Divergent selection shapes population trait and plasticity differences
We interpret cases when both the QST-FSTanalysis showed large divergences from neutral expectation and phenotype-climate correlations were significant as strong evidence for climate-driven selection. Cases with only one of these tests showing population differences provide partial evidence for climate-driven selection (Table 4). For instance, there were four cases showing QST > FST but non-significant trait-climate correlations. These inconsistencies between the two tests could be due to divergent selection that is not related to the climatic gradients we tested. There were also four cases showing QST ≈ FST and significant trait-climate correlations. In these cases, QSTconfidence intervals overlapping with FST could be due to the bootstrap sampling of genotypes with less genetic variation compared to the full sampling design, thus lowering the QST estimate. There were also more significant plasticity-climate correlations (4 out of 5) than significant QST-FST differences for plasticity (2 out of 5). Finally, there was a range of results across the three gardens within the QST-FST analysis itself. Together, these tests provide a continuum of support for selection on traits and trait plasticity, and highlight which traits may be under the strongest selection and potentially the most important to investigate under climate change.
We found the largest QST values for spring bud flush, consistent with other studies showing high phenological divergence across latitudinal clines (Hurme 1999; Howe et al. 2003; Hallet al. 2007; Evans et al. 2016). Spring bud flush is highly differentiated among P. fremontii populations, with a difference of up to eight weeks observed in flush timing (Grady et al. 2015; Cooper et al. 2019; Blasini et al. 2020). We also found large population differences in fall bud set timing of ~2-5 weeks across the common garden gradient, reflected in moderate QST values in two out of the three gardens. The strong population differences in phenology found here agree with Fischer et al. (2017), who showed leaf phenology accounted for >80% of the variation in tree and forest productivity among Fremont cottonwood genotypes. We found larger population differences in bud flush compared to bud set. This result is intriguing given that spring bud flush is primarily governed by temperature, while fall bud set is mostly cued by precise day length periods (Thomashow 2001; Howe et al. 2003). While day length is driven by latitude and is constant from year to year, temperature can vary. The fixed environmental cue of day length should allow populations to become highly locally adapted and differentiated in bud set compared to a variable environmental cue such as temperature. However, the strength of photoperiod-driven selection on bud set may be relaxed in all but the highest elevation populations, where the trade-off for longer growing seasons is selected for in areas that very rarely or never experience killing frosts (Howe et al. 2003). Both phenology traits also showed strong relationships with provenance climate across the gardens, except for bud flush in the coldest garden, where low temperatures prevented an earlier flush in the southern, warm-adapted populations (Fig. 3a).
Our detection of selection was dependent, in part, on the environmental conditions of each garden. Bud flush, bud set, and SLA all exhibited divergent selection (QST > FST) in two out of three gardens (Fig. 4). Tree growth traits exhibited even larger variation in QST among gardens. For example, we observed high population differentiation in height expressed in the hottest garden (QST = 0.44). When populations were planted in the moderate and cool gardens, these population differences diminished, but became more strongly predictable from home climate (Fig 3). QST estimates for trunk diameter also decreased with decreasing garden temperature. This variability in QST across gardens suggests that phenotypes shaped by selection pressures across a species’ range can be expressed differently in different growing environments, with some environments enhancing and others dampening population phenotypic differences (Oke et al. 2015; Akman et al. 2021). Particularly for growth traits, this may represent an interaction between the selection pressures that have shaped existing variation across the species range and novel selection pressures imposed in a common garden experiment or under future climate change.
The larger population-level trait differences exhibited in the hottest common garden for most traits (except SLA) could be driven by the maladaptation of the cold-adapted, northern populations to the extreme thermal conditions experienced in this hot garden. This climate transfer from northern to southern Arizona represents an extreme warming treatment, a scenario that may be imposed on populations under severe heat waves with climate change (Cook et al . 2015). Similarly, Evans et al. (2016) found that the relationship between QST and FST changed through time, with tree height displaying high population differentiation (QST > FST) under the growing conditions in one year but not the next. Long-term common garden experiments can demonstrate how population differences are expressed both across different environments and through time. Given the intensification of extreme events and climate variability going forward (Jentsch et al. 2007; Ganguly et al. 2009; Garfin et al. 2013; Williams et al. 2020), these types of field trials should be expanded to evaluate the correspondence between the degree of existing climate adaptation and the potential for future climate survival, either through phenotypic plasticity, selection on remaining genetic variation, or a combination of the two (Nicotra et al.2010; Josephs 2018).
Our QST-FST comparison revealed support for divergent selection acting on phenotypic plasticity in bud flush and tree height, and showed partial evidence for selection on plasticity in the other three traits (Table 2; Fig. 5). This is in contrast to previous studies that found no evidence of selection on trait plasticity using QST-FST type comparisons (Lindet al. 2011, De Kort et al. 2016), and low overall support for selection on plastic responses to temperature (Arnold et al.2019). Our results suggest that for some traits, differences in plasticity among populations across a wide environmental gradient are larger than expected from neutral genetics, where some populations show minimal plasticity and others exhibit high plasticity. Conversely we found some evidence for stabilizing selection in DRC plasticity, indicating that the difference in the magnitude of plasticity for this trait across our populations was smaller than expected by FST, however it was not below the FSTconfidence interval in all 100 plasticity permutations. The mosaic of natural selection acting on trait plasticity across our populations shows how plasticity itself can evolve in response to different climates.
The mosaic of natural selection acting on trait plasticity across our populations shows how plasticity itself can evolve in response to different climates. We found significant plasticity-climate relationships in phenology and growth traits, where the sign of the correlation switched between these two types of traits (Fig. 3). Specifically, we found trees sourced from colder environments were significantly more plastic in height and DRC compared to the warm provenance populations, but were not as plastic with regard to their bud set and bud flush (Fig. 3). This is an example of a multivariate plasticity response, where plasticity in one trait may be affecting the plasticity in another trait (Nielsen & Papaj 2022). The higher plasticity in phenology traits measured in populations from hotter provenances is counterintuitive because colder source populations experience much more predictable fall freezing events and higher yearly temperature variation (see TD in Supplemental Table 1), and theory predicts plasticity will increase under predictably variable environments (Chevin & Lande 2010). However, our climate transfer of southern Arizona populations to the northernmost cold garden represents an extreme climate event (over 15°C colder in the coldest month for the populations from the hottest source locations, Supplemental Table 1) far outside of some populations’ normal temperature range, which can result in large, maladaptive plastic responses (Chevin & Hoffman 2017). The higher phenological plasticity seen in hot-adapted populations did not translate into increased growth or growth plasticity, likely due to maladaptive phenological plasticity that pushed these trees outside of the appropriate growing season window (Cooper et al. 2019). The high bud set plasticity of warm-adapted populations meant that these trees did not set bud until late in the growing season, when freezing temperatures damaged non-dormant tissues. The subsequent frost damage translated to lower growth compared to cold-adapted trees that set buds earlier in the season and avoided frost damage. The increased height of the cold adapted populations in the coldest garden relative to the warm populations produced the significant differences in height plasticity. Therefore, our result of higher height plasticity in populations sourced from cold locations can be partially explained by the warm populations’ maladaptive plasticity in phenology.
In comparing our results to previous findings of no divergent selection on plasticity in other systems, it is important to consider both climate means and variances. In this system, higher growth plasticity observed in populations sourced from colder, high elevation locations could also be due to adaptation to increased climate variability, compared to the central and southern Arizona populations. Specifically, the temperature difference between the mean warmest month and the mean coldest month was the largest for the three populations collected on the Colorado Plateau compared to the rest of the populations below the Mogollon Rim of the Plateau (see TD in Supplemental Table 1). This follows the theory that higher levels of plasticity should occur in more variable environments (Lande 2009).
Finally, our estimates of heritability in both traits and trait plasticity also indicate that these components of the phenotype can evolve in response to selection, at least under some environmental conditions. Broad-sense heritability values for the five traits were moderate, with a mean value across all gardens of 0.21 (Table 2). Our phenology heritability measures (H2 = 0.04-0.48 for bud flush and H2 = 0.19-0.30 for bud set) were lower than previously found in some Populus studies (H2 = 0.94 for bud flush and H2 = 0.91 for bud set in P. trichocarpa x deltoides , Frewen et al. 2000). Heritability values for growth traits (H2= 0.11-0.27 for height and 0.08-0.21 for DRC) and SLA (H2 = 0.10-0.35) were fairly consistent with other reported Populus estimates (H2 = 0.03-0.42 for height and H2 = 0.09-0.25 for diameter at breast height in P. tremuloides , Ding et al. 2020; H2 ≈ 0.2-0.6 for SLA in P. nigra , Guet et al. 2015). The range of heritability estimates for the same trait across the three gardens highlights the environment-dependent nature of heritability. This is especially apparent in our bud flush results, where we found the lowest value in the cold garden (H2= 0.04) and the highest value in the warmest garden (H2 = 0.48). There was also no trend toward higher or lower heritability estimates in a particular common garden. These results suggest that in some environments evolutionary potential is limited but can increase as environmental conditions and associated selection pressures change. Furthermore, these heritability increases are not necessarily associated with a specific direction of change (i.e., increasing or decreasing temperature). Broad-sense heritability for the five trait plasticities ranged from 0.09-0.18, a similar result to bud burst plasticity found in another riparian deciduous tree, black alder (H2 = 0-0.129, De Kortet al. 2016). Our results of genetic variation in trait plasticity combined with the evidence for selection based on QST-FST analysis and non-zero heritability estimates show selection on the heritable components of phenotypic plasticity may lead to evolving plasticity across the landscape among these Arizona populations of Fremont cottonwood.