Introduction:
Understanding the processes shaping phenotypic diversification in nature is a central objective of ecology and evolutionary biology (Schluter 2000, Bolnick et al. 2011). The effects of phenotypic variation in foundation species can be far-reaching, influencing everything from species interactions to the evolution of complex communities (Whithamet al. 2020). Trait variation within widespread species can be extensive due to historic demographic processes and spatially and temporally heterogeneous landscapes exerting different selection pressures across their range (Whitlock 2008). Over time, subpopulations can become genetically and phenotypically differentiated due to neutral processes, such as drift, gene flow, and mutation, as well as the adaptive process of natural selection (Wright 1931; Spitze 1993; Holsinger & Weir 2009; Leinonen et al. 2013). The relative importance of these stochastic versus selective forces is still debated but is crucial for understanding the probability and rate of phenotypic divergence in the past and future (O’Hara 2005; Hangartner et al.2012; Leinonen et al. 2013). Forest ecosystems provide evidence of significant genetic differences, a high degree of local adaptation, and ecological consequences for associated species and communities (Savolainen et al. 2007; O’Neill et al. 2008; Leimuet al. 2008; Hereford 2009), including species of Populus(Whitham et al. 2006; Grady et al. 2011; Grady et al. 2013; Evans et al. 2016; Fischer et al. 2017; Cooperet al. 2019). Understanding the processes underlying genetic and phenotypic divergence in these species, especially in relation to past and future adaptation to climatic variation, is essential both for selecting current stock for restoration and forecasting the potential for further adaptation in response to climate change (Grady et al. 2015; Evans et al. 2016).
One way to test whether natural selection is the mechanism responsible for generating phenotypic differences among populations is to compare QST, the variation in quantitative traits, to FST, the variation in neutral genes (Wright 1951; Lande 1992; Spitze 1993). QST is the quantitative genetic analog to FST and measures the proportion of additive genetic variance in a trait attributed to among-population differences. If QST exceeds the neutral expectation of FST, there is evidence that directional selection is responsible for population-level phenotypic differentiation. If QST ≈ FST, the null model that population differences are due to genetic drift alone cannot be rejected. Finally, if QST is lower than FST, this suggests uniform or stabilizing selection acting to constrain among-population divergence (Spitze 1993). QST-FST comparisons have been primarily used to detect selection and evaluate the degree of local adaptation among populations, but have increasingly been used as a management and conservation tool (Leinonen et al . 2013). For example, QST has been used to designate populations as separate conservation units (Leinonen et al. 2008), to assess the adaptive potential of invasive species, measure the rates of evolution in different environments, and look at the constraints on adaptation due to increased habitat fragmentation (Leinonen et al. 2013). The surge in both experimental and theoretical studies comparing molecular and quantitative genetic variation has revealed a major role of natural selection in shaping intraspecific variation in quantitative traits (McKay & Latta 2002; Leinonen et al. 2008; Leinonen et al. 2013), with approximately 70% of all studies showing QST > FST (Leinonenet al. 2008). QST studies are often used as an exploratory analysis to first see the selective patterns across a suite of traits, and then target those traits with the highest levels of differentiation to examine their genetics and responses to selection more closely (Leinonen et al. 2008; Whitlock 2008).
The pattern of phenotypic variation in tree species along climate gradients often appears consistent with local adaptation in response to selection by climatic conditions. For example, phenological traits are closely linked to temperature and photoperiod, and show strong latitudinal clines in multiple tree species (Howe et al. 2003; Savoleinen et al. 2007; Evans et al. 2016; Cooper et al. 2019). Within Populus , growth and phenology traits differ among genotypes (Frewen et al. 2000; Howe et al. 2000; Fischer et al. 2017; Davis et al . 2020), with evidence of variation and adaptive differences among populations (Grady et al. 2011; Evans et al. 2014; McKown et al. 2014; Cooperet al. 2019). In Populus fremontii specifically, there are large population differences in phenology expressed in common garden experiments at both the cold and hot edge of the species’ tolerance (Cooper et al . 2019), as well as clear correspondence between a population’s source climate and its mortality and productivity in cold vs. hot conditions (Grady et al. 2011, 2013, 2015). Population structure in P. fremontii has also been attributed to differences in spring and winter precipitation, which can affect flowering phenology, and therefore gene flow, across its range (Cushman et al . 2014; Ikeda et al. 2017). However, to definitively show that phenotypic variation among populations is due to divergent selection by their home climate, we need approaches that integrate molecular and phenotypic assessments in common garden environments.
The role of selection by past climatic conditions in shaping intraspecific variation in foundation species is especially important to quantify in the American Southwest, where the effects of climate change are pronounced (Garfin et al. 2013, Williams et al. 2020). Fremont cottonwood is especially sensitive to drought and high temperature, as is evidenced by stand-level mortality at the Bill Williams National Wildlife Refuge on the lower Colorado River (Fig. 1). Mortality in these trees is associated with the megadrought that Williams et al. (2020) identify as being the second worst drought in the past 1200 years in the American Southwest. Recent studies by Hultine et al. (2020a) and Blasini et al. (2020) suggest that these trees are at the edge of their thermal tolerance where water is essential for evaporative cooling. Thus, current climatic gradients will be exacerbated by ongoing climate change, leading to new selection pressures on functional traits that may be locally adapted to a narrower range of environmental conditions.
In this study, we use trait data from three experimental common gardens spanning the climatic range of P. fremontii to quantify phenotypic divergence (QST) and compare it to neutral genetic divergence (FST). Common gardens are necessary to ensure that among-population variance components reflect genetic differences and are not inflated by environmental effects (Leinonenet al. 2013). Reciprocal experimental gardens can indicate whether populations are locally adapted to their current environments, reveal traits that vary across environmental gradients as a result of phenotypic plasticity (Kawecki & Ebert 2004; Franks et al.2014), and quantify the intensity of selection across space (Whitlock 2008). Our use of multiple common gardens adds to the QST literature by examining how population-level trait differentiation is expressed across environmental gradients. Plastic responses to environmental stress or release from stress may mask or amplify genetically determined trait differences that have emerged as a result of divergent selection (Oke et al. 2015). Therefore, it is important to assess phenotypes in multiple growing conditions in order to demonstrate how the environment can modify the degree to which we can detect evidence of selection.
The three gardens used in this study contain cloned cuttings from 16 populations of P. fremontii collected throughout Arizona. Both the collection and garden sites span an elevational gradient of almost 2000 m, consistent with the species’ range and including a difference of 12°C mean annual temperature and > 500 mm in mean annual precipitation. The benefit of these experimental gardens is enhanced by the development of genomic data based on the identification of 1000s of single nucleotide polymorphisms (SNPs) in the Fremont cottonwood genome. These data can provide improved estimates of FST, owing to their greater coverage of the genome and potentially lower mutation rate than microsatellites, which have been routinely used to estimate FST. SNPs are an ideal type of marker for quantifying molecular divergence because mutation rates and the effects of drift on SNP variation are considered to be more similar to loci that control quantitative traits (Edelaar & Bjorklund 2011). Thus, the only difference between quantitative trait loci driving QSTand the loci used in FST estimates should be that only the latter conform to neutral molecular evolution (Leinonen et al. 2013).
In order to address whether natural selection by climatic conditions is an agent of phenotypic diversification across the range of Fremont cottonwood, we evaluated three hypotheses: 1) Genetic variation in multiple tree traits (phenology, specific leaf area, height, and trunk diameter) will be evident among populations and genotypes in each of the three common gardens, although the magnitude of the genetic component may vary across environments and among traits. 2) QSTvalues will be significantly higher than the neutral expectation of FST, suggesting divergent selection has outweighed drift in shaping trait differences. Again, this drift-selection balance may vary among traits, and our ability to detect selection on these traits may vary across common gardens. 3) Mean population phenotypes will show strong associations with their climate of origin, especially for the most differentiated traits (those with high QST). This pattern is expected when phenotypic differentiation is strongly shaped by selection due to climate.