Genetic control of flower shape in Antirrhinum majus

In Antirrhinum, development along a radial axis of floral meristems results in flowers with four types of organs, sepals, petals, stamens and carpels, arranged in concentric whorls. Within each whorl there are several organs which differ in shape according to their positions relative to a dorsoventral axis of the flower (e.g., 5 petals in whorl 2, see Fig. 1). The pattern of these differences is such that the flower has only one plane of symmetry, coinciding with the dorsoventral axis (Coen, 1991; Coen and Nugent, 1994; Luo et al., 1996).

Two systems of genes characterized in Antirrhinum determine either the radial or the dorsoventral patterns (for reviews, see Coen and Meyerowitz, 1991; Schwarz-Sommer et al., 1990). When inactivated, genes determining the radial pattern typically cause the differences between whorls to be reduced whilst dorsoventral differences may be retained. Mutations in genes controlling the dorsoventral pattern reduce differences between organs within whorls, without altering the radial pattern. In extreme cases, this may result in all organs within the same whorl being identical, giving radially symmetric flowers. Mechanisms controlling the radial and dorsoventral patterns may therefore be separate, at least to some extent, perhaps providing flexibility in the evolution of flower structure and shape. These mechanisms may also differ in that they generate differences in cell and regional identities which are sharply defined along the radial axis but more subtle along the dorsoventral axis. Here, we further address the problem of how genes control the dorsoventral pattern.

The best characterized dorsoventral mutants analysed in Antirrhinum give rise to ventralized phenotypes (Carpenter and Coen, 1990; Luo et al., 1996). For example, whereas wild-type flowers have five petals of three identities, one ventral, two laterals and two dorsals, in mutants with radially symmetric flowers all petals have ventral identity. Ventralization in these mutants indicates that the genes affected act in dorsal regions, consistent with the finding that one of these genes, cycloidea (cyc), is expressed only in dorsal parts of the flower (Luo et al., 1996). In addition, ventralization suggests that one role of dorsal-acting genes may be to restrict other determinants of the dorsoventral pattern to ventral domains. So far, however, no such determinants have been identified.

We describe a semidominant mutation, divaricata (div), which, in contrast to previously described mutations, affects ventral regions of the flower and confers a lateralized phenotype. We show by the analysis of mutant combinations that the dorsal genes cyc and dichotoma (dich) negatively regulate the div gene. In addition, by analysing the effects of chromosomal duplications we show that the div gene acts in a dosage-dependent manner, suggesting that the level of its product may be critical for determining subtle differences in regional identities in ventral regions of the corolla.

The cyc allele used in this work was contained in line JI25 which has radially symmetric (peloric) flowers (Carpenter and Coen, 1990). This line carries the cyc-25 allele and is also mutant at the unlinked dich locus, as shown by test-crosses to a single dich mutant (obtained from Drs U. Wobus and K. Hammer, Gatersleben, Germany). In a Dich⁺ background, the cyc-25 allele gives a semipeloric phenotype, similar to that of cyc-608 (Carpenter and Coen, 1990). The three possible double mutants with cyc, dich and div were obtained in 150 F₂ progeny derived from crossing JI25 to a single div mutant homozygote. The genotypes of the double mutants were determined by test-crosses to single mutants. However, the triple mutant could not be obtained in this F₂. The cyc;dich double mutants (peloric) occurred only in combination with Div⁺;Div⁺ or to Div⁺;div. This was as expected because dich is linked to div (about 25 cM in the sequence dich-pal-div; see Stubbe, 1996). The triple mutant div;cyc;dich was therefore obtained in F₃ families derived from cyc;dich homozygotes that were heterozygous for div. The genotype of the triple mutant was then confirmed by test-crosses to single and double mutants. Further genetic tests involving the cyc, dich, pal and div loci gave results consistent with dich being responsible for the peloric/semipeloric difference in a cyc mutant background. For example, in progenies from crosses between double heterozygotes for dich and div which were homozygous at cyc (semipeloric) and the div;dich;cyc homozygote (peloric), the peloric/semipeloric difference segregated linked to div, giving 20 recombinat gametes out of 94 tested, in close agrreement with the known distance between div and dich.

To select for changes at the div locus, progenies from crosses betweenDiv⁺ and div homozygotes were screened for deviations from the expected phenotype of heterozygotes (see Fig. 5). The Div⁺ progenitors in these crosses were derived from line JI2 (Martin et al., 1985) which carries an insertion of the transposon Tam3 at pallida (pal), a gene controlling flower colour which is linked to div. The pal allele carrying Tam3 (pal-2) gives ivory flowers with clones of red cells unlike Pal⁺ which gives red flowers. 200 Pal⁺ revertants, most of which were heterozygous Pal⁺/pal-2, were grown at 15°C, a temperature at which Tam3 transposition is stimulated (Carpenter et al., 1987), and crossed to a div line carrying pal-35, an allele giving pale colored flowers (Almeida et al., 1989).

In 4000 progeny from these crosses, we selected four plants with wild-type flower shape and colour, all of which subsequently gave mostly normal F₂ progeny segregating for div and pal. Two of these four F₂ families segregated for the three different pal alleles, indicating that their F₁ progenitors carried a duplication of the pal locus. This was confirmed by probing Southern blots of DNAs from the F₁ wild-types digested with EcoRI and NruI with a 1.6 kb EcoRI-BamHI fragment of the pal clone pJAM501 (Coen et al., 1986; Almeida et al., 1989). To show that the div locus was contained in the duplicated region, the same blots were probed with a 3.4 kb EcoRI-HindIII fragment of the rcp clone pJAM5A (Robbins et al., 1989) and a cDNA clone of the flo gene (pJAM101, Coen et al., 1990).

In progenies from the F₁ wild-types we identified 2 Div+/Div+/div, 4 Div⁺/div/div and 2 Div⁺/Div⁺/Div⁺ genotypes by Southern analysis as above. A Div⁺/Div⁺/Div⁺ genotype carrying three doses of rcp^D (see Fig. 6), the three different pal alleles and two doses of flo^D and one of flo^d (hence containing pal-35-Div⁺ and Div⁺-flo^d recombinant segments) was crossed to line JI25 which contains cyc-25 and is Div⁺ but carries rcp^d. Thus, in the progeny from this cross the dosages of pal, rcp and flo could be readily determined by Southern analysis. Probing this progeny with a cyc clone showed that all plants carried one copy of Cyc⁺ and one copy of cyc-25 irrespective of Div⁺ dosage.

Dorsoventral differences are particularly striking in the corolla of the wild-type Antirrhinum flower, which comprises five petals of three identities: two dorsals, two laterals and one ventral (Fig. 1). Each of the petals in lateral or dorsal positions is asymmetric whilst the ventral petal is bilaterally symmetrical. The five petals are united for part of their length forming a tube which ends in a sharp border with the petal lobes. This border can be represented as a wavy line with peaks or troughs at petal boundaries, reflecting a continuous variation in tube length along the dorsoventral axis (Fig. 2). In addition to tube length, a conspicuous marker for dorsoventral differences in the tube is provided by two stripes of yellow hairs on the internal surface of the tube, located to either side of the boundaries between ventral and lateral petals.

The wild-type dorsoventral pattern is altered by a semidominant mutation, div (see Stubbe, 1966), which causes ventral regions of the corolla to adopt a lateral identity whilst the dorsal petals remain unaltered (Fig. 2). In div mutant flowers, each half of the ventral petal becomes nearly a mirror-image of its adjacent part of the lateral petal. In mutant homozygotes, the domain affected includes the ventral petal and adjacent parts of the lateral petals, resulting in the loss or drastic reduction in the width of the stripes of hairs that mark the junctions between ventral and lateral petals. The altered domain is narrower in heterozygotes in which only the ventral petal is clearly affected. The boundaries of the region altered by div in the lateral petals cannot be precisely defined as the effect of div is strongest at the most ventral position decreasing in a graded manner towards lateral positions.

In contrast to the div gene which is needed for ventral identity, the cyc and dich genes together are needed for dorsal and lateral identities. In flowers from single cyc mutants the petals at ventral and lateral positions have ventral identity and those in dorsal positions have laterodorsal identity (Luo et al., 1996). The dich mutation only affects the shape of the dorsal petals which have reduced asymmetry relative to wild-type (see Stubbe, 1966; Luo et al., 1996). The effect of dich becomes clearer in cyc;dich double mutants which have radially symmetric flowers in which all petals have ventral identity (Luo et al., 1996; see also materials and methods and Fig. 3). Thus, in a cyc background, the dich mutation causes the normally asymmetric petals at dorsal positions to adopt the symmetric ventral identity.

To investigate how the div gene might interact with the cyc and dich genes, we constructed all possible combinations of the three mutants. cyc;dich;div triple mutants have radially symmetric flowers in which all petals have lateral identities, showing that the domain affected by the div gene extends all round the flower in a cyc;dich mutant background (Fig. 3). This indicates that the dorsal genes normally act negatively on the div gene.

The phenotypes of double mutants allow the effects of cyc and dich on the div gene to be distinguished. In div;cyc double mutants the ventral and lateral petals are symmetrical and have lateral identity, as in the triple mutant, but the dorsal petals have laterodorsal identity (Fig. 4). The dorsal petals can be divided in two domains. One, close to their junction with the lateral petals is altered in div;cyc double mutants relative to single cyc mutants, with an identity similar to that of of the triple mutant in the same region. Thus, the domain affected by the div gene extends towards lateral and dorsal positions in a cyc back- ground. In the other domain, close to the most dorsal position, the div;cyc double mutant is similar to the single cyc mutant, indicating that the effect of div does not reach the most dorsal position (Fig. 4). This reflects a negative action of Dich⁺ on the div gene, because in the triple mutant the domain affected by div extends all round the flower. However, Dich⁺ only acts on the div gene in a cyc mutant background, because in div;dich double mutants the dorsal petals have the same identity as in the single dich mutant and the ventral and lateral petals have the same identity as in the single Div mutant (not shown). Therefore, both the cyc and dich genes can act negatively on the div gene. However, cyc acts on div in lateral and dorsal petals irrespective of dich whereas dich only acts on div in dorsal petals in a cyc mutant background.

The domain affected by cyc includes the entire lateral petal in a Div⁺ background (petals in lateral positions are converted from a bcd to an ee identity; see Figs 2B and 3B). However, in a div mutant background, the domain affected by cyc includes only the region of the lateral petals close to their junctions with the dorsal petals. In this case, petals in lateral positions are converted from a bc to a cc identity (Figs 2B and 3B). Thus, the domain affected by cyc appears to be narrower in a div mutant background. One possible explanation is that the mRNA expression domain of the cyc gene could become more restricted in a div mutant background. To test this possibility, we determined the pattern of cyc mRNA expression in the div mutant by in situ hybridization. The pattern of cyc expression in div flowers could not be distinguished from that of wild-type. This was particularly clear in transverse sections of flower buds at stages when all organs have been initiated and each of the five petals can be easily defined. As in wild-type, at such stages cyc mRNA was found only in dorsal petals of the div mutant ending abruptly three to five cells away from the junction with the lateral petals (Fig. 5). Therefore, the reduced effect of cyc on the lateral petals in div mutants is not due to an altered pattern of cyc mRNA expression.

The semidominance of div suggests that Div⁺ may act in a dosage-dependent manner in ventral regions. Alternatively, div might be a neomorphic or gain-of-function allele. To distinguish between these possibilities, we determined the effects of different div allele dosages in plants carrying three copies of the div locus. If Div⁺ acts in a dosage-dependent manner and div is a loss-of-function allele, Div⁺/Div⁺/div genotypes should have the wild-type phenotype and Div⁺/div/div genotypes should have the same phenotype as normal heterozygotes. In addition, three doses of Div⁺ might have a ventralizing effect. However, if div is a neomorphic or gain-of-function allele, increases in the dosage of the div mutant allele should have a lateralizing effect on ventral regions. To obtain such genotypes, we carried out a mutagenesis experiment in which progenies of crosses between Div⁺ and div homozygotes were screened for deviations from the expected phenotype of heterozygotes (Fig. 6A).

Out of about 4000 F1 progeny from these crosses, most plants had the expected phenotype of div heterozygotes and none had the phenotype of div homozygotes. However, four exceptional plants had the wild-type flower shape, suggesting that they might carry an extra dose of Div⁺ giving a Div⁺/Div⁺/div genotype. To test this, we took advantage of differences between their progenitors in genetic and molecular markers flanking the Div⁺ and div alleles (Fig. 6A). Southern analysis using probes for three flanking markers (pal, rcp and flo, Fig. 6B) showed that the exceptional F1 progeny with a wild-type phenotype had inherited one copy of the chromo- some segment carrying the div mutant allele and two copies of the segment containing the Div⁺ allele.

This was particularly clear in the case of pal, a gene affecting flower colour, because the F1 wild-types carried three different pal alleles, each of which gives a distinctive restricion pattern (Fig. 6B). In addition, genetic evidence for the duplication at pal could be obtained in progenies derived from the F1 wild- types, because each of the three pal alleles confers an unique flower colour phenotype. Differences in the restriction patterns at each of the other two loci flanking div, provided a means of determining their dosages from changes in the relative intensities of the bands that they originated.

Determining the dosages of the markers above in progenies derived from the F1 wild-types allowed four plants with Div⁺/div/div genotypes to be identified (Fig. 6B), all of which gave the same phenotype as normal Div⁺/div heterozygotes, consistent with Div⁺ acting in a dosage-dependent manner and div being a loss-of-function allele. In addition, we obtained two plants with three copies of the wild-type Div⁺ allele conferring a novel phenotype in which the lateral petals appeared to be slightly ventralized. To obtain a clear phenotypic marker for ventralization conferred by three doses of Div⁺, we crossed a Div⁺/Div⁺/Div⁺ genotype to a cyc mutant. The heterozygous Cyc⁺/cyc progeny, were either wild-type or had flowers with a notch in the lateral petals close to their junctions with the dorsal petals (Fig. 7). This phenotype is typical of a weak ventralization as it is also observed in plants doubly heterozygous for cyc and rad, mutations that both have ventralizing effects (Carpenter and Coen, 1990). Analysis of the genotypes of six plants with this phenotype showed that they all carried three doses of Div⁺ whereas six of their sibs with the wild-type phenotype carried the normal two doses.

We have analysed mechanisms by which genes acting differentially along the dorsoventral axis of the Antirrhinum corolla control its shape. The wild-type corolla has five petals of three identities, two dorsals, two laterals and one ventral. This pattern is altered by a semidominant mutation, div, which causes a ventral region of the corolla to adopt a lateral identity. As a result, each half of the ventral petal becomes nearly a mirror-image of the adjacent part of the lateral petal, an effect reminiscent of that of segment-polarity mutations in Drosophila (Nüsslein-Volhard and Wieschaus, 1980).

In contrast to the div gene which is needed for ventral identity, the cyc and dich genes together are required for lateral and dorsal identities. Plants mutant for both cyc and dich have radially symmetric flowers in which all petals have ventral identity. The expression of ventral identity all round the flower in these mutants, suggests that one role of Cyc⁺ and Dich⁺ might be to restrict Div⁺ to ventral regions. To test this possibility we introduced the div mutation in a cyc;dich mutant background. Consistent with our prediction, cyc;dich;div triple mutants have radially symmetric flowers in which all petals have lateral identities, showing that the domain altered by div extends all round the flower in a cyc;dich background.

The two dorsal genes, cyc and dich, differ in their domains of action and in the contexts in which they negatively regulate div. Whereas the cyc gene can by itself affect the action of Div⁺ in dorsal and lateral petals, the dich gene affects a dorsalmost region of the corolla and only acts on Div⁺ in a cyc mutant background. This is consistent with the view that the cyc and dich genes can substitute for each other to some extent (Luo et al., 1996).

Interestingly, the region affected by cyc appears to become more restricted in a div mutant background. The petals in lateral positions adopt ventral identities in cyc mutants but retain lateral identities in cyc;Div double mutants. This is not due to an altered pattern of cyc mRNA expression, because in div mutant flowers, as in wild-type, cyc mRNA was found only in dorsal petals ending abruptly three to five cells away from the junction with the lateral petals. A more likely explanation is that Div⁺ mediates a non-cell-autonomous process in which Cyc⁺ activity spreads from dorsal to lateral regions. However, the action of Div is not necessary for all the effects of cyc on lateral petals because their symmetry is still altered by cyc in a div mutant background.

The semidominance of div reflects the action of Div⁺ in a dosage-dependent manner, as shown here by determining the effects of different div allele dosages in plants carrying three copies of the div locus. Consistent with Div⁺ acting in a dosage-dependent manner, we found that Div⁺/Div⁺/div genotypes have the wild-type phenotype and Div⁺/div/div genotypes have the same phenotype as normal heterozygotes. In addition, we found that plants carrying three doses of Div⁺ have flowers in which the lateral petals are slightly ventralized. This may reflect a shift in the balance between the activities of the div gene and of genes acting dorsally because in a back-ground heterozygous for cyc, Div⁺/Div⁺/Div⁺ genotypes had the same weakly ventralized phenotype as that of plants doubly heterozygous for cyc and rad, mutations that both have ventralizing effects (Carpenter and Coen, 1990).

Action of Div⁺ in a dosage-dependent manner indicates that the level of its product may be critical for determining petal identities in ventral and lateral regions. One possibility is that the level of Div⁺ product peaks at the most ventral position and declines gradually along the dorsoventral axis to a low point as a result of the negative action of Cyc⁺ and Dich⁺ in dorsal regions. Regional identities at different positions would in this case reflect directly their levels of Div⁺ product. This would be in line with the dosage-dependent action of genes expressed as gradients (Driever and Nüsslein-Volhard, 1988). Graded effects could alternatively reflect Div⁺ acting in a relay system of signals which would interact with a Div⁺ product in a concentration-dependent manner to determine petal identities. Semidominance would then reflect a fine balance between the concentrations of the Div⁺ product and of the other signals. In this case, Cyc⁺ might act through these other signals to determine asymmetry in lateral petals. Isolation of the div gene might allow these possibilities to be distinguished.

We are indebted to Enrico Coen for discussions throughout this work and constructive criticism of the manuscript. We thank Rosemary Carpenter for providing Antirrhinum lines and M. J. Leandro and C. Couceiro for discussions. The financial support of the EC and of the Portuguese Government (JNICT) is acknowledged.