A small BAC library (36,864 clones) was constructed and screened by PCR to identify 16 clones that contained coding sequences for previously identified ESTs (Table 1). Altogether, these clones span more than 1.7 megabases (non-redundant) of the axolotl genome. BAC clones were end-sequenced using the ABI-Sanger method and then 454 sequencing technology was used to generate several thousand, high quality sequence reads for each clone (Table 2). Sequence assembly statistics (N50 and average sequence coverage) indicate that high quality assemblies were generated for each BAC; seven BAC assemblies yielded a single long contig and three BAC assemblies yielded two contigs separated by single gaps. The sequence coverage provided by the assemblies approximated the estimated size of BAC clones on agarose gels (data not shown). The remaining assemblies consisted of 6 or fewer large contigs. The reason why a few contigs yielded incomplete assemblies is not clear because different numbers of high quality reads were obtained for each BAC and contig numbers within assemblies were not correlated with sequencing depth. However, in only one case was it clear (while editing and annotating contigs) that repetitive sequences confounded contig assembly of a BAC (clone H3_4F24). Indeed, very few repetitive DNA sequences were identified overall within axolotl BACs, with retrotransposons representing the largest fraction (Table 3). These results suggest that genic regions of the axolotl are not completely structured by repetitive sequences. Annotated BAC assemblies have been deposited in GenBank [GenBank: EU686400–EU686415].

To further investigate the structure of genic regions within the axolotl genome, introns and exons were identified within BACs and compared to orthologous sequences from humans. BLAST analysis confirmed the presence of targeted EST sequences within 14 of 16 BAC assemblies. The length of orthologous coding sequences between axolotl and humans is highly conserved, as is the location of exon/intron boundaries (Table 1; Additional file 1). However, axolotl introns are strikingly longer than human introns: within five genes for which orthology could be firmly established, axolotl introns average 9454 bp while human introns average only 1938 bp (N = 32 introns compared). Further comparisons show that axolotl introns are approximately 14× larger than orthologous introns from chicken (N = 32) and 12× larger than orthologous introns from Xenopus tropicalis (N = 25; purinergic receptor P2X3 was not identified in the X. tropicalis assembly) (Additional file 1, Figure 1). Thus, non-coding genic regions are contributing significantly more to axolotl genome size than they are to vertebrates with "average-sized" genomes.

It is possible that axolotl introns are large because they contain DNA sequence classes that are unique or over-represented in comparison to other vertebrates. To test this idea, all axolotl introns and orthologous human introns were searched for self-similarity, repetitive DNAs (transposons and retrotransposons), and non-coding RNAs (including miRNAs and snoRNAs). Examination of individual self-self intron alignments and alignment of the concatenated intron dataset revealed that axolotl introns do not contain extensive tracts of repetitive DNA and are composed of largely unique sequence (Additional files 2 and 3). Multiple retroelement types were identified in axolotl introns in the selected genes but none were identified in the orthologous human introns (Table 3). Although the human genome contains many repeat classes, the only repeats identified in this sample of human introns were DNA transposons. The proportion of nucleotides accounted for by interspersed repetitive sequences is significantly higher in axolotl introns, relative to human introns (1.82% vs. 0.38%, Z = 25.6, p << 0.0001). A total of 70 candidate miRNA precursors and 21 snoRNAs (16 HACA type snoRNAs and 5 CD type snoRNAs) were identified from sense DNA strands of the axolotl (Additional files 4 and 5). The miRNAs totaled 7 kb and the snoRNAs totaled 2.7 kb for a total contribution of 2.7% to overall intron length. By way of comparison, computational searches of 39 orthologous human introns (58,313 bp) identified 6 candidate miRNAs, 1 CD type snoRNA, and no candidate HACA type snoRNAs (Additional files 4 and 5); none of these human introns contain annotated miRNAs or snoRNAs within the current human genome assembly 22. Thus, the density of predicted small, intronic ncRNAs is significantly higher in axolotls than in humans (Table 4). These analyses show that axolotl introns contain a greater diversity of transposable elements and potentially functional DNA sequence elements than human introns.

The high density of predicted miRNA structures within axolotl introns could be an artifact of the methods that were used to identify candidate miRNAs, or could represent other complex hairpin sequences that do not enter miRNA processing. To investigate this further, predicted miRNA sequences were aligned to 773,450 small RNA sequences that were recently characterized from amputated and regenerating axolotl limbs (unpublished data). This new axolotl miRNA database will be described elsewhere. Two of the predicted miRNAs from axolotl introns had stem regions that aligned perfectly with mature miRNA sequences from the axolotl limb miRNA database (Figure 2): AMmiRNA16 aligned to a single 24 bp sequence and AMmiRNA23 aligned to three independently sampled 26 bp sequences. These perfect alignments suggest that some of the predicted elements within axolotl introns are likely to be bona fide miRNA genes.

Two of the BACs in this study were selected because they contain transcripts with no known homolog in other vertebrates (Table 1). Results from microarray analyses predict a role for these "no-hit" EST contigs in two unique salamander developmental processes: metamorphosis and regeneration. The no-hit transcript that is encoded on H3_4A11 (Mex_Nohits_2574_Contig_1) is significantly downregulated during spinal cord regeneration, while the no-hit transcript that is encoded on H3_61C19 (Mex_Nohits_221_Contig_2) is significantly upregulated during spinal cord regeneration and downregulated during skin metamorphosis 2324. Although some no-hit ESTs are truncated versions of known genes, it is possible that many of the ~2000 no-hit transcripts in the Ambystoma EST database correspond to novel axolotl genes. Annotation of axolotl no-hit EST/BAC alignments supports the later hypothesis. Two novel genes, Axnovel_1 and Axnovel_2, were identified within H3_4A11 and H3_61C19, respectively. These novel genes correspond to the no-hit transcripts described above. Unexpectedly, a group of no-hit ESTs aligned to a second region of H3_4A11 that is distinct from Axnovel_1. These alignments predict a third novel gene (Axnovel_3) that has introns and is spliced (Figure 3). None of these three genes show sequence similarity to any known vertebrate gene.

To determine if these novel genes encode proteins or non-coding RNAs, EST/BAC sequence alignments were manually curated and searched for open reading frames (ORFs) using ORF finder at NCBI 25. In all three cases the longest ORF was oriented 5' to 3' relative to the EST sequences. Axnovel_2 and Axnovel_3 can be translated into long ORFs (Axnovel_2 – 786 bp and Axnovel_3 – 360 bp) that are initiated with a start methionine and terminated by a stop codon. Manual curation of Axnovel_3 revealed several small exons that were not identified by automated sequence alignments (Figure 3). The coding sequence spans eight small 5' exons that range in length from 21 to 60 bp and extends 48 bp into a longer 3' exon that contains the presumptive 3' UTR of this gene. Nearly the entire length of Axnovel_1 (249 of 313 bp) can be translated into a single ORF with a stop codon. The only in-frame methionine codon is located in the middle of the ORF, however the first codon of the longest ORF is CTG, so it is possible that this gene uses an alternative CUG start codon 26. Interestingly, orthologous EST sequences have also been sampled for Axnovel_1 in A. tigrinum tigrinum, a close relative. The A. t. tigrinum contig shares >98% nucleotide identity with Axnovel_1 and also encodes a 5' CUG. It is unclear if Axnovel_1 is translated into a functional protein or if it functions as a ncRNA; however maintenance of gene structure and sequence identity between salamander species that diverged several million years ago supports the idea that it is functional.

The most likely mechanism for the origin of novel, functional genes in the Ambystoma genome is gene duplication, as there is no evidence for whole genome duplication in A. mexicanum. It is important to consider the possibility that the large Ambystoma genome may have been shaped by a higher rate of gene duplication and fewer gene losses, and thus contain a greater overall number of genes. If paralogous loci are abundant in the Ambystoma genome, then many salamander genes are expected to show relatively more, many-to-one orthology relationships with genes from other vertebrates. To test this hypothesis, paralogs were predicted for a high quality, human-salamander ortholog dataset (N = 577), wherein primary axolotl orthologs were required to cover > 89% of the annotated length of each primary human ortholog. Approximately 86% (N = 498) of the human-axolotl gene pairs in this dataset were predicted to be 1:1 orthologs (Additional file 6). Many: many ortholog groups were predicted for 15 human-axolotl gene sets (Additional file 7) and include members from gene families that are notorious for gene duplication and gene conversion events (e.g. globins, tubulins, and actins). Of the remaining gene pairs, 2.6× more paralogs were predicted for axolotl primary orthologs (Additional files 8 and 9). Specifically, one or more human paralogs correspond to 25 human primary orthologs, yielding 32 different paralogs overall. In comparison, 39 primary axolotl orthologs correspond to 84 different axolotl paralogs. The list of axolotl specific paralogs include annexin A1 (N = 4), ferritin heavy polypeptide (N = 4), H3 histone family 3A (N = 3), calmodulin 2 (N = 2), and matrix metalloproteinase 1 (N = 2). The largest number of axolotl paralogs (N = 28) was identified for paternally expressed 10 isoform RF1 (peg10), an imprinted mammalian gene that shows sequence similarity to retrotransposons. As these axolotl paralogs exhibit higher sequence similarity to fish pol polyproteins 27 than human peg10, they probably correspond to an active retrotransposon family in the axolotl genome. Overall, these data predict 2% more duplicated loci in the axolotl genome versus the human genome (39/577 vs. 25/577), and more paralogs are predicted on average for axolotl duplicated loci (2.3 vs. 1.3). These estimates support the hypothesis of more lineage specific genes in the axolotl genome than the human genome. Assuming these genes also contain longer introns, the genic portion of the axolotl genome is predicted to exceed the total genome size of some vertebrates (see below).

A BAC Library was constructed from partially digested and size selected genomic DNA that was isolated from the erythrocytes of a single A. mexicanum female. Methods for DNA isolation and BAC library construction followed 52. 36,864 colonies were robotically picked into ninety-six 384 well plates. BAC pools were constructed and screened by PCR with 96 EST primer sets to identify 16 BACs that contained protein-coding loci.

DNA was isolated for each of the 16 BAC from 200 ml of overnight culture using a Plasmid Maxi Kit (Qiagen). All of the BACs were sequenced in a single 454 GS20 sequencing run, on one plate that was divided into subregions using a sixteen-lane gasket. Clones were also end sequenced using BigDye 3.1 chemistry and electrophoresis on an ABI capillary sequencer. Sequences were screened by Crossmatch 53 to remove vector (pCC1BAC), contaminating E. coli sequences (NC_002695.1), and additional contaminating sequences (gis:146575, 215104, 469217, 520486, 2501752) that were identified by a preliminary search of all BAC sequences against the NCBI nr database. After automated assembly using Phrap 53 (force level = 1, all other parameters set to default), all contigs over 10 kb were aligned to one another to identify contigs that contained presumptively overlapping sequence. These were visually inspected using SeqMan (DNASTAR Lasergene) and manually joined when appropriate. Contiguous sequences of assembled BACs were searched (blastn) 54 against the complete set of all known salamander transcripts at Sal-Site 55 and human RefSeq (blastx and tblastp) 56 to identify and annotate gene regions within BACs. For multispecies comparisons, the locations of orthologous intron breaks were identified by aligning (blat) 57 human RefSeq proteins to genome sequence of human (Build 36.2), chicken (galGal3), X. tropicalis (xenTro2), and salamander (current study). Self/self and all vs. all alignments of salamander intron sequences were performed using the program dottup (EMBOSS package) 58.

427,188 bp of sequence from 48 axolotl introns was searched using several algorithms. Salamander BACs, predicted introns, and orthologous introns were searched for known repetitive elements using RepeatMasker 59 and libraries of known repeat elements 60. Candidate miRNAs and snoRNAs were identified on the basis of predicted structural motifs within intronic regions of BACs. To identify candidate miRNAs, BAC sequences were searched using the program ProMiR II 61. These candidate miRNAs were further tested for probable functionality using the program MiPred 62. Candidate snoRNAs were identified using CDSeeker and ACASeeker functions of snoSeeker 63, with "modified site" files containing known human methylation and pseudouridine sites.

Reciprocal best-Blastx searching between the Ambystoma EST assembly and human RefSeq databases identified primary ortholog pairs between A. mexicanum and human. To ensure that axolotl EST contigs yielded a high quality dataset for paralog prediction, the analysis was limited to 577 ortholog pairs in which primary A. mexicanum orthologs were required to cover >89% of the annotated length of each primary human ortholog. This limited the analysis to relatively short genes whose overall lengths have been conserved during evolution. The axolotl EST contigs were assembled previously using PaCE 64 and CAP3 65 using a 90% nucleotide identity threshold to assemble ESTs into contigs 21. Singleton contigs were excluded from the analysis. Orthologous relationships were determined using an informatics approach followed by manual annotation. The conservative Inparanoid approach 6667 was used first to identify presumptive orthologs and paralogs. Primary A. mexicanum orthologs were searched (blastx) against all A. mexicanum contigs 55 and human primary orthologs were searched (blastx) against all human RefSeq entries 56. If these within species searches identified amino acid sequences that were more similar to the primary ortholog, relative to the between species comparison, they were tentatively considered lineage specific paralogs. This parsed human/A. mexicanum primary orthologs among 1:1, 1: many, many: 1, or many: many orthology classifiers. This analysis predicted 178 axolotl paralogs and 62 human paralogs. To complement the Inparanoid approach, A. mexicanum primary orthologs were searched (blastn) against EST contigs that have been assembled for a close relative: A. t. tigrinum. If 2 or more A. t. tigrinum genes were found to reciprocally best match a primary A. mexicanum ortholog and significantly similar A. mexicanum contigs, the A. mexicanum primary ortholog was considered a duplicated locus and the corresponding A. mexicanum contigs were considered paralogs. This approach verified 19 of the Inparanoid predictions for axolotl paralogs and suggested 18 novel paralogs that were not predicted by Inparanoid. Because the approaches described above do not: 1) differentiate paralogs from splice variants, 2) evaluate the quality of EST contigs, or 3) identify all paralogs, it was necessary to manually inspect the quality of all overlapping sequence alignments for presumptive 1:1 orthologs and paralogs. As a result of manual annotation, approximately 50% of the predicted human and axolotl paralogs were discarded and 4 of the 1:1 orthologs were re-classified as duplicated axolotl loci (anterior gradient 2 homolog, integral membrane protein 2b, stress-associated endoplasmic reticulum protein 1, and parvalbumin). We note that axolotl gene sequences in this analysis are substantially under sampled by comparison to the known list of human gene sequence. This sampling difference is expected to yield a minimum estimate of the true abundance of axolotl/amphibian paralogs.