A contradictory GLABRA3 allele helps define gene interactions controlling trichome development in Arabidopsis

Arabidopsis leaf trichomes are unicellular, and are composed of a stalk and two to four branches. Trichome development can easily be observed on young leaves where the mature trichomes are found at the tip and in various stages of development toward the base(Hülskamp et al., 1994; Szymanski et al., 1998). Trichome development can be divided into six stages(Szymanski et al., 1999). At stage one an epidermal cell ceases to divide and undergoes radial expansion. During subsequent stages developing trichomes expand out of the leaf surface and produce two to four branches. They then cease to expand and form a thick cell wall containing surface papillae during the sixth and final phase. There is a pattern to trichome development in that trichomes rarely develop next to one another and this pattern appears to be controlled by cell-cell communication (Larkin et al.,1996). The development of leaf trichomes in Arabidopsisis being used to address questions concerning cell fate specification, pattern formation and cellular differentiation(Hülskamp et al., 1998; Marks et al., 1991; Szymanski et al., 2000).

Genetic analyses have identified three genes that are involved in promoting trichome initiation: GLABROUS1 (GL1), a R2R3-Myb-type transcription factor; TRANSPARENT TESTA GLABRA1 (TTG1), a protein with WD-40 repeats; and GLABRA3 (GL3), a bHLH type transcription factor (Oppenheimer et al.,1991; Payne et al.,2000; Walker et al.,1999). Severe loss-of-function mutations in either GL1 or TTG1 result in greatly reduced trichome initiation. Mutations in GL3 moderately reduce trichome initiation and result in smaller trichomes that have fewer branches(Koornneef et al., 1982; Payne et al., 2000). The Arabidopsis genome contains another gene closely related to GL3,ENHANCER OF GLABRA3 (EGL3). Plants containing mutations in both GL3 and EGL3 display the greatly reduced trichome initiation phenotype that is exhibited in gl1 and ttg1 mutants(Zhang et al., 2003). Mutations in EGL3 alone only slightly alter trichome initiation and development. Unlike gl1 mutants, which only affect trichome development, ttg1 and gl3 egl3 mutants display additional defects that include the production of extra root hairs and the loss of both pigmentation and seed coat mucilage production(Galway et al., 1994; Koornneef, 1981; Zhang et al., 2003).

The expression patterns of GL1 and GL3 have been characterized. Both genes are diffusely expressed at low levels in fields of epidermal cells on young leaves, and then expressed at higher levels in early stage trichomes (Larkin et al.,1993; Zhang et al.,2003). Additional studies have shown that the GL1 and GL3 proteins probably act in a complex with TTG1. It is thought that a threshold level of the GL1 GL3 TTG1 activator complex is needed to irreversibly push a cell into the trichome pathway (Payne et al.,2000; Schellmann et al.,2002; Szymanski et al.,2000). The predicted consequence of reaching the threshold is the generation of an autoregulatory loop that results in the up-regulation of the complex genes, as well as genes needed for the initial stages of trichome differentiation, including TRYPTICHON (TRY). TRY, which encodes a single-repeat R3 Myb, limits trichome initiation and prevents neighboring cells from becoming trichomes(Hülskamp et al., 1994; Schellmann et al., 2002). In yeast assays, TRY physically interacts with GL3 and may function to reduce the ability of the GL1 GL3 TTG1 activation complex to activate gene expression in developing trichomes and those cells that neighbor trichomes(Zhang et al., 2003). To mediate this latter role, it has been proposed that once translated in developing trichomes, TRY protein diffuses through the connecting plasmodesmata to neighboring cells(Schellmann et al., 2002).

A model similar to the trichome initiation model has been proposed for root hair (H-cell) development where the WER, TTG1 and GL3/EGL3 proteins form an initiation complex to promote the N-cell fate (non hair)(Schiefelbein, 2003; Zhang et al., 2003). The CAPRICE (CPC) gene, which encodes a single-repeat R3 MYB homologous to TRY and is expressed in N-cells, acts as a negative regulator of the N-cell fate (Lee and Schiefelbein, 2002; Wada et al., 1997). CPC protein has been shown to be able to travel to H-cells and to down-regulate the expression of N-cell genes like WERin H-cells (Lee and Schiefelbein,2002; Wada et al.,2002).

This paper presents the characterization of a new allele of GL3.In contrast to previously described gl3 mutants, the trichomes of the new mutant display greatly increased levels of endoreduplication and over expand early during development. This novel contradictory gl3phenotype has provided a new insight into trichome development.

The gl1 mutant used in this study was isolated from the Weigel Activation tagged T-DNA lines (Weigel et al., 2000). The try and egl3 mutants were identified using the SIGnAL website(http://signal.salk.edu)and were derived from SALK lines 29760 and 77439, respectively. gl3-1was obtained from Dr Koornneef in a Ler background and was backcrossed three times into a Col background before being used in these analyses. Plants were grown on soil (LG3 and LP5, Sungrow Horticulture,Bellevue, WA) at 22°C and constant illumination of 80μE/m²/second.

Bulk segregation analysis of an F₂ population derived from a cross between Col sst and Ler was used to determine the general chromosomal location of sst(Lukowitz et al., 2000). To more finely map the chromosomal location of sst, DNA samples from 302 sst F₂ plants were tested with PhyC, CiW9 and CiW10 SSLP marker primer pairs (Research Genetices). Data from these reactions placed the sst locus between PhyC and CiW9 on chromosome 5. Additional SSLP markers were created utilizing information from the insertion/deletion (InDel)polymorphism collection generated by Cereon Genomics (now Monsanto Co.)(Jander et al., 2002). InDels 454480 and 457742 were ultimately used to limit the sst locus to either TAC clone K1O13 or P1 clone Myc6. Because the Myc6 clone contained the candidate GL3 gene, sst was tested for allelism to GL3.

The gl3-1 mutation causes a C to T transition at codon 378, which results in the loss of an AciI site. Using forward 5′-TCAGTACGGAGCCTTTTCCAACAGC-3′ and reverse 5′-CTTTAACATTCCTTGTGATGGTGCC-3′ primers, a 509 bp genomic fragment was amplified using the Extract-N-Amp PCR system (Sigma). Digesting the samples with AciI generated fragments of 174 bp and 73 bp for Col compared with a 247 bp fragment for gl3-1.

The gl3-sst mutation causes a C to T transition at codon 78 leading to the elimination of a DdeI site. Using forward 5′-GCGAATTCGCCATGGCTACCGGACAAAACAGAAC-3′ and reverse 5′-GCGGAAGCTCGTCTGGTGAC-3′ primers, a 661 bp fragment was generated by PCR. Digesting the samples with DdeI generated fragments of 460 bp and 89 bp for Col compared with a 549 bp fragment for gl3-sst.

SALK line 77439 contains a T-DNA insert in the third exon of At1g63650. To distinguish between the insertional mutant and wild loci two sets of primers were used. One set of primers (forward 5′-CCCCGGAGGAGCGTTATCCAATGGA-3′ and reverse 5′-CCCTTAAGTGACGATAAATACACTCCG-3′) was used to detect the wild-type gene and another set of primers (forward 5′-GCGTGGACCGCTTGCTGCAACT-3′ and reverse 5′-CCCTTAAGTGACGATAAATACACTCCG-3′) was used to detect the T-DNA insertion.

The plant specimens were frozen in liquid nitrogen and processed with an Emitech K1150 cryo-preparation system where the specimens were sputter coated with 20 nm gold and imaged using a cold stage with a Hitachi S3500N scanning electron microscope. For X-ray elemental analysis, the samples were sputter coated with 20 nm of nickel and analyzed with the EDAX, Inc. Falcon System.

The p928 LacOP construct, which contains 256 repeats of the Lac Operator sequence, was generated by moving a 12.5 kb fragment containing the repeated sequence from pSV2-dhfr 8.32 (Robinett et al., 1996) into the T-DNA vector p928 (gift from Dr Mitra,University of Nebraska).

The coding sequence of a modified LacI repressor protein and a C-terminal SV40 nuclear localization sequence used for the pEGAD LacI construct originated from a 1276 bp EcoRI-DraI fragment from p3′SSdimer-Cl-EGFP (Robinett et al.,1996). This fragment was cloned into the EcoRI and blunt-ended HindIII sites of the 35S::GFP-containing T-DNA vector pEGAD (Cutler et al., 2000) to create plasmid pEGAD LacI.

The promoter regions from MYB5 (forward 5′-GAGCTCTGCTGGAGAAATTCATCCCAA-3′ and reverse 5′-ACCGGTCTCCGCCGTCTTCAACAAAGC-3′), ATML1 (forward 5′-GAGCTCGATCCATAGTTTCTAAAATGTGC-3′ and reverse 5′-ACCGGTGGATTCAGGGAGTTTCTTTAACC-3′), TRY (forward 5′-AAGCTTGAGCTCGTTAGTTTAATGGGTCGA-3′ and reverse 5′-GGATCCACCGGTGATGAAATTTTGAGTTTGG-3′) were amplified from Columbia genomic DNA using JumpStart Taq Redimix (Sigma) and cloned into pCR2.1 using the Topo-TA cloning system (Invitrogen). Each promoter was then excised as a SacI-AgeI fragment and was swapped with the 35S promoter of pEGAD (Cutler et al.,2000). pEMYB5::GFP-cGL3 was generated by excising the coding region of GL3 from pD2CD-7 (a gift from Dr Lloyd, University of Texas) as an EcoRI-BamHI fragment and cloning into the corresponding sites of pEGAD MYB5 Pro. pEATML1::GFP-cGL1 was generated by excising the coding region of GL1 from pGL1-A(a gift from Dr Lloyd) with EcoRI and BglII and cloning it into the corresponding sites of pEGAD ATML Pro. pETRY::GFP-cTRY and pEMYB5::GFP-cTRY were generated by excising the coding region of TRY from pCR-TRY-BD as an EcoRI/BamHI fragment and cloning it into the appropriate sites of pEGAD TRY Proand pEGAD MYB5 Pro, respectively

The binary constructs were introduced into Agrobacterium tumefaciens C58C1 by electroporation. The Arabidopsis Columbia strains were transformed with Agrobacterium by dipping and selected on soil with Basta (Clough and Bent,1998). GFP flourescence was detected in mature trichomes using a Nikon Eclipse E800 microscope with a Cool Cam color CCD camera (Cool Camera Co., Decatur, GA) and Imago Pro Plus version 3.0 software (Media Cybermetics,Silver Spring, MD).

Mature leaf tissue from Columbia wild-type, gl3-1 and gl3-sst was stained with DAPI (4′6-diamidino-2-phenylindole;Sigma) as previously described (Szymanski and Marks, 1998). The staining was visualized with a Nikon Diaphot-200 inverted microscope with DIC optics and a Kodak MDS290 photo capture system.

pGL3-AD was constructed by cloning an EcoRI/BamHI fragment containing the GL3 coding region from pD2DC-7 [(Payne et al.,2000) a gift from Dr Lloyd] into the EcoRI and BglII sites of pGAD424 (Clontech). psst-AD was generated by swapping a 702 bp BamHI/BglII fragment from pCR-sst-N with the corresponding fragment from pGL3-AD. The GL3 coding region from the gl3-sst mutant(pCR-sst-N) was generated by PCR amplifying a 1.08 kp fragment from gl3-sst cDNA using the primers, forward 5′-GCGAATTCGCCATGGCTACCGGACAAAACAGAAC-3′ and reverse 5′-CGGATCAAGAACGTTGTCGATGTG-3′. The fragment was cloned into pCR2.1 (Invitrogen) and the presence of only the gl3-sst specific C to T change at positon 235 was confirmed by DNA sequencing. The coding regions for GL1, TTG1 and TRY used to make GAL4 binding domain fusions originated from pGL1-B, pWS10 and pCR-TRY-BD. Both pGL1-B and pWS10 were gifts from Dr Lloyd(Payne et al., 2000), whereas pCR-TRY-BD was a PCR-generated TRY coding region from Columbia cDNA (forward 5′-GAATTCGCCATGGATAACACTGACCGTCGT-3′ and reverse 5′-GGATCCCTAGGAAGGATAGATAGAAAAGCG-3′) that was cloned into pCR2.1 (Invitrogen) and verified by DNA sequencing. The pGL1-BD,pTTG1-BD and pTRY-BD were generated by cloning a 0.62 kb EcoRI/PstI fragment, 1 kb EcoRI/SalI fragment and 0.32 kb EcoRI/BamHI fragment, respectively,into the corresponding sites of pBridge (Clontech).

The construct used for the GL1-TRY competition assay,pGL1-BD/TRY-free, was generated by cloning a 0.32 kb NotI/BamHI fragment from pCR-TRY-FR into the NotI and BglII sites of pGL1-BD. pCR-TRY-FR was generated by cloning a PCR-generated TRYcoding region from Columbia cDNA (forward 5′-GCGGCCGCAGCCATGGATAACACTGACCG-3′ and reverse 5′-GGATCCCTAGGAAGGATAGATAGAAAAGCG-3′) into pCR2.1(Invitrogen).

The appropriate pGAD424- and pBridge-based constructs were transformed sequentially into the yeast strain Y190 using electroporation. The cells were selected on plates containing SD synthetic medium (2% glucose, 1× yeast nitrogen base) lacking first only Leu, then Leu and Trp. Liquid cultures of SD synthetic medium lacking Leu and Trp were used to measure β-galactosidase(β-gal) activity (Ausubel et al.,1995). The cells were grown to an OD₆₀₀=0.7-1.0,pelleted by centrifugation and suspended in z-buffer (60 mM Na₂HPO₄, 40 mM NaH₂PO₄, 10 mM KCl,1 mM MgSO₄, 50 mM β-mercaptoethanol, pH 7.0). The cells were permeabilized by adding a final concentration of 0.005% SDS and 3.5%chloroform (v/v). ONPG (o-nitrophenyl-D-galactopyranoside; Sigma) was added as a substrate. After incubation at 30°C, the reaction was stopped with sodium carbonate and measured for activity with an OD₄₂₀. Units ofβ-gal activity was determine by the equation U=1000×[OD₄₂₀]/time (in seconds) × volume (in ml)× [OD₆₀₀]. For each comparison, multiple independent yeast isolates (n=4-8) were tested multiple times (n=3-6).

A new trichome mutant with a highly pleiotropic trichome phenotype was found in an EMS-treated Col population(Fig. 1B). Trichomes on wild-type leaves typically consist of a stalk and three to four branches(Fig. 1A,D). The trichomes on the leaves of the mutant varied from nearly normal(Fig. 1E) to lacking a normal stalk and exhibiting multiple abnormal branches(Fig. 1F-I), to lacking branches (Fig. 1J). For this reason, the mutant was tentatively named shapeshifter (sst). The sst trichomes also varied in their degree of maturation. During the final phase of development, wild-type trichomes develop papillae on the outer surface of the cell wall (Fig. 1M). While the normal appearing sst trichomes developed a few papillae (Fig. 1O), most mutant trichomes lacked papillae (Fig. 1N). In addition to the morphological defects in trichome development, sst leaves exhibited fewer trichomes that were abnormally clustered (Table 1).

The F₁ plants from crosses between sst and Col shared two trichome abnormalities with homozygous sst mutants. First, there were fewer trichomes per leaf and second, the F₁ trichomes consistently had more branches than wild-type trichomes(Table 1 and Fig. 1C). Neither of the alterations in the F₁ phenotype was as extreme as seen in the homozygous mutant. The segregation analysis of the F₂ population indicates that the severe sst phenotype was recessive and probably caused by a mutation at a single locus (53 sst; 155 wild type + extra branches; fits 1:3, χ²=0.025, P>0.8). These findings were further supported by the analysis of selected F₃populations. It was found that all selected F₂ plants with wild-type trichomes only produced wild-type F₃ plants, sstF₂ plants only produced sst F₃s, whereas the F₂ plants with trichomes containing extra branches gave rise to segregating F₃ populations (data not shown).

The sst mutation was mapped to a chromosomal region containing the candidate gene GL3. To determine if sst represented a new allele of GL3, sst was crossed with the gl3-1 mutant. The F₁ plants had a distinct phenotype as shown in Fig. 1L. The trichomes on these plants tended to be less branched than wild-type trichomes, but the branches were much larger in diameter than gl3-1 mutant trichomes(Fig. 1K). These results suggested that sst is an allele of GL3. To obtain molecular evidence that sst was a new GL3 allele, the GL3locus in the sst mutant was isolated by PCR and sequenced. The GL3 locus of sst contained a single base change of C to T in codon 78, which converts Leu to Phe. As a final indicator that sstwas a new allele of GL3, a green fluorescent protein (GFP)-tagged version of the wild-type GL3 coding region was moved into sst plants (see Materials and methods). This gene fusion was placed under the control of the Arabidopsis MYB5 gene promoter, which has previously been shown to drive gene expression in trichomes(Li et al., 1996). This construct rescued the phenotype resulting from both sst and gl3-1 mutations (Fig. 2). These results demonstrated that sst represented a new GL3 allele, which hereafter will be called gl3-sst.

The gl3-sst trichomes exhibit a dramatic increase in cell expansion during early stages of trichome development(Fig. 3A-F; note differences in bar scales between A and D, B and E, C and F). This resulted in most stage two gl3-sst trichomes being much larger than wild-type trichomes(Fig. 3E). The increased surface area of early stage mutant trichomes appeared to support a larger number of branch initiation sites (Fig. 3F). The degree to which the branches expanded varied greatly(compare Fig. 1F with 1H). While trichome expansion eventually ceased, very few gl3-ssttrichomes developed papillae. In contrast to the excessive expansion seen for gl3-sst trichomes, gl3-1 trichomes expanded less during early stages (Fig. 3G,H).

The elemental composition of Col and gl3-sst trichome cell walls was compared using SEM EDAX, which can determine the major elemental composition of selected surfaces (see Materials and methods). The papillae on Col trichomes contained carbon, oxygen, magnesium, calcium and phosphorus,whereas the smooth wall regions between the papillae were lacking phosphorus(see Supplemental data Fig. S1). For the walls of gl3-sst trichomes that contained papillae, it was found that both the papillae and intervening regions lacked phosphorus and magnesium. The gl3-sst trichomes that lacked papillae also lacked calcium, as well as phosphorus and magnesium. These cell walls were similar in composition to the mature guard and pavement cells surrounding the trichomes on wild-type leaves, which only showed peaks for carbon and oxygen.

The DNA content of mature Col trichomes ranges from 16C to 64C(Hülskamp et al., 1994; Melaragno et al., 1993; Szymanski and Marks, 1998). The analysis of images of DAPI-stained gl3-sst trichome nuclei indicated that the gl3-sst trichomes contain much more DNA than wild-type or gl3-1 trichomes (compare Fig. 4A, B and C, captured at the same magnification and exposure settings). Interestingly, the DAPI-stained gl3-sst mutant nuclei appeared to be composed of numerous lobes collapsed upon one another. To observe the nuclei in living plants, Col and gl3-sst plants expressing the nuclear-localized N7 GFP fusion were used for confocal analysis (Cutler et al.,2000) (see Materials and methods). This analysis showed that wild-type N7 nuclei were single spheres(Fig. 4D), whereas the gl3-sst N7 nuclei were composed of a series of interconnected lobes (Fig. 4E,F).

The gl3-sst trichome nuclear defect was further characterized to determine if each nuclear lobe in gl3-sst trichomes contains a distinct set of chromosomes. This question was addressed using the two component Lac Operator/Repressor-GFP (Lac OP/I-GFP) reporter system that has been used to visualize interphase chromatin(Robinett et al., 1996) (see Materials and methods). Fig. 5shows representative confocal images of Col and gl3-sst nuclei from plants that contained integrated Lac Operator sequences and that expressed the Lac I-GFP fusion protein. To compare cells that undergo different degrees of endoreduplication, the trichome nuclei(Fig. 5A,B) were compared to the nuclei of guard cells and epidermal pavement cells(Fig. 5C,D). All nuclei exhibited a low level of green fluorescence, which allowed the outline of the nuclei to be visualized. The guard cells have been shown to contain 2C DNA(Melaragno et al., 1993) and a single faint spot, corresponding to the binding of Lac I-GFP to the integrated Operator, was seen in most of the guard cell nuclei of Col and gl3-sst leaves (Fig. 5C,D). Epidermal pavement cells typically have a DNA content ranging from 4C to 16C and, as shown in Fig. 5C,D, these cells typically contained brighter spots. Col trichomes contain an average of 32C and, as shown in Fig. 5A, the integrated Lac operator was most often visualized in interphase chromatin as a single intense spot (n=3 independent lines), although occasionally closely clustered bright spots were seen (not shown). Shown in Fig. 5B is an image of a gl3-sst trichome nucleus (n=5 independent lines) with a relatively large focus of fluorescence. Two lines (one gl3-sst and one Col) had trichome nuclei with two intense spots and these lines probably have two Lac OP inserts. Lines expressing Lac I-GFP, but lacking the integrated Lac OP repeat did not have spots. These lines were first generation transformants for the Lac OP repeat, so the copy number of Lac OP units has not been assessed. This analysis also showed that the morphology of non-trichome epidermal cell nuclei in gl3-sst was not affected by the mutation (compare Fig. 5C,D).

It was previously noted that EGL3 is functionally redundant to GL3 and it has been shown that gl3-1 egl3 plants lack trichomes (Payne et al., 2000; Zhang et al., 2003). To determine if EGL3 is needed for the gl3-sst trichome development, an egl3 mutant was crossed to gl3-sst. The egl3 mutant used for this analysis was the previously uncharacterized SALK line 77439, which contained a T-DNA insertion in the third exon of EGL3. Thus, to determine if the T-DNA insertion mutant line behaved in the same matter as previously described egl3 mutants, the T-DNA mutant was also crossed to gl3-1. As previously described for egl3 and gl3-1 interactions, it was found that plants homozygous for the insertion and gl3-1 were glabrous. In the F₂ population of plants derived from the cross between the insertional mutant and gl3-sst, 24 plants had a wild-type phenotype and another 10 closely resembled gl3-sst. Progeny from five of the F₂gl3-sst-like lines were analyzed. One of the five F₃ populations (n=25 plants) consisted solely of plants with a reduced number of gl3-sst-like trichomes. These plants were found to be homozygous for both gl3-sst and the T-DNA insertion (see Materials and methods for genotyping). In addition to the reduced number of leaf trichomes, it was found that the double mutants lacked stem trichomes(see Supplemental data Fig. S2).

The mutation responsible for gl3-sst allele is in a region that encodes a protein domain which mediates interaction with GL1 and TRY proteins(Payne et al., 2000; Zhang et al., 2003). To determine if the mutation affected protein-protein interactions, the yeast two-hybrid system was used to compare the interactions of gl3-sst and GL3 proteins with GL1, TTG1 and TRY. To study the interactions, the coding regions of the GL3 and gl3-sst were fused to the activation domain(AD) of GAL4, whereas GL1, TRY and TTG1 were fused to the binding domain (BD) of GAL4. Yeast containing either empty pGAD424 (AD) or pBridge (BD) vectors in conjunction with any of the corresponding trichome protein fusions did not exhibit significant β-gal activity (data not shown), whereas yeast containing either sst-AD or GL3-AD and GL1-BD, TRY-BD or TTG1-BD exhibited β-gal activity (Fig. 6A-C). However, there was a significant difference in activity level between the strains containing either sst-AD or GL3-AD. Yeast isolates containing GL1-BD and GL3-AD exhibited approximately fourfold higher levels ofβ-gal activity than isolates containing GL1-BD and sst-AD(Fig. 6A). Yeast containing TTG1-BD and GL3-AD exhibited approximately a twofold higher level ofβ-gal activity than isolates containing TTG1-BD and sst-AD(Fig. 6B). Finally, it was found that TRY-BD interacted equally well with either GL3-AD or sst-AD(Fig. 6C).

In a previous model, it has been posited that TRY and GL1 compete for the same binding site on the GL3 protein to form alternate types of complexes(Marks and Esch, 2003). To test this model and to determine if the mutation in gl3-sst alters the proposed competitive interaction between TRY and GL1, the pBridge system(Clontech) was used to create a competitive yeast interaction assay. In addition to expressing a GAL4 binding site fusion protein, pBridge can also express a second protein (free), which is under the control of a methionine repressible promoter and does not contain either a BD or AD domain. To determine if TRY can compete with GL1 for a binding site on GL3, TRYwas cloned into the free site of the GL1-BD construct to create the plasmid GL1-BD/TRY-free. In the absence of methionine, where TRY-free expression would be at maximum levels, it was found that yeast containing GL3-AD or sst-AD and GL1-BD/TRY-free exhibited very little β-gal activity(Fig. 6D). As the concentration of methionine was increased, which should depress TRY-free expression, the levels of β-gal activity increased(Fig. 6D). The differences in the levels of β-gal activity were not due to the presence or absence of methionine, as yeast containing either GL3-AD or sst-AD and GL1-BD were able to express β-gal to the same level in the presence or absence of 1 mM methionine (data not shown). These results are consistent with the model,which predicts that free TRY protein should be able to inhibit the interaction between GL3-AD and GL1-BD. Furthermore, with the reduction in GL1 GL3 interactions in the gl3-sst plants, it would take less TRY protein to reduce the level of the activation complex.

To corroborate the yeast interaction data, the cellular location of the GL3, GL1 and TRY proteins was ascertained. This analysis made use of the GFP-GL3 fusion construct that was used to rescue the gl3-1and gl3-sst mutants. In addition, GFP fusions with GL1 (GFP-GL1) and TRY (GFP-TRY) were generated (see Materials and methods). The L1 layer-specific promoter from ATML1, which is upregulated in developing trichomes, was used to drive the expression of GFP-GL1 and the TRY promoter was used to express the GFP-TRY fusion(Schellmann et al., 2002; Sessions et al., 1999). Both of these constructs were able to rescue their respective mutants (see supplemental Fig. S3). Strong gl1 mutants typically lack all adaxial leaf surface trichomes. The expression of the GFP-GL1 construct in the gl1background restored the development of wild-type-like trichomes. trymutants typically have trichomes that are extra branched, and 10-15% of try trichomes develop next to one another. The GFP-TRYconstruct was able to rescue both of these phenotypes. As shown in Fig. 7A-I, all of the fusion proteins can be clearly detected in the trichome nuclei. The filter set used to collect the images limits the fluorescence emission to the green range. At the higher gain and exposure setting on the CCD camera, the primarily red chlorophyll autofluorescence was still detected in the green range. This background autofluoresecence was detected in the mesophyll cells of all samples at the higher gain and exposure settings. In addition, a low background of green autofluorescence was emitted from control Col wild-type trichome cell walls at the higher settings. Importantly, Col wild-type trichome nuclei lacked detectable fluorescence(Fig. 7J-L). These results place TRY, GL3 and GL1 proteins in the trichome nuclei, where they could interact as observed in the yeast system.

To determine if increased TRY expression could alter the gl3-sst phenotype, mutant plants were transformed with a GFP-TRY construct under the control of the MYB5 promoter,which drives expression in trichomes (see Materials and methods). This construct has the ability to rescue the phenotype of try mutant plants (data not shown). The phenotype of gl3-sst plants containing the TRY gene was modified. Firstly, it reduced the number of leaf trichomes on these plants. On the third leaf, the Col plants had 71±9 trichomes, the gl3-sst mutant had 44±5 trichomes, and the TRY-expressing gl3-sst line had only 26±5 trichomes. Secondly,extra TRY expression greatly reduced the prevalence of clustered trichomes. The percentage of trichome clustering observed on the third leaf was 14.5% for the gl3-sst mutant as compared with 1.5% for the TRY-expressing gl3-sst line. No trichome clusters were detected in the Col plants.

A new recessive trichome mutant with a novel phenotype has been described. The trichomes on the new mutant were highly variable in shape and for this reason, the mutant was named shapeshifter (sst). Genetic mapping of sst placed it in the same chromosomal region as the GL3 gene. Three lines of evidence were used to show that sstrepresents a new GL3 allele. First, the analysis of F₁ and F₂ populations derived from a cross between sst and gl3-1 plants failed to produce any wild-type plants. Second,sequencing of the sst GL3 locus revealed a base substitution. Third,a wild-type GL3 transgene was able to rescue the sstphenotype.

The gl3-sst mutant is phenotypically distinct from the other loss-of-function gl3 mutants. The trichomes on previous gl3mutants typically had more slender stalks and fewer branches than wild-type trichomes (Koornneef et al.,1982; Payne et al.,2000). In contrast, the gl3-sst trichomes have a swollen stalk and vary widely in branch number. Another difference is in the degree to which the nuclear DNA of gl3-1 and gl3-sst trichomes undergoes endoreduplication. Wild-type Col trichome nuclei typically contain a range between 16C and 64C with an average of 32C, while gl3-1trichomes average 16C (Hülskamp et al., 1994; Melaragno et al.,1993; Szymanski and Marks,1998). While the exact levels of nuclear DNA were not determined,it was clear from the intensity of DAPI staining and the increased nuclear size that gl3-sst trichomes exhibited higher than wild-type levels of endoreduplication.

The EGL3 gene, located on chromosome one, is closely related to GL3 (Payne et al.,2000; Zhang et al.,2003). To determine if this gene has a role in the gl3-sst phenotype, the genetic interactions between an EGL3T-DNA mutant and both gl3-1 and gl3-sst were studied. As previously reported, it was found that the trichomes that develop on gl3-1 mutants required EGL3 function(Zhang et al., 2003). Furthermore, it was found that gl3-sst plants produced trichomes in the absence of a functional EGL3 gene. This showed that the gl3-sst allele is functional.

Many of the mutant gl3-sst trichome defects could be due to alterations in the microtubule cytoskeleton. For example, the swollen appearance of early stage gl3-sst trichomes is similar to trichomes on wild-type plants treated with microtubule destabilizing agents such as oryzalin (Mathur and Chua,2000; Szymanski et al.,1999). Likewise, the abnormally lobed structure of gl3-sst trichomes nuclei is similar to nuclei in plants treated with antimicrotubule drugs. In this case, the lobate nuclear structure is thought to result from spindle microtubule disassembly, which results in a disruption of karyokinesis.

The majority of the cell walls of the mature gl3-sst leaf trichomes had a `glassy' appearance when viewed by stereoscopic microscopy. SEM analysis revealed that the walls lacked the papillae that are normally associated with wild-type trichomes. This suggested that most gl3-ssttrichomes failed to undergo a final maturation step. To further analyze these differences, the elemental composition of the trichome cell wall was ascertained. Compared to wild-type trichomes, the gl3-ssttrichomes showed two forms of cell wall immaturity. In gl3-ssttrichomes with a papillate surface, magnesium, which was found throughout the cell wall of wild-type trichome, and phosphorus, which was localized within the wild-type trichome papillae, were both absent. In papillae-less gl3-sst trichomes, calcium, in addition to magnesium and phosphorus, was missing from the cell wall, which suggested that the deposition of these elements play an important role in trichome cell wall maturation.

Extra endoreduplication occurred in gl3-sst trichome nuclei. In both Col wild-type and gl3-sst trichomes, the Lac OP/I-GFP reporter experiments suggested that the chromatids of endoreduplicated chromosomes are tightly associated with one another. Comparable results have also been reported using a similar version of the Lac OP/I-GFP system to study the nature of chromosomes in pavements cells(Kato and Lam, 2001; Kato and Lam, 2003). However,Kato and Lam (Kato and Lam,2003) reported the presence of multiple Lac I-GFP binding sites in pavement cells, which is contradictory to the idea that the sister chromatids are associated with one another as in polytene chromosomes. In the present study, only single binding sites were observed in most nuclei, whether from trichome, pavement, or guard cells. Overall our results suggest that the chromatids of endoreduplicated chromosomes are highly associated with one another (or are polytene). Furthermore, the results provided evidence that the highly lobed gl3-sst nuclei contain single sets of chromosomes that have not undergone karyokinesis.

It was previously shown that GL1 and TTG1 can interact with GL3 in yeast two-hybrid assays (Payne et al.,2000). Thus, the interactions between either GL1 or TTG1 with either GL3 or gl3-sst were compared. The location of the mutation in gl3-sst in the 78th codon predicted that gl3-sst might display an aberrant interaction with GL1, and indeed, this result was seen. The mutation resulted in a 75%reduction in the interaction between gl3-sst and GL1, based upon reproducible quantitative beta-gal assays. In addition, it was found that the gl3-sst interaction with TTG1 was reduced by 50%. Since GL1, GL3 and TTG1 proteins have been proposed to form an activating complex that regulates genes needed for trichome development, it is likely that the gl3-sst phenotype is due to a reduction in this complex.

A model based on the dynamic interactions between GL1, GL3 and TTG1 activating complex has previously been presented(Szymanski et al., 2000). The model proposed that the loss of trichomes on plants that overexpress the CPC gene were due to the ability of CPC to inhibit the physical interaction between GL3 and GL1. Because it has subsequently been shown that TRYencodes a CPC-like protein, the model was modified to propose that TRY functions to inhibit the physical interaction between GL3 and GL1(Marks and Esch, 2003). In this paper, evidence is presented that supports this model. Competitive yeast interaction assays demonstrated that TRY can interact with GL3 and that this interaction can prevent the GL3-GL1 interaction. Interestingly, the mutation in gl3-sst does not influence the interaction between gl3-sst and TRY.

TRY has been found to be expressed in young trichomes and it has been predicted that TRY protein can migrate from initiating trichomes into neighboring cells (Schellmann et al.,2002). Given the interaction data described, it is likely TRY would be able to disrupt any residual GL1 GL3 complexes in neighboring cells and that may be the mechanism by which TRY acts to inhibit the initiation of neighboring trichomes.

Further support for TRY movement comes from the analysis of CPC. It has recently been shown that CPC protein is produced in the non root hair cells and can move into the adjacent files of cells that become root hairs(Wada et al., 2002). In the root, CPC is probably inhibiting the formation of a complex between WER and GL3/EGL3. Direct movement of TRY could not be detected in this study, because the level of fluorescence from the GFP-TRY fusion was below the background autofluorescence emitted by chlorophyll in the mesophyll. Given the expression pattern of TRY and the recent CPC findings, the movement of TRY into cells neighboring developing trichomes would not be unexpected.

It has been posited that a cell enters the trichome pathway only after a critical threshold concentration of GL1 GL3 TTG1 complex has accumulated(Szymanski et al., 2000). Once the threshold is reached, it is predicted that an auto regulatory loop is generated that results in the up-regulation of the complex genes. Over time,as the level of the activator increases, it is predicted that different genes needed for trichome differentiation would be induced. In order for trichome initiation to proceed, it is predicted that genes needed to positively influence trichome development would be expressed first and then genes needed either to modulate or inhibit trichome development would be expressed later. Genes activated early during trichome initiation, when the activator complex would be relatively low, would be induced because they have a higher affinity for the complex than genes that are activated later.

In terms of this threshold model it is interesting to note that different promoters can be used to functionally drive the activator genes. For example,it may seem surprising that the MYB5::cGL3 construct could rescue the gl3-1 mutant trichome phenotype. However, the MYB5 promoter is upregulated in developing trichomes and gl3-1 mutants do initiate trichomes, albeit undersized ones (Li et al., 1996). Thus, GL3 protein would be expressed in the trichomes of gl3-1 plants containing the MYB5::cGL3 construct. Given that native MYB5 expression increases as trichomes develop, it is probable that the levels of GL3 protein would increase. This increase would allow the sequential activation of positive and negative regulators of trichome development as predicted by the threshold model. The ability of the ATML1::cGL1 construct to rescue the gl1 mutant appears to be further complicated by the fact that gl1 mutants do not initiate trichomes. The ATML1 promoter is expressed at a low level in all the epidermal cells of the developing leaf and in situ analyses have shown that ATML1 expression is enhanced in young developing trichomes(Sessions et al., 1999). In terms of the threshold model it is possible that the ATML1 promoter provides a threshold level of GL1 protein in all the epidermal cells. However,only the few cells that achieve the threshold level of GL3 enter the trichome pathway. Once a cell enters the pathway the levels of ATML1expression would be enhanced and the level of GL1 protein would increase,again allowing the sequential activation of genes as predicted by the threshold model.

Trichome development in gl3-sst provides some additional evidence for the threshold model. This support comes from two aspects of the gl3-sst phenotype, which may seem paradoxical. First, there are fewer trichomes on the leaves of gl3-sst plants, which could indicate increased trichome inhibition. In contrast, there are more trichome clusters,which suggest that trichome inhibition is relaxed. Given the reduced interaction between gl3-sst and both GL1 and TTG1, the reduction in trichome number could be the result of fewer cells reaching a critical concentration of activation complex. The reduced interaction between gl3-sst and GL1 or TTG1 could also result in increased trichome clustering if the inhibition signal emitted from gl3-sst trichomes was reduced. This latter possibility is supported by the observation that increased TRY expression in gl3-sst can prevent trichome clustering. Other aspects of the gl3-sst phenotype such as excess expansion and extra endoreduplication could be due to the unregulated activation of genes needed to mediate these cellular phenomenon during early trichome development. However, levels of activator complex needed to stimulate the expression of genes that either modulate or inhibit these early processes may not be reached in gl3-sst trichomes. Layered on top of this relatively simple threshold model, there is probably an added level of complexity. The activator complex possibly does not consist of just GL1, GL3/EGL3 and TTG1, but contains additional components as well. It is more likely that gene regulation in developing trichomes is controlled by both the quantity of activator complex and its composition. Thus, some aspects of the gl3-sst phenotype could result from aberrant ratios of complexes containing or lacking GL1, TTG1 or other components.

While the pleiotropic nature of the gl3-sst mutant trichomes may appear confusing, this very feature provides an excellent resource for studying all aspects of trichome development. These mutants are deficient in trichome initiation, trichome cell patterning, cell expansion,endoreduplication and maturation. Through future genetic screens for suppressors and enhancers of the various components of the phenotype and through detailed expression profiling of the mutant, the understanding of many processes involved in cellular differentiation will be advanced.

We would like to thank Dr Andrew Belmont (University of Illinois) for providing the plasmids containing the Lac Operator and Lac I sequences that were used to build the Lac OP/I-GFP constructs used in this study. We thank the Salk Institute Genomic Analysis Laboratory for providing the sequence-indexed Arabidopsis T-DNA insertion mutants and ABRC for distributing the seeds. We thank Dr David Ehrhardt and ABRC for seeds containing the N7 GFP fusion and the plasmid pEGAD. We thank Drs John Larkin and Alan Lloyd for useful discussions and material support for this work. We thank the University of Minnesota College of Biological Sciences Imaging Center Facility for excellent technical support. M.H., S.N., B.I. and J.N. were supported in part by REU supplements to IBN0091052 awarded to M.D.M.