Postglacial colonization history reflects in the genetic structure of natural populations of Festuca rubra in Europe

Abstract We conducted a large‐scale population genetic survey of genetic diversity of the host grass Festuca rubra s.l., which fitness can be highly dependent on its symbiotic fungus Epichloë festucae, to evaluate genetic variation and population structure across the European range. The 27 studied populations have previously been found to differ in frequencies of occurrence of the symbiotic fungus E. festucae and ploidy levels. As predicted, we found decreased genetic diversity in previously glaciated areas in comparison with nonglaciated regions and discovered three major maternal genetic groups: southern, northeastern, and northwestern Europe. Interestingly, host populations from Greenland were genetically similar to those from the Faroe Islands and Iceland, suggesting gene flow also between those areas. The level of variation among populations within regions is evidently highly dependent on the postglacial colonization history, in particular on the number of independent long‐distance seed colonization events. Yet, also anthropogenic effects may have affected the population structure in F. rubra. We did not observe higher fungal infection rates in grass populations with lower levels of genetic variability. In fact, the fungal infection rates of E. festucae in relation to genetic variability of the host populations varied widely among geographical areas, which indicate differences in population histories due to colonization events and possible costs of systemic fungi in harsh environmental conditions. We found that the plants of different ploidy levels are genetically closely related within geographic areas indicating independent formation of polyploids in different maternal lineages.

Rapid range expansion from the refugia has often been found to be associated with a founder effect and, consequently, decreased genetic diversity and genetic drift (Eidesen et al., 2013;Hewitt, 1999;Petit et al., 2003). Additionally, human-mediated dispersal has potentially affected the genetic structure of several grass species used in the agriculture in Europe, as observed in European forage grass species (Balfourier, Imbert, & Charmet, 2000). As a result of admixture, increased genetic diversity has been detected in contact zones, where different natural or human-induced colonization routes meet (Petit et al., 2003;Taberlet et al., 1998). As a result of multiple introductions from different origins and recombination between divergent genomes, species can form admixed, possibly highly variable populations.
The symbiosis is facultative for the host and fitness effects can vary from antagonistic to mutualistic depending on the prevailing environmental conditions while Epichloë is completely dependent on its host in terms of survival, nutrition, reproduction, and transmission abilities (Clay & Schardl, 2002;Saikkonen et al., 1998). Genetic variability of both the host and fungus is a prerequisite for populations to adapt to changing selection pressures (Gundel et al., 2010). Vertical transmission of Epichloë can form highly specialized maternal lineages, since only one fungal genotype is typically transmitted to the seed progeny. In seeds, cross-fertilization of the host introduces new genetic combinations of host genotypes to Epichloë, and can cause a genetic mismatch between the systemic fungus and the host, if the fertilizing pollen is genetically distant, resulting in incomplete transmission (Gundel et al., 2010;Saikkonen et al., 2010Saikkonen et al., , 2004.
In this study, our main two objectives are to gain further knowledge of genetic diversity patterns and seed flow in F. rubra s.l. occurring in southern and northern Europe, and of the genetic diversity patterns of host populations with differing infection rates of symbiotic fungi. We hypothesized that genetic diversity would be reduced, especially at the edges of the European range due to potential founder effects, bottlenecks, and selection, while greater diversity would be expected near refugial areas and contact zones.
We expect to observe higher fungal infection rates in grass populations with lower levels of genetic variability, and vice versa, because of possible genetic mismatches between the host and the fungus. We aimed at answering the following questions: (a) How is genetic diversity distributed among populations and regions? What is the level of geographic population genetic structuring and how has the postglacial colonization shaped the geographic distribution of haplotypes? (b) Is there association between E. festucae infection rates and the levels of genetic diversity in the host plant populations? (c) Does genetic structure within ploidy levels differ among geographic areas? 2 | ME THODS

| Study system
We used perennial, rhizomatous grass Festuca rubra L., s.l. (red fescue, Poaceae) as a model for estimating genetic variability and population structure across the Europe range to evaluate the genetic resources to create novel plant gene combinations for use of crop varieties. F. rubra is an important turf grass that is widely cultivated in temperate regions (Gould & Shaw, 1983;Hand, Spangenberg, Forster, & Cogan, 2013;Inda et al., 2008). The western Mediterranean F. rubra group is estimated to have diverged during Pleistocene (c. 1.6 MYA; Inda et al., 2008). Inda et al. (2008) suggest in their biogeographical study on Loliinae that F. rubra has experienced postglacial expansion into northern Eurasian latitudes and more recently colonized holarctic areas from Eurasia. Wind pollinated, outcrossing, and self-incompatible F. rubra reproduces sexually by seeds and effectively by asexual vegetative (tillers) and pseudoviviparous propagules (Ahlholm et al., 2002;Dirihan et al., 2016). Clonal tillering is the predominant reproduction strategy facilitating the efficient establishment of a genotype within a habitat, but only a small part of the tillers produce inflorescences yearly (Jónsdóttir, 1991;Saikkonen et al., 2010). There is no marked seedbank of F. rubra (Wäli et al., 2009), and thus, environmental conditions at the time of seed ripening are likely to determine germination rates and the survival of seedlings. The success of F. rubra can be connected with the systemic and vertically transmitted symbiotic fungus, Epichloë festucae Leuchtm., Schardl & Siegel (Clavicipitaceae;Ahlholm et al., 2002;Saikkonen et al., 2010;Wäli et al., 2007;Wäli et al., 2009). E. festucae infections have been observed to be common in F. rubra in dehesa grasslands in Spain (Dirihan et al., 2016;Zabalgogeazcoa et al., 1999) and in northern latitudes in subarctic regions (Dirihan et al., 2016;Wäli et al., 2007). Sexual reproduction of E. festucae (stroma formation, choke disease) has regularly been observed in Spain (Zabalgogeazcoa et al., 1999;pers. comm. Pedro Gundel), but not in the subarctic (Wäli et al., 2007).  (Foggi & Müller, 2009;Inda et al., 2008;Lu, Chen, & Aiken, 2006). The taxonomic identities of the plant samples in the regions are given in Supporting information Appendix S1. The nuclear ITS region (ITS1-5.8S-ITS2) is commonly used for phylogenetic analyses in Loliinae (Inda et al., 2008). The ITS sequence data of the same samples used in this study revealed no sequence differences among the F. rubra subspecies and their hybrids occurring in Finland and northern Atlantic islands, while minor ITS sequence differences were found between F. rubra taxa and F. rothmaleri (Saikkonen et al., in prep). Hereafter, we refer to the studied taxa as F. rubra.

| Plant material
The sampling procedure and E. festucae detection (infected E+/ uninfected E−) has been described in Dirihan et al. (2016). The frequency of endophyte infections (E+) was 44.9% in our sample set (N = 566). The percentages of the endophyte infections varied from 0% to 82.5% among regions and from 0% to 95.8% among populations (Supporting information Appendix S1; see also Dirihan et al., 2016 for their larger sample set). The frequency of endophyte infections was 82.5%, 73.1%, 55.4%, and 38.2% in Spain, Kevo in northern Fennoscandia, Iceland, and the Faroe Islands, respectively. No infections were observed in southern (Hanko) and northern (Kilpisjärvi) Fennoscandia, and only one plant was infected in Greenland and Norway each. In addition to variation in E. festucae infections, the studied populations were found to include plants with different ploidy levels, and the proportions of tetraploids (2n = 4x = 28), hexaploids (2n = 6x = 42), and octoploids (2n = 8x = 56) were 12.7, 82.2, and 5.1%, respectively, in our data set (N = 566; Supporting information Appendix S1; see also Dirihan et al., 2016 for their larger sample set).
Genomic DNA of F. rubra was extracted from fresh leaves of F. Drawbacks of cpSSR markers include primarily size homology; however, it is not a significant problem at the intraspecific level (Ebert & Peakall, 2009;Estoup, Jarne, & Cornuet, 2002;Navascués & Emerson, 2005;Provan et al., 2001). The forward primers of each cpSSR primer pair were end-labeled with two different phosphoramidite fluorescent dyes, either 6-FAM or HEX. The samples were analyzed by multiplexing markers with different labels and expected fragment sizes (2-3 samples per PCR reaction). All primer pairs produced bands that matched the expected sizes. PCR amplifications were performed, as described in von Cräutlein et al. (2014). Each genotyping plate included individuals from several populations, and negative and positive controls. In the case of rare alleles, PCR amplifications were repeated to ensure the existence of rare alleles.
The PCR products were run on an ABI 3130xl DNA Sequencer using the GeneScan 500 ROX Size Standard (Applied Biosystems) at the Institute of Biotechnology, University of Helsinki, Finland. The amplified fragment lengths were assigned to allelic sizes at the accuracy of one base pair with Peak Scanner version 1 software (Applied Biosystems). We also calculated these statistics separately for plants with different ploidy levels (N = 566, 2n = 28, 42, 56). The percentage of polymorphic loci (%P), the mean number of observed alleles over loci (Na), the mean effective number of alleles over loci (N e ), the mean number of unique alleles over loci (N p ), and the unbiased haploid genetic diversity over loci (uh) were computed using GenAlEx 6.5 software (Peakall & Smouse, 2006. We also tested for significance of differences in allelic diversity estimates (N e , uh) of the loci (altogether 13 cpSSR loci) between regions, BAPS clusters, and endophyte infection status groups (infected E+/uninfected E−) in four regions (Faroe Islands, Iceland, Kevo, Spain) with sufficient number of E+ plants (see Table 1). These tests were conducted by employing related samples nonparametric Friedman's two-way analysis of variance by ranks or related samples Wilcoxon signed rank nonparametric test for two groups (Sokal & Rohlf, 1995;Wilcoxon, 1945), arranged for paired ob- The genetic relationships of individuals were assessed using a model-based approach using the Bayesian Analysis of Population Structure (BAPS) software, version 6.0 (Corander, Cheng, Marttinen, Sirén, & Tang, 2013;Corander & Tang, 2007), by applying a nonspatial genetic mixture analysis with known populations to the cpSSRs results of the whole sample set, with the clustering of linked loci option (Corander & Tang, 2007). Allele frequencies and the number of genetically diverged groups in a population were treated as random variables. To determine the most probable A hierarchical analysis of molecular variance (AMOVA) was used to estimate the degree of genetic differentiation among regions, among populations, and among the six clusters (K = 6) obtained by BAPS program by using Arlequin software version 3.5 (Excoffier & Lischer, 2010), which estimates genetic structure indices using information on the allelic content of haplotypes and allele frequencies (Excoffier, Smouse, & Quattro, 1992 (Weir & Cockerham, 1984) by using Arlequin software version 3.5 (Excoffier & Lischer, 2010). F ST is unbiased with respect to sample size, and it adjusts allele frequency estimates with respect to sample sizes (Weir & Cockerham, 1984; see Whitlock, 2011). The significance of the fixation indices was computed with 10,100 nonparametric permutations. Note. Population and sample information are given in Supporting information Appendix S1. N, sample size; %P, percentage of polymorphic loci; N e , mean effective number of alleles over all loci; N p , mean number of private alleles over all loci; uh, unbiased haploid genetic diversity over all loci; MLG, number of unique multilocus haplotypes; eMLG, number of estimated unique multilocus haplotypes.

| Statistical methods
TA B L E 1 Allelic diversity estimates and numbers of unique haplotypes in Festuca rubra across eight regions and six BAPS clusters based on 13 cpSSR loci; BAPS clusters: S1 = Spain1-group, S2 = Spain2-group, F = Fennoscandia-group, A = Atlantic-group

| Allelic diversities and numbers of multilocus haplotypes
A total of 238 cpDNA multilocus haplotypes (MLG) were identified in the sample set (N = 588) based on the 13 combined cpDNA microsatellite loci. Among regions, the different haplotype numbers in relation to sample sizes (eMLG) and allelic diversity estimates (N e and uh) were clearly highest in Spain and lowest in Greenland, but relatively high genetic diversities were also observed in the Faroe   Fennoscandia-group (F) and the other clusters, and lowest between Spain1-and Spain2-groups, which co-occurred abundantly in the same populations (Supporting information Appendix S5).
Significant differences were detected in the allelic diversity estimates (N e , uh) among the six clusters (Table 1).

The largest clusters
Atlantic-group and Fennoscandia-group possessed clearly the lowest levels of allelic diversity and numbers of different haplotypes, whereas considerably higher levels of allelic diversity and numbers of different haplotypes were found in both Spain-groups. However, the highest levels of genetic diversity and private allele numbers were found in the smallest clusters A-F and F-A.
F I G U R E 3 A gene flow network identified for the six clusters (K = 6) as obtained by BAPS in Festuca rubra. Gene flow is shown by weighted arrows, which indicate relative average amounts of ancestry coming from the source cluster but present now among individuals assigned to the target cluster. Cluster S1 = Spain1-group; cluster S2 = Spain2-group; cluster F = Fennoscandia-group; cluster A = Atlantic-group. Estimated ancestral admixture (gene flow) is indicated by weighted arrows (Tang, Hanage, Fraser, & Corander, 2009) The results of the admixture analysis shown as a gene flow network of six clusters (K = 6) are summarized in Figure 3. Intercluster ancestral gene flow varied from 0.02% to 3.3% (Figure 2).

| Genetic differentiation between regions and populations
The AMOVA analysis showed that 23.5% of genetic variation lies among regions, 10.6% among populations within regions and the remaining two-thirds (65.9%) within populations (Table 2) In the northern Atlantic area, no significant genetic differentiation was detected among the three islands, which indicates effective gene flow among the Faroe Islands, Iceland, and Greenland (Table 2)

| Genetic structure and endophyte infection rates
The

| Genetic structure and ploidy levels
Among the six clusters obtained by BAPS (K = 6), all three ploidy levels (2n = 28, 42, 56) occurred in Atlantic-, Fennoscandia-, and similarly as it has been found for several other plant and animal species in Europe (Fjellheim, Rognli, Fosnes, & Brochmann, 2006;Hewitt, 1996Hewitt, , 1999Hewitt, , 2000Hewitt, , 2004Keppel et al., 2012). The gene flow plot produced by BAPS and low levels of differentiation among the Spanish populations suggest that the gene pools of Spain 1-and Most haplotypes originating from the Northern Atlantic islands fell into the highly differentiated Atlantic-group, which showed decreasing amounts of genetic diversity toward north. Rapid range expansion from refugia to previously glaciated northern areas has often been found to be associated with a decreased genetic diversity in the north (Eidesen et al., 2013;Hewitt, 1999 (Brochmann et al., 2003;Jimenez-Mejías et al., 2012;Schönswetter, Elven, & Brochmann, 2008;, although the Atlantic ocean has also been seen as an impenetrable barrier to seed dispersal (Eidesen et al., 2013;Hultén, 1958). Interestingly, even though plants from eastern and western Greenland have often been found to be genetically different as a result of the Greenlandic ice cap barrier for gene flow (Alsos, Torbjorn, Normand, & Brochmann, 2009;Eidesen et al., 2013), our cpDNA data show that seed dispersal has occurred between western Greenland and northern Atlantic islands. The extremely low genetic diversity level observed in Greenland is likely due to founder effect and genetic drift and very rare occurrences of colonizations to the island. However, plant dispersal from ice free areas in eastern or western Greenland has been suggested to be considerable (Eidesen et al., 2013). Yet, based on our results on F. rubra, this is unlikely, since almost no private alleles and unique haplotypes were found in the Greenlandic populations. It is also possible that long-distance dispersal events have occurred from northern America, which can be resolved in future studies with circumpolar sampling.  (Schönswetter et al., 2008). In F. rubra, based on the gene flow plot obtained by BAPS, the Atlanticgroup has received genetic material in small quantities from the Fennoscandia-group and Spain-2-group, which suggests that the glacial origin of the Atlantic-group might be in some southern European peninsula, similarly as detected in C. bigelowii (Schönswetter et al., 2008) and Ranunculus glacialis L. (Schönswetter, Paun, Tribsch, & Niklfeld, 2003), but further sampling from different potential refugial areas is needed to confirm that. The accumulation of different genetic groups in the northern Atlantic islands may be a consequence of natural long-distance seed dispersal from other sources not studied here or a consequence of anthropogenic seed dispersal through seed trade and movement of farm animals and cattle feed between the continent and islands (see Linder et al., 2018). Humaninduced seed dispersal is also suggested to be a substantial factor to shape the population genetic structure of forage and turf grass L.
perenne in Europe (McGrath et al., 2007). Moreover, the occurrence of the distinct and genetically diverse cluster A-F that contributes to increased genetic diversity in the Faroe Islands and Iceland may indicate periglacial survival in ice free areas in the Faroe Islands, as suggested in many studies (Brochmann et al., 2003;Jimenez-Mejías et al., 2012;Westergaard, Alsos, Popp et al., 2011).  (Brochmann et al., 2004). Such effect might be stronger in the RIS2 population with a predominance of octoploids than in RIS1 with a prevalence of hexaploids. Similarly, among other arctic polyploid plants, most genetic variation has been found as fixed heterozygosity and as variation among different populations (Brochmann et al., 2004). In contrast, the subarctic populations RBS3 and RIS3 were found to be diverged from all other populations because of abundant occurrences of rare and genetically diverse haplotypes of F-A cluster, which has increased the amount of genetic diversity and numbers of unique haplotypes.
One explanation for the genetic distinctiveness of the F-A cluster is the glacial survival in nunataks or in other in situ glacial refugia in Norway (Eidesen et al., 2013;Brochmann et al., 2003, and references therein). There is fossil and molecular evidence that supports the idea of refugia in Scandinavia for Sagina caespitosa (Westergaard, Alsos, Popp et al., 2011) and conifer trees (Parducci et al., 2012). However, representatives of the F-A cluster can be found at low frequencies also in the Atlantic islands, which indicate long-distance seed dispersal among the northern areas. Thus, the origin of this rare cluster remains unclear. Moreover, we cannot exclude the possibility of an anthropogenic origin for F-A cluster haplotypes, as cultivar haplotypes might have dispersed to natural habitats in northern Finland from seed mixtures sown along roadsides (Wäli et al., 2007).

| E. festucae infection rates versus population genetic structures
In   Saikkonen et al., 2010Saikkonen et al., , 2004. This view is supported by an observation that only little genetic variation is found in the subarctic E. festucae populations in Kevo, thus indicating that only few fungus genotypes are mainly vertically transmitted from maternal plants to offspring in the region (Wäli et al., 2007; see also Sullivan & Faeth, 2004;Zhang, Ren, Ci, & Gao, 2010). Moreover, wind dispersed pollen is likely to generate new genotypes in the populations of F. rubra, causing asymmetric rates of gene flow and interactions of incompatible genotypes, as suggested for Neotyphodium and its host Festuca arizona by Sullivan and Faeth (2004). Lower diversities of symbiotic fungi compared to the host diversities have been observed in the natural populations of Neotyphodium species and their host Achnatherum, which indicates more restricted gene flow distances in fungi than in their plant hosts (Zhang et al., 2010). Seeds produced by outcrossing of genetically distant parents can cause incompatibilities between the fungus and the grass host, and consequently, the vertical transmission of the fungus with seeds would fail (Gundel et al., 2010;Saikkonen et al., 2010 Gundel et al. (2010), possible losses of infections in populations located in southern Fennoscandia may be caused by genetic mismatches between the host and fungus, since, in that region, two genetically highly distant genetic groups, Fennoscandia-and Spanishgroups occur abundantly (see also Saikkonen et al., 2010Saikkonen et al., , 2004. In contrast in the populations in Kilpisjärvi (RIS1, RIS2) and Greenland, the possible losses of infections may be due to the observed extremely low genetic variability among the plants causing material or energetic costs to maintain endophytes under stressful environmental conditions when seed production can be lost by fitness-depressed host plants (Gundel et al., 2010). Nevertheless, lack of infections can also be caused by introductions of noninfected seed material into populations without fungal gene flow. Interestingly, only one infected plant was observed in Greenland, the haplotype belonging to Spain2group, whereas all other haplotypes belonged to genetically very distant Atlantic-group, which may indicate its recent introduction.

| Ploidy levels versus population genetic structures
In F. rubra, the frequencies of plants with higher ploidy levels were clearly greater in Atlantic-and Fennoscandia-groups than in Spaingroups, which may suggest the selective advantage of higher ploidy levels in the genetically poorer populations in the north. Sampoux and Huyghe, (2009) have proposed for the fine leaved fescue lineages that cytotypes with high ploidy levels might have been efficient colonizers and competitors. Therefore, such lineages would have expanded more effectively in the northern areas. Polyploidy can facilitate a species' competitive ability and ability to colonize novel environments. They may have a broader ecological range than their diploid relatives, especially in northern latitudes (Brochmann et al., 2004;Linder & Barker, 2014). In F. rubra, almost all distinct ploidy levels (tetra-, hexa-and/or octoploids) occur among Atlantic-, Fennoscandia-, and Spain1-and Spain2-groups, which shows that plants of different ploidy levels are closely related which suggests independent formation of polyploids in different maternal lineages in different geographic areas. Similarly in Dupontia (Poaceae), several different ploidy levels were detected in three diverged genetic groups in a circumarctic area (Brysting, Fay, Leitch, & Aiken, 2004).

| CON CLUS IONS
This is the first comprehensive study that illustrates the genetic structure and geographic differentiation over a wide area in natural populations of F. rubra differing in the infection rate of the symbiotic fungus E. festucae and ploidy level. It shows that the level of genetic variation in different geographic regions is evidently highly dependent on the postglacial colonization history.
The relationship between the host and the symbiotic fungus in the sense of genetic variability is not consistent but appears to dif-

CO N FLI C T O F I NTE R E S T
None declared.