Plant neighborhood control of arbuscular mycorrhizal community composition


Author for correspondence:
Christine V. Hawkes
Tel:+1 512 471 6049


  • • Arbuscular mycorrhizal fungi (AMF) are important root symbionts that can provide benefits to plant hosts, yet we understand little about how neighboring hosts in a plant community contribute to the composition of the AMF community. We hypothesized that the composition of the plant neighborhood, including the identities of both host and neighbor, would alter AMF community composition.
  • • We tested this in a glasshouse experiment in which a native perennial grass (Nassella pulchra) and three annual grasses (Avena barbata, Bromus hordeaceaous and Vulpia microstachys) were grown in two neighborhoods: conspecific monocultures and heterospecific perennial–annual mixtures. To identify AMF taxa colonizing plant roots, we used a combination of terminal restriction fragment length polymorphism and cloning.
  • • Both host and neighbor were important in structuring AMF communities. Unique AMF communities were associated with each plant host in monoculture. In heterospecific neighborhoods, the annual neighbors V. microstachys, A. barbata, and B. hordeaceus influenced N. pulchra AMF in different ways (synergistic, controlling, or neutral) and the reciprocal effect was not always symmetric.
  • • Our findings support a community approach to AMF studies, which can be used to increase our understanding of processes such as invasion and succession.


The presence and community composition of arbuscular mycorrhizal fungi (AMF) can have important and well-established consequences for plant community dynamics. AMF affect the outcome of plant–plant competition (Bray et al., 2003), plant community dominance (Grime et al., 1987), plant diversity, and ecosystem productivity (van der Heijden et al., 1998a,b). Conversely, the ecological factors that affect AMF community composition may be as complex as the plant communities that host them. Plant host identity (e.g. Vandenkoornhuyse et al., 2002), plant priority (N. T. Hausmann & C. V. Hawkes, unpublished), soil type (Lekberg et al., 2007), habitat fragmentation (Mangan et al., 2004), and seasonality (Pringle & Bever, 2002) are all involved in structuring AMF communities. Of these, the role of the plant host in determining the community of AMF has been both the most studied and the most controversial (Bever et al., 1996; Smith & Read, 1997; Aldrich-Wolfe, 2007).

The strength and specificity of plant host effects on AMF remains an open question that will help us define the mechanisms underlying AMF distributions, understand the nature of plant–AMF relationships, and develop AMF as a tool in restoration and conservation. Delineating the degree of preferential associations between AMF and plant hosts is also essential to our understanding of the relative importance of top-down and bottom-up controllers between plant and AMF communities. Recent studies support a role for host plant identity in AMF growth and community structure. AMF are entirely dependent on plant hosts for carbon and have unequal growth rates on different plant species (e.g. Bever et al., 1996; StreitwolfEngel et al., 1997; Klironomos, 2002, 2003). Co-occurring plants can host distinct AMF communities (e.g. Vandenkoornhuyse et al., 2002; Gollotte et al., 2004; Scheublin et al., 2004; Douhan et al., 2005; but see Aldrich-Wolfe, 2007), which has been attributed to ecological host specificity and AMF functional diversity (McGonigle & Fitter, 1990; Helgason et al., 2007). Nonnative plants, in particular, host unique suites of AMF relative to native plants (Batten et al., 2006; Hawkes et al., 2006; Mummey & Rillig, 2006). In most of these examples, however, the host-associated AMF community is not entirely unique – a subset of AMF taxa are present across multiple hosts. Although AMF may not be host-specific per se, host identity clearly does play a role in structuring AMF communities. To understand the importance of host effects on AMF at the community level, we must explicitly examine interactions among plant hosts.

When multiple plant hosts are present, the plant neighborhood can be an important modifier of plant host effects on AMF. The local neighborhood in which a plant grows has long been known to affect plant growth, above- and belowground competition, facilitation, and so forth (e.g. Weiner, 1984; Casper et al., 2003), with the balance of local positive and negative interactions between close neighbors resulting in plant spatial distributions at the community level (e.g. Miriti et al., 2001). If interactions between local plant neighbors alter direct effects of plant hosts on AMF or change the available pool of AMF, then these neighborhoods also have the potential to structure AMF communities. Most of the evidence for plant neighbor effects on root and soil AMF comes from exotic plant invasions (Mummey et al., 2005; Batten et al., 2006; Hawkes et al., 2006; Mummey & Rillig, 2006). For example, Mummey et al. (2005) found that AMF in roots of one common grass were strongly affected by the presence of a highly invasive forb. Other studies have shown that less invasive exotics have the same influence on AMF in native roots (Hawkes et al., 2006). The evidence to date suggests that local plant neighborhoods can affect AMF, but this is largely based on studies of uncontrolled field associations, single nonnative neighbors, or multiple simultaneous neighbors.

Understanding the relative importance of plant host and neighbor identity within plant neighborhoods is a key first step in resolving top-down controllers of AMF communities. Specifically, are host effects on AMF maintained regardless of neighborhood? If host effects on AMF are malleable in the presence of neighbors, does the identity of the neighbor matter? Repeatable and unique root AMF communities associated with different hosts across multiple neighborhoods would suggest host specificity as the dominant factor determining AMF community composition. Alternatively, root AMF communities that are easily altered by neighbors would indicate a hierarchy of plant effects on AMF, perhaps determined by the relative strength of underlying mechanisms such as carbon supply or soil microclimate. Interactions of host and neighbor effects are likely and would be evident in different effects of different neighbors on AMF composition or abundance.

In this study, we explored how plant host and neighbor identity interact to structure AMF community composition. We focused on California grasslands where native perennial grasses have given way to exotic annual grasses over the past 150 yr (Mack, 1989; Heady et al., 1991). We focused on one native perennial (Nassella pulchra) and three annual grasses (one native, Vulpia microstachys; two exotic, Avena barbata and Bromus hordeaceus). These species co-occur in California grasslands where the native perennials compete poorly against the dominant exotic annuals (e.g. Hamilton et al., 1999; Grman & Suding, 2009). Because recovery of native perennials is a key restoration target, we were interested in mixtures of N. pulchra with different annuals. We also know that the combined presence of the exotic annuals, A. barbata and B. hordeaceus, can change the N. pulchra root AMF community based on previous experimental fieldwork where the grasses were maintained for 4 yr (Hawkes et al., 2006). Nassella pulchra maintained a distinct AMF community in monoculture, but the AMF in N. pulchra roots shifted dramatically to resemble the AMF in annual roots when all three were grown in mixture (∼80% overlap). By contrast, the AMF in annual grass roots showed very little (∼12%) change when grown with N. pulchra (Hawkes et al., 2006). These apparent neighbor effects were confounded with differences in the timing of germination between species, the abundance of each species, and long-term histories of each plot. Here we wanted to explicitly test how root AMF communities were affected by plant hosts and how those host effects were altered by different local neighborhoods.

We hypothesized that the interactions of host and neighbor drive patterns of root-associated AMF colonization and community composition during plant neighborhood establishment. We tested this hypothesis in a glasshouse experiment in which we created heterospecific neighborhoods (mixtures) by growing N. pulchra with each annual and conspecific neighborhoods (monocultures) by growing each species alone. Each species was evaluated as both a host and a neighbor. We addressed three questions. Do different plant hosts support different root AMF communities? How do root AMF communities change when in conspecific vs. heterospecific plant neighborhoods? How do different annual neighbors (heterospecific neighborhoods) change root AMF community composition in perennial N. pulchra plant roots compared with conspecific neighborhoods? We expected to find an interaction of plant host and neighbor, consistent with different neighbors having different effects on hosts and AMF. Moreover, based on annual suppression of perennials in the field (Dyer & Rice, 1997; Hamilton et al., 1999; Brown & Rice, 2000; Grman & Suding, 2009), we expected asymmetry in these effects, with N. pulchra root AMF more strongly impacted by annual neighbors than vice versa.

Materials and Methods

Plants and soils

We used four grasses: the native perennial Nassella pulchra (Hitch.) Barkworth, two exotic annuals, Avena barbata Link and Bromus hordeaceous L., and one native annual, Vulpia microstachys (Nutt.) Munro (nomenclature follows Hickman, 1993). In March 2005, we collected soil from the UC Hopland Research and Extension Center in Mendocino County, California, USA. Soils were sandstone-derived gravelly loam (Squawrock-Witherell complex). The area was largely dominated by a mixture of exotic annual grasses, including A. barbata and B. hordeaceus, which potentially limited the initial mycorrhizal pool and made the soils a conservative choice for the experiment. The soil was air-dried overnight and then sieved to 4 mm before being mixed 1 : 1 with sterile, coarse sand. Live roots were chopped to 1 cm and mixed in with the soil. Within a week of collection, soils were used to fill 700-ml pots to which we simultaneously added four pre-germinated seeds. Plants were grown in University of California Berkeley glasshouses, with pot positions randomized every other week over 10 wk. All pots were watered regularly without fertilizer and without supplemental lighting.

Experimental design

We used the glasshouse experiment to test how host and neighbor identity affected root AMF community composition, root colonization, and plant biomass. We created conspecific neighborhoods (monocultures of each species) and heterospecific neighborhoods (mixtures of N. pulchra with each annual) (Table 1). This experimental design allowed us to analyze both N. pulchra and the annual plants as hosts and neighbors (see ‘Statistical analyses’ section below). Treatments were replicated seven times with the exception of N. pulchra monocultures, which were replicated 21 times for independent comparisons with each annual. Thus, we had a total of 63 pots (Table 1), each containing four individual plants (four of one species in monocultures, and two per species in mixtures).

Table 1.  Experimental design and treatments included in each analysis
Plant species*nHost*Test 1Test 1aTest 1bTest 1cTest 2Test 3
Host × neighborNassella– Vulpia pairsNassella– Avena pairsNassella– Bromus pairsConspecific neighborhoodsNassella × neighbor
  • *

    Plant species included in each treatment abbreviated by genus name: Nassella pulchra, Vulpia microstachys, Avena barbata and Bromus hordeaceus.

  • Number of replicate pots in each treatment.

  • In these cases, seven replicates were used out of 21 for independence.

Nassella Nassella21NassellaXXXXXX
Vulpia Vulpia  7VulpiaXX  X 
Avena Avena 7AvenaX X X 
Bromus Bromus  7BromusX  XX 
NassellaVulpia 7NassellaXX   X
NassellaAvena 7NassellaX X  X
AvenaX X   
NassellaBromus 7NassellaX  X X
BromusX  X  


We included an additional nine replicates of each treatment to measure above- and belowground biomass. To develop allometric relationships for root and shoot biomass over a range of plant sizes, we destructively harvested three of these additional replicates at 3, 7, and 10 wk to capture a range of plant sizes. Roots and shoots of each plant were dried at 60°C to a constant mass and then weighed to the nearest tenth of a milligram.

The final harvest was 10 wk after planting, which should have allowed ample time for AMF networks to establish (Hart & Reader, 2002). We chose this time-frame to mimic the critical period of establishment of the annual grasses in the field. Germination occurs in early winter, with most seeds germinating within 2 wk of the first winter rains (Bartolome, 1979). Seedling mortality occurs largely between germination and the end of January (often as a result of mid-winter drought), after which populations are stable through late April when plants mature and dry (Heady, 1958). Thus, our timeline represents the period during which host and neighbor effects might be important for both seedling establishment and development of the initial AMF network in soil.

Roots of each individual were harvested by carefully untangling them while briefly immersed in water. Only those roots attached to a host were retained. We pooled entire root systems for the two individuals of the same species in mixture pots and two individuals in monoculture pots (84 total root samples) to ensure sufficient root material for our analyses. Half of the roots were stored in 95% ethanol for analysis of colonization rates and half were frozen at −80°C for DNA extraction. For colonization, roots were stained with acid fuchsin and hyphae, arbuscules, and vesicles were scored using the line intersect method at ×200 with 100 intersections (McGonigle et al., 1990). Shoots were oven-dried as above. We identified AMF taxa in roots using a combination of cloning, to generate an experiment-wide library of known sequences, and terminal restriction fragment length polymorphism (T-RFLP), to fingerprint individual roots that could be keyed to the library.

DNA extraction and PCR conditions

Root DNA was extracted using a modified Cetyl trimethylammonium bromide, phenol-chloroform protocol (Kennedy et al., 2003) and cleaned with a Qiagen DNeasy kit following the manufacturer's instructions (Qiagen, Valencia, CA, USA). We amplified a ∼550-bp fragment of 18S rDNA with the universal primer NS31 and the putative AMF-specific primer AM1 (Helgason et al., 1998; for discussion of primer efficacy see Daniell et al., 2001; Douhan et al., 2005). In separate PCR reactions, we created untagged PCR products for cloning and sequencing and tagged PCR products intended for T-RFLP. Tagged PCR products were labeled on both ends with different fluorescent labels (HEX-NS31 and 6-FAM-AM1; Sigma-Aldrich, St Louis, MO, USA). Regardless of fluorescent labels, all fragments were amplified in 20-µl reactions using a final concentration of 1.5 µm MgCl2, 20 µm Tris-HCl (pH 8.4), 50 µm KCl, 1 µm bovine serum albumin (BSA), 0.2 µm each dNTP, 0.5 µm each primer, and 0.5 U Taq DNA polymerase (Invitrogen, Carlsbad, CA, USA). PCR was carried out for 10 cycles at 95°C for 1 min, 58°C for 1 min and 72°C for 2 min, and 24 cycles at 95°C for 30 s, 58°C for 1 min and 72°C for 2 min, followed by a final cycle at 95°C for 30 s, 58°C for 1 min and 72°C for 10 min (Helgason et al., 1998).

Cloning and sequencing

We generated an experiment-wide clone library using pooled amplicons from all root samples as well as those from a complimentary study that included all the same plant species and soils but addressed a separate question about priority effects on root AMF (N. T. Hausmann & C. V. Hawkes, unpublished data). Pooled PCR products were gel-purified to improve the efficiency of the ligation reaction and were ligated into the pCR®2.1-TOPO® vector before transformation into chemo-competent Escherichia coli cells following the manufacturer's instructions (Invitrogen). We screened putative positive clones for inserts of the correct size with the universal M13 primers (Invitrogen) and sequenced 450 of the positive transformants. PCR products of clones were cleaned with ExoSAP-IT®(USB, Cleveland, OH, USA) and forward and reverse cycle sequencing reactions were carried out with BigDye®Terminator v3.1 (Applied Biosystems, Foster City, CA, USA) and M13 sequencing primers. Cleaned sequencing products were run on a 3730 DNA Analyzer (Applied Biosystems, Foster City, CA, USA); forward and reverse strands were auto-aligned and hand-edited to remove ambiguous bases using sequencher™ 4.2.2 (Gene Code Corporation, Ann Arbor, MI, USA). We eliminated 21 putative chimeras using the Ribosomal Database Project II Chimera Check v. 2.7 (Michigan State University, MI, USA). All sequences were deposited in GenBank (accession numbers EU123332–EU123465, inclusive).

To separate AMF sequences from nontarget sequences, we used National Center for Biotechnology Information BLAST (Altschul et al., 1990). We found that 234 (52%) of the clones matched AMF sequences, 79 (17.5%) matched other fungi (primarily ascomycetes), 60 (13.3%) were unresolved, and the remaining 77 (17.1%) matched only weakly with other sequences in the database. Unresolved and unmatched sequences were discarded. Using ClustalX version 1.83.1 (Jeanmougin et al., 1998), we automatically aligned AMF sequences, their closest BLAST matches, and additional AMF sequences from a phylogenetically representative set of Glomeromycota (Schϋβler et al., 2001) downloaded from GenBank. Sequences were then hand-aligned using SeAl version 2.0a11 (Rambaut, 1996). We similarly aligned the nonmycorrhizal fungal sequences in a separate analysis so the fragments generated by these sequences could be removed from the analyses.

To define AMF phylotypes, we performed separate Bayesian phylogenetic analyses on the AMF and non-AM fungal sequences (Supporting Information Figs S1, S2). To identify the optimum DNA substitution model we used MrModelTest version 2.2 (Nylander, 2004). All phylogenetic analyses were performed with MrBayes version 3 (Ronquist & Huelsenbeck, 2003). Nucleic acid positions with data in fewer than 25% of taxa were ignored. We ran Markov-chain Monte Carlo Bayesian analysis for 10 million generations under the invariable gamma model with equal state frequency rates, using four chains simultaneously with a temperature of 0.05. We saved every thousandth tree and discarded the first 5000 trees as burn-in; remaining trees were used to estimate posterior probabilities. We defined phylotypes as terminal clades with a posterior probability > 0.74. Clones that did not fit into a supported monophyletic group were grouped paraphyletically according to the closest clade with high support. We narrowly defined phylotypes at this stage because phylotypes with identical T-RFLP profiles would be pooled at later stages of the analysis to form operational taxonomic units (OTUs). We identified 24 phylotypes in our AMF clone sequence database (naming follows Schwarzott et al., 2001). Rarefaction analysis (analytical rarefaction version 1.3 software; S. Holland; estimated that at least 95% of the AMF diversity was sampled (data not shown).

T-RFLP analyses

T-RFLP analysis of clones was used to generate fragment length profiles of known sequences to use in identifying T-RFLP profiles of the experimental samples. Computer-generated restriction maps of ‘expected’ fragment lengths suggested that four enzymes would generate sufficient polymorphic fragments to identify AMF and nontarget sequences. The enzymes were HinFI (G^ANTC), MboI (^GATC), TaqI (T^CGA) (Fermentas, Hanover, MD, USA), and Hsp92II (CATG^) (Promega, Madison, WI, USA); all four have been used previously in AMF T-RFLP analyses (Mummey et al., 2005). We excluded three of the eight potentially identifying fragments (AM1-HinFI, AM1-Hsp92II and AM1-MboI) because they generated either too few fragments or too many fragments outside of detection limits (< 75 bp or > 500 bp) to be informative.

To generate empirical fragment lengths from the clones, we re-amplified clones (including non-AMF) using the fluorescent NS31/AM1 primers as described for PCR conditions above. PCR products were digested and fragment lengths were analyzed on the ABI 3730. We verified that 1926 of the 1944 clone fragments matched predicted lengths within 5 bp. Differences between predicted and empirical fragment lengths have been reported previously (Hartmann & Widmer, 2008). Clones without clear matches were removed from the analysis.

The 24 AMF phylotypes were further collapsed into 15 AMF OTUs based on clone fragment profile similarity, a conservative and likely underestimation of the true AMF diversity. Most phylotypes had only one identifying T-RFLP fragment per enzyme. Clones with identical T-RFLP profiles were grouped into an OTU using the TRAMPR package in R (Fitzjohn & Dickie, 2007). With the exception of one group (Glom/Acau), only sister phylotypes had identical T-RFLP profiles, which was to be expected based on our conservative phylotype grouping method. Pooling across phylotypes is an unfortunate artifact of the T-RFLP/cloning method that could be avoided either by using a more polymorphic gene (with its own limitations) or by creating costly and time-consuming clone libraries of individual samples. We followed the same grouping procedure for nontarget fungal sequences and verified that no complete T-RFLP profiles were shared between AMF and nontarget sequences (Tables S1, S2). In this way, we effectively removed T-RFLP peaks generated from nontarget sequences.

T-RFLP analyses of samples were performed much in the same way as the clones, with the following exceptions. To minimize PCR bias, we pooled three replicate PCR reactions for each sample and gel-purified before digestion. Runs were only accepted with total fluorescence of 5000–50 000 rfu and at least one peak with minimum 1000 rfu. T-RFLP profiles not meeting these criteria were rerun with adjusted concentrations of sample DNA. To identify AMF OTUs in sample T-RFLP profiles based on clone T-RFLP profiles (Table S1), we again used TRAMPR (Fitzjohn & Dickie, 2007). We considered an OTU present if all five identifying fragments (NS31-HinFI, NS31-Hsp92II, NS31-MboI, NS31-TaqI, and AM1-TaqI) were present within 1.5 bp of the known fragment length, supporting high confidence (P < 10−4) in the species identifications (Dickie & FitzJohn, 2007).

Statistical analyses

To examine the effect of our treatments on shoot biomass, per cent root length colonized by hyphae, arbuscules and vesicles, and root AMF richness, we tested: host and neighbor effects for each N. pulchra–annual neighbor pair with two-way, fixed-factor ANOVAs; host effects based on conspecific neighborhoods using one-way ANOVA; and effects of different neighborhoods on N. pulchra using one-way ANOVA (Table 1). Shoot biomass was included as a covariate in AMF ANOVAs, but was found not to be significant and was removed. Full ANOVA statistics are reported in Table S3. Percentages were arcsine square root transformed for normality and homogeneity of variance and are presented in figures as back-transformed values with asymmetric 95% confidence intervals; untransformed data are reported as means with ±1 SE. Means were considered significantly different at P ≤ 0.05. Pearson's bivariate correlations were calculated for the relationship between shoot and root biomass.

In order to treat the two plant species in each of the heterospecific neighborhood treatment pots independently as host and neighbor, we ran a Monte Carlo analysis to test for within-pot effects. For each N. pulchra–annual plant pair, we calculated pot-level pairwise Bray–Curtis similarities for root AMF communities. Treatment averages of pot-level similarities were compared against a null distribution of similarities generated from 1000 random pairings of N. pulchra with each annual. All within-pot AMF community similarities were statistically indistinguishable from the null distributions, allowing us to treat individuals from the same pot as independent.

AMF community composition in roots was analyzed using multivariate statistics. To determine whether root AMF differed significantly by treatment, we used multiple response permutation procedure (MRPP), a Monte Carlo approach that compares dissimilarities within and among groups (McCune & Grace, 2002). If two treatments differ in composition, their dissimilarities (calculated with a Bray–Curtis dissimilarity matrix for presence/absence of each AMF OTU in each sample) ought to be less than the average pairwise dissimilarities between two random collections of sampling units drawn from the entire population. MRPP can be structured with multiple, nested, one-way analyses in order to test complex models (Biondini et al., 1985). For MRPP, we report chance-corrected within-group agreement (A), a measure of effect size, and P-value. The structure of our MRPP analyses mirrored the ANOVAs (Table 1), except that we added an analysis of experiment-wide main effects of host, neighbor, and their interaction.

To visualize the differences among AMF communities tested with MRPP, we used nonmetric multidimensional scaling analysis (NMS) with Bray–Curtis dissimilarity matrices calculated from presence/absence of AMF OTUs (Clarke, 1993; McCune & Mefford, 2006). Correlations between the NMS axes and other variables were identified using a secondary matrix that included shoot biomass, root AMF community richness, and root colonization by hyphae, arbuscules, and vesicles. Additionally, we performed a Dufrene–Legendre indicator species analysis to test for associations of AMF taxa with treatments. Dufrene–Legendre analysis compares the relative frequency of a species in a treatment to a null distribution of frequencies across treatments (Dufrene & Legendre, 1997). Because of the limitations of cloning and T-RFLP methods, we were not able to calculate true abundances of AMF OTUs. Thus, abundance was calculated as the frequency of AMF OTU presence across treatment replicates. The NMS and MRPP analyses were run with pc-ord version 5.15 (McCune & Mefford, 2006); all other analyses were performed in R (R Development Core Team, 2007).


AMF community composition

The 15 AMF OTUs in this study came from three major family groups: Glomeraceae, Acaulosporaceae, and Diversisporaceae. Paraglomeraceae was absent, as the AM1 and NS31 primers are known to exclude this group (Daniell et al., 2001; Douhan et al., 2005). We did not find any Gigasporaceae taxa in this study, even though they have been found associated with these grasses in previous work (Hawkes et al., 2006). We amplified a large number of nontarget fungal sequences, a recognized problem with these primers (Douhan et al., 2005), which we identified and removed from the analysis based on their distinct T-RFLP profiles (Table S2).

We were able to confidently identify AMF in 73 out of 84 root samples. All AMF taxa were found in at least 30% and at most 75% of root samples in the experiment. AMF OTU richness, defined as the total number of OTUs present in a single root sample, was consistent across treatments in this study, with an average of 5.56 ± 0.40 OTUs except for V. microstachys (P = 0.011). When V. microstachys was grown in mixture, fewer AMF taxa were present compared with monocultures (monoculture, 8.3 ± 1.2; mixture, 3.9 ± 1.4; P = 0.037); N. pulchra root AMF richness, however, was unaffected by V. microstachys neighbors (monoculture, 4.4 ± 1.3; mixture, 2.4 ± 0.5; P = 0.181). Shoot biomass was never a significant covariate of AMF richness.

Host and neighbor effects on AMF community composition

Neighborhood influenced root AMF community composition in the overall experiment-wide analysis (Test 1, Table 1), with different patterns based on plant host (MRPP A = 0.052, P = 0.004) and the interaction of host by neighbor species (MRPP A = 0.107, P < 0.001), but not neighbor alone (MRPP A = 0.011, P = 0.198). Among hosts, AMF communities in N. pulchra roots were largely unique, separating from both exotic annual hosts along the first axis of the NMS, which explained ∼62% of the variation in the analysis (Fig. 1). The AMF community in roots of V. microstachys was intermediate between these two groups along NMS axis 1. Among the annuals, A. barbata and B. hordeaceus had more similar root AMF compared with each other than with V. microstachys root AMF, which separated along NMS axis 2 where a third of the community variation was explained.

Figure 1.

Host effects on arbuscular mycorrhizal fungal (AMF) root communities based on the experiment-wide nonmetric multidimensional scaling ordination. Plant hosts are plotted as the average (± 1 SE) ordination coordinates based on samples in each treatment and are designated by their genus name. AMF operational taxonomic units (OTUs) that were associated with r2 > 0.40 on at least one of the axes are plotted. Final stress for a two-dimensional solution was 11.65. Host (A = 0.052, P = 0.004) and the interaction of host and neighbor (A = 0.107, P < 0.001) were significant in multiple response permutation procedures, but the main effect of neighbor was not (A = 0.011, P = 0.198).

In paired comparisons of N. pulchra with annual neighbors, root AMF composition was influenced by significant interactions of host and neighbor for N. pulchraV. microstachys (MRPP A = 0.113, P = 0.028) and N. pulchraA. barbata (MRPP A = 0.088, P = 0.027), but not N. pulchraB. hordeaceus (MRPP A = −0.014, P = 0.584). In N. pulchraV. microstachys (Fig. 2a) and N. pulchraA. barbata (Fig. 2b), there was strong separation of root AMF communities between conspecific neighborhoods. In addition, both of these pairs showed greater similarity of root AMF in heterospecific neighborhoods to the root AMF composition of the annual monoculture, particularly for A. barbata. For N. pulchraB. hordeaceus (Fig. 2c), there was substantial overlap of root AMF among all combinations (Fig. 2c). Neither host nor neighbor was a significant main effect in the paired comparisons with these species.

Figure 2.

Nonmetric multidimensional scaling ordination of arbuscular mycorrhizal fungal (AMF) communities in Nassella pulchra–annual pairs, comparing monocultures and mixtures of (a) N. pulchraVulpia microstachys, (b) N. pulchraAvena barbata, and (c) N. pulchraBromus hordeaceus. For each graph, host and neighbor are plotted as the average (±1 SE) ordination coordinates based on samples in each treatment and are identified by the first initials of their genera (e.g. NV = N. pulchra host, V. microstachys neighbor), with monocultures represented by open symbols and perennial–annual mixtures by closed symbols. The r2 values for axis 1 are indicated in the bottom right-hand corner of each panel. AMF operational taxonomic units (OTUs) that were associated with r2 > 0.40 on at least one of the axes are plotted. Note that in (a), GlomIX perfectly overlaps GlomX (not shown) and GlomII perfectly overlaps GlomVII (not shown). Final stress values for the two-dimensional solutions were (a) 8.62, (b) 11.73, and (c) 11.33. The interaction of host and neighbor was significant in multiple response permutation procedures for N. pulchraV. microstachys (A = 0.113, P = 0.028) and N. pulchraA. barbata (A = 0.088, P = 0.027), but not N. pulchraB. hordeaceus (A = −0.014, P = 0.584).

Neighbor identity significantly altered the AMF community in N. pulchra roots (MRPP A = 0.102, P = 0.009) in comparisons of conspecific with heterospecific neighborhoods (Fig. 3). With V. microstachys neighbors, the N. pulchra root AMF community was unique and highly variable compared with all other neighborhoods along NMS axis 1. With A. barbata neighbors, the N. pulchra root AMF community also separated from other neighborhoods along both NMS axes and was less variable than with V. microstachys neighbors. By contrast, N. pulchra root AMF did not change when grown with B. hordeaceus compared with a conspecific neighbor.

Figure 3.

Nonmetric multidimensional scaling ordination of annual neighbor effects on Nassella pulchra arbuscular mycorrhizal fungal (AMF) communities. Each combination of N. pulchra host grown with an annual neighbor is plotted as the average (± 1 SE) ordination coordinates based on samples in each treatment. AMF operational taxonomic units (OTUs) that were associated with r2 > 0.40 on at least one of the axes are plotted. Final stress for a two-dimensional solution was 11.91. Neighbor was a significant factor in multiple response permutation procedures (A = 0.102, P = 0.009).

The differences observed at the community level are reflected in differential fidelity of AMF OTUs to hosts and neighborhoods revealed by the indicator species analysis (Table 2) and the NMS analyses (Figs 1–3). GlomVI appeared significantly more often in N. pulchra roots when it was grown in conspecific vs. heterospecific neighborhoods (Table 2). Avena barbata was closely associated with several AMF OTUs regardless of neighbor, including Glom/Acau, GlomII, GlomVII, GlomVIII, and GlomXII (Table 2, Figs 1, 2b). Moreover, GlomII appeared more frequently in A. barbata roots and in N. pulchra roots when grown with an A. barbata neighbor (Table 2, Fig. 3). Roots from monocultures of V. microstachys had GlomIX, GlomX, and GlomXII significantly more often compared with any other N. pulchra–V. microstachys combination (Table 2, Fig. 2a). No AMF OTUs were preferentially associated with N. pulchra–B. hordeaceus monocultures and mixtures, consistent with the high overlap among these treatments in NMS plots (Table 2, Fig. 2c).

Table 2.  Dufrene–Legendre indicator species analyses for arbuscular mycorrhizal fungal (AMF) operational taxonomic units (OTUs)* associated with plant hosts in monocultures and Nassella pulchra in paired mixtures with annual neighbors (Vulpia microstachys, Avena barbata and Bromus hordeaceus)
  • *

    We report the relative abundance of each AMF OTU. Bold type indicates P < 0.05. Italicized type indicates P < 0.10.

Host monocultures
Nassella 0.620.850.590.320.350.680.530.820.530.380.47
Vulpia 0.830.670.670.670.830.580.670.750.670.670.58
Avena 1.000.791.000.500.570.790.931.000.930.930.79
Bromus 0.770.540.850.390.540.300.620.690.770.620.54
Host × neighbor interaction: Nassella pulchra and Vulpia microstachys
Host × neighbor interaction: Nassella pulchra and Avena barbata
Host × neighbor interaction: Nassella pulchra and Bromus hordeaceus
Total abundance0.650.660.630.370.440.540.560.710.580.500.49

AMF colonization of plant roots

All plant roots were > 88% colonized by AMF hyphae, confirming that the time-frame of the experiment was sufficient for colonization. Across hosts in conspecific neighborhoods, root colonization of N. pulchra by hyphae was slightly greater than that of all other host plants (P < 0.010; Fig. 4a). Vulpia microstachys had the lowest rate of arbuscule colonization (P < 0.001; Fig. 4b). Vesicles were less common in N. pulchra and V. microstachys than in A. barbata and B. hordeaceus, with A. barbata and B. hordeaceus having nearly double the rate of vesicle colonization found in N. pulchra and V. microstachys (P < 0.001; Fig. 4c).

Figure 4.

Arbuscular mycorrhizal fungal (AMF) root colonization of plant hosts in monoculture (a–c) and in paired mixtures of Nassella pulchra and annuals (d–f) for hyphae (a, d), arbuscules (b, e), and vesicles (c, f). Bars are means ± 95% confidence intervals. Letters indicate significant differences among treatments (P < 0.05) in post hoc comparisons; lines are additionally used to designate cases where only host was a significant main effect.

Both host and neighbor identity influenced root colonization in N. pulchra–annual pairs (Fig. 4d–f). Hyphal colonization was affected by neighborhood, with significant interactions of host and neighbor in all three pairs (P ≤ 0.035; Fig. 4d). While colonization in N. pulchra did not change, hyphal colonization in the annuals depended on the neighborhood: V. microstachys and B. hordeaceus had 4% greater, while A. barbata had 6% fewer hyphae in roots when the neighbor was N. pulchra rather than conspecifics (Fig. 4d). By contrast, root colonization by arbuscules was significantly affected only by the main effects of host and neighbor in N. pulchraV. microstachys (P < 0.004) and N. pulchraA. barbata (P < 0.001), and only by that of the host in N. pulchra–B. hordeaceus (P < 0.022). Arbuscules in heterospecific neighborhoods were two to three times more abundant in N. pulchra hosts than in annuals (Fig. 4e). Vesicles responded to the experimental combinations only in N. pulchra–B. hordeaceus pairs, with main effects of both host (P = 0.002) and neighbor (P < 0.001). Vesicle colonization increased when B. hordeaceus was either the host or the neighbor (∼8%) compared with N. pulchra (∼5%; Fig. 4f). However, colonization of N. pulchra overall was unaffected by the identity of the annual neighbor (P > 0.322).

Plant biomass

Across hosts in conspecific neighborhoods, N. pulchra was smaller than all three annual grasses (P < 0.001; Fig. 5a). When grown with annual neighbors, N. pulchra biomass was reduced by the presence of A. barbata and B. hordeaceus, but not V. microstachys (P < 0.001; Fig. 5b,c). We found annual neighbor effects on N. pulchra biomass both in paired comparisons (Fig. 5b) and in comparisons of N. pulchra in monoculture with heterospecific neighborhoods (Fig. 5c). Neighbor also had a significant effect on the size of A. barbata (P = 0.024) and B. hordeaceus (P < 0.001), with greater biomass being found when these annuals were in heterospecific neighborhoods with N. pulchra than when they were in monoculture (Fig. 5b). Vulpia microstachys biomass did not change when it was grown with N. pulchra neighbors (P = 0.344; Fig. 5b). Shoot biomass and root biomass were significantly correlated for all species in this study (r2 = 0.64–0.92), and thus shoot biomass should scale predictably for whole-plant biomass.

Figure 5.

Plant biomass for (a) hosts in monoculture, (b) Nassella pulchra–annual pairs, and (c) N. pulchra with different neighbors. Bars are means ± 1 SE. In (a) and (c), different letters are used to designate significant differences in post hoc comparisons (P < 0.05). In (b), significant differences are indicated by lines where only host was a significant main effect, by lines and letters where both host and neighbor were significant main effects, and by letters only when the interaction of host and neighbor was significant. Bar colors indicate treatments as in Fig. 4.


We found that the plant neighborhood was as important as the identity of a plant host in structuring root AMF communities. Top-down effects were specific to the host and neighbor, with the presence of some neighbors affecting the root AMF community more strongly than others. These belowground interactions were asymmetric and varied independently of plant biomass. Based on these results, we conclude that host specificity is neighbor-specific, with the local plant neighborhood as much a factor in root AMF communities as the individual plant host. Neighbor effects included skewing towards AMF of one plant host (controlling), creating a novel root AMF community not previously present in either host (synergistic), and having no effect (neutral). The potential exists for a hierarchy of neighbor effects on AMF that may be somewhat predictable from aboveground positive and negative interactions between neighbors. Distinguishing the relative importance of plant hosts and neighbors in terms of their effects on AMF will help us to understand how interactions among plant neighbors play a role in plant–AMF feedbacks and processes such as invasion and succession.

Unique AMF communities were associated with roots of different hosts. For example, in plant monocultures, AMF in the perennial N. pulchra were different from those in the annuals, and among the annuals, native V. microstachys AMF were different from AMF in the exotics A. barbata and B. hordeaceus. We tested hosts at a single time of year on a single soil type collected from an area largely dominated by exotic grasses, which may limit the available pool of AMF and mask variation that would otherwise be observed in the field (e.g. Husband et al., 2002; Lekberg et al., 2007). In using the presence/absence of AMF OTUs, we also could be looking at differences among our communities based on some AMF with low abundance. Nevertheless, the finding that different hosts support different root AMF communities is consistent with other observations of co-occurring plants supporting significantly different AMF (e.g. Vandenkoornhuyse et al., 2002, 2003; Johnson et al., 2004). The host specificity of root AMF observed in monocultures, however, did not consistently extend to mixtures.

AMF communities in plant roots were altered in both composition and colonization by the presence of certain neighbors. Moreover, the symmetry of reciprocal neighbor effects on root AMF varied with each neighbor pair. Nassella pulchra and, to a lesser extent, V. microstachys had distinct root AMF when in monoculture, but the AMF in N. pulchra and V. microstachys roots converged on a new, synergistic community when the two species were grown together. The root AMF community in N. pulchraV. microstachys mixtures was, however, more similar to the root AMF in the V. microstachys monoculture. A stronger effect of the annual was observed for N. pulchra–A. barbata mixtures, where A. barbata dominated the root AMF community by functionally converting N. pulchra AMF to an A. barbata AMF community; N. pulchra did not have the reciprocal effect on A. barbata. By contrast, AMF communities in B. hordeaceus and N. pulchra roots were largely invariant across treatments so that the presence of a neighbor was essentially neutral. Other studies have found changes in AMF communities during plant succession (Johnson et al., 1991; Husband et al., 2002) and exotic plant invasions (e.g. Mummey et al., 2005; Batten et al., 2006; Hawkes et al., 2006; Mummey & Rillig, 2006; Stinson et al., 2006) that also partly represent changes in plant neighborhoods, although this was not tested explicitly.

The composition of the plant neighborhood beyond the host clearly matters to overall AMF community composition, with individual plant neighbors able to cause community-wide shifts in root AMF. The mechanism of these effects may be either abiotic or biotic. Certain plants may change soil nutrient pools or other abiotic soil characteristics that affect AMF. We did not measure soil properties in this study, but some labile pools could have changed over the course of 10 wk. Based on the patterns of plant neighbor effects on biomass, however, we suspect that direct biotic effects were more important here. For example, effects of the three annual neighbors on N. pulchra root AMF were proportional to their effects on N. pulchra biomass. Differential effects of annual neighbors may thus be related to interaction strength, either positive or negative. A hierarchical model of neighbor effects is further supported by similarly strong shifts in root AMF observed in previous fieldwork with mixtures of A. barbata, B. hordeaceus, and N. pulchra, which the current results suggest were probably attributable to A. barbata only (Hawkes et al., 2006). More work is necessary to understand AMF controllers in the presence of multiple neighbors.

In the context of a plant neighborhood, we can begin to hypothesize about the AMF niche of host plants and the host niche of AMF. In N. pulchra roots, distinct AMF communities emerged depending on neighbor identity, suggesting that the plant has a broad AMF niche. Both exotic annuals appear to have a narrower AMF niche in these soils, given the similarity of their root AMF communities in conspecific and heterospecific neighborhoods. At the same time, from the fungal perspective, N. pulchra may represent a preferred host for the standing AMF community in these grasslands. The preference for N. pulchra is supported by substantially greater arbuscule colonization of N. pulchra roots compared with the exotic annuals when a choice between them was provided in mixtures. When N. pulchra was present as an alternative host, arbuscule colonization decreased by 50% in B. hordeaceus and 85% in A. barbata compared with colonization when the exotic annuals were grown in monoculture. Arbuscule colonization can reflect functional differences in plant relationships with AMF, which is supported by a small negative correlation between arbuscule colonization and plant biomass (r2 = 0.16, P < 0.001) (but see StreitwolfEngel et al., 1997; Smith et al., 2003; Smith et al., 2004). Further study is needed, however, to examine interactions of the annuals with multiple neighbors. Recent evidence suggests that some AMF have broader niches and are more generalist than others (Helgason et al., 2007; Lekberg et al., 2007). As we learn more about the ecology of individual AMF taxa, we can begin to explore the generalist or specialist nature of AMF species and their host associations (Johnson et al., 2006; Kiers & van der Heijden, 2006).

Although we did not explicitly test for effects of exotic and native neighbors, our results are consistent with a native–exotic split in terms of neighbor effects on root AMF and support a growing body of literature that shows that exotic plants can alter the composition of AMF communities (e.g. Mummey et al., 2005; Batten et al., 2006; Hawkes et al., 2006). The native annual neighbor V. microstachys had a synergistic effect on N. pulchra root AMF compared with the controlling or neutral effects of exotic annual neighbors. There was also no preference for AMF root colonization of N. pulchra when it was grown with the native annual – instead, arbuscule colonization increased slightly in the presence of N. pulchra while V. microstachys biomass did not change. The apparent preference of AMF for natives is surprising given the long-term prevalence of exotic grasses in this system and points to an evolutionary context for plant–AMF interactions that may underlie differences in host and neighbor effects between natives and exotics.

The absence of mutualists can prevent the establishment and naturalization of introduced species (Parker et al., 2006) and, conversely, the presence of mutualists may facilitate invasions (Richardson et al., 2000). We found that the exotics were able to associate with many of the same AMF species as the native grasses in our study and their biomass increased when they grew with the native N. pulchra, leading us to speculate that the exotics are able to exploit the pool of native AMF. Of course, this study represents a snapshot taken > 100 yr after the initial arrival of these exotic grasses – the exotic grass interactions with local AMF may have been very different upon arrival, their interactions may have changed with time, or the AMF community may have been altered by long-term annual dominance. Moreover, in this experiment we used sieved, homogenized soils that disrupted the AMF soil network, which could change the relative abundance of AMF available to colonize roots. While we can study the current impact of exotic plants on the AMF community, to understand the role of AMF in the invasion process, it will be necessary to examine plant invasions as they occur in a field setting.

Local interactions among plant neighbors are as important as host effects in top-down control of root AMF communities. To understand the generality and predictability of these results will require identification of the underlying drivers. We suggest that neighbor effects on AMF reflect the strength of plant interactions (positive or negative) and that the degree of interaction strength reflects a hierarchy of effects on AMF. Whether alternative AMF communities are stable and how they affect plant fitness will partly determine local plant and fungal diversity and, ultimately, community trajectories (Chase, 2003). In the current study, it is unclear whether AMF preferences provide an advantage or disadvantage to N. pulchra, but the presence of neighborhood-dependent root AMF communities during N. pulchra establishment suggests that AMF could play a role in neighborhood interactions. Understanding the consistency of these AMF communities and their impacts on the establishment of native perennials may help guide restoration practices in the future.


The authors thank T. Bruns, T. Dawson, M. Firestone, W. Sousa, and three anonymous reviewers for comments on earlier versions of the manuscript. Help with laboratory work was provided by C. Benemann, V. Boukili, S. Kivlin, E. Limm, and D. Svehla. Y. Valles and D. Vieites were essential to the phylogenetic analyses. NH was supported by an NSF doctoral dissertation improvement grant (DEB 0508926) and a Dolores Zohrab Liebmann Fellowship.