Abstract
Insects and plants engage in a multitude of complex interactions. In antagonistic cases, such as herbivory, insects often specialize on a few closely related plant species to overcome physical and chemical defences. More rarely, herbivorous insects can feed on a range of plant species. Leaf-cutter ants are generalist herbivores that forage from a variety of plant species, which the ants bring to the fungus they farm, Leucoagaricus sp. While we show that anti-herbivory plant compounds can harm Leucoagaricus sp. in vitro, it is unknown how the ants’ fungus gardens are able to incorporate a large diversity of plants with differing plant chemistry. Here, we investigate the fungus garden bacterial community’s ability to degrade plant secondary compounds. We cultured fungus garden bacteria, sequenced the genomes of 42 isolates, and found genes involved in plant secondary compound degradation, including monoterpene epsilon-lactone hydrolase. Some of these genes show in situ expression in metatranscriptomes, such as limonene-1,2-monooxygenase. A majority of the bacterial isolates grew unhindered in the presence of plant secondary compounds and, using GC-MS, isolates from the genera Pseudomonas, Klebsiella, Enterobacter, and Bacillus could degrade either α-pinene, β-caryophyllene or linalool.
Additionally, using a headspace sampler, sub-colonies of fungus gardens reduced α-pinene and linalool over a 36-hour period, while Leucoagaricus sp. strains alone only reduced linalool, not α-pinene. Our study provides evidence that Leucoagaricus sp. has a variable ability to tolerate and degrade plant secondary compounds, indicating that it may depend on bacteria to detoxify the diversity of plant chemistry the system encounters.
Introduction
The struggle between plants trying to avoid consumption and organisms trying to exploit plants for energy define herbivorous interactions and provide an arena for antagonistic coevolution1–4. Herbivorous insects, one of the most well-studied and largest taxa of herbivores, are thought to have had major impacts on shaping the diversity of the plant world5,6 and vice versa2,7. Plants have developed physical defences, such as trichomes, surface waxes, or leaf toughness, which can hinder herbivory. In response, herbivorous insects have morphological and physiological adaptations for consuming plant matter, such as claws for gripping trichomes8, producing adhesive material to attach to slippery wax layers9, and sclerotized mandibles and enlarged head musculature10 for feeding on tough plant material.
Plants also produce an extraordinarily diverse set of plant secondary compounds (PSC), some of which are used to deter herbivores11. Not only are the chemicals diverse in structure and toxicity, the mechanisms behind the release of plant chemicals are complex, depending on the type of damage, the taxon inflicting the damage, the species of plant, and environmental conditions (e.g. hours of sunlight, temperature, moisture). Additionally, the synthesis, storage, and release of these compounds are variable, which can affect the rapidity and severity of a plant’s response to herbivory. Some chemical responses are constitutively present, with toxins being slowly released through open stomata, leaf cuticles, and gland walls. Other chemicals are sequestered in specialized structures and released upon herbivore damage12. Many plant species produce toxins, like hydrogen cyanide, where the enzyme and substrate involved are compartmentalized separately. Upon wounding, the two components mix and hydrogen cyanide is released13. Finally, chemicals can be released as a direct response to being induced by herbivore or pathogen attack14. Herbivorous insects engage in multiple strategies to overcome chemical plant defences, including harbouring gut microbiota that aid in the detoxification of PSC. Microbes associated with mountain pine beetles15–17, red turpentine beetles16, pine weevils18, gypsy moths19, apple maggot flies20, termites21, and coffee berry borers22 have been found to play an important role in PSC degradation. The role of gut microbes in PSC detoxification indicates that microbes have the potential to determine the diet range of herbivorous insects23.
Leaf-cutter ants are dominant herbivores in most of the Neotropical ecosystems and are able to forage from a diverse array of plants. In a long-term study, active colonies of two species of leaf-cutter ants, Atta colombica and Atta cephalotes, were observed cutting leaves from 67–77% of all plant species recorded in the foraging area24. In a year-long study by Wirth and colleagues25, one colony of A. colombica foraged from the leaves or flowers of 126 species, representing 91 genera and 52 families. In mature colonies of Atta sp., the ants forage ravenously, forming massive foraging columns that create distinctive trails over time (Figure 1A). The leaf-cutter ants do not directly consume the plant substrate, but rather use it to feed Leucoagaricus sp. (Agaricales: Agaricaceae), the obligate fungal mutualist of the ants. In return, Leucoagaricus sp. degrades the leaf substrate and serves as food for leaf-cutter ants, providing usable energy and nutrients in the form of specialized hyphal swellings known as gongylidia26,27. This process occurs in structures known as fungus gardens (Figure 1B), which are maintained in underground chambers. Mature colonies of the genus Atta can be composed of hundreds of fungus garden chambers, supporting the nutrition of millions of larvae, pupae, and emerged workers.
The fungus garden, which functionally serves as the ants’ external gut, includes the main fungal symbiont Leucoagaricus sp. and a diverse and abundant community of bacteria. In contrast, the internal gut of leaf-cutter ants has a reduced bacterial community, with adult worker guts containing primarily Wolbachia or Mollicutes28,29. Culturing, scanning electron microscopy30 (Figure 1C), and metagenomics of fungus gardens demonstrate a consistent presence of garden bacteria and have established a core bacterial community that mostly consists of Proteobacteria, the majority being in the class Gammaproteobacteria31–36. While there is variation in the bacterial diversity and abundance between colonies and geographic regions, certain bacterial genera are consistently present and readily cultured, such as Enterobacter, Pseudomonas, Pantoea, Klebsiella, and Burkholderia. With the exception of Klebsiella and Pantoea, which fix nitrogen for the system37, the role of the fungal garden’s bacterial community remains a mystery. However, metagenomic studies have yielded clues, suggesting several potential roles, including a likely lead in PSC detoxification36.
In this study, we examine the ability of garden bacteria from leaf-cutter ants to metabolize PSC. Focusing on bacteria isolated from fungus gardens of fungus-growing ants collected in Brazil, we investigate the potential of garden bacteria to tolerate and degrade PSC. First, we isolate strains of Leucoagaricus sp. and determine susceptibility to eight PSC. Then, we sequence the genomes of 42 isolates of garden bacteria and predict the presence of genes involved in PSC degradation. Additionally, we analyse previously generated garden bacteria metagenomes, Leucoagaricus sp. genomes, and metatranscriptomes to investigate the presence and expression of genes involved in PSC degradation. We also expose garden bacteria to eight PSC and determine susceptibility. Then, using Gas Chromatography-Mass Spectrometry (GC-MS), we quantify the in vitro ability of 15 isolates of garden bacteria to degrade four PSC. Finally, we measure reduction of two PSC by fungus gardens from our laboratory colonies of Atta cephalotes using a headspace sampler coupled to a GC.
Results and Discussion
Sensitivity of Leucoagaricus strains to PSC
Fungi generally have extensive degradative capabilities38 and Leucoagaricus sp. is known to produce laccases, which are used to treat incoming plant material and degrade plant phenolic compounds39. However, plant phenolics represent one class of plant chemicals so we tested if Leucoagaricus sp. could tolerate eight PSC: α-pinene, β-caryophyllene, eucalyptol, farnesol, limonene, linalool, p-cymene, and terpinolene. These compounds have been detected in fungus gardens collected in the field36 and/or are known to be present in plants observed being foraged by leaf-cutter ants. We focused mostly on monoterpenes because terpenes often take a prominent place in the profile of blends released by damaged plants40. Five different strains of Leucoagaricus sp. from Atta sexdens, Atta laevigata, Atta bisphaerica, Atta capiguara, and Paratrachymyrmex diversus colonies were tested for their ability to grow in the presence of eight different PSC (Figure 2A/B). Of note, while P. diversus is not within the leaf-cutter ant lineage and largely collects substrate like seeds, insect frass, and dry plant debris for its garden, the species has been observed occasionally collecting fresh leaf and flower material as substrate for its fungus garden41,42. The Leucoagaricus sp. from P. diversus was the most generally inhibited, with complete growth inhibition from terpinolene, eucalyptol, linalool, and p-cymene, and high inhibition from α-pinene and limonene. The Leucoagaricus sp. from P. diversus and A. laevigata colonies were also the only Leucoagaricus sp. that were inhibited by β-caryophyllene, a sesquiterpene. The Leucoagaricus sp. from an A. sexdens colony also exhibited high sensitivity to PSC, with complete inhibition occurring from limonene, terpinolene, eucalyptol, and linalool. The Leucoagaricus sp. from an A. capiguara colony was the most resistant to the PSC tested, only exhibiting high inhibition in the presence of linalool, while exhibiting low to no inhibition in the presence of the remaining seven compounds (Figure 2B).
Multiple studies have investigated the toxicity of various plant compounds to the fungal cultivars of leaf-cutter ants43–46. LaPointe et al.45 isolated Leucoagaricus sp. from Acromyrmex landolti (a species in the other genus of leaf-cutter ants) and A. laevigata and found that growth of the isolates was reduced when grown on agar with aqueous plant extract or leaf homogenate compared to growth of the fungus on agar alone. In another study by Howard et al.46, four terpenoids were tested for potential toxic effects against Leucoagaricus sp. from A. cephalotes and found that the four compounds had variable effects on fungal growth with certain compounds having little/no effect and others inhibiting the fungus greatly. Our results, combined with previous studies, suggest that the fungal cultivars of fungus-growing ants have varying levels of resistance to PSC, indicating that other microbes may be involved in detoxification.
Gene content of bacterial isolates, garden bacteria metagenomes, garden metatranscriptomes, and Leucoagaricus sp
As the fungus garden is the external gut of leaf-cutter ants, the fungus gardens’ microbiome likely fills roles similar to other gut microbiomes, including helping detoxify PSC. Taking advantage of the external and aerobic nature of the leaf-cutter ant fungal “gut”, we cultured and sequenced bacteria from fungus gardens and tested the potential of isolates to degrade PSC. Garden bacteria were isolated from multiple fungus gardens of the two genera of leaf-cutter ants, Atta sp. and Acromyrmex sp., as well as three genera of other fungus-growing ants, Paratrachymyrmex sp., Cyphomyrmex sp., and Apterostigma sp., collected in Brazil (Table S1). After obtaining pure isolates, 42 bacterial genomes were sequenced (Table S2), including genera from both abundant and consistent genera (e.g., Pseudomonas, Enterobacter), as well as more sparse genera (e.g., Chitinophaga, Bacillus). To assess the in silico potential of the 42 isolates for PSC degradation, we analysed the genomes (Figure 3A) for the presence of the diterpene degradation pathway47, the cymene pathway (ko1220:module 00419), the saxA gene, the α-pinene/limonene degradation pathway (ko00903), the geraniol pathway (ko00281), the cumate pathway (ko01220: module 00539), the trans-cinnamate pathway (ko1220:module 00545), 20 cytochrome p450s known to be involved in the microbial transformation of PSC48, and the squalene hopene cyclase gene (shc) (Dataset S1, Dataset S2). In addition to the 42 isolate genomes, we used the same genes and pathways to analyse 12 existing fungus garden bacteria metagenomes from Brazil (Dataset S1, Dataset S2), metatranscriptome data generated from the top portions of field-collected fungus gardens from two Atta cephalotes colonies and one Atta colombica colony (Table S3, Table S4), and two published genomes of Leucoagaricus sp. strains (Dataset S1, Dataset S2).
No genes belonging to the diterpene degradation cluster, the cymene pathway, or saxA were found in the 42 bacterial isolates analysed (Dataset S1, Dataset S2). However, the metagenome annotation did predict the presence of the cymene degradation pathway: one metagenome had the complete pathway, two metagenomes had ¾ of the pathway, five metagenomes had one gene from the pathway, and the remaining four had none of the genes (Dataset S2).
The α-pinene/limonene degradation pathway and geraniol pathway shared three genes between the two pathways: K01692 (enoyl-CoA hydratase), K01825 (fadB, 3-hydroxyacyl-CoA dehydrogenase), and K01782 (fadJ, 3-hydroxyacyl-CoA dehydrogenase). All isolates had at least one out of three shared genes, while one Pantoea, Klebsiella, Enterobacter, and Acinetobacter had all three. For the unique genes in the pathways (9 genes out of 12 for α-pinene/limonene and 12 genes out of 15 for geraniol), four of seven Enterobacter isolates had the highest proportion of unique α-pinene/limonene genes at 22%. All Burkholderia isolates, the Acinetobacter isolate, four of seven Enterobacter isolates, one of four Klebsiella isolates, and one of 10 Pantoea isolates were predicted to contain the monoterpene epsilon-lactone hydrolase gene (K14731), which is involved in monocyclic monoterpene degradation49. Most isolates also contained an aldehyde dehydrogenase (K00128) involved in the α-pinene/limonene degradation pathway (Figure 3A). In the 12 metagenomes, the proportion of genes present from the α-pinene/limonene pathway ranged from 22% to 56% (Dataset S2). Additionally, we detected the expression of certain genes in the α-pinene/limonene pathway in the metatranscriptomic dataset (Table S4, Figure S2A). In the first A. cephalotes colony (FG1), we detected 33% of the α-pinene/limonene pathway in a range of 0.506-2831 transcripts per million (TPM), and all three shared genes in a range of 0.337-3.051 TPM. Of note, we detected limonene 1,2-monoxygenase (K14733) expression at 0.506 TPM, which is the first step of limonene transformation. In the second A. cephalotes colony (FG2), we detected 22% of the α-pinene/limonene pathway with 59.99 TPM and 3724 TPM, and all three shared genes in a range of 0.168-11.54 TPM. In the A. colombica colony (FG3), we detected 22% α-pinene/limonene pathway with 9.062 TPM and 2238 TPM, and one of the shared genes with 4.178 TPM. In all three metatranscriptomes monoterpene epsilon-lactone hydrolase (K14731) was expressed, reflecting its presence in the individual bacterial isolate genomes and confirming the gene’s expression in situ.
For the unique genes in the geraniol pathway, Acinetobacter had the highest proportion of genes at 75%, while most other isolates were predicted to have between 8% and 17% of genes (Figure 3A). Acinetobacter contained all the genes except for the geraniol dehydrogenase (K19653, K17832) and one of the 3-hydroxyacyl-CoA dehydrogenases (K00022). The other isolates were predicted to have hydroxymethylglutaryl-CoA lyase and acetyl-CoA acyltransferase, in addition to the shared genes described above. In the metagenomic dataset, the proportion of completeness for the geraniol degradation pathway was between 17% and 75% (Dataset S2). In the metatranscriptomic dataset (Table S4, Figure S2A), we found that FG1 expressed 50% of the geraniol pathway in a range of 0.115-25.47 TPM, including citronellol dehydrogenase (K13774) at 0.147 TPM, which is involved in the first step of geraniol and citronellol transformation. In FG2, we detected 42% of the geraniol pathway in a range of 0.033-27.18 TPM and in FG3, we detected 16% of the geraniol pathway with 1.334 TPM and 9.208 TPM.
Ten of the isolates contained one gene from the cumate pathway, with seven Burkholderia and two Pseudomonas isolates predicted to contain different components of the p-cumate 2,3-dioxygenase enzyme (K16303 or K16304) and one Pantoea isolate predicted to have 2,3-dihydroxy-p-cumate/2,3-dihydroxybenzoate 3,4-dioxygenase (K10621). On the other hand, more isolates had genes involved in the trans-cinnamate transformation pathway. Ten of twelve Burkholderia, all Klebsiella, and one of three Pseudomonas isolates had at least 50% of the genes necessary for cinnamate degradation (Figure 3A). Not all of the genes are necessary for a complete pathway, as there are two routes from trans-cinnamate to trans-2,3-dihydroxy-cinnamate. In the metagenomes, cumate and trans-cinnamate pathways had a range of 0%-100% and 50%-80% completeness, respectively (Dataset S2). We did not detect the expression of genes from these pathways in the metatranscriptomic dataset.
Twenty cytochrome p450s known to be involved in isoprenoid transformation were analyzed48. Most of the cytochrome p450s were not detected in the 42 bacterial isolates tested (Dataset S1). However, four cytochrome p450s were predicted to be present: CYP102A1, CYP106A2, CYP107H, and CYP108. CYP102A1 was detected in four out of 12 Burkholderia isolates, but none of the other genera (Figure 3A). CYP102A1 functions as a fatty acid hydroxylase and small mutations can result in the catalysis of oxidation of substances such as terpenes50. CYP106A2 was detected in Bacillus (ICBG1751) and is known to be a steroid hydroxylase, as well as a bacterial diterpene hydroxylase51. CYP107H and CYP108 were both detected in one Pantoea isolate (ICBG870). CYP107H is involved in a pathway that cleaves fatty acyl chains for biotin biosynthesis, while CYP108 is involved in α-terpineol hydroxylation52. We saw an increase in the amount of cytochrome p450s predicted to be present in the metagenomes, including CYP111, which catalyses the 8-methyl hydroxylation of linalool53. The other cytochrome p450s detected only in the metagenomes were CYP105A3, CYP101, CYP105A1 (Dataset S1), responsible for hydroxylation of compactin54, oxidation of camphor55, and hydroxylation of vitamin D356 and diterpene resin acids7, respectively. We also detected the cytochrome p450s described above in the individual bacterial genomes: CYP102A1, CYP107H, CYP106A2, and CYP108 (Dataset S1). We did not detect the expression of these 20 cytochrome p450s in the metatranscriptomic dataset.
Squalene-hopene cyclase (shc) is responsible for synthesizing hopanoids, which integrate into biological membranes and increase structural order, resulting in a reduction of permeability and an increase in stability of baterial membranes58. The presence of this enzyme in garden bacteria isolates could explain the ability to tolerate stressful conditions, such as growth in the presence of PSC. All Burkholderia isolates and the Asaia isolate are predicted to have squalene-hopene cyclase, while none of the other isolates were predicted to contain this gene (Figure 3A). shc was detected in 11/12 metagenomes (Dataset S1).
In addition to the garden bacteria genomes, metagenomes, and metatranscriptomes, we analyzed two existing Leucoagaricus sp. for the same PSC degradation pathways. Overall, the two genomes, one from Atta cephalotes (Leucoagaricus sp. [Ac12]) and one from Cyphomyrmex costatus (Leucoagaricus sp. [SymC.cos]), lacked most of the gene sets analyzed (Dataset S1, Dataset S2). This could be partially due to most of the genes used for annotation being of bacterial origin and so the annotation did not lend itself well to fungal genomes.
The presence of cytochrome p450s, α-pinene/limonene degradation genes, geraniol degradation genes, cumate degradation genes, trans-cinnamate degradation genes, p-cymene degradation genes, and squalene-hopene cyclase indicate that garden bacteria may be able to metabolize PSC, or at least survive in the presence of PSC. With the data available we were able to predict that both individual garden bacteria isolates and garden bacteria metagenomes contained the genes necessary to degrade or transform PSC that could harm the fungus gardens. Additionally, the higher completeness of pathways observed in the metagenomes suggests that while individual strains may not have the entire pathway for the degradation of a compound, as a community, garden bacteria have the capabilities to reduce PSC within the fungus garden. The metatranscriptomic dataset supports this hypothesis, as we found expression of genes involved in α-pinene/limonene and geraniol degradation. These metatranscriptomes came directly from field gardens, indicating that bacteria in fungus gardens in the natural environment are actively expressing genes involved in PSC/monoterpene degradation. In combination with the lack of genes predicted in Leucoagaricus sp. genomes within these pathways, we infer that fungus garden bacteria could be involved in detoxifying PSC based on their genomic content.
Bacterial tolerance of PSC
In addition to in silico analyses of the garden bacteria isolates, we assessed their tolerance by exposing garden bacteria isolates to the eight PSC described above (Figure 3B). Most bacterial isolates were able to grow uninhibited in the presence of a high concentration of the different compounds. Linalool was the most inhibitory against the bacterial isolates, causing some degree of inhibition against all isolates except Pseudomonas and three Burkholderia isolates. While none of the isolates were predicted to contain genes involved in the degradation pathway, all isolates except for Acinetobacter and Bacillus were completely resistant to p-cymene, perhaps due to an alternative to degradation, such as efflux pumps. Farnesol, β-caryophyllene, and terpinolene did not inhibit the majority of bacterial isolates, causing small zones of inhibition in the Bacillus isolate as well as one to two other isolates. α-pinene and limonene caused slightly more inhibition, especially in Bacillus and two Klebsiella isolates, but most bacteria were resistant or only slightly susceptible to these two compounds. These trends were also seen in a larger set of samples (Figure S1). Like the tolerance of Leucoagaricus sp., bacterial isolates have varying susceptibilities to PSC.
in vitro degradation of PSC by bacterial isolates
Fifteen bacterial isolates that represent the diversity of garden bacteria across multiple fungus-farming ant lineages, as well as mostly being from genera that are consistently abundant between fungus gardens, were grown in liquid culture supplemented with one of four PSC (α-pinene, β-caryophyllene, eucalyptol or linalool) and degradation was assessed with GC-MS. To account for isolates unable to grow in the presence of certain PSC, we grew all bacterial isolates in two ways: (A) adding the compound during exponential phase (Figure 4) and (B) adding the compound after stationary phase (Figure S3). All Enterobacter isolates (ICBG810, p=0.0004; ICBG643, p=0.0018; ICBG832, p=0.0062) significantly reduced α-pinene (t-test, Bonferroni correction: α=.0033) during exponential growth. Also, one of two Klebsiella isolates (ICBG873, p=0.0026) and one of two Bacillus isolates (ICBG1751, p=0.0008) significantly reduced α-pinene. Additionally, other bacterial isolates showed a varying range of reductions of α-pinene between vials, resulting in large variability and lack of significance with the Bonferroni corrected alpha-value. No isolates significantly reduced α-pinene within stationary growth. One of three Pseudomonas isolates (ICBG639, p<0.0001) significantly reduced β-caryophyllene in the exponential environment. The same Pseudomonas isolate (ICBG639, p=0.0014) also significantly reduced β-caryophyllene at the stationary phase. Linalool was reduced by two isolates, Burkholderia (ICBG637; exponential p=0.0036, stationary p=0.0008) and Pseudomonas (ICBG967; exponential p=0.0020, stationary p=0.0024). Finally, eucalyptol, the fourth compound tested, was not reduced by any of the isolates tested (Figure S4). In all cases, no breakdown products were detected by GC-MS, which could be due to complete degradation of the compounds into components of central metabolism59,60.
Some genera isolated, such as Pseudomonas and Burkholderia, have been implicated in plant compound degradation in other systems, such as in bark beetles and mountain pine beetles16,61. In the current study, two isolates of Pseudomonas were found to significantly reduce either β-caryophyllene or linalool in vitro. In addition to Pseudomonas, in the bark beetle system, Serratia isolates were able to reduce 3-carene and (-)-β-pinene and Rahnella isolates were able to reduce 3-carene, (-)-α-pinene, (+)-α-pinene and (-)-β-pinene16. While Serratia and Rahnella were not cultured in our study, metagenomic and culturing studies from fungus gardens of leaf-cutter ants have detected Serratia (Figure S5) and Rahnella32,33,36, suggesting that other isolates within fungus garden bacterial communities could be involved in metabolizing PSC that simply were not tested in this study.
Fungus gardens exposed to α-pinene or linalool did not experience a shift in bacterial community composition
We exposed sub-colonies of Atta cephalotes to a low and high dose of α-pinene or linalool for 48 hours and then extracted bacterial DNA for 16S rRNA amplicon sequencing to determine if there was a shift in bacterial community, as previously observed for mammalian herbivore gut microbiomes62. At the doses and exposure tested, we did not see a change in abundance of certain expected community members known to degrade PSC (i.e., Pseudomonas), compared to the control sub-colonies (data not shown). However, the bacterial genera observed are the same as genera observed from the fungus gardens of leaf-cutter ants collected and immediately processed in the field31–33,36 (Figure S5). The colonies used in the experiments have been in lab for approximately 7 years (RM120223-02), 5 years (CR14) and 1 year (CF180406-01, CF180405-02, HH180403-03), which indicates that fungus garden bacterial members – at the genus-level – are maintained even with a large change in environment: tropical rainforest in Costa Rica to a controlled lab setting in Wisconsin. Specifically, we confirmed that the genera we cultured directly in the field for our in vitro assays were found in our laboratory colonies. All isolates tested in the GC-MS experiment that had been cultured from the Brazil colonies were detected in the 16S data except for Bacillus (i.e. Burkholderia, Pantoea, Pseudomonas, Enterobacter, and Klebsiella were present at varying abundances (Figure S5B). For our subsequent experiments (described below), this means we can assume that our results are relatable to environmental fungus gardens.
Degradation of α-pinene and linalool by Atta cephalotes fungus gardens
To examine if our in vitro determination of PSC tolerance and reduction reflects an in vivo role in fungus garden and the associated bacteria’s behaviour in the environment, we use sub-colonies of fungus gardens from Atta cephalotes colonies and measured headspace concentrations of two PSC, α-pinene and linalool (i.e., the compounds that were degraded by garden bacteria and largely inhibited Leucoagaricus growth). The headspace of the control vials was compared to the headspace of vials containing fungus gardens and compound. When sub-colonies from three A. cephalotes colonies were exposed to α-pinene (Figure 5A), there was significant reduction of α-pinene in the fungus garden samples at 12 hours (only α-pinene vs fungus garden + α-pinene, p<0.0001; cotton + α-pinene vs fungus garden + α-pinene, p=0.1191), 24 hours (only α-pinene vs fungus garden + α-pinene, p<0.0001; cotton + α-pinene vs fungus garden + α-pinene, p<0.0001) and 36 hours (only α-pinene vs fungus garden + α-pinene, p<0.0001; cotton + α-pinene vs fungus garden + α-pinene, p<0.0001), compared to most control vials (mixed regression model with time and treatment as fixed effects and ant colony as random effects, α = 0.05). Additionally, the 36-hour sub-colonies had significantly reduced α-pinene compared to the 12-hour sub-colonies (p<0.0001) and the 24-hour sub-colonies (p<0.0001), suggesting active metabolizing by the fungus garden over this time period. Leucoagaricus sp. strains were also tested in a similar fashion. Vials containing Leucoagaricus sp. were exposed to α-pinene and the headspace was measured after 36 hours of exposure (Figure 5B). Compared to the control vials, Leucoagaricus sp. strains did not reduce α-pinene significantly (p=0.2786, Welch two sample t-test).
When sub-colonies from five A. cephalotes colonies were exposed to linalool (Figure S6A), there was a significant reduction of linalool in the headspace between controls and 12-hour sub-colonies (only linalool vs fungus garden + linalool, p<0.0001; cotton + linalool vs fungus garden + linalool, p=0.0728), 24-hour sub-colonies (only linalool vs fungus garden + linalool, p<0.0001; cotton + linalool vs fungus garden + linalool, p=.0021), and 36-hour sub-colonies (only linalool vs fungus garden + linalool, p<0.0001; cotton + linalool vs fungus garden + linalool, p=0.0049). However, we do not see any significant differences between the 12-hour, 24-hour, and 36-hour linalool amounts, suggesting there may be a limit to the degradation possible with this compound that was reached at 12 hours post-exposure. For the headspace sampling with Leucoagaricus sp., linalool was significantly reduced (p=0.0036, Welch two sample t-test), which was surprising due to the high inhibition by this compound in the plate assay (Figure 2B, Figure S6B). This could be due to a difference in dosage, which has been shown to have an effect on bacterial tolerance and degradation of PSC17. The tolerance assays contained high amounts of compound, whereas the headspace sampling had lower amounts of compound and Leucoagaricus sp. Was allowed to obtain a higher biomass (grown for longer on PDA). Additionally, while we saw significant reduction of linalool by Leucoagaricus sp., strains from CF180405-02, CF180406-01, and AB1 had linalool amounts that were encompassed by the amount found in the control. Therefore, we conclude that Leucoagaricus sp. has variable abilities to tolerate and degrade common PSC at the strain level. Overall, the observed reductions of PSC indicate that the fungus gardens of leaf-cutter ants have the capability to reduce PSC that enter the system and Leucoagaricus sp. may depend on other microbes to aid in detoxifying these compounds.
Conclusions
In addition to the important morphological and physiological features of insects and plants, both organisms form symbioses with microorganisms to outcompete one another. In herbivorous insect systems, gut bacteria can be imperative in determining the plant range of its host. This study shows that bacteria associated with the fungus gardens of leaf-cutter ants can metabolize PSC, potentially enabling leaf-cutter ants to forage from a greater variety of plant sources. While earlier work with metagenomic sequencing in leaf-cutter ant fungus gardens showed the potential of the bacteria to degrade anti-herbivory compounds, our work combines in silico, in vitro, and in vivo evidence that indicates that garden bacteria can degrade PSC, some of which may otherwise inhibit the growth of Leucoagaricus sp. Like other herbivorous insect systems, the gut microbiome of leaf-cutter ants is demonstrably important for dictating palatable plant substrate, with the unique feature of the ants’ gut being external in the form of fungus gardens. Overall, this study demonstrates the intricacy and nuance with which microbes serve as an interface between herbivores and the plants they consume.
Materials and Methods
Fungal tolerance assay
We selected compounds for testing based on leaf extracts from plant families that have been foraged by leaf-cutter ants25, detection of terpenes in fungus gardens of Atta laevigata36, and commercial availability: 98% (1R)-(+)-α-pinene (Acros Organics), >90% β-caryophyllene (TCI), 99% eucalyptol (Sigma-Aldrich), 95% farnesol(Sigma-Aldrich), 96%(S)-(-)- limonene (Sigma-Aldrich), 97% linalool (48.2% (R)-(-)-linalool/51.8% (S)-(+)-linalool) (Sigma-Aldrich), 99+% p-cymene (Acros Organics), and 85% terpinolene (Sigma-Aldrich). The eight PSC were tested against five strains of fungal cultivar from Atta sexdens, Atta laevigata, Atta bisphaerica, Atta capiguara, and Paratrachymyrmex diversus colonies (isolation information in Table S1). A 6 mm fungal plug of Leucoagaricus sp. was put onto a 60 mm Oxoid Malt Extract Agar (OMEA; per L: 30 g malt extract, 5 g mycological peptone, 15 g agar) plate and allowed to grow for two weeks. Then, for each PSC, the cultivar was exposed to a sterile disc with 15 μL of compound that had been allowed to dry in a biological safety hood for five min. The edges of fungal growth and the edge of the disc were 1 cm apart. Each compound was done in triplicate (three plates per plant defence compound per cultivar) and inhibition was monitored over the course of two weeks, with pictures being taken on day 2 and 14. Inhibition was determined by a qualitative scale where 0 =no inhibition, 1=no/slight inhibition at day 2 and resume normal growth by day 14(compared to control), 2=no/slight inhibition at day 2 and resume slow growth by day 14 (compared to control), 3= mostly inhibited by day 2, no additional growth by day 14, 4=complete inhibition at day 2 and day 14 (Figure 2A).
Sampling and bacterial isolations
Fungus-farming ant colonies were collected in January 2017 in the following general locations: Anavilhanas, AM; Ducke Reserve, AM; Itatiaia, RJ; Botucatu, São Paulo State; Ribeirão Preto, São Paulo State. Details regarding the exact GPS coordinates and environment of the samples can be found in Table S1. In lab, pure isolates were obtained after several rounds of subculturing based on morphology. A total of 317 isolates were obtained. We identified 117 isolates to genus-level by 16S rRNA sequencing as previously described63. Briefly, colonies were lysed and PCR was performed with 16S rRNA primers 27F (5’-GAG AGT TTG ATC CTG GCT CAG-3’) and 1492R (5’-GGT TAC CTT GTT ACG ACTT-3’). Samples were sequenced using Sanger sequencing at the University of Wisconsin – Madison Biotech Center (Madison, WI) and analysed using 4Peaks and CLC Sequence Viewer 7. The 16S rRNA gene sequence were matched using BLAST64 and the SILVA database (https://www.arb-silva.de) for the nearest genus-level identification.
DNA Extraction, Assembly, and Annotation
DNA from 42 bacterial isolates was extracted using the Promega Wizard Genomic DNA Purification Kit using the Gram-Negative and Gram-Positive Bacteria Protocol. Quality control was done using Qubit BR dsDNA kit (Invitrogen, USA). Genomic DNA libraries for Illumina MiSeq 2×300bp paired-end sequencing were prepared by the University of Wisconsin-Madison Biotechnology Center. Reads were corrected with MUSKETv1.1, paired-ends were merged with FLASH v1.2.7, and assembled with SPAdes 3.11.0. Genome statistics can be found in Table S2. Species-level identification was determined by uploading the genomes to JSpeciesSW, performing a Tetra Correlation Search, and taking the first result. If there were conflicts in the top 5 results (i.e. different genera), the top 5 genomes were pulled and ANI with pyani (github.com/widdowquinn/pyani) using the ANIm analysis was performed. Then, the genome with the highest percent similarity was selected as our isolate’s taxonomic status. Predicted proteins putatively involved in PSC degradation were identified from each genome in one of two ways: 1) Using the Kyoto Encyclopedia of Genes and Genomes (KEGG) Automatic Annotation Server (KAAS). Specifically, genes encoding enzymes putatively involved in monoterpene degradation or aromatic compound degradation, as defined in the KEGG limonene and α-pinene degradation pathway (ko00903), geraniol degradation (ko00281), degradation of aromatic compounds (ko01220, modules 00419,00539, 00545) were investigated after annotation. 2) DIAMOND v0.9.21.12265 BLASTP against the Uniprot Swiss-Prot and TrEMBL databases (www.uniprot.org/downloads), downloaded on July 18, 2019. We only kept the top 5% of hits (--top 5) that had an e-value below 1e-05 for each query sequence. Then, using the grep command, we looked for the following Uniprot accession numbers that corresponded to cytochrome p450s: P00183, P14779, Q2L6S8, P18326, Q59079, Q59831, Q06069, P53554, P33006, U5U1Z3, A9F9S4, Q59723, Q59990, Q9K498, Q53W59, Q8VQF6, A9FZ85, Q88LH7, Q88LH5, Q88LI2, Q65A64; the 20 genes in the diterpene degradation cluster: Q9X4W9, Q9X4W8, Q9X4X8, Q9X4X7, Q9X4X6, Q9X4X5, Q9X4X4, Q9X4X2, Q9X4X1, Q9X4X0, Q9X4W7, Q9X4W6, Q7BRJ3, Q7BRJ4, Q7BRJ5, Q7BRJ6, Q7BRJ7, Q7BRJ8, Q7BRJ9, Q9X4X; saxA: A0A0N7FW12; squalene-hopene cyclase: P33990, P54924, P33247. We also performed DIAMOND BLASTP using a custom database with solely these gene sequences with a query coverage cut-off of 75% (--query-cover 75) and e-value cut-off of 1E-05. If there was alignment in both the Uniprot analysis and the custom analysis, then genes were predicted to be present. The annotation methods were used on individual bacterial genomes, two publicly available Leucoagaricus sp. genomes (BioProject: PRJNA179280 and PRJNA295288), and the publicly available leaf-cutter ant garden bacteria metagenomes from Brazil (Gold Analysis Project ID: Ga0157357 - Ga0157368). For the metagenomes, we used the metagenomes KAAS option instead of the complete/draft genome KAAS option.
Metatranscriptomic sequencing of fungus gardens
Samples were collected directly from the field into RNAlater buffer. Samples were taken from the top sections of three different colonies: two Atta cephalotes colonies from La Selva and one Atta colombica colony from Golfito (Table S3). Total RNA extraction was identical to a method previously described66. cDNA library construction and Illumina HiSeq2000 sequencing were performed at the University of Wisconsin Biotechnology Center (Madison, WI). Metatranscriptomes were uploaded to MG-RAST and processed with their SOP67. We downloaded the reads post-processing (quality reads) and then metatranscriptomes were analysed using prodigal V2.6.2, DIAMOND v0.9.21.122, and kallisto v. 0.43.168. First, we ran prodigal on the assembled fna files of the garden bacteria metagenomes (downloaded from JGI) with the metagenomic flag (-p meta). Then, a kallisto index was created with all of the combined prodigal garden bacteria metagenome output. The kallisto quant command was used to pseudo-align the garden bacteria index against the metatranscriptome reads. This gave a transcripts per million (TPM) value of bacterial transcripts in the metatranscriptome. Then, we used DIAMOND to blastp the metagenome coding regions against the Uniprot and KEGG databases described above. Using grep, we found the genes of interest (same as in the bacteria isolate annotation) and connected the gene of interest, the metagenomic transcript it mapped to, and the TPM in the metatranscriptome. For genes with multiple transcripts and different TPMs, we recorded the unique values (Figure S2) and summed the TPMs for total expression (Table S4). We also did the same workflow for four housekeeping genes (gyrB: K02470, rpoB: K03043, rpoD/sigA: K02086, rpsL: K02950).
in vitro garden bacterial tolerance
We tested the effect of the eight PSC on 46 bacterial isolates using Whatman 6mm discs. Bacterial isolates were grown overnight (16-24 hours) until turbid (OD600 ∼1-2). 100μL of overnight culture was spread and plated on Yeast Malt Extract Agar (YMEA; per L: 4 g yeast extract, 10 g malt extract, 4 g dextrose, 15 g agar). A disc with 15μL of PSC was deposited in the center of the bacterial lawn. Each PSC was done in triplicate (3 plates per plate secondary compound per bacterial isolate). After 48 hours, pictures were taken of the plates using an Epson scanner and then uploaded into Fiji69. Fiji Version 1.0 was used to measure the zones of inhibition caused by each PSC (in centimetres). The average of the three zones of inhibition was calculated. The zones of inhibition were all scaled in reference to the largest zone observed so that 0 indicates inhibition (zone of inhibition = 3cm) and 1 indicates no inhibition (100% growth; zone of inhibition = 0cm)
Gas Chromatography-Mass Spectrometry (GC-MS) of bacterial isolates incubated with PSC
Bacterial isolates were prepared two ways for GC-MS: addition of compound during exponential growth or stationary growth. For both methods, bacterial isolates were grown overnight (16-24 hours) in 10% tryptic soy broth (TSB). All shaking was done at 300 rpm at room temperature. All experiments included an extra vial to read the OD600 to ensure bacterial growth in the presence of compound, which also served to confirm earlier patterns of compound tolerance by bacterial isolates (data not shown). (A) Exponential phase The OD600 was taken to calculate how much overnight culture would be needed to reach an OD600 of 0.08 (i.e. C1V1= (0.08)(1 mL)). The appropriate amount of overnight culture was inoculated into vials containing 10% TSB and 2.5 μL/mL of one of four PSC, α-pinene, β-caryophyllene, eucalyptol, linalool, that were added using a glass manual GC syringe (10 μL, Thermo Scientific). The bacterial cells and plant compound were left shaking for another two days at room temperature. This was done with 15 bacterial isolates representing six genera, as well as negative controls (no bacteria), in triplicate (16 × 4 compounds × 3 replicates). (B) Stationary phase 10 μL of overnight culture was pipetted into 987.5 μL of 10% TSB in vials. After two days of incubation at room temperature while shaking at 300 rpm, 2.5 μL/mL of one of four PSC (α-pinene, β-caryophyllene, eucalyptol, linalool) were added directly to the vials using a glass manual GC syringe (10 μL, Thermo Scientific). Then, the bacterial cells and plant compound were left shaking for another two days at room temperature. This was done with the same number of samples listed above.
For both method A and B, PSC were extracted by pipetting 1mL of hexane into each vial and shaking the vials overnight. After the phase were allowed to separate, 500 μL of the hexane-PSC phase was removed and put into new vials containing 500 μL hexane and 5 μL/mL toluene (as the internal standard). We then analysed the abundance of each PSC using GC-MS. Specifically, the GC system consisted of a Thermo Fisher Trace 1310 Gas Chromatograph coupled with Thermo ISQ LT Single Quadrupole Spectrometer. We injected 1 μL of each mono-/sesqui-terpene sample directly, with a split flow ratio of 30:1. We used an oven profile of 40°C, followed by a ramp of 3°C min-1 to 115°C (monoterpenes) or 130°C (sesquiterpenes) and then 30°C min-1 to 250°C with a 2 min hold. Peaks were integrated and analysed using the Chromeoleon Chromatography Data System Software.
Signal peaks from the GC were integrated and standardized based on the toluene internal standard (peak area/internal standard peak area) for each vial. Additionally, we used standard curves of the four pure PSC to measure changes in concentration in the samples when compared to controls. Standard curves were made to incorporate the possible ranges of concentrations (0 μL/mL to 3.5 μL/mL) within the experiment. Proportional change of bacterial treatments versus the nonbacterial control was then calculated. Specifically, we took the average of the non-bacterial control standardized peak areas and subtracted the control average from all the bacteria-compound peak areas. Then, we divided the adjusted value by the non-bacterial control average to obtain the percent change [(bacterial standardized peak area – average of control standardized peak areas)/average of control standardized peak area]. We then analysed the standardized values in JMP Pro 13 by performing one-sampled Student’s t-tests for each compound with a null hypothesis of μ=0, representing no change between compound abundance in bacteria-treated and the non-bacterial control. Since we were performing 15 separate statistical tests for each compound (between non-bacterial control and each of the 15 bacterial isolates), we used a Bonferroni correction to avoid false positives (α=.05/15=.0033).
Bacterial community changes when fungus gardens are exposed to PSC
Two different doses of compound were used in this experiment: 5 μL in 5.08 cm flame-sealed 50 μL capillaries and 25 μL in 5.08 cm flame-sealed 100 μL capillaries. For each dose there were three treatment groups: (A) control (no compound) (B) α-pinene and (C) linalool. For each treatment, approximately 5 g of the top-middle layer of fungus garden were taken from the same five Atta cephalotes colonies used in the headspace experiment (5 colonies × 3 treatments × 2 doses = 30 sub-colonies). The fungus garden piece was placed in a 60 mm glass petri dish to ensure stability and to make final collection easier. Then, the fungus garden piece was placed into a 12 oz glass jar with metal lids (Nakpunar, New Jersey, USA). When all fungus garden pieces had been randomly assigned a treatment for each dose, glass capillaries containing either (A) nothing (B) α-pinene and (C) linalool were placed in a glass tube to hold the capillary upright, then put in the appropriate jar containing fungus garden.
DNA extraction + 16S rRNA sequencing After 48 hours, fungus garden pieces were collected into 50 mL conical tubes and weighed. Total DNA was extracted using a bacterial enrichment method previously described 32. Briefly, the fungus garden material was homogenized using a sterile mortar and pestle. Then, the homogenized material was submerged in 1X PBS containing 0.1% Tween 80, followed by centrifugation for 15 min at 500-700rpm. This results in a layered mixture containing leaf-material and fungal mass at the bottom, and bacteria in the middle/top. This washing step was repeated until the bacterial layer became more transparent (about 3-5 times). Then the bacterial sample was spun down, the pellet was resuspended in 1X PBS + .1% Tween 80 and passed through a 40 μm filter to remove any larger, non-bacterial debris. The filter was flushed with an additional 5 mL 1XPBS + .1% Tween 80 then the entire sample was spun down again. Total DNA was extracted using a Qiagen DNeasy Plant Mini Kit. DNA was submitted to the Biotechnology Center for library preparation and sequencing. Specifically, amplicon libraries spanning the V4 region of the 16S ribosomal gene were constructed and then sequenced using Illumina MiSeq 2×300bp at the University of Wisconsin – Madison Biotechnology Center.
Sequence reads were processed, aligned, and categorized using DADA2 1.12.170. The DADA2 pipeline (https://benjjneb.github.io/dada2/tutorial.html) was followed almost exactly in July 2019. The only change was made in the filtering step using altered truncLen and trimLeft parameters since we sequenced 2×300bp reads (truncLen=c(225,280); trimLeft=c(10,10)). The statistical analysis was performed as described in the DADA2 pipeline provided above. Additional microbiome analyses, including the violin plots in Figure 5C, were adapted from https://bioconductor.org/help/course-materials/2017/BioC2017/Day1/Workshops/Microbiome/MicrobiomeWorkflowII.html. The adapted code can be found in github.com/cfrancoeur/PSC.
Headspace sampling of fungus gardens with PSC
Atta cephalotes colonies (Table S5) were used in this experiment to create sub-colonies. 20 mL 18 mm Restek (Bellefonte, PA, USA) vials (cat#23082) with magnetic screw-thread caps (cat#23090) were prepared three ways with α-pinene or linalool: (1) Empty vials with a 20 μL Accu-Fill 90 micropet cut to 2.54 cm and flame-sealed containing 1 μL of PSC (n=6). (2) Vials with approximately 0.3-0.4g of cotton with a 2.54 cm flame-sealed micropet containing 1 μL of PSC (n=3). (3) Vials with approximately 0.3g of fungus garden material with all ants manually removed. Vials contained a 2.54 cm flame-sealed micropet containing 1 μL of PSC (n=3 subsamples × 3-5 different Atta cephalotes colonies). Three colonies were used in the α-pinene experiment and five colonies were used in the linalool experiment. Three colonies with three subsamples was determined to have enough statistical power for the observed differences. We also prepared samples of vials with only fungus garden (i.e. no exposure to PSC) during certain runs to ensure that there were no detectable PSC innate to the system. This was done for three separate time points based on exposure to a PSC: 12 hours post-exposure, 24 hours post-exposure, and 36 hours post-exposure. At these given time points, each respective set of vials were destructively sampled and analysed by a Shimadzu HS20 Headspace Sampler coupled to a Shimadzu GC-2010 Plus with a flame ionization detector. Specifically, vials were loaded into the headspace sampler and injected into a column with a 50:1 split flow ratio. For the vials with α-pinene, the headspace sampler and oven were at 60°C, followed by a 20°C/min ramp up to 140°C. For the vials with linalool, which has a higher boiling point than α-pinene, the headspace sampler and oven were kept at 70°C, followed by a 25°C/min ramp up to 205°C. Then, compounds were identified using retention time (α-pinene =3.2 minutes, linalool=6.2 minutes) and areas under the curve were calculated in Shimadzu’s LabSolutions software to determine the relative difference in α-pinene or linalool between vials. α-pinene and linalool are compounds that the garden bacteria can degrade and are inhibitory against Leucoagaricus sp.
Since we took subsamples (sub-colonies) from each Atta cephalotes colony (3 subsamples × 3 time points × 5 colonies), we employed a linear mixed-effects model to account for the correlation (non-independence) between subsamples. Specifically, to test if ant colony had an effect on the observed value, we used the lmer package v. 3.1-0, holding time and treatment as the fixed effects and ant colony as the random effect. Before the analysis, we divided the values by 1,000,000 to rescale the response for the lmer optimization procedure. For the α-pinene treatment, the colony variance is reported as 0, indicating that the variability with respect to ant colony is much smaller than the variability with respect to the residual error. For the linalool treatment, the colony variance was 0.000226, indicating that some of the variability observed was due to the sampling from different colonies. We then used the estimated marginal means (EMMs) with the emmeans package v. 1.3.5 for linear regression analysis of the data, using the pairs() method. Marginal means were compared pairwise between exhaustive two-way level combinations of treatment (control, cotton, fungus garden) and of time (12hr, 24hr, 36hr). Assumptions of normality, linearity, and homoscedasticity for linear regression were examined by plot diagnostics and were met for each analysis. All the code used in this analysis is available at github.com/cfrancoeur/PSC.
Headspace sampling of Leucoagaricus sp
Leucoagaricus sp. strains were isolated by plating small pieces of healthy fungus garden on Potato-Dextrose Agar (PDA). Laboratory fungus-farming ant colonies are kept in a temperature-controlled 28°C room in separate large plastic containers. Five Atta cephalotes colonies collected over the course of several years (2012-2018) from Costa Rica were used for this experiment. Additionally, several isolates from Brazilian Atta gardens used in the Leucoagaricus tolerance experiment (Figure 2) were also included.
2 mL of PDA was pipetted into 20 mL 18 mm Restek vials with magnetic screw-thread caps and left to solidify on a slant. Then, 3×3 mm pieces of freshly growing Leucoagaricus sp. strains were placed onto the slant and grown for one month at room temperature in the dark. Three vials were prepared for each of the Atta cephalotes cultivars (n=5 strains × 3 vials × 2 compounds) and one vial was prepared for three additional Leucoagaricus sp. strains: AB1, AL2, AS1 (n=3 strains × 1 vial × 2 compounds). After the month of growth, 20 μL Accu-Fill 90 micropet (Becton, Dickinson and Company, N.J.) cut to 2.54 cm and flame-sealed were filled with (A) nothing (B) 1 μL α-pinene or (C) 1 μL linalool and then added to the vials. After 36 hours of exposure, the headspace of the vials was analyzed with the same methodology described for the sub-colony headspace sampling. Signal peaks were statistically compared using a Welch two sample t-test, comparing the peaks from the control vials to the vials containing Leucoagaricus sp.
Data Availability
All sequencing data has been uploaded to NCBI under the following BioProject numbers: PRJNA564151, PRNJNA429666, PRNJNA429667, PRNJNA429668, PRNJNA565936, PRJNA577467. Individual accession numbers for each dataset are included in Table S2, Table S3, and Table S5.
Author contributions
C.F., L.K., K.G., A.J.B, R.D.M-S., A.A.P-T., K.K-R., C.R.C designed research; C.F., A.J.B., R.D.M-S. performed research; C.F., K.G. analysed data, and C.F., L.K., K.G., C.R.C. wrote the paper.
Acknowledgments
We are grateful to members of the Landick Lab and Amador-Noguez Lab at UW-Madison, Sherry Cao and Mehmet Tatli, for training on the Shimadzu Headspace sampler. We thank Michael Liou and Michael Howe for their assistance with the statistical analyses. We also thank the staff and scientists at the La Selva Biological Station for allowing field collections of Atta cephalotes and the scientists Allan Artavia and Miguel Pacheco in the Pinto-Tomás lab at University of Costa Rica for their assistance with field collections. We are also grateful to the Comisión Institucional de Biodiversidad of the University of Costa Rica (Resolución N° 009) and the Ministerio de Ambiente y Energía for providing collection permits in Costa Rica. Finally, we thank Heidi Horn for training on collecting fungus-farming ant colonies. This material is based upon work supported in part by the Great Lakes Bioenergy Research Center, U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research under Award Numbers DE-SC0018409 and DE-FC02-07ER64494, the National Institutes of Health grant U19 TW009872, the National Science Foundation grant DEB-1927155, and Vicerrectoría de Investigación (810-B0-501), Universidad de Costa Rica, Ministerio de Ciencia, Tecnología y Telecomunicaciones (MICITT), Costa Rican Government and Florida Ice and Farm Company.