QST - FST: Comparison of quantitative trait differentiation to neutral genetic expectation
We found evidence of selection (QST ≠ FST) driving phenotype differences in over half of the traits measured across the three gardens (Table 2, Fig. 4). Overall, the mean QST value across all traits and all gardens (0.42) was above the FST confidence interval (0.144-0.205), consistent with directional selection shaping trait differences and increasing local adaptation among these populations. Bud flush traits exhibited the highest levels of population differentiation. For example, QST for bud flush in the hot and mid gardens was 0.87 and 0.86, respectively, while QST values for bud set were more moderate (0.26-0.54). The confidence intervals for QST crossed those of FST in two phenology measurements (bud set in Agua Fria and bud flush in Canyonlands), suggesting no difference from the neutral expectation of genetic drift of these traits in these environments. Divergent selection on specific leaf area was apparent in the mid (QST = 0.54) and cold gardens (0.79), but not detectable in the hot garden of Yuma (0.31). Tree growth traits showed relatively lower QST values compared to leaf traits across gardens. Tree height showed significant divergent selection when measured at the hottest common garden in Yuma, Arizona (QST = 0.44), but was not statistically different from FST in the two cooler gardens of Agua Fria and Canyonlands (0.14 and 0.26, respectively). Contrary to our hypothesis, we found evidence of stabilizing selection for basal trunk diameter in the coldest garden, where the QST value (0.01) fell below the FST confidence interval. However, this result should be interpreted with caution since dominance reduces estimates of QST and is therefore a poor indicator of stabilizing selection (Cubry et al. 2017). This trait was indistinguishable from the neutral expectation of FST in the warm and mid gardens (QST = 0.50 and 0.03, respectively).
Mean QST values for all trait plasticities except DRC were above the FST confidence interval (Table 2), suggesting overall divergent selection acting on plasticity. We found the strongest evidence for divergent selection on plasticity for bud flush (mean QST = 0.84) and height (mean QST = 0.66) where the QST 95% confidence interval distribution never crossed the FSTconfidence interval (Fig. 5). In the analyses of bud set (mean QST = 0.44) and SLA (mean QST = 0.69), the lower QST confidence interval distribution overlapped with FST, indicating that for some of the 100 possible plasticity datasets, trait plasticity differences among populations were not distinguishable from the neutral expectation of drift. Specifically, 67% of the permutations fell within the FST interval for bud set, while only 6% overlapped for SLA. Similar to DRC, the QST values for DRC plasticity were much lower than the other plasticities (mean QST = 0.07), however the upper confidence interval did overlap with FST 76% of the time (Fig. 5). These three cases of plasticity QST confidence interval distributions overlapping with FST represent a weaker detection of selection. However, these results do provide partial evidence for SLA, where 94/100 permutations were non-overlapping.