Parallel evolution of TCP and B-class genes in Commelinaceae flower bilateral symmetry

Background Flower bilateral symmetry (zygomorphy) has evolved multiple times independently across angiosperms and is correlated with increased pollinator specialization and speciation rates. Functional and expression analyses in distantly related core eudicots and monocots implicate independent recruitment of class II TCP genes in the evolution of flower bilateral symmetry. Furthermore, available evidence suggests that monocot flower bilateral symmetry might also have evolved through changes in B-class homeotic MADS-box gene function. Methods In order to test the non-exclusive hypotheses that changes in TCP and B-class gene developmental function underlie flower symmetry evolution in the monocot family Commelinaceae, we compared expression patterns of teosinte branched1 (TB1)-like, DEFICIENS (DEF)-like, and GLOBOSA (GLO)-like genes in morphologically distinct bilaterally symmetrical flowers of Commelina communis and Commelina dianthifolia, and radially symmetrical flowers of Tradescantia pallida. Results Expression data demonstrate that TB1-like genes are asymmetrically expressed in tepals of bilaterally symmetrical Commelina, but not radially symmetrical Tradescantia, flowers. Furthermore, DEF-like genes are expressed in showy inner tepals, staminodes and stamens of all three species, but not in the distinct outer tepal-like ventral inner tepals of C. communis. Conclusions Together with other studies, these data suggest parallel recruitment of TB1-like genes in the independent evolution of flower bilateral symmetry at early stages of Commelina flower development, and the later stage homeotic transformation of C. communis inner tepals into outer tepals through the loss of DEF-like gene expression.

The genetic basis of flower bilateral symmetry is best understood in the model species A. majus, and involves the action of four asymmetrically expressed transcription factors [8,9,[19][20][21]. Dorsal identity is specified by the class II TCP genes CYCLOIDEA (CYC) and DICHOT-OMA (DICH), and the MYB gene RADIALIS (RAD), whereas ventral (abaxial) identity is conferred by the MYB gene DICHOTOMA (DICH). CYC and DICH are derived from a recent duplication event at the base of the Antirrhineae tribe [22], and have distinct but overlapping functions, as inferred from their mutant phenotypes [8,9]. Whereas wild type A. majus plants have five petals, four stamens and a dorsal staminode, cyc single mutants often have extra dorsal petals that are reduced in size, and a fully developed dorsal stamen. By contrast, dich single mutants only lack the internal asymmetry of wild type dorsal petals. Together with the fully ventralized, and, hence, radially symmetrical flowers of cyc:dich double mutants, these data demonstrate a role for CYC in dorsal stamen abortion, petal growth and organ number determination, and a role for DICH in shaping dorsal petal growth [8,9].
A similar range of dorsal identity functions has been assigned to CYC-like genes of other core eudicots, including Linaria vulgaris, P. sativum, I. amara, Lotus japonicus, and G. hybrida [10][11][12][13]20,23]. Furthermore, although monocots do not have a strict CYC gene ortholog due to two separate duplication events at the base of core eudicots [24], the observation that the class II TCP gene RETARDED PALEA1 (REP1) in Oryza sativa (rice, Poaceae) is expressed only in the dorsally positioned palea, suggests further recruitment of TCP genes in the evolution of monocot flower bilateral symmetry [17]. Indeed, it was recently found that bilaterally symmetrical flowers of Costus spicatus (Costaceae; Zingiberales) and Heliconia stricta (Heliconiaceae) have asymmetric TCP gene (CsTB1a and HsTBL2b, respectively) expression at early to late stages of flower development [18]. However, in contrast to the core eudicots, CsTB1a and HsTBL2b expression is restricted to the ventral, rather than the dorsolateral, side of the flower [18].
The monocot order Commelinales contains species with both radially and bilaterally symmetrical flowers, is sister to the Zingiberales and comprises five families: Commelinaceae (for example, Commelina and Tradescantia), Hanguanaceae (Hanguana only), Philydraceae (for example, Philydrella and Helmholtzia), Haemodoraceae, and Pontederiaceae [46,47] (Figure 1). Phylogenetic analyses suggest that showy, insect-pollinated and bilaterally symmetrical flowers are plesiomorphic to the Commelinales/ Zingiberales clade [47]. However, since Hanguana and Cartonema (Commelinaceae) are both progressively sister to the two major Commelinaceae clades, and have radially symmetrical flowers, it has been suggested that radial flower symmetry is ancestral to Commelinaceae [29,48]. Based on the prevalence of radial flower symmetry in its sister families -including Joinvilleaceae and Flagellariaceae -an ancestral state of bilateral flower symmetry has similarly been invoked for Poaceae [48][49][50]. Thus, flower bilateral symmetry probably evolved independently in the Commelinaceae (Figure 1), Philydraceae, Zingiberales and Poaceae, as well as in other monocots outside the commelinid clade (for example, Orchidaceae) [29]. A critical question in evolutionary developmental biology is whether similar changes at the level of gene networks, genes and/ or gene regions underlie these convergent shifts in floral form.
Flower bilateral symmetry in the genus Commelina involves differential organ development in all of the whorls (Figures 1 and 2) [51][52][53][54][55]. In C. communis and C. dianthifolia, each hermaphroditic flower comprises two whorls of three tepals, two whorls of three stamens/staminodes and a whorl of three fused carpels (Figure 2A, B) [54]. In the first whorl of both species, all three outer tepals are membranous, but the lateroventral two are distinct from the dorsal one in both size and shape ( Figure 2G). In the second whorl of C. dianthifolia all inner tepals are large and showy, but vary slightly in overall size and shape along the dorsoventral axis ( Figure 2E, H). By contrast, in the second whorl of C. communis, only the dorsolateral inner tepals are large and showy; the ventral inner tepal is small and membranous, similar to outer tepals ( Figure 2D, G) [54]. In the stamen whorls of both species, the dorsal (outer whorl) or dorsolateral (inner whorl) organs are underdeveloped staminodes that function to attract pollinators, but produce few sterile pollen grains ( Figure 2D, G, H). By contrast, the long lateroventral stamens in the outer whorl, and the medium ventral stamen in the inner whorl are showy and produce viable pollen ( Figure 2D, G, H). Finally, in the fifth whorl of both species, the dorsal carpel is underdeveloped and sterile, while the two ventral carpels are fertile [54].
Based on detailed analyses of flower development, and the known role of class II TCP genes in core eudicots, Hardy et al. (2009) recently hypothesized a role for TB1like TCP genes in establishing bilateral symmetry in early developmental stages of Commelina flower development [54]. Furthermore, we propose a second, non-exclusive hypothesis, that dorsal-specific expression of DEF/GLOlike genes defines the dorsoventral axis of symmetry in mid-to late-stage C. communis corollas. In this study, we use gene expression and micromorphological data to test three predictions of these hypotheses that: (1) TB1-like genes are asymmetrically expressed in bilaterally symmetrical C. communis and C. dianthifolia, but not radially symmetrical Tradescantia (formerly Setcreasea) pallida,   [55]. Focal taxa are in bold. Inner tepal and stamen morphology is depicted for each species as bilaterally symmetrical (gray) or radially symmetrical (black). The ancestral state of Commelinaceae inner tepal and stamen symmetry is reconstructed as radial and bilateral, respectively. A shift to radial stamen symmetry is inferred at the base of Tradescantia, whereas a late-developmental shift to weak bilateral symmetry is inferred for C. dianthifolia and C. virginica.  corollas consistent with a role in differential organ growth, (2) DEF and GLO orthologs are co-expressed in staminodes, stamens and the dorsolateral inner tepals, but not in outer tepals or the outer tepal-like ventral inner tepal of C. communis, and (3) cellular morphology of the C. communis ventral inner tepal is more similar to outer tepals than dorsolateral inner tepals, suggesting asymmetric loss of inner tepal identity in the second whorl.

Plant material
Inflorescence material was harvested from wild-collected, flowering individuals of C. communis and T. pallida in Missouri and Kansas (USA) during the summer months, and from C. dianthifolia plants grown under standard greenhouse conditions at the University of Kansas. For scanning electron microscopy (SEM) and in situ hybridization, whole inflorescences were fixed overnight in formalin acetic alcohol (FAA: , and gradually taken through an alcohol series to 100% ethanol. For quantitative reverse transcriptase polymerase chain reaction (qRT-PCR), flower organs from 10 to 16 flowers were individually dissected twice from 3 to 5 mm long buds, pooled according to identity/position, and snap-frozen in liquid nitrogen. Total RNA was extracted using TriReagent (Applied BioSystems, Carlsbad, CA, USA), DNA contamination was removed using TURBO DNase (Applied BioSystems), and 1 μg of RNA was reverse transcribed using the iScript Select cDNA synthesis kit (Bio-Rad, Hercules, CA, USA). Genomic DNA was extracted from leaf material using the DNeasy Plant Mini Kit (Qiagen, Valencia, CA, USA).

SEM
Ethanol-dehydrated inflorescences were dissected to reveal internal floral organs as necessary. Tissues were critical point dried in a Tousimis critical point drier, mounted on stubs, sputter-coated with gold and viewed with a D. Leo field emission scanning electron microscope (VTT, Espoo, Finland). For micromorphological analyses of inner and outer tepals, both adaxial and abaxial surfaces were analyzed at regions proximal, medial and distal to the floral axis.

Gene isolation and phylogenetic analysis
In order to isolate and sequence all TB1-like orthologs from genomic DNA of C. communis, C. dianthifolia and T. pallida, we used the degenerate forward primers, CYCF1, CYCF2, CYC73aaF, CYC73bF, which were previously designed to amplify CYC/TB1 genes from a wide range of angiosperm taxa, in combination with the reverse primers CYCR1 and CYCR2 [56,57]. Amplicons were cloned into the pGEM-T vector (Bio-Rad), and 5 to 10 colonies per successful PCR reaction were sequenced. To determine if the Commelinaceae sequences corresponded to CYC/REP/TB1-like genes, and to determine orthology and copy number, amino acid sequences spanning the TCP to R domain from C. communis, C. dianthifolia and T. pallida were aligned by MAAFT [58,59] and then by eye in MacClade [60] with previously designated CINCI-NATTA (CIN)-like, CYC-like, REP-like and TB1-like TCP genes [24,61]. Nucleotide alignments were subjected to maximum likelihood (ML) analyses in GARLI 0.951 and Bayesian analyses in MrBayes 3.1.2 following model optimization in MrModelTest [62][63][64]. ML analyses were run using 10 random addition sequences with 500 bootstrap replicates. Bayesian analyses were run twice for 10 million generations, sampling every 1,000 generations, with 25% of trees discarded as burn-in. Newly generated TB1-like sequences were submitted to Genbank with accession numbers [Genbank: JQ622131-JQ622135 and JQ622142]. DEF and GLO orthologs were isolated and sequenced from cDNA of C. communis, C. dianthifolia and T. pallida using previously described degenerate primers [65]. Amplicons were cloned and sequenced as previously described, and aligned nucleotide sequences were subjected to ML and Bayesian analyses as described for the TB1-like genes. Newly generated DEF-and GLO-like genes were submitted to Genbank with accession numbers [Genbank: JQ622136-JQ622141].

qRT-PCR
To compare patterns of TB1-like and B-class gene expression in mid-stage floral organs of T. pallida, C. dianthifolia and C. communis, qRT-PCR analyses using DyNAzyme II Hot Start DNA Polymerase (GE Healthcare, Pittsburgh, PA, USA) and SYBR Green I (Invitrogen, Carlsbad, CA, USA) were conducted on a DNA Engine Opticon 2 realtime PCR machine (MJ Research. Waltham, MA, USA) as previously described [66]. Two primers pairs per target gene were designed in Primer 3, and the most efficient primer pair was selected for expression analysis (Additional file 1) [67]. Where possible, primer pairs were designed to span introns to rule out DNA contamination. β-ACTIN (ACT) and EUKARYOTIC TRANSLATION ELONGATION FACTOR 1 ALPHA 1 (EF1alpha) showed little transcriptional variation across different tissues of C. communis and C. dianthifolia, and T. pallida, respectively, and were, therefore, selected as reference genes. A negative cDNA control containing RNA from dorsolateral inner tepals and lacking the reverse transcriptase enzyme was initially used as a negative control template for all qRT-PCR analyses. Furthermore, when no transcription was detected in floral material, genomic DNA was used as a positive control. For each gene the mean and standard deviation was determined for three to four technical replicates. Results were analyzed separately for one (C. dianthifolia) or two (C. communis and T. pallida) biological replicates to account for differences in developmental staging of the pooled material, and compared for consistent results between organ type and organ position. Analyses were not carried out for C. dianthifolia outer tepals due to difficulty obtaining good quality RNA from these organs.

In situ hybridization
To better determine the spatio-temporal pattern of gene expression, in situ hybridization was carried out in waxembedded inflorescence tissues of the most closely related species, C. communis (DEF, GLO and TB1a) and C. dianthifolia (DEF and TB1a). Antisense and sense gene-specific probes of CcDEF (for both species), CcGLO (C. communis only), CcTB1a (C. communis only) and CdTB1a (C. dianthifolia only) were generated using T7 and SP6 RNA Taq polymerase (Roche Applied Science, Indianapolis, IN, USA) according to the manufacturer's instructions. Each probe spanned ca. 400 bps, and included part of the C-terminal domain and the 3'untranscribed region. As an experimental control, an antisense probe was generated for CcHistone4 as described previously [57]. In situ hybridization was performed as described previously [68,69].

Flower development and perianth micromorphology
Comparative morphological analyses revealed a similar progression of flower development for C. communis and C. dianthifolia at early to mid stages. In both species, development proceeded asymmetrically, with the ventral/ lateroventral organs developing more rapidly than the dorsal/dorsolateral organs ( Figure 2J, K). In the outer tepal and stamen whorls, this asymmetry was maintained into late stage development for both species (Figure 2D, E, G, J, K). By contrast, at mid to late stages of development the ventral inner tepal of C. communis, but not C. dianthifolia, arrested prematurely. This resulted in strong asymmetry in the mature corolla of C. communis, with the ventral inner tepal resembling the small colorless outer tepals ( Figure  2D, G versus 2E, H). Unlike C. communis and C. dianthifolia, flowers of T. pallida were radially symmetrical from early to late stages of development ( Figure 2F, I, L).
To determine if the reduced outer tepal-like ventral inner tepal of C. communis also resembled outer tepals at the micromorphological level, detailed SEM analyses were conducted on mature floral organs at the cellular level. C. communis outer tepals, dorsolateral inner tepals and ventral inner tepals were all characterized by long wavy cells on both the adaxial and abaxial surfaces ( Figure 3A). Additionally, the abaxial surface of both outer tepals and ventral inner tepals had stomata; no stomata were found on either surface of the dorsolateral inner tepals, or the adaxial surface of outer tepals and ventral inner tepals ( Figure 3A). To distinguish between the alternative possibilities that stomata define outer tepal identity or determine ventral inner tepal identity, similar analyses were conducted on mature C. dianthifolia (both outer and inner tepals) and T. pallida (only inner tepals) flowers ( Figure 3B, C). As in C. communis, C. dianthifolia outer tepals were marked by stomata on the abaxial surface. However, unlike C. communis ventral inner tepals, no stomata were evident on either surface of the dorsolateral and ventral inner tepals in C. dianthifolia or T. pallida ( Figure 3B, C).

Gene duplication in the TB1-like gene lineage
Cloning and phylogenetic analyses revealed two CYC/ REP/TB1-like homologs (hereafter TB1a and TB1b) in C. communis, C. dianthifolia and T. pallida that formed two distinct clades ( Figure 4). Maximum likelihood and Bayesian analyses yielded single trees with similar topologies that largely tracked the known species phylogeny of monocots. The Commelinaceae TB1a genes formed a well-supported clade (73% ML bootstrap (MB); 99% posterior probability (PP)) sister to the TBL2 genes of Zingiberales species (100% PP), and together these clades were sister to the REP1 genes in Poaceae (98% PP) (Figure 4). Thus, orthology between Commelinacae TB1a, Zingiberales TBL2 and Poaceae REP1 genes was strongly supported by Bayesian PPs. Commelinaceae TB1b genes similarly formed a well-supported clade (99% PP) sister to the TB1-like genes of Zingiberales, Poaceae and Liliaceae species. However, the relationship between genes of each family was not well-supported (Figure 4). On a broader scale, both REP1 and TB1 clades grouped with CYC-like genes from eudicots (100% MB; 90% PP) (Figure 4), even when the more distantly related PCF-like TCP genes were used as outgroups [61] (data not shown).

Differential expression of TB1-like genes
In order to test the prediction that TB1 orthologs are expressed asymmetrically in mid stage bilaterally symmetrical C. communis and C. dianthifolia tepals, but not radially symmetrical T. pallida tepals, expression analyses were carried out on dissected flower buds using qRT-PCR ( Figure 5 and Additional file 2). In C. communis, TB1a expression was significantly higher in ventral versus dorsolateral outer tepals (average 10.4-fold difference (SD = 9.7) for two biological replicates), and ventral versus dorsolateral inner tepals (average 12.6-fold difference (SD = 9.7) for two biological replicates) ( Figure 5 and Additional file 2). Similarly, in C. dianthifolia, TB1a expression was significantly higher in ventral versus dorsolateral inner tepals (9.1-fold difference for one biological replicate) ( Figure 5). No TB1a expression was detected in either outer or inner tepals of T. pallida for either biological replicate ( Figure 5 and Additional file 2).
Unlike tepals, the correlation between TB1a expression and stamen differentiation was only weakly supported. In C. communis, although TB1a was asymmetrically expressed in dorsal staminodes versus ventral stamens, the direction of differential expression contrasted between replicates ( Figure 5 and Additional file 2). This suggests dynamic expression of TB1a at mid stage stamen development. By contrast, TB1a expression was weak but consistently higher in dorsal staminodes versus ventral stamens of C. dianthifolia (2.3-fold difference for biological replicate one ( Figure 5); detectable in dorsal staminodes but undetectable in ventral stamens for biological replicate two (data not shown)), and was not significantly different between dorsal and ventral stamens of T. pallida (1.2-fold difference for one biological replicate) ( Figure 5). In contrast to qRT-PCR analyses, no TB1a expression was detectable in early to late stage flowers of C. communis, C. dianthifolia or T. pallida using in situ hybridization (data not shown). This suggests that TB1a transcripts are below the level of detection using this method.

Arabidopsis tha t t liana TCP C C 4 P P Pha P P la l l enopsis hy h brid TCP C C 1
Antirrhinum maj a us CIN

Pha P P la l l enopsis hy h brid TCP C C 3 P P Arabidopsis tha t t liana TCP C C 3 P P
Allium cepa_ e TCP

Dendr d d obium hy h brid TCP C C 1 Pha P P la l l enopsis hy h brid TCP C C 2 P P Pha P P la l l enopsis hy h brid TCP C C 5 P P Costu t t s sp s icatus t t TB T T L B B 1a
Lilium long l l iflo fl fl rum TCP C C 1

Commelina communis TB T T 1b Zea may a s TB T T 1 Zingiber offici o nalis TB T T L B B 1a OmTB T T L B B 1a
Sorghum bicolo l l r TB T T 1 QRT-PCR analyses of TB1b genes revealed different expression patterns relative to TB1a paralogs ( Figure 5 and Additional file 2). In T. pallida, TB1b was undetectable in all floral organs and leaves, despite amplification of genomic DNA ( Figure 5 and Additional file 2). In C. communis, TB1b expression was highest in leaves, with low ( Figure 5) to no (Additional file 2) detectable expression in all other organs. For both replicates, no significant differences in transcript levels were detected between dorsal and ventral organs ( Figure 5 and Additional file 2). Finally, for C. dianthifolia, no TB1b expression was detected in any floral organ, again despite amplification of genomic DNA ( Figure 5).

Differential expression of B-class genes
To test the prediction that B-class gene expression is exclusively reduced or absent from the outer tepal-like ventral inner tepal of C. communis, DEF and GLO genes were isolated from all three focal species, and both qRT-PCR and in situ hybridization analyses were carried out on inflorescence tissues. ML and Bayesian analyses supported the isolation of both DEF and GLO orthologs from each species, and relationships among a larger set of DEF and GLO genes largely tracked the known species tree for monocots (Additional file 3). As predicted, GLO was expressed in inner tepals and stamens of C. communis, C. dianthifolia and T. pallida ( Figure 5 and Additional file 2). Furthermore, analyses revealed low (T. pallida based on one biological replicate) to no (C. communis based on two biological replicates) expression in outer tepals of representative Tradescantia and Commelina species ( Figure 5 and Additional file 2). In C. communis, GLO was expressed similarly in dorsolateral and ventral inner tepals, with expression being significantly higher in dorsal staminodes than ventral stamens (2.1-fold difference for two biological replicates (SD = 0.27)) ( Figure 5 and Additional file 2). A similar asymmetric expression pattern was detected in the stamen whorls of C. dianthifolia ( Figure 5). However, in T. pallida there was no significant difference in expression between ventral versus dorsal inner tepals and stamens for one biological replicate ( Figure 5 and Additional file 2).
Consistent with qRT-PCR analyses, in situ hybridization analyses revealed GLO expression in the inner tepal and stamen whorls of young C. communis flower buds, with transcripts being absent from the center of the flower ( Figure 6A). The second to fourth whorl expression was maintained during mid to late stages of Relative quantitative gene expression in dissected flower tissues. TB1a expression is significantly stronger in ventral versus dorsal tepals of C. communis and C. dianthifolia but not T. pallida, and is expressed asymmetrically in different directions in stamens of all three species (top panel). TB1b expression is low or absent in floral tissues, but is expressed strongly in leaves (second top panel). GLO expression is significantly stronger in staminodes versus stamens of C. dianthifolia and C. communis, and is low to undetectable in outer tepals and gynoecia (second bottom panel). DEF expression is significantly stronger in dorsal versus ventral inner tepals of C. communis, being undetectable in the latter (bottom panel). b-ACT and EF1a were used as internal controls. Experiments were conducted with three to four technical replicates. Bars show mean and standard deviation. Second biological replicates were analyzed separately and are shown in Additional file 2. White bars, whole organs; blacks bars, dorsal organs; gray bars, ventral organs. Asterisks denote significant differences between dorsal and ventral organ comparisons. nd, not detected.
C. communis flower development, with transcripts becoming detectable in the gynoecium from mid to late stages ( Figure 6B, C). Since qRT-PCR for all three species, and in situ hybridization for C. communis, revealed no correlation between GLO expression and inner tepal morphology, in situ hybridization experiments were not carried out for C. dianthifolia or T. pallida GLO orthologs.
QRT-PCR analyses revealed DEF gene transcripts in inner tepals and stamens of C. communis, C. dianthifolia and T. pallida ( Figure 5 and Additional file 2), with low to no expression of DEF in outer tepals and gynoecia of C. communis and T. pallida ( Figure 5 and Additional file 2). In C. communis, DEF expression was significantly higher in dorsal staminodes versus ventral stamens (3.5fold difference for two biological replicates (SD = 0.32)), and in contrast to dorsolateral inner tepals, was completely absent from ventral inner tepals for two biological replicates ( Figure 5 and Additional file 2). In C. dianthifolia, DEF expression was not significantly different between dorsal staminodes and ventral stamens. However, in striking contrast to C. communis, DEF was significantly higher in ventral versus dorsolateral inner tepals (2.3-fold difference for one biological replicate). Finally, in T. pallida, DEF expression was not significantly different between dorsal and ventral stamens, and was highly variable in dorsolateral versus ventral inner tepals between two biological replicates ( Figure 5 and Additional file 2).
Consistent with qRT-PCR, in situ hybridization analyses of DEF showed much higher expression levels in dorsal compared to ventral inner tepals of early to late stage C. communis flower buds ( Figure 6D-I). However, in the stamen whorls, this dorsoventral gradient of DEF expression only became evident following late-stage differentiation of the ventral stamens ( Figure 6D-I). Comparable analyses in C. dianthifolia revealed DEF expression in both dorsolateral and ventral inner tepals, dorsal staminodes and ventral stamens during early to midstages of development ( Figure 6J-M). Strong expression was maintained in both the dorsolateral and ventral inner tepals into late development ( Figure 5N). However, DEF expression became gradually weaker in ventral stamens relative to dorsal staminodes during late stage anther differentiation ( Figure 6M, N). No antisense transcripts of DEF were detectable in outer tepals or gynoecia of C. communis or C. dianthifolia. Furthermore, no staining was observed in control sections of either species using DEF or GLO sense probes ( Figure 6O and data not shown).

Discussion
A major conclusion from this study is that TB1-like genes have been independently recruited in establishing Commelinaceae corolla bilateral symmetry in early development, similar to several core eudicot lineages and rice [8][9][10][11][12][13]17]. Furthermore, micromorphological evidence and expression of the B-class MADS-box gene CcDEF suggests that late stage asymmetry between the dorsolateral and ventral inner tepals of C. communis is due to a homeotic transformation of the ventral inner tepal into an outer tepal. In the sections below, we discuss the implications of these results as they relate to our understanding of convergent trait evolution.

Parallel evolution of TB1-like genes across angiosperms
Bilateral flower symmetry can be achieved through the asymmetric loss or reduction of organs (structural zygomorphy) or the modification of organs (presentation zygomorphy) within one or more whorls, and can be present in early and/or late stages of development [48]. Although both types of zygomorphy are found within the monocots, structural zygomorphy is particularly prevalent within the Asparagales (for example, Orchidaceae), Arecaceae, Dasypogonaceae, Zingiberales (for example, Zingiberaceae), Commelinales (for example, Commelinaceae) and Poales (for example, Poaceae) [18,27,29,48,70]. In most cases, bilateral symmetry affects the inner tepals and stamen whorls. However, outer tepal, inner tepal, stamen and gynoecial zygomorphy have all evolved in the monocots multiple times independently [48].
Maximum parsimony character state reconstructions support a gain of tepal bilateral symmetry within the Commelina-containing Commelineae clade of Commelinaceae ( Figure 1). Available data suggest that members of the Commelineae have flowers that are strongly bilaterally symmetrical in early development, but vary in the strength of structural tepal zygomorphy in late development [54,71] (this study). For example, although all are strongly zygomorphic in early development, at anthesis C. communis inner tepals are highly differentiated along the dorsoventral axis, whereas C. dianthifolia and Murdannia nudiflora inner tepals are only slightly differentiated in size and shape. Based on evidence from core eudicots, it has been hypothesized that asymmetric expression of CYC/TB1-like genes underlies early stage structural zygomorphy in the Commelineae [54]. Consistent with this hypothesis, TB1a is expressed asymmetrically in tepals of C. communis and C. dianthifolia, but not T. pallida, at mid stages of flower development. However, in contrast to the asymmetric expression of CYC and DICH in the model core eudicot A. majus and REP1 in O. sativa, expression is significantly higher in the ventral, as opposed to dorsolateral, tepals [8,9,17] (this study). It is predicted that the asymmetrical expression of TB1a in C. communis and C. dianthifolia is initiated in early flower development and has no effect on late stage development when the ventral inner tepal of C. communis arrests growth (see next section).
has been assigned to IaCYC of the rosid core eudicot Iberis amara (Brassicaceae) in mid to late stage dorsal petal development [8,9,11]. By contrast, at mid to late stages of A. majus flower development, CYC expression in the dorsal petals actually increases cell proliferation and elongation relative to the ventral petal [8,9]. Together with results from Commelinaceae, these data support developmentally and taxonomically distinct roles for CYC in establishing early to late stage perianth organ growth, and suggest that parallel recruitment of CYC/TB1 genes in floral bilateral symmetry is not limited to the dorsal side of the flower. Indeed, a ventral pattern of expression was recently demonstrated for CsTBL1a in bilaterally symmetrical C. spicatus (Costaceae, monocot) flowers [18].
Further studies testing the involvement of CYC/TB1 genes in transitions from radial to bilateral flower symmetry will require functional tests, and should aim to elucidate whether changes in CYC/TB1 expression are due to cis-regulatory or upstream changes. In A. majus, the NAC family protein CUPULIFORMIS (CUP) has been implicated in the positive regulation of CYC [72]. Thus, expression analyses of CUP-like genes in perianth organs of Commelinaceae and other monocots varying in flower symmetry might be a good starting point to address this question.

Parallel evolution of B-class genes in monocots
It has been hypothesized that changes in the expression of organ identity genes can explain the evolution of flower bilateral symmetry in certain monocots by altering the identity of a subset of organs within a floral whorl. In the case of orchids, one of several DEF-like genes evolved an asymmetric expression pattern following gene duplication, and is implicated in modification of the ventrally positioned inner tepal into the characteristic lip [25,27,[73][74][75]. Our data also support a role for DEF-like gene evolution in modification of the C. communis ventral inner tepal. However, unlike orchids, the modification of DEF-like gene expression in C. communis was not preceded by gene duplication and is presumably not associated with changes in protein function at the biochemical level. Furthermore, unlike C. dianthifolia and T. pallida, the ventral inner tepals of C. communis flowers have stomata on their abaxial surface similar to outer tepals. This micromorphological marker correlates both with the general outer tepallike appearance of the ventral inner tepal and the complete absence of DEF-like gene expression during early to late stages of development.
Together, these studies suggest parallel evolution of DEF-like genes in the independent origins of monocot flower bilateral symmetry. Future studies in other monocot families will be required to determine the generality of these results, and to test whether evolution of DEF-like gene expression can explain petaloidy in the stamen whorl [29]. Furthermore, a key question remains as to whether evolution of DEF-like gene expression can be explained by cis-regulatory or upstream changes. In A. thaliana, SUPERMAN (SUP), FLORAL ORGAN NUM-BER1 (FON1), CUP-SHAPED COTYLEDON1 (CUC1), and CUC2 proteins negatively regulate both B-class genes in the sepal whorl [76]. Thus, expression of these proteins in the ventral inner tepal of C. communis could explain the loss of DEF-like expression in this organ. However, this scenario would require that SUP-, FON1and CUC1/2-like proteins do not regulate the C. communis GLO-like gene, which is expressed normally in the ventral inner tepal.

Conclusions
Together with other studies, our gene expression data on three morphologically diverse species of Commelinaceae suggest a parallel role for two major transcription factors in the independent evolution of angiosperm flower bilateral symmetry. In the case of class II TCP genes, changes in expression are correlated with early developmental shifts from radial to bilateral flower symmetry in several core eudicots and commelinid monocots (rice, Costus and Commelina). This supports a conserved role for class II TCP genes in organ growth across angiosperms, and suggests either evolutionary constraint on the flower symmetry gene network or the involvement of few genes in the establishment of floral meristem symmetry. Evolution of DEF-like genes is also implicated in shifts from radial to bilateral flower symmetry, at least within divergent monocots (Phaelaenopsis and C. communis). However, in contrast to class II TCP genes, changes in both function (Phaelaenopsis) and regulation (C. communis) of B-class genes are implicated in late developmental shifts in within-whorl organ identity. Further studies are required to test the generality of class II TCP and B-class gene evolution in diversification of monocot flowers, and to decipher whether/why the same genes have been the targets of repeated selection across millions of years of angiosperm diversification.

Additional material
Additional file 1: Primers used for RT-PCR. Primer sequences used for amplification of target and housekeeping genes from Commelina communis, C. dianthifolia and Tradescantia pallida are listed in table format.
Additional file 2: Second biological replicate for relative quantitative gene expression in dissected flower tissues. In agreement with the first biological replicate ( Figure 5), GLO expression is significantly higher in dorsal staminodes versus ventral stamens of Commelina communis, but is similar across inner tepals of both Tradescantia pallida and C. communis (left). DEF is again undetectable in ventral, but not dorsal, inner tepals of C. communis. By contrast, DEF expression is higher in ventral versus dorsal inner tepals of Tradescantia pallida (second from left). Similar to the first biological replicate, TB1a is undetectable in T. pallida inflorescence tissues, and is expressed significantly more strongly in both outer and inner ventral versus dorsal tepals of C. communis (second from right). TB1b is undetectable in inflorescence organs of C. communis, although the primers amplify genomic DNA, and is undetectable in inner tepals and gynoecia of T. pallida (right). b-ACT and EF1a were used as internal controls. Experiments were conducted with three to four technical replicates. Bars show mean and standard deviation. White bars, whole organs; blacks bars, dorsal organs; gray bars, ventral organs. Asterisks denote significant differences between dorsal and ventral organ comparisons. gDNA, genomic DNA; Inf, whole inflorescence; Dot, dorsal outer tepal; Vot, ventral outer tepal; Dit, dorsal inner tepal; Vit, ventral inner tepal; Dst, dorsal stamens/staminodes; VSt, ventral stamens; Gyn, gynoecia; nd, not detected.
Additional file 3: Best maximum likelihood phylogeny of monocot B-class genes. Maximum likelihood bootstrap values above 69% (left) and Bayesian posterior probabilities above 89% (right) are shown for each branch. Asterisks denote 100% support. Dashes indicate bootstrap support below 70% only when the corresponding posterior probability is shown. The focal genes of this study are highlighted in gray boxes.