Drivers of adaptive evolution at the per-transcript level
Drought, freeze and geography associated variables jointly explained 50% of the expression variance for the Cold-condA category, 17% for the Warm-condA category and 19% for the Ad-Pl category (Table 1). The full model with all three predictors was significant only for the Cold-condA category and this was driven by geography and drought on the first RDA axis (p geo = 0.004,p drought = 0.011).
Estimates of genomic ancestry for the 30 maternal trees ranged from 0.18 to 1.00 (Fig. S2b), with 100% genomic ancestry from P. flexilisset at a value of 1.0. Survival estimates ranged from 0.37 to 1.00 at the Cold garden and from 0.23 to 0.90 at the Warm garden. Across both gardens, estimated survival of maternal trees was positively correlated with their genomic ancestries (Pearson’s r ; Warm garden: r= 0.24, p = 0.18; Cold garden: r = 0.40, p = 0.02). On average (± 1 sd), population-level expression values were weakly correlated with population-level estimates of ancestry (-0.02 ± 0.31 in Cold garden and 0.02 ± 0.31 in Warm garden). Correlation coefficients for individual transcripts, however, ranged from -0.6 to 0.8 for Cold-condA, from -0.8 to 0.8 for Warm-condA and from -0.6 to 0.4 for Ad-Pl. The cumulative distribution of correlation coefficients between ancestry and transcript abundance for all categories was not significantly different from their respective matched background set of transcripts (Cold-condA: D = 0.05, p = 0.702; Warm-condA:D = 0.03, p = 0.84; Ad-Pl: D = 0.13, p = 0.74). Several transcripts in the Warm-condA and Cold-condA categories displayed significant associations with ancestry (p < 0.05), but none were significant after multiple test corrections. No transcript in the Ad-Pl category was significantly associated with ancestry.
Expression values on average (± 1 sd) were weakly correlated with survival estimates (Cold garden: r = -0.008 ± 0.32; Warm garden:r = 0.029 ± 0.33). Correlation coefficients for individual transcripts, however, ranged from r = -0.68 to 0.76 for Cold-condA, from r = -0.89 to 0.73 for Warm-condA and fromr = -0.42 to 0.55 for Ad-Pl category. The cumulative distributions of correlation coefficients for all categories were not significantly different from their respective backgrounds (Cold-condA:D = 0.073, p = 0.817; Warm-condA: D = 0.065,p = 0.11; Ad-Pl: D = 0.17, p = 0.39). Similar to the case of ancestry, we identified several transcripts for the Warm-condA and Cold-condA categories as significantly associated with survival (p < 0.05), although none passed the multiple testing correction. No transcript in the Ad-Pl category was significantly associated with survival.
While no GO terms were enriched in either Q STcategories, significant depletion was noted for GO terms related to signal transduction and cell communication in Cold-condA, and for hydrolase activity in Warm-condA (Table S2). Several transcripts related to freeze and drought tolerance, as well as cell wall modulation were classified under the Cold-condA category. These included VIN3 ,PPR , XTH and Aquaporins such as TIP1 (Sung & Amasino, 2005; Raimund, 2015; Tucker et al ., 2018). Similarly, the Warm-condA category contained transcripts related to drought stress response and photosynthesis, including numerous ERF family genes,SWEET , WRKY, psbE and multiple auxin responsive elements, such as ARF2 (Lata et al ., 2015; Zhang et al ., 2020). Transcripts classified in the Ad-Pl category, included auxin responsive elements, such as SAUR50 (Sun et al ., 2016), which is critical for light signaling, α-expansin and ABA signaling pathway family genes, such as LPPD , which have known roles in numerous plant developmental and stress response pathways (Marowa et al ., 2016). Thus, even without clear relationships between population-level transcript abundances, survival, and genomic ancestry by QST category, there were several notable examples of functionally sensible genes in eachQST category consistent with adaptive responses to drought and freeze gradients. Together with the analyses presented in the previous section these results provide support for hypothesis H1 and H2 at the per-transcript level but render minimal support for H3.