Introduction.

Many of the biological processes in a plant are regulated at the level of transcription. Changes in gene expression have been shown to underlie the response to environmental cues and stresses (such as light, temperature, and nutrient availability), the defense response against pathogens, the regulation of metabolic pathways, the regulation of photosynthesis, or the establishment of symbiotic relationships, to name a few. In plants, as well as in animals, development is based on the cellular capacity for differential gene expression (reviewed in: Scott, 2000; Benfey and Weigel, 2001). Accordingly, many of the genes identified in screens for Arabidopsis mutants with altered, for example, flower or root development have been found to encode transcription factors. Alterations in gene expression are also emerging as a major source of the diversity and change that underlie the morphological evolution of eukaryotic organisms (Doebley and Lukens, 1998; Cubas et al., 1999b; Carroll, 2000; Tautz, 2000). In particular, morphological changes that occurred during plant domestication and crop improvement in agriculture have been associated with mutations in transcription factors (Peng et al., 1999), alterations in their expression (Doebley et al., 1997; Wang et al., 1999b), or changes in the expression of other types of regulatory proteins (Frary et al., 2000). Related transcription factors, such as the Arabidopsis MYB proteins WEREWOLF (WER) and GLABROUS1 (GL1), have been shown to be functionally equivalent, and owe their particular roles in plant development to differences in their expression patterns (Lee and Schiefelbein, 2001).

The availability of the Arabidopsis genome sequence (Lin et al., 1999; Mayer et al., 1999; Arabidopsis Genome Initiative, 2000; Salanoubat et al., 2000; Tabata et al., 2000; Theologis et al., 2000) allows a global, or genomic, analysis of transcriptional regulation in plants. Whereas the mechanisms of transcription are largely common across eukaryotes, their components vary among kingdoms. The complement of genes coding for transcriptional regulators in Arabidopsis has been described (Arabidopsis Genome Initiative, 2000; Riechmann et al., 2000). Their systematic functional characterization can be pursued with a variety of reverse genetic methods (Riechmann and Ratcliffe, 2000). In addition, gene expression profiling technologies, such as DNA microarrays, allow monitoring transcription factor activity at a genome-wide level. These studies should eventually lead to an understanding of the interplay of the transcription factors with the genome whose expression they control.

This chapter intends to provide a genomic perspective on transcriptional regulation in Arabidopsis. The first section briefly reviews the different types of proteins directly involved in transcription in eukaryotes, and our current understanding on how they function. The following sections consist of a description of the Arabidopsis complement of genes and proteins involved in transcriptional control, in particular sequence-specific DNA-binding transcription factors and chromatin-related proteins. Transcriptional regulators often act in a combinatorial fashion, and this mode of action is reviewed in the context of Arabidopsis promoters and cis-regulatory sequences, and of protein-protein interactions. Finally, genome-wide functional analyses of transcription factors, the characterization of the Arabidopsis promoterome, and of the transcriptome by gene expression profiling experiments, are considered. The availability of the genome sequence of different prokaryotic and eukaryotic organisms has provided for new ways of searching for unity and diversity among biological systems, and given birth to the field of comparative genomics. Although the subject of this book is Arabidopsis, reference is made in this chapter to other eukaryotic organisms, in order to situate the Arabidopsis genome information in a broader biological context.

2. Transcription machinery: concepts, components, and mechanisms

In eukaryotic organisms, regulation of gene expression proceeds through mechanisms that are fundamentally different from those in prokaryotes, which explains both the large number and diversity of proteins that are involved in the process, as well as how it can be tightly regulated to facilitate the diversification in expression patterns that is required for biological complexity (Struhl, 1999). In a prokaryote such as E. coli, the ground state for transcription is non-restrictive, that is, the RNA polymerase complex is not limited in its ability to gain access to the DNA and initiate RNA synthesis (Struhl, 1999). Negative regulation is rare, and exerted by sequence-specific repressors. Furthermore, it has been estimated that the global structure of the E. coli gene regulatory network possesses low complexity. On average, a transcription factor would regulate three genes, and an E. coli gene would be under the direct control of two transcription factors (Thieffry et al., 1998). There is a prominence of promoters controlled by a single regulator, and whereas many of the regulators regulate themselves (usually through auto-inhibitions), very rarely do they regulate other transcription factors (Thieffry et al., 1998).

In contrast, the ground state for transcription in eukaryotes is restrictive, as a result of the packing of the DNA into chromatin, which blocks the recognition of the core promoters by the basic transcription machinery (Kornberg, 1999; Struhl, 1999). The effects of chromatin structure on promoter accessibility makes chromatin modifying activities necessary for eukaryotic transcription, and has important implications for the way transcription factors act. In addition to the components of the basic transcription machinery and to scores of sequence-specific DNA-binding transcription factors, eukaryotic genomes contain a variety of genes that code for chromatin-related proteins. Furthermore, transcriptional regulators in eukaryotes operate following a combinatorial logic (an efficient way of increasing the number and diversity of gene regulatory activities), and the complexity of the regulatory networks can be great.

Prokaryotic sequence-specific DNA-binding transcription factors often recognize binding sites longer than 12 base-pairs (bp) (see RegulonDB,  http://www.cifn.unam.mx/regulondb/ and DPInteract,  http://arep.med.harvard.edu/dpinteract) (Robison et al., 1998; Salgado et al., 2001), whereas binding sites for eukaryotic transcription factors are usually shorter, 5 to 10 bp long. A combinatorial mechanism composed of factors that recognize short sequences is probably a more economical way (requires a reduced number of factors) of selectively regulating the expression of tens of thousands of genes, than a mechanism based upon factors that are each dedicated to control a small number of genes and operate through longer target sites. Thus, the DNA binding characteristics of eukaryotic transcription factors, and the mechanisms of transcription themselves, might be operationally and evolutionarily related to features of eukaryotic genomes such as the vast increase in genome size and in the number of genes to be regulated.

Briefly, the proteins involved in transcription in eukaryotes can be classified into four different functional groups: (1) the basic transcription apparatus and intrinsic associated factors (also known as general transcription factors, or GTFs); (2) large multi-subunit coactivators and other cofactors; (3) sequence-specific DNA-binding transcription factors; and (4) chromatin-related proteins. In contrast to the components of the basal transcription machinery, which in general are highly conserved, coregulators and transcription factors have diverged largely among eukaryotes. The roles that the proteins in these four classes play can be summarized as follows (after Lee and Young, 2000; Lemon and Tjian, 2000, and  http://web.wi.mit.edu/young/pub/regulation.html) (for an extensive coverage of the mechanisms of eukaryotic transcription, see: Latchman, 1998; Elgin and Workman, 2000; White, 2001).

(1) The basic transcription apparatus, and intrinsic associated factors. In eukaryotic organisms, there are three different RNA polymerases, which are responsible for the synthesis of rRNA (Pol I), mRNA (Pol II), and tRNA, 5S rRNA, and other small RNA molecules (Pol III). The focus of this chapter is the transcription of protein-encoding genes, which is carried out by Pol II exclusively. Pol II is a multi-subunit enzyme (Cramer et al., 2001) that requires accessory factors to recognize promoter sequences and accurately initiate transcription. These general transcription factors (GTFs) include TFIIA, TFIIB, TFIID, TFIIE, TFIIF, and TFIIH. GTFs carry out a variety of different functions, from positioning the polymerase on the promoter (TFIIB) to unwinding its DNA (TFIIH). TFIID is a multi-subunit complex that is generally responsible for promoter recognition. It contains the TATA-box binding protein (TBP) and several TBP-associated factors (TAFs) (reviewed in Green, 2000). The TAF subunits of TFIID are critical for the responsiveness of the basic apparatus to transcriptional activators. However, individual TAFs are not essential for transcription of all genes in a genome. TAFs contribute to the specificity and variety of transcriptional responses: distinct TAFs can be targeted by different classes of activators, and individual TAFs can function as promoter selectivity factors. Furthermore, some TAFs can form part of other multi-subunit regulatory complexes, in addition to TFIID, such as the histone acetylation SAGA complex; and whereas most of the TAFs are ubiquitously expressed, some are expressed in a tissue or cell-type specific manner, which can lead to the formation of different TAF-containing complexes (for a review on gene-selective roles of GTFs and TAFs, see Veenstra and Wolffe, 2001).

(2) Large multi-subunit coactivators, and cofactors that bind sequence specific transcription factors. This heterogeneous class of regulatory proteins includes cofactors that interact with sequence-specific transcription factors and modulate their DNA binding or interaction with the core machinery, as well as large multi-subunit coactivators such as the Mediator complex, initially identified in yeast. Multi-subunit coactivators interact with Pol II and/or with multiple types of activators, serving as a modular adapter to regulate transcription initiation (Hampsey and Reinberg, 1999). The Mediator (or Mediator-like) complex is found in organisms from yeast to humans, but its number of subunits vary, and the complex from one organism might contain subunits that have no orthologs in another (Malik and Roeder, 2000; Rachez and Freedman, 2001).

(3) Sequence-specific DNA-binding transcription factors (activators and repressors). These are transcription factors of the classic type: usually defined as proteins that show sequence-specific DNA binding and are capable of activating and/or repressing transcription. They are responsible for the selectivity in gene regulation, and are often themselves expressed in a tissue, cell-type, temporal, or stimulus-dependent specific manner. Transcription factors are modular proteins, with distinct and functionally separable domains, such as DNA-binding and activation domains. Most known transcription factors can be grouped into families according to their DNA binding domain (Luscombe et al., 2000). Transcription factors can interact directly with different components of the general machinery and with coactivators, affecting complex formation. They can also interact with chromatin remodeling complexes.

(4) Chromatin-related proteins. This group includes factors that covalently modify histones (such as histone acetylases and deacetylases), and remodeling complexes that hydrolize ATP for reorganizing chromatin structure (such as the SWI/SNF and ISWI complexes). Histone acetylation is generally a characteristic of transcribed chromatin, whereas deacetylation is associated with repression. Accordingly, histone acetyltransferase activities are found in coactivators, and deacetylase activities in corepressors. Chromatin proteins usually form part of multi-subunit complexes.

Using the regulation of the HO endonuclease gene in yeast (Saccharomyces cerevisiae) as a paradigm, the steps leading to transcriptional activation can be summarized as follows (Cosma et al., 1999; Cosma et al., 2001). Upstream sequences are recognized by a transcription (enhancer-binding) factor, with accessibility to its targets sites despite the packing of the DNA into chromatin fibers. This transcription factor recruits the SWI/SNF complex, which then recruits SAGA, and results in the remodeling of chromatin and localized histone acetylation, which facilitates the access of additional transcription factors to cis-regulatory sequences. The secondary activators direct gene transcription through multiple interactions with cofactors and the core machinery, recruiting the RNA polymerase complex to the transcription initiation site. The specific order in which the different chromatin-modifying complexes are recruited can vary among promoters and organisms, but the dual role of activators, first enlisting chromatin modifying activities and then inducing localization of the basal transcription apparatus, appears to be widespread in eukaryotes, including plants (see below, and: Agalioti et al., 2000; Brown et al., 2001; Merika and Thanos, 2001).

In many instances, the correct functioning of a gene requires the termination of the activation of its transcription to be as rapid or precise as its initial triggering. Termination of activation can be accomplished by several mechanisms, among them the targeted destruction of transcription factors after their interaction with the basal transcription machinery. Phosphorylation of a transcription factor molecule by kinases that form part of the Pol II holoenzyme (such as Srb10 or TFIIH) would mark it for ubiquitin-mediated destruction, effectively preventing it from engaging into another Pol II initiation event, and freeing the promoter sequence to interact with another transcription factor molecule (reviewed in Tansey, 2001).

In addition to the mechanisms of transcriptional control that the classes of proteins described in this section mediate, there are at least two other possible levels of regulation of gene expression in eukaryotes: DNA methylation and nuclear organization. DNA methylation is associated with suppressed gene expression, and is reviewed in other chapters of this book (see also: Finnegan et al., 2000; Habu et al., 2001). Nuclear organization could provide for a higher level of regulation of gene expression, where different transcriptional functions might be segregated into distinct compartments (for models and reviews, see: Francastel et al., 2000; Lemon and Tjian, 2000; Cremer and Cremer, 2001; Misteli, 2001).

Of all the groups of proteins involved in transcription, the most numerous one is that of sequence-specific DNA-binding transcription factors. They are the principal factors upon which the mechanisms for selectivity of gene activation are built, and the basic (although not the only) protein components of the combinatorial logic of transcription.

3. Transcription factor gene content of the Arabidopsis genome

The analysis of the Arabidopsis genome sequence indicates that it codes for at least 1,572 transcription factors, which account for ∼6% of its estimated ∼26,000 genes (Arabidopsis Genome Initiative, 2000; Riechmann et al., 2000) (Table 1). This observation, however, represents an underestimate of the total number of transcription factors, given that, at present, approximately 40% of the proteins predicted from the genome sequence cannot be assigned to functional categories on the basis of sequence similarity to proteins of known biochemical function (Lin et al., 1999; Mayer et al., 1999; Arabidopsis Genome Initiative, 2000; Salanoubat et al., 2000; Tabata et al., 2000; Theologis et al., 2000). Some of those uncharacterized proteins are expected to be transcriptional regulators and, in fact, novel classes of transcription factors are still being discovered (for example: Boggon et al., 1999; Schauser et al., 1999; Kawaoka et al., 2000; Nagano et al., 2001; Windhövel et al., 2001). Therefore, the total number of transcription factor genes present in Arabidopsis (as well as, for the same reasons, in any other of the sequenced eukaryotic genomes) will be uncertain for some time.

A question pertaining to genome-wide surveys is whether all the proteins identified by sequence similarity searches do indeed belong to the functional groups into which they are being catalogued. In the case of transcription factors, the answer depends on the particular gene family that is considered. If the conserved DNA-binding domain that defines a gene family is poor in sequence information (for example, some zinc-coordinating motifs), the ratio of false positives in the searches can be relatively high (although it can often be reduced by additional sequence comparison strategies that are beyond the scope of this chapter, see: Riechmann et al., 2000). On the other hand, many families are defined by long DNA-binding domains (50 to 70 amino acids) with multiple residues being highly conserved (that is, the domains are rich in sequence information). The three-dimensional structure of these domains might have been solved, and revealed the contacts between some of the conserved residues and the DNA. In cases like these, such as for example the homeobox and the AP2/ERF (APETALA2/ethylene response factor) families, it is reasonable to expect all the members of the gene family to be transcription factors (activators or repressors) (the AP2/ERF family was initially referred to as AP2/EREBP, for AP2/ethylene responsive element binding protein). However, there are cases in which a family of bona fide transcription factors might also contain members that have additional functions (for reviews on multifunctional transcription factors: Ladomery, 1997; Wilkinson and Shyu, 2001). For example, the Drosophila homeodomain protein Bicoid directs anterior embryo development both by regulating transcription and by interacting with Caudal mRNA and inhibiting its translation, thus restricting Caudal (which is another homeodomain protein) accumulation to the posterior part of the embryo through posttranscriptional control. Both DNA- and RNA-binding are specified by the Bicoid homeodomain, but by distinct subregions or residues in it (Niessing et al., 2000). The Arabidopsis MYB-related protein AtCDC5 is known to be homologous to the S. cerevisiae CEF1 and S. pombe Cdc5 proteins (Hirayama and Shinozaki, 1996; Ohi et al., 1998). Cdc5 proteins are essential for the G2/M progression, but their molecular functions are not completely understood, as they are required for pre-mRNA splicing and associate with core components of the splicing machinery, but also show sequence-specific binding to double stranded DNA and transactivation potential (Burns et al., 1999; Lei et al., 2000). It is thus possible that Cdc5 proteins are another example of factors with several distinct molecular functions. Two other members of the Arabidopsis MYB-related family, AtTRP1 and AtTBP1, have been identified as telomere-binding proteins (Chen et al., 2001a; HwaNg et al., 2001), although a possible role in transcription cannot be ruled out because the 5′ regions of some Arabidopsis genes contain two or more non-contiguous telomeric repeats (Regad et al., 1994). These examples illustrate the limitations of using sequence similarity to assign potential roles to proteins that are otherwise uncharacterized, and also how the determination of their molecular functions can be elusive. Similar cases might occur within some of the zinc-coordinating protein families, since the same or related motifs can be involved in DNA- and RNA-binding, and may be present in proteins withfunctions involving nucleic acid binding but not necessarily transcriptional regulation. For example, vertebrate Y-box proteins contain a zinc-coordinating cold-shock domain, and are often dual DNA- and RNA-binding proteins that can regulate transcription and/or translation (reviewed in: Matsumoto and Wolffe, 1998; Sommerville, 1999).

With these caveats in mind, the Arabidopsis complement of transcription factors has been the subject of an extensive genome-wide descriptive analysis, which also included a comparison with those of Drosophila melanogaster, Caenorhabditis elegans, and Saccharomyces cerevisiae (Riechmann et al., 2000). The main conclusions of that study are summarized here.

The 1,572 transcriptional regulator genes identified in the Arabidopsis genome are classified into more than 45 different gene families (Table 1; Figure 1), all of which are scattered throughout the genome. In addition, there are a few single-copy or “orphan” genes, such as LEAFY (LFY). Transcriptional regulators represent approximately 4.6, 3.5, and 3.5% of the genes in Drosophila, C. elegans, and yeast, respectively (Riechmann et al., 2000). Thus, the Arabidopsis content of transcription factors is 1.3 times that of Drosophila, and 1.7 times that of C. elegans and yeast (Riechmann et al., 2000). The large number and diversity of transcription factors in Drosophila were proposed to be related to its substantial regulatory complexity (Adams et al., 2000). Applying the same logic to Arabidopsis suggests that the regulation of transcription in plants is as complex as that in Drosophila. Furthermore, if the estimated total number of genes in humans, 30,000–40,000, and of transcription factor genes, 1,850–2,000, are correct (International Human Genome Sequencing Consortium, 2001; Tupler et al., 2001; Venter et al., 2001), then the transcription factor gene content of Arabidopsis and of H. sapiens are similar (∼6% in Arabidopsis, versus 4.6–6.6% in humans). It should be noted, however, that there is a substantial degree of uncertainty about these estimates of gene numbers in humans (see, for example: Hogenesch et al., 2001; Wright et al., 2001).

Transcription factors, the networks that they form, and the genes that they regulate, have been proposed as a possible objective measurement (connectivity of gene-regulation networks) of the biological complexity of an organism (Szathmáry et al., 2001). From that point of view, the large number of transcription factors in Arabidopsis was interpreted in the context of the complexity of secondary metabolism in plants (Szathmáry et al., 2001), but it might also be related to the complex interactions between plants and the environment (both biotic and abiotic) as well as to the degree of duplications in the genome (see below, and Arabidopsis Genome Initiative, 2000).

The extent to which the Arabidopsis complement of transcription factors represents that of other plants is still an open question. Since the evolutionary divergence between the monocot and dicot lineages is a relatively recent event, which perhaps occurred ∼200 million years ago (Savard et al., 1994), it could be expected that the overall composition of monocot and dicot transcription factor complements would be similar. In fact, the largest transcription factor families in Arabidopsis also appear to be the most prevalent ones in monocotyledonous plants. For example, the phylogenetic comparison of a subset of maize and Arabidopsis MYB-(R1)R2R3 sequences shows that the amplification of the gene family occurred prior to the separation of monocots and dicots (Rabinowicz et al., 1999). In addition, within phylogenetically well-studied families of transcription factors, such as the MADS-box family, many examples of orthology can be identified between Arabidopsis genes and those from rice or maize, and even from gymnosperms (reviewed in: Theissen et al., 2000; Ng and Yanofsky, 2001) (see also  http://www.mpiz-koeln.mpg.de/mads). Putative orthologous MADS-box genes have regularly maintained conserved functions, even after substantial sequence divergence (Theissen et al., 2000). However, it is also apparent that diversity in transcriptional regulators will be found within the plant kingdom, and between monocots and dicots. Many MADS-box gene duplication and diversification events occurred after separation of the moss and fern lineages from the lineage that originated the flowering plants (Münster et al., 1997; Hasebe et al., 1998; Krogan and Ashton, 2000; Svensson et al., 2000), and at least two clades of MADS-box genes appear to have been amplified in the phylogenetic lineage that led to grasses with respect to Arabidopsis (Theissen et al., 2000). Similarly, whereas most of the amplification of the MYB-(R1)R2R3 gene family occurred prior to the separation between monocots and dicots, several subgroups in maize appear to have originated recently or undergone duplication (some of these duplications are likely to be associated with the allotetraploid origin of the maize genome, but others do not reflect it: Rabinowicz et al., 1999). These recent expansions could have allowed a functional diversification that might not be present in Arabidopsis.

An issue that impinges on the question of the similarity of the Arabidopsis complement of transcription factors with that of other plants is the degree of completeness of the current characterization (i. e., sequence determination and analysis) of the Arabidopsis genome, in particular if that question is to be addressed on a gene-by-gene basis. TRM1 is a maize C2H2 zinc finger transcription factor involved in the repression of rbcS gene expression in mesophyll cells that is related to the mammalian transcription activator-repressor YY1 (Xu et al., 2001). A BLAST search of the higher plant DNA sequences available in GenBank (July 2001) identifies homologous genes in other monocots (Triticum aestivum) as well as in dicotyledonous plants (Nicotiana tabacum, Solanum tuberosum), but not in Arabidopsis. It is possible that an Arabidopsis TRM1 homolog resides in one of the still unsequenced segments of the genome (see  http://www.arabidopsis.org). Similarly, there are a few Arabidopsis transcription factor genes represented by ESTs or BAC-end sequences that still cannot be identified in the genome sequence. The limitations of the current sequencing technologies make it impractical or impossible to determine the sequence of eukaryotic genomes to absolute completeness. Thus, a failure to identify a particular gene in the genome sequence of an organism should not be taken as a definitive proof of the absence of that gene. In addition, gene sequences might diverge more than expected, which might result in the identification of homologous genes requiring more sophisticated sequence analysis than a standard BLAST search. For example, a homolog of the mammalian tumor suppressor gene p53 can be identified in the sequence of the C. elegans genome, despite initial reports that no p53-like gene was present in that organism (Derry et al., 2001; Schumacher et al., 2001).

The genome-wide comparison of transcription factors among eukaryotic organisms (Arabidopsis, Drosophila, C. elegans, and S. cerevisiae, encompassing the plant, animal, and fungal kingdoms) reveals the evolutionary generation of diversity in the regulation of transcription (Riechmann et al., 2000). Each of these eukaryotic kingdoms has its own set of particular transcription factor families and genes. Members of kingdom-specific families represent 45% of the Arabidopsis complement of transcriptional regulators, whereas those of families that are present in all four organisms account for 53% (Figure 2). In each organism, a minority (2–5%) of its transcription factors belong to families that are present in two of the three kingdoms: in animals and yeast (SOX/TCF, Fork head, and RFX1-like transcription factors) or in plants and animals (TULP, CPP, and E2F/DP families) (Figure 2) (Riechmann et al., 2000). This distribution of genes and gene families reflects the evolutionary history of eukaryotes. According to molecular phylogenetic analyses, plants, animals and fungi all diverged from a common ancestor during a short period of time, ∼1.5 billion years ago (Wang et al., 1999a; Philippe et al., 2000; Nei et al., 2001). Thus, most of the transcription factor families are either shared by the three kingdoms (those that were present in the common ancestor), or specific to each one (those families that arose independently following divergence).

Many of the Arabidopsis transcription factor gene families are large (Table 1). However, none has been so disproportionately amplified as the nuclear hormone receptors in C. elegans (∼38% of its transcription factors), the C2H2 zinc finger proteins in Drosophila (∼46%), or the C6 and C2H2 families in yeast (∼25% each one) (Figure 3) (Riechmann et al., 2000). The three largest families of transcription factors in Arabidopsis, AP2/ERF, bHLH (basic-region helix-loop-helix), and MYB-(R1)R2R3, each represent only ∼9% of the total, and there are several other families with comparable numbers of genes (Figure 3) (Riechmann et al., 2000). The two transcription factor families that have been more substantially amplified in plants, as compared to animals and yeast, are the MYB and the MADS families. Another difference between the Arabidopsis complement of transcription factors and those of the other eukaryotes is that less than 25% of it consists of zinc coordinating proteins, whereas zinc coordinating transcription factors represent ∼51% of the total in Drosophila, ∼64% in C. elegans, and ∼56% in yeast (Riechmann et al., 2000).

The Arabidopsis transcription factors that belong to families that are common to all eukaryotes do not share significant similarity with those from the other kingdoms, except in the conserved DNA binding domains that define the respective families (Riechmann et al., 2000). Furthermore, diversity in protein sequence and structure is increased by domain shuffling (Figure 1). Shuffling of some of the DNA-binding domains that are present in all eukaryotes has generated novel transcription factors with plant-specific combinations of modules, as for example in the homeodomain, MADS, and ARID protein families (Figures 1 and 4) (Riechmann et al., 2000).

The Arabidopsis genome contains many tandem gene duplications and large-scale duplications on different chromosomes (Arabidopsis Genome Initiative, 2000; Blanc et al., 2000; Vision et al., 2000). Whereas some of these duplications have been followed by rearrangements and divergent evolution, up to 40 to 60% of the Arabidopsis genes might comprise pairs of highly related sequences (the percentage depending on the parameters used in the analyses) (Arabidopsis Genome Initiative, 2000; Blanc et al., 2000). Transcription factor genes follow these general observations. A comparison of the transcription factor complement to itself (all-against-all) revealed that, on average, closely related genes account for ∼45% of the total number in the major families (a pair of proteins was considered highly similar if they showed >60% amino acid sequence identity along at least two-thirds of the length of one of them) (Riechmann et al., 2000). The pairs or groups of closely related genes most often correspond to duplications in different chromosomes (∼65% on average), or to duplications in the same chromosome but at very large distances (∼22%), than to tandem repeats (∼13%) (Riechmann et al., 2000). In addition, clusters of three or more homologous transcription factor genes are very rare in the genome (Riechmann et al., 2000). This distribution indicates that it will be feasible to generate double or triple mutants for the majority of the pairs or groups of highly related genes that, because of their sequence similarity, might have overlapping or partially redundant functions (which might not be revealed by single mutant analyses; see below).

The analysis of ∼120,000 Arabidopsis expressed sequence tags (ESTs) (sequences available in GenBank in January 2001) suggests that, in terms of overall expression and considered as a whole, transcription factor genes are not substantially different from the rest of the genes in the genome. Approximately half of the ∼26,000 predicted genes are matched by an EST (Arabidopsis Genome Initiative, 2000; Theologis et al., 2000). Similarly, when the major Arabidopsis transcription families are considered, ∼47% of the genes are represented by an EST (Table 2). This observation is in contrast to the sometimes common assumption that, because of their regulatory nature, genes of this class are generally expressed at low levels.

4. Chromatin remodeling proteins.

Chromatin structure is an important element of the mechanisms that determine gene expression patterns in eukaryotes, because nucleosome assembly eliminates the accessibility of promoter sequences for the basal transcription machinery. The unfolding of packed chromatin is necessary for gene expression and, conversely, repression requires the formation and maintenance of condensed chromatin structures. Gene silencing and epigenetic phenomena, in which chromatin structure and histone modifications play a role, are by themselves the subject of other chapters in this book.

As summarized above, one of the mechanisms of transcription factor action is the recruitment of chromatin remodeling complexes to target promoters. This mechanism has been deduced from research on transcription in yeast and mammalian cells, but studies on the regulation of the β-phaseolin (phas) gene in bean (Phaseolus vulgaris) suggest that it also operates in plants (reviewed in Li et al., 2001a). The phas gene is silenced in vegetative tissues as a consequence of the positioning of a nucleosome over the TATA boxes of the promoter, making them inaccessible to TBP, whereas nucleosome displacement allows the gene to be highly expressed during seed development (Li et al., 1998). Such modification in chromatin structure results from the presence of the seed-specific transcription factor PvALF, a member of the ABI3/VP1 family (Li et al., 1999). However, PvALF is not sufficient for phas transcriptional activation, which does not occur in the absence of abscisic acid (ABA) (Li et al., 1998). Thus, a plausible model is that PvALF mediates chromatin reconfiguration, then allowing the binding of ABA-responsive transcription factors and the recruitment and assembly of the basal transcription machinery on the phas promoter (Li et al., 2001a).

The remodeling or reconfiguration of chromatin involves different types of enzymes, such as members of the SWI2/SNF2 subfamily of the DEAD/H box superfamily of nucleic-acid stimulated ATPases, and proteins that covalently modify histones, such as acetyltranferases (HATs) and deacetylases (HDACs), kinases, and methyltransferases (for reviews: Kadonaga, 1998; Elgin and Workman, 2000; Fry and Peterson, 2001; Jenuwein, 2001; Urnov and Wolffe, 2001). All eukaryotes appear to contain several proteins belonging to each one of these types, and each type can be further divided into different structural subclasses. Such structural diversity allows different proteins with the same biochemical activity to be involved in specialized cellular functions. The chromatin proteins with enzymatic activity usually form part of multi-subunit complexes, which might be necessary for their specificity and functionality.

In general, histone acetylation is a characteristic of transcribed chromatin, whereas deacetylation is associated with repression. HATs acetylate the ϵ-amino groups of specific lysine residues in the amino-terminal tails of the histone proteins that form the octamer around which the DNA wraps in the nucleosomes. Histone deacetylase-containing complexes reverse this covalent modification (reviewed in Khochbin et al., 2001). The molecular mechanisms by which histone acetylation affects chromatin structure and influences transcription could involve the destabilization of interactions between the DNA and the histone octamer (by neutralizing positive charges), interference with the high-order packing of chromatin, or the modification of interactions between histones and other proteins (reviewed in: Marmorstein, 2001a; Marmorstein and Roth, 2001; Roth et al., 2001). Other types of post-translational modifications, such as phosphorylation and methylation, also occur on histones, and can regulate chromatin structure and transcriptional activation and repression (reviewed in Marmorstein, 2001a). Methylation of specific lysine residues in histone tails is a relatively stable modification, thus providing a stable epigenetic mark for transcriptional regulation and gene silencing via heterochromatin assembly (reviewed in: Jenuwein, 2001; Rice and Allis, 2001). Lysine methyltransferase activity has been demonstrated for several eukaryotic SET domain proteins (Table 3). In addition, histone tails can also be methylated at arginine residues by a different class of enzymes, that act as coactivators of transcription (Chen et al., 1999).

A “histone code” hypothesis has been proposed, suggesting that distinct covalent histone modifications might be used by the cell, sequentially or in combination, to generate a “code” that could be read by other proteins to produce different transcriptional outputs (Strahl and Allis, 2000). Reading the “histone code” would necessitate protein domains that recognize, in a receptor-ligand type of interaction, the different covalent modifications that can occur on histones (Strahl and Allis, 2000). Binding activities have been identified in several of the protein domains that are frequently found in chromatin-related proteins, such as the bromodomain and the chromodomain, which can recognize acetylated- and methylated-lysine residues of the histone tails, repectively (Table 3). A further level of complexity in regulatory mechanisms is inferred from the observation that the same covalent modifications that can be found on histones also occur on other proteins involved in transcriptional control. For example, histones are not the only targets for HATs, as HAT-catalyzed acetylation can also regulate the activity of transcription factors and co-factors (reviewed in: Sterner and Berger, 2000; Chen et al., 2001b). Lastly, another group of enzymes involved in chromatin remodeling is that of the DNA-dependent ATPases of the SWI2/SNF2 type. Yeast SWI2/SNF2 is the catalytic subunit of the multiprotein SWI/SNF remodeling complex, which can mediate the repositioning of nucleosomes by sliding histone octamers to other sites on the same DNA molecule, as well as by transferring them to other DNA molecules (reviewed in: Vignali et al., 2000; Flaus and Owen-Hughes, 2001).

A catalogue of known and putative chromatin proteins in Arabidopsis and maize has been compiled in ChromDB, a database that aims to present information on the entire complement of chromatin proteins in plants ( http://chromdb.biosci.arizona.edu/). ChromDB lists over 220 different Arabidopsis chromatin genes, including SWI2/SNF2 homologs (22 genes), HATs (12 genes; 10 are listed as HATs and 2 as TAF_II250 homologs; Table 4), HDACs (17 genes; Table 4), and SET-domain-protein genes (29 genes), and also includes histones (50 genes) and homologs of subunits of global transcription factors.

The definition and identification of the complement of chromatin proteins in Arabidopsis, or in any other eukaryotic organism, and in particular of the subset of those proteins that might be involved in transcriptional control, is complicated by several factors. First, chromatin remodeling is mediated by multiprotein complexes, some of which have already been purified and characterized from yeast and animal (mammalian, Drosophila) cells, but none from plants. Some of these complexes (for example, the yeast SAGA and human PCAF complexes) show a remarkable conservation in subunit composition, but there are also cases of proteins and complexes that are specific to one kingdom (Sterner and Berger, 2000, and see below). Thus, biochemical studies will be needed to obtain a complete description of the Arabidopsis complement of chromatin proteins. Another complication for the identification of bona fide chromatin proteins arises from their multi-domain architecture. Chromatin proteins frequently combine different domains or motifs of distinct molecular functions (Table 3). However, those domains are not necessarily unique to chromatin proteins, as they (or related sequences) can be present in other types of proteins. For example, an Arabidopsis protein that contains sequences related to the chromodomain is localized to the chloroplast and forms part of the chloroplast signal recognition particle pathway (Klimyuk et al., 1999). Lastly, the structure of chromatin influences not only transcription, but also other nuclear processes that are physically associated with the genome, such as replication, recombination, and DNA repair. Thus, that a protein is chromatin-related does not necessarily imply that it is involved in transcriptional control. For these different reasons, the identification and description by sequence similarity searches of the complement of chromatin proteins involved in transcriptional regulation, and of their biochemical and molecular functions, is more complicated than that of the sequence-specific DNA-binding transcription factors discussed above.

It is apparent from the content in known chromatin genes of the Arabidopsis genome that chromatin remodeling is important in plants for the control of gene expression. That some of the molecular mechanisms for chromatin reconfiguration and transcriptional control are conserved among plants and the other eukaryotic kingdoms can be deduced from the presence of orthologous genes. Furthermore, similarities or functional equivalence at the molecular or physiological level has been demonstrated in some cases, as illustrated with the following examples.

An RPD3-type maize histone deacetylase has been shown to complement a S. cerevisiae null mutant in the homologous RPD3 gene (Rossi et al., 1998).

The Arabidopsis gene BUSHY (BSH), which codes for a protein with high sequence similarity to S. cerevisiae SNF5 (a component of the SWI/SNF remodeling complex), can partially complement a snf5 mutation in yeast (Brzeski et al., 1999).

Arabidopsis homologs of human CBP/p300 proteins recapitulate the binding specificity of p300 for the adenoviral oncoprotein E1A, in addition to being capable of activating transcription in mammalian cells (Bordoli et al., 2001).

The Arabidopsis protein that is orthologous to yeast GCN5 possesses HAT activity, and can interact with Arabidopsis ADA2 proteins, suggesting that a complex analogous to yeast SAGA (of which GCN5 and ADA2 form part) and human PCAF also exists in plants (Stockinger et al., 2001).

PICKLE (PKL; also initially referred to as GYMNOS) is an Arabidopsis protein of the SWI2/SNF2-type that appears to be involved in the repression of several important developmental regulators, such as LEC1 and meristematic genes (Eshed et al., 1999; Ogas et al., 1999). PKL is homologous to human Mi-2, a component of the NuRD complex. By virtue of its different subunits, the NuRD complex combines both ATP-dependent chromatin remodeling and HDAC activity. The homology between PKL and Mi-2 suggests that a NuRD-like complex might exist in plants; thus, a plausible mechanism for gene repression by PKL is via histone deacetylation mediated by NuRD (reviewed in Ahringer, 2000).

The Polycomb group (PcG) and the trithorax group (trxG) of proteins in Drosophila and mammals control the cellular inheritance of mitotically stable states of gene expression, homeotic genes in particular. PcG and trxG proteins (repressors and activators, respectively) are thought to regulate transcription by modulating the structure of chromatin (reviewed in: Brock and van Lohuizen, 2001; Francis and Kingston, 2001; Mahmoudi and Verrijzer, 2001). The proteins within each group (Pc or trx) can be unrelated in sequence; rather, their relationship to each other comes from the fact that they operate together in the form of multi-subunit complexes of a genetically defined function (Gould, 1997). Homologous or related proteins for some PcG and trxG factors have been identified in Arabidopsis, and in some cases functionally characterized. Three Arabidopsis proteins show homology to the Drosophila SET-domain PcG protein Enhancer of zeste (E(z)), CURLY LEAF (CLF), CURLY LEAF LIKE (CLK), and MEDEA (MEA) (Goodrich et al., 1997; Grossniklaus et al., 1998) (Table 5). CLF is a repressor of floral organ identity (i.e., homeotic) gene expression in vegetative tissues (Goodrich et al., 1997), whereas MEA is involved in the maternal control of embryogenesis (Grossniklaus et al., 1998). Another Arabidopsis PcG protein involved in seed development is FERTILIZATION-INDEPENDENT ENDOSPERM (FIE), which shows homology to PcG proteins with WD repeats, such as Drosophila extra sex combs (esc) (Ohad et al., 1999). Animal E(z) and esc proteins have been shown to interact and to co-localize in unique complexes. Similarly, Arabidopsis FIE and MEA also interact, which provides a molecular explanation for the similarities between the fie and mea mutant phenotypes (Spillane et al., 2000; Yadegari et al., 2000). Other Arabidopsis proteins, such as EMBRYONIC FLOWER2 (EMF2), FERTILIZATION-INDEPENDENT SEED 2 (FIS2), and VERNALIZATION 2 (VRN2) are related to a different Drosophila PcG protein, Suppressor of zeste 12 (Su(z)12) (Luo et al., 1999; Birve et al., 2001; Gendall et al., 2001; Yoshida et al., 2001).

Despite these similarities, however, novel features in the chromatin-mediated regulation of gene expression have also evolved in plants. Plants contain what appears to be a kingdom-specific family of histone deacetylases, the HD2 class (Lusser et al., 1997; Aravind and Koonin, 1998; Dangl et al., 2001) (Table 4). Orthologs for some Arabidopsis chromatin proteins are not found in yeast or animals. This is the case, for example, of MOM1, a SWI2/SNF2-related protein that is involved in the maintenance of transcriptional gene silencing (Amedeo et al., 2000; Arabidopsis Genome Initiative, 2000). In addition, homologous chromatin proteins can show structural variation among the different eukaryotic kingdoms, and some of those variations appear to be specific to plants. In fact, eukaryotic chromatin proteins are a prominent example of evolutionary innovation by domain shuffling, deletion, and accretion (International Human Genome Sequencing Consortium, 2001). For example, Arabidopsis CBP/p300-like proteins lack the bromodomain and the CREB-binding region that are highly conserved in animal CBP/p300 proteins (Bordoli et al., 2001) (Table 4; CBP/p300 proteins are not found in yeast). Instead, one of these Arabidopsis proteins (PCAT1) contains a repeated motif of unknown function that does not show sequence similarity to any other known amino acid motif (Bordoli et al., 2001).

Other Arabidopsis chromatin genes that have already been genetically or functionally characterized further show the importance of chromatin-mediated regulation of gene expression in multiple aspects of the plant life cycle (Table 5). Reduction of AtHD1 (an HDAC-coding gene, also referred to as AtRPD3A) transcript levels by using antisense RNA caused pleiotropic developmental alterations, suggesting a global role for AtHD1 in regulating gene expression during development (Wu et al., 2000a; Tian and Chen, 2001). Similarly, reduction of AtHD2A activity (which codes for an HDAC of the plant-specific HD2 class) resulted in aborted seed development (Wu et al., 2000b). In another study, mutants in the HDAC gene AtHDA6 were isolated, which were morphologically wild-type but showed deregulated expression of transgenes, suggesting that AtHDA6 might be specifically involved in (transgene) silencing processes (Murfett et al., 2001). In addition to MOM1 and PKL, mentioned above, another Arabidopsis gene coding for a SWI2/SNF2-type protein that has been functionally characterized is DECREASE IN DNA METHYLATION1 (DDM1). DDM1 is required to maintain normal cytosine methylation patterns and to stabilize transposon behavior (Jeddeloh et al., 1999; Miura et al., 2001; Singer et al., 2001).

In summary, genetic studies on a variety of biological processes in Arabidopsis, the determination of its genome sequence, and biochemical studies performed in maize, have all started to illuminate the different physiological functions that chromatin remodeling might play in plants. However, our understanding of chromatin remodeling at the molecular level, and on how it influences plant nuclear processes, is extremely limited, and mostly derived from comparisons with the better-studied systems of yeast, Drosophila, and mammalian cells. If chromatin research in these model organisms is to be viewed as an example, it is clear that biochemical studies will be essential to understand chromatin in plants.

5. The combinatorial nature of transcriptional regulation: promoters, cis-elements and trans-acting factors.

Whereas plants and animals (or, to be more precise, Arabidopsis and Drosophila, C. elegans, and humans) might have comparable contents of transcription factors (3.5–6.6% of the total number of genes; see above), the organization of the regulatory sequences on which these transcription factors act can be different in the two kingdoms. In animals, the regulatory sequences that determine the correct temporal and spatial expression of a gene can extend over tens of kilobases (kbs) of DNA (for a review, see Bonifer, 2000). In contrast, regulatory sequences of plant genes usually span much shorter DNA intervals, often less than 1 or 2 kbs. This is reflected in the compact organization of the Arabidopsis genome, in which gene density is high. Out of the sequenced 115.4 megabases (Mb) of the 125 Mb genome, 51.2 Mb (or 44% of the sequenced regions) correspond to predicted exons and introns (Arabidopsis Genome Initiative, 2000). On average, there is one gene per 4.5 kb of DNA: the gene length (exons plus introns) is approximately 2 kb, and ∼2.5 kb correspond to intergenic regions. Considering the whole genome, transposons account for ∼20% of the intergenic DNA, resulting in an average of 2 kb of DNA for the 5′ and 3′ regions of a particular gene (Arabidopsis Genome Initiative, 2000). Other plants, maize for example, have genomes that are much larger than that of Arabidopsis, but with a similar organization of promoter sequences: in the maize genome, active genes are usually distributed in compact gene-rich islands, with much of the genomic DNA corresponding to repetitive sequences made up of retrotransposons (SanMiguel et al., 1996; Fu et al., 2001). As a result, regulatory sequences in Arabidopsis, and in plants in general, are easier to identify and delimit experimentally than in, for example, humans (for an introduction to the problem in mammals, see: Gumucio et al., 1993; Hardison et al., 1997; Bonifer, 2000; Fickett and Wasserman, 2000; Scherf et al., 2001). Compact Arabidopsis 5′ promoter sequences often recapitulate faithfully the expression of the native gene when assayed in transgenic plants by reporter gene fusions, that is, in a chromatin context. However, this is not always the case, because regulatory elements can also be localized downstream of the transcription start site: in introns, in the 5′ untranslated region, or in 3′ sequences (Larkin et al., 1993; Sieburth and Meyerowitz, 1997; Deyholos and Sieburth, 2000; Yu et al., 2001). For example, the large second intron of the MADS-box floral organ-identity gene AGAMOUS (AG) is essential for the correct expression of the gene, and contains binding sites for at least two AG regulators, LFY and WUSCHEL (WUS) (Sieburth and Meyerowitz, 1997; Bomblies et al., 1999; Busch et al., 1999; Deyholos and Sieburth, 2000; Lohmann et al., 2001).

In spite of the structural differences between animal and plant cis-regulatory and promoter regions, regulation of gene expression is often in both cases the result of multiple inputs, reflecting, or taking advantage of, the combinatorial nature of the mechanisms of eukaryotic transcription. Multiple stimuli can converge through different cis-acting elements on a promoter to coordinately regulate the expression of the corresponding gene (Arnone and Davidson, 1997; Yuh et al., 1998). The cis-acting elements are usually organized in a modular fashion: both in animals and plants, the regulatory region of a gene can be partitioned into discrete subelements, each one containing one or several binding sites for transcription factors and performing a certain regulatory function (Benfey and Chua, 1990; Arnone and Davidson, 1997). The modular nature of cis-regulatory systems is exemplified by the 2.3 kb promoter region of the sea urchin developmentally regulated Endo16 gene, one of the best characterized eukaryotic promoters (Yuh et al., 1998, 2001). It consists of six different regulatory modules, which provide different regulatory functions that are integrated through interrelations between the modules, and result in the spatial expression, and repression, of the gene, as well as on its variable rates of transcription (Yuh et al., 1998, 2001). This cis-regulatory system therefore acts like an information processing unit, and computational models for the modes of action of some of its modules have been established (Arnone and Davidson, 1997; Yuh et al., 1998, 2001). The view of cis-regulatory regions as information processing systems in which the output of developmental (or other) inputs is hardwired, is probably applicable to the eukaryotic genome as a whole (Arnone and Davidson, 1997; Davidson, 2001).

In plants, regulation of gene expression by systems of cis-acting modules, and the fact that these modules can interact synergistically (i. e., that combinations of modules direct gene expression in a manner not observed with the modules in isolation), was first described for the cauliflower mosaic virus (CaMV) 35S promoter (Benfey and Chua, 1990). The CaMV 35S promoter directs high levels of expression in most tissues and developmental stages when introduced as a transgene in plants, but can be dissected into subdomains that confer tissue-specific expression (Benfey and Chua, 1990; Benfey et al., 1990a, b).

The combinatorial interaction of cis-elements has also been demonstrated, for example, for Arabidopsis light-regulated promoters. Several consensus cis-sequences that are necessary for high activity in the light have been identified in the promoters of photosynthesis-associated nuclear genes (such as the rbcS and cab genes). These consensus sequences are referred to as ‘light responsive elements (LREs)’. Minimal promoters, sufficient to confer light-dependent expression, contain several LREs, but no single LRE is found in all light-regulated promoters (in fact, some LREs are also present in promoters that are not regulated by light) (Argüello-Astorga and Herrera-Estrella, 1998). LREs function combinatorially: whereas they cannot confer proper light responsiveness in isolation, paired combinations of them are able (1) to respond to a wide spectrum of light through the phytochrome signal transduction pathways, (2) to respond to the chloroplast developmental state, and (3) to confer a photosynthetic-cell specific expression pattern, therefore satisfying the strict definition of light-inducible (Puente et al., 1996; Chattopadhyay et al., 1998b). Thus, it is the combination of LREs in the promoter what serves as the integration system for the coordination of different light and developmental inputs to regulate the expression of the photosynthesis-related genes (Puente et al., 1996; Chattopadhyay et al., 1998b). Similarly, the promoter of the meristem identity gene LEAFY serves as the convergence point for different signals that control flowering time in Arabidopsis, including both environmental cues (daylength pathway) and endogenous signals (gibberellins) (Blázquez and Weigel, 2000), in accordance with the concept of promoters acting as information processing systems.

The combinatorial and synergistic function of cis-elements in eukaryotic promoters is logically accompanied by the combinatorial mode of action of the trans-acting factors that bind to those sites, and allows for the generation of regulatory diversity by a limited number of factors and binding sites. The requirement of several, often adjacent, cis-elements for the regulation of gene expression can be related to direct interactions between the proteins that bind to those elements. Direct interactions among transcription factors, however, is not the only molecular mechanism by which they can function combinatorially to regulate gene expression, since they can also interact with other components of the transcription machinery and with other classes of regulatory proteins. For example, LFY and WUS cooperatively participate in the regulation AG expression, yet they bind independently to AG cis-regulatory sequences and a direct interaction between the two proteins has not yet been detected (Lohmann et al., 2001).

Several examples of direct interactions between different Arabidopsis transcription factors have been reported, although the number is still small. In addition to increasing the regulatory repertoire, direct interactions between transcription factors are one of the mechanisms by which proteins with very similar DNA binding domains might achieve regulatory specificity (see, for example, Grotewold et al., 2000). Direct interactions can occur between members of the same protein family, to form dimeric complexes that bind to palindromic DNA sequences (such as in the case of the MADS domain proteins: Huang et al., 1996; Riechmann et al., 1996a; Riechmann et al., 1996b), or between transcription factors of different families. Examples of the latter include Arabidopsis, maize, and petunia proteins of the MYB and bHLH families (Table 6), interactions between bZIP and ABI3/VP1 proteins in rice and Arabidopsis (Hobo et al., 1999; Nakamura et al., 2001), between soybean C2H2 zinc finger and bZIP proteins (Kim et al., 2001), and between Dof and bZIP transcription factors in Arabidopsis and maize (Chen et al., 1996; Vicente-Carbajosa et al., 1997). The interaction between bZIP and ABI3/VP1 proteins (TRAB1 and VP1, respectively, in the case of rice, and ABI3 and ABI5 in Arabidopsis) provides a mechanism for VP1-mediated, ABA-inducible gene expression (Hobo et al., 1999; Nakamura et al., 2001). The interaction between bZIP and Dof proteins might mediate the endosperm-specific expression of seed-storage proteins (Vicente-Carbajosa et al., 1997).

Within the MADS domain family, interactions are not limited to the formation of protein dimers, but also include the formation of ternary complexes. APETALA1 (AP1), APETALA3 (AP3), PISTILLATA (PI), and AGAMOUS (AG) are MADS domain proteins that, together with AP2, control the development of floral organs in Arabidopsis (Bowman et al., 1991; Coen and Meyerowitz, 1991; Goto et al., 2001; Jack, 2001; Theiben, 2001). AP3 and PI bind to DNA forming a heterodimer, whereas AP1 and AG can both bind to DNA as homodimers or as heterodimers with other MADS domain proteins (Huang et al., 1996; Riechmann et al., 1996a; Riechmann et al., 1996b). The activity of AP1, AP3, PI, and AG, however, requires of floral cofactors, also MADS domain proteins, that are encoded by the SEPALLATA genes, SEP1, SEP2, and SEP3 (Pelaz et al., 2000; Honma and Goto, 2001; Pelaz et al., 2001a; Pelaz et al., 2001b). In yeast two-hybrid experiments, AP3 and PI together, but neither one of the two proteins individually, can physically interact with AP1 and with SEP3 (Honma and Goto, 2001). The ectopic expression of AP3, PI, and AP1, or of AP3, PI, and SEP3 converts vegetative leaves into petaloid organs (Honma and Goto, 2001), whereas AP3 and PI alone are not sufficient for such organ conversion (Krizek and Meyerowitz, 1996). These results indicate that the formation of ternary complexes might be necessary for the function of AP3 and PI. The role that ternary complex formation might play in AP3 and PI function could be several fold: from providing an activation domain that AP3 and PI appear to lack, but that AP1 and SEP3 have (Honma and Goto, 2001), to increasing the DNA-binding specificity/affinity of the complex versus that of the protein dimers, given that the organ identity activity of AP1, AP3, PI, and AG is independent of their individual DNA-recognition properties (Riechmann and Meyerowitz, 1997a). These results with the Arabidopsis floral organ identity proteins parallel and expand previously obtained data for the Antirrhinum majus MADS-domain proteins SQUAMOSA, DEFICIENS, and GLOBOSA (SQUA, DEF, and GLO, which are AP1, AP3, and PI orthologs, respectively). DEF and GLO were also found to form ternary complexes with SQUA, and the three proteins together to bind DNA with increased affinity versus SQUA or DEF/GLO alone (Egea-Cortines et al., 1999).

Whereas these isolated examples illustrate the importance of interactions between transcription factors for the regulation of transcription, and how the combinatorial logic can operate, they do not convey the scope of the regulatory interactions in which transcription factors could be involved. For this, the whole complement of proteins should be considered (see below).

6. Genome-wide analyses of transcriptional regulation.

The future of biological sciences in the “post-genome era” has been anticipated as an endeavor to generate a collection of comprehensive “functional maps” (corresponding to the “transcriptome”, the “phenome”, the “interactome”, the “localizome”, and so on), that would be compiled into a “biological atlas” which would represent the modular nature of biological processes in a holistic manner and allow the formulation of new hypothesis (Greenbaum et al., 2001; Kim, 2001; Vidal, 2001). These maps could be visualized as two-dimensional matrices in which one axis represents all the genes or proteins that can be tested in an organism, and the other a comprehensive series of mutant backgrounds, conditions to which the organism can be exposed, etc. (Vidal, 2001). For instance, a yeast transcriptome map of this type is already being developed (Hughes et al., 2000b). The interactome would represent the map of physical interactions among all the proteins of a proteome (reviewed in Walhout and Vidal, 2001) (for attempts to construct the yeast interactome, see: Uetz et al., 2000; Ito et al., 2001). The localizome map would describe in what cells and cellular compartments, and when, all the proteins of an organism's proteome can be found; and to produce the phenome, a collection of mutants encompassing all the genes in a genome would be screened in a large series of phenotypic assays (for C. elegans and yeast, see: Ross-Macdonald et al., 1999; Fraser et al., 2000; Gönczy et al., 2000; Maeda et al., 2001).

Such view is also appropriate when considering Arabidopsis transcriptional regulation at a global level in a cellular and organismal context, for whose understanding several of those functional maps would be required: the genome-wide transcriptome map, the interactome and the phenome of the transcriptional regulators, as well as other “-ome” maps not previously considered, such as the promoterome. Intrinsic to this view is the realization that none of these different “-ome” maps would lead, in isolation from the others, to a comprehensive or even logical understanding of transcriptional regulation and of its role as a major determinant of cellular and organismal functions and phenotypes.

To generate these functional maps, the systematic investigation of transcription factor function and transcriptional regulation in Arabidopsis can be pursued with a variety of tools for functional genomic analyses, including reverse genetics methods, gene expression profiling experiments, and protein-protein interaction screens (reviewed in Riechmann and Ratcliffe, 2000). Whereas the availability of the Arabidopsis genome sequence allows us to compile lists of proteins that are involved in the regulation of transcription, and of putative promoter and cis-acting sequences, a global understanding of this process is still in its infancy. However, somewhere along the way of generating these functional maps, and once a sufficient amount of data has been collected, it should be possible to start decoding the “language” of transcriptional control, and to eventually be able, for instance, to build synthetic promoters directing gene expression in novel, designed spatial and temporal patterns (for an example of an initial attempt to design an artificial expression cassette in plants simply by statistical analysis of nucleotide sequences, see Sawant et al., 2001).

Conclusion

Genomics research, and functional genomics in particular, has often been hailed as the provider of a new paradigm in biology research; it has also been sometimes reviled by describing such type of research as consisting of little else than “fishing” experiments. Both opposing views are based on one common premise: the contrast between a frequently hypothesis-free genomics and the hypothesis-driven research that has been so much favored in molecular biology over the past decades. But genomics might not fit either one of these two disparate views. In many aspects, genomics bears many similarities with classic biology disciplines, such as genetics (in which the whole genome is blindly mutagenized at random to “fish” for interesting, novel phenotypes), and taxonomy and phylogeny (in which lists of elements are compiled and the relationships between them have to be established). It is in many instances the methodic collection of unanticipated data what allows the formulation of new hypothesis. However, if not radically different in concepts, genomics certainly changes the scales of biology research, and provides new dimensions to it. The relative simplicity of the Arabidopsis thaliana genome, together with the availability of many modern genetic and genomic research tools in that species, indicates that Arabidopsis is a premier organism to elucidate the complex logic of transcription at a genome-wide level in multicellular eukaryotes.