A recent duplication revisited: phylogenetic analysis reveals an ancestral duplication highly-conserved throughout the Oryza genus and beyond

Background The role of gene duplication in the structural and functional evolution of genomes has been well documented. Analysis of complete rice (Oryza sativa) genome sequences suggested an ancient whole genome duplication, common to all the grasses, some 50-70 million years ago and a more conserved segmental duplication between the distal regions of the short arms of chromosomes 11 and 12, whose evolutionary history is controversial. Results We have carried out a comparative analysis of this duplication within the wild species of the genus Oryza, using a phylogenetic approach to specify its origin and evolutionary dynamics. Paralogous pairs were isolated for nine genes selected throughout the region in all Oryza genome types, as well as in two outgroup species, Leersia perrieri and Potamophila parviflora. All Oryza species display the same global evolutionary dynamics but some lineage-specific features appear towards the proximal end of the duplicated region. The same level of conservation is observed between the redundant copies of the tetraploid species Oryza minuta. The presence of orthologous duplicated blocks in the genome of the more distantly-related species, Brachypodium distachyon, strongly suggests that this duplication between chromosomes 11 and 12 was formed as part of the whole genome duplication common to all Poaceae. Conclusion Our observations suggest that recurrent but heterogeneous concerted evolution throughout the Oryza genus and in related species has led specifically to the extremely high sequence conservation occurring in this region of more than 2 Mbp.


Background
The analysis of an increasing number of complete genome sequences has allowed in-depth studies of the role of sequence redundancy in genome evolution [1][2][3][4]. Gene duplication has been considered for a long time to be a source of novel functions, and to have played a significant part in genome functional evolution and species divergence. Hypotheses on the evolution of genes duplicated by whole genome duplication (WGD), segmental or local events were proposed in 1970 by Ohno [5] and models for the evolution of these duplicated genes have since been elaborated. Following the unexpected observation that Arabidopsis thaliana is a paleopolyploid, a whole genome duplication (WGD) having occurred some 35-40 million years ago (MYA) [6], it was shown that extant plant genomes probably all result from successive cycles of WGD/diploidization [7]. Major losses [8,9], structural and functional divergence [10,11] or concerted evolution [12], have all been observed in eukaryotic genomes, in particular after whole genome duplication events.
Analysis of the complete sequences of the genomes of rice (Oryza sativa) subspecies indica and japonica suggested two independent duplications: a WGD that occurred between 53 and 94MYA, and which is thus common to all cereals, and a more recent segmental duplication between the distal regions of the short arms of chromosomes 11 and 12 [13]. The relative chronology of this latter duplication and speciation events within the Oryza genus are unclear. It was first identified by genetic [14] and physical mapping [15], with an estimated length of 2.5-3 Mbp. Other authors, using synonymous substitution rates between conserved gene pairs for dating, described a duplication of 5.44 (chromosome 11) and 4.27 Mbp (chromosome 12) 5MYA [8], 3.3 Mbp 7.7MYA [16], 3.3 Mbp 20MYA [17] or 6.5 and 4.8 Mbp 21MYA [13], while Goff et al. [18] calculated 25MYA using protein/protein alignments. According to Gaut [19], the divergence between Erhartoideae (Oryza) and the other Pooideae (such as wheat) is approximately 46MYA. Molecular dating places the divergence of the Oryza-Leersia clade with other genera at ~20MYA, that of the Oryza and Leersia genera at 14.2MYA, and divergence of the most basal species in the genus (O. granulata) at 10MYA [20], in agreement with fossil reports [21]. Recent data using other genes has confirmed this divergence time [22]. The evolutionary dynamics of the duplication have been studied between the two subspecies O. sativa ssp. japonica and O. sativa ssp. indica [23]. These authors concluded that this region could be affected by concerted evolution.
Previous studies on the evolution of large-scale gene duplication were based on the available genome sequences from widely-divergent species and little is known about the short-term evolution of duplicated copies and their role in species divergence within a genus. The model species Oryza sativa L. and its wild relatives represent an ideal system to answer questions about gene and genome evolution [24,25]. Genomic data and the wellcharacterized phylogeny available for this genus enable a comparative approach of the evolutionary history of this duplication between several closely-related species.
Adopting a phylogenetic approach, we isolated and sequenced orthologous duplicated pairs from the region of interest in a set of 7 representative Oryza genomes, including tetraploid O. minuta and the surrogate parental species O. punctata and O. officinalis, as well as in the closely-related species Leersia perrieri and Potamophila parviflora. We demonstrate the presence and strong conservation of the duplication both within the genus and in close outgroup species. Its presence in the more distant species Brachypodium distachyon and Sorghum bicolor [26] suggests that its origin is concomitant with the cereal ancestral genome duplication and that the specific mechanisms that have led to the high levels of sequence conservation within this region of the Oryza genomes are probably recurrent.

Sequence conservation in subtelomeric regions of chromosomes 11 and 12
The duplicated subtelomeric regions of Oryza sativa ssp. japonica chromosomes 11 and 12 have been described as being highly conserved [8,13,17]. Additional file 1 shows a dot plot between the first 2.5 Mbp of these chromosomes. Sequence conservation is particularly high within the first 2 Mbp. Beyond this point, large-scale conservation is no longer detectable, similarity being limited to individual genes or blocks of genes, which are visible on the zoom of this region. The loss of colinearity is due to sequence divergence and the movement of transposable elements since the duplication event.

Phylogenetic analysis
Phylogenetic trees based on duplicated sequences can have two topologies, depending on the relative dates of the duplication and speciation events. If duplication predates speciation, we expect to find one copy of each gene pair from all species in one branch of the tree and the other copies in a second branch. In contrast, if the duplication follows speciation we expect to find the paralogous gene pairs as terminal nodes. If the duplication between chromosomes 11 and 12 occurred within the Oryza genus, we would expect to find two gene copies for post-duplication species, with a "speciation after duplication" topology, and only one for species having diverged before the duplication. Using primer pairs selected as described in Methods, we amplified and sequenced gene fragments from seven Oryza species and the closely-related Leersia perrieri and Potamophila parviflora (Figure 1), corresponding to nine genes (named A to I for simplification) selected along the duplicated region (Table 1). Among these, five (B, D, E, H, I) were retained for genus-wide analysis according to the following criteria: minimum length of 500 bp, amplification of both exonic and intronic sequences to clone the more variable intronic regions and their distribution on the duplicated fragment. The four remaining sequences (A, C, F, G) were amplified on a reduced set of species (O. brachyantha and/or O. granulata, L. perrieri and P. parviflora). Putative functions were verified by BLASTX alignment against Viridiplantae proteins.

Figures 2, 3 and 4 and
Additional files 2 and 3 show phylogenetic trees constructed using the maximum likelihood (ML) method, and bootstrap values for both ML and maximum parsimony (MP) analysis. Trees were rooted with homologous copies identified in either Sorghum bicolor [26] or Brachypodium distachyon http://www.brachypo dium.org genomes, but the branches leading to these outgroup are not proportional to their divergence. Two copies of each sequence were isolated almost systematically in all species including L. perrieri and P. parviflora. The fact that two separate copies were isolated for all genes in Leersia is not surprising considering its position in our trees, where all copies of all amplified sequences from this species are grouped in the same clade as the Oryza species.
For sequences A, C, F and G all paralogous copies group together, showing a "duplication after speciation" topology, except for the two gene C copies of O. granulata (See Additional files 2 and 3: ML trees inferred from genes A (1), C (2), F (3) and G (4)). For sequences F and G, only one copy from P. parviflora and O. granulata, respectively, were isolated. Bootstrap values are not strong for internal nodes, certainly because of the weak phylogenetic signal for these sequences ( Table 2), but terminal nodes grouping the paralogous copies are strongly supported.
For sequences B and D (Figure 2), two copies were isolated for all species, and all paralogous pairs are grouped together, except for O. officinalis gene B copy 1 which diverged earlier. Their strong conservation rate is reflected by the weak support for internal nodes, particularly for sequence B. For sequence D, we isolated only one copy from O. australiensis but, given the number of clones sampled (21), the second copy has probably been deleted or is too divergent to be amplified. Moreover, this single copy is sister to one of the tetraploid O. minuta copies, which was not expected. Overall, sequences B and D clearly show a "duplication after speciation" topology type.
On the other hand, sequences E and I display a "duplication before speciation" topology ( Figure 3). Only one copy of gene E was isolated for O. brachyantha and O. granulata, but for all other species the two copies are separated. One copy forms a monophyletic clade with the Nipponbare chromosome 11 sequence, while the second and third copies of O. punctata and O. minuta are grouped with the Nipponbare paralog on chromosome 12. The second copies from O. officinalis, O. australiensis, L. perrieri and P. parviflora, as well as the single copies from O. brachyantha and O. granulata, are grouped in a second, large clade, being more divergent from the copy 1 in these more ancient species. This is in agreement with the hypothesis of an independent divergence of the two paralogous sequences after duplication, the O. brachyantha and O. granulata single copies belonging to the "chromosome 12" clade.
In the ML tree of gene I we clearly observe separation between paralogs from O. brachyantha, O. australiensis, O. officinalis, O. punctata, O. minuta and Nipponbare, each paralogous set for these species forming a monophyletic group. However, neither the ML nor the MP trees allow clarification of the relationships between the copies of the older species, P. parviflora, L. perrieri and O. granulata. This analysis is complicated by the fact that we isolated only The genes in bold were amplified on the complete sample set whereas the others were amplified on the more distant species only. CDS size is given for the multiple alignement of codons sequences.
Phylogenetic relationships, genome type and accessions number of representative Oryza species analysed Figure 1 Phylogenetic relationships, genome type and accessions number of representative Oryza species analysed. !! !! ∀ one paralog for P. parviflora and O. granulata. We observed a 221 bp repeat element insertion, accompanying a deletion in copy 2 of O. brachyantha, but no topology change was observed when excluding this large indel event before analysis. This repeat sequence belongs to the MITE castaway-like family (BLASTN against the TIGR-Oryza-repeat v3.3 database, e-value = 7.2e -5 ).
For gene H, we obtained peculiar results ( Figure 4).

Paralogous pair divergence
To investigate potential bias in paralog divergence, we first compared the sequence data sets ( Table 2). The number of parsimony informative sites and indel events are given for information. The mean rates of synonymous (dS) and non-synonymous (dN) substitutions are the means for all sequence comparisons in each data set. Mean dS varies from 0.040 for sequence B to 0.150 for sequence F, mean dN varies from 0.010 for sequences C and H to 0.1 for sequence E, and mean K varies from 0.084 for sequence B to 0.236 for sequence F. There seems to be no correlation between the two kinds of observed topologies and the global divergence values of the data set, indicating that these genes are evolving at equivalent rates, whatever the proportion of within-species concerted evolution.
We show divergence values between each paralogous pair in Additional file 4. It would have been interesting to ML trees inferred from genes B (1) and D (2) Figure 2 ML trees inferred from genes B (1) and D (2). Numbers above branches indicated bootstrap support for ML and MP respectively. If only one number is present, that means incongruence between the two methods and only the ML bootstrap is shown. Numbers of clones sequenced for each copy are in parentheses. Oryza minuta (allotetraploid) copies are underligned.
compute combined data set analysis, at least for a complete sampling matrix, in order to increase information support, but this was not possible as the paralogous pairs were not isolated for all species. We were particularly interested in the dS values, to examine global neutral evolution of our duplicated pairs, and the dN/dS ratio, to verify the neutrality hypothesis and detect signatures of positive selection. The K ratio, the rate of nucleotide substitution calculated for orthologous non-coding sequences, is expected to be higher than the dN value and approximately equal to the dS rate, as non-coding sequences are also considered to evolve without selective pressure. However, if mechanisms leading to homogenization of paralogous pairs between both chromosomes 11 and 12 operate indiscriminately on both coding and non-coding sequences, we ML trees inferred from genes E (1) and I (2) Figure 3 ML trees inferred from genes E (1) and I (2). Numbers above branches indicated bootstrap support for ML and MP respectively. If only one number is present, that means incongruence between the two methods and only the ML bootstrap is shown. Numbers of clones sequenced for each copy are in parentheses. Oryza minuta (allotetraploid) copies are underligned.
would expect that the intron sequences diverge more slowly between paralogs than between inter-species orthologs. If these mechanisms apply only to coding sequences, dN and dS rates between paralogs should be

Evolutionary dynamics of duplicated genes in O. minuta
In polyploid species, the evolution rates between duplicated copies are expected to change, either by accumulation of deleterious mutations in one of the redundant copies, leading to pseudogenization, or accumulation of positive mutations leading to neofunctionalization, or possibly subfunctionalization. Four copies for each sequence in the 11-12 duplication should be present in the tetraploid species O. minuta, two from the B genome and two from the C genome [27], except if gene loss has occurred early in the diploidization process. Thus, genes are three times redundant and we assessed whether this redundancy can influence their evolution. We tested to see (1) if we could detect accumulation of mutations and pos-ML tree inferred from gene H Figure 4 ML tree inferred from gene H. Numbers above branches indicated bootstrap support of ML and MP respectively. If only one number is present, that means incongruence between the two methods and only the ML bootstrap is shown. Numbers of clones sequenced for each copy are in parentheses. Oryza minuta (allotetraploid) copies are underligned. The genes in bold were amplified on the complete sample set whereas the others were amplified on the more distant species only. Mean dS, mean dN and mean K are the average synonymous, non-synonymous and non-coding substitutions rates values for all the pairwise comparisons in one data set.
itive selection due to relaxed selection constraint or (2)  To compare with the data of Lu et al. [25], we calculated the number of synonymous and non-synonymous substitutions in the tetraploid and its parental genomes, with Nipponbare (copies 1 and 2) as outgroup (Table 3)

Analysis of the duplicated region in Sorghum and Brachypodium
The amplification of two copies for most genes we selected in the 11-12 region, not only for species from the Oryza genus, but also from the related Leersia perrieri and Potamophila parviflora, was concordant with the recent results of Paterson et al. [26]. These authors detected a duplicated segment, also showing strong conservation, in the corresponding regions of Sorghum bicolor chromosomes 5 and 8 and suggested that the duplication event occurred before the cereal divergence. We  Figure 5). Sequence comparison with the current assembly of the 4× coverage of the Brachypodium distachyon genome identified only one contig, super-contig 7 (~17.7 Mbp). However, closer inspection showed that these hits corresponded to two different regions of this contig, the first 3 Mbp and the last 0.5 Mbp. ACT visualization of sequence conservation shows that the duplicated region at the end of the contig (beginning at 17 Mbp) is inverted compared with the sequences of chromosomes 11 and 12 ( Figure 6).

Discussion
The rice genus underwent two episodes of rapid diversification [28] and thus rapid speciation which, with the fact that the 11-12 subtelomeric region is highly conserved, explains the poorly resolved internal node in some of our trees. This leads to unclear phylogenetic relationships between Oryza species and the outgroup Leersia perrieri, in contrast to the observations of Guo and Ge [20]. Moreover, Leersia presents similar characteristics to O. brachyantha, which is on the boundary of the genus [29]. We identified more than two copies of the H sequence for L. perrieri, each sister to one Oryza species copy. L. perrieri was identified as a diploid species (2n = 24) [30], and we have obtained independent confirmation (A. d'Hont, personal communication). Two copies of Adh2 and Gpa1 were also isolated in this species [20], both of "Leersia" type. These genes and gene H may have been duplicated since the divergence of Leersia from the other Oryza branches but more sequence information from this species is necessary to draw precise conclusions. While we cannot exclude mechanisms of "birth and death" in the generation of new gene copies elsewhere in the wild species' genomes, our approach, including amplification on mapped BAC clones in all Oryza species, strongly suggests that the gene copies are effectively on the orthologous regions of these genomes.
Isolation of paralogous pairs in seven Oryza species and two outgroups confirmed, firstly, that the duplication is not specific to the genus and, secondly, that the gene sequences are highly conserved between species. Wang et al. [23] described a high level of concerted evolution in this duplication in the two Oryza sativa subspecies, japonica and indica, which they dated to 5-7MYA, but showed that this conservation was heterogeneous along the segment. Similarly, our analysis shows different phylogenetic topologies throughout the duplication in the Oryza genus. All species display the same evolutionary mechanisms for the first sequences on the duplication, with a "duplication after speciation" topology. While we cannot formally exclude independent duplication in all species, widespread concerted evolution is the most parsimonious explanation. Paralogous pair divergence is similar, showing high conservation of the sequences. Even the allotetraploid species, Oryza minuta, shows no evidence of relaxed selective pressure, despite the putative presence of four copies of each gene. This conservation throughout the genus and in related species suggests that concerted evolution in this subtelomeric region is a recurrent proc-Graphical representation of the syntenic regions between rice and sorghum ess. Moreover our analysis of the K ratio between paralogous and orthologous copies indicated that the concerted mechanism involved would occur on the whole genes, and not only on the coding sequences.
Recently, Paterson et al. [26] described a duplicated segment in the corresponding regions of the sorghum genome and suggested that the apparent segmental duplication in Oryza sativa resulted from the older pan-cereal duplication. These observations and our results indicate that we are no longer looking at the short-term evolution of recently-duplicated genes, as has been suggested [8,13,16,17], and that previous dating based on molecular clock calculations were biased by the weak divergence rate. However, these authors describe a much larger conserved, duplicated region in rice and the exact extent and degree of conservation remain to be determined. Our results rather suggest that recurrent gene conversion is probably limited to a relatively short region, with much higher conservation in the immediate sub-telomeric region and a gradient of sequence divergence. This may explain the relatively high divergence times (17MYA for rice/rice duplicates and 34MYA in sorghum) calculated by Paterson et al. [26] In this context, a similar duplication in the Brachypodium distachyon genome is expected. Indeed, Oryza and Brachypodium both belong to the BEP (Bambusideae-Ehrhartoideae-Pooideae) clade, whereas sorghum belongs to the PACC (Panicoideae-Arundinoideae-Chloridoideae-Centothecoideae) clade [31]. These clades diverged between 50 and 70 MYA [19], soon after the divergence of the grasses. We identified two regions orthologous to the 11-12 duplication on the first Brachypodium genome release, confirming its presence in this species, although future assemblies using deeper coverage will be needed to confirm the chromosome locations.
Gene conversion and unequal crossing-over events are the mechanisms proposed to explain such a level of conservation after tens of millions of years, but more in-depth genomic and cytological work would help to determine the type and frequency of these events. An inversion event, which constitutes a major chromosomal locus rearrangement, was detected on sorghum chromosome 8 and potentially in one of the Brachypodium (end of the supercontig 7) duplicated regions. Inversions can be a source of genomic novelties as well as sequence divergence [32] and such an event in a region which has undergone concerted evolution suggests that it is probably recent.
In the more proximal region of the duplication (genes F, H and I), gene pairs appear to be less influenced by concerted evolution as we observed "duplication before speciation" topologies and isolated single copies for ancient species. Moreover the neutral dS rate was stronger for these genes. This could be explained either by divergence of one of the sequences, making amplification of both copies with primers designed on Oryza sativa impossible, or loss of one copy, as for the majority of duplicated genes in rice through the diploidization process [8]. A clear rupture in highly-conserved colinearity can be observed in the dot plot of the 11-12 region in Oryza sativa (Additional file 1). Wang et al. [23] proposed a first model of the distribution and order of crossing over events throughout the duplication explaining the heterogeneity in sequence similarity between japonica paralogs. We will be able to extend this model to wild species with finer genome analysis, but our results on gene H (L-Galactono, 4-lactone dehydrogenase) already suggest recent conversion events specific to two species (O. sativa and O. punctata).
Genetic recombination is influenced by chromosomal location [33]. The subtelomeric location of the 11-12 duplication could be one factor explaining its evolution. However, the subtelomeres of rice have rather been described as dynamic regions where duplications have spawned new copies of genes [34]. In agreement with our observations, Wang et al. [35] recently described gene conversion occurring at a higher frequency towards the terminal regions of rice and sorghum chromosomes, showing wholly converted genes at an average distance of 3 Mbp from the telomeres in rice and a similar tendency in homologous regions of sorghum. However, these calculations are biased by the over-representation of two duplicated regions, between chromosomes 3 and 10 and the 11-12 duplication, which represent between them 82% of wholly converted genes and, to a lesser extent, high levels of conversion in the orthologous regions in sorghum.
Rather than being a genome-wide phenomenon, these observations suggest that as-yet unknown selective pressures have contributed to the maintenance of high sequence identity within these two specific regions, and particularly the 11-12 duplication.
Our results suggest the presence of two duplicated chromosomal fragments, currently found on all Oryza chromosomes 11 & 12, sorghum chromosomes 5 & 8, and Brachypodium contig 7, which have been homogenized through concerted evolution since the ancestral WGD, dated after the Eudicot-Monocot divergence (between 150 and 200 MYA [36,37]). Wang et al. [23] proposed a stochastic evolution of gene pairs, in which conversion acts as an occasional, sometimes frequent interruption to independent evolution of paralogs. Our observations on genes in the subtelomeric 11-12 region throughout the Oryza genus and in related species, suggesting continuous concerted evolution affecting the same gene pairs in widely-divergent species, are not in agreement with this hypothesis. As suggested above, they rather indicate mechanisms acting preferentially in specific duplicated regions, and most notably in the duplication between chromosomes 11 and 12.

Conclusions
Our observations suggest recurrent but heterogeneous concerted evolution has led to the extremely high sequence conservation occurring in this region of more than 2 Mbp. The detection of paralogous copies for almost all genes in all the species studied indicates a specific mechanism which has led to conservation in this duplicated region throughout the Oryza genus and in related species. It will be interesting to compare detailed structure of both distal ends of chromosomes 11 and 12 with other rice genomic regions (chromosomes 3 and 10). More detailed comparative analysis will allow a clearer understanding of the selection or structural pressure which tends to conserve this particular region. . We also included two closely-related species, Leersia perrieri and Potamophila parviflora. Information on the samples used for phylogenetic reconstruction is displayed in Figure 1.

Species sampling and amplification
Translations of sequences annotated as coding sequences from genes in the first 2.5 Mbp of chromosomes 11 and 12 were used to isolate informative paralogous genes on the Nipponbare genome from the Rice Annotation Genome database [38]. These sequences were aligned with all O. sativa japonica cDNA sequences using TBLASTN [39] at an e-value of 10 -5 to select only genes for which there is proof of expression.  Table 2.

Phylogenetic analysis
Sequences were first aligned with Muscle [41] then refined manually in the data matrix using Seaview [42] and Bioedit http://www.mbio.ncsu.edu/BioEdit/ BioEdit.html. Maximum parsimony analyses were performed on PAUP [43], using the Heuristic algorithm with default parameters. Analyses were conducted either with insertions/deletions included or with indels excluded and recoded according to Simmons and Ochoterena [44]. We used PhyML 3.0 [45] for maximum likelihood analysis and the automated tool provided by the Datamonkey webserver [46] for selection of the DNA substitution model. For all the sequence data sets, the Hasegawa-Kishono-Yano (HKY85) model, which does not assume equal base frequencies and accounts for the difference between transitions and transversions with one parameter, was selected. Bootstrap support was estimated with 1000 bootstrap for both methods.
To characterize the matrix data set and the divergence of the paralogous copies, we calculated pairwise non-synonymous (dN) and synonymous (dS) nucleotide substitutions per sites in the coding regions with the modified Nei-Gojobori method [47] in MEGA4, with overall transition/transversion bias for each CDS also estimated in MEGA4 [48]. We define the number of substitutions per site in the non-coding regions (introns) as the K rate. It was calculated with DnaSP [49], along with the deletion/ insertion polymorphism (DIPs). dS and K variation was tested with a Wilcoxon test, with α = 0.05, and mean K was tested with a parametric Z-test. Selective pressure (dN/dS >1) was tested using a Codon-Based Z-test in MEGA4, with all positions containing alignment gaps eliminated in pairwise sequence comparisons. Significance levels were corrected for multiple tests (672 comparisons overall) for diploid analyses using the Benjamini&Hochberg procedure [50].

Sequence analyses
Dot plots were carried out with the Dotter programme [51] using default parameters. Syntenic regions were identified by TBLASTN alignment against Sorghum bicolor [26] and Brachypodium distachyon (first public release, http:// www.brachypodium.org/) genomic sequences with a cutoff of e -15 using all CDS on the duplicated region (3 Mbp) on chromosomes 11 (546 CDS) and 12 (548 CDS). Largescale sequence conservation was analysed using the Artemis Comparison Tool (ACT [52]) to project BLASTN alignments carried out on the Health Protection Agent Double Act server http://www.hpa-bioinfotools.org.uk/ pise/double_act.html, after masking repeated sequences with RepeatMasker [53]. Dot-plots on subsequences of 1 Mbp were used to detect small-scale rearrangements.