Evolution, 58(10), 2004, pp. 2252-2265 EVOLUTION OF ANT-CULTIVAR SPECIALIZATION AND CULTIVAR SWITCHING IN APTEROSTIGMA FUNGUS-GROWING ANTS PALLE VILLESEN,^'^ ULRICH G. MUELLER,2.3.4 XED R. SCHULTZ,^ RACHELLE M. M. ADAMS,^ AND AMY C. BOUCK^ ^Department of Ecology and Genetics, University of Aarhus, DK-8000 Aarhus C, Denmark E-mail: palle@birc.dk ^Section of Integrative Biology, Patterson Laboratories, University of Texas, Austin, Texas 78712 ^E-mail: umueller@mail.utexas.edu "^Smithsonian Tropical Research Institute, Apartado 2072, Balboa, Republic of Panama ^Department of Systematic Biology, MRC 188, National Museum of Natural History, Smithsonian Institution, Washington, District of Columbia 20560-0188 E-mail: schultz@lms.si.edu ^Department of Genetics, University of Georgia, Athens, Georgia 30602 Abstract.?Almost all of the more than 200 species of fungus-growing ants (Formicidae: Attini) cultivate litter- decomposing fungi in the family Lepiotaceae (Basidiomycota: Agaricales). The single exception to this rule is a subgroup of ant species within the lower attine genus Apterostigma, which cultivate pterulaceous fungi distantly related to the Lepiotaceae. Comparison of cultivar and ant phylogenies suggests that a switch from lepiotaceous to pterulaceous fungiculture occurred only once in the history of the fungus-growing ants. This unique switch occurred after the origin of the genus Apterostigma, such that the basal Apterostigma lineages retained the ancestral attine condition of lepi- otaceous fungiculture, and none o? the Apterostigma lineages in the monophyletic group of pterulaceous fungiculturists are known to have reverted back to lepiotaceous fungiculture. The origin of pterulaceous fungiculture in attine ants may have involved a unique transition from the ancestral cultivation of litter-decomposing lepiotaceous fungi to the cultivation of wood-decomposing pterulaceous fungi. Phylogenetic analyses further indicate that distantly related Apterostigma ant species sometimes cultivate the same cultivar lineage, indicating evolutionarily frequent, and possibly ongoing, exchanges of fungal cultivars between Apterostigma ant species. The pterulaceous cultivars form two sister clades, and different Apterostigma ant lineages are invariably associated with, and thus specialized on, only one of the two cultivar clades. However, within clades Apterostigma ant species are able to switch between fungi. This pattern of broad specialization by attine ants on defined cultivar clades, coupled with flexible switching between fungi within cultivar clades, is also found in other attine lineages and appears to be a general phenomenon of fungicultural evolution in all fungus-growing ants. Key words.?Apterostigma, Attini, fungiculture, fungus-growing ants, Pterulaceae. Received April 3, 2003. Accepted July 5, 2004. Fungus-growing ants (Formicidae, tribe Attini) are well In contrast, the lower attine ants (eight genera, including the known for their habit of cultivating fungi for food, a behavior genus Apterostigma) cultivate morphologically unspecialized that originated in the common ancestor of attine ants about fungi (called G3 fungi) from which the Gl fungi arose (S.A. 50-60 million years ago (Wilson 1971; Schultz and Meier Rehner, pers. comm.; P. Abbott, N. Gerardo, and U. Mueller, 1995; Mueller et al. 2001). Since then, attine ants have un- unpubl. data), and which are therefore evolutionarily ple- dergone an adaptive radiation that has resulted in 13 extant siomorphic with respect to the G1 group (Chapela et al. 1994; genera containing more than 210 described species, all ob- Mueller et al. 1998). There is only one known exception to ligately dependent on the farming of mutualistic fungi, usu- this lepiotaceous cultivation of Gl and G3 fungi: some but ally in nests excavated in the soil (Schultz and Meier 1995; not all ant species in the genus Apterostigma cultivate fungi Weiterer et al. 1998; Brand?o and Mayh?-Nunes 2001). Do- that are distantly related to the Lepiotaceae (Fig. 1). These mestication of fungi from free-living populations, as well as so-called G2 fungi are characterized by a unique micromor- cultivar exchanges between ant species in the same and in phology (Chapela et al. 1994) and are cultivated by their different genera, have occurred multiple times during the Apterostigma hosts in hanging "veiled gardens" in which evolutionary history of the attine ants (Chapela et al. 1994; the cultivated fungus also forms a thin mycelial envelope that Mueller et al. 1998; Mueller 2002; Mueller and Gerardo surrounds the garden proper. G2 fungi were grouped into the 2002). family Tricholomataceae (Basidiomycota: Agaricales) in two Distinct modes of fungal cultivation characterize different previous phylogenetic analyses (Chapela et al. 1994; Mon- attine ant lineages. The great majority of species cultivate calvo et al. 2000), but recent work suggests a close affinity lepiotaceous fungi (Basidiomycota: Agaricales: Lepiotaceae) with the family Pterulaceae (Basidiomycota: Agaricales) from one of two distinct groups, termed Gl and G3 fungi by (Munkacsi et al. 2004). Chapela et al. (1994). The fungi of the higher attine ants This study elucidates the origin and evolution of the cul- (calledGl fungi, zcultivated by the ant genera Sencomyrmex, tivation of pterulaceous fungi by ants in the genus Apter- Trachymyrmex, and the leafcutter ants Atta and Acromyrmex) ostigma. Phylogenetic analyses of both fungi and ants support possess several derived morphological features that probably a single origin of pterulaceous fungal cultivation. Our anal- arose during a long history of coevolution with attine ants, y sis divides the pterulaceous cultivars into two distinct 2252 ? 2004 The Society for the Study of Evolution. All rights reserved. FUNGICULTURE EVOLUTION IN APTEROSTIGMA ANTS 2253 Boletus recipes bolete mushroom Amanita muscaria fly agaric Pterulaceous cultivars G2 - Apterostigma cultivars G4 - Apterostigma cultivars Termitomyces microcarpus (?ingal symbiont of termites) Pleurotus ostreatus oyster mushroom Pholiota squarrosa shaggy scalycap Agaricus arvensis horse mushroom Macrolepiota procera parasol mushroom Lepiotaceous cultivars G3 - lower attine cultivars Gl- higher attine cultivars 0.1 FIG. 1. Relationship between fungi cultivated by fungus-growing ants. ME tree showing the distant phylogenetic position of the fungi cultivated by some Apterostigma ants (G2 and G4) relative to the lepiotaceous fungi cultivated by most attine ants (Gl and G3). For comparison, some well-known fungi are indicated. The tree was constructed using 899 fungal sequences from this study and Moncalvo et al. (2002). Sequences were aligned using ClustalW (Thompson et al. 1994), poorly aligned regions were discarded using GBLOCKS (Castresana 2000), and the phylogeny was constructed using FastME (Desper and Gascuel 2002). Figure is provided as an illustration of the position of attine-ant fungi and conveys substantially the same information as figure 1 in Munkacsi et al. (2004). 2254 PALLE VILLESEN ET AL. FIG. 2. Apterostigma gardens. (A) Garden of the G2-cultivator Apterostigma collare hanging in a sheltering crevice formed by tree branches. A mycelial veil surrounds the G2 garden proper. The single nest entrance is visible as a hole in the veil slightly to the lower right of the garden's center. (B) Garden of the G3-cultivator Apterostigma auriculatum in a cavity underneath a log on the forest floor. The garden is sessile and is not surrounded by a veil. G4 gardens (not shown) are also unveiled and sessile, and thus resemble the G3 nest architecture shown on the right. clades, the veiled fungi termed the G2 group and a novel unveiled group termed G4. Phylogenetic analyses further re- veal a faithfulness of ant lineages to cultivate fungi from only one of the two major clades of pterulaceous cultivars (G2 or G4). Within each of these ant clades, however, single Ap- terostigma species are associated with a wide diversity of cultivars, resulting from evolutionarily recent and possibly ongoing cultivar exchanges between different Apterostigma ant species. MATERIALS AND METHODS Collection and Sampling Strategy Collections of Apterostigma ants and their gardens were made as part of extensive ant-cultivar surveys in Central and South America (Mueller et al. 1998). Three types o? Apter- ostigma gardens could be distinguished during field collec- tion: (1) G2 gardens, comprised of a G2-type fungus and characterized by a hanging, veiled garden architecture (gar- dens are suspended and surrounded by a mycelial envelope; Chapela et al. 1994; Fig. 2); (2) G3 gardens, comprised of a G3-type fungus and characterized by a sessile, unveiled gar- den architecture (gardens are spongelike and rest on the bot- tom of the garden cavity; Chapela et al. 1994; Mueller et al. 1998; Fig. 2); and (3) a heretofore unknown type of garden, termed G4, characterized by a sessile, unveiled garden ar- chitecture (thus resembling G3 gardens), but also possessing micromorphological characteristics of G2 fungi (e.g., abun- dant clamp connections; Chapela et al. 1994; U. Mueller, unpubl.). A total of 25 Apterostigma cultivar isolates from 25 nests of the following nine Apterostigma ant species were available for molecular analysis; A. auriculatum: Gamboa, Panama (w = 1); A. cf. gonoides: km 7 El Llano-Carti Suitupo Road, Panama (? = 1); A. dentigerum: Gamboa, Panama (w = 2); Pipeline Road, Panama (M = 1); Tule, Panama {n = 1); Fort Sherman Military Reservation, Panama {n = 1); A. dorotheae: Paramaketoi, Guyana (H = 2); Kurupukari; Guyana (n = 2); A. manni: Barro Colorado Island, Panama (? = 1); Pipeline Road, Panama (n = 2); Gamboa, Panama (n = 1); A. urichii: Kurupukari, Guyana (n = 1); A. pilosum sp. 1: Gamboa, Panama (M = 4); Kurupukari, Guyana (n = 1); A. pilosum sp. 2: El Llano, Panama {n = 1); and A. pilosum sp. 4: Gam- boa, Panama {n = 3). In all figures presented here, samples are indicated with reference to geographic origin (PA: Pan- ama, GU: Guyana) and sample number followed by the ant species name from which the fungus was obtained. The most recent revision of the genus Apterostigma (Lattke 1997) merges several previously recognized species into an unre- solved "pilosum-species complex," a default group that may be polyphyletic. Three species included in our study fall into the "pilosum-species complex" and are therefore listed as A. pilosum species 1, 2, and 4, respectively. Lattke's (1997) revision splits the genus Apterostigma into two broad groups, the auriculatum group (represented in our collection only by the single species A. auriculatum), and the pilosum group (represented in our collection by the remaining eight ant spe- cies). (Note that the "pilosum group" is different from the "pilosum species complex," which is a default complex in- cluded within the pilosum group [Lattke 1997]). Details of the collection and fungal-isolation methods are described in Mueller et al. (1996, 1998). Isolated fungi were grown in liquid culture, then preserved through lyophilization and cryostorage at ? 80?C. DNA Extractions Fungal mycelium was homogenized under liquid nitrogen, and the homogenate was extracted using a modified phenol- chloroform-CTAB procedure (Doyle and Doyle 1987): 1 hour incubation at 60?C with 2xCTAB (2% hexadecyltrimetyl am- monium bromide, 0.75 M NaCl, 50 mM Tris pH 8.0, and 10 mM EDTA), followed by one extraction with phenoLchlo- roform, one additional extraction with chloroform: isoamyl- alcohol (24:1), and precipitation with cold 100% ethanol. After resuspension of the DNA in water, the concentration FUNGICULTURE EVOLUTION IN APTEROSTIGMA ANTS 2255 of DNA was estimated on agarose gels with ethidium bro- mide, then diluted (5-10 ng/jxl) for polymerase chain reaction (PCR) amplification. Whole ants were extracted using a DNA-binding mem- brane kit without organic extraction or ethanol precipitation. Either the DNA/RNA isolation kit by Amersham Biosciences (Piscataway, NJ; catalog no. US73750) with the Amersham standard protocol, or the Qiagen DNeasy animal tissue ex- traction kit (Qiagen, Inc., Valencia, CA; catalog no. 69506) with the Qiagen standard protocol except that 50 |JL1 and 200 |JL1 of extraction buffer were used for the first and second elutions, respectively, and the Proteinase K digestion was carried out at 37?C for 45 min. RFLP Screen of Cultivated Fungi Cultivar DNA was analyzed in a two-step process, follow- ing the methods in Mueller et al. (1998). Each cultivar isolate was first profiled with internal transcribed spacers-restriction fragment length polymorphism markers, then representative RFLP types were selected for sequencing analysis. ITS-PCR reactions were carried out in a volume of 25 |JL1. Five |JL1 of ITS-PCR product were digested in a total volume of 20 \L\ using the following restriction enzymes in separate diges- tions: Nla III, Hha I, Dra I, RSA, and Hae III. Digested PCR product were electrophoresed in 2% agarose gels and visu- alized with ethidium bromide. Fungal DNA Sequencing and Cloning Target rDNA sequences from nuclear ITS and 25S gene regions were amplified using fungal primers ITS4 and ITS5 for the internal transcribed spacer regions (ITSl and ITS2, plus the intervening 5.8S region; White et al. 1990), and primers LROR and LR5 for the first 900 bp of the 25S region (large subunit, LSU; Rehner and Samuels 1994; Mueller et al. 1998). DNA sequencing of the ITS region showed multiple ITS-amplification products (high levels of heterozygosity or within-array ITS diversity), making it necessary to clone PCR products in most cases: 10 random PCR-clones were screened using RFLP and all different RFLP types were subsequently sequenced. This resulted in a varying number of sequences per sample, representing the sequence diversity within that fungal cultivar. Cloned sequences are represented as the sam- ple name followed by an underscore (i.e., GU34_2 corre- sponds to sample Guyana, sample 34, clone 2). ITS-PCR products were cloned into competent Escherichia coli cells, using a TA Cloning Kit (Invitrogen, Carlsbad, CA), then prepared for sequencing. ITS and LSU sequences were gen- erated on an ABI PRISM 377 DNA Sequencer (Applied Bio- systems, Foster City, CA) using the BigDye Terminator Cy- cle Sequencing Kit. Contigs were assembled from forward and reverse sequences using the DNASTAR software pack- age (Madison, WI). Sequences are deposited at Genbank un- der accessions AY367562-AY367612 (ITS) and AY367613- AY367633 (LSU). Ant DNA Sequencing Ant sequence data were generated for a 984 bp fragment (representing 328 amino acid residues) of the mitochondrial TABLE L Primers used to amplify cytochrome c oxidase subunit I (COI) DNA sequences from ants. Numbers in parentheses indicate annealing position of the 5' primer base in the COI coding sequence of the honey bee. Apis mellifera (Crozier and Crozier 1993). Forward primers: LCO1490 (17): 5' GGT CI13 (187): 5' ATA CI21 (684): 5' CTT Jerry (691): 5' CAA Reverse primers (position indicates annealing position of 5' primer base): HC02198 (728): 5' TAA CI14 (997): 5' GTT CI22 (1259): 5' ACT Ben (1121): 5' GCW CI24 (1293): 5' TCC CAA CAA ATC ATA AAG ATA TTG G 3' ATT TTT TTT ATA GTT ATA CC 3' TAT CAA CAT TTA TTT TGA TTT TT 3' CAT TTA TTT TGA TTT TTT GG 3' ACT TCA GGG TGA CCA AAA AAT CA 3' TCT TTT TTT CCT CTT TC 3' CCA ATA AAT ATT ATA ATA AAT TGA 3' ACW ACR TAA TAK GTA TCA TG 3' TAA AAA ATG TTG AGG AAA 3' cytochrome c oxidase subunit I (COI) gene, corresponding to positions 211 to 1195 of the Apis mellifera COI coding sequence (Crozier and Crozier 1993). Extracted DNA were generally amplified in two steps: In the first step, a fragment of approximately 1277 bp was amplified using the primers LCO1490 (forward) and CI24 (reverse) with a 45?C annealing temperature and 1-min ramping between the annealing and extension steps. Then, using the product of the initial am- plification as template, two shorter overlapping fragments were reamplified with a variable (48?C to 52?C) annealing temperature and no ramping, using as primers, for the 5' fragment, either LCO1490/HCO2198 or CI13/CI14, and, for the 3' fragment, primer combinations CI21/CI24, CI21/CI22, Jerry/CI24, or Jerry/Ben3R. Primer sequences are listed in Table 1. Sequences are deposited at Genbank under acces- sions AY398280-AY398304. Phylogenetic Analyses For analyses of fungi, several species closely related to the Apterostigma cultivars were selected as outgroups based on a recently published phylogeny of the Agaricales (Petersen and Krisai-Greilhuber 1999, Moncalvo et al. 2000), and re- spective sequence information was obtained from Genbank. For analyses of ants, species from two genera (Myrmicocrypta and Mycocepurus) closely related to Apterostigma (Schultz and Meier 1995; Lattke 1997, 1999) were used as outgroups. Fungal sequences were initially aligned in Megalign (Dynastar) and further aligned by hand; unalignable regions were excluded from the analysis. Alignment of the 984 bp of ant nucleotide sequence was trivial, as COI is a protein-coding gene and amino acid number is highly conserved across all ants (T. Schultz, unpubl.). DNA sequence alignments can be obtained from http: //www.daimi.au.dk/~biopv/cv/apterostigma/. Four dataseis were separately analyzed (1) large sub- unit(LSU) (25S) for cultivated and free-living related fungi, (2) internal transcribed spacers (ITS) for G2-cultivars, (3) ITS for G4-cultivars, and (4) COI for attine ants {Mycoce- purus, Myrmicocrypta, and Apterostigma). Parsimony (MP) analyses.?Maximum-parsimony analy- ses were conducted in PAUP 4.0b4a (Swofford 2001) using the heuristic search option with tree bisection-reconnection (TBR) (LSU, G2-ITS, and ants) or branch-and-bound (G4- 2256 PALLE VILLESEN ET AL. TABLE 2. Cultivar group (G2, G3, and G4), garden architecture, and ITS-RFLP type for 22 Apterostigma cultivars isolated from 22 colonies of nine Apterostigma species. Apterostigma auriculatum is the only Apterostigma species that is known to cultivate a lepiotaceous G3-fungus (Mueller et al. 1998); one A. auriculatum cultivar was included for comparison with the two other groups (G2 and G4) of pterulaceous Apterostigma cultivars. G3 and G4 gardens have a sessile and unveiled architecture, whereas G2 gardens are hanging and veiled (see Fig. 2). Micromorphological similarities between G2 and G4 fungi had suggested a possible close phylogenetic link between G2 and G4 fungi (see Materials and Methods). Cultivar type (micromorphology) Garden architecture RFLP type G3 G4 G2 site, unveiled nging, veiled A B C D E F G L M N O Ant species A. auriculatum A. pilosum sp. 4 A. cf. gonoides A. manni A. dentigerum A. dorotheae A. pilosum sp. 1 A. pilosum sp. 2 A. urichii Total 1 1 1 1 112 1112 1 2 1 2 1 1 1 1 3 1 1 1 Total 1 3 1 4 4 4 3 1 1 22 ITS) branch swapping and 500 (for fungi) and 1000 (for ants) random-taxon-addition replicates; successive-approxima- tions weighting (SW) analyses used 200 (fungi) and 500 (ants) replicates. Heuristic-search bootstrap analyses (Fel- senstein 1985) used TBR branch-swapping and consisted of 1000 (for fungi) and 5000 (for ants) pseudoreplicates, with 10 random-taxon-addition replicates per pseudoreplicate. Maximum-parsimony-based comparisons of unconstrained and constrained MP trees used the K-H test (Kishino and Hasegawa 1989), the Templeton Wilcoxon signed-ranks test (Templeton 1983), and the winning-sites test (Prager and Wilson 1988) in PAUP 4.0b8. Maximum-likelihood (ML) analyses.?Nucleotide substi- tution models for ML analyses were evaluated with the data and MP tree(s) using the likelihood-ratio test implemented in ModelTest 3.0 (Posada and Crandall 1998). Maximum- likelihood analyses were conducted in PAUP 4.0b4a (Swof- ford 2001). For the LSU and ant data, heuristic searches employed the adopted model and an optimal MP tree as the branch-swapping starting tree and consisted of five iterative subsearches, each using updated model parameter values based on the results of the preceding search and each using successively more intensive branch-swapping regimes (Cur- rie et al. 2003; Sallum et al., 2002; Mueller et al., 1998). For the G2-1TS data, heuristic searches employed the adopted model and a series of five TBR branch-swapping subsearches in which model parameters were successively updated. For the G4-1TS data, heuristic searches employed the adopted model and 100 random-taxon-addition subsearches. Heuristic ML bootstrap analyses employed TBR branch-swapping and consisted of 100 (LSU), 1000 (G2-1TS, G4-ITS), and 1500 (ants) pseudoreplicates. Maximum-likelihood-based compar- isons of unconstrained and constrained ML trees used the K- H test (Kishino and Hasegawa 1989) and the S-H test (Shi- modaira and Hasegawa 1999; Goldman et al. 2000). Bayesian analyses.?Bayesian analyses were conducted in MrBayes 2.01 (Huelsenbeck and Ronquist 2001). Analyses of the LSU data employed the GTR + F + I model (Rod- riguez et al. 1990; general time-reversible model of nucle- otide substitution with a proportion of sites invariant and gamma-distributed rates); analyses of the G2-1TS and G4- ITS data employed the HKY model (Hasegawa et al. 1985); and analyses of the ant data employed the SSR + F model (Huelsenbeck and Ronquist 2001; site-specific rate classes with gamma-distributed rates) incorporating three codon-po- sition rate classes. All analyses included four separate runs, each consisting of 300,000 Markov chain Monte Carlo (MCMC) generations and four simultaneous MCMC chains (three heated), and each with a "burn-in" of 100,000 gen- erations. For each analysis, post-burn-in trees from all four runs were pooled to calculate posterior probabilities. Bayes- ian topology-based hypothesis tests consisted of enumerating the proportion of pooled post-burn-in trees consistent with each of the competing hypotheses. RESULTS Restriction Fragment Length Polymorphisms Typing The restriction fragment length polymorphisms (RFLP) typing of 25 fungal cultivars from nine ?i??sr&n? Apterostigma species (A. auriculatum, A. cf. gonoides, A. manni, A. dentig- erum, A. dorotheae, A. pilosum sp. 1, A. pilosum sp. 2, A. pilosum sp. 4, and A. urichii) is consistent with the three distinct fungal groups (G2, G3, and G4) noted during the field collection (see Materials and Methods; Table 2). Each of the three fungal groups is cultivated by its own distinct, nonoverlapping set of ant species (Table 2). Apterostigma auriculatum is the only known Apterostigma species to cul- tivate a G3-fungus; because this fungus was previously shown to be lepiotaceous (Mueller et al. 1998), it is not included in the subsequent phylogenetic analyses of the pter- ulaceous cultivars. Of the Apterostigma ant species cultivat- ing pterulaceous fungi, some ant species, such as manni, den- tigerum, and dorotheae (Table 2), cultivate several different RFLP cultivar types within their particular cultivar group (G2 vs. G4 fungi, respectively), indicating cultivar diversity with- FUNGICULTURE EVOLUTION IN APTEROSTIGMA ANTS 2257 in single ant species. Furthermore, the same RFLP cultivar type is shared in four cases between different ant species, suggesting either cultivar exchange between ant species via lateral transfer between nests (e.g., garden stealing), or in- dependent domestication of the same free-living fungal lin- eage, as has been hypothesized for other lower attine species (Mueller et al. 1998). Phylogenetic Analyses Cultivar 25S (LSU) dataset.?Maximum-parsimony analy- ses of the LSU data produced 16 equally parsimonious trees (MPTs) with parsimony-informative length = 505, CI = 0.485, RI = 0.845; successive approximations weighting fa- vored a subset of two of these trees. For both of these trees, the likelihood ratio test found the TrN + I + F model (Tamura and Nei 1993; but with a proportion of sites invariant, and gamma-distributed rates) to be significantly better fitting than the next less complex model (P = 0.000010), with both trees equally likely. ML analysis with this model identified a single tree (Fig. 3) with log likelihood of -4321.41458. A second analysis using the most complex (i.e., most general) model available, GTR + I + F (Rodr?guez et al. 1990), identified exactly the same topology. Bayesian analyses using this model identified a tree entirely congruent with the ML tree. The Apterostigma cultivars are monophyletic and bootstrap pro- portions and Bayesian posterior probabilities (Fig. 3) strongly support the historical divergence o? Apterostigma cultivars into two clades. These clades, respectively called G2- and G4- cultivars, are congruent with the G2- and G4-cultivar groups previously revealed by morphological and RFLP data (see above). The genus Gerronema is reconstructed as the sister clade to the Apterostigma cultivars under both likelihood and parsimony criteria. This relationship was suggested in Mon- calvo et al. (2000), although with very low support. Our current taxon sampling fails to recover the hydropoid clade as de- scribed in other papers (e.g. Moncalvo et al. 2000), thus the relationships among the outgroup taxa in our phylogeny may be dependent on our particular taxon sampling. This potential criticism of reconstructions among outtgroup taxa does not apply to the interpretation of the phylogenetic relationships among the Apterostigma cultivars discussed below. Cultivar G2-ITS dataset.?MP analyses of the G2-ITS data produced 20 MP trees with parsimony-informative length = 105, C.I. = 0.733, R.I. = 0.927. Successive approximations weighting identified a subset of two of these trees. For both of these trees, the likelihood ratio test found the HKY model (Hasegawa et al. 1985) to be significantly better fitting than the next less-complex model (P < 0.000001), with both trees equally likely. Maximum Likelihood analysis using this mod- el identified a single tree (Fig. 4) with log likelihood of ? 2139.6617. Bayesian analyses using this model identified a nearly identical tree, differing with regard to the position of one clade (not shown). Single species of Apterostigma (e.g., dentigerum, dorotheae, pilosum sp. 1) cultivate diverse sets of fungi drawn from different clades of this phylogeny; thus single Apterostigma species appear polymorphic for their cultivars, possibly as a result of cultivar exchanges between different G2-cultivating Apterostigma lineages (Fig. 4). Cultivar G4-ITS dataset.?MP analyses of the G4-ITS da- taset produced a single MPT with parsimony-informative length = 49, C.I. = 0.898, R.I. = 0.968. Successive ap- proximations weighting identified the same tree. The likeli- hood ratio test found the HKY model (Hasegawa et al. 1985) to be significantly better fitting to the data and this tree than the next less complex model {P = 0.000006). ML analysis using this model identified a single tree (Fig. 5) with log likelihood of ?1274.70873. Bayesian analyses using this model identified the same tree. Apterostigma pilosum sp. 4 and A. manni each cultivates fungi from the two main clades of this phylogeny; these iwo Apterostigma species thus appear polymorphic for their cultivars, possibly as a result of cultivar exchanges between different G4-cultivating Apterostigma lin- eages (Fig. 5). Ant dataset.?MP analyses of the ant data set identified two equally parsimonious trees (MPTs) with parsimony-in- formative length = 1396, C.I. = 0.437, R.I. = 0.658. Suc- cessive approximations weighting identified the same two trees. Maximum Parsimony analyses under the constraint that the G4-cultivating ants are monophyletic identified four MPTs with parsimony-informative length = 1404, C.I. = 0.432, R.I. = 0.655. Successive approximations weighting identified two trees (SWTs), a subset of the four MPTs. Maximum-parsimony comparisons of the unconstrained and constrained trees indicate that the topologies in which the G4-cultivating Apterostigma species are constrained to be monophyletic are significantly worse-fitting to the data under both the K-H (P = 0.0454) and Templeton (P = 0.0455) test criteria, and marginally significantly worse fit- ting under the winning sites test criterion {P = 0.0768). Based on the parsimony criterion, G4-cultivating ants thus do not appear to be monophyletic, but paraphyletic with respect to the G2-cultivating ant clade (Fig. 6). For both the unconstrained and constrained topologies, the likelihood ratio test found the GTR + I + F model (Rodriguez et al. 1990) to be significantly better fitting to the data and the MPTs than the next less complex model (P < 0.000001), with both unconstrained trees equally likely, and both con- strained trees equally likely. Unconstrained ML analysis iden- tified a single tree (Fig. 6) with log likelihood of -7105.27198. This tree is identical in ingroup topology to the MPTs. Like- lihood analyses under the constraint that the G4-cultivating ants are monophyletic, and using one of the constrained SWTs as a starting tree for branch-swapping, identified a single tree with a log likelihood of ?7108.0347. This tree is identical in ingroup topology to the constrained MPTs. Maximum-likelihood comparisons of the unconstrained and constrained trees indicate that the topologies in which the G4-cultivating Apterostigma species are constrained to be monophyletic are not significantly worse fitting to the data, using the K-H test with both normal (P = 0.3167) and 1000 resampling of estimated likelihood (RELL) bootstrap (P = 0.303) approximations, and using the S-H test with 1000 RELL bootstrap approximations (P = 0.150) (Kishino and Hasegawa 1989; Shimodaira and Hasegawa 1999; Goldman et al. 2000). Likelihood tests are thus inconclusive about the paraphyly of G4-cultivating ants. The 50% majority-rule consensus of the pooled post-burn- in trees, identified by Bayesian analysis using the SSR + F model, is identical in ingroup topology to the trees found in 2258 PALLE VILLESEN ET AL. 100 95,82 T ? Collybia polyphylla Collybia dryophila Lentinula edades Rhodocollybia maculata Marasmiellus ramealis 95 52, <50 100 98,95 100 , 80,56 Micromphale per?orans |? Omphalotus nidiformis '? Neonothopanus namb?* Campanella subdendrophora Armillar?a tabescens Marasmius pyrocephalus ? Gloiocephala menieri Strobilurus trullisatus Rhodotus palmatus Xerula furfuracea 100 97 <50, <50 95 56,65 92,99 / rC 100 100 69,50 98,94 75,88 Pleurotopsis longincua Baeospora myriadophylla - Hydropus scabripes Marasmius deledans Crinipellis maxima 100 [- Gerronema strombodes Gerronema subclavatum r PA cultivar A. 100, 100 ?j 100 100 100, 100 100,100 r. 100 92, 100 manni ? PA cultivar A. manni PA cultivar A. pilosum sp. 4 PA cultivar A. cf. gonoides PA cultivar A. manni PA cultivar A. manni PA cultivar A. pilosum sp. 4 GU cultivar A. dorotheae GU cultivar A. urichii GU cultivar A. dorotheae PA cultivar A. sp. PA cultivar A. pilosum sp. 1 PA cultivar A. pilosum sp. 2 PA cultivar A. dentigerum PA cultivar A. pilosum sp. 1 - GU cultivar A. dorotheae PA cultivar A. dentigerum PA cultivar A. dentigerum PA cultivar A. dentigerum PA cultivar A. dentigerum PA cultivar A. dentigerum FiG. 3. Maximum-likelihood phylogeny o? Apterostigma cultivars and free-living fungi currently placed in the family Tricholomataceae (Basidiomycota: Agaricales), based on 25S rDNA sequences. Numbers indicate Bayesian posterior probabilities (above branches), par- simony bootstrap proportions (below branches, left), and likelihood bootstrap proportions (below branches, right). GU (Guyana) and PA (Panama) indicate collection locations of the Apterostigma nests. The Apterostigma cultivars are monophyletic and divided into two well- supported clades, respectively called G2 and G4 cultivars. 0.005 substitutions/site the MP and ML analyses (Fig. 6). Ants in the basal attine genera Myrmicocrypta and Mycocepurus, as well as the basal Apterostigma species A. auriculatum, all cultivate lepiota- ceous fungi (G3 cultivars), whereas all other Apterostigma species in this phylogeny cultivate pterulaceous fungi (G2 and G4 cultivars). The monophyly of the pterulaceous-cul- tivating Apterostigma species indicates that the transition away from lepiotaceous cultivation and to pterulaceous cul- tivation occurred only once. The basal, paraphyletic position of G4-cultivating ants among the pterulaceous-cultivating ants further suggests that the direction of this transition was from G3 to G4 fungi, that G2 cultivation evolved subsequent to this original transition, and that G2-cultivating Aptero- stigma ants probably arose from G4-cultivating ancestors. Examination of the Bayesian post-burn-in trees indicates that the hypothesis that the G4-cultivating Apterostigma species are paraphyletic with respect to the G2-cultivating species has a posterior probability of >99.98%. The alternative hypothesis FUNGICULTURE EVOLUTION IN APTEROSTIGMA ANTS 2259 Apterostigma species 100 86,96 94 GU32 cultivar - A. pilosum sp. 1 - GU39_10 cultivar-/\. dorotheae - GU39_11 cultivar->i. dorotheae GU39_2 cultivar - A. dorotheae GU39_3 cultivar - A. dorotheae GU39 4 cultivar - A. dorotheae GU39_5 cultivar - A. dorotheae ? GU39_6 cultivar - A. dorotheae GU39_7 cultivar - A. dorotheae ? GU39_9 cultivar - A. dorotheae GU34_2 cultivar - A. dorotheae ? GU34_7 cultivar - A. dorotheae ? PA408_9 cultivar - A. dentigerum ? PA430_2 cultivar - A .dentigerum PA431_1 cultivar-^, dentigerum ? PA431 6 cultivar->A. dentigerum 64,62 ^ PA204_1 cultivar - A. dentigerum PA204_4 cultivar - A. dentigerum PA430_1 cultivar - A. dentigerum PA326_2 cultivar - A. dentigerum GU51 11 cultivar-A dorotheae 100 100 94,100 100 93,95 100 67,66 94 100, 100 GU51_1 cultivar - A. dorotheae GU58_1 cultivar - A. dorotheae GU58_10 cultivar->A. dorotheae PA85 cultivar - A. sp. - PA407_8 cultivar - A. sp. - PA410_1 cultivar-,A. pilosum sp. 2 - PA409_2 cultivar - A. pilosum sp. 1 PA407_10 cultivar - A. sp. PA410_8 cultivar - A. pilosum sp. 2 GU56 1 cultivar - A. urichii \ GU56 8 cultivar - A. urichii 100 62,100 ?- C PA411_2 cultivar - A. dentigerum 95,1001? PA411_1 cultivar-A dentigerum PA412 1 cultivar - A. pilosum sp. 1 100 97, 100 ? PA412_4&16 cultivar - A. pilosum sp. 1 PA412_7 cultivar - A. pilosum sp. 1 PA412_8 cultivar - A. pilosum sp. 1 PA412_2 cultivar - A. pilosum sp. 1 0.005 substitutions/site i? li i. a* 3i "O C Q. "53 Q. 5: 3. Ct> & O P O 3 p 3 3- Cf Go =::CQ- ~ CD to C 3 ?8 I FIG. 4. Maximum-likelihood phylogeny of the G2 Apterostigma cultivars, based on ITS sequences. The tree is arbitrarily rooted. Sequences are labeled as collection location of Apterostigma nests (GU, Guyana and PA, Panama) followed by sample number. For many cultivars, several ITS sequences were generated after cloning of PCR products (see Methods); these ITS clones are indicated by an underscore (e.g., "_2") with the cultivar from which they were derived. Numbers indicate Bayesian posterior probabilities (above branches), parsimony bootstrap proportions (below branches, left), and likelihood bootstrap proportions (below branches, right). Single species of Apterostigma appear polymorphic for their cultivars, possibly as a result of cultivar exchange between ant species. that the G4 cultivators are monophyletic has a posterior prob- ability of <0.02%. Thus, Bayesian MCMC analyses strongly support the conclusion that the G4-cultivating Apterostigma spe- cies are transitional between the plesiomorphic G3 cultivators and the derived G2-cultivating ant clade. Phylogenetic analyses of fungal LSU sequence data under MP, ML, and Bayesian criteria reconstruct two distinct clades of symbionts cultivated by Apterostigma ants (Fig. 3): the G2 group, previously recognized by Chapela et al. (1994), and a novel G4 group, which is the sister clade to the G2 fungi and is recognized here for the first time. Any given Aptero- stigma ant species cultivates fungi exclusively in either the G2 or the G4 group, indicating broad specialization of Ap- terostigma species on one of the two cultivar groups. The closest free-living (i.e., nonsymbiotic) relatives of the G2 and G4 Apterostigma cultivars in our analysis are the fungi 2260 PALLE VILLESEN ET AL. Apterostigma species PA44 2 cultivar - A. manni PA21 1 cultivar - A mann/ PA91 1 cultivar-JA. manni PA91 2 cultivar - A. manni PA2 cultivar - A. piiosum sp. 4 100 100, 100 PA426_5 cultivar - A. piiosum sp. 4 ? PA403_1 cultivar - A. piiosum sp. 4 ? PA403_9 cultivar - A. piiosum sp. 4 PA406_1 cultivar - A. cf. gonoides PA406_3 cultivar - A. cf. gonoides PA86 cultivar - A. manni 0.005 substitutions/site i* ?A i> o w c 3 -8 ni ? ^ r? G4 > G2 sequence is strongly supported by the Bayesian analysis (>99.98 % posterior probability), were as the alternative topology, in which the G4 and G2 cultivators are sister groups, has very low support (<0.02 % posterior probability). Extended phylogenetic analyses of Apterostigma ant spe- cies and their fungal cultivars will directly test whether cul- tivation of lepiotaceous cultivars (G3 cultivation) is plesiom- orphic for the genus Apterostigma, a scenario favored by Mueller et al. (1998, 2001). The current taxonomy of the ant genus (Lattke 1997) indicates only a division into two broad groups, the auriculatum group and the pilosum group (the pilosum group is different from the "pilosum species com- 2262 PALLE VILLESEN ET AL. 100 99,100 100 98,100 Myco?pums sp. Mycocepurus tardas - Mycocepurus tardus 100 60,56 100 99,100 Mycocepurus tardus Myrmicocrypta sp. 2 Myrmicocrypta sp. 2 Myrmicocrypta sp. 1 Myrmicocrypta buenziii Myrmicocrypta buenziii 100 100 95,94 99,100 100 - A. auriculatum A. auriculatum 100 <50, <50 <50, 62 100 B 99,100 100 100 <50, 57 99,100 100 100 90,65 0.05 substitutions/site 89,90 100 87,91 76,98 Apterostigma ant species A. cf. goniodes - A. manni - A. manni A. sp. A. pilosum sp. 4 A. pilosum sp. 4 A. pilosum sp. 4 A. pilosum sp. 2 - A. dentigerum - A. dentigerum A. cf. dorottieae - A. cf. dorotheae A. pilosum sp. 1 A. pilosum sp. 1 100 98,100 93 70,92 O O < o ?\ Q O c_ < Q} ??F O ?I Q to o ??H