Complex host-pathogen coevolution in the Apterostigma fungus-growing ant-microbe symbiosis

Background The fungus-growing ant-microbe symbiosis consists of coevolving microbial mutualists and pathogens. The diverse fungal lineages that these ants cultivate are attacked by parasitic microfungi of the genus Escovopsis. Previous molecular analyses have demonstrated strong phylogenetic congruence between the ants, the ants-cultivated fungi and the garden pathogen Escovopsis at ancient phylogenetic levels, suggesting coevolution of these symbionts. However, few studies have explored cophylogenetic patterns between these symbionts at the recent phylogenetic levels necessary to address whether these parasites are occasionally switching to novel hosts or whether they are diversifying with their hosts as a consequence of long-term host fidelity. Results Here, a more extensive phylogenetic analysis of Escovopsis lineages infecting the gardens of Apterostigma ants demonstrates that these pathogens display patterns of phylogenetic congruence with their fungal hosts. Particular clades of Escovopsis track particular clades of cultivated fungi, and closely-related Escovopsis generally infect closely-related hosts. Discordance between host and parasite phylogenies, however, provides the first evidence for occasional host-switches or acquisitions of novel infections from the environment. Conclusion The fungus-growing ant-microbe association has a complex coevolutionary history. Though there is clear evidence of host-specificity on the part of diverse Escovopsis lineages, these pathogens have switched occasionally to novel host fungi. Such switching is likely to have profound effects on how these host and parasites adapt to one another over evolutionary time scales and may impact how disease spreads over ecological time scales.


Background
Most parasites are intimately dependent on one or a few hosts. Because of this host fidelity, parasites are expected to track speciating hosts by speciating themselves. This process, known as cospeciation, will lead to cocladogenesis, the topological matching of symbiont phylogenies.
Parasite and host phylogenies are rarely identical, however; forces such as duplication (parasite speciation in the absence of host speciation), sorting events (host speciation without commensurate parasite speciation), and host-switching (parasites begin to use a new host) [1,2] can generate discordance between the phylogenies of hosts and their symbionts. Despite these complications, congruent phylogenies are known in host-parasite associations [3][4][5] and in host-mutualist associations as well [6][7][8].
The fungus-growing ant-microbe symbiosis is a novel example of a system in which cocladogenesis occurs between a host and both its mutualistic and parasitic symbionts. Research over the last decade has demonstrated the congruence of the phylogenies of fungus-growing ants, the fungi that they cultivate (i.e., their fungal cultivars) and the cultivar-attacking pathogen Escovopsis at ancient phylogenetic levels [9][10][11]. Genetic analyses of more recently diverged, younger lineages demonstrate discrepancies between ant and cultivar associations, which are likely due to a combination of lateral transfer of cultivars between colonies and occasional domestication of free-living fungi by the ants [12][13][14][15]. To date, the two published phylogenetic studies of the Escovopsis-cultivar association indicate no discordance between the phylogenies of the cultivars and Escovopsis [10,16].
Ancient codiversification of fungus-growing ants and their cultivars is driven by the intimate dependence of the ants on fungus as their primary food source and the intimate dependence of the fungus on ants for protection, nutrition and dispersal [17]. In ants, the ability to cultivate fungi for food arose only once, about 50-60 million years ago, and gave rise to roughly 200 described, extant species of fungus-growing ants (Tribe Attini) [18]. The long coevolutionary history of these mutualists has led to the specialization of each ant species on the cultivation of a unique, narrow range of cultivated fungi, most of which are in the family Lepiotaceae. As depicted in Figure 1A, these lepiotaceous cultivars form two morphologically and molecularly distinct groups ('G1' and 'G3'; [9]). There has been one switch to a distantly related cultivar; most ants in the genus Apterostigma now cultivate fungi in the family Pterulaceae [19], which is distantly related to the family Lepiotaceae. The pterulaceous cultivars fall into two monophyletic, morphologically distinct cultivar groups ('G2' and 'G4' in Figure 1A; [11]). One Apterostigma species, A. auriculatum, has retained the ancestral state of growing lepiotaceaous cultivars [11].
Currie et al. [10] demonstrated that, at ancient levels, the phylogeny of Escovopsis (Ascomycota: Hypocreales) (Figure 1B), a genus of specialized, highly pathogenic microfungi that attack the ants' fungal cultivars, matches that of the ants' diverse cultivars and consequently that of the ants themselves. Escovopsis has only been found associated with nests of attine ants. Upon establishing infection, Escovopsis consumes the ants' cultivated fungi and can devastate attine colonies [20][21][22]. Though infection rates vary across host species, infections are prevalent in colonies of many attine genera throughout their geographic ranges [16,20,21]. Escovopsis is thought to track the cultivars because of the coevolutionary specialization of each Escovopsis lineage on attacking and overcoming defenses of only a narrow range of cultivar hosts [16,23].
Ancient phylogenetic congruence between cultivars and Escovopsis suggests that these pathogens may be tightly tracking their speciating hosts by speciating themselves, and that Escovopsis lineages have not switched to novel cultivar hosts over evolutionary time. To test for hostswitching, however, it is necessary to include extensive sampling across the diversity of both hosts and symbionts. Previous studies of Escovopsis host-fidelity have included few samples of Apterostigma-associated Escovopsis despite the fact that they are an extremely diverse group of fungus-growing ant pathogens. Currie et al. [10], the most extensive phylogenetic analysis of Escovopsis to date, included only two Apterostigma-associated Escovopsis, which were morphologically similar and were isolated from ant colonies that raised closely-related fungi. Not surprisingly, these isolates formed a single monophyletic "Apterostigma Escovopsis" clade ( Figure 1B). However, unlike the other fungus-growing ant genera, which each raise cultivars in a single cultivar group, Apterostigma ants raise cultivars in three groups (G2, G3 and G4 in Figure  1A), which are each attacked by morphologically distinct Escovopsis types [23]. More extensive sampling of these diverse Apterostigma pathogens, therefore, can reveal the extent to which Escovopsis species are host-faithful, tracking their particular hosts without host-switching.
Through extensive geographic sampling and phylogenetic analysis of Apterostigma-associated Escovopsis, we ask whether host and pathogen phylogenies are still congruent when genetic analyses are extended to include the diversity of the Apterostigma-associated Escovopsis. First, do Apterostigma-associated Escovopsis form a monophyletic clade as the Apterostigma ants do, or are Apterostigma-associated Escovopsis polyphyletic like their cultivars? Second, do the Apterostigma Escovopsis form three distinct clades that correspond to the three cultivar groups (G2, G3 and G4) raised by the different species of Apterostigma ants? Based on earlier findings that Escovopsis is highly cultivartype specific [16], we hypothesize that more extensive sampling will reveal that the Apterostigma-associated Escovopsis are not monophyletic like their associated anthosts, because the Escovopsis that infects lepiotaceous Apterostigma cultivars (i.e. the cultivars raised by A. auriculatum) will be more closely-related to Escovopsis isolated from lepiotaceous gardens of non-Apterostigma ant species than to Escovopsis isolated from pterulaceous Apterostigma gardens ( Figure 1C). This would support findings of Currie et al. [10], depicted in Figure 1B, that the pterulaceousattacking Escovopsis form a monophyletic clade distinct from the Escovopsis that infects lepiotaceous cultivars, but would contradict their findings of complete congruence between ant, cultivar, and Escovopsis phylogenies. We further hypothesize that more extensive sampling will reveal that pterulaceous-attacking Escovopsis will fall into two clades associated with the two pterulaceous cultivar groups (G2 and G4). Overall, in looking at the fungusgrowing ant-microbe symbiosis as a whole, we predict four host-specific Escovopsis clades that are each specialized at attacking one of the four known fungus-growing ant cultivar groups (G1, G2, G3 and G4) ( Figure 1C).

Diversity of Apterostigma-associated Escovopsis
Of 623 colonies from which microbes were sampled, at least one fungal symbiont (either cultivar or Escovopsis) was isolated from each of 410 colonies. For the purpose of this study, based on field identification of the ants, garden architecture and growth form of cultivar isolates, each col-ony was classified as either a G2, G3 or G4 colony, which raise respectively G2, G3 and G4 cultivars (see introduction).
Escovopsis infection of these 410 colonies was common and pathogen phenotypes were diverse. G2 colonies and G4 colonies had much higher infection rates than G3 colonies (G-test with Yate's Correction: G2 vs. G3, G = 36.0, df = 1, p < 0.0001; G4 vs. G3, G = 6.3, df = 1, p < 0.0001; G2 vs G4, G = 0.1, df = 1, p = 0.8). More than 50% of G2 and G4 colonies were infected with at least one Escovopsis type, whereas only 11% of G3 colonies were infected (Table 1). Escovopsis samples isolated from infected colonies were classified into four morphotypes based on spore-color: brown, yellow, white and pink. These types have different micromorphological conidiophore structures (Currie, unpublished) and likely represent different Escovopsis lineages. In the absence of proper species descriptions, we will refer to the different Escovopsis line-Current symbiont phylogenies and hypothesized Escovopsis relationships Figure 1 Current symbiont phylogenies and hypothesized Escovopsis relationships. (a) Cultivar phylogeny simplified from [9][10][11] (b) Escovopsis phylogeny from [10]. This phylogenetic reconstruction, the most complete to date, includes very few Apterostigma-associated pathogens. (c) Hypothesized Escovopsis phylogeny in which there are four distinct Escovopsis clades corresponding to the four known cultivar clades. In this hypothesized phylogeny, the Pterulaceae-attacking pathogens are distinct from the Lepiotiaceae-attacking pathogens. * indicates Apterostigma-associated symbionts. Note that the only Apterostigmaassociated symbionts outside the G2/G4 clade are those isolated from colonies of A. auriculatum, the only Apterostigma sp. that does not cultivate pterulaceous fungi [11,13]. See introduction for further details. ages by their characteristic spore-color (white, pink, yellow, brown). While white and pink Escovopsis were each specific to a single cultivar group, brown Escovopsis infected both G2 cultivars and G4 cultivars, and yellow Escovopsis infected both G2 cultivars and G3 cultivars. The yellow Escovopsis isolates associated with these two clades, however, are micromorphologically distinct from one another (Currie, unpublished) and likely are two separate species. A small percentage of colonies (10% of G2 colonies) were infected by multiple Escovopsis morphotypes ( Table 1).

Phylogenetic relationships of Apterostigma-associated Escovopsis
The results of parsimony, likelihood and Bayesian analyses were highly concordant. Three well supported clades were identified that correspond to brown, white and pink Escovopsis ( Figure 2). Yellow Escovopsis is not monophyletic; G2-attacking yellow Escovopsis is genetically distinct from the single isolate of G3-attacking yellow Escovopsis. Overall, as predicted, these diverse Apterostigma-associated Escovopsis do not form a monophyletic clade. Both the yellow and pink Escovopsis isolated from Apterostigma colonies with G3 cultivars are nested within other G3-attacking Escovopsis and are distinct from other Apterostigma-associated Escovopsis. Contradictory to our predictions, G2-attacking and G4-attacking Escovopsis do not form separate, monophyletic clades. Within the brown Escovopsis, there are two clades of G4-associated Escovopsis.
Parametric-bootstrapping verified the polyphyly of isolates of G4-associated Escovopsis. The null hypothesis of a single origin of G4-associated Escovopsis was rejected at p < 0.001. This implies that brown Escovopsis has switched multiple times between G2 and G4 hosts.

Discussion
Phylogenetic patterns of the fungus-growing ant microbe symbiosis reveal a coevolutionary history of host-fidelity punctuated by occasional host-shifts. All known Escovopsis lineages have some limitation to their host-range. For example, we here show that pink Escovopsis attacks only lepiotaceous G3 cultivars (including A. auriculatum's cultivars), white Escovopsis attacks only G2 cultivars, and though Escovopsis with yellow spores attacks both G2 and G3 cultivars, the yellow Escovopsis lineages associated with each of these host groups are morphologically and genetically distinct (Figure 2). Despite this specificity, however, there is not complete congruence of host and pathogen phylogenies as suggested by previous studies [10], indicating that Escovopsis host ranges have shifted and may continue to shift (Figure 3). This complex history parallels that of other symbiotic associations in which extensive sampling reveals that codiversification is interrupted often by host-switches [13,[24][25][26]. In fact, it appears that the cases where cocladogenesis persists over evolutionary time are mostly vertically-transmitted endosymbionts [27][28][29][30], whereas most ectosymbionts, such as Escovopsis and the fungal cultivar, show patterns of switching and absence of strict cocladogenesis with their hosts.
Adaptive processes may explain the host-fidelity of most Escovopsis types, which leads to the host-specific Escovopsis clades revealed here. Gerardo et al. [23] demonstrated through microbial bioassays that Escovopsis lineages are attracted to chemical signals released by their host cultivars. For example, in microbial bioassays, isolates of yellow Escovopsis from G2-Apterostigma colonies grow more rapidly towards chemical signals produced by G2 than by G4 and G3 cultivars, which is consistent with the hostrange of yellow Escovopsis. Unlike yellow Escovopsis, brown Escovopsis from G2-Apterostigma colonies is equally attracted to G2 and G4 cultivars. It is possible that this host-attraction would make it easier for brown Escovopsis to switch between G2 and G4 hosts than it would be for yellow Escovopsis to switch between hosts, because brown Escovopsis would be equally likely to move through G2 and G4 fungal gardens, find healthy cultivar and establish infection. This is consistent with the phylogenetic results here, where is seems that brown Escovopsis has switched between G2 and G4 hosts.
Microbial bioassays have also revealed that cultivars can defend themselves against some Escovopsis but not others [23]. G3-Apterostigma cultivars can inhibit isolates of both brown Escovopsis and G2-associated yellow Escovopsis but cannot inhibit isolates of pink Escovopsis, possibly explaining why brown and G2-associated yellow Escovopsis do not attack G3 cultivars in nature, while pink Escovopsis does. Brown and yellow Escovopsis are not, however, inhibited by most isolates of G2 and G4 cultivars, explaining why natural infection is possible in these host-parasite combinations. Overall, both cultivar defense against Escovopsis and Escovopsis' attraction to host cultivars may maintain Escovopsis' specialization and prevent rampant hostswitching.
Because Escovopsis species are host-specific, we hypothesized that wider sampling of Escovopsis would reveal four Escovopsis clades that correspond to four cultivar and ant clades ( Figure 1). However, contradictory to our hypothesis (Figure 1c), G4-associated Escovopsis are not monophyletic (Figure 3b). Furthermore, there is a lack of congruence of cultivar and Escovopsis phylogenies at deeper nodes ( Figure 3). Whereas previous analyses [9][10][11] have indicated that Apterostigma ants, their pterulaceous cultivars and their associated Escovopsis are distantly-related to the highly-derived leafcutter ants and their associated microbes (including G1 cultivars), some Apterostigma-associated Escovopsis lineages, namely brown Escovopsis, are sister to the Escovopsis attacking G1 cultivars ( fig 3b). This suggests that an Escovopsis lineage switched between these two distantly-related, ecologically-distinct fungal hosts (i.e. Apterostigma colonies and leafcutter colonies).
Discordance of host and pathogen phylogenies suggests that Escovopsis lineages have switched hosts over the evolutionary history of their host association, but the available evidence does not allow inference regarding the frequency at which switching occurs. It is also unclear whether switching involves the acquisition of novel Escovopsis strains by the ants from their environment, or whether it involves the direct transmission of Escovopsis between colonies by some unknown mechanism. Further research on the exact mechanism of Escovopsis transmission would be helpful in revealing the likelihood of pathogen exchange between colonies containing distantlyrelated cultivars.

Conclusion
Phylogenetic analyses coupled with extensive sampling of host and parasites reveal a more complete picture of the complexity of the Escovopsis-cultivar association in colonies of fungus-growing ants, which consists of specialized pathogen species that occasionally switch between distantly-related hosts. Clades of closely-related Escovopsis attack specific cultivar groups, causing the matching of cultivar and Escovopsis phylogenies at some scales. Discordance of host-parasite phylogenies, however, arises due to host-switching ( Figure 3). These results reveal the need for additional sampling across the fungus-growing ant microbial symbiosis as a whole. To date, there has not been extensive sampling and analysis of the pathogens that attack the diverse G3 cultivars grown by many fungus-growing ant species [13]. There are also few published genetic analyses of the cultivars of the leafcutter ants, agricultural pests in much of the Neotropics, and the leafcutter-associated Escovopsis. Broad sampling and genetic analyses across the symbiosis will give insight into how labile these associations are over both ecological and evolutionary time.

Collections and infection prevalence
From 2001-2004, we sampled 632 Apterostigma colonies collected across their geographic range in order to isolate fungal symbionts (cultivar and Escovopsis). All fungi were cultured following procedures of [16]. Escovopsis samples from Panama, Costa Rica, and Argentina were maintained as live cultures on potato dextrose agar with 50 mg/L each of penicillin and streptomycin until spores and mycelium could be directly frozen at -80 degrees. Fungal samples from Ecuador were only temporally maintained live after collection and were then stored in 95% alcohol prior to export from the country. DNA extraction of frozen samples followed a CTAB extraction protocol modified from [31].
Infection prevalence in the three colony-types (G2, G3 and G4) was determined by dividing the number of colonies infected with Escovopsis by the total number of colonies from which either Escovopsis or cultivar was successfully isolated (colonies from which no microbes were isolated were excluded from these analyses). We then used log-likelihood ratio tests (a.k.a. G-tests) to compare rates of infection across colony-types. These tests were performed in R (ver 2.3.1, [32]) using the function g.test.r [33] with the William's correction applied.

Samples for phylogenetic reconstruction
To determine the relationship amongst Escovopsis strains isolated from Apterostigma spp. colonies, samples for phylogenetic reconstruction were selected to include all Escovopsis morphotypes isolated from Apterostigma spp. colonies. Because colonies with G2 cultivars are commonly found and are frequently infected with Escovopsis, we sequenced more Escovopsis strains from G2 (n = 44) than from G3 (n = 5) or G4 (n = 4) colonies. We also Escovopsis types are identified here according to spore-color. 1 For the G2-cultivar colonies, the number of infections based on spore-color exceeds the number of total infections because many colonies were infected with two or three Escovopsis types. 2 Though sharing a similar spore-color, the G2-attacking and G3-attacking yellow Escovopsis have distinct micromorphologies (Currie, unpublished).
Escovopsis phylogeny based on EF-1 alpha sequencedata Figure 2 Escovopsis phylogeny based on EF-1 alpha sequencedata. Each branch is labeled with likelihood bootstrap values (above), Bayesian posterior probabilities (below, left) and parsimony bootstrap values (below, right). Unlabeled branches have values of less than 50 for at least two analyses of support. * indicates that all three support values are 95 or greater. Each Escovopsis node is labeled with a sample code, the name of the associated ant species, and the country of origin (AR, Argentina; CR, Costa Rica; GU; Guyana; EC, Ecuador; PA, Panama). Labels for isolates from Apterostigma colonies are in bold. # emphasizes the G4-associated Escovopsis, which are not monophyletic. ^ marks the two Apterostigma Escovopsis isolates included in a previously published analysis [10]. Bars along the right side indicate the spore-color of the sample and the host-cultivar clade. Not all outgroups are shown for clarity.  Figure 2.  Sequences were assembled in SeqMan II (ver 5.05, DNAS-TAR), aligned using Clustal W WWW [34]and edited manually in MacClade (ver 4.06, [35]).

Phylogenetic analyses and hypothesis testing
Parsimony analyses were performed in PAUP* (ver 4.0b10, [36]) using heuristic searches with TBR branch swapping and 10,000 random addition sequence replicates (multrees = yes). In order to obtain estimates of clade support, non-parametric bootstrapping was performed with heuristic searches of 5000 replicate datasets and 10 random addition sequence replicates per dataset (multrees = no).
For maximum likelihood and Bayesian analyses, a model of sequence evolution was estimated for the data set using MODELTEST ver. 3.7 [37]. The chosen model, K81uf + pinvar + Γ, was used for all maximum likelihood analyses and parametric hypothesis testing. Because it is not possible to implement this model in Mr. Bayes, a more complex model of sequence evolution, GTR + pinvar + Γ, was used in all Bayesian analyses.
For maximum likelihood analysis, we performed a successive approximation search using PAUP* to estimate the topology [38]. Starting parameter values estimated from a parsimony tree (TBR branch swapping, 100 random addition sequence replicates, multrees = no) were used in an initial maximum-likelihood search. Parameters were then re-estimated from the resulting tree and the search was repeated with these new parameters. This procedure was repeated until the resulting tree was identical in topology to that from the previous iteration. Non-parametric bootstrapping was performed with heuristic searches of 1000 replicate datasets starting from a neighbor-joining tree (multrees = yes).
For Bayesian analyses, using Mr. Bayes (ver 3.0b4, [39]), four separate Markov Chain Monte Carlo (MCMC) runs were performed starting from random trees for each of four simultaneous chains. Runs were five million generations with a burn-in of 100,000 generations, default prior distribution for model parameters, and the differential heating parameter set to 0.2. The joint posterior probabilities and parameter estimates of each run were congruent, suggesting the chains were run for a sufficient number of generations to adequately sample the posterior probability landscape.
Phylogenetic analysis with no topological constraints indicated two origins of G4-associated Escovopsis ( Figure  2). To test the hypothesis of monophyly of Escovopsis isolated from G4 colonies, we compared the observed, optimal tree (alternative hypothesis) to trees constrained to represent the null hypothesis of a single origin of G4 Escovopsis. Sequence evolution parameters were estimated by using maximum likelihood under the K81uf + pinvar + Γ. We used parametric bootstrapping procedures to evaluate 500 simulated datasets generated using Seq-Gen (ver 1.2.5, [40]).