We used a set of 40 in vitro cultivated isolates of G. intraradices originating from one population in Tänikon, Switzerland, to address the question of occurrence of recombination in AMF. For reference, we included the isolate DAOM181602 originating from Québec, Canada. All isolates were clonally subcultured to obtain sufficient quantities of clean DNA 17. Sequences of 11 polymorphic nuclear markers were used for a total alignment length of 3037 bp. These markers were initially developed to reveal sequence length polymorphisms and were identified using repeat-finding software. However, sequencing of the different alleles showed that the large majority of the detected polymorphism was found in regions flanking the repeat motifs. We found 75 polymorphic sites and 72 indel mutations (for details on polymorphism at each locus see additional file 1). The high degree of sequence polymorphism identified within the population corroborates earlier data on polymorphism in coding regions 1819 and random genetic markers 17. The distinct combinations of alleles at the 11 loci identified 17 genotypes in the population (labeled I–XI and XIII–XVIII). Isolate DAOM181602 was a unique genotype not found in the Swiss population (labeled XII). A sequence alignment of all 11 loci from the 18 genotypes is available as additional files 2 and 3. Identifying organisms by multi-locus sequence typing was pioneered by Maiden et al. 20 for bacteria and successfully applied to fungi to e.g. infer sources of human pathogen outbreaks 21.

We first estimated the phylogenetic relationships among the genotypes to identify potentially recombining genotypes and clonal lineages. Then, a series of statistical tests were performed on sequences from all genotypes to test whether there is significant evidence of recombination in the population. By applying a neighbour-network algorithm 22 reticulate paths connecting the core genotypes could be observed (I–X, DAOM181602, XIII, XIV; Fig. 1). A reticulate pattern suggests that recombination among genotypes may have contributed to the evolution of the genotypes, but multiple alternative mechanisms such as lack of phylogenetic resolution or homoplasy may also contribute to a reticulate pattern 23. However, if recombination occurred within the population it is most likely among the core group of genotypes connected through a network (Fig. 1). The core genotypes weakly clustered into two subgroups, but multiple branches connect the two subgroups (I, IV–X and II, III, XIII, XIV, DAOM181602 respectively; Fig. 1). Five genotypes (XI, XV–XVIII) were distinguished by a comparatively large number of mutations from the set of core genotypes with very high bootstrap support (100%; Fig. 1). Three more genotypes were significantly differentiated from the core genotypes, but branch lengths were considerably shorter (IX, DAOM181602, XIV; Fig. 1). The reticulate pattern among the genotypes was conserved if all indel polymorphisms were excluded from the analysis (see additional file 4). This confirms that potential recurrent mutations, due to a higher indel mutation rate observed in repeat-rich regions, do not qualitatively alter the results of the analysis. The reticulate pattern among the core genotypes was still observed when the strongly differentiated genotypes (XI, XV–XVIII) were excluded, and this further supports the evidence of recombination (see additional file 4).

To examine evidence of recombination, we used multiple sequence-based tests, because detection abilities can vary strongly among tests depending on the degree of polymorphism and phylogenetic divergence of sequences 2425. We performed six tests of recombination on concatenated sequences of 11 loci. The concatenation of loci was necessary in order to overcome the limiting number of polymorphic sites within individual loci. Firstly, we used the Φ_w test that has been shown to reliably discriminate between recurrent mutations and recombination 26. The test uses a sliding-window procedure to assess phylogenetic compatibilities of nearby polymorphic sites in the concatenated sequences. A total of 72 sites were informative and significant evidence of recombination was found among the genotypes (p < 0.0001). We then used two phylogenetic (Bootscanning and RDP) and three nucleotide substitution based (Geneconv, MaxChi and Chimaera) tests 272829303132. These tests had already been applied to several fungal datasets in a critical analysis of statistical power of different recombination tests 24. As sliding-window based tests are sensitive to the phylogenetic signals of neighbouring polymorphic sites, the concatenation order of our loci is likely to affect the detection abilities of the different tests. A pair of concatenated loci that would have been inherited clonally within the population is not expected to show evidence of recombination. However, for a pair of concatenated loci that show a reshuffling within the population, the tests are likely to provide evidence for recombination (i. e. a significant recombination breakpoint). As not all pairs of loci are expected to show similar levels of clonal inheritance or reshuffling, multiple concatenation orders are useful to control for a potential bias through the arbitrary nature of concatenation. In total, we used three different concatenation orders, two of which were random orders and the third separates loci with strong signals of recombination. We found that five isolates from the population (II, III, XI, XIII and XIV) and the commercially cultivated isolate DAOM181602 showed recombination breakpoints detected by two or more tests in all three concatenation orders. Furthermore, isolate XVIII showed recombination breakpoints by at least one test in all three concatenation orders. Isolate X did not show any recombination breakpoints in any of the concatenation orders. The graphical results showing significant recombination breakpoints are shown in additional files 5, 6, 7. Detailed results with significance thresholds for each predicted recombination breakpoint of the concatenation order shown in additional file 5 can be found in additional file 8. The presented recombination breakpoint analysis is not suitable to distinguish between intragenic recombination or reassortment of loci as sequences of individual loci are too short to perform the tests on each locus separately.

Recombination creates conflicting phylogenies among distant loci through reshuffling of DNA sequences among different genomes. To test this prediction, we used a further, independent recombination test. The partition homogeneity test creates artificial datasets by sampling randomly among all observed sites of the genotypes and then swapping sites among loci 3334. The length of maximum parsimony trees for all loci are calculated and summed. If recombination occurred among the genotypes in the population, the actual summed tree length should be shorter than the summed tree lengths based on the artificially created datasets, because recombination should have introduced incongruence among loci. The partition homogeneity test showed that the actual summed tree length was 8 steps shorter than the shortest observed summed tree length in the artificially created dataset (p < 0.001, Fig. 2). Results were qualitatively similar if all indel polymorphisms were excluded (see additional file 9). Such conflicts in phylogenetic congruence among loci are most likely explained by recombination among the genotypes. The extent and frequency of recombination cannot, however, be inferred from these results. Results from the partition homogeneity test were shown to be sensitive to bias in base composition and mutation rate across loci 3536, but as our study included individuals only from one population, we expect this influence to be negligible.

The index of association measures the degree of linkage among different loci. This test has been widely used to detect linkage disequilibrium, following the initial application to bacterial genotypes 37. In our study, we calculated the index of association for the complete set of genotypes as well as for the two different subsets of genotypes. According to the phylogenetic relationships among the 18 genotypes (Fig. 1), including all 18 genotypes in the analysis of index of association would be unlikely to reveal recombination because of a number of lineages that appear not to recombine with the core isolates. Indeed, analysis using all 18 genotypes revealed strong linkage disequilibrium (I_A = 4.076, p = 0.0001), suggesting a population that deviates significantly from recombinant. If the five genotypes showing the strongest evidence for clonal evolution are excluded (see Fig. 1; bootstrap value = 100%; genotypes XI, XV–XVIII excluded) the linkage disequilibrium is no longer significant (I_A = 0.226, p = 0.087). No linkage disequilibrium was detected either (I_A = 0.167, p = 0.183) if all significantly separated genotypes are excluded (see Fig. 1; bootstrap value > 90%; genotypes IX, XI, XII, XIV–XVIII excluded), indicating that this subset of genotypes is potentially recombining. Two previous analyses of index of association in AMF populations showed opposite results from each other, although they studied different AMF species to those in the present study. Stukenbrock and Rosendahl 13 found strong linkage among three loci in three Glomus species, suggesting a strict clonal evolution. Vandenkoornhuyse, Leyval and Bonnin 11 found no evidence for linkage disequilibrium in two Glomus species based on ISSR loci. Our analysis showed that in a partially recombining AMF population, choosing various subsets of genotypes depending on the evidence of clonality changes the outcome of linkage analyses, but that groups of genotypes without significant linkage disequilibrium can indeed be detected.

Taken together, the graphical and statistical tests of recombination strongly suggest that recombination occurred among some of the genotypes in the field. The network analysis of the genotypes does not provide direct evidence for recombination, as other processes than recombination could lead to a reticulate pattern 22. However, the analysis suggested that a core group of genotypes are the most likely candidates to have undergone recombination. The index of association analysis showed that, indeed, the strongest signal of recombination is among this core group of genotypes. The analysis of phylogenetic congruence among all genotypes using the partition homogeneity test strongly suggested the occurrence of recombination within the population. Furthermore, by combining the results of the six sequence-based tests using a sliding-window analysis, five isolates of the population showed significant recombination breakpoints supported by two or more tests in all three tested concatenations. Interestingly, the commercial inoculum DAOM181602 grouped into the core group of recombining G. intraradices isolates, with evidence supported by multiple tests. Recurrent mutations or homoplasy are unlikely to explain these results as excluding indel polymorphism and using robust methods do not qualitatively influence the results. A previous study of genetic diversity in the same population used length polymorphism data at the same loci as in this study to identify different isolates 14. The reticulate relationships among the different genotypes suggested the occurrence of recombination as pointed out by Young 16. However, complete sequencing of all loci in all genotypes and appropriate analyses would be required to reach a clear conclusion 16. Homoplasy in the dataset would artificially increase signals of recombination as a pair of genotypes showing alleles of identical length but distinct sequence at a particular locus would be considered as potentially recombinant. The present study was based on sequences from all loci and all identified genotypes, therefore, allowing the control of size homoplasy. In future studies, data on genomic locations of the different loci could elucidate the frequency of recombination events and distinguish between inter- and intra-chromosomal recombination.

An alternative hypothesis could potentially explain patterns of recombination in sequences, as in some AMF it has been shown that isolates harbour genetically different nuclei 910. The previous use of linkage disequilibrium analyses (i.e. index of association) on AMF spores using AFLP were shown to be misleading if only genetic variation within isolates was considered 10. The sensitivity of fingerprinting techniques may artificially increase linkage disequilibrium and potentially mask recombination. Kuhn et al. 10 concluded that genetic heterogeneity found among nuclei within isolates has most likely arisen by accumulation of mutation, instead of recombination. To avoid the potentially confounding factor of within-isolate variation, we deliberately chose markers that did not reveal any significant within-isolate polymorphism (see Supporting Information in 14). We have found other markers showing multiple alleles within single spore isolates. These were excluded from the analysis to rule out confounding within-isolate recombination but are the subject of further research. Thus, the genotypes included in our study describe the unique, or at least the overwhelmingly predominant, nuclear genotype present in each isolate. The evidence for this is that only single alleles were found at each locus within isolates using capillary electrophoresis (see Supporting Information in 14). This argues against a potential explanation that is an alternative to recombination, namely that differences between isolates arise as a result of shifts in the relative frequency of pre-existing nuclear variants.

Our study reveals that a five isolates of G. intraradices isolates show strong evidence for recombination. However, we also identified five genotypes (XI, XV–XVIII) that were distinguished by a comparatively large number of independent mutational events, suggesting a clonal evolution in these lineages (Fig. 1). Statistical support for a clonal evolution of these genotypes was shown by very high bootstrap values separating these five lineages from the core group of genotypes (100%; Fig. 1). Several statistical tests identified putative recombination breakpoints in all these genotypes. However, these findings were not corroborated by different concatenation orders (additional files 5, 6, 7). This suggests that recombination between these latter genotypes and other genotypes in the population is either absent or rare. However, a much higher sampling effort in the field could yield genotypes that recombined with those lineages. Very few morphological criteria are known to distinguish isolates from the different clonal lineages, but the clonal lineages show clear differences in hyphal and spore densities produced in root organ cultures 17. Furthermore, genetic differences among a subset of the currently studied AMF population were shown to affect symbiotic functions with host plants 38. Knowledge of the effects on host plants by different clonal lineages and recombining genotypes would be important to understand the potential effects of recombination on AMF – host plant interactions. Recombination may have reduced genetic and phenotypic differentiation among the core genotypes, while clonal evolution in the separate lineages may have led to the fixation of a large number of genetic differences. In order to consider genotypes of different lineages as belonging to one biological species (i.e. forming an interbreeding population), all individuals should have the potential to undergo genetic exchange among each other. Our data, however, suggests that several lineages are independently evolving and may give rise to cryptic speciation.

The detection of recombination in a population of AMF is in strong contrast to the previous assumption of ancient asexuality in the phylum of Glomeromycota. The main argument for asexuality was based on morphological stasis over 400 million years, suggested by fossil evidence, and a lack of observed sexual structures 35. Together with bdelloid rotifers and ostracods 5, AMF represent what Maynard Smith termed "evolutionary scandals" as accumulation of deleterious mutations and a slow rate of adaptation should strongly disadvantage asexual reproduction in comparison to sexuality in the long term 39. While several theories tried to address this issue, recent evidence suggests that very few true asexuals might actually exist 8. A more likely scenario is that genomes of most organisms undergo at least sporadic recombination events to cope with the deleterious effects of long term asexuality. Genomic signatures of recombination seem to be more pervasive than previously thought 8, highlighting the fundamental role of sex in the evolution of genomes.

Sexual reproductive cycles involving the fusion of genetically different hyphae (i.e. mating) are thought of being the main mechanism in fungi allowing recombination among different genomes 40. But the lack of laboratory evidence of sexual structures involved in mating proved to be inconclusive about the occurrence of recombination in the species. Over the past decade, molecular evidence strongly suggested the occurrence of recombination in a number of fungal species that were presumed to strictly reproduce clonally. The human pathogen Coccidioides immitis, the pathogenic Aspergillus flavus and A. fumigatus, as well as the ant cultivated fungi in the fungus-growing ant symbiosis were all shown to undergo at least sporadic recombination events or even to possess functional genes required for meiosis 41424344. In the case of Cryptococcus neoformans, where a known sexual cycle exists, population genetic structures were found to be almost completely clonal. Nevertheless, reproduction between individuals of the same mating type was shown to allow recombination to occur 45. In a related species, a similar mechanism was shown to be at the origin of a major human pathogen outbreak 46.

In AMF, vegetative incompatibility was shown to occur when mycelia from different locations come into contact 4748. Such mechanisms, in combination with a lack of a known sexual cycle, were thought to exclude the possibility of recombination between genetically different isolates 49. Our study suggests that these mechanisms need to be further researched within AMF populations as they are the most likely mechanisms allowing the mixing of nuclei and subsequent recombination among different AMF in the soil.

A total of 40 G. intraradices isolates using in this study originated from an agricultural field site in Tänikon, Switzerland 51. The same set of isolates was used for population genetic analyses using AFLP, simple sequence repeat and mitochondrial markers 1417. Species identity of all isolates was verified by Croll et al. 14 by sequencing the internal transcribed spacer region and subsequent comparison with deposited sequences of the same species. The G. intraradices isolate DAOM181602 originated from a field site at Pont Rouge, Quebec, Canada (Biosystematics Research Centre, Ottawa, Canada). G. intraradices is widely used as a commercial AMF inoculum for agricultural applications (Premier Tech Inc., Canada) and is the first AMF isolate to be entirely sequenced 52. It is a haploid fungus with a compact genome of 15 Mb and is the only AMF for which ploidy has been measured 53. Ploidy could be different for other AMF species 54.

In vitro cultures of each isolate were established from single spores with Ri T-DNA-transformed carrot root on standard M growth medium 55. Five to ten two-compartment plates were inoculated by clonal subculturing to allow the proliferation of the fungus in one compartment that is kept root-free, while remaining connected to the roots in the other compartment 1756. Root-free fungal compartments of all plates were pooled per single spore line for extraction of hyphae and spores. Freshly isolated hyphae and spores of each isolate were separately dried overnight at 48°C and ground into a fine powder using a Retsch MM300 mixer mill (Retsch, GmbH). The DNA was extracted using a modified version of the Cenis method for fungal DNA extraction with an additional step of 1:1 dilution with a solution of 24:1 of chloroform isoamyl alcohol before the final precipitation, to remove remaining impurities 57.

We applied tests of recombination on 18 different G. intraradices genotypes that were identified by Croll et al. 14 on the basis of variation at 11 nuclear loci. In the Swiss population, 17 genotypes were identified. The isolate DAOM181602 showed a unique genotype. Ten loci were shown to contain short regions of repetitive DNA, indel polymorphism and SNPs in the flanking regions of the repeats (see Supplementary Material in Croll et al. 14). In addition, one nuclear gene intron was used 14. The function of the gene is currently unknown. Each marker revealed only one allele per single spore isolate 14. We identified other markers with multiple alleles per isolate but chose not to include these in the analysis to avoid confounding potential intra-isolate recombination with among isolate recombination (data unpublished). Croll et al. 14 identified alleles at each locus based on length differences and sequenced each allele that was seen to be different due to length polymorphism. For this study, we extended the sequencing, by sequencing all 11 loci in all the 18 genotypes identified by Croll et al. 14 so that any additional polymorphisms that were not based on length differences could be detected and also to check whether alleles with the same length in different isolates were also the same sequence. PCR amplifications were performed according to Croll et al. 14. PCR products were purified using the MinElute PCR purification kit (Qiagen, Inc.). Purified and quantified PCR products were directly cycle sequenced with BigDye Terminator v1.1 (Applied Biosystems, Inc.) following the supplier's instructions. Cycle sequence products were purified by ethanol precipitation. An ABI PRISM™ 3100 Genetic Analyzer was used for automated sequencing. All sequence profiles were visually checked using 4 peaks software (A. Griekspoor and T. Groothuis, http://mekentosj.com/4peaks). Sequences of all loci were aligned using the ClustalW algorithm implemented CLCbio Free Workbench 4.0 software http://www.clcbio.com and alignments were checked manually. Independent PCR and re-sequencing of alleles was performed to check for potential genotyping errors for all loci and in several isolates. No sequence variation was found in these tests. Identical labels (roman numerals I–XVIII) were used to name the genotypes as in the previous study 14. The sequences at all 11 loci of the 18 genotypes were deposited: locus Bg32 [GenBank: EU534209–26]; Bg42 [GenBank: EU534227–44]; Bg62 [GenBank: EU534245–62]; Bg196 [GenBank: EU534263–80]; Bg235 [GenBank: EU534281–98]; Bg273 [GenBank: EU534299–316]; Bg276 [GenBank: EU534316–34]; Bg303 [GenBank: EU534335–52]; Bg348 [GenBank: EU534353–70]; Bg355 [GenBank: EU534371–88]; nuclear intron [GenBank: EU534389–406]. A sequence alignment of all genotypes with concatenated loci as used for the data analyses is available in additional files 2 and 3.

The neighbour-net algorithm implemented in SplitsTree 4.8 was used to construct a network based on the uncorrected p distance using concatenated sequences of all loci 22. The robustness of the branching pattern was assessed with 1000 bootstrap replicates. Analyses were performed in two ways: (1) coding indel and repeat polymorphisms with Gapcoder using 1/0 to code for presence or absence of gaps in the alignment 58 and (2) ignoring all indel polymorphisms (see additional file 4). Ignoring gaps reduces the risk of including polymorphisms that may be due to recurrent mutations occurring in short repeat motifs, confounding potential signatures of recombination. Nevertheless, the majority of indel mutations are not in repetitive stretches of DNA and are, therefore, unlikely to be recurrent mutations.

Following Posada 24 and Posada and Crandall 30, we applied several recombination tests to evaluate the robustness of our results. To test for evidence of recombination based on phylogenetic compatibilities of nearby polymorphic sites along concatenated sequences, the Φ_w test implemented in Splitstree 4.8 was used 2659. The test was run with the default settings of a window size of 100 and k = 2. Five additional recombination tests were run in order to predict putative recombinant regions in the concatenated sequences: (1) Geneconv 2932 was performed scanning sequence triplets and treating indel blocks as single polymorphisms. (2) MaxChi 28 was used to scan all possible sequence triplets with 30 variable sites per window, alignments gaps (indels) were not considered as this could generate false positives 27. (3) General recombination detection (RDP test implemented in RDP software; 27) was performed with a window size of 10 and without specifying a reference sequence. (4) Bootscanning 31 was used with a window size of 100 and a step size of 20, standard distances were used for calculations and p values were binomial. (5) Chimaera 30 is a modification of the MaxChi test and was run with 30 variable sites per window. All five tests were run with RDP v2.08 27, setting the significance threshold to p = 0.05. Multiple comparison corrections were performed by a Bonferroni correction 27.

The partition homogeneity test 3334 implemented in PAUP 4.0b10 60 was used to estimate the degree of incongruence among loci created by recombination. To compare the degree of incongruence, 1000 datasets were artificially created by re-sampling all sites without replacement, regardless of the loci boundaries. The summed tree length of the maximum parsimony trees of each locus found for the original dataset was compared to the summed tree lengths of the maximum parsimony trees of each locus for the artificially created dataset. Firstly, the analysis was performed giving equal weight to substitutions and indel polymorphisms, giving a total of 77 parsimony-informative sites. An additional 70 variable sites were parsimony-uninformative. Secondly, the analysis was performed considering only substitutions in the sequences. This provided 32 parsimony-informative characters and 43 parsimony-uninformative characters (for results see additional file 9).

The index of association (I_A) measures the degree of linkage equilibrium among a set of genotypes. I_A = V_O/V_E -1, where V_O is the observed variance of the number of loci being different within all pairs of genotypes and V_E is the variance of the number of loci being different within all pairs of genotypes expected under complete linkage equilibrium. Therefore, I_A = 0 if genotypes recombine freely. The I_A and the significance threshold of a deviation from linkage equilibrium (with 10'000 randomizations) were calculated with the program MultiLocus v1.3b developed by P.-M. Agapow and A. Burt 61.