for each genotype and environment. In doing so, bias can be introduced. Therefore, we only included studies that match experimental treatments to each genotype’s native environment (i.e., the environment from which genotypes were collected). Furthermore, a challenge in the meta-analysis was the presence of nonlinear reaction norms in experimental designs with continuous environmental treatments that are frequently observed in common garden experimental designs. Thus, we only included studies that used categorical experimental environments.

After compiling studies, we measured CovGE and GxE magnitude on phenotypic data. More methods can be found in the manuscript published in Ecology Letters in 2022.

See Related Dataset Albecker et al. (2022) for model code.