Differences among six woody perennials native to Northern Europe in their level of genetic differentiation and adaptive potential at fine local scale

Abstract The ability of perennial species to adapt their phenology to present and future temperature conditions is important for their ability to retain high fitness compared to other competing plant species, pests, and pathogens. Many transplanting studies with forest tree species have previously reported substantial genetic differentiation among populations within their native range. However, the question of “how local is local” is still highly debated in conservation biology because studies on genetic patterns of variation within and among populations at the local scale are limited and scattered. In this study, we compare the level of genetic differentiation among populations of six different perennial plant species based on their variation in spring flushing. We assess the level of additive genetic variation present within the local population. For all six species, we find significant differentiation among populations from sites with mean annual temperature ranging between 7.4°C and 8.4°C. The observed variation can only be partly explained by the climate at the site of origin. Most clear relationship between early flushing and higher average spring temperature is observed for the three wind‐pollinated species in the study, while the relations are much less clear for the three insect‐pollinated species. This supports that pollination system can influence the balance between genetic drift and natural selection and thereby influence the level of local adaptation in long‐lived species. On the positive side, we find that the native populations of woody plant species have maintained high levels of additive genetic variation in spring phenology, although this also differs substantially among the six studied species.


| INTRODUCTION
Phenology is important for fitness of perennial species. Higher spring temperatures are expected to prolong the growing seasons due to earlier bud burst (Dragoni et al., 2011;Menzel & Fabian, 1999;Menzel et al., 2006;Richardson et al., 2010Richardson et al., , 2013Vitasse, Delzon, Dufrêne, et al., 2009), while lack of synchrony between phenology and occurrence of spring frost events increases risk of damage to early flushing plants (Duputié, Rutschmann, Ronce, & Chuine, 2015). Yet again, raised temperatures with no frost events and changes in daily temperature fluctuations could also result in later bud burst and influence time between bud development and growth cessation (Chmura et al., 2011;Rohde, Bastien, & Boerjan, 2011;Way, 2011). Several studies have documented substantial variation among populations in their phenology reflecting their geographic origin (Alberto, Derory, et al., 2013;Chuine & Beaubien, 2001;Salmela, Cavers, Cottrell, Iason, & Ennos, 2013). It is therefore fair to assume that genetic variation within species combined with divergent selection has played an important role for the ability of many tree species to thrive in a large distribution range across strong environmental gradients (Brousseau et al., 2016;Kawecki & Ebert, 2004;Pluess et al., 2016). However, species differ in their response and spring and autumn phenology may show different patterns. A study based on 59 tree species from similar climatic clines thus showed a relative clear pattern with respect to bud set in autumn but less clear pattern in spring bud burst (Alberto, Aitken, et al., 2013). The expectation of local adaptation often leads restoration and conservation programs to focus on gene pools at the local scale rather than the regional scale (Stanturf et al., 2015), but only few studies have actually assessed the variation at the local scale (within ~100 km).
Variation among individual genotypes in their phenology triggers selection at the population level in favor of the new conditions, if the variation is in fitness traits and expressed with moderate or high heritability (Alberto, Aitken, et al., 2013;Falconer & Mackay, 1996). Genetic differentiation within and among population is therefore important for the species ability to adapt to climate change (Kubisch, Holt, Poethke, & Fronhofer, 2014). However, natural selection is not the only evolutionary force, because the combined actions of natural selection, gene flow and genetic drift drive the level of genetic variation within and among populations in fitness traits (Nadeau, Meirmans, Aitken, Ritland, & Isabel, 2016). The realized patterns therefore depend on effective population sizes, force of natural selection and limitations in pollen and seed dispersal (Aitken & Whitlock, 2013;Savolainen, Pyhäjärvi, & Knürr, 2007). Natural selection tends to increase the level of genetic variation among populations, but reduce the variation within populations leading to local adaptation. Fragmentation into small populations can create drift that also increase variation among population variation, but without supporting local adaptation. Small populations can rather reduce fitness due to inbreeding, at least in outcrossing species. Pollination and seed dispersal across the landscape among populations will counteract the fragmentation, but seed and pollen dispersal across environmental gradients will also counteract natural selection and thereby reduce the ability of populations to adapt to specific local conditions (Kubisch et al., 2014). The level of gene flow is expected to differ among woody species depending on occurrence, landscape features, mating system, pollen, and seed dispersal vectors.
We therefore expect that the degree of local differentiation-and the degree to which this reflect local adaptation-will differ among woody plant species depending on their major life-history traits.

| Objectives of this study
On the above background, the objectives of this study were to use data on spring phenology to 1. Assess the degree to which woody perennials reveal fine-scale genetic differences among geographically and climatic close (differences in mean annual temperature around 1°C) populations, and test whether differences (if any) influence the fitness revealing signs of local adaptation.

Estimate levels of genetic variation within and among populations
in spring phenology and compare the levels of population differentiation (Q ST ) and additive genetic variation (V A ) among species.
Based on the results, we discuss the implications for dynamic conservation of genetic resources and the ability of woody plant species to adapt to future climate conditions.

| MATERIALS AND METHODS
Six broadleaved species in Denmark were studied, of which three are Description of the field trials, location of populations, and climate information at the population site and information on families within each population are provided in Appendix S1.

| Assessment of phenotypic data and estimation of variance components
We used spring phenology assessed as bud burst scores using a scale from 0 to 8, where class 0 was closed winter buds and class 8 was fully where Y ijkl is the trait measured for tree l, μ is the overall mean of the trait, B i is the fixed block effect, P j is the fixed population effect, λ ij is the fixed population by block interaction, f k(j) is the random effect of family within population, τ ijk is the random effect of plot k, and ε ijkl is the residual. Q ST values were estimated from model (1) as well having population effects as random.
The significance of the genetic variances of traits within sites and genotype by environment interaction across sites was tested using the log likelihood ratio (Gilmour et al., 2009). The significance of the populations was tested using the Satterthwaite approximation (Satterthwaite, 1946) in the procedure GLM in the statistical software program SAS (SAS Institute Inc. 2011).

| Quantitative genetic analysis
We estimated the additive genetic variance (V A ) as 4σ f 2 and narrow sense heritability within sites as; the estimated family variance, σ τ 2 is the estimated plot variance, and σ ε 2 is the estimated within plot variance. The phenotypic variance was estimated as ⌢ V P = σ 2 f +σ 2 τ +σ 2 ε . Families were considered to represent groups of half-sibs. This assumption will overestimate the additive genetic variance, heritability, and expected response to selection, if the offspring within progeny groups on average are more related than half-sibs (Falconer & Mackay, 1996). While the assumption of half-sibs may provide a good fit for most trees species (Kjaer, McKinney, Nielsen, Hansen, & Hansen, 2012;Larsen & Kjaer, 2009), the situation may be more complicated for R. dumalis, where polyploidy may be present as the variance between families is biased due to a fraction of the dominance genetic variance, even if the families consist of pure half-sibs (Roberts, Gladis, & Brumme, 2009).
The degree of differentiation among populations was estimated as the Q ST values (Spitze, 1993) using the formula; where ⌢ V POP is the estimated variance between populations, and ⌢ V A is the estimated additive genetic variance. (1)

| Support to the hypothesis of location adaptation
Climate data for the locations of origins were estimated with the ClimateEU v4.63 software package following the methodology described by (Hamann, Wang, Spittlehouse, & Murdock, 2013). Weighted regression analysis to test relationship between phenology of populations and climate at the population sites was carried out using procedure REG in SAS (SAS Institute Inc. 2011) having the predicted values of populations as dependent variable and climate variables as explanatory variables. One divided by the error variance for the predicted population values were used as weights. The climate variables tested for bud burst were monthly minimum temperatures and differences between monthly maximum and minimum temperatures in March, April, May and June, because these data are expected to provide good proxies for the risk of frost occurring after flushing. A backward selection approach (5% level) was used for the selection of climate variables that could best explain the variation in flushing time. The relationship between geographic distances and difference in average budburst was tested by a Mantel test as implemented in R version 3.2.2 (Dray & Dufour, 2007).

| Fitness effects of spring phenology
Genetic correlations between bud burst and height growth were estimated based on individual tree data to assess whether the height as a proxy for realized fitness in the present climate varied with the phenology. Genetic correlations could be estimated between survival and phenology based on plot means, because the plants were grown in family plots in all trials (see Appendix S1). Genetic correlations were estimated according to Falconer and Mackay (1996) as , where σ fxy is the family (within population) covariance between trait x and y, σ fx 2 is the family (within population) variance for trait x, and σ fy 2 is the family (within population) variance for trait y.

| Presence of genetic variation within and among populations in bud burst
Genetic variation both within and among populations was significant in all six species, but with large differences among the species. The additive genetic variance for bud burst (V A ) thus ranged from 0.07 in C. sanguinea to 0.34 in Q. petraea (Table 1)

T A B L E 1 Family variation and genetic parameters for bud burst in different species
F I G U R E 2 Difference among species in genetic variation within populations (V A = additive genetic variance) and population differentiation (Q ST value) for bud burst insect-pollinated species, wind-pollinated species

| Relationship of phenology with growth, fitness, distance between populations and climate at original population site
The

| DISCUSSION
Our results show that fine-scale local genetic differentiation in fitness traits such as phenology indeed can be present and to a larger extent than previously anticipated in the Danish gene conservation strategy (Graudal, Kjaer, & Canger, 1995). However, the analysis across the six different species did not provide unique patterns of local adaptation, as the absence of significant regression between bud burst time and spring temperature in the three insect-pollinated species suggests that the variation among population is not always simply reflecting local adaptation. It is rather more likely a result of natural selection and neutral processes simultaneously acting as predicted by theory (Nadeau et al., 2016).The fact that populations in all studied six species were significantly differentiated in their spring phenology suggests that local populations of woody perennials more often than not are genetically differentiated, and this can be the case even if they only are separated by 10 to 35 km ( Figure 4) and growing in areas with low variation in altitude and where the spring temperature varies only between 1°C and 2.5°C (Figure 3). This finding should be compared to the prediction of Northern Europe becoming 2-4°C warmer by  insect-pollinated species with small and scattered populations (Petit & Hampe, 2006;Smith & Donoghue, 2008). The results correspond to the expectation that long-distance gene flow through wind pollination will maintain connectivity among populations (lowering Q ST ) and counteract loss of genetic variation (maintaining high h 2 ) within populations (Sork, 2016). High levels of gene flow among populations across landscapes have been reported in the wind-pollinated Quercus (Gerber et al., 2014). The insect-pollinated C. sanguinea was in the opposite corner of Q. petraea in Figure 2, reflecting a much higher proportion of genetic variation located among populations. C. sanguinea in general occurs in scattered and small populations in Denmark (Ødum, 1968).
Seed dispersal by birds (Krüsi & Debussche, 1988) may generate longdistance gene flow across the landscape, but the observed level and distribution of genetic diversity in the present study point toward genetic drift rather than natural selection as important driver behind the observed population differentiation. The true difference between Q. petraea and C. sanguinea in level of genetic variation may be even higher, because the assumption of half-sib relationship within seed from single tree collection is likely to create a bias toward overestimation of V A in the rose as discussed in M&M above. M. sylvestris is again a different case in Figure 2, with low level of genetic differentiation but also relatively low additive genetic variance. Pollinating insects have been reported to visit primarily trees in immediate vicinity, but long-distance pollination events are also likely to occur (Larsen & Kjaer, 2009;Reim et al., 2015) and seeds moved by birds and deer that feed on the wild apples also create long-distance gene flow (Larsen & Kjaer, 2009). Studies based on neutral SSR markers revealed low level of population differences within Denmark (F ST = 0.03) (Larsen, Asmussen, Coart, Olrik, & Kjaer, 2006), which is very close to the es-  (Hynynen et al., 2010), the results suggest that B. pubescens possesses the resilience at the population level to change phenology corresponding to the predicted increase in temperature. A similar conclusion can probably be drawn for Q. petraea also (Figure 3). For many of the other studied species, the generation time is possibly less than 20 years (e.g., R. dumalis, M. sylvestris, C. avellana, and C. sanguinea), and for these species, it therefore also seems reasonable to assume that populations can evolve in spring phenology with a speed that can match the predicted increase in temperature. The response will only be realized if early bud burst has substantial influence on fitness (survival and reproduction) or because of directional selection implemented in domestication programmes. The estimates in the present study refer to the latter situation, because the heritability was estimated in managed progeny test and the realized heritability in natural population may be substantially lower due to higher environmental and developmental heterogeneity. In Denmark, all major woody plant species are included in domestication programs based on breeding seed orchards in order to support effective and rapid selection for continued fitness ).
Our finding on genetic differences among geographically close populations (cf.  especially, the species associated with small population sizes and limited gene flow. Populations at the low latitudinal limit of the species range have in general maintained high biological diversity over time as shown for, for example, Abies alba (Bergmann, Gregorius, & Larsen, 1990;Larsen, 1986) and southern populations should therefore be considered as a source of gene pool for assisted migration of genotypes under the future climate predictions (Hampe & Petit, 2005). The populations in the present study are close to the northern distribution range of the species (except for B. pubescens) and genetic diversity may therefore be lower compared to populations closer to the refugial areas. But we still observed substantial level of genetic variation in the fitness-related trait. For species occurring in extremely small fragmented populations assisted migration at a more localized landscape level may still be desirable to counteract inbreeding within these populations due to random drift. Adaptive potential of a species is not only determined by the presence of genetic variation within and among populations as studied here, but also by the presence of phenotypic plasticity in adaptive traits (Nicotra et al., 2010). Final conclusions on the adaptive potential of species to climate change will therefore also depend on the role of plasticity in local adaptation in the studied species.