The plants used in this study were originally collected as a part of research examining the occurrence and ecological importance of Epichloë festucae in wild populations of Festuca rubra L. s.l. across Europe in 2011 (Dirihan et al., 2016; Leinonen et al., 2019; von Cräutlein et al., 2019; Saikkonen et al., 2020; Vázquez de Aldana et al., 2020). The initial plants were split and a copy of each individual was maintained in pots with a mixture of peat and sand in the greenhouses at the Ruisalo Botanical Garden of Turku University. Splitting grass plants is a common way of generating identical genetic copies of both the host and the endophyte as the endophyte systemically infects the above ground plant tissue. The initial plants were tested for endophyte infection with methods described in Dirihan et al. (2016). In this study, we focused on three European regions of the host distribution extreme range with the highest occurrence of E. festucae infections (Dirihan et al., 2016). We examined a total of 240 individual Epichloë-infected plants originating from 15 natural populations from six islands in the Faroe Islands (n = 71), from six populations located in two different habitat types, meadows (n = 73) and riverbanks (n = 30), in Finland, and from three populations located in two different habitat types, Mediterranean oak forest (n = 22) and semiarid oak grassland (n = 45), in Spain (n = 66). More detailed information on the plants, occurrence of associated Epichloë-fungus and collection sites, including population geographic locations, coordinates, altitudes, habitat features and estimate of grazing intensity, can be found in Table 1, Dirihan et al. (2016); von Cräutlein et al. (2019). The frequencies of Epichloë- individuals in the studied populations ranged from 5% to 81% and the number of individual isolates varied accordingly (Table 1; Dirihan et al., 2016).

Epichloë festucae was isolated from the plants in 2013 and 2014. Three leaves from each tiller were selected from pots and tillers were surface sterilized. A leaf was cut in five segments and inoculated on autoclaved Petri dishes containing 5% potato dextrose agar (PDA). Plates were stored at room temperature until mycelium emerged, after which a small sample of mycelium were transferred to a new PDA plate on a piece of sterilized cellophane. Total genomic DNA was extracted from pure cultures of mycelium growth using the E.Z.N.A Plant DNA Kit (Omega Bio-Tek, Norcross, GA, United States) according to the procedures described in von Cräutlein et al. (2014). The same DNA samples were used for the analyses based on the SSR mating type and alkaloid gene markers.

Genetic structure and the amount of clonality in the 240 E. festucae isolates were investigated using 14 polymorphic SSR markers, which were developed based on the searches for ≥10 mono- and dinucleotide repeats, and for ≥8 tri-, tetra-, penta-, and hexanucleotide repeats in the unplaced genomic scaffold sequences of E. festucae (for section “Materials and Methods” see von Cräutlein et al., 2014). The forward primers of each SSR primer pair were end-labeled with two different phosphoramidite fluorescent dyes, either HEX or 6-FAM. The samples were analyzed by multiplexing markers (2-4 primer pairs/reaction) with different labels and expected fragment sizes. Allele sizes ranged from 92 to 340 bp depending on the primer pairs (Supplementary Table 1). The details of PCR amplifications are described in von Cräutlein et al. (2014). Each genotyping plate included negative and positive controls and samples from several populations. The PCR products were run on an ABI 3130xl DNA Sequencer using GeneScan 500 ROX Size standard (Applied Biosystems) at the Institute of Biotechnology, University of Helsinki, Finland. Peak Scanner version 1 software were used (Applied Biosystems) to assign the allelic sizes of the amplified fragments. The detailed information on SSR markers, including e.g., names and locations of markers in the E. festucae chromosomes, are described in Supplementary Table 1 and whole SSR data set with host plant IDs in Supplementary Table 2. As E. festucae is haploid, the samples were expected, and did produce one allele per locus. However, in the rare exception where multiple alleles were observed in at least one SSR locus of Faroe Islands (five isolates), Finland (six isolates), and Spain (six isolates), the samples were not included in the study.

The genetic loci involved in alkaloid biosynthetic pathways essential for the production of ergot alkaloids (EAS), indole-diterpenes (IDT) and lolines (LOL) are complex gene clusters in Epichloë taxa, whereas the pyrrolopyrazine alkaloids (PPZ, previously referred as PER) production is dependent on the alleles of the perA gene (Tanaka et al., 2005; Schardl et al., 2012, 2013b; Berry et al., 2015, 2019). The genes encoding different alkaloid classes have recently been identified (Schardl et al., 2013b) allowing us to predict alkaloid production based on presence or absence of a key alkaloid genes within the pathway (Takach et al., 2012; Schardl et al., 2013b,c; Charlton et al., 2014; Takach and Young, 2014; Berry et al., 2015; Shymanovich et al., 2015; Vikuk et al., 2019). These genes are upregulated in planta (Young et al., 2006, 2015; Chujo and Scott, 2014) and the gene clusters are devoid of known pathway specific regulatory genes, unlike other fungal secondary metabolite clusters that often contain a gene that encodes a regulatory function in the form of pathway specific transcription factors (e.g., aflR required for aflatoxin production; Woloshuk et al., 1994).

The presence of selected key genes from the loci for alkaloid production and mating type idiomorphs (genes mtAC and mtBA) were examined in a total of 198 E. festucae isolates originating from six populations in the Faroe Islands (n = 60), six populations in Finland (n = 91) and three populations in Spain (n = 46) (Table 1).

A multiplex PCR method was used to determine the mating type (A or B) and key genes present at each alkaloid loci (Charlton et al., 2014). The primers including two additional primers for IDT genes (idtK and idtF) used for mating type and alkaloid gene profiling, expected product sizes and six different multiplex sets are described in Supplementary Material in Charlton et al. (2014). The PCR amplification methods are described in Charlton et al. (2014). PCR products were analyzed by gel electrophoresis on a 1.5% agarose gel and visualized with ethidium bromide by UV transillumination. The combination of two samples, MS2K-35 and SPGD-31 was positive for all examined alkaloid genes and used as a positive control on each PCR plate. The presence of a nonsense mutation and inframe stop codon in the first exon of idtF gene were determined in isolates that contained the IDT genes required for terpendole C or lolitrem B productions. The primers idtF-M-F (5′-GGGCCATCCTATCTTACAC-3′) and idtF-M-R (5′-ACGAAGCCTTGAATCCAC-3′) were designed based on the idtF gene sequences with and without the mutation (GenBank accession numbers: EU530694 and MF464362). The PCR product of each alkaloid gene locus was sequenced using the following methods: the PCR products were separated on a 1.2% agarose gel, extracted from the gel and purified with E.Z.N.A. Gel Extraction Kit (Omega, Bio-Tek). The purified PCR products were submitted to Macrogen Inc., for Sanger sequencing with both upstream and downstream primers. The obtained sequences were visualized and manually corrected using Chromas version 2.6.5 (2018). The sequence similarity searches were performed in GenBank using BLAST sequence analysis tool (NCBI). All sequences showed the closest match (100% similarity) with E. festucae and the alkaloid gene region in question. The sequences were submitted to the European Nucleotide Archive. Accession numbers are available in Supplementary Table 3 for the positive controls of the alkaloid genes and for the presence or absence of deletion in the idtF gene in the set of samples.

The presence or absence of a set of key alkaloid genes within the alkaloid class pathway detected here are described in Charlton et al. (2014). Ergovaline was expected to be produced, if five examined EAS genes (dmaW, easC, easA, cloA, and lpsB) produced PCR bands of the expected size, and chanoclavine was expected if only dmaW and easC were present. Peramine, a pyrrolpyrazine-1-one, was assumed to be produced, if all three markers including perA5′, perAT2 and perAR, produced PCR bands, and if the reductase domain (perAR) was absent, then pyrrolpyrazine-1, 4-dione (PPZ-1 diones) were expected (Berry et al., 2019). Ergovaline and peramine chemotypes of some of the F. rubra – E. festucae symbionts in the present study (n = 27) determined by Vázquez de Aldana et al. (2020) were compared with the alkaloid gene profiles obtained in this study to confirm prediction of the alkaloid production. The first stable indole-diterpene intermediate, paspaline, is predicted if idtG and idtQ are both present. Isolates that can produce early pathway terpendoles, such as terpendoles E and I, also contain a functional idtF, and isolates that can produce late pathway terpendoles, such as terpendole C, contain idtF and idtK in addition to idtG and idtQ. Lolitrem B (LTB), the end product of IDT biosynthesis in E. festucae, was predicted to be produced, if in addition to the genes mentioned above also idtJ produced a PCR band. The prerequisite of the production of terpendole C and lolitrem B was also that the sequence of idtF-gene was functional, without the deletion in the first exon of the gene that causes an inframe stop codon (Young et al., 2009; Shi et al., 2017; Yi et al., 2018). In addition, presence of idtP gene was checked from our unpublished data in 109 of the F. rubra – E. festucae combinations in the present study. The host samples used in alkaloid gene, idtF mutation and idtP-gene detections are provided in Supplementary Table 4. Lolines were predicted to be produced, if examined LOL genes (lolC, lolA, lolO, and lolP) produced PCR bands.

The number of multilocus alkaloid gene genotypes (aMLG) based on the presence (1) or absence (0) of each key alkaloid gene was determined using haploid binary data and multilocus options within populations, across populations within regions and across the whole data set. For the alkaloid gene data set, pairwise PHi_PT values were used to estimate population pairwise differentiation levels within regions, two populations, FAS4 and RBS3, were excluded from the analysis, because of low numbers (<5) of isolates per population using the GenAlex version 6.5. (Peakall and Smouse, 2006, 2012).

The number of multilocus SSR genotypes (nMLG), the number of expected genotypes based on rarefaction (eMLG) and random association among loci (indices of associations: I_A and r_d; clone corrected data, 999 permutations) were computed for each population, across Finnish, Faroese and Spanish populations within regions and across the whole data set. Tests of random association among loci were not performed for the Faroese and Finnish populations due to lack of statistical power because of a low sample size after clone correction (Fincham and Day, 1963). Minimum spanning networks (MSN) using Bruvo’s distance were computed across all populations (Bruvo et al., 2004). The analyses were computed with R 3.5.1 (R Core Team, 2016) package poppr (v2.8.5.; Kamvar et al., 2014).

To study the genetic relationships of isolates, a Bayesian Analysis of Population Structure (BAPS) software, version 6.0, was used by applying a non-spatial mixture clustering analysis at individual level (sampling unit) with linked loci option (Corander and Tang, 2007; Corander et al., 2008) using multilocus SSR data set including all the isolates in order to represent the distribution of allele frequencies in this randomized sample set (n = 240). The partition of optimal K numbers, which refer to number of groups into which the SSR data can be clustered, was conducted by performing 150 iterations of K from 2 to 30, which resulted in the number of genetically diverged clusters in optimal partition to be 15 [log(marginal likelihood) value = −1901,6]. Fixed K model was used, because BAPS identified several small clusters (nine clusters with an average of 2.7 individuals). The number of clusters for the fixed K model was determined based on the uppermost hierarchical levels of genetic structure shown in a UPGMA tree based on the Kullback-Leibler divergence matrix. The individual level mixture clustering analysis with linked option was conducted using the fixed K mode with 150 iterations of K = 4.

Genetic diversity indices of the SSR were calculated with two data sets by using all isolates (n = 240) and using all unique MLGs within regions (n = 103) with haploid data option and are based on the numbers of allele frequencies at each locus. The percentage of polymorphic loci (P%), the average effective numbers of alleles (Ne), the average numbers of unique alleles (Np) and unbiased genetic diversity (uh) estimates were calculated for each locus (n = 14), over the entire sample set (n = 240), over the MLGs (n = 103), for each region (n = 3), for each population (n = 15) and for each genetic group obtained by BAPS analysis (K = 4). A principal coordinate analysis (PCoA) was used to plot the major patterns in SSR data sets based on the whole data set (n = 240) and separately for each region (Faroe, Finland, and Spain) without clone corrections in relation to the genetic similarities of the isolates using pairwise individual-by-individual haploid genetic distance matrixes. Tests for significance were run with 999 random permutations. Analyses were conducted using GenAlex v. 6.5.

A hierarchical analysis of molecular variance (AMOVA; Weir and Cockerham, 1984; Excoffier et al., 1992; Weir, 1996) was used to estimate the degree of differentiation among regions and populations, and pairwise Fst values were used to estimate population differentiation levels among populations, regions, habitats within region and genetic clusters (K = 4) in SSR data set (n = 240, all isolates included). Moreover, the degree of differentiation among regions were also estimated with the data set with unique MLGs (n = 103). The analysis was conducted using Arlequin software, version 3.5 (Excoffier and Lischer, 2010). The significance of the fixation indices was run with 999 non-parametric permutations.

The Bayesian mixture analysis was conducted using fixed K mode of K = 4 [log (marginal likelihood) value = −2194.4]. At K = 4, the isolates are mainly distributed according to the geographical regions (Figure 1). Fin-group (n = 104) represents mainly Finnish isolates (n = 101) and three isolates from the Faroe Islands (FAS3). Far-group (n = 69) contains mainly Faroese isolates (n = 67) and two isolates from Finnish populations (KS3, RBS1). The Spanish isolates were partitioned in one large and one smaller genetic group. Sp1-group includes fourteen isolates from SPGD population. Sp2-group (n = 53) represents all isolates from SPLV (n = 23) and SPPOR (n = 21) but also eight isolates from SPGD and one isolate from the Faroe Islands (FAS6).

Consistent with the results of Bayesian mixture analysis, a clear geographic pattern of three groups was detected among all the Finnish, Faroese and Spanish isolates (n = 240) according to the principal coordinate analyses based on pairwise individual-by-individual haploid genetic distance matrixes, in which the first and second axes explained 44.5% and 18.6% of the variation, respectively (Figure 2A). In Faroe Islands, three individuals differed from the main group, which was highly mixed with individuals from different populations, the first and second axes explained 35.4% and 16.7.% of the variation, respectively (Figure 2B). In Finland, two individuals differed from the main group, which was also mixed with individuals from different populations, the first and second axes explained 36.2% and 27.7% of the variation, respectively (Figure 2C). In Spain, individuals were distributed in one small and one large group and individuals of the larger groups were moderately mixed, the first and second axes explained 19.3% and 13.2% of the variation, respectively (Figure 2D).

Relatively low levels of genetic variation indices and numbers of private alleles were detected in most of the Finnish and Faroese populations, although the levels of genetic variation differed widely among the populations (Table 2). In contrast, relatively high levels of genetic variation indices and high numbers of private alleles were found in all Spanish populations (Table 2). Accordingly to the regions, genetic diversity estimates differed between the geographic locations of genetic groups (BAPS clusters) being highest in Spanish Sp2-group and clearly lower in Far- and Fin-groups (Table 2).

Overall using all the isolates (n = 240), the AMOVA analysis estimated that 66.6% of the genetic variation occurs among regions, 36.1% within populations and only 7.3% among populations within regions indicating high differentiation among the regions and low differentiation among the populations (Table 3). Based on the data set of unique MLGs (n = 103) for the regions, 44% of the genetic variation occurs among populations and 56% within populations (Table 3).

In the Faroe Islands (using all the Faroese isolates), most of the variation (73.1%) was detected within populations and moderate levels of genetic differentiation (26.9%) among the six populations (Table 3). The population pairwise Fst values varied from 0.004 to 0.447 being on average 0.25 ± 0.14 and significant differentiation levels were found among 67% of the population pairs (p < 0.05; n = 10; Supplementary Table 5A).

In Finland (using all the Finnish isolates), majority of the variation (95.5%) was within populations and only low level (4.5%) genetic differentiation was found among the six populations (Table 3). No significant differentiation was detected between the populations located in meadow (n = 3) and riverbank (n = 3) habitats (Fst = 0.067, p = 0.175, see Table 1). The population pairwise Fst values varied from 0.011 to 0.202 being on average 0.065 ± 0.060 and 20% of population pairs showed significant differentiation between populations (p < 0.05; n = 3; Supplementary Table 5A).

In Spain (using all the Spanish isolates), most of the variation (73.6%) was detected within populations and moderate level genetic differentiation (26.4%) observed among the three populations (Table 3). A significant differentiation was detected between the populations located in Mediterranean forest (SPGD) and dehesa grassland (SPLV and SPPOR) habitats (Fst = 0.295, p < 0.001, see Table 1). The population pairwise Fst values are shown in Supplementary Table 5A.

Among the genetic groups (K = 4, including all the isolates, n = 240), AMOVA analysis revealed that 75% of the genetic variation is distributed among the genetic groups and 25% within clusters indicating very high genetic differentiation among the groups (Table 3). Using the data set of unique MLGs for the region (n = 103), 53.3% of the genetic variation occurs among populations and 46.7% within populations (Table 3). The pairwise Fst values of genetic groups are shown in the data sets with all the isolates and with MLGs in Supplementary Table 5B.

The clonal structure of the isolates according to populations within regions is visualized in Figure 3. In total, 103 fungal multilocus nuclear genotypes (nMLG) were observed in the SSR data set (n = 240) in fifteen populations collected from Spain, the Faroe Islands and Finland (Table 2). High frequencies (92.4%) of unique nMLGs and eMLGs occurred in the Spanish populations and only four common genotypes were present in two populations, of which three nMLGs in Mediterranean oak population (SPGD) (Table 2). In the Faroe Islands, fifteen (65.2%) nMLGs were unique in the region and eight common nMLGs consist of 78.9% of the isolates (n = 56) located in different Faroese populations, three largest clone sizes are 22,11 and 8 isolates/nMLG. In Finland, fifteen (78.9%) nMLGs were unique in the region and four common nMLGs consist of 85.4% of the isolates (n = 88) located in different Finnish populations, two largest clone sizes are 64 and 19 isolates/nMLG.

The indices of associations (I_A and r_d) were utilized to estimate the relative importance of reproduction types, i.e., either sexual or asexual reproduction predominated in the regions (Kamvar et al., 2014). Significant linkage disequilibrium was estimated in Finland, the Faroe Islands and Spain and separately in each Spanish population suggesting linked loci and predominance of asexual reproduction in all three regions, although indices of association were clearly lower in Spain than in Finland and the Faroe Islands (Table 2).

The frequencies of mating types according to the populations and genetic groups (K = 4) are shown in Figure 1 and Table 4. The ratio of mating types A and B was 0.92, therefore the proportions of each mating type are close to equal among populations in Spain. In SPLV and SPPOR populations most of the isolates were of mating type B, whereas most of SPGD isolates were of mating type A. In contrast, majority of the isolates in Finland and the Faroe Islands were of mating type A, although both mating types were present in half of the populations in both Finland and Faroe Islands. Mating type genes were not detected in four of the isolates. Based on the genetic groups (K = 4) Fin-, Far-, and Sp2 -groups shared the mating types A and B, but only mating type A occurred in Sp1-group (Table 4).

The alkaloid gene profiles of different aMLGs with mating types, isolate numbers and predicted and confirmed chemotypes are presented in Table 5 and host information and alkaloid profiles of the E. festucae in Supplementary Table 6. Overall, 38 multilocus alkaloid genotypes (aMLGs) and 20 unique chemotypes were identified among the Spanish, Faroese and Finnish E. festucae isolates based on the seventeen genetic loci or alleles associated to ergot alkaloid (EAS), indole-diterpene (IDT), pyrrolpyrazine (PPZ), and loline (LOL) production (n = 198; Figure 1 and Table 5). Common to all isolates was the absence of genes required for the biosynthesis of lolines. Most of the isolates (94,4%, n = 187) are predicted to produce at least one end product of EAS, PPZ and/or IDT alkaloid class. Most (90.4%, n = 179) of the isolates had all the EAS genes targeted, and thus, they are likely to have a functional EAS pathway and are predicted to produce ergovaline. About one fifth (16.7%, n = 33) of the isolates contained all PER markers and they are expected to produce peramine, the pyrrolopyrazine-1-one PPZ-1-ones (Berry et al., 2019). Isolates (75.8%, n = 150) that lack the reductase domain, are predicted to make pyrrolopyrazine-1, 4-diones (PPZA2). Chemotypically determined ergovaline and peramine production in the same F. rubra-E. festucae-symbiosis revealed that 72.7% (n = 11) and 100% (n = 10) of the isolates corresponded with the predicted alkaloid profiles, respectively (Table 5; Vázquez de Aldana et al., 2020). Three isolates contained all functional IDT genes and, thus, they are predicted to produce lolitrem B. The earlier IDT pathway intermediates paspaline, terpendoles I and C are predicted to be produced by 1.5% (n = 3), 43.9% (n = 87), and 8.6% (n = 17) of the isolates, respectively.

The numbers of isolates within the aMLGs differed largely from unique genotypes (one isolate/aMLG) to common genotypes (2-71 isolates/aMLG) (Table 5). Most of the aMLGs (81.6%) were only located in one region. Altogether, 31.6%, 36.8%, and 50% of the aMLGs were observed in the Faroe Islands, Finland and Spain, respectively.

In Spain, 40.4% of the isolates possessed different aMLGs (n = 19) and considerably higher amounts of predicted chemotypes were detected in three Spanish populations compared to northern populations (Figure 1 and Table 5). The observed 13 different chemotypes are predicted to produce different combinations of final products of ergovaline or lolitrem B and their early pathway products and peramine or pyrrolopyrazine-1, 4-diones (Figure 1 and Table 5). All isolates are expected to be toxic to mammals due to presence of all EAS and/or IDT genes, except one isolate (gt-21) (Table 5). In contrast, the frequency of PER and PPZA-2 chemotypes that potentially deter insect feeding differ among the populations. Majority of the SPLV (88.2%) and SPPOR (85.7%) isolates can be expected to produce peramine (PER), which is the case of 37.5% in SPGD isolates (Table 5). However, 62.5% of SPGD isolates are expected to produce pyrrolpyrazine-1, 4 diones (PPZA-2), which is the case of one isolate in both SPLV and SPPOR. The most common aMLG, gt-40 (n = 15), contained about one third (31.9%) of the Spanish isolates and occurred in all populations (SPGD n = 2; SPLV n = 10; SPPOR n = 3) (Table 5). Gt-40 was positive for twelve out of thirteen alkaloid markers and is predicted and confirmed to produce ergovaline and peramine and is predicted to produce terpendoles (Table 5). Significant pairwise differentiation levels based on the occurrence of the alkaloid genes were estimated between the Mediterranean oak forest (SPGD) and both dehesa populations SPLV and SPPOR (Phi_PT = 0.150, p = 0.005; Phi_PT = 0.101, p = 0.037, respectively), but not between the dehesa populations of SPLV and SPPOR (Phi_PT = 0.054, p = 0.106) (Supplementary Table 5C).

In the Faroe Islands, one fifth (20%) of the isolates possessed different aMLGs (n = 12). Overall, five unique aMLGs and seven common aMLGs with 91.7% of the fungal isolates were present in the region. Several IDT genes were absent in the Faroese isolates (Table 5). Most (88.3% and 90%) of the isolates are expected to produce ergovaline and pyrrolopyrazine-1, 4-diones (PPZA-2), respectively (Figure 1 and Table 5). The most common aMLG, gt-27 (n = 30) was observed in half of the Faroese isolates in five populations and tested positive for seven out of thirteen markers including all EAS genes and lacked perAR marker and all five IDT markers (Table 5). Second most common Faroese aMLG, gt-33 (n = 9) contained 15% of Faroese isolates in four populations and tested positive of eight out of thirteen markers including all EAS genes and lacked perAR marker and all IDT genes except idtG (Table 5). No significant pairwise differentiation levels based on the occurrence of the alkaloid genes were detected among the Faroese populations in alkaloid gene variation (p > 0.05) except between FAS1 and FAS5 (0.147, p = 0.039) (Supplementary Table 5C).

In Finland, only 15.4% of the isolates possessed different aMLGs (n = 14). Overall, seven unique and seven common aMLGs with 92.3% of the isolates were detected in the region (Table 5). Majority (91.2%, and 87.9%) of the isolates contained all the EAS and IDT early pathway genes (idtG, idtQ, and idtP) and they are expected to produce ergovaline and/or terpendole I, respectively (Figure 1 and Table 5). Most (91.2%) of the isolates are predicted to make pyrrolopyrazine-1, 4-diones (PPZA-2) and one isolate peramine (gt-38; Table 5). The most common aMLG, gt-39 (n = 70) was found in 76.9% of Finnish isolates in all six populations (Table 5). The gt-39 was positive for 11 out of 13 markers, lacked perAR marker and idtJ gene, and had non-functional idtF gene (determined from ten isolates of gt-39). Thus, gt-39 is likely to produce ergovaline due to presence of all five detected EAS genes, pyrrolopyrazine-1, 4-diones (PPZA-2) due to presence of perA5 and perAT2 markers and the indole-diterpenes pathway intermediate terpendole I due to presence of idtG, idtQ and idtP (Table 5). No significant pairwise differentiation based on the occurrence of the alkaloid genes was detected among the Finnish populations (p > 0.05; Supplementary Table 5C).

Because the occurrence of the host plant is a prerequisite for the dispersal of associated symbiotic microbes, the genetic diversity of microbe should primarily mirror forces driving the postglacial distribution of the host species and secondarily the microbes, or the host and the microbe in concert as a phenotypic unit. During the last ice age, ending about 20,000 years ago, glaciers extended over much of northern Europe and also over much of Canada and some of the northern United States. Presently, F. rubra has a broad circumarctic-circumboreal distribution due to its great adaptive ability to colonize new pockets of land exposed from the retreating ice sheet during the postglacial distribution history (Inda et al., 2008; Braun et al., 2020). Taxonomically F. rubra is a morphologically variable species complex showing extensive hybridization, polyploidy, as well as phenotypic and genetic variation (Jenkin, 1955; Markgraf-Dannenberg, 1980; Ainscough et al., 1986; Aiken and Fedak, 1992; de la Fuente et al., 2001; Catalan et al., 2004; Catalan, 2006; Soreng et al., 2017). Thus, distinguishing taxonomic entities as species, subspecies and varieties within the complex is challenging (Saikkonen et al., 2019).

Epichloë festucae symbiotic plants can be commonly detected in all three geographic regions examined in this study, but the frequencies of symbiotic plants vary among regions, and populations and habitats within regions irrespective of phenotypic and genetic variation of the host (Dirihan et al., 2016; von Cräutlein et al., 2019). The highest overall endophyte frequencies were found in Spain, where 69% of plants harbored E. festucae. In contrast, 36% and 30% of grasses were Epichloë symbiotic in Faroe Islands and northern Finland, respectively (Dirihan et al., 2016). Although the fungus is capable of horizontal transmission by sexual spores, contagious spreading within the host populations and long-distance migration among populations or geographic regions appears to be strongly constrained in the Faroe Islands and Finland where the sexual life cycle appears to be extremely rare (Wäli et al., 2007). The present results support this as mating type representation were skewed toward the presence of MTA. Similar geographic patterns of genetic population structures of F. rubra and E. festucae suggest that the symbiont has primarily migrated with the host (see von Cräutlein et al., 2019). To date, there has been very little focus on mating type frequency. The recent studies have mainly concentrated on the asexual Epichloë species and it appears that only one mating type gene occur in most of the studied species, like E. festucae var. lolii (Hettiarachchige et al., 2015). In contrast, mating type gene frequency is found in equilibrium in Epichloë typhina where sexual stage is active (Bushman et al., 2019).

Here we propose that the classic theory of island biogeography (MacArthur and Wilson, 1967) and the geographic mosaic of coevolution (Thompson, 2005) provide a useful framework to understand distribution, and genetic diversity and geographic differentiation of E. festucae populations. Similarly to macro-organisms, a positive species-area relationship has been detected to lead to higher microbial diversity in large and less isolated sampling areas in studies using bacteria diversity in water-filled treeholes (Bell et al., 2005), foliar fungi in birch trees living in fragmented environments (Helander et al., 2007), ectomycorrhizal fungi on “tree islands” (Peay et al., 2007), and soil bacteria and fungi in land-bridge islands as models (Li et al., 2020). Analogously to the prediction that species diversity should reflect “island” size and isolation, we detected highest genetic and genotype diversity of E. festucae in Spain which can be treated as a “continent” from where potential F. rubra and E. festucae colonists dispersed into Faroe Islands and Fennoscandia. Similarly, the reduction of genetic variation detected in E. festucae during the postglacial distribution history appears to be formed by local selection pressures imposing the symbiotum across the examined geographic regions.

In Spain, we found the highest genetic diversity that can be explained by the occurrence of populations near glacial refugia and biodiversity center of fine fescues (Saint Yves, 1930) and occasional sexual reproduction. The two genetically distinct groups of E. festucae (Sp1 and Sp2) were distributed unevenly in the examined two habitats. The genetically more diverse Sp2-group predominated in Mediterranean savannah-like grasslands, dehesa, located higher in altitudes near Salamanca (populations SPLV and SPPOR). Sp2-group was detected also in Mediterranean oak forest (population SPGD) to a lesser extent. In contrast, the genetically less diverse Sp1-group was prevalent in Mediterranean oak forest and absent in dehesa. The alkaloid gene assemblage differed also among the habitat types and larger variation of alkaloid chemotypes occurred in dehesa compared to Mediterranean oak forest. Nearly all examined fungal lineages had the potential to produce ergot alkaloids with known anti-vertebrate and anti-invertebrate properties as well as peramine or PPZA-2 with anti-invertebrate properties (Ball et al., 1997; Berry et al., 2019; Caradus and Johnson, 2020; Hudson et al., 2021). More than 60% of the fungal isolates from xerophytic forest grasses had genetic potential to produce PPZA-2 and more than 30% of them also terpendoles. Much larger variation of isolates with both anti-vertebrate and -invertebrate properties occurred in dehesa grassland plants and almost all isolates had genetic potential to produce both ergovaline and peramine whereas only two isolates had potential of PPZA-2. Moreover, low genetic differentiation levels between dehesa populations support the similarities in population genetic structure revealed by both SSR and alkaloid markers, as detected in the studies of Arroyo Garcia et al. (2002), Vázquez de Aldana et al. (2010) suggesting adaptation to similar herbivores selection pressures in dehesa habitat. Much lower genetic differentiation was observed among the host grass populations compared to its fungal symbiont populations in Spain, which maybe due the outcrossing of genetically distant host individuals (von Cräutlein et al., 2019). These results suggest that prevailing selection pressures driving distinct prevalence of genetic structures in two different habitats in Spain is operating either on the fungus or fungus-grass genotype combination rather than on the host grass individually, although some variation is observed within dehesa populations (see also Vázquez de Aldana et al., 2010).

In concordance with the presumption that diversity should decrease with the distance to the source regions, some genetic diversity appears to be lost during the colonization of exposed land following retreating ice sheet in North Europe. However, our previous study on the host grass supports the hypothesis that some of the grass individuals may have survived on nunataks in Faroe Islands (Dirihan et al., 2016; von Cräutlein et al., 2019). The present study on E. festucae in the Faroe Islands do not support the same for the symbiotic fungus. Most Faroese isolates fell into locally adapted Far-group, had lost several IDT genes and were predicted to produce the same alkaloids across islands. Only three non-local SSR isolates were detected, two isolates from the Fin-group in FAS3 population and one isolate from Sp2-group in FAS6 population. This suggests that the long-distance co-dispersal of microbes with their hosts occur but not as efficiently into the region as its host F. rubra which possess a contact zone of various maternal lineages in the Faroe Islands especially since gene flow from the other locations have been found to be more effective in maritime than inland locations as seeds might arrive by floating and by birds (Saikkonen, 2000; Golan and Pringle, 2017; von Cräutlein et al., 2019). The genetic mismatches between the host and the fungal genotypes can affect infection losses of new plant genotypes during the establishment process and consequently reduce the number of novel fungal genotypes in the region (Saikkonen et al., 2010b; von Cräutlein et al., 2019). Thus, relatively large proportions of non-infected F. rubra individuals in Faroese populations may reflect the process of infection losses. In addition, similarly to Spanish populations, the Faroese populations were more differentiated from each other revealed by SSR markers than the Finnish ones, although predominantly two relatively large identical genotypes were present in most of the populations indicating fungal gene flow via host seeds among the islands. The observed genetic differentiation of the populations may have been caused by varying selection pressures driving populations in different directions, gene flow from other locations and predominance of vertical transmission of E. festucae in mainly clonally dispersing F. rubra in Faroe Islands (Harberd, 1961; Heide, 1990; Saikkonen et al., 1998, 2002; Zhang et al., 2010; Leinonen et al., 2019).

In the species northernmost distribution range in subarctic Finland, the examined E. festucae populations present distinct genetic Fin-group with relatively low genetic diversity compared to the Southern Spanish populations. Similarly to its host populations of F. rubra, one large genotype were observed also in the fungal populations indicating longevity and expansion of local host-fungus genotypes (von Cräutlein et al., 2019). F. rubra clones have been detected to be centuries old and occupying large areas (Harberd, 1961) as selection favors the presence of one dominant genotype in a clonal population (Milgroom, 1996). Only two non-local isolates were found in the region, even though relatively high frequencies of non-local host cpDNA haplotypes were observed in the mostly clonal host grass populations (von Cräutlein et al., 2019), which can be due to reproductive differences between the species (Sullivan and Faeth, 2004). Non-local plants can have reduced probability to flower in northern latitudes preventing vertical transmission of non-local E. festucae strains (Leinonen et al., 2019). Moreover, infections can be lost by long-distance seed dispersers during the establishment process because infected plants may have a lower fitness compared to uninfected plants especially in harsh conditions (Leinonen et al., 2019). In contrast to Spanish populations with genetically distinct habitat-specific groups, no differentiation was observed among populations divided in two habitats, meadows and riverbanks, in Finland, which may due to efficient local expansion of dominant host-fungus genotypes.

The alkaloid genotyping has proven to be very powerful to identify potential bioactive alkaloids produced by populations of Epichloë (Takach et al., 2012; Charlton et al., 2014; Young et al., 2014; Shymanovich et al., 2017). Genome sequencing of the first two E. festucae isolates and other related Epichloë species revealed considerable diversity within the genus based on the presence or absence of the alkaloid genes (Schardl et al., 2013a,b,c, 2014; Winter et al., 2018). In the current study, 20 unique chemotypes were predicted from the 198 isolates. These chemotypes ranged from individual alkaloids (e.g., chanoclavine, CC, paspaline, PAS, or the pyrrolopyrazine-1, 4-diones, PZZA-2), to more complex chemotypes representing up to three different classes of alkaloids. Sequencing of genetic loci associated with each alkaloid has revealed repetitive AT-rich transposable elements are associated with alkaloid diversity. In some cases, such as that of the pyrrolopyrazine perA gene, transposable elements have disrupted the gene causing the loss of the reductase domain, which results in production of pyrrolopyrazine-1, 4-diones rather than the expected peramine (Schardl et al., 2013b,c; Berry et al., 2015, 2019). In addition, the ergot alkaloid and indole-diterpene loci are located in what appears to be unstable AT-rich regions of the genome in the subtelomere region. Interestingly, we observed in each Spanish population one isolate able to produce lolitrem B. To our knowledge, the production of lolitrem B is rarely observed in F. rubra-E. festucae symbiotum and only mentioned in Young et al. (2009). The ergot alkaloid, ergovaline, and the pyrrolopyrazines, peramine and pyrrolopyrazine-1, 4-diones, were most commonly observed within the populations. However, isolates tested for ergovaline and peramine in a previous study (Vázquez de Aldana et al., 2020) did not consistently detect ergovaline when expected. As explanation, the ergovaline pathway has been reported to be silent, with no gene expression observed (Schardl et al., 2013b,c; Charlton et al., 2014; Young et al., 2015). Sequencing this population of E. festucae may provide greater insights into the evolution of these biosynthetic genes. The more limited predicted chemotypes found in Finland and the Faroe Islands versus that of Spain, may be due to selection pressure since the Spanish populations may have an advantage of many different chemotypes.