A BAC end view of the Musa acuminata genome

Background Musa species contain the fourth most important crop in developing countries. Here, we report the analysis of 6,252 BAC end-sequences, in order to view the sequence composition of the Musa acuminata genome in a cost effective and efficient manner. Results BAC end sequencing generated 6,252 reads representing 4,420,944 bp, including 2,979 clone pairs with an average read length after cleaning and filtering of 707 bp. All sequences have been submitted to GenBank, with the accession numbers DX451975 – DX458350. The BAC end-sequences, were searched against several databases and significant homology was found to mitochondria and chloroplast (2.6%), transposons and repetitive sequences (36%) and proteins (11%). Functional interpretation of the protein matches was carried out by Gene Ontology assignments from matches to Arabidopsis and was shown to cover a broad range of categories. From protein matching regions of Musa BAC end-sequences, it was determined that the GC content of coding regions was 47%. Where protein matches encompassed a start codon, GC content as a function of position (5' to 3') across 129 bp sliding windows generates a "rice-like" gradient. A total of 352 potential SSR markers were discovered. The most abundant simple sequence repeats in four size categories were AT-rich. After filtering mitochondria and chloroplast matches, thousands of BAC end-sequences had a significant BLASTN match to the Oryza sativa and Arabidopsis genome sequence. Of these, a small number of BAC end-sequence pairs were shown to map to neighboring regions of the Oryza sativa genome representing regions of potential microsynteny. Conclusion Database searches with the BAC end-sequences and ab initio analysis identified those reads likely to contain transposons, repeat sequences, proteins and simple sequence repeats. Approximately 600 BAC end-sequences contained protein sequences that were not found in the existing available Musa expressed sequence tags, repeat or transposon databases. In addition, gene statistics, GC content and profile could also be estimated based on the region matching the top protein hit. A small number of BAC end pair sequences can be mapped to neighboring regions of the Oryza sativa representing regions of potential microsynteny. These results suggest that a large-scale BAC end sequencing strategy has the potential to anchor a small proportion of the genome of Musa acuminata to the genomes of Oryza sativa and possibly Arabidopsis.


Background
Until novel technologies that will enable extremely lowcost genomic DNA sequencing are developed, funding bodies are very selective when choosing new plant genomes to sequence. Current technologies are only able to produce the sequence of a mammalian-sized genome of the desired data quality for $10 to $50 million or more. The initial goal of many genome projects is often to gain a glimpse of the genome of interest at a low cost and in an effective manner. In plants there is often some advantage in leveraging the finished genomes of Arabidopsis thaliana and Oryza sativa through comparative genomics. A. thaliana was chosen as model for the dicotyledons due to its small genome size (125 Mb) [1] and rice [2] (O. sativa) was the first cereal and monocot to be sequenced [3].
Musa species (bananas and plantains) comprise very important crops in sub-Saharan Africa, South and Central America and much of Asia. The Musa species Musa acuminata (AA genome) and Musa balbisiana, (BB genome), both with 2n = 22 chromosomes represent the two main progenitors of cultivated banana varieties. The haploid genome of Musa species was estimated as varying between 560 to 800 Mb in size [4][5][6], over four times larger than that of the model plant A. thaliana (125 Mb) [7] and over 30% larger than that of O. sativa (390 Mb) [2].
Comparative genomics in the monocots have focused on the extent of synteny between closely-related species of monocots belonging to the family of Poaceae [8]. Extensive micro and macro synteny has been shown between O. sativa, barley, maize and wheat [9,10] and the degree of conservation often varies between different chromosomal locations. Synteny between distantly related plants is more bioinformatically challenging to elucidate and probably occurs less frequently.
In order to understand the sequence content and sequence complexity of the Musa genome, it is necessary to sequence a large number of randomly selected clones that are representative of the entire genome. An alternative approach is to end-sequence a large number of Bacterial Artificial Chromosomes (BACs) randomly selected from a BAC library [11]. This latter approach does not provide a truly random sampling of the genome since regions in which the restriction site for the particular enzyme used for library construction is under-represented will also be under-represented. Nevertheless, BAC end sequencing does provide a quasi-random sampling of the genome and carries with it the advantage that BAC clones that appear to contain targets of interest provide excellent material for other analyses such as fluorescent in situ hybridization (FISH) to metaphase or pachytene chromosomes or in depth sequencing for gene discovery. A large collection of BAC end-sequences (BES) is also an essential component of a genome sequencing project. Here, we examined whether Musa BES can lead to insights into the Musa genome composition using bioinformatic comparisons to protein, repeat, expressed sequence tags (ESTs) and other databases. From the BES, we investigate the Musa gene density, GC content, protein and SSR content and putative comparative-tile BACs that represents potential regions of microsynteny between the O. sativa and Musa species.

Results and discussion
Sequence searches, simple sequence repeats, GC profiling and protein discovery will be discussed first, followed by an analysis of genome mapping to O.sativa and A. thaliana to identify comparative tile BACs from the Musa library that will be likely collinear (i.e. showed microsynteny).

BAC end sequencing
End sequencing of BACs from a HindIII BAC library constructed from leaves of the wild diploid 'Calcutta 4' clone [12], generated 6,252 high quality reads with an average length of 707 nucleotides, giving a total length of ~ 4.4 Mb that included 2,979 paired end reads ( Table 1). All sequences have been submitted to GenBank, with the accession numbers DX451975 -DX458350.

Database sequence searches
Comparison of the BES with the TIGR non-identical amino acid database revealed that 11% of the sequences contained "genic" regions by virtue of good matches, excluding transposons/repeats (36%). Using a stringent threshold of 1e-5, 80% identity and 80% coverage resulted in 2.6% BES matches to chloroplast/mitochondria (Table 2). Of the protein matches, the top BLAST match in over 50% of cases was to O.sativa and in 30% to A. thaliana proteins, consistent with the closer relatedness between Musa and O. sativa when compared to Musa and A. thaliana. This is also consistent with matches to the TIGR Plant Gene Indices where the highest level of homology was shown to O. sativa followed by barley, wheat and other monocots ( Figure 1). Of the BES analysed, 36% were found to contain sequences homologous to transposable elements or repeats. The majority of transposable elements belonged to the Ty1 copia type (742) followed by the Ty3 gypsy (211) types of retrotransposons (Table 2) consistent with previous data that class I retrotransposons contributing to most of the nucleotide [13] and from studies using papaya BAC end sequences. We also found 111 matches to miniature inverted repeat transposable elements (MITEs), the most abundant being adh-11-like (46), followed by adh type D-like (22) and adh type G-like (12). Gene density predictions calculated from the number BES with protein matches (686) at E = 1e-15 estimates the presence of a gene every 6.4 kb ( Table  3) which is consistent with previous gene density studies from one Musa BAC studied [14]. In contrast, a second BAC from the same study gave a gene density of a gene in every 10 kb, however upon closer examination one half of the BAC consisted of transposon related genes while the other half was non-transposon related. The discrepancy between the data suggests that the gene organization resembling Gramineae where genes are clustered in generich regions separated by gene-poor DNA containing abundant transposons. In comparison with other plant genomes, gene density appears to be similar to reports for the automatic annotation for O. sativa of 6.2 kb per gene [15] and different from A. thaliana with 4.5 kb per gene [6].

Functional annotation
Gene Ontology (GO) is a controlled vocabulary of functional terms that allows consistent annotation of gene products [16]. In order to assign putative functional roles to the Musa acuminata sequences, we used the GO assignments of the A. thaliana proteome [16]. Among the 686 BES that did not contain a match to the repeat or transposon databases but contained a match the TIGR comprehensive protein database, 664 had matches to A. thaliana proteins and were given GO assignments based on the top matches. The genes are shown to cover a broad range of GO categories (Figure 3).

GC profile
GC profiling was performed on the matching region between the BES and the top protein hit. Any BES not containing a match from the start codon was excluded. In par-  sativa and maize showed a higher GC value at the 5' end within the first 150 bp from the predicted start site, which gradually decreased towards the 3' end. This result is consistent from previous reports where it has been shown that Gene Ontology assignments for Musa BES Figure 2 Gene Ontology assignments for Musa BES.
Mean GC content as a function of position (5' to 3') across 129 bp sliding windows Figure 3 Mean GC content as a function of position (5' to 3') across 129 bp sliding windows.

Sliding window position, bp from ATG
grasses have high mean GC content and asymmetrical distributions, while the eudicots have lower GC content and more symmetrical distributions [17,18]

GC content
The GC content for organisms varies between the genomic, intron and exon regions and can be as low as 22% (Plasmodium falciparum) to more than 70% (Zea mays). GC content was determined on the matching region between the BES and the top protein hit. The mean GC content of all BES was 39% and coding sequence GC content was 47% consistent with previous studies which was shown to have an overall GC content to be 38% and within exons to be 49% based on 2 BACs [14]. This and the previous section have shown that BES with protein matches can allow GC content and GC profiling to be calculated with some degree of accuracy. Further confirmation using a larger dataset was carried out using ESTs,-2,280 Musa ESTs [19] was downloaded from GenBank, clustered and assembled to give 1,123 unique sequences of which 179 were contigs. The unique sequences generated 1,056 potential open reading frames containing an average GC content of 51%. These results are consistent with previous studies on GC content within monocots and dicots [17].

Simple sequence repeats
Simple sequence repeats (or microsatellites) are a class of molecular markers that are often polymorphic and are widely used for generating genetic maps [20]. A total of 352 potential SSR markers were discovered within the BAC end-sequences ( Table 4). The most abundant SSRs in all four size categories were AT-rich. This is in agreement with previous reports of microsatellite abundance in other species: poly(AT)/(TA) and AT-rich trinucleotide repeats were the most abundant repeats of their class in A. thaliana and in yeast [21]. Similar to observations for Rosaceae ESTs [22], dinucleotide repeats represent the most abundant of the four microsatellite classes. None of the SSRs present in this study has been reported previously and no matches were found with previous identified Musa SSRs [23,24].

Musa BAC end tiling on the O. sativa and A. thaliana genome
For a relatively uncharacterized species where there may be synteny with some chromosomal regions of well sequenced model species, high throughput BAC end sequencing offers the potential to 'tile' the genome of the uncharacterized species onto to that of the sequenced species. BES mapping to O. sativa and A. thaliana were carried out in order to further characterize our BAC library and to test whether a BAC end sequencing approach might be effective for Musa in the manner described above. When  Musa BACs that fulfil the criteria of having top blast hits to the same chromosome and having no homology to mitochondria and chloroplast were deemed candidate putative comparative-tile-BACs, and potentially represent regions of highly conserved gene content and organization. The predicted size of the Musa BACs (and thus the distance between the end-sequences) was compared to the span by which the paired matches are separated in the O. sativa and A. thaliana genomes respectively. Separations in the Musa BES matches that exceeded our arbitrary cut off of 500 Kb, may represent expansions of the syntenic regions and due to rearrangements during the evolution of the two genomes.

Conclusion
In this study, 2 major ideas were examined. Firstly, that Musa BES can lead to insights into the Musa genome with specific reference to gene density, GC content, protein and SSR discovery; and secondly, that the sequences can be used to identify regions of potential microsynteny between Musa and other species. The BAC end-sequences were shown to contain homology to proteins, expressed sequence tags, transposons, repeat sequences and to be useful for simple sequence repeat identification and estimation of gene statistics and GC content. Proteins encoded in these BES were shown to cover a broad range of GO categories. Although there is only limited microsynteny between Musa and O. sativa, the results suggest that a large-scale BAC end sequencing strategy has the potential to anchor at least a small portion of the genome of Musa onto that of the sequence of the O. sativa. Largescale BAC end sequencing would show whether there are more regions of microsynteny between the reference genome and the genome of interest and if there was support for whole genome sequencing due to unique gene features and genome characteristics. BAC end data would be one useful indicator along with existing EST or genomic sequences for funding bodies to use when selecting new plant genomes to sequence and assess the potential of leveraging the finished genomes of A. thaliana and O. sativa through comparative genomics. We expect that a similar analysis using other plant or animal species would provide insights into the genome in a very cost effective and efficient manner through database searches and synteny to model species.

BAC end sequencing
The BES were generated from a Musa bacterial artificial chromosome (BAC) library constructed from leaves of the wild diploid 'Calcutta 4' clone (Musa acuminata subsp. Burmannicoides 2n = 2 × = 22) with an average insert size of 100 kb [12].
DNA template was prepared in 384-well format by a standard alkaline lysis method. End sequencing was performed using Applied Biosystems (ABI) Big Dye terminator chemistry and analyzed on ABI 3730 xl machines. Base calling was performed using TraceTuner and sequences were trimmed for vector and low quality sequences using Lucy [25].

BAC end database searches
Sequences were compared to all entries in the TIGR Plant Gene Indices [26] using blat and to the TIGR non-identical amino acids database that contains non-identical protein data from a number of databases including GenBank, RefSeq and Uniprot using blastx (cut-off value 1e-5). The BAC end-sequences were also compared with repetitive sequences in the TIGR Repeat Database [27] and an inhouse transposon database using blastx with a cut-off value of 1e-5. The BAC end-sequences were compared with the TIGR rice genome sequence assembly and the A. thaliana genome sequence from TAIR using blastn with a cut-off value of 1e-10. To identify comparative tile BACs from the Musa library that were likely collinear (i.e. showed microsynteny) with the reference genomes, the searches against the Musa genomic sequence were parsed for the top pair of BES for which both ends had the highest significant match to a stretch of O. sativa or A. thaliana sequence and where the two regions on the Musa genome were between 100 kb and 500 Kb apart. The BAC end data sets for O. sativa, A. thaliana, maize and M. truncatula used for GC profiling was originally downloaded from Gen-Bank and then the vector trimmed and cleaned sequences were downloaded from estinformatics.org [28].