Linking symbiont community structures in a model arbuscular mycorrhizal system


Author for correspondence:
James F. Meadow
Tel: +1 406 994 4227


  • The influence of plant communities on symbiotic arbuscular mycorrhizal fungal (AMF) communities is difficult to study in situ as both symbionts are strongly influenced by some of the same soil and environmental conditions, and thus we have a poor understanding of the potential links in community composition and structure between host and fungal communities.
  • AMF were characterized in colonized roots of thermal soil Mimulus guttatus in both isolated plants supporting AMF for only a few months of the growing season and plants growing in mixed plant communities composed of annual and perennial hosts. Cluster and discriminant analysis were used to compare competing models based on either communities or soil conditions.
  • Mimulus guttatus in adjacent contrasting plant community situations harbored distinct AMF communities with few fungal taxa occurring in both community types. Isolated plants harbored communities of fewer fungal taxa with lower diversity than plants in mixed communities. Host community type was more indicative than pH of AMF community structure.
  • Our results support an inherent relationship between host plant and AMF community structures, although pH-based models were also statistically supported.


Understanding the drivers and controls that structure biotic communities is one of the most fundamental goals in ecology (Begon et al., 2006), and a major focus of microbial ecology in recent years (Fierer & Jackson, 2006; Ramette et al., 2009). Soil microbes, in particular, exist in a complex heterogeneous environment and their presence is often a result, to varying degrees, of dispersal, environmental tolerance and ecological interactions (Martiny et al., 2006; Miransari, 2011; Unterseher et al., 2011). The relative importance of these controls is poorly understood for the majority of soil organisms, especially microorganisms (Fierer et al., 2009).

Arbuscular mycorrhizal fungi (AMF) form a root symbiosis with the majority of vascular plants in all terrestrial soil systems, and influence processes from the scale of individual microbial interactions to ecosystems (Smith & Read, 2008). The relationship between the two symbionts is generally regarded as mutualistic and primarily entails the exchange of plant photosynthates for immobile soil nutrients, especially phosphorus. Plants might receive additional benefits in improved water relations, pathogen resistance and improved survival when establishing in existing mycorrhizal networks (Newsham et al., 1995; Ruiz-Luzano, 2003; van der Heijden, 2004), and plant growth and community composition are affected by mycorrhiza and AMF community structure (van der Heijden et al., 1998, 2003). A small number of AMF taxa (< 300 described) appear to engage in mycorrhiza rather nonselectively with the majority of plant taxa; this taxonomically lopsided relationship might be one reason why AMF communities have previously been assumed to be somewhat less influenced by plant community composition. Two-way controls on communities have been shown using feedback experiments (e.g. Bever, 2002), and host plant identity, as well as biome, have been cited as important in affecting AMF community composition and diversity (Mummey et al., 2005; Hausmann & Hawkes, 2010; van de Voorde et al., 2010; Davidson et al., 2011). Recent advances in the molecular technology required to study AMF communities in situ have shown that AMF diversity and community structure can change with differences in soil conditions, such as pH, texture and nutrient levels, and that fungal communities are sometimes controlled more by environmental characteristics than by dispersal limitations (Öpik et al., 2006; Redecker, 2006; Lekberg et al., 2011). Meta-analysis of molecular AMF studies has revealed potential generalists and specialists within the Glomeromycota (Öpik et al., 2010), and Kivlin et al. (2011) reported that plant community type, soil temperature and moisture were all associated with changes in AMF community composition. The relative importance of host versus soil environment, however, is difficult to study in natural systems since plant and fungal communities both respond to variation in soil conditions; the result is that we have an incomplete understanding of the relative influences of soil conditions and host communities as drivers of AMF community structure and composition.

Geothermal soils in Yellowstone National Park (USA) are a unique example in soil formation, in that these soils are largely a product of geothermal influences, including steam and elevated temperatures, periodic inundation by chemically diverse geothermal waters, and rapid taphonomy of plant and microbial biomass by mineral-saturated waters; mineral precipitates are the primary parent materials for many of these soils (Rodman et al., 1996; Channing & Edwards, 2004). These complex soil formation factors result in a mosaic of highly variable soil conditions across a relatively small spatial scale. The heterogeneous and depauperate nature of biotic communities in these thermal systems presents an elegant study system for addressing ecological questions without some of the confounding influences of distance, atmospheric differences and dispersal barriers. Plants living in these soils often exist in conditions that are far beyond the tolerance limits of most vascular plants (Bunn & Zabinski, 2003), resulting in a subset assemblage of thermal-tolerant plants within the larger context of the Greater Yellowstone Ecosystem. One of these thermal-tolerant plants, a facultative-thermal, annual forb, is Mimulus guttatus, and, when M.  guttatus grows in geothermal soils, it is often heavily colonized by AMF (Bunn & Zabinski, 2003). Mimulus guttatus appears to have a wide tolerance for geothermal soil conditions and occurs in two very different plant community situations in geothermal soils: as a patch within an existing plant community consisting of grasses and other forbs, including annual and perennial plants (subsequently referred to as ‘communal’ sites; Fig. 1); and as individual, isolated plants emerging in essentially bare soil with few or no other plants (subsequently referred to as ‘isolated’ sites; Fig. 1). In addition, these disparate plant communities often occur within 10 m of one another, offering a unique opportunity to analyze the influence of plant community differences and edaphic conditions on the soil microbial community. Vectors for AMF dispersal have previously been identified in Yellowstone National Park thermal soils by Lekberg et al. (2011), indicating that dispersal limitations between paired sites are not likely to play a major role in structuring these communities. The reason for the difference in plant community types is not known.

Figure 1.

Paired site framework. All paired sites consisted of a communal (left) and an isolated (right) component. Hypothesized arbuscular mycorrhizal fungal (AMF) community differences are based on the discrepancy in carbon (C) supply (central panels) between these two host community situations, where communal sites might foster fungal networks throughout the growing season with some winter support, whereas isolated plant mycorrhizal support consists of C from a single plant during a short growing season window.

From the perspective of the obligate mycorrhizal fungal community interacting with M.  guttatus, these plant community types present two very different symbiont life history situations (Fig. 1). In the case of AMF colonizing hosts in mixed communities, host availability occurs during the majority of the year from different host plant species, and with some element of winter root support in the case of sites that remain unfrozen. For those in isolated sites, all host support comes in a relatively short several month window, the duration of the M.  guttatus life cycle, and symbiosis then shuts down for the remaining majority of the year. Given interspecific differences in sporulation and spore tolerance to harsh environmental conditions (Klironomos et al., 2001), the AMF communities in these two situations might differ either as a result of taxon-level variation in dispersal limitations (e.g. insufficient sporulation rates for some taxa to adequately disperse into isolated soils before M.  guttatus germination) or of spore tolerance to storage during the months between host life cycles. AMF taxa in both host community types experience some degree of harsh, geothermal soil conditions, although mixed plant host communities are present throughout most of the year and for the entire regular growing season, and this host community structure allows for year-round mycorrhizal networks rather than obligatory spore production and storage. The severe filter exerted by isolated plants existing for only 2 months in otherwise bare soil would, however, probably favor those AMF taxa that are either prodigious spore producers that are dispersing into isolated sites at some point before M.  guttatus germinates, or are especially tolerant of long-term storage as spores in geothermal soil conditions. In either case, the communities associated with these isolated sites are potentially a subset of AMF taxa occurring in communal plants, or, alternatively, a unique set of AMF that are selected for by conditions imposed by an isolated host, or a combination of the two. A substantive difference in AMF communities between host community types, if sufficiently independent of soil characteristics, would indicate an inherent link in symbiont community structures associated with taxon richness and some predilection among AMF taxa in terms of association with ephemeral or perennial plant community types.

In the present study, we assess the relative roles of pH and host community structure as controls on AMF community composition in this model system in geothermal soils in Yellowstone National Park by contrasting AMF communities in the roots of M.  guttatus living in either communal or isolated plant community types. We hypothesize that isolated M.  guttatus will harbor depauperate AMF communities in comparison with communal M.  guttatus, and fungal communities associated with isolated plants will be distinct from those associated with communal plants.

Materials and Methods

Plant and soil collection

Five paired sites were included in this study from Imperial Meadow (M1, M2 and M3; centered around 44°3304′′N, 110°5105′′W) and the Rabbit Creek (R1 and R2; 44°3055′′N, 110°4916′′W) drainage in Lower and Midway Geyser Basins, respectively, in Yellowstone National Park (WY, USA). Sites were selected based on the presence of appropriate adjacent paired patches of Mimulus guttatus DC. Isolated sites consisted of either sparse or clumped M.  gutattus plants with no other plant taxa growing within 1 m. All of these plants were small relative to optimal growth conditions (≤ 10 cm tall; Dorn, 2001) and, although mycorrhizal networks could conceivably extend beyond 1 m, the distance to surrounding plants helps to reduce this possibility. Between six and 20 plants were sampled individually from each patch (five pairs of patches), and sample number was determined by the patch size to avoid sampling > 20% of the individual plants from each patch. Plants sampled from a given patch were all with 10 m of one another; although M. gutattus density was comparable between isolated and communal sites, total plant density was inevitably higher in communal sites. Soil temperature was measured in the rooting zone of each plant, and whole plants were extracted with 5–10 g of rhizosphere soil, transported to the laboratory within 8 h and frozen (−80°C) until being processed.

For processing, individual plants were allowed to thaw at room temperature (c. 23°C) for 30 min before roots were carefully separated from soil. Soil was oven dried overnight (60°C) and analyzed for pH (Hendershot et al., 2008; 2 : 1 water extract), total carbon (C) and nitrogen (N) (LEKO combustion, St. Joseph, MI; Yeomans & Bremner, 1991) and texture (micro pipette extraction; Miller & Miller, 1987). As a result of the strong influence from adjacent geothermal spring features, pH is considered to be a driving variable in these soils (Rodman et al., 1996), and is preferentially used during analysis in this study. The use of pH as a primary variable in ecological analyses of soil microbial communities is also well established in non geothermal systems (e.g. Rousk et al., 2010).

DNA extraction and PCR

AMF colonization rates were estimated for plants from each site (Koske & Gemma, 1989), and washed roots were cut into small segments, with size (c. 0.5 cm) based on colonization rates in that more than half of all segments at a given length were colonized at least once. Eight root segments were randomly picked per plant for molecular analysis.

We extracted DNA from root pieces by first denaturing plant and fungal tissues in 80 μl 1 mM TBE (Tris/Borate/EDTA) buffer solution (95°C for 2 min), after which root pieces were manually crushed using sterile micro pestles. A Chelex 100 suspension (20 μl) was added and samples were vortexed and placed on ice for 2 min before a second denaturation, a final vortexing and a final icing. Samples were then centrifuged (5000 RCF for 5 min) to pellet cell contents, and the supernatant was drawn and diluted 50 times for use as template in nested PCR.

The first PCR was conducted with eukaryote-specific primers (NDL22-0061) to target the variable D2 region of the large ribosomal subunit, and the second PCR utilized the Glomeromycota-specific FLR3–FLR4 primer pair (van Tuinen et al., 1998; Gollotte et al., 2004). Even though this primer combination can amplify non-Glomeromycotan sequences, and some AMF groups might be missed (Mummey & Rillig, 2007), we detected taxa from most major AMF groups and no non-Glomeromycetes. All PCR was performed using GoTaq Green Master Mix (Promega, Madison, WI, USA), and with the following thermocycling conditions for PCR-1: 1 min at 95°C; 30 cycles of 1 min at 95°C, 1 min at 53°C and 1 min at 72°C adding 4 s to the elongation step for each cycle; followed by a final elongation of 5 min at 72°C. The PCR-2 program was the same, but with an annealing temperature of 56°C and only 25 cycles without the stepped elongation times. All PCR was performed with 50 μl volume, including 2 μl of 50 times diluted DNA extract or 50 times diluted PCR-1 product, for PCR-1 and PCR-2, respectively. This sequencing approach, including small-root-segment PCR coupled with direct sequencing, was used to avoid cloning and to attempt to capture single colonizations. The trade-off associated with direct sequencing is that more root segments can be analyzed with less effort and cost compared with cloning and sequencing, although mixed sequences are lost. Using this approach, approximately half of all root segments are expected to result in positive PCR products. In addition, the majority of these positive products are expected to return singular AMF sequences. The first indication of multiple AMF sequences in the same PCR product is the appearance of multiple bands on agarose gel, although some combinations of AMF taxa produce segments that are of nearly equal length. Thus, PCR products were visualized using electrophoresis on agarose gel, and products showing single bands were cleaned using the QIAquick PCR Purification Kit (Qiagen, Chatsworth, CA, USA) and submitted for direct sequencing at the Idaho State University Molecular Research Core Facility ( Some PCR products, even those with apparently singular bands on agarose gel, were composed of multiple sequences, and these were filtered out on sequence trace visualization.


All sequence traces were manually screened using FinchTV version 1.4.0 for Mac (, and sequences showing signs of contamination, indicating mixed colonizations, were discarded. All screened sequences were queried using BLAST (Altschul et al., 1990). Operational taxonomic units (OTUs) were designated by clustering at 97.5% sequence similarity based on a neighbor-joining tree; representative sequences from each taxon cluster were used in final alignment and phylogenetic reconstruction. Although 97% similarity effectively separated most OTUs, a bimodal clade within Glomus A was resolved at 97.5%, and this had identical results for all other OTUs. As our rDNA dataset was relatively small, and as known difficulties in assigning AMF OTUs based on sequence similarities occur with all candidate primer sets, we felt that manual clustering and assessment of an effective similarity cutoff provided a more conservative approach to OTU selection than the use of automated clustering algorithms and a hard similarity cutoff. This also preserves potentially novel OTUs, as AMF in Yellowstone National Park thermal soils have not been extensively studied. Close GenBank sequences, as well as INVAM and BEG isolates, were included with each OTU representative to better distinguish the clades. Sequences were aligned with ClustalW in Mega 5.0 for Mac (, and maximum likelihood model tests were performed using PhyML (PhyML 3.2 for Linux; Paradis et al., 2004; Guidon et al., 2010) in R (version 2.11.1 for Linux; R Development Core Team, 2010) with the ape package (Paradis et al., 2004). Bayesian inference phylogenetic trees were produced using BEAST ( for Linux and phylogenetic tree images were produced in R using ape. Trees were run under a general time-relaxed model with a five-parameter Γ site distribution and with some sites assumed as invariant (G+Γ+I ) based on PhyML results (Akaike information criterion (AIC) = 15 928). Trees were run for 107 iterations, saving every thousandth result, and compiling after a 20%‘burn-in’. Tentative OTU names were assigned based on a newly proposed AMF phylogeny (Schüßler & Walker, 2010; Krüger et al., 2011). Partial rRNA sequences from these samples were deposited in GenBank under accession numbers JN836499, JN8365019 and JN83651128 (Supporting Information Table S1).

Statistical analysis

Relative abundance data were calculated for each fungal taxon in each site by combining presence in plants and applying a sample total standardization (Notes  S1, S2); sample total standardized data were used for all analyses other than initial rarefaction. All statistical analyses were performed in R. Network visualizations were performed using the bipartite package (Dormann et al., 2008), and individual sites were summarized with α-diversity metrics (richness, Shannon–Wiener diversity and Shannon–Wiener evenness). Tests of these indices were performed as Kruskal–Wallis rank-sum tests. We tested for a bias in sampling effort by comparing the total number of sequences per site, the number of plants that returned clean AMF sequences per site and the average numbers of sequences per plant for each site, all with two-sample t-tests. We also created species accumulation curves for each community type using the ‘rarefaction’ method implemented in the vegan package (Oksanen et al., 2011).

Raup–Crick probabilistic dissimilarity values were computed for multivariate analyses as instituted in the vegan package (Raup & Crick, 1979; Chase et al., 2011). Agglomerative hierarchical clustering was conducted using the flexible-β approach (αı =  −0.625) to cluster sites by community similarity. Silhouette and heat plots were created with the optpart package (Roberts, 2010) to assess goodness-of-clustering for each solution. Analysis of similarities (ANOSIM), to assess significance of clustering solutions, was performed with the vegan package. Four clustering solutions were tested with agglomerative hierarchical clustering, and these correspond to Fig. 4: (a) community type; (b) bimodal pH, with M2c in the high-pH cluster; (c) balanced ranked pH, exchanging M2c and M1i; and (d) unbalanced ranked pH, including M3i in the low-pH cluster. The last three clustering solutions (b, c and d) represent three different views of two-cluster solutions based on pH rather than community type.


Soil analysis

All soil physical and chemical characteristics measured, with the exception of soil temperature, were predictably correlated with pH (Pearsons correlation coefficient r = 0.72, 0.64 and 0.57 for pH compared with total N, total C and clay, respectively) and none were distributed as evenly as pH (Table 1). Although temperature has previously been explored as a primary driver of plant and fungal communities in Yellowstone National Park thermal soils and has been noted to fluctuate during growing seasons (Bunn & Zabinski, 2003; Bunn et al., 2009), the soils used in this study had a relatively narrow range of average temperature, from 24.2 to 29.8°C. This narrow range, and the fact that we collected soil temperature data at only a single time point, did not allow the complete investigation of temperature as a primary gradient of interest in driving AMF community patterns in this study. There was also no difference in average soil temperature between the two community types (t = −0.62, P = 0.56; from a two-sample t-test). Isolated plant communities were all found growing in high-pH soils (mean of 8.97) with low percentages of total C and N (mean of 1.17% and 0.16%, respectively), whereas plants growing in communal sites were generally in more neutral to slightly acidic soils with a single notable exception. Although pH is quite different between plant community types (P = 0.015; Table 1), one communal site, M2c, has a high mean pH (pH 9.0) which is more consistent with isolated sites in this study, and its other measured soil characteristics follow suit.

Table 1.   Mean soil parameter values (values represent averages of measurements (± SE) from rooting zone soil under individual plants within communal or isolated sites)
SiteCommunity typeNo. of PlantsNo. of SequencesSoil temp (°C)pHa%Nb%Cb% Sandc% Siltc% Clayc
  1. Reference method used: aHendershot et al. (2008); bYeomans & Bremner (1991); cMiller & Miller (1987).

M1Isolated81726.75 (± 0.49)8.12 (± 0.11)0.21 (± 0.01)1.09 (± 0.05)84.38 (± 0.84)14.50 (± 0.76)1.12 (± 0.23)
M2Isolated144229.79 (± 1.13)9.75 (± 0.10)0.09 (± 0.00)0.89 (± 0.02)89.50 (± 0.54)10.50 (± 0.54)0.07 (± 0.07)
M3Isolated51324.60 (± 0.68)8.67 (± 0.24)0.15 (± 0.01)1.04 (± 0.04)74.60 (± 1.78)22.60 (± 1.36)2.60 (± 0.68)
R1Isolated113325.91 (± 0.51)9.34 (± 0.10)0.18 (± 0.01)1.54 (± 0.06)76.45 (± 1.40)22.82 (± 1.20)0.73 (± 0.27)
R2Isolated184424.22 (± 0.55)8.99 (± 0.07)0.18 (± 0.01)1.22 (± 0.04)78.72 (± 0.63)20.22 (± 0.57)1.22 (± 0.10)
M1Communal5928.44 (± 0.80)6.18 (± 0.22)0.33 (± 0.08)3.41 (± 1.26)81.57 (± 1.67)16.86 (± 1.32)1.57 (± 0.37)
M2Communal41026.25 (± 1.03)9.00 (± 0.12)0.19 (± 0.02)1.64 (± 0.19)84.00 (± 2.35)15.75 (± 2.53)0.50 (± 0.29)
M3Communal91925.22 (± 0.36)6.36 (± 0.15)0.29 (± 0.02)2.63 (± 0.18)65.55 (± 0.87)31.45 (± 1.05)3.18 (± 0.31)
R1Communal82127.38 (± 1.19)6.97 (± 0.27)0.53 (± 0.06)7.01 (± 0.93)58.00 (± 2.04)39.12 (± 1.92)2.88 (± 0.52)
R2Communal132026.77 (± 0.68)6.72 (± 0.22)0.45 (± 0.02)4.95 (± 0.26)53.67 (± 1.92)36.20 (± 1.32)10.53 (± 0.95)

AMF molecular identification and phylogeny

DNA was extracted from 1128 root segments for PCR, and, by design, approximately half of all root segments showed positive PCR products on agarose gel, 478 of which showed singular bands and were submitted for sequencing. After manually quality filtering for mixed sequences and putative chimeras, 228 sequences were used for final analysis. These were individually searched using BLAST and assigned tentative taxonomic identifiers based on nearest BLAST result or known INVAM or BEG isolate. Pairwise alignments and phylogenetic construction with sequences from known isolates resulted in 28 OTUs in eight genera. Posterior support for every genera-level clade is 1.0, and similarly high support is consistent throughout most major phylogenetic groupings (Fig. 2). We also found no difference between isolated and communal sites in the number of plants returning AMF sequences (t = −1.23, P = 0.26; from a two-sample t-test), but communal plants did harbor more OTUs per plant (1.69 and 1.27 OTUs per communal and isolated plant, respectively; t = −2.38, P = 0.056; from a two-sample t-test of average OTUs per plant in each site). Isolated plants returned an average of 0.52 more clean sequences per plant than communal plants (mean sequences per plant, 2.63 and 2.11 for isolated and communal plants, respectively; t = 1.96, P = 0.09; from a two-sample t-test), and an average of 14 more clean sequences were found per site in isolated sites as a result (29.8 and 15.8 sequences per site for isolated and communal sites, respectively; t = 2.04, P = 0.09; from a two-sample t-test). These tests together indicate an approximately even sampling effort across community types in terms of the number of plants sampled, but the large difference in the number of sequences returned per site is perhaps an indication of the larger number of mixed colonizations in communal plants that were discarded (Fig. S1a,b). More sequences per plant and per site could potentially bias richness and diversity towards isolated plants, but this was certainly not the case in the present study, as communal plants were far more OTU rich and diverse. It is also likely, given the number of singletons detected in communal plants, that this sequencing approach resulted in an underestimate of OTU richness in communal plants, as indicated by the steep terminus of the communal species accumulation curve (Fig. S1c).

Figure 2.

Bayesian inference phylogeny. Arbuscular mycorrhizal fungal (AMF) taxa used in these analyses are shown in bold and with points at branch tips indicating plant community affiliation; black points were found only in communal sites, white were found only in isolated sites, and gray were found in both community types. Clade bars indicate genera sensuSchüßler & Walker (2010) and Krüger et al. (2011). Names assigned during this study follow this convention, whereas isolates used for alignment retain names given in the National Center for Biotechnology Information (NCBI) database. Node support values are posterior probabilities resulting from 107 iterations, saving every thousandth result, and compiling after a 20%‘burn-in’. An ascomycete (Mortierella polycephala; NCBI accession number AF113464) was used as an out group and is not shown.

AMF diversity and community composition

Of the AMF taxa found, 18 were found only in communal sites and seven only in isolated sites, whereas only three taxa were found in both site types, including Rhizophagus intraradices and Funneliformis mosseae (synonymous with Glomus intraradices and Glomus mosseae, respectively; Schüßler & Walker, 2010; Krüger et al., 2011). Bipartite network visualization of the dataset (Fig. 3) reveals a striking pattern of diversity differences between communal and isolated sites, regardless of their spatial proximity in the study area. Sites and fungal OTUs (boxes along the top and bottom, respectively, of Fig. 3) are ordered using a CCA-based χ2 algorithm to reduce the number of interaction cross-overs, and are thus an indication of underlying relationships. Isolated sites appear to be dominated by only a single or two fungal OTUs (represented by the number and relative width of connections in Fig. 3), whereas communal sites show little sign of domination by any one OTU, but, rather, a more even and diverse fungal community. Kruskal–Wallis rank-sum tests of taxon richness, Shannon–Wiener diversity and Shannon–Wiener evenness all conclusively illustrate these differences (Fig. S2). Taxon richness in communal sites is twice as high as that in isolated sites, with 3.8 more fungal taxa in communal sites (χ2 =  7.45, P = 0.006), and Shannon–Wiener diversity and evenness are 1.9 and 1.2 times higher, respectively, in communal than in isolated sites (χ2 = 6.82, P = 0.009 for diversity and χ2 =  5.77, P = 0.016 for evenness). It is notable here that the single uncharacteristically high-pH communal site (M2c) has diversity indices that are all consistent with other communal sites, rather than with comparably high-pH isolated sites (Fig. 3).

Figure 3.

Full bipartite network visualization of all sites in the study. Boxes above are sites composed of all plants sampled from that site. Boxes below are individual arbuscular mycorrhizal fungal (AMF) operational taxonomic units (OTUs). Interaction width is proportional to the relative abundance of fungal OTUs found at each site after sample total standardization (Supporting Information Notes S1, S2). White sites were isolated, black sites were communal. White OTUs were found only in isolated sites, black were found only in communal sites, and gray were found in both site types. The order of sites and OTUs in the figure is a result of a CCA-based χ2 algorithm that minimizes the number of interaction cross-overs.

Cluster and discriminant analysis

Two different clustering approaches were employed for comparison: one relying on community dissimilarity data using the Raup–Crick probabilistic metric, and another on environmental data. The first strategy resulted in clustering consistent with plant community type (Fig. 4a), but, when pH was used instead, clusters were created with plant community types mixed in two-cluster solutions. The community-type-based clustering solution was tested against three other possible combinations of pH-based two-cluster solutions with ANOSIM and, although all four models were supported to some extent, the community-type solution showed the lowest P value (P = 0.009), though the R statistic was perhaps more telling (Fig. 4). The R statistic from ANOSIM is based on the difference in mean ranks between groups and within groups, and ranges from −1 to 1, with ‘0’ denoting completely random grouping and positive values indicating systematic grouping (Legendre & Legendre, 1998). The amount of variation explained by the clustering solutions was highest for the community-based solution, although all pH-based solutions returned positive R statistics.

Figure 4.

Comparison of clustering solutions. Trees represent clustering solutions based on: (a) plant community type; (b) bimodal pH; (c) balanced ranked pH; and (d) unbalanced ranked pH. Clade bars on trees represent community type homogeneity within each cluster; green are communal sites, orange are isolated sites, and blue include both community types. Silhouette plots (second column) represent within-group goodness-of-clustering from Raup–Crick dissimilarity. ‘Set-to-Set’ heat plots indicate goodness-of-clustering across clusters, with white representing the highest degree of similarity and red representing the highest degree of dissimilarity. P values are from analysis of similarities (999 permutations); R statistic represents variance of dissimilarity explained by the clustering solution, with possible values from −1 to 1, where 0 equals random assignment and 1 equals systematic grouping.


Symbiotic organisms are limited not only by their own environmental tolerances, but also by the environmental tolerances of their potential symbionts, and this is especially true of plants and their obligately biotrophic AMF colonizers. Given the complexities of these relationships, the relative importance of either factor on the assembly of symbiotic communities is poorly understood. Our objective in this study was to elucidate drivers of AMF community composition and structure to assess a structural link between symbiont communities, and thus to assign relative influence to these drivers. Because the traits of plant and fungal communities tend to be driven by some of the same soil conditions, disentanglement of the relative influences from soil and ecological interactions becomes intractable. Thus, studies attempting to elucidate drivers of AMF community composition often focus on either of the two drivers: edaphic factors irrespective of plant community compositional properties (Schechter & Bruns, 2008; Lekberg et al., 2011), and across an edaphic gradient in which plant communities are constant (Wu et al., 2007), or, conversely as a factor of plant community compositional (van de Voorde et al., 2010) and structural (Börstler et al., 2006) differences across relatively constant edaphic conditions. Missing from this literature is an in situ comparison of the relative influences of soil and host community structure on AMF communities. The current study was an attempt to address this problem; the differences we observed in AMF communities were better predicted by host community type than by soil pH. Isolated plant communities used in this study were all found in high-pH soils, whereas only a single communal site (M2c) fit this description, and all other communal sites were found in near-neutral pH soils; thus, these two predictors were, unfortunately, nearly completely confounded. Although limited to a single site, and therefore insufficient for robust conclusions, the overlap between pH and plant community types in the high-pH communal site provided an opportunity to compare the relative influences of soil and plant community structure on AMF community composition.

We hypothesized that AMF communities existing in the roots of communal M.  guttatus would be distinct from isolated communities, such that a richer and more diverse AMF community would be found compared with the relatively depauperate AMF community living in isolated plant roots. The most striking example of such a difference in AMF community composition and structure can be seen in Fig. 3. Although taxon richness is clearly a factor in distinguishing community types, the dominance by one or two fungal OTUs in isolated sites is perhaps more salient for the identification of fungal community structural differences between isolated and communal sites. This characteristically even distribution was also observed in the lone communal site with more harsh soil characteristics (M2c), an indication that the effects of host vegetation type on fungal communities are more important than the influence of pH. Although the overlap in pH between isolated and communal sites was limited to a single example, this site sheds light on a larger pattern of AMF community assembly; these findings are consistent with van de Voorde et al. (2010), who showed experimentally that plant community assembly history can have a major influence on symbiotic communities living in the roots of these plant communities.

In addition, we found surprisingly little overlap between fungal communities in these contrasting host situations. Indeed, only three of the 28 fungal taxa were found in both plant community types, seven taxa were found only in isolated sites and 18 were found only in communal sites, even though paired sites are adjacent and can be assumed to have few barriers to AMF dispersal. Our findings can probably be attributed, at least in part, to island effects as AMF are not neutral in their dispersal abilities, although this sheds light on the autecological differences between AMF taxa, especially given the low degree of OTU overlap between community types (three OTUs), relative to OTUs found multiple times in only one community type (10 OTUs). One of the three fungal taxa found in both community types, R. intraradices (Rhi1), was found primarily in isolated sites and made only rare appearances in communal sites, indicating that this taxon is either well suited to high-pH soil conditions or most successful in short-lived plant communities where dispersal and aggressive colonization are favored. Given that this taxon is found world-wide in highly variable soil conditions (Rosendahl et al., 2009; Öpik et al., 2010), the latter seems more apt, and the fact that R. intraradices was differentially dominant in the two community types might indicate a more conserved ‘niche’ for this taxon than is often assumed. The other two of the three shared taxa show no particular affinity for either community type, and one of these, F. mosseae (Fun4), was among the most commonly detected fungal taxa in this study. This seems to contradict the idea that isolated AMF communities would be a subset of the surrounding communal assemblages, but rather lends credence to the hypothesis that AMF taxa that occur in isolated plants are particularly suited to these conditions, seemingly more so than to life in communal sites in which competition among other colonizing AMF might play a greater role, although some common taxa might be able to establish in either situation.

As this study was designed around the detection of differences in AMF community composition as a function of host community type, a major hurdle was the separation of soil characteristics, which probably play a major role in structuring these thermal plant communities, from the effects of the plant community structure itself. Host community type was only reasonably broken up into two clusters, and thus we presented a comparison of the community-based clustering solution with three different options for clustering based on ranked pH. Although all three pH-based clustering solutions resulted in positive R values, indicating that communities were clustered by pH better than expected by random assignment, even the most significant pH-based clustering solutions (Fig. 4b; R = 0.556; P = 0.015) accounted for substantially less of the variation than explained by the community-based solution (R = 0.728; P = 0.009), which was the most indicative of fungal community clustering. Clearly, these results cannot be interpreted to infer that pH has no effect on AMF communities as all four models show some degree of support. Indeed, ANOSIM tests of all three pH-based clustering solutions resulted in P values that were low; the R statistic provides a better interpretation of the goodness-of-clustering, and both measures (R statistic and the resulting P value) indicate that host vegetation type plays a role in structuring AMF communities and, in this case, is more closely associated with variation in AMF communities than is pH.

In addition to the modest soil condition overlap, the scope of our study was also limited by our use of only a single plant species at a given point in time, which was inherent in our study design; our choice of M.  guttatus was based on its ephemeral life history strategy and its appearance in completely isolated thermal soils near simultaneously emerging communal patches. Given that this study only explored the mycorrhizal community associated with a single plant species, more information will certainly be gained as more plant taxa are investigated for similar community structure links. Use of a single plant species helped to reduce the possibility that the differences in fungal community composition were simply a result of potential M.  guttatus specificity issues. As geothermal soils can be subject to seasonal and diurnal variation in environmental conditions, it is important to note that our study represents only a snapshot of conditions encountered during the flowering period of M.  guttatus. Differences in AMF community structure and composition are probably attributable to some combination of the three factors discussed thus far: long-term spore storage in harsh soil conditions limiting the establishment of some AMF taxa in isolated plant roots; a short isolated growing season limiting the accumulation of AMF taxa in isolated plant roots; and host specificity issues associated with surrounding plant taxa in communal sites. This study was not designed to disentangle these three potential contributions to fungal community characteristics, but rather to detect a community structure link between host and fungal communities.

By comparing the relative strengths of associations between AMF community and either soil chemical characteristics or plant community traits, we found that host community structure was more predictive of the observed differences in AMF community structure than was pH, and that AMF communities in isolated plant patches were less species rich, less even and less diverse. Although this is consistent with our first hypothesis, a competing pH-based model was also supported, and a more robust conclusion will rely on more overlap in soil conditions and more host species than just one.

This study adds to the growing body of evidence that plant communities exert some control over their associated AMF communities, and that AMF taxa are not neutral in their appearance in ecological communities. Our results also illustrate a potential link in symbiont community structure that might help soil ecologists understand AMF communities and their assembly patterns in both natural and managed systems.


We are grateful to the Yellowstone National Park permit personnel for help with research permits, to Ylva Lekberg for help with molecular methods and to Rosie Wallander for help with soil analysis. We also thank three anonymous referees for their valuable comments. J.F.M. was supported by a GK-12 Graduate Fellowship from the National Science Foundation and the Big Sky Institute, and by additional support from U. S. Department of Agriculture National Research Initiative Competitive Grant to C.A.Z.