•The role of mycorrhizal networks in forest dynamics is poorly understood because of the elusiveness of their spatial structure. We mapped the belowground distribution of the fungi Rhizopogon vesiculosus and Rhizopogon vinicolor and interior Douglas-fir trees (Pseudotsuga menziesii var. glauca) to determine the architecture of a mycorrhizal network in a multi-aged old-growth forest.
•Rhizopogon spp. mycorrhizas were collected within a 30 × 30 m plot. Trees and fungal genets were identified using multi-locus microsatellite DNA analysis. Tree genotypes from mycorrhizas were matched to reference trees aboveground. Two trees were considered linked if they shared the same fungal genet(s).
•The two Rhizopogon species each formed 13–14 genets, each colonizing up to 19 trees in the plot. Rhizopogon vesiculosus genets were larger, occurred at greater depths, and linked more trees than genets of R. vinicolor. Multiple tree cohorts were linked, with young saplings established within the mycorrhizal network of Douglas-fir veterans. A strong positive relationship was found between tree size and connectivity, resulting in a scale-free network architecture with small-world properties.
•This mycorrhizal network architecture suggests an efficient and robust network, where large trees play a foundational role in facilitating conspecific regeneration and stabilizing the ecosystem.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Mycorrhizal networks (MNs), or the mycorrhizal fungal mycelia that connect two or more plants, are increasingly recognized as mediators of interactions among trees through their effects on tree survival, growth and competitive ability (Simard & Durall, 2004; Selosse et al., 2006; Whitfield, 2007). Mycorrhizal networks provide a source of mycorrhizal fungal inoculum for establishing seedlings (Finlay & Read, 1986; Nara, 2006), and a potential conduit for interplant transfer of water, carbon and nutrients (Simard et al., 1997; He et al., 2005; Smith & Read, 2008; Warren et al., 2008). Three major challenges of MN research are determining the architecture, function and ecological significance of MNs; considerable debate exists on all levels (Whitfield, 2007). Historically, interest in MNs has focused on their formation and function in controlled artificial systems (Wu et al., 2001) and in natural ecosystems (Simard et al., 1997; Lerat et al., 2002). However, little remains known regarding the architecture of MNs in the field. Architecture includes the physical components (e.g. nodes and links) of the network and their genetic complexity, the relationships among the components, and the spatial extent and topology of the components and their relationships. Describing these attributes is a prerequisite to understanding how MNs function (e.g. in fungal colonization of plants, mycelial growth dynamics, or nutrient uptake/exchange between plants) and how they affect plant populations, communities, and forest dynamics (e.g. in tree regeneration, competition and mortality) (Nara & Hogetsu, 2004; Simard & Durall, 2004; Selosse et al., 2006).
Network theory provides a useful framework for describing the structure, function and ecology of MNs (Bray, 2003; Southworth et al., 2005; Selosse et al., 2006). The architecture of networks is described using the node degree (i.e. number of links a node forms with other nodes), degree distribution (i.e. the distribution of links among nodes), clustering coefficients (i.e. the accessibility of links to nodes), path lengths (i.e. the number of link steps separating nodes) and the contribution of nodes to network connectivity with respect to their topological position or centrality (Table 1) (Albert et al., 2000; Bray, 2003; Southworth et al., 2005). With MNs, architecture depends largely on the taxonomic, spatial and temporal scales investigated (Southworth et al., 2005; Selosse et al., 2006). For example, whether one uses plants, fungi or both as nodes can lead to different structural interpretations (Southworth et al., 2005). In this study, we describe MN architecture by representing trees as nodes and mycorrhizal fungi as links.
Table 1. Glossary of terms relating to mycorrhizal networks assuming that trees are nodes or hubs linked via shared colonization by an individual fungal genet
(i.e. betweenness centrality) The degree of a node, relative to the distribution of degrees in the network; describes the contribution of nodes to network connectivity with respect to their topological position in the network
(i.e. connectivity) The number of links a node has to other nodes
The probability distribution of degrees among nodes in a network
An individual fungal genotype or clone
A node with a high degree of connectivity relative to the average degree distribution
A path (connection) between two vertices (nodes); in this case an undirected path (edge) representing a fungal genet eij that colonizes the roots of two trees i and j or more (‘hyperlink’)
Mycorrhizal network (MN)
An underground network where the mycelia of mycorrhizal fungi link the roots of plants of the same or different species
Points or vertices in a graph that are joined by a physical, functional, or physiological pathway
Node clustering coefficient
The density of links a node has relative to that of its neighbours; indicates the cliquishness among nodes in the network
The variation among node degrees divided by the maximum variation possible in the network
Network clustering coefficient
The mean clustering coefficient among all nodes in the network; indicates the accessibility of links to nodes
The number of linkages in a network, relative to the maximum possible number of linkages
(i.e. eccentricity) The longest of the shortest paths between any two nodes in the network
Mean path length
The minimum number of link steps separating two nodes, averaged across all node pairs in the network
(i.e. geodesic) The number of link steps separating two nodes
A network model where nodes link randomly to other nodes, resulting in a probability degree distribution following a Gaussian curve; random networks are generally less cliquish and more easily traversed than regular networks
A network characterized by nodes with limited connectivity, high clustering coefficients and a probability degree distribution following a Gaussian curve; regular networks are generally cliquish and not easily traversed
A network in which some nodes (i.e. hubs) have a high centrality and degree of connectivity relative to the average among all nodes in the network, resulting in a skewed node degree distribution; they are both cliquish and easily traversed
A network with a small topological diameter (eccentricity) relative to the total number of nodes in the network where nonneighbouring nodes can still reach each other through a small number of link steps
The architecture of biological networks often follows one of three models: regular, random or scale-free (Table 1). In both regular and random networks, links tend to be distributed equally among nodes, but the topology of regular networks is generally more cliquish and harder to traverse than that of random networks. In scale-free networks, some nodes (i.e. hubs) are highly linked and more central to the network, resulting in a skewed node degree distribution (Albert et al., 2000; Bray, 2003). Scale-free networks are both cliquish and easily traversed, and tend to have a ‘small world property’, where most nodes can be accessed from every other node by a small number of hops or steps (i.e. a small path length). They also tend to be more robust to perturbations than regular or random networks. For example, the random deletion of a node would usually have little effect on the overall connectivity of the network, unless hubs were specifically targeted for removal (Albert et al., 2000; Bray, 2003).
Ectomycorrhizal fungi are diverse in their form, function and ecology (Smith & Read, 2008), and the networking capabilities of individual species or genets are expected to vary accordingly. For example, the host specificity of a fungal species and the size, longevity, morphology and continuity of its individual genets are all factors that influence the architecture and function of the resulting MN. The physical traits of Rhizopogon vesiculosus and Rhizopogon vinicolor (Basidiomycota, Villosuli group sensuKretzer et al., 2003) genets, combined with their known benefits to hosts in accessing water and nutrients, make them ideal candidates for studies of mycorrhizal networking. They have been consistently reported as dominant members of ectomycorrhizal communities throughout Douglas-fir (Pseudotsuga menziesii) forest development (Chu-Chou & Grace, 1981; Jones et al., 1997; Twieg et al., 2007) and have been shown to increase seedling establishment, growth, and resistance to root pathogens and drought (Molina et al., 1999). They produce durable tuberculate mycorrhizas and form rhizomorphs estimated to span from metres (e.g.R. vinicolor) to several decametres (e.g. R. vesiculosus) in length (Kretzer et al., 2004, 2005). Previous studies have shown that rhizomorph-forming fungi, including species of Rhizopogon, are particularly adept at translocating nutrients and water through continuous mycelial networks (Brownlee et al., 1983; Egerton-Warburton et al., 2007; Warren et al., 2008).
The old-growth interior Douglas-fir forests (Pseudotsuga menziesii var. glauca) where we studied MN architecture undergo gap-phase regeneration (Newsome et al., 1991), resulting in self-perpetuating, multi-cohort, climax forests (Vyse et al., 2006). The regenerative capacity of these forests has decreased over the last decade, however, as a result of increasing temperatures, summer moisture deficits, and disturbances (Hamman & Wang, 2006; Griesbauer, 2008; Klenner et al., 2008). Understanding MN architecture may provide an insight into how MNs affect forest regeneration, thus improving our predictions of forest dynamics and resilience to disturbance.
In this study, we used multi-locus microsatellite DNA markers to discriminate among individuals of interior Douglas-fir and genets of two ectomycorrhizal fungal species, R. vesiculosus and R. vinicolor. We considered the presence of a single Rhizopogon genet on roots of two different trees as a network link, and then used this approach to characterize the architecture of an MN. Our aims were: to determine if R. vesiculosus and R. vinicolor colonize multiple trees or tree cohorts; to describe the size and spatial pattern of the Rhizopogon spp. genets; to compare the networking qualities of R. vesiculosus and R. vinicolor; and to determine the architecture of the MN. Based on previous work showing that R. vesiculosus forms larger genets than R. vinicolor (Kretzer et al., 2004, 2005), we predicted that R. vesiculosus would link more trees than R. vinicolor. We also predicted that the greater connectivity of large trees compared with small trees would result in a scale-free network where large trees serve as hubs.
Materials and Methods
This study was conducted in a dry, cool interior Douglas-fir (Pseudotsuga menziesii var. glauca (Beissn.) Franco) forest near Kamloops, Canada (51°51′7′′N latitude, 120°31′46′′W longitude). The monotypic, multi-storied, multi-cohort interior Douglas-fir forest was located on a mesic, permesotrophic (zonal) site in the Thompson Dry, Cool Interior Douglas-fir biogeoclimatic variant (IDFdk2) (Lloyd et al., 1990) (Supporting Information Figs S1 and S2). The elevation of the study plot was 1035 m above sea level, with a 10–40% slope and a southeast aspect. Approximately 44 cm of precipitation falls annually in the IDFdk2, with the majority as snow during winter months (Lloyd et al., 1990). Mean maximum and minimum temperatures are 21.0°C and −4.2°C (mean annual = 3.4°C), respectively, with an average growing season of 166 d (April–October) (Environment Canada; Canadian climate norms, 1971–2000). The 30 × 30 m plot in which we conducted the study contained 67 live trees (722 stems ha−1) with a basal area of 22.3 m2 ha−1. Trees were grouped into four cohorts (age groups) based on correlations between age and stem diameter in a subsample of trees and the diameter distribution of the whole stand. Additional details pertaining to tree measurements and mapping, as well as soil properties and per cent cover and frequency of plants within the study site, are provided in Methods S1.
We sampled tuberculate Rhizopogon-like (Basidiomycota; Rhizopogon vesiculosus and Rhizopogon vinicolor sensuKretzer et al., 2003) mycorrhizas (when encountered; n =401) from four sides of every tree in the plot, within the dripline or between trees where canopy cover was sparse. We used a dispersed, nonrandom sampling approach to maximize the probability of sampling roots from every tree in the plot. The location of each sample, derived from its distance and azimuth to two or more neighbouring trees, was plotted in ArcGIS (V9.1; ESRI, Redlands, CA, USA) as shown in (Fig. 1 and Fig. S3). The depth (cm) of each tubercule sample, measured from the surface of the forest floor, was recorded and used to compare the mean depths of occurrence of R. vesiculosus and R. vinicolor genets. Fresh needle or cambium tissue was collected from the 67 trees within the plot and from an additional 64 border trees to provide reference DNA for the identification of tree roots. Reference DNA was collected from border trees if their height was greater than their distance from the plot boundary, ensuring that all trees with roots potentially inside the plot were identified. After collection, mycorrhiza and tree tissue samples were placed in a refrigerated cooler for transport and stored at 4°C for no more than 2 wk before further processing. Tuberculate mycorrhizas were washed in deionized H2O and their identity confirmed as Rhizopogon spp. based on morphological characters under a stereomicroscope at ×20 magnification. Tree cambial tissue was placed immediately on dry ice and extracted within 24 h of collection.
Molecular methods and analysis
Both tree and fungal DNA was extracted from samples using the Qiagen DNeasy® plant extraction kit and protocols (Qiagen Inc., Mississauga, Canada), followed by PCR amplification and analysis of microsatellite DNA loci (for details see Methods S1 and Table S1). Douglas-fir trees were distinguished using microsatellite markers developed by Slavov et al. (2004) targeting the loci PmOSU (Oregon State University)_1C3, PmOSU_1F9 and PmOSU_2D4. Genets of R. vesiculosus/R. vinicolor were distinguished by the loci Rv02, Rv15, Rv46, Rv1.34, Rve2.77 and Rve3.21 for both species, plus Rve2.10, Rve1.21 and Rve2.44 for R. vesiculosus, and Rv53 and Rv2.14 for R. vinicolor (Kretzer et al., 2004). Microsatellite loci were amplified together in PCR multiplex groups of two to four primer sets each, with forward primers labelled at the 5′ end with 6FAM (Integrated DNA Technologies Inc., Coralville, IA, USA), NED, VIC or PET (Applied Biosystems, Foster City, CA, USA) fluorescent dyes. A ‘PIG-tail’ tack-on sequence was applied to the 5′ end of reverse primers targeting Douglas-fir microsatellite DNA to reduce stutter banding as per Brownstein et al. (1996). PCR amplifications were conducted in 10-μl reaction volumes containing ultrapure H2O, GeneAmp 10× PCR Buffer (Applied Biosystems), 10 mg ml−1 bovine serum albumin (BSA), 1 mM GeneAmp dNTP mixture (Applied Biosystems), 2.5 mM MgCl2, 0.25–0.35 mM of each primer, 10 ng of DNA template and 5 U ml−1 AmpliTaq Gold polymerase (Applied Biosystems). Temperature cycles began with a 95°C hot-start for 10 min; followed by 30 cycles of 45 s at 93°C, 30 s at 55°C, and 30 s at 72°C; and a final hold for 7 min at 72°C. Microsatellite loci were analysed using a DNA sequencer (3130XL genetic analyser; Applied Biosystems) and then genotyped using Gene Mapper software (V4.0; Applied Biosystems). Two or more samples were considered to represent an individual tree or fungus if they had matching alleles at all microsatellite loci analysed. Every tree, including border trees that were not encountered as roots in the plot, had a unique genotype based on the targeted microsatellite loci. More than 30 tree genotypes were compared among needle, bud, root and cambium tissue sources to confirm consistency among tissue types. For Rhizopogon, additional tubercule samples were collected in the vicinity of the study site to determine allele frequencies and to test the power of genet delineations within the site (these samples represented an additional 37 R. vesiculosus and 33 R. vinicolor genets) (Table S1). The estimated probability that two unrelated individuals could have identical multilocus genotypes by chance (probability of identity) was 1.4 × 10−6 for Douglas-fir trees (n =131), 2.7 × 10−5 for R. vesiculosus (n =51) and 6.4 × 10−6 for R. vinicolor (n =46). All population genetic statistics were calculated using GenAlEx software version 6.1 available at http://www.anu.edu.au/BoZo/GenAlEx/.
Statistics and network modelling
All statistical tests were performed using JMP software (V6; SAS Institute Inc., Cary, NC, USA) unless otherwise noted, with α = 0.05. Descriptive statistics are reported as a range with mean and SD. The number of trees colonized by individual genets was compared between R. vesiculosus and R. vinicolor using a Wilcoxon rank-sum test. The maximum width and geometric area of Rhizopogon genets were calculated using ArcGIS (V9.1) and compared between Rhizopogon species using two-sample t-tests. The mean depth of occurrence of genets was compared between Rhizopogon species using one-way ANOVA with subsampling in the R statistical computing environment available at http://www.r-project.org. Associations among tree cohort class, tree height, tree diameter, fungal genet frequency and tree node degree were tested using Spearman’s rank correlation tests. The mean node degree and centrality of trees were compared between cohorts using ANOVA.
Mycorrhizal network architecture was modelled using Pajek (de Nooy et al., 2005). Network measures were simplified by the exclusion of loops (a tree linked to itself through a single fungal genet) and multiple links between nodes (two or more trees linked through more than one genet) unless otherwise noted. The network model was classified as regular, random or scale-free based primarily on the degree distribution, centrality and clustering coefficients of nodes (Bray, 2003; Southworth et al., 2005). The spatial orientation of tree nodes was integrated with graph-theoretical measures for network analysis and was used to produce a spatially explicit model of network topology.
Characteristics of trees and fungi involved in the network
A total of 56 Douglas-fir genotypes encountered as roots forming mycorrhizas with Rhizopogon species were matched to reference tree boles based on microsatellite DNA analysis (Table 2). They included 45 of the 67 trees inside the 30 × 30 m plot, with an additional 11 genotypes matching tree boles outside the plot (Fig. 1). We found that up to 19 trees, and trees from all age classes, were linked together by a single Rhizopogon genet (Table S2). All 44 of the colonized trees from the youngest two cohorts shared one or more Rhizopogon genets with trees from older cohorts (Table S2). In addition, young trees from which Rhizopogon species were not sampled were primarily distributed within the area occupied by established fungal genets (Fig. 1). There were no associations between individual Rhizopogon spp. genets and any specific tree cohort class or genotype.
Table 2. Size and networking characteristics of tree nodes by cohort class in a 30 × 30 m plot of interior Douglas-fir (Pseudotsuga menziesii var. glauca) forest
Values are means ± 1 SD.
aRoot lengths were estimated based on the occurrence of Rhizopogon spp. mycorrhizas; cohort 1 root data are based on one sample only. NA, not applicable.
Estimated age (yr)
Number of tree boles in 30 × 30 m plot
Number of trees with roots colonized by Rhizopogon in the plot
1.6 (± 0.3)
9.0 (± 5.3)
20.4 (± 4.4)
27.4 (± 2.7)
Diameter at breast height (cm)
Mean diameter at breast height
0.7 (± 0.68)
8.1 (± 5.2)
24.5 (± 6.2)
46.4 (± 5.3)
Maximum observed root lengtha (m)
Mean of maximum lengths
4.5 (± 4.3)
7.6 (± 5.1)
15.1 (± 5.9)
Rhizopogon vesiculosus genets per tree
Mean R. vesiculosus genets per tree
Rhizopogon vinicolor genets per tree
Mean R. vinicolor genets per tree
Node degree range
Mean node degree (ANOVA: P <0.01)
0.7 (± 1.6)
5.7 (± 7.8)
16.3 (± 13.0)
19.8 (± 13.8)
Mean node centrality (ANOVA: P <0.01)
Of the 401 Rhizopogon spp. tubercules collected, 338 were matched to both a tree and fungal genet (Table S2). We encountered a total of 14 R. vesiculosus genets and 13 R. vinicolor genets (Fig. 1). Nine genets from each species were found on roots from more than one tree and were thus considered a link. Rhizopogon vesiculosus genets were linked with a greater number of host trees (2–19 trees; mean = 10.2 ± 6.6; n =9) than R. vinicolor genets (3–10 trees; mean = 4.4 ± 2.2; n =9) (P =0.03). Also, R. vesiculosus genets had a larger span (0–20.9 m; mean = 13.9 ± 5.4 m) than R. vinicolor genets (0–12.1 m; mean = 5.4 ± 3.7 m) (P <0.01) and covered a larger geometric area (R. vesiculosus: 0–135.3 m2, mean = 35.9 ± 42.8 m2; vs R. vinicolor: 0–10.0 m2, mean = 3.4 ± 3.7 m2; P <0.01) (Fig. 1). We found that R. vesiculosus genets also occurred at greater mean depths (1–34 cm; overall mean = 10.8 ± 1.1 cm; n =14) than R. vinicolor (1–18 cm; overall mean = 8.7 ± 2.0 cm; n =13) (F =6.52, df = 1, P =0.01).
Mycorrhizal network architecture
Fifty-five of the 56 tree genotypes identified from Rhizopogon spp. tubercules were linked to one or more trees in the plot (Table S2). The maximum distance between any two trees in the network (43.2 m) was traversed through only two fungal links, and the maximum path length between any two trees in the network was three fungal links, regardless of the trees’ spatial locations or size (representing a small-world property; Table 1 and Table S3; Fig. 2). The degree to which a tree was linked with other trees (i.e. node degree) was positively correlated with its cohort class (ρ = 0.57, P <0.01), height (ρ = 0.58, P <0.01), diameter (ρ = 0.60, P <0.01) and maximum root length (ρ = 0.78, P <0.01). Additionally, there was a strong positive association between the node degree of a tree and the number of Rhizopogon spp. genets colonizing it (ρ = 0.91, P <0.01). Although trees from all cohorts were linked, large mature trees acted as hubs with a higher degree of connectivity (ANOVA comparing node degree among cohorts: F =9.94, df = 3, P <0.01) and a more central position in the MN (ANOVA comparing node centrality among cohorts: F =6.70, df = 3, P <0.01) (Table 2). This resulted in a skewed distribution of node degrees, characteristic of a scale-free model, with the connectivity of some large trees well above average (Fig. 3). The tree with the highest node degree (k =47) and centrality (0.07) in the MN was a mature tree (94 yr old) located 4.2 m outside the plot boundary (marked with an arrow in Figs 1 and 2; see also tree no. 23 in Table S2). This tree was colonized by eight R. vesiculosus genets and three R. vinicolor genets in the plot.
Mycorrhizal networks link multiple tree cohorts
We uncovered an extensive network that linked trees of all ages in an uneven-aged old-growth forest, where 62% of Douglas-fir trees from the two youngest cohorts were established within the network of veteran trees. The MN was comprised of R. vesiculosus and R. vinicolor, each with unique horizontal and vertical spatial patterning in the soil. Our finding that multiple tree cohorts rather than a single age class were included in the MN implies that Rhizopogon fungi have a diverse energy source that is secure over space and time. While linked trees collectively supply carbon to the fungus, young trees in turn gain access to an established fungal inoculum source (Finlay & Read, 1986; Nara, 2006). Our study demonstrates that the mycorrhizal symbiosis is not just between two or more organisms, but is a complex assemblage of fungal and plant individuals that spans multiple generations (Selosse et al., 2006).
Fungal genet parameters
The maximum span of genets in our study was 20.1 m for R. vesiculosus and 12.1 m for R. vinicolor, supporting the finding of Kretzer et al. (2004, 2005) that R. vesiculosus has larger genets (≥ 13.4 m) than R. vinicolor (2.0 m maximum) in 40- to 80-yr-old second-growth Douglas-fir forests in Oregon. The span of R. vesiculosus genets in our study was also in the range reported for other rhizomorph-forming species such as Suillus bovinus (17.5 m; Dahlberg & Stenlid, 1994) and Suillus variegatus(27 m; Dahlberg, 1997) in a > 100-yr-old Pinus sylvestris forest. Studies of ectomycorrhizal fungi have reported genet sizes ranging from ≤ 1 m span (Gherbi et al., 1999) to ≥ 40 m (Bonello et al., 1998), and often attribute these size differences to the life history traits of the fungi. For example, species forming small genets are thought to be short-lived, early-successional species dispersed primarily through sexual reproduction (e.g. spores), while large genet sizes are considered to represent late-stage, perennial species spreading primarily through vegetative growth (Dahlberg & Stenlid, 1994; Bergemann & Miller, 2002). However, most of these studies were based on the occurrence of sporocarps, which may misrepresent the size and distribution of fungal mycelia belowground (Gardes & Bruns, 1996; Zhou et al., 2001). Moreover, associations among species presence, genet size and stand successional stage are not always consistent (Guidot et al., 2001; Redecker et al., 2001; Dunham et al., 2003; Twieg et al., 2007). Thus, more work is needed to determine whether the size differences we observed between R. vesiculosus and R. vinicolor are attributable to differences in genet longevity, growth rate, or mode of dispersal.
The number of host trees linked by a single genet was greater for R. vesiculosus (mean = 10.2) than R. vinicolor (mean = 4.4), which is consistent with their size differences. The potential for linkages between Douglas-fir trees and R. vinicolor genets in our study was similar to that reported by Lian et al. (2006) for Tricholoma matsutake in an 85-yr-old Pinus densiflora forest (4.5 pine trees per genet); however, the average number of host trees linked through associations with R. vesiculosus genets was more than twice that of T. matsutake or R. vinicolor. Moreover, Rhizopogon genets in our study linked recent natural regeneration with old trees, with important implications for old-growth forest dynamics. By contrast, T. matsutake linked similar-aged trees in an even-aged forest. In addition, we found that R. vesiculosus tubercules occurred at greater mean depths than those of R. vinicolor, showing evidence of vertical niche differentiation, with implications for fundamental processes such as nutrient uptake (Dickie et al., 2002; Neville et al., 2002; Tedersoo et al., 2003). For example, a greater depth range may provide R. vesiculosus with greater access to water and nutrients deeper in the soil during dry periods (Baier et al., 2006; Genney et al., 2006), thereby promoting the water stress tolerance of its hosts.
We represented Rhizopogon genets as continuous links across space, although it is possible that some genets may be fractioned as a result of clonal propagation or disruption over time. The continuity of Rhizopogon genets was supported by a concurrent study using fine-scaled sampling of 20-cm3 soil blocks in contiguous transects, where we found the mycelia of both species to colonize and form continuous connections between multiple trees within a 2-m span (K. J. Beiler et al., unpublished). Previous studies have shown that strand-forming fungi similar to Rhizopogon provide continuous transport pathways across ecologically significant distances (Brownlee et al., 1983; Finlay & Read, 1986) and that severed linkages in the hyphal network are able to reconnect via anastamosis (Bebber et al., 2007; Fricker et al., 2008; Rotheray et al., 2008). There was also a high degree of linkage redundancy between trees in our study. For example, most trees were colonized more than once by individual Rhizopogon genets, and multiple links between trees (the same trees linked through two or more genets) represented 22% of all linkages in the network (Tables S2 and S3).
We found that the node degree of a tree was positively correlated with its size and cohort class. This is not surprising given that the probability of a tree encountering fungal genets increases with the extent of its root system, but to our knowledge this has never been empirically demonstrated. Accordingly, the association between tree size/cohort class and connectivity resulted in a scale-free network as we had predicted. This highlights the importance of large mature trees in the architecture of the MN, where they accounted for most of the connectivity and centrality among nodes in the network. In addition, the network was easily traversed both spatially and topologically (i.e. path lengths were small relative to the number of nodes), giving the network a small-world property. These results are in contrast to those of Southworth et al. (2005), who described a random network in Quercus garryanaforests of California. The difference in results may be a consequence of different approaches in defining a link. We defined a link as a single fungal genet found on two different trees, whereas Southworth et al. defined a link as the presence of a morphotype (near species-level identification based on the morphology of a mycorrhiza) on a tree, such that the node degree of a tree was equal to its morphotype richness. Stand structure also differed between these studies: our interior Douglas-fir forest was dense and had a skewed diameter distribution, with a few large, old hub trees and many young trees, whereas Southworth et al.’s savanna forest had fewer, larger, widely spaced trees that were interspersed with Ceanothus shrubland.
In our study, the tree with the highest node degree was directly linked to 47 other trees through its association with eight R. vesiculosus genets and three R. vinicolor genets in the plot. This corresponded to 84% of potential linkages between this tree and all other trees encountered in the plot, and was three times higher than the mean node degree among trees. The influence of this tree as a network component, despite only a portion of its roots being sampled (the bole of the tree was located 4.2 m outside the plot), suggests that it would be an even stronger hub at a larger spatial scale. Overall, MNs are probably even more extensive and complex than what we describe here because we examined only two of up to 65 ectomycorrhizal fungal species previously described in interior Douglas-fir forests of British Columbia (Twieg et al., 2007).
Implications for network functioning and forest management
The scale-free architecture and small-world property of the MN suggest that resources could be efficiently shuttled to expanding mycelial fronts, including those associated with regenerating seedlings (Bray, 2003). For example, carbon and water resources have been shown to move from multiple plants into a common mycelium (Leake et al., 2001;Wu et al., 2001; Querejeta et al., 2003). When a young seedling connects into this common mycelium, as shown in this study, it has direct access to these resources, potentially facilitating its establishment. This premise is supported by field studies showing greater establishment of germinants when linked into the MN of larger trees (Nara, 2006; McGuire, 2007), and by those showing belowground resource transfer among plants facilitated by mycorrhizal fungi (Simard et al., 1997; Lerat et al., 2002; Teste et al., 2009). Even if resource transfer does not occur between trees, seedlings would benefit from colonization by Rhizopogon genets already being supplied with carbon by other trees in the network. In turn, Rhizopogon genets would benefit from colonizing a range of hosts because their large carbon demands are unlikely to be met by young understorey trees alone. This is implied by the results of Högberg et al. (1999), where ectomycorrhizal fungi specific to pine hosts acquired 87–100% of their carbon from overstorey trees and relatively little from understorey trees.
The architecture of our network also suggests that it is a relatively robust system, where it would be resilient against random perturbations (Albert et al., 2000; Bray, 2003) but susceptible to targeted removals of hub trees (e.g. by high-grade harvesting, insects or disease) (see Figs S4, S5). For example, if hub trees were removed in our study (equal to 20% of networking trees), the capacity for resource transfer and MN facilitation of forest regeneration would be greatly reduced compared with 20% of networking trees removed at random (Figs S4c,d, S5c,d). Because large trees sustained a greater number of fungal genets than small trees, the removal of hub trees may also have a large effect on the genetic diversity of R. vesiculosus and R. vinicolor populations. In summary, we found that most trees in a multi-cohort old-growth forest were linked in a scale-free MN, where large trees served as hubs, with implications for understorey regeneration and functional continuity in the stand. To ensure that old-growth Douglas-fir forests remain resilient and self-regenerative following disturbance, our findings support a management approach that conserves large trees or groups of trees and their mycorrhizal fungal associates.
We thank Sally Aitken and Valerie LeMay for suggestions during project development, Jenna Benson, Tanja Bergen, Jessie Brown and Fraser McIntosh for their assistance with sample collection in the field, and Bill Clark and Mary Ann Olson for help with primer optimization. We also thank Melanie Jones, Nusha Keyghobadi, John Klironomos, Francis Martin, and three anonymous reviewers for their comments on earlier versions of the manuscript. This project was funded by the Province of British Columbia through the Forest Investment Account–Forest Science Program, a University Graduate Fellowship from the University of British Columbia, and the Natural Sciences and Engineering Research Council of Canada.