Environments and Hosts Structure the Bacterial Microbiomes of Fungus-Gardening Ants and their Symbiotic Fungus Gardens

The fungus gardening-ant system is considered a complex, multi-tiered symbiosis, as it is composed of ants, their fungus, and microorganisms associated with either ants or fungus. We examine the bacterial microbiome of Trachymyrmex septentrionalis and Mycetomoellerius turrifex ants and their symbiotic fungus gardens, using 16S rRNA Illumina sequencing, over a region spanning approximately 350 km (east and central Texas). Typically, microorganisms can be acquired from a parent colony (vertical transmission) or from the environment (horizontal transmission). Because the symbiosis is characterized by co-dispersal of the ants and fungus, elements of both ant and fungus garden microbiome could be characterized by vertical transmission. The goals of this study were to explore how both the ant and fungus garden bacterial microbiome are acquired. The main findings were that different mechanisms appear to explain the structure the microbiomes of ants and their symbiotic fungus gardens. Ant associated microbiomes had a strong host ant signature, which could be indicative of vertical inheritance of the ant associated bacterial microbiome or an unknown mechanism of active uptake or screening. On the other hand, the bacterial microbiome of the fungus garden was more complex in that some bacterial taxa appear to be structured by the ant host species, whereas others by fungal lineage or the environment (geographic region). Thus bacteria in fungus gardens appear to be acquired both horizontally and vertically.


Introduction
Symbioses with microorganisms have been crucial to the success of insects, if not eukaryotic life in general [24,25,55,64,80]. One of the most striking discoveries is that many symbioses often consist of more than a single host and a single symbiont, and are best viewed as a community of macroand microorganisms [10,22,45,54,74,105]. One of the central issues facing symbiology (the study of symbioses) is understanding how these complex entities are organized and function in a dynamic world, especially with regard to the mechanisms that maintain specificity and homeostasis between hosts and microbial symbionts [20,22,34,38,80,98]. For example, insects often have a microbiome that is structured toward particular life history, e.g., with microorganisms often supplying key nutrients or resources that essentially "upgrade" specialized but otherwise poor diets [23,25,29] or employing symbionts as defense against pathogens [2,15,47]. Ants are among the most ecologically important insects, which can be attributed to their large colonies, population sizes, and diversity in diets [28,53,81,101]. While ants have a digestive system similar to most omnivorous insects, they may feed on a variety of food sources and employ symbionts to upgrade their diets [6,19]. For example, some ants may harbor bacteria that fix nitrogen or involved in N cycling generally [9,42,77].
The fungus gardening ant (subtribe Attina) symbiosis has long been known as a complex system of interacting macro-and micro-symbionts. Fungus-gardening ants cultivate an external fungal symbiont as their primary food source, but the symbiosis is also characterized by hundreds of other associated microorganisms [39,69,87,91,92]. For example, Actinobacteria cultivated via exocrine glands on some attine species' cuticle produce secondary metabolites that appear to have a defensive function against predatory Blake Bringhurst and Mattea Allert contributed equally to this work. 1 3 fungal species (Escovopsis spp.) [56,68,72]. The relationship between Actinobacteria and Escovopsis spp. appears to have a long coevolutionary history with the fungus-gardening ant symbiosis [13,79], although some Actinobacteria may be environmentally acquired [1,75,96,104]. Likewise, Burkholderia bacteria associated with fungus gardens also appear to have antimicrobial function [33]. Certain yeasts may be antagonists towards Actinobacteria and nitrogen fixing bacteria in the fungus garden may provide amino acids to the ants [58,82,86]. Microfungi in gardens may also serve important nutritional, detoxification and defensive functions against specialist and general competitors and predators of the fungus garden [69,86]. Much of our understanding has focused on the species found in the tropics, which may or may not be representative of the interactions occurring in all fungus-gardening symbioses. For instance, the specialized predatory fungus (Escovopsis) is not thought to be very common in temperate attines [87] and other bacterial taxa may be just as important if not more important than Actinobacteria [43]. As a result, detailed intraspecific natural history studies of microbiomes of ants and their symbiotic fungi may increase our understanding of the roles of microbial associates [47].
After more than a decade of research, much has been learned about the basic structure of the microbial communities of fungus-gardening ants. The bacterial communities of fungus-gardens typically are characterized by few genera that make up the vast majority of the microbiome; for example, Atta colombica and Atta cephalotes fungus gardens contained primarily Enterobacter, Klebsiella, Citrobacter, Escherirchia, and Pantoea, i.e., bacteria that are often found in herbivore digestive systems [3]. Probably because attine ants have outsourced most digestion to the external fungal symbiont, the gut microbiome appears to be less diverse and may have a relatively minor role with regard to digestion [89,90]. The function of many bacterial taxa residing in fungus gardens are poorly understood; for instance, notably, one report found that fungal inocula containing a high amount of Mesoplasma were more likely to cause colony decline, though it was unclear whether this was correlative or causal [67]. Ishak et al. [43] examined the microbes present in the fungus gardening ants Trachymyrmex septentrionalis; the results indicate that Actinobacteria (e.g., Pseudonocardia sp., Kribbela sp., Amycolatopsis sp., and Streptomyces sp.) and Mollicutes were the most abundant bacterial taxa in the fungus of T. septentrionalis. Ishak et al. [43] suggested that T. septentrionalis ants within the same colony and colonies have distinct microbiomes, whereas another study demonstrated specificity of bacterial microbiomes among colonies of Acromyrmex echinatior [1]. A recent comparative study illustrated that five different attine species and their colonies harbor distinct microbial communities from one another [88]. Like many symbioses, attine symbionts are obtained from a mix of horizontal and vertical transmission. For example, the fungus garden is typically vertically transmitted from parent colony to gynes (new queens) [5,70,76,91]. Actinobacteria are also thought to be primarily vertically transmitted, since clades of ants typically grow similar strains of Pseudonocardia [13,71]. On the other hand, some studies have documented sharing of fungal symbionts among ant species and environmental acquisition of bacteria [5,50,76,92]. Closely related ant genera may even have different mechanisms of gut bacteria acquisition [108]. Unfortunately, there have been very few studies that have taken a comparative approach to determine how the microbiomes of these symbioses are acquired outside of major study systems such as Panamanian Acromyrmex species [1,104,108].
Although research is accumulating on fungus gardeningant microbiomes, not many studies have examined how geography, host species, or type of fungal symbiont may influence the structure of microbial communities [50,67,88]. For example, ants in the "higher attine genera" (Mycetomoellerius, Paratrachymyrmex, Seriocomyrmex and Trachymyrmex) grow conservatively five lineages of fungi and may host switch to varying degrees [5,32,44,59,71,100]. It remains unknown how the microbiome may be altered as members of the symbiosis are added or lost, such as by fungal symbiont switching observed among many host ant species [5,32,59,70,71,76,92]. Unfortunately, there have been very few microbiome studies that examine both ants and fungus gardens across ecological scales [43,50]. Studies have focused on fungus garden bacterial microbiome structure [3,67] or solely on ants [88][89][90]. Similarly, Ishak et al. [43] examined temporal patterns of microbiome structure, whereas Kellner et al. [50] examined spatial components.
Groundbreaking -yet limited -studies like the latter are needed because the ants, fungus and many of their bacterial associates are clearly interacting with one another across ecological and evolutionary scales [5,76,91,93,94]. Therefore conclusions drawn from the microbiome of only one partner of the symbiosis are limited.
To unravel what factors influence the microbiome of T. septentrionalis and M. turrifex ants and their symbiotic fungus, we examined the bacterial microbiome of the ants and symbiotic fungi of T. septentrionalis and M. turrifex across a portion of their shared ranges. The broad goal of this study was to characterize the bacterial microbiome of T. septentrionalis and M. turrifex and their symbiotic fungus garden and determine potential drivers of this variation. Specifically, we had four main questions: (1) Do the ant species harbor different bacterial communities? (2) Do the two ant species influence the bacterial community structure of the fungus gardens? (3) Are fungal lineages characterized by different microbiomes? (4) Are there geographical differences in the structure of ant or fungal bacterial microbiomes?

Study Area
Samples of both ants and fungus garden materials were obtained during May and June 2016 from field colonies from sites in two broad regions in east-central Texas. Samples were obtained from four sites in northeast Texas in close proximity to Tyler, Texas ("East Texas") (approximately 32.29° N 95.24° W) and from three sites approximately 300 km away in Bastrop, Texas (approximately 29.39°N 97.32°W) ("Central Texas") ( Fig. 1). Both species of ants co-occur at most of the same locations ( Fig. 1, Table 1). Although all of these sites occur in the Post Oak Savanna Ecoregion, which is characterized by sandy soils and an overstorey of post oaks (Quercus stellata) and occasionally loblolly or shortleaf pine (Pinus taeda and P.echinata, respectively) and understory shrubs such as yaupon holly (Ilex vomitoria) and eastern redcedar (Juniperus virginiana), East Texas tends to be wetter than Central Texas (115 cm vs. 90 cm of precipitation per year, respectively) [21]. The latter is especially known for extreme droughts where less than 30 cm of rain may fall in a year [73]. Central Texas is also the western limit of T. septentrionalis whereas East Texas (and western Louisiana) is the eastern limit of M. turrifex [59,84,95,97].

Sampling Methods
We collected up to five adult ants from each colony of both species; these ants were pooled into a single DNA extract to account for heterogeneity among individual ants in the colony [43]. Ants were collected directly from inside fungus gardens with ethanol and flame-sterilized forceps, meaning that the ants collected were likely garden workers (i.e., not foragers who could pick up bacteria inadvertently while outside the nest). Roughly an equal number of T. septentrionalis and M. turrifex colonies from our samples of the East Texas and Central Texas populations were utilized (N = 13 for T. septentrionalis and N = 11 for M. turrifex) ( Table 1). Five T. septentrionalis from central Texas and eight from east Texas were used, whereas our samples included only two M. turrifex colonies from central Texas (the remaining were collected in east Texas) ( Table 1). A small sample of fungus garden material was collected similarly with flame and ethanol sterilized forceps from the same garden chambers where the ants were collected. Lastly, seven soil samples were taken from within the nest fungal chambers (N = 4 from East Texas and N = 3 from Central Texas) to act as a negative control, and make sure the microbe communities observed with the ant or fungal samples were not a relic of soil contamination. All samples were preserved immediately upon collection in 100% ethanol.

Fungal Genotyping
Each fungus garden was genetically identified to phylotype (clade) using the terminology and methods of Luiso et al. [59]. Briefly, gongylidia (swollen hyphal tips diagnostic feeding structures of higher fungus-gardening ants) were plucked off the fungus with flame-sterilized forceps, placed in an aqueous solution of Chelex, and heated in a thermal cycler [59]. Before PCR amplification, the DNA extract was diluted (1:10) using nuclease-free water (higher concentrations of DNA inhibited PCR reactions). The PCR Fig. 1 Map of Texas displaying the East Texas (red) and Central Texas (yellow) sample collection sites was performed using the ITS 4 and ITS 5 primers to amplify the 18S rRNA ITS gene [59,103]. The resulting PCR products were sent to the DNA Sequencing Facility at the University of Texas at Austin for Sanger sequencing on an Applied Biosystems 3730 DNA Analyzer. The resulting ITS sequences were cleaned up and aligned in Geneious 10.1.2 [49]. Sequencing errors or misreads in the DNA sequences were manually corrected. Sequences were identified using BLAST and personal databases of ITS sequences (Table 1) [59].

DNA Extraction, PCRs, and Sequencing of Microbiomes
DNA extraction and sequencing of the 16S samples was performed at MRDNA in Shallowater, Texas (http:// www. mrdna lab. com/). DNA extractions were conducted using Qiagen DNEasy Powersoil Pro kits (Qiagen item no. 12888). The sequencing provider conducts regular negative controls. DNA sequences were amplified from whole ants and fungus using primers Gray28F 5′GAG TTT GATCNTGG CTC AG and Gray519R 5′GTNTTACNGGGCKGCTG that span the V1-V3 hypervariable regions of the 16S rRNA gene. They were processed using the HotStarTaq Plus Master Mix Kit (Qiagen, USA) under the following conditions: 94 °C for 3 min, followed by 28 cycles of 94 °C for 30 se, 53 °C for 40 s, and 72 °C for 1 m, after which a final elongation step at 72 °C for 5 min was performed. After the samples were amplified and checked for adequate genetic yields, the sub-samples were pooled back together and purified using calibrated Ampure XP beads. The purified and pooled PCR product was used to create a DNA library and sequenced using the Illumina MiSeq platform in PEx300 mode. Sequences have been deposited in the NCBI Sequence Read Archive under the BioProject PRJNA789907.

Microbiome Bioinformatics
Initial sequence cleanup was performed by removing short sequences with < 150 bp, sequences with ambiguous base calls, chimeras, sequences with runs exceeding 6 bp, and singleton sequences [26]. The resulting sequences were then inputted into Qiime2-2020.6, after having their barcodes and linker and reverse primers removed [8]. Sequences were demultiplexed using the demux plugin (https:// github. com/ qiime2/ q2-demux). The dada2 plugin was then used to merge the forward and reverse sequences, as well as perform basic quality control [14]. When using the dada2 plugin, sequences were truncated down to 260 base pairs as the average quality score dipped below 20 beyond this point. Taxonomy classification with 99% similarity was performed utilizing the SILVA 132_QIIME database [83,106].
To do this, we created our own taxonomic classifier using the "feature-classifier fit-classifier-naive-bayes" command and the SILVA database. This classifier was used to assign sequences a taxonomic classification using the featureclassifier plugin [7] with the "classify-sklearn" command. Sequences associated with mitochondria and chloroplasts were removed using the "qiime taxa filter-table" and "qiime taxa filter-seqs" commands. To ensure an equal diversity comparison across all samples, the sequences of each sample had to be rarified to reduce the effects of samples with more sequences having more potentially unique sequences (McMurdie and Holmes, 2014). For this, a sampling depth of 1700 sequences per sample was utilized using the "qiime feature-table rarefy" command. An OTU table of the rarified samples was created by inputting a tabulated taxonomic bar plot, created using the "taxa barplot" command, of our sequences into Qiime2 View (https:// view. qiime2. org). This OTU table can be found on Dryad Digital Repository https:// doi. org/ 10. 5061/ dryad. 7wm37 pvv1. The Qiime2 processing pipeline has been deposited on GitHub (https:// github. com/ bsbri nghur st/ TS-and-MT-Micro biome-Files).

Statistical Methods
The OTU table was used to calculate the percent abundance of each OTU for each sample. From here, the mean percent abundance and 95% confidence interval of each OTU for a sample type (such as T. septentrionalis ants or East Texas fungus) was calculated. Taxonomic bar plots were constructed using cumulative abundances percentages of OTUs with percentages greater than 1% across the associated samples to compare between sample types. The VEGAN R package was used to calculate the Shannon's Diversity Index (H) for every ant, fungal garden, and soil sample [78]. The respective average Shannon's Diversity Index values were then compared among groups (ant host, fungal clade, or region), using either Welch's two sample t-tests or ANOVAs, for the ant and fungal garden samples. A Tukey's Honest Significant Difference test was performed using the multcomp package in R as the post-hoc analysis to compare Shannon's Diversity Index values among groupings with three different groups [40]. EstimateS v9.1.0 was used to create rarefaction curves to compare OTU richness among groups of ant and fungal garden samples as well as compare sampling adequacy [16]. Rarefaction curves were plotted along with their respective 95% confidence interval. These rarefaction curves were additionally extrapolated out to 20 samples if a variable category did not contain over 20 samples to further illuminate differences between the variable categories due to their small sample sizes.
Additionally, to characterize microbiome structure that may be independent of overall abundance, we conducted an indicator species analysis (ISA) using the indicspecies R package (with each analysis having 9,999 permutations). ISA is a method that examines which bacterial taxa contribute most toward the overall variation in microbiome structure of the ant and fungal garden samples [12]. ISAs were performed at the full microbiome level and within the major bacterial classes. Significant OTUs outlined by each ant or fungal class-based ISA had their cumulative abundances summed, where then taxonomic bar plots were created to show the proportion of the summed OTU abundances within each ant/host ant species, region, or fungal garden clade, with the proportions additionally separated out by the specific OTUs. As these taxonomic bar plots show the proportion of the summed OTU abundances within certain categories, with the y-axes depicting the specified proportion of the summed significant OTU abundances, the summation of the proportions for both, or all, categories equal 1, or 100%. To control for multiple hypothesis testing, analyses of composition of microbiomes (ANCOMs) were performed as differential abundance tests for the OTUs in addition to the ISAs [60]. ANCOMs compared the impact of the significant variable (either ant/host ant species, fungal garden clade, or geographic region), found through the multivariate statistics (details in following section), at the full microbiome level in Qiime2-2020. 6. The results of the ANCOM provided each OTU a W statistic, which was the number times a selected OTU's log-ratio was found to be significantly greater than another OTU's log-ratio, as well as if that W statistic was significant. As an OTU's W statistic is based off pairwise comparisons with other OTUs, the threshold for significance is based on the number of OTUs inputted into the ANCOM. For control measures in the ANCOMs, OTUs with less than 10 sequences across the ant or fungus samples were removed as well as OTUs that appear in only one ant or fungus sample. Additionally, to conform with the ANCOMs, pseudocounts of 1 were added to any zero that appeared in the ant or fungus OTU tables. Finally, the venn R package was used to create a Venn diagram to compare the presence and absence of OTUs found between different sample types [27].

Multivariate Statistics
Since our ant and fungal samples were confounded by multiple variables (host ant species (N = 2), fungal garden clade (N = 3), and geographic region (N = 2)), we explored how much variation in microbiome structure was explained by these variables. To see any initial patterns in the data, we first constructed non-metric multidimensional scaling (NMDS) plots based off of Bray Curtis distances using the VEGAN R package to visualize patterns between the ant and fungal microbiomes [78]. To look for statistical significance 1 3 in these patterns, we then created OTU tables with a subset of either the ant or fungal samples. A detrended correspondence analysis (DCA) was performed on each of the two OTU tables (ants or fungus) utilizing the VEGAN R package that then determined whether a canonical correspondence analysis (CCA) or redundancy analysis (RDA) was the best tool [78]. A threshold value of 4 for the DCA Axis 1 was used during the DCA to denote whether a CCA or RDA was to be used, with a value greater than 4 denoting that a CCA was optimal [78]. Each CCA or RDA was performed under a reduced model and used 10,000 permutations. Due to unequal sample sizes among the main effects (ant species, region and fungal clade), we did not test for any interactions due to the risk of a type I error.
To further quantify differences between the microbial communities of the ants and their fungus gardens, CCAs or RDAs were performed on major bacterial classes identified by the significant OTUs in the ISAs to determine whether host ant, fungal lineage or region played any role in community structure of these groupings. These tests were performed on OTU tables consisting of the subset of OTUs corresponding to significant taxonomic groups revealed by the ISA. Further ISAs were performed on all the OTUs within these bacterial classes to highlight which taxa might be driving the overall patterns.

Bacterial Distribution
After the raw data was processed and filtered through Qiime2, the dataset consisted of 1,003,775 sequences, with 481,514 sequences associated with the 24 ant samples, 470,669 sequences associated with the 24 fungal samples, and 51,592 sequences associated with the seven soil samples. After rarifying the data, 974 OTUs were detected, with 275 of these OTUs found within the ant samples, 699 OTUs found within the fungal samples, and 476 OTUs found the soil samples.

Regional Analysis of Ant, Fungus, and Soil Microbiome
Initial multivariate analysis of the ant and fungal samples indicated that ants, fungus, and soil samples formed visually distinct clusters ( Fig. S1; stress = 0.185). Moreover, ants, fungus, and soil bacterial microbiomes were significantly different from one another (CCA; F = 2.952, df = 2, p < 0.0001). The ant (H = 1.939), fungus garden (H = 1.930), and soil samples (H = 4.000) had significantly different Shannon's Diversity Index values (F = 12.59, df = 2, p = 3.49 × 10 −5 ), with the soil samples having a significantly higher average in comparison to the ant and fungus garden samples (Fig. S2). These results confirm that microbial communities of ants and fungus we were analyzing in this study are not contaminants originating from the surrounding soils. In all subsequent analyses, soils were removed since they were of no primary interest.

Ant Microbiome Composition
The two ant species were the primary driver of their microbiomes (CCA; F = 3.6942, df = 1, p < 0.0001), whereas fungus garden clade (F = 1.1906, df = 2, p = 0.2028) and region (F = 0.9101, df = 1, p = 0.5952) did not appear to have significant roles in structuring the bacterial microbiome of ants. These findings were corroborated with NMDS plots showing that the ant samples formed distinct clusters based on species (Fig. 2a; stress = 0.191), and not region ( Fig. 2b; stress = 0.191). Additionally, NMDS plots showed that T. septentrionalis had a less variable microbiome than M. turrifex, since T. septentrionalis ant samples were more clustered together while M. turrifex ant samples were more variable (Fig. 2a).
Ant microbiomes appeared to be structured by different bacterial taxa. The most common taxa in the microbiome of T. septentrionalis ants were one strain of Solirubrobacter (30.122%), a strain of Luteimonas (14.552%), and an unknown member of Burkholderiaceae (9.814%) (Table 2a). In contrast, the three most common taxa in the microbiome of M. turrifex ants were a strain of Amycolatopsis (20.417%), an unknown member of Burkholderiaceae (14.449%), and an undescribed member of Microbacteriaceae (9.797%) ( Table 2b). Such findings were in concordance with cumulative abundances as well (Fig. 3). Additionally, Pseudonocardia strains consisted of 7.048% of the M. turrifex ant microbiome whereas it was nearly absent in (< 1%) T. septentrionalis ants (Table 2a, (Table 3a, b, Table S1). After filtering the OTU table in preparation for the ANCOM, 39 OTUs were found in the ant samples. As such, the level of significance in the ANCOM for the ant samples was based on 39 OTUs. As ant species was the only variable found to be significant by the CCA, the ANCOM analysis of these two ant species found that a strain of Amycolatopsis (W = 36) associated with the M. turrifex ant microbiome, whereas two strains of Solirubrobacter (W = 36; W = 32), a strain of 1 3 Aeromicrobium (W = 30), Naumannella (W = 30), and Ponticoccus (W = 29) were associated with the T. septentrionalis ant microbiome (Table S2a).
As a next step, we probed whether bacterial diversity in the classes identified as significant in the ISA were explained by (1) geographic region, (2) ant species or (3) fungal garden clade (Table S1). The ISA reported that four bacterial classes (Actinobacteria, Alphaproteobacteria, Bacteroidia, and Gammaproteobacteria) were important in describing the variation among ants. The Actinobacteria and Gammaproteobacteria communities of the ants were primarily explained by the ant species (CCA ;  Table 4a). However, the Alphaproteobacteria and Bacteroidia communities of the ants were not explained by any variable (RDA; Table 4a). The ISA found 12 significant OTUs within the Actinobacteria community that differed between the two ant species. OTUs of Pseudonocardia and Amycolatopsis associated with M. turrifex and OTUs of  Naumannella, undescribed Intrasporangiaceae, and Ponticoccus within T. septentrionalis were important drivers in overall discrimination between the two ant associated Actinobacteria communities (Fig. 4a, b). Other significant OTUs driving the differentiation in the Gammaproteobacteria between the two ant species included a Lutimonas OTU in T. septentrionalis and Stenotrophomonas in M. turrifex (Fig. S3). Despite community level differences in taxa comprising the bacterial microbiomes of the two ant species, overall diversity of the two species was similar. Although M.

Fungal Microbiome Composition
Unlike ant bacterial microbiomes, geographic region appeared a greater role in structuring the bacterial microbiome of the fungus garden (CCA; F = 1.4075, df = 1,   (Table 5a). The three most common taxa in the microbiome of Central Texas fungus were Mesoplasma (12.941%), a strain of Tyzzerella (7.840%), and a strain of Pantoea (7.042%) ( Table 5). Such findings were in concordance with cumulative abundances as well (Fig. 5).
The ISA found that two OTUs defined East Texas fungus, whereas 35 OTUs defined the Central Texas fungi; with most of these indicator OTUs found in the families Acetobacteraceae, Acidobacteriaceae, Burkholderiaceae, Cthoniobacteraceae, Enterobacteriaceae, Sphingobacteriaceae, and Weeksellaceae (Table 3, Table S3). After filtering the OTU table in preparation for the ANCOM, 73 OTUs were found in the fungus samples. As such, the level of significance in the ANCOM for the fungus samples was based around 73 OTUs. As geographic region was the only variable found to be significant by the CCA, the ANCOM of these two fungal regions found only that Tyzzerella (W = 58) was associated with the Central Texas fungal microbiome (Table S2a). Except for a single undescribed OTU in the genus Spiroplasma, no members of the Entomoplasmataceae (the family of the most common taxa, Mesoplasma) were significant members of the fungus garden bacterial microbiomes (Table 3, Table S3).
The ISA reported that five bacterial classes (Acidobacteriia, Actinobacteria, Alphaproteobacteria, Bacteroidia, and Gammaproteobacteria) were important in describing the variation among the fungus gardens (Table 3 and S2). As with the ants, the following analyses probed whether variation in the OTUs in these classes were explained by (1) geographic region, (2) host ant species, or (3) fungal clade. The Acidobacteriia community of the fungus gardens was Fig. 4 Bar plots depicting the proportion of the significant OTU abundances within the Actinobacteria community (identified using an indicator species analysis) of ant samples that are driving the Actinobacteria community differences between the two ant species, with separate plots made for the significant taxa corresponding to (A) M. turrifex and (B) T. septentrionalis ants  Table 4b). The Actinobacteria community of fungus gardens was significantly explained by the host ant species and region, but the fungal clade may play a role in determining community variation, although the test was not statistically significant (CCA ; Table 4b), but low enough to warrant concern about a type II error. The Gammaproteobacteria community of fungus gardens was explained by all three variables (CCA; Table 4b). The Alphaproteobacteria and Bacteroidia communities for fungus gardens was not explained by any variable (RDA; Table 4b).
The ISA within the Acidobacteriia community found that an Edaphorbacter OTU and an undescribed OTU from the Acidobacteria subgroup 2 as being significant to Central Texas fungal samples (Fig. 6a). Similarly, regional differences among the Actinobacteria were driven by two OTUs, with an OTU of Mycobacterium found to be significant in Central Texas fungal gardens and Naumannella found to be significant for East Texas fungus gardens (Fig. 6b). Three OTUs within Gammaproteobacteria appeared to explain regional community differentiation, with OTUs within the Burkholderia-Caballeronia-Paraburkholderia complex and a Massilia OTU found significant for Central Texas fungal samples (Fig. 6c). Additionally, an OTU within Gammaproteobacteria (a member within the Burkholderia-Caballeronia-Paraburkholderia complex) was found to significantly drive community differences of the three fungal lineages, but this significance was due to the driver OTU being found primarily within the Clade B3 samples, which was relatively undersampled (n = 3) (Fig. S4a).
Since Mesoplasma was very common in the fungus garden samples, yet seemed to poorly characterize fungus garden microbiome structure (per the ISA), sequences of Mesoplasma were removed to examine for patterns among the other bacterial taxa. When sequences associated with Mesoplasma were removed, the sampling depth of the sequences was reduced to 1050 due to the high abundances of Mesoplasma in some samples. However, two samples had less than 1050 sequences and were removed from further calculations with Mesoplasma sequences excluded. Without Mesoplasma, the fungus microbiome was significantly determined by region (CCA; F = 1.3826, df = 1, p = 0.0061), whereas fungal clade (F = 0.9233, df = 2, p = 0.7121) and host ant species (F = 0.8869, df = 1, p = 0.8839) were not significant factors in structuring fungus garden microbiomes. These findings were corroborated with an NMDS plot showing that the fungal samples formed clusters based around their region (Fig. 7, stress = 0.173).
Comparing extrapolated rarefaction curves for fungus garden microbiomes (Mesoplasma excluded) show that Central Texas fungi contain significantly more OTUs in comparison to East Texas fungi (Fig. 8a). The average Shannon's Diversity Index for Central Texas (H = 3.094) Fig. 6 Bar plots depicting the proportion of the significant OTU abundances within the (A) Acidobacteriia, (B) Actinobacteria, and (C) Gammaproteobacteria communities (identified using an indicator species analysis) across all fungal samples that are driving the community differences between the two collection regions, east (East) and central (Central) Texas fungus was greater than that of East Texas (H = 2.065) fungus (t = − 2.140, df = 15.221, p = 0.049). Rarefaction curves of fungus garden microbiome based on host ant species showed no significant differences (Fig. 8b).
Rarefaction curves based on fungal clade showed significant differences; with Clade B4 fungus having more expected OTUs than Clade B5 fungus (Fig. 8c). The similarity between the rarefaction curves of fungus grown  (Fig. 8e).
Comparing extrapolated rarefaction curves for fungus garden microbiomes (Mesoplasma included) show that Central Texas fungi contain significantly more OTUs in comparison to East Texas fungi (Fig. S5a); the average Shannon's Diversity Index for Central Texas (H = 3.033) fungus was greater than that of East Texas (H = 1.476) fungus (t = − 2.727, df = 12.332, p = 0.018). Rarefaction curves of fungus garden microbiome based on host ant species showed no significant differences (Fig. S5b). Rarefaction curves based on fungal clade showed significant differences; with Clade B4 fungus having more expected OTUs than Clade B5 fungus (Fig. S5c). The similarity between the rarefaction curves of fungus grown by the different ant host species was also reflected in the Shannon's Diversity Index values for fungus grown by T. septentrionalis (H = 1.661) and M. turrifex (H = 2.248) (t = 0.985, df = 21.87, p = 0.3354) (Fig. S5d). However, there was not a statistical difference in the average Shannon's Diversity Index between Clade B3 (H = 3.042), Clade B4 (H = 2.173), and Clade B5 (H = 1.640) fungal samples (F = 1.197, df = 2, p = 0.323) (Fig. S5e). In summary, the comparisons of analyses with or without Mesoplasma suggest that Mesoplasma did not appear to influence overall diversity patterns.
An examination of the taxa shared among ant species, fungus gardens and region, suggested some sharing between fungus gardens and with ants, but very little sharing among the ants' species (Fig. 9). For example, 30 taxa were found to be shared among fungus gardens from central and east Texas and T. septentrionalis and M. turrifex ant microbiomes (Fig. 9, Table S4). Out of the 30 shared OTUs, only Mesoplasma was in abundances greater than 1% across all ant and fungus sample types. Three OTUs (all ambiguous or uncultured taxa; Table S4) were shared between fungus gardens of the two regions and T. septentrionalis ants, whereas 59 OTUs were shared just between fungus gardens of both regions and M. turrifex ants (Fig. 9, Table S4). However, no taxa had abundances greater than 1% for both the comparison between both regional fungal gardens to T. septentrionalis and the comparison to M. turrifex ants. Two OTUs were shared just between the Central Texas fungus gardens and both ant species, whereas 7 OTUs were shared just between the East Texas fungal gardens and both ant species (Fig. 9, Table S4). No taxa had abundances greater than 1% for both the comparison between both ant species to Central Texas fungal gardens and the comparison to East Texas fungal gardens.

Discussion
The most significant finding in this study is that different processes appear to structure the bacterial microbiomes of the fungus-gardening ant symbiosis. Each ant species had a unique bacterial microbiome, whereas fungal lineage appeared to have little to no role in structuring the bacterial microbiome of the fungus garden. This study provides more evidence that individual ant species possess unique bacterial microbiomes [37,88]. Unlike an earlier report that found similarities of bacterial microbiome structure in the ants and fungus gardens of the attine Mycocepurus smithii, fungus gardens of T. septentrionalis and M. turrifex possess bacterial microbiomes that are very distinct from their host ant species [50]. Other than OTUs of two bacterial classes (Acidobacteria and Alphaproteobacteria) associated with fungus gardens that appeared to segregate among fungal lineages and the two ant species, there appeared to be few if any bacteria that could be codispersed with the fungus (Table 4), but some of this could be within colony transfer as the ants walk around the garden shedding bacteria [18]. The bacterial microbiome of the fungus garden had a lesser defined structure that may in part be determined by the environment.
The ant associated bacterial microbiomes were very distinct. While T. septentrionalis and M. turrifex had a high abundance of Actinobacteria as part of their microbiome, the taxonomic identity of these bacteria differed between the two species. Mycetomoellerius turrifex ant microbiome was characterized by Actinobacteria in the genera Pseudonocardia and Amycolatopsis, whereas the ant microbiome of T. septentrionalis was characterized by Ponticoccus and Microlunatus. Additionally, T. septentrionalis appeared to have an extensive coverage of Solirubrobacter. The function of Solirubrobacter is unknown, however, it has been reported in soil crust [85], agricultural soils [52], and earthworm burrows [99]. Actinobacteria are commonly known to produce antibiotic and are commonly seen in fungus-gardening ants and other insects where they are thought to serve as defensive symbionts [30,46,48]. For example, Actinobacteria are thought to be part of a generalized defense against Escovopsis pathogens [17,56] and of the ants [11,63]. Considering the overall differences in ant microbiome structure and the differences in indicator species, it would appear that each species has either (1) a different strategy of dealing with pathogens or (2) experiences different pathogens altogether. The identity of pathogens in either population is largely unknown and unexplored, though studies on other attines have shown that different attine taxa can share the same or similar pathogens [35,51].
Assuming that the significant ant species effect of ant bacterial microbiomes is an outcome of vertical transmission, the contrasting microbiomes found in T. septentrionalis and M. turrifex could reflect different phylogeographic histories. The relative uniformity of bacterial communities associated with T. septentrionalis ants could reflect low diversity from recent population expansion or strong purifying selection of microbial consortia, whereas the high variation among M. turrifex ant microbiomes could be due to relaxed selection on bacterial consortia and/or high diversity due to an ant population expansion that happened in the more distant past. The genetic diversity of T. septentrionalis ants is generally low in Texas (and lowest in Central Texas), which likely represents recent expansion, a conclusion supported by mtDNA, microsatellites and genotyping-by-sequencing approaches [62,95].
Variation in ant species microbiomes may involve responses to stress, since stress is thought to structure microbiomes [41,65,107]. For example, microbiomes may become more variable in stressful environments as hosts lose control over their symbionts or variable microbiome structure may be a mechanism to deal with environmental stress [65]. The region studied (east and central Texas) encompasses the eastern and western range limits of M. turrifex and T. septentrionalis. Recent distribution models indicated that different variables predicted their respective distributions; for example, the M. turrifex distribution was explained by annual mean temperature whereas T. septentrionalis was predicted by winter precipitation [97]. How the microbiomes react to these environmental variables remains unknown, but could be explored experimentally by exposing ant colonies to temperature or humidity stress.
Interestingly, we found very little support that each ant species are structuring the microbiome of their fungus gardens, since the gardens of each species were not different from one another in terms of their associated bacterial community. This is surprising, considering the differences in ant-associated Actinobacteria, which hypothetically would produce different secondary metabolites [4,36,66] that could in turn influence the abundances of other bacterial taxa. One possible explanation is that regions could differ in plant communities and thus the types of substrates the ants are feeding the garden [31,32,94]. While there were significant differences among the OTUs encountered in each region, most belonged to the same bacterial taxa (e.g., Actinobacteria or Acidobacteria, Table 3, S2), which suggests common mechanisms in the acquisition of these bacteria and some genetic differences among bacteria that could be explained by dispersal limitation. A previous study examined the ant and fungal microbiome of Mycocepurus smithii and discovered that the fungal microbiome was influenced by the environment rather than the fungal lineage, which suggested that the ants were actively acquiring bacteria from the environment [50]. As a result, most of the impact of ant species or fungal lineage could be obscured by fungal substrates and environments.
This study along with others [37,67,89,90] revealed that Mesoplasma to be common in ants and fungus gardens, yet variation in Mesoplasma (or Enterobacteriales, generally) was not a significant predictor or microbiome structure or ant host, fungus garden or region. Phylogenetic analyses suggest that there are attine specialized Mesoplasma distinct from army ant Mesoplasma, but the resolution of standard Illumina reads appears to be too low to warrant interspecific comparisons among attine species [67]. The function of Mesoplasma remains unknown, but it might contribute to colony mortality, it might be opportunistic, or it might be a permanent mutualist or a contextdependent mutualist [90].

Conclusion
This study has shown that two distantly related, co-occurring fungus gardening ant species possess complex microbiomes that are acquired by different mechanisms. Much of the bacterial microbiome of the fungus garden appears to be acquired from the environment whereas the ant associated microbiome has a strong ant-host signal. The factors driving the environmental signal could be explored by incorporating more sites and more colonies within sites. We caution that the strong ant-host signal does not necessarily imply vertical transmission and could be a result of active uptake by the ant hosts that is independent of vertical inheritance. A limitation of the current study is that it did not incorporate intraspecific molecular markers (i.e., within host or fungus garden lineage); as a result, this study did not possess the power to explicitly test for possible "phylosymbiosis" of ant or antfungus and bacterial symbionts, which predicts phylogenetic concordance among hosts and symbionts [57,102]. Two distinct processes could determine the apparent species-specific ant microbiome. Bacteria could be inherited vertically via founding queens, which would explain the strong cophylogenetic signal between ants and Pseudonocardia bacteria [13]. On the other hand, the species specific effect we report here could also be explained by horizontal transmission if ants are actively acquiring bacteria in a species specific manner [75]. As a result, while there is a strong ant species effect driving microbiome structure, some of this pattern could result from behavioral differences where the ants selectively acquire or "screen" bacterial lineages over others. Therefore, the comparison of M. turrifex and T. septentrionalis may be somewhat crude and may overemphasize coevolutionary signals, considering the large genetic distances between these two species [100]. Future studies should compare microbiomes of closely related species, such as those found in western North America and the northern Neotropics [5,100] and employ intraspecific genetic markers [5,61,62]. Moreover, co-phylogenetic concordance does not necessarily imply coevolution/coadaptation [22]. To discriminate among the latter, cross-fostering experiments (switching host ants onto novel fungi [93,94],) could demonstrate whether fungal associated (e.g., Acidobacteria and Alphaproteobacteria, Table 4) are adapted to certain hosts.