An analysis of synteny of Arachis with Lotus and Medicago sheds new light on the structure, stability and evolution of legume genomes

Background Most agriculturally important legumes fall within two sub-clades of the Papilionoid legumes: the Phaseoloids and Galegoids, which diverged about 50 Mya. The Phaseoloids are mostly tropical and include crops such as common bean and soybean. The Galegoids are mostly temperate and include clover, fava bean and the model legumes Lotus and Medicago (both with substantially sequenced genomes). In contrast, peanut (Arachis hypogaea) falls in the Dalbergioid clade which is more basal in its divergence within the Papilionoids. The aim of this work was to integrate the genetic map of Arachis with Lotus and Medicago and improve our understanding of the Arachis genome and legume genomes in general. To do this we placed on the Arachis map, comparative anchor markers defined using a previously described bioinformatics pipeline. Also we investigated the possible role of transposons in the patterns of synteny that were observed. Results The Arachis genetic map was substantially aligned with Lotus and Medicago with most synteny blocks presenting a single main affinity to each genome. This indicates that the last common whole genome duplication within the Papilionoid legumes predated the divergence of Arachis from the Galegoids and Phaseoloids sufficiently that the common ancestral genome was substantially diploidized. The Arachis and model legume genomes comparison made here, together with a previously published comparison of Lotus and Medicago allowed all possible Arachis-Lotus-Medicago species by species comparisons to be made and genome syntenies observed. Distinct conserved synteny blocks and non-conserved regions were present in all genome comparisons, implying that certain legume genomic regions are consistently more stable during evolution than others. We found that in Medicago and possibly also in Lotus, retrotransposons tend to be more frequent in the variable regions. Furthermore, while these variable regions generally have lower densities of single copy genes than the more conserved regions, some harbor high densities of the fast evolving disease resistance genes. Conclusion We suggest that gene space in Papilionoids may be divided into two broadly defined components: more conserved regions which tend to have low retrotransposon densities and are relatively stable during evolution; and variable regions that tend to have high retrotransposon densities, and whose frequent restructuring may fuel the evolution of some gene families.

found that in Medicago and possibly also in Lotus, retrotransposons tend to be more frequent in the variable regions. Furthermore, while these variable regions generally have lower densities of single copy genes than the more conserved regions, some harbor high densities of the fast evolving disease resistance genes.

Conclusion:
We suggest that gene space in Papilionoids may be divided into two broadly defined components: more conserved regions which tend to have low retrotransposon densities and are relatively stable during evolution; and variable regions that tend to have high retrotransposon densities, and whose frequent restructuring may fuel the evolution of some gene families.

Background
The legume family (Fabaceae) is one of the largest and most diverse plant families and is considered to have its origin in the tropics about 65-70 Myr ago [1]. The ability of many legumes to fix nitrogen in association with rhizobia bacteria gives them special importance in natural environments and agriculture.
The family is divided into three subfamilies, Mimosoideae, Caesalpinoideae, and Papilionoideae. Most agriculturally important species fall within two Papilionoid subclades that diverged some 50 Myr ago, the Phaseoloids and Galegoids. The Phaseoloids are an essentially tropical group including bean, cowpea, soya, and pigeon pea. The Galegoids are essentially temperate and include clover, pea, lentil, fava bean, chickpea and the model legumes M. truncatula and L. japonicus [1,2].
Two notable exceptions are lupin (Lupinus spp.) and peanut (Arachis hypogaea). They fall within the Genistoid and Dalbergioid clades respectively, which are more basal in their divergence within the Papilionoids than the Phaseoloids and Galegoids [2]. A simplified tree representation of Papilionoid phylogeny is represented in Fig. 1. Although the importance of lupin species is growing, especially in Australia, the importance of peanut is much greater. Global annual peanut production is about 36 million tonnes, with >90% being produced in the developing world. It is of great social importance, especially in Africa, and in Asia where it provides more calories than soya [3].
Cultivated peanut is an allotetraploid of recent origin, with an AB genome. Its polyploidy and low genetic diversity have hampered advances in its genetics and genomics.
To avoid these problems we recently generated a map from two diploid A-type genome relatives of peanut, A. duranensis and A. stenosperma, the former being the most probable A genome donor to peanut [4], [5], [6]. In this work we aimed to align this map with the substantially sequenced genomes of Lotus and Medicago and enhance our understanding of the Arachis genome. In addition, because of its out-group status (Fig. 1), the results were likely to be informative about the structure and evolutionary history of legume genomes generally.
To compare genomes, orthologous loci need to be mapped in the species of interest. Using single copy genes facilitates the identification of orthologs, however, they generally have low polymorphism, making marker development difficult. The solution is to target the more variable introns. Furthermore if polymerase chain reaction (PCR) primers are designed to bind conserved sequences that flank introns, they work in a range of species and serve as a toolkit for the development of comparable "anchor markers" across a range of plant species. Previously we developed software for the design of such primers and published a set of anchor markers ideally suited to the comparison of legume genomes [7].
The level of synteny between two species depends upon the genomic restructuring events that occurred since their evolutionary divergence. Transposons have an important role in such restructuring. During evolution, transposons can undergo both massive amplifications and reductions in number, inflating and reducing the size of genomic regions. Also, the presence of transposons of almost identical sequence in multiple positions of a genome promotes recombination and genome restructuring. Transposons are divided into two classes. Class I transposons, also known as retrotransposons, have a "copy and paste" replication mechanism that involves an RNA intermediate. Class II transposons, also known as DNA transposons do not have an RNA intermediate, and usually transpose by a "cut and paste" mechanism that does not involve replication [8]. Retrotransposons are further divided into two classes, those with Long Terminal Repeats (LTR retrotransposons), and those without (non-LTR retrotransposons). LTR retrotransposons are the most abundant transposons in plant genomes. Because of this, and because of the special propensity of their terminal repeats to promote recombination events, they are considered to have a special role in plant genome evolution [9].
Here we report the placement of 102 anchor markers on the map of the Arachis A-genome and analyses of synteny of Arachis with Medicago and Lotus. In addition, we investigated if there was a correlation between genome conservation and transposons. Finally, we briefly investigated gene distribution in relation to synteny and retrotransposon distribution.

Plant material
Genetic mapping was done with a previously described F 2 population of 93 plants derived from a cross of A. duranensis (K7988) with A. stenosperma (V10309) [4].

Marker development
Legume anchor markers were developed essentially as described in [7]. Key features of the marker development were: i) Identification of ESTs from multiple legume species, usually Lotus, soya and Medicago, with single strong blast detected sequence similarites against all predicted Arabidopsis proteins and the alignment of these ESTs.
ii) Alignment of ESTs to a corresponding genomic region from Lotus or Medicago and inference of intron positions.
iii) Identification of conserved intron-flanking sequences, and design of primers to bind these conserved sequences.
The rationale behind this procedure is that: i) Markers to unique sequences within a genome facilitate the comparison of genetic maps, and genes that are single copy in Arabidopsis have a high probability of being single copy in legume genomes (also see results).
ii) Introns are more variable than coding regions, and therefore they are better for marker development.
iii) Primers that bind to sequences that are conserved are more likely to be transferable to other species.
The primers were used in PCR with the progenitors of the Arachis mapping population. Polymorphisms were identified by size-or sequence variation. In the latter case, most markers developed were Cleaved Amplified Polymorphic Sequence (CAPS) or derived Cleaved Amplified Polymorphic Sequence dCAPS [10].
Other markers used in map construction have been [4], or will be, published elsewhere (unpublished data).

Genetic Mapping
Linkage analysis was done with Mapmaker Macintosh version 2.0 [11] essentially as described in [4].
A tree represention of the phylogeny of the Papilionoids with triangles representing the major clades, and the two subc-lades of the Galegoids; the Robinioids and the IRLC (plastid DNA Inverted Repeat Lacking Clade) Figure 1 A tree represention of the phylogeny of the Papilionoids with triangles representing the major clades, and the two subclades of the Galegoids; the Robinioids and the IRLC (plastid DNA Inverted Repeat Lacking Clade). Names of some notable genera are placed within the triangles. Note that Arachis which is a member of the Dalbergioids, represents a more basally diverged Papilionoid than the Galegoid legumes, and therefore serves as an out-group to a Lotus vs. Medicago comparison. The figure is a simplified and stylized phylogeny based on a tree in [2].

Analysis of synteny
Points of correspondence of Arachis with the model legumes were assigned by using Arachis marker sequences as queries in blast searches against the genome sequence and genetic mapping for Lotus and, for Medicago, by blast similarity searches against the genome sequence using CViT blast (Chromosome Visualization Tool, http://www.med icago.org/genome/cvit_blast.php). For legume anchor markers, and other markers with a single strong blast detected sequence similarity to Arabidopsis predicted proteins in addition to the the Arachis nucleotide sequence, we also used the Arabidopsis protein sequence in blastn, tblastn and blastp searches. A single strong blast detected sequence similarity was defined as E-value for the best blast alignment being at least 10 6 times lower than the Evalue for the second best blast alignment.
Most of the Lotus Bacterial Artificial Chromosome/Transformation competent Bacterial Artificial Chromosome (BAC or TAC) clones identified in blast searches using Arachis marker sequences had been positioned on the Lotus map based on the intraspecific cross between ecotypes Gifu and Miyakojima MG-20 [12]. Where a map position was absent, or only a whole genome shotgun or EST sequence was available, the corresponding BAC/TAC clone was isolated and/or sequenced and mapped in Gifu × MG-20 and/or in L. filicaulis × L. japonicus Gifu [13]. All map positions are given with respect to the former map. Tandem Repeat Occurrence Locator http://wsmartins.net/ webtroll/troll.html was used for identifying the microsatellites for genetic mapping.

Tables of correspondences between
Arachis markers and the model legume genomes were used in a Microsoft Excel file, together with logical functions, to create plots of the Arachis genetic map against Lotus and Medicago (Additional file 1).
In addition, to visualize a comparison of the Arachis genome simultaneously with Medicago and Lotus, colored coded blocks representing Arachis marker correspondences with Lotus and Medicago were placed along the sides of Arachis LGs. Colors were assigned to the model legume chromosomes following a rainbow spectrum along the X and Y axes of the previously published Lotus vs. Medicago genome plot [14], in such a way that corresponding groups of Lotus and Medicago are represented by corresponding or neighboring colors in the spectrum ( Fig. 2 and Additional file 2).
A synteny score was assigned to each point of correspondence between Arachis and the model legumes according to its number of neighbors within a continuous color block. For instance, the top marker in Arachis linkage group 9 (Ar9) has a Arachis-Lotus synteny score of 3, and the last

Arachis
LGs with affinities to Lotus and Medicago chromo-somes represented as colored blocks, and synteny blocks (SBs) indicated Figure 2 Arachis LGs with affinities to Lotus and Medicago chromosomes represented as colored blocks, and synteny blocks (SBs) indicated. Arachis LGs are numbered according to [4] with size in cM indicated below. Marker positions are indicated as horizontal lines across LGs, anchor markers as red lines, other markers as black lines.
Colors were assigned to the model species chromosomes so that syntenous chromosomes are represented by corresponding colors. SBs are numbered according to [14], with the addition of SB11 identified in this study. A full genetic map is in Additional file 2.

Mt Lj
Ar3 only Lj marker on Ar6 has a Arachis-Medicago synteny score of 8 ( Fig. 2 and Additional file 2). Synteny scores of markers were plotted along the model genomes, with values averaged over 6 Mbp windows for the Medicago genome sequence, and over 10 cM windows for the Lotus genetic map (Fig. 3c, 3d).

Analysis of transposon distributions
Lotus transposon sequences from a Kazusa Lotus transposon data set were used as a target blast database in Repeat-Masker http://www.repeatmasker.org/, run on the Lotus BAC/TAC sequences. Results were summarized as percentage sequence coverage for Class I and Class II transposons for each BAC/TAC using Perl http://www.perl.org/. This result, together with the genetic map positions of the BAC/TACs was used to plot the retrotransposon distribu- . Markers mapped to intervals are plotted as horizontal lines. fLj indicates that the Lotus chromosome has been inverted. The reference orientation of Lj5 has recently been inverted, thus Lj5 in this plot is equivalent to fLj5 in [14].
tions within the Lotus genome with values averaged over 10 cM window size (Fig. 3d).
Medicago Class I and Class II transposon sequences from GIRI http://www.girinst.org/ were submitted to CViT blast. The output was used to plot transposon distribution along the Medicago genome with values averaged across a 6 Mbp window size. For Medicago, consensus sequences of the extended NBS domains from the CNL and the TNL subfamilies of plant disease resistance genes from [15] were used in CViT blast searches, and the distribution of blast detected similarites (E-value ≤ 1E-20) plotted on the Medicago genome with values averaged over 6 Mbp window size (Fig. 3b).

Marker development
Introns in anchor genes showed high levels of polymorphism between the progenitors of the Arachis mapping population (c. 1 single nucleotide polymorphism/90 bp and 1 insertion/deletion polymorphism/1.6 kbp). Eighty seven intron-based anchor markers were developed and genotyped, most being CAPS or dCAPS (Additional file 3).

Mapping
Using a minimum LOD score of 3.0 and a maximum recombination fraction (θ) of 0.35, 351 markers mapped into 10 linkage groups (LGs). Of the 87 legume anchor markers genotyped, 81 were assigned to a LG. In addition, 21 other sequence characterized mapped markers were homologs of single copy genes in Arabidopsis and thus were effective anchors, making a total of 102 mapped anchor markers. Anchor markers were distributed through all of the LGs, except the smallest, Ar10. Most other mapped markers are microsatellites or sequence characterized AFLP.
LGs were numbered according to [4], however, the inclusion of new markers resulted in 10 LGs, corresponding to the number of chromosome pairs in Arachis, instead of the previous 11.

Synteny analysis
Of the legume anchor markers, only six (Leg050, Leg168, Leg33MGm, Leg20MGm, Leg069, Leg034), had strong multiple affinities to the Lotus genome and four to the Medicago genome (LegTC987, Leg202, Leg304, Leg299). This clearly implies that anchor marker genes (represented in a single copy in Arabidopsis) tend to also be represented as single copy genes in the model legumes making them ideal markers for the comparison of genomes.
Using sequence information from the 102 anchor markers and 228 other mapped and sequence characterized markers, we were able to make 128 and 126 points of correspondence between Arachis and Medicago and Lotus respectively.
Following the previously published dot plot presentation of genome similarity between Lotus and Medicago [14], Lotus LG5 (Lj5) are syntenic, are the first in their genetic map/genome orders in the dot plots, and thus have a red color code. Ar9 has a predominance of correspondences with Mt1 and Lj5 which is therefore also placed first in the order for the dot plots. The predominance of "red" associations of Ar9 in Fig. 2 (and Additional file 2) makes this easily visible. Similarly, progressing through Ar8, Ar4, Ar6, Ar10, Ar3, Ar2, Ar1, Ar7 and Ar5 the main affinities with the model legumes roughly follow through the spectrum with the last, Ar5, having predominantly violet (Lj2, Mt5) affinities. The main exception to this is the lower region of Ar8 which has Mt2 -Lj3 affinities. This region corresponds to the main off-axis region of synteny in the Arachis vs. Lotus plot, and which we named synteny block 11 (SB11).

The affinities of Arachis
LGs and Lotus chromosomes in the plot shows a diagonal tendency (Fig. 3f), although a number of factors make the diagonal less clear than for the Arachis vs. Medicago plot (see below): i) Areas of suppressed recombination in Lj6, and the top of Lj1, creates vertical lines of points in the Ar4 -Lj6 and Lj1 -Ar8 squares, if physical instead of genetic distances could have been used, diagonal lines or clusters could well have been formed. ii) There is a substantial off-axis Ar8 -Lj3 cluster in the plot (SB11), this cluster presumably reflects an ancient translocation event, and apparently corresponds to an off-axis area (Mt2 -Lj3) in the Lotus vs. Medicago plot [14]. iii) Lack of resolution in some regions of the genetic maps means that some points are superimposed, for instance the Arachis vs. Lotus region SB2 appears to be supported only by two points in the plot, but in fact is supported by four ( Fig. 2

and Additional files 1 and 2). Nevertheless, for most of the Arachis
LGs the overall affinities are clear, co-linearities are visible between Ar6 and the lower portion of Lj1, and between Lj2 and Ar7. Some areas of the plot show double associations: for instance Ar6 and Ar2 with the bottom of Lj1; and the bottom of Ar2 and the top of Ar1 with the middle of Lj4.
Points of correspondence between Arachis and Medicago are all from blast searches of the pseudomolecules of Medicago genome. This approach is an alternative to the direct comparison of linkage positions used in the Arachis-Lotus comparison. These two somewhat different approaches were used because of the different forms of the data from the genome projects, and because the Lotus vs. Arachis comparison was partly based on blast searches against genome data and partly on genetic mapping. Correspondences between Arachis LGs and Medicago chromosomes were found in seven of the 10 Arachis LGs. Of the remaining three, two (Ar4 and Ar2) have "shattered", or fragmented, synteny. The remaining Ar10 shows no clear affinities, although it has been tentatively assigned a position within the genome plot according to two blast detected sequence similarities with Medicago. The plot of Arachis LGs vs. Medicago chromosomes shows a diagonal tendency (Fig. 3a). There are two apparent regions of colinearity: lower region Mt2 -central region Ar8; lower region Mt5 -lower region Ar7. In the remainder of the plot, points often cluster within squares. The clearest example being Ar6 -Mt7: intra-chromosomal rearrangements since the Arachis -Medicago evolutionary divergence have disrupted co-linearity, but a clear Ar6 -Mt7 correspondence is preserved.
In addition to the synteny blocks previously identified in the Lotus vs. Medicago comparison, an unnumbered offaxis region of synteny visible in the Lotus vs. Medicago plot was also apparent in the Arachis vs. model legumes plots. This region consisting of Ar8 -Mt2 -Lj3 affinities was named SB11.
An Excel file with all marker data, synteny information and genome plots of Arachis vs. Medicago/Lotus is available as Additional file 1. We hope that the file allows the reader to get a good 'feel' for the data, because within the file, parameters can be changed and the effects on the plots visualized.  Fig. 3d). Retrotransposon coverage is very high (c. 30%) in Lotus in exactly the region of the missing synteny block SB1 (Fig. 3d).

Analysis of transposon, single copy genes and disease resistance gene homolog distributions and synteny with
To study gene distributions we focused on two groups of genes with distinct evolutionary pressures: single copy genes, the distribution of which was already apparent from the distribution of anchor marker correspondences, and one of the highest copy number and best characterised of plant gene families, nucleotide binding site encoding resistance gene homologs; Coiled-Coil -Nucleotide Binding Site -Leucine Rich Repeat (CNL) and Drosophila Toll and mammalian Interleukin (IL)-1 receptor -Nucleotide Binding Site Leucine Rich Repeat (TNL) gene subclasses [15].
In Lotus the density of anchor marker, and hence of single copy gene, correspondences with Arachis is not evenly distributed; synteny blocks tend to have higher densities of anchor markers and hence single copy genes (anchor markers, are shown as red lines across LGs in Fig. 2, as red marker names in the expanded version of Fig. 2 presented in Additional file 2, and by solid red dots in Fig. 3a, 3f). The analysis of resistance gene homologs showed that, in the currently available Lotus genome sequence [16] there are 54 CNL and 174 TNL genes encoding disease resistance gene homologs. Of these, 33 CNL and 54 TNL are in mapped TACs and BACs, and the remainder in unmapped clones or whole genome sequence. These genes are not evenly distributed and tend to form clusters, for instance in the upper region of Lj3 and the lower region of Lj2. A number of these clusters are located within regions of high retrotransposon densities. For instance, in the upper region Lj3 there are 13 TNL and 4 CNL encoding genes and in the top of Lj3 there are five CNL encoding genes (Fig. 3e). Overall there was no correlation of CNL genes and retrotransposons, however, the correlation coefficient of TNL genes and retrotransposons was 0.2.

Analysis of transposon and disease resistance gene homolog gene family distributions and synteny with Arachis in Medicago
In Medicago there are about 2.8 retrotransposons to each DNA transposon. LTR retrotransposons outnumber non-LTR retrotransposons by about 16 fold. Retrotransposons are very unevenly distributed in the Medicago genome sequence. There is a clear tendency that Medicago regions with high synteny scores with Arachis have low retrotransposon density, and that regions with low synteny scores have high retrotransposon densities (Fig. 3c). The Correlation coefficient of synteny score and density of blast detected retrotransposon sequence similarities (E-value ≤ 1E-60) per 6 Mb of Medicago genome was -0.35. Retrotransposon densities are moderately high in Medicago in exactly the region of the missing synteny block SB1 -the top of Mt1 (Fig. 3a, 3c, ref [14]).
In Medicago, as was the case in Lotus, the density of the anchor marker correspondences with Arachis is not evenly distributed; there is a trend that synteny blocks have higher densities of anchor markers, and hence of single copy genes (Fig. 2, 3 and Additional files 1 and 2).
In the Medicago genome sequence, there are 177 CNL and 156 TNL encoding sequences [15]. Of these, four CNLs and two TNLs are on unmapped BACs. The genes are not evenly distributed and tend to be clustered. Particularly striking are two superclusters, much higher in copy number than any cluster identified in Lotus (Fig. 3b, note that the scale for resistance gene homolog distributions in Medicago and Lotus are different). One of these superclusters is mostly of CNL encoding sequences at the top of Mt3, and one of TNL encoding sequences at the bottom of Mt6 [15]. The Mt3 supercluster contains 73 CNL and 9 TNL encoding sequences, the Mt6 supercluster contains 57 TNL encoding sequences. Other lower copy number clusters are also apparent, for instance on Mt4, Mt8 and Mt5, these clusters are more similar in copy number to the clusters observed in Lotus. The correlation coefficients of TNL and CNL encoding sequences with retrotransposons are 0.39 and 0.33 respectively.

Plots of the overall profiles of synteny with Arachis and retrotransposons in the Medicago and Lotus genomes
The comparison of the plots of synteny score with Arachis and retrotransposon density along the genomes of Medicago and Lotus shows similarites in the patterns of the peaks within corresponding groups (Fig. 3c, 3d). The profiles of synteny with Arachis plotted along the model legume genomes is most conserved, with a similarity of peaks (corresponding to synteny blocks) both with respect to positions and even relative heights. The main exception being that the order of the synteny blocks that are off-axis in the genome plots (SB7 and SB11) is changed. These rearrangements presumably correspond to chromosome translocation events in one of the three evolutionary lines leading to the model legumes and Arachis. Peak structure of retrotransposon densities is similar especially when considering the time of evolutionary divergence between Lotus and Medicago. However, relative peak heights of retrotransposon densities are markedly different, for instance while the highest retrotransposon concentrations in Medicago are at the top of Mt3 and the lower portion of Mt6, the highest densities in Lotus are in the non-corresponding groups Lj5, Lj3 and Lj4.

Discussion
Integrated maps that link crop and model plants allow knowledge gained from independent research on different plants to be accumulated [17]. This work links the genetic map of peanut, one of the world's most important grain legumes, to the model legumes Lotus and Medicago, a first step for its inclusion in an integrated genetic system for legumes.
Genome restructuring progressively breaks down syntenic relationships between species over evolutionary time. In addition, whole genome duplications occur periodically during plant evolution, followed by progressive diploidization [18]. These events split and obfuscate syntenic relationships. For the legumes, useful levels of synteny have been shown between the Galegoid models and the Phaseoloids, which diverged some 50 Myr ago [19], [20], [21]. However, between the models and the more basally diverged Genistoid lupin, syntenic relationships are more complex [22,23]. The divergence of the Dalbergioid clade to which Arachis belongs is placed at a similar, or slightly later, date in evolutionary time than the divergence of the Genistoids. Therefore the degree to which we could detect macro-synteny between the models and Arachis was uncertain.
Initial inspection of the Arachis vs. model legume plots show surprising degrees of synteny considering the time of species divergence. Although there are some regions of double affinities between Arachis and the model legumes, most synteny blocks have a single main affinity and not two affinities interleaved (Figs. 2, and 3). These patterns indicate that the last universal legume whole genome duplication predated the divergence of Arachis from the Galegoids and Phaseoloids sufficiently that the common ancestral genome was substantially diploidized. Synteny at the macro-level between Arachis and the model legumes will be useful for many genomic regions.
The different cross-species plots showed fascinating similarities, and the power of the Arachis out-group (Fig. 1) allows new inferences to be made: In the Lotus vs. Medicago genomic plot [14] distinct conserved synteny blocks and non-conserved regions are observed. To explain this, we could hypothesize rearrangements/deletions within the non-conserved regions, either in Medicago or Lotus, or in both. Although either explanation is possible, the philosophical principle of Ockham's razor guides us to prefer the simplest explanation: that regions lack synteny due to disruption in Medicago or Lotus but not both. However, with the addition of the Arachis out-group, the power of inference is increased: we see similar patterns of synteny (and disruption) in all possible species by species comparisons (with the notable exception of SB1). Therefore, the evidence from the Arachis out-group strongly argues against the simplest explanation for patterns of Medicago and Lotus synteny and disruption. The inference, instead, is that certain legume genomic regions are consistently more stable during evolution than others. Additional evidence for this was found in our recently completed study, where 104 anchor markers mapped in bean are used to detect genomic regions that are syntenic with Lotus, Medicago and Arachis. In this study, large syntenic segments are also found to be conserved between all species. These syntenic segments correspond to synteny blocks 2, 3, 4, 6, 7, 8, 9, 10 and 11 (the latter consisting of a block of bean LG5 -Lj3 -Mt2 -Ar8 associations). SB1 and SB5 are also evident, but the former is small, covering only 2.3 cM in LG1 of bean, and the latter is fragmented into three sections covering bean LG1 and LG6 ( [21]; Arachis vs. bean marker correspondences from this reference are also summarized in Additional file 1).
We sought an explanation for these observations and began by analyzing transposon distributions in Lotus and Medicago. We found that retrotransposons are very unevenly distributed in both the model legume genomes and that the retrotransposon-rich regions tend to correspond to variable regions, intercalating with the synteny blocks (Fig. 3c, 3d). This tendency is particularly evident for Medicago, its higher retrotransposon content [14], and higher proportion of anchored BACs compared to Lotus may account for this. Overall, considering the time of evolutionary divergence, the patterns of synteny blocks, variable regions and retrotransposon distributions are substantially similar. In addition, it is notable that SB1, which is conserved between Medicago and Lotus, but was not evident between Arachis and the model legumes, is positioned on local peaks in the densities of retrotransposons in the model legumes. Considering this, we suggest that the euchromatic gene space of the model legumes, and by inference possibly most Papilionoid legumes, can be divided into two broadly defined components: regions that remain relatively stable, and regions that experience high rates of genome restructuring. The former tend to be syntenic across taxa and to have low retrotransposon densities, and the latter tend to show little synteny, and to have high retrotransposon densities. The proposed genome model is similar to the pan-genome concept, originally from bacteria but recently suggested for plants [24]. It should be noted that this model appears consistent with the data used here, from diploid genomes. However, plants which have undergone rapid genome restructuring after polyploidy, such as soya, may differ.
In comparing syntenic and more variable regions, another trend seems clear. Variable regions have lower densities of anchor-marker correspondences and therefore single copy genes (Figs. 2, 3a, 3f). However, regions without synteny are not simply "holes" in the dot plot, because there are correspondences in these regions; but they are scattered. In contrast to the low concentrations of single copy genes, some of the retrotransposon rich regions in the model legumes host high densities of some genes in multigene families, and we illustrated this with the plant disease resistance gene homologs.
The fast evolving nature of the repetitive fraction of the genome has been documented in many plant species. This evolution involves rapid expansions and reductions in transposon numbers [25], such that, for instance, even closely related species of Arachis can be distinguished using whole genome in-situ hybridization [6]. Therefore, it seems likely that the restructuring within the variable regions, which tend to be transposon rich, frequently includes amplification of some sequences and elimination of others ("birth and death"). Considering this, it would be expected that natural selection would tend to select against single copy or essential genes, and high densities of transposons inhabiting the same genomic regions. On the other hand, it may be expected that the colocalization of certain fast-evolving multigene families and high densities of transposons could be advantageous. The presence of high densities of resistance gene homologs, a gene family for which rapid birth and death, and frequent diversifying selection has been well established, in some of the variable retrotransposon rich regions, supports this view. A detailed phylogenetic analysis of NBS-LRR genes in Medicago also provides support to the hypothesis that restructuring in the variable regions has driven the evolution of some resistance gene clusters; NBS-LRR encoding genes in retrotransposon rich regions are, on average, more recent in origin, and have more unusual domain rearrangements than those in synteny blocks [15].
Retrotransposon rich genomic regions may play a similar role in legumes as in trypanosomes, where they interrupt synteny and are associated with gene family expansions and the evolution of new gene diversity [26,27]. The legume retrotransposon rich regions may also be similar to pericentromeres, exceptional genomic regions that are also retrotransposon rich: in animals they contain segmental duplications implicated in gene creation, and in plants they harbor rearrangements and insertions uncommon in euchromatin [28]. However, the size of the retrotransposon rich regions described here, extending in some cases to entire or nearly entire euchromatic chromosome arms, and the association of some of them with disease resistance genes, seems notable. For applied science, the presence of clusters of resistance gene homologs in regions of low synteny also has important implications. Synteny between genomes may often not enable predictions of the locations of orthologous resistance genes, although the genome model presented here may aid in the identification of the resistance genes for which synteny is more likely to be preserved.

Conclusion
In summary, we have presented evidence that the last whole genome duplication within the legumes preceded the divergence of Arachis and the model legumes. We also show that levels of macrosynteny between Arachis and the model legumes within ten synteny blocks will be useful for studies of Arachis, for instance, to aid in gene cloning and candidate gene identification. We also show that the retrotransposon profile in the model legumes is negatively correlated with the maintenance of macrosynteny during legume evolution, and that a very substantial percentage of gene space lies outside identifiable synteny blocks. We suggest that the gene space in Papilionoids may be divided into two broadly defined components: regions that are more conserved in evolution which tend to have low retrotransposon densities; and variable regions that tend to be retrotransposon rich, and whose frequent restructuring may fuel the evolution of plant disease resistance genes and perhaps other multigene familes.