The AP2 transcription factors DORNRÖSCHEN and DORNRÖSCHEN-LIKE redundantly control Arabidopsis embryo patterning via interaction with PHAVOLUTA

Angiosperm growth results from two developmental growth phases: an embryonic phase whereby the primary body plan is established together with the apical-basal and radial axes of symmetry, and a postembryonic phase, in which new organs are initiated via the shoot and root meristems. Cotyledon development during dicot embryogenesis marks the start of organogenesis and the change from radial to bilateral symmetry at the transition from the globular- to heart-stage embryo. Studies of seedling-lethal Arabidopsis mutants have defined three embryonic domains wherein subsequent cell fate is largely determined by position in the embryo(Mayer et al., 1991): (1) the majority of the cotyledons, together with the shoot apical meristem, are derived from the apical domain, recognisable as the derivatives of cell divisions of the upper tier of cells at the octant stage; (2) the central domain mainly gives rise to the hypocotyl and root apical initials; and (3)the basal region, derived from the hypophysis, generates the root quiescent centre (Harada, 1999).

Several Arabidopsis genes are important for the development of the apical region of the embryo and give rise to cotyledon phenotypes when mutated. Combined mutations at the CUP-SHAPED COLYLEDON (CUC1,2 and 3) loci result in cotyledon fusion accompanied by an absence of the shoot apical meristem (SAM)(Aida et al., 1997; Vroemen et al., 2003; Hibara et al.,2006). Polar auxin transport essentially contributes to the establishment of both bilateral symmetry(Liu et al., 1993) and apical-basal polarity (Friml et al.,2003), and an abnormal cotyledon number at low penetrance is also observed in plant hormone response mutants(Rashotte et al., 2006; Saibo et al., 2006). In Arabidopsis, auxin polarity and embryonic patterning have been extensively studied in mutants of PIN gene family members that encode plant-specific proteins involved in auxin efflux. PIN proteins are functionally redundant, and higher-order pin mutants reveal defects in embryonic cell division patterns that are reflected postembryonically mostly in cotyledon defects such as fusion or monocotyledony(Friml et al., 2003; Furutani et al., 2004; Vieten et al., 2005). Additionally, pinoid, a mutant in a serine/threonine protein kinase that affects localisation of PIN proteins(Friml et al., 2004), or mutations in MONOPTEROS (MP) and BODENLOS(BDL), which encode the auxin responsive factor ARF5 and its inhibitor IAA12, respectively, all cause cotyledon defects and/or disrupted embryo domains (Bennet et al.,1995; Hardtke and Berleth,1998; Hamann et al.,1999). A general feature of auxin signalling and cucmutants is the incomplete penetrance of cotyledon defects, suggesting that pathways leading to bilateral symmetry and cotyledon establishment are considerably redundant.

Redundant control of embryo patterning is also demonstrated by the Arabidopsis class III HD-ZIP gene family. PHAVOLUTA(PHV) and PHABULOSA (PHB) are well-known representatives of this HD-ZIP subclass and were identified as dominant gain-of-function alleles due to mutations in a highly conserved microRNA target site (McConnell et al.,2001; Bao et al.,2004). Amongst this family, only REVOLUTA (REV)has a loss-of-function phenotype (Talbert et al., 1995); however, higher-order knockouts reveal redundant functions in embryo and cotyledon patterning(Prigge et al., 2005).

The DORNRÖSCHEN (DRN) (also known as ENHANCER OF SHOOT REGENERATION1; ESR1) gene contributes to Arabidopsismeristem organisation (Kirch et al.,2003) and cytokinin-independent shoot regeneration(Banno et al., 2001). DRN expression is highly dynamic and is observable from the two- to four-cell stage in the embryo proper, before focussing to the emerging cotyledons and becoming restricted to the SAM at the torpedo stage. During postembryonic development, DRN remains detectable in the L1 layer of the SAM, from where expression extends into emerging lateral organs(Kirch et al., 2003). A paralogous gene, DRNL, is linked to DRN on chromosome 1 and has also been named ESR2 (Ikeda et al., 2006), SOB2(Ward et al., 2006) and BOLITA (Marsch-Martinez et al.,2006). To elucidate further the role of DRN and DRNL in Arabidopsis development, we characterised loss-of-function mutants and, following a yeast two-hybrid screen with DRN,established that both proteins are capable of heterodimerising with members of the class III HD-ZIP family. In view of the highly redundant control of early embryo patterning involving multiple independent gene pathways, we show here that DRN and DRNL are two additional factors that control Arabidopsis embryogenesis. DRN not only acts upstream of auxin polar transport and response, but also functions redundantly with DRNL and interacts with PHV in planta.

Two independent insertion mutants in the DRN gene (At1g12980)containing a dSpm element were obtained from the SLAT lines in Columbia (Tissier et al.,1999): drn-1 (Kirch et al., 2003) and drn-2 (SM_3.35017). For the DRNL gene (At1g24590), a drnl-1 mutant in Ler was identified from an enhancer-inhibitor insertional mutagenesis screen as line ITS75 (Speulman et al., 1999);the drnl-2 EMS allele in Ler was a gift from T. Jack(Dartmouth College, Hanover, NH). An insertion line (899_CO2) for the PHV gene (At1g30490) was obtained from the SAIL population(Sessions et al., 2002). All gene insertions were confirmed by PCR (primers DRNF and Spm8 for DRN;DRNLF and ITIR3 for DRNL; PHVIntR and LB3 for PHV; Table 1) followed by sequencing of the amplicons. Homozygous mutants were confirmed by the absence of the wild-type gene using primers flanking the insertion (DRNF and DRNR, DRNLF and DRNLR, or PHV and PHVIntR). For drnl-2, homozygosity was confirmed using a dCAPS marker with primers DRNLCAPS1 and DRNLCAPS2, which give a wild-type amplicon of 184 bp that is cleaved in the mutant by AccI into two fragments of 159 and 25 bp. drn-1 and drnl-1 mutant lines were back-crossed three times.

Plants were cultivated on soil in the greenhouse or in sterile culture on 0.5×MS medium supplemented with 1% sucrose under long-day conditions (16 hours light, 8 hours dark) at 22°C.

Cotyledons were cleared with cold acetone for 20 minutes and decolourised with 100% ethanol overnight. Ovules were dissected from siliques and cleared overnight with Hoyers Solution (2.5 g gum arabic, 100 g chloral hydrate, 5 ml glycerol and 30 ml water). Microscopy was performed using a Zeiss Axiophot microscope equipped with an Axiocam HR CCD camera using differential interference contrast optics.

Non-radioactive in situ hybridisations and the preparation of dioxygenin-labelled RNA probes by T7 RNA polymerase essentially followed the protocols of Kirch et al. (Kirch et al.,2003) or Bradley et al.(Bradley et al., 1993). Probes were as follows: for DRN, from nucleotide +377 (relative to the ATG at +1) to the stop codon and including 78 bp of the 3 UTR; for DRNL,nucleotide +348 to the stop codon; and for PHV, from nucleotide +1585 to +2475.

The DR5::GFPer and pPIN1::PIN1:GFP reporter lines (gifts from J. Friml, ZMBP, Tübingen, Germany) were crossed into the drn-1 mutant background. Homozygous drn-1 plants harbouring the DR5::GFPer construct or segregating F₂drn-1embryos from a cross between drn-1 and a pPIN1::PIN1:GFPline were monitored for GFP expression using a Leica confocal microscope.

The construction of a meristem-enriched cDNA library has been described by Cole et al. (Cole et al.,2006). As bait, 348 bp of the DRN open reading frame(ORF) encoding the N-terminal 116 amino acids of the DRN protein was amplified by PCR, sequenced, directionally cloned into pGBKT7 (Clontech) using NcoI and BamHI, and transformed into the yeast strain Y187(Clontech Palo Alto, CA). The two-hybrid screen was performed by yeast mating,according to the manufacturer's protocol (Clontech PT3024-1), and with quadruple selection (β-glucuronidase, -Leu, -Trp, -Ade).

The ORFs encoding full-length DRN or DRNL and PHVs (the C-terminal 451 amino acids from amino acid 391 to the end) were cloned in-frame into the BamHI site of pUC-SPYCE or pUC-SPYNE(Walter et al., 2004). For control experiments, GFP fusions were created in the pRT-ΩNotI/AscI vector(Überlacker and Werr,1996). Transient expression in leek epidermal cells was performed according to Cole et al. (Cole et al.,2006). YFP/GFP fluorescence was visualised using a MZFLIII stereomicroscope (Leica) after UV excitation and using a GFP filter. All images were processed using Photoshop software (Adobe).

Epitope-tagged proteins or peptides were synthesised via the EasyXpress in vitro transcription/translation system (Qiagen) based on the T7 promoter. Templates of the required proteins were obtained via nested PCR reactions on the respective gene ORFs in pUC-SPYCE/NE containing an HA or myc epitope-coding sequence preceding the YFP subdomains. The T7 promoter and 6xHis tag were added to DRN and DRNL coding regions by nested PCR using primers from the EasyXpress Kit. For co-expression, the relevant amplicons were mixed. Control reactions were performed with single amplicons. Immunoprecipitation (IP), gel electrophoretic analyses and detection of epitope-tagged proteins essentially followed the protocols of Cole et al. (Cole et al.,2006). Peroxidase activity was detected via chemiluminescence and documented on Kodak X-Omat AR film. C-terminal fragments of class III HD-ZIP proteins, the PAS-like domain and the AP2 domain and were amplified by PCR from cDNA derived from various plant tissues, and epitope-coding sequences or the T7 promoter were added by nested PCR using the EasyXpress Kit. Polypeptide termini: PHV, amino acid 624 to end; PHB, amino acid 636 to end; REV, amino acid 644 to end; CNA, amino acid 618 to end; AtHB8, amino acid 619 to end; and PHV PAS-like domain, amino acids 721 to 792. The AP2 domain of DRN and DRNL extended from amino acids 54 and 55 to 115 and 116, respectively.

The position of the insertion in the drn-1 allele is after nucleotide +327 (relative to the ATG) of the DRN gene, within the AP2-domain coding region. For drn-2, the insertion is after nucleotide +7 of the DRN ORF, and for drnl-1 it is after nucleotide +777. The drnl-2 allele has a base substitution from C to T at position +278, resulting in an A to V substitution at amino acid 93. This conserved A residue has recently been shown in Brassica napus ERF/AP2 proteins to be essential for DNA binding(Liu et al., 2006), suggesting that the drnl-2 AP2 domain is unable to bind target genes. For the phv 899_CO2 allele, the insertion is after nucleotide +84.

The major phenotype of the drn mutant is abnormal cell division,observable from the globular embryo stage onwards(Table 2). The wild-type globular embryo contacts the single file of cells known as the suspensor through the hypophysis, which divides asymmetrically in the globular embryo to give: (1) an upper lens-shaped cell (see Fig. 1A,B), which is the progenitor of the quiescent centre; and (2) a subtending daughter cell, which will generate the columella stem cells. It is at the transition from globular to heart stage that cell divisions parallel to the surface at lateral marginal positions give rise to the emerging cotyledon lobes(West and Harada, 1993). drnl mutants showed no embryo development defects. By contrast,approximately half (48.0%) of homozygous drn-1 embryos exhibited a phenotype consisting of abnormal development in the hypophysis region, or no obvious distinction between embryo proper and suspensor. In a few percent of drn mutants, the hypophyseal cell divided periclinally at an early stage (Fig. 1C), resulting in the absence of the lens-shaped cell and a subsequent haphazard cellular organisation of this region (Fig. 1D,F). At a similar frequency, suspensors of drn mutants had double or triple cell files (Fig. 1E), presumably resulting from supernumerary cell divisions perpendicular to the normal plane of division. Both drn and drnl single mutants showed defective cotyledon development phenotypes at incomplete penetrance, including monocotyledonous seedlings, seedlings with partially fused cotyledons, tricots or various tricot fusion combinations(Table 2 and Fig. 1G-I). Completely fused or cup-shaped cotyledons were occasionally observed (<1%) in both drnmutant alleles, and these plants produced a functional SAM and developed normally (Table 2 and Fig. 1J,K). Also rare (<1%)were plants with a single cotyledon-like structure, no hypocotyl and a rudimentary root (Fig. 1L,M),reminiscent of mp (Berleth and Jürgens, 1993) and bdl(Hamann et al., 1999)seedlings, which are mutants in ARF5(Hardtke and Berleth, 1998)and its IAA12 partner protein(Hamann et al., 2002). The lower penetrance in drn mutants of cotyledon defects as compared with cell division defects indicates that embryo cell patterning defects are not manifested in post-germination development.

Inappropriate cotyledon development in drn or drnlmutants resulted in subsequent alterations of leaf phyllotaxy: a single leaf initiated opposite single fused cotyledons(Fig. 1N), or, in the case of tricots, three leaf primordia initiated between the cotyledons(Fig. 1O). Cleared monocotyledonous drn-1 and drnl-1 cotyledons either showed a wild-type vasculature pattern (Fig. 1P) with a single midvein (Fig. 1Q), or two or more mid-veins(Fig. 1R), showing that the cotyledon derived from either a single cotyledon primordium or from the fusion of two discrete primordia. Both alternatives suggest the improper recruitment of cells into developing cotyledon primordia, and the improper establishment or maintenance of organ boundaries during organogenesis.

In view of the drn embryo mutant phenotype, we reinvestigated the drn-D overexpression mutant(Kirch et al., 2003) for more subtle phenotypes. The main phenotype is an enlarged SAM and vegetative meristem arrest coupled with the initiation of radialised leaves(Kirch et al., 2003). However,ectopic DRN expression also affects embryonic development;polycotyledony was occasionally observed (>1%; see Fig. S1A in the supplementary material). Postembryonic developmental phenotypes were also observed, including the fusion of leaf margins, fusion between stems and leaves and variable floral organ number (>2-3% of plants; see Fig. S1B-E in the supplementary material), implying that DRN contributes to organogenesis and patterning throughout development.

The incomplete penetrance and similarity of drn and drnlsingle-mutant cotyledon phenotypes suggest redundant gene functions. We therefore created double mutants between drn-1 and either drnl-1 or the stronger allele drnl-2. Embryonic cell patterning defects were observed in essentially all (90 of 96) drn-1 drnl-1 double-mutant embryos (Fig. 1S,T), considerably higher than the frequency for drn-1single-mutant embryos (Table 2). drn-1 drnl-1 double mutants also showed pleiotropic cotyledon phenotypes, including those observed in the single mutants(Table 2). However, the most significant feature of the drn-1 drnl-1 double mutant was a large increase in phenotypic penetrance: the mp-like phenotype was observed in 20% of mutants, and cotyledon defects in 30%, raising the total penetrance of plants with a phenotype to about 50%(Table 2). Double-homozygous drn-1 drnl-2 plants are sterile, but approximately a quarter of the progeny of drn-1 drnl-2/DRNL plants had pin-like embryos,with a complete absence of cotyledons (Fig. 1U). These plants were genotyped as double-homozygous mutants and directly initiated leaves from a functional SAM(Fig. 1V).

In view of the genetic redundancy observed between DRN and DRNL, we investigated whether the expression patterns of these genes overlapped in Arabidopsis embryos using RNA in situ hybridisation. DRN is expressed from the four-cell stage(Kirch et al., 2003) and throughout the globular embryo (Fig. 2A). At the transition stage, DRN expression was localised to the apical cell tiers (Fig. 2B), and throughout the heart stage it became increasingly restricted to the lobes of the developing cotyledons(Fig. 2C). From the mature heart stage throughout the torpedo stage and until embryo maturity, DRN expression was confined to the SAM(Fig. 2D). DRNLexpression began in the early globular embryo(Fig. 2E) and was then restricted to the apical portion of the globular embryo(Fig. 2F). During the heart stage, the DRNL expression domain pre-patterned that of the emerging cotyledons, and at the elaborated heart stage, transcripts were confined to the sub-epidermal cells at the tip of the cotyledons(Fig. 2G); DRNLtranscription ceased after the heart stage(Fig. 2H). These data show that the expression of DRN and DRNL overlap in the apical hemisphere of the globular-stage embryo and in sub-epidermal cells of the developing cotyledons.

The anomaly between the expression domains of DRN and DRNL in the central and apical embryo region and in the hypophysis region, where defective cell division phenotypes are observed in drn-1 and drn-1 drnl-1 mutants, suggests that both genes have non cell-autonomous functions. Although movement of DRN and DRNL proteins cannot be excluded, given the similarities between drn and drnl mutant phenotypes and those of many auxin-signalling mutants,auxin is a good candidate for mediating DRN and DRNLfunction. To test this, we crossed the DR5::GFP reporter construct,which indirectly measures auxin distribution and response, into the drn-1 mutant background. In wild type, a DR5::GFP maximum was established in the hypophysis and upper suspensor cell at about the 32-cell stage (Fig. 2I). In drn-1 mutants, DR5::GFP showed a more centralised, diffuse expression (Fig. 2J), and a subsequent maximum in the basal domain often appeared that was asymmetrical and not confined to the hypophysis and upper suspensor cells(Fig. 2K). In wild-type heart-stage embryos, DR5::GFP maxima were observed at the base of the embryo and at the tips of the developing cotyledons in the L1 cell layer(Fig. 2L). Heart-stage drn embryos showed either a similar DR5::GFP maximum to that of wild type in normal cotyledons, or an absence of expression in phenotypically abnormal cotyledons (Fig. 2M). The abnormal DR5::GFP expression pattern in drn mutants demonstrates that DRN functions upstream of auxin-mediated responses necessary for root and cotyledon specification.

We used PIN1 as an additional marker to address auxin transport in the drn mutant embryo. We analysed F₂ embryos from a cross between drn and a PIN1::PIN1-GFP transgenic line,segregating drn and wild-type embryos and allowing a simultaneous comparison of PIN1 expression to be made in both genotypes. From a total of 149 embryos analysed, 21 showed a drn mutant phenotype. At about the 32-cell stage, PIN1 in the embryo centre was laterally localised, whereas in the wild type it was basally localised (compare Fig. 2N with 2O). Slightly later, the disorganised cells in the hypophysis region of the drnmutant also expressed PIN1 with a variable cellular localisation, including significant lateral concentration, whereas in the wild type the distribution was basal (compare Fig. 2P with 2Q). The altered expression of both DR5::GFP and PIN1::PIN1-GFP in the drn mutant unequivocally places DRN function upstream of auxin transport and response in the early embryo.

To elucidate further the role of DRN in embryo development, we isolated putative interacting protein partners in a yeast two-hybrid screen using the N-terminal 116 amino acids of the DRN protein, including the AP2 domain, as bait. The C-terminal part of the DRN protein was excluded as it demonstrated weak auto-activation activity. Amongst 177 positive clones sequenced from a total of 1576 following quadruple selection for Trp, Leu, Ade and theβ-glucuronidase reporter gene, were sequences encoding several kinases,metabolic enzymes and proteins of unknown function. Since DRN belongs to a plant-specific family of transcription factors, we focussed our analyses on potential transcription factor partners. Of note was PHV (AtHB9; At1g30490),which was independently isolated eight times, based on cDNA termini sequences. All isolated PHV clones encoded C-terminal parts of the protein, extending maximally from amino acid 754 to the last amino acid, 841.

The PHV expression pattern is described elsewhere(Prigge et al., 2005), but to support the functional significance of the putative protein-protein interactions between PHV and DRN or DRNL, we demonstrated that PHV is temporally and spatially co-expressed with DRN and DRNL in the proembryo, and is concentrated apically in the developing globular embryo before becoming localised to the adaxial side of the developing cotyledons (Fig. 2R-U).

We used two methods to substantiate the affinity of the protein-protein interaction between DRN and PHV shown by the yeast two-hybrid screen. Firstly,we performed co-immunoprecipitation (CoIP) experiments using epitope-tagged full-length DRN, DRNL and PHV. These results(Fig. 3A,B) confirmed the protein-protein interaction between DRN and PHV and between DRNL and PHV in vitro. DRNL could not be co-immunoprecipitated by DRN(Fig. 4B), demonstrating that DRN and DRNL are not capable of heterodimerisation.

Additionally, we used bimolecular fluorescence complementation (BiFC)(Walter et al., 2004) in a transient assay in leek epidermal cells to verify the observed biochemical interaction between DRN or DRNL and PHV, in vivo, using full-length DRN and PHVs as the C-terminal part of PHV was sufficient to sustain interaction with DRN in the two-hybrid screen. GFP expression was observed in control experiments (Fig. 3C-E) and YFP expression was reproducibly obtained in multiple independent experiments when the PHVs BiFC construct was co-bombarded with DRN or DRNL BiFC constructs(Fig. 3F,G). A series of negative controls were performed, including co-bombardment of empty YFP vectors, and no evidence for DRN heterodimerisation with DRNL was observed(data not shown). More importantly, no YFP fluorescence was observed following co-bombardment of DRN in pUC-SPYCE and SHOOTMERISTEMLESS (STM) in pUC-SPYNE,nor in the reciprocal cloning combination (data not shown). These results not only demonstrate that that both DRN and DRNL can form stable heterodimers with the PHV protein in planta, but that these interactions are specific.

Considering the biochemical dimerisation between DRN and PHV in vitro and in planta, we asked whether genetic evidence would support the hypothesis of an active DRN-PHV protein dimer by constructing a drn-1 phv double mutant. The phv mutant showed wild-type embryo development. However,the drn-1 phv mutant showed embryo cell division defects similar to those observed in the drn-1 drnl-1 double mutant and at almost complete penetrance as compared with that of the drn single mutant(Table 2). Additionally, the penetrance of drn-1 phv cotyledon defects was higher than that of drn single mutants, with slightly more plants with an mpphenocopy (Table 1). The increased penetrance of embryo cell defects of drn-1 phv plants over drn single mutants suggests that both genes contribute to the same embryo developmental pathways.

The Arabidopsis class III HD-ZIP protein family also contains the closely related homologue of PHV, PHABULOSA (PHB), in addition to REVOLUTA(REV), CORONA (CNA; also known as ATHB-15 - The ArabidopsisInformation Resource) and ATHB8 (Sessa et al., 1994; Prigge et al.,2005). As these family members act redundantly(Prigge et al., 2005), we investigated whether DRN and DRNL could interact with other class III HD-ZIP members apart from PHV. We used equivalent C-terminal parts of all five proteins corresponding to the PHV sequences obtained from the two-hybrid screen (see Materials and methods). His-tagged fulllength DRN and DRNL proteins were detected following CoIP via the HA-tagged C-terminal fragments of all five members of the class III HD-ZIP family(Fig. 4B). DRN and DRNL can thus dimerise with all Arabidopsis class III HD-ZIP proteins, a finding supported by their overlapping expression patterns in the embryo(Prigge et al., 2005)(Fig. 2). We also tested the specificity of PHV interactions by investigating whether the full-length PHV protein could co-precipitate STM, another protein whose expression domain overlaps that of PHV in the embryo. A full-length STM-GFP fusion protein could not be co-precipitated via the HA tag of PHV(Fig. 4C), confirming that PHV specifically interacts with DRN and DRNL.

The C-terminal region of class III HD-ZIP members of Arabidopsisis highly conserved (Fig. 4A)and therefore indicative of conserved function. Using the SMART protein motif search of EMBL(http://smart.embl-heidelberg.de),putative homology was found to several characterised domains, including the PAS domain - named after Drosophila period (PER), vertebrate aryl hydrocarbon receptor nuclear translocator (ARNT) and Drosophilasingle minded (SIM) - which is found in a superfamily of more than 1000 eukaryotic proteins directly or indirectly involved in signal transduction(Mukherjee and Bürglin,2006). Homology to the PAS domain resides between amino acids 722 and 792 of the PHV protein, within the sequence we have shown to dimerise with DRN and DRNL. Although this homology is weak, probably representing the low sequence identity between PAS domains(Gilles-Gonzalez and Gonzalez,2004), this region spans five conserved α-helices of the PAS domain. For this reason, we investigated whether this region of the PHV protein was sufficient to maintain high affinity binding to the full-length DRN protein. In pulldown assays, DRN could dimerise with the 70-amino-acid PHV PAS fragment (Fig. 4D) and,reciprocally, the PHV PAS fragment was able to precipitate DRN or DRNL(Fig. 4D), identifying a functional protein-protein interaction domain within this region.

To establish which region of the DRN protein is responsible for dimerisation with PHV, we performed CoIP experiments with subfragments of the N-terminus of DRN used for the two-hybrid screen: (1) the N-terminal polypeptide (amino acids 1-53) excluding the AP2 domain; and (2) the conserved 61 amino acid AP2 consensus domain (Kim et al., 2006) (DRN amino acids 54-115). The N-terminal polypeptide of DRN was not able to co-precipitate PHV (data not shown), whereas the AP2 domain could (Fig. 4F),confirming that it is sufficient to mediate heterodimer formation with PHV. The same experiment was performed using the DRNL AP2 domain (amino acids 55 to 116), which was also able to precipitate the PHV polypeptide(Fig. 4F). The AP2 domain is generally considered to comprise a GCC-box DNA-binding motif(Riechmann and Meyerowitz,1998; Sakuma et al.,2002) composed of three-stranded β-sheets and anα-helix, occurring either in tandem repeats such as in the founding family member APETALA2 (Okamuro et al.,1997), or as a single domain as in DRN or DRNL. The 3D structure of the Arabidopsis ERF1 AP2 domain(Allen et al., 1998; Marchler-Bauer et al., 2005)shows that the DNA-binding domain resides in the N-terminal β-sheet region of the domain, leaving residues within the C-terminus to orientate on the opposite face to the DNA-binding face so as to be sterically potentially able to interact with other proteins. The cysteine and serine residues in DRN and DRNL AP2 domains (Fig. 4E;marked in yellow Fig. 4G) are noteworthy: the cysteine conserved in both DRN and DRNL is unique amongst plant AP2 domain proteins (see Kim et al.,2006) and falls within the α-helix of the RAYD domain,considered to be a conserved structural motif relevant to the function of all AP2 proteins (Okamuro et al.,1997).

Two independent drn or drnl mutant alleles share similar pleiotropic cotyledon phenotypes at low penetrance; however, DRN has a more pronounced role in patterning of the hypophysis and suspensor regions because drnl mutants were unaffected in this aspect of embryo development. The drnl-2 allele showed the highest cotyledon phenotype penetrance and is presumably a stronger allele than drnl-1. The mutated Ala residue has been shown in Brassica napus ERF/AP2 proteins to be essential for DNA binding (Liu et al., 2006), suggesting that the mutated drnl-2 protein is unable to bind target genes. Importantly, the penetrance of hypophysis or cotyledon phenotypes significantly increased in drn-1 drnl-1 double mutants, thereby demonstrating a role for DRN and DRNL in both the apical and basal domain of the Arabidopsis embryo and that the genes act redundantly. The penetrance of drn-1 drnl-1double-mutant phenotypes was still incomplete compared with that of drn-1 drnl-2 double mutants, probably reflecting different drnl-1allele strengths. The complete lack of cotyledons at almost complete penetrance in drn-1 drnl-2 mutants demonstrates that both genes together are essential for cotyledon initiation.

Functional contributions of DRN and DRNL can be separated in the apical and basal embryo domains: loss of function of either gene in the apical domain is manifest in the cotyledon phenotype observed in single and double mutants. This correlates with the expression pattern of both genes in apical cell tiers concomitant with cotyledon initiation. No cell patterning defects were observed in the basal embryo domain up to the 16-cell stage of drn-1 drnl-1 mutants, despite transcriptional activity of both genes throughout the proembryo. However, about 50% of drn single-mutant embryos and essentially all drn-1 drnl-1 double-mutant embryos exhibited subsequent cell division defects in the hypophyseal region and subtending suspensor cells in the absence of either DRN or DRNL transcription in these cells. The spatial separation of gene expression and hypophyseal phenotype, therefore, must involve either mobile DRN or DRNL proteins or other interdomain signaling components. This is similar to the situation observed for BDL and MP(Berleth and Jürgens,1993; Hamann et al.,1999), which are expressed in proembryo cells but not in the hypophysis, which in mp and bdl mutants fails to undergo the asymmetric division that gives rise to the quiescent centre precursors and columella initials. MP and BDL therefore affect the hypophysis in a non-cell-autonomous manner and signalling between embryo proper and underlying hypophysis is essential for normal root development. A transient indirect response to auxin together with a postulated additional factor operating downstream of MP and BD underlies this cell-to-cell signalling (Weijers et al.,2006). Expression of DR5::GFP in drn mutants is more informative in explaining the spatial anomaly between DRN/DRNL gene expression domains and phenotype. In wild type, an auxin maximum is established in the hypophysis and upper suspensor cell at the late globular stage and is necessary for hypophysis cell fate specification (Friml et al.,2003). This maximum as reported by DR5::GFP is absent in drn mutant globular embryos, similar to the situation in bdlmutants (Weijers et al.,2006), and might explain the cell division defects in the hypophysis region. The absence of local DR5::GFP response at the cotyledon tips and in provascular strands in drn mutant cotyledons correlates with phenotypic defects in cotyledon initiation.

PIN1, an additional auxin marker, marks all cell boundaries up to the 16-cell stage in wild type, before the polarity in expression is established that concentrates PIN1 basally in the provascular cells facing the hypophysis(Steinman et al., 1999; Friml et al.,2003). The variable polarity of PIN1 in cells of the hypophysis region of drn mutants suggests an alteration in directed auxin flow,which might explain the absence of DR5::GFP accumulation here at an appropriate temporal phase. Although variable PIN1 polarity correlates with the abnormal cell divisions in the hypophysis region, it is distinct from the domain of DRN transcription. This strengthens our conclusion that drn cell division phenotypes arise either from movement of the DRN protein or that DRN functions upstream of auxin transport and involves additional interdomain signaling components. The altered PIN1 distribution in drn embryos unequivocally places DRNfunction upstream of the auxin transport fundamental for embryo apical-basal patterning (Friml et al.,2003; Weijers et al.,2005), a conclusion that could not have been derived from altered DR5 activity alone. This is supported by the pin-like phenotype of drn-1 drnl-2 embryos, which phenocopies pin mutants deficient in polar auxin transport.

Interdomain signalling in Arabidopsis embryos is also suggested by analyses of embryo-lethal mutants, in which normal embryo development involves the inhibition of the embryonic potential of the suspensor by the embryo proper (Marsden and Meinke,1985). Aberrant hypophyseal cell divisions in drn single mutants rarely lead to a defective root phenotype, which suggests an ability of the lower hypophysis cell to generate columella stem cells or organise the RAM, independent of the number of precursor cells. This supports the hypothesis that additional and partially redundant pathways are activated as cell number increases during embryogenesis, and that these can compensate for early developmental defects (Laux et al.,2004). It should be remembered that stereotypic cell division patterns of Arabidopsis embryos are not representative of plants in general and a single row of suspensor cells is not the rule in dicots; for example, Phaseolus multiflorus has a massive suspensor consisting of multiple cell files merging into the embryo proper(Wardlaw, 1955).

The vasculature of monocotyledonous drn mutants shows that cotyledons may be single or arise from cotyledon fusion, presumably via disruption of cell recruitment into correctly initiated cotyledons and disruption in the maintenance of cotyledon boundaries during organogenesis. This, together with supernumery cotyledons in drn and drnlmutants, shows that both genes have roles in both cotyledon initiation and boundary maintenance. Leaf primordia initiate at auxin concentration maxima via the dynamic expression of PIN1, which acts as an instructive signal for organ initiation (Reinhardt et al.,2003). In drn and drnl mutants, disruption of leaf phyllotaxis suggests that cotyledon misdevelopment at least temporarily reprogrammes the phyllotactic inductive signals for leaf initiation, perhaps via further disruption of auxin responses.

We have demonstrated biochemically and via transient in planta assays that DRN and DRNL can heterodimerise with the C-termini of all members of the class III HD-ZIP family. The in vivo relevance of these interactions is supported by the expression patterns of the gene family members: all except ATHB8are co-expressed with DRN and DRNL in the apical embryo domain (Emery et al., 2003). In the absence of evidence for heterodimerisation between DRN and DRNL, both proteins may individually compete for the same class III HD-ZIP interaction partners, which also act redundantly(Prigge et al., 2005).

The conserved C-terminus region is characteristic for class III HD-ZIP proteins, although a function has yet to be assigned to it. Mukherjee and Bürglin (Mukherjee and Bürglin,2006) identify a MEKHLA domain within eukaryotic PAS-containing proteins, specific to class III HD-ZIP genes of higher plants and Chlamydomonas reinhardtii. It contains the PAS domain and an additional 150 amino acids and is hypothesised to represent a discrete functional unit involved in a signalling pathway(Mukherjee and Bürglin,2006). Our data unequivocally show that a 71 amino acid region containing five α-helices within the PAS-like domain is sufficient to mediate interactions between DRN and PHV. These α-helices and the associated five- to six-stranded antiparallel β-barrels form a pocket that may contain various prosthetic groups(Mukherjee and Bürglin,2006). PAS domains have been reported to mediate protein-protein interactions (Taylor and Zhulin,1999; Card et al.,2005) and contain two highly conserved S1 and S2 regions(Zhulin et al., 1997). Bacterial PAS domains sense oxygen via a bound heme molecule and propagate a signal via the His kinase pathway (Hao et al., 2002). Plant PAS proteins possibly function similarly, or the MEKHLA domain might provide a docking structure to recruit other proteins into a transcriptionally functional complex. Although plant HD-ZIP proteins efficiently bind DNA as homo- or heterodimers(Sessa et al., 1993; Johannesson et al., 2001), it can be envisaged that higher order protein complexes containing DRN and/or DRNL and possibly other proteins are formed and co-ordinately act as a transcriptional unit in the control of embryo patterning. In vitro CoIP experiments demonstrate heterodimerisation between full-length DRN and PHV proteins, and BiFC experiments performed with the PHV C-terminus and full-length DRN further confirm the interaction in planta. PHV and PHB are involved in adaxial/abaxial leaf patterning(McConnell et al., 2001; Emery et al., 2003) and are also expressed in the SAM and during early stages of leaf development where DRN remains active postembryonically(McConnell et al., 2001; Kirch et al., 2003),suggesting that class III HD-ZIP proteins and AP2-class transcription factors such as DRN or DRNL might be partners throughout the plant life cycle.

We have supported DRN-PHV protein interaction data with genetic data showing a combined genetic effect of PHV and DRN in embryonic patterning: enhanced embryo cell division defects of the drn-1 phv double mutant show that both genes contribute to the same embryonic patterning pathways. This might reflect partial redundancy between both gene functions; but if DRN and PHV act in a common protein complex, and as both DRN and DRNL are functionally redundant and can interact with other partially redundant class III HD-ZIP family members, it is more likely to reflect redundancy involving other heterodimer combinations between different members of both protein families.

Our finding that the DRN and DRNL AP2 domain alone is sufficient to mediate heterodimerisation with PHV is to the best of our knowledge the first experimental evidence that plant AP2 domains have a role in protein dimerisation as well as in DNA binding. AP2 proteins are plant-specific and with 144 AP2/ERF members, comprise one of the largest transcription factor families in Arabidopsis (Sakuma et al., 2002). They are key regulators in diverse developmental processes such as flower formation [AP2(Jofuku et al., 1994)], ovule development [AINTEGUMENTA (Elliott et al.,1996; Klucher et al.,1996)] and abiotic stress [TaDREB1(Shen et al., 2003)]. Both the tandem repeat unit and the single AP2 domain comprise functional DNA-binding motifs. The cysteine and serine residues in the DRNL and DRN AP2 domains,unique among Arabidopsis AP2 domains, reside within the coreα-helix of the RAYD element, which has been proposed to mediate protein-protein interactions (Okamuro et al., 1997). Two alternative effects of dimerisation which would affect the regulation of target genes are that either the interaction between DRN or DRNL and class III HD ZIP proteins sterically interferes with the DNA-binding activity of the AP2 domain, or it contributes to the DNA-binding specificity/affinity of the AP2 domain, as is known for protein-DNA interactions mediated by the homeodomain(Moens and Selleri, 2006). Based on overlapping transcription patterns, a heterodimeric complex between DRN-PHV or DRNL-PHV or related HD-ZIP III proteins could control the transcription of target genes required for normal cotyledon development. PHV and DRN do not promiscuously interact with other proteins, such as STM, that are co-expressed in the early embryo, supporting the specificity of complexes involving DRN and PHV.

We show in this paper that two Arabidopsis AP2-domain containing paralogues control embryo development in specific and early embryo expression domains. Both proteins can form protein-protein interactions with class III HD-ZIP proteins and with PHV via a PAS-like domain in the C-terminal region of PHV and PHB, and the AP2 domain of DRN and DRNL. Our data suggest that transcriptional complexes involving DRN and DRNL act redundantly with their class III HD-ZIP partners to control embryo organogenesis and patterning. The robustness of embryonic patterning as suggested by the low penetrance of mutant phenotypes therefore finds a biochemical basis in the promiscuity of transcription factor interactions. The identification of DRN and DRNL as partners of class III HD-ZIP proteins therefore enables biochemical access to signal transduction cascades in the embryo on the basis of genetic pathways.

We thank Thomas Jack, Anwesha Nag and Yingzhen Yang for the drnl-2allele prior to publication; Hans Sommer for the library for two-hybrid screening; Klaus Harter for providing vectors for BiFC; and J. Friml for the DR5::GFPer and PIN1::PIN-GFP constructs. This project was funded by the Deutsche Forschungsgemeinschaft through SFB 572.