Estimating the effect of environment on among population
expression trait differentiation
To further address H1 & H2 at the per-transcript level, we identified
climatic drivers of adaptive differentiation for eachQ ST category (Fig. 2) using redundancy analysis
(RDA) as implemented in the vegan v.2.5.6 package in R (Oksanen et
al ., 2013). Following Menon et al . (2021), we classified the
environmental variables obtained from ClimateNA into drought- and
freeze-associated variables, as these two axes are likely the primary
selective agents across the hybrid zone populations. To reduce the
dimensionality of the environmental dataset, we performed principal
component analysis (PCA) separately for the drought- and
freeze-associated variables and retained PC axes that explained at least
90% of the variance. We used the population-level estimates of gene
expression as the response matrix with the drought-associated PCs,
freeze-associated PCs, and geography as the predictor matrices.
Geography was represented by scaled and centered estimates of latitude
and longitude. For each Q ST category, we
evaluated the effect of the full model that included all three
predictors, as well as the marginal effects of terms when the full model
was significant (p < 0.05).