Accepted Article

Received Date : 13-Feb-2014 Revised Date : 24-Apr-2014 Accepted Date : 25-Apr-2014 Article type

: Original Article

Landscape genomics and a common garden trial reveal adaptive differentiation to temperature across Europe in the tree species Alnus glutinosa

ORIGINAL PAPER

Hanne De Kort1, Katrien Vandepitte1, Hans Henrik Bruun2, Déborah Closset-Kopp3, Olivier Honnay1, Joachim Mergeay4 1: Plant Conservation and Population Biology, Biology Department, University of Leuven, Kasteelpark Arenberg 31, B-3001 Heverlee, Belgium 2: Ecology and Evolution Section, Department of Biology, University of Copenhagen, Universitetsparken 15, 2100 København Ø 3: Research unit of “Ecologie et Dynamique des Systèmes Anthropisés”, Jules Vernes University of Picardy, 1 Rue des Louvels, F-80037 Amiens Cedex, France 4: Research Institute for Nature and Forest, Gaverstraat 4, B-9500 Geraardsbergen, Belgium

E-mail:

[email protected];

[email protected];

[email protected];

[email protected]; [email protected]

This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/mec.12813 This article is protected by copyright. All rights reserved.

Accepted Article

Corresponding author: Hanne De Kort; [email protected]; Tel: +32 (0)16 32 15 20; Fax: +32 (0)16 32 19 68

Running title: Adaptability of A. glutinosa to climate

Abstract The adaptive potential of tree species to cope with climate change has important ecological and economic implications. Many temperate tree species experience a wide range of environmental conditions, suggesting high adaptability to new environmental conditions. We investigated adaptation to regional climate in the drought-sensitive tree species Alnus glutinosa (Black alder), using a complementary approach that integrates genomic, phenotypic and landscape data. A total of 24 European populations were studied in a common garden and through landscape genomic approaches. Genotyping-By-Sequencing was used to identify SNPs across the genome, resulting in 1990 SNPs. Although a relatively low percentage of putative adaptive SNPs was detected (2.86% outlier SNPs), we observed clear associations among outlier allele frequencies, temperature, and plant traits. In line with the typical drought avoiding nature of A. glutinosa, leaf size varied according to a temperature gradient and significant associations with multiple outlier loci were observed, corroborating the ecological relevance of the observed outlier SNPs. Moreover, the lack of isolation-by-distance, the very low genetic differentiation among populations and the high intra-population genetic variation all support the notion that high gene exchange combined with strong environmental selection promotes adaptation to environmental cues.

This article is protected by copyright. All rights reserved.

Accepted Article

Introduction The capacity to adapt to climatic conditions affects the long term survival and the geographical distributions of populations and species, and is therefore considered to be a key aspect of conservation and global change biology (Davis & Shaw 2001; Jump & Peñuelas 2005; Berg et al. 2010; Alberto et al. 2013). The adaptive potential of tree species in particular may have major ecological and economic consequences in the face of climate change, as trees dominate many terrestrial ecosystems at latitudes where the impact of climate change is expected to be most pronounced (Kremer et al. 2012). Nevertheless, although tree populations commonly possess low levels of neutral genetic differentiation, suggestive of high gene flow among populations and/or large effective population sizes, they often show substantial genetic differentiation at quantitative traits, suggesting substantial adaptive potential (McKay & Latta 2002; Savolainen et al. 2007; De Kort et al. 2012). Climate has been alternating between long glacial periods (up to 100,000 years) and shorter warmer interglacial periods (10,000-15,000 years), often with abrupt transitions between periods (Bennett 1990; Stewart et al. 2010). Many tree species survived these sudden climate changes and persisted for tens of thousands of years in relatively restricted refugia, indicating a considerable capacity to cope with strong selective pressures (Bennett et al. 1991; Svenning et al. 2008; Shafer et al. 2011). The successful post-glacial migration of many tree species out of these refugia has been suggested to be largely driven by standing adaptive genetic variation (Petit & Hampe 2006; Barrett & Schluter 2008; Temunović et al. 2013), likely maintained by efficient gene flow from refugial populations (Savolainen 2011; Kremer et al. 2012; Hendry 2013). Considering that these past processes gave rise to extant patterns of climatic adaptation, the current distribution of adaptive over neutral genetic differentiation may lend highly valuable insights in the processes governing the adaptive potential of extant tree

This article is protected by copyright. All rights reserved.

Accepted Article

populations in the light of climate change (Hof et al. 2011; Hoffmann & Sgrò 2011; Quintero & Wiens 2013). Although tree species are considered to have high adaptability to climate, ongoing habitat loss and fragmentation have been demonstrated to considerably reduce population genetic diversity and to increase inbreeding in many tree species, which may threaten the ability to cope with the synergistic effects of habitat reduction and climate change (Brook et al. 2008; Hof et al. 2011; Meier et al. 2012; Vranckx et al. 2012). Moreover, the complex genetic architecture of plant traits and antagonistic correlations among genes and phenotypic traits may limit the ability of populations to respond to divergent selection pressures (Aitken et al. 2008; Wagner & Zhang 2011; Hendry 2013). Therefore, it is unclear how well tree populations may respond to climate change, emphasizing the need for a better understanding of the genetic and phenotypic consequences of climatic variability (Garzón et al. 2011; Reed et al. 2011). Common garden trials have a long history in measuring adaptive differentiation of quantitative traits among tree populations originating from regions with different climates (Savolainen et al. 2007; De Kort et al. 2012). Although growing plants in the same environment is assumed to control for phenotypic plasticity, environmental maternal effects and the common garden environment may still induce plastic plant responses that can bias inferences on adaptive divergence (Cano et al. 2004; Kawecki & Ebert 2004; Gienapp et al. 2008; Scheepens et al. 2010). The emergence of relatively time- and cost-efficient high throughput genome-wide DNA sequencing, however, offers the opportunity to screen genomes for signs of adaptive differentiation (i.e. outlier loci), unconfounded by phenotypic plasticity (Morin et al. 2004; Allendorf et al. 2010; Narum et al. 2013). Combined with detailed information on climatic variables, such an approach can reveal relevant associations between putative adaptive loci and environmental factors, simultaneously accounting for nonThis article is protected by copyright. All rights reserved.

Accepted Article

adaptive geographic patterns affecting allele frequencies (cfr. landscape genomics; Joost et al. 2007; Coop et al. 2010; Manel et al. 2010; Orsini et al. 2013). Nevertheless, as phenotypic traits with continuous trait values are often under polygenic control, with many loci exerting small effects (Zhao et al. 2011; Le Corre & Kremer 2012; Hendry 2013), population genomic screens can miss the signature of adaptive differentiation in such traits. Hence, complementary approaches integrating genomic, phenotypic and landscape information may be very fruitful for testing hypotheses on adaptability in non-model organisms. For this study, we explored patterns of local adaptation to climate in 24 populations of Alnus glutinosa, a wind-pollinated and water-dispersed tree species of riparian ecosystems. The species is widely distributed across Europe, experiencing a wide range of environmental conditions (Meusel et al. 1965). Phylogeographical analyses based on chloroplast DNA and fossil pollen revealed that most of central and northern Europe was colonized from a refugium close to the Carpathian Mountains (Huntley & Burks 1983; King & Ferris 1998). This finding suggests considerable adaptability to different climatic conditions during postglacial migration. Moreover, ancient tetraploidisation probably allowed the species to maintain high genetic diversity and adaptability in the rear-edge climate relicts of northern Africa, by increasing the effective population size, which further corroborates the adaptive potential of the species (Lepais et al. 2013). A previous landscape genetic study based on 163 AFLP markers already suggested a signal of adaptation to climate conditions in Europe (Cox et al. 2011), but lacked complementary common-garden trials to confirm adaptation at the level of the phenotype and to evaluate the contribution of putatively adaptive loci in shaping adaptive phenotypic differentiation. To further assess the magnitude of genomic adaptation to climate and to elucidate the ecological relevance of genome-wide outliers, we combined a large common garden trial with two independent landscape genomic approaches. Genotyping-BySequencing (GBS) was used to identify Single Nucleotide Polymorphisms (SNPs) across the

This article is protected by copyright. All rights reserved.

Accepted Article

genome, which were then subject to a Bayesian outlier detection method and a locus-specific landscape genomic approach. Multivariate redundancy analyses were applied (i) to estimate the relative contribution of geography (isolation-by-distance, IBD) and climate (isolation-byadaptation, IBA) in shaping neutral vs. adaptive genetic patterns, and (ii) to assess the relative importance of neutral and adaptive loci in shaping variation in phenotypic plant traits.

Materials & Methods Study species Black alder (Alnus glutinosa (L.) Gaertn, Betulaceae) is a widespread deciduous tree distributed across Europe (Meusel et al. 1965). It is a monoecious, self-incompatible and wind-pollinated species, and its seeds are mainly dispersed by water (McVean 1953; Chambers & Elliott 1989). Due to low nutrient demands, high growth rates, and the ability to stabilize riverbanks and prevent erosion, A. glutinosa contributes particularly to riverine biodiversity by providing habitats for a specific flora and fauna both on the tree itself and in the flooded root system (Dussart 1999; Brandle & Brandl 2001). A. glutinosa is a typical drought avoiding and flooding tolerant species, generally occurring near wetlands and river banks (Braun 1974; Herbst et al. 1999). Sample collection Seeds were collected from a total of 356 trees across 24 European populations (Fig. 1, Table 1). Eight regions (three populations per region) were sampled, including five Belgian regions (Sandy Region, Eastern Brabant, Western Brabant, Campine, and the Ardennes, a southernBelgian region with a slightly colder and wetter climate), a French region (Picardy), a Danish region (Sjælland) and an Italian region (Tuscany). This hierarchical sampling design (Fig. 1)

This article is protected by copyright. All rights reserved.

Accepted Article

was adopted to assess the magnitude of adaptive differentiation at two spatial scales; the small scale has been tackled previously in the context of Belgian seed zone delineation and showed limited adaptation within Belgium based on analyses of pairwise population differentiation (De Kort et al. unpublished manuscript). Here, we investigate the large European scale which allows disentangling of the effects of different environmental factors on the magnitude of local adaptation. Seeds from the 356 mother trees were stratified on moist filter paper at 4°C for 12 weeks to break dormancy. Approximately 100 seeds per mother tree (seed family) were subsequently planted on potting soil in 48 cell division trays to germinate. Three seedlings from each of 15 families per population then were randomly selected from the germinated pool and transplanted into plastic pots (18 cm diameter x 21 cm depth) filled with universal Saniflor potting soil, to set up in a common garden (see below). After one year of growth, leaves were collected from one plant per seed family and dried on silica gel prior to DNA extractions. SNP genotyping DNA from 320 individuals was extracted using DNeasy Plant Extraction kits (Qiagen Inc., Valencia, CA, USA), at a concentration of approximately 100 ng/μl, as measured using a NANODROP2000 spectrophotometer (ISOGEN LIFE SCIENCE, Belgium). DNA integrity was evaluated on 1.5% agarose gels. De Novo Genotyping-By-Sequencing (GBS) was used for constructing reduced representation libraries for the Illumina next-generation platform. A PstI GBS library comprising 320 samples, 57 replicated samples and 7 blanks was prepared according to Elshire et al. (2011), and the resulting library was diluted and sequenced on an Illumina HiSeq 2000 (Cornell University Genomics Core Laboratory). Briefly, Raw DNA sequences were analyzed with the Universal Network Enabled Analysis Kit (UNEAK) pipeline, implemented in TASSEL v3.0 (For details, see Lu et al. 2013). To reduce the impact

This article is protected by copyright. All rights reserved.

Accepted Article

of sequencing errors, an error tolerance rate parameter of 0.03 was implemented because the read counts of rare tags, which are assumed to result from sequencing error, were less than 3% according to the frequency distribution of all tags (sensu Lu et al. 2013). The resulting genotypes were filtered to those with sequencing depth between 3 and 127 (using VCFtools v0.1.10, Danecek et al. 2011). Final filtering of the data involved elimination of individuals with greater than 90 % missing data and SNPs with more than 20 % missing data (VCFtools v0.1.10). Overall, this yielded genotypes at 1990 polymorphic loci (a total of 31 monomorphic loci were excluded) with on average 3.47 % missing data per genotype. SNP coverage (i.e. the number of reads/site/individual) and allele frequencies were on average 65.81 (±62.83) and 0.18 (±0.12) respectively (Supporting information 1). The 57 replicated samples rendered a mean genotyping error rate, i.e. the mean number of erroneously genotyped markers per sample, of 5.5%, which possibly reduced the power to detect outlier loci to some extent. The ancient tetraploidisation of A. glutinosa is not expected to affect genotyping results as tetraploid individuals are restricted to northern Africa, which is far from our sampling sites (Lepais et al. 2013). Outlier detection and overall genetic structure Detection of outlier SNPs was based on the Bayesian likelihood approach implemented in BAYESCAN (Foll & Gaggiotti 2008). This approach compares population-specific allele frequencies with a common migrant gene pool, thereby allowing different migration rates, which reduces the amount of false positives considerably (Narum & Hess 2011). The use of a model with a common migrant gene pool may be particularly suited here, assuming that the Alder populations originate from the same Carpathian refugium. Log10 values of the posterior odds (PO) > 0.5 were taken as ‘substantial’ evidence for selection (Jeffreys 1961; Foll & Gaggiotti 2008). The false discovery rate (FDR) was set at 0.05 to adjust the log10(PO) significance thresholds corresponding to the 0.5 values. By separating the detected outlier This article is protected by copyright. All rights reserved.

Accepted Article

SNPs, a neutral and an outlier dataset were constructed. The neutral dataset was used for an AMOVA with 10,000 permutations to estimate FST among all populations (ARLEQUIN 3.5, Excoffier & Lischer 2010). Genetic clustering among populations (Principal Coordinates Analysis, PCoA) was assessed for both the neutral and the outlier dataset using GENALEX 6.5 (Peakall & Smouse 2006). Levels of genetic diversity within populations (He) and Pairwise FST values between populations were calculated based on all SNPs using 10,000 permutations (ARLEQUIN 3.5). Landscape genomics A total of 8 climatic variables related to precipitation, temperature and relative air humidity were downloaded from Worldclim.org (temperature and precipitation) and from the SAGE (Center for Sustainability and the Global Environment) website (humidity; Supporting information 2). To remove collinearity among climatic variables, we applied a Principal Components Analysis (PCA) on all climatic variables. The climate related variables could be reduced to two main principal components retaining 92.35% of the original variation (Supporting information 2). These two principal components (PC1 positively related to temperature and PC2 positively to precipitation) were used for the landscape genomic analyses. Latent fixed mixed modeling (LFMM) was used to associate each outlier SNP with temperature (PC1), precipitation (PC2), latitude and longitude while accounting for neutral population structure (Frichot et al. 2013). Neutral population clustering into two distinct clusters was derived from the genetic PCoA analysis (Fig. 2). No further clustering of individuals into more subtle genetic clusters could be observed when visualizing different PC axes or by performing the PCoA on subsets of the populations. An MCMC algorithm was used for each of the 4 environmental variables, using 1000 sweeps for burn-in and 10,000

This article is protected by copyright. All rights reserved.

Accepted Article

additional sweeps to compute LFMM parameters (|z|-scores) for all loci (Frichot et al. 2013). Significance was determined using a standard Gaussian distribution and a Bonferroni correction for multiple testing. A SNP effect was considered significant when its |z|-score was greater than 4, corresponding to a p-value of p