Determinants of root-associated fungal communities within Asteraceae in a semi-arid grassland


  • Jeannine Wehner,

    1. Dahlem Center of Plant Sciences, Plant Ecology, Institut für Biologie, Freie Universität Berlin, Berlin, Germany
    2. Berlin-Brandenburg Institute of Advanced Biodiversity Research (BBIB), Berlin, Germany
    Search for more papers by this author
  • Jeff R. Powell,

    1. Hawkesbury Institute for the Environment, University of Western Sydney, Penrith, NSW, Australia
    Search for more papers by this author
  • Ludo A. H. Muller,

    1. Dahlem Center of Plant Sciences, Botanik, Institut für Biologie, Freie Universität Berlin, Berlin, Germany
    Search for more papers by this author
  • Tancredi Caruso,

    1. School of Biological Sciences, Medical Biology Centre, Queen's University Belfast, Belfast, UK
    Search for more papers by this author
  • Stavros D. Veresoglou,

    1. Dahlem Center of Plant Sciences, Plant Ecology, Institut für Biologie, Freie Universität Berlin, Berlin, Germany
    2. Berlin-Brandenburg Institute of Advanced Biodiversity Research (BBIB), Berlin, Germany
    Search for more papers by this author
  • Stefan Hempel,

    1. Dahlem Center of Plant Sciences, Plant Ecology, Institut für Biologie, Freie Universität Berlin, Berlin, Germany
    2. Berlin-Brandenburg Institute of Advanced Biodiversity Research (BBIB), Berlin, Germany
    Search for more papers by this author
  • Matthias C. Rillig

    Corresponding author
    1. Dahlem Center of Plant Sciences, Plant Ecology, Institut für Biologie, Freie Universität Berlin, Berlin, Germany
    2. Berlin-Brandenburg Institute of Advanced Biodiversity Research (BBIB), Berlin, Germany
    Search for more papers by this author


  1. While plant–fungal interactions are important determinants of plant community assembly and ecosystem functioning, the processes underlying fungal community composition are poorly understood.
  2. Here, we studied for the first time the root-associated eumycotan communities in a set of co-occurring plant species of varying relatedness in a species-rich, semi-arid grassland in Germany. The study system provides an opportunity to evaluate the importance of host plants and gradients in soil type and landscape structure as drivers of fungal community structure on a relevant spatial scale. We used 454 pyrosequencing of the fungal internal transcribed spacer region to analyse root-associated eumycotan communities of 25 species within the Asteraceae, which were sampled at different locations within a soil type gradient. We partitioned the variance accounted for by three predictors (host plant phylogeny, spatial distribution and soil type) to quantify their relative roles in determining fungal community composition and used null model analyses to determine whether community composition was influenced by biotic interactions among the fungi.
  3. We found a high fungal diversity (156 816 sequences clustered in 1100 operational taxonomic units (OTUs)). Most OTUs belonged to the phylum Ascomycota (35.8%); the most abundant phylotype best-matched Phialophora mustea. Basidiomycota were represented by 18.3%, with Sebacina as most abundant genus. The three predictors explained 30% of variation in the community structure of root-associated fungi, with host plant phylogeny being the most important variance component. Null model analysis suggested that many fungal taxa co-occurred less often than expected by chance, which demonstrates spatial segregation and indicates that negative interactions may prevail in the assembly of fungal communities.
  4. Synthesis. The results show that the phylogenetic relationship of host plants is the most important predictor of root-associated fungal community assembly, indicating that fungal colonization of host plants might be facilitated by certain plant traits that may be shared among closely related plant species.


Plant roots interact with a range of soil fungi, which can influence plant growth and fitness (Lu & Koide 1994; Raaijmakers et al. 2008; Smith & Read 2008; Porras-Alfaro et al. 2011), plant community composition (van der Heijden et al. 1998), as well as ecosystem functioning (Bradley, Gilbert & Martiny 2008; Fisher et al. 2012). Depending on the identities of the host plant and fungus, these interactions can be of a mutualistic, neutral or parasitic nature. In grasslands, arbuscular mycorrhizal (AM) fungi are dominant symbiotic fungal partners and are known to increase nutrient status (Marschner & Dell 1994), improve water relations (Auge 2001) and protect host plants against pathogens (Borowicz 2001; Wehner et al. 2010; Veresoglou & Rillig 2012). In addition to AM fungi, which extend into the rhizosphere, plant roots are often colonized by fungal endophytes that reside completely within plant tissues (Rodriguez et al. 2009). One example of these endophytes in grasslands is the ‘dark septate endophytes’, which have been shown to provide benefits to the host plant but whose ecological function remains unclear (Jumpponen 2011; Porras-Alfaro et al. 2011). However, due to their microscopic nature and the difficulty of isolating many root-associated fungi, a large fraction of species remains unknown (Peay, Kennedy & Bruns 2008) and molecular methods are often essential to describe these fungal communities.

In general, community dynamics can be driven by local environmental gradients as well as by large-scale processes that determine the colonization and extinction of species within a region (Chase & Leibold 2003; Ricklefs 2006). For micro-organisms, some have suggested that assembly processes are driven by a mixture of environmental filtering and random sorting from a regional species pool (Martiny et al. 2006). By contrast, if interspecific interactions among microbial taxa are important, assemblages will display non-random coexistence patterns (aggregation or segregation of species; Pan & May 2009).

Some of the local environmental factors that have been shown to affect fungal community composition include nitrogen availability, soil moisture and pH (Mulder & de Zwart 2003; Cox et al. 2010; Fujimura & Egger 2012). Root-associated fungal community composition also seems to depend on dispersal limitation (Lekberg et al. 2006) and on host species identity; many fungal pathogens of plants are host specific, whereas contradictory observations exist regarding host specificity of mycorrhizal fungi (Zhou & Hyde 2001). Evidence for host specificity has been found for AM fungi (Vandenkoornhuyse et al. 2003), for ectomycorrhizal (ECM) fungi (Morris et al. 2008) and also for endophytes in general (Wearn et al. 2012). Host range may also be linked to plant traits that facilitate infection and carbon/nutrient exchange; these traits may be constrained within phylogenetic lineages, resulting in covariance between host phylogenetic relationships and fungal species/communities (Espiau et al. 1997; Hibbett, Gilbert & Donoghue 2000; Gilbert & Webb 2007; Ishida, Nara & Hogetsu 2007; Refrégier et al. 2008). Once the effects of these various environmental and host filters have exerted their influence on the fungal assemblages, interactions among fungal species may further influence assembly processes (Pan & May 2009). Such interactions among fungi can be negative (competition) or positive (facilitation) or random (predominant environmental filtering). For example, a potential negative interaction among fungi has been observed in that the presence of AM fungi in roots was negatively correlated with endophyte colonization (Wearn et al. 2012). By contrast, Pan & May (2009) found mainly positive interactions by examining endophytic fungi only; host infections with one fungus increased the vulnerability to infection by other fungi.

In this study, which is the first investigating root-associated fungal (Eumycota) communities of co-occurring, phylogenetically relatively closely related plant species in grasslands, we performed 454 pyrosequencing of internal transcribed spacer 1 (ITS1) amplicon libraries to characterize fungal diversity in roots. As a next-generation parallel sequencing technology, 454 pyrosequencing increases the likelihood of detecting rare phylotypes due to the higher sampling depth, which would be difficult to reach with the classical cloning-Sanger sequencing methodology (Öpik et al. 2009). Specifically, we tested the hypothesis that host phylogeny, spatial structure and soil simultaneously affect the community composition of root-associated fungi of 25 different Asteraceae species in a semi-arid grassland. Since many ecological drivers exhibit spatial autocorrelation, we accounted for the potentially confounding effects of unmeasured drivers (dispersal limitation, other environmental variables) by explicitly estimating spatial structure in fungal communities, which allowed us to estimate the independent effects of host plant phylogeny and soil type. We also used null model analysis, testing whether species co-occurrence patterns follow a random distribution, to determine whether negative (segregation) and positive (aggregation) interactions among root-associated fungi may have observable effects on assembly processes.

Materials and methods

Description of the study site

All plants were collected from a site in the nature protection area ‘Oderhänge Mallnow’, located in north-eastern Germany approximately 120 km east of Berlin (52.4636° N, 14.4574° E). The site is a dry grassland habitat with over 200 plant species combining elements of both steppes and more mesic habitats, an annual precipitation of approximately 500 mm and a mean annual temperature of 8.7 °C (Deutscher Wetterdienst 2010). It can be characterized as Adonido-Brachypodietum or rather Potentillo-Stipetum with Adonis vernalis and Stipa capillata as character species (Hensen 1997). The area is part of a large (60 km long and up to 20 km wide), a post-glacial region with dry grassland habitats occurring along the Oder river (called ‘Oderbruch’). The investigated grassland is grazed by sheep twice a year. The plant species were collected within an area of approximately 2 km2 characterized by a gradient in soil type from sandy to loamy (see Fig. S1). The pH ranged from 5.5 in sandy areas to 8.3 in more loamy areas. The sandy areas had an average water content of 9.6% and an average C/N ratio of 19.4 compared to a water content of 16.6% and a C/N ratio of 15.9 in the loamy areas. The soil factor exemplified a relatively steep gradient within a small spatial scale. Despite our soil assaying strategy having been conducted at a crude scale, it allowed us to estimate the importance of edaphic properties, in general, in driving fungal community assembly at the relevant spatial scales. The landscape is also variable, with hills and valleys remaining from the final battles of World War II, which may have important consequences for fungal dispersal.

Sampling and DNA preparation

To ensure correct taxonomic identification, we sampled the 25 plant species of the Asteraceae and one species of the Campanulaceae (to serve as a reference taxon outside the Asteraceae; Table 1) during their flowering periods in either May, July or September 2010. We have chosen the Asteraceae because they are a large and diverse family of plants, often exhibiting high local diversity, making this an ideal group to target the questions that we address here. The family is at least 38 million years (Myr) old, but the fossil record suggests that it may be much older (43–53 Myr; Bremer & Gustafsson 1997).The distributions of the different plant species are very patchy, and three specimens of each species, randomly selected in an area of approximately 10 m2 and containing a single soil type, were sampled at a given time. Our intention was to address whether fungal community similarity was linked to the degree that host plants shared a common evolutionary history, which required us to establish a gradient of phylogenetic distances by sampling a large number of species. The important level of replication here was in the number of species, not the number of individuals within a species; replication within a species was performed as a means to determine reproducibility of host-associated communities.

Table 1. List of sampled plant species
Plant speciesFamily/Subfamily
Achillea pannonica ScheeleAsteraceae/Asteroideae
Anthemis tinctoria L.Asteraceae/Asteroideae
Artemisia campestris L.Asteraceae/Asteroideae
Aster linosyris (L.) Bernh.Asteraceae/Asteroideae
Bellis perennis L.Asteraceae/Asteroideae
Conyza Canadensis (L.) CronquistAsteraceae/Asteroideae
Erigeron muralis Lapeyr.Asteraceae/Asteroideae
Helianthus tuberosus L.Asteraceae/Asteroideae
Helichrysum arenarium (L.) MoenchAsteraceae/Asteroideae
Leucanthemum vulgare Lam.Asteraceae/Asteroideae
Senecio jacobea L.Asteraceae/Asteroideae
Senecio vernalis Waldst. & Kit.Asteraceae/Asteroideae
Solidago virgaurea L.Asteraceae/Asteroideae
Carlina vulgaris L.s.str.Asteraceae/Carduoideae
Centaurea jacea L.s.l.Asteraceae/Carduoideae
Centaurea scabiosa L.s.l.Asteraceae/Carduoideae
Centaurea stoebe L.s.l.Asteraceae/Carduoideae
Chondrilla juncea L.Asteraceae/Cichorioideae
Cichorium intybus L.Asteraceae/Cichorioideae
Hieracium pilosella L.Asteraceae/Cichorioideae
Hieracium umbellatum L.Asteraceae/Cichorioideae
Hypochaeris radicata L.Asteraceae/Cichorioideae
Leontodon hispidus L.Asteraceae/Cichorioideae
Picris hieracioides L.Asteraceae/Cichorioideae
Taraxacum sp. F.H. Wigg.Asteraceae/Cichorioideae
Campanula rotundifolia L.Campanulaceae

Whole plants were excavated, taking care that the root system was kept as intact as possible, and for each individual plant, the geographical location and the local soil type were recorded. After excavation, the roots were immediately cooled in the field and afterwards stored at −20 °C until further processing. Only plant species for which a permit was obtained were sampled. Roots were gently washed with deionized water to remove adhering soil particles. We did not surface-sterilize the root systems since we were concerned that interspecific variation in root anatomy could result in bias in the efficacy of the procedure. We acknowledge that rhizosphere-associated fungi (in contrast to those directly interacting with the roots) may have been sequenced but argue that these taxa are much less abundant relative to those directly associated with the roots and the likelihood of being represented in the resampled communities is very small. Roots were cut into 0.5 cm pieces, randomly chosen and directly filled in the 2 mL bead tubes (approximately 50–100 mg) provided as part of the PowerSoil DNA Isolation kit (MO BIO Laboratories Inc., Carlsbad, CA, USA) without any further processing. The DNA was extracted following the manufacturer's instructions except for an initial incubation at 65 °C for 10 min followed by a 10-min vortexing step. A soil DNA isolation kit was preferred to a plant tissue DNA isolation kit as preliminary tests indicated that it provides DNA of higher purity although it yields a lower DNA concentration (data not shown).

For pyrosequencing, triplicate PCR amplifications of the fungal ITS region covering ITS1, 5.8s rDNA and ITS2 were performed for each sampled individual in 50 μL reactions, each containing 5 ng template DNA, 50 μm of each deoxynucleoside triphosphate (dNTP), 200 nm of each of the forward (ITS1F; Gardes & Bruns 1993) and the reverse primer (ITS4; White et al. 1990) and 0.5 U Taq DNA polymerase (1000 U; Fermentas, St. Leon-Rot, Germany) in 1× PCR buffer (GenTherm; Rapidozym, Berlin, Germany). The PCR temperature profile consisted of an initial denaturation at 94 °C for 2 min 30 s, followed by 25 cycles of 94 °C for 30 s, 55 °C for 30 s and 72 °C for 45 s, and a final extension at 72 °C for 10 min. PCR products were examined by agarose gel electrophoresis and quantified using a NanoPhotometer (Implen, München, Germany). The PCR products were diluted to the same concentration and pooled per individual. The pooled PCR products were purified to remove the non-incorporated ITS primers of the first PCR which would compete with the longer primers used for the tagging step, using the NucleoSpin Gel and PCR Clean-up kit (Macherey-Nagel, Düren, Germany).

Amplicon libraries for pyrosequencing were prepared in a subsequent PCR amplification using the purified PCR products as templates. These PCRs were performed under the same conditions as described above, except the ITS1F primer was extended with a Roche 454 A pyrosequencing adapter and an error-correcting barcode sequence (Hamady et al. 2008), and the ITS4 primer was extended with a Roche 454 B sequencing adapter (see Table S1). In total, we had 78 samples, split across three partitioned sections (each section representing 1/8th of the plate) in one pyrosequencing run. One sample from each species was included in each of the three sections, requiring the use of 26 different barcodes.

The PCR temperature profile consisted of an initial denaturation at 94 °C for 2 min for 30 s, followed by 10 cycles of 94 °C for 30 s, 55 °C for 30 s and 72 °C for 45 s, and a final extension at 72 °C for 10 min. PCR products were examined by agarose gel electrophoresis and quantified using a NanoPhotometer and purified using the NucleoSpin Gel and PCR Clean-up kit (Macherey-Nagel). Equimolar pools of the pooled PCR products stemming from each of the three replicates of each plant species were made, and sequencing was performed by the Genome Sequencing and Analysis Core Resource at Duke University (Durham, NC, USA) on a Roche FLX 454 pyrosequencing instrument.

Data analysis

Generating the phylogenetic tree for the host plants

A two-loci phylogeny was reconstructed for the host plant species and two outgroup species (Campanula rotundifolia and Calycera sp.) using DNA sequences of the plastid maturase K (matK) spacer region and the nuclear ITS region retrieved from GenBank (see Table S2 for accession numbers). Sequences not available in GenBank were replaced with available sequences of congeneric species (Cadotte, Cardinale & Oakley 2008). The sequences were aligned with mafft version 6 (Katoh et al. 2002) and species trees were estimated with beast version 1.6.1 (Drummond & Rambaut 2007) assuming a GTR + I + G model of nucleotide substitution. Trees were sampled every 1000 generations for a total of 10 million generations, and the maximum clade-credibility tree with node heights based on sample averages, generated from the final 9001 sampled trees, was calculated. Due to missing intrageneric sequence information, branch lengths within the genera Senecio and Centaurea were standardized to branch lengths within Hieracium where intrageneric information was available. To assess the effects of this adjustment, subsequent analyses were performed with a set of 100 trees containing randomized intrageneric branch lengths for the three Centaurea species (see Fig. S2 for the phylogenetic tree).

Fungal diversity

To reduce sequencing errors, for example due to the presence of ambiguous bases or homopolymers, sequences were denoised using the PyroNoise algorithm as implemented in mothur version 1.25.1 (Schloss et al. 2009). Subsequently, reads without a valid barcode or primer sequence were excluded, and the ITS1 region was extracted from the remaining raw sequence reads using the Fungal ITS extractor Perl script (Nilsson et al. 2009), which only considers reads with complete ITS1 regions.

Operational taxonomic units (OTUs) were generated using Bayesian clustering as implemented in crop version 1.33, which generates OTUs without using a hard cut-off regarding sequence similarity for species discrimination thus reducing common overestimation of the number of OTUs (Hao, Jiang & Chen 2011). Briefly, the method splits all sequences into blocks and a centre sequence is identified that characterizes the cluster. Pairwise distances among sequences are then calculated; the probability that a sequence belongs to a cluster is a function of its distance to the centre sequence. The process is repeated after pooling/splitting clusters into separate blocks in a Markov Chain Monte Carlo chain (here, 10 000 iterations). More detail is provided in the original publication. We also clustered DNA sequences using CD-HIT (Huang et al. 2010) to validate the results of the algorithm implemented in crop. CD-HIT uses a greedy heuristic clustering algorithm that is fast (ideal for this purpose). The number of clusters predicted using CD-HIT and the crop algorithm were very different (details for CD-HIT see Table S5), with CD-HIT resulting in a much higher number of clusters compared to crop, which might represent an overestimation of OTUs, and therefore, we decided to use the more conservative crop output in subsequent analyses.

The number of reads that were obtained per sample varied widely. Because measures of diversity are highly dependent on sample size (Smith & van Belle 1984), we performed a procedure to assess sample-based richness and abundance by standardizing each sample to 1000 reads using the bootstrapping method (Efron 1979), which provides a conservative approach to control for inconsistencies in the number of reads. In brief, 1000 subsamples of 1000 reads were obtained with replacement from each plant sample and fungal community OTU richness was assessed for each subsample. The mean across all 1000 subsamples was then assessed (the distribution of richness estimates did not deviate from normality) and was rounded to the closest integer. Subsequently, OTU information on relative number of sequence reads for the most abundant taxa (aggregating to the richness estimate obtained from the bootstrapping procedure; when two sequences were equally represented, then ranking was based on a randomization procedure) was extrapolated to sum up to 1000 reads. Singletons were filtered out during this resampling procedure. All further analyses were done on the resampled data set.

Taxonomic identities of the OTUs were obtained by comparing their representative sequences with all fungal ITS1 sequences in the GenBank nucleotide sequence data base (as of January 2013) using Basic Local Alignment Search Tool (blast; Altschul et al. 1997). Sequences belonging to fungal species in GenBank are often not confirmed by morphological characteristics, and many identifications may be erroneous; however, this imperfect approach is currently the best means of assigning tentative identities to DNA sequences (Porter & Golding 2011), and curated data bases of ITS are only available for specific groups of fungi (i.e. UNITE;, for ECM fungi). Subsequent analyses of the blast results to assign OTUs at phylum or order level were performed using the software package megan version 4 (Huson et al. 2007). OTUs not of fungal origin were excluded.

All subsequent analyses were done in r version 2.14.1 (R Development Core Team 2011). First, we calculated rarefaction curves (function rarefaction for package ‘vegan’, Oksanen et al. 2009) for the different individuals of each plant species to gauge adequacy of sampling depth. To examine the distribution of fungal OTUs among the three replicates of one plant species, we generated a dendrogram after performing cluster analysis based on Canberra distances.

Variance partitioning and null model analysis

The spatial positions of the plant individuals were used to run principal coordinate analysis of the neighbourhood matrix (PCNM; Borcard et al. 2004). The resulting eigenvectors account for the multivariate spatial autocorrelation of our plant individuals at all spatial scales that could be resolved by the sampling strategy (Borcard & Legendre 2002; Desdevises et al. 2003). This was done to separate the plant individuals spatially and test for the influence of space on fungal community composition. Space is a proxy for all the factors that are spatially structured and determine the spatial pattern in species distributions but have not been measured explicitly.

To determine the effect of host plant phylogeny on fungal community composition, we first generated a phylogenetic distance matrix of the phylogenetic tree of the Asteraceae. This distance matrix was used to calculate a constrained analysis of principal coordinates to estimate phylogenetic inertia (Desdevises et al. 2003). The principal coordinates represent the phylogenetic relationship among the terminal taxa at all levels. Coordinate axes that were determined to be significant predictors of fungal community composition using the forward.sel function in r (package ‘packfor’) were extracted and included in the variance partitioning as one variance component (Desdevises et al. 2003). Variance partitioning was done using redundancy analysis (RDA), with the resampled fungal community matrix as response variable and three explanatory matrices: the PCNM eigenvectors (‘spatial effect’), the significant coordinate axis of plant phylogeny (‘host phylogeny’) and the two-level categorical factor (sand/loam) of the soil (‘soil’). We also tested for the effect of the three different sampling times (that we needed because of different flowering time), and sampling time was included as fourth factor in the variance partitioning. The variance partitioning accounted for the variance explained by each of our predictors separately but also for their combined effects.

Constrained RDA followed by a pseudo-F test was used to assess the significance of the explanatory variables. Calculations concerning the variance partitioning were also done in package ‘vegan’.

To specifically address the importance of biotic interactions, which are not adequately assessed through variance partitioning (Smith & Lundholm 2010), we conducted a null model analysis (Gotelli 2000; Gotelli & Ulrich 2010) to test for species co-occurrence patterns. We used the C-scores (a checkerboard index assessing the extent of species segregation/aggregation, compared to random species distribution, Stone & Roberts 1990) to measure the extent to which fungal species co-occur. The higher the index, the more species are negatively associated with each other, whereas a low index indicates positive interaction. To create null distributions of the C-score, we used an algorithm equivalent to SIM 9 in Gotelli (2000), which is a randomization algorithm where both row and column sums are fixed. This specific algorithm is considered to be ideal for testing for patterns of co-occurrence arising from species interactions. The analysis was implemented in r using the package ‘bipartite’ (Dormann et al. 2009) and the function oecosimu in the package ‘vegan’ and 5000 random fungal community matrices for comparing the central tendency of null distribution to the observed C-scores.


Sequence recovery

A total of 158 721 sequence reads met our quality criteria across all host plants and from these, 156 816 ITS1 sequences, with a length varying between 101 and 530 base pairs (bp; median of 170 bp), were used for further analysis. Two plant individuals, one from Carlina vulgaris and one from Hypochaeris radicata, resulted in only a single sequence and were therefore excluded from our analyses. Excluding these individuals, the number of ITS1 sequences varied from 7060 and 52 (median of 1615) among the different samples.

The Bayesian OTU clustering resulted in a total of 1793 OTUs, of which 811 OTUs were singletons. After resampling, 1100 non-singleton OTUs remained; note that these numbers do not add up since some of the singletons from the original data set might be picked up twice during resampling, with replacement, and are then no longer singletons.

Of these 1100 OTUs, 966 OTUs were confirmed to be of fungal origin (identification is based on megan version 4) and used in subsequent analyses. Each plant species on average hosted 65 OTUs in their roots, but variation was high (standard deviation: 19.08). Per individual, the highest number of OTUs was found in the roots of one Hieracium pilosella (114 OTUs) and one Taraxacum sp. (107 OTUs), whereas the lowest number of OTUs was recorded from one Helianthus tuberosus (18 OTUs; Table S3). Rarefaction curves approached saturation for most individuals (Figs S3 and S4).

Identification of the main groups of fungi

The taxonomic identification with megan version 4 found 87.8% of the OTUs and 94.1% of the sequences to be of fungal origin. The remaining 12.2% of all OTUs (134 OTUs and 5.9% of the sequences; Fig. 1) could not be assigned to a known organism group. Most of the fungal OTUs belonged to the phylum Ascomycota (394 OTU, 35.8% of all OTUs, 56.3% of the sequences) with the most abundant sequence matching Phialophora mustea (Table S6) and the Basidiomycota (201 OTUs, 18.3% of all OTUs, 16.4% of the sequences) with Sebacina as most abundant genus (Table S6). We also found 7.2% of the OTUs (79 OTUs and 2.4% of the sequences) belonging to Glomeromycota (Fig. 1), with the AM fungus Rhizophagus irregularis as one of the 50 most abundant OTUs. Chytridiomycota (15 OTUs, and 0.14% of the sequences) accounted for 1.4% of OTUs (Fig. 1). About 25.2% of the OTUs (277 OTUs and 18.9% of the sequences) could be assigned to fungi but not to a certain phylum and are classified in Fig. 1 as ‘unassigned fungal OTUs’.

Figure 1.

Relative abundance of fungal phyla in the operational taxonomic unit (OTU) taxa definitions that were delineated with crop.

Effects of species identity, spatial structure, host plant phylogeny and soil on fungal communities and biotic interactions

The cluster analysis gives a broad overview of variation in fungal community composition among the different plant species (Fig. 2). Variation in fungal community composition is generally associated with plant species identity; for some plant species, all three individuals appeared in a single cluster (e.g. H. pilosella, Leucanthemum vulgare, Anthemis tinctoria), while only a few plant species displayed large intraspecific variation and did not cluster (e.g. Achillea pannonica, Bellis perennis). In most cases, at least two individuals appeared in a single cluster (e.g. Solidago virgaurea, Centaurea jacea).

Figure 2.

Canberra distance-based cluster dendrogram of internal transcribed spacer fungal communities as these were sampled from the replicated (three replicates) host plants.

To separate the direct contribution of these host plant effects from other drivers of fungal community composition, we partitioned variance in fungal communities to ‘spatial effect’, ‘host phylogeny’ and ‘soil type’ components. Each class of variables accounted for a significant proportion of the multivariate variance in OTU distributions (constrained RDA followed by pseudo-F test; < 0.005). The three main predictors together explained 30% of the variance in fungal community composition (Table 3), with ‘host phylogeny’ (20%) explaining most, followed by space with 9% and soil type with 1%. Sampling time, which exhibited some colinearity with host plant phylogeny, accounted for 1% of the explained variance, suggesting that compositional differences were driven by host plant phylogeny and that these effects were not strongly confounded by sampling time. We plotted fungal community composition on the phylogeny of the Asteraceae to visualize the main pattern: closely related plant species share similar fungal communities (Fig. 3).

Figure 3.

Phylogenetic comparative analysis of fungal community composition metrics of the plant hosts. The phylogenetic tree is a reconstruction of the evolutionary history of the Asteraceae based on the internal transcribed spacer and matk region. The two metrics (white and black bullets) of the fungal community are the axes loadings of the first and second non-metric multidimensional scaling (NMDS) axes, respectively, following ordination analysis that was applied to the entire fungal community (NMDS stress = 0.25). For NMDS1, the black and white dots represent NMDS loadings between −0.51 and 0.13 and between 0.13 and 0.75, respectively. For NMDS2, the black and white dots represent NMDS loadings between −0.51 and −0.02 and between −0.02 and 0.45, respectively. Thus, dots of the same colour and combinations of dots at the terminal branches visualize similarities in fungal community composition in relation to positions in the plant phylogenetic tree (for statistical analysis see text).

The null model analysis yielded a significantly lower C-score than expected by chance (positive effect size; Table 2): this implies species segregation (Gotelli 2000), that is, most pairs of species co-occur less often than expected by chance.

Table 2. Results of the null model analysis on the presence/absence matrix of total fungal community composition and the Glomero-, Asco- and Basidiomycota subset per plant individual
 Observed C-scoreEffect sizeMean of the expected C-scoreP-value
  1. The analysis was done on the complete resampled data set (76 individuals and 1100 fungal operational taxonomic units). The P-value indicates more observed checkerboards than expected suggesting negative species co-occurrences.

Fungi vs. plant individual9.853.509.780.00099
Glomeromycota vs. plant individual12.173.9311.640.00099
Ascomycota vs. plant individual14.115.8813.890.00099
Basidiomycota vs. plant individual7.

Looking only at the Ascomycota subset, we found a similar pattern: factors explaining the most were host plant phylogeny (15%) and space (7%). The other two predictors explained < 1% of the variance (Table 3). For the Basidiomycota and Glomeromycota subset, we could not find the same pattern as for all fungi together (Table 3). In the null model analysis of the subsets, for Ascomycota and Glomeromycota OTUs demonstrated segregation, but a neutral pattern was observed for the Basidiomycota.

Table 3. Results of the variance partitioning showing the explained variance of ‘spatial effect’, ‘host phylogeny’, ‘soil type’ and ‘season’
 All fungiAscomycota (adjusted R2 of the individual fractions)Basidiomycota (adjusted R2 of the individual fractions)Glomeromycota (adjusted R2 of the individual fractions)
  1. The analysis was done on all fungal operational taxonomic units (OTUs) and subgroups (79 OTUs of Glomeromycota, 390 OTUs of Ascomycota, 201 OTUs of Basidiomycota) defined by megan version 4, transformed by Hellinger transformation.

Spatial effect0.
Host phylogeny0.
Soil type0.010.0020.010.00


Alpha diversity

It is common for 454 sequencing studies to report considerably higher diversity than cloning and sequencing studies (Öpik et al. 2009; Becklin, Hertweck & Jumpponen 2012). Our 156 816 sequences resulted in a large number of OTUs, considering that we focused on root-associated fungi of plants in a single plant family (the Asteraceae) in a small area. We attribute this to the high plant diversity in the investigated area, which is a biodiversity hotspot in Germany with about 200 plant species. The grassland is quite heterogeneous consisting of small hills with steep slopes and interspersed woodland stands, which may provide a range of different microhabitats for fungal colonization.

The number of sequence reads varied strongly between the individuals. This variability could be caused by errors during the amplification or during sequencing. For example, a fungal community on a certain plant individual, which has an unequal distribution of short and long ITS fragments could introduce amplification errors during PCRs, since shorter DNA fragments are preferentially amplified (Bellemain et al. 2010). Another potential factor causing this variation could be the barcodes used (Hamady et al. 2008). However, we used a resampling procedure to correct for these differences in our analysis.

The plant species with the highest number of OTUs was H. pilosella, while the species with the lowest number was H. tuberosus. The low number of OTUs in Helianthus might be due to its root system, a tuber-forming root with a relatively coarse architecture, compared to the highly branched root system of Hieracium, which might therefore be more susceptible to fungal infection (Newsham, Fitter & Watkinson 1995).

Taxonomic composition of the fungal communities

Most of the OTUs (35.8%) belonged to the Ascomycota, which are commonly found to be the dominant root-colonizing fungal group in semi-arid grasslands (Porras-Alfaro et al. 2011). The most abundant OTU in our sampling area was identified as closely related to P. mustea, the function of which is not well understood. Phialophora species are known to form a complex group of fungi with endophytes, saprobes and plant pathogens (Yan, Rogers & Wang 1995; Ko et al. 2011). The second most abundant fungus was identified as related to Paraphoma chrysanthemicola, section Phoma (de Gruyter et al. 2010), a common dark septate endophyte in grasslands (Porras-Alfaro et al. 2011). The roles of these endophytes in ecosystems still remain unclear but under controlled condition, they can enhance plant performance (Newsham 2011).

Surprisingly, we also found a relatively large number of Basidiomycota (18.3%), which are typically more frequent in forest soils (Buée et al. 2009). The relative abundance we report is almost double the relative abundance that Porras-Alfaro et al. (2011) retrieved in their study, conducted in a semi-arid grassland. The most abundant genus of Basidiomycota and the third most abundant overall was Sebacina (Order Sebacinales). Sebacinalean fungi are common endophytes in many plant roots (Selosse, Dubois & Alvarez 2009; Weiss et al. 2011) and may enhance plant growth and pathogen protection (Schäfer, Khatabi & Kogel 2007). The AM fungus R. irregularis (Phylum Glomeromycota) was relatively abundant, and among the 50 most abundant sequences. In total, Glomeromycota accounted for 7.2% of all OTUs. Glomeromycota are widespread and dominant in grasslands, and Asteraceae generally form symbioses with them (Hempel et al. 2013).

We also found that 22.6% of OTUs were from fungi of uncertain identity, which likely reflects the fact that only a low percentage of fungi have been formally described (Mueller & Schmit 2007).

Predictors explaining community composition

The multivariate distribution of fungal species, which is an aspect of beta-diversity (Legendre, Borcard & Peres-Neto 2005), could be explained to a certain extent by space, host phylogeny, soil type and sampling time; all four predictors had a significant influence on the fungal communities.

However, the most important factor explaining total fungal community composition was ‘host phylogeny’, accounting for 20% of the variance and suggesting that closely related species shared more similar fungal communities than expected by chance (i.e. root-associated fungal communities exhibited conservatism across the host phylogeny). The cluster analysis showed that fungal community composition tended to be more similar among individuals of the same plant species. Such a relationship between host identity/phylogeny and associated fungi has been observed for plant pathogenic fungi, like the anther smut fungi and their caryophyllaceous hosts (Refrégier et al. 2008) and necrotrophic leaf fungi (Gilbert & Webb 2007). Similar relationships have also been observed for mycorrhizal fungal associations in Orchis (Jacquemyn et al. 2011) and ECM fungi (Tedersoo et al. 2013).

Fungal colonization of host plants is likely to be limited or facilitated by certain plant traits, which may be shared among closely related plant species. For example, root traits linked to the architecture of the root system (e.g. root length, root diameter, ratio of fine and coarse roots) and foraging behaviour, which exhibit phylogenetic constraints (Kembell & Cahill 2005), are likely to influence fungal colonization due to variation in surface area-to-volume ratios and the frequency of infection points (Newsham, Fitter & Watkinson 1995). For example, very dense root systems with many fine roots might be more susceptible to fungal infections compared to less branched, thick roots (Newsham, Fitter & Watkinson 1995).

On the other hand, this phylogenetic signal in fungal community composition could be simply due to patterns of co-occurrence, with fungi and their hosts limited by the same or covarying environmental conditions. However, our results suggest this is not the case since 20% of the variation accounted for phylogeny was independent of the other predictors. Furthermore, in a recent study, Tedersoo et al. (2013) found similar effects of host plant phylogeny on fungal community composition within the Salicaceae, which indicates that our pattern might be independent of the investigated system.

The second most important factor for structuring the fungal communities was space, which could be due to a strong fragmentation of the landscape, dispersal dynamics or an interaction between these dynamics and environmental drivers not accounted for by our soil factor. The alternation of hills and valleys might limit dispersal and gene flow between the communities on the roots of the different host plants. For example, for AM fungi, habitat filtering and dispersal limitation are both drivers of assembly processes (Kivlin, Hawkes & Treseder 2011; Caruso et al. 2012), but the total amount of variation in species turnover explained by environmental parameters (edaphic variables, host species, etc.) may be small due to broad tolerances in some taxa (Powell et al. 2011).

Despite the statistical significance of differences in soil type, we only detected comparably small effects of soil type. This may imply generally weak niche differentiation with respect to this factor or be due to the coarse categorical factor used to represent soil type (i.e. unmeasured edaphic variables may account for more variation). Edaphic variables are likely to be spatially structured, and we contend that this variation, if important, would be accounted for by the spatial component. Our results are in line with those of others who have observed that soil type and other soil chemical properties have a small but significant influence on fungal community composition (Verbruggen et al. 2012 for AM fungi).

By looking only at fungal subsets, we found Ascomycota showing the same strong phylogenetic pattern as total fungi, whereas for Glomeromycota and Basidiomycota, the spatial effect is more important than host plant phylogeny. Thus, our subgroups contributed differently to our variance partitioning pattern: Ascomycota mostly determined the large variance accounted for by host plant phylogeny, while Basidiomycota determined the fraction of variance accounted for soil type. All subgroups contributed more or less equally to the variance explained by space. The results of the subgroups have to be carefully interpreted because of potential differences in sampling depth among the subgroups. Glomeromycota for example represent only 2.4% of the sequences which may not be a representative sampling amount. The same might be true for Basidiomycota. Another aspect could be a real difference in host specificity, which is especially known for Glomeromycota (Smith & Read 2008). Studies on host specificity of Basidiomycota mostly addressed ECM fungi, which we may not have here.

Spatial position and soil type may be confounded by species identity because species usually were sampled in the same soil type and area. Because we were limited to sampling plants in the areas where they could be found and as their distributions were patchy, there was a degree of non-independence among the factors we tested. That said, this had little impact on the analysis as the variation was generally attributed to individual components and not to the dimensions in which multiple components overlapped. The results are therefore conservative as they accounted for possible sources of non-independence among the tested factors. Indeed, this is the strength of this study since we are studying these patterns under the natural conditions in which they occur.

It was not the aim of this study to test for effects of species identity on fungal community composition. However, we tested for this effect and results suggested that host species identity is a relatively poor predictor of fungal community relative to host plant phylogeny as it explains a similar proportion of variation but with substantially more costs associated with degrees of freedom (data not shown).

All in all, we could explain 30% of the variance with the three predictor variables (when correcting for co-linearity in sampling time), which indicates that assembly processes in root-associated fungal communities are predictable to a certain extent, at least in our system. Additional sources of variation could include, for example, neighbour effects or stochastic colonization followed by priority effects (Hausmann & Hawkes 2009; Mummey, Antunes & Rillig 2009; Dickie et al. 2012) or biotic interactions among the fungal species (Pan & May 2009). It was clear from the null model analysis of species co-occurrence that the root fungal communities (especially Ascomycota and Glomeromycota) were not random assemblages but demonstrated strong spatial segregation. This latter result suggests that negative interactions among fungi led to limiting similarity in their composition (Gotelli 2000) and indicate competition among fungi. Next to environmental filtering, which allows only certain fungi to persist (as shown in variance partitioning), these negative interactions could be an additional explanation for the pattern of community composition found in our study. These processes likely combine to contribute to the coexistence of similar species (Shigesada, Kawasaki & Teramoto 1979) and may thus help explain the observed level of fungal diversity in our system. Analysis of the Basidiomycota, however, suggested that interactions among these species had negligible effects on community assembly. Basidiomycota communities may have, therefore, been driven by environmental factors, but variance explained by our environmental factors was low. Further investigations might be necessary to explain the factors driving Basidiomycota community composition.


Our sequence data set enabled analysis of the impact of host phylogeny, space and soil on root-associated fungal community composition. The variance explained by the predictors is quite high and significant. Therefore, we can conclude that the fungal community assembly in this system is predictable to a certain extent. However, other factors, for example, community-level interactions among plant species, interactions between root-associated fungi and other soil microbes, fine-scale soil heterogeneity, and stochastic colonization and extinction events, may be also important. We found patterns of species co-occurrence consistent with the hypothesis that negative interactions might prevail in the assembly of root-inhabiting fungal assemblages. This limiting similarity, in combination with environmental filtering, is likely to be an important reason for the high level of diversity observed in this semi-arid grassland.


We thank Michael Ristow (Universität Potsdam) for help with the identification of the plant species in Mallnow, Ruth Lintermann for her help during field and laboratory work, Dr Erik Verbruggen for helpful comments on this manuscript, Sebastian Horn for his help with compiling crop, Jan Treiber (Senckenberg Museum Görlitz) for his help with the maps and Landesumweltamt Brandenburg for the permission to work in Mallnow. This study was partially supported by the Dahlem Center of Plant Sciences at Freie Universität Berlin.

Data accessibility

Raw sequence data are submitted to NCBI (accession PRJNA202091).