Evolution of regulatory interactions controlling floral asymmetry

One approach to studying the evolution of novel traits is to analyse mutants that eliminate the trait. Such mutants have been termed atavistic as they give phenotypes resembling the presumed ancestral condition (Cantu and Ruiz,1985; Hall, 1995). The genes identified by atavistic mutations can provide a convenient starting point for analysing how changes in gene interactions can lead to phenotypic novelty (Babu et al., 2004; Gibson and Honeycutt, 2002; Tautz, 2000). By comparing the network of interactions in species with and without the relevant trait, it should be possible to pinpoint some of the key steps underlying evolutionary change. Here, we adopt this approach for the case of dorsoventral asymmetry in flowers, a trait that is thought to have arisen many times independently during evolution (Donoghue et al.,1998; Endress,1999; Stebbins,1974) and for which key regulatory genes giving atavistic mutant phenotypes have been identified.

Several genes controlling floral asymmetry have been identified in Antirrhinum majus. Antirrhinum flowers are asymmetric along their dorsoventral axis, having distinct dorsal, lateral and ventral organ types. Asymmetry is most evident in the petal and stamen whorls and depends on the action of the duplicate genes CYCLOIDEA (CYC) and DICHOTOMA (DICH). CYC and DICH are both expressed in the dorsal domain of the flower meristem and continue to be expressed at later stages in dorsal floral organs, although expression of CYC occurs in a wider region than that of DICH(Almeida et al., 1997; Luo et al., 1999; Luo et al., 1996). cyc dich double mutants have radially symmetric ventralised flowers, while single cyc or dich mutants have partially ventralised flowers, consistent with sub-functionalisation of paralogs following the CYC-DICH duplication (Gubitz et al., 2003; Hileman and Baum,2003). Developmental analysis of single and double mutants has shown that CYC and DICH can enhance or repress organ growth,depending on developmental stage and organ type, with CYC having a stronger phenotypic effect than DICH(Almeida et al., 1997; Luo et al., 1999; Luo et al., 1996).

CYC and DICH encode transcription factors belonging to the TCP family, many members of which influence patterns of plant cell growth and proliferation (Almeida et al.,1997; Crawford et al.,2004; Kosugi and Ohashi,1997; Kosugi and Ohashi,2002; Kosugi et al.,1995; Luo et al.,1999; Luo et al.,1996; Nath et al.,2003; Palatnik et al.,2003; Tremousaygue et al.,2003). Members of this family can be grouped into two classes, I and II, based on sequence similarity in the TCP domain and the consensus DNA-binding sequence. A good candidate for a direct target of CYC and DICH is RADIALIS (RAD). RAD is expressed in the dorsal domain of floral meristems, in a manner that depends on CYC and DICH. Plants mutant for RAD have almost fully ventralised flowers, retaining only slight dorsal identity in their uppermost regions, suggesting that many of the effects of CYC and DICH are mediated through RAD. RAD acts antagonistically to DIVARICATA (DIV), which promotes ventral identity(Almeida et al., 1997; Corley et al., 2005; Galego and Almeida, 2002). RAD and DIV encode related MYB-like proteins that are thought to compete for common protein or DNA targets.

Unlike Antirrhinum, Arabidopsis has radially symmetrical flowers. The TCP1 gene of Arabidopsis is the closest homologue to CYC/DICH and is expressed asymmetrically in the dorsal domain of young flower meristems and axillary meristems(Cubas et al., 2001). This indicates that the common ancestor of Antirrhinum and Arabidopsis had a CYC/TCP1-related gene that was asymmetrically expressed even though the flowers were presumably radially symmetric. In contrast to CYC and DICH, TCP1 is only transiently expressed at very early stages of flower development.

To understand the evolution of asymmetry, we analysed how CYC acts in Antirrhinum and Arabidopsis. We show that CYC binds to DNA, and use random binding site selection to define the consensus binding site. Sequences matching this site are found in the RAD promoter and intron, and are bound by CYC, indicating that RAD may be a direct target of CYC. CYC is also able to activate the RAD gene of Antirrhinum in the context of Arabidopsis but is unable to activate endogenous RAD-like genes of Arabidopsis, which lack sequences identical to the consensus binding site. CYC is nevertheless able to act in Arabidopsis, reducing leaf and increasing petal size, through changes in cell proliferation and expansion. By activating CYC protein at specific times, we show that CYC can act at various stages of Arabidopsis organ growth. Taken together, our results suggest that ancestral TCP1/CYC-like genes played a role in regulating a network of target genes involved in growth and development. Some of these interactions could have been retained in both Antirrhinum and Arabidopsis,accounting for the common developmental effects of CYC expression in both species. In addition, an interaction between CYC and RAD was established or preserved specifically in the Antirrhinum lineage.

Arabidopsis seeds were sterilised and grown on growth media [GM;1× Murashige and Skoog salt mixture, 1% (w/v) sucrose, 100 μg/ml inositol, 1 μg/ml thiamine, 0.5 μg/ml pyridoxin, 0.5 μg/ml nicotinic acid, 0.5 mg/ml MES, 0.8% (w/v) agar, pH 5.7] with or without 50 μM kanamycin (Kan). The media was supplemented with 10 μM dexamethasone (DEX;stock solution 25 mM in ethanol) or ethanol 0.04% (v/v) as a control. Plants were grown on a mix of John Innes Potting Compost No 1, vermiculite and grit in the ratio 1.5:1:1 (by volume) at 20°C with 16 hour light/8 hour dark cycles. All Arabidopsis transgenic lines (35S::CYC:GR, RAD::RAD) were in the Columbia background, which was also used as the wild-type control.

The pRSET-b vector (Invitrogen) was used for production of recombinant CYC protein in E. coli (plasmid pJAM2093). BL21(DE3) SBET pLysE(Schenk et al., 1995; Studier and Moffatt, 1986) E. coli cells were transformed with the construct and the recombinant His-CYC fusion protein was expressed under the control of the T7 promoter and purified from the soluble fraction or from inclusion bodies produced in bacteria, as described previously (Kosugi and Ohashi, 1997; Kosugi and Ohashi, 2002). The conditions for the DNA-binding reaction and electrophoresis, and the DNA probes used were as described(Kosugi and Ohashi, 2002). For the EMSA with the RAD promoter and intron, the DNA probes were obtained by PCR using the primer combinations 5733/5734 and 5735/5736 with the plasmid pJAM2283, respectively (5733,5′-ACCGTAGAACATTATAGACAACA-3′; 5734,5′-CACCAACAAAACCTTCCACATAG-3′; 5735,5′-GCTATAACGTCGATGTGTCTC-3′; 5736,5′-ATTCTAAAAACCACGAGAGTCC-3′). For the EMSA with the RADpromoter and intron, the primer combinations 5733-5734 and 5735-5736 were used, respectively.

Random binding site selection for the His-CYC protein was performed using the oligonucleotide BS18N, as described previously(Kosugi and Ohashi, 1997; Kosugi and Ohashi, 2002). The DNA-protein complex was separated by polyacrylamide gel electrophoresis and the DNA recovered from the gel was amplified by PCR. After the fifth round of selection, the DNA amplified by PCR was cloned into pGEMT-easy vector(Promega) for sequencing.

The binary plasmid coding for the CYC protein fused to the rat glucocorticoid receptor (GR) (Lloyd et al., 1994) under the control of cauliflower mosaic virus (CaMV)35S promoter was obtained as follows. The CYC open reading frame was amplified by PCR using the primers 5′-CYC(5′-CGGGATCCATGGTTGGGAAG-3′) and 3′-CYC(5′-GAAGATCTTTGATGAACTTGTGCT-3′) from plasmid pJAM2095. The forward primer introduced a Kozac sequence and a BamHI site before the start codon and the reverse primer removed the CYC stop codon and created a BglII restriction site at the C terminus. This PCR fragment was digested with BamHI and BglII and cloned into the BamHI restriction site of the plasmid GR-pBluescript (pRS020)(Sablowski and Meyerowitz,1998) resulting in an in-frame translational fusion at the C terminus of CYC with the rat glucocorticoid hormone-binding domain (pJAM2387). The CYC cDNA was sequenced to check for PCR errors and to check if the protein was in frame with the GR sequence. This fusion was called CYC:GR. The double CaMV 35S promoter and 35S polyadenylation signal from CaMV were isolated as a 1.56 kb KpnI/EcoRV fragment from pJIT60(Guerineau, 1993) and cloned into a SacI filled in/KpnI sites in pGreenII 0029 (plasmid pJAM2388). A 1.7 kb XbaI/BamHI fragment containing the CYC:GR fusion was isolated from plasmid pJAM2387 and the ends filled in. This fragment was cloned into the SmaI dephosphorylated site of plasmid pJAM2388,adjacent to the double 35S promoter and the orientation was checked by restriction mapping. This construct was called 35S::CYC:GR (plasmid pJAM2389).

Arabidopsis plants were transformed with the 35S::CYC:GR construct by floral dipping (Clough and Bent,1998). The Agrobacterium strain used was C58C1 pGV101 pMP90. Kan-resistant transformants (T1 generation) were selected on GM plates. Approximately, two thirds of T2 plants (segregating 3:1 on Kan) showed a phenotype when also germinated on DEX. T3 progeny of plants that showed a phenotype on DEX-containing media were hemizygous, whereas the progeny of plants that did not show a phenotype were homozygous. 35S::CYC:GR homozygous plants were crossed to other homozygous independent lines and to wild type. Double hemizygous plants never showed a phenotype when grown in DEX-containing media, whereas the hemizygous lines obtained from the cross with wild type showed the phenotype.

For Northern-blot analysis, total RNA was extracted with TRI-Reagent(Sigma), according to the manufacturer's instructions, from 35S::CYC:GR T2 seedlings grown on Kan and DEX, with and without a phenotype. The RNA was blotted onto Hybond N+ (Amersham) according to the manufacturer's instructions and probed with CYC or NPTII (neomycin phosphotransferase II) probe.

For RT-PCR analysis, RNA was extracted using RNeasy Plant mini kit(Qiagen), from duplicate samples of tissue from 21-day-old 35S::CYC:GR RAD::RAD plants grown on GM media, 6, 18 and 48 hours after transfer to 10μM DEX plates. cDNA was prepared from 2 μg of total RNA using Thermoscript RT-PCR system (Invitrogen). Aliquots of the cDNA were used as template for PCR with gene-specific primers.

Scanning electron microscopy (SEM) analysis was carried out on `2-Ton'epoxy (Devcon) replicas of Arabidopsis petals and leaves as previously described (Carpenter et al.,1995; Green and Linstead,1990). These replicas were sputter coated with gold palladium,analysed and photographed with a Philips XL 30 FEG SEM. Petal and leaf areas were calculated using a program written in MatLab called CellFinder. Medians were obtained by analysis of variance. The analyses were carried out using the statistical package GenStat for Windows 7th Edition (VSN International,Oxford, UK).

To determine whether CYC would bind to the consensus DNA-binding sequences for TCP proteins (Kosugi and Ohashi,1997; Kosugi and Ohashi,2002), the CYC open reading frame was fused in-frame with a HIS-Tag epitope and the recombinant protein purified from E. coliextracts using a metal affinity resin. The protein was present in both the insoluble and soluble fractions. Electrophoretic mobility shift assays (EMSA)were carried out using a 21 bp double-stranded oligonucleotide probe that contained the overlapping consensus binding sites for both class I and class II TCP proteins (GTGGTCCCAC) (Kosugi and Ohashi, 2002). Incubation of the probe with CYC protein from the soluble fraction showed band retardation(Fig. 1A, lane 1), indicating that CYC protein can bind to these DNA sequences. Mutations in the core of the oligonucleotide abolished binding (Fig. 1A, lanes 2 and 3), indicating that these nucleotides were important for interaction with CYC. According to the sequence of the TCP domain, CYC belongs to the class II subfamily of TCP proteins(Kosugi and Ohashi, 2002) and should therefore bind preferentially to a class II binding site GTGGNCCC. However, when various mutations were introduced in the oligonucleotide to test this, the DNA-binding behaviour of CYC protein was more similar to class I than class II proteins (Fig. 1A, lanes 4 and 5). In particular, mutations that disrupt class II binding did not disrupt CYC binding (Fig. 1A, lane 4). It was only when mutations that affect class I binding were introduced that CYC binding was affected, suggesting that despite sequence similarity to class II proteins, CYC DNA-binding specificity resembles that of class I proteins (Fig. 1A, lane 5). To determine the optimal sequence specificity for the binding of CYC protein, we conducted a random binding site selection analysis using CYC recombinant protein. After five rounds of selection, the program MEME(meme.sdsc.edu)was used to align the sequences of the selected oligonucleotides and the consensus binding site GGNCCCNC was obtained(Fig. 1B), confirming that the CYC binding site is more similar to that of TCP class I than class II proteins.

In Antirrhinum, genetic analysis has suggested that RADcould be a potential target gene of CYC(Corley et al., 2005). To investigate whether this interaction could be direct, the RADpromoter and intron were screened for CYC-binding sites. Two sites matching the CYC consensus were found 1.5 kb upstream of the start codon of the RAD open reading frame, within 24 bp of each other(Fig. 2A). A third matching site was found in the intron. PCR fragments containing promoter and intron putative binding-sites were amplified and used as probes in an EMSA with recombinant CYC protein (Fig. 2B). CYC was able to bind to these sequences, supporting the idea that RAD is a direct target of CYC.

To investigate whether CYC expression could affect growth in Arabidopsis, the coding region of CYC was fused to that of the rat glucocorticoid receptor and expressed under the control of the 35S promoter(35S::CYC:GR). Sixteen transformants were obtained, of which 10 had a clear developmental phenotype when grown on the glucocorticoid inducer dexamethasone(DEX). Northern-blot analysis on three independent transformants segregating for the construct showed that the severity of the phenotype correlated with the level of CYC mRNA accumulation(Fig. 3). Curiously, only plants hemizygous for the construct showed strong expression, indicating that expression was silenced in homozygotes. To check whether this was a consequence of transgene dosage or allelic suppression, 35S::CYC:GR homozygotes were crossed to other independent homozygous lines and to wild type. Double hemizygous plants did not show a phenotype when grown in DEX media, whereas the hemizygous lines obtained from the cross with wild type showed the phenotype. These results indicate that two doses of the transgene result in gene silencing, a phenomenon that has been recorded previously for some transgenes (de Carvalho et al.,1992).

One aspect of the phenotype of DEX-induced 35S::CYC:GR plants was that the petals were bigger than those of wild type(Fig. 4). To determine the stage at which this size change became evident, flowers were staged according to the number of nodes below the first flower with visible petals, termed flower 1 (Butenko et al., 2003; Patterson, 2001). Petals from 35S::CYC:GR flowers became larger than wild-type petals after flower 2(Fig. 4A-C). Measurements of surface area of mature petals (Fig. 4C) showed that 35S::CYC:GR petals were about 1.5 times larger in area for all four petals. In addition, floral organ senescence was delayed as petals remained attached to the silique for a prolonged period, in contrast to wild-type organs, which abscise at flower 5 or 6. To investigate the correlation between petal area and senescence further, we measured petal areas in two mutants with delayed or deficient abscission: inflorescence deficient in abscission (ida), in which floral organs remain attached to the plant body after the shedding of mature seeds(Butenko et al., 2003); and ethylene-response gene (etr1-1), an ethylene response mutant that shows delayed floral organ abscission(Bleecker et al., 1988; Bleecker and Patterson, 1997; Patterson and Bleecker, 2004). In neither of these mutants was petal size affected to the same degree as 35S::CYC:GR plants, although there may have been a slight increase in size for etr1-1 (data not shown).

To determine whether the increase in petal area reflected changes in cell proliferation or expansion, the adaxial epidermal layer of mature petals was analysed by SEM (Fig. 5). The average cell area from the tip and middle regions of the petal epidermis was determined. This showed that the cells of 35S::CYC:GR petals were about 1.6(middle region) or 1.8 (tip region) times bigger than the epidermal cells of wild-type petals. These values are comparable with the overall increase in petal area (1.5 times), indicating that the effect of CYC on petal area could be largely accounted for by increased cell expansion.

To determine the developmental stage at which CYC exerts its effects on petal development in Arabidopsis, 35S::CYC:GR plants were transferred to DEX after bolting, when the first two flowers had opened in the main inflorescence. The inflorescence was then allowed to develop further. All flowers showed delayed senescence and abscission, indicating that CYC can have an effect late in development even after the flowers have opened.

35S::CYC:GR plants grown in DEX-containing media, were dwarfed and exhibited small oval-shaped leaves (Figs 6, 7, 8). Mature leaves of 35S::CYC:GR plants were on average eight times smaller in surface area than those of wild type (Fig. 7A,C). To investigate whether the difference in leaf size was due to a difference in cell size or in cell number, the adaxial epidermal layer of leaf 4 was analysed by SEM at different stages of development(Fig. 7B,D). At maturity, cells of 35S::CYC:GR leaves were about four times smaller than those in wild-type plants, suggesting that contrary to what was observed in the petals, CYC has a role in repressing cell expansion in the leaves. However, if reduced cell size was solely responsible for the reduction in overall leaf size, the leaves of 35S::CYC:GR plants should be four times smaller rather than observed value of eight times. This indicates that 35S::CYC:GR leaves also had about half the number of cells of wild type. To determine whether this reflected an early arrest of cell proliferation, epidermal cells of younger leaves were analysed. At day 14, cells of 35S::CYC:GR leaves were larger than those of wild type,suggesting that cell division had arrested early and the cells were starting to differentiate (Fig. 7B,D). Moreover, evidence of recent cell divisions could be seen in the epidermal cells of wild-type leaves (formation of new stomata), whereas in 35S::CYC:GR leaves no such events could be seen. Thus, CYC reduces both cell proliferation(by promoting early arrest of cell division) and cell expansion in leaves.

To determine the developmental stages at which CYC can affect leaf growth,35S::CYC:GR plants were grown in media with DEX and transferred to media without DEX at day 10 and left to grow to maturity(Fig. 8B). At the time of transfer, the first two rosette leaves were about 2 mm wide and the third leaf was 0.5 mm wide. Analysis of consecutive rosette leaves of mature plants showed that from the third leaf onwards, the leaves increasingly resembled leaves of plants grown continually in the absence of DEX (compare Fig. 8A with 8B).

This showed that removal of CYC could restore leaf growth at early stages of development, before they were about 0.5 mm wide. Moreover, the effect was greater the earlier that CYC was removed.

Conversely, when plants were grown without DEX and moved onto DEX at day 10(Fig. 8C), growth repression was first detected for leaves 3-4 and became progressively more pronounced in later leaves (compare Fig. 8C with 8D). These results suggest that activation of CYC can influence leaf development at early stages, before leaves are 0.5 mm wide, leading to a reduction in final leaf size.

The Arabidopsis genome contains six RAD-like genes none of which containing the consensus CYC binding-site within 2 kb upstream or downstream of the coding regions (C.B., M.M.R.C. and E.C.). To investigate whether the effects of CYC in Arabidopsis might nevertheless involve activation of these genes, RT-PCR was performed on 35S::CYC:GR plants 6, 24 and 48 hours after being transferred to DEX-containing media. No increase in expression of the RAD-like genes was observed, suggesting that CYC does not affect expression of any of these genes.

To test whether CYC could activate the Antirrhinum RAD gene in the context of Arabidopsis, 35S::CYC:GR Arabidopsis plants were crossed to transgenic plants containing RAD under the control of its own promoter (RAD::RAD) (C.B., M.M.R.C. and E.C.). When plants with both transgenes were grown in DEX, they showed the same phenotype as 35S::CYC:GR plants. The double transgenics were also grown without DEX and transferred, at 21 days after germination, to media containing DEX. Fig. 9 shows that RADexpression was upregulated 6 hours after the induction of CYC, confirming that CYC can bind to the RAD promoter in vivo, and showing that this interaction can be reconstituted in Arabidopsis.

We have shown that CYC most likely acts as a transcription factor that binds to the RAD target gene. The interaction between CYC and RAD can be reconstituted in Arabidopsis, as induction of CYC leads to increased expression of the Antirrhinum RAD gene in Arabidopsis. Moreover, CYC expression affects growth and development of leaves and petals in Arabidopsis, indicating that it is able to influence developmental target genes in this species, as well as in Antirrhinum. In the case of leaves, CYC acts from an early stage in development to inhibit cell proliferation and at a later stage to reduce cell expansion, resulting in an overall reduction in leaf size. In petals, CYC acts at a late stage to promote cell expansion, leading to an increase in size of all petals.

In contrast to the situation in Antirrhinum, endogenous RAD-like genes are not activated by CYC in Arabidopsis. This suggests that although CYC can exert effects in Arabidopsis, the network of interactions involving CYC has changed since the Antirrhinumand Arabidopsis lineages diverged. The most recent common ancestor of Arabidopsis and Antirrhinum most likely had a TCP1/CYC gene that was asymmetrically expressed in axillary meristems(Cubas et al., 2001). How might the network of interactions have diverged in the lineages derived from this common ancestor?

In considering the evolutionary changes in a regulatory network, it is convenient to distinguish between two types of target process influenced by a transcription factor. One is the set of potential targets that could be influenced by expression of the transcription factor at any developmental stage and in any part of the plant. This class of targets can be revealed through constitutive expression of the transcription factor. Evolutionary changes in potential targets may occur, for example, through changes in the structure of the transcription factor (i.e. binding site) or changes in the promoters of target genes. The other type of target process is the set of actual targets that are influenced only where and when the transcription factor is normally active. Given a set of potential targets, changes in the expression pattern of the transcription factor can lead to modifications to the defined subset of actual targets.

The ability of CYC to influence both Antirrhinum and Arabidopsis development suggests that a range of potential developmental targets were present in their common ancestor. A subset of these would have been the ancestral actual targets in the dorsal regions of axillary meristems. Although these actual targets are unknown, they are unlikely to have involved morphological asymmetry of the flower as the ancestral condition is thought to have been radial symmetry.

A key step in the lineage leading to Antirrhinum may have been co-option of RAD into the network of potential targets(Fig. 10, step a). This could have arisen though mutations in the RAD promoter or intron, bringing RAD under the control of CYC. This would account for why over-expression of CYC in Arabidopsis does not lead to activation of RAD-like genes (as they lack appropriate recognition sites) even though it can activate the Antirrhinum RAD gene. It is also consistent with the more restricted role of RAD in floral symmetry compared with CYC and DICH.

In addition to co-option of RAD, other alterations in the regulatory interactions may also have occurred in the Antirrhinumlineage. For example, CYC has developmental effects on stamen arrest as well as a slight effect on petal lobe asymmetry independent of RAD. These effects may have arisen through divergence in some of the target processes influenced by CYC(Fig. 10, step b). It is also possible that the target sequence recognised by CYC diverged to some extent(the CYC consensus binding site is not typical of type II TCP proteins).

Another evolutionary change may have involved persistent expression of CYC in the dorsal domain of Antirrhinum floral meristems [in contrast to transient expression of TCP1 in Arabidopsis(Cubas et al., 2001)], leading to a change in expression or range of actual targets. Given the phenotypic effects of CYC on petal size in Arabidopsis, it is possible that such persistence could create an asymmetric Arabidopsis flower with larger dorsal petals. Persistence in the Antirrhinum lineage may have arisen through evolution of an autoregulatory loop(Fig. 10, step c), as CYC contains promoter sequences matching the consensus CYC-binding site (the TCP1 promoter does not contain such sites). However,persistent expression could also have arisen through interaction with other factors that are not directly dependent on CYC. In addition, we cannot rule out the possibility that the ancestral CYC/TCP1 gene was persistently expressed in dorsal floral meristems and that expression became transient in the lineage leading to Arabidopsis.

Compared with CYC, little is known about the role of TCP1and its downstream targets in Arabidopsis. Nevertheless, it seems that some of the potential target processes influenced by CYC may be conserved in Arabidopsis. Expression of CYC in leaf primordia influences both cell division and expansion. Moreover, this effect is cumulative with increased duration of CYC expression. This is comparable with the effects of CYC on organ development in Antirrhinum,consistent with some overlap between potential target processes influenced by CYC in the two species but also some divergence(Fig. 10, step d).

Expression of CYC also influences petal development in Arabidopsis. All petals are equally affected, reflecting ectopic expression of CYC all around the developing flower. The effect of CYC largely involves cell expansion late in development, in contrast to CYC in Antirrhinum, which also has early developmental effects. Moreover,CYC expression does not affect stamen development in Arabidopsis,unlike the situation in Antirrhinum. Thus, the potential developmental targets in floral organs have probably diverged, consistent with the inability of CYC to switch on RAD-like genes in Arabidopsis.

Taken together, our results indicate that interactions involving CYC have changed in many ways since the separation of the Antirrhinum and Arabidopsis lineages. These changes may themselves represent a subset of the changes that have occurred in the whole regulatory network in which CYC is embedded. Nevertheless, some elements of the network have been preserved, such as the initial asymmetric expression pattern and the modification of developmental target gene activity. Thus, floral asymmetry has most likely arisen through a process of tinkering(Jacob, 1977; Jacob, 2001) with the strengths and pattern of connections in a regulatory network, in which some common elements may still be discernable.

We thank Shunichi Kosugi for providing the oligonucleotides with the TCP class I and II binding sequences and for technical help. We are grateful to Robert Sablowski for providing the plasmid pRS20 containing the glucocorticoid receptor sequence, Mark Chadwick for making the pRSET-CYC construct and Reidunn Aalen for providing the ida seeds. This work was funded by a grant from the BBSRC.