Exon-Enriched Set of Single-Nucleotide Polymorphisms Shows Associations with Climate in European Beech (Fagus sylvatica L.)

Abstract

European beech is an ecologically and commercially important species, which is expected to decline in several regions because of heat and drought stress associated with climate change. Knowledge of the genetic basis of the adaptation to climate is needed to guide assisted migration. Genetic variation at 1704 single-nucleotide polymorphisms (SNPs) resulting from ddRAD sequencing, primarily located in gene exons, was studied in 181 specimens representing 123 populations distributed over most of the whole range. Bayesian analysis of population structure yielded two clusters exhibiting a clear longitudinal cline and correlated with indicators of low temperatures and temperature fluctuation. Five SNPs were significantly associated with climatic variables related primarily to heat and temperature ranges. Two alternative explanations are offered for the observed response patterns: (i) differential sensitivity to heat and (ii) response mediated by vegetative phenology.

1. Introduction

European beech (Fagus sylvatica L.) is a dominant tree species of a wide range of forest ecosystems in Europe [1]. In spite of its success in colonizing a major part of Europe, beech populations have always been negatively affected by human activities (replacing natural beech stands by conifer monocultures [2]) or human-induced factors such as air pollution [3]. Currently, it is increasingly threatened by climate change, especially on the rear edge of the distribution range [4,5]. Physiological drought associated with prolonged periods without rainfall and high evapotranspiration driven by increasing temperatures is causing the decline of beech in several parts of southern and central Europe, especially in younger age classes [6,7,8]. Field surveys have documented mass mortality of adult beech stands in several regions [9,10,11], which are likely to increase dramatically after the extreme summer drought in 2022.

Climate change is currently a major threat for forest ecosystems and thus an important challenge facing both nature conservation and forestry at the national, European, and worldwide levels. Global surface temperature over land already increased by 1.59 °C in 2011–2020 compared to 1850–1900 and is predicted to increase by further 1.4 °C until the end of the century even under the most optimistic (although hardly realistic) scenario [12]. The increase in temperature is not a problem in and of itself, but is associated with weather extremes such as heatwaves, long drought periods, and extreme storms. Repeating these facts may be perceived as annoying by the general public (especially when influenced by ‘climate skeptics’); however, specialists in the field need to not only realize the urgency but also prepare and accomplish mitigation measures. Assisted migration, i.e., the transfer of genetic material from populations, which in the past have experienced climatic conditions expected on the target sites in the future (and are thus expected to be adapted to future climates), is among the most frequently proposed solutions for forests [13].

The knowledge about patterns of adaptive genetic variation in forest trees at the molecular level is steadily increasing, but still is quite limited. The availability of completely sequenced reference genomes of several species has fostered research in this field [14]. The candidate gene approach has long been preferred in genomic studies of forest trees, as the rapid decay of linkage disequilibrium in most forest tree species (especially those with large populations and thus commercially important) makes traditional approaches for mapping quantitative trait loci difficult [15,16]. Currently, it is gradually being replaced by studies on a larger scale in terms of genome coverage, employing numbers of single-nucleotide polymorphisms (SNPs) in the order of thousands or tens of thousands. In forest trees with generally large genomes, whole-genome resequencing remains mostly unaffordable from the financial point of view; targeted sequence capture [17,18] or reduced-representation sequencing [19,20] are used as alternatives. Particularly double-digest restriction site-associated DNA sequencing (ddRAD) was found to be a useful tool for SNP discovery and adaptation studies in trees [21,22,23].

Studies on European beech have not been an exception in this development. Candidate gene resequencing was preferred at the beginning of beech genomics studies [24,25,26]. Recently, reduced-representation sequencing including ddRADseq has become more frequently used [20,27,28], but whole-genome resequencing studies appear as well [29].

Our earlier studies on SNP associations with climate relied on the candidate gene approach [30,31,32]. In order to increase the number of polymorphisms, this study is based on the ddRAD sequencing approach in populations covering a major part of the distribution range of European beech colonized during the Holocene from a common source in expectation that the patterns arising from adaptation are not confused with differences caused by different population history. The objective of the study was revealing SNPs responsive to climatic trends.

2. Materials and Methods

2.1. Experimental Material

Material for this study was collected mainly in international provenance experiments with European beech and a small part in natural populations. To minimize the effects of history and demography on genetic variation, populations originating from the glacial refugium located close to the eastern foothills of the Alps (Slovenia and Istria) were chosen [33]. Provenance trials included two Slovak plots of the experiment coordinated by the Institute of Forest Genetics of the J. H. von Thünen Institute in Grosshansdorf, Germany, which was established in two series in 1995 and 1998 with 2-year-old seedlings (100 and 32 provenances covering a major part of the distribution range of beech in Europe were planted at the localities Vrchdobroč and Tále in central Slovakia, respectively). At these plots, material from 62 and 32 provenances, respectively, was collected for this study. Another provenance trial serving as a source of material was a plot of the international experiment with southeast European provenances at the locality Medvednica, Croatia, where 14 provenances were sampled. The remaining material was collected in 15 natural populations (Figure S1 and Table S1). Annual and seasonal climatic variables as well as climatic indices for the provenances originated were derived by the ClimateEU v. 4.63 software [34] and were complemented by selected bioclimatic variables from the WorldClim high-resolution interpolated climate database [35]. The bioclimatic variables were derived from meteorological data within the period 1960–1990 at the 30″ resolution. The list of the climatic variables used is provided in Table S1.

2.2. DNA Extraction and Sequencing

Twigs with dormant buds were collected from trees, for which physiological data were available, surviving until December 2020 (1 to 2 trees per population). Total genomic DNA was extracted from the silica-dried buds using a modified CTAB protocol following Doyle and Doyle [36] (1987). DNA concentration and quality were assessed with a Qubit 4 Fluorometer (Thermo Fisher, Waltham, MA, USA) and NanoDrop (Thermo Fisher, Waltham, MA, USA), and samples were sent for sequencing to IGA Technology Services, Udine, Italy.

Double-digest restriction site-associated DNA sequencing (ddRAD) was applied to reveal single-nucleotide polymorphisms (SNPs) in the nuclear genome. ddRAD libraries were produced by IGA Technology Services using a custom protocol, with minor modifications with respect to Peterson’s double-digest restriction site-associated DNA preparation [37]. A total of 300 ng of genomic DNA was double-digested with 2.4 U of both SphI and MboI endonucleases (New England BioLabs, Ipswich, MA, USA) in 30 µL reaction supplemented with CutSmart Buffer and incubated at 37 °C for 90 min, then at 65 °C for 20 min. Fragmented DNA was purified with AMPureXP beads (Agencourt, Beverly, MA, USA) and ligated with 200 U of T4 DNA ligase (New England BioLabs) to 2.5 pmol of overhang barcoded adapter for rare cut sites and to 5 pmol of overhang barcoded adapter for frequent cut sites in 50 µL reaction incubated at 23 °C for 60 min and at 20 °C for 60 min followed by 20 min at 65 °C. Samples were pooled and bead-purified. Targeted fragment distribution was collected on BluePippin instrument (Sage Science Inc., Beverly, MA, USA) with a setting in the range of 520–640 bp. Gel-eluted fraction was amplified with indexed primers using Phusion High-Fidelity PCR Master Mix (New England BioLabs) in final volume of 50 µL and subjected to the following thermal protocol: [95 °C, 3 min]—[95 °C, 30 s—60 °C, 30 s—72 °C, 45 s] × 10 cycles—[72 °C, 2 min]. After cleanup, libraries were checked with both Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA, USA) and Bioanalyzer DNA assay (Agilent Technologies, Santa Clara, CA, USA). Libraries were sequenced with 150 cycles in paired-end mode on NovaSeq 6000 instrument following the manufacturer’s instructions (Illumina, San Diego, CA, USA).

2.3. Data Analysis

The initial bioinformatic analyses performed by IGA Technology Services consisted of demultiplexing raw Illumina reads using the process_radtags utility included in Stacks v2.61 [38]. Alignment to the reference F. sylvatica genome (version 1.2 chromosome-level assembly; http://thines-lab.senckenberg.de/beechgenome/ accessed on 21 June 2022) [12] using BWA-MEM [39] with default parameters and selection of uniquely aligned reads (with a mapping quality > 4). Detection of the covered loci from the aligned reads using the gstacks program in Stacks v2.61. Filtering of detected loci using the population program in Stacks v2.61, with option −R = 0.75 in order to retain only loci that are represented in at least 75% of the dataset. To remove loci suspected of technical errors, a cut-off max-obs-het = 0.8 was used in order to process a nucleotide site at a locus with an observed heterozygosity at maximum of 80%. After the initial bioinformatic analyses, SNPs and samples were additionally filtered based on several criteria: SNPs with quality value Q < 13 (Q = 26 was the actual lowest value); SNPs and samples with >20% missing data; SNPs with <5% minor allele frequency (MAF); and SNPs and samples with read depth further than ±2 or 3 standard deviations from the mean, respectively. These additional steps reduced number of samples from 191 to 181 and SNPs from 346,503 to 25,068. After this, SNPs not located in exons were removed, using genome annotation provided by the reference genome [12], which reduced the number of SNPs to 4449. Finally, only one SNP per locus was retained, the one with lowest number of missing values, for a final number of 1740 SNPs.

To reveal potential substructure, the Bayesian clustering algorithm implemented in STRUCTURE version 2.3.4 [40] was applied to cluster individuals into K groups, employing the admixture model with correlated allele frequencies. K values were tested from 1 to 10. A burn-in period of 100,000 iterations was followed by 50,000 iterations for estimation of the membership coefficients. Fifteen independent Markov chains were run for each K. The choice of the true number of clusters K was based on the method of Evanno et al. [41] (2005), and 10 runs with the highest posterior probability were aggregated by CLUMPP ([42]; Figure S2). Cluster proportions were correlated with the index of continentality after Conrad and Pollak [43]: k = 1.7A/sin(θ + 10) − 14; where k is the index of continentality, A the average annual temperature range (BIO7), and θ is the latitude.

Associations between SNPs and the phenotypic traits were assessed using TASSEL v. 5.0 [44]. The SNP/trait associations were tested under the mixed linear model (MLM) to correct for neutral genetic structure. To do this, the MLM function was supplied with kinship and PCA matrices generated by the respective functions in TASSEL from the filtered SNP data. All settings we left at default values for the analysis. Second, latent factor mixed modeling (LFMM) [45] was used as implemented in the LEA package in R [46]. Since LFMM requires imputation of missing genotypes, this was performed using the impute function of the same package. The imputation was performed 24 times, and these datasets were tested using the LFMM function. Median p-values were calculated from the 24 iterations, and then q-values were calculated to adjust for multiple testing using the R package qvalue [47]. Finally, redundancy analysis (RDA) was performed using a process adapted from Capblancq et al. [48] using the R package vegan [49]. Missing genotypes had to be imputed again: for this purpose, the 24 imputed datasets from LFMM were used, p-values were calculated as a median of these 24 datasets, and q-values were again calculated using the package qvalue. To assess correlations between SNPs and environmental variables in RDA, angles were calculated between the vectors of SNP scores on the first 5 axes of RDA (with 0 being the origin of the coordinate system) and environmental variables. The angles were calculated for each of the 24 imputed datasets, and the median direction was calculated using the R package circular [50]. Correlations between SNPs and environmental variables were calculated as cosine of these angles, and correlation coefficients |r| > 0.6 were considered.

3. Results

3.1. Population Structure

The ΔK measure after Evanno et al. [41] clearly peaked at K = 2, with a secondary peak at K = 4. By principle, ΔK cannot indicate K = 1 as the most probable number of clusters (i.e., the absence of any structure). However, the curve of log-posterior probabilities indicates that K = 2 is the appropriate estimate of the number of gene pools (Figure S2a). Except for a few outliers in Normandy and Marne in France, the distribution of cluster proportions at K = 2 showed an almost continuous West–East cline (Figure 1), and closely correlated with the index of continentality (r = 0.5474, p < 0.0001), as well as with several climatic factors associated with climate continentality such as temperature fluctuations, temperature extremes, and degree days (Table S2).

Increasing the number of groups to K = 4 did not bring any improvement of the picture—two clusters predominate (displaying also a longitudinal trend) which the remaining two clusters are irregularly distributed and represented at low frequencies (Figure S2b).

3.2. Climatic Associations

Out of 25,068 SNPs, which passed filtering, 4409 were located in exons, out of which 1704 SNPs quite regularly distributed across the whole genome were retained (Figure 2a). An association study revealed 61 SNPs, which were significantly associated with at least one climatic variable: LFMM identified 55 such SNPs, RDA 11 SNPs, while Tassel did not reveal any (

Table 1. Overview of loci exhibiting consensual and partly consensual associations with climatic variables.

Reference Chromosome	SNP Position	SNP	Climatic Variable	r 1	Annotation [12]
2	42540393	G/A	BIO2	−0.809	>Bhaga_2.g4653 XP_018816688.1 probable aminotransferase TAT2 isoform X1|transferase activity|metabolic process
8	947644	G/A	AHM	−0.695	>Bhaga_8.g118 VVA10054.1 AT2G17540
8	24354137	A/G	bFFP	0.914	>Bhaga_8.g2937 XP_030953202.1 SUPPRESSOR OF ABI3-5 isoform X1|nucleus|RNA binding|metal ion binding|mRNA splicing, via spliceosome
8	36882971	T/A	BIO3	0.682	>Bhaga_8.g4420 XP_030954219.1 ABC transporter B family member 19|integral component of membrane|ATP binding|ATPase activity|ATPase-coupled transmembrane transporter activity|transmembrane transport
6	41874431	C/T	AHM 2 DD > 18 3	−0.695	>Bhaga_6.g4735 XP_030953127.1 pentatricopeptide repeat-containing protein At4g39530-like

¹ Correlation between SNP and a climatic variable; ² LFMM; ³ RDA BIO2—mean diurnal range; AHM—annual annual heat/moisture index; bFFP—the Julian date of the beginning of frost-free period; BIO3—isothermality; and DD > 18—degree days above 18 °C.

and Figure 3). Their distribution was not as regular as in the former case. Only two such SNPs were located on the largest chromosome 1 or chromosomes 10 and 11, while 10 such SNPs were located on chromosome 6 and 9 on chromosome 8 (Figure 2b). Nevertheless, only in four cases was the association confirmed simultaneously by both LFMM and RDA (Table 1). In addition, in one case, an SNP showed significant associations with two different climatic variables as inferred by LFMM and RDA.

4. Discussion

4.1. Choice of Populations and Climatic Proxies

In populations with different initial genetic structures, identical selection pressure can lead to different outcomes. Therefore, our study deliberately focused on populations originating from the same glacial refugium. In contrast to many other tree species, European beech colonized most of its current distribution range from a single source (not necessarily a single refugium, but a single refugial area), localized at the eastern foothills of the Alps [33,51]. The only areas where other refugia formed the present-day genetic structures of beech populations are the Apennine Peninsula and the southern and southeastern Balkans; populations located in these areas were excluded. Magri [51] suggested further minor refugia in the regions covered by our study: the Pyrenees, southwestern France, Moravia, and the Romanian Apuseni Mts. However, these refugia probably did not substantially contribute to the current gene pools. First, none of the studies based on neutral or quasi-neutral markers [33,52,53,54,55,56,57] has indicated such contribution. Second, the expansion of these refugia during the Holocene, as reconstructed by Magri [51], was very limited compared to the main Slovenian refugium. Therefore, provenances from these areas were retained in our dataset.

Concerning the choice of climatic variables, we tried to focus on variables which covered the whole spectrum of climatic adaptation drivers. We included annual means, because such variables are conventionally measured and widely reported, which means that they are available for a dense network of meteorological stations and can be reliably interpolated. On the other hand, tree species very rarely respond to such simple climatic characteristics, as they show just a loose correlation with climatic drivers of physiological stress [58]. However, we realize that seasonal temperature or precipitation averages and climatic indices better reflect those aspects of the climate which operate directly upon the organism or operate through indirect mechanisms. Therefore, we included all variables provided by ClimateEU [34], which included not only annual or seasonal temperature and precipitation averages, but also variables related to vegetative and reproductive phenology such as degree days or boundaries of frost-free period. To account for temperature extremes, we complemented this dataset by selected bioclimatic variables taken from the WorldClim database [35].

4.2. Climatic Associations

In contrast to rangewide allozyme- and nSSR-based studies, which did not indicate any longitudinal trends in the populations originating from the Slovenian refugium [33,54,56,59], the present study revealed a clear longitudinal cline in the Structure cluster proportions. This discrepancy can be attributed to the character of the studied markers. Nuclear microsatellites are commonly treated as typical neutral markers [60]. Allozyme genes are also generally considered neutral [61], although they encode protein molecules and may thus be subject to selection. Their frequencies sometimes correlate with environmental gradients (e.g., [62]), but not necessarily with longitude. The set of SNPs used in this study deliberately focused on polymorphisms in gene exons. Among them, there was a certain proportion of missense mutations with a potential environment-dependent effect on fitness. In the interior of Europe, the gradient of longitude is collinear with climate continentality, which is one of the factors determining beech occurrence [63] and is, in turn, positively correlated with the occurrence and severity of temperature extremes (Table S3). In our case, Structure cluster proportions were correlated primarily with the indicators of low temperatures (mean coldest month temperature, degree days below 0°, and extreme minimum temperature). Because of threats associated with climate change, contemporary research on the genetic basis of adaptation focuses mainly on the effects of drought and heat, and several SNPs responsive to precipitation and temperature gradients have been discovered [25,30,64]. However, climate change is a recent phenomenon, while the current gene pools reflect adaptive processes in the evolutionary past. Winter frosts cause xylem cavitation and loss of hydraulic conductivity. Both physiological mechanisms considerably affect the growth and survival of beech. Consequently, low temperatures have been an important driver of beech distribution [65,66,67].

Climatic variables used in the association study were highly intercorrelated (Table S3), and the use of such variables may inflate associations [68]. To avoid this, only SNPs identified by both LFMM and RDA were considered climate-responsive. The identified SNP climate associations showed a certain discrepancy with the overall pattern identified by Structure. The adaptative capacity to heat and drought stress is considered crucial for the future of beech in a major part of its distribution range [69,70]. Two SNPs were significantly associated with the annual heat-to-moisture deficit as a climatic index related to aridity, but the surrogates of drought such as precipitation totals surprisingly did not show a direct relationship to Structure cluster proportions. On the other hand, both the distribution of Structure groups and several SNPs responded to temperature fluctuations (temperature seasonality and annual temperature range) reflecting temperature extremes. High temperatures are known to induce changes in leaf anatomy and cause the degradation of chlorophyll and damage to the photosynthetic apparatus. The response to heat varies among beech populations, which implies a genetic basis of sensitivity to high temperatures [71,72,73]. Alternatively, response patterns can be mediated by phenology. In addition to the photoperiod, the budburst date is controlled by the accumulation of chilling temperatures in the winter and forcing temperatures in the spring [74]. Both driving factors are related to climate continentality, which results in a clear longitudinal trend of leaf flushing [75], which is collinear with the distribution of Structure clusters. The observed close association of an A/G polymorphism on chromosome 8 with the Julian date of the beginning of the frost-free period yields further support to this hypothesis, as budburst timing is the main mechanism of avoiding late frost damage in beech [76].

Genes with SNPs exhibiting significant associations with climate are involved in a wide range of functions such as enzymatic activities, splicing, or transmembrane transport. However, an assessment of the exact mechanisms underlying these associations is difficult. For instance, the TAT2 tyrosine aminotransferase is known to be involved in the biosynthesis of tocopherol [77], which in turn plays a role in the low-temperature adaptation of Arabidopsis thaliana [78]. However, we do not dare to speculate if the G/A polymorphisms in the TAT2-controlling gene is causally associated with the diurnal temperature range in this way.

5. Conclusions

Climate change is an urgent threat to forest ecosystems, requiring immediate action. Assisted migration is currently the most often suggested mitigation measure [79]. The main information source to guide such transfers is provenance research. Large-scale international provenance experiments are available for most major tree species, where populations of different origins are replicated over several plantation sites and their performance (growth, survival, stem form, and phenology) is tested under different climates. Phenotypic response patterns to climatic transfer (climatic difference between the site of origin and the site of plantation) are then used to choose appropriate FRM for future climates [80]. The problem is that environment-induced phenotypic variation cannot be completely eliminated in provenance experiments, and the traits used to measure performance are typically polygenic. Detecting the physiological and ultimately the genetic basis of these traits can provide a more reliable keystone to guide adaptation measures to climate change.