TAXONOMY CLASSIFICATION
Evolutionary ecology
1 | INTRODUCTION
Genetic diversity is a prerequisite for evo­lutionary change in all organisms; preservation of a species’ genetic diversity likely increases its chances of surviving over evolutionary time when facing environmental changes. Plant evolutionary biologists, foresters, and conservation geneticists have long been interested in the genetic differences among populations and the degree to which these may contribute to local adaptation (see Table 1 for the definition of population genetic terms cited in this mini review). This interest traces back to the common garden experiments of Turesson et al. (1922) and the reciprocal transplants of Clausen et al. (1941). For decades, common garden and reciprocal transplant experiments have been instrumental in advancing our understanding of how natural selection shapes geographic phenotypic variation (reviewed in Flanagan et al., 2018; Sork, 2018). As putatively neutral molecular genetic markers (i.e., allozymes and DNA-based dominant and codominant loci) became available, plant biologists were able to compare the levels of genetic diversity and the degree of divergence seen at phenotypic traits with those at single gene markers (Reed & Frankham, 2001; De Kort et al., 2013; Leinonen et al., 2013; Marin et al., 2020).
Applications of the knowledge of traditional marker-based neutral genetic variation (NGV hereafter) to the conservation and restoration of plant species have been somewhat controversial due to the assumed evolutionary neutrality of used markers and their limitations to be informative about the adaptive potential (García-Dorado & Caballero, 2021; Teixeira & Huber, 2021). Although levels of NGV might not be always predictive of adaptive genetic variation (AGV hereafter; Teixeira & Huber, 2021), it is possible that NGV under the current conditions may become AGV under changed environmental conditions. However, NGV, largely corresponding to within-population genetic variation from allozymes to nucleotide sequences as reflected in the percentage of polymorphic loci (%P ), allelic richness (AR ), or gene diversity (Hardy-Weinberg expected heterozygosity,H e), is regarded to be a poor “proxy” of levels of AGV in quantitative traits (i.e., narrow- and broad-sense heritabilities [h 2 andH 2]; Reed & Frankham, 2001; Depardieu et al., 2020).
The same applies to the relationship between measures of among-population genetic differentiation (e.g., Merilä & Crnokrak, 2001). The comparison between F ST ([Wright, 1951] or its analogs estimated from neutral genetic markers [Merimans & Hedrick, 2010]; see Holsinger & Weir [2009] for different definitions and interpretations of F ST) and Q ST (F ST analog for quantitative traits; Spitze, 1993; Depardieu et al., 2020), i.e.,Q STF ST comparisons or relationships, was formalized with the adoption ofQ ST in the 1990s. Q STcreates an explicit quantitative prediction of the expectation for quantitative traits under neutrality which, thus, solidified the inference that quantitative traits typically show greater genetic divergence among populations than expected under neutrality (Merilä & Crnokrak, 2001; De Kort et al., 2013; Leinonen et al., 2013). Assuming that the used genetic markers are neutral, this supports the view that the divergence of quantitative traits among populations is predominantly driven by natural selection. Although F ST is generally a poor predictor of Q ST, many researchers still follow or in part support the assumption that levels of NGV would be indicative of those of AGV (e.g., Oostermeijer et al., 1994; Hamrick & Godt, 1996; Ottewell et al., 2016; DeWoody et al., 2021; García-Dorado & Caballero, 2021, but see Teixeira & Huber, 2021).
Although there is already an ongoing transition from conservation genetics to conservation genomics (Allendorf et al., 2010; Sork, 2018), genomic data for many rare plants are still scarce, and hence, conservation managers and practitioners need to continuously utilize information on NGV, if any, to support their decision making. Comparative (i.e.,Q STF ST comparisons) and, particularly, theoretical studies of NGV and AGV within and among populations in a variety of organisms are very abundant in the literature (e.g., Reed & Frankham, 2001, 2003; Hendry, 2002; McKay & Latta, 2002; Leinonen et al., 2013 and references therein; Li et al., 2019). So far, however, there have been few studies that have appliedQ STF ST comparisons to conservation, even though several such applications are possible (Reed & Frankham, 2003; but see McKay et al., 2001; Petit et al., 2001; Gravuer et al., 2005; Rodríguez-Quilón et al., 2016).
As a different issue from the above, there have been increasing recommendations in lowering the gap between conservation science and practice (sometimes coined as “the conservation genetics gap”, “the research-implementation gap”, or “the science-practice gap”) (Taylor et al., 2017; Britt et al., 2018; Dubios et al., 2019; Fabian et al., 2019; Holderegger et al., 2019). It is agreed that conservation researchers should communicate with practitioners to integrate their genetic findings into conservation implementation (Ottewell et al., 2016; Chung et al., 2021). To achieve this, a generally and clearly written narrative coveringQ STF ST in seed plants might be needed to lower the threshold for plant conservation practitioners to employ population genetics information in conservation practice.
With this in mind, we first introduce the current knowledge about within-population genetic variation and among-population differentiation both in NGV and AGV in seed plants to highlight the distinction between the approaches used for each system to identify NGV and AGV. Next, we introduce the known general application ofQ STF ST comparisons to plant biology. We also provide management suggestions as to how to capture germplasms (e.g., seeds) covering most AGV and NGV based on the analyses of molecular and quantitative trait data.
2 | COMPARISON OF WITHIN-POPULATION GENETIC VARIATION: NEUTRAL MARKERS VERSUS ADAPTIVE TRAITS
As neutral genetic markers reflect demographic processes (including past demographic histories) within local populations, they are informative for the management and conservation of genetic purposes. Small populations are generally susceptible to the loss of NGV and less adaptive to novel environments due to the loss of AGV through genetic drift (Reed & Frankham, 2003). The degree of individuals’ heterozygosity (estimated as the number of loci for which each individual is heterozygous) is often correlated with fitness (Oostermeijer et al., 1994; Reed & Frankham, 2003). Even when there is a real relationship between an individual’s heterozygosity and fitness, this does not imply that there should be a relationship betweenH e and h 2 at the population level. These are determined by somewhat different processes.
In a meta-analysis of 71 (60 out of these with allozymes) published datasets, H e is only weakly correlated withh 2 or H 2: r = 0.217 (–0.88 to 0.90, SD ± 0.433), indicating that neutral marker-based measures only explain 4% of the variation in quantitative traits (Reed & Frankham, 2001). In addition, the correlation between allozymeH e and h 2 for 17 metric characters in seven populations of the annual Phlox drummondii is highly variable, ranging from r = –0.714 to r = 0.355 (recalculated from Schwaegerle et al., 1986). Likewise, the correlation between microsatellite H e andH 2 of five phenotypic traits in seven populations of the endangered herb Psilopeganum sinense ranges from r = –0.707 to 0.261 (Ye et al., 2014). However, caution is needed because, at a degree of freedom of five with seven populations, the critical value of r for α = 0.05 is very high at r = 0.75, giving a very low power; when Bonferroni correction is applied across the five phenotypic traits, this becomes even higher at r= 0.87. Similar results revealing a weak correlation between NGV and AGV are available from other wild plant species as well: the rare perennial herb Scabiosa canescens and its common congener S. columbaria (allozymes, H e vs.H 2; Waldmann & Andersson, 1998); the annualClarkia dudleyana (allozymes, H e vs.CV G, coefficient of genetic variation of quantitative traits; Podolsky, 2001), the annual Hordeum spontaneum (allozymes, H e vs.H 2, Volis et al., 2005), and the selfing annualSenecio vulgaris (amplified fragment length polymorphisms [AFLPs], H e vs. H 2; Steinger et al., 2002).
These studies suggest that NGV has a limited ability to predict AGV within populations. Reed & Frankham (2001) listed six factors that could be responsible for the low correlation between NGV and AGV. Namely, these are differential selection, non-additive genetic variation, different mutation rates (µ ), low statistical power, environmental effects on quantitative characters, and impact of regulatory variation. In addition, various forms of natural selection affecting the level of neutral polymorphism at linked sites may also contribute to the lack of a relationship between NGV and AGV. The most dramatic effect on neutral variation occurs when beneficial alleles at loci contributing to AGV spread into a population, a process known as a “selective sweep” (Nielsen, 2005; Stephan, 2019). Selective sweep leads to a dramatic reduction of local H e andAR along the chromosome segment (Kreitman, 2001).H e and AR for non-neighboring or unlinked neutral regions are likely not affected by such events (Nielsen, 2005) because linkage disequilibrium between NGV and AGV decays gradually under the influence of recombination.
It should be noted that, however, invoking selective sweep as a factor that lowers the correlation between NGV and AGV could be problematic. The sweeping of one beneficial allele means that the AGV in that gene also disappears. Therefore, since AGV and NGV can be both high when a selective sweep does not occur, but they are both reduced after a sweep, a positive correlation between AGV and NGV can be still maintained. Therefore, we need to ask whether there are other forms of natural selection in which NGV is lowered without reducing AGV. One such scenario, the hitchhiking effect of fluctuating selection, was provided by Barton (2000): fluctuating environment causing the adaptive alleles to oscillate between low and high frequencies, thus maintaining AGV without fixation or loss, is expected to reduce the levels of the surrounding NGV. The feasibility of such an evolutionary scenario is receiving growing attention, as fitness is indeed found to fluctuate rapidly and widely in natural populations (Bell, 2010; Messer et al., 2016) and population genomic studies have revealed seasonal oscillations of allele frequencies at a large number of sites (Bergman et al., 2014; Machado et al., 2021).
Under balancing selection, different alleles affecting fitness are maintained via heterozygote advantage, rare-allele advantage, or temporally/spatially heterogeneous selection. By definition, such loci harbor high levels of AGV (Aguilar et al., 2004; Charlesworth, 2006). The level of NGV is also expected to be elevated at sites closely linked to the loci of stable balanced polymorphism (Charlesworth, 2006). However, only very closely neighboring neutral sites may experience such an increase in polymorphism because meiotic recombination quickly erodes linkage disequilibrium around the selected loci (Fijarczyk & Babik, 2015). This suggests that a high level of AGV can be maintained by balancing selection without a proportional increase in NGV on the genomic average. Therefore, balancing selection should also contribute to the lack of a positive correlation between NGV and AGV.
To summarize, heterozygosity at adaptive and neutral loci is expected to be impacted by different evolutionary factors, which may explain why estimators of NGV are poor surrogates for AGV within plant populations.
3 | COMPARISON OF AMONG-POPULATION DIFFERENTIATION: NEUTRAL MARKERS VERSUS ADAPTIVE TRAITS
Since sessile plants are subject to spatially divergent selection, elucidating the effects of local adaptation on population differentiation has become more important in light of adaptation to changing environments, including global climate change (Ehrich & Raven, 1969; Savolainen, 2011; Colautti et al., 2012). A commonly used way to infer the impact of divergent selection on plant population differentiation is by comparing Q ST (reflecting differentiation caused by both neutral and selective forces) versusF ST estimates (reflecting differentiation due to neutral processes including genetic drift) (Whitlock, 2008). The neutrality expectation depends on the assumption that mutation rates (µ ) are substantially lower than migration rates (m ) (Hendry, 2002). Neutral markers having high µ (e.g., microsatellites) are not recommended to be used inQ STF ST comparisons (Hendry, 2002; Edelaar et al., 2011). However, Li et al. (2019) suggested the use of microsatellites by discarding the most variable loci (i.e., outliers).
The Q STF ST comparisons (i.e., elucidation of the relative magnitudes ofQ ST and F ST) have already provided valuable insights into responses of plant traits to spatiotemporal environmental heterogeneity (Kremer et al., 1997; Merilä & Crnokrak, 2001; McKay & Latta, 2002; Volis et al., 2005; Savolainen et al., 2007; Leinonen et al., 2008, 2013). TheQ STF ST relationship can have three different outcomes that have different interpretations (Merilä & Crnokrak, 2001; Leinonen et al., 2008):Q ST > F ST,Q STF ST, andQ ST < F ST. First, if Q ST >F ST, the observed trait differentiation exceeds neutral expectations and the fraction not explained by neutral processes is likely to have been caused by disruptive (divergent) selection. Second, if Q STF ST, trait differentiation is indistinguishable from the effects of drift, and, thus, there is no evidence for selection (Lande, 1992). Finally, ifQ ST < F ST, trait divergence among populations is less than expected due to genetic drift alone; this pattern is sug­gestive of spatially uniform or stabilizing selection (favoring average phenotypes) across populations.
Using several simple generalized linear models, Leinonen et al. (2008) carried out a meta-analysis of 55 animal and plant studies that used the same populations for both F ST andQ ST estimation. Their results confirmed the main conclusions of Merilä & Crnokrak (2001), who found a low but significant positive correlation between Q ST andF ST (Spearman rank correlation,r s = 0.39, P = 0.017; Leinonen et al., 2008), and, on average, Q ST >F ST (P < 0.001). Leinonen et al. (2008) suggested that genetic differentiation due to natural selection and local adaptation is the “norm,” not the exception. The positive correlation between the degree of adaptive phenotypic divergence and differentiation at neutral loci is mainly caused by limited gene flow and enhanced local genetic adaptation, known as “isolation by adaptation” (Nosil et al., 2007). Leinonen et al. (2008) further found that the study design (viz. , wild, broad sense, and narrow sense), marker type (restriction fragment length polymorphisms, random amplified polymorphic DNAs, microsatellites, allozymes, and AFLPs), and trait type (morphological traits and life-history traits) rarely explain any significant variance in the Q ST data. Furthermore, Leinonen et al. (2008) pointed out two potential biases in finding that 70% of Q ST values exceed the associated F ST values. First, a sampling bias due to the deliberate selection of populations from contrasting environments to be investigated, as well as focus on populations previously known to be phenotypically divergent. Second, a publication bias favoring studies reporting Q ST >F ST outcomes, possibly because of difficulties interpreting Q STF ST andQ ST < F STpatterns. For example, Q ST <F ST could be due to canalization, which is a process or tendency in which “species genetic backgrounds share the same genetic constraints” (Lamy et al., 2012) and “a fundamental feature of many developmental systems” (Hall et al., 2007). To partially distinguish canalization and uniform selection, Lamy et al. (2012) suggested “a bottom-up approach” that combines information fromQ STF ST comparisons and phylogenetic reconstruction. For a given trait, ifQ ST < F ST and phylogenetically closely related species occurring under different environmental conditions exhibit trait conservatism, then canalization could be inferred as an alternative to the classical uniform selection hypothesis (cf. fig. 3 in Lamy et al. [2012]). Well-known examples of canalization in plants are leaf shape in Arabidopsis thalianaand cavitation resistance found in all Pinus species (Hall et al., 2007; Lamy et al., 2011). The R package “driftsel” (Ovaskainen et al., 2011; Karhunen et al., 2013; 2014) can be used to differentiate between stabilizing selection, diversifying selection, and random genetic drift, allowing to circumvent a lot of the problems with the traditional Q STF STcomparisons.
The study by De Kort et al. (2013) was the first meta-analysis ofQ STF ST comparisons (401 cases that included each Q ST value per trait for each entry) exclusively focusing on plants. The authors compiled 51 entries representing 44 plant species from 18 families covering 17 entries for annuals, 19 for herbaceous perennials, and 15 for woody species. De Kort et al. (2013) found that averageQ ST values were significantly larger than the corresponding F ST values (0.345 versus 0.214, Wilcoxon signed-rank test, P = 0.003: paired t -test,P = 0.000, recalculated from original data from De Kort et al., 2013). The authors also found that the excess ofQ ST relative to F ST was significantly negatively correlated with F ST(β = –0.484, P < 0.01). A weak but positive overall relationship between pairwise Q ST andF ST values (r s = 0.278,P = 0.048; β = 0.464, P = 0.003, recalculated from De Kort et al., 2013) suggests that F ST in neutral markers could be to some degree predictive ofQ ST in quantitative traits. These correlations are what one would expect because (i) Q STreflects both neutral forces and natural selection caused by environmental differences and F ST only measures neutral processes including genetic drift and gene flow, (ii)Q ST and F ST estimates are based on the same (among-population) partition of total genetic variation, differing only in the data used in estimation—quantitative adaptive loci (the former) and neutral loci (the latter), and (iii) divergent selection that causes Q ST could also lead to the increase of F ST by restricting gene flow (“isolation by adaptation”; Nosil et al., 2007). In addition, De Kort et al. (2013) found a significant positive correlation between the average inter-population distance and theirQ STF STdifference values (P < 0.05), suggesting that isolation by distance plays an important role in adaptive evolution. The authors’ meta-analysis suggests that plant species are generally differentiated by natural selection in various types of traits (viz ., fitness [reproductive and physiological traits] and non-fitness [biomass-related and phenological traits] both in early life and in the adult stage). For example, the authors detected a largerQ STF ST difference values for non-fitness traits than for fitness traits, confirming the expectation that the former respond, in general, faster to directional selection than the latter (Merilä & Sheldon, 1999; Leinonen et al., 2008). Finally, De Kort et al. (2013) found slightly higherQ STF ST difference values for annuals than perennials (0.143 versus 0.123), but the difference was not significant. This may not support the prediction (De Kort et al., 2013) that perennials can respond to selection slower than annuals.
In summary, these differences in F ST andQ ST are a product of the different evolutionary forces such as drift, gene flow, and selection (Slatkin, 1973), which are further complicated by potential biasing effects caused by phenotypic plasticity, environmental maternal effects, non-additive genetic interaction, pleiotropic effects, and, as mentioned above, different µ in F ST andQ ST (for more details see De Kort et al., 2013).