A multi-organ transcriptome resource for the Burmese Python (Python molurus bivittatus)

Background Snakes provide a unique vertebrate system for studying a diversity of extreme adaptations, including those related to development, metabolism, physiology, and venom. Despite their importance as research models, genomic resources for snakes are few. Among snakes, the Burmese python is the premier model for studying extremes of metabolic fluctuation and physiological remodelling. In this species, the consumption of large infrequent meals can induce a 40-fold increase in metabolic rate and more than a doubling in size of some organs. To provide a foundation for research utilizing the python, our aim was to assemble and annotate a transcriptome reference from the heart and liver. To accomplish this aim, we used the 454-FLX sequencing platform to collect sequence data from multiple cDNA libraries. Results We collected nearly 1 million 454 sequence reads, and assembled these into 37,245 contigs with a combined length of 13,409,006 bp. To identify known genes, these contigs were compared to chicken and lizard gene sets, and to all Genbank sequences. A total of 13,286 of these contigs were annotated based on similarity to known genes or Genbank sequences. We used gene ontology (GO) assignments to characterize the types of genes in this transcriptome resource. The raw data, transcript contig assembly, and transcript annotations are made available online for use by the broader research community. Conclusion These data should facilitate future studies using pythons and snakes in general, helping to further contribute to the utilization of snakes as a model evolutionary and physiological system. This sequence collection represents a major genomic resource for the Burmese python, and the large number of transcript sequences characterized should contribute to future research in this and other snake species.


Background
A major innovation enabled by next-generation sequencing technologies has been the ability to assemble extensive genomic resources for non-traditional model species. This expanding ability has in turn enabled a renaissance in the use of diverse model species to deliver novel insights not previously possible. Among the emerging model species archetypes are species that demonstrate extreme phenotypes. There is widespread interest in generating necessary genomic resources to facilitate research on these new models of extreme vertebrate phenotypes.
One such group for studying extreme phenotypes are the snakes. Snakes have become increasingly prominent model systems [1], primarily because they represent a vertebrate model system that possesses numerous important extreme adaptations at the morphological and developmental [2][3][4], physiological and metabolic [5,6], and molecular levels [7][8][9][10][11]. The Burmese python (Python molurus bivittatus) in particular has become a focal model system for studying extreme physiological remodelling and metabolic fluctuations that accompany feeding [12][13][14]. A major problem in studying snakes, however, is that they are highly divergent from other model vertebrate systems that already have genomic resources. The closest vertebrate to snakes with an available complete genome sequence is the Anolis lizard (just now being formally published [15]), which last shared a common ancestor with the python~166 MYA [9,16].
Otherwise, the next closest vertebrates with complete genomes are birds (chicken, finch), which last shared a common ancestor with snakes~275 MYA [16]. Although some studies have utilized high-throughput sequencing with short reads to study snake transcriptomics, prior to the Anolis genome they have been constrained to using bird reference genomes, and have not produced sets of assembled and annotated transcripts [17,18]. Other than the Anolis genome, the only existing genomic/transcriptomic resource relevant for studying snakes is a transcriptome data set for the garter snake (Thamnophis sirtalis) [19]. Although more closely related than the lizard, this species is also highly evolutionarily distant from the python, as these two species last shared a common ancestor 60-100 MYA [9,16]. Thus, to advance prospects for research utilizing pythons as a model system, a python-specific transcriptome set is needed.
Here, we have assembled a moderate-sized set of transcriptome data from 454 pyrosequencing to create a robust transcriptome reference for future studies utilizing pythons as models for research. We specifically chose to use the more expensive per-base 454 platform for its longer read lengths, which should favor higher assembly accuracy and de novo assembly of transcripts. Since our primary goal was to establish a relatively large well-annotated baseline set of snake transcript sequences, we sequenced cDNA libraries generated from multiple sources (heart and liver) and various time points before and after feeding. These sequences were assembled into a combined set of annotated transcript contigs.

Results and Discussion
Sequencing and contig assembly In sum, 983,979 reads totaling >210 megabases (Mbp) were sequenced from python cDNA libraries from heart and liver tissue (Table 1). Combining all cDNA sequences from python heart and liver samples, we assembled 37,245 contigs with a total length of 13,409,006 bp, and with an average GC content of 41.52%. This assembly included 669,607 of our reads, leaving 314,372 singleton reads not incorporated into contigs. Among these contigs, most were sampled by multiple reads, a large number (805) had 100 reads or more ( Figure 1). The top 5,000 contigs had lengths greater than 573 bp, and the top 1,000 contigs were longer than 1,420 bp (Figure 1 and 2A).
In the bulk of the data, there is a clear correlation between contig size and read number ( Figure 1A), as is expected from random sequencing. The nucleotide-level contig coverage (estimated based on the average read length of~235 bp) had a mean of 10.5 and a median of 3.5, with 8,084 contigs having > 5 fold average nucleotide coverage. Most contigs are probably close to but not quite full length, since most are covered 2-12 fold with reads at the nucleotide level ( Figure 1B).

Annotation of contigs
We annotated genes based on BLAST similarity to known genes in a hierarchical fashion, first based on  best tBLASTx hits to known Ensembl Anolis and chicken genes that are thought to be one-to-one orthologs with human genes. Transcript contigs were also matched to known genes based on BLASTx searches against the Genbank non-redundant (nr) protein database (and annotated based on matches), and any remaining genes were annotated based on megaBLAST hits to the entire nr nucleotide collection. Of the 37,245 assembled transcript contigs, 13,286 were matched to some known gene through this hierarchical process, and were thus annotated based on similarity to known genes. Thus, we were able to assign some annotation to 35.7% of all contigs. Compared to the length distribution of all contigs (Figure 2A), the distribution of contig lengths for those with any annotation shows a notable enrichment for the annotation of longer contigs (>1,000 bp; Figure 2B). Among the contigs that were annotated, 3,822 had a best BLAST match to known chicken genes that are one-to-one human orthologs, and 4,302 hit known Anolis lizard one-to-one human orthologs. Ensembl gene IDs were assigned to transcript contigs based on hits with chicken and Anolis genes, and human orthologs were assigned to each contig based on the Ensembl orthologous gene relationship estimates. We considered the annotation of our contigs to be "high confidence annotations" when Ensembl IDs from Anolis and chicken BLAST hits both linked back to the same human ortholog; 3,046 of our contigs fell into this class ( Figure 2C).
For contigs with high-confidence annotations, we compared the protein sequence divergence between our python contigs and the lizard and chicken matches. It is estimated that the python and the Anolis lizard last shared a common ancestor~166 MYA [9,16], whereas the chicken and python last shared an ancestor~275 MY [16] ( Figure 3A). Thus, as expected, the protein sequence divergence between the lizard and python proteins (mean = 0.73) is notably less than that between chicken and python proteins (mean = 0.66), although the variation in divergence is quite large ( Figure 3B-C). There also is a fairly consistent linear relationship (R 2 = 0.35389; Figure 3C) between the python-lizard and python-chicken protein divergence. This indicates that, overall, the protein sequence divergence between python and lizard BLAST matches tends to be proportional to that between the corresponding python and chicken BLAST matches. This correlation fits the expectation that the chicken and lizard BLAST matches tend to be orthologs of one another, wheras poor correlations might indicate BLAST matches to lizard and chicken paralogs.  Comparisons of the distribution (A), and linear relationship (B) of protein sequence similarity of the python-chicken versus python-lizard presumed orthologous protein pairs. Only 'high confidence annotated contigs', in which the lizard-snake and chicken-snake matches were to Ensembl orthologous genes, were used for comparisons. Evolutionary context for snakes in relation to major amniote lineages (C); images from Wikimedia Commons.

Gene ontology (GO) analysis
For the purpose of GO annotation, we were able to associate GO terms to 12,370 python contigs that were BLASTx-matched with known proteins in the NCBI nr database. The frequencies of second-level GO term annotations for our set of 12,370 python matched contigs are shown in Figure 4. Pythons are important models for studying physiological and metabolic remodelling. It is therefore notable that our set of annotated genes includes high frequencies of genes with Biological Process GO annotations that include metabolism, development, cell organization, and morphogenesis that are all likely categories of genes likely to be important for later functional studies of adaptations in pythons (Figure 4).

Data deposition and accessibility
Raw data from heart cDNA libraries is accessioned in the NCBI Sequence Read Archive (SRA: SRX018167). A minority of the data analysed here, from liver cDNA, were published previously, although not previously assembled [20], and related raw data is accessioned in the SRA (SRA: SRX057862). The set of assembled transcript contigs from this study, together with an extensive table with coordinated information and annotation for contigs, are available online via the journals website (as Additional File 1 and Additional File 2, respectively); these files are also available at http://www.snakegenomics.org/SnakeGenomics/Processed_Data.html.

Conclusions
Our ultimate goal is to use the python, and other snake species, as models for studying extreme adaptation at various biological levels, from the extreme evolution of proteins [8,10,11] to the extreme systems biology of physiological redesign accompanying feeding [12,21]. We therefore consider it a necessary first step to establish baseline resources, such as this transcriptome set. Here, we chose to use the relatively long sequences available from the 454 platform to conduct de novo assembly of transcripts for the python because having such longer sequences is expected to generally favor longer and more accurate transcript assemblies. Additionally, having longer transcript reconstructions is also expected to lead to greater success in identification of orthologous genes in other more well-studies model species, particularly in the case of the python, which is more than~160 MY diverged from the next closest reference genome of the Anolis lizard ( Figure 3C).
The results of our de novo assembly did indeed produce a relatively large number of long reconstructed transcripts, with nearly 2,000 contigs greater than 1 kb in length. Contrary to expectation, however, we had relatively low success in matching these contigs to known vertebrate genes, with~35% of contigs matching known genes. Similarly, in a recent analysis of 454-based transcriptome data from diverse tissues for the garter snake, only 34% of transcript contigs were matched to known genes [19]. These numbers for snakes are relatively low compared with percentages of gene identification from other recent transcriptome projects. For example, a recent study on the heart transcriptome of the vole was able to identify~43% of transcripts based on homology with known mouse transcripts [22].
One obvious explanation for the difficulty in identifying transcripts to known genes for snakes is the relatively low numbers of known genes deposited in NCBI for snakes and reptiles in general. For example, of thẽ 2.35 million vertebrate proteins on NCBI, 1.61 million are from mammals, compared to~195,000 for birds, 90,000 for squamate reptiles (lizards and snakes), and 24,000 for snakes. Furthermore, because a large proportion of proteins deposited from reptiles are from phylogenetic studies (with one gene sequenced from many species), the diversity of proteins represented is even lower than might be expected from the above numbers. This paucity of genetic information for reptiles highlights the importance for deposition of data from studies like this one, and further argues for the need for additional data to complement our knowledge of amniote genetic diversity.
There are ongoing initiatives to sequence the genomes of the Burmese python [20], as well as the garter snake [1], which should collectively contribute substantial information on reptilian and snake genomics helping to fill a void in our current knowledge of the genomics of amniotes. The genome project for the python will include the addition of more transcriptome data from diverse tissues, and the transcriptome set here will be combined with future data for annotating the python genome [20], and serve as a valuable reference for thousands of annotated python genes in the meantime.

RNA isolation and cDNA library creation
Tissues were procured from a total of 4 animals (one sample per tissue, each tissue from a distinct animal) obtained from commercial pet trade breeders under approved animal care protocols, and stored in RNAlater or snap-frozen in liquid nitrogen prior to RNA extraction. Prior to tissue extraction, two animals were fed and then euthanized either 1 day or 3 days after feeding [13], following existing IACUC protocols in place at the University of Texas Arlington and The University of Colorado.
Total RNA was extracted using Trizol Reagent (Invitrogen), following the manufacturer's protocol. Extracted RNA was enriched for mature mRNA transcripts using three successive rounds of purification with Oligo dT 25 beads (PureBiotech), precipitated using linearized acrylamide (Ambion) sodium acetate, and ethanol, and analyzed using a BioAnalyzer pico-RNA chip (Agilent).
Following adapter ligation, ligation products were purified using RNA Clean beads three successive times, and then with streptavidin beads (PureBiotech). Samples were then melted from the streptavidin beads using 0.1M NaOH and precipitated (as above). Completed libraries were then quantified and checked for appropriate size distribution using the DNA-nano chip on a BioAnalyzer (Agilent). Where necessary, libraries were PCR amplified using Platinum Taq polymerase (Invitrogen) using a minimal number of amplification cycles (less than 25 cycles).

454-sequencing of cDNA libraries
All cDNA libraries were sequenced using the 454 GS FLX sequencer using the LR70 sequencing kit and 70 × 75 mm PicoTiterPlate (Roche). Emulsion PCR kits II and III (Roche) were used for sequencing cDNA libraries to obtain sequence from both ends of transcripts, because cDNA libraries were directional (with kit II sequencing from the 5' end, and kit III sequencing from the 3' end).
Assembly of cDNA contigs, and identification of orthologous genes All of our python cDNA data were assembled into contigs using the Newbler de novo assembler algorithm of the gsassembler (Roche 454). Contig coverage was estimated by multiplying the number of reads per contig by the average read length divided by contig length. All contigs were compared to the set of Anolis (lizard) and chicken Ensembl protein-coding genes that are estimated by Ensembl Compara to be one-to-one orthologs with Human genes using BLASTx. When contigs had hits to both chicken and Anolis one-to-one orthologs, Ensembl IDs were used to link back to the predicted human ortholog using Ensembl Compara's one-to-one ortholog predictions. If both chicken and Anolis hits liked to the same human gene, these were considered 'high-confidence annotated contigs'. Contigs were also compared to the complete NCBI nr database first using BLASTx against all proteins (at an E-value threshold 10 -5 ). If contigs had no hits to nr proteins, they were compared at the nucleotide level to all nr sequences mega-BLAST. We preferentially annotated contigs (with best BLAST hits) based first on similarity to Anolis and chicken one-to-one orthologs, then based on nr proteins, and finally on nucleotide comparisons where available.
For gene ontology analysis, results of the NCBI nr protein database BLASTx search were used to connect python transcript contigs with known gene ontology annotations. Gene ontology annotations were identified using the Blast2GO bioinformatics suite based upon the BLASTx output [23]. For the purpose of annotating and displaying GO annotations, we used GO-slims, which depicts second level GO terms that are most conducive to graphical interpretation.

Links
Consortium for Snake Genomics website and data clearinghouse [http://www.snakegenomics.org]