Introduction

Ants are critical components of terrestrial ecosystems around the world (Hölldobler and Wilson, 1990). Among ants, leaf-cutters in the genus Atta (Figure 1a) are particularly dominant, with mature colonies achieving immense sizes and housing millions of workers (Hölldobler and Wilson, 2008, 2010). Ranging from the southern United States to Argentina, species of leaf-cutter ants can construct elaborate subterranean nests containing hundreds of chambers and displacing up to 40 000 kg of soil (Hölldobler and Wilson, 2010). The ant societies housed within these nests are equally impressive, with an intricate division of labor observed between different castes of workers (Hölldobler and Wilson, 2010). Associated with this division of labor is substantial worker-size polymorphism: the dry weight of individual workers in the same colony can differ by 200-fold (Hölldobler and Wilson, 2010). The success of leaf-cutter ants is largely attributed to their obligate mutualism with a basidiomycetous fungus (Leucoagaricus gongylophorus) that they culture for food in specialized gardens (Figure 1b) (Weber, 1966; Hölldobler and Wilson, 2008, 2010). Fresh plant forage collected by the ants serves to nourish the fungus, which in turn produces nutrient-rich hyphal swellings (gongylidia) that feed the colony (Weber, 1966). The symbiosis between leaf-cutter ants and their cultivar is thought to have originated 8–12 million years ago, and numerous adaptations in both the ants and the fungus have occurred over this long history of agriculture (Weber, 1966; Chapela et al., 1994; Schultz and Brady, 2008).

Figure 1
figure 1

Leaf cutter ants forage on plant material (a) that they use as manure for specialized fungus gardens (b). Pyrosequencing of bacterial 16S genes from fungus gardens of the leaf-cutter ants A. colombica and A. cephalotes recovered 8000–12 000 sequences representing 25–274 OTUs (OTUs, 97% identity cutoff) (c). Microbial community composition was then investigated by directly pyrosequencing 382–441 Mbp of DNA from the same leaf-cutter ant fungus gardens and using the Genome Abundance and Average Size tool to estimate the relative abundance of different microbial groups (d) (photo credits: a, Jarrod J Scott; b, Austin D Lynch).

Full size image

The fresh-foliar biomass leaf-cutter ants integrate into their fungus gardens is composed largely of recalcitrant lignocellulosic polymers. The ants presumably gain indirect access to the carbon stored in plant cell walls through the metabolic activities of their fungus gardens, which act as an ancillary digestive system (Pinto-Tomas et al., 2009). Despite being a critical aspect of leaf-cutter ant biology, the process through which fungus gardens degrade plant forage has only recently been intensely investigated (De Fine Licht et al., 2010; Schiott et al., 2010; Suen et al., 2010; Semenova et al., 2011). Originally it was thought that the fungal cultivar primarily degraded cellulose, and that this was the main polymer converted into nutrients for the ants (Martin and Weber, 1969). However, the cellulolytic capacity of this fungus has come into question, as it has been shown that pure cultures cannot grow on cellulose as a sole carbon source (Abril and Bucher, 2002). This has led to the suggestion that cellulose is not deconstructed in leaf-cutter ant fungus gardens, but rather that the fungal cultivar uses a variety of hemicellulases to deconstruct primarily starch, xylan and other plant polymers (Gomes De Siqueira et al., 1998; Silva et al., 2006a, 2006b; Schiott et al., 2008).

Another model posits that plant cell wall degradation in fungus gardens is partially mediated by lignocellulolytic bacteria. There is some support for this model. Importantly, recent work has found evidence for substantial cellulose deconstruction in the fungus gardens of Atta colombica and the presence of lignocellulolytic bacteria in these ecosystems (Suen et al., 2010). Another study, employing the culture-independent analysis of membrane-lipid markers, has supported the hypothesis that a distinct community of predominantly Gram-negative bacteria resides in fungus gardens (Scott et al., 2010), and the presence of symbiotic nitrogen-fixing bacteria in the genera Pantoea and Klebsiella has also been shown (Pinto-Tomas et al., 2009). Together with culture-dependent investigations recovering microbial groups with a broad array of metabolic activities (Bacci et al., 1995; Santos et al., 2004), these experiments have led to the suggestion that fungus gardens represent specialized fungus–bacteria consortia selected for by the ants, and that the bacteria have essential roles, including plant biomass degradation, nutrient biosynthesis, and competitive or antibiotic-mediated exclusion of pathogens (Mueller et al., 2005; Haeder et al., 2009; Pinto-Tomas et al., 2009; Suen et al., 2010).

Using a combination of metagenomics and metaproteomics, we provide insights into the microbial activities in leaf-cutter ant fungus gardens. Culture-independent investigations have previously been performed on leaf-cutter ant fungus gardens (Scott et al., 2010; Suen et al., 2010), but to date only a small quantity of bacterial sequences (∼6 Mb) from the fungus gardens of a single ant species have been characterized. Here, by expanding on previous work, we sought to document the non-eukaryotic component of fungus gardens, describe the similarity of communities from different ant species and examine potential microbial activities in situ. To this end, we generated three 16S pyrotag libraries of over 8000 sequences each and over 1.2 Gbp of raw 454 Titanium community metagenomic data from the bacterial component of A. cephalotes and A. colombica fungus gardens. To account for potential differences in microbial communities due to the extent of plant biomass degradation, we individually examined the top and bottom strata of A. colombica fungus gardens, which correspond to where the ants integrate fresh forage and remove partially degraded plant substrate, respectively. We then conducted metaproteomic analyses on whole fungus gardens to identify proteins produced in these ecosystems and examine the physiology of resident bacteria in more detail. We found that similar bacterial communities inhabit all fungus-garden samples analyzed, and that the metabolic potential of resident bacteria includes nutrient biosynthesis, hemicellulose and oligosaccharide degradation, and other functions that potentially enhance plant biomass processing in these ecosystems. Below we discuss a novel framework for understanding the complex interplay between leaf-cutter ants and the symbiotic communities residing in their fungus gardens.

Materials and methods

Sample processing for community metagenomes and 16S pyrotag libraries

Fungus gardens from healthy A. cephalotes and A. colombica colonies were collected from nests near Gamboa, Panama, in April 2009. Whole A. cephalotes gardens were combined for subsequent analyses, whereas fungus gardens of A. colombica were laterally bisected to separate the top and bottom strata. Immediately after collection, the bacterial fraction of the samples was isolated and DNA was extracted as previously described (Suen et al., 2010). Briefly, plant, ant and fungal material were removed from all samples through a series of washing or centrifugation steps using 1 × PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 2 mM KH2PO4). DNA was subsequently extracted from the remaining bacterial fraction using a Qiagen DNeasy Plant Maxi Kit (Qiagen Sciences, Germantown, MD, USA). One community metagenome and one 16S library were generated from each of the three samples using 454 Titanium-pyrosequencing technology (Margulies et al., 2005). Draft genomes of three bacteria isolated from Atta fungus gardens were also generated to supplement the reference databases used for the phylogenetic binning of metagenomic data. Technical details for the sequencing, assembly, and annotation of all data can be found in the Supplementary Information.

Metaproteomics

Metaproteomic analysis was conducted on fungus-garden material collected in Gamboa, Panama, from a nest of A. colombica distinct from that used for metagenomic analyses. Moreover, we also conducted metaproteomic analyses on a lab-reared colony of A. sexdens for comparison. Detailed methods can be found in Supplementary Information. Briefly, proteins were extracted from whole fungus-garden material, and the resulting protein solution was digested into peptides and subsequently analyzed by liquid chromatography tandem mass spectrometry (LC-MS/MS). The resulting peptide tandem mass spectra were compared with predicted protein datasets of the three community metagenomes individually. Peptide matches were filtered using Sequest (Eng et al., 1994) scores, MS-GF spectral probabilities (Kim et al., 2008), and false discovery rates. We restricted our functional analyses to peptides mapped to proteins phylogenetically binned as bacterial, and the IMG-ER and KEGG annotations of these proteins were inspected to identify those potentially involved in biomass degradation or nutrient cycling (Table 5). Peptides mapping to these select proteins were inspected manually (Figure 5, Supplementary Dataset 6).

Results

Community metagenomes and 16S pyrotag libraries

Pyrosequencing of the V6–V8 variable region of the bacterial 16S rRNA gene for the same three samples yielded between 8000–12 000 reads (termed ‘pyrotags’) each (Table 2). Previous attempts to recover Archaeal 16S sequences from fungus gardens were unsuccessful (Suen et al., 2010), and amplification of these genes was not attempted here. Pyrosequencing of community DNA from three samples, representing both the individual top and bottom strata of A. colombica fungus gardens as well as the combined strata of A. cephalotes gardens, each yielded 382–441 Mb of raw sequence data (Table 1). Reads from each library were assembled into community metagenomes comprising 40–100 Mbp of sequence data.

Table 1 Sequencing statistics of the community metagenomes
Full size table

Microbial diversity in fungus gardens

Clustering of sequences in the 16S pyrotag libraries from the A. colombica top, A. colombica bottom and A. cephalotes fungus-garden samples recovered 204, 274 and 25 operational taxonomic units (OTUs, 97% identity cutoff), respectively. The majority of the OTUs were most similar to sequences of Gammaproteobacteria and Firmicutes (22–217 OTUs, 72–89% of OTUs, and 2–35 OTUs, 5–12% of OTUs, respectively), and only OTUs corresponding to those phyla were represented in all three samples (Figure 1c). Phyla represented in lower abundance and more sporadically included the Betaproteobacteria (⩽10 OTUs, ⩽4.9% of OTUs), Alphaproteobacteria (⩽7 OTUs, ⩽3.4% of OTUs), Bacteriodetes (⩽4 OTUs, ⩽2% of OTUs), Acidobacteria (⩽1 OTU, ⩽1% of OTUs) and Actinobacteria (⩽1 OTU, ⩽4% of OTUs) (Figure 1c). Most pyrotags corresponded to the Gammaproteobacterial family Enterobacteriaceae (79–99% of individual pyrotags, Table 2). Although taxonomic profiles were similar in all three pyrotag libraries, the bacterial diversity of each of the A. colombica samples was greater than that recovered from the A. cephalotes fungus garden sample.

Table 2 Family-level classification of partial-length 16S sequences recovered from Atta colombica and Atta cephalotes fungus gardens
Full size table

Community metagenomic analyses recovered primarily bacterial sequences (71–80% of total assembled bp) (Table 3). Consistent with the 16S pyrotag libraries, the majority of sequences in all three data sets matched most closely to Gammaproteobacteria (69–72%), especially Enterobacteriaceae (53–70%). To refine taxonomic resolution and infer the relative abundance of microbial groups, raw reads were phylogenetically classified using the Genome relative Abundance and Average Size (GAAS) tool (Angly et al., 2009). Estimates based on GAAS analyses indicate that the Gammaproteobacteria were particularly abundant, with the genus Enterobacter comprising over 50% of the bacterial population in all the three metagenomes (Figure 1d). The community metagenomes also contained representative sequences from the genera Klebsiella (3.8–4.9%), Pantoea (1.8–15.6%), Escherichia (5.3–6.3%), Citrobacter (3–5.8%), Pseudomonas (0.04–4.2%) and Lactococcus (0.01–2.2%). BLAST-based classification of the assembly indicated that ∼1% of the sequences corresponded to bacteriophage in each of the community metagenomes (Table 3), whereas the GAAS tool estimated that 15.1%, 16.8% and 6.8% of the A. colombica top, A. colombica bottom and A. cephalotes metagenomes could be comprised of bacteriophage, respectively.

Table 3 Phylogenetic classification of all assembled contigs and singletons in the leaf-cutter ant fungus garden metagenomes
Full size table

Consistent with the GAAS- and BLAST-based analyses, the largest phylogenetic bins created by phymmBL were assigned to the genera Enterobacter, Pantoea, Klebsiella, Escherichia, Citrobacter and Pseudomonas. The Enterobacter bins were by far the largest, containing 15.3–29.5 Mb of sequence. The majority of these sequences were most similar to the draft genome of Enterobacter FGI 35, a strain isolated in this study from an A. colombica fungus garden. The Pantoea bins were the next largest, containing between 5–7.2 Mb of sequence each.

Metabolic potential of bacterial lineages

To compare the coding potential of different bacterial groups in fungus gardens, we analyzed genus-level phylogenetic bins of sequences constructed from the community metagenomes. Comparison of the coding potential in the bins with the KEGG database (Kanehisa et al., 2008) recovered well-represented sugar metabolism pathways in most of the Enterobacteriaceae bins (Figure 2). Moreover, pathways involved in B-vitamin and amino-acid metabolism were found to be highly represented in both the Pseudomonas and Enterobacteriaceae bins. The Lactococcus bins showed relatively low representation in most of these pathways. Clustering of phylogenetic bins from each of the metagenomes by their KEGG pathway representation indicated that bacterial members corresponding to the same genus, with the exception of Citrobacter, had similar metabolic profiles.

Figure 2
figure 2

Reconstruction of KEGG pathways recovered from phylogenetic bins generated from the leaf-cutter ant fungus-garden metagenomes. KEGG profiles normalized by the number of predicted proteins in each phylogenetic bin were used for the clustering analysis. Pathways involved in the metabolism of carbohydrates, amino acids and B-vitamins were among the most highly represented and are shown here.

Full size image

To examine how leaf-cutter ant fungus garden microbial communities differed from other environments, we predicted Clusters of Orthologous Groups (COGs (Tatusov et al., 2001)) from all contigs and reads from the three fungus-garden metagenomes and compared these with COG profiles from all other metagenomes available on the Integrated Microbial Genomes/Microbiomes (IMG/M) database (Markowitz et al., 2008) (Figure 3). COG profiles for the three fungus-garden metagenomes were found to be highly similar. Compared with all other metagenomes on IMG, many COG categories were overrepresented in fungus gardens (Fisher's exact test, P<0.01), including amino-acid transport and metabolism, carbohydrate transport and metabolism, and inorganic ion transport and metabolism (Figure 3). Specific COGs involved in carbohydrate transport and metabolism were analyzed in more detail to investigate possible bacterial roles in polysaccharide degradation, and sugar transporters and phosphotransferase system components in particular were found to be significantly overrepresented in the fungus-garden metagenomes (Fisher's exact test, P<0.01) (Supplementary Dataset S2).

Figure 3
figure 3

Comparison of the COG category distributions of the three combined fungus-garden metagenomes (A. colombica top, A. colombica bottom, and A. cephalotes combined) and all other metagenomes available in IMG. The average COG values are shown ±s.d.

Full size image

To further investigate potential bacterial roles in plant-polymer deconstruction, we compared all predicted proteins in the three community metagenomes with the carbohydrate active enzymes (CAZy) database (Cantarel et al., 2009) and identified numerous enzymes potentially involved in this process (Table 4). The largest proportion of the identified proteins were most similar to oligosaccharide-degrading enzymes (176–566 CAZymes, 28–30%), and relatively few were found to be predicted cellulases (4–5 CAZymes, 0.2–0.6%). Compared with other well-known lignocellulose-degrading communities such as the Tammar wallaby foregut (Pope et al., 2010) and termite hindgut (Warnecke et al., 2007), fungus gardens contained relatively fewer cellulases and hemicellulases, but similar numbers of oligosaccharide-degrading enzymes.

Table 4 Partial list of CAZymes identified in the leaf-cutter ant fungus-garden metagenomes, as compared with those found in the termite hindgut and wallaby foregut
Full size table

Comparison of Enterobacter populations

To identify the similarities between the Enterobacter populations across different metagenomes, we performed a fragment recruitment analysis comparing all predicted genes from the Enterobacter FGI 35 phylogenetic bins from each metagenome with the draft Enterobacter FGI 35 genome (Figure 4). The fragment-recruitment analysis identified near-uniform coverage of >95% nucleic acid identity BLAST hits across the 33 Enterobacter FGI 35 contigs, with the exception of four regions between 18–66 kb large that we termed variable regions I–IV. Moreover, we found that there was also near-uniform coverage of 70–85% identity BLASTN hits across the draft genome. Investigation of the coding potential in these conserved regions identified genes required for the synthesis of thiamine, pyridoxine, nicotinate, nicotinamide, pantothenate, folate and 19 amino acids. Only the later stages of the histidine biosynthetic pathway could be identified, although the full pathway is present in other Enterobacter contigs. These regions also encoded ABC transporters and phosphotransferase system components predicted to uptake cellobiose, xylose, glucose, sucrose, β-glucosides, arbutin or salicin, N-acetylmuramic acid, mannitol, mannose, sorbitol, galactitol, L-ascorbate, fructose, ribose, L-arabinose, methyl-galactoside, sulfate, sulfonate, spermidine/putrescine, 2-aminoethylphosphonate, iron and other nutrients. The variable regions were found to contain primarily hypothetical genes and genes of unknown function, although some phage integrases were also identified.

Figure 4
figure 4

Fragment recruitment analysis of genes phylogenetically binned to Enterobacter FGI 35 against the draft Enterobacter FGI 35 genome. Each point indicates the best BLASTN match of a gene. Tick marks on the bottom indicate contig boundaries of the FGI 35 draft genome. Regions showing little or no coverage in the recruitment are marked on the top.

Full size image

Metaproteomics

Individual searches of the metaproteomic data against the predicted protein databases of each community metagenome recovered a total of 1186 redundant and 869 non-redundant peptides. Of all the distinct peptides recovered, 129 were found in both laboratory and field samples, while 351 were unique to the laboratory sample and 389 were unique to the field sample. A total of 747, 238 and 201 peptides were recovered for the searches against the A. cephalotes, A. colombica top and A. colombica bottom datasets, respectively. These peptides were mapped onto a total of 653 proteins, of which 354 were predicted from contigs or singletons that were phylogenetically binned as bacterial (see Supplementary Information for details on the phylogenetic binning procedure). The majority of bacterial proteins identified were predicted to belong to the Enterobacteriaceae, and functions predicted from these proteins included a variety of metabolic processes (Table 5, Supplementary Dataset 5). Figure 5 highlights the overlap observed between laboratory-reared samples and field-collected samples for one peptide mapped to a predicted glycosyl hydrolase. Details for all mass spectra and the annotations for the bacterial proteins they mapped to can be found in Supplementary Datasets 3 and 5, respectively.

Table 5 A subset of bacterial proteins identified in leaf-cutter ant fungus gardens using liquid chromatography-tandem mass spectrometry
Full size table
Figure 5
figure 5

Example of overlap between field-collected and laboratory-reared fungus-garden samples for the glycoside hydrolase family 3 peptide N.AIADLLFGDVNPSGK.L. (a) Theoretical b and y ions, identified m/z values are highlighted in red. (b) Field sample MS/MS spectra; instrument, LTQ Orbitrap (high mass accuracy MS, low mass accuracy MS/MS); parent ion p.p.m. error, 4.03; peptide charge state, +2; retention time, 42.72 min; XCorr, 4.19; MS-GF spectra probability, 1.42 × 10−13. (c) Laboratory-kept sample MS/MS spectra; instrument, LTQ Orbitrap (high mass accuracy MS, low mass accuracy MS/MS); parent ion p.p.m. error, 3.06; peptide charge state, +2; retention time, 33.36 min; XCorr, 4.08; MS-GF spectra probability, 1.67 × 10−12. Manual annotation of additional peptides of interest can be found in Supplementary Dataset 6.

Full size image

Discussion

Leaf-cutter ants are dominant New World herbivores, foraging on up to 17% of the foliar biomass in some ecosystems (Costa et al., 2009). In 1874 Thomas Belt established that leaf-cutters do not consume leaf material directly, as had been previously assumed, but instead use it as manure to cultivate a fungus for food in specialized gardens (Belt, 1874). For over a hundred years after Belt's pioneering discovery it was believed that the fungus gardens of leaf-cutter ants represented a monoculture of the fungal cultivar that degraded plant cell-wall material and converted it into nutrients for the ants (Weber, 1966; Martin and Weber, 1969). However, both the lignocellulolytic capacity of the cultivar and the view that fungus gardens are composed solely of the fungal mutualist have been recently challenged (Gomes De Siqueira et al., 1998; Abril and Bucher, 2002; Scott et al., 2010; Suen et al., 2010). In this study, we explored the hypothesis that bacteria are common constituents of fungus gardens that could be participating in plant biomass degradation and nutrient cycling.

Our work demonstrates that a distinct community of bacteria resides in the fungus gardens of A. colombica and A. cephalotes leaf-cutter ants. Our identification of similar bacterial groups in fungus-garden samples taken from different ant species and garden strata supports this conclusion. Moreover, this is consistent with our finding that relatively few bacterial genera comprise the majority of the metagenomic sequences recovered in this study (see below). This, combined with the previous work on nitrogen fixation, plant biomass degradation and membrane-lipid profiles in these ecosystems, indicates that bacteria are long-term residents of fungus gardens and not merely allochthonous organisms introduced from leaf material or the surrounding soil (Bacci et al., 1995; Pinto-Tomas et al., 2009; Scott et al., 2010; Suen et al., 2010). Thus, the term ‘fungus garden’ may be misleading, as these environments are composed of a fungus–bacteria community.

The bacterial component of the microbial ecosystem in fungus gardens appears to be dominated by only a few groups. Specifically, the genera Enterobacter, Klebsiella, Citrobacter, Escherichia and Pantoea represent over two-thirds of the bacterial component in each of the community metagenomes (Figure 1d). This narrow genus-level diversity is likely the result of both the nutrient composition of the plant–fungal matrix and the meticulous hygienic practices of the ants. For example, leaf-cutters continuously weed their gardens to remove areas infected with microbial pathogens (Currie and Stuart, 2001), and also apply antimicrobials derived from both glandular secretions and symbiotic actinobacteria (Currie et al., 1999; Fernández-Marín et al., 2006). The extent of plant biomass degradation could also affect microbial diversity, but if this was a critical factor we would expect to find distinct communities between top and bottom garden strata, which contain fresh leaf material and largely degraded biomass, respectively. The similarity between different strata observed here, consistent with previous work reporting little difference between 16S libraries constructed from these two regions (Suen et al., 2010), indicates that the extent of plant biomass degradation is not a major contributor to community structuring. The consistent presence of bacterial groups within the Enterobacteriaceae throughout different garden strata and leaf-cutter ant species implicates them as having a consistent role in fungus gardens, and suggests that these environments represent highly structured communities rather than a random collection of opportunistic microbes. Although it remains a possibility that while removing the fungal matrix and plant debris from fungus gardens our analysis excluded microbial groups adhering to fungal or plant biomass, thereby skewing the composition of the metagenomes, our results are generally consistent with previous culture-independent investigations that either analyzed whole fungus gardens or utilized different methods to isolate bacterial cells (Scott et al., 2010; Suen et al., 2010). Moreover, our processing of fresh rather than frozen fungus-garden material may be partially responsible for our success in removing fungal or plant debris from our samples.

Bacteria of the genus Enterobacter appear to be particularly prevalent in fungus gardens. In contrast to the narrow genus-level diversity observed in these environments, multiple species of Enterobacter appear to be present in all the gardens analyzed. Our fragment-recruitment analysis demonstrates that populations of bacteria with >95% and 70–85% nucleic acid identity to the reference Enterobacter FGI 35 genome exist in these environments (Figure 4). The four large gaps identified in the recruitment plot likely represent prophage or other variable elements in the reference genome. Because Enterobacter FGI 35 was isolated from an A. colombica fungus garden, the near-uniform coverage of genes at >95% identity across all metagenomes indicates that highly similar strains of Enterobacter are present in all of the samples analyzed. The near-uniform coverage of genes at 70–85% identity likely represents multiple distinct species, as it is improbable that genes from a single population of bacteria would have such a large range of nucleotide identity to a single reference genome. Genes 70–85% identical to the Enterobacter FGI 35 genome may represent divergent Enterobacter species or even novel Enterobacteriaceae for which an appropriate reference for phylogenetic binning does not exist. That different species of leaf-cutter ant harbor abundant Enterobacter populations indicates that this group may be an important constituent of the fungus-garden community.

The overall functional potential of the metagenomes includes a diversity of bacterial genes associated with plant biomass degradation, supporting previous work that has suggested a role for bacteria in this process. The vast majority of CAZymes identified in the metagenomes are associated with oligosaccharide degradation or simple sugar metabolism, suggesting that bacteria are processing partially degraded plant material. We also found KEGG pathways involved in hexose and pentose sugar metabolism to be highly represented in the Enterobacteriaceae, indicating that sugar monomers can be readily metabolized by many of these bacteria. Moreover, our KEGG, COG and metaproteomic analyses recovered numerous sugar transporters (Figure 2, Table 5, Supplementary Dataset 2), including a large number of cellobiose-specific phosphotransferase system components that are known to be involved in the uptake of the by-products of cellulose hydrolysis (Figure 1, Table 5, Supplementary Dataset 2). Together, these data suggest that bacterial community members are metabolizing predominantly partially degraded plant material, although it remains a possibility that unidentified bacterial lignocellulases also have a role in the degradation of more recalcitrant biomass.

Bacterial lineages in fungus gardens were also found to possess diverse biosynthetic pathways. Pathways involved in amino-acid and B-vitamin metabolism were particularly well-represented in the detected Enterobacteriaceae and Pseudomonas sequences, and biosynthetic pathways for thiamin, pyridoxine, nicotinate, nicotinamide, pantothenate, folate, and all 20 amino acids could be reconstructed from the Enterobacter bins. As mentioned above, enzymes involved in the metabolism of oligosaccharides and simple sugars were also identified in many of these groups, indicating that they may convert carbon-rich plant biomass into amino acids, B-vitamins, proteins or other nutrients. Previous work has indicated that bacteria have a role in the introduction and cycling of nitrogen in fungus gardens (Pinto-Tomas et al., 2009). Together with our work, this suggests that the combined metabolism of resident bacteria may enrich the nutrient composition of fungus gardens through the conversion of carbohydrate-rich oligosaccharides into a variety of other nutrients that could promote the growth of the fungal cultivar or even nourish the ants themselves.

Our metaproteomic analysis recovered peptides mapping to bacterial proteins predicted to participate in biomass degradation and nutrient biosynthesis, supporting the results of our metagenomic characterization and further indicating that bacteria are involved in these processes (Table 5, Supplementary Datasets 3, 5, and 6). Our manual inspection of the metaproteomic data identified multiple peptides belonging to glycoside hydrolases, sugar transporters and amino acid and B-vitamin biosynthetic pathways. That multiple peptides could be assigned to proteins with similar predicted functions indicates that these processes may be prevalent in fungus gardens. Moreover, many of the mapped peptides originated from both laboratory-reared and field-collected samples, including one that belonged to a family 3 glycosyl hydrolase (Figure 5). Although these data should be interpreted cautiously due to the few bacterial proteins identified overall, this may indicate physiological similarities between bacteria in laboratory-reared versus field-collected colonies.

Not all bacteria in fungus gardens were found to have substantial biosynthetic capacity, and in particular the Lactococcus groups appeared to have limited coding potential in the majority of pathways analyzed. This may be a result of lower sequencing coverage, as only a relatively small fraction of the metagenomes was predicted to belong to these groups. Alternatively, these groups may not be contributing substantially to nutrient cycling and are able to subsist on free sugars and other nutrients available in fungus gardens. Importantly, the by-products of Lactococci metabolism may acidify fungus gardens and contribute to the maintenance of the lower pH in these ecosystems, which has previously been observed at 4.4–5.0 (Powell and Stradling, 1986). Regulation of the pH of fungus gardens to this narrow range has been hypothesized to be critical to the growth of fungal cultivar, but the mechanism through which this occurs has remained unknown (Powell and Stradling, 1986). Few peptides from our metaproteomic data sets were recovered from this group, indicating that they may be present in low abundance.

In addition to bacteria, we also found that fungus gardens contain substantial populations of bacteriophage (Figure 1). These organisms could play key roles by limiting bacterial abundance or decreasing ecosystem productivity. Moreover, because fungus gardens contain numerous closely related genera in the Enterobacteriaceae, bacteriophage could provide a common mechanism for gene transfer between lineages. The presence of bacteriophage in fungus gardens adds to the number of organisms that are shaping these ecosystems and introduces a new layer of complexity into the ecology of fungus gardens.

Metagenomics and metaproteomics have previously been shown to be invaluable tools for analyzing microbial communities (Ram et al., 2005; Gill et al., 2006; Woyke et al., 2006; Kalyuzhnaya et al., 2008; Wilmes et al., 2008; Allgaier et al., 2009; Verberkmoes et al., 2009; Burnum et al., 2011), including those associated with herbivores (Warnecke et al., 2007; Brulc et al., 2009; Pope et al., 2010; Burnum et al., 2011). Here we use these techniques to provide insight into the fungus gardens of leaf-cutter ants. Our work shows that relatively few genera dominate the bacterial fraction of these communities, and that the genus Enterobacter appears to be particularly prevalent. We show that bacteria have diverse metabolic potential associated with the degradation of plant biomass, and we confirm the production of two bacterial glycoside hydrolases in situ. Moreover, we show that bacteria in fungus gardens likely participate in the biosynthesis of amino acids, B-vitamins and other nutrients that potentially enhance the growth or biomass-processing efficiency of the fungal cultivar. This is consistent with a model of synergistic biomass degradation by a fungus–bacteria consortium. Our work enhances our knowledge of how leaf-cutter ants process massive quantities of plant biomass in their ancillary digestive systems, and underscores the importance of symbiotic communities on the evolution and ecology of herbivores.