Ongoing coevolution in mycorrhizal interactions

Authors


Author for correspondence:
Jason D. Hoeksema
Tel: +1 662 915 1275
Email: hoeksema@olemiss.edu

Abstract

Contents

 Summary286
I.Introduction287
II.Background: local coevolutionary selection and the hierarchy of candidate coevolving traits inmycorrhizal interactions287
III.Question 1: is coevolutionary selection ongoing between plants and mycorrhizal fungi, and, if so, what form(s) does it take?290
IV.Question 2: how does community context affect local coevolutionary selection in the mycorrhizal interaction?291
V.Question 3: what are the relative roles of the three elements of the geographic mosaic theory of coevolution (selection mosaics, hotspots and coldspots, and trait remixing) and nonreciprocal selection from biotic and abiotic factors in driving trait diversification of plants and mycorrhizal fungi?294
VI.Conclusions297
 Acknowledgements297
 References298

Summary

Coevolution can be a potent force in maintaining and generating biological diversity. Although coevolution is likely to have played a key role in the early development of mycorrhizal interactions, it is unclear how important coevolutionary processes are for ongoing trait evolution in those interactions. Empirical studies have shown that candidate coevolving traits, such as mycorrhizal colonization intensity, exhibit substantial heritable genetic variation within plant and fungal species and are influenced by plant genotype × fungal genotype interactions, suggesting the potential for ongoing coevolutionary selection. Selective source analysis (SSA) could be employed to build on these results, testing explicitly for ongoing coevolutionary selection and analyzing the influence of community context on local coevolutionary selection. Recent empirical studies suggest the potential for coevolution to drive adaptive differentiation among populations of plants and fungi, but further studies, especially using SSA in the context of field reciprocal transplant experiments, are needed to determine the importance of coevolutionary selection compared with nonreciprocal selection on species traits.

I. Introduction

Coevolution, the evolutionary changes in species driven by natural selection imposed by interacting species reciprocally on each other, has a long history of study in ecology and evolution. Throughout their writings, Darwin and Wallace both recognized the importance of the coevolutionary process in evolutionary diversification, emphasizing the primacy of species interactions as drivers of adaptive evolutionary change in species. Indeed, it has been argued that Darwin’s book on orchids provided the initial catalyst for all subsequent studies of coevolution and the evolution of extreme specialization (Ghiselin, 1984; Thompson, 1994). Since Darwin and Wallace, studies on coevolution have shown that species interactions can drive rapid and sustained evolutionary change in species at multiple spatial and temporal scales, generating genetic diversity within populations, leading to adaptive differentiation among populations, and often leading to ecological speciation (Schluter, 2009). In fact, it has been argued that much of the diversity on earth is a consequence of coevolutionary diversification in species interactions (Ehrlich & Raven, 1964; Thompson, 1994, 2005). Clearly, studies of coevolution in species interactions can provide insight into the fundamental processes generating and maintaining biodiversity, including genetic and phenotypic diversity within and between species.

Some interactions between plants and mycorrhizal fungi may be among the oldest of coevolving species interactions; fossil and molecular evidence supports the hypothesis that the arbuscular mycorrhizal (AM) symbiosis originated around the time that plants first colonized land (460–475 Mya; reviewed by Morton, 1999; Cairney, 2000; Brundrett, 2002). Although we cannot directly confirm the role of coevolutionary selection in the development of the various types of mycorrhizal symbiosis, it is highly likely that reciprocal selection between mycorrhizal plants and fungi has driven the reciprocal evolution of plant and fungal traits governing the interaction, at many stages in geological time (Brundrett, 2002). As complementary traits evolved in plants and fungi, such as fungal dependence on host plant carbon for energy and hosts providing a more hospitable environment for fungi (Brundrett, 2002), each evolutionary change in one partner would have made further evolution in the other partner more advantageous, promoting coevolutionary selection for further complementarity in traits. Complex suites of traits adapted to the symbiosis would not evolve in either partner without repeated bouts of reciprocal selection for complementary traits in both partners (Brundrett, 2002). Although multiple reviews have addressed the potential role of coevolution in the early evolution of mycorrhizal interactions (Morton, 1999; Cairney, 2000; Halling, 2001; Brundrett, 2002), it is still an open question whether ongoing coevolutionary processes continue to contribute to diversification in mycorrhizal plants and fungi.

Cairney (2000) argued that, although coevolutionary selection may have been important in the origins of the various mycorrhizal symbioses, plants and mycorrhizal fungi have mostly exerted ‘diffuse’ coevolutionary selection pressures on each other, and are now more likely to be undergoing parallel evolution in response to abiotic environmental variation, rather than exerting important recent or ongoing coevolutionary selection on each other (also see Helgason & Fitter, 2009). This viewpoint provides a valid (although not mutually exclusive) alternative to the hypothesis that ongoing coevolution is important for contemporary trait evolution in plants and their mycorrhizal symbionts. I suggest, however, that the necessary evidence – specifically, data from experiments actually estimating selection on plant and mycorrhizal fungal traits from coevolution and other sources – has mostly not yet been gathered to powerfully distinguish between those possibilities. Here, I review the potential for ongoing coevolution between plants and mycorrhizal fungi, with four main goals: to suggest a theoretical framework to guide research on ongoing coevolution in the mycorrhizal symbiosis; to pose three key research questions within this framework; to assess current empirical knowledge regarding the answers to those questions; and ultimately to inspire empiricists to creatively conduct the missing experiments. Section II provides background on coevolutionary selection and the hierarchy of candidate coevolving traits in plants and mycorrhizal fungi. Sections III, IV, and V then each develop a key research question on ongoing coevolution in mycorrhizal interactions, followed by some general conclusions in Section VI.

II. Background: local coevolutionary selection and the hierarchy of candidate coevolving traits in mycorrhizal interactions

1. Local coevolutionary selection

Coevolutionary selection between two species within a single local community, that is local coevolutionary selection, has the potential to act as a potent evolutionary force on the traits of species by virtue of the feedbacks driven by its inherent reciprocality. The reciprocality of coevolutionary selection means that the fitnesses of two interacting species both depend not only on their own genotype (and associated traits), but also on the distribution of genotypes (and traits) in the other species. Specifically, there is a genotype by genotype (G × G) interaction effect on the fitnesses of both species. As a result of this reciprocality, adaptive changes in the traits of one species in response to selection may trigger subsequent adaptive changes in the traits of the second species, which in turn feed back to cause adaptive evolution of the first species, and so on. Put another way, the relative fitness of genotypes in one species is context-dependent and the context itself can also evolve (Wade, 2003). For example, the relative fitnesses of different genotypes of a hummingbird and a hummingbird-pollinated plant depend not only on hummingbird bill shape (which differs among hummingbird genotypes), or on flower morphology (which differs among plant genotypes), but on the match between the two, generating a G × G interaction for both hummingbird fitness and plant fitness. If hummingbird bill shape adapts in response to selection driven by the average flower morphology in the plant, reciprocal selection on flower morphology (driven by the new average hummingbird bill shape) changes as a result.

2. Heritability and genetic architecture of candidate coevolving traits

The hummingbird flower example illustrates the point that coevolution occurs not between two species, but between phenotypic traits in two species (Kiester et al., 1984; Wade, 2003; Ridenhour, 2005). Thus, the first step in investigating the potential consequences of ongoing coevolution for mycorrhizal interactions must be to identify phenotypic traits of plants and mycorrhizal fungi that are potentially the targets of ongoing coevolutionary selection, and to assess the degree to which they are genetically variable among individuals and thus subject to natural selection (see Supporting Information Notes S1 for a discussion of related key concepts from quantitative genetics, and their application in analyzing mycorrhizal interactions).

3. The hierarchy of candidate coevolving traits in mycorrhizal interactions

Although most analyses of adaptive evolution have focused on single traits or small numbers of traits, it is now widely recognized in evolutionary biology that environments impose natural selection not on single traits, but simultaneously on complex suites of species traits (Lande & Arnold, 1983; Phillips & Arnold, 1989; Kingsolver et al., 2001; Blows, 2007). Furthermore, we now recognize that traits can be discerned at multiple levels of a hierarchy, ranging from the proteins produced directly by genes to higher level traits that result from integration of multiple components of an organism’s physiology (Whitham et al., 2006). Thus, consideration of potential traits under coevolutionary selection in mycorrhizal interactions may begin with lower level traits that are probably not far removed from the controlling genes in plants and fungi, but must also consider higher level traits, including those controlling the community structure of mycorrhizal interactions. For all of these candidate traits, we need studies of their phenotypic and genetic variability within and among populations, heritability, and genetic architecture (including correlations with other traits), so that that we can assess their potential to evolve in response to ongoing selection.

A priori, the most obvious candidate traits that may have at one time undergone coevolutionary selection in mycorrhizal interactions are the lower level traits controlling the formation of the mycorrhizal interface, for example the ability to form an arbuscule in AM symbioses or a Hartig net in ectomycorrhizal (EM) symbioses. However, I know of no studies estimating the heritability of these lower level traits, and it has been suggested that they are fixed within species of plants and mycorrhizal fungi (Morton, 1999), although screening of whole soil fungal communities for allelic variants of key functional genes (e.g. Artz et al., 2009) may increase our power to detect variation in such traits. Lower level traits governing specificity in signaling and recognition between interacting partners in mycorrhizal interactions deserve special attention as candidate coevolving traits because a high degree of specificity in such traits could allow coevolutionary cycling to drive fluctuating polymorphisms at corresponding recognition loci in plants and fungi, as has been found for the major histocompatibility complex (MHC) loci in vertebrates (Hughes & Nei, 1988) and self incompatibility loci in fungi (May & Matzke, 1995). For example, if plants are capable of some degree of specificity in recognition of fungi, then selection should strongly favor recognition by plants of beneficial vs nonbeneficial (pathogenic or mycorrhizal) taxa of fungi, and subsequent exclusion or rejection of nonbeneficial fungi (Kiers & van der Heijden, 2006; Bever et al., 2009). In turn, pathogenic or nonbeneficial mycorrhizal fungi would be under selection to avoid recognition, which would drive further reciprocal evolution in the recognition genes of the plant.

Some evidence points to the existence of some specificity in mycorrhizal signaling: AM plants can apparently distinguish between broad groups of soil fungi (AM vs ericoid vs pathogenic; Genre et al., 2009) and can under some circumstances allocate preferentially to more beneficial AM fungal partners (Bever et al., 2009), not all plant and AM or EM mycorrhizal fungal taxa exhibit identical preferences for each other (e.g. Molina et al., 1992; Halling, 2001; Vandenkoornhuyse et al., 2003), and molecules contained in root exudates have been shown to serve as signaling molecules for both the EM symbiosis (Lagrange et al., 2001) and the AM symbiosis (Reinhardt, 2007). However, we still have much to learn about specificity in mycorrhizal signaling as it relates to coevolution. In particular, the degree of specificity in recognition of signaling molecules by different members of the rhizosphere community is still poorly understood, and we have a pressing need for studies that quantify phenotypic variation and heritability of signaling and recognition traits within species of both plants and mycorrhizal fungi, including characterization of the range of specificity in those traits.

Mycorrhizal colonization intensity, measured as the proportional or absolute amount of root length or fine root tips colonized, is a higher level trait that potentially integrates all of the lower level traits discussed so far and may be controlled in part by genes in both plants and mycorrhizal fungi. Numerous studies have now demonstrated significant genetically based variation for these traits within plant species (e.g. Dixon et al., 1987; Graham & Eissenstat, 1994) and within fungal species (e.g. Lamhamedi et al., 1990; Burgess et al., 1994). Although only a handful of such studies have been designed to allow the estimation of the heritability of those traits (e.g. Rosado et al., 1994a,b; Kropp, 1997), the few examples suggest that, for some pairs of plant and fungal species in natural populations, sufficient additive genetic variation exists for coevolutionary selection to potentially act on these traits in both plants and mycorrhizal fungi (but see Gahoonia & Nielsen, 2004).

Some of the studies of variation in colonization intensity provide insight into the genetic architecture governing those traits in plants and fungi. Specifically, the distributions of such traits observed among progeny from controlled crosses (Lamhamedi et al., 1990; Kropp, 1997; Tagu et al., 2005) suggest that EM colonization intensity of tree hosts is under polygenic control, rather than under control by a small number of genes of major effect. Moreover, several studies (e.g. Toth et al., 1990; Zhu et al., 2001) have demonstrated reduced mycorrhizal colonization intensity in highly selected modern cultivars of crop plants compared with older cultivars, suggesting that artificial selection for crop traits (such as disease resistance) may have resulted in correlated selection against mycorrhizal colonization intensity. These results parallel those of a recent study showing how artificial selection has reduced the ability of modern cultivars of soybean (Glycine max) to benefit from root symbiosis with rhizobia bacteria (Kiers et al., 2007). Intriguingly, one study found that putative quantitative trait loci (QTLs) in the genome of the tree Populus trichocarpa controlling mycorrhizal colonization intensity by the EM fungus Laccaria bicolor mapped onto the P. trichocarpa genome very near a previously discovered QTL that contributes to mediation of poplar resistance to fungal rust pathogens (Tagu et al., 2005). If such findings are confirmed for poplar or other plants, they raise the possibility that one or more closely linked loci may have simultaneous (pleiotropic) effects on both mycorrhizal colonization and disease resistance traits of plants. This type of pleiotropy could potentially constrain coevolution of plants with mycorrhizal and/or pathogenic fungi, because the two types of partners may impose conflicting selection pressures on the same genes (Griswold & Whitlock, 2003).

A second class of higher level traits that potentially integrates the lower level traits discussed so far is the set of ecological interaction strengths for a particular interaction between an individual plant and a fungal genet. Relevant plant traits, for example, include the response of the plant to the presence of the fungus, in terms of absolute or relative performance. Many studies have now found significant genetic variation within plant or mycorrhizal fungal species (e.g. Boerner, 1990; Hetrick et al., 1995; Eason et al., 2001; Schultz et al., 2001; Koch et al., 2006; Piculell et al., 2008; Hoeksema et al., 2009; Karst et al., 2009; Seifert et al., 2009) for their ecological effect on or response to each other. For example, Koch et al. (2006) found significant variation in host-plant growth promotion among isolates from a natural population of the AM fungus Glomus intraradices. Schultz et al. (2001) found substantial differences between two ecotypes of the grass Andropogon gerardii for responsiveness to infection with AM fungi. These examples demonstrate the presence of genetic variability within or between natural populations for ecological effect and response traits of mycorrhizal interactions, allowing for the possibility that such traits could be subject to ongoing natural selection. As discussed for colonization intensity above, however, we understand very little about the specificity of these traits, and are in need of studies that test whether they vary depending on the identity of the partner species.

At the highest levels of the trait hierarchy we find traits that are referred to as ‘community phenotypes’ whereby the genotype of one species determines the composition of other species with which it interacts (Whitham et al., 2006). Some of the best examples of community phenotypes are emerging from studies of poplar trees (Populus species), where it has been shown that different plant genotypes host different communities of arthropods, bacteria, fungi, and vertebrates (Whitham et al., 2006). Some of these community phenotypes of poplar have high heritabilities, suggesting that these higher level traits have a genetic basis and that poplar populations could respond to natural selection by changing the communities of other species with which they interact. Whether such adaptations result from selection pressure by those same communities, and whether such adaptation would result in reciprocal selection pressure on those same communities, has not been determined. Several studies report significant genetic variation within plant species for mycorrhizal fungal community structure (e.g. Taylor et al., 2004; Korkama et al., 2006). Such studies provide intriguing evidence of a genetic basis in plants for mycorrhizal fungal community structure, raising the possibility that a plant population could adapt to changing environments via a shift in the species composition of the mycorrhizal fungal guild with which they are interacting. Fungal guild composition would only be considered a coevolving trait, however, if both plant and fungal fitnesses depend on fungal guild composition, and also on the traits of one or more members of the fungal guild.

III. Question 1: is coevolutionary selection ongoing between plants and mycorrhizal fungi, and, if so, what form(s) does it take?

1. Theoretical framework

Local coevolutionary selection between two species can take a variety of different forms, causing diversifying, stabilizing, or directional selection on the traits mediating species interactions. In part, which form of coevolutionary selection operates may depend on the ecological outcome of the interaction and the patterns of symmetry in fitness benefits between different combinations of interacting genotypes. For example, parasitic ecological outcomes can generate fluctuating polymorphisms for traits in both host and parasite species, driven in part by negative frequency-dependent coevolutionary selection that favors rare genotypes in one or both of the host and the parasite species (reviewed by Thompson, 2005). If the ecological outcome between two species is a consistent mutualism, in which both species consistently benefit from the presence of the other, coevolutionary selection can favor complementarity of traits in the two species via stabilizing coevolutionary selection. This pattern of selection is driven by positive frequency dependence, resulting in fixation of single mutualistic genotypes within local populations of each species at equilibrium (Law, 1985; Law & Koptur, 1986; Parker, 1999). Bever (1999) showed, however, that if the fitness benefits derived by hosts and symbionts in a mutualism are asymmetrical among genotypic combinations, that is, no particular combination of host and symbiont genotypes is best for both the host and the symbiont, then negative frequency-dependent coevolution can result, producing diversifying selection on both hosts and symbionts, analogous to the dynamics predicted for parasitic interactions.

In many putative mutualisms, ecological outcomes may actually span a continuum between mutualism and parasitism, changing over time or space depending on the traits of the participants and the ambient environmental conditions (Bronstein, 2001). Indeed, many putative mutualisms that involve the exchange of costly benefits may be under constant risk of sliding into parasitism, because directional selection will always favor individuals that can cheat the interaction by receiving benefits without returning the favor (Maynard Smith, 1989; Williams & Nesse, 1991; Herre et al., 1999). In such cases the dominant form of coevolutionary selection on traits in the interaction may also change over time and space, if variation in ecological outcome causes corresponding variation in the differential success of genotypes. To predict the effects of coevolution between plants and mycorrhizal fungi on the evolution of traits within and between populations, an essential question that needs to be answered is Question 1: is coevolutionary selection ongoing between plants and mycorrhizal fungi, and, if so, what form(s) does it take?

2. Answering Question 1

Indirect evidence from distributions of ecological outcomes  Ecological outcomes in mycorrhizal interactions may range from mutualism to parasitism depending on the environmental context and depending on plant or fungal genotypes (Lewis, 1973; Smith & Smith, 1996; Johnson et al., 1997; Egger & Hibbett, 2004; Jones & Smith, 2004; Karst et al., 2008). This observation suggests that, if ongoing coevolutionary selection is acting on traits governing mycorrhizal interactions, it is likely to be spatially and temporally variable, and unlikely to typically follow the predictions of simple mutualistic coevolutionary models that assume interactions are always mutualistic and that assume symmetry in fitness benefits among different combinations of partners (Law, 1985; Law & Koptur, 1986; Parker, 1999). Rather, selection may be influenced by a mixture of beneficial and parasitic ecological outcomes, by a mixture of symmetries of benefit between different combinations of plants and fungi, and by the effective degree of symbiotic intimacy of the interaction (Guimaraes et al., 2007), which is perhaps uniquely ambiguous for mycorrhizal interactions (Thompson, 2005).

Indirect evidence from tests of G × G interactions within populations  An indirect method to test for ongoing local coevolutionary selection in mycorrhizal interactions is to test for G × G interaction effects on plant and fungal fitness within a population. Ideally, such an experiment would be carried out using large numbers of genotypes (e.g. clones, open-pollinated families, half-sib families, and full-sib families) of both plants and fungi sampled randomly from a population of interest, paired experimentally in all factorial combinations, and grown in natural conditions over a time span allowing the estimation of lifetime fitness for each species in each combination of genotypes. Although this ideal experimental design has not yet been achieved for any plant–mycorrhizal interactions, tests for G × G interactions under artificial conditions, using small numbers of genotypes, and only estimating components of fitness still provide valuable information on the potential for coevolutionary selection in an interaction (e.g. Dixon et al., 1987; Rosado et al., 1994b; Otero et al., 2005; Piculell et al., 2008; Hoeksema et al., 2009). Finding significant G × G interaction effects on the fitness of both plants and fungi suggests that ongoing coevolutionary selection may be occurring, although no information is necessarily provided about which traits are the targets of coevolutionary selection, or the form of coevolutionary selection (e.g. positive vs negative frequency dependence). Moreover, the coevolutionary consequences of G × G interactions between the genomes of species will depend on the degree of co-transmission of those genomes among generations, as a result of nonrandom genotype associations (Ridenhour, 2005; Wade, 2007).

Direct evidence from selective source analysis (SSA)  Direct evidence for ongoing coevolutionary selection between plants and mycorrhizal fungi could be obtained by directly estimating coevolutionary selection on candidate coevolving traits. Statistical methods for conducting such an analysis are now available in the form of selective source analysis (SSA) (Ridenhour, 2005). While traditional selection analysis estimates total natural selection on a trait (Lande & Arnold, 1983) as the relationship between fitness and the value of a trait without regard to the source of selection pressure, SSA is designed to estimate selection on a trait resulting from multiple environmental sources, including interactions with other species. For example, SSA on a single quantitative trait of a plant, including selection by a quantitative trait of a mycorrhizal fungal species, requires the observation of multiple instances of interaction between the two species, with measurements of the plant trait (x), the fungal trait (y), and plant fitness (wp) for each instance. Also, partial regression coefficients (φi) must be estimated for the following multiple regression model:

image(Eqn 1)

where α is the intercept and the partial regression coefficients (φ1 ... φ5) represent the selection gradients on the plant from different environmental sources. In general, coefficients of linear terms represent directional selection gradients, and coefficients of squared terms represent stabilizing or diversifying selection. For example, φ1 and φ3 are the directional and stabilizing/diversifying (respectively) selection gradients on trait x of the plant, from sources other than the fungus, while φ2 and φ4 estimate selection gradients on the plant resulting from variation in the fungal trait, independent of the value of trait x of the plant. Most importantly for detection of coevolutionary selection, φ5 is the selection gradient on plant trait x resulting from its covariance with fungal trait y. Conducting the same analysis for fungal fitness (wf) would yield estimates for selection from the same sources on the fungus. Finding nonzero estimates for φ5 in both analyses would represent evidence for ongoing coevolutionary selection between traits x and y in the mycorrhizal interaction. Comparing results across the two analyses provides clues about the form of coevolutionary selection; for example, a positive φ5 for fungal fitness and a negative φ5 for plant fitness would suggest negative frequency-dependent coevolutionary selection. In contrast, finding a nonzero estimate for φ1 but not for φ5 in both analyses would support the hypothesis that selection from sources other than the interaction is more important than coevolutionary selection for ongoing evolution of the trait in question. Notes S2 provide additional details on SSA and its potential application to mycorrhizal interactions.

IV. Question 2: how does community context affect local coevolutionary selection in the mycorrhizal interaction?

1. Theoretical framework

Local coevolutionary selection between two focal species may be influenced by additional species interacting with one or both of the focal species. For example, the presence of red squirrels (Tamiasciurus hudsonicus) alters the coevolutionary selection that conifers and crossbills (Loxia curvirostra) exert on each other’s cone and bill morphology, respectively (Benkman et al., 2001). When coevolutionary selection between two species is altered in this way by the presence of additional species, some researchers have referred to the result as ‘diffuse coevolution’ (Strauss et al., 2005). This terminology, at least when it is carefully defined and used sensuStrauss et al. (2005), does not imply that coevolution is unimportant when it is diffuse; rather, the terminology is used to distinguish whether multiple species have independent or interactive selective effects on another species in a coevolving interaction. Thompson (2005) argues that one-to-one vs diffuse coevolution is an artificial dichotomy that distracts from the rich variety of potential forms that multi-specific coevolution, that is, coevolution between multi-specific guilds of species, can take.

Regardless of terminology, it is clear that the presence of multiple species on either side of an interaction, regardless of host specificity, does not indicate that ongoing coevolution is unimportant or weak for all species involved (Thrall et al., 2007). Rather, community context can modify the ways in which species exert ongoing selection pressures on each other’s traits (Strauss et al., 2005; Thompson, 2005), and ongoing coevolution has the potential to be an important driver of trait diversification in diverse multi-specific interactions, such as mycorrhizal interactions. Low-diversity symbiotic interactions, in which individuals of the same two species maintain persistent contact with each other throughout their lifetimes, may allow a greater chance for the evolution of extreme specialization between two species; however, extreme specialization is only one potential consequence of ongoing coevolution in an interaction (Thompson, 2005). At the other extreme, free-living interactions, in which individuals on both sides of the interaction encounter multiple individuals of different species throughout their lifetimes, may not favor the evolution of extreme specialization between two species. Rather, they may favor coevolutionary alternation in diverse interactions of hosts and their enemies, development of multi-specific mutualistic networks through the accumulation of species converging on a core set of traits, or some combination of these two extremes (Thompson, 2005). Thus, the observation of high diversity and numerous generalist taxa participating in mycorrhizal interactions cannot be used as stand-alone evidence that ongoing coevolution is unlikely or unimportant in mycorrhizal interactions, as argued by Cairney (2000). Rather, the diversity of taxa participating in mycorrhizal interactions leads to a corresponding variety of ways in which ongoing coevolution may operate (if it operates at all) in mycorrhizal interactions, and what we really need is to answer Question 2: how does community context affect local coevolutionary selection in the mycorrhizal interaction?

2. Answering Question 2

Hypotheses on how community context affects coevolution in mycorrhizal interactions  A pervasive characteristic of mycorrhizal interactions is that they are not strictly pairwise, with guilds of multiple fungal species being typical for both AM and EM interactions, and guilds of multiple plant species typical for AM interactions and occasional for EM interactions (Smith & Read, 2008). Moreover, these interactions are embedded in a broader community of interspecific interactions, including pathogens and herbivores attacking plants, pollinators and frugivores providing mutualistic services to plants, fungivores consuming mycorrhizal fungi, and saprobic fungi competing with mycorrhizal fungi for mineral nutrients. Thus, in order to understand the potential role of ongoing coevolution in the interaction between any particular pair of plant and fungal species, we need to consider how the presence of other species within the plant and fungal guilds, as well as other types of species interaction effects on the plants and the fungi, may influence those coevolutionary dynamics. Below, in the specification of Hypotheses A1-A3, I will argue that there is no a priori reason to assume that community influences diminish the importance of ongoing coevolution in mycorrhizal interactions, but that such influences may simply add to the number and complexity of ways in which selection (coevolutionary or not) could influence trait evolution.

Specifically, I suggest that there are multiple distinct ways in which diversity and species composition in guilds of plants and mycorrhizal fungi may affect ongoing coevolution among those species. Using the simplified example of a two-species fungal guild interacting with a single plant species, I suggest three hypotheses.

Hypothesis A1: multiple fungi have identical selective effects on plant traits.  Multiple species within guilds of mycorrhizal fungi exert identical selection pressures on the same plant traits. As a result, ongoing coevolutionary selection between traits of the plant and traits of particular mycorrhizal fungi is unaffected by the relative abundances of species in the fungal guild. Thus, in an interaction between a single plant species and two mycorrhizal fungal species, a coevolving trait in the plant (such as overall susceptibility to fungal colonization) is subject to the same net selection gradient from traits of the fungal guild, regardless of the species composition of the fungal guild, or interactions among the fungal species. In turn, the coevolving trait in each member of the fungal guild (such as compatibility with the plant) experiences the same selection pressure from the plant. This hypothesis can be considered a null model for how guild composition influences coevolution in mycorrhizal interactions.

Hypothesis A2: multiple fungi have selective effects on different plant traits.  Multiple species within a guild of mycorrhizal fungi exert selection pressure on different plant traits. As a result, ongoing coevolutionary selection between any particular plant trait and any particular mycorrhizal fungal species will be unaffected by the composition of the rest of the fungal guild, unless there are ecological interactions between fungal species, such as competition, or unless multiple plant traits are genetically correlated. Under Hypothesis A2, the net selective pressure exerted by a mycorrhizal fungal guild on a particular plant trait at any point in space and time may be dominated by the numerically most abundant or frequent member of the guild, as seems to be the case in the well-studied parasitic interactions between snails and a diverse guild of trematode parasites in New Zealand (Lively & Dybdahl, 2000). If fungal community composition changes over space or time, the plant traits that are the main targets of selection would be expected to change accordingly.

Hypothesis A3: multiple fungi have different selective effects on the same plant traits.  Multiple species within a guild of mycorrhizal fungi exert different selective pressures on the same plant trait(s). As a result, fluctuations in the relative abundances of different species in the fungal guild are likely to significantly alter selection on the focal plant trait(s). If plants cannot recognize specific partner fungal species and cannot choose to interact with better partners, and fungal species vary from beneficial to parasitic on the plant, then selection by the fungal guild may act on the plant trait of overall compatibility or susceptibility to the mycorrhizal interaction. The net selective effect of the fungal guild on this plant trait will depend on the average degree of benefit derived by the plant from allowing fungal colonization. Thus, directional selection on that trait could vary, depending on the relative abundance of beneficial and parasitic fungal genotypes in the fungal guild, both within and between species. Alternatively, if plants can recognize different mycorrhizal fungal taxa, and such recognition is expressed as a single higher level trait, that is, a community phenotype, then selection on that trait by the fungal guild could vary from stabilizing to diversifying, depending on the relative abundance of beneficial and parasitic fungal genotypes.

Diversity and ‘diffuse’ coevolution in mycorrhizal inter-actions  None of the three hypotheses outlined above implies that coevolution is less important in mycorrhizal interactions, simply because diverse species guilds are involved, even though selection could be diffuse (sensuStrauss et al., 2005) under Hypotheses A2 and A3. Although the traits that are the main targets of selection (Hypothesis A2) or the net form of selection on a particular trait (Hypothesis A3) may change with guild composition, coevolution does not stop when multiple species are introduced into a guild. Rather, diversity in interacting guilds introduces the possibility that coevolution between a particular plant trait and a particular mycorrhizal fungal trait may vary with community context. Sometimes, the selective effect of one species on a trait may be diminished by the presence of another species in the guild, and sometimes it may be enhanced. Indeed, one possible form of diffuse coevolution is a case in which the overall strength of the selective effect exerted on a trait by two different species is synergistically enhanced by the interaction of the two species, over and above their independent selective effects. Finally, all three hypotheses show how ongoing coevolutionary selection can be important for trait evolution in a diverse interaction, without selection for specificity in either guild.

Indirect evidence for the three hypotheses  Coevolutionary theory predicts that free-living mutualisms, in which individuals in one guild interact with multiple individuals of different species in another mutualist guild during their lifetimes, should not continue to coevolve in a strictly pairwise fashion. Rather, long term, these interactions will accumulate species in one or both interacting guilds, which experience selection for convergence on a core set of traits making the interaction possible (Thompson, 2005). In addition, theory on coevolution in diverse host–parasite interactions suggests that, as mutualistic interactions accumulate more species, they may be more susceptible to cheaters, who exploit the interaction by receiving benefits without providing benefits in return (Thrall et al., 2007). Most detailed studies of free-living mutualisms have confirmed this prediction (reviewed by Thompson, 2005). Although it is not clear whether mycorrhizal interactions are predominantly free-living or symbiotic (Thompson, 2005), some mycorrhizal interactions seem to conform to the pattern expected for free-living interactions, with multiple species on one or both sides of the interaction, including taxa that vary in their function in the interaction (Sanders, 2002; Klironomos, 2003). This observation provides indirect evidence that rejects Hypothesis A1, but does not distinguish between Hypotheses A2 and A3. For example, Klironomos (2003) paired plant and fungal species in all combinations within multiple AM communities, and found that no plant performed best with the same fungal isolate and that plant growth responses to fungi ranged from strongly mutualistic to strongly parasitic within communities, depending on which partner species were paired.

Evidence for functional complementarity and niche partitioning among different AM mycorrhizal fungal species suggests that different traits could potentially be under selection in interactions between particular pairs of plant and fungal species (Hypothesis A2). Hart & Reader (2002) and Maherali & Klironomos (2007) have found evidence for functional complementarity among the major lineages of AM fungi, whereby fungi in the Glomeraceae exhibit higher root colonization rates and confer greater pathogen protection to host plants, compared with fungi in the Gigasporaceae which exhibit higher rates of extraradical hyphal growth and confer greater phosphorus (P) uptake to host plants. Consequently, Glomeraceae may exert selection for decreased pathogen resistance in plants, whereas Gigasporaceae may exert selection for decreased P uptake traits of plants. Moreover, these results suggest that diverse guilds of AM fungi may coexist through niche complementarity, which could evolve as a consequence of coevolutionary displacement (sensuThompson, 2005) between competing fungal species.

Direct evidence for the three hypotheses from selective source analysis  Although such an analysis has not yet been conducted, SSA could theoretically be used to distinguish among multiple hypotheses (including the three posed above) for how community context influences ongoing coevolution in mycorrhizal interactions. In Eqn 1, φ1 and φ3 are the directional and stabilizing/diversifying (respectively) selection gradients on trait x of the plant, from environmental sources other than the focal fungus, such as an abiotic factor or a second fungal species. If a second fungal species is also driving selection on the focal plant trait (as in Hypothesis A3), then one or both of those selection gradients would be estimated as nonzero. If a second fungal species is driving selection on a different trait in the plant besides trait x (as in Hypothesis A2), any resulting variation in plant fitness would be captured in the error term (not shown) for Eqn 1. In such a system, with one plant species having two candidate coevolving traits (x1 and x2) interacting with two fungal species each having single candidate coevolving traits (y and z, respectively), an expanded SSA analysis could be used to distinguish among Hypotheses A1–A3, and to elucidate different scenarios involving correlated selection on multiple traits (within Hypotheses A2 and A3) (for details of such an expanded SSA analysis, see the final section of Notes S2).

V. Question 3: what are the relative roles of the three elements of the geographic mosaic theory of coevolution (selection mosaics, hotspots and coldspots, and trait remixing) and nonreciprocal selection from biotic and abiotic factors in driving trait diversification of plants and mycorrhizal fungi?

1. Theoretical framework: the geographic mosaic theory of coevolution

Although local coevolutionary selection is an important building block of coevolutionary dynamics, studies on coevolution in a wide range of species interactions in the last few decades have now demonstrated that coevolution is inherently geographically structured. Consequently, coevolution can drive adaptive differentiation among populations, and coevolution in one population can dominate trait evolution in another. Adaptive differentiation among populations may simply contribute to genetic diversity within species, or may provide the first step towards ecological speciation (Schluter, 2009). Because most species are collections of genetically variable populations distributed among environments that often differ in abiotic factors and/or community composition, the pattern and strength of natural selection imposed by species on each others’ traits are also expected to be highly variable. This recognition that coevolution is an inherently geographically structured process has been synthesized by John Thompson as the geographic mosaic theory of coevolution (GMTC) (Thompson, 2005).

Specifically, the GMTC proposes that coevolution exhibits three inherently geographic characteristics, beyond local coevolutionary selection: coevolutionary hotspots and coldspots, geographic selection mosaics, and trait remixing (Fig. 1a). Coevolutionary hotspots are communities in which the pairwise interaction between two species exhibits coevolutionary selection, that is, there is a G × G interaction effect on the fitnesses of both species, and these hotspots are expected to be embedded in a geographic matrix of coldspots. In coldspots, selection is not reciprocal; coldspots include communities in which only one of the species occurs (i.e., structural coldspots). Geographic selection mosaics occur when the fitness functions of coevolutionary selection differ among environments; that is, there is a genotype by genotype by environment interaction (G × G × E) for the fitness of at least one of the two species. Trait remixing is the suite of processes that potentially influence the geographic distributions of alleles at loci underlying coevolving traits (i.e. the geographic structure of G for both species), including mutations, gene flow, genetic drift, and population extinction/recolonization dynamics.

Figure 1.

 Alternative hypotheses on the relative importance of the three elements of the geographic mosaic theory of coevolution (GMTC) and nonreciprocal selection. Each circle represents a separate local community. Arrows inside circles represent selection among plants (P), mycorrhizal fungi (F), and the abiotic environment (E). Arrows outside circles represent gene flow among communities. (a) Hypothesis B1: all three of the GMTC elements – hotspots/coldspots, selection mosaics, and trait remixing – are important and interact to play prominent roles in driving diversification within and across communities in which two species occur. Reciprocal selection is shown in some communities (hotspots) and not in others (coldspots). The dashed arrow indicates that plant selection on fungi differs, compared with the first community, implying a selection mosaic. The main effect of the abiotic environment is to modify interspecific selection exerted by plants on mycorrhizal fungi. (b) Hypothesis B2: coevolutionary selection is important but is consistent among populations; that is, true geographic selection mosaics (G × G × E interactions) and hotspots/coldspots are relatively unimportant or uncommon. Coevolution can still be geographically variable, if the distribution of genotypes differs among communities, as a result of trait remixing processes. (c) Hypothesis B3: ongoing local coevolutionary selection is consistently less important compared with unidirectional selection exerted on species by abiotic environmental factors, or exerted by one partner species on the other.

Theoretical studies of geographically variable coevolution have shown how the three key elements of the GMTC can interact to generate a wide variety of outcomes of coevolution, including patterns of specificity, local adaptation/maladaptation, and patterns of genetic variation within and between populations (e.g. Gomulkiewicz et al., 2000; Nuismer et al., 2000; Nuismer, 2006). Moreover, empirical studies in a wide range of systems have found support for key elements of the theory, such as the important influence of hot spots and cold spots of reciprocal selection (reviewed by Thompson, 2005), although it remains challenging to rigorously demonstrate key phenomena central to the GMTC, especially coevolutionary selection mosaics (Gomulkiewicz et al., 2007). Furthermore, nonreciprocal adaptation of species to both biotic and abiotic environmental factors has been repeatedly demonstrated to be an important force driving diversification in mycorrhizal fungal and especially plant species (e.g. Bohrer et al., 2003; Saenz-Romero et al., 2006). Thus, we need to answer Question 3: what are the relative roles of the three elements of the GMTC (selection mosaics, hotspots and coldspots, and trait remixing) and nonreciprocal selection from biotic and abiotic factors in driving trait diversification of plants and mycorrhizal fungi among populations?

2. Answering Question 3

Alternative hypotheses on the relative importance of the three GMTC elements and nonreciprocal selection

Hypothesis B1: all three of the GMTC elements – hotspots/coldspots, selection mosaics, and trait remixing – are important and interact to play prominent roles in driving diversification with and across communities in which two species occur (Fig. 1a).  For example, even when coevolutionary selection is only occurring in a subset of communities (hotspots), genotype frequencies and associated trait values in coldspots can be influenced by coevolution in hotspots if gene flow (trait remixing) among communities is significant and selection in hotspots is sufficiently strong (e.g. Gomulkiewicz et al., 2000). If two hotspots differ in the specific trajectory of local coevolutionary selection (a selection mosaic), the relative influence of the two hotspots on overall coevolutionary dynamics in a collection of communities will depend on the relative strengths of selection in the two hotspots. Implicit in this first hypothesis is that coevolutionary selection between species is likely to be more important than nonreciprocal selection from any source (including other species or abiotic factors) in driving ongoing diversification of species because of its potential for rapid and ongoing feedbacks. Thompson (2005) argues that such sustained feedbacks are unlikely for G × E interactions (where E is an abiotic environmental variable), because feedbacks would require rapid changes in abiotic factors (E) in response to the evolution of species.

Hypothesis B2: coevolutionary selection is important but is consistent among populations; that is, true geographic selection mosaics (G × G × E interactions) and hotspots/coldspots are relatively unimportant or uncommon (Fig. 1b).  Under this hypothesis, G × G interaction effects on the fitnesses of two species are similar across all communities in which the two species co-occur, and thus coevolutionary dynamics can be understood by following the dynamics of reciprocal selection along with trait remixing processes. Models of this hypothesis show that it can still generate complex geographic patterns of coevolution, as reciprocal selection can differ geographically among communities as a consequence of the presence of different genotypes, as a result of trait remixing processes (e.g. Gandon & Michalakis, 2002; Gandon & Nuismer, 2009).

Hypothesis B3: ongoing local coevolutionary selection is consistently less important compared with unidirectional selection on species exerted by abiotic environmental factors (Cairney, 2000; Helgason & Fitter, 2009), or exerted by one partner species on the other (Fig. 1c).  Under this hypothesis, hotspots and selection mosaics among hotspots are relatively unimportant. Rather, differing forms of selection among coldspots, whereby different coldspots exhibit different unidirectional selective effects of abiotic factors and species on particular focal species, are predicted to be more important for trait evolution than mosaics of reciprocal selection.

Evidence from tests of G × G interactions among populations and G × E interactions  Among-population cross-inoculation experiments can be used to test for significant main (G) and interactive (G × G) effects of plant and fungal genotypes (representative of populations) on the fitness of either or both species. In these experiments the fitnesses of plants and mycorrhizal fungi are estimated (in a common environment) in reciprocal pairings between plant genotypes from multiple plant populations and fungal genotypes from multiple fungal populations. Significant G × G interactions among populations for fitness components of a focal species indicate that geographically variable selection exerted by the other species may be driving differentiation among populations of the focal species by favoring particular combinations of plant and fungal genotypes. However, the results of such empirical studies cannot necessarily be used to determine the relative importance of key elements of the GMTC, such as hotspots of reciprocal selection and selection mosaics, compared with nonreciprocal selection exerted by one species on another or direct selection from the abiotic environment. Thus, by themselves, such studies cannot be used to distinguish among Hypotheses B1–B3 above.

To illustrate this point, consider one particular pattern of G × G interaction that can be detected in such an experiment: local adaptation, wherein fitness is highest for one or both species in local (sympatric) combinations of plant and fungal populations, compared with nonlocal (allopatric) combinations of plants and fungi from different populations. In an EM interaction between three Pinus species and the fungus Rhizopogon occidentalis, Hoeksema & Thompson (2007) found evidence for a clinal pattern of local adaptation by fungal populations to populations of Pinus host plants, which may indicate that geographically variable coevolutionary selection is driving adaptive divergence among fungal populations. However, such a pattern could be generated with or without true geographic selection mosaics (see Hypothesis B2 above), and virtually any pattern of local adaptation and maladaptation between species can result from geographically variable coevolution, depending on the details of the underlying model (Gomulkiewicz et al., 2007). Moreover, local adaptation of one species to another could occur without any reciprocal selection. For example, trait remixing processes or nonreciprocal adaptation to abiotic factors could generate genetic structure in a focal species, followed by local adaptation of a second species to populations of the focal species (see Hypothesis B3 above). The latter scenario is consistent with the results from the study by Hoeksema & Thompson (2007) mentioned above. Although they found evidence for a clinal pattern of local adaptation by fungal populations to populations of Pinus host plants, Pinus host plants exhibited no G × G interaction for fitness across fungal populations, suggesting that pines may not be adapted to local populations of the fungus.

One difficulty in interpreting the results of such geographic G × G experiments is that we do not know how the results are influenced by the particular common environment chosen for the experiment, or the degree to which that environment is representative of any of the native environments of the genotypes included in the experiment. Similarly, tests for G × E interactions, in which the ecological outcome of mycorrhizal interactions is tested across multiple genotypes of either the plant or the fungus in multiple abiotic environments, provide information on selection exerted by the abiotic environment on either the plant or the fungus but are not informative regarding the existence of G × G interactions within populations or how G × E interactions are influenced by community context. As far as I am aware, all published tests of plant and mycorrhizal fungal adaptation to heavy metal-contaminated soils fall into this latter category (reviewed by Meharg & Cairney, 2000). These experiments have shown that plant and mycorrhizal fungal populations can adapt rapidly in response to direct abiotic selection from heavy metal-contaminated soils, but have not yet shown that such adaptive processes are stronger than local adaptation between genotypes of plants and mycorrhizal fungi or whether coevolution between plants and mycorrhizal fungi might differ between contaminated and noncontaminated soils; that is, whether these soils might drive a true selection mosaic in coevolving mycorrhizal interactions. Thus, those experiments do not yet allow us to assess the relative importance of direct abiotic selection (G × E interactions), coevolutionary selection (G × G interactions), and selection mosaics (G × G × E interactions) in driving adaptive differentiation of plants and fungi among environments and do not distinguish among Hypotheses B1–B3 above. As Taylor (2000) states, the experimental system of mycorrhizal fungal interactions on and off heavy metal-contaminated soils may indeed be particularly useful for testing the predictions of the GMTC, but the critical experimental tests have not yet been performed.

Evidence from tests of G × G × E interactions, reciprocal transplants, and selective source analysis  In order to determine the relative importance of key elements of the GMTC, such as hotspots of reciprocal selection and selection mosaics, compared with nonreciprocal selection exerted by one species on another or direct selection from the abiotic environment, we need experiments that explicitly test for G × G × E interactions for the fitness of interacting species. Such tests involve testing G × G interactions for the fitness of interacting species, across multiple environments (E). Nuismer & Gandon (2008) explain how such experiments allow adaptive divergence among populations to be partitioned into its three components: G × G, G × E, and G × G × E interactions. In such experiments, finding significant G × G × E interactions for the fitness of a focal species suggests the existence of a selection mosaic (sensuThompson, 2005); that is, the specific nature of coevolutionary selection (G × G) on that species differs among environments. Finding that G × G interactions are significant and identical in form, regardless of environment, would suggest that selection mosaics and hostpots/coldspots per se are unimportant; that is, supporting Hypothesis B2 above. Significant G × E interactions in such experiments, in the absence of G × G or G × G × E interactions, suggest the importance of direct abiotic selection on plant and fungal traits, compared with coevolutionary selection.

In one version of a G × G × E experiment, Johnson et al. (2010) estimated plant and fungal performance in all reciprocal combinations of plant populations (G), whole AM fungal guilds of species (G), and sterile soils (E) for the interaction between the grass A. gerardii and its AM fungi from three populations, in a glasshouse experiment. As far as I am aware, this innovative experiment represents the most complete G × G × E experiment conducted to date for any interspecific interaction and demonstrates the potential utility of mycorrhizal interactions for testing general coevolutionary theory (also see Bohrer et al., 2003). They found a G × G × E interaction for arbuscule formation, such that it was highest in local combinations of plants, fungi, and soils, suggesting that a selection mosaic (different G × G interactions for different E) has resulted in local adaptation of plants and fungi to each other (supporting Hypothesis B1). This result builds on the conclusions of Schultz et al. (2001), who found indirect evidence in the same system for differential selection by the plant–AM interaction on mycorrhizal dependence of A. gerardii in high- vs low-fertility soils. However, Johnson et al. also found that one component of fungal fitness – the mass of extraradical hyphae produced by the AM fungi – depended only on a G × E interaction, wherein extraradical hyphal mass was greatest for local combinations of fungi and soils regardless of plant genotype. This result supports Hypothesis B3 by suggesting the potential importance of direct selection exerted by the soil environment on fungi, regardless of fungal genotype, although the fungal trait(s) that are the target for this selection are unknown. To integrate those two results, we need a better understanding of the relationships among fungal fitness, arbuscule formation, and extraradical hyphal biomass for AM fungi.

Piculell et al. (2008) explored the potential for selection mosaics in plant–EM fungal interactions by estimating plant and fungal performance in reciprocal combinations of plant genotypes and fungal genotypes, across two different artificial environmental gradients – soil types (laboratory or field), and the presence or absence of nonmycorrhizal microbes – for the interaction between bishop pine (Pinus muricata) and the fungus R. occidentalis. Although they did not find any statistically significant G × G × E interaction effects on the performance of the plant or the fungus, their results suggest the potential for selection mosaics in this plant–EM interaction. For example, fungal colonization intensity, a potential component of fungal fitness, depended on an interaction between plant genotype and soil type. Similarly, plant root:shoot allocation, a potential component of plant fitness, depended on an interaction between fungal genotype and soil type. Additional experiments are badly needed in which plant and fungal fitness are estimated for the same combinations of plant and fungal genotypes, across multiple environments, especially realistic environmental gradients. Tests for differing G × G interactions across gradients in soil fertility, temperature, moisture, and atmospheric CO2 availability would be particularly useful for testing how mycorrhizal interactions may coevolve differently in response to anthropogenic disturbances.

Ideally, G × G × E experiments can be conducted in multiple common gardens in the field (Nuismer & Gandon, 2008), with the same genotypic combinations of interacting species replicated across all environments. In such an experiment, ‘E’ is representative of the entirety of the selective environment, apart from the contribution of the genotypes of the two species. In glasshouse experiments testing for G × G × E effects (Piculell et al., 2008; Johnson et al., 2010), important environmental factors may be absent, preventing a direct assessment of the relative importance of coevolutionary selection compared with selection from other environmental factors. If values for the traits of interacting species can be measured in the experiment, then SSA could be used within each common garden, to partition sources of selection explicitly among traits of interacting species and specific aspects of the environment hypothesized to be important selective sources. Comparing results of SSA among sites in reciprocal transplant common gardens could allow assessment of the relative importance of hotspots/coldspots, selection mosaics, one-way selection between species, and direct selection from abiotic environmental factors (Gomulkiewicz et al., 2007). Gomulkiewicz et al. (2007) caution, however, that confirming the absence of G × G interactions (or G × G × E interactions) may be difficult in such experiments, because a lack of a statistically significant interaction may be a result of insufficient statistical power.

VI. Conclusions

Although many of the necessary empirical studies have not yet been conducted, the few existing empirical studies are consistent with the possibility that ongoing coevolution plays a fundamental role in driving trait diversification within and between populations of plants and mycorrhizal fungi. Existing evidence suggests that higher level traits, including mycorrhizal colonization intensity and plant responsiveness to mycorrhizal colonization, exhibit substantial heritable genetic variation within plant and fungal species, and thus have the potential to respond to coevolutionary selection. More studies are needed, however, of genetic variation for lower level traits controlling the formation of the mycorrhizal phenotypic interface and for higher level traits such as specificity and community composition, as well as genetic correlations among multiple traits. Ecological studies suggest that the exact form of local coevolutionary selection in mycorrhizal interactions is likely to be influenced by a mixture of beneficial and parasitic ecological outcomes and by asymmetries in benefit between different combinations of plants and fungi.

Although a few empirical studies show the potential for local coevolutionary selection in mycorrhizal interactions, SSA could be employed to test explicitly for ongoing coevolutionary selection. SSA could also be used to distinguish among three alternative hypotheses presented here on the influence of community context on local coevolutionary selection. All three hypotheses show how species diversity in mycorrhizal interactions does not necessarily diminish the importance of coevolution for trait evolution in those interactions. Recent geographically structured empirical studies suggest the potential for coevolution to drive adaptive differentiation among populations of plants and fungi, but additional studies, especially using SSA in the context of field reciprocal transplant experiments, are needed to distinguish among three alternative hypotheses presented here on the relative importance of geographically variable coevolutionary selection, geographically uniform coevolutionary selection, and nonreciprocal selection on traits of mycorrhizal plants and fungi.

Acknowledgements

Tiffany Bensen, Anjel Craig, Justine Karst, Bridget Piculell, John Thompson, Sumi Weerasooriya, and three anonymous reviewers provided helpful comments on an earlier draft of the manuscript.

Ancillary