Insights into the Musa genome: Syntenic relationships to rice and between Musa species

Background Musa species (Zingiberaceae, Zingiberales) including bananas and plantains are collectively the fourth most important crop in developing countries. Knowledge concerning Musa genome structure and the origin of distinct cultivars has greatly increased over the last few years. Until now, however, no large-scale analyses of Musa genomic sequence have been conducted. This study compares genomic sequence in two Musa species with orthologous regions in the rice genome. Results We produced 1.4 Mb of Musa sequence from 13 BAC clones, annotated and analyzed them along with 4 previously sequenced BACs. The 443 predicted genes revealed that Zingiberales genes share GC content and distribution characteristics with eudicot and Poaceae genomes. Comparison with rice revealed microsynteny regions that have persisted since the divergence of the Commelinid orders Poales and Zingiberales at least 117 Mya. The previously hypothesized large-scale duplication event in the common ancestor of major cereal lineages within the Poaceae was verified. The divergence time distributions for Musa-Zingiber (Zingiberaceae, Zingiberales) orthologs and paralogs provide strong evidence for a large-scale duplication event in the Musa lineage after its divergence from the Zingiberaceae approximately 61 Mya. Comparisons of genomic regions from M. acuminata and M. balbisiana revealed highly conserved genome structure, and indicated that these genomes diverged circa 4.6 Mya. Conclusion These results point to the utility of comparative analyses between distantly-related monocot species such as rice and Musa for improving our understanding of monocot genome evolution. Sequencing the genome of M. acuminata would provide a strong foundation for comparative genomics in the monocots. In addition a genome sequence would aid genomic and genetic analyses of cultivated Musa polyploid genotypes in research aimed at localizing and cloning genes controlling important agronomic traits for breeding purposes.


Background
Taken together, Musa species (bananas and plantains) comprise the fourth most important crop in developing countries [1]. The fruit is a staple food in sub-Saharan Africa, South and Central America and much of Asia, while the leaves are used for sheltering and wrapping food and the male bud can be eaten as a vegetable. Musa is a member of the monocot order Zingiberales, a Commelinid lineage that diverged from the line leading to rice (Poales) in the mid-cretaceous period over 100 million years ago ( Figure 1) [2,3]. 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. Table bananas are sterile, parthenocarpic and diploids AA or triploid with the AAA genome constitution, and represent only a fraction of world production, although they are an important cash crop. Cooking bananas and plantain cultivars, mostly consumed in the countries of production, generally have an AAB or ABB genome constitutions [4]; these are boiled, fried, dried, or sometimes ground into flour.
Knowledge concerning the genetic diversity, the origin of cultivars [5][6][7][8][9][10][11][12] and Musa genome structure [13][14][15] has greatly increased over the last few years. The haploid genome of Musa species was estimated as varying between 560 to 600 Mb in size [16,17], just four times larger than that of the model plant Arabidopsis (125 Mb) [18] and 30% larger than that of rice (390 Mb) [19]. Genetic maps have been developed [20][21][22][23] and recently, BAC resources were generated for both M. acuminata [24,25] and M. balbisiana [26]. A cytogenetic map based on BAC-FISH is being anchored to genetic maps in order to better characterize structural variation among M. acuminata genomes [22]. These resources will pave the way for studies of Musa genome structure and evolution through comparisons with other monocot and eudicot genomes.
The utility of genomic comparisons of monocot and eudicot plants (e.g. [27][28][29][30]) is growing with the availability of the complete genome sequences of rice [19], Arabidopsis [18] and poplar [31], and active genome sequencing projects for a growing number of other angiosperms [32]. Most genome-scale comparative investigations within the monocots have focused on analyses of closely-related species of monocots belonging to the family of Poaceae [27,[33][34][35][36]. Numerous papers have described extensive microsynteny between rice, barley, wheat, maize, Sorghum and sugarcane [27,35,[37][38][39][40][41][42], although the degree of conservation varies between different chromosomal locations. Fewer attempts have been made to investigate the synteny between distantly-related plants. In addition, whereas extensive genomic resources have been developed for rice and other cereal species in the grass family (Poaceae), there is relatively little data on gene content or genome structure for non-grass monocots ( Figure 1). Recently, the first two BAC clones genomic sequences [43], and a BAC end sequencing study of the M. acuminata genome [44] have been published. Here we present data on the genomic structure and organization of 1.8 Mb of Musa genomic nuclear DNA (including the two BAC sequenced previously [43]), show for the first time the existence of microsynteny between Musa, rice and Arabidopsis, characterize the extent of microsynteny between the two Musa species representing the progenitors of most cultivated genotypes, analyze monocot EST sequences and discuss the evolutionary implication of these results. The BAC clones sequenced in this study were identified by hybridization with gene sequences previously selected to correspond to one or a few loci in Musa, rice and Arabi-Current understanding of relationships among monocot orders [118] Figure 1 Current understanding of relationships among monocot orders [118]. Families are shown in bold caps and genera with EST sequences in dbEST [119]. The number of sequences in GenBank (as of 10/08/07) are shown in parentheses for each order and the shaded box highlights Commelinid orders. The nodes with < 75% bootstrap support are grey.
dopsis, thus possibly contain orthologous sequences with these distantly related plant species.

Selection of Musa BAC clones using broad-spectrum
Sorghum cDNA and Musa RFLP probes As part of a program aiming at selecting conserved probes from monocotyledons and Arabidopsis thaliana towards comparative genetic mapping studies, genomic and cDNA probes from various species were tested by Southern hybridization on DNA of various monocotyledons including Musa and rice. Among the probes found conserved between rice, M. acuminata cv. Madang, M. balbisiana cv. Pisang Klutuk Wulung (PKW) and Arabidopsis that revealed a single or low copy locus hybridization pattern, nine were selected. These nine probes that correspond to Sorghum bicolor cDNAs (SbRPG) were used to screen a M. acuminata cv. Calcutta-4 bacterial artificial chromosome (BAC) library (Table 1 and Additional file 1). Of these nine SbRPG genes, four encoded nuclear genes targeted to the chloroplast and/or implicated in photosynthetic-related functions, supporting the notion that this class of genes is under strong pressure for functional conservation. All Musa BACs identified were subjected to HindIII fingerprinting. This enabled us to separate the Musa BACs into groups likely to be derived from different regions of the Musa genome. Overall, a good correlation was observed between the number of loci identified in rice, Sorghum and Musa by Southern blot, BAC fingerprint (for Musa) and analysis of whole genome sequence (for rice) for these nine SbRPG genes (Table 1), all of which were found to be in single or low-copy in both Musa and rice.
One BAC clone was selected for sequencing for probe SbRPG132. For probes SbRPG373, SbRPG661 and SbRPG851, which were found to be present in one or two copies in rice, two Musa BACs with distinct HindIII fingerprints that might be derived from homeologous regions were selected for sequencing with the aim of studying the evolution of lineage-specific duplications in both Musa and rice (Table 1). Two BAC clones from M. acuminata cv. Calcutta-4 (Musa A) and two BACs from M. balbisiana cv. PKW (Musa B) isolated using the genetically-mapped RFLP single-copy probes CIR560 and CIR257 [23] were also fully sequenced with the objective of studying the extent of synteny between Musa A and B species as well as against the rice genome. These RFLP probes were selected because they corresponded to genomic clones encoding genes of known function, CIR257 for a GA-20 oxidase and CIR560 for a beta 1-3 glucanase, previously shown to be associated to traits of agronomic importance in controlling plant height [45,46] and stress response [47][48][49][50], respectively.

Analysis of 1.8 Mb of Musa genomic sequences reveals particular features for the Musa genes
Musa genome statistics A total of 13 BACs (Table 1) were sequenced, generating over 1.4 Mb of unique Musa sequence. In order to provide a uniform set, data four additional BAC sequences (see Additional file 2 and [43]) were included in our annotation pipeline. These analyses revealed 443 predicted genes (on a total of 1.8 Mb of Musa genomic sequence from 17 BAC inserts), after elimination of all putative protein coding genes smaller than 100 amino acid residues. Approximately half of the gene models had matches in GenBank. Their classification based on similarities to genes found in the public sequence databases is presented in Additional file 3 and an annotation overview of the Musa genes in Additional file 4. Gene models were also compared against the Musa EST database donated to the Global Musa Genomics Consortium by Syngenta and maintained at the MIPS (Munich Information Center for Protein Sequences, Munich Germany), revealing that at least 10% of the predicted genes had a perfect match with EST sequences, thus probably being expressed in Musa tissues. Analysis of gene size, exon-intron structure and base composition for these 443 predicted genes is summarized in Table 2. The annotation revealed that, with the exception of MA4_78I12, the BACs analyzed were gene-rich (an average density of one gene per 4.1 kb). Our annotation of MuH9 revealed a total of 23 gene models for an average gene density of one gene per 3.6 kb compared with one gene per 6.9 kb based upon the earlier annotation [43]. In the case of MuG9, our pipeline predicted a total of 14 gene models in the first 52 kb of this BAC followed by a region of ~21 kb containing only transposons (5). Thus the gene density in the non-transposon containing region is one gene per 3.7 kb, very similar to MuH9. Previous annotation of this BAC [43] predicted 7 genes in the same 52 kb region for a density of one gene per 7 kb, with the remainder being transposon-related. The difference between the two annotations is due mainly to the larger number of hypothetical genes identified by the TIGR pipeline as well as some gene splits (e.g. MuH9-5 is split into three genes). Like the last ~21 kb of MuG9, BAC MA4_78I12 found to be mainly composed of class II transposable elements and also contains 7 interspersed predicted genes of which only the homolog of the SbRPG661 probe had a match in public databases. BAC-FISH experiments showed that BAC MA4_78I12 hybridized to all M. acuminata chromosomes except their extremities (Figure 2A). This pattern of hybridization is similar to what we observed by genomic in situ hybridization (GISH) using total genomic DNA as a probe and suggested that the extremities of the chromosomes are poor in repeated sequences [14]. Two gene-rich BACs (MA4_54N07 and MA4_82I11) were also analyzed by BAC-FISH and each hybridized at the extremity of one chromosome (see Figure 2B for MA4_54N07).

Base Composition and GC Distribution along the Musa genes
The GC content of Musa coding sequences was compared with those of other monocots (rice, onion, asparagus) and dicots (Arabidopsis) using two data sets -unigene clusters and -singleton ESTs found in the TIGR plant transcript assembly database (TC/ESTs; [51]) and the 443 annotated genes (CDS) from the 17 Musa spp. sequenced BAC clones ( Figure 3). The GC distributions of TC/EST ( Figure 3A) and CDS regions of the Musa BACs ( Figure 3B) were found to be asymmetrical and bimodal as compared to Arabidopsis and onion which are clearly symmetrical and unimodal (this report, [52,53]). The Musa GC content distribution resembles that of rice and other Poales with higher average GC content than eudicots (see also Table  2) and a long tail towards high GC values. We next examined GC content along the direction of transcription from the ATG start codon for each predicted Musa CDS using a sliding window of 129 bases ( Figure 4). By manual inspection of the data, we were able to identify two categories of GC profiles from the Musa CDS: the first set shows a marked "rice-like" gradient of GC composition from 5' to 3' end and a higher GC content than Arabidopsis all along the CDS ( Figure 4A), and the second set is "Arabidopsis-like" lacking a significant GC gradient from 5' to 3' ( Figure 4B).

Analysis of Musa repetitive elements
Several approaches were used to characterize the genomic sequence with respect to repeats. Database searches of the predicted genes against a non-redundant protein database (see Methods section) revealed a total of 78 transposable element (TE)-related sequences. Excluding TE-rich BAC MA4_78I12, there are on average ~2.6 retrotransposons (TE of class I) per 100 kb. Only one TE of class II encoded protein was detected. BAC sequences were also screened for previously characterized Musa RADKA repeats [15]; an average of 1.8 RADKA-related repeats (GenBank Acces-sions AF399938-AF399941, AF399943-AF399946 and AF399948) per 100 kb were identified. In an attempt to identify as yet uncharacterized repeats, the BAC sequences were also analyzed by RepeatScout [54]. After removing repeats having similarity to Arabidopsis or rice proteins, Musa CDS, RADKA sequences and transposable elements, six repeats with at least three copies were identified (data not shown). Five of these sequences have no significant hits to genes in GenBank while the sixth matches Gen-Bank accession X99496 with a strong similarity to a part of the Musa ycf2 chloroplast gene. Analysis of individual BACs with PrintRepeats [55] shows that each BAC contains only a small number of regions that are repeated within the BAC, an observation that is supported by the relative ease with which the BAC sequences could be closed and finished.

Microsynteny analysis between Musa and either rice or Arabidopsis
The 443 Musa predicted proteins were aligned against the rice and Arabidopsis proteomes. The results showed that 268 and 224 Musa proteins have hits with an E-value threshold of 1e-10 against the rice and Arabidopsis proteomes, respectively. The relative positions of the homologous genes identified in the rice and Arabidopsis genomes were compared to the order of the corresponding Musa genes with i-ADHoRe software [56]. Using this stringent approach, we were able to identify nine Musa BAC sequences showing microsynteny among the 17 Musa BACs analyzed: eight cases with rice and one case with Arabidopsis (Additional file 5).
The i-ADHoRe analyses identified syntenic blocks of three to ten genes (Additional file 5). We then refined the analyses by conducting reciprocal BLASTP searches between  the genes in the orthologous regions. This analysis extended the number of genes included in these syntenic blocks. The most interesting cases of synteny conservation were found between BAC MBP_91N22 and rice chromosome 1 ( Figure 5A), BAC MA4_25J11 and rice chromosomes 1 and 5 ( Figure 5B), BAC MA4_8L21 and rice chromosome 3 ( Figure 5C), BAC MuH9 and rice chromosome 4 (Additional file 6), and BAC MA4_42M13 and rice chromosome 2 (Additional file 7). Between five and eleven genes were found in common between the syntenic Musa and rice orthologous regions. In most cases the common genes were found in the same order and orientation in rice and Musa. However, several additional genes were typically found between the shared orthologs. Interestingly, the number of genes without orthologs within otherwise syntenic regions is much higher in rice as compared to Musa. This could be explained by differences between the rice and Musa lineages in the rate of translocation, duplication and gene death. Note also that Musa BAC MBP_91N22 displays conservation of synteny with two very distant segments on rice chromosome 1.
In the case of BAC MA4_25J11, two rice orthologous regions were found with i-ADHoRe analyses and reciprocal BLASTP searches (rice chromosomes 1 and 5), revealing a duplication of this region in the rice genome. It is interesting to note that the two rice orthologous regions on chromosomes 1 and 5 have lost different sets of genes compared to Musa, as has been observed previously in other duplicated regions in angiosperms [37,57,58]. Phylogenetic analyses on the 10 Musa genes from BAC MA4_25J11 and co-orthologs [59] found on rice chromosomes 1 and 5 revealed that these regions were the product of the genome-wide duplication that has been hypothesized to have occurred early in the history of the Poaceae [56,[60][61][62]. Duplicate maize, sugarcane, Sorghum, wheat, and barley genes occur in two separate clades in trees for loci 4, 6 and 7 ( Figure 6 and Additional Distribution of GC content in Musa and its comparison with other plant species The only significant case of microcolinearity found between Musa and Arabidopsis involved three consecutive genes (Additional file 9). Interestingly, this Musa-Arabidopsis syntenic block was not found to be conserved in rice.

Syntenic relationships between two regions of M. acuminata and M. balbisiana
We also investigated conservation of synteny between two regions of the genomes of M. acuminata and M. balbisiana species. Hybrids between these two species represent the majority of cultivated Musa genotypes worldwide. To carry out this pilot study, we selected BACs from orthologous regions of two single-copy, genetically-mapped RFLP probes (CIR560 and CIR257) encoding genes of agronomic interest. In both cases a high level of sequence conservation was found (see Figure 7 and Additional file 10) over the entire length of the sequenced regions in common between the two pairs of BACs analyzed (82.9% of nucleotide sequence identity for the MA4_82I11-MBP_81C12 pair and 87.6 % for the MA4_54N7-MBP_91N22 pair). The overall levels of sequence identity in genic regions were similar between the two pairs of orthologous BACs: 96.0 % for the MA4_82I11-MBP_81C12 pair and 96.3 % for the MA4_54N7-MBP_91N22 pair (based on the aligned orthologous gene pairs defined in Table 3; see the following paragraph for further details). A high degree of synteny was found between the orthologous sequences in both gene content and gene orientation. However, we observed some incongruence between the gene predictions of the orthologous BACs whose protein products have no match in public databases (i.e. hypothetical protein genes). In contrast, the predicted structures of genes that are homologous to known sequences were largely congruent between the orthologous BACs. Given the high levels of sequence conservation between the two Musa species, such variation of gene structure and exon/intron boundaries is unlikely for most functional genes. Hence, this analysis supports that further validation of gene models through additional EST sequencing or targeted RT-PCR is required.

Divergence between M. acuminata and M. balbisiana
In order to evaluate the degree of divergence between the two Musa genomes, we obtained maximum likelihood estimates for K s values comparing pairs of orthologous genes identified in the M. acuminata and M. balbisiana BACs. We restricted our analysis to those genes (detailed in Table 3) that were intact and matched known gene sequences. For example the gene model for the 14 th locus in the M. acuminata genome (Figure 7; L14) is similar to a pectinesterase related protein, but the gene model was excluded from the analysis because the predicted coding sequence contained several in-frame stop codons indicating that this sequence is a pseudogene. The estimated K s values ranged from 0.0231 (Additional file 10; L19) to 0.0960 (L17), far below saturation levels (i.e. K s << 1), with an average of 0.0410 (Table 3). Applying an average synonymous substitution rate of 4.5 per 10 9 years for nuclear genes in the Zingiberales (see below), this suggests that M. acuminata and M. balbisiana diverged approximately 4.6 Mya ago.

Evidence for a large-scale duplication event in the Musa ancestor
We also estimated the K s values between 1,446 pairs of duplicated Musa genes identified among 15,661 ESTderived unigenes found to be part of the paralog sets [63].  Phylogenetic analyses on three of the ten M. acuminata genes from MA4_25J11 BAC clone Figure 6 Phylogenetic analyses on three of the ten M. acuminata genes from MA4_25J11 BAC clone. These three Musa genes have homologous genes in rice chromosomes 1 and 5 and the locus numbers are taken from Figure 5B. Stars indicate duplication events in the most recent common ancestor of major grain lineages (i.e. rice, wheat and maize). MA4_25J11 BAC clone was isolated by the SbRPG132 probe. assess spikes in the accumulation of duplicated genes [64]. If we assume that gene duplications and gene deletions are random and have relatively steady rates during the course of evolution, such a distribution is expected to show an L shape [64][65][66][67][68]. The distribution of K s values for duplicated Musa genes exhibits a large peak centered around K s = 0.55 ( Figure 8) indicating an increase in the number of gene duplications that occurred in the Musa ancestor circa 61 Mya (assuming a synonymous substitution rate of 4.5 per 10 9 years; see below). This ancient burst of duplications is likely the result of one or more large-scale duplication events. Alternatively, the observed duplications could be associated with a burst of transposon activity as has been hypothesized for some duplicate gene pairs in Arabidopsis [69]. However, analyses of K s plots for duplicated rice genes were unable to detect the 60 Mya duplication event in the Poacae that is evident in analyses of gene trees and duplicated blocks in the rice genome (e.g. [70,61]; this study). This may be due in part to the slower substitution rate we estimate for the Zingiberales relative to the Poaceae (see below).
We also analyzed the 18,612 ginger (Zingiber officinale; Zingiberaceae, Zingiberales) EST-derived unigenes available on the TIGR Plant Transcript Assemblies web site [71] (sequences generated by David Gang, University of Arizona) and found no evidence of large-scale duplication in the K s distribution for paralogous pairs (Figure 8). Moreover, the modal K s for reciprocal best matches between the Musa and Zingiber unigene sets is 0.78 (Figure 8), larger than the mode for Musa paralogous pairs. The age of the most recent common ancestor for the Musacaceae and Zingiberaceae is estimated at 87 Mya [3,72,73]. This implies an average synonymous substitution rate of 4.5 per 10 9 years (0.78 synonymous substitutions per site/ (2*87,000,000 years)), intermediate between rates estimated for the Poaceae (6.1-6.5 per 10 9 years) and palms in the order Arecales (2.61 per 10 9 years; [74]. We must emphasize that all of these rate estimates are approximate, based on rough estimates of minimum divergence times. However, regardless of ambiguities in substitution rate calibrations, our results indicate that the predicted largescale duplication that occurred in the Musa lineage (K s = 0.55) post-dates the divergence of lineages leading to Zingiber and Musa (K s = 0.78), but occurred well before the separation of Musa A and Musa B (K s = 0.0410).
K s values were also computed on 1,034 pairs of homologous genes identified between the Musa ESTs and the rice genome sequences. As expected, the distribution of K s values between rice-Musa homologs form a single peak centred around K s = 1.7 ( Figure 8). Using this K s value to estimate the age of the Poales-Zingiberales split is less straightforward than described above for the Musa-Zingiber split, because synonymous substitution rates clearly vary between these Commelinid monocot lineages.
BAC fingerprint analyses revealed that whereas SbRPG854 hybridized to a single locus in the Musa genome, SbRPG probes SbRPG132 hybridized to 6 regions, SbRPG663 hybridized to 5 loci, and two loci were identified for SbRPG373, SbRPG661 and SbRPG851 (    doxin), SbRPG851 (phosphoglycerate kinase) and SbRPG373 (hypothetical protein), respectively. Phylogenetic analyses suggested that the SbRPG851 Musa homologs duplicated prior to the divergence of the Poales and the Zingiberales, (probably independent from the large-scale duplication described above), and the SbRPG661 and SbRPG373 Musa homologs are sister to each other in the gene tree, suggesting the duplications arose after the divergence of the Poales and the Zingiberales (data not shown).
We also analyzed the degree of conservation between genomic regions surrounding SbRPG661, SbRPG851 and SbRPG373 duplicated genes in Musa and rice and found no synteny in regions anchored by these homologs. This absence of synteny could be explained by duplication events and subsequent gene losses or by the translocation of the focal genes.

Analysis of Musa genes reveals some particular features
Sequencing and annotation of ~1.8 Mb of Musa genomic sequence indicated that most of the BACs analyzed were gene rich with a low content of transposable element. Our analyses of 443 Musa genes predicted revealed that Musa genes generally have a "rice-like" bimodal GC distribution with a very asymmetrical and long tail towards high GC content as in previous studies [43,44]. However, a second class of "Arabidopsis-like" genes was found with an overall low GC content and no significant gradient along the coding sequence. In contrast to a previous comparison of grass and non-grass monocots [52,53], our analyses suggest that Zingeberales genes share some characteristics with the genomes of both eudicots and members of the Poaceae. This result suggests that the Musa genome is more similar to cereal genomes relative to onion, asparagus and the basal-most monocot lineage, Acorus.

Syntenic relationships between distantly-related monocots
Whereas widespread conservation of synteny has been well established for members of the grass family (Poaceae), gene order has not been generally conserved between rice and Arabidopsis (e.g. reviewed by [75]. Few studies have compared genome structure between the members of the Poaceae and other monocot families, but recent comparisons between onion, garden asparagus and rice have failed to find evidence of conservation of macroor micro-synteny [76,77]. However the genomic tag approach developed by [78] has allowed detecting anchor points between grasses and monocots. In this study we were able to identify microsyntenic regions in the Musa and rice genomes that have persisted over some 117 million years of evolution since these two lineages diverged [2]. However, in all syntenic regions detected, the shared genes were separated by intervening genes reflecting the occurrence of numerous insertions and deletion of genes in both rice and Musa. Insertions and deletions have been observed between rice and Arabidopsis regions showing micro-colinearity [58] and to a much lower extent between colinear regions among Poacea genomes [37,79]. Further sequencing of the Musa and other monocot genomes will provide more insight on the extent of lineage-specific gene gain and loss in otherwise syntenic regions.

A first insight into syntenic relationships between Musa A and B
We focused our pilot study on two genomic regions containing genes of agronomic importance for Musa and rice to gain insight into the extent of conservation between the two cultivated species, M. acuminata (A genome) and M. balbisiana (B genome). Our data revealed an extremely high level of colinearity between the two Musa genomes in both regions. However several insertions and deletions occurred during the period of divergence (~4.6 Mya) of the two Musa species. The high level of microsynteny between the two genomes is likely to accelerate gene isolation in M. balbisiana once the construction of the whole genome physical map of M. acuminata has been completed by the Global Musa Genomics Consortium.

Unveiling the paleopolyploid nature of Musa species
There is accumulating data supporting that polyploidy is one of the most important evolutionary mechanisms influencing the structure and content of angiosperm genomes [80]. Our work indicates ancient polyploidization in the lineage leading to Musa approximately 60 Mya. Similar lineage-specific events were described in the Poaceae [81,82], Brassicaceae [56,83,84], Populus [31], Solanaceae, Leguminoceae [64], Papaveraceae, Acorus, the Magnoliids and the Nymphaceae [65]. Polyploidy has clearly been an important source of genetic variation across the angiosperms as retained duplicate genes typi-cally show divergent patterns of gene expression [85,86].
In Musa, as in other plant species, novel phenotypes can emerge from this genomic amalgam, including some with high visibility to natural selection, such as organ size and disease resistance.
Of particular interest is the "composite" nature of the duplicated rice regions relative to the syntenic Musa BAC MA4_25J11; different sets of genes were lost in rice chromosome 1 and 5, respectively as compared to Musa. This type of evolution is likely to reflect a dynamic of duplication [62] and independent evolution in both monocot lineages including recurrent cycles of genome duplication followed by diploidization. This phenomenon was also identified by [58] in their analysis of differential gene loss following duplication events in rice and Arabidopsis. Furthermore, our phylogenetic analyses of gene sets including the genes on Musa BAC MA4_25J11, rice orthologs and related genes found in the Arabidopsis genome and TIGR gene indices corroborate previous results suggesting that a genome-wide duplication in the common ancestor of all major cereal lineages is responsible for the large duplicated segments observed in the rice genome [61,62,87]. This finding illustrates how comparative analyses of distantly-related monocot species can complement studies on cereal genomes.

Is rice a good model to study the structure and evolution of Musa genomes?
The use of rice as a reference species to accelerate mapbased cloning projects by extrapolating marker position data and increasing marker density in targeted regions has a proven efficiency among cereal crops (e.g. barley, wheat, Sorghum), with a perceivable trend towards decreased efficiency when phylogenetic distance increases. Our analyses of the amount of microsynteny between rice and Musa suggest that there are cases in which predictions based upon microsynteny are useful but also this may not be general. In addition although our data showed that Musa genome is more similar to grain genomes relative to onion, asparagus and the basal monocot, Acorus, the differences observed confirmed that cereal genomes are not representative of all monocots [52,53,76,77]. This work also highlight that comparative analyses between distantly-related species such as rice and Musa are very important to improve our understanding of monocot genomes and more generally of angiosperms genome evolution.

Conclusion
In conclusion, this study represents the first effort to investigate the existence and extent of microsynteny between rice and Musa, two-distantly related monocot species. Our analyses revealed a higher degree of synteny than has been reported for other comparisons between the rice and species outside of the grass family. In addition, we identified evidence for an extensive microsynteny between the two Musa species representing the progenitors of most cultivated genotypes. In addition, we identified evidences for an ancient genome-scale duplication event in the lineage leading to Musa and highlighted the complexity of analyzing the structure and evolution of plant genomes following independent cycles of genome duplication and diploidization.

Selection of Musa BAC clones
Nine probes known from previous data to be conserved between rice, Musa acuminata cv. Madang, Musa balbisiana cv. PKW and Arabidopsis and revealing single or very low copy number locus were selected. These nine probes (SbRPG) correspond to Sorghum cDNA developed by Rustica Prograin Génétique and CIRAD [88]. These cDNAs and two Musa genomic probes CIR257 and CIR560 [20] were used to screen high density filters of the M. acuminata Calcutta-4 BAC library [25] according to standard protocols [89](see Table 1). The probes CIR257 and CIR560 were also used to screen M. balbisiana cv. PKW BAC library [26]. BAC DNA of positive clones was isolated using a Qiagen Robot 9600 and Qiagen 96-well BAC DNA isolation kit and digested with the restriction enzyme HindIII. The HindIII fingerprints were then hybridized with the corresponding probe to determine the number of loci.

BAC-FISH analysis
Chromosome preparations were made as described in D'Hont et al [14] from root tips of M. acuminata cv. Calcutta-4 cultivated in glasshouse. Fluorescent in situ hybridizations (FISH) were performed as described in D'Hont et al [14], with 30 ng of BAC DNA labeled with digoxigenin or biotin as probes and 50 ng/μl of sheared salmon sperm DNA. The chromosomes were counterstained with DAPI (4'.6-diamidino-2-phenylindole).  [90,91], and contig gaps were filled if necessary.

Sequence annotation
Annotation of the BAC assemblies was carried out using the TIGR annotation pipeline, a collection of software known as Eukaryotic Genome Control (EGC) that serves as the central data management system. Each BAC sequence was processed through a series of algorithms for predicting genes (Genscan+, Genemark.hmm, Glimmer) [92][93][94], splice sites [95,96] and tRNAs [97]. The AAT package [98] was used for homology search against nucleotide and protein databases, that include plant-specific cDNA and EST sequences, TIGR plant gene indices [99], a non-redundant amino acid database filtered from public sources, and SwissProt [100]. Protein models generated by the searches and predictions are further searched against Markov model (HMM) databases, including PFAM [101], and automatically assigned a putative name based on domain hits or homology to previously identified proteins. Gene structures and names were manually inspected and refined as necessary. Annotated gene models were scanned for Musa transposable element nucleotide sequences downloaded from GenBank and then compared to a curated database of transposable elementencoded proteins [102]. The top match from each hit was used to classify the transposable element.

Comparison of BACs with one another
In order to determine whether the BACs selected by hybridization actually arose from duplicated regions of the M. acuminata (A) genome or homeologous regions of the M. balbisiana (B) genome, or to identify duplicated regions in the M. acuminata (A) genome (pairs of BACs hybridizing with the same probes), each BAC was compared against all other BACs using MUMmer [103]; Dotter [104]; [105] and an all-by-all BLASTP search [106]. The sequence identity of the overlapping sequences between BACs: MA4_82I11 and MBP_81C12 or MA4_54N7 and MBP_91N22, was computed with Stretcher from the EMBOSS package [107].

Synteny search
The 443 Musa predicted proteins were aligned against the rice and Arabidopsis proteomes using the BLASTP program (e-value < 1e-10) [108]. The i-ADHoRe software [56] which looks for regions where the gene order is similar between two genomic sequences was used with the following parameters: a gap size and a cluster gap of 40-40, a q value of 0.9, three anchor points and a probability cutoff of 0.001. For four BACs (MA4_25J11, MA4_8L21, MBP_91N22 and MuH9), we tried to extend the regions of synteny between Musa and rice found by i-ADHoRe by conducting reciprocal BLASTP searches between the genes corresponding on the homologous regions.

Phylogenetic analyses
Musa genes were used in BLASTX searches to query a database of rice and Arabidopsis gene family clusters [109]. Translated blast searches (tBLASTX) against the TIGR plant gene indices [110] were also performed and inferred protein sequences with e-values < 1e-30 were compiled with homologous Musa, rice and Arabidopsis sequences. Amino acid alignments of the compiled sequences were constructed using MUSCLE [111] and manually adjusted. Parsimony analyses were performed on the amino acid alignments using PAUP* v4.0b10 [112].

Contruction of unigenes
Musa EST sequences were provided by the Global Musa Genomics Consortium [113]. These sequences were first assembled into unigenes using the TGICL package [114] to eliminate sequence redundancy. Because unigenes are derived from EST sequences and so have no annotated open reading frames and may contain frameshift sequencing errors, the following approach was taken. Each unigene was aligned against the rice proteome (downloaded from GenBank) using BLASTX. The best match was considered significant if the alignment length was >100 amino acids and the Expect value (E) was <1e-15. The open reading frame was then extracted from the unigene sequence using the Genewise program (which can infer frameshift sites; [115] with the corresponding best match protein as a guide.

Estimation of the level of synonymous substitution between two sequences
For each pair of coding sequence, the two translation products were aligned using the MUSCLE program [111] and the resulting alignment was used as a guide to align the nucleotide sequences. After removing gaps and N-containing codons, the level of synonymous substitution (K s ) was estimated using the maximum likelihood method implemented in CODEML [116] under the F3x4 model [117].

Distribution of the age of duplication of Musa genes
All-against-all nucleotide sequence similarity searches were done among the open reading frame extracted from the unigene sequences using BLASTN [106]. Sequences aligned over >300 bp and showing at least 40% identity were defined as pairs of paralogs. Then we estimated K s for each pair of paralogs. We systematically discarded one sequence from a pair of paralogs showing no synonymous substitutions (K s = 0) as well as all K s values involving this sequence to avoid the inclusion of redundant entries of the same gene in the analysis (see [64] for further details). A gene family of n members results from n-1 gene duplication events. However, the number of possible pairwise comparisons within a gene family (n × (n-1)/2) can be substantially larger than the number of gene duplications, which results in multiple estimates of the ages of some duplications. To eliminate the redundant K s values, pairs of duplicated sequences were grouped into gene families using a single linkage clustering method. Then we used the hierarchical clustering method described in [64] to reconstruct the approximated phylogeny of each gene family: (1) Initially, all sequences in the family were treated as a separate clusters. (2) Then, the K s values for all possible pairs of clusters were compared. (3) The pair of clusters having the smallest K s value was replaced by a single new cluster containing all their sequences. (4) The median K s value was chosen to represent the duplication event that gave rise to the two merged clusters. (5) Steps 2 to 4 were repeated until all sequences were contained in a single cluster. When two clusters A and B contained more than one sequence, their associated K s value in step 2 corresponded to the median K s obtained for all possible pairs of any sequence from A and any sequence from B. Additional material