Where the wild things are: looking for uncultured Glomeromycota

Authors


Summary

  • Our knowledge of Glomeromycotan fungi rests largely on studies of cultured isolates. However, these isolates probably comprise one life-history strategy – ruderal. Consequently, our knowledge of arbuscular mycorrhizal (AM) fungi may be biased towards fungi that occur primarily in disturbed habitats and associate with disturbance-tolerant host plants. We can expect to see a signal for this in DNA-based community surveys: human-impacted habitats and cultivated plants should yield a higher proportion of AM fungal species that have been cultured compared with natural habitats and wild plants.
  • Using the MaarjAM database (a curated open-access database of Glomeromycotan sequences), we performed a meta-analysis on studies that described AM fungal communities from a variety of habitats and host plants.
  • We found a greater proportion of cultured AM fungal taxa in human-impacted habitats. In particular, undisturbed forests and grasslands/savannahs contained significantly fewer cultured taxa than human-impacted sites. We also found that wild plants hosted fewer cultured fungal taxa than cultivated plants.
  • Our data show that natural communities of AM fungi are composed largely of uncultured taxa, and this is particularly pronounced in natural habitats and wild plants. We are better poised to understand the functioning of AM symbioses associated with cultivated plants and human-impacted habitats.

Introduction

Arbuscular mycorrhizas (AMs) are ancient and essential mutualisms between plants and fungi that have allowed plants to inhabit a wide variety of environments. These fungi (Phylum Glomeromycota) receive their entire carbon provisions from the host plant while imparting several benefits to the plant: most notably enhanced nutrient uptake, and increased abiotic and biotic stress tolerance (Smith & Read, 2008). Although there is evidence that these fungi associated with land plants from their inception (c. 460 million yr ago) (Pyrozinski & Malloch, 1975; Redecker et al., 2000; Strullu-Derrien et al., 2014), they show a remarkably depauparate species richness, with only 244 currently described taxa (Schüßler, 2013). This is significantly fewer than the numbers of taxa in other, less ancient fungal phyla: there are 30 000 species in Basidiomycota and 64 000 in Ascomycota, and 900 and 700 species, respectively, in the closest relatives Zygomycota and Chytridiomycota (all values from Kirk et al., 2008).

Such low species richness is difficult to understand, given that AM fungi associate with a large diversity of plant taxa (estimated 74% of plant species) (Brundrett, 2009). Recently, however, a surfeit of metagenomic studies from natural ecosystems has revealed greater species numbers at both local and global scales than previously thought (Moora et al., 2011; Öpik et al., 2013). Concomitant with this burgeoning diversity is the number of undescribed Glomeromycotan taxa – those known on the basis of DNA sequences only, with no known counterparts among morphologically described species (Öpik et al., 2014). Given our knowledge about the Glomeromycota, it is difficult to know what these ‘undescribed taxa’ represent. Do they represent sequences from groups that have been already encountered, or are these new lineages? Would these undescribed taxa show novel functions among AM fungi? How many AM fungal species are actually out there? Answers to these questions may change the way we think about the Glomeromycota.

Although we are learning more about AM fungi in natural communities, a more complete understanding of this phylum and the full taxonomic and functional diversity therein may be impeded by certain biases inherent in the scientific culture. Habitats and hosts that are either economically important or easy to work with dominate the existing literature. Thus, the number, identity and functioning of fungi that are described may under-represent the full biological complexity within this group.

The habitat bias

Molecular descriptions of AM fungal communities are providing greater sensitivity of detection and increased resolution of species composition compared with the earlier prevailing spore morphology-based surveys. Based on morphologically described species alone, estimates of global species richness in the Glomeromycota have increased from c. 150 species in 1993 (Walker & Trappe, 1993) to the current 244 species (Schüßler, 2013). When environmental sequences are considered, estimates range from 348 small subunit (SSU) ribosomal RNA (rRNA) gene-based virtual taxa (VTs) (Öpik et al., 2013) to over 1600 internal transcribed spacer (ITS)-based species hypotheses (SHs) (Kõljalg et al., 2013). Evidently, even the taxonomic richness of this phylum is still being documented, and therefore knowledge of the biology of this group of fungi is far from complete.

A large proportion of Glomeromycota detected in ecological surveys belong to ‘uncultivated’ sequence types (van der Heijden et al., 2008). While it is clear that these fungi belong to the Glomeromycota, there are no related sequences available in the public databases that originate from cultured specimens, and it is highly probable that even total sequencing of culture collections and herbaria would not close this gap (cf. Hibbett et al., 2011). Is this discrepancy because these ‘undescribed taxa’ represent isolates that cannot be easily cultured? Or have we simply failed to collect data from many ecosystems, thus ignoring species characteristic of nonmodel habitats? Relatively few habitat types have traditionally dominated the AM fungal research and literature – a recent search revealed that 11 of the 24 most cited experimental AM field studies sampled from old-field (post-agricultural) research sites (Web of Science; search terms: arbusc* mycorr* diversity; December 2013). If understudied habitats represent a reservoir of ‘undescribed taxa’, then current AM fungal diversity estimates might be decidedly low.

The glasshouse bias

Getting an AM fungus into culture can be challenging. To circumvent culturing, researchers have used spore communities in soil to study AM fungal natural diversity. However, this approach tends to underestimate true levels of diversity (Clapp et al., 1995; Hempel et al., 2007) because AM fungi differ in the amount and rate of sporulation (Carvalho et al., 2001; Oehl et al., 2004, 2009; Escudero & Mendoza, 2005). In fact, sporulation of some taxa has never been observed under experimental conditions (Merryweather & Fitter, 1995; Clapp et al., 2002; Husband et al., 2002; Rosendahl & Stukenbrock, 2004).

As an alternative to studying spore communities in situ, many researchers rely on ‘trap cultures’ whereby model host plants are grown with soil from an environmental sample under controlled conditions in order to trigger sporulation and provide more material for spore-based AM fungal identification. It is generally accepted, however, that these trap cultures are a poor approximation of natural AM fungal communities (Sýkorová et al., 2007; Hazard et al., 2013). The act of culturing is akin to severe disturbance; soil mixing fragments the soil mycelia and destroys connections to the host plant. Such conditions should favour the growth of disturbance-tolerant fungi, such as those that are more resilient to mycelial damage, invest more heavily in intra-radical structures, and can regenerate from root fragments (Chagnon et al., 2013). Disturbances of this kind are common in human-impacted ecosystems, although they occur to a lesser extent in natural systems.

In addition to disturbance, the prevailing practice of using a single host, which is usually a disturbance-tolerant cultivated or generalist plant species, may favour the growth of more generalist fungi over those with specific host requirements. Although trap-culturing of AM fungi is an invaluable method for obtaining cultures for experimentation, it is highly biased towards a limited functional group of AM fungi – the disturbance-tolerant generalist species. Ultimately, because we depend so heavily on cultured AM fungi to understand Glomeromycotan biology via comparative studies and controlled experiments, our knowledge may be confined to the disturbance-tolerant (ruderal) and generalist fungi. Thus, knowledge about other, nonmodel, disturbance-intolerant, specialist AM fungal species from nonmodel hosts and habitats may be systematically lacking.

Given that human-impacted ecosystems (characterized by recurrent heavy disturbance) probably contain more ruderal fungi compared with natural ecosystems (typically with lower disturbance intensity and frequency), we expect that DNA-based AM fungal community surveys of human-impacted systems should yield a higher proportion of species that have been cultured, compared with community analyses of undisturbed, natural systems. Further, host growth strategy may be an important predictor of the proportion of undescribed AM fungal taxa in natural systems. That is, we expect agricultural cultivated plants (disturbance tolerant) to host more AM fungal species that have been cultured as compared with wild plants. To test these hypotheses, we performed a meta-analysis of AM fungal community studies and asked whether disturbed systems had a higher proportion of described (cultured) AM fungal taxa than natural, undisturbed systems. We also explored the proportion of cultured AM fungal taxa found associated with plant hosts of different growth strategies.

Materials and Methods

Data collection

Publications included in this meta-analysis were selected from the 249 publications represented in the MaarjAM database (Öpik et al., 2010) and collected on 6 August 2013. MaarjAM is a curated, online, open-access AM fungal-specific sequence and meta-data database that places SSU rRNA gene sequences into groups called virtual taxa (VTs); the database also includes other markers such as the ITS and large subunit (LSU) rRNA gene, but these are not subjected to VT delimitation. VTs are delimited phylogenetically on the basis of bootstrap support of clades and sequence similarity ≥ 97% (Öpik et al., 2010, 2014). The VT concept implemented in this database provides an opportunity to examine natural AM fungal communities at approximately species-level taxonomy, using a consistent and stable taxon assignment approach (Öpik et al., 2010, 2014). The MaarjAM database incorporates representative sequences of AM fungal sequence groups for every host plant species at every site (where possible) in the International Nucleotide Sequence Databases (INSD) (http://www.insdc.org/). The MaarjAM database also incorporates data for sequenced cultured morphospecies from taxonomic publications. In this manner, sequences from both cultures and environmental samples are subjected to uniform VT delimitation (Öpik et al., 2014). This results in some VTs containing several morphospecies as a result of the relatively low variation of the central fragment of the SSU, in particular in Diversisporales (Öpik et al., 2010, 2013). Currently, SSU rRNA gene sequences are available for 84 of 244 morphospecies and are incorporated in MaarjAM in 60 VTs; an additional 288 VTs contain no sequences from morphospecies (Öpik et al., 2014).

Our approach used data already present in the MaarjAM database to assess differences in the proportion of cultured:uncultured VTs across different habitats and host identities. In this analysis, a VT is considered cultured if it contains at least one sequence from a cultured morphospecies. In this way, the occurrence of the cultured species can be recorded on the basis of DNA sequences, even in the absence of cultures from respective habitats and hosts.

Publications were included in the meta-analysis if they met the following criteria.

(1) Experimental set-up

DNA sequences must have been extracted from naturally occurring field sites. No sequences were obtained from glasshouse or trap culture experiments. Only control treatment sequence data were used when experimental conditions were tested in a study. For example, DNA extractions from inoculation or soil enrichment experiments were excluded in order to represent only unmanipulated AM fungal communities. In experimental designs with a temporal component, multiple sampling events were pooled into a single representative community. As in other meta-analyses, multiple ecologically distinct sites may have been sampled within an experiment. To meet our criteria, an experimental location was treated as an independent site when the location differed in vegetation type and abiotic environmental conditions.

(2) Sequences

Publications targeting only the SSU rRNA gene were included in the analysis. The SSU rRNA gene is a relatively conserved region of DNA that contains sufficient intraspecific variation and phylogenetic signal to allow delimitation of AM fungal sequence groups approximately at the species level, or slightly higher (Öpik et al., 2014). For this analysis, only sequences covering the central fragment of the gene were considered, because this marker region is most commonly used in AM fungus community surveys. Publications with sequences not covering this region or not targeting all Glomeromycota, such as those obtained using general fungal primers, AM fungal primers with lower level taxon specificity, or primers targeting the ITS region or LSU rRNA gene, were excluded. Accessions without representative DNA sequences available and no INSD accession number (referred to in MaarjAM as ‘dummy’ accessions) were excluded from the analysis in order to maintain consistent comparisons between publications.

We considered AM fungi identified from soil and/or plant root samples when considering communities from different habitats, but only AM fungi identified from plant roots when considering communities from different plant hosts. Finally, we considered only data sets generated using Sanger sequencing, because of the low number of next-generation sequencing studies available.

Once we had identified appropriate data sets, accessions containing sequences and related meta-data were extracted from the MaarjAM database. The sequences were used with their VT identities as present in MaarjAM.

(3) Categorical variables

Publications must have included ecological, environmental, and taxonomic meta-data to accurately determine the categorical explanatory variables. The categorical variables for this meta-analysis were as follows.

  1. Cultured versus uncultured AM fungi. In this study, two groups of AM fungal virtual taxa were distinguished: those that included sequence(s) from morphologically identified species maintained in a culture collection were considered ‘cultured’ VTs; those composed entirely of environmental sequences, with no sequence from a cultured AM fungus, were considered ‘uncultured’ VTs.
  1. Habitats. The habitat classification scheme was modified from that used in the MaarjAM database meta-data (described in Öpik et al., 2010) in order to increase the statistical power of the analysis. Four categories of habitat were used: human impacted (agricultural and industrial), forest, grassland/savannah, and desert/shrubland.
  2. Cultivated versus noncultivated plant species. We considered domesticated plants bred for food production as cultivated hosts and all other plant species noncultivated host plant species.
  3. Plant functional group. The following six categories were used: C3 grasses, C4 grasses, nitrogen (N)-fixing plants, woody plants, myco-heterotrophs and forbs (other than N-fixing plants and myco-heterotrophs).

Meta-analysis

Two sets of independent variables (moderators) were used in the meta-analysis as follows: habitat (with two levels: human-impacted and natural habitats, or four levels: human-impacted habitats, forest, grassland/savannah and desert/shrubland), and host plant species (with two levels: cultivated and noncultivated plants; or for host plant functional groups with six levels: C3 grass, C4 grass, N-fixing plant, woody plant, myco-heterotroph and forb). The data matrices were compiled as occurrences of VTs (presence/absence): (1) at each site, as defined in the experimental set-up; and (2) associated with each host plant species. Individual VTs were categorized either as cultured or uncultured AM fungal taxa, as described in the categorical variable definition of cultured versus uncultured AM fungi. The total count of cultured and uncultured VTs was calculated for each replicate (site or host species at a site) within a moderator.

For each study, we calculated data points for each site and also for each host species as follows:

display math(1)

This calculation methodology was used to create the data points for three data sets: (1) all data, (2) only Glomerales, and (3) only Diversisporales. AM fungi in the orders Paraglomerales and Archaeosporales were not included as sub-data sets because of the small number of records in the MaarjAM database resulting from infrequent detection in original case studies. Replication levels were variable among these data sets because Glomerales VTs tend to dominate over Diversisporales in the original case studies. Only replicates (sites or hosts) with at least one cultured or uncultured VT were analysed for the Glomerales or Diversisporales.

To adjust for differing sampling effort between published studies, we weighted each value with a modified nonparametric weighting scheme, calculated as follows: weighting factor = (N × N)/(N + N), where N equals the number of replicate samples taken at each site or for each host plant. This nonparametric weighting scheme has been used widely in the ecological literature (Hoeksema & Forde, 2008; Veresoglou et al., 2012).

Habitat effect

To determine the effect of habitat type on the proportion of cultured AM fungal taxa per site, first the cultured AM fungal VT proportion associated with human-impacted habitats was compared with the cultured AM fungal VT proportion in natural habitats. The parameter value in human-impacted habitats was used as the intercept with which the coefficient for natural habitats was compared. Similarly, we determined differences among the three natural habitat types compared to the human-impacted habitat intercept.

Host effect

To test whether host type affected the proportion of cultured AM fungal taxa per plant species, first the cultured AM fungal VT proportion associated with cultivated plant species was compared with that associated with noncultivated plants. The parameter value for cultivated plant species served as the intercept against which other categories were compared. Similarly, we determined differences among the six plant functional groups compared to the cultivated plant species intercept.

Statistical analysis

Statistical analyses were performed using R v3.0.2 (R Core Team, 2013). All meta-analysis procedures were calculated with the ‘metafor’ package (Viechtbauer, 2010) using a Knapp and Hartung adjustment for mixed effects models to test for coefficient significance in all analyses (Knapp & Hartung, 2003). In this approach, one moderator level is assigned to be the model intercept and all other coefficients are evaluated against this to determine the significance of differences.

Results

In this meta-analysis, data from 70 publications were used to investigate the effect of disturbance and host identity on the proportion of cultured AM fungal VTs across habitats and host plant species. The habitat data set incorporates a total of 162 sites from 70 publications and the plant host data set incorporates 279 plant species from 62 publications (Supporting Information Tables S1–S2). There was no estimated residual heterogeneity in any of the analyses; therefore, these values are not reported.

Habitat effect

The average percentage of cultured Glomeromycota VTs per site was lower in natural than in human-impacted habitats (= 0.008) (Fig. 1a). This was also true when only Glomerales and Diversisporales were examined (Glomerales: = 0.001; the same trend in Diversisporales: = 0.062) (Fig. 1b,c).

Figure 1.

The proportion of cultured arbuscular mycorrhizal (AM) fungal taxa among the total AM fungal taxa per site in human-impacted versus natural systems for all Glomeromycota (a), Glomerales (b), and Diversisporales (c). Symbols represent the weighted model coefficient percentage of cultured taxa (cultured taxa/all taxa) per site ± SE. The number of sites for each habitat category is shown in the figure, above respective moderator levels. Differences between moderator levels (human-impacted versus natural systems) were tested using a linear mixed-effect model. The statistical differences comparing the natural habitat moderator level to the human-impacted moderator level intercept were tested. Significance: **,  0.01;  0.10.

The average percentage of cultured Glomeromycota and Glomerales VTs per site was lower in forests and grasslands/savannahs, but not in deserts/shrublands, when compared with human-impacted habitats (Glomeromycota: grassland/savannah, = 0.010; forest, = 0.002; Glomerales: grassland/savannah, = 0.001; forest, = 0.001) (Fig. 2a,b). In the case of Diversisporales, only forest habitats (= 0.021) had a lower average percentage of cultured VTs per site compared with the human-impacted habitats (Fig. 2c).

Figure 2.

The proportion of cultured arbuscular mycorrhizal (AM) fungal taxa among all AM fungal taxa in different habitat types for all Glomeromycota (a), Glomerales only (b), and Diversisporales only (c). Symbols represent the weighted model coefficient percentage of cultured taxa (cultured taxa/all taxa) ± SE. The number of sites for each moderator level is shown in the figure, above respective moderator levels. The statistical differences among natural habitat moderator levels (desert/shrub, forests and grassland/savannahs (grass/sav)) compared to human-impacted moderator level intercept were tested using linear mixed-effect models. Significance: ***,  0.001; **,  0.01; *,  0.05.

Host effect

The average percentage of cultured Glomeromycota VTs per plant species was lower in the case of noncultivated than cultivated plant species (= 0.002) (Fig. 3a). This was also true when only Glomerales was examined (= 0.001)(Fig. 3b), but not in the case of Diversisporales (= 0.380) (Fig. 3c).

Figure 3.

The proportion of cultured arbuscular mycorrhizal (AM) fungal taxa among all AM fungal taxa in cultivated and noncultivated host plant species for all Glomeromycota (a), Glomerales only (b), and Diversisporales only (c). Symbols represent the weighted model coefficient percentage of cultured taxa (cultured taxa/all taxa) ± SE. The number of host species in each moderator level is shown in the figure, above respective moderator levels. Statistical differences between moderator levels (cultivar and noncultivar) were tested using a linear mixed-effect model. The statistical comparison relates the moderator level to the cultivated plants intercept. Significance: ***,  0.001; **,  0.01.

The average percentage of cultured Glomeromycota VTs per plant species was lower in the case of forbs (= 0.004), N-fixing plants (= 0.017), woody plants (= 0.001), and myco-heterotrophs (= 0.001) as compared with cultivated plant species (Fig. 4a). C3 and C4 grasses were associated with a similar percentage of cultured AM fungal taxa as cultivated plant hosts. Again, the same was true for the Glomerales (forbs, = 0.001; N-fixing plants, = 0.003; woody plants, = 0.001; myco-heterotrophs, = 0.001) (Fig. 4b), but not for the Diversisporales (Fig. 4c).

Figure 4.

The proportion of cultured arbuscular mycorrhizal (AM) fungal taxa among all AM fungal taxa in different host types for all Glomeromycota (a), Glomerales only (b), and Diversisporales only (c). Symbols represent the weighted model coefficient percentage of cultured taxa (cultured taxa/all taxa) ± SE. The number of host plant species in each moderator level is shown in the figure, above respective moderator levels. Statistical differences among moderator levels (C3 grass, C4 grass,forb, nitrogen-fixing plant, woody plant and myco-heterotroph) compared to the cultivated host plant intercept were tested using a linear mixed-effect model. Significance: ***,  0.001; **,  0.01; *,  0.05.

Discussion

Our understanding of Glomeromycotan biology has seen major breakthroughs in recent years: the publication of the first genomes (Tisserant et al., 2013; Lin et al., 2014), taxonomic reorganization (Oehl et al., 2011; Redecker et al., 2013), re-evaluation of their (a)sexual nature (Lee et al., 2014; Riley et al., 2014), and a reassessment of fundamental symbiotic functioning (Kiers et al., 2011; Fellbaum et al., 2012). All this research relies on using cultured fungal specimens.

In this meta-analysis we demonstrate that cultured fungi constitute only a minor fraction of AM fungi in natural and human-impacted habitats. Even in human-impacted habitats the proportion of cultured taxa is less than half of all species found within a site. These results highlight that our current understanding of Glomeromycota biology may be confined to a very small group of species with specific growth strategies. AM fungi (and their functions) that do not exist within model hosts and habitats may play important roles in global ecosystems, but this is yet to be explored.

Are there more easily cultured fungi in disturbed systems?

We found support for our hypothesis that disturbed systems harbour more cultured AM fungi than natural habitats. Furthermore, the proportion of cultured AM fungi among the total number of species per site differed depending on habitat type. In this study, AM fungal communities described from forests and grasslands contained fewer cultured AM fungi compared with human-impacted (disturbed) habitats.

We predicted that human-impacted systems would select for fungi that are more easily cultured. There is ample evidence that disturbance selects for phylogenetically conserved fungi (Johnson et al., 1991; Helgason et al., 1998, 2007). These fungi typically exhibit traits that may be associated with disturbance tolerance, such as restricted soil mycelia (McGonigle & Miller, 1996; Hart & Reader, 2004) and germination from hyphal fragments (Klironomos & Hart, 2002).

To determine if there was a phylogenetic basis underlying this trend, we analysed the data at the level of AM fungal order, as there is evidence that AM fungi differ broadly in terms of life history strategy between the Glomerales and the Diversisporales. Specifically, members of the Glomerales (Glomeraceae and Clairoideoglomeraceae) are thought to exhibit more r-selected traits whereas members of the Diversisporales (Gigasporaceae, Acaulosporaceae, Pacisporaceae and Diversisporaceae) tend towards more K-selected traits (such as large extraradical mycelium, spore size and density and increased hyphal diameter; Hart & Reader, 2002; Sýkorová et al., 2007; Chagnon et al., 2013). These differences suggest that Diversisporales may contain a higher proportion of uncultured fungi in undisturbed systems, as fungi in this group should be more abundant in undisturbed systems, and less amenable to culturing. In fact, we observed the opposite pattern. Our results showed that the Glomerales contained significantly more uncultured taxa in forests and grasslands/savannahs compared with human-impacted systems, whereas there was no difference in the proportion of uncultured taxa among different types of sites for the Diversisporales.

There are several possible reasons for the observed difference between the Glomerales and Diversisporales. First, it could be partially artifactual, that is, related to the lower species resolution power of the marker used, and the higher number of morphospecies grouped into a single VT in Diversisporales (Öpik et al., 2010). There is evidence to suggest that SSU rDNA does not provide good resolution to the species level in Diversisporales compared with other Glomeromycotan groups (Öpik et al., 2013). This would result in fewer VTs and possibly a higher proportion of total VTs in this order being categorized as ‘cultured’. The discrepancy may also be attributable to the fact that more Diversisporales morphospecies have been sequenced for the SSU rRNA gene (39%; 38 of 97 species) compared with Glomerales (only 23%; 25 of 109 species) (table 2 in Öpik et al., 2010). However, it is reasonable to assume that the most easily cultivated Glomeromycota have been sequenced by now for the commonly used markers. Thus, the proportion of sequenced species may reflect the greater ease of cultivability of Diversisporales over Glomerales, meaning that the observed difference in the ratio of cultured:uncultured taxa reflects the difference in the proportion of cultivable species between these groups. Furthermore, the cultivated species of Glomerales are represented in few clades of that order, while those in Diversisporales are more evenly distributed across the order (Öpik et al., 2010, 2013).

Taken together, the findings suggest that there might be functionally divergent groups present within Glomerales, the largest order of Glomeromycota, which may be the reason why our functional trait-based expectation was not supported by the data.

It is interesting that we detected no difference in the proportion of cultured fungi between human-impacted sites and deserts/shrublands. The simplest explanation for this is that there has been a considerable history of research in these environments and that these fungi are well represented in culture collections simply as a result of sampling intensity. But this trend may also have an ecological basis: because these systems are so resource-limited, there may be selection for a fungus to associate with the first host it encounters, that is, a more generalist host affiliation. This would result in fungi that are more easily grown in trap cultures, and thus better represented among cultured fungi. We do not yet understand, however, the role of stress in the fidelity of the host/AM fungus relationship.

Why are there more cultured fungi with certain plant groups?

We found support for the hypothesis that host identity affected the proportion of uncultured AM fungal taxa in a given data set; that is, fungal communities growing with cultivated plants contained more cultured taxa compared with plants rarely used in trap cultures or glasshouse experiments (i.e. forbs, N-fixing plants, woody plants and myco-heterotrophs). There are several reasons why this might be true. First, it is unlikely that all fungi in an environmental sample will be able to form an association with a trap plant. While traditionally believed to be generalist and nonspecific in terms of host range, research from natural systems reveals a scenario where natural AM fungal communities preferentially associate with a subset of available hosts (Vandenkoornhuyse et al., 2003; Öpik et al., 2009; Davison et al., 2011; Chagnon et al., 2012; Osanai et al., 2013). Thus, the fungi cultured on trap plants represent only a fraction of the total community that is able to colonize the trap plant. To compound this host effect even further, the plants used in trap cultures are most often those that grow well under glasshouse conditions (having a fast growth rate and being disturbance tolerant). When we recently surveyed the literature (search terms: arbusc* mycorr* diversity; Web of Science, November 2013), we found that of the 14 most cited AM fungal glasshouse studies, which include trap cultures or manipulative experiments, 10 used one of the following plant hosts: Trifolium, Sorghum, Allium, Prunella and Plantago (Bever et al., 1996; Streitwolf-Engel et al., 1997; van der Heijden et al., 1998; Van Tuinen et al., 1998; Eom et al., 2000; Hodge et al., 2001; Klironomos, 2003; Oehl et al., 2003, 2004; Munkvold et al., 2004).

This study indicates that forbs, N-fixers, woody plants and myco-heterotrophs were associated with proportionately fewer cultured AM fungal taxa. This may be because such hosts are underrepresented in glasshouse and field studies. While there has been some work with N-fixing hosts, these studies have typically been carried out on agricultural hosts (i.e. Medicago and Trifolium). Woody AM hosts, despite their wide distribution in many ecosystems, are grossly underrepresented, probably because long-lived perennials are hard to work with, but also because most of these species are tropical/subtropical, which are less represented in the literature (Öpik et al., 2010, 2013; Kivlin et al., 2011).

It is worth noting that the Diversisporales did not show any difference among plant hosts in terms of the proportion of cultured AM fungal taxa. That is, cultured and uncultured taxa from this group were equally represented across all plant functional groups. It may be that these fungi are less strict in their host requirements, which would make them more easily cultured, but this has not been tested in the literature. This finding is surprising because some members of this order are notoriously difficult to obtain, and keep, in culture, and particularly root organ culture (Klironomos & Hart, 2002).

Is our knowledge about the Glomeromycota biased towards cultured AM fungi?

Our results suggest that c. 60% of AM fungi that are currently known from DNA sequences are uncultured. Natural, undisturbed systems and wild plants are associated with more uncultured AM fungi than disturbed systems and cultivars. However, this value varies dramatically depending on the type of system and the identity of the host. Given that the vast majority of cultured AM fungal isolates originate in cultivated plants and human-impacted sites, is our understanding of AM fungi limited to these systems? Given that some systems and hosts are almost entirely undescribed by cultured AM fungi, our knowledge about natural systems and wild plants may be more limited than previously thought.

Are these fungi uncultured because they are unculturable or because they occur in sites or on hosts that have been overlooked? In this study, we suggest that culturability may be related to the life history traits of the AM fungi. However, we cannot yet conclude which traits are important determinants of culturability. The evidence on trait diversification in AM fungi is limited to basic information about mycelial investment (Hart & Reader, 2002; Johnson, 2010), spore size and propagule types (Klironomos & Hart, 2002), but more likely it is based on a diverse suite of traits, pertaining to dispersal, colonization and persistence in an environment. More information on such traits, and their variability in different environmental contexts, is needed to make better predictions about AM fungal community assembly and functioning.

Acknowledgements

The authors wish to thank the Irving K. Barber School of Arts and Science URA for funding received by P.D.Z., and the Natural Sciences and Engineering Research Council for funding received by M.M.H. M.Ö. receives funding from the Estonian Research Council (grants 9050 and IUT20-28) and the European Regional Development Fund (Centre of Excellence FIBIR). We are grateful to the three anonymous referees, S. Hart and the Editor for useful comments on the manuscript.

Ancillary