To survey the fire ant gene repertoire, we generated ESTs from a normalized cDNA library derived from ants of all developmental stages and castes (workers, queens, and males) of both the monogynous and polygynous social forms. First, we sequenced the 5' ends of 22,560 clones from the cDNA library. This yielded a total of 28,113 sequence reads, since about one-fourth of all clones were sequenced twice. From this set we then removed artifactual sequences and sequences smaller than 200 base pairs (bp; after vector and primer clipping), identifying 21,715 high-quality ESTs of 522 bp average length (Table 1).

To find redundant transcripts, the 21,715 ESTs were assembled into contiguous sequences (contigs, Table 1) using the Paracel Clustering Package. A total of 14,170 ESTs were assembled into 4,319 contigs, while the remaining 7,545 ESTs remained singleton sequences. In sum, there were 11,864 gene sets, hereafter referred to as assembled sequences, that putatively represent different transcripts. However, this number is expected to overestimate the true number of transcripts represented because some non-overlapping ESTs may represent the same gene and because assembly may have failed in case of alternative splicing, sequence polymorphism or sequencing errors. Assessed with a second independent method, the number of putatively different fire ant transcripts was indeed estimated at 'only' 9,770 (see below). The average length of all assembled sequences was 600 bp.

Since some of the cDNA clones were sequenced several times, 1,262 of the 4,319 contigs are due to re-sequencing, that is, composed of sequences of a single re-sequenced clone. The remaining 3,057 contigs are 'true contigs', that is, derived from at least two independent cDNA clones (Table 1).

To obtain a tentative estimate of the percentage of 5' truncated transcripts, we compared the fire ant assembled sequences to a set of 3,951 proteins listed on the eukaryotic orthologous groups (KOG) database 23 that are highly conserved among Drosophila melanogaster, Caenorhabditis elegans and Homo sapiens. In total, 1,827 fire ant assembled sequences had a highly significant blastx hit (E ≤ 1e-20) to the Drosophila KOG proteins. Among these, 749 (41%) had regions of similarity that started within the 20 first amino-terminal amino acid residues of their Drosophila homologs with either an in-frame methionine at the same position as the fruitfly start methionine (588) or upstream of the alignment start (161). This suggests that up to 41% of the assembled sequences might have an intact 5' end, whereas the remaining 59% are probably 5' truncated.

The number of 3' truncated transcripts was harder to estimate because most cDNA clones (52.8%) were not sequenced all the way through to their 3' end (that is, the 5' sequence reads were shorter than most cDNA clones). Nevertheless, since 39.3% of all fire ant ESTs ended with a polyA sequence, up to 39.3% of our ESTs may have an intact 3' end. This is, however, likely to be an overestimate, as not all polyA sequences are true polyA tails.

Consistent with the expectation that the fire ant cDNA clones were sequenced from the 5' end, 92.2% of all assembled sequences with significant similarity to a gene in the non-redundant (nr) database were encoded on the plus strand. This estimate was obtained by counting how many times the open reading frames (ORFs) of the fire ant assembled sequences matched that of their best homologs in other organisms (see next section). However, a small percentage of the ant assembled sequences (7.8%) appeared to be encoded on the minus strand. This could be due to non-specific annealing of the SMART adaptors, to transcription of an adjacent gene pointing in the opposite orientation, or to the presence of antisense transcripts in our library.

To assess overall sequence quality, we computed the number of unresolved bases, marked as N by the base-calling program phred, present in all ESTs and assembled transcripts. The majority of sequences (83.7% of assembled sequences and 81.3% of all ESTs) had no unresolved bases. Another 15.8% of assembled sequences and 17.5% of ESTs had between one and three unresolved bases. Finally, a small percentage of sequences (0.5% of assembled transcripts and 1.2% of ESTs) had more than four unresolved bases.

We used the blastx algorithm to compare the 11,864 fire ant assembled sequences to the nr database. Of these, 2,936 (24.7%) and 3,964 (33.4%) assembled sequences matched known or predicted protein-coding genes at a cutoff expectation value (E) of 1e-20 and 1e-5, respectively (Figure 1a). By contrast, 6,431 (54.2%) had no similarity at all to genes in the nr database (E > 1). For many of these 6,431 clones, the lack of detectible similarity may be because the sequenced region does not encompass a long enough ORF to meet the blastx comparisons' cutoff of 1. This may result from 5' truncation of cDNA clones (causing ESTs to consist mostly or entirely of 3' untranslated region), from a long 5' untranslated region, or from priming in intron regions of the pre-mRNAs. Alternatively, transcripts may lack large ORFs because they are short or because they are noncoding RNAs (that is, transcripts other than rRNA or tRNA that do not code for proteins). Noncoding RNAs are now thought to make up a considerable portion of the polyadenylated transcripts found in libraries such as ours 2425. For instance, in humans 57% of all polyadenylated transcripts might be noncoding RNAs 26.

Figure 1b depicts the 'best hit' for the 3,964 fire ant assembled sequences displaying significant similarity to known or predicted protein-coding genes. The best hit was a honey bee gene 61.6% of the time. This was expected, as the honey bee is the most closely related species with a fully sequenced genome. Due to the paucity of non-honey bee hymenopteran sequences in GenBank, for only 106 (2.7%) assembled sequences was the best hit a known ant gene; and only 41 (1.0%) assembled sequences were most related to a gene from hymenopteran species other than ants or the honey bee. An additional 953 (24.0%) fire ant assembled sequences were most similar to genes from non-hymenopteran insect species. Of these, 359 and 417 had best matches to fruitfly and mosquito genes, respectively. Interestingly, a subset of 320 genes (8.1%) shared their closest similarity with vertebrates, which is an observation that has also been made for the honey bee 27. Other assembled sequences were most similar to genes from Nematoda (11) or other Animalia (26). Several had best matches to bacteria (4) or protozoa (13), possibly because these sequences were derived from microbes that infect fire ants or that have a commensal relationship with them. Alternatively, these sequences could be due to microbial contaminations acquired during sample collection. Finally, 17 assembled sequences appeared to be derived from viruses, including the recently identified S. invicta SINV-1 and SINV-1A viruses 2829.

Interestingly, for 1,341 fire ant assembled sequences the best hit was a non-hymenopteran gene (bacterial, viral and protozoan hits excluded). This could be due to extensive sequence divergence between ant-bee gene pairs or gene loss in the bee. We examined these two alternatives using the recently completed and annotated honey bee genome sequence 30. Most fire ant genes with a non-hymenopteran best hit (80.5%; 1,080/1,341) had a significant blastx hit to an annotated honey bee gene (Additional data file 1). Using tblastx, blastn or Ensembl (v38 Apr 2006 31) honey bee gene predictions, an additional 69 fire ant genes showed evidence for a potential honey bee homolog (Additional data file 1). Thus, for these 1,149 assembled sequences, sequence divergence is the likely reason for a non-hymenopteran best hit. Such sequence divergence could be due to directional selection in the honey bee lineage. The remaining 192 (14.3%) assembled sequences do not display significant similarity to the honey bee genome (Additional data file 1). This could be because some ant sequences are too short to meet the significance threshold for similarity (1e-5), extreme sequence divergence, or putative gene loss in the honey bee lineage.

We also used the blastx analysis described as an alternative method to estimate the number of unique fire ant genes sequenced. A total of 3,366 fire ant assembled sequences matched 2,772 different honey bee proteins, suggesting that 82.4% (2,772/3,366) of the fire ant assembled sequences may be unique. Thus, the 11,864 fire ant assembled sequences may represent 9,770 different genes. Assuming that the fire ant and the honey bee have a similar total number of genes (that is, 13,448 to 20,998 predicted genes, Ensembl v38 April 2006 31), this would represent approximately 46.5% to 72.7% of the genes in the fire ant genome.

In addition to the above-mentioned blastx searches to identify putative protein-coding genes, we carried out two other genomic analyses. First, to identify potential noncoding RNAs among the fire ant assembled sequences, we compared all assembled sequences via blastn to known noncoding RNAs from the NONCODE database 32 and the miRBase microRNA collection 33. Consistent with the view that noncoding RNAs are often poorly conserved across taxa 25, the vast majority of fire ant sequences had no significant hits in these databases (E > 1e-5). Only one fire ant transcript (SiJWG03CAD.scf) was highly similar (E = 3e-14) to a known human microRNA (miRBase ID: hsa-mir-594). Second, we identified 772 assembled sequences conserved between the fire ant and the honey bee that fulfilled the following conditions: no resemblance to any known protein in the nr database (blastx, E > 1e-5), a good blastn hit against the honeybee genome (E ≤ 1e-5), and no significant blastn hit against other organisms (E > 1e-5). This list of genes (Additional data file 2) is likely to include transcripts with conserved untranslated region sequence motifs and some additional noncoding RNAs. However, it may also contain ant protein-coding genes that failed to have a blastx hit because they are truncated or because their honey bee homolog failed to be predicted during genome annotation.

Provisional functional annotation of the fire ant assembled sequences was done by adopting the GO annotation of the best-matching homologs in the nr database. At a blastx E-value cutoff of 1e-5, 3,964 fire ant assembled sequences displayed matches to proteins in the nr database. Of these, 3,035 (76.6%) could be annotated into at least one of the three main GO categories (biological process, molecular function, or cellular component) and 1,617 (40.8%) were in all three. The distribution of the fire ant assembled sequences among the main subcategories is summarized in Table 2 and the full GO assignments are in Additional data file 3. The most frequently identified molecular functions were 'binding' and 'catalytic activity' and those for biological process were 'physiological process' and 'cellular process' (Table 2). In addition to the annotation through blastx searches, GO classifications were assigned to fire ant assembled sequences based on the Prosite protein domains they contain (Table 2, Additional data file 4). These two GO annotations were then contrasted with the GO annotation of the D. melanogaster genome: The relative counts of fire ant genes were significantly different (hypergeometric distribution: p < 1e-8) from the relative counts of Drosophila genes in up to 23 second-level GO categories (Table 2). This could indicate that these gene categories are over- or underrepresented in the fire ant genome relative to the Drosophila genome. Alternatively, these gene categories may simply be biased in cDNA libraries relative to genomes, for instance, because they contain mainly highly or mainly lowly expressed genes. GO groupings and subcategories can be further explored using the AmiGO feature 34 of the Fourmidable database. As the annotations are automated, all functional assignments are tentative and considered at the 'inferred from electronic annotation' (IEA) level of evidence (see 35).

The ants are classified within the order Hymenoptera, a group of insects including ants, bees and wasps. To identify Hymenoptera-specific genes, we looked for fire ant sequences that exhibited similarity only to genes from the honey bee or other Hymenoptera species. Using stringent criteria, we identified 148 fire ant sequences with strong similarity to the honey bee genome (tblastx, E < 1e-10) but no similarity to other known sequences (tblastx against non-hymenopteran sequences of the EMBL Nucleotide Sequence Database release 88; E > 1).

As the fire ant sequences are not necessarily full-length, the region of ant-bee homology, while apparently Hymenoptera-specific, may be part of a larger and phylogenetically conserved protein. To investigate this possibility, we examined the surrounding honey bee genomic sequence (±5,000 bp) of each candidate Hymenoptera-specific gene. Genes predicted by homology with other organisms were found near most of our putative ant-bee pairs. These regions of ant-bee homology may simply be fragments of known genes that diverged in ants and bees. However, for 23 ant-bee gene pairs (Table 3, Figure 2, Additional data file 5), the predicted neighboring genes are either specific to bees or are transcribed in the opposite direction. Unless the region of ant-bee homology is part of a conserved gene with a large intron (that is, >5,000 bp), these 23 ant-bee gene pairs are strong candidate Hymenoptera-specific genes.