We isolated three to four different cDNA sequences of DEF-like genes from each of the orchid species Vanilla planifolia (Vanilloideae), Phragmipedium longifolium (Cypripedioideae), Spiranthes odorata (Orchidoideae) and Gongora galeata (Epidendroideae), whereas in Hypoxis villosa (Hypoxidaceae), we found only two different sequences. In contrast, in almost all the orchid species analyzed only a single GLO-like sequence was identified, with the exception of Habenaria radiata, where two GLO-like genes have been isolated 61. In Hypoxis villosa two different sequences were found (Table S1 in Additional file 1).

The molecular phylogeny of all monocot DEF-like sequences in GenBank (Figure 1) shows strong support for the existence of four orchid-specific clades of orthologous genes that we designate according to the first reported sequence in the literature 5758: PeMADS2-like genes (clade 1), OMADS3-like genes (clade 2), PeMADS3-like genes (clade 3) and PeMADS4-like genes (clade 4). The phylogenetic reconstruction indicates that clades 1 and 2, as well as clade 3 and 4, are sister clades (Figure 1). The topology of each clade reproduces the phylogeny of the Orchidaceae, indicating that the genes within these clades are orthologues (different genes that originated in speciation events) 42. The phylogenetic analysis indicates that the sequences in each of the four orchid DEF-like clades have features specific to each clade and suggest that, following gene duplication, they might have acquired different functions. In particular, the C-domain of the corresponding proteins is a region possibly involved in protein – protein interactions in multimeric complexes 68. This region experienced a remarkable level of clade-specific diversification (Additional file 2), whereas in the M-, I- and K-domains most of the substitutions involve amino acids of similar chemical properties. In the positions encoding the MIK-region, nucleotide identity ranges from 68 to 92%, but as expected, it drops to a range between 46 to 90% in the region encoding the C-terminal domain of sequences in clades 1, 3 and 4. This reduced identity is particularly pronounced in clade 2, where it ranges from 28 to 57% as a result of a large number of substitutions (e.g. PeMADS5 and SpodoDEF2) or early stop codons (e.g. OMADS3 and GogalDEF2) that eliminated the positions encoding for the otherwise highly conserved C-terminal 'PI-derived' and 'paleoAP3' motifs (Additional file 2). Finally, the monophyletic status of the individual clades of DEF-like genes in Orchidaceae (Asparagales) and in the orders Liliales, Poales and Commelinales is supported with posterior probabilities of at least 99% (nodes indicated with arrows, Figure 1). However, the phylogenetic relationships between these clades are undefined by these data.

In contrast with the ancient clades of orchid DEF-like genes, all GLO-like sequences of orchids form a single clade supported with PP = 1.0 (Figure 2). Only HrGLO1 together with OrcPI and HrGLO2 with SpodoGLO1, representatives of the subfamily Orchidoideae, form two distinct and equally well-supported subclades (PP = 1.00), one of which appears more closely related to GLO-like genes from the subfamily Epidendroideae, whereas the other is closer to the single sequence found in Phragmipedium longifolium (Cypripedioideae) (Figure 2). It remains to be seen whether this duplication is specific to the subfamily Orchidoideae or took place after the divergence of the Cypripedioideae and Vanilloideae, as suggested by our analysis. Despite this apparent subfamily-specific duplication and divergence, the sequences of orchid GLO-like genes are between 65 to 94% identical at the nucleotide level in the MIK-region and between 75 to 92% in the C-terminal domain, where there are neither large deletions nor early stop codons (Additional file 3). As with, to the clades of orchid DEF-like genes, GLO-like sequences also reproduce the phylogenetic relationship of the species from which they were isolated.

Each region of the alignments is distinctly variable and thus has different proportions of phylogenetic informative positions. For example, this is reflected in the higher support levels and phylogenetic resolution in the tree obtained with the MIK-region of DEF-like genes (Figure 1), compared with the one estimated only with the C-terminal domain or with both regions together (Additional files 4 and 5). This indicates that different domains of these transcription factors are subject to different selective regimes. In contrast, the high degree of conservation in all regions of the GLO-like genes is reflected on the relatively similar phylogenetic reconstruction obtained with complete sequences (Figure 2), or individual regions (Additional files 6 and 7).

The apparent lack of family-wide duplications in the GLO-like genes within Orchidaceae contrasts with the two well-supported (PP = 0.92 and 1.0, respectively) clades formed by the other Asparagales sequences (Figure 2). According to their species composition, we estimate that these two clades A1 and A2 at least go back to the origin of the Hypoxidaceae and are present in the relatively advanced family Asparagaceae (Figure 2). Most importantly, clade A1 is clearly associated with the corresponding clade of the family Orchidaceae (PP = 0.89).

The grasses belong to an order (Poales) that also shows a well-supported and group-specific internal duplication of their corresponding GLO-like genes 6369. Furthermore, the clades corresponding to the Liliales, Zingiberales, Arecales and Alismatales are also monophyletic (supported with PPs ≥ 0.91) and contain species-specific gene duplications (Figure 2).

For every pair of sequences tested, the Maximum Likelihood-based (Relative Rate Test) RRT in HyPhy employs a (Likelihood Ratio Test) LRT to evaluate whether the data fit the null hypothesis of a molecular clock where the branch lengths of a phylogeny are equal or an alternative hypothesis where branch lengths are different.

After adjusting all the results of the RRTs for multiple comparisons, we found that 128 of 231 pairs of orchid DEF-like genes rejected the null hypothesis of a constant rate of substitution (P ≤ 0.027). In order to compare the rate differences between the pairs of DEF-like sequences that rejected the null hypothesis, we grouped them according to the clade where they belong (Figure 1) and estimated the median of the relative rates of synonymous and nonsynonymous substitution of these groups (Additional file 8). In the case of nonsynonymous substitutions, the most significant comparisons involve species from orchid Clades 1 and 2 with rate between 0.10 and 0.30 substitutions/nucleotide (Additional file 8A). This means that the genes from clade 2 have a rate of nonsynonymous substitutions about three times higher than those of sequences in clades 3 and 4 (Additional file 8A). On the other hand, the relative rate of synonymous substitution fluctuates independently of the clades compared (Additional file 8B).

In contrast, all 45 pairwise comparisons of orchid GLO-like genes failed to reject the null hypothesis (P ≥ 0.05) and have a median rate of nonsynonymous and synonymous substitution of 0.023 and 0.534 substitutions per nucleotide, respectively. Similarly, the RRT with DEF- and GLO-like genes from Poales did not reject the null hypothesis of constant rates of substitution.

The analysis of variation of ω along codon sites showed that both DEF- and GLO-like genes fit a scenario where positive selection is absent (M7, where ω < 1) (Table S2 in Additional file 1). This result is supported by the estimation of ω in DEF-like genes with the FEL method from HyPhy. In this analysis, 156 of 249 codons (63%) in the alignment of DEF-like genes are under purifying selection with a P ≥ 0.95, whereas the rest of the codons are under neutral evolution (data not shown). However, the tests we applied are conservative and cannot detect positive selection when it affected only a small proportion of sites in a few branches of the phylogeny. For this reason, we implemented a series of tests that focus on specific branches of the phylogenies of DEF- and GLO-like genes (See "Methods" and Table 1).

To determine whether the ω ratio varies substantially in clades of DEF- and GLO-like genes from the order Poales and the family Orchidaceae we compared the null hypothesis (H0) of a "single ω ratio for all branches" (M0) with several nested alternative hypotheses (H#) represented by M2, where one or more clades of interest are free to adopt different values estimated from the data (Figure 3).

In the case of DEF-like genes, we tested five hypotheses where H0 assumes that ω adopted the same value along all branches of the phylogeny. Using LRT we compared this scenario with four relevant alternative hypothesis: H1, where a specific ω is estimated for the branches of clade 1 (C1); H2, where clade 2 (C2) has a different ω to the rest of the branches; H3, where orchid clades C1, C2 and C3 have estimated ω values different to the rest of the branches and H4, where the branches of each of the four clades of the orchid DEF-like genes and the single clade of Poales DEF-like genes (P1) have ω estimates significantly different from the rest of the branches (Figure 3).

Hypotheses H0 and H3, which assume no significant differences between all or some of the groups of branches tested, are significantly (≤ 0.0001) rejected in favor of H4, where specific ω values were estimated for the different groups of DEF-like genes from Orchidaceae and Poales (Figure 3). Most importantly, in hypothesis H4 where ω is estimated for each group of branches, these values converge and indicate that although all clades tested are under strong purifying selection, the group of branches in orchid clade C2 has an ω value of 0.1978, which is more than twice those estimated for other groups of branches (Figure 3).

Similarly, in the corresponding analysis of monocot GLO-like genes we tested a null hypothesis H0 where all clades have the same estimated ω value and six alternative hypotheses where this rate is estimated for one or more relevant groups of branches. To determine whether gene duplication is associated with a distinct pattern of evolution, we estimated ω on both the single clades of GLO-like genes from orchids (C1) and Liliales (L1) as well as the paralogous clades from Poales (P1, P2) and the remainder of Asparagales (A1, A2) (Figure 4).

Comparisons of H0 vs H1 and H7 rejected the scenario where ω is the same in all branches, thus favoring the alternative hypotheses where the orchid clade has ω = 0.043 (P ≤ 0.0001) (Figure 4). Like the DEF-like sequences, all clades of GLO-like genes are under strong purifying selection, but this is especially pronounced in the case of the single orchid clade (Figure 4).

We complemented the previous branch-based analysis with two tests for detecting rate shifts in paralogous clades 70. Specifically, we tested paralogous clades of DEF-like genes 1 and 2 as well as 3 and 4 and the GLO-like clades P1 and P2 (Tables 2, 3 and 4). The tests showed that in clades 1 and 2 there have been significant (P ≤ 0.0005) shifts in the value of ω following gene duplication (Table 2). Both models MC and MD estimated that approximately 42% of the sites of both clade 1 and 2 have divergent ω values, (ω_{2 C1} = 0.21 and ω_{2 C2} = 0.50, respectively), and that these site classes are on a background of stronger purifying selection (ω₁ = 0.033 in 55% of the sites) and neutral evolution (ω₀ = 1.28 or ω₁ = 1.0 in approximately 2% of the sites) (Table 2). These results are consistent with those estimated by the previous analysis with M2, where clade 1 is under more stringent purifying selection than clade 2 (Figure 3). Also in agreement with the previous analysis of M0 vs. M2, the analysis of DEF-like genes in clades 3 and 4 showed that 75% of their sites are under stronger purifying selection (ω_{2 C4} = 0.025 and ω_{2 C3} = 0.017) than genes in clades 1 and 2 (Table 3).

A similar set of branch-site tests used to study the divergence of the two clades of Poales GLO-like genes (Table 4) showed that MC, the model where shifts in ω take place after gene duplication, fits the data significantly better than other hypotheses (P < 0.0001). This result confirms that although P1 and P2 are under purifying selection (ω₀= 0.211) as previously shown with M0 vs M2, ω is especially stringent when considering the clades individually (ω_2P1 = 0.0148 and ω_2P2 = 0.01608, respectively) (Table 4).

We estimated ω in the branches preceding the origin and subsequent duplications of DEF- and GLO-like genes from Orchidaceae and Poales by employing the branch-site models MA and MA1 (Figure 3, Table 5). As discussed in "Methods", MA and MA1 assume that branches in a phylogeny are divided in foreground and background while their codon sites are sorted in classes 0, 1, 2a and 2b according to their ω value (Table 1). In site classes 0 and 1, both foreground and background branches have 0 < ω₀ < 1 or ω₁ = 1, respectively. However, in the foreground branches, MA assumes that site classes 2a and 2b have ω₂ > 1 whereas in MA1 these site classes have ω₂ = 1 (Table 1). Therefore, comparing MA with MA1 is a test of positive selection on branches and sites.

The branch-site analysis showed positive selection on the branch preceding the divergence of DEF-like genes in the Poales (Table 5). In this branch, the corresponding Bayes Empirical Bayes (BEB) analysis, included in MA, detected seven codon sites under positive selection (PP ≥ 0.90), most of which are in the region encoding the K-domain (Figure 5). The codon substitutions in these sites represent both radical and preferential changes in the chemical properties of the amino acids that they encode (Figure 5). The positions under positive selection that we detected in this lineage are different from those identified by Hernández-Hernández et al. (2007) 22 in other nodes of the phylogeny of class B genes as well as from sites in AP3 from Arabidopsis thaliana that are critical for protein – protein interactions 71.

Furthermore, we tracked the occurrence of positive selection along the branches representing the divergence of DEF-like genes in Poales following the emergence of E. elephas and S. angusifolia (Table 5, Figure 3). Although in these analyses the foreground branches have ω > 1 and sites under positive selection were identified, the LRT does not reject the null model MA1.

Similarly, we tested for positive selection in branches that in Orchidaceae precede and follow the duplications of DEF-like clades 1, 2, 3 and 4 (Figure 3, Table S3 in Additional file 1). The LRTs of these analyses did not reject the null hypothesis where over 95% of the codon sites are under purifying selection (ω₀ = 0.084).

Further analyses in different nodes of the evolution of GLO-like genes in the Orchidaceae and Poales did not reject the null hypothesis where most of the sites have ω₀ = 0.112 (Figure 4 and Table S4 in Additional file 1).

The phylogenetic analyses presented here strongly support the existence of four ancient, orchid-specific clades of DEF-like genes. These four clades and their internal branches are generally supported with PP values > 0.99. Our data strongly suggest that the sequences within each clade are orthologues, because they reproduce the systematic relationships reported for the four most derived subfamilies of the Orchidaceae: (Vanilloideae(Cypripedioideae(Orchidoideae, Epidendroideae))) 4478. This indicates that the clades are the result of ancient gene duplication events that at least precede the origin of the subfamily Vanilloideae, which according to recent estimates, emerged between 71 to 62 MYA 79. The ancient origin, conservation and unique expression (see below) of these distinct groups of orchid-specific genes suggest that they have key roles in determining the organ identities behind the characteristic orchid floral morphology 4262. Our data show that clades 1 and 2, as well as clades 3 and 4, are sister to each other. Although clades 1 and 2 seem most closely related to some other DEF-like genes in the Asparagales (Figure 1), the data available do not allow unequivocal reconstruction of the deeper phylogenetic relationships between orchid paralogous groups and the rest of the monocot sequences.

Exceptionally, we found that members of subfamily Orchidaceae have two lineages of GLO-like genes (Figure 2). We argue that these two lineages may be the result of a subfamily-specific duplication because exhaustive RACE on several cDNA pools from species in subfamilies Vanilloideae (Vanilla planifolia), Cypripedioideae (Phragmipedium longifolium) and Epidendroideae (Gongora galeata) yielded only a single GLO-like gene. The fact that GLO-like genes from Cypripedioideae and Epidendroideae seem to associate with different lineages of Orchidoideae (Figure 2), may be the result of differential gain and loss of duplicate genes during the evolution of these subfamilies.

The phylogeny of DEF-like genes provides the evolutionary context needed to interpret functional information on these sequences and suggests a particular model of perianth organ determination and evolution 4262. Mapping the known expression patterns onto the phylogenies of orchid DEF-like sequences shows that genes belonging to the same clade have the same or very similar expression domains 42. These clade-specific expression patterns suggest that the duplication events that gave rise to the ancestors of these clades were followed by transcriptional and functional differentiation of each paralogue.

Specifically, in addition to expression in the gynostemium, genes from both clades 1 and 2 are expressed in the outer and inner tepals, whereas the expression of genes in clades 3 and 4 is limited to the inner tepals or to the lip, respectively 42. Considering both the expression patterns of each clade and the events of gene duplication that generated them, it seems likely that after the first gene duplication, the ancestor of genes in clades 3 and 4 produced the differences between inner (expression "on") and outer tepals (expression "off"). The resulting morphology might still exist in the basal genus Apostasia (subfamily Apostasioideae) whose flowers do not yet possess elaborate lips 45. Similarly, the distinction between lateral inner tepals and the lip emerged after a second gene duplication affecting the ancestor of clades 3 and 4, followed by changes in the cis-regulatory regions of the duplicated genes that resulted in differential expression and determination of lip identity by clade 4 genes 42. Validating this scenario involves further characterization of the patterns of expression of DEF- and GLO-like genes in the basal and relatively species-poor families Apostasioideae, Vanilloideae and Cypripedioideae.

Although genes in clade 1 and 2 are expressed in all the perianth organs, it is unlikely that they have a completely redundant role in the determination of perianth organ identity. According to our phylogenetic reconstruction, these clades were already present in the Vanilloideae, an orchid subfamily that emerged at least 62 million years ago 79. Retention of duplicate genes for such a long time is alone a strong argument for functional diversification. Furthermore, our analyses show that the members in these clades have substantial differences in their C-terminal domains, distinct rates of nonsynonymous substitution and significant differences in their respective patterns of purifying selection (Figures 1, 3, Tables 2, 3, Additional file 2). Specifically, in the sequences of clade 2 DEF-like genes, a relatively high proportion of non-synonymous substitutions followed the duplication that generated this clade and eventually caused the truncations in the open reading frames that characterize the sequences analyzed here. Despite the high level of divergence of clade 2 orchid DEF-like genes, our detailed characterization of ω along the phylogeny of all orchid class B genes did not indicate specific branches or sites where positive selection took place. Our estimation of ω along sites, branches and specific clades documents a scenario of prevalent purifying selection (Figures 3, 4, Tables 2 to 4) that agrees with previous analyses of class B genes and other floral organ identity genes 13142280. However, we cannot completely rule out the occurrence of positive selection, since its effects on few sites during a brief evolutionary period can be masked by ensuing and continuous purifying selection 8182.

The uniform expression of GLO-like genes in the perianth organs of orchids 596061, indicates that although these genes are probably essential for proper flower development, they may not play a role in determining the different organ identities in the orchid perianth 42. However, further work is needed to determine whether the subfamily Orchidoideae-specific gene duplication (Figure 2) is associated with differential patterns of expression and function and thus with the specific morphology of this subfamily.

Similarly to DEF-like sequences, all clades of GLO-like genes were under strong purifying selection, but this is especially pronounced in the case of the single clade of genes from orchids (Figure 4). We think the differences in the ω ratio reflect the distinct selective constraints affecting duplicated class B genes. Specifically, the ratios corresponding to the generally monogenic clades of the Orchidaceae (ω = 0.043) and Liliales (ω = 0.0907) are lower than those of the rest of the Asparagales and the Poales, where there are two GLO-like loci (ω = 0.1206 and ω = 0.157, respectively). We reasoned that in orchids, and to a lesser extent in GLO-like sequences from Liliales, the lower rate of substitution reflects the stable co-evolutionary interaction between the product of a single GLO-like gene with several DEF-like interaction partners. Also, duplicated loci in clades A1 and A2 from the rest of the Asparagales, as well as P1 and P2 from the Poales, may be completely or partially redundant and thus under less stringent purifying selection than their single-copy homologues in other species (Figure 4).

The fact that each of the clades of DEF- and GLO-like sequences replicates the systematic relationships of the four most advanced subfamilies of the Orchidaceae 4478 suggests that these sequences may be useful for reconstructing the phylogeny of the Orchidaceae, provided the analysis exclusively involves orthologous sequences.

The order Poales contains 18 families, of which three are represented in our analyses of DEF-like genes: Restionaceae (Elegia elephas), Joinvilleaceae (Joinvillea ascendens) and Poaceae with the rest of the species (Streptochaeta angustifolia, Oryza sativa, Triticum aestivum, Hordeum vulgare, Zea mays). Recent phylogenetic analyses showed that the early-diverging lineage of the Restionaceae is the sister group of a clade containing Joinvilleaceae and the more derived Poaceae; the latter comprising most of the species of Poales 4765. Previous studies 63 sampled species representing the morphological transition from the typical monocot flower to the grass floret. Specifically, the flowers of the basal Elegia elephas and Joinvillea ascendens are actinomorphic, possessing two trimerous whorls of tepals and three or six stamens, respectively 63. In contrast, the floret of the early-diverging grass Streptochaeta angustifolia has 12 bracts. The trimerous arrangement of the six bracts VII to XII have been interpreted as the first and second perianth whorls 63. Most importantly, the expression of class B gene SaAP3 in the last three bracts and in the six stamens suggests that these bracts are second whorl organs, possibly a transitional form preceding the evolution of actual lodicules 6383. Moreover, the study of Whipple et al (2007) suggests that class B genes control the identity of second whorl organs in a broader sense than only petal identity. The morphologies outlined above contrast with the grass floret found in the rest of the Poaceae species analyzed here. Specifically, the floret is formed by one lemma and one palea subtending a flower formed by two or three lodicules, and the male and female reproductive organs. The grass floret is also a zygomorphic structure like the orchid flower, mostly due to frequent differential suppression of stamens from different whorls 8485 and suppression of the adaxial lodicule in most derived grasses (e.g. Hordeum, Oryza) 4884.

In this context, the evidence for positive selection in the branch that represents the divergence of Poales DEF-like genes from the rest of the monocot genes (Figure 3, Table 5) suggests that during the morphological divergence of grasses a series of nonsynonymous substitutions took place before the emergence of the characteristic grass floret. Positive selection probably continued along the branches following the divergence from Elegia elephas and Streptochaeta angustifolia (Table 5), but the corresponding signal eventually became masked by positions under purifying selection that probably encode amino acids essential for a grass-specific network of class B gene targets. The facts that class B transcription factors SILKY1 (DEF-like) and ZMM16 (GLO-like) from maize share conserved heterodimerization specificity with Arabidopsis thaliana class B proteins APETALA3 and PISTILLATA in vitro and rescue the corresponding null mutants makes it conceivable that the residues mediating the interaction between these two sets of class B proteins already existed in the most recent common ancestor of monocots and eudicots 76. This not only suggests that the mechanisms of organ identity determination for second whorl organs were established early during the evolution of the angiosperms, but also that the subsequent morphological divergence in the grasses is probably associated with the lineage-specific substitutions in the K-domain that we detected. Substitutions in this domain may have the potential for changing higher order complex formation of class B proteins and thus their binding specificity to downstream target genes, eventually enabling them to coordinate the development of novel perianth structures.

Methods for detecting positive natural selection, like the ones we employed here, are powerful tools to generate and experimentally verify interesting hypothesis on the evolution of new gene functions and phenotypes. Recently, molecular adaptation of polygalacturonase inhibitor protein, TRIM5α, feruloyl esterase A and salicylic acid methyltransferase has been tested on proteins encoded by genes where nucleotides under positive selection were modified via site-directed mutagenesis or by domain-swapping experiments generating genes encoding chimeric proteins 86878889. In the future, similar approaches could be employed to assess the functional consequences of positive selection in DEF-like transcription factors from Poales. For example, hypothesis on the transcriptional activity of these proteins could be evaluated by substituting sites under positive selection with the ones found in present-day orchids or the ones inferred to have existed in their common ancestor. The interactions of chimeric or mutagenized DEF-like proteins from Poales with other MADS-domain transcription factors and DNA could be assayed in vitro via yeast-two-hybrid and electrophoretic mobility shift assays 90, respectively, or in planta employing Fluorescence Resonance Energy Transfer (FRET) 91.

Previous phylogenetic analyses of the DEF- and GLO-like proteins identified specific motifs characteristic for certain plant lineages 3335. Similarly to what we observed in the different lineages of orchid DEF-like genes, most of these characteristic motifs evolved in the C-terminal domain of the encoded proteins. This domain is highly variable and in class B proteins it might be involved in the formation of multimeric transcription factor complexes 3992. The evolutionary importance of these differences in the C-terminal domain is supported by their long evolutionary conservation (Additional file 2). In particular, the family specific C-terminal deletions observed in orchid DEF-like proteins from clade 2 are exceptional if one considers that in all published DEF-like proteins, C-terminal deletions are only species specific (M. Mondragón-Palomino, unpublished results). The importance of these orchid-specific deletions is highlighted by recent findings on the occurrence of frameshift mutations in the regions encoding C-terminal motifs of the members of the clades of DEF- and GLO-like genes 9394. For example, the study by Kramer et al94 indicates that the C-terminal motif 'euAP3' resulted from a translational frameshift caused by a single nucleotide deletion in the ancestral motif 'paleoAP3'. Because the paleoAP3 motif is found in DEF-like proteins throughout the angiosperms (e.g. in TM6-like proteins of eudicots), whereas the euAP3 motif is found only in DEF-like proteins of higher eudicots, Lamb and Irish 95, suggested that there is a causal relationship between duplication of DEF-like genes, mutations in the C-domain, functional differentiation of DEF-like genes and the emergence of specific floral morphologies. Although the previous evidence argues for the functional importance of the C-terminal deletions in the orchid clade 2 proteins, there is surprisingly little experimental evidence for a function of the C-terminal domain, not only in orchids but also in other species. Specifically, recent independent studies with truncated DEF-like proteins from Arabidopsis thaliana and Chloranthus spicatus suggest that the C-terminal domain is not essential for floral identity 9697. Interestingly, yeast-two hybrid experiments involving OMADS3 from orchid Oncidium "Gower Ramsey", a clade 2 protein lacking the conserved C-terminal motifs PI-derived and paleoAP3 (Additional file 2), showed that this protein forms homodimers 57, opening up the possibility of novel regulatory functions for the proteins of clade 2. However, results derived from experiments using DEF-like proteins with truncated C-terminal domains of other monocots do not lend support to a particular role of this domain in dimerization 6890.

Pre-anthesis flower buds of Hypoxis villosa (Hypoxidaceae) and orchids Vanilla planifolia (Vanilloideae),Phragmipedium longifolium (Cypripedioideae), Spiranthes odorata (Orchidoideae) and Gongora galeata (Epidendroideae) were collected from the living collections of the Halle (Halle an der Saale), Wilhelma (Stuttgart) and Heidelberg Botanical Gardens. They were preserved and transported in liquid nitrogen or RNAlater (Sigma) and stored at -80°C. The choice of the orchid species was based on their subfamily membership, the availability of blooming individuals and the number of plants present in living collections. In this study, we did not include material from the subfamily Apostasioideae because there were no specimens available in the living orchid collections outside of their natural range of distribution in Southeast Asia. We included H. villosa because this species and the rest of the family Hypoxidaceae (Asparagales) show important similarities to the orchids 9899 and members of the Hypoxidaceae are frequently used as point of comparison in recent molecular phylogenies of the orchids.

Bud material was used for total RNA isolation with Biomol's reagent, following the protocol of the manufacturer. Total RNA was employed for cDNA synthesis by using a poly-T primer with Fermentas MuLV Reverse Transcriptase. MADS-box gene specific sequences were isolated by 3' RACE from at least two different cDNA pools from each species with primer pair RQVT2 (5'-CGR CAR GTG ACS TTC TSC AAR CG-3') and AB07 (5'-GAC TCG AGT CGA CAT CTG-3') under conditions specified by the manufacturer for Taq polymerase (Fermentas). PCR products of about 700 bp were cloned into vectors pGEMT (Promega) or pJET1 (Fermentas) following the protocols of the manufacturers. The ligation products where electroporated into E. coli XL1 Blue, and 50 to 100 positive clones from each ligation were selected and sequenced in both directions multiple times with vector-specific primers on an ABI 3730xl DNA Analyzer using Big Dye Terminator chemistry. The resulting sequences were assembled and managed with the program SEQUENCHER (4.5, Gene Codes Corporation). The level of identity and phylogenetic association of these sequences with previously identified monocot DEF- and GLO-like sequences was determined with different strategies using BLAST 100 to all plant sequences in GenBank. The sequences that where unequivocally identified as those of DEF- or GLO-like transcripts were employed to generate specific primers to isolate the 5' end of the sequence with the 5'-RACE kit (Roche). Assembly of the 5' and 3' partial sequences allowed us to design specific primers to isolate the corresponding full-length sequences.

We assembled two datasets by retrieving with BLAST and keyword searches all monocot DEF- or GLO-like nucleotide and amino acid sequences deposited in NCBI's GenBank until September 20^th 2007 (Table S1 in Additional file 1). The amino acid sequences were aligned using the program T-coffee 101102 to the corresponding DEF- and GLO-like conceptual amino acid translations of the cDNA sequences that we isolated from orchids. In these alignments we included as outgroup representatives the following sequences from basal angiosperms: KjAP3 (Kadsura japonica),AP3-1 (Illicium floridanum),IhAP3-1 (Illicium henryi),BsAP3 (Brasenia schreberi),NjAP3-3 (Nuphar japonica),EfAP3 (Euryale ferox),NymAP3 (Nymphaea sp.), NtAP3_1 (Nymphaea tetragona),AP3-1 (Nuphar variegata) and AmAP3 (Amborella trichopoda) for the phylogeny of DEF-like sequences. Similarly, for the phylogeny of GLO-like genes we employed the sequences of KjPI (Kadsura japonica), PI (Illicium floridanum), IhPI-1 (Illicium henryi), BsPI (Brasenia schreberi), CcPI (Cabomba caroliniana), NtPI (Nuphar tetragona), EfPI (Euryale ferox), NjPI-1, NjPI-2 (Nuphar japonica) and AtPI (Amborella trichopoda). In an initial step, the alignments were assessed with GBLOCKS 103 and T-coffee. Subsequently, after manual improvement using GenDoc 104, we used amino-acid sequence alignments as a template to align the corresponding nucleotide sequences. The nucleotide alignment of DEF-like genes contained 61 sequences and was 762 bp long, whereas the GLO-like matrix contained 74 sequences and was 696 bp long.

We analyzed the alignments of DEF- and GLO-like sequences on their complete length or separated in the positions encoding the MIK region and the C-terminal domain, as defined in 71. The program MODELTEST (3.7) 105 was employed to determine which model of nucleotide substitution fits better to each alignment according to the corrected Akaike Information Criterion 106. The parameters of the best model were employed to reconstruct the phylogenies of DEF- and GLO-like genes with Mr. Bayes (3.1.2) 107. We performed preliminary runs of Bayesian Inference to determine the point where the Likelihood estimates resulting from each generation converges on a single value. Based on this, we performed all analyses of Bayesian inference for 2,000,000 generations, sampling every 100th and with a burn-in of 3,000 generations. Finally, we obtained a consensus tree with the rest of the results and used the posterior probabilities (PP) to estimate the statistical support for each node.

We determined the level of substitution saturation of DEF- and GLO-like sequences by plotting their genetic distance versus the number of transitions and transversions. Because the number of transitions relative to that of transversions decreases with increasing divergence, the point of substitution saturation corresponds with the distance value where the plots diverge from linearity and reach a plateau 108. The pairwise genetic distances of both DEF- and GLO-like genes were obtained in PAUP 4 beta 10 (Swofford 2002) using the nucleotide substitution model GTR+I+G which best fitted the data as estimated by analysis with MODELTEST (see previous section). The pairwise number of transitions and transversions corresponding to each dataset was obtained using DAMBE 4.5.56 108 and the plots were drawn with MicroSoft Excel.

We employed the Maximum Likelihood (ML) pairwise Relative Rate Test (RRT) as implemented in the program HyPhy (v. 1.0) 109 to estimate the relative rate of substitution between the orchid and Poales DEF- and GLO-like sequences and their corresponding outgroup sequence TGDEFA and TGGLO from Tulipa gesneriana (Liliaceae), respectively. We chose these as representatives of the outgroup because the phylogenetic analysis presented in this paper showed that their relationship to the sequences from the rest of the monocots is well supported and their level of nucleotide substitution is not saturated.

The Maximum Likelihood-based RRT in HyPhy employs a Likelihood Ratio Test (LRT) to evaluate whether the sequence data fits the null hypothesis of a molecular clock (equal rates) represented by a tree of three taxa where all parameters are constrained to be equal along its branches, or an alternative hypothesis where such parameters are free to adopt different values. In order to infer the parameters of these phylogenetic hypotheses for each pair of sequences, we used the Muse-Gaut model 110 of codon substitution with nucleotide equilibrium frequencies based on their position in the codon (in Hyphy nomenclature MG94W9). The resulting parameter estimates were compared through series of Likelihood Ratio Tests that involved all pairs of sequences in relation to the corresponding outgroups. There is no simple procedure to reduce the effect of multiple comparisons in this series of P-values because the RRTs share the same outgroup and thus are not independent. Alternatively, the developers of HyPhy recommend adjusting the P-values of the LRTs for False Discovery Rate by implementing the Benjamini-Hochberg procedure (S. Kosakovsky-Pond, Hyphy on-line discussion forum). The False Discovery Rate (FDR) is the proportion of false positive results among all significant results. With the Benjamini-Hochberg correction we enforced a FDR = 0.05 by ranking all n P-values from smallest to largest, then dividing them by n and multiplying the result by 0.05 111. The LRT comparisons that rejected the null hypothesis of neutrality are those where the original P-value was smaller than the corrected value.

For consistency with analyses described in the following section, the RRT analysis excludes sequences from basal angiosperms and all positions with indels.

The ratio (ω) of the rate of nonsynonymous substitutions at nonsynonymous sites (dN) to synonymous (dS) substitutions at synonymous sites was estimated to figure out whether the coding region of a gene is under negative (purifying) selection (ω < 1), positive selection (ω > 1) or evolves neutrally (ω = 1). Variations of ω along the evolutionary history of a gene family indicate changes in both the associated selective regime and functional constraints. Because different selective regimes might affect only some branches or sites during relatively short evolutionary time 6, we analyzed the heterogeneity of selective pressures in DEF- and GLO-like genes with the program codeml from the PAML package 112113, by comparing five pairs of models (abbreviated M) that describe the pattern of codon substitution at the levels of: (a) codon sites (M7 vs M8), (b) specific branches in a phylogeny (M0 vs M2) and (c) sites from specific clades (M1a vs MC and M3 vs MD) or (d) sites from specific branches (MA and MA1). Table 1 summarizes the main features and parameters of each model of codon substitution and the datasets analyzed with them. Each of these models serves to estimate ω and other parameters. Comparing the likelihood of these estimates via a Likelihood Ratio Test (LRT) determines whether a model that considers positive selection fits the data better than one assuming neutral or purifying selection. A detailed description of each test is provided elsewhere 70112114.

We investigated the occurrence of positive selection along codon sites by comparing the model M8 with model M7 112. Then, we characterized the variation in selection pressure among branches of DEF- and GLO-like genes with a series of tests comparing the null hypothesis M0 "one ω ratio for all branches", with different alternative hypothesis based on M2 "different ω ratios", where independent ω values are estimated for specific branches 115 (Table 1).

We implemented clade and site tests of models M1a vs MC and M3 vs MD, to determine whether functional constraints differ significantly between clades after gene duplication in the two paralogous groups of DEF-like genes in Orchidaceae (Clades 1 and 2) or (Clades 3 and 4) (Figure 1), or the duplicate GLO-like genes from Poales (Figure 2). Clade models MC and MD assume that there were variations in the selective pressures among the amino acids encoded by a gene and that some of these sites also experienced changes in selective regimes at different points of their evolutionary history, such as changes occurring after gene duplication 70 (Table 1).

With branch-site models A and A1 116, we tested for the occurrence of positive selection on individual codons along specific branch groups of DEF- and GLO-like genes from the orchids and Poales. Model A assumes that the branches in the phylogeny are divided a priori in foreground and background clades, where only the former may have experienced positive selection (Table 1). In the null model MA1 ω₂ = 1, thus comparing model MA with MA1 is a direct test of positive selection on the foreground clades.

The relative fit of the parameters estimated by each of these models of codon substitution is represented by a maximum likelihood value that can be compared between nested hypotheses by a Likelihood Ratio Test (LRT). The LRT statistics are assumed to be χ² distributed with degrees of freedom equal to the difference in the number of parameters between models. The application of these models requires a phylogenetic assumption, but since the process of ML inference with program codeml is computationally intensive, for most of the tests we employed unrooted trees inferred as previously described from a smaller dataset. This dataset does not include sequences of basal angiosperms and sequences with large sections of indels (gaps) because we did not want to consider them as additional character states. The alignment of DEF-like genes is 747 bp long and contains 41 sequences, while the alignment of GLO-like genes is 666 bp long and has 61 sequences. The tests involving clade-sites models required the use of three alignments that only included members of clades 1 and 2, 3 and 4 from the Orchidaceae DEF-like genes or sequences from clades 1 and 2 of the Poales, respectively (Table 1). The phylogenies corresponding to these alignments were constructed as described previously.

To detect variations in parameter estimation, we performed and compared the analyses of each model at least twice. As with the RRTs with HyPhy, all analyses with codeml were conducted without considering positions with indels. In addition to the tests performed with PAML, we carried out an independent estimation of ω in DEF-like genes with the on-line implementation of the Hyphy package http://www.datamonkey.org. Specifically, we employed the FEL (Fixed Effects Likelihood) method to estimate the rates of nonsynonymous and synonymous substitution across codons with the same alignment and phylogeny previously used to test M0 with PAML.