CRABS CLAW and SPATULA, two Arabidopsis genes that control carpel development in parallel with AGAMOUS

Carpels are the female reproductive unit of flowers. They enclose the ovules that bear the female gametophyte, and they provide a screen that is penetrated only by appropriate male gametes. At maturity, they generate the fruit that harbours the developing seeds. The identity of carpels and other floral organs is specified by three classes of homeotic genes (Coen and Meyerowitz, 1991; Weigel and Meyerowitz, 1994). These encode three identity functions, named A, B and C, that are active within the developing flower in concentric overlapping fields. When A function acts alone, sepal identity results, A and B together specify petals, B and C in combination result in stamen development, and C function alone specifies carpels. A and C functions have additional roles in mutually preventing each others action, with A blocking C function in the perianth (sepals and petals), and C preventing A function in the reproductive organs (stamens and carpels).

Only one C function gene is known. In Arabidopsis this is AGAMOUS (AG). In ag mutant flowers (Bowman et al., 1989, 1991), A function is no longer blocked from the central whorls. Petals now develop in place of stamens, and, instead of carpels, four sepals arise that constitute the outer whorl of another inner flower. The role of AG in specifying carpel identity can also be seen indirectly in mutants of A function genes including APETALA2 (AP2) (Bowman et al., 1989, 1991; Kunst et al., 1989). In these, AG is no longer under negative control in the perianth and it is ectopically expressed there (Drews et al., 1991). Rather than developing into sepals, the first whorl organs now display the properties of normal fourth whorl carpels. Similar sepal-to-carpel transformations are observed when AG is constitutively expressed (Mizukami and Ma, 1992). (Continued proliferation in the center of the floral meristem in ag mutants is the result of disruption of a different function of AG, imposition of determinacy; Mizukami and Ma, 1995; Sieburth et al., 1995.)

Even so, AG function is not absolutely required to generate all carpel properties. When AG activity is in turn removed from ap2 A function mutants, floral organs nevertheless retain many carpel properties including style, stigma and ovule production (Bowman et al., 1991). Other carpel genes must be involved.

To identify further genes involved in carpel development, we undertook a screening program for carpel-specific mutants (Alvarez and Smyth, 1998). We argued that such genes are likely to include primary targets of AG regulation, as well as genes acting independently of AG. Mutants of two genes, CRABS CLAW (CRC) and SPATULA (SPT), were isolated which have different developmental defects largely exclusive to the gynoecium. In this study, the relationships between the two genes, CRC and SPT, and with AG and other organ identity genes are defined. Both CRC and SPT function were found to be inextricably associated with the carpel development program. Analysis of the interactions of CRC and SPT with AG has further revealed that, rather than being direct downstream targets of AG, CRC and SPT play relatively independent roles in controlling different aspects of carpel development in parallel with AG.

The Landsberg erecta ecotype was used throughout. Plants were grown under daylight supplemented with continuous cool white fluorescent light at 20-25°C. Seeds were mutagenised with 25 mM ethyl methane sulphonate (EMS) for 12 hours, or 25 krad of γ irradiation from a ⁶⁰Co source, and mutants identified by screening second generation plants. The loci of mutants were mapped by crossing them to a tester line carrying ten mutant markers, and linkage was detected among F₂segregants.

Multiple mutant plants were generated by intercrossing homozygous mutants (or heterozygotes if plants were both male and female sterile), and the desired mutant combinations recognised among the phenotypic categories seen in the F₂segregants. Their genotypes were confirmed by checking that they arose in the expected Mendelian proportions, and, if possible, by progeny testing. The agamous mutant used was ag-1, a null allele (Yanofsky et al., 1990). The apetala2 and pistillata mutant alleles used, ap2-2 and pi-1, are also null alleles (Jofuku et al., 1994; Goto and Meyerowitz, 1994). For phenotypic analysis, the first 15–20 flowers on the primary inflorescence shoot were examined unless otherwise specified.

For light microscopy and scanning electron microscopy (SEM), samples were immersed in 2% glutaraldehyde in 0.025 M sodium phosphate buffer (pH 6.8) and vacuum infiltrated for up to one hour. For sections, specimens were then washed, dehydrated in an ethanol series, and infiltrated and embedded in LR White resin. Sections of 2 μm were cut, dried onto slides and stained with 0.3% toluidine blue (w/v) in 1% sodium tetraborate (w/v, pH 9) for 30 seconds. For SEM, glutaraldehyde-fixed tissues were further fixed in 1% OsO₄ before dehydration through a graded ethanol series and critical point drying using liquid CO₂. Specimens were coated with platinum in an Eiko 1B.5 sputter coater, dissected if necessary, and viewed using a Hitachi s570 scanning electron microscope. To view the vasculature, cleared whole gynoecia were prepared. They were fixed in 3:1 ethanol:glacial acetic acid (v/v) for 24 hours, rinsed in 0.1 M potassium phosphate buffer (pH 7.5) for 2 hours, cleared in 1 M NaOH at room temperature for 12 hours, and washed in the same buffer. They were then mounted whole in buffer on a slide, and observed using dark-field optics.

The mature wild-type gynoecium of Arabidopsis has been described in detail elsewhere (Hill and Lord, 1989; Smyth et al., 1990; Sessions and Zambryski, 1995). In brief, it consists of a stigma, style and ovary joined to the floral receptacle by a very short stem (Fig. 1A,B). The ovary is divided medially into two locules. Each of the lateral walls is occupied largely by valve tissue. These are joined to two vertical strips of tissue, the repla, that run medially. In the mature fruit, the valve separates from the replum exposing the seeds. Internal to each replum, a septum grows across the ovary and fuses post- genitally with septal tissue growing from the other side. A row of ovules arises from placentae on each side of the elongating septum. The upper region of the gynoecium narrows into a short style that is capped by a dry stigma. Inside the style and septum, a core of transmitting tract cells develops. These secrete an extra-cellular matrix that supports pollen tube growth. Studies of mutants have revealed that the gynoecium consists of two congenitally fused carpels joined along their edges where replum, septum and placental tissues arise (Okada et al., 1989).

Two alleles of the CRABS CLAW (CRC) gene were isolated, a stronger EMS allele crc-1 and a weaker γ-induced allele crc-2. The CRC locus was mapped to chromosome 1, just below APETALA1. The stronger mutation crc-1 results in a gynoecium that is shorter overall than wild type but wider both laterally and medially (Fig. 1C,D). The two carpels are characteristically unfused in their upper third. Above the point of fusion, each carpel extends outward in the lateral plane, before bending inwards such that each half of the greatly abbreviated style is now nearly horizontal, and the two clusters of stigmatic papillae intermingle. All cell types of the wild type are present in appropriate locations. The only ectopic structure seen is an occasional ovule that arises from the replum outside the ovary. Overall the total number of ovules per gynoecium is reduced relative to wild type. One further characteristic is the occasional presence of more than two carpels, especially in the first formed flowers (Table 1). These are inserted medially. The only other mutant defect observed was the abolition of nectary development (see Bowman and Smyth, 1999). The weaker allele, crc-2, shares all these defects but they are less severe, especially the reduction in gynoecial length.

Two EMS alleles of the SPATULA (SPT) gene were obtained, a weaker mutant allele spt-1 and a stronger allele spt-2. The SPT locus is on chromosome 4, less than 1 map unit below APETALA2. spatula mutants have a spectrum of changes different from that seen in crc mutants (Fig. 1E,F). The spt-2 gynoecium at anthesis is narrower in the stylar region, and there is reduced development of stigmatic papillae at the tip. It is also somewhat flatter, being expanded laterally. In some cases the two carpels are unfused on one or both sides in the upper-most regions. The style is hollow, and the septum is under-developed and unfused, especially toward the top. There is no transmitting tract, as cells that secrete extra-cellular matrix are absent from the style and septum. The length of the gynoecium is not significantly different from wild type, although it carries about 20% fewer ovules. However only a quarter of these go on to develop as seeds. This appears to be the result of inefficient pollen tube growth associated with loss of the transmitting tract. Again, the weaker allele, spt-1, shares all the disruptions but mostly to a lesser degree. Defects in other floral or vegetative organs were not detected.

To test if CRC and SPT share redundant functions, the double mutant was created. Some redundancy seems likely, as the mature crc-1 spt-2 gynoecial organs are markedly different from either single mutant (Fig. 1G,H). They are now boat- shaped, elongated at the tips and typically unfused except at the base. While the tissue morphology of the carpel wall closely resembles that of wild type, there is a marked reduction in other carpel tissues including stigmatic papillae, style, septum, transmitting tissue and ovules. In addition, there tends to be a greater loss of flower meristem determinacy than in crc- 1 single mutants, with stamens occasionally being observed interior to the fourth whorl carpels (Table 1). Plants heterozygous for spt-2 and homozygous for crc-1 show slightly increased loss of carpel properties compared with crc-1 (results not shown), suggesting that the level of SPT function is important in crc mutant background. This effect is not allele specific.

Mutations of either B class gene PISTILLATA (PI) or APETALA3 (AP3) cause carpelloid organs to arise in what would normally be third whorl stamen positions (Bowman et al., 1989, 1991). To test whether CRC and SPT are ectopically active in these third whorl organs, the phenotype of flowers doubly mutant for each of crc-1 and spt-2 with the null allele pi-1 (Bowman et al., 1989; Goto et al., 1994; Hill and Lord, 1989) were investigated.

In pi-1 mutant flowers the third whorl is highly variable in structure, consisting of carpels and carpelloid filaments that demonstrate various degrees of connate fusion with each other and with fourth whorl carpels (Bowman et al., 1989; Hill and Lord, 1989; Fig. 2A). In crc-1 pi-1 double mutants, the third whorl organs are less fused allowing a significant reduction in carpelloid properties to be deduced (Fig. 2B). Compared with unfused carpels in other genotypes (ap2-2 for example, see below), they have little style and stigmatic tissue, and relatively few ovules arise from their edges. In flowers of plants homozygous for spt-2 and pi-1, an even greater diminution in carpelloid development is observed (Fig. 2C). These reductions suggest that both CRC and SPT are now active in the third whorl of pi mutants and promote their feminisation. That is, B function normally acts to prevent the inappropriate action of CRC and SPT in the third whorl of the Arabidopsis flower.

Mutations in the A class gene AP2 cause carpelloid organs to develop in the outermost whorl (Bowman et al., 1989; 1991; Kunst et al., 1989). To determine whether CRC and SPT contribute to this ectopic carpellody, double mutants were made between each of crc-1 and spt-2 and the null mutant ap2- 2 (Bowman et al., 1991; Jofuku et al., 1994). Before this, however, a detailed analysis of the carpelloid elements within the first whorl medial organs of ap2-2 single mutants was carried out for later comparisons (Fig. 3A-D).

Firstly, the shape of the unfused carpels is linear with parallel edges. The edges, especially around the apex, are involuted, and the whole organ bends inwards so that it appears hooded (Fig. 3A). Secondly, the wall of the valve region is closely similar in structure to the ovary wall of the central gynoecium (Fig. 3A,B). Even so, tissue with leaf-like properties (including occasional stellate trichomes) and sepal characteristics can occur, especially in the first produced flowers. Thirdly, there is a remarkable change in the distribution of stylar and stigmatic cells. Instead of being confined to the tip of the gynoecial tube, these now arise along the edges of the unfused organ as well (Fig. 3A,B). They are intercalated between replum-like cells and a flange of septum cells (Fig. 3C), and appear as a lobe of tissue projecting from the ovary wall (Fig. 3B). The stigmatic cells typically decrease in extent toward the base of the carpel (Fig. 3D). These changes occur in all unfused carpels, including those often present in the centre of the ap2-2 flower. It seems that, when a carpel is unfused, the valve and replum differentiate relatively normally, as do the septum and the ovules internally, but that the lateral edges now enter the style and stigma differentiation program that is normally confined to the apex.

The effect of disruption of CRC function in these ectopic carpels was then examined in crc-1 ap2-2 double mutant flowers (Fig. 3E,F). The medial outer whorl organs develop ostensibly as unfused carpels but tend to be slightly broader and more leaf-like than in ap2-2 single mutants (Fig. 3A). There is a greater abundance of phylloid epidermal characteristics on their abaxial surface, such as irregularly shaped cells, intervening stomata and stellate trichomes. Stylar epidermal cells with their prominent cuticular ridges are less abundant, and stigmatic papillae are also reduced and restricted to the upper regions (Fig. 3E,F). The number of marginal ovules is greatly reduced, but a significant flange of septal tissue typically remains.

When first whorl organs of the spt-2 ap2-2 double mutant are examined, there is a more striking loss of stigmatic papillae, and septal tissue is also mostly lost (Fig. 3G,H). Involution of the margins is also reduced compared with ap2-2 single mutants (Fig. 3A), and there is a blunt-ended prominence at the apex associated with stylar cells. Similar disruptions are seen in the fourth whorl. (In passing, in both crc-1 ap2-2 and spt-2 ap2-2 there is a reduced degree of connate and adnate fusion of first whorl organs, and an increased number of free standing stamens interior to the first whorl.)

The phenotypes of the crc-1 ap2-2 and spt-2 ap2-2 double mutants demonstrate that CRC and SPT are both active in the outer whorl of ap2-2 mutant flowers, and that they promote the same carpel attributes as they do in the fourth whorl of wild- type flowers. Thus A function normally plays a role in suppressing their carpel promoting activity in the outer whorl. Together with the pi-1 double mutant analysis, the results reinforce the proposal that wherever normal carpels arise in the Arabidopsis flower, CRC and SPT function necessarily occur.

The flowers of crc-1 ag-1 and spt-1 ag-1 double mutants are indistinguishable from the ag-1 single mutant in each case (except that the crc-1 ag-1 double mutant lacks nectaries). This is consistent with the ag-1-induced disruption to the fourth whorl (where another flower arises) occurring before any disruptions resulting from crc-1 and spt-2 mutation.

Indirect genetic evidence for some interplay between CRC and AG functions was seen, however, in that in a crc-1 homozygous mutant background, ag-1 heterozygotes now have a distinguishable phenotype. In particular, there is a marked decrease in flower meristem determinacy. This is reflected by an increased number of carpels in the fourth whorl (more than half of flowers now have three or more carpels) (Fig. 4A), and proliferation of further organs occurs within the carpels of nearly half of all flowers (Fig. 4B; Table 1). These organs occur as alternating groups of stamens and carpels separated by an elongating pedicel. The mature gynoecium may also be elevated on a small gynophore (Fig. 4A), and there are reductions in style and stigma development and in ovule numbers. This effect is not allele specific. Thus AG is semi- dominant in a crc mutant background, and it may be that CRC function quantitatively promotes the AG determinacy function to some degree.

The carpel identity role of AG has been inferred in part from the observation that ectopic first whorl carpels arise when AG function is released from AP2 inhibition, as in ap2-2 mutants. Significantly, however, carpelloid features have been reported to develop in these organs even when AG function is also removed in ap2-2 ag-1 double mutants (Bowman et al., 1991). We have therefore further characterised these AG-independent carpelloid properties in these flowers before investigating which, if any, are controlled by CRC and SPT.

In ap2-2 ag-1 double mutants the linear, hooded shape of the outer medial organs and internal carpelloid organs is retained (Fig. 5A). Also, the lobes of style cells and stigmatic papillae, the septal flange, and the ovules that line the sides of the ap2-2 single mutant organs (Fig. 3A-D) are also frequently present, especially in earlier flowers (Fig. 5A,B). The first major change seen associated with loss of AG function involves the specialised cells of the ovary wall. The outer epidermis now shows variable vegetative properties, including leaf-like cells in combination with the very long cells characteristic of sepals (Fig. 5B). Internally, too, there is a striking change. The ovary wall of wild-type carpels (Fig. 5C) (and of crc-1 spt-2 double mutants; Fig. 5D) is made up of six specialised cell layers. The first whorl carpels of ap2-2 single mutants have a very similar structure (Fig. 5E). In ap2-2 ag-1 first whorl organs, however, the adaxial layer normally made up of many thin cells that will ultimately be lignified in the mature fruit (Spence et al., 1996), is frequently replaced by a row of palisade mesophyll cells (Fig. 5F), just as is seen in the equivalent adaxial region of leaves.

A second major effect of removing AG function is a reduction in development of a stylar prominence. When the carpelloid organs are unfused, the tip now has a tapered, leaf- like point that is typically devoid of stigmatic papillae (Fig. 5G, compare Fig. 3A). The same conclusion can be reached from fused organs (Fig. 5H), where examination of the vasculature confirms severe stylar reduction. In wild-type gynoecia (Fig. 5I), the medial vascular bundles extend into the style and branch extensively producing a number of blunt- ended xylem elements (Fig. 5J; Sessions and Zambryski, 1995). In ap2-2 ag-1 flowers, apical extension of the medial vasculature occurs only sporadically (Fig. 5K). Instead, it extends to the shoulders of the organ before converging with the central vascular bundle.

Thus AG function, at least when no longer repressed by A function, does not specify the full carpel development program, but determines the identity of the ovary wall and promotes growth of the style.

To test the role of CRC and SPT in specifying the remaining AG-independent carpelloid properties, a line generating ap2-2 pi-1 ag-1 triple mutant plants was grown and the crc-1 and spt- 2 mutants incorporated into it separately and together. The ABC triple mutant was used instead of the AC double to avoid the lateral patches of stamen tissue that are sometimes seen in first whorl medial organs of ap2-2 ag-1 double mutants (Bowman et al., 1991).

All organs of the ap2-2 pi-1 ag-1 triple mutant genotype are carpelloid except for the lateral first whorl organs (Fig. 6A). Adjacent organs frequently fuse, even though they may not strictly occupy the same whorl but rather lie in alternate positions. Where they fuse, the development of stigmatic papillae no longer occurs from the lateral margins, but is typically restricted to the unfused commissural position at the apex (Fig. 6A). A replum also develops between the fused organs, and a septum and ovules usually arise internally. The septum houses transmitting tissue that is capable of supporting pollen tube growth (not shown).

When CRC function is compromised in addition to AG, leaving SPT function solely present (in the crc-1 ap2-2 pi-1 ag-1 quadruple mutant), the organs are still hood-shaped but are now somewhat shorter and broader and tend to be unfused (Fig. 6B). They continue to show some irregular development of stigma, style, septum and ovule development along their edges, although the extent is significantly reduced. On the other hand, when SPT function is mutant (in the spt-2 ap2-2 pi-1 ag- 1 quadruple) and only CRC function presumably remains, all organs are longer with parallel edges (Fig. 6C), and they are always unfused. There is a marked loss of marginal carpelloid tissues, with no sign of stigma, style, septum or transmitting tract development. The only structures seen are short stalked outgrowths on the margins that occasionally show some ovule- like differentiation of integuments, or a small cap of stigma- like cells.

Finally, when CRC and SPT functions are both compromised (in crc-1 spt-2 ap2-2 pi-1 ag-1 pentuple mutants), the organs are now more ovate, and marginal outgrowths are further reduced (Fig. 6D). Trichomes may occur on the abaxial surface, especially in the first-formed flowers. The organs are not fully leaf-like, however, in that the epidermal morphology on both outer (abaxial) and inner (adaxial) surfaces retains some sepal-like properties, especially in later formed flowers.

These observations are summarised in Table 2. The conclusions are that, in the absence of AG function, CRC strongly promotes the longitudinal growth of floral organs, whereas SPT promotes both involution of the edges and development of the carpel-specific tissues that arise from them.

In this study, we make deductions about the role of wild-type genes from their mutant phenotypes. It is necessary to spell out several caveats involved in this approach. First, the changes seen in individual mutants may reflect the primary disruption, and/or they may be secondary downstream changes. For instance, CRC appears to enhance AG and SPT activity as evidenced by the partial dominance of ag and spt in crc homozygotes. Thus, in crc single and multiple mutant combinations, some elements of the phenotype may arise due to reductions in the activity of other non-mutant genes.

Second, epistasis can have a range of interpretations. If one mutant is epistatic to another, the first gene may normally activate the action of the second, or the second gene may negatively regulate the first. Alternatively, the epistatic gene may normally provide the conditions necessary for the second to function. We argue that the latter is the trivial explanation for ag being epistatic to crc and spt, in that the ag single mutant results in the inappropriate activity of A class genes such as AP2 that promote perianth organs and suppresses CRC and SPT function in carpel morphogenesis.

A third situation where qualification is necessary occurs when the phenotype of a double mutant includes properties not seen in either single mutant. This could mean that the genes have partially overlapping functions, and it is only when both are mutant that the function is revealed. The potential complication here is that the further properties lost in the double mutant may be far downstream or an indirect consequence of the initial disruption point of one or both genes. For instance, to uncover the effects of CRC and SPT in the ag mutant background required us to introduce the ap2-2 mutant. Our analysis then requires us to assume that the loss of AP2 does not significantly modify carpel morphogenesis. We further assume that in the ap2-2 ag-1 background any other genes that may become active in the absence of AG activity do not antagonise carpel morphogenesis. In this regard we concede that additional A class genes may be involved, as evidenced by the sepalloid characteristics remaining when CRC, SPT and A, B and C functions are all disrupted.

Bearing in mind these caveats, we make the following deductions.

A number of genetic tests have demonstrated that CRC and SPT are active wherever carpelloid organs arise in the flower, even in the absence of the carpel gene AG. They appear to perform the same function in ectopic carpels as they do in the wild-type gynoecium. There is no genetic evidence that they are active in other floral organs.

The C function gene AG encodes a transcription factor of the MADS family (Yanofsky et al., 1990), but it seems unlikely that AG directly activates CRC and SPT expression. The presence of many carpel characteristics in the organs of ap2-2 pi-1 ag-1 flowers that are ultimately lost when either CRC or SPT are also mutant (see Table 2) demonstrates that neither CRC nor SPT lie directly downstream of AG function.

The fact that CRC and SPT activity promote carpel development in the outer whorl organs of ap2-2 flowers suggests that their action in this region of wild-type flowers is prevented by A class genes. This is similar to the cadastral function that AP2, and the product of another A function gene LEUNIG (Liu and Meyerowitz, 1995), have on the C function gene AG (Bowman et al., 1989; 1991; Drews et al., 1991). However, unlike AG, which reciprocally antagonises A function, the absence of carpel features attributable to CRC and SPT activity in ag-1 mutant flowers suggests that CRC and SPT do not have an equivalent cadastral function with respect to A function. Together, these results suggest that the activities of CRC and SPT in promoting carpel development will be restricted to wherever AG alone is active. In this regard, although CRC and SPT are not formally downstream of AG activity, AG appears essential for their action, acting as a buffer against potentially negative AP2 action. For this reason CRC and SPT activities in promoting carpel morphogenesis are inseparable from AG activity when AP2 activity is not compromised.

Double mutants between crc-1 and spt-2 and the B function gene pi reveal that CRC and SPT are active in the third whorl carpels of pi mutants. It is only here where B class activity is absent (and the potentially inhibiting A function is still blocked by C function) that the carpelloid promoting activities of CRC and SPT become active. This implies that B class activity is antagonistic to the activity of CRC and SPT in promoting carpellody.

When AP2 function is compromised, most carpel features may arise independent of AG function. The only carpel tissue types not seen in first whorl medial organs of ap2-2 ag-1 double mutants are those of the ovary wall or valve (Table 2). These are now almost fully vegetative. Their outer carpel-like epidermis is absent, and cells resembling the epidermis of leaves and sepals are present. Trichomes may also be abundant, especially in the first-formed flowers (see also Bowman et al., 1991). The thin longitudinal cells underlying the inner epidermis that will become lignified in the fruit (Spence et al., 1996) are also absent, and the mesophyll cells that replace them often include an incipient palisade layer as seen in leaves.

Even though the valve is not carpel-like, the organs retain their carpel-like shape. They are hooded, and their margins are parallel (Table 2). The only difference in shape is the absence of a flattened extension at the organ’s incurved tip. This appears to represent reduced growth of the ‘stylar prominence’ (although style cells themselves can still differentiate on the lateral margins). Examination of the vasculature of carpelloid organs in ap2-2 ag-1 flowers has revealed that they lack the specialised xylem cells normally present in styles (Sessions and Zambryski, 1995). It may be that AG promotes development of the stylar prominence by activating growth- promoting genes such as KNAT1. This shoot meristem gene is expressed in the style, and when expressed ectopically in leaves it induces lobing in which the vascular strands are ramified in a style-like fashion (Chuck et al., 1996).

The main role of CRC seems to be to control aspects of carpel growth. Differentiation is not affected in that all carpel tissue types arise in crc mutants. The mature gynoecium is shorter and wider than in wild-type, the increased width being evident from its inception (J. A., unpublished). This suggests that CRC suppresses excess lateral expansion but promotes longitudinal growth of the developing carpel. This proposal is reinforced by observations in the ap2-2 mutant background when crc is also mutant (Table 2). The unfused first whorl organs consistently become ovate and more leaf-like in shape. By contrast, when CRC function is not disrupted, the organs are linear with parallel edges. The mechanism by which CRC may produce such proposed growth effects is unclear (see Bowman and Smyth, 1999, for discussion). A detailed analysis of cell division patterns within wild-type and mutant carpels would help resolve this question. The role of CRC in promoting ovule initiation may also be direct or indirect, although the former is supported by the observation that ectopic external ovules occasionally arise from the replum in crc mutant gynoecia.

A distinctive feature of the crc-1 mutant phenotype is that the carpels are unfused in the upper third of the gynoecium. From growth analysis, we postulate that this may be a secondary consequence of the increase in width of the mutant gynoecium (J. A., unpublished). Results from multiple mutant combinations support this, in that whenever crc is present in mutant form, unfused carpelloid organs are ovate (Table 2). Thus the upper region of crc single mutant carpels may be narrower and less likely to show congenital fusion than lower regions.

Unlike CRC, disruption to SPT function results in the complete loss of a specific tissue type. The transmitting tract within the style and septum is absent in any plant homozygous for spt. There is also a marked reduction in the extent of stigma production and some reduction of the style. Internally, too, core tissues of the style and septum are reduced, especially in the apical region. The number of ovules is also reduced. These reductions are more apparent in the unfused first whorl carpels of ap2-2 pi-1 ag-1 mutants. When SPT function remains, all these tissue types may be present, although variable in extent (Table 2), but if the spt-2 mutant is incorporated as well, they almost all disappear. The sole carpelloid structure remaining is the short filiform outgrowths at the margins that occasionally show vestigial ovule-like differentiation. This indicates that SPT normally promotes the development of all these marginal carpel tissues.

Organ fusion is mostly eliminated when the spt-2 mutant is introduced into the ap2-2 ag-1 pi-1 mutant background. It may be that SPT normally promotes congenital fusion of the carpels from an early stage, and that precursors of the septum at the margin play a direct role in promoting fusion through positive adhesion. Alternatively, SPT may promote growth at the carpel margins, and changes to fusion and marginal tissue development may be the joint consequences of disruption to this lateral growth. If the latter proposal is correct, the loss of marginal tissue types may be a secondary, downstream consequence of reductions in the substrate required for their initiation.

Notwithstanding the inferences made to date, multiple mutant phenotypes suggest that the functions of the three carpel genes overlap to some degree.

While the phenotype of spt-2 ag-1 ap2-2 pi-1 mutants highlights the role of SPT in marginal carpel tissue development hinted at by the spt single mutant phenotype, it also suggests a redundant role for AG in this process. The reduction in septum, stigmatic papillae and ovule development is clearly greater in spt-2 ag-1 ap2-2 pi-1 organs than is seen in spt-2 single mutants.

For CRC and SPT, the non-additive phenotype of the crc-1 spt-2 double mutant gynoecium also implies some overlap in the developmental processes lost in each. However, we know that AG shows a dosage response when CRC function is compromised, so the severity of the crc spt defects may be a consequence of AG activity also being marginally reduced. It is insufficient, however, to impair the identity of the ovary wall. A feature of the Arabidopsis carpel is an infolding at the margins which gives the carpel its hooded shape. AG and SPT appear to play the primary role in promoting this process, although their functions are redundant. In crc-1 spt-2 double mutants, where AG is alone active, the carpelloid organs continue to demonstrate infolding at the lateral margins. Also when AG activity is alone reduced in the carpel, leaving SPT and CRC activity (in ap2-2 pi-1 ag-1 mutants), the carpelloid organs continue to exhibit strongly involuted carpel margins (Table 2). It is only where both AG and SPT activity are absent (in spt-2 ap2-2 pi-1 ag-1 mutant flowers) that organs exhibit little evidence of marginal involution. A role for AG in promoting marginal involution is consistent with the leaf curling due to ectopic AG activity in the curly leaf (clf) mutation (Goodrich et al., 1997) as well as through over expression of AG in 35S-AG plants (Mizukami and Ma, 1992).

One function of AG, separable from its organ identity role, is to suppress growth at the centre of the flower meristem (Mizukami and Ma, 1995; Sieburth et al., 1995). The minor and variable reduction in the process of flower meristem determinacy seen in crc-1 single mutants, and the more extensive reduction in crc-1 spt-2 double mutants (Table 1), most likely occur as a consequence of CRC and SPT normally acting to enhance AG activity in this function. Alternatively, these genes may act separately from AG, but their role in suppressing growth at the centre of the flower can only be detected when AG activity is present. Both possibilities are consistent with AG regulation of flower meristem determinacy being dose dependent in crc mutant background (Table 1). In passing, it is worth noting that the increase in organ numbers seen when crc and spt are each combined with pi-1 or ap2-2 may also be a consequence of reduced flower meristem determinacy. This requires more detailed investigation.

If the functions of CRC, SPT and AG are simultaneously disrupted, the carpels are converted almost completely to leaf- like organs. This implies that these genes are necessary to specify the normal tissue types and growth characteristics of carpels. But how is positional information established that specifies where and when the genes will act?

Several Arabidopsis genes are already known where the mutant gynoecium reveals major changes in pattern elements, including ETTIN, MONOPTEROS, PINOID and PIN- FORMED (Nemhauser et al., 1998), and these genes may be involved in the spatial regulation of one or more of AG, CRC and SPT. This seems particularly likely for ETTIN with regard to SPT activity, in that the stigmatic lobe and septal tissues that SPT promotes arise ectopically in ettin mutants (Sessions and Zambryski, 1995; Sessions et al., 1997). It will be interesting to examine genetic interactions between mutants of ettin and spt in detail (Alvarez and Smyth, 1998).

The ap2 null mutant phenotype (Bowman et al, 1991; Kunst et al., 1989; present study) also provides evidence for spatial regulation of carpel tissue development by uncovering a striking ‘edge effect’ not previously highlighted. When carpels are unfused, as in first whorl medial positions of ap2-2 mutants, and less frequently toward the centre of the flower, stylar and stigmatic papillae arise inappropriately as a flange of tissue inserted between the normally positioned replum and septum. On the other hand, whenever ap2-2 carpels are fused, even in the first whorl, ectopic cell types do not arise from the fusion zone. It seems that an unfused carpel edge is somehow sensed during development, and that style and stigmatic cells arise from it. The edge is normally restricted to the apex of the gynoecial cylinder, but when unfused carpels arise, the edge now extends continuously around the periphery. These generalisations extend to ap2-2 ag-1 and ap2-2 pi-1 ag-1 genotypes as well. SPT function may somehow be involved, as development of stigmatic papillae is lost from the margins of unfused spt-2 ap2-2 carpels. Future cloning of SPT and analysis of its expression will shed light on this proposal.

In this study we have shown that simultaneous disruption of AG, CRC and SPT functions (in the absence of AP2 function) results in the loss of almost all carpel features within floral organs. In their place, leaf-like organs arise. From our analysis, we propose that AG does not control the full carpel program but defines only the identity of the carpel wall and growth of the stylar prominence. On the other hand, SPT promotes growth of carpel margins and the differentiation of specialised tissues from them, and CRC promotes carpel elongation. The three genes carry out these functions relatively independently, and all are required simultaneously for normal carpel ontogeny. An understanding of how they bring this about requires further study of their products and mechanisms of action. To this end, the cloning of SPT is in progress, and the nature of the CRC gene and its product are described in the following paper (Bowman and Smyth, 1999). CRC encodes a regulatory protein, and its pattern of expression in normal and mutant plants throws light on some of the interpretations made here.

We are indebted to John Bowman for many discussions and speculations. Angela Atkinson, Gerd Bossinger, Stan Paul Case, Bob Elliott, Megan Griffith, Catherine Guli, Marcus Heisler and Cameron Johnson also gave us much constructive criticism and beneficial feedback over the years. John Bowman, Marcus Heisler and Yuval Eshed commented usefully on the manuscript. We thank Nancy Garavelas for providing us with the crc-2 mutant, and Nich Collins for preliminary mapping of CRC and SPT using morphological markers. Gunta Jaudzems and Joan Clark provided excellent facilities for microscopy. This work was supported by the Australian Research Council.