Assessing per-transcript level signatures of adaptive evolution
We evaluated the among population component of additive genetic variance (QST ) for each transcript given our quantitative genetic study design (Spitze, 1993). Under neutrality,QST should be similar to its genome-wide equivalent (FST ) (Leinonen et al ., 2008). Under spatially divergent selection pressures driving local adaptation, however, Q ST should statistically exceedF ST. Using the per transcript variance components obtained from equation (1), we estimated QST as:
\(Q_{\text{ST}}=\ \frac{\sigma_{a}^{2}}{6\sigma_{w}^{2}\ +\ \sigma_{a}^{2}}\), (2)
where \(\sigma_{a}^{2}\) represents the among population variance component and \(\sigma_{w}^{2}\) represents the within population variance component. The constant 6 originates from our assumption that the mixed-sib design is a 50:50 mixture of half and full siblings (Gilbert & Whitlock, 2015). Confidence intervals (CI, 95%) per transcript per garden were obtained using 1000 parametric bootstrap estimates generated using the bootMer function implemented in the lme4 v.1.1-28 package in R (Bates et al ., 2015). To address our hypotheses concerning garden specific patterns of adaptive differentiation at the per-transcript level (H1 & H2), we compared the 0.025 quantile of QST for each transcript with the 0.95 quantile of FST and identified transcripts exhibiting signatures of local adaptation. Since we are primarily interested in the architecture underlying GEI, subsequent analyses and our main results focus on three QSTcategories: (a) conditionally adaptive in cold garden (Cold-condA), transcripts with QST >FST only in high elevation garden; (b) conditionally adaptive in the warm garden (Warm-condA), transcripts withQST > FST only in low elevation garden; and (c) adaptive plasticity (Ad-Pl), transcripts with QST >FST across both gardens and a significantQST reaction norm (p < 0.05) assessed using a t -test (Fig. 2). For eachQST category, we performed gene ontology (GO) enrichment analyses using a hypergeometric test with 1000 permutations to compute the family-wise error rate as implemented in the GOfuncR v1.12 package (Grote, 2020) in R. The background for GO enrichment analyses was the full set of GO terms across all annotated and non-contaminant transcripts from EnTAP (Table S1; Methods S1).