Phylogenetic relationships among host plants explain differences in fungal species richness and community composition in ectomycorrhizal symbiosis



  • Geographic and taxonomic host ranges determine the distribution of biotrophic organisms. Host phylogenetic distance strongly affects the community composition of pathogens and parasites, but little is known about the host phylogeny effect on communities of mutualists, such as plant–pollinator and plant–mycorrhizal fungi systems.
  • By incorporating phylogenetic eigenvectors into univariate and multivariate models, we aimed to determine the relative contribution of host phylogeny and environmental variables to mycorrhizal traits and community composition of ectomycorrhizal (EcM) fungi in Salicaceae at the local scale.
  • Host phylogeny explained 75% of the variation in fungal species richness and 20% of the variation in community composition. We also re-analyzed a system involving eight hosts from Japan, in which host phylogeny explained 26% and 9% of the variation in fungal richness and community composition, respectively. [Correction added after online publication 21 May 2013: in the preceding sentence the values 9% and 26% have been transposed.]
  • Phylogenetic eigenvectors that differentially account for clades and terminal taxa across the phylogeny revealed stronger host effects than did the treatment of host species as categorical or dummy variables in multiregression models, and in comparison with methods such as Mantel test and its analogs. Our results indicate the usefulness of the eigenvector method for the quantification of the host phylogeny effect, which represents an integrated complex function of taxonomic sampling effect and phylogenetic distance per se.


The species composition of natural communities is driven by a combination of historical, deterministic and neutral processes. Within communities, species exhibit ecological traits and interact in a manner that is often predictable by their phylogenetic relations, a phenomenon termed ‘niche conservatism’ (Webb et al., 2002; Wiens et al., 2010). The integration of phylogenetic and ecological analyses allows phylogenetic autocorrelation to be accounted for in the so-called comparative analyses of trait evolution (Wiens et al., 2010) and the relative importance of phylogeny in the response of a species to environmental change and community formation to be estimated (Diniz-Filho et al., 1998, 2012; Desdevises et al., 2003; Covain et al., 2008; Kühn et al., 2009). Similar to spatial autocorrelation resulting from dispersal limitation, insights into phylogenetic relations among organisms have provided a new paradigm for understanding the evolutionary ecology of traits, species and entire communities (Cavender-Bares et al., 2009; Weber & Agrawal, 2012). Both spatial and phylogenetic autocorrelation can be addressed using an eigenvector-based approach, which is more sensitive to clade-specific effects relative to the conventional matrix-based approaches, such as the Mantel test and its analogs (Diniz-Filho et al., 1998, 2012; Desdevises et al., 2003). These eigenvectors are orthogonal and approximate distances among sampling units over various spatial and phylogenetic scales. They can be implemented in variation partitioning (Desdevises et al., 2003) and multiregression modeling (Kühn et al., 2009; Diniz-Filho et al., 2012).

The distribution of obligate mutualists and antagonists is influenced by their geographic and taxonomic host range (Poulin et al., 2011). The taxonomic host range has been shown to vary typically from species to family levels in antagonistic organisms, including plant pathogenic fungi (Gilbert & Webb, 2007), folivorous beetles (Ødegaard et al., 2005; Weiblen et al., 2006) and helminths of fish (Poulin, 2010). Accordingly, phylogenetic distance in communities of parasites and pathogens has been found to increase with host phylogenetic distance (Ødegaard et al., 2005; Weiblen et al., 2006; Gilbert & Webb, 2007; Poulin, 2010). However, all of these studies have not taken into account the potential change in the distance effect among lineages or other confounding effects of spatial, temporal and environmental variables, which may account for much of the unexplained variation (Kühn et al., 2009; Weber & Agrawal, 2012).

In contrast with antagonists, the effect of host phylogeny on the structure of communities of mutualistic organisms remains poorly understood. It is logical to assume that, as a result of simultaneous processes of co-evolution, niche differentiation and niche conservatism among closely related taxa, phylogenetic distance among host organisms will affect the communities of their mutualistic associates (Cavender-Bares et al., 2009). Furthermore, mutualistic and antagonistic interaction networks exhibit similar patterns of nestedness and modularity, suggesting that the same ecological and evolutionary processes shape these associations (Fontaine et al., 2011). Mutualistic networks are often compartmentalized by morphological adaptations of one or both partners, which are generally phylogenetically conserved in plant–animal networks (Rezende et al., 2007).

Ectomycorrhizal (EcM) symbiosis involving plant roots and fungi provides a good system to assess the effect of host phylogeny on the evolution of mycorrhiza-related traits and fungal communities (Hoeksema, 2010; Põlme et al., 2013). First, certain traits, such as mutualistic benefits and preferential associations, are inheritable features that are expressed at the plant intraspecific level (Piculell et al., 2008; Hoeksema et al., 2012; Velmala et al., 2013). In natural and plantation forests, host plant clones or species generally account for the strongest predictor of EcM fungal community composition (Korkama et al., 2006; Ishida et al., 2007; Morris et al., 2008; Tedersoo et al., 2008; Bahram et al., 2012). However, EcM fungal communities respond strongly to edaphic and climatic variables (Morris et al., 2008; Taylor, 2008; Tedersoo et al., 2008; Bahram et al., 2012), which may constrain or blur the host effect. Second, both phytobionts and mycobionts differ in their taxonomic range of partners, although the causes for partner selectivity remain poorly understood (Molina et al., 1992; Suvi et al., 2010). Although most fungal species exhibit a broad host range, patterns of specificity are inferred to be most evident at the plant genus level (Molina et al., 1992; Ishida et al., 2007). Among EcM host plants, the Salicaceae family is the most species-rich EcM plant lineage in Europe, and comprises genera and species with contrasting ecological traits (Savage, 2010), rendering this monophyletic group well suited to address the relative effects of host phylogeny and environmental factors. In both Populus deltoides and Salix spp., EcM colonization depends on the moisture–oxygen balance in the soil (Lodge, 1989). The species richness of EcM fungi varies with soil type and host taxon in Salicaceae (Fontana, 1962; Paradi & Baar, 2006; Krpata et al. 2008; Ishida et al., 2009; Ryberg et al., 2010; Bahram et al., 2011; Hrynkiewicz et al., 2012).

Based on the community ordination analysis and Mantel test of pairwise community similarity indices, Ishida et al. (2007) hypothesized that communities of EcM fungi are determined by the taxonomic relatedness of host trees. Here, we explicitly test the relative effects of host phylogeny and environmental variables on community composition and species richness of EcM fungi in Salicaceae at the local scale. Based on previous studies, we hypothesize that, in particular, host phylogeny, flooding duration and soil pH all influence the richness and community composition of EcM fungi. We also re-analyze the ‘multihost’ EcM fungal community dataset of Ishida et al. (2007) to quantify the effects of host phylogeny and raw phylogenetic distance on EcM fungal species richness and community structure. We compare the relative performance of the traditionally used Mantel test and phylogenetic eigenvector approach in generalized least-squares (GLS) model selection in addressing the host phylogeny effect.

Materials and Methods

Salicaceae dataset

The sampling of Salix and Populus roots was performed in 31 sites across Estonia, northern Europe. The sites were located 0.3–300 km distant from each other (Table 1, Supporting Information Table S1), and were chosen to represent contrasting flooding regime and soil acidity. In 2011, we recorded the number of months during which plants were flooded – that is, the stem and surrounding soil were under water. In each site, a single healthy adult tree or bush belonging to Salicaceae was sampled, ensuring that there were no other species of Salix within a distance of 10 m and no species of adult Populus within a distance of 25 m to avoid the intermingling of roots from different Salicaceae species. Other EcM tree species were easily distinguished on the basis of root morphology, especially taste, color, branching pattern and diameter. In total, 23 individuals of Salix spp. and eight individuals of Populus spp. were sampled to cover much of the phylogenetic breadth of this family in the North European boreal zone. In each site, three root samples (15 × 15 cm2 to 10 cm depth) were collected at 1–1.2 m distance from the stem and 1.5–2 m from each other using a spade or a knife. In the laboratory, the roots were carefully removed from the soil under tap water. The living roots of Salicaceae were initially distinguished by morphology (occurrence and size of ectomycorrhizas, color, taste and branching patterns of the coarse roots); the other roots were discarded. The cleaned roots were laid on large Petri dishes filled with tap water. EcM root tips were sorted into morphotypes on the basis of color, texture, presence and abundance of cystidia, extraradical mycelium and rhizomorphs. The root tips of each morphotype were stored in CTAB buffer (1% cetyltrimethylammonium bromide, 100 mM Tris–HCl (pH 8.0), 1.4 M NaCl and 20 mM EDTA) for molecular analyses.

Table 1. Characteristics of study sites
SiteHost speciesLatitudeLongitudeFlooding duration (months)Fungal species richness
Rannaküla Salix caprea 58.376222.915108
Järvselja base S. dasyclados 58.267227.305846
Supilinn S. cinerea 58.233626.422594
Tähtvere S. alba 58.234326.420647
Sirgu 1 S. fragilis 58.215126.545037
Sirgu 2 S. cinerea 58.215026.544543
Raadi S. fragilis 58.393726.737304
Puura-Tamme S. phylicifolia 57.512227.045224
Tüki S. cinerea 58.421926.582914
Vorbuse S. alba 58.436926.651125
Sillaotsa-Selli 1 S. cinerea 58.262026.160172
Sillaotsa-Selli 2 S. starkeana 58.262426.160073
Vedu S. triandra 58.577526.748515
Põrguauk S. caprea 58.284926.366808
Vahessaare S. daphnoides 58.276526.3528015
Sirvaku S. viminalis 58.194826.790804
Virksimetsa S. aurita 58.255527.024294
Keeri S. pentandra 58.320126.4837110
Ülejõe S. triandra 58.393926.7090611
Oiu S. pentandra 58.397125.9771211
Lemme S. fragilis 57.967124.404309
Lülli S. dasyclados 58.461126.811653
Luunja S. alba 58.341526.7754110
Härjanurme Populus berolinensis 58.321126.3915011
Eerika P. alba 58.363526.6690011
Põlva P. berolinensis 58.069227.092808
Raadi park P. alba 58.395226.7359010
Tusti P. suaveolens 58.391725.7656013
Raadi P. balsamifera 58.396026.756609
Haage P. tremula 58.365226.6225510
Järvselja P. tremula 58.283427.3231211

For the molecular identification of fungal symbionts, one to four representative root tips from each morphotype per sample were separately subjected to DNA extraction, PCR and sequencing. DNA was extracted using a DNeasy 96 Plant Kit (Qiagen, Crawley, West Sussex, UK) as recommended by the manufacturer. PCR targeted the internal transcribed spacer (ITS) region of ribosomal DNA, which represents the standard barcoding marker for the identification of fungi (Schoch et al., 2012). PCR was performed with a combination of fungal-specific primer ITSOF-T, ascomycete-exclusive LB-W and Pezizales-specific LR3-Pez (see Table S2 for all primer sequences and references). In the case of no product, the eukaryote primer ITS4 was used in combination with ITSOF-T or 58SF. DNA resulting in poor-quality sequences was re-amplified with the primer ITSOF-T in combination with the taxon-specific primers ITS4-Tom, ITS4-Russ, ITS4-Clav, ITS4-Sord, ITS4-Seb or ITS4B. The PCR included 5 μl of 5 × HOT FIREPol Blend Mastermix Ready to Load (Solis Biodyne, Tartu, Estonia), 3 μl of template DNA, 0.5 μl of each 20 μmol ml−1 primer and 16 μl dsH2O. PCRs were run under the following conditions: an initial 15 min at 95°C, followed by 35 cycles of 95°C for 30 s, 55°C for 30 s, 72°C for 1 min, and a final cycle of 10 min at 72°C. PCR products were separated by electrophoresis through a 1.5% agarose gel, purified using Exo-Sap enzymes (Sigma, St Louis, MO, USA) and subjected to sequencing with primers ITS5, LF340 and⁄or ITS4. Sequences were assembled into contigs and checked for quality using Sequencher 4.10 software (GeneCodes Corp., Ann Arbor, MI, USA). Species were delimited on the basis of a 98.0% ITS2 sequence similarity threshold (Tedersoo et al., 2006). All unique sequences from each individual plant were submitted to UNITE (Abarenkov et al., 2010a; accessions UDB008785–9000) and the International Sequence Databases (INSD; accessions JX316520–6728).

Soil pH and organic matter were determined from pooled rhizosphere soil of the three samples. Soil pH was determined in a dilution series of dsH2O. Organic matter concentration was determined as mass loss on ignition at 450°C for 180 min.

To confirm the identification of host plant species, we amplified and sequenced the plastid trnH intron in any doubtful cases. To further address the phylogenetic relationships among host plants, we targeted the following fast-evolving loci: plastid trnH intron, plastid rbcL intron, nuclear ITS and nuclear external transcribed spacer (ETS). We utilized both root tip and leaf material, and processed the DNA as described above for fungi. The primer sequences are given in Table S2. The ITS region of Salicaceae failed to amplify or sequence on many occasions, and it was therefore excluded from further analyses. As there were numerous taxonomic conflicts and potential quality issues in Salicaceae sequences in INSD, we relied solely on the sequence data generated by ourselves. However, we downloaded sequences of trnH and rbcL from INSD for Manihot esculenta and Elaeocarpus sylvestris (Malpighiales), which served as outgroups for phylogenetic analyses. As there were no intraspecific sequence differences within Salicaceae, we used a single representative sequence from each species in the phylogenetic analyses. Sequences of genes were trimmed to the same length, concatenated and aligned using MAFFT 6 with an option of considering DNA secondary structure in iterative alignments (Katoh & Toh, 2008).

Multihost dataset

The multihost dataset comprises abundance data of fungal species from two pooled root samples obtained by the excavation of traced roots. These samples were collected from 60 tree individuals belonging to eight host species (representing both Fagales and Pinaceae) in two adjacent mixed forest sites in Japan. Fungal species were identified by combining morphotyping, terminal restriction fragment length polymorphism (T-RFLP) and ITS sequence analyses (Ishida et al., 2007). From the original dataset, we excluded species belonging to putative endophytes and saprotrophs (cf. Tedersoo et al., 2010), as well as T-RFLP peaks representing unidentified organisms. For each of the eight host species and Psilotum nudum (Pteridophyta; outgroup), sequences of three genes – plastid rbcL, plastid matK and nuclear 5.8S-ITS2 – were downloaded from INSD, trimmed and aligned using MAFFT 6 as described above.

Statistical analyses

Maximum likelihood phylogenetic analyses were run to recover the phylogenetic relationships among hosts in both datasets as implemented in RAxML 7.2.8 (Stamatakis et al., 2008), including the GTR + Gamma evolutionary model, data partitioning among three genes and 100 fast bootstrap replicates. The branch lengths of the best-scoring trees were imported to the Ape package of R (Paradis et al., 2004; R Core Development Team, 2012) to generate ultrametric phylograms that account for the differential rates of evolution among clades. We refrained from dating the nodes, because this introduces an extra source of error, especially for the Salicaceae in which molecular dating studies are unavailable. Based on the ultrametric phylograms, pairwise patristic distance matrices were calculated. Pairwise patristic distance is a form of phylogenetic distance that measures the length of all branches connecting two terminal taxa (Fig. 1a). Spatial autocorrelation was assessed by transforming geographical coordinates into a Euclidean distance matrix. Moran's I and Mantel r statistics were calculated to detect autocorrelation in phylogenetic and spatial distance matrices. Nonsignificant values of these indices suggested that spatial autocorrelation was absent from both datasets, and it was not therefore considered in further analyses. Phylogenetic distance was significant in all cases. To further test and quantify the relative effect of host phylogeny, the phylogenetic distance matrices were translated into phylogenetic eigenvectors according to Dray et al. (2006) using the Vegan package of R (Oksanen et al., 2012). These phylogenetic eigenvectors are orthogonal and represent the phylogenetic relationships among terminal taxa at different taxonomic levels (Dray et al., 2006; Diniz-Filho et al., 2012; see Fig. 1). These vectors were forward selected in the Packfor package (Dray et al., 2009; the significance level α was set to 0.1 to include only potentially important vectors in the GLS model selection).

Figure 1.

Conceptual differences between the phylogenetic eigenvector approach and regression based on the Mantel variogram to quantify the effect of phylogeny on traits and communities. (a) An ultrametric phylogram representing nine ingroup species (divided into three clades indicated by colored symbols) and an outgroup species that is excluded from the calculation of phylogenetic eigenvectors. Phylogenetic eigenvectors are built on the basis of the pairwise patristic distances among species (patristic distance between species #4 and #8 is indicated by a thicker line in the phylogram). Phylogenetic eigenvectors are orthogonal; they represent the amount of phylogenetic variation proportional to their eigenvalue and are listed according to the decreasing amount of explained variance. For x species, x – 1 eigenvectors are calculated; the first four vectors are shown for simplicity. The size of the circles associated with each species represents their coordinates on an eigenvector ranging from −1 to 1; open circles, negative values; closed circles, positive values. Significant eigenvectors are selected as covariates for further multiregression or multivariate analyses, and their adjusted determination coefficients are interpreted as the amount of variation explained. (b) Mantel variogram representing the relative phylogenetic distance (patristic distance or other distance measures) among organisms and distance among communities. The black regression line indicates the linear fit between the distance matrices. Symbols and colors represent the clades in the phylogram. Circles represent phylogenetic distances that are shared among clades (constituent clades are indicated by matching colors). Colored linear regression lines indicate differences in phylogenetic distance decay in the three clades to demonstrate the overall poor performance of Mantel variogram analysis without accounting for clade-specific patterns, which are addressed in the phylogenetic eigenvector approach.

To test for the presence of spatial autocorrelation, we ran Mantel tests and calculated Moran's I statistics for multivariate (community) and univariate (EcM colonization, richness) traits, respectively, in both the Salicaceae and multihost datasets, using a Euclidean distance matrix that was calculated on the basis of GPS coordinates. Because no spatial autocorrelation was detected (P  > 0.1 in all cases), it was not considered in further analyses.

The cumulative number of EcM fungal species recovered in three root samples per tree served as a proxy for fungal small-scale species richness in the Salicaceae dataset. In the multihost dataset, two pooled samples from each tree served to represent species richness (Ishida et al., 2007). GLS models with corrected Akaike information criterion (AICc) were applied to select the best multiregression models for explaining the species richness and EcM colonization (only Salicaceae dataset) as implemented in the Nlme package of R (Pinheiro et al., 2011). In the Salicaceae dataset, all phylogenetic eigenvectors, flooding duration, growth form (tree vs bush), soil pH, organic matter concentration and soil particle size (clay, silt + loam, sand, gravel), were included as independent variables for model selection. No significant multi-collinearity was present in the environmental variables (variation inflation factors, VIF < 10). We also ran a separate model without phylogenetic eigenvectors using host genus as a discrete factor with two levels. To check the robustness of the findings, we performed additional GLS analyses involving only Salix species.

In the multihost dataset, the effects of phylogenetic eigenvectors and site on species richness were subjected to GLS model selection. For the multihost dataset, we also ran GLS analyses for the two sites separately to address the robustness of our findings. For both datasets, we also performed analyses using host species as dummy variables to detect the effect of individual species, but neglecting the phylogenetic relations among species.

To address the effect of these phylogenetic and environmental variables on EcM fungal community composition, we generated a community distance matrix based on the Hellinger-transformed species frequency and binary fungal species occurrence data for Salicaceae and multihost datasets, respectively. The studied variables were used to construct multivariate models in the Adonis routine of the Vegan package of R. Adonis represents a nonparametric multivariate analysis of variance (MANOVA) that allows the simultaneous testing of multiple factors and covariates based on permutation tests, and provides their partial coefficients of determination. In parallel, global nonmetric multidimensional scaling (GNMDS) ordination plots were prepared to demonstrate the trends in community composition of EcM fungi in the Salicaceae dataset. Significant variables were fitted into the GNMDS space using the function envfit. For both analyses, the Bray–Curtis distance measure and 1000 permutations were used.


Salicaceae dataset

The 93 root samples of all 31 plants included EcM root tips, except for three samples from a single Salix alba tree individual that was flooded for 11 months. Molecular analysis of 474 EcM root tips provided identification to 97.2% of morphotypes. As a result of our conservative approach, similar morphotypes were usually merged into the same species by molecular typing. Morphotypes represented more than one species on 12 (1.4%) occasions. Clustering of fungal ITS2 sequences from 30 EcM Salicaceae individuals revealed 129 species of EcM fungi which were assigned to 27 phylogenetic lineages (Table S3). Altogether, 82 species were recovered from only one plant individual. The eight sampled Populus individuals hosted 69 fungal species, whereas the 22 EcM Salix individuals harbored 79 species (19 species overlapping) of EcM fungi. Among the four most frequent EcM fungal species (> 6), two (Cortinarius cf. alnetorum and Cenococcum geophilum) were present on both Salix and Populus spp., whereas two (Genabea fragilis and Tomentella sp.) colonized exclusively Salix spp. (statistically nonsignificant). In the overall fungal community of Salicaceae, the /tomentella–thelephora, /inocybe and /tuber–helvella lineages were the most species rich, comprising 42 (32.6%), 17 (13.2%) and nine (7.0%) species, respectively. Two species of Parascutellinia were recovered as EcM symbionts for the first time. Blast searches against public sequence databases over the PlutoF workbench (Abarenkov et al., 2010b) suggested that at 98% ITS2 sequence similarity, 75.4% of species have been recovered in previous studies from fruit bodies, EcM root tips and soil (Table S3).

The molecular identification of plants from root tip DNA revealed a single misidentified root cluster belonging to Alnus incana, which was excluded from all analyses. From both root tips and leaf samples, we obtained high-quality sequences of the nuclear ETS region (except Populus spp.), plastid trnH intron and plastid rbcL intron that effectively separated most species of Salicaceae. Salix starkeana, S. phylicifolia and S. cinerea all had identical sequences of the three genes. No topological incongruences among nodes were observed between nuclear and plastid genes.

The species richness of EcM fungi was structured phylogenetically (IM = 0.592; < 0.001), but not geographically (IM =0.069; = 0.386). Phylogenetic distance classes corresponding to species-, section- and genus-level distances exhibited significantly positive Moran's I values (Fig. S1a). Seven phylogenetic vectors had a significant effect on the species richness of EcM fungi, which cumulatively explained 75.1% of the variation (Fig. 2a; Table 2). These phylogenetic eigenvectors represented the split between the two genera (#1, #2), subgenera of Salix and Populus (#1, #2, #3) and closely related Salix species (#8, #10, #13, #14). Only 11.6% of the variation in EcM fungal species richness was ascribed to the effect of genus in a separate analysis (F1,27 = 6.69; = 0.017). Neither the edaphic variables nor the flooding duration affected EcM colonization or species richness, although preliminary tests indicated a strong correlation between flooding duration and species richness (Pearson r = −0.597; = 31; < 0.001). Much of the flooding information overlapped with that of phylogenetic eigenvector #3 (r = −0.495; = 31; = 0.005). This vector was also strongly correlated with soil texture (= 0.735; = 31; < 0.001). Other phylogenetic eigenvectors did not display significant correlations with environmental parameters. Within Salix, four phylogenetic eigenvectors explained 47.7% of the variation in EcM fungal species richness.

Table 2. Determination coefficients, F values and P values of the best models in the univariate generalized least-squares (GLS) analyses and multivariate permutational ANOVAs (Adonis)
  R 2 adjusted F valueP value
I. Salicaceae dataset
Species richness (= 31)
Eigenvector #10.24410.670.006
Eigenvector #30.1035.580.027
Eigenvector #20.1066.410.012
Eigenvector #80.0835.820.022
Eigenvector #140.0816.530.022
Eigenvector #100.0757.050.010
Eigenvector #130.0596.750.015
Fungal community (= 30)
Eigenvector #20.0642.070.001
Eigenvector #10.0541.730.009
Soil particle size0.0531.710.002
Eigenvector #60.0431.380.043
Eigenvector #120.0401.280.014
II. Multihost dataset (Ishida et al., 2007)
Species richness (= 60)
Eigenvector #50.16512.700.002
Eigenvector #20.0998.760.007
Fungal community (= 60)
Eigenvector #20.0392.440.001
Eigenvector #10.0291.770.008
Eigenvector #50.0261.600.016
Figure 2.

Ultrametric phylograms and their relation to significant phylogenetic eigenvectors (circles): (a) Salicaceae dataset; (b) multihost dataset of Ishida et al. (2007). Values above the branches indicate bootstrap support > 70. The size of the circles associated with each species represents their coor-dinates on an eigenvector ranging from −1 to 1. Open circles, negative values; closed circles, positive values.

The community composition of EcM fungi was significantly phylogenetically structured (Mantel test: rM = 0.244; < 0.001), but there was no spatial autocorrelation in the dataset (rM = 0.020; = 0.411). Overall, host phylogenetic distance per se had a weak negative effect on EcM fungal community similarity (F1,443 = 27.5; R2 = 0.058; < 0.001; Fig. 3a). Based on significantly positive Mantel r values, fungal communities of plants belonging to the same species, section and genus were significantly more similar to each other than expected (Fig. 3b). Adonis permutation analysis revealed that four phylogenetic vectors and soil particle size had a significant effect on EcM fungal community composition, explaining 20.1% and 5.3% of the variation in the fungal community, respectively (Table 2; Fig. S2). The split between the two genera and among Salix subgenera (eigenvectors #1 and #2), the split between P. alba and P. tremula (eigenvector #6) and the split between S. aurita and related species (eigenvector #12) mostly contributed to the divergence in EcM fungal community (Fig. 2a; Table 2). In a separate analysis, plant genus explained only 5.6% of the variation in the EcM fungal community (F1,27 = 2.04; = 0.005). Within Salix, a single significant phylogenetic eigenvector (#2) explained 13.5% of the variation in the EcM fungal community (F1,15 = 2.03; = 0.002).

Figure 3.

Phylogenetic distance decay among communities of ectomycorrhizal (EcM) fungi with increasing host phylogenetic and taxonomic distance. (a, b) Salicaceae dataset; (c, d) multihost dataset of Ishida et al. (2007). In the upper Mantel variograms, the linear regression line is fitted with data (a): F1,443 = 27.5; R2 = 0.058; < 0.001; (c): F1,1768 = 9.24; R2 = 0.005; = 0.002. In the bottom correlograms, distance classes were determined on the basis of taxonomic information; closed circles indicate significant Mantel r values for distance classes.

Other soil variables and flooding duration had no effect on the community structure of EcM fungi at this geographic scale. When host species were included as dummy variables, these had no significant effect on fungal species richness, EcM colonization or fungal community composition.

Multihost dataset

In the multihost dataset of Ishida et al. (2007), we included 145 EcM fungal species of the 204 T-RFLP taxa of different organisms for subsequent analyses. EcM fungal species richness was structured phylogenetically (IM = 0.197; = 0.002), but not geographically (IM = 0.033; = 0.406). As based on Moran's I values, host phylogenetic distance had a negative effect on fungal community similarity only at the plant species level (Fig. S1b). Nonetheless, the phylogenetic eigenvectors #2 (representing the order-level split in Fagales) and #5 (representing the family-level split in Betulaceae) had a significant effect on EcM fungal species richness among host trees, explaining 26.4% of the variation (Fig. 2b; Table 2). When host species were included as dummy variables, these explained 21.9% of the variation in EcM fungal species richness.

The EcM fungal community composition was structured by host phylogeny (rM = 0.101; = 0.026), but not by spatial variables (rM = 0.029; = 0.142), and there was no significant site effect. The overall host phylogenetic distance exhibited a marginally negative linear relationship with fungal community similarity (F1,1768 = 9.24; R2 = 0.005; = 0.002; Fig. 2c). Fungal communities were more similar than expected at the species and family levels, but not at the genus level (Fig. 2d). Three phylogenetic eigenvectors had a significant effect on EcM fungal community composition, explaining 9.4% of the variation (Table 2). The eigenvectors #1, #2 and #5 represented the split among hosts at the kingdom, Fagales order and Betulaceae family level, respectively (Fig. 2b). Host species included as dummy variables also had a significant effect on community composition, but these significant dummies explained only 6.0% of the variation. When both study sites were analyzed separately, two significant phylogenetic eigenvectors explained 9.6–13.7% of the variation in the fungal community, indicating the robustness of our findings.


Salicaceae dataset

Salicaceae species hosted a highly diverse community of EcM fungi at the local scale, although sample-scale species richness is relatively low in Salix spp. (see Tedersoo et al., 2012a). In spite of our limited sampling at each site, fungal species richness per plant individual was comparable with that of previous studies (Paradi & Baar, 2006; Ishida et al., 2009; Hrynkiewicz et al., 2012), suggesting that a substantial proportion of species were recovered from each individual. Populus species that support higher mycobiont richness (Bahram et al., 2011) were also represented by more species in the three samples per individual plant. The high explanatory power of the univariate model (adjusted R2 > 0.7) suggests that sampling was sufficient to capture critical differences in species richness among plant taxa. The overall phylogenetic structure of Salicaceae-associated EcM fungi resembles that of previous studies, with disproportionately high richness of the /tomentella–thelephora lineage and individual groups of Pezizales (Ascomycota), but relatively low representation of the /russula–lactarius lineage (Ishida et al., 2009; Ryberg et al., 2010; Bahram et al., 2011; Hrynkiewicz et al., 2012; Tedersoo et al., 2012a). In temperate forests, most common species of EcM fungi tend to associate with several host genera (Molina et al., 1992; Ishida et al., 2007). Most of the fungal species recovered from Salicaceae spp. were previously found from multiple other hosts in the Northern Hemisphere, indicating that at least many common associations are nonspecific over such a large geographic scale. However, field observations of fruit bodies suggest that there are many, potentially rare, Salicaceae-specific taxa among agaricoid Basidiomycota (Hansen & Knudsen, 1992; Molina et al., 1992). Although direct evidence for mycobiont specificity in Salicaceae is lacking, the paucity of association with many common EcM fungal lineages, such as /amphinema–tylospora, /piloderma, /clavulina and /boletus, irrespective of habitat type, is suggestive of partner selectivity.

Of the environmental predictors, only soil texture had a marginal effect on the EcM fungal community of Salicaceae, which is consistent with the results at the global scale (Tedersoo et al., 2012a). The lack of flooding effect on species richness and community composition was unexpected, because it represented the steepest environmental gradient in this study, and EcM fungi are aerobic organisms. Long-term flooding reduces EcM colonization in both Salix and Populus species (Lodge, 1989), and it probably accounts for the low species richness in S. alba riparian forests in the Netherlands (Paradi & Baar, 2006).

The effect of host phylogeny

The phylogeny of host plants was the strongest predictor of EcM fungal species richness and community composition of EcM fungi in the Salicaceae and multihost datasets. Host phylogeny explained 75% of the variation in EcM fungal species richness and 20% of the variation in fungal community composition. In the re-analyzed multihost dataset (Ishida et al., 2007), host phylogeny explained 26% and 9% of the variation in species richness and community composition, respectively. These data strongly support the hypothesis that taxonomic relatedness among host plants governs the ‘host effect’, which is the most influential factor in many EcM fungal communities (Taylor, 2008). Moreover, for both species richness and community composition in these datasets, models incorporating host phylogeny explained a greater proportion of total variation relative to models neglecting phylogenetic relations (based on adjusted R2). With current statistical tools, we are unable to separate the species (taxonomic sampling) effect from pure phylogeny effect because of their potential interaction.

Taken together, these two datasets of EcM fungi provide the first explicit evidence for a host phylogeny effect on communities of mutualistic organisms at the local scale, and are consistent with overall host phylogenetic distance effects on communities of antagonists, such as insect herbivores (Ødegaard et al., 2005; Weiblen et al., 2006), fungal plant pathogens (Gilbert & Webb, 2007; Liu et al., 2012) and fish parasites (Poulin, 2010). In communities of antagonists, the overall host phylogeny effect, as revealed from Mantel variograms, ranges from 7% to 40%, which is greater than that observed for EcM fungi based on the same method (6% and < 1% in the Salicaceae and multihost EcM datasets, respectively). It is tempting to speculate that phylogenetic distance has a stronger effect in antagonistic interactions because of greater specificity and co-evolution resulting from the evolutionary arms race. However, all these studies differ by phylogenetic breadth, methods of collection and statistical analysis, which prevent straightforward comparisons without accounting for sampling biases. The substantial effect of host phylogeny on small-scale fungal richness and EcM colonization is consistent with previous results from other types of root symbioses. For example, growth benefits of arbuscular mycorrhizal fungi are phylogenetically conserved in both plants (Reinhart et al., 2012) and fungi (Maherali & Klironomos, 2007), and the breadth of orchid–fungal interactions (Jacquemyn et al., 2011; Martos et al., 2012) is phylogenetically conserved in plants. In EcM symbiosis, the potential capacity to produce enzymes to access nitrogen and phosphorus from organic material is phylogenetically conserved both among and within EcM fungal lineages (Tedersoo et al., 2012b). These examples, as well as the genetic determination and conservation of growth benefits, root characteristics and fungal communities among offspring within species (Piculell et al., 2008; Velmala et al., 2013), indicate that mycorrhiza-related ecological traits are phylogenetically conserved in plants and fungi at the scale of 10−1–108 yr. Taken together, the observations that overall phylogenetic distance is a poor predictor of mycorrhizal richness and community, and the high explanatory power of several phylogenetic eigenvectors, indicate that host phylogeny may exhibit differential effects in various clades of host plants and that the effects may be nonlinear (Ødegaard et al., 2005).

Mantel correlograms revealed that the effect of host phylogeny was most prominent at the species to genus level in Salicaceae. The genus-level split between Populus and Salix contributed 15% and 28% of the overall phylogeny effect in EcM fungal species richness and community composition, respectively. This was confirmed by the multivariate model, in which the phylogenetic eigenvectors #1 and #2, both corresponding to the genus-level and Salix subgenus-level splits, explained most of the variation in community data. Based on experimental synthesis trials involving a few congeneric plant species and observations of fruit bodies, previous studies have suggested the host plant genus to be the main taxonomic level at which specificity occurs (Molina et al., 1992). In the multihost dataset, genus-level community similarity was also inferred to be strongest in the original analysis of Ishida et al. (2007), but not in the present re-analysis, which used replicated individuals of host species instead of data pooled by species. The lack of genus-level effect observed here probably stems from the relatively low community similarity among two Betula spp. We strictly included only EcM fungi and used a Bray–Curtis dissimilarity metric, but our results were robust to differences in distance measures. However, the inclusion of other (non-EcM and unidentified) T-RFLP taxa blurred the phylogeny effect (not illustrated). The re-analysis allowed us to determine the effect of host phylogeny and revealed that, in addition to a strong species effect, phylogenetic distances corresponding to a family level influence fungal community composition (phylogenetic eigenvectors #1 and #2 contrasting Pinaceae and Betulaceae, respectively, with other taxa; Fig. 2b). In both datasets, the effect of host phylogeny is probably underestimated, because the Salicaceae dataset exhibits a relatively short phylogenetic distance gradient and low replication within species. Only eight host species from two host lineages (Pinaceae and Fagales) were sampled in the multihost dataset. We ascribe the relatively greater host effect of Salicaceae to the presence of host preference and/or specificity patterns in this family (see the first paragraph of the Discussion). The host phylogeny effect is probably strongest in plant communities that are highly diverse at both the genus and higher taxonomic levels, and that include members displaying elevated specificity for their mycorrhizal symbionts (e.g. various habitats including Alnus spp. or Nyctaginaceae spp; Suvi et al., 2010; Põlme et al., 2013).

Methodological considerations

Here, we focused on the use of phylogenetic eigenvectors built on pairwise patristic distances among species as an alternative to the more commonly used Mantel variograms (Ødegaard et al., 2005; Weiblen et al., 2006; Gilbert & Webb, 2007) to infer the relative effect of host phylogeny on mycorrhizal traits and fungal communities of plants. Rather than addressing the absolute phylogenetic distance among all species, the eigenvector-based approach enables us to study the effect of phylogeny over different phylogenetic scales within the same group of species and allows the effect of phylogenetic distance to vary among clades (Diniz-Filho et al., 1998). As a result of such flexibility, eigenvector-based app-roaches exhibit type II error rates lower than those in the Mantel test and type I error rates comparable with those in the Mantel test (Diniz-Filho et al. 1998; Legendre & Fortin, 2010). Furthermore, these eigenvector-based analyses reveal the influential clades in the phylogeny that remain hidden within Mantel variograms and correlograms (Figs 1b, 3). This advantage, in turn, allows the postulation of specific hypotheses for these particular evolutionary events and addresses the correlation between environmental and phylogenetic variables to reveal their shared effect (Desdevises et al., 2003; Kühn et al., 2009).


The inclusion of phylogenetic eigenvectors in multifactorial univariate and multivariate analyses provides a useful alternative to the Mantel test in assessing the relative effect of phylogeny on the evolution of traits and communities. This method revealed that host phylogeny is a strong predictor of species richness and community composition of fungi in EcM plants. Furthermore, our study indicates that host phylogenetic distance may have a positive relationship with community distance in mutualistic organisms.


We thank P. Gerhold, F. Martin (the Editor) and six anonymous referees for insightful comments on earlier versions of the manuscript. This study received funding from Estonian Science Foundation grants 8235, 9286, PUT171, and the Centre of Excellence ‘Frontiers in Biodiversity Research’ (FIBIR).