SEARCH

SEARCH BY CITATION

Keywords:

  • assembly rules;
  • boreo-nemoral forest;
  • habitat selection;
  • host selection;
  • seasonality

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We investigated whether arbuscular mycorrhizal fungal (AMF) communities in plant roots are random subsets of the local taxon pool or whether they reflect the action of certain community assembly rules. We studied AMF small subunit rRNA gene sequence groups in the roots of plant individuals belonging to 11 temperate forest understorey species. Empirical data were compared with null models assuming random association. Distinct fungal species pools were present in young and old successional forest. In both forest types, the richness of plant–AMF associations was lower than expected by chance, indicating a degree of partner selectivity. AMF communities were generally not characteristic of individual plant species, but those associated with ecological groups of plant species – habitat generalists and forest specialists – were nonrandom subsets of the available pool of fungal taxa and differed significantly from each other. Moreover, these AMF communities were the least distinctive in spring, but developed later in the season. Comparison with a global database showed that generalist plants tend to associate with generalist AMF. Thus, the habitat range of the host and a possible interaction with season played a role in the assembly of AMF communities in individual plant root systems.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Arbuscular mycorrhizal fungi (AMF) (phylum Glomeromycota) colonize the roots of most terrestrial plants, facilitating mineral nutrient uptake from soil in exchange for plant-assimilated carbon (Smith & Read, 2008). While AMF have repeatedly been shown to exhibit host-specific growth responses (Bever, 2003) and to induce differential growth responses in host plant species (van der Heijden et al., 1998; Klironomos, 2003), the degree of selectivity in natural associations remains unresolved. Many observations suggest that there is little or no partner specificity (i.e. exclusive association with a single taxon) in plant–AMF combinations (Smith & Read, 2008, p. 35). Indeed, the low number of AMF species [over 200 morphospecies – A. Schüßler's Glomeromycota phylogeny, http://www.lrz.de/~schuessler/amphylo/, or nearly 300 small subunit (SSU) rRNA gene sequence groups – Öpik et al., 2010] compared with the very large number of potential host species (perhaps 200 000 or 80–90% of terrestrial plants) implies that each fungal taxon must have many hosts.

However, the results of studies assessing plant–AMF compatibility using spore inoculation (Giovannetti & Hepper, 1985; Bever et al., 1996; Smith & Read, 2008) and more recently molecular studies addressing AMF communities in plant roots have started to reveal a more complex picture. Although most AMF taxa may be capable of colonizing a number of plant species, there is significant evidence to suggest that not all combinations of plant–fungal associations are equally likely to occur (Fitter, 2005; Helgason & Fitter, 2005). Öpik et al. (2010) summarized current knowledge about the global distribution of SSU-rRNA-gene-based AMF molecular taxa and found that they are unevenly distributed among vascular plant superorders. Case studies have also described divergent AMF communities in the roots of different coexisting plant species (Helgason et al., 2002; Vandenkoornhuyse et al., 2002, 2003; Scheublin et al., 2004; Pivato et al., 2007; Santos-Gonzalez et al., 2007; Sýkorová et al., 2007b; Öpik et al., 2008; Li et al., 2010). In principle, such findings could be related to the abundance (Dumbrell et al., 2010a), life history (Sýkorová et al., 2007a) or ecology (Chaudhary et al., 2008; Dumbrell et al., 2010b) of either fungal or plant partners. For example, Öpik et al. (2009) found that plant species and fungal taxa that could be characterized as habitat generalists frequently associated with one another, while forest specialist plants and fungi were similarly associated.

Although information about AMF communities in the roots of different plant species is accumulating, the related question of whether such communities are truly a nonrandom selection of taxa from the relevant local fungal taxon pool remains unclear. Community ecology has for decades considered so-called assembly rules, i.e., the processes governing the nonrandom component in community composition, which can be deduced by comparing certain parameters of an observed data set with the same parameters in multiple randomized data sets (Gotelli & Graves, 1996; Gotelli, 2000; Watkins & Wilson, 2003). However, the approaches used hitherto to study AMF communities have not sampled a sufficiently large number of plants (individuals and species) for the underlying pool of AMF taxa to be described with accuracy. Moreover, the analytical approaches used previously have generally been limited to simple permutation of data matrices to infer the importance of experimental or environmental factors in shaping AMF communities. In order to evaluate whether AMF communities in the roots of plant species are nonrandom assemblages, data are required from a representative number of plant individuals and species, collected from the same area and thus exposed to the same AMF taxon pool.

Here, we report new information on AMF communities at the level of plant individuals belonging to six forest understorey species. Using this information and published data from a further five species from the same location, we investigate whether the AMF communities in plant roots are random subsets of the local taxon pool or whether they reflect the action of community assembly rules. We compare the actual distribution of AMF taxa among plant individuals against null models that represent random colonization of plants by AMF taxa. If there are rules governing community assembly, we aim to investigate how these are related to the seasonal and successional dynamics of plant communities. Further, we test Öpik et al.'s (2009) hypothesis in the framework of assembly rules by considering whether AMF and plants exhibit deviations from random assembly at the level of ecological groups.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Study area and sampling

The study was carried out in Koeru boreo-nemoral forest, Estonia (58°58′N; 26°03′E), which is home to one of the best-studied AMF communities (Öpik et al., 2008, 2009; Moora et al., 2011; Powell et al., 2011; Unterseher et al., 2011), and from where 46 AMF SSU rRNA gene sequence groups have been recorded previously (according to Öpik et al., 2010; 51 according to the nomenclature in Öpik et al., 2009). The study area consists of 130 ha of spruce Picea abies forest (Hepatica nobilis Mill. site type) on a calcaric cambisol. Detailed descriptions of the area are provided by Moora et al. (2007, 2009) and Zobel et al. (2007).

In a previous study (Öpik et al., 2008), we described AMF communities in 90 individuals belonging to five plant species: Fragaria vesca, Galeobdolon luteum (note that Lamiastrum galeobdolon is a synonym for this species), H. nobilis, Oxalis acetosella and Trifolium pratense (Table 1). This information is used in the current analysis. In addition, we present new data from 123 individuals belonging to six further plant species: Geranium pratense, Geum rivale, Hypericum maculatum, Paris quadrifolia, Veronica chamaedrys and Viola mirabilis (Table 1). All sampled plant species were classified into two groups according to their habitat preference as presented in the BiolFlor database (Klotz et al., 2002): typical forest plant species (five species, hereafter called forest specialists) and plant species growing in a wide range of habitats including grasslands and forests (six species, hereafter called generalists; Table 1). This classification is consistent with the habitat preferences of the study species in Estonia according to Paal (1997), as well as descriptive studies of forest (Moora et al., 2007; Aavik et al., 2009) and grassland vegetation (Pärtel et al., 1999; Aavik et al., 2008).

Table 1.  Sampling of plant species and individuals in relation to sampling month and forest habitat type
PlantYoung forest standsOld forest standsTotalSources
JuneJulyOctoberJuneJulyOctober
  1. Numbers of plant individuals from which DNA was successfully amplified are shown, with the number of recovered AMF SSU rRNA gene clones in parentheses. Numbers of plant individuals from which DNA was successfully amplified are shown (six individuals per sampling month per stand were subjected to analyses if not stated otherwise). NP indicates that the species was not present, while NA indicates that it was not analysed. Sequencing data for different plant species were generated in this study or taken from Öpik et al. (2008).

Generalists
 Fragaria vesca2 (32)0(0)3 (44)4 (28)5 (44)5 (58)19 (206)Öpik et al. (2008)
 Geranium pratense0 (0)3 (31)5 (57)1 (9)3 (42)4 (51)16 (190)This study
 Geum rivale0 (0)5 (65)5 (63)0 (0)2 (27)3 (53)15 (208)This study
 Hypericum maculatum1 (11)4 (41)2 (29)2 (26)2 (28)2 (30)13 (165)This study
 Trifolium pratenseNA3 (31)4 (26)NPNPNP7 (57)Öpik et al. (2008)
 Veronica chamaedrys5 (49)5 (51)4 (55)2 (28)6 (77)7 (96)29 (356)This study
Forest specialists
 Galeobdolon luteumNPNPNPNP3 (36)3 (39)6 (75)Öpik et al. (2008)
 Hepatica nobilis5 (41)6 (74)6 (64)5 (31)4 (50)5 (48)31 (308)Öpik et al. (2008)
 Oxalis acetosella2 (27)6 (67)3 (26)4 (33)5 (19)6 (48)26 (220)Öpik et al. (2008)
 Paris quadrifolia3 (44)5 (73)2 (29)4 (48)6 (91)NP20 (285)This study
 Viola mirabilis4 (71)6 (93)6 (80)2 (18)6 (74)6 (80)30 (416)This study
Total22 (275)43 (526)40 (473)24 (221)42 (488)41 (503)212 (2486) 

Sampling of the six additional plant species followed the same procedures as described in Öpik et al. (2008). Briefly, this consisted of sampling two forest ecosystems of different age and management intensity: three mature old growth spruce forest stands (old forest stands) and three early successional stands in areas that were clear-cut approximately 25 years ago (young forest stands). Individual plants were sampled from a 10 × 10 m plot in each stand at the beginning of June, the end of July and the beginning of October in 2003. These sampling times were chosen to represent the beginning of the plant growth season, the mid-season and the end of the growth season in the study ecosystem. Entire plant individuals of each species were excavated from the sampling area if present. Roots were cleaned in the laboratory, dried with silica gel and stored before analysis. Generally, a maximum of two individuals from each plant species per plot per sampling time were included in subsequent analyses (Table 1; a third individual of V. chamaedrys was sampled from one old forest stand in October).

Molecular analyses

Molecular analyses followed those described by Öpik et al. (2008), with the sole difference being the cloning vector used (pCR2.1-TOPO vector as opposed to pGEM-T Easy). In brief, a 20-cm subsample of the root system of each plant individual was pulverized and DNA extracted using the Nucleospin® 96 Plant kit (Macherey-Nagel, Düren, Germany). Subsequent PCR reactions contained Expand High Fidelity Buffer (Roche Applied Science, Mannheim, Germany) with 15 mM MgCl2, 100 nM of each of the dNTPs, 200 nM of each of primers NS31 and AM1 (Simon et al., 1992; Helgason et al., 1998), 20 mg mL−1 bovine serum albumin, 0.7 U of Expand High Fidelity enzyme mix (Roche Applied Science) and 5 μL of DNA. Thermocycling conditions were as follows: 94 °C for 2 min; 10 cycles of 94 °C for 15 s, 58 °C for 30 s and 72 °C for 45 s; 20 cycles of 94 °C for 15 s, 58 °C for 30 s and 72 °C for 45 s+5 s per cycle; and 72 °C for 7 min using a DNAEngine PTC Dyad thermocycler (MJ Research, Reno, NV).

PCR products were purified using the MinElute PCR Purification kit (Qiagen, Crawley, UK), and cloned and sequenced following the method of Griffiths et al. (2006). Purified PCR products were inserted into the pCR2.1-TOPO vector (Invitrogen, Paisley, UK) by the College of Life Sciences Cloning Service at the University of Dundee, Dundee, UK. From each sample, 48 white or light blue colonies were selected and stored. Colonies were grown in 1 mL of 2 × Luria–Bertani broth with 0.15 mg mL−1 ampicillin in deep-well microtitre plates. Plasmids were purified using a Multiscreen Plasmid Minipreparation Kit (Millipore, Bedford, MA). Sequencing of 16 clones per sample was performed using the BigDye® Terminator v3.1 Cycle Sequencing kit (Applied Biosystems, Warrington, UK) with vector primers directed against the SP6 or T7 promoter regions. Sequencing reactions were purified using 96- or 384-well Geneclean plates (Genetix, Queensway, New Milton, UK) and run on an ABI Prism 3700 DNA Analyzer (Applied Biosystems). For some samples, 16 further randomly chosen clones were sequenced in order to obtain enough sequences of good quality.

Phylogenetic analysis

The new sequences, those from Öpik et al. (2008) and Glomeromycota reference sequences from the MaarjAM database (http://maarjam.botany.ut.ee/; Öpik et al., 2010) were aligned automatically using the mafft multiple sequence alignment web service in jalview version 2.6.1 (Waterhouse et al., 2009), and subjected to neighbour-joining (NJ) analysis (F84 model with gamma substitution rates) in topali version 2.5 (Milne et al., 2004). Sequence groups containing the Koeru sequences of interest in this study (i.e. the new sequences and those from Öpik et al., 2008) were delimited based on the NJ tree at ≥97% sequence similarity. Representative sequences from each sequence group were subjected to blast searches with default settings against the MaarjAM (status 28.01.2011, Öpik et al., 2010) and GenBank databases. Fifty-seven sequences appeared to belong to Ascomycetes and Basidiomycetes and were excluded from further analyses. In total, 2486 good-quality Glomeromycota sequences were retained in the analysis (Table 1). For subsequent analyses, the phylogenetically delimited sequence groups were assigned to the respective virtual taxa (VT) from the MaarjAM database (Supporting Information, Table S1) on the basis of the NJ tree and the blast against MaarjAM. Representative sequences of detected sequence groups were submitted to the European Molecular Biology Laboratory Nucleotide Sequence Database (accession numbers FR728433–FR728626 and FR837672-FR837673). For illustrative purposes, a phylogenetic tree containing sequences from major clades of Glomeromycota and the representatives of each sequence group was generated using beast (version 1.5.3; Drummond & Rambaut, 2007). The HKY+I+G nucleotide substitution model was selected on the basis of AIC (jmodeltest; Posada, 2008). A burn-in corresponding to 10% of samples was discarded and trees were drawn every 1000 generations from three independent runs of 10 000 000 generations. The results are summarized on a maximum clade credibility tree.

Statistical data analyses

Null model construction

Selectivity in plant–AMF associations was analysed in the framework of assembly rules (Gotelli, 2000; Watkins & Wilson, 2003). The nonrandom component of plant–AMF associations was detected by comparing the characteristics of an observed data matrix with those from multiple randomized matrices. The observed data consisted of a matrix with 212 individual plants in rows (i) and 40 fungal VT in columns (j). Thus, cell ij of the matrix was filled with the count of clones corresponding to VT j derived from plant individual i.

The choice of the randomization algorithm is important for ensuring that randomized matrices represent the desired null model (i.e. the null hypothesis that the observed pattern is a product of chance). Ecological mechanisms that are excluded from the null model can generate deviation from a random pattern. We used two algorithms in order to impose slightly different constraints and assumptions:

  • 1
    Randomization algorithm 1 – ‘permatswap’: Clones were assumed to be independent; thus, different clones from a single sample could be separately redistributed throughout the matrix during randomization. We fixed row and column sums and matrix fill, so that in both the original and every randomized matrix (i), plant individuals were associated with the same number of fungal clones; (ii) the overall counts of particular VT did not vary; and (iii) the overall number of plant–AMF associations did not vary. Thus, while the number of plant–AMF associations was strongly constrained by the randomization procedure, the precise identity of the associations was not. The quasiswapcount algorithm, implemented using the permatswap function in the r package vegan (Oksanen et al., 2010), was used for this randomization.
  • 2
    Randomization algorithm 2 – ‘permatfull’: AMF form individual structures that may extend spatially over areas that exceed the size of the sampling units used in this study (Rosendahl & Stukenbrock, 2004). In order to reflect the potential nonindependence between clones, we replicated the analysis using a less constrained, but sample-based procedure (i.e. permutation of matrix cells) within the permatfull function in r package vegan. Using this approach, column sums were constrained during randomization, and matrix fill remained constant as a result of the permutation, but row sums were allowed to vary. Thus, the overall count for each VT and the overall number of plant–AMF associations were held constant, but the numbers of fungal clones associated with plant individuals varied.

Although located in a homogeneous habitat, sampling was conducted in six spatially distinct stands within the forest. Therefore, both algorithms were further constrained to perform randomization separately within each stand. The combinations of constraints used by both randomization algorithms are considered to be conservative because of expected low type I error (Gotelli, 2000; the approaches correspond to Gotelli's Sim2 and Sim9).

Factors influencing AMF community composition in Koeru

For an overview of AMF communities in Koeru, the effects of certain factors were investigated using the permutational multivariate analysis of variance (permanova) function adonis from vegan, with the Bray–Curtis distance (BC) used as a measure of community dissimilarity. Separate models were constructed containing the following factors: forest successional stage (young or old), season (June, July or October) and either plant species (11 species) or plant ecological group (generalist or forest specialist). Rather than using the default permutation test to assess the significance of effects, we repeated the permanova procedure 999 times on data matrices randomized according to the permatswap algorithm (which meets the requirement of constant row fill), and the number of randomized pseudo-F statistics more extreme than the observed value was used to estimate P.

Selectivity in associations between plants and AMF

Because our focus was primarily on selectivity in associations between AMF and plants, the following questions were investigated:

  • 1
    Richness: Are AMF richness per plant species or plant ecological group, and plant richness per AMF VT lower than expected by chance? Low richness may be indicative of host preference. Richness measures from the original data matrix were compared with analogous measures from 999 randomized matrices.
  • 2
    Community assembly (BCdiff): Do the AMF communities associated with particular plant species and ecological groups differ from random subsets of the local fungal taxon pool? Species- or ecological group-level BC was calculated between the observed matrix and each of 999 randomized matrices (BC[observed vs. random]=BCor). In parallel, BC was calculated for 999 randomized vs. randomized matrices (BC[random vs. random]=BCrr). The latter calculation provided a population of BC measures that might be expected to arise by chance. The vector of 999 BCrr values was subtracted from the vector of 999 BCor in a random pairwise manner to produce a final vector of 999 values (BCdiff=BCor−BCrr). BCdiff has an expected value of 0 if the community composition is random.
  • 3
    Pairwise comparison of communities (BCpair): Are plant species and ecological groups associated with different communities of AMF? BC was calculated between the AMF communities associated with all pairs of plant species and between habitat generalist and forest specialist plants in the observed matrix (BC[group1 vs. group2]=BCpair). Each value was compared with 999 analogous measures from randomized matrices.

We applied these analyses to subsets of the data matrix representing different successional stages or seasons in order to assess whether the assembly rule varied in relation to these factors. For approaches (i) and (iii), 95% quantiles within the 999 randomized values were calculated, and the P values were approximated based on the number of randomized values more extreme than the observed value. For approach (ii), 95% quantiles of BCdiff were calculated, and P values were approximated based on the number of BCdiff values below 0.

Indicator species analyses

To identify the organisms involved in nonrandom associations, we tested whether any fungal VT could be characterized as ‘indicator species’ for certain plant species or ecological groups. We used Dufrêne–Legendre indicator analysis (Dufrêne & Legendre, 1997) as implemented by the indval function from the r package labdsv (Roberts, 2010). This index varies from 0 to 1 and would be maximal if all examples of a VT were distributed among all individuals of only one plant species or ecological group. The indval function uses unconstrained permutation; hence, the randomization procedure is not directly comparable to those used in other analyses. For each indicative fungal VT, the previously recorded habitat range was compiled (in August 2010) using the metadata that accompany accessions in the MaarjAM database. The proportion of accessions coming from forest habitat was calculated for each taxon. All analyses were carried out using r 2.10.1.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Glomeromycota diversity in Koeru forest

Forty VT were detected in a combined set of new and previously published (Öpik et al., 2008) sequences from 11 plant species in Koeru forest. The sequences belonged to Glomeraceae (27 VT), Gigasporaceae (2), Acaulosporaceae (7) and Diversisporaceae (4) (Fig. S1, Table S1). Seven of the detected VT were new to Koeru forest (not detected by Öpik et al., 2008, 2009) and included two VT not previously detected elsewhere (VT 315, VT 316). This analysis brings the number of known Glomeromycota VT recorded at the site to 53. The AMF community composition at Koeru was significantly influenced by plant identity – both in models containing plant species (permanova: pseudo-F=1.96, P<0.01) and plant ecological group (pseudo-F=5.06, P<0.01) – and forest successional stage (in a model containing plant species: pseudo-F=5.10, P<0.01; in a model containing plant ecological group: pseudo-F=4.98, P<0.01), but not by season or any interaction between factors.

Richness of partners in plant–fungal associations

Plant species were on average associated with 16.8 AMF VT, which was significantly fewer than expected from either randomization approach (P<0.01; Fig. 1; Table S2). The mean AMF richness was higher among forest specialist (18.8) than generalist (15.2) plants. Fungal VT were also associated with significantly fewer of the studied plant species (4.6 on average) than expected by chance (P<0.01; Fig. 1). Significantly fewer associations between plant species and fungal VT than expected by chance were recorded in both successional stages and all seasons using permatswap (all P<0.01) and in the young forest successional stage and the June and July sampling times (P<0.05) using permatfull (Table S2). Overall, four out of 11 plant species (two generalists and two forest specialists) were associated with fewer fungal VT than expected using permatswap, while there was only one species-level deviation (H. maculatum – generalist) from random using permatfull (Table S2). Meanwhile, four out of 40 fungal VT were associated with fewer plant species than expected using permatswap, while permatfull did not reveal any VT-level deviations from random (Table S2). When the data were split by season or successional stage, fewer taxa differed significantly in the richness of their associations (Table S2).

image

Figure 1.  Richness of the associations between plant species and AMF sequence groups. The mean richness of sequence groups associating with study plant species and the mean number of study plant species associating with sequence groups are shown. Bars show the distribution of mean richness values resulting from 999 randomizations of the data matrix; the solid line indicates the observed richness value. (a, b) Randomized values from the permatswap algorithm; (c, d) values from the permatfull algorithm (see Materials and methods for further details). In all cases, observed richness was lower than expected from randomization.

Download figure to PowerPoint

Composition of AMF communities associated with plant species and plant ecological groups

In general, the composition of AMF communities associated with plant species did not differ significantly from random subsets of the local AMF pool (as measured by BCdiff), except the community associated with V. mirabilis (95% quantiles of BCdiff=0.01−0.245, P<0.05 using permatswap; Table S3). However, the fungal communities associated with the two ecological groups of plants differed significantly from random (Fig. 2). When the data were restricted to particular seasons or successional stages, the only deviations from random occurred among the plant ecological groups during the July sampling time using the permatswap algorithm (forest specialists: 95% quantiles of BCdiff=0.007−0.153, P<0.05; generalists: 95% quantiles of BCdiff=0.009−0.201, P<0.05; Table S3).

image

Figure 2.  Difference in BC between real and randomized AMF communities (BCdiff) associated with habitat generalist and forest specialist plant species. For a precise calculation of BCdiff see Materials and methods. Bars indicate the values of BCdiff resulting from 999 randomizations. (a, b) Randomized values from the permatswap algorithm; (c, d) values from the permatfull algorithm. The solid line at 0 indicates the expected value if the AMF communities associated with each group are not different from random. In all instances, the 95% quantiles within the BCdiff values were >0 (P<0.05).

Download figure to PowerPoint

The AMF communities associated with pairs of plant species differed significantly in certain cases (i.e. BCpair; permatswap: P<0.05 in 14/55 pairs; permatfull: P<0.05 in 7/55 pairs; Table S4). Pairwise differences were most frequently observed between plant species corresponding to the two different ecological groups: using permatswap, 16% of the within-ecological group comparisons (i.e. a forest specialist species vs. forest specialist species or generalist species vs. generalist) exhibited significant differences, compared with 33% of the between-ecological group comparisons (i.e. a forest specialist species vs. generalist species; Table S4).

Correspondingly, the AMF communities associated with the two ecological groups of plants were clearly distinct from each other (P<0.01 using both randomization algorithms; Fig. 3). Significant differences between the two groups were observed in both young and old stands using both randomization algorithms (P<0.01; Table 2). Meanwhile, forest specialist and generalist plant species were not associated with significantly different AMF communities in June, but later in the season, AMF communities emerged that differed significantly between the ecological groups of plants (P<0.05 using both randomization algorithms; Table 2).

image

Figure 3.  BC between the AMF communities associated with forest specialist and habitat generalist plant species (BCpair). Bars denote the BC values from 999 randomized data sets; the solid line indicates the BC distance observed in the real data set. (a) Randomized values from the permatswap algorithm; (b) values from the permatfull algorithm (see Materials and methods for further details). In both cases, the distance between AMF communities in the roots of habitat generalist and forest specialist plants was greater than expected at random.

Download figure to PowerPoint

Table 2.  BC between observed AMF communities associated with habitat generalist and forest specialist plants (BCpair) in the entire data matrix and subsets of the matrix split according to successional stage or sampling month
Data subsetBCpairpermatswappermatfull
  1. The columns permatswap and permatfull show the 95% quantiles of BCpair measures taken from 999 matrices randomized according to the two different algorithms (permatswap and permatfull; see Materials and methods for further explanation); significant differences between observed and randomized measures of BCpair are taken to reflect community assembly rules acting on the AMF communities associating with the different plant ecological groups. **P<0.01; *P<0.05.

All data0.330.104–0.202**0.121–0.236**
Young forest stands0.310.122–0.25**0.124–0.257**
Old forest stands0.400.13–0.297**0.173–0.342**
June0.370.29–0.4520.254–0.48
July0.460.179–0.333**0.181–0.355**
October0.370.201–0.342*0.188–0.355*

Indicator VT

Certain AMF VT were significantly indicative of particular plant species or plant ecological groups in the overall data matrix or in subsets of the data split according to successional stage or season (Table 3). Notably, fewest fungal taxa (2) were at all indicative of any plant or plant ecological group during the first sampling time (June; Table 3). Accessions in the MaarjAM database (Öpik et al., 2010) representing the VT that were important indicators of forest plants overall in this study were more frequently recovered from forest habitats (18/29 or 62% of accessions) than those associated with VT that were indicative of generalist plants (28/141 or 20% of accessions; χ2=19.6, d.f.=1, P<0.01).

Table 3.  AMF sequence groups (VT) that are significantly indicative of plant species and ecological groups, specifically habitat generalists and forest specialists
Plant species/ecological groupAll dataYoung forest standsOld forest standsJuneJulyOctober
  1. Sequence groups that can be considered importantly indicative are taken as those with an indicator value ≥0.25 as suggested by Dufrêne & Legendre (1997) and are shown in bold type. **P<0.01; *P<0.05. VT are sequence groups delimited over currently published datasets in the MaarjAM database on the basis of bootstrap support and sequence similarity ≥97% (http://maarjam.botany.ut.ee/; Öpik et al., 2010).

Habitat generalists
 Fragaria vescaVT115 0.25** VT194 0.24* VT46 0.20*VT219 0.25*
 Geranium pratenseVT315 0.25* VT316 0.25*VT191 0.34**VT191 0.78* VT219 0.55*
 Geum rivale
 Hypericum maculatum
 Trifolium pratenseVT114 0.29*
 Veronica chamaedrys
Generalists pooledVT113 0.49** VT115 0.35* VT219 0.09* VT37 0.06*VT113 0.53** VT115 0.39*VT113 0.44* VT219 0.20* VT37 0.15**VT113 0.50** VT37 0.11* VT219 0.11*VT113 0.54** VT115 0.44*
Forest specialists
 Galeobdolon luteumVT67 0.26** VT65 0.17 *VT67 0.20*VT67 0.67**
 Hepatica nobilisVT199 0.28*
 Oxalis acetosella
 Paris quadrifoliaVT60 0.19*VT135 0.58** VT160 0.41* VT166 0.29*
 Viola mirabilisVT160 0.22*
Forest specialists pooledVT199 0.21* VT143 0.16* VT140 0.06*VT140 0.12*VT199 0.35**VT199 0.31*

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Arbuscular mycorrhizal fungi have traditionally been regarded as low-specificity mutualists that associate with a wide range of plant partners (Smith & Read, 2008). Nonetheless, it is becoming increasingly clear that distinct AMF communities are present in the rhizosphere (Bever et al., 1996, 2001; Eom et al., 2000) and associate with the roots of different plant species (Helgason et al., 2002; Vandenkoornhuyse et al., 2002, 2003; Scheublin et al., 2004; Pivato et al., 2007; Santos-Gonzalez et al., 2007; Sýkorová et al., 2007b; Mummey & Rillig, 2008; Öpik et al., 2008; Li et al., 2010). However, to the best of our knowledge, no previous studies have made a statistical comparison of the AMF associating with a large number of plant species from the same plant community. Thus, it is largely unclear to what extent plant-species- or plant-ecological-group-level AMF communities simply reflect a random sample of the local AMF taxon pools.

The assembly rules approach attempts to identify the processes influencing biological communities by comparing the parameters of real community matrices with the same parameters in matrices that have been randomized according to a chosen algorithm. The randomization algorithm is not merely a mathematical construct, but also the expression of a specific ecological hypothesis to be tested. While this approach has a long history in plant and animal ecology (Gotelli & Graves, 1996; Gotelli, 2000; Manly, 2007), it has only recently been adopted by microbial ecologists (Horner-Devine et al., 2007). Although the majority of studies considering assembly rules have analysed species presence–absence matrices, quantitative approaches that incorporate taxon abundance are expected to be ecologically more sound (Watkins & Wilson, 2003). Here, we present the first attempt to address assembly rules in AMF communities. We applied a quantitative statistical approach that used the counts of clones as a proxy for the relative abundance of AMF taxa in a sample. Potential biases in PCR resulting from possible preferential amplification of VT and plant-species-specific PCR inhibitors may mean that the sequence counts for each fungal VT do not necessarily reflect their natural relative abundance. Nonetheless, the constraints we imposed on randomization (in the permatswap algorithm) meant that counts of clones associated with each fungal VT and plant individual were fixed. Thus, any bias should not have resulted in artefactual identification of preferential associations between particular plant and fungal taxa. Moreover, it is worth noting that in our previous work (Öpik et al., 2009), we found significant and important concordance between measures of AMF taxon relative abundance in samples generated by independent analysis of the same plant root samples with independent PCRs, followed by cloning and Sanger sequencing or with pyrosequencing.

A further methodological issue that could potentially influence interpretation of the results is the unbalanced nature of the dataset with respect to the factors potentially influencing community composition (season and successional stage). Nonetheless, we used a permanova procedure that accounts for unbalanced designs, and partial reanalysis of the selectivity analyses using only those species that were present in every successional stage and season (i.e. excluding G. luteum, G. rivale and T. pratense) produced very similar results to those based on the full data (Supporting Information).

The permanova analysis showed that the AMF community present at the Koeru site did not differ between seasons, indicating that the same suite of taxa was consistently present. However, different AMF communities were associated with forest stands of different ages, effectively presenting plants in the two forest stand types with different fungal species pools. A classic study of AMF spores derived from soil samples along a successional gradient from grassland to forest similarly revealed changes in the AMF community, which the authors attributed to the characteristics of the host plant communities or the soil (Johnson et al., 1991). However, AMF communities can be robust to successional change; for example, Liu et al. (2009) found no differences in the AMF communities associating with a chronosequence of Caragana korshinskii plantations spanning 35 years. In our study, the use of a constant set of host plant species and the lack of an interaction between plant identity and successional stage suggests that the effect of successional stage was not a direct result of host plant identity.

The permanova also provided evidence for selectivity in associations between AMF and plants, both at the level of plant species and plant ecological groups. In our more detailed analysis of selectivity, both the number of AMF VT associated with particular plant species and the number of plant species associated with particular AMF VT were significantly lower than expected, indicating some level of partner selectivity. While we found only limited evidence of plant-species-level differences in associated AMF communities, significantly different AMF communities were associated with generalist and forest specialist plant groups. This result lends support to the contention that ecological groups of plants harbour different AMF communities (Öpik et al., 2009). While the contrast in the absolute richness of fungal species associating with forest specialist and generalist plant species is present, but fairly small, Öpik et al. (2009) detected a twofold difference in richness by utilizing a 454 sequencing approach that yielded counts of sequences that were orders of magnitude higher than those presented here. This comparison suggests that the (forest) specialist fungi might be infrequent, and that there can be substantial quantitative limitations in the ability of different approaches to detect nonrandom patterns in AMF communities. Our results also encourage further consideration of the mechanisms underlying selectivity in AM associations. Under experimental conditions, plant species ‘pick up’ different fungal communities from the same soil environment (Gollotte et al., 2004; Aldrich-Wolfe, 2007; Uibopuu et al., 2009), with the identity of neighbouring plant species potentially able to influence this process as well (Hausmann & Hawkes, 2009, 2010). AMF taxa differ in terms of their symbiotic function (Helgason et al., 2007), and there is evidence of preferential allocation of photosynthates by host plants to the more beneficial fungal partners (Bever et al., 2009). Moreover, studies of mutant plant individuals have shown that mutations in a single gene can induce barriers to colonization by AMF that affect AMF species to different degrees (Smith & Read, 2008). However, there also exists the potential for AMF host preference. AMF exhibit variation both in their mutualistic effects and in their competitive abilities, and there is some evidence of a trade-off between these characteristics (Bennett & Bever, 2009). Thus, successfully colonizing AMF may not always be the most beneficial for the host plant.

The taxon composition of AMF communities in soil (Dumbrell et al., 2011) and plant roots (Bever et al., 2001; Daniell et al., 2001; Husband et al., 2002; Öpik et al., 2003; Liu et al., 2009) is known to undergo seasonal changes. Our results suggest a novel detail in these seasonal dynamics: that it reflects a gradual organization of community assembly. In both the full and the reduced (balanced) data sets, the distinctiveness of AMF communities associated with plant ecological groups was the least in spring, but developed later in the season. Similarly, no AMF taxa were significantly indicative of the plant ecological groups in spring, but were by the later sampling times. In the study ecosystem, root growth of herbaceous plants halts in the winter and root functioning in spring is mostly in the form of new roots. On this basis, the fungal communities associating with individual plants presumably reform each year to some degree. Our results may therefore indicate that the colonization of new plant roots by AMF in spring is largely random, with selectivity only becoming apparent later in the season. This interpretation would be consistent with the feedback model of Bever et al. (1997), and would have the practical consequence that snapshot surveys of AMF communities may be unrepresentative of broader patterns. It should be noted, however, that the permanova did not produce a significant interaction between season and ecological group, and neither did a similar, but less powerful analysis by Öpik et al. (2008) when analysing the previously published subset of the data. Therefore, only very tentative conclusions can be drawn.

Accessions in the MaarjAM database (Öpik et al., 2010) corresponding to VT that were important indicators of forest specialist plants overall were more frequently recovered from forest habitats than those associated with habitat generalist plants. This difference is largely attributable to the widespread VT113, which was a good indicator of generalist plants and is infrequently recovered from forest habitats (16/106 or 15% of its accessions in MaarjAM). Nonetheless, it supports Öpik et al.'s (2009) finding from the same site that generalist plants tend to associate with generalist AMF taxa, while forest specialist plants associate with a mix of forest specialist and generalist AMF taxa.

The characteristics of those fungal taxa that were significantly and importantly indicative of plant ecological groups can also shed further light on the associations revealed in this analysis. AMF Glomus VT113 was the best indicator of generalist plant species. It is also the dominant fungus at the Koeru boreo-nemoral forest site (Öpik et al., 2008, 2009, this study), was the most abundant taxon in a large-scale study of AMF associating with an experimentally introduced plant species across European wooded habitats (Moora et al., 2011) and is the most commonly recorded VT in the MaarjAM database (Öpik et al., 2010). The molecular taxon contains sequences from cultures of Glomus fasciculatum (BEG53) and Glomus intraradices (unidentified cultures), which are both known to be generalist fungi (e.g. Börstler et al., 2010; Oehl et al., 2010) as is VT113 (Öpik et al., 2006, 2010). Glomus VT115, which includes sequences from cultures of G. intraradices, Glomus irregulare and Glomus vesiculiferum, is closely related to VT113, and it has not been distinguished from VT113 in several studies (cf. MaarjAM database). Studies of G. intraradices have also illustrated that single AMF species can exhibit complex responses to different host plants (e.g. Klironomos, 2003). Glomus intraradices genotype richness can also vary between locations and is positively related to disturbance, and there may be little overlap of genotypes between sites (Börstler et al., 2010), suggesting the presence of considerable functional diversity within the taxon. The currently unresolved taxonomic complexity of the G. intraradices species aggregate (Stockinger et al., 2009) calls for further investigation to clarify whether it represents a true generalist species with multiple ecotypes or whether the apparent widespread nature of the species is attributable to several cryptic species of more limited distribution. Glomus VT199, which includes sequences from cultures of Glomus hoi, is a dominant taxon in Koeru (Öpik et al., 2008, 2009, this study) and was a significant indicator of forest specialist plants. Elsewhere, it is commonly recorded in temperate forests and grasslands. This taxon has shown selectivity towards Acer among forest plants in field samples, but not under pot experiment conditions (as Glo9, Helgason et al., 2002). Although somewhat fragmentary, this information also supports the notion that the fungal taxa preferentially associating with generalist plants tend to be widespread or phenotypically variable, i.e., generalists themselves.

Selectivity in associations between plant and AMF taxa may shed light on other aspects of plant and fungal ecology. It potentially has important consequences for the study of plant community interactions, both for the formulation of new hypotheses and for designing studies that are not confounded by the effects of specific mutualists. The association of generalist plants with generalist AMF taxa may also have an important role to play in determining the success of biotic invasions (Moora et al., 2011).

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We are grateful to Frank Wright for his comments on earlier versions of the manuscript. The project was supported by Estonian Science Foundation grants 7371, 7366, 7738, SF0180098s08 (University of Tartu), a Marie Curie European Reintegration Grant within the seventh European Community Framework Programme (GLOBAM, PERG03-GA-2008-231034, M.Ö.) and by the European Regional Development Fund (Centre of Excellence FIBIR), EU FP6 project ECOCHANGE and EU FP7 project SCALES. SCRI is supported by grant in aid from the Scottish Government RERAD (T.J.D.).

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Figure S1. Maximum clade credibility tree of SSU rRNA gene sequences of Glomeromycota obtained from Koeru in this study (indicated in bold) or earlier (in italics; Öpik et al., 2008; Powell et al., 2011), and reference sequences.

Table S1. Counts of clones of AMF sequence groups recovered from different plant species.

Table S2. Richness of associations between plant species and AMF VT (according to Öpik et al., 2010).

Table S3. Deviation of observed AMF communities associated with plant species and ecological groups from random assemblages created either with permatswap or permatfull algorithms.

Table S4. Comparison of AMF communities associating with pairs of plant species.

Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

FilenameFormatSizeDescription
fem_1103_sm_figs1.eps1293KSupporting info item
fem_1103_sm_supplinfo-fig.doc20KSupporting info item
fem_1103_sm_supplinfo-tables.doc568KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.