Functional diversification of MYB23 and GL1 genes in trichome morphogenesis and initiation

One important evolutionary mechanism that creates new gene functions is the duplication and functional diversification of genes. In principle,diversification of duplicated genes can take place at the level of transcriptional regulation or by a diversification of protein function. The relative contribution of these two possibilities to diversification of duplicated genes is an important issue in evolutional biology.

A particularly attractive solution to addressing the mechanisms of gene diversification in evolution is provided by the large family of MYB-related R2R3 transcription factors in plants (Jin and Martin, 1999; Stracke et al., 2001). It is estimated that this family has rapidly evolved within the last 500 million years after the divergence of the vascular plants from the bryophytes (Rabinowicz et al.,1999). It has been suggested that the amplification of the R2R3 MYB genes occurred in conjunction with the development of new plant-specific cellular functions (Martin and Paz-Ares,1997) and that the role of MYB genes is to provide plasticity to plant metabolism and development (Romero et al., 1998). This view is supported by the findings that members of the R2R3 MYB-family are involved in many different biological processes,including the trichome and root-hair differentiation(Lee and Schiefelbein, 1999; Oppenheimer et al., 1991),cell-shape determination (Noda et al.,1994), regulation of leaf form(Waites et al., 1998), control of secondary metabolism (Mol et al.,1998), pathogen response (Yang and Klessig, 1996), drought stress response(Urao et al., 1993),protection from UV radiation (Jin et al.,2000) and hormone signalling(Gubler et al., 1995).

A particularly well-suited model for studying the diversification of duplicated genes is the sub-class of the three paralogous R2R3 MYB genes, WEREWOLF (WER), GLABRA1 (GL1) and AtMYB23 (MYB23). WER and GL1 act in two different functionally non-overlapping developmental processes. WER is important for root-hair development and GL1 for trichome development on aerial organs (Lee and Schiefelbein,1999; Oppenheimer et al.,1991). WER and GL1 proteins with 57% of sequence identity are functionally equivalent and changes in the cis-regulatory sequences completely account for the functional diversification of the two proteins(Lee and Schiefelbein, 2001). Overexpression studies and the analysis of the expression pattern suggested that MYB23 could have a similar function to GL1(Kirik et al., 2001). The identification of two myb23 mutants enabled us now to study the function of MYB23 and its functional relationship to GL1. The myb23 single mutants exhibit reduced trichome branching but no obvious effect in trichome initiation. The gl1 myb23 double mutants,however, are devoid of trichomes at the leaf edges, which are not effected in the gl1 single mutants indicating a functional redundancy of GL1 and MYB23. Promoter and protein-coding region swap experiments showed that the two proteins are functionally equivalent with respect to the regulation of trichome initiation, but not with respect to trichome branching. This indicates that changes in the regulation of trichome initiation are evolved at the level of the cis-regulatory regions and that diversification with respect to the regulation of branching evolved at the level of cis-regulatory regions, as well as at the level of altered protein function.

Plants were grown at 24°C with 16 hours of light per day. The isolation of the gl1, ttg1, gl2, gl3, cpc and try mutant alleles used in this study has been described: gl1-1 and gl3-1 (in Landsberg background) (Koornneef et al.,1982); ttg1-1 (in Landsberg background)(Koornneef, 1981); cpc-1 (in Wassilewskija background)(Wada et al., 1997); and try-JC (in the Columbia background)(Larkin et al., 1999). gl2-4AA allele was kindly provided by ZIGIA, MPIZ Cologne. The myb23-1 mutant was isolated from the SALK T-DNA collection (Columbia ecotype), the myb23-2 mutant was isolated from a Wisconsin T-DNA population (Wassilewskija ecotype). Lines homozygous for multiple mutations and/or transgenes were constructed by crossing single mutant or transgenic plants, examining the F2 progeny for putative mutant phenotypes, and confirming the desired genotype in subsequent generations by backcrossing to single mutants and/or PCR-based tests.

Trichomes were analyzed on the first two leaves of soil-grown plants.

The histochemical analysis of plants containing the GUS reporter constructs was performed essentially as described previously(Vroemen et al., 1996).

The DISCUS software package (Carl H. Hilgers-Technisches Büro,Königswinter, Germany) was used to measure the fluorescence intensity of DAPI-stained nuclei. The relative fluorescence units (RFU) were calibrated with wild type trichome nuclei that were previously reported to have an average of 32C (Schnittger et al.,1998; Szymanski and Marks,1998). By defining the average fluorescence intensity of the wild-type trichomes as 32 RFU our RFU values should roughly correspond to C values (Walker et al., 2000). As a control for this procedure, gl3 (16C) and try (64C)mutants were included.

RNA was isolated from rosettes of 2-week-old plants and subjected to RT-PCR, which was essentially performed as described previously(Kirik et al., 2002). MYB23 gene-specific primers were used for RT-PCR (do-s2,5′-AGAAGAATGAGAATGACAAGAG; and do-1, 5′-TACGTCAGTTGGTGTTGCGTGGAC). Amplifications of the translation elongation factor EF1αA4 cDNA (primers EF1α-UP: 5′-ATGCCCCAGGACATCGTGATTTCAT and EF1α-RP:5′-TTGGCGGCACCCTTAGCTGGATCA) were used as a control.

The MYB23::cMYB23 was constructed by fusing the 2032 bp 5′regulatory region (Kirik et al.,2001) with the MYB23 cDNA and cloning the fusion in the pGPTV-BAR vector (Becker et al.,1992).

The GL1 regulatory sequences used in these experiments are identical to the ones published previously(Lee and Schiefelbein, 2001)and include a 1.4 kb 5″ fragment and a 1.8 kb 3′ fragment. The MYB23 regulatory sequences include a 3.1 kb 5′ fragment and a 1.0 kb 3′ fragment. Constructs encoding the GL1 and MYB23 proteins include the entire transcriptional unit of each gene from the start to the stop codons.

In the GL2::GL3 construct a 2.1 kb 5′ GL2regulatory region (Szymanski et al.,1998) drives the expression of the full-length GL3 cDNA. Details of the transgene constructs are available upon request.

RNA in situ hybridisation was essentially performed as described previously(Larkin et al., 1993). Antisense and sense strand digoxigenin-labelled RNA probes were derived from the 450 bp 3′ fragment of the MYB23 cDNA that includes 180 bp of the 3′ UTR. This cDNA region does not comprise the conserved MYB-domain coding sequence and corresponds to the region with the lowest sequence similarity to other related MYB genes.

Fusions with the GAL4 activation domain and GAL4 DNA-binding domain were performed in the pACT and pAS plasmids (Clontech). TRY, GL3 and N-terminal truncation of GL3 (GL3-96 aa) were fused to the GAL4 activation domain in the pACT plasmid. For the GL1 and MYB23 fusions with GAL4 DNA-binding domain in the pAS vector, we used truncated fragments missing 27 amino acids and 25 amino acids at the C terminus respectively. All used constructs and empty vectors did not show any self activation in yeasts. The interaction strength was determined by measuring the activity of the lacZ reporter gene using the ONPG assay (Clontech).

Previous studies have indicated that the MYB23 gene is involved in the regulation of trichome development(Kirik et al., 2001). In order to study the function of the MYB23 gene, we have screened the available T-DNA insertion lines. Two alleles of the myb23 mutant were isolated. The myb23-1 allele was isolated from the SALK T-DNA collection (Columbia ecotype) and has an insertion in the second intron of the gene (Fig. 1A). The myb23-2 allele was isolated from the Wisconsin T-DNA insertion library (Wassilewskija ecotype) and has a T-DNA insertion 4 bp after the stop codon. RT-PCR analysis revealed that the MYB23 expression is not detectable in the myb23-1 allele and it is strongly reduced in the myb23-2 allele (Fig. 1B).

When compared with the corresponding wild types the myb23-1 and myb23-2 alleles displayed reduced trichome branching(Fig. 1C,D; Table 1). To prove that this phenotype is specific for the mutations in MYB23 we transformed myb23-1 plants with the MYB23::MYB23 construct containing the MYB23 protein-coding region, placed under the control of a 3.1 kb fragment 5′ to a start codon and a 1.0 kb fragment 3′ to the stop codon. This construct rescued the number of trichome branches to wild-type levels showing that the MYB23 gene is required for proper trichome branching(Table 1).

As overexpression of the MYB23 gene causes ectopic trichome initiation (Kirik et al.,2001), we also analyzed epidermal patterning in shoots. Both myb23-1 and myb23-2 alleles displayed wild-type pattern of the trichomes, indicating that the MYB23 gene does not have an essential function in trichome patterning.

The MYB23 and GL1 proteins share 63% of identical amino acid residues,which suggests that these two genes may have redundant or overlapping functions (Kirik et al.,2001). Plants that harbour a null gl1-1 mutant allele,resulted from a deletion of the entire GL1 protein-coding region, are not completely glabrous, a small number of trichomes develop at the edges of late rosette leaves and petioles (Koornneef et al., 1982; Oppenheimer et al.,1991) (Fig. 2A). We did not find any changes in trichome production at the leaf edges of the myb23 mutants (data not shown). To test whether MYB23function at the edges of leaves and petioles is masked by genetic redundancy,we created the gl1 myb23-2 double mutant. The double mutant plants were completely glabrous (Fig. 2A). The introduction of the MYB23::MYB23 construct in the double mutant resulted in a rescue of the leaf edge and petiole trichomes(Fig. 2B). Thus, the gl1 myb23-2 double mutant uncovers the redundancy of MYB23 and GL1 in trichome initiation at the edges and petioles of the rosette leaves.

High sequence similarity of the MYB23 and GL1 genes and similar expression patterns (Kirik et al.,2001) raised the question of why gl1 mutants are not rescued by the wild-type copy of MYB23. To what extend are differences in the transcriptional control or in the protein functions responsible for the redundancy? To address these questions, we exchanged the regulatory and protein-coding regions of the GL1 and MYB23genes.

The finding that the above-described MYB23::MYB23 construct rescues the myb23 mutant phenotype indicates that the 3.1 kb of 5′ regulatory sequence and 1 kb of 3′ sequence contain all regulatory sequences necessary for normal MYB23 gene function. For the GL1 gene, we used 1.4 kb of the 5′ regulatory region and 1.8 kb of the 3′ region that was previously used in a complementation construct (Lee and Schiefelbein,2001). As a reference for the subsequent experiments, the GL1::GL1 construct and the MYB23::MYB23 constructs were introduced into the gl1 and the gl1 myb23-2 backgrounds. Whereas the GL1::GL1 construct rescued completely the gl1mutant phenotype, the MYB23::MYB23 construct showed no effect in gl1 mutants but induced leaf edge and petiole trichomes development in the double mutant (Fig. 2B). To test whether MYB23 protein is functionally equivalent to the GL1 protein,we transformed gl1 mutant and the gl1 myb23-2 double mutant with the GL1::MYB23 and the MYB23::GL1 constructs. The GL1::MYB23 gl1 and GL1::MYB23 gl1 myb23-2 transgenic plants showed a wild-type trichome number and pattern(Fig. 2B), indicating that the MYB23 protein posses the same biochemical activities necessary for trichome initiation as the GL1 protein. No effect of the MYB23:GL1 construct was found in gl1 plants but gl1 myb23-2 plants were rescued back to the gl1 phenotype (Fig. 2B). Together, these data demonstrate that during trichome initiation the GL1 and MYB23 proteins are functionally interchangeable and difference in function of these genes during trichome patterning is due to differences in the transcriptional regulation.

Reduced trichome branching of the myb23 mutants demonstrates that MYB23 function is necessary for the proper trichome cell morphogenesis. Can GL1 protein substitute this morphogenetic function of the MYB23 protein? We tested this by introducing the MYB23::GL1 construct into myb23-1 plants. The resulting transgenic plants showed only weak rescue of the branching phenotype (Table 1). As MYB23::MYB23 showed full rescue of trichome branching it is conceivable that GL1 protein is less active than MYB23 in promoting trichome branching.

To study the importance of the transcriptional regulation in functional specification of the GL1 and MYB23 genes, we introduced the GL1::MYB23 construct into the myb23-1 mutant background. These plants displayed no rescue of trichome branching(Table 1), indicating that transcriptional control is pertinent for the regulation of branching by MYB23. Taken together, the functional diversification of the MYB23 and GL1 genes with respect to branching occurs at the level of both the transcriptional regulation and protein function.

Although the initial studies of the expression of MYB23 suggested that it is expressed similarly as GL1(Kirik et al., 2001), the above-described studies suggest differences in transcriptional regulation during early stages of trichome development.

To follow the expression of GL1 and MYB23 during trichome development, we used GL1::GUS and MYB23::GUS reporter lines(Kirik et al., 2001; Larkin et al., 1993). The GL1::GUS reporter line was shown previously to adequately reflect the transcription of the GL1 gene(Larkin et al., 1993). For MYB23, we narrowed the relevant promoter region down to a 1.9 kb region that was previously used to analyze the expression of the AtMYB23 gene (Kirik et al.,2001). We used this fragment to drive the expression of the MYB23 cDNA and found that this fragment is sufficient to rescue the myb23 mutant phenotype in the gl1 myb23 background (data not shown). This shows that the MYB23::GUS reporter provides a pertinent proxy of the MYB23 gene expression.

A comparison of the reporter expression MYB23::GUS and GL1::GUS in wild type revealed marked differences in the spatial and temporal pattern. Initially, GL1::GUS is expressed ubiquitously in the young leaf primordia and becomes more prominent in developing trichome cells (Fig. 3B)(Larkin et al., 1993). This expression ceases when trichomes begin to initiate branches. By contrast,using the same staining conditions we did not find a ubiquitous expression of MYB23::GUS in young leaf primordia. Expression is confined to developing trichome cells where it persists at high levels throughout all stages of trichome development (Fig. 3B).

To further analyze the transcriptional regulation of the GL1 and MYB23 genes, we compared their expression in different mutants affected in trichome initiation (Hulskamp et al., 1999; Szymanski et al., 2000). In try and cpc mutants, which display increased trichome nest frequency and trichome number, respectively,we did not detect expression changes using RT-PCR(Fig. 3A). In gl1 and ttg1 mutants, which show strongly reduced trichome production, MYB23 mRNA was not detectable. In gl3 mutants, which have less trichomes compared with wild type, we found that amount of the MYB23 RNA was strongly reduced(Fig. 3A). By contrast, GL1 expression was not changed in ttg1 and gl3mutants. These results show that MYB23 expression correlates with the presence of trichomes on the leaves whereas GL1 expression is independent of them.

Comparable results were obtained using the respective promoter::GUS fusions. The ubiquitous expression of the GL1::GUS reporter is not affected in ttg1 and gl1 mutants(Fig. 3B). Expression of both the GL1::GUS and MYB23::GUS reporters was not affected in the gl3 mutant (data not shown), suggesting that reduced expression of MYB23 detected by RT-PCR is due to reduced number of trichomes in this mutant. As MYB23 is not initially ubiquitously expressed, we focused on the rare trichomes that occasionally formed at the leaf edges of gl1 and ttg1 mutants. In the gl1 mutant, both reporter constructs were active in trichomes. Rare trichomes in the ttg1 mutant showed expression of the GL1::GUS but not MYB23::GUS reporter, indicating that MYB23 gene transcription depends on TTG1 activity, whereas the GL1 gene transcription is not TTG1 dependent.

To verify the expression pattern of the MYB23::GUS reporter, we localized MYB23 transcript by in situ hybridization(Fig. 4). We detected a strong hybridization signal in developing trichomes of young leaf primordia(Fig. 4A) that was also persistent in the older trichomes (Fig. 4B). In contrast to the GL1 mRNA in situ localization reported by Larkin et al. (Larkin et al.,1993), we did not find any epidermis-specific accumulation of the MYB23 transcript, which is not confined to developing trichome precursors.

To further dissect the function of MYB23 in trichome development,we created double mutants of myb23 with gl2, try, gl3 and cpc.

In gl2-4AA mutants, the majority of trichomes fail to grow out and only trichomes at the leaf edges develop and branch. In the gl2-4AA myb23-2 double mutant, leaf edge trichomes do not grow out(Fig. 5H,I). Thus myb23 mutants enhance the morphogenesis defect in the gl2mutant. Conversely, overexpression of MYB23 in trichomes using strong GL2 promoter partially rescued the gl2 mutant phenotype(Fig. 5J-M). This result was surprising as GL2 is thought to act downstream of GL1 and therefore was expected to also act downstream of MYB23. Both observations together suggest that MYB23 has a function in trichome cell morphogenesis at the same genetic level as GL2.

The try myb23-1 double mutant showed an intermediate branching phenotype. When compared with try (average 5.0 branches) and myb23-1 (average 2.6 branches) mutants, the double mutant exhibits in average 3.4 branches (Fig. 5E,F; Table 1). As the increased trichome branching in try mutants is correlated with an increased ploidy levels (64C DNA content versus 32C in wild type)(Hülskamp et al., 1994)(Fig. 6), we analyzed whether MYB23 function is also required for the extra DNA endoreduplication rounds in the try trichomes. Both myb23 alleles showed ploidy levels similar to wild type (Fig. 6). However, the ploidy level of the try myb23-1 double mutant displayed an intermediate value (average is 44C) and differed significantly (P<0.001) from the try mutant (average is 67C) and from the myb23 mutant (P<0.005; average is 34C). Thus, although the single mutant phenotypes suggest that regulation of branch number by the MYB23 gene is independent from the regulation of DNA endoreduplication, the try myb23 double mutant phenotype revealed that MYB23 function becomes limiting in the regulation of DNA endoreduplication when nuclei undergo additional endoreduplication cycles.

The same two processes affected in the try mutant – cell morphogenesis and DNA endoreduplication in trichomes – are also impaired in the gl3 mutant. gl3 trichomes are smaller, underbranched and have reduced nuclear DNA content(Hülskamp et al., 1994). The gl3 branching phenotype appears to be epistatic in the gl3 myb23-2 double mutant (Fig. 5B,C; Table 1),suggesting that the MYB23 and GL3 may genes act in the same pathway.

To further elucidate the role of MYB23 and GL3 in trichome branch formation, we overexpressed the two genes under the control of the GL2 promoter, which drives a strong trichome-specific expression in leaves (Szymanski et al.,1998). Plants transformed with the GL2::GL3 construct exhibited enlarged, strongly over-branched trichomes with up to 10 branches(Fig. 5G; Table 1). The trichome nuclear size was drastically increased (Fig. 5G) with a ploidy level of on average 110C, suggesting that those nuclei have undergone six endoreduplication cycles, two more than wild-type trichome nuclei (Fig. 6).

Arabidopsis plants transformed with the GL2::MYB23construct did not exhibit any changes in trichome size, branching and DNA endoreduplication level (average is 31.6 C; n=66), indicating that the wild-type level of MYB23 is not a limiting factor for endoreduplication cycles and branch initiation in wild-type plants. To test whether the inhibitory activity of TRY may suppress the possible effect of the GL2::MYB23 construct, we introduced this construct in the try mutant background. Overexpression of MYB23 under the GL2 promoter in try background resulted in ectopic trichome development but did not have any significant effect on the trichome branching and DNA content (try GL2::MYB23: 57C, n=64; try:64C, n=65; Student's t-test: P=0.16).

Taken together, our data show that MYB23 is required for proper branch formation but, in contrast to GL3, increased MYB23 levels do not trigger additional branch formation. This suggests that the MYB23 concentration in trichomes is not limiting for induction of trichome branches.

The functional equivalence of the GL1 and MYB23 proteins during trichome initiation suggests that these proteins have similar biochemical properties. The GL1 protein directly interacts with GL3, GL3 binds to the WD-40 protein TTG1 but GL1 does not interact with TTG1. It has been suggested that these three proteins form the activator complex(Payne et al., 2000; Szymanski et al., 2000). To test whether MYB23 interacts with the GL3 protein, we made fusion constructs of the GL3 cDNA with the GAL4 activation domain (AD) and MYB23 cDNA with the GAL4 DNA binding domain (DB). Yeast two hybrid assays revealed an interaction between GL3 and MYB23(Fig. 7) but not with TTG1(data not shown). Payne et al. (Payne et al., 2000) demonstrated that GL3 interacts with GL1 through the N-terminal end of the GL3 protein. We found that the truncation of 96 amino acids from the N-terminal end of the GL3 protein also abolished the interaction between GL3 and MYB23 (Fig. 7). These results suggest that MYB23 interacts with the same GL3 protein domain as GL1.

Since try mutant trichomes are increased in size and in trichome branch number, TRY acts as a negative regulator of trichome branching. To address the mechanism of the inhibitory role of TRY in trichome branching we tested the interactions of the TRY protein with GL3, MYB23 and GL1 (Fig. 7). Our interaction assay did not reveal an interaction between TRY and MYB23 and between TRY and GL1. We found an interaction between TRY and GL3 proteins that was abolished when the N-terminally truncated version of GL3 was used in assay, indicating that TRY, GL1 and MYB23 may bind to the same region of GL3. This suggests that MYB23 and TRY proteins compete for binding to GL3 similar as it was shown for GL1 (Esch et al., 2003). As the binding strength of TRY and GL3 is lower than that of MYB23 and GL3 – as revealed in yeast two-hybrid studies – effective competition of TRY with GL3 would require that either the concentration of TRY or the combined concentration of all the TRY-like genes is higher than that of MYB23.

What molecular mechanisms fuel the evolution of plant form? In principle,evolutionary changes are accompanied by functional diversification of regulatory genes and may occur through changes of protein function or by changes of the regulation of genes. In particular the latter has been favoured because the extreme morphological differences of organisms do not correspond to the low divergence of proteins. In fact, many examples are known where protein function is conserved over large evolutionary distances(Doebley and Lukens, 1998).

Functional specification of paralogous genes provides a valuable experimental system with which to study the molecular mechanisms of gene diversification. In this study, we have investigated the functional diversification of MYB23 and GL1. They share a sequence identity of 63% over the entire protein and 92% in the actual MYB domain. MYB23 and GL1 are located on chromosomes III and V in the regions that originated from chromosomal duplication(The Arabidopsis Genome Initiative,2000). It is therefore likely that the two genes are derived from gene duplication. Our genetic analysis revealed that MYB23 and GL1 have a partially redundant function with respect to trichome initiation, and that MYB23 has an additional role in the regulation of trichome branching. The analysis of their functional specification by promoter and protein-coding region swap experiments, expression studies and protein-protein interaction studies gave insights in the functional diversification of the two genes.

In contrast to GL1 and WER genes, which have completely non-overlapping expression domains and unrelated functions(Lee and Schiefelbein, 2001), GL1 and MYB23 are both expressed in trichomes and regulate their development. Our swapping experiments, however, revealed that their actual regulation is significantly different and not interchangable. In this study, we have not specifically addressed the regulatory role of introns,which have recently been shown to play a role in GL1 regulation(Wang et al., 2004). However,as GL1::MYB23 and GL1::GL1 or MYB23::MYB23 and MYB23::GL1 constructs rescued the corresponding mutants equally well,it is unlikely that the coding region and the introns carry any information relevant for the regulatory differences between GL1 and MYB23.

A comparison of their temporal and spatial expression revealed two differences. First, GL1 but not MYB23 is initially expressed ubiquitously in young leaves; and second, MYB23 is expressed throughout all stages of trichome development, whereas GL1 expression ceases long before trichome development is completed. Both aspects can be correlated with their respective functions. The initial ubiquitous expression of GL1 is thought to be important for early pattern formation(Hulskamp, 2004; Larkin et al., 2003). It is postulated that the initial ubiquitous expression of GL1 and other positive regulators of trichome initiation triggers a patterning mechanism that leads to a regular pattern of differentiated cells in which the positive regulators, including GL1 are upregulated. Consistent with the finding that MYB23 is not expressed in these initial stages, MYB23 has only a subtle role in trichome initiation and the expression of GL1 under the MYB23 promoter cannot rescue gl1 mutants.

In addition, the extended expression of MYB23 in developing trichomes is consistent with the primary role of MYB23 gene in trichome branching. This extended expression of MYB23 is functionally relevant, as trichome branching in myb23 mutants is not rescued when MYB23 is expressed under the control of the GL1promoter.

Some of these regulatory changes can be correlated with different responses to known trichome patterning genes. Although GL1 expression is independent from GL3, GL1 and TTG1, the expression of MYB23 in leaf trichomes is regulated by TTG1 but not by GL3 and GL1.

Despite the relative low sequence identity of GL1 and MYB23, the two proteins can functionally replace each other during trichome initiation. Trichome initiation on the whole leaf surface was rescued, with the GL1 promoter driving either MYB23 or GL1 protein; trichome initiation at the leaf edges and petioles of the gl1 myb23 double mutant was rescued with the MYB23 promoter driving either GL1 or MYB23 proteins.

In accordance with the ability of the MYB23 and GL1 proteins to rescue the trichome initiation phenotype of each mutant, yeast two-hybrid assay showed that MYB23 protein interactions with known trichome patterning proteins are similar to those found with GL1. Therefore it came as a surprise that protein functions are not fully conserved during trichome branching. When GL1is expressed under the MYB23 promoter, trichome branching was not completely rescued, indicating that GL1 cannot fully replace the protein function of MYB23 in this developmental context. This raises the question of whether the regulation of trichome branching by MYB23 involves the same downstream genes as during trichome patterning. Although the current data do not allow the determination of this, some observations suggest that this is the case. Notably, mutations in GL3 and TTG1 not only lead to patterning defects but also to reduced branching; mutations in TRY, a gene that inhibits trichome initiation, also result in increased branching. A conceivable scenario is that, similar as postulated in the context of pattern formation, TTG1 and GL3 form a branch-promoting complex together with MYB23 and that TRY counteracts this by competing with MYB23 for the binding with GL3. One possible explanation for the difference between GL1 and MYB23 proteins in this context is that MYB23 protein may endow the trimeric complex with higher activity in the promotion of cell growth and DNA endoreduplication.

We thank laboratory members and anonymous reviewers for critical reading of the manuscript and helpful suggestions. We thank Irene Klinkhammer for the technical assistance. We also acknowledge the Signal Program (La Jolla, CA)and the Arabidopsis Biological Resource Center (Ohio State University,Columbus, OH, USA) for providing the myb23-1 T-DNA insertion line. The myb23-2 allele was isolated with the help of the Wisconsin Arabidopsis Knockout Facility. We thank the ZIGIA group, MPI of Plant Breeding, Cologne, for providing the gl2-4AA mutant. This work was supported by the SFB 572 to M.H., NSF IBN-0316312 to J.S. and NSF DBI-0115684 to D.G.O.