Sample preparation and sequencing

No animal ethics approvals are required to conduct research on ants. Research in Brazil was covered by Brazilian Council of Research and Scientific Development permit Processo CNPq 001884/2012-3 and Instituto Chico Mendes de Conservação da Biodiversidade (ICMBio) collecting permit 14789–6. Research in Peru was covered by Ministerio de Agricultura Instituto Nacional de Recursos Naturales INRENA Autorización No. 034-2004-INRENA-IFFS-DCB; Modificación a la Autorización No. 034-2004-INRENA-IFFS-DCB; Ministerio de Agricultura Instituto Nacional de Recursos Naturales INRENA Autorización No. 12 C/C-2004-INRENA-IANP; Carta No. 553-2014-MINAGRI-DGFFS/DGEFFS; Ministerio de Agricultura Instituto Nacional de Recursos Naturales INRENA Autorización No. 088-2005-INRENA-IFFS-DCB; CARTA No. 0217-2012-SERNANP-JRNTAMB; Resolución del Jefe de la Reserva Nacional Tambopata No. 020-2012-SERNANP-JEF; Permiso para Fauna y Flora Silvestre No. 004905-AG-INRENA; Permiso para Fauna y Flora Silvestre No. 006771-AG-INRENA; and Permiso para Fauna y Flora Silvestre No. 009154-AG-DGFFS.

Specimens of three Sericomyrmex species were collected from live colonies in the field and preserved in RNAlater. Specimens of A. megacephala were collected from a live laboratory nest and also preserved in RNAlater. Voucher specimens for each of the species sequenced are deposited in the Department of Entomology, National Museum of Natural History, Washington, DC, USA (Table 1), as recorded in the NMNH K-EMu database (http://collections.nmnh.si.edu/search/ento/). In order to ensure sufficient quantities of RNA, we included ten workers per sample for each Sericomyrmex species and five workers for the larger species, A. megacephala, crushing them with sterilized wooden sticks to enable the RNAlater to rapidly penetrate the integument. All specimens were stored at -20°C until RNA extraction. We extracted RNA using the Promega SV Total RNA Isolation System, following the standard protocol. The Institute for Bioscience and Biotechnology Research (IBBR) at the University of Maryland prepared libraries using the standard Illumina TruSeq RNA Sample Preparation protocol (all libraries were normalized and redundant rRNA was removed), and performed 100bp pair-end sequencing on Illumina HiSeq.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Specimen data.

https://doi.org/10.1371/journal.pone.0151059.t001

Data cleaning and assembly

We concatenated the raw data and performed a quality check with FastQC [17] before and after trimming (Fig 1). Raw reads were cleaned and trimmed with Trimmomatic [18] and the resultant cleaned reads were assembled with Trinity [19,20] using the default parameters. Basic assembly statistics (number of transcripts, average transcript length, and N50) were obtained from Trinity. We processed the resulting assemblies in TransDecoder [21] to identify candidate coding regions within the transcripts. Predicted peptides were then filtered with CD-HIT-EST (98% global similarity) [22] to remove redundant sequences. After removal of duplicates, peptides were further filtered in order to select only one peptide per putative transcript by choosing the longest ORF (Open Reading Frame) per Trinity subcomponent with a custom-made Python script [23]. This step ensured that we had removed variation in the coding regions of our assemblies due to alternative splicing, closely related paralogs, and allelic diversity. Peptide sequences for the seven sampled ant genomes were filtered at 98% similarity in CD-HIT-EST. To calculate coverage we multiplied the number of raw reads with the expected length (100 bp), and divided that by total number of bases in the assembly [24].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Workflow summary.

https://doi.org/10.1371/journal.pone.0151059.g001

Orthology Assignment and Alignment

We identified orthologs using OMA stand-alone v.0.99w software [25,26] with default settings on 100 cores on the Smithsonian Lattice high-performance computing cluster (Linux-based with AMD processors). We constructed an amino acid supermatrix by concatenating the set of OMA groups containing all taxa (1,317). Each putative ortholog group (from now on “gene”) was aligned individually using MAFFT [27]. Aligned genes were then trimmed with Gblocks [28] to cull regions of dubious alignment.

We separately ran an orthology assignment in OMA for a data subset containing only the Sericomyrmex transcriptomes. For this we used TransDecoder to identify the nucleotide sequences for the candidate protein-coding regions and performed the same filtering as described above with the larger dataset on the resultant nucleotide sequence data. Individual gene alignments were trimmed with Gblocks [28] and analyzed in Geneious v.8 [29].

Phylogenetic analyses

Maximum-likelihood inference was conducted with PhyML-PCMA [30,31]. We selected 10 PCs (principal components) in the PhyML-PCMA analyses and used empirical amino-acid frequencies. PhyML-PCMA estimates a model through the use of a principal component (PC) analysis; in this case using 10 PCs. Bootstraps were calculated in PhyML-PCMA for 100 replicates. Concomitantly, tree searches were conducted in PhyloBayes MPI 1.4e [32] using the site-heterogeneous CAT + GTR model of evolution. Three independent chains were run for 1295–1445 cycles, and the initial cycles discarded as burn-in were determined for each analyses using the “tracecomp” executable, with convergence assessed using the maximum bipartition discrepancies across chains (maxdiff < 0.1).

Divergence dating

Divergence dates were estimated using the Bayesian relaxed molecular clock approach as implemented in PhyloBayes v.3.3f [32] applying an auto-correlated model of clock relaxation [33,34]. Three calibration constraints (S1 Table), based on a recent study of myrmicine ants [35], were used with soft bounds [36] under a birth-death prior in PhyloBayes. PhyloBayes was run for 32,664 cycles, sampling posterior rates and dates every 5 cycles. The initial 5000 cycles were discarded as burn-in. We excluded the outgroup taxon Apis mellifera from final analyses because we considered it to be an inappropriate outgroup for the purposes of divergence dating. All other taxa are members of a single subfamily, the Myrmicinae, which occupies a highly derived position within the ant family Formicidae, whereas A. mellifera, the honey bee, is a highly derived member of a highly derived family within the superfamily Apoidea. The most recent common ancestor of the lineages containing the Apoidea and the Formicidae dates to at least 140 mya, whereas the ancestor of the Myrmicinae, the focus of our dating analyses, dates to ~99 mya [35,37,38]

Annotation and Ortholog Hit Ratio

We used BLASTx [39] (cut-off E-value 1e^-5) to compare the unfiltered, assembled transcripts (Trinity output) against the non-redundant (nr) protein database of NCBI. Resulting xml files were used as input for CLC Workbench (CLC Inc., Aarhus, Denmark). We performed functional annotation and mapping to GO terms, using Blast2GO with default settings, in order to summarize functional categories of the genes, annotate our data set, and determine the quality of our transcriptomes. We also used the resultant xml output to run Orthology Hit Ratio calculations following previously described methods and scripts [40]. This analysis estimates the degree to which a transcriptome is fully sequenced and assembled by comparing the length of the contigs that recovered BLAST hits with the length of their top BLAST hits [41,42].

Gene search

Based on previous studies of attine ant genomes and ant transcriptomes [10–12,43] we considered the following genes or gene families of particular interest: genes in the arginine metabolism pathway (arginase, nitric oxide synthase, argininosuccinate synthase, and argininosuccinate lyase), detoxification genes (cytochrome P450 monooxygenase), hexamerins (hex 70a, hex 70b, hex 70c and hex 110), desaturase, RYamide, and chitinases. The nucleotide or protein sequence for each of these genes was downloaded from NCBI GenBank or, in the case of the desaturase gene, obtained from the supplemental material of a previous study [43]. We used sequences from the closely related species Atta cephalotes and Acromyrmex echinatior for searches for arginase, nitric oxide synthase, and cytochrome P450. For chitinase searches we assembled protein and nucleotide sequences that were used in similar analyses in a study of the Atta cephalotes genome [10], for a total of ten different chitinase-like proteins from Acromyrmex echinatior, Nasonia vitripennis, and A. mellifera (S4 Table). For argininosuccinate synthase and argininosuccinate lyase, which are lost in leaf-cutter ants, we used copies from the closely related myrmicine ants Solenopsis invicta, Pogonomyrmex barbatus, and Wasmannia auropunctata. For hexamerin searches we used Apis mellifera sequences in order to repeat the methods of previous studies [10] and because some of the attine ant hexamerin sequences were unavailable. For searches for ryamide, which is absent in ants, we used a Drosphila melanogaster sequence. Full sequences of all genes used in our analyses, with GenBank accession numbers, can be found in S1 Text. We used BLAST+ standalone [44] to manually create a BLAST database for each of the transcriptomes separately (using unfiltered, assembled transcripts) and searched the created databases for each of the query sequences with the cut-off E-value 1e^-5. Returned hits were then BLASTed against NCBI to confirm that they had returned the same protein.

Orthology assignment

We used OMA standalone software [26] to place predicted genes into orthologous groups. OMA’s advantage over standard bidirectional best-hit approaches is that OMA’s algorithm uses evolutionary distances instead of scores, considers distance inference uncertainty, includes many-to-many orthologous relations, and accounts for differential gene losses [26,45]. The number of orthologs obtained for the complete dataset (ten ingroup and three outgroup taxa) was 1,317 and the number of phylogenetically informative genes (n>4) was 11,447. The final concatenated matrix contains a total of 649,095 amino-acid sites present in all taxa.

Functional annotation

The percentage of reads that had at least one BLAST hit varied between 50% and 55.8% of total transcripts (Table 2), which is slightly higher than in similar transcriptome studies [40,43]. Final number of annotated sequences varied between 21,555 and 26,808 sequences. Our data were annotated with a wide range of functional categories represented on all levels of the Gene Ontology database and were comparable to other similar studies, with no biases toward any particular category (Fig 2, Table 2). For all four transcriptomes the top BLAST hit species was recovered as Acromyrmex echinatior, which is a higher-attine leaf-cutting species, a very close relative of Sericomyrmex, and a close relative of Apterostigma. This was also the species with the largest number of total BLAST hits for the three Sericomyrmex species. For A. megacephala the top BLAST hit species was the non fungus-farming ant Cerapachys biroi.

Phylogeny and dating

Results from phylogenetic analyses of this dataset (Fig 4) are congruent with existing phylogenies, including a monophyletic Sericomyrmex clade that is the sister of Trachymyrmex zeteki, and with A. megacephala as the sister taxon to all other attine ant taxa included in this study. Considering the large number of characters in this dataset, the branches subtending the three Sericomyrmex species are reconstructed as remarkably short in both analyses. This result indicates a very low genetic divergence between Sericomyrmex species. To investigate this further we ran a separate orthology search including only the three Sericomyrmex species, which revealed 4,217 orthologous genes present in all three samples, varying in length from 303 to 10,530 bp. Of those 4,217 loci, 92.71% (3,910 sequences) are 99% to 100% pairwise identical, as defined by Geneious [29]. Of the remaining orthologs, 290 are between 90% and 99% pairwise identical. Combining these results, 99.59% of all orthologs shared by these three Sericomyrmex morphospecies are at least 90% pairwise identical. This result is unexpected given that we selected the most molecularly and morphologically divergent Sericomyrmex samples available for RNA extraction. Based on morphology and multiple gene sequences, of the samples chosen for this study S. cf. luederwaldti and S. cf. mayri were considered a priori to be more similar, so the high similarity of their transcriptome sequences is less surprising. However, S. cf. parvulus is morphologically distinctly different from the other two species and is considered to be a member of one of the most basally diverging lineages of Sericomyrmex species (Ješovnik, unpublished). Even though phylogenetic analyses of the transcriptome data agreed with prior analyses regarding these relationships, i.e., that S. cf. luederwaldti and S. cf. mayri are more closely related to each other than either is to S. cf. parvulus, the degree of divergence separating S. cf. parvulus from those two species is much lower than expected based on divergences separating similarly related taxa, e.g., some leaf-cutter species. In general, the transcriptomes indicate that even morphologically divergent Sericomyrmex species are separated by remarkably low genetic distances, which means that accurately recovering species boundaries may require more variable data from non-coding regions. Importantly, this pattern, also supported by our divergence-dating analyses (see next paragraph) indicates that Sericomyrmex has rapidly radiated into multiple, morphologically distinguishable species that occupy a broad geographic distribution [4,46] with only a small degree of accompanying genetic divergence, in contrast to most other comparable ant groups, including its sister taxon, the Trachymyrmex iheringi species group. This conclusion must obviously be investigated further, given that transcriptomes from only three Sericomyrmex species were included in this study. It will be particularly important to test whether this rapid radiation may have been driven by coevolution with a specialized clade of higher attine fungi, as has been suggested for Atta [2,47], especially since recent research suggests the possibility of high symbiont fidelity between Sericomyrmex species and a single species of fungal cultivar [48]. Another potentially important factor in this radiation could be major changes in genetic architecture such as chromosome duplications and rearrangements, because data from two species indicate that, compared to most other attine ants, Sericomyrmex has an unusually high number of chromosomes and an unusually large genome [49,50].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Time-calibrated phylogeny of attine ants.

Highlighted taxa are novel transcriptomes obtained for this study. This topology was recovered in both ML and Bayesian analysis. Asterisks (*) at nodes indicate bootstrap values of 100 and posterior probabilities of 1.0. Green error bars at nodes indicate minimum and maximum age estimates; the time scale at the bottom is in million of years.

https://doi.org/10.1371/journal.pone.0151059.g004

The divergence dating analysis, which included ant species only (the outgroup A. mellifera was excluded in the dating analysis), recovered the stem-group age (i.e., the earliest possible origin) of fungus-farming ants as 66.7 million years ago (mya) and the crown-group age (i.e., the latest possible origin) as 53.6 mya. The crown-group age estimate is consistent with previous estimates of 50–60 mya [2,9,35,50,51], whereas the stem-group age estimate is slightly younger than a stem-group estimate from a recent study (74.6 mya) of a large genomic data set [12]. However, these results are in some cases imprecise and are not directly comparable because neither included taxa from the closest non-fungus-farming relatives (the sister group) of the attine ants [35] and only this study included representatives of both clades resulting from the basal-most divergence at the origin of the fungus-farming ants. Our age estimate (33.3–53.6 mya) of this divergence between the Paleoattini (including the genera Myrmicocrypta, Apterostigma, and Mycocepurus) and the Neoattini (including all other attine ant genera) should be interpreted with caution because our taxon sampling is arguably inadequate for the task. Our taxon sampling for the higher attine ants is better and represents the first genomic data set that includes Sericomyrmex. We estimate the crown-group age of the higher attine ants to be 29.2 (31.3–27.4) mya; the crown-group age of the Sericomyrmex+Trachymyrmex zeteki clade (i.e., the Trachymyrmex iheringi group) as 22.6 (25.3–20.4) mya; and the crown-group age of the leaf-cutting ants (genera Atta and Acromyrmex) to be 17.9 (20.4–15.6) mya (Table 3). For comparison, Schultz and Brady [2] recovered ~16 mya, ~10.5, and ~8 mya mya for those same nodes, respectively, and Nygaard and colleagues estimate ~17 mya for the leaf-cutter origin [12]. Our dating analyses, based on transcriptome data, are most likely overestimating divergence times due to the very small taxon sample. This possible overestimation notwithstanding, the crown-group age of Sericomyrmex is reconstructed as very recent, estimated at 4.8 mya for the age of the ancestor of all extant species. For comparison, the stem- and crown-group ages of the clade containing the attine host/parasite species pair Mycocepurus goeldii and M. castrator are estimated to be 2.04 and 3.31 million years, respectively [52], and crown-group divergences between sister species pairs in the Cyphomyrmex wheeleri group range from 5.3 to 7.0 mya [3]. The branch separating Sericomyrmex from the most recent common ancestor it shares with the Trachymyrmex iheringi group (represented here by T. zeteki) is very long (Fig 4), which might be explained by taxon undersampling, especially if there are more extant Sericomyrmex species that could split that long branch if they had been included. We consider this unlikely, however, because the three sampled Sericomyrmex species were chosen to represent the full range of diversity in the genus both morphologically and molecularly and because other phylogenomic studies currently in preparation, employing much denser taxon sampling, corroborate that the genus Sericomyrmex is the sister group of the entire T. iheringi clade, i.e., that the branch subtending Sericomyrmex is unoccupied by other Trachymyrmex species.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Crown-group and stem-group age estimates.

In units of million years ago, with standard error given in parentheses. An asterisk (*) indicates a posterior probability of 1.

https://doi.org/10.1371/journal.pone.0151059.t003

Gene family searches

In order to further investigate the sequenced transcriptomes, we chose several genes of interest based on previous ant genome and ant transcriptome studies [10,11,43] and performed separate searches for those genes. In general, the similarity of the results of our gene BLAST searches to those of other attine ant genomic studies serves as a confirmation of the completeness of our transcriptomes.

Arginine metabolic pathway genes.

The first genomic studies of the higher-attine leaf-cutting ant genera Atta and Acromyrmex demonstrated losses of two genes in the arginine anabolic pathway, whereas all other ants for which data are available, including species in the same subfamily (Myrmicinae) as the attine ants, have functional copies of those genes [10,11]. It was suggested that the leaf-cutting ants, or even all higher attine ants, had become obligately dependent on their fungi for arginine. Such a metabolic division of labor is perhaps not surprising in the highly derived higher attine ants, given that their fungi are also obligate symbionts, and similar examples are known from other mutually obligate symbioses [53–55]. However, a more recent study, which included one lower attine ant species that is relatively closely related to the higher attines, indicates that those two genes, argininosuccinate synthase and argininosuccinate lyase, were lost prior to the origin of the higher attines [12], begging the question of at what point in attine evolution the ants became obligately dependent on their fungal cultivars for the amino acid arginine.

Our BLAST searches for argininosuccinate synthase and argininosuccinate lyase in the three Sericomyrmex species did not retrieve any hits, which, because Sericomyrmex species are higher attines, was to be expected. However, BLAST searches also failed to find these genes in A. megacephala. A. megacephala is a member of the Paleoattini, one of two sister clades formed by the basal-most divergence in the fungus-farming ant phylogeny. The other sister clade, the Neoattini, contains Sericomyrmex and all previously studied attine ant species, so the absence of these two genes in a paleoattine indicates that the dependence of attine ants on their fungal cultivars for arginine arose in the common ancestor of all fungus-farming ants. This result is unlikely to be due to inadequate transcriptome data for two reasons. First, our Ortholog Hit Ratios are high, and, second, searches for two other enzymes in the arginine metabolic pathway, the catabolic genes arginase and nitric oxide synthase, produced hits in all of our transcriptomes (Table 4). Functional copies of these two genes are likewise present in the genomes of Atta and Acromyrmex [10,11].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Gene searches.

Number of gene copies and isoforms per species.

https://doi.org/10.1371/journal.pone.0151059.t004

Desaturases

The desaturase gene family plays an important role in the synthesis of cuticular hydrocarbons (CHC), which are one of the key elements of nestmate recognition in social insects. Using the desat1 gene of the ant Formica exsecta [43] as a query sequence, we found five copies of desaturase in A. megacephala and eight copies in each of the three Sericomyrmex species, all with various numbers of isoforms (Table 4). When the found copies were BLASTed back to the NCBI protein database, the highest similarity was found with delta(Δ)11 desaturase genes of other ants. This finding is comparable to that of a similar study of the Atta cephalotes genome, in which six out of eleven identified desaturase-like genes matched Δ11 desaturase [10].