The PRETTY FEW SEEDS2 gene encodes an Arabidopsishomeodomain protein that regulates ovule development

Ovules, the precursors of seeds, have been characterized in many species. In Arabidopsis, the ovule consists of maternal structures (the funiculus, inner and outer integuments, and nucellus), and the embryo sac,which is located within the nucellus (Fig. 1). Ovules initiate from the placenta and the distal portion of each ovule primordium develops into the nucellus(Fig. 1A). In the center of the nucellus, a large megaspore mother cell (MMC) forms and undergoes meiosis. The largest resulting megaspore divides, develops, and differentiates to form the embryo sac (Fig. 1C). The integuments initiate from the central region of the ovule primordium(Fig. 1B) and grow to enclose the nucellus. At anthesis, the embryo sac is mature and a small opening between the integuments, the micropyle, forms to allow the pollen tube growth to deliver sperm nuclei to the central cell and egg cell.

Using a model plant system, some of the loci regulating the intricate developmental events described above were identified(Skinner et al., 2004). Characterization of Arabidopsis ovule mutants has enabled the construction of a model for the genetic control of ovule development that includes parallel independent regulatory pathways for gametophyte differentiation and integument morphogenesis(Baker et al., 1997). Based on genetic and molecular analyses, many genes that regulate ovule development also regulate flower organogenesis. During the evolution of plants, ovules arose before the four floral whorls. Consequently, it has been hypothesized that the functions of some ovule regulatory genes, whose functions were redundant with other genetic loci, diverged during floral evolution and adapted to regulate one or more processes during floral development(Broadhvest et al., 2000; Gasser et al., 1998). Evidence for this proposed evolutionary pathway is supplied not only by the phenotype of mutants, but also by the degree of genetic redundancy that still exists in ovule development.

Four-fifths of the known loci that control ovule development regulate the same, or a similar process, in another plant organ. For example, the WUSCHEL(WUS) homeodomain protein controls integument initiation(Groß-Hardt et al.,2002), as well as maintaining a pluripotent cell identity in the central region of floral meristems (Laux et al., 1996; Mayer et al.,1998). In wus mutants, shoot meristems are initiated repetitively, but expire prematurely, because the pluripotent population of cells in the center of the meristem loses this stem cell activity(Laux et al., 1996). Ectopic expression of the SHOOT MERISTEMLESS (STM) and WUSgenes was sufficient to activate cell division in differentiated tissues. Accompanying microarray analyses indicate that STM and WUS each activate a different subset of developmental genes(Gallois et al., 2002). More significantly, when appropriate developmental signals are present, WUS activity can initiate the formation of leaf, flower, or embryo primordia(Gallois et al., 2004).

AGAMOUS (AG) is another locus that regulates both ovule and flower development. Based on analyses in the apetala2 mutant background, genetic data reveal that AG and the closely related SEEDSTICK, SHATTERPROOF1, and SHATTERPROOF2 loci all specify ovule identity (Pinyopich et al.,2003; Western and Haughn,1999). In flowers, the AG gene specifies organ fate and limits stem cell proliferation (Bowman et al., 1991b). AG establishes carpel identity in the fourth whorl and, in conjunction with APETALA3 activity, specifies stamen identity in the third whorl (Bowman et al.,1989; Yanofsky et al.,1990). In ag mutants, indeterminate flowers consisting of only sepals and petals form; stamens and carpels are absent. The gain-of-function studies with AG indicate that this gene is sufficient to initiate stamens and carpels(Kempin et al., 1993; Mizukami and Ma, 1992). In developing flowers, WUS induces AG expression(Lenhard et al., 2001; Lohmann et al., 2001). Later in flower development, however, repression of WUS by AG is essential to terminate floral meristem proliferation(Lenhard et al., 2001; Lohmann et al., 2001).

We previously reported that the PRETTY FEW SEEDS2 (PFS2)gene regulates ovule patterning (Park et al., 2004). Based on genetic analyses, we report here that PFS2 encodes a homeodomain gene that is a member of the WUS clade of transcription factors. The PRESSED FLOWER (PRS) and NARROW SHEATH (NS) genes are also members of the WUS clade of transcription factors (Haecker et al.,2004; Nardmann et al.,2004). In the maize ns1 ns2 double mutant, the leaf blade is significantly narrower than wild type, which results from diminished recruitment of cells from the lateral domain of the shoot meristem during leaf initiation (Nardmann et al.,2004). Consistent with this hypothesis is the expression of NS transcripts in the lateral edges of leaf primordia as they emerge from shoot meristems (Nardmann et al.,2004). In the Arabidopsis prs mutants, the lateral sepals and the cell files that normally form on the margins of the medial sepals are often absent (Matsumoto and Okada,2001). In addition, prs mutants lack stipules and sometimes exhibit defects in lateral stamen development(Nardmann et al., 2004). PRS is expressed in the margins of sepal and leaf primordial where it is proposed to induce cell proliferation(Matsumoto and Okada, 2001)and recruit meristem cells into these primordia(Nardmann et al., 2004). This function is similar to that proposed for NS.

Loss-of-function analyses indicate that PFS2 plays a key role during ovule patterning by regulating cell proliferation of the maternal integuments and differentiation of the MMC. The molecular and genetic analyses described here indicate ovule development is sensitive to the level of PFS2 activity.

Wild-type plants and recessive pfs2 mutants(Park et al., 2004) were grown at ambient laboratory temperatures with fluorescent lighting (∼100μmol/m²/s), as recommended by Kranz and Kirchheim(Kranz and Kirchheim, 1987). To generate genetic lines segregating for ag and pfs2,pollen from pfs2-1 mutants was used to fertilize putative ag-1 heterozygotes. Gametophytes that were mutant for both of these loci sometimes exhibited developmental defects. This was determined by examining cleared ovules using differential interference contrast microscopy(Herr, 1971).

Tissues were fixed overnight in 50 mM cacodylate buffer, pH 7.0, 5%glutaraldehyde. Samples were then dehydrated in a graded ethanol series, which was gradually replaced with acetone. These samples were infiltrated with epoxy resin that was polymerized overnight at 60°C(Spurr, 1969). Sections (0.5μm) were stained with thionin and acridine orange(Paul, 1980) and visualized using a Zeiss Axioscope microscope (Thornwood, NY). Bright-field images were captured using a Kodak MDS 290 digital camera (Rochester, NY) that was connected to the microscope. Images were cropped and manipulated for publication using Adobe Photoshop (Adobe Systems, Inc., San Jose,California).

SEM samples were prepared as described by Robinson-Beers et al.(Robinson-Beers et al., 1992),except as noted here. Pistils were sputter coated with platinum. Samples were examined and images captured using the Hitachi S-4000 FE-SEM (Tokyo, Japan),which was operated at an acceleration voltage of 5 kV.

Genomic DNA from pfs2-1 mutants was digested using Hind3 restriction enzyme. DNA fragments were ligated into covalently closed circles using T4 DNA ligase and transformed into SoloPack Gold Supercompetant Cells(Stratagene, La Jolla, CA). Plasmids were selected using ampicillin and the flanking DNA was sequenced using the M13 primer.

The pSOP3 plasmid contains the promoter and coding sequence for PFS2. This gene was PCR-amplified from genomic DNA using the following primers: GTTATGGATCCAAAAATATGTG and GGGACAGAGATCTTTTGAGT. To create pSOP3, the PCR product was digested using BamHI and BglII and inserted into the compatible site in pCAMBIA 2200(Hajdukiewicz et al., 1994). This plasmid was moved into Agrobacterium tumefaciens strain AGL by triparental mating (Figurski and Helinski,1979). Plants were transformed with this A. tumefacienscontaining this plasmid using established transformation techniques(Clough and Bent, 1998). Transformants were selected on 50 μg/ml kanamycin. The ovule phenotype of transformants was evaluated using microscopic techniques described above.

The pSOP2 plasmid contains the PFS2 coding sequence under control of the cauliflower mosaic virus 35S promoter (35S). Using gene-specific primers (GTTCGGAATTCCACAACAAC and GGGACAGAGATCTTTTGAGT), and genomic DNA as a template, the PFS2 coding sequence was PCR-amplified. The PCR product was digested using EcoRI and BglII, and inserted into these sites in pBH6 (Hauser et al.,2000). The 35S::PFS2 fragment was excised by cleaving the DNA with NotI and was cloned into this restriction site in pMLBart. The resulting plasmid, pSOP2, was used to transform Arabidopsisplants using the floral dip method (Clough and Bent, 1998). T1 transformants were selected by spraying seedlings with Finale (AgrEvo, Wilmington, DE), which had been diluted 1:1000 in water. In resistant plants, the presence of the transgene was verified by PCR.

The PFS2 cDNA was cloned into by RT-PCR, by the method of Kawasaki(Kawasaki, 1990). RNA was purified from inflorescences using RNeasy spin columns (Qiagen, Valencia, CA). First-strand template was primed with oligo dT₁₇, then PCR amplified using the following primers: ATGGGCTACATCTCCAACAA and TCAGTTCTTCAGAGGCATGA. The PFS2 cDNA was TA cloned into pCRII as recommended by the manufacturer (Invitrogen, Carlsbad, CA), and the resulting clones were sequenced to identify one without any mutations. The resulting plasmid is called pSOP15.

Total RNA was isolated from fresh tissue using the RNeasy plant RNA isolation kit (Qiagen, Valencia, CA). RNA concentration was measured using RiboGreen dye (Molecular Probes, Eugene, OR) and a fluorometer. Using Accupower RT PreMix (Bioneer Corp., Rockville, MD), the RNA template was primed with dT₁₇ and reversed transcribed. Qualitative differences in PFS2 transcripts were determined by PCR using the following gene-specific primers: ATGGGCTACATCTCCAACAA and TCAGTTCTTCAGAGGCATGA. Relative AG expression levels were determined using the following primers:ATGGCTGACAAGAAGATTAGG and AACGAAGTCAGTTGAGACAA. For normalization purposes,the following primers were used to amplify glyceraldehyde-3-phosphate dehydrogenase C (GAP) transcripts: ATGGCTGACAAGAAGATTAGG and AACGAAGTCAGTTGAGACAA. Different primers (CTGGAGATGATGCACCAAGA and GGAAGGTACTGAGTGATGCT) were used to amplify ACTIN11 transcripts. Amplification of transcripts was evaluated using 25, 30, and 35 PCR cycles to identify conditions where PCR reagents were not limiting. PCR products were separated by size using agarose gels, stained with ethidium bromide, and visualized under a UV light. Images from these gels were captured using a ChemImager 4400 (Alpha Innotech Corp., San Leandro, CA), and relative band intensities were measured using the associated software.

Except for the modifications noted below, previously described methods for in situ hybridization were used(Vielle-Calzada et al., 1999). To generate templates for probe synthesis, the insert in pSOP15 was PCR amplified. The T7 RNA polymerase initiation sequence was placed in front of one of the gene-specific primers, which allowed direct synthesis of digoxigenin-labeled probe from PCR products. To PCR-amplify the template for antisense probe synthesis, the following primers were used:TTCCACACACAAACCGACCACA and TAATACGACTCACTATAGGGAAAGTCCGGTTGTCCCTCGTTT. The template for the sense probe was amplified using TAATACGACTCACTATAGGG-TTCCACACACAAACCGACCACA and AAAGTCCGGTTGTCCCTCGTTT primers. RNA probes were synthesized using the Dig-RNA labeling kit (Roche Applied Science, Indianapolis, IN). The 244-bp cRNA products were synthesized and added to the hybridization solution, so the final concentration was 500 ng/ml. Probes were not hydrolyzed. Slides were hybridized at 45°C and washed at 50°C. For color detection, 1 mM levamisole (Sigma, St Louis, MO)was added to Western Blue substrate (Promega, Madison, WI). Slides were evaluated under bright-field and DIC optics and images were captured and modified as before.

The recessive pfs2 mutant exhibits defects in early ovule development including embryo sac differentiation(Park et al., 2004). Reciprocal crosses between pfs2-1 mutants and wild-type plants confirmed that defects in reproduction derive from sporophytic defects in ovule development (Table 1). The crosses show that pfs2 pollen was as virile as wild-type pollen,indicating that pollen development was unaffected in the pfs2-1 mutant background (Table 1). In pfs2 ovules, the embryo sac was occasionally absent(Fig. 2B), but more frequently,it exhibited morphological defects and contained fewer than the normal complement of seven cells (Fig. 2C). Wild-type integument cells were columnar(Fig. 2A), but pfs2 integument cells exhibited less directional cell expansion(Fig. 2B). The reduction in directional growth and cell division in this mutant resulted in integuments that commonly were shorter than wild type(Fig. 3B). Only 4% of the pfs2 mutants developed functional embryo sacs as shown by analyses of ovule anatomy and subsequent plant fertility trials.

In addition to aberrant ovule phenotypes, the leaves and petals of pfs2 mutants displayed abnormalities. When compared to wild-type petals, the edges of pfs2 mutant petals were wavy and crenulated(Fig. 2G). In pfs2 mutants, the leaves appeared narrower than wild type due to leaf curling(Fig. 2E), but the width of pfs2 leaves was not significantly different than wild type (data not shown).

Since the pfs2-1 mutant derives from a T-DNA mutated line, it was necessary to determine if the T-DNA mapped to the mutant locus. In a segregating population, genetic data indicated that the T-DNA was tightly linked with the pfs2 mutation; no recombinants were found in 68 mutant plants. Accordingly, the DNA flanking the T-DNA insert in pfs2-1 was isolated by plasmid rescue and sequenced (see Materials and methods). The rescued DNA corresponds to the At2g01500 locus, which encodes a homeodomain transcriptional factor. In a genome-wide survey of WUS homeodomain genes, Haecker et al.(Haecker et al., 2004)annotated At2g01500 as WOX6. In the pfs2-1 and pfs2-2 alleles, T-DNA inserts were incorporated into the second and third intron of this homeodomain gene, respectively(Fig. 4A).

A wild-type copy of this homeodomain gene (At2g01500) was used to transform pfs2-1 mutants to determine if it would complement the mutant phenotype. The pSOP3 construct contains the PFS2 promoter, exons,introns, and untranslated sequences. This construct was transformed into pfs2 mutants and the transformants were evaluated microscopically. While pfs2 mutants containing the pSOP3 transgene were fully fertile (Fig. 3A), 60% of the transformants exhibited subtle defects in ovule morphology. In these plants,the outer integument was shorter than in wild type and exposed the inner integument and nucellus (Fig. 3D). These subtle morphological defects in ovules might result from gene dosage effects or inappropriate expression of the transgene in some transformants. Nonetheless, restoration of full fertility to pfs2 mutants containing a wild-type copy of this homeodomain gene indicates that the PFS2 locus encodes the At2g01500 protein.

Fig. 4B shows the alignment of the proteins most similar to PFS2, including PRS, WUS, and PFS2-LIKE(At3g18018). Within the homeodomain and WOX (TL-LFP) domain, these four proteins share 95% amino acid identity(Fig. 4B). The PFS2 and PFS2-LIKE proteins share short motifs at the N- and C-termini(Fig. 4B). Outside these amino acid regions and domains, the similarity among these proteins drops precipitously.

To relate the observed pfs2 mutant phenotype with the expression of PFS2 in wild-type plants, the relative abundance and localization of PFS2 was measured using RT-PCR and in situ hybridization(Fig. 5). In developing seeds, PFS2 transcripts were found in the embryo, suspensor, and endosperm nuclei, but were absent in the integuments(Fig. 5A,B). In seedlings, PFS2 transcripts were found in the shoot apical meristem and leaf primordia, but were not expressed in expanded cotyledons or mature leaves(Fig. 5C and data not shown). In reproductive structures, transcripts could be detected in floral apical meristems, floral primordia, stamens, and pistils(Fig. 5D,E). Although the pfs2 mutant does not have male defects, the PFS2 gene expresses in the tapetum in anthers (Fig. 5E). In the gynoecium, PFS2 expression was present throughout developing ovules (Fig. 5E-G). In addition, PFS2 mRNA was weakly expressed in petals, sepals, and the walls of carpels(Fig. 5H,J). In summary, PFS2 mRNA was expressed in differentiating primordia, but absent in organs that were nearly mature or fully developed.

In pfs2 mutants, despite the wide distribution of transcripts,only leaves, petal margins, and ovules exhibited an abnormal phenotype. Either PFS2 has a limited role in the development of primordia from shoot meristems,where this gene product was expressed, or a redundant gene complements the absence of activity there. The PFS2 gene, similar to other known WOXgenes, exhibits an expression domain that is much broader than the tissues/organs that display aberrant phenotypes in mutant individuals(Haecker et al., 2004).

To determine if PFS2 was active outside the ovules, this gene was expressed under the control of the 35S cauliflower mosaic virus promoter. In transgenic plants overexpressing PFS2 (PFS2 OE), there were a variety of phenotypes. The most severe phenotypes corresponded to the highest levels of PFS2 expression – many of these plants were unable to form reproductive structures. In the most severely effected plants,the apical meristems enlarged, but primordia that formed remained undifferentiated for weeks (Fig. 6A). Occasionally a leaf primordium emerged, but it did not mature into a leaf (Fig. 6B). Other transformants showed disorganized or irregular leaf orientations or phyllotaxy(Fig. 6C). These plants often became highly fasciated because of the inability of the primordia to separate from the apical meristem and adjacent primordia. Most transformants failed to reproduce because plant development terminated before reproductive structures formed.

Plants expressing the PFS2 gene at lower levels formed flowers. In transformants that made flowers, the stamens sometimes had stigmatic papillae on the tips (Fig. 6D). Stronger floral phenotypes included an increase in the number of carpelloid stamens,especially in the second and third whorls(Fig. 6E). Carpelloid stamens often formed at the expense of petal development. In the ovules from these plants, the outer integuments did not undergo, or exhibited reduced,asymmetric cell expansion (Fig. 6F), and the nucellus of these plants contained only small parenchymatous cells (Fig. 6G). Expression of PFS2 suppressed differentiation of the MMC and outer integument. Thus, each of the PFS2 OE phenotypes showed defects in differentiation of primordia from meristematic regions.

The leaf phenotype of pfs2 mutants resembles the previously reported phenotypes for curly leaf (clf) mutants(Goodrich et al., 1997) and transgenic plants that overexpress AG(Kempin et al., 1993; Mizukami and Ma, 1992). CLF activity represses AG transcription in leaves. In the clfmutant, ectopic expression of AG in the leaves induced leaf curling(Goodrich et al., 1997). Since the leaves in pfs2 mutants also curl(Fig. 7C), the relative expression of AG was assayed in various genetic backgrounds. RT-PCR data demonstrated that, as PFS activity increases, the relative abundance of AG transcripts decreases(Fig. 7E). The leaf phenotype of ag pfs2 double mutants was examined to determine if AG activity was necessary for leaf curling. In ag pfs2 double mutants, leaf curling diminished but was not completely eliminated(Fig. 7D). Thus, analyses of these single- and double-mutant phenotypes reveal that AG activity was partially responsible for leaf curling(Fig. 7).

When the relative abundances of AG transcripts were similarly measured in inflorescences similar results were found; when compared to wild-type flowers, AG expression decreased in PFS2 OE flowers and increased in pfs2 mutant flowers(Fig. 7E). Analyses of AG localization indicate that this decrease in AG expression derives primarily from reduced expression in young floral primordia(Fig. 7G). Samples from PFS2 OE and wild-type inflorescences were simultaneously prepared and then hybridized on the same slide to allow signal intensities to be compared. In wild-type ovules at anthesis, AG transcripts localize to the endothelium, which surrounds the gametophyte(Bowman et al., 1991a). In PFS2 OE ovules, which fail to form a gametophyte, AGtranscripts localized to the nucellus and inner integument(Fig. 7G). This observation indicates that gametophyte identity either was not established or was established and then lost from this region of the ovule. Normally AGtranscripts are present in developing gynoecia and stamens, but in PFS2 OE plants the AG transcripts were also detected in some of the petal primordia (Fig. 7I). Previous publications(Lenhard et al., 2001; Lohmann et al., 2001) show that expanding the domain of AG expression leads to many of the floral phenotypes observed in Fig. 6E.

The results presented here indicate that the PFS2 gene encodes a homeodomain transcription factor. Restoration of complete fertility to pfs2 mutants containing a wild-type copy of this homeodomain gene substantiates this conclusion. In a subset of these transformants, however,subtle defects in ovule morphology were observed(Fig. 3D). Similar results for other mutant lines that were complemented with a transgene were attributed to altered expression of the transgene(Elliott et al., 1996). The pSOP3 complementation construct lacked 3′ untranslated sequences, which affects the expression of some genes(Larkin et al., 1993). The absence of these sequences from pSOP3 may account for the observed morphological defects. Based on the mutant phenotype and double mutant interactions, it was hypothesized that the PFS2 gene either affects ovule differentiation or ovule patterning(Park et al., 2004). Since both of these hypotheses are consistent with the data collected to date, they will be discussed further.

Defects in patterning of the nucellus in ovule primordia, not only affect development of the distal portion of ovule primordia, but also the chalaza and integuments. In nzz ovule primordia, the distal nucellus domain is not established, which results in inappropriate expression of chalaza-specific genes in this region (Sieber et al.,2004). Similarly in the nucellus of pfs2 mutants, defects in the patterning of the ovule primordia were observed; parenchyma cells often proliferate in the area normally occupied by the MMC(Park et al., 2004). This could result either from the failure to form a boundary between the nucellus and chalaza or a breakdown in ovule patterning. Since PFS2 transcripts were present in the chalaza and the nucellus(Fig. 5), PFS2 could not act alone to (1) establish a boundary between the chalaza and nucellus or(2) specify the identity of the nucellus or MMC during patterning. If PFS2 regulates one of these functions, its activity would need to be modulated by a region-specific factor.

While patterning defects are found in pfs2 mutants, these could derive from defects in regulating the timing of differentiation. In pfs2 mutant ovules, the decreased size of the MMC, gametophyte, and integuments may be due to premature maturation of cells. In pfs2mutants, the integuments are shorter than wild type, which could result from premature differentiation of integument primordia. The cell in the center of the nucellus normally develops into the MMC, but if this cell differentiates before this identity is specified, then meiosis will not occur. Similarly,defects in megagametogenesis occur if the functional megaspore or one of its mitotic products differentiates too early. This is consistent with observations of the mutant phenotype and accounts for both decreased integument length and defects in MMC and gametophyte differentiation. In addition, maintenance of cells in an undifferentiated state by PFS2 explains the overexpression phenotypes. Differential growth of the outer integument is regulated by the INNER NO OUTER (INO) locus, which establishes abaxial polarity (Baker et al.,1997; Balasubramanian and Schneitz, 2002; Villanueva et al., 1999). As a result of PFS2 overexpression, abaxial polarity in the outer integument was disrupted, which led to reduced growth of the outer integument. This often induced a phenocopy of the inomutant (Fig. 6F). In PFS2 OE plants, the MMC did not differentiate and small cells filled this region (Fig. 6G). Thus,analyses of these data led to the hypothesis that PFS2 helps to maintain cells in an undifferentiated state.

The genes that share the highest similarity to PFS2 are NS,PRS, and WUS, which have been proposed to recruit meristematic cells into emerging primordia, promote cell proliferation, and establish a pluripotent population of stem cells (Laux et al., 1996; Matsumoto and Okada, 2001; Nardmann et al.,2004). Each of these processes affects cell proliferation or inhibition of cellular differentiation or both. PFS2 is expressed in developing primordia, but is absent from mature tissues(Fig. 5). PFS2overexpression interfered with differentiation and maturation of leaves, outer integuments, and floral primordia (Fig. 6). The broad expression pattern of PFS2 transcripts and the overexpression phenotypes indicate that the PFS2 locus is active in these regions. While different plant regions were affected in PFSOE plants, these data are consistent with the observation that PFS2 delays differentiation and maturation of primordia. This proposed function is similar to the functions of other members of the WUS clade of genes.

Comparable WUS and PFS2 overexpression phenotypes lend further support to this hypothesis. When WUS was expressed under control of the APETALA3 promoter, the result was the development of supernumerary stamens and carpelloid stamens(Lenhard et al., 2001; Lohmann et al., 2001). Overexpression of PFS2 in flowers yielded a similar phenotype(Fig. 6E). Depending on the stage of development and location in the floral meristem, WUS both activates and represses AG expression(Lenhard et al., 2001; Lohmann et al., 2001). The repression of WUS expression by the AG protein establishes a complex feedback loop that correctly establishes stem cell identity and floral determinacy during floral development(Lenhard et al., 2001; Lohmann et al., 2001). Since AG establishes and maintains determinant growth of floral primordia, its relative decrease in abundance in PFS2 OE plants could account for the emergence of extra organ primordia from developing flowers. This proposal will be rigorously tested in future work.

We thank Daynet Vega and Lorraine Keller for technical assistance, John Davis for the gift of pCAMBIA 2200, Bart Janssen for pMLBART, Yuval Eshed for ag-1 mutant seed, and Margaret Joyner, Paris Grey and two anonymous reviewers for helpful comments. This research was supported by the National Research Initiative of the USDA Cooperative State Research, Education and Extension Service (grant number 2002-35100-12109), the NSF Floral Genome Project (DBI-0115684), and a UF dissertation fellowship to S.O.P.