MBE Advance Access originally published online on May 26, 2006
Molecular Biology and Evolution 2006 23(8):1480-1492; doi:10.1093/molbev/msl022
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Review |
Cross-Species Annotation of Basic Leucine Zipper Factor Interactions: Insight into the Evolution of Closed Interaction Networks
* Department of Biological Sciences, Purdue University and Department of Neuroscience, Johns Hopkins Medical School
E-mail: deppmann{at}jhmi.edu.
 |
Abstract |
|---|
 TOP Abstract IntroductionMethodology of Predicting Dimer...A New Approach for...Conclusions and PerspectiveSupplementary MaterialAcknowledgementsReferences |
|---|
Dimeric basic leucine zipper (bZIP) factors constitute one of the most important classes of enhancer-type transcription factors. In vertebrates, bZIP factors are involved in many cellular processes, including cell survival, learning and memory, cancer progression, lipid metabolism, and a variety of developmental processes. These factors have the ability to homodimerize and heterodimerize in a specific and predictable manner, resulting in hundreds of dimers with unique effects on transcription. In recent years, several studies have described dimerization preferences for bZIP factors from different species, including Homo sapiens, Drosophila melanogaster, Arabidopsis thaliana, and Saccharomyces cerevisiae. Here, these findings are summarized as novel, graphical representations of closed, interacting protein networks. These representations combine phylogenetic information, DNA-binding properties, and dimerization preference. Beyond summarizing bZIP dimerization preferences within selected species, we have included annotation for a solitary bZIP factor found in the primitive eukaryote, Giardia lamblia, a possible evolutionary precursor to the complex networks of bZIP factors encoded by other genomes. Finally, we discuss the fundamental similarities and differences between dimerization networks within the context of bZIP factor evolution.
Key Words: bZIP • dimer • transcription • DNA binding
|
Introduction |
|---|
|
TOPAbstractIntroductionMethodology of Predicting Dimer...A New Approach for...Conclusions and PerspectiveSupplementary MaterialAcknowledgementsReferences |
|---|
Basic leucine zipper (bZIP) proteins are an exclusively eukaryotic class of enhancer-type transcription factor. Across species, bZIP factors are involved in many processes that are critical to the function of an organism. During animal embryogenesis, for example, bZIP factors are necessary for the proper development of organs and tissues such as the liver, bone, heart, and fat (Wang et al. 1992
; Darlington et al. 1998; Eferl et al. 1999). In adult animals, bZIP factors are involved in processes such as metabolism, circadian rhythm, and learning and memory (Darlington et al. 1995, 1998; Yamaguchi et al. 2000; Sanyal et al. 2002). These factors play similarly critical roles not only in animals but also in other phylogenetic kingdoms. In plants, bZIP factors are important for seed development and flower maturation (Jakoby et al. 2002), and in yeast, bZIP factors are necessary for sexual differentiation and entry into stationary phase (Takeda et al. 1995; Watanabe and Yamamoto 1996). In addition to these species-specific effects, more generalized functions of bZIP factors have also been described, such as the initiation of cellular responses to UV damage or osmotic stress (Hai et al. 1999; Toone et al. 2001). The conserved presence of bZIP factors across eukaryotes, together with their critically important role in a myriad of cellular functions, underscores the importance of this class of enhancer-type transcription factors.
The leucine zipper motif that characterizes these proteins is classically defined as having leucine residues spaced 7 amino acids apart repeating at least 3 times (LxxxxxxLxxxxxxL) (Landschulz et al. 1988). Upon dimerization, parallel leucine zipper motifs interact through a coiled coil, hydrophobic interface that juxtaposes 2 adjacent basic regions (Vinson 1989; O'Shea et al. 1991; Pu and Struhl 1991; Ellenberger et al. 1992). Interestingly, once bound to DNA, these adjacent basic regions undergo a major change in conformation (O'Neil and DeGrado 1990; Patel et al. 1990; Weiss et al. 1990). Collectively, these basic regions constitute a motif that interacts with eukaryotic promoter DNA in a sequence-specific manner (Hill et al. 1986; Oliphant et al. 1989). Given that bZIP domains are so stereotyped, the fundamental question arises: How do these factors influence such a broad range of cellular functions? The answer to this question has to do with a bZIP factor's dimerization and DNA-binding preferences as well as its transactivation (or repression) properties. Therefore, unique pairings of bZIP factors often result in unique pairings of DNA-binding preferences and transactivation domains. Accordingly, the larger the array of bZIP factors in a given genome, the greater the potential for complex transcriptional programs affecting the unique functions of individual cells, tissues, organs, and the species itself.
The notion that every bZIP dimer encodes a unique function emphasizes the need to define all potential bZIP dimers. The human genome contains 56 genes encoding proteins with bZIP motifs (Tupler et al. 2001). If one assumes that all bZIP factors encoded within the human genome are capable of interacting with each other, then there is the potential to generate approximately 1600 different dimers, all with a unique effect on transcriptional response. Of course, interaction between bZIP factors is much more discriminating than this theoretical situation. We calculate based on experimental and predictive data that there are actually about 340 unique dimers in humans, approximately 5-fold less than if bZIP dimerization occurred promiscuously. Nevertheless, even this restricted interaction generates a generous spectrum of transcriptional activities and allows for incredible flexibility in how these factors contribute to genetic programs. A similar scenario almost certainly applies to bZIP proteins encoded by other genomes.
The goal of this review is to compile experimental and predictive data published in the last several years in order to provide a comprehensive summary of unique bZIP dimers from several phylogenetic kingdoms. Previous evaluations of such complex dimer networks have resulted in cumbersome and complicated tables that have proven difficult to interpret (Deppmann et al. 2004; Bornberg-Bauer et al. 2005). In order to simplify these data, we have developed a network mapping approach that summarizes the dimerization potentials of all bZIP factors encoded by a given genome. Using this graphical convention, we present a comparative analysis of the bZIP dimerization networks from Homo sapiens, Arabidopsis thaliana, Drosophila melanogaster, and Saccharomyces cerevisiae. In addition to providing a useful tool for investigators studying bZIP factors, these analyses reveal fundamental differences in the organization of each of these networks. Among these differences is the observation that animal bZIP factor dimerization networks extensively heterodimerize, whereas the plant dimerization networks almost exclusively homodimerize. It is clear that by comparing dimerization networks, additional insight into the evolution of bZIP factor interactions can be obtained.
|
Methodology of Predicting Dimer Formation |
|---|
|
TOPAbstractIntroductionMethodology of Predicting Dimer...A New Approach for...Conclusions and PerspectiveSupplementary MaterialAcknowledgementsReferences |
|---|
Although there is no substitute for experimentation, predictive analysis serves to rationalize previously observed bZIP interactions (e.g., the cAMP responsive element binding protein homodimer [CREB]) and to focus investigators on potentially unexpected interactions. In recent years, the biophysical studies from several groups have clearly defined the rules governing bZIP interactions (O'Shea et al. 1992
; Krylov et al. 1994, 1998; Keating et al. 2001; Acharya et al. 2002). Using these rules, we are now able to predict with high certainty if 2 bZIP factors will interact (Fong et al. 2004; Grigoryan and Keating 2006). Indeed, predictive and experimental analyses of bZIP interactions in the human proteome are in nearly complete agreement (Vinson et al. 2002; Newman and Keating 2003; Fong et al. 2004). Given that much of the annotation presented in this review is derived from predictive data, we will begin by describing the methodology used to generate these predictions.
Conventions Used to Describe Leucine Zipper Interactions
A leucine zipper is an 
-helical stretch of amino acids in which leucines are positioned every 7 residues (Landschulz et al. 1988). One repeating unit of a leucine zipper is referred to as a heptad. The positions within the heptad are designated a, b, c, d, e, f, and g, with the characteristic leucine residue occurring in the d position. In terms of amino acid sequence, the heptad begins with position g and is followed by positions a, b, c, d, e, and f. Positions a and d reside in the hydrophobic dimerization interface, whereas positions b, c, and f reside on the hydrophilic face of these amphipathic helices. In the 3-dimensional structure of the -helix, the e and g positions flank the hydrophobic face of the molecule and can impact leucine zipper dimerization through electrostatic interactions (Vinson et al. 1989; O'Shea et al. 1991, 1992).
The Hydrophobic Dimerization Interface
As mentioned, amino acids present at the a and d positions make up the hydrophobic dimerization interface. In humans, the d position is almost invariantly a leucine; however, the a position is much more variable (Deppmann et al. 2004). This variability is in large part responsible for determining whether a leucine zipper will prefer homodimerization or heterodimerization. Interestingly, the key residues affecting dimerization preference at this site are not hydrophobic amino acids but rather are lysine and asparagine. In general, asparagine at the a position favors homodimer formation, whereas lysine at this position favors heterodimer formation (Acharya et al. 2002). The reasons for this are relatively straightforward. Heterotypic interaction between asparagine and most other residues in the a position is thermodynamically unfavorable, whereas interaction between asparagine and another asparagine or lysine is mildly favorable. At this position, it is not the favorable interactions that determine the ability of leucine zippers to homo- or heterodimerize, rather the unfavorable interactions that limit its possibilities. This notion is perhaps best illustrated by examples of lysine at the a position. The reason that lysine at the a position favors heterodimerization is not because it has great affinity toward other factors but because this homotypic electrostatic interaction is so unfavorable. However, a heterotypic interaction between lysine and any of the hydrophobic residues is mildly favorable, resulting in an apparent preference for heterodimerization. The contribution of several amino acids to the stability and specificity of homotypic and heterotypic a–a interactions has been examined previously and is the subject of ongoing investigation (Acharya et al. 2002).
Electrostatic Interactions
Charged amino acids at the g and e positions contribute to leucine zipper stability as well as regulate dimerization specificity (Alber 1992; Baxevanis and Vinson 1993). Electrostatic interactions between the g position amino acids on a given leucine zipper and the e position amino acids on opposing leucine zippers result in either attractive or repulsive interactions. These attractive or repulsive g–e interactions can mediate either homodimerization or heterodimerization. Interactions between acidic and basic residues of the g–e positions are generally favorable (e.g., E 
R, E K, D R, R E, K E, and K D). Interactions where both the g and e positions contain acidic (or basic) amino acids are unfavorable (e.g., E E, E D, K K, and R K). Interactions between glutamine and charged residues also are considered unfavorable (e.g., E Q, Q E, Q K, R Q, and K Q). Lastly, the Q Q interaction has a coupling energy of 0 but imparts increased stability to a dimer (Krylov et al. 1998).
By using previously reported thermodynamic properties of a–a and g–e interactions (some of which are reviewed above), it is possible to predict if 2 bZIP factors will interact and/or to rationalize dimerization that has been established through direct experimentation. Because the proof of principle for this has been demonstrated on a genome-wide scale several times (Fassler et al. 2002; Vinson et al. 2002; Deppmann et al. 2004), the current challenge is to annotate these interactions.
|
A New Approach for Representing Closed Interaction Networks |
|---|
|
TOPAbstractIntroductionMethodology of Predicting Dimer...A New Approach for...Conclusions and PerspectiveSupplementary MaterialAcknowledgementsReferences |
|---|
In the last few years, there have been several comprehensive studies describing bZIP interactions in several different species (Fassler et al. 2002
; Vinson et al. 2002; Newman and Keating 2003; Deppmann et al. 2004). Although these studies have been extremely useful by providing a complete annotation of the factors and their dimerization preferences, the tabular form in which the data are organized is challenging to interpret for those unfamiliar with the field. Therefore, to make the complexity of these networks more accessible, we use a graphical convention to represent these data (fig. 1). All bZIP factors encoded by a given genome are arranged around a circle. The shape of a particular bZIP factor is based on its predicted dimerization properties (black circle for homodimer only, gray square for homodimer or heterodimer, white diamond for heterodimer only). Factors with highly similar leucine zippers and dimerization properties are grouped, and in the majority of cases, their proteins are considered paralogs. If the factors within such a group are predicted to dimerize (quasihomodimers), the group is bracketed. Dimerization between nonparalogous factors (or groups of factors) are considered heterodimers and are connected by a line. Because leucine zippers within paralogous groups are similar, it is expected that these factors will share dimerization preferences. However, even highly similar leucine zippers could have subtle differences in dimerization preference. For the human network, these subtle differences are represented in a previously published experimentally derived interaction grid (Newman and Keating 2003). For the other networks presented in this review, similar insight can be gleaned from predictive interaction grids (described below). For several of the circular maps presented here, the inclusion of tinted, semitransparent ovals in the background of the map indicates the consensus DNA-binding sequences preferred by a particular collection of dimers. The details of this added feature on the interaction maps will be discussed together with the human bZIP proteins. It is important to note that although these maps are as comprehensive as possible, they are "works in progress" and will require updates as new dimerization or DNA-binding information becomes available.
|
|
In some cases, we exclude possible, but improbable, dimers from the network map. However, we acknowledge that weak interactions may be of interest to investigators who are studying bZIP factors under special circumstances such as 2 monomers expressed at very different molar concentrations. Therefore, beginning with D. melanogaster, we provide a grid summarizing the complex predictive data set for interaction between all bZIP factors in a genome (fig. 2B). Data in these grids are compiled by summing the previously published thermodynamic coupling energies of a–a and g–e interactions across all relevant heptads (Krylov et al. 1994
, 1998; Acharya et al. 2002; Deppmann et al. 2004). We assign all potential dimers a color representing strong (black), medium (crosshatched), weak (dark gray), or repulsive (light gray). A detailed explanation of how these grids are generated can be found in supplementary material (see Supplementary Material online). This methodology agrees with a recently published online predictive tool (http://compbio.cs.princeton.edu/bzip/) (Fong et al. 2004). We present a grid for each species, except Homo sapiens, where an interaction grid has already been published for human bZIP proteins based on experimentation (Newman and Keating 2003; Fong et al. 2004).
|
Although the dimerization maps presented in this review are based on experimentation whenever available, the accuracy of predicting interactions is limited. This point was recently addressed by Singh and colleagues who performed an analysis of the reliability of bZIP dimerization predictions by comparing experimental and predictive data for human bZIP factors (Fong et al. 2004
). Using their methodology, which agrees well with that used here, they found that prediction is reliable 92% of the time. Of course, for the purposes of annotation, this number is almost certainly higher because both prediction and published experimental data are used. The reliability of predictive analysis should increase if and when coupling energies for rare a–a, g–e, and d–d interactions are defined. Another limitation of prediction as well as many biophysical studies is that these analyses do not take in to account interaction outside of the bZIP domain or the effects of posttranslational modifications such as phosphorylation, which is thought to destabilize dimerization (Szilak et al. 1997).
The Human Dimerization Network
The human genome encodes 56 bZIP proteins, many of which have well-characterized dimerization and DNA-binding properties (Tupler et al. 2001; Vinson et al. 2002). The human bZIP dimerization map is presented in figure 1. Because the DNA-binding properties of these factors have been studied extensively, it was possible to integrate the interaction data with the DNA-binding preferences of the dimers (tinted ovals). The DNA target sites represented include the cAMP-responsive element (CRE), the TPA-responsive element (TRE), the MAF recognition element (MARE), the CAAT box, the PAR site, and the CRE-like element (CRE-L) (Montminy et al. 1986; Angel et al. 1987; Oikarinen et al. 1987; Liou et al. 1990; Andrews et al. 1993; Hunger et al. 1994; Kataoka et al. 1994; Clauss et al. 1996; Ubeda et al. 1996) (table 1). It is important to note that we have listed the consensus sequences for each of these elements and that, in fact, there is great variability in the actual sequences of these elements as they appear in the promoters of target genes (FitzGerald et al. 2004; Zhang et al. 2005). Additionally, if each partner protein of a heterodimer (e.g., Jun and ATF2) displays a different DNA-binding preference as a homodimer, it is likely that the DNA-binding preference of one or both constituents will be altered (Hai and Curran 1991). Variability of bZIP factor DNA-binding preference is a commonality of all factors discussed in this review, and inclusion of DNA-binding preferences in any of the network maps presented here should be considered a generalization.
|
|
The human map presented in figure 1 consists of a similar number of homodimerizing, heterodimerizing, and homo/heterodimerizing factors. The factors clustered as binding to TRE and MARE sites display a high degree of inter- and intrafamily heterodimerization. Factors clustered as preferring CRE or CAAT DNA sequences form homodimers and interfamily heterodimers. PAR factors and CRE-L factors form stable homodimers (Drolet et al. 1991
; Vinson et al. 2002). It is also important to note that factors capable of homodimerizing are also likely to dimerize efficiently with paralogous factors (denoted by brackets) because of highly similar leucine zippers. Dimerization between 2 paralogs is referred to as a quasihomodimer. We will use this human map as a point of reference for the comparisons made throughout the remainder of this review.
Based on figure 1 and the many experimental and predictive studies performed on human bZIP factors over the past 15 years, we conclude that these proteins can dimerize in many different combinations, with the potential for enormously complex transcription programs. An example of this complexity is seen when we consider what is known regarding the interactions of the Jun, Fos, and ATF2 bZIP proteins. Jun proteins can dimerize with Fos or ATF2 proteins generating complexes that bind preferentially to TRE or CRE DNA, respectively. Experiments manipulating the dimerization preference of Jun such that only Jun:Fos or Jun:ATF2 dimers can form show that both heterodimers contribute to cellular transformation (van Dam et al. 1998). However, although Jun:ATF2 complexes deregulate cell growth by promoting autocrine growth (growth in the absence of serum), Jun:Fos dimers induce anchorage independence (growth without attachment) (van Dam et al. 1998). The genetic targets of these complexes are now being defined. Mercola and colleagues identified promoters that are occupied by Jun:ATF2 dimers following the induction of genotoxic stress (Hayakawa et al. 2004), and Curran and colleagues published a list of gene targets that are misregulated as a consequence of cellular transformation by viral Fos (Ordway et al. 2005). Unfortunately, because these 2 studies used different experimental conditions, it is impossible to obtain a meaningful comparison of the genes regulated by Jun:ATF2 and Jun:Fos. The example of closely related dimers mediating distinct functions is almost certainly not limited to the Jun, Fos, and ATF2 proteins but rather is more likely a common feature of the bZIP interaction network.
It is not difficult to accept that different heterodimers have different functions. Harder to envision is how functional differences arise between dimers generated from paralogs possessing identical dimerization and DNA-binding preferences. Birrer and colleagues used a doxycyclin inducible system in Rat1a cells to identify differences in cellular response and target gene regulation following overexpression of c-Jun, JunB, and JunD (Leaner et al. 2003). The assumption was that all Jun dimers would have an overlapping, if not an identical, impact on these cultured cells. Yet the results showed that c-Jun and JunB expression correlated with cellular transformation, whereas JunD exerted antiproliferative effects when expressed in the cells (Leaner et al. 2003). The authors further conclude that, in an overexpression paradigm, c-Jun and JunB induced a similar set of genes upon which JunD had no influence. Most recently, the functional redundancy of Jun paralogs has begun to be addressed in mouse models where individual knockouts of the 3 jun genes have shown different phenotypes, with lethality associated only with the loss of c-jun (Johnson et al. 1993; Jochum et al. 2001). The differences in these phenotypes are likely due to their critical functions within nonoverlapping tissues in the context of both homodimers and heterodimers. Moreover, because ablation of a given Jun family member would potentially affect many different heterodimers (fig. 1), the critical functional unit for these phenotypes may not be a homodimer. In support of this notion, replacing the c-Jun gene with cDNA for JunB completely rescues heart outflow tract defects and the massive liver apoptosis characteristic of the c-jun null embryo (Passegue et al. 2002). Therefore, the functional contribution of a monomer within a homodimer versus a heterodimer remains complicated. However, these complexities speak to the usefulness of prediction and annotation of interaction data (fig. 1).
The D. melanogaster Dimerization Network
The bZIP network for D. melanogaster is annotated in figure 2. As reported previously, most Drosophila bZIP factors have human orthologs (Fassler et al. 2002). In fact, dJun and dFos substitute for their mammalian orthologs in reporter gene assays (Zhang et al. 1990) and use similar regulatory mechanisms as human bZIP factors. Phosphorylation by conserved signaling pathways regulates the transcriptional activity of Drosophila CREB and Jun (Sluss and Davis 1997; Horiuchi et al. 2004), and a redox mechanism similar to that described for mammalian bZIP factors controls DNA binding (Xanthoudakis and Curran 1994; Deppmann et al. 2003; Jindra et al. 2004). In light of these commonalities, it is not surprising that this network shows a very similar profile of dimerization preferences as the human network (figs. 1 and 2A). The most notable difference between the 2 networks is the number of bZIP paralogs. For example, Drosophila contains only 1 JUN gene and 1 FOS gene, whereas Homo sapiens has 3 JUN and 4 FOS/FRA genes, respectively. With the exception of the Par family, this reductive trend is observed for all other families of bZIP factors as well. The evolutionary significance of this difference is not clear but suggests that Drosophila bZIP factors may reflect the basic components of an ancestral network that were duplicated and diversified to give rise to the more complex network observed in humans. As a companion to the dimerization network map presented in figure 2A, a grid representing all possible interactions among Drosophila bZIP proteins is presented in figure 2B.
In all the species discussed in this review, there are published examples of alternative splicing generating bZIP variants that impact the function of their full-length proteins. The distilled nature of the Drosophila network, combined with the power of Drosophila genetics, allows for an opportunity to briefly explore diversity created by alternative splicing. Taghert and colleagues investigated splice variants of Crc, the Drosophila ortholog of human ATF4 (Hewes et al. 2000). Three Crc isoforms have been identified: 2 that contain the bZIP domain and 1 that is missing the bZIP domain. The authors found that isoform-specific mutations within the Crc gene define 2 lethal complementation groups and concluded that this reflects "overlapping, but distinct, functions" of the variant proteins. Another example of bZIP splice variants involves Pdp1, the mammalian Par family ortholog in Drospohila. The Pdp1 gene has 4 separate start sites and produces at least 6 different Pdp1 mRNAs (Reddy et al. 2000). Most of the Pdp1 isoforms contain the bZIP domain but display mutually exclusive expression patterns during development, indicating that each may have a separate function in these distinct body regions.
The Yeast Dimerization Network
Yeast bZIP factors have provided substrate for some of the first and most detailed analysis of this class of transcription factors. In fact, the yeast factor GCN4 was used to define the bZIP domain as being sufficient for dimerization and interaction with DNA in a sequence-specific manner (Hope and Struhl 1986). Moreover, the nature of DNA-binding specificity was first defined using GCN4 and a variety of innovative mutagenesis strategies (Hill et al. 1986; Oliphant et al. 1989; Kim et al. 1993). The initial identification of the leucine zipper and the first X-ray crystallographic structure of a bZIP dimer bound to DNA were performed using the yeast bZIP factor GCN4 (Landschulz et al. 1988; O'Shea et al. 1991; Ellenberger et al. 1992). Like the H. sapiens bZIP factors, the consensus DNA-binding preferences for most of the S. cerevisiae bZIP proteins have been defined (Fernandes et al. 1997; Garcia-Gimeno and Struhl 2000) and are presented along with the network in figure 3A and in table 1.
Of the networks discussed in this review, the S. cerevisiae bZIP network is the smallest (fig. 3A). Eleven of the 14 factors in this network are exclusively homodimeric. This is in striking contrast to the human and Drosophila networks which have a high proportion of heterodimerizing factors (figs. 1 and 2). The predicted dimerization properties of the yeast bZIP proteins largely agrees with experimental analysis (Fernandes et al. 1997; Toone and Jones 1999; Garcia-Gimeno and Struhl 2000; Proft et al. 2001). However, it remains unclear what the dimer partners are for YAP4, –6, and –8 because these proteins are not predicted to form homodimers, and the interaction grid (fig. 3B) does not indicate that they have potential partner proteins with which to heterodimerize.
Yeast provides an excellent opportunity to study nuances in bZIP dimer function using genetic approaches. We have already reviewed studies with the human Jun paralogs that addressed some issues regarding how highly related dimers with identical DNA-binding preferences can display differences in function. However, because the mammalian cells used in these experiments express a variety of dimer partners for the Jun proteins, it is difficult to relate the differences observed to Jun homodimer or heterodimer formation. These issues have been addressed in yeast where Church and colleagues sought to define functional differences between YAP1 and YAP2 homodimers (Cohen et al. 2002). Using microarray analysis, they compared gene expression profiles from haploid cells lacking either of these paralogs. The experiments were well controlled because YAP1 and YAP2 do not form heterodimers with other bZIP proteins and in the context of cells deficient for either gene do not dimerize with each other to form quasihomodimers (fig. 3A). Results demonstrated that these highly similar homodimers influence the transcription of gene sets containing minimal overlap (Cohen et al. 2002). Undoubtedly, these distinct target genes contribute to the additional functional differences noted for the YAP1 and YAP2 proteins (Toone and Jones 1999; Rodrigues-Pousada et al. 2004).
Saccharomyces cerevisiae possesses a simplified network of bZIP dimerization and can be manipulated genetically to learn much about network function. Because yeast display a noticeable lack of heterodimer formation, the system is more useful for addressing fundamental questions regarding the function of bZIP homodimers. However, additional studies on the 3 bZIP proteins that prefer not to homodimerize may provide insight into how heterodimerization may have evolved as a mechanism through which a protein network of limited size can expand its function within a cell.
The Plant Dimerization Network
Using data from previous studies, we compiled the dimerization network for A. thaliana (Jakoby et al. 2002; Deppmann et al. 2004) (fig. 4A), the largest network discussed in this review. It is important to note that plant bZIP factors are slightly different from the others discussed in the amino acids used to drive interaction as well as the heterogeneous lengths of leucine zippers represented in the genome (Deppmann et al. 2004). These unique features notwithstanding, plant bZIP factors are amenable to predictive analysis. Upon initial examination, it is apparent that the most prominent feature of this network is the large number of groups consisting of paralogous bZIP proteins predicted to form only homodimers (fig. 4A). Although the companion grid reveals several possibilities for heterodimerization (fig. 4B), the summed thermodynamic determinants (coupling energies) reveal that the constituents of most putative heterodimers are much more likely to homodimerize. This indicates that these heterodimers are unlikely to form in vivo.
The plant bZIP network generates a large amount of dimer diversity in a manner similar to what was noted in the human network. Heterodimerization among the H. sapien bZIP proteins could produce up to 340 unique dimers. Similarly, A. thaliana uses large groups of paralogous factors with nearly identical bZIP domains to generate 175 possible dimer combinations. Clearly, combining these 2 mechanisms of dimer formation would produce the maximum amount of dimer diversity for an organism. This is illustrated by the networks of S. cerevisiae and D. melanogaster, which have the potential of forming approximately 13 and 54 unique dimers, respectively.
An obvious question that arises is how specific dimer formation is achieved within a network that contains such a high degree of quasihomodimer formation. The principal regulators of leucine zipper protein interaction are affinity, specificity, and local protein concentration. In the case of the plant bZIP proteins, the affinity and specificity of interaction are virtually the same, leaving local protein concentration, or additional regulatory mechanisms, as the most likely modulators of dimerization. A number of studies on the TGA family of 10 quasihomodimeric bZIP factors provides examples of this regulation (Schiermeyer et al. 2003; Bensmihen et al. 2005). Although many TGA factors have nonoverlapping tissue distribution, the ones that do overlap seem to bind one and other with similar affinity and compete for binding with similar elements (Schiermeyer et al. 2003; Bensmihen et al. 2005). However, Pontier and colleagues have demonstrated that the stability of certain TGA paralogs can be modulated by phosphorylation (Pontier et al. 2002) and, under these conditions, cause a biased representation of dimers in the cell. Another influence on dimer stability in plants is provided by the non–bZIP protein, NPR-1, which specifically interacts with leucine zippers to stabilize a subset of TGA homodimers and increase the absolute levels of these homotypic complexes in vivo (Despres et al. 2000; Niggeweg et al. 2000).
In previous sections, we discussed some evidence for functional differences between heterodimers, paralagous homodimers, and alternatively spliced variants. Conceptually, it makes sense that different quasihomodimers would display unique functionalities as well. However, from our examination of the literature, we have not found a report that compared quasihomodimer function in plants or in another organism. Currently, the tethering approach used by Neuhold and Wold (1993) to force exclusive dimerization of particular basic helix-loop-helix transcription factors could be applied to examine the function of specific bZIP quasihomodimers in cells. Although Yaniv and colleagues have validated this experimental approach for studying bZIP dimers (Bakiri et al. 2002), differences in quasihomodimer function were not the topic of their study.
Giardia lamblia Contains a Single bZIP Factor
To this point, we have focused on annotating bZIP interactions, discussing functional differences between different types of dimers, and comparing dimerization networks from several organisms. In doing so, our observations elicit obvious questions about the origins of bZIP factors and how bZIP interaction networks may have evolved. In an attempt to address these questions, we examined bZIP factors in the protozoan, Giardia lamblia. This organism lacks mitochondria and is among the most ancient eukaryotes studied to date (Subramanian et al. 2000). Using various known bZIP factors as a query and the TBlastN program at http://www.mbl.edu/Giardia, the G. lamblia genome (McArthur et al. 2000) was mined for bZIP motifs. The search identified the sequence in figure 5A. This sequence was used as the probe in a second Blast search that yielded no other related sequences. We conclude from this analysis that there may be only a single bZIP factor in this organism. It should be noted that another protozoan, Dictyostelium discoideum, has been reported to have 19 bZIP factors (Eichinger et al. 2005). This is not inconsistent with the notion that the Giardia bZIP factor may be ancestral because many characteristics of Giardia (Best et al. 2004) indicate that these eukaryotes are more primitive than Dictyostelium.
|
Dimerization prediction was performed on the G. lamblia bZIP factor revealing a strong preference for homodimer formation. The leucine zipper of this factor contains 4 attractive g–e interactions in the first and second heptads and 4 attractive a–a interactions across all 4 heptads (fig. 5B). As the only identified bZIP factor in Giardia, it is not surprising that it contains the prototypic determinants for homodimerization. Interestingly, a predictive analysis was recently performed on bZIP factors in Dictyostelium (Huang et al. 2006
). Although this was not a genome-wide analysis, the authors suggest that 8 out of 18 factors with highly similar leucine zippers dimerize with one another with similar predicted affinity. Our cursory examination of the aligned leucine zippers presented in Huang et al. (2006) reveals that there are likely between 3 and 5 groups of quasihomodimerizing paralogs, which would result in a network similar to Arabidopsis (fig. 4A). Comparing the increasingly complex networks of Giardia, Dictyostelium, and Arabidopsis reveals a progression from a network with a solitary homodimerizing bZIP factor to a complex network of homodimerizing and quasihomodimerizing groups of paralogs. This comparison suggests that these homodimeric networks share a common evolutionary lineage. Discovery of a potentially ancestral bZIP factor is truly significant and allows for speculation on bZIP factor evolution. Once the bZIP factors for other protozoans (and perhaps Archeabacteria) are annotated, it will be possible to propose an origin for this motif and to perform follow-up studies on how simple dimerization networks function in primitive eukaryotes. This information may shed light on when and how bZIP factors gained heterodimerizing features as well as additional properties that are critical to their roles as transcriptional regulators in higher eukaryotes.
Evolution of bZIP Factor Dimerization Networks
In an effort to visualize the evolutionary relationship between the bZIP domains presented in this review, a phylogenetic analysis was performed (Supplemental Figure 2, Supplementary Material online). As one might expect, there is significant grouping between H. sapiens and D. melanogaster bZIP domains. Most of the heterodimerizing factors from these 2 organisms cluster to the same region of the tree, indicating a common origin in a single, ancestral heterodimerizing factor. However, it is intriguing that there is no interspecies clustering between H. sapiens and A. thaliana bZIP motifs. The absence of human/plant bZIP orthologs strongly suggests that bZIP proteins in these 2 kingdoms have undergone completely divergent evolution. Interestingly, this analysis positions the Giardia bZIP factor on a branch containing the yeast, human, and fly bZIP factors (Supplemental Figure 2, Supplementary Material online).
Through the course of evolution, organisms undergo local, regional, and global genomic duplications (Prince and Pickett 2002). Moreover, duplicate genes are thought to diverge from each other by incorporating changes that result in a loss or gain of unique functionality (Force et al. 1999). All bZIP factors are likely to be derived from an ancient primordial bZIP factor, perhaps similar to the one identified in G. lamblia (fig. 5). This primordial factor may have been duplicated successively during the course of speciation with new functionalities arising as a result. This principle was recently discussed in some detail for bZIP and basic helix-loop-helix factors by Weiner and colleagues (Bornberg-Bauer et al. 2005). Using this logic, we can consider how the D. melanogaster and H. sapiens networks are related. Because most Drosophila bZIP proteins have human orthologs, it is possible that these factors reflect the building blocks of what is now the human bZIP network. Presumably, several rounds of gene duplication and sequence divergence underlie the observed clustering of similar, human bZIP domains.
It is clear from experimentation that only a fraction of bZIP factors in the genome of animals are expressed in any given cell. It seems possible that the smaller network of yeast represents a rationale for the minimum number of bZIP factors required in a particular animal cell. Because higher organisms have a huge diversity of cell types, it may be necessary to substitute and add factors into a set similar to those resident in yeast depending on the needs of the cell. Interestingly, classes of factors that yeast do not have, such as C/EBP, may have evolved to serve specific needs like adipose tissue development (Darlington et al. 1998). By this same logic, it seems possible that classes of factors common between yeast and higher organisms such as AP-1 and ATF may be necessary for more general housekeeping functions. Consistent with this notion is the broad tissue distribution of AP-1 and ATF/CREB factors in vertebrates and their roles in processes such as cell proliferation, metabolism, and stress response (Karin et al. 1997). Of course, even within these common families, significant modifications have evolved, such as addition of paralogs and the ability to heterodimerize.
|
Conclusions and Perspective |
|---|
|
TOPAbstractIntroductionMethodology of Predicting Dimer...A New Approach for...Conclusions and PerspectiveSupplementary MaterialAcknowledgementsReferences |
|---|
This review summarizes the rules governing bZIP factor dimerization and utilizes a novel method to represent the bZIP dimerization networks from 4 different species. In the process of annotating these networks, several interesting facts emerged about the organization of these networks that impact our understanding of how bZIP proteins and their dimerization properties might have evolved. The comparison of 2 complex organisms, plant and human, suggests that network complexity can be achieved by distinct mechanisms. The plant network achieves diversity by having several large sets of paralogous factors that homodimerize/quasihomodimerize, whereas the human network has smaller groups of paralogs, which can homodimerize, quasihomodimerize, and/or heterodimerize.
We included selected examples from the wealth of experimental literature on this topic to suggest that every dimerization network is unique, and the function of every dimer within a network is unique. What we have touched on only briefly are the many mechanisms operating in cells that can bias dimerization, including the role of posttranslational modifications and the impact of accessory molecules on driving dimer formation and/or enhancing dimer stability. These examples of regulation, taken together with the fact that only a fraction of these factors are expressed at any given time, underscores the point that these networks are likely much more complex and dynamic than the static diagrams represented here. Although the structural features of leucine zippers can be exploited to predict dimerization properties and to organize bZIP proteins within a phylogenetic tree, the next challenge is to understand the origins of the complex regulatory mechanisms that handle the increased functional demands placed on bZIP transcription factors as eukaryotic organisms evolved.
|
Supplementary Material |
|---|
|
TOPAbstractIntroductionMethodology of Predicting Dimer...A New Approach for...Conclusions and PerspectiveSupplementary MaterialAcknowledgementsReferences |
|---|
Supplementary material and a color version of Figure 1 are available at Molecular Biology and Evolution online (http://www.mbe.oxfordjournals.org/).
|
Acknowledgements |
|---|
|
TOPAbstractIntroductionMethodology of Predicting Dimer...A New Approach for...Conclusions and PerspectiveSupplementary MaterialAcknowledgementsReferences |
|---|
The authors would like to thank Rosa Carrasco for discussions regarding G. lamblia. We are also grateful to David Ginty, Anthony Harrington, Narendrakumar Ramanan, Aruna Sathyamurthy, and Larry Zweifel for critically reading the manuscript. The authors are supported by National Institutes of Health (NIH) NS-053187 (C.D.D.), NIH NS-34814 (to David Ginty), and PHS CA-78264 (E.J.T.).
|
Footnotes |
|---|
Douglas Crawford, Associate Editor
|
References |
|---|
|
TOPAbstractIntroductionMethodology of Predicting Dimer...A New Approach for...Conclusions and PerspectiveSupplementary MaterialAcknowledgementsReferences |
|---|
Acharya A, Ruvinov SB, Gal J, Moll JR, Vinson C. 2002. A heterodimerizing leucine zipper coiled coil system for examining the specificity of a position interactions: amino acids I, V, L, N, A, and K. Biochemistry 41:14122–31.[CrossRef][Medline]
Alber T. 1992. Structure of the leucine zipper. Curr Opin Genet Dev 2:205–10.[CrossRef][Medline]
Andrews NC, Erdjument-Bromage H, Davidson MB, Tempst P, Orkin SH. 1993. Erythroid transcription factor NF-E2 is a haematopoietic-specific basic-leucine zipper protein. Nature 362:722–8.[CrossRef][Medline]
Angel P, Imagawa M, Chiu R, Stein B, Imbra RJ, Rahmsdorf HJ, Jonat C, Herrlich P, Karin M. 1987. Phorbol ester-inducible genes contain a common cis element recognized by a TPA-modulated trans-acting factor. Cell 49:729–39.[CrossRef][ISI][Medline]
Bakiri L, Matsuo K, Wisniewska M, Wagner EF, Yaniv M. 2002. Promoter specificity and biological activity of tethered AP-1 dimers. Mol Cell Biol 22:4952–64.
Baxevanis AD, Vinson CR. 1993. Interactions of coiled coils in transcription factors: where is the specificity? Curr Opin Genet Dev 3:278–85.[CrossRef][Medline]
Bensmihen S, Giraudat J, Parcy F. 2005. Characterization of three homologous basic leucine zipper transcription factors (bZIP) of the ABI5 family during Arabidopsis thaliana embryo maturation. J Exp Bot 56:597–603.
Best AA, Morrison HG, McArthur AG, Sogin ML, Olsen GJ. 2004. Evolution of eukaryotic transcription: insights from the genome of Giardia lamblia. Genome Res 14:1537–47.
Bornberg-Bauer E, Beaussart F, Kummerfeld SK, Teichmann SA, Weiner J III. 2005. The evolution of domain arrangements in proteins and interaction networks. Cell Mol Life Sci 62:435–45.[CrossRef][ISI][Medline]
Choi H, Hong J, Ha J, Kang J, Kim SY. 2000. ABFs, a family of ABA-responsive element binding factors. J Biol Chem 275:1723–30.
Clauss IM, Chu M, Zhao JL, Glimcher LH. 1996. The basic domain/leucine zipper protein hXBP-1 preferentially binds to and transactivates CRE-like sequences containing an ACGT core. Nucleic Acids Res 24:1855–64.
Cohen BA, Pilpel Y, Mitra RD, Church GM. 2002. Discrimination between paralogs using microarray analysis: application to the Yap1p and Yap2p transcriptional networks. Mol Biol Cell 13:1608–14.
Darlington GJ, Ross SE, MacDougald OA. 1998. The role of C/EBP genes in adipocyte differentiation. J Biol Chem 273:30057–60.
Darlington GJ, Wang N, Hanson RW. 1995. C/EBP alpha: a critical regulator of genes governing integrative metabolic processes. Curr Opin Genet Dev 5:565–70.[CrossRef][ISI][Medline]
Deppmann CD, Acharya A, Rishi V, Wobbes B, Smeekens S, Taparowsky EJ, Vinson C. 2004. Dimerization specificity of all 67 B-ZIP motifs in Arabidopsis thaliana: a comparison to Homo sapiens B-ZIP motifs. Nucleic Acids Res 32:3435–45.
Deppmann CD, Thornton TM, Utama FE, Taparowsky EJ. 2003. Phosphorylation of BATF regulates DNA binding: a novel mechanism for AP-1 (activator protein-1) regulation. Biochem J 374:423–31.[CrossRef][ISI][Medline]
Despres C, DeLong C, Glaze S, Liu E, Fobert PR. 2000. The Arabidopsis NPR1/NIM1 protein enhances the DNA binding activity of a subgroup of the TGA family of bZIP transcription factors. Plant Cell 12:279–90.
Drolet DW, Scully KM, Simmons DM, Wegner M, Chu KT, Swanson LW, Rosenfeld MG. 1991. TEF, a transcription factor expressed specifically in the anterior pituitary during embryogenesis, defines a new class of leucine zipper proteins. Genes Dev 5:1739–53.
Eferl R, Sibilia M, Hilberg F, Fuchsbichler A, Kufferath I, Guertl B, Zenz R, Wagner EF, Zatloukal K. 1999. Functions of c-Jun in liver and heart development. J Cell Biol 145:1049–61.
Eichinger L, Pachebat JA, Glockner G, et al. (97 co-authors). 2005. The genome of the social amoeba Dictyostelium discoideum. Nature 435:43–57.[CrossRef][Medline]
Ellenberger TE, Brandl CJ, Struhl K, Harrison SC. 1992. The GCN4 basic region leucine zipper binds DNA as a dimer of uninterrupted alpha helices: crystal structure of the protein-DNA complex. Cell 71:1223–37.[CrossRef][ISI][Medline]
Fassler J, Landsman D, Acharya A, Moll JR, Bonovich M, Vinson C. 2002. B-ZIP proteins encoded by the Drosophila genome: evaluation of potential dimerization partners. Genome Res 12:1190–200.
Fernandes L, Rodrigues-Pousada C, Struhl K. 1997. Yap, a novel family of eight bZIP proteins in Saccharomyces cerevisiae with distinct biological functions. Mol Cell Biol 17:6982–93.[Abstract]
FitzGerald PC, Shlyakhtenko A, Mir AA, Vinson C. 2004. Clustering of DNA sequences in human promoters. Genome Res 14:1562–74.
Fong JH, Keating AE, Singh M. 2004. Predicting specificity in bZIP coiled-coil protein interactions. Genome Biol 5:R11.[CrossRef][Medline]
Force A, Lynch M, Pickett FB, Amores A, Yan YL, Postlethwait J. 1999. Preservation of duplicate genes by complementary, degenerative mutations. Genetics 151:1531–45.
Garcia-Gimeno MA, Struhl K. 2000. Aca1 and Aca2, ATF/CREB activators in Saccharomyces cerevisiae, are important for carbon source utilization but not the response to stress. Mol Cell Biol 20:4340–9.
Grigoryan G, Keating AE. 2006. Structure-based prediction of bZIP partnering specificity. J Mol Biol 355:1125–42.[CrossRef][ISI][Medline]
Hai T, Curran T. 1991. Cross-family dimerization of transcription factors Fos/Jun and ATF/CREB alters DNA binding specificity. Proc Natl Acad Sci USA 88:3720–4.
Hai T, Wolfgang CD, Marsee DK, Allen AE, Sivaprasad U. 1999. ATF3 and stress responses. Gene Expr 7:321–35.[ISI][Medline]
Hayakawa J, Mittal S, Wang Y, Korkmaz KS, Adamson E, English C, Ohmichi M, McClelland M, Mercola D. 2004. Identification of promoters bound by c-Jun/ATF2 during rapid large-scale gene activation following genotoxic stress. Mol Cell 16:521–35.[CrossRef][ISI][Medline]
Hewes RS, Schaefer AM, Taghert PH. 2000. The cryptocephal gene (ATF4) encodes multiple basic-leucine zipper proteins controlling molting and metamorphosis in Drosophila. Genetics 155:1711–23.
Hill DE, Hope IA, Macke JP, Struhl K. 1986. Saturation mutagenesis of the yeast his3 regulatory site: requirements for transcriptional induction and for binding by GCN4 activator protein. Science 234:451–7.
Hope IA, Struhl K. 1986. Functional dissection of a eukaryotic transcriptional activator protein, GCN4 of yeast. Cell 46:885–94.[CrossRef][ISI][Medline]
Horiuchi J, Jiang W, Zhou H, Wu P, Yin JC. 2004. Phosphorylation of conserved casein kinase sites regulates cAMP-response element-binding protein DNA binding in Drosophila. J Biol Chem 279:12117–25.
Huang E, Blagg SL, Keller T, Katoh M, Shaulsky G, Thompson CR. 2006. bZIP transcription factor interactions regulate DIF responses in Dictyostelium. Development 133:449–58.
Hunger SP, Brown R, Cleary ML. 1994. DNA-binding and transcriptional regulatory properties of hepatic leukemia factor (HLF) and the t(17;19) acute lymphoblastic leukemia chimera E2A-HLF. Mol Cell Biol 14:5986–96.
Jakoby M, Weisshaar B, Droge-Laser W, Vicente-Carbajosa J, Tiedemann J, Kroj T, Parcy F. 2002. bZIP transcription factors in Arabidopsis. Trends Plant Sci 7:106–11.[CrossRef][ISI][Medline]
Jindra M, Gaziova I, Uhlirova M, Okabe M, Hiromi Y, Hirose S. 2004. Coactivator MBF1 preserves the redox-dependent AP-1 activity during oxidative stress in Drosophila. EMBO J 23:3538–47.[CrossRef][ISI][Medline]
Jochum W, Passegue E, Wagn