Fourteen BAC clones from an Ae. aegypti genomic library were isolated using PCR primers specific to single-copy genetic markers 9. Shotgun sequences from each BAC were assembled into scaffolds using both the TIGR assembler 2 and Seqman 10. Scaffolds resulting from the different methods of assembly (see Materials and methods) were consistent with the others. Mate-pair inconsistencies were usually from sequences that were in repeat regions of the scaffolds. A small number of single-copy chimeric clones were observed and their elimination, along with other mate-pair inconsistencies, did not change the assembled sequences.

The majority of sequence gaps were filled using primers designed to the unique sequence flanking gaps. Some primers designed to close these gaps did not produce any PCR products and sequencing reactions with these primers using the BAC clones as template terminated at the same region or were unreadable due to polymerase slippage. All remaining gaps in the 14 BACs were flanked by highly repetitive sequence. Assembled BAC sequences were compared with the genome assembly (BLASTN) to see if they assembled in a similar manner. The gaps present in the BAC clone assemblies were either coincident with gaps present in the genome assembly or sequence diverged in the genome assembly when gaps were not present in the same region (as discussed below).

Contigs from each BAC clone were oriented on the basis of end sequences and mate-pairs. Three BACs (BAC4, BAC7 and BAC8) were each assembled into continuous sequences with no gaps. The remaining BAC sequences assembled into sets of oriented scaffolds with gaps (arbitrarily replaced by 100 Ns) (Table 1). The only BAC clones that showed differences with the assembly made at TIGR were BAC8 and BAC9. Assembled contigs seem to have been mixed during their assembly and a careful assembly (using Seqman) separated the two BAC clones into their respective scaffolds. This was verified with PCR spanning gaps and comparison to the genome assembly. The 14 BAC assemblies totaled 1,571,625 bp (approx 0.12% of the 1.3 Gb genome). The average G+C content of all scaffolds was 37.75%. Although all the sequences had a G+C around the average, BAC3 had the lowest at 27% and BAC2 had the highest at 47% (see Table 1).

Repeat masking resulted in the masking of approximately 20% of the sequence. As repeat masking here was based on protein homology, the total sequence consisting of transposon sequence is likely to be higher. Manual annotation and similarity searches with in silico predictions and EST hits with transcribed transposon sequences increased the repeat/transposon content to approximately 35%. The Feilai element 11 was the most common element, comprising approximately 38% of the repeats. Almost all transposons identified were retrotransposons.

In silico gene prediction was performed initially on the raw assembled scaffolds. A preliminary (BLASTX) analysis of these predicted transcripts (data not shown) demonstrated that there was a significant amount of over-prediction, gene-splitting and incorporation of random and transposon-based ORFs into gene models. Masking of repeat sequences before gene prediction reduced the number of gene models and this dataset was used as evidence for manual annotation.

Gene models predicted by Genscan 12 and FGENESH 13 before repeat masking often included exons derived from transposon ORFs. An Aedes gene was often split into two predictions, with the incorporation of unmasked transposon-based and other random ORFs. In addition, the ab initio generated sets of gene models (by Genscan and FGENESH) were different. However, some predicted exons did match Aedes ESTs. Several hundred ESTs were identified from the Aedes database (e < -100) as well as from the Drosophila and Anopheles datasets (e < -50). A preliminary BLAST analysis of Aedes ESTs (e = 0.0) demonstrated that a large portion of them (around 30%) mapped to transposon ORFs.

The 14 BACs were manually annotated in Apollo 14 using various tiers of evidence like ESTs and comparison to other dipteran peptides (see Materials and methods). Transcripts from the Anopheles and Drosophila genomes were used in conjunction with Aedes ESTs to limit the number of exons to those that had similarity to gene models in the other dipteran genomes. Annotations that did not possess similarity to the two dipteran genomes were also analyzed to include ORFs that may be specific to the Aedes genome as well as those that may have diverged significantly from their Anopheles or Drosophila homologs.

There were a total of 51 manual annotations (Table 2) among the 14 BAC sequences, with BAC2 having no annotated transcripts. Fifty of 51 manual annotations were found in the Ae. aegypti 1.0 Genebuild (AaegL1.1) 4 and 41 of these were identical (see Table 2). The remaining varied in several ways including differences in the 3' or 5' exon (seven transcripts), different intron/exon structure (two transcripts) or the annotation was missing in that region of the genome (one transcript). In all cases, the differences in the manually annotated models were based on Aedes EST comparisons, comparisons to annotations and ESTs in the Drosophila and Anopheles genome as well as confirmation by sequencing of PCR amplicons in a few cases. A number of transcripts differed in the length of the 3' or 5 ' UTRs. These differences were usually 10-20 bp long and not considered discordant with the gene build unless they differed by entire exons. All annotations had nucleotide matches in the Aedes genome and most had hits to Aedes ESTs. The genomic region encompassing BAC11 had two extra transcripts (AAEL03517 and AAEL02535). A protein comparison revealed that both genome-annotated transcripts were exons from a rhabdovirus nucleocapsid protein. These were not included in the list of manual annotations.

To confirm the annotation and expression of a subset of these annotations, primers were designed to all manually annotated transcripts where the prediction lacked necessary evidence. PCR was performed on cDNA obtained from all stages of the mosquito (see Materials and methods). These sequences were utilized to correct or confirm manual annotations when the curator presented multiple possible gene models or splice sites for a particular sequence. All 20 amplicons sequenced were identical to a curated gene model (see Table 2).

Eight of the 14 BAC clones had annotations present more than once in the genome assembly. This was unexpected as these BACs were specifically isolated using validated single-locus genetic markers 9. These replicated transcripts present in AaeL1.1 were virtually identical and usually present along with the same flanking transcripts in different supercontigs. To see if intergenic sequence were also replicated, the assembled BAC scaffolds were compared to the Aedes genome assembly scaffolds containing the identical transcripts. Though replicated transcripts were virtually identical, intergenic/intron sequences were usually identical on one replicate while they varied slightly on the other. These eight blocks of sequence were present in complete or partially replicated segments in different parts of the Aedes genome assembly, with only one replicate possessing identical intergenic sequence and the rest having slightly variable intergenic sequences.

Some replicated blocks were 'hybrids' of the BAC clone and the genomic duplication. This is seen in BAC14, where all five transcripts are found on two supercontigs in the same order and structure. Intergenic sequences from the first two transcripts are identical to that on supercont1.140 while the remaining transcripts have intergenic sequences corresponding to that on supercont1.146. This is also seen with BAC9, where the last transcript and its intergenic sequence are found on one scaffold while the remaining transcripts and their intergenic sequnce correspond to another scaffold - even though all transcripts are found on both scaffolds.

BAC1 was the most complicated with the five transcripts, being found on four supercontigs. All transcripts were seen in supercont1.789 while the remaining usually terminated at the end of a scaffold or had gaps which did not include all transcripts. These three transcripts were also seen with different intergenic sequences on supercont1.1137. The fourth transcript had the 3' end matching up to this scaffold and the 5' end on supercont1.393. The fifth transcript was found on supercont1.1393, whereas a sixth transcript with identical intergenic sequence was not found in the genome, although transcripts matching it but with varying intergenic sequence were found. These replicated regions were usually flanked by highly repetitive DNA and/or gaps or were present at the end of a supercontig.

When compared with the Anopheles and Drosophila gene sets (Table 3), 50 and 43 Aedes transcript annotations had orthologous transcripts in the Anopheles and Drosophila gene sets, respectively. The genes from the two other dipteran genomes that were similar to the manual annotations were almost always orthologs of each other (determined by reciprocal BLASTs) 4. Although most Aedes annotations had a one-to-one relationship in the other genomes, some matches were to genes from multigene families. In some cases, the primary BLAST match was much better than the rest and in these cases, an ortholog was postulated. In cases where a number of transcripts matched the manual annotation with similar e-values, orthologs could not be predicted. A single manual annotation did not have any similarity in either genome, and when compared to other dipteran datasets with less stringent parameters it demonstrated similarity to an Ae. albopictus salivary protein.

To compare gene sizes between the two mosquitoes, the amount of sequence covered by the orthologous genes in Aedes and Anopheles were compared. Single-exon genes were usually the same size; however, the size of multiexon genes was directly proportionate to the number of introns in Aedes. On average, Aedes genes were about 3.9 times the size of their Anopheles orthologs. Only one Aedes BAC sequence demonstrated any degree of synteny with Drosophila. BAC11 had two adjacent transcripts that were found to be next to each other in the Drosophila genome. Of the 11 BACs with more than one annotated transcript, nine sequences demonstrated synteny with the Anopheles genome. Overall, 38 of the 50 transcripts included in these BACs demonstrated synteny in 10 blocks.

For a summary of each BAC clone assembly and analysis please see Additional data file 1.

The quality of these BAC assemblies is critical for a valid assessment of the genome assembly and the automatic gene-annotation pipeline. To enable this assessment, each BAC clone was individually assembled using two assembly algorithms and the resulting duplicated assemblies were compared to make sure that contigs were identical. In addition, all BAC sequences were assembled together to ensure that they sorted independently into the contigs corresponding to individual BAC clones. These stringent assemblies revealed that the sequence of BAC9 (GenBank: AC149799), which was submitted to GenBank before this analysis, had contigs in it that were from BAC8 (GenBank: AC149798). A stringent analysis of these BACs in particular enabled their correct assembly. It was interesting to note that gaps present in the final BAC scaffolds were identical to those present in the genome assembly. We believe that the high repeat content of the sequence in the remaining gaps produces tertiary structures that are not conducive to sequencing. A high G+C content may also contribute to this phenomenon. As a result, we were unable to close several gaps. The 14 final assemblies were confirmed both with PCR, sequencing and a comparison to the genome assembly.

Assembled and oriented BAC scaffolds were masked for repeat sequence to characterize the transposon content as well as to enable a more efficient in silico gene model prediction. Gene-prediction algorithms cannot distinguish transposon ORFs, resulting in their being annotated along with species-specific ORFs. Resulting gene models may not be indicative of real genes, as genes could be split, merged or have extra exons. Initial repeat identification demonstrated that the Aedes genome has an unusually high repeat content 15. Repeat masking 1617 was performed using multiple repeat datasets to maximize the number of repeats identified. An initial analysis of in silico gene annotations derived from the masked sequences revealed that a number of transposons were not identified as a result of the incomplete cataloging of the Aedes transposon dataset. This is seen with BAC2, where there were no transcripts annotated on the assembled sequence but gene prediction on repeat-masked sequence suggested the presence of up to 18 transcripts that are derived from unmasked transposon ORFs. The high repeat content of this genome is particularly interesting and impacted on the sequencing, assembly, in silico and manual annotation presented in this study. The proper identification of a genome's repeat content is vital as it impacts on these analyses that form the basis of genomic studies.

Manually curated genes are generally considered to be the highest tier of gene models for genome annotation and training datasets. Annotations were based on several sets of data that include manual inspection of species-specific ESTs and comparative data. A portion of the ESTs mapped to transposons, complicating the manual annotation. These transposon-related ESTs can be attributed either to active transposition or to genome-related transposition silencing. As a result, in silico gene prediction on unmasked sequence resulted in a higher number of predicted genes (around 4 times more), while the presence of unidentified repeat sequences on masked sequence resulted in over-prediction as well. Although most of the ORFs from the 51 final manually annotated gene models were present in these predictions, transposons present in intergenic sequences led to the splitting and merging of exons along with transposon ORFs. Though the resulting gene predictions from the two ab initio gene-prediction programs were not alike, they did capture similar exons. These in silico predicted exons were helpful in determining splice sites, along with EST and comparative evidence during manual annotation. The large repeat content in this genome highlights the importance of proper repeat identification and masking before gene prediction in annotation pipelines.

Gene models (see Table 2) were predicted only if they had supporting EST and comparative evidence and did not overlap with sequence that was homologous to transposons. We do not believe we have eliminated any 'domesticated' transposons, although this remains a possibility.

PCR performed on a cDNA library confirmed expression of a subset of transcripts, enabled a sequence comparison of the expressed transcripts with the manual annotations and also introduced an annotation quality-control step. To enable the most thorough expression analysis, the cDNA library was derived from RNA extracted from all stages of mosquito development (see Materials and methods). This molecular verification points to the importance of manual annotations in a genome-annotation pipeline that can not only verify the quality of the auto-annotation but also provide a set of high-quality transcripts that can be used to develop and improve it.

All manual annotations were compared to the Aedes genome assembly and Genebuild - AaegL1.1 (see Table 2). Almost all manually annotated transcripts were found in the Aedes gene build. Differences between the manually annotated models and the transcripts from the gene build included a transcript missing, extra transcripts in the gene build and differences in annotation (see Table 3). When looking at nucleotide similarity (BLASTN), only one transcript on BAC7 (number 20, see Tables 2, 3) did not have a match in the gene build, even though it had perfect nucleotide match in the genome. This annotation belonged to a multigene family (histone H3) and had several almost identical annotated transcripts elsewhere in the Aedes genome. The sequence flanking this gene model consisted of transposon sequence, and the entire region was labeled as repetitive in the genome assembly 4. This transcript, present in multiple copies in the genome as well as being flanked by transposon sequence, was masked before mapping of ESTs to the assembled genome and consequent gene annotation. This points to the importance of differentiating multicopy gene sequences versus those that are homologous to transposons and to the necessity of a comprehensive catalog of the Aedes transposon dataset.

This set of manually annotated transcripts enables a quality check of the Aedes genome auto-annotation. Approximately 12% of the manually annotated transcripts possessed minor differences from their auto-annotation counterparts, indicating a high-quality genome annotation effort. These differences, as well as the identification of a rhabdovirus nucleocapsid incorporation, highlights the importance of manual annotation and points to a few issues an auto-annotation pipeline may have.

The 14 BACs were identified from single-locus genetic markers 9. However, eight of these blocks of genomic sequence possessed transcripts (including the single-copy markers) that were replicated in the genome assembly, along with flanking transcripts, in the same order and structure (see Table 2). A further analysis of the single-copy genetic markers in Severson et al. 9, reveals that 26 of the 146 single-copy genetic markers used are present more than once in the genome assembly (data not shown). The high percentage of repeated single-copy markers from a well-known study presents the possibility that these duplicated assembly regions may have resulted from actual segmental duplications, haplotype polymorphisms or misassemblies.

If these regions represented segmental duplications, they would have to be physically close to each other - as the genetic markers have been extensively used and the genetic positions calculated have been well characterized and fall out as one genetic locus 9. However, the genome assembly has these repeated single-copy markers sometimes localizing to different supercontigs (suggesting a greater distance between them). These different supercontigs sometimes also have markers on them that localize to different linkage groups. This suggests that even though there may be a number of repeated markers present close to each other, a certain degree of misassembly would explain how a single-copy genetic marker would be duplicated on another supercontig or present along with a genetic marker from another linkage group. These events can be explained by the high repeat content of this genome and the presence of repeats flanking these regions, further complicating their proper assembly. It was interesting to note that shotgun sequences from identical repeats were some of the only discrepancies in our assemblies in this study. However, the relatively small size of these assemblies enabled us to completely assemble the BACs correctly.

If these regions represent haplotype polymorphic regions, they should demonstrate genetic drift and therefore a certain amount of sequence variation. These differences would result in the haplotype regions assembling into two scaffolds and therefore complicating the assembly. This phenomenon is seen in polymorphic regions of the A. gambiae genome (demonstrating 95-99% similarity) that assembled independently of each other (48 and R. Bruggner and M. Hammond, personal communication). Strains used for sequencing are usually inbred to eliminate usual genomic variation to enable an easier assembly and analysis (the strain of Ae. aegypti used for genome sequencing (LVP^ib12) was inbred for 12 generations from an already inbred strain). However, this cannot eliminate the presence of balanced polymorphisms where homozygous regions result in lethality - a phenomenon extensively used in Drosophila genetics. Haplotype polymorphic regions are expected in genome assemblies; however, their negative effects on assembly and analysis can be minimized by proper strain selection and inbreeding. The replicated regions seen here were not precise duplications, as a comparison of the entire nucleotide sequence revealed intergenic differences between the replicated blocks. A comparative analysis revealed that 23 of the 28 transcripts encompassed by these 'replicated' BACs were single copy in both the A. gambiae and D. melanogaster genomes, again suggesting a single-copy nature. The variation seen between replicated regions, the 'hybrid' nature seen between the BAC sequence and the genomic replicates, the characterization of the markers and encompassed genes as being single copy in Aedes 9, as well as in Anopheles and Drosophila, lead us to believe that these replicated regions in the genome assembly represent polymorphic haplotypes coupled with some misassembly resulting from flanking repeat sequence. There remains the possibility that some of these regions are actually duplicated in the genome and are present close to each other.

The replication of an unusually high percentage of genomic blocks experimentally shown to contain single-copy sequences (57% (8 of 14)), indicates the presence of an assembly issue which affects the number of gene predictions in the gene build and the relation of various scaffolds to each other. This phenomenon also emphasizes the importance of strain selection and proper inbreeding to enable an easier genome assembly. The proper characterization of these probable haplotype regions would enable a better genome assembly and mapping of scaffolds to linkage groups.

All manually annotated transcripts were compared to the Drosophila and Anopheles gene sets (see Table 3). Only one annotation (number 19) did not show homology to Anopheles or Drosophila proteins with the search parameters used. This transcript did demonstrate similarity to an Aedes salivary protein (D7cclu23-like salivary protein). When the search parameters were relaxed, the primary hit to Anopheles is an odorant-binding protein (OBP49). A salivary- or odorant-related gene would be expected to have significantly diverged from Anopheles and even further diverged from Drosophila homologs and would not show a high degree or any similarity in the stringent comparative searches used.

Of the remaining 50 transcripts, 50 and 43 demonstrated similarity to the Anopheles and Drosophila gene sets, respectively. Seven manual annotations that did not have any similarity to the Drosophila genome (but did to the Anopheles genome) may have either been lost in the lineage that gave rise to the higher dipterans or have significantly diverged from their homologs. Most transcripts had a one-to-one relationship with a gene in the other dipteran genomes. In general, most manually annotated transcripts were similar in length and amino-acid identity to the other dipteran transcripts. The transcripts that had similarity to Anopheles transcripts had an average of 72% identity to the manually annotated transcripts with a range 32-100% identity. The 49 Drosophila transcripts had an average of around 60% identity with a range 31-97%.

The 51 transcripts represent a gene density of one gene every 30.8 kb, which is considerably lower than in A. gambiae 7 or D. melanogaster 6. Aedes genes possessed larger intergenic sequences, resulting in multi-exon genes being about four times as large as their Anopheles counterparts. This lower gene density seen in Aedes can be related to its larger genome size, with a much higher repeat/transposon content.

To look for syntenic relationships, the 11 blocks of transcripts (those with more than one annotation) were compared with the Anopheles and Drosophila genome (see Table 3). Ten of the blocks had transcripts in them that were similarly clustered in the Anopheles genome, whereas only one cluster of two adjacent transcripts was found in Drosophila. Overall, 38 transcripts in 10 blocks demonstrated the closer relationship and shorter divergence times between the two mosquitoes 18.

Syntenic studies between genomes can have important applications, including the verification of transcripts and gene annotations. Transcript 38 did not have similarity to any anopheline annotation but possessed significant similarity to a Drosophila transcript. This BAC sequence demonstrated synteny to the Anopheles genome and was the only transcript missing, although the nucleotide sequence corresponding to this sequence was present. Further investigation revealed that the transcript corresponding to it was removed in the last Anopheles gene build. The presence of this transcript in both the Drosophila and Anopheles genome, as well as the corresponding nucleotide sequence in the Anopheles genome, suggests that this anopheline transcript needs to be reinstated.