Inferring dispersal patterns of the generalist root fungus Armillaria mellea


Author for correspondence:
Kendra Baumgartner
Tel: +1 530 754 7461


  • Investigating the dispersal of the root-pathogenic fungus Armillaria mellea is necessary to understand its population biology. Such an investigation is complicated by both its subterranean habit and the persistence of genotypes over successive host generations. As such, host colonization by resident mycelia is thought to outcompete spore infections.
  • We evaluated the contributions of mycelium and spores to host colonization by examining a site in which hosts pre-date A. mellea. Golden Gate Park (San Francisco, CA, USA) was established in 1872 primarily on sand dunes that supported no resident mycelia. Genotypes were identified by microsatellite markers and somatic incompatibility pairings. Spatial autocorrelation analyses of kinship coefficients were used to infer spore dispersal distance.
  • The largest genotypes measured 322 and 343 m in length, and 61 of the 90 total genotypes were recovered from only one tree. The absence of multilocus linkage disequilibrium and the high proportion of unique genotypes suggest that spore dispersal is an important part of the ecology and establishment of A. mellea in this ornamental landscape.
  • Spatial autocorrelations indicated a significant spatial population structure consistent with limited spore dispersal. This isolation-by-distance pattern suggests that most spores disperse over a few meters, which is consistent with recent, direct estimates based on spore trapping data.


The extent of gene dispersal is a fundamental factor in population biology. Dispersal determines the degree of connectivity among subpopulations, and thus affects population evolutionary potential (Slatkin, 1987). From an ecological perspective, dispersal allows species to colonize new environments. For many plant-symbiotic fungi, such as ectomycorrhizal (ECM) or plant-pathogenic fungi, dispersal propagules are spores that are often dispersed by wind. The finite lifespan of the hosts or the host tissues colonized by these fungi necessitates dispersal. Therefore, dispersal is critical for these fungal species to persist from one leaf/root to another, or from one generation of host to another.

With fungi, dispersal has traditionally been estimated by direct methods, such as spore trapping, or by monitoring the spatiotemporal progression of the symptoms/signs of colonization (McCartney et al., 2006). These direct methods are subject to the following technical limitations: first, spore traps are ineffective at low spore concentrations, as is the case when traps are far (> 100 m) from the inoculum source (Aylor, 2003), and, second, monitoring the progression of colonization at a field scale does not accommodate dispersal at the landscape scale. Indirect, genetic methods can be used to overcome such limitations. The theory of isolation by distance (IBD; Wright, 1943) posits that genetic similarity between neighboring individuals is higher than between more distant individuals. For a fungus with a limited capacity for spore dispersal, we would thus expect to find a nonrandom distribution of genotypes (i.e. spatial genetic structure) within the spatial scale of local dispersal. To estimate the spatial scale of local dispersal, spatial autocorrelation methods (Hardy & Vekemans, 1999; Smouse & Peakall, 1999) have provided indirect estimates of gene dispersal that are consistent with direct estimates of propagule dispersal in several plant species (Vekemans & Hardy, 2004). The assessment of dispersal and demographic connectivity among individuals is of fundamental importance to define the management unit for a threatened species or to guide strategies aimed at controlling an invasive or pathogenic species.

Contemporary spore dispersal in fungi has recently been characterized using neutral genetic markers and individual-based methods, and, in a few cases, spore dispersal has been suggested to be spatially limited. For instance, using spatial autocorrelation analysis, Carriconde et al. (2008) revealed that neighboring genotypes of the ECM fungus Tricholoma scalpturatum were more closely related than expected under the assumption of a random spatial distribution, suggesting that spore dispersal was spatially limited in this wind-dispersed species. Similarly, the distribution of genotypes of the Pacific golden chanterelle (Cantharellus formosus) were shown to be nonrandom at distances up to 400 m, suggesting restricted spore dispersal at such spatial scales (Dunham et al., 2006). By contrast, a random distribution of genotypes of Rhizopogon vinicolor and R. vesiculosis within study sites of 50 m × 100 m suggested that spore dispersal, perhaps facilitated by small mammal mycophagy, occurred at spatial scales larger than the boundaries of the study sites in these two species (Kretzer et al., 2005). This latter study illustrates the more common finding of an absence of spatial genetic structure in ECM fungi (reviewed by Carriconde et al., 2008), suggesting extensive spore dispersal. Indirect estimates of gene dispersal have contributed to a better understanding of fungal population structures.

Species of the fungal genus Armillaria are important components of many natural forests and timber plantations of temperate, boreal and tropical regions of the world (Baumgartner et al., 2011). Some Armillaria spp. function mainly as saprotrophs (Kile et al., 1991) and play key roles in carbon and mineral cycling in forest ecosystems. Other species, such as A. ostoyae and A. mellea, are best known as virulent plant pathogens that are the causal agents of Armillaria root disease (Morrison et al., 1991). Armillaria ostoyae is one of two pathogens responsible for the majority of coniferous timber losses in the world (Kile et al., 1991). Similarly, A. mellea is an aggressive pathogen of numerous horticultural crops (e.g. Vitis, Juglans, Citrus, Prunus) and ornamentals, and is known to infect > 500 species (Raabe, 1962).

For the more aggressive species (e.g. A. mellea), the fungus colonizes living woody roots, kills them and uses them as a substrate. It can persist saprotrophically in the form of a mycelium within roots after the host tree dies or is cut down. These colonized roots serve as an inoculum source when the roots of replanted trees grow into direct contact and are colonized by this resident mycelium. Mycelium can also spread below ground within the form of a specialized structure, known as a rhizomorph. Rhizomorph growth and the growth of mycelium through susceptible roots have long been considered as the two main modes of infection and spread of Armillaria root disease (Redfern & Filip, 1991). Indeed, the potential for such subterranean expansion of an individual genotype over many hectares has been documented among different Armillaria species in different forest ecosystems around the world (Adams, 1974; Shaw & Roth, 1976; Korhonen, 1978; Anderson & Ullrich, 1979; Kile, 1983; Legrand et al., 1996; Prospero et al., 2003). In one study, in vitro growth rates were used to estimate the age of a genotype of A. gallica that covered 15 ha of forest, and this genotype is considered to be one of the oldest organisms on Earth (Smith et al., 1992).

As a result of sexual reproduction, Armillaria species produce mushrooms (also known as sporocarps or basidiocarps) that release basidiospores. The epidemiological importance of basidiospores is thought to be low, relative to colonization by a resident mycelium. In orchards, vineyards and timber plantations, a disease center is typically inhabited by one to a few diploid individuals (i.e. resident mycelia) that originate from the previous forest stand (e.g. Rizzo et al., 1998; Baumgartner & Rizzo, 2002; Prospero et al., 2008). Furthermore, inoculation attempts with basidiospores in the field often fail (e.g. Rishbeth, 1970) and the haploid mycelia that germinate from basidiospores are rarely detected in nature (e.g. Peabody et al., 2000). By contrast, other studies suggest that basidiospores are important. First, the neutral genetic structure of Armillaria populations is probably a result of sexual reproduction (Prospero et al., 2008; Baumgartner et al., 2010). Second, in some temperate hardwood and conifer forests, a disease center among naturally established host plants is inhabited by multiple, intermingled, diploid individuals of Armillaria (Ullrich & Anderson, 1978; Rishbeth, 1988; Legrand et al., 1996). Finally, in conifer plantations of New Zealand, basidiospores of A. novae-zelandiae have been shown to colonize freshly cut wood of Pinus radiata (Hood et al., 2008).

Little is known about the spatial scale of basidiospore dispersal of Armillaria. A recent study investigated basidiospore dispersal gradients of A. novae-zelandiae in podocarp–hardwood forests of New Zealand using spore traps (Power et al., 2008). These direct estimates of spore dispersal suggest that the vast majority of basidiospores are deposited within a few meters of the sporocarps, with only a small proportion traveling up to 150 m from the forest edge (the farthest distance sampled).

In the present study, we investigated the spatial distribution of A. mellea genotypes in Golden Gate Park, a human-made, urban woodland in San Francisco, CA, USA. This study site is unique among other studies of fungal genetic structure because it represents a potentially closed system; the park is bordered by a large urban area, which is further surrounded on three sides by the Pacific Ocean. The park consists almost entirely of planted hosts that were established in the early 1870s on land covered primarily by sand dunes, with trees restricted to the eastern end of the park (the exact locations of which are unknown). According to the First Biennial Report of the San Francisco Park Commissioners (McLaren, 1872), ‘The Golden Gate Park contains c. 1000 acres (405 ha), of which 270 acres (109 ha) at the eastern end is good arable land, covered in many places with trees and shrubbery’. Throughout 730 acres (296 ha) of the park, which was referred to in the Commissioner’s 1872 Report as ‘a waste of drifting sand’, the planting of trees pre-dates the presence of A. mellea. There is no evidence that sand dunes supported woody plant hosts to harbor A. mellea, and therefore we hypothesize that genotypes recovered from hosts planted in such areas were originally established by spores. This approach is different from past Armillaria studies in orchards, vineyards and timber plantations (e.g. Rizzo et al., 1998; Baumgartner & Rizzo, 2002), in which resident genotypes spanning the root systems of multiple, adjacent hosts probably influenced subsequent colonization by spores. We used microsatellite markers (Baumgartner et al., 2009) and spatial autocorrelation analyses of kinship coefficients to evaluate the relative contributions of spores and mycelial growth to the spread of Armillaria root disease, and to test the hypothesis of spatially limited spore dispersal.

Materials and Methods

Study site and sampling scheme

The Armillaria mellea (Vahl) P. Kumm. collection was gathered from Golden Gate Park, San Francisco, CA, USA (37°46′8.54″N, 122°29′0.67″W) from April 2001 through December 2002, within a rectangular area of c. 4.5 km × 1 km (Fig. 1). We systematically surveyed for symptomatic hosts and A. mellea sporocarps. Trees, shrubs and stumps were examined along 80 parallel transects, each of which was 1 km in length, oriented north to south and positioned c. 50 m apart. Symptoms of Armillaria root disease on living hardwood and coniferous trees and shrubs included stunted shoots with chlorotic and/or dwarfed leaves, a sparse canopy and resinosis, the latter of which is a symptom specific to the trunks of conifers. Stumps and symptomatic trees and shrubs viewed from each transect were sampled for signs of Armillaria by removing the soil from their root collars to observe the root surface for rhizomorphs. Bark was then removed from the exposed root collars with a knife to collect mycelial fans and/or decayed wood. During the mushroom season, from late October through early January, transects were walked again to survey for sporocarps. The spatial coordinates (x, y) of each sample were recorded with a hand-held, global positioning system. When possible, the plant identity was recorded for each host from which the sample was collected.

Figure 1.

Spatial distribution of Armillaria mellea isolates recovered from symptomatic trees and sporocarps in Golden Gate Park, San Francisco, CA, USA. Each white circle (61 in total) represents one of 90 multilocus genotypes (MLGs). Colored symbols represent repeated MLGs (29 in total), each of which was shared by two or more isolates. Black arrows point to the two largest MLGs (red squares, 342.8 m; purple hexagons, 322.1 m). The arboretum (light green) and other intensively managed groves (dark green) within the park are featured.

DNA isolation and PCR amplification

In the laboratory, portions of decayed wood, mycelial fans or sporocarp stipes were transferred to 1% water agar (WA) containing benomyl 50WP (4 μg ml−1) and streptomycin sulfate (100 μg ml−1), incubated in darkness at 25°C for 7 d, and further purified by hyphal tip subculture, as detailed in Baumgartner & Rizzo (2001). Species identity was determined on the basis of diagnostic patterns of restriction fragments resulting from AluI digestion of the nuclear ribosomal DNA intergenic spacer region I (IGS-I) (Harrington & Wingfield, 1995). A total of 166 isolates were identified as A. mellea (AluI fragments of 320 and 150 bp; western US restriction pattern; Harrington & Wingfield, 1995). Eleven and three isolates were identified as A. gallica (AluI fragments of 399, 240 and 183 bp) and A. nabsnona (AluI fragments of 534 and 200 bp), respectively.

Genomic DNA was extracted from fresh mycelium after 7 d of incubation in potato dextrose broth (PDB) at 25°C and 150 rpm, in darkness. All 166 isolates of A. mellea were genotyped with nine microsatellite markers specific to western US A. mellea isolates (Baumgartner et al., 2009). Eight loci (Am024, Am036, Am059, Am080, Am088, Am091, Am109, Am125) were chosen for population genetic analyses based on positive amplification. Scoring errors were minimized by genotyping all isolates twice.

Identification of genotypes

Multilocus genotypes (MLGs) were obtained for each isolate by combining alleles at the eight microsatellite loci. To verify the integrity of the microsatellite loci for the delineation of diploid individuals of the A. mellea population in Golden Gate Park, we used two approaches.

First, somatic incompatibility tests (Shaw & Roth, 1976) were conducted to confirm that isolates with identical MLGs were part of the same diploid individual. As somatic compatibility is thought to be governed by multiple loci in Armillaria (Worrall, 1997), these tests provided an additional level of discrimination. Somatic compatibility is thought to be a measure of ‘self-recognition’ between two diploid isolates of the same Armillaria species. Pairings of isolates with the same MLG were conducted in all possible combinations, in duplicate. Self-pairings were used as positive controls and a pairing between isolates with different MLGs was used as a negative control. Agar plugs from a pair of isolates, taken from the margin of 14-d-old cultures, were placed 4 mm apart on malt extract agar and incubated for 6 wk at 25°C in darkness. After 6 wk, two isolates separated by a zone of inhibition or a dark dividing line were considered to be somatically incompatible (i.e. different MLGs), whereas those that merged into a single colony to resemble a self-pairing were considered to be somatically compatible (i.e. the same MLG). All isolates with identical MLGs were somatically compatible. Therefore, for the purposes of this study, a unique, diploid individual of the A. mellea population is referred to as an ‘MLG’.

Second, isolates that differed by only one microsatellite allele were paired if the different allele was caused by a scoring error or, possibly, by a somatic mutation within the priming site of a microsatellite locus. A somatic mutation that occurs through mitotic division during the expansion of a long-lived, diploid individual of A. mellea could result in the detection of different allelic profiles from distant locations within a sprawling mycelium. In order to identify such pairs of MLGs from our collection, we calculated a pair-wise, squared genetic distance matrix, according to Smouse & Peakall (1999) for co-dominant data, using GenAlex 6.2 (Peakall & Smouse, 2006). The frequency distribution of the genetic distances was then plotted in histograms. In populations exhibiting only sexual reproduction, the frequency distribution is expected to be unimodal and normally distributed (Arnaud-Haond et al., 2005, 2007). In populations exhibiting somatic mutation, the distribution is expected to be bimodal. A first peak close to zero (i.e. low genetic distance between MLGs) can characterize MLGs that arose from somatic mutations. Accordingly, pairs of MLGs that differed by a single allele (i.e. pair-wise squared genetic distance = 1) were paired to determine whether they were somatically compatible, using the methods described above.

As the mycelial networks of the isolates collected occur below ground and can span multiple root systems, we used the spatial coordinates of the host(s) from which they were recovered to define their boundaries. Isolates recovered from different trees that had the same MLG and were also somatically compatible were considered to represent samples of the same mycelium that were assumed to have originated from one sexual reproduction event on one tree, and subsequently expanded below ground as mycelium to colonize additional trees. The genotypic richness (R) was estimated as = (− 1)/(− 1), where G is the number of MLGs and N is the sample size (Dorken & Eckert, 2001). This index varies from zero in a monoclonal population to unity when each isolate represents a unique MLG.

Population genetic statistics

For each locus, the number of alleles (Na), expected heterozygosity (HE) and inbreeding coefficient (FIS) were estimated using GENEPOP 4.0 (Raymond & Rousset, 1995). We tested whether FIS values differed significantly from zero (departure from Hardy–Weinberg equilibrium) with Fisher’s exact tests using a Markov chain algorithm to estimate unbiased P values, as implemented in GENEPOP. To conduct these analyses, two datasets were used: one including all copies of each MLG (all isolates) and a clone-corrected dataset including a single copy of each MLG (clone corrected).

The hypothesis of random mating was tested using pair-wise and multilocus linkage disequilibrium analyses. As the presence of repeated MLGs generates nonrandom associations between loci, we used clone-corrected datasets. Pair-wise linkage disequilibrium was tested for each pair of loci, with a log-likelihood ratio G statistic, using the Markov Chain algorithm in GENEPOP. Multilocus linkage disequilibrium analysis was tested by computing the index of multilocus gametic disequilibrium rd, which is based on the index of association (IA) (Brown et al., 1980), but is independent of the number of loci (Agapow & Burt, 2001). Departure from the null hypothesis of the random association of alleles, consistent with random mating (no linkage disequilibrium: rd = 0), was assessed by comparing the observed value of rd with that expected under the assumption of random mating. This latter value was obtained by permuting alleles (1000 permutations) between MLGs independently per locus, using Multilocus 1.3 (Agapow & Burt, 2001).

Spatial genetic structure

To identify genetic clusters or demes among MLGs, we used a Bayesian method of assignment in STRUCTURE (Pritchard et al., 2000; Falush et al., 2003). Genetic clusters or demes (i.e. geographically localized, panmictic groups of individuals) can occur as a result of spatially limited gene dispersal (Guillot et al., 2009). STRUCTURE uses a Markov chain Monte Carlo algorithm to assign MLGs to a genetic cluster, assuming Hardy–Weinberg equilibrium and minimizing linkage disequilibrium among loci within clusters. The likelihood of the posterior probability distributions was computed for each number of clusters K from 1 to 20. To check for consistency of likelihood values for each K value between runs, each K was simulated six times, with a run length of 200 000 iterations after the specified burn-in (200 000 iterations), using the admixture model of genetic ancestry and the correlated model of allele frequency. The number of clusters K was determined as the maximal log-likelihood probability of the data estimated for different values of K.

A spatial autocorrelation approach was used to investigate the spatial genetic structure of A. mellea. This approach is based on estimates of coefficients that measure genetic similarity between MLGs, within specific ranges of geographic distance (Vekemans & Hardy, 2004), and was implemented in SPAGeDI (Hardy & Vekemans, 2002). Genetic similarity between MLGs was measured as a pair-wise kinship coefficient (Fij) (Ritland, 1996), which has been shown previously to be the most powerful method for the detection of spatial genetic structure when using variable genetic markers, such as microsatellites (Vekemans & Hardy, 2004). Average Fij values were estimated across loci over a set of spatial distance classes. We defined the number of distance classes containing an even sample size (50 distance classes allowed a minimal number of 250 pair-wise comparisons per distance class). Autocorrelograms were constructed by plotting Fij as a function of distance class. In order to test whether Fij was significantly different from that expected under a random spatial distribution, 95% confidence intervals for Fij were obtained by permuting locations among isolates 1000 times. To quantify the extent of spatial genetic structure of the A. mellea population, we regressed Fij on the logarithm of pair-wise spatial distance to obtain the regression slope (blog). The significance of the observed value of blog was assessed by random permutations (1000) of locations among MLGs in order to obtain a simulated distribution of expected values of blog under the hypothesis of random spatial distribution of MLGs. As blog depends on the sampling scale, we estimated Sp statistics as –blog/(1 − F(1)), where F(1) is the mean kinship coefficient among pairs of MLGs belonging to the first distance class (Vekemans & Hardy, 2004). Sp can be used to compare spatial genetic structure among different studies with different sampling schemes.

The comparison of spatial autocorrelograms both including and excluding repeated MLGs provided an estimate of the clonal subrange (Alberto et al., 2005). The clonal subrange corresponds to the spatial distance at which both autocorrelograms converge. Technically, the clonal subrange is the spatial scale at which the probability of clonal identity approaches zero and beyond which clonality does not affect spatial genetic structure. Because A. mellea does not produce asexual spores, the clonal subrange represents the characteristic maximum size reached by an MLG through vegetative, mycelial growth. We evaluated the contribution of mycelial growth to spatial genetic structure using a dataset that included all isolates (including repeated MLGs). We also used a dataset including all isolates, but only considering kinship values for pairs of different MLGs (excluding repeated MLGs). This was achieved by grouping isolates sharing the same MLG into the same category for analysis with the software SPAGeDI, and restricting pair-wise comparisons to among different categories (i.e. different MLGs). Thus, all the spatial information for isolates sharing the same MLG was maintained for comparisons involving distinct MLGs. This procedure makes it possible to avoid the use of central coordinates for each MLG (average spatial coordinates over isolates sharing the same MLG), which relies on the unverified hypothesis of isotropy (uniform growth) in A. mellea. For each distance class, we estimated the probability of genotypic identity F(r) (Harada et al., 1997) with GENCLONE (Arnaud-Haond & Belkhir, 2007). F(r) corresponds to the fraction of isolates sharing the same MLG and separated by spatial distances belonging to each distance class.


Genotypic structure and richness

We amplified 36 alleles across the eight microsatellite loci, and a total of 90 unique MLGs were identified among 166 isolates. The allelic profiles of these MLGs were characteristic of western US populations of A. mellea (Baumgartner et al., 2010). The genotypic richness of the sample was 0.54. The frequency distribution of pair-wise genetic distance among all MLGs was unimodal and normally distributed (Fig. 2), suggesting that sexual recombination, and not somatic mutation, probably contributed to the distinction of unique MLGs. Moreover, pairings between MLGs differing by only one microsatellite allele (first bar in Fig. 2) revealed that 13 of 14 pairs were incompatible; MLGs differing by one allele were indeed different diploid individuals, except in a single case. These data suggest that we accurately identified diploid individuals and that somatic mutations do not frequently obscure genotype identity for the markers used here. Most MLGs were known from only one isolate and were therefore small in size (Fig. 3). However, 29 MLGs (32%) were represented by at least two isolates and three MLGs were represented by > 10 isolates. The largest two MLGs spanned maximum distances of 322.14 and 342.75 m, respectively (Fig. 1). Some MLGs infected up to six plant genera (Table 1).

Figure 2.

Frequency distributions of pair-wise genetic distance among all Armillaria mellea multilocus genotypes (MLGs). The unimodal, normal distribution of frequencies indicates that sexual recombination, and not somatic mutation, probably contributed to the distinction of MLGs.

Figure 3.

Frequency distributions of linear dimensions of Armillaria mellea multilocus genotypes (MLGs). The linear dimension was estimated as the maximal spatial distance between the farthest isolates sharing the same MLG, for those that were detected more than once in the collection. MLGs detected once in the collection (i.e. from one isolate) were assumed to measure < 1 m (distance class 0–1).

Table 1.   Linear dimensions and plant species from which we recovered repeated Armillaria mellea multilocus genotypes
CloneNumber of isolatesLinear dimensions (m)Plant species
  1. Plant species from which A. mellea has not been reported previously are shown in bold. NA, plant was unidentifiable (i.e. a rotten stump or a dead, standing tree with no foliage).

A373.2Cupressus macrocarpa, Camellia japonica, Quercus sp.
C469.5Cupressus macrocarpa, Camellia sp., Pinus radiata, NA
E487.6Pinus radiata, Pinus ponderosa, NA
F228.6Ilex sp., NA
G23.1Salix sp., Pinus sp.
I232.6Pinus sp., NA
K650.5Pinus sp., Quercus wislizenii, Eucalyptus sp., Arbutus menziesii, NA
L218.8Pinus ponderosa, NA
M27.6Rhododendron sp., NA
N292.7Pinus sp., NA
O324.4Quercus sp., NA
Q25.1Cupressus macrocarpa, NA
S343.9Pseudotsuga sp., Heteromeles arbutifolia
T242.5Quercus rugosa, NA
U220.4Pinus radiata, NA
V1057.8Pinus sp., NA
W1146.0Pinus sp., Juniperus sp., Eucalyptus sp., Robinia pseudoacacia, NA
X425.4Pinus sp., Eucalyptus sp., NA
Y27.9Pinus sp., NA
Z2322.1Pinus sp., Callistemon sp.
AA16342.8Pinus sp., Heteromeles arbutifolia, Rhododendron sp., Callistemon sieberi, Azara microphylla, Peumus boldus, NA
BB218.2Pinus sp., NA
CC28.5Pinus sp.
DD218.1Eucalyptus sp., Taxodium mucronatum
EE273.1Quercus sp., NA

Hardy–Weinberg equilibrium and gametic linkage disequilibrium

When using a single representative copy of each MLG (clone-corrected dataset), the expected heterozygosity (HE) ranged from 0.22 (locus Am080) to 0.72 (locus Am109), and the sample showed a significant heterozygote deficit over all loci (FIS = 0.022; < 0.001; Table 2). When repeated MLGs (all isolates) were maintained in the analysis, a significant heterozygote excess was also detected (FIS = − 0.02; < 0.001). Two loci, Am036 and Am080, consistently deviated from Hardy–Weinberg expectations and exhibited a significant heterozygote deficit in both analyses (< 0.01; Table 2). By contrast, Am059, Am109 and Am125 showed significant heterozygote excess, but only in the dataset that used all isolates.

Table 2.   Number of alleles (Na), expected heterozygosity (HE) and inbreeding coefficient (FIS) of eight microsatellite loci for Armillaria mellea in Golden Gate Park
LocusNaAll isolatesClone-corrected
  1. Analyses were conducted with two datasets, one including all copies of each multilocus genotype (MLG) (all isolates) and a second including a single copy of each MLG (clone-corrected). Significant departures from Hardy–Weinberg equilibrium are coded as follows: ***, < 0.001; **, < 0.01; *, < 0.05.

Am05930.49− 0.28***0.49− 0.05
Am08840.36− 0.050.35− 0.01
Am09160.63− 0.100.63− 0.03
Am10990.68− 0.01*0.720.03
Am12530.47− 0.20*0.48− 0.22
Multilocus360.46− 0.02***0.450.02***

When testing for gametic linkage disequilibrium and after applying Bonferroni corrections for multiple pair-wise comparisons, the null hypothesis of the random association of alleles could not be rejected for any of the pairs of loci (from a total of 28 pairs). Similarly, the index of multilocus gametic disequilibrium rd revealed no significant departure from random recombination of alleles across all loci (rd = 0.004, = 0.33).

Spatial genetic structure

Bayesian analyses conducted in STRUCTURE assigned all MLGs to one genetic cluster, as the maximum log-likelihood of the data was reached for K = 1. Thus, we did not detect any population subdivision among MLGs. The spatial autocorrelograms either including or excluding repeated MLGs converged at an average spatial distance of 131.8 m (distance class of 110–158 m; Fig. 4), indicating that this distance constitutes the clonal subrange of A. mellea in Golden Gate Park. This is the linear spatial dimension above which subterranean expansion of a diploid individual does not affect the overall spatial distribution of MLGs, and corresponds to a probability of genotypic identity F(r) of 0.0075. F(r) did not decrease steadily over increasing spatial distances (Fig. 4). The clonal subrange occurred at a spatial distance at which F(r) first reached a value below 1%, but did not match a null value of F(r) that occurred at spatial distances larger than the two largest MLGs identified in the sample (322.14 and 342.75 m, respectively). The unexpected increase in F(r) between 200 and 300 m (Fig. 4) may signal the occurrence of dispersal by clonal fragmentation. Consequently, we removed isolates that made up the two largest, and probably fragmented, MLGs to recalculate the clonal subrange. Accordingly, the clonal subrange was estimated to be 92.7 m, the size of the third largest MLG in the dataset.

Figure 4.

Analysis of clonal structure in Armillaria mellea using spatial autocorrelograms. The pair-wise kinship coefficient Fij is plotted as a function of spatial distance classes among pairs of isolates. Two distinct analyses were performed: Fij was estimated from a dataset including all isolates (open circles) and from a dataset including only pairs of isolates with different multilocus genotypes (closed circles). The probability of clonal identity F(r) is also plotted against spatial distance class (closed squares, dashed line). The spatial distance at which the two correlograms merge (131.8 m) provides an estimate of the radius of the clonal subrange, and corresponds to an F(r) value close to zero (F(r) = 0.0075).

When all isolates were considered, we observed a sharp decrease in average kinship coefficient over the first four spatial distance classes, a pattern in agreement with an IBD model. The kinship coefficient Fij was significantly positive for the first three distance classes (Fij = 0.19, 0.05 and 0.01, < 0.011, for average spatial distances of 17.3, 48.6 and 86.2 m, respectively). At larger spatial distances, Fij was either significantly negative or did not differ significantly from a random spatial distribution. The significant value of the regression slope (blog = − 0.018, < 0.001) indicated a significant spatial genetic structure: average Fij values between pairs of isolates decreased linearly with an increase in the natural logarithm of the spatial distance (Fig. 5a). The Sp statistic quantifying spatial genetic structure was 0.022.

Figure 5.

Spatial autocorrelograms plotting the pair-wise kinship coefficient Fij as a function of the logarithm of spatial distance class among pairs of Armillaria mellea isolates. Two distinct analyses were used: (a) Fij was estimated from a dataset including all isolates; (b) Fij was estimated from a dataset including only pairs of isolates with different multilocus genotypes. Dotted lines delimit 95% confidence intervals around the null hypothesis of a random spatial distribution of isolates. The solid black lines are regression lines.

When comparisons between isolates sharing the same MLG were excluded from the analysis, the spatial autocorrelogram had a similar shape (Fig. 5b), except that the kinship coefficients estimated at distance classes below the clonal subrange were lower than when all isolates were considered (note the different scales for the y axis of Fig. 5a,b). Fij values decreased over the first three distance classes and had significant positive values at the first distance class (Fij = 0.046, < 0.001, for an average spatial distance of 30.4 m). The regression slope between Fij and the logarithm of spatial distances was significantly different from that expected under a random spatial distribution of MLGs (blog = − 0.003, = 0.003). The Sp statistic quantifying spatial genetic structure was 0.004.


Using a combination of microsatellite genotyping, somatic incompatibility testing and spatial coordinates, we gained insights into the colonization process of an organism that spreads by both aerial dispersal of sexual spores and subterranean, vegetative expansion of individual genotypes. Genetic similarity among neighboring isolates of A. mellea was higher than among more distant isolates. Vegetative, below-ground growth and wind dispersal of basidiospores, which, from our findings, are both spatially limited, contributed to the nonrandom distribution of diploid individuals. The high proportion of unique MLGs and physical linkage equilibrium detected among the microsatellite loci support the important role of sexual reproduction, relative to vegetative growth, in shaping the population genetic structure of A. mellea in Golden Gate Park.

Over all loci, a significant heterozygote deficit was detected, and this is in agreement with a previous study of other A. mellea subpopulations in the USA (Baumgartner et al., 2010) and analyses of other Armillaria species (Saville et al., 1996; Prospero et al., 2008). Inbreeding as a result of subdivision of the sample into multiple subpopulations that do not freely exchange migrants (i.e. Wahlund effect; Wahlund, 1928) can be ruled out, because we did not find evidence of population structure among MLGs. However, a weak population subdivision, as expected under the IBD pattern identified in this study, is not typically detected by assignment methods, such as that implemented in STRUCTURE (Waples & Gaggiotti, 2006). We hypothesize that significant heterozygote deficit may reflect inbreeding as a result of sibling matings or of somatic recombination through haploid–diploid matings, as suggested by Baumgartner et al. (2010). The likelihood of sibling matings is supported by our finding of a higher genetic similarity among neighboring MLGs relative to distant MLGs. Although this process may occur in Armillaria populations, our estimates of high genotypic richness and multilocus linkage disequilibrium are nonetheless consistent with a reproductive life history predominated by outcrossing.

In contrast with previous beliefs with regard to the spread of Armillaria root disease (Redfern & Filip, 1991), our results reveal a more important role of spore dispersal, relative to mycelial growth, in the colonization of new habitat. In Golden Gate Park, a large proportion of MLGs (61 of the 90 in total) were recovered from only a single tree. This contrasts with the presence of one to a few diploid individuals, spanning the roots of many adjacent trees, as documented in past studies of A. mellea populations in California (Rizzo et al., 1998; Baumgartner & Rizzo, 2002). At such study sites, the hosts (grapevines and pear trees) were planted on land that was cleared of infected forest stands, and thus the colonization of the successive crops was primarily a consequence of root infection by resident mycelia. Assuming that resident mycelia were initially rare to absent from much of Golden Gate Park, the planted trees constituted a potential site for the establishment of infections by spores; colonization of the park was mainly by new infection foci. A similar colonization strategy was revealed in the root pathogen Onnia tomentosa, which was shown to initially colonize conifer stands by basidiospores, and then to expand to colonize adjacent trees through mycelial growth (Germain et al., 2009). Similarly, the presence of many small genotypes of the ECM fungus Suillus pungens in a post-fire stand of Pinus muricata suggested that the recolonization of the new seedlings occurred mainly by spores, and not by the resident mycelium that dominated the mature stand before the fire (Bruns et al., 2002). This is in contrast with other tree pathogens, such as Phellinus tremulae and, to a lesser extent, Heterobasidion annosum, whose populations are typically characterized by small genotypes, independent of the forest age, suggesting that spores are the primary dispersal mode for these two pathogens (Holmer et al., 1994; Garbelotto et al., 1999). Because A. mellea can occur on the roots of asymptomatic trees (Baumgartner & Rizzo, 2001), and we sampled only symptomatic trees, our collection may have been biased towards the detection of small, unique MLGs. However, the intensive sampling carried out over the course of 2 yr (309 total samples from mycelial fans and sporocarps, exhaustive examination of symptomatic trees) was the most feasible approach to mapping the A. mellea population of the 409-ha park.

A strong spatial genetic structure was detected when all isolates were maintained in the spatial autocorrelation analyses. These results reflect the spatial grouping of most repeated MLGs up to c. 130 m, reflecting the contribution of mycelial growth to the clustering of repeated MLGs at such distances. However, we detected two MLGs with greater spatial dimensions (322.1 and 342.8 m in length) in the arboretum of Golden Gate Park. These largest two MLGs may represent resident mycelia that pre-dated the planting of the park, although the exact locations of the original forested areas are unknown. A thick layer of wood chips (c. 30 cm) and frequent irrigation in the arboretum (P. Ehrlich, Presidio of San Francisco, Golden Gate National Recreation Area, pers. comm.) lead to excessive soil moisture, which is known to increase mycelial growth of A. mellea within susceptible root tissue (Redfern & Filip, 1991). Alternatively, it is possible that humans have moved infected plant material within the arboretum (e.g. grinding down an infected stump and then using the wood chips as mulch 300 m away), thereby creating the appearance of a large MLG that is actually made up of discontinuous fragments of the same MLG on distant trees. Further evidence in support of human-mediated fragmentation of these two large MLGs is the unexpected increase in the probability of genotypic identity at large distances (Arnaud-Haond et al., 2007).

The examination of the convergence point of the two spatial autocorrelograms that either included or excluded repeated MLGs provided an estimate of the average clonal subrange of A. mellea. The clonal subrange has previously been calculated primarily for plant species, for example, Cymodocea nodosa, a rhizomatous seagrass that spreads both by vegetative and sexual propagules (Alberto et al., 2005). We estimated the clonal subrange for a fungal species with analogous modes of dispersal. Spatial autocorrelograms including and excluding repeated MLGs merged at an average distance of 131.8 m, indicating that this is the characteristic maximum size (clonal subrange) of an A. mellea MLG in Golden Gate Park. Indeed, all MLGs were < 100 m in length, with the exception of two (322.1 and 342.8 m), the latter of which probably resulted from human-mediated fragmentation in the heavily managed arboretum of the park (see previous paragraph). After excluding these two largest MLGs from the analysis, a clonal subrange of 97.2 m was calculated, reflecting the spatial patterns of vegetative growth of A. mellea in a human-managed, ornamental landscape, where the host population probably pre-dates much of the A. mellea population. This clonal subrange can theoretically be used as a minimum sampling distance between trees, when one seeks to maximize the genotypic richness of an A. mellea population. Nonetheless, our measurements are specific to the population studied here, and confirmation in other environments (e.g. natural forests, plantations) is required. In addition, the comparisons of clonal subrange values in different habitats could allow an assessment of the effects of habitat characteristics (e.g. age, host density, levels of disturbance) on the reproductive mode of A. mellea (clonality vs sexual reproduction).

When repeated MLGs were removed from the spatial autocorrelation analysis, a steady decrease in kinship coefficient was observed for the first three distance classes (up to an average spatial distance of 119 m), with a significant and positive spatial autocorrelation detected at the first distance class (average spatial distance of 30 m). This IBD pattern is weak overall and only detectable at the shortest spatial distance class. One of the assumptions of the IBD model is that the population is at equilibrium between genetic drift and gene dispersal (Rousset, 1997). This assumption can be violated, for instance, in a founder population. In our case, we presume that the A. mellea population of Golden Gate Park is probably at drift dispersal equilibrium, in part because over 100 yr have passed since the establishment of the park (McLaren, 1872). Moreover, the fungus reproduces sexually each year, based on our observations of sporocarps. Consequently, the large number of sexual generations that have occurred in the A. mellea population of Golden Gate Park have probably brought the population to drift dispersal equilibrium, and thus its structure could be shaped by IBD (Hardy & Vekemans, 1999).

Theoretically, a nonrandom spatial distribution of genotypes, which is indicative of spatial genetic structure, can be caused by various factors, such as limited gene dispersal in space (i.e. IBD), local adaptation (e.g. host specialization) or the presence of physical barriers to gene flow (e.g. a mountain range) among genetically differentiated clusters of genotypes (Guillot et al., 2009). Among these factors, limited gene dispersal has the most support as the contributing factor to spatial genetic structure. Indeed, there was no evidence of genetically differentiated clusters of MLGs within the park (e.g. no clustering of genotypes infecting pine trees). Furthermore, in several instances, a single MLG infected multiple tree species, and this is consistent with the broad host range of A. mellea (Raabe, 1962) and the lack of reports of host specialization from the literature. Our finding of significant spatial genetic structure is, rather, in agreement with direct estimates of basidiospore dispersal in A. novae-zelandiae: most basidiospores are deposited within a few meters from the sporocarps, whereas deposition up to 150 m is much less frequent (Power et al., 2008). The IBD pattern suggests that spore dispersal mainly occurs over a few meters. This very localized dispersal may provide ecological benefits. All spores from a given sporocarp of A. mellea can undergo plasmogamy with their diploid parent (Anderson & Ullrich, 1982). The diploid nucleus often displaces the haploid nucleus within the haploid mycelium (Rizzo & May, 1994). In this way, a diploid–haploid mating between parent and progeny expands the parent’s mycelium, and thus its substrate base. The parental genotype persists, and thus we may have underestimated local spore dispersal, because such haploid–diploid matings following local dispersal of haploid spores are undetectable by our genotyping methods.

Our findings suggest that spore dispersal is spatially restricted in A. mellea, but rare instances of long-distance dispersal may promote genetic homogeneity between spatially separate subpopulations. Indeed, gene flow through spore dispersal has been suggested as a means of preventing genetic differentiation among subpopulations of A. ostoyae and A. mellea at spatial scales beyond hundreds of kilometers (Prospero et al., 2008; Baumgartner et al., 2010). Long-distance spore dispersal is thus plausible and is probably important for the colonization of new habitat by A. mellea. Similarly, low levels of genetic differentiation have been found among distant subpopulations of wood decay fungi (Stenlid & Gustafsson, 2001). For example, distant subpopulations of the forest pathogen Heterobasidion annosum are not genetically differentiated (Stenlid et al., 1994), in spite of evidence that most spores are deposited within a few meters of a sporocarp, and only 0.1% of the total released spores are estimated to be trapped 100 m away (Kallio, 1970). Like H. annosum, large population sizes, high spore production per sporocarp and rare, long-distance spore dispersal for A. mellea probably counteract genetic divergence at a large spatial scale. Alternatively, although individual-based methods, such as spatial autocorrelations, characterize contemporary gene dispersal, population-based methods estimate evolutionary averages of dispersal. The latter approach depends on the establishment of an a priori definition of a subpopulation that may not accurately reflect the population structure. The spatial distribution of genetic diversity is also affected by historical events (e.g. establishment by multiple founder events vs a single, massive invasion by a large panmictic group). Moreover, gene flow that causes genetic homogeneity between subpopulations is the product of the migration rate (i.e. the fraction of dispersers entering the subpopulation each generation) and the effective population size (Slatkin, 1987). Pathogen populations are typically characterized by very large population sizes (e.g. millions of spores are produced per mushroom), and therefore subpopulations may be genetically homogeneous in the presence of very low migration rates (Rieux et al., 2011; Travadon et al., 2011). Our study illustrates that, when the distribution of genotypes is examined at a local spatial scale, an indirect estimate of the scale at which fungal spore dispersal occurs can be obtained from genetic and spatial data, and thus provides an alternative to the labor-intensive, direct trapping methods.


The authors thank Peter Ehrlich (Presidio of San Francisco, Golden Gate National Recreation Area, San Francisco, CA, USA) for references on the history of Golden Gate Park, and C. Dutech (INRA, Bordeaux, France) for valuable comments on previous versions of the manuscript. This research was funded by the United States Department of Agriculture – Agricultural Research Service (USDA-ARS) and the National Science Foundation Biocomplexity Program.