We began our analysis with a comparison of the solved structures of BTB domains from the Protein Data Bank (PDB) 31, which included examples from BTB-ZF proteins, Skp1, ElonginC and T1 domains (Figures 1, 2, 3). A three-dimensional superposition showed a common region of approximately 95 amino acids consisting of a cluster of 5 α-helices made up in part of two α-helical hairpins (A1/A2 and A4/A5), and capped at one end by a short solvent-exposed three stranded β-sheet (B1/B2/B3; Figure 1). An additional hairpin-like motif consisting of A3 and an extended region links the B1/B2/A1/A2/B3 and A4/A5 segments of the fold. Because of the presence or absence of secondary structural elements in certain examples of the fold, we use the designation A1–A5 for the five conserved α-helices, and B1–B3 for the three common β-strands. We refer to this structure as the core BTB fold. When present, other secondary structure elements are named according to the labels assigned to the original structures. Thus, the BTB-ZF family members the promyelocytic leukemia zinc finger (PLZF) and B-cell lymphoma 6 (BCL6) contain additional amino-terminal elements, which are referred to as β1 and α1, Skp1 protein contains two additional carboxy-terminal helices labeled α7 and α8, ElonginC is missing the A5 terminal helix, and the T1 structures from Kv proteins are formed entirely of the core BTB fold (Figures 1 and 2). Sequence comparisons based on the structure superpositions show less than 10% identity between examples from different families, except for Skp1 and ElonginC, which is in the range of 14% to 22%; however, all structures show remarkable conservation with Root mean square deviation (RMSD) values of 1.0 to 2.0 Å over at least 95 residues (Figure 3). Despite these very low levels of sequence relatedness, 15 positions show significant conservation across all of the structures, and 12 of these correspond to residues that are buried in the monomer core (Figure 2). Most of these highly conserved residues are hydrophobic and are found between B1 and A3, with some examples in A4. In addition to this common set, conserved residues are also found within specific families (Figure 2), and some of these participate in family-specific protein-protein interactions.

The four known structural classes of BTB domains show different oligomerization or protein-protein interaction states involving different surface-exposed residues (Figures 2 and 4). There is little overlap between the interaction surfaces of the homodimeric, heteromeric and homotetrameric forms of the domain, which are represented here by examples from the BTB-ZF, Skp1/ElonginC and T1 families, respectively. Contacts involving the amino-terminal extensions of the BTB-ZF class and the carboxy-terminal elements of the Skp1 families form a significant fraction of the residues involved in protein-protein interaction in each of those respective systems, but additional contributions from the 95 residue core BTB fold are involved. There are multiple examples of conserved surface-exposed residues that participate in family-specific protein-protein interactions. For example, the B1/B2/B3 sheet is found in all BTB structures and, therefore, is part of the core BTB fold, but participates in very different protein interactions in the T1 homotetramers, the ElonginC/ElonginB and Skp1-Cul1 structures. Inspection of T1 residues in this area shows sequences such as the 'FFDR' motif in B3 have diverged from the other BTB families to become important components of the tetramerization interface 29 (Figure 2). In Skp1, B3 has a distinctive 'PxPN' motif that is involved in Cul1 interactions 24 (Figure 2). Thus, the solvent-exposed surface of the BTB fold is extremely variable between families, forming the basis for the wide range of protein-protein interactions.

The connection between A3 and A4 (drawn as a dashed line in Figure 1b) is variable in length and in structure, and makes key contributions to several different types of protein-protein interactions. The region adopts an extended loop structure in the T1 domain and ElonginC, where it makes important contributions to the homotetramerization and to the von Hippel-Lindau (VHL) interfaces, respectively (Figure 4). In PLZF and BCL6, this segment forms strand β5 and associates with β1 from the partner chain to form a two-stranded antiparallel sheet at the 'floor' of the homodimer 522. In Skp1, this region includes a large disordered segment followed by a unique helix α4, but it is not involved in any protein-protein interactions 23242526.

We searched the Ensembl and Uniprot databases for BTB proteins 3233. In order to effectively eliminate redundant sequences and partial fragments, and to reduce sampling bias due to uneven database representation, we limited our search to the known and predicted transcripts from 17 fully sequenced genomes. We carried out HMMER 34 searches with a panel of hidden Markov models (HMMs) describing the four known families of BTB structures. As expected from the low sequence similarities, searches with family-specific HMMs could not retrieve sequences from the other families in a single iteration. For example, the HMM trained on the BTB domains from BTB-ZF proteins could not immediately retrieve sequences from T1-Kv proteins. Additional sequences were added to each of the family-specific HMMs in several cycles, and the results were compiled into final multiple sequence alignments. The retrieved sequences were manually inspected and class-specific HMMs were used to define the start/end sites of specific families of BTB domains. We have assembled this collection of over 2,200 non-redundant BTB domain sequences in a publicly available database 35.

In addition to the genome-centric analyses, we searched the NCBI nr database with PSI-BLAST 3637. Beginning with the sequence of the BTB domain from the BTB-ZF protein PLZF, T1 sequences were retrieved with e-values below 10 after four PSI-BLAST iterations carried out with a generous inclusion threshold of 0.1, as previously reported 30. Skp1 and ElonginC sequences could not be retrieved with e-values below 10 starting with BTB-ZF or T1 sequences, even with a PSI-BLAST inclusion threshold of 1.0. Based on searches with representative BTB sequences from each of the major families, BTB sequences were consistently retrieved from eukaryotes and poxviruses, but no examples from bacteria or archaea were found (data not shown), with the remarkable exception of 43 BTB-leucine-rich repeat proteins in the Parachlamydia-related endosymbiont UWE25 38. In general, plant and animal genomes encode from 70 to 200 distinct BTB domain proteins, while only a handful of examples are found in the unicellular eukaryotes. We identified an intermediate number, 41, in the social amoeba Dictyostelium discoideum 39 (Figure 5).

The distribution of BTB families varies widely according to species (Figure 5). The overall number of BTB domain proteins and their family distribution is similar in the mammalian and fish genomes that we considered, with 25 to 50 examples from each of the BTB-ZF, BTB-BACK-kelch (BBK) and T1-Kv families, and another 40 to 50 proteins with other architectures. We expect that this distribution is similar across all vertebrate genomes. The family distribution in the insects (as exemplified by Drosophila and Anopheles) is generally similar to that of the vertebrates, but with fewer overall examples. In contrast, Caenorhabditis elegans contains very few BTB-ZF and BBK proteins, but a large number of meprin and tumor necrosis factor receptor associated factor homology (MATH)-BTB and Skp1 proteins. In Arabidopsis, there are 21 BTB-nonphototropic hypocotyl (NPH)3 proteins, which appear to be a plant-specific architecture. Only five and six BTB domain proteins were found in Saccharomyces cerevisiae and Schizosaccharomyces pombe, respectively.

Based on these observations, the domain most likely underwent domain shuffling followed by lineage-specific expansion (LSE) during speciation events. The most commonly observed architecture across several different families consists of a single amino-terminal BTB domain, a middle linker region, and a characteristic carboxy-terminal domain that is often present as a set of tandem repeats (Figure 6). Along with domain shuffling and domain accretion, LSE is considered one of the major mechanisms of adaptation and generation of novel protein functions in eukaryotes, and is frequently seen in proteins involved in cellular differentiation and in the development of multicellular organisms 40. For example, both BTB-ZF proteins and Kruppel-associated box (KRAB)-ZF proteins have essential roles in development and tissue differentiation and have undergone LSE in the vertebrate lineage 304142.

We attempted to construct a phylogeny based on BTB domain sequences, but we could not consistently cluster the entire collection. Due to the very low levels of sequence similarity between some of the families (Figure 3), we were unable to support phylogenies with significant bootstrap values despite many attempts with several different approaches and algorithms, including distance, maximum parsimony or maximum likelihood methods.

We eventually turned to BLASTCLUST as a more appropriate tool to subdivide this highly divergent set of sequences 37 (Figure 6). BTB domain sequence/structure families correlate with the protein architectures, and the BTB-NPH3, T1, Skp1 and ElonginC families could be distinguished at an identity threshold of 30% with this method. Domain sequences from BTB-ZF, BBK, MATH-BTB and RhoBTB proteins formed distinct clusters only at higher cutoffs, and are thus more closely related (Figure 6). The BTB domain sequences from vertebrate BTB-ZF and BBK proteins are more closely related, and cannot be separated by BLASTCLUST.

The majority of BTB domains from the BTB-ZF, BBK, MATH-BTB, RhoBTB and BTB-basic leucine Zipper (bZip) proteins contain a conserved region amino-terminal to the core region, which likely forms a β1 and α1 structure as seen in PLZF 2243 and BCL6 5. We refer to this as the 'long form' of the BTB domain, which has a total size of approximately 120 residues. Note that many of the protein domain databases, such as Pfam 44, SMART 45 and Interpro 46, recognize only the 95 residue core BTB fold, and do not detect all of these additional elements, even though at least half of the metazoan BTB domains correspond to the long form. The long form BTB domain sequences also are more highly related to each other than to the BTB-NPH3, T1, Skp1 and ElonginC families, as based on the BLASTCLUST analysis (Figure 6). These groupings were consistently observed even when only the residues from the core fold were included in the analysis, and so the sequence clustering is not simply due to the presence or absence of the amino-terminal elements. We predict that most long form BTB domains are dimeric, and that several of these associate into higher order assemblies via inter-dimer sheets involving β1, as discussed below.

BTB-ZF proteins are also known as the POK (POZ and Krüppel zinc finger) proteins 47. Many members of this large family have been characterized as important transcriptional factors, and several are implicated in development and cancer, most notably BCL6 4849, leukemia/lymphoma related factor (LRF)/Pokemon 47, PLZF 50, hypermethylated in cancer (HIC)1 5152 and Myc interacting zinc finger (MIZ)1 53.

In the BTB-ZF setting, the domain mediates dimerization, as shown by crystallographic studies of the BTB domains of PLZF 22 and BCL6 5. This is confirmed in numerous solution studies 52243545556. An important component of the hydrophobic dimerization interface in PLZF and BCL6 is the association of the long form elements β1 and α1 from one monomer with the core structure of the second monomer. The dimerization interface has two components: an inter-molecular antiparallel β-sheet formed between β1 from one monomer and β5 of the other monomer; and the packing of α1 from one monomer against α1 and the A1/A2 helical hairpin from the other monomer. The strand-exchanged amino terminus is likely to have arisen from a domain swapping mechanism 57. We believe that most BTB domains from human BTB-ZF proteins can dimerize, because 34 of these 43 domains are predicted to contain all of the necessary structural elements in the swapped interface including β1, α1 and β5 (Additional data file 1). As well, many highly conserved residues are found in predicted dimer interface positions 22. Nine human BTB-ZF proteins lack β1, and thus cannot form the β1–β5 interchain antiparallel sheet, and we expect that these domains are also dimeric due to the presence of α1 and the conservation of interface residues. In PLZF and BCL6, the BTB domain forms obligate homodimers 522, and disruption of the dimer interface results in unfoldfed, non-functional protein 6.

In nearly all BTB-ZF proteins, the long form BTB domain is at or very near the amino terminus of the protein, and the Krüppel-type C₂H₂ zinc fingers are found towards the carboxyl terminus of the protein. These two regions are connected by a long (100–375 residue) linker segment (Figure 6). Sequence conservation is largely restricted to the BTB domain and the carboxy-terminal ZF region, as exemplified by BCL6 from human and zebrafish, which are 78%, 37% and 85% identical across the BTB, linker and ZF regions, respectively. The linker region frequently contains low complexity sequence and is predicted to be unstructured in most cases. Except for proteins that are highly related over their full lengths, the linker regions do not identify significant matches in sequence searches of the NCBI nr set. This architecture suggests a model in which the dimeric BTB domain connects the DNA binding regions from each chain via long, mostly unstructured tethers. Thus, we expect that the DNA binding ZF domains can bind two promoter sites, but that the exact spacing and orientation of these sites is not critical, as long as they are within a certain limiting distance. The linker is not without function, however, as it interacts with accessory proteins that take part in chromatin remodeling and transcription repression, such as the BCL6-mSin3A and PLZF-ETO interactions 658.

The BTB domains from some BTB-ZF proteins can mediate higher order self-association 59606162, and the formation of BTB oligomers in the BTB-ZF proteins has important implications for the recognition of multiple recognition sequences on target genes. In Drosophila GAGA factor (GAF), oligomerization of BTB transcription factors is thought to be mechanistically important in regulating the transcriptional activity of chromatin 6162, and the BTB domain is essential in co-operative binding to DNA sites containing multiple GA target sites 62. Several other BTB transcription factors also bind to multiple sites 526063. The formation of chains of BTB dimers involving the β1/β5 'lower sheet' has been observed in two different crystal forms of the PLZF BTB domain 2243, although the significance of this is unclear as BTB dimer-dimer associations are very weak and are not observed in solution under normal conditions (unpublished results and 43). Higher-order association could be a property of a subset of BTB domains, with GAF-BTB representing domains that have a strong propensity for polymerization, whereas in cases such as PLZF-BTB, the self-association of dimers is observed only at very high local protein concentrations, such as those required for crystal formation. Interestingly, many Drosophila BTB domains have characteristic hydrophobic sequences in the β1 and β5 regions 1. In many of these, the β1 region contains at least three large, hydrophobic residues in a characteristic [FY]×[ILV]×[WY][DN][DN][FHWY] sequence that is not present in BTB-ZF proteins from other species. This conserved segment has high β-strand propensity, consistent with the presence of interchain β1 contacts across dimers. Exposed hydrophobic residues in this sheet region may drive strong dimer-dimer associations in these Drosophila BTB-ZF proteins, an idea that is supported by modeling studies 64.

Heteromeric BTB-BTB associations have been described between certain pairs of BTB domains from this family, including PLZF and Fanconi anemia zinc finger (FAZF) 65, and between BCL6 and BCL6 associated zinc finger (BAZF) 66. Heteromer formation in BTB transcription factors may be a mechanism for targeting these proteins to particular regulatory elements by combining different chain-associated DNA binding domains in order to generate distinct DNA recognition specificities 67, as seen in retinoic acid receptor/retinoid X receptor transcription factors 68.

In addition to the architectural roles resulting from BTB-BTB associations, many BTB domains in this family interact with non-BTB proteins, and this effect is central to their function in transcriptional regulation. For example, BCL6 is able to associate directly with nuclear co-repressor proteins such as nuclear co-repressor (NCoR), silencing mediator for retinoid and thyroid hormone receptors (SMRT) and mSin3a 5586970717273. A 17 residue region of the SMRT co-repressor binds directly with the BCL6 BTB domain in a 2:2 stoichiometric ratio in a complex that requires a BCL6 BTB dimer 5. This peptide is an inhibitor of full-length SMRT, and reverses the repressive activities of BCL6 in vivo 48. Remarkably, the interaction with this peptide appears to be specific to the BCL6 BTB domain, and there is no significant sequence conservation in the BCL6 peptide binding groove relative to other human BTB-ZF proteins. In these other proteins, this groove may be a site for as yet uncharacterized BTB-peptide or BTB-protein interactions.

In all organisms studied, BTB domains from BTB-ZF proteins show high conservation of the residues Asp35 and Arg/Lys49 (PLZF numbering; Additional data file 1). These residues are found in a 'charged pocket' in the BTB structures of PLZF and BCL6, and have been shown to be important in transcriptional repression 674. The structure of the BCL6-BTB-SMRT co-repressor complex, however, did not show interactions between this region and the co-repressor 5. Mutation of Asp35 and Arg49 disrupts the proper folding of PLZF 6, and these residues are thus important for the structural integrity of the domain. Interestingly, Asp35 and Arg/Lys49 are also conserved in the BTB domains from BBK, MATH-BTB and BTB-NPH3 proteins (Figure 2 and Additional data file 1).

Many members of this widely represented family of proteins are implicated in the stability and dynamics of actin filaments 75767778. With few exceptions, all of the 515 BTB-kelch proteins in our database also contain the BTB and carboxy-terminal kelch (BACK) domain. These BBK proteins are composed of a long-form BTB domain, the 130 residue BACK domain 79, and a carboxy-terminal region containing four to seven kelch motifs 808182. Most BBK proteins have a region of approximately 25 residues that precede the BTB domain, unlike BTB-ZF proteins where BTB is positioned much closer to the amino terminus (Figure 6; Additional data file 1). We predict that this amino-terminal region in the BBK proteins is unstructured, although it is shown to have a functional role in some proteins 75. Notably, the distribution of BBK proteins parallels that of the BTB-ZF proteins across genomes. We did not find BBK proteins in Arabidopsis thaliana or in the yeasts.

The sequences of BTB domains from BBK proteins are most closely related to those from BTB-ZF proteins (Figure 6), suggesting that they adopt similar structures. Indeed, BTB domains from BBK proteins have been shown to mediate dimerization 758384 and have conserved residues at positions equivalent to those at the dimer interface in BTB-ZF proteins (Additional data file 1). There are reports of BTB-mediated oligomerization in BBK proteins, consistent with the role of some these proteins as organizers of actin filaments 757784. Because most of the BTB sequences from BBK proteins are predicted to contain the β1, α1 and β5 long form elements, oligomerization of these proteins may occur via dimer-dimer associations involving the β1 sheet, as proposed for the BTB-ZF proteins. There are, however, no strongly characteristic sequences or enrichment of hydrophobic residues in the β1 region.

In Pfam, the POZ domain superfamily (Pfam Clan CL0033) includes BACK, BTB, Skp1 and K_tetra (T1) sequences 44. The known structures of BTB, Skp1 and T1 domains show the conserved BTB fold, and the inclusion of the BACK domain in this Pfam Clan suggests that the BACK domain also adopts this fold. Secondary structure predictions for BTB, Skp1 or T1 domain sequences, however, consistently reflect the known mixed α/β content of the BTB fold, whereas the BACK domain is predicted to contain only α-helices 79. Further clarification of this issue will require the experimental determination of the structure of the BACK domain.

Skp1 is a critical component of Cul1-based SCF complex, and forms the structural link between Cul1 and substrate recognition proteins 858687. Skp1 proteins are only distantly related to other BTB families (Figures 3 and 6), and are composed of the core BTB fold with two additional carboxy-terminal helices. These latter helices form the critical binding surface for the F-box region of substrate-recognition proteins. Many Skp1 sequences have low complexity insertions after A3, which are disordered in several crystal structures, followed by helix α4, which is unique to this family 23242526 (Figures 1 and 2). Skp1 proteins are found in all organisms studied, with significant expansions in C. elegans and A. thaliana (Figure 5). Interestingly, the Cul1-interacting surface of Skp1 does not overlap with the dimerization surface seen in BTB-ZF structures, and is mostly separate from the tetramerization surface in the T1 domains (Figure 2; Additional data file 1). Therefore, a unique surface of the BTB fold in the Skp1 proteins has adapted to mediate interactions with Cul1.

ElonginC is an essential component of Cul2-based SCF-like complexes, also known as VCB (for pVHL, ElonginC, ElonginB) or ECS (for ElonginC, Cul2, SOCS-box) E3 ligase 8889. This protein serves as an adaptor between ElonginB and the VHL tumor suppressor protein, which interacts with hypoxia inducible factor (HIF)-1α and targets it for degradation 89909192. In any given organism, the sequence identity between ElonginC and Skp1 is approximately 30% or less, but these proteins are nonetheless more closely related to each other than to other BTB sequences (Figure 3). The structure of ElonginC showed that it is composed entirely of the core BTB fold, but lacks the terminal A5 helix 27289394. We found ElonginC proteins in all organisms studied (Figure 5). Like Skp1, ElonginC is significantly similar to other BTB sequence classes only in the buried positions of the monomer core (Figure 2). A β-strand in the A3/A4 connecting region participates in the ElonginC-VHL interaction, and the sequence in this region is characteristic of ElonginC 28.

The T1 domain from voltage-gated potassium channels modulates channel gating and assembly 102995. This domain is a distant homolog to all other BTB domains, and segregates into a unique cluster at less than 30% sequence identity with BLASTCLUST. The T1 domain is found in a large number of voltage-gated potassium channel proteins in all metazoan genomes surveyed (Figure 5). T1 sequences have been classified according to their sequence similarity into nine Kv families (Kv1 through Kv9) 9697. The full-length protein sequences are composed of a disordered amino-terminal region, the T1 domain, a transmembrane ion transduction domain (Pfam PF00520), and a long carboxy-terminal region with some predicted secondary structure (Figure 6).

Structurally, the T1 domain is composed of the core BTB fold without any amino- or carboxy-terminal extensions (Figures 1 and 2; Additional data file 1). The T1 domain mediates homo-tetramerization in numerous crystal structures 11299899. Despite the very low levels of sequence similarity to the other BTB domain families, several of the characteristic buried residues are conserved (Figure 2). It is striking that most of the residues found in the polar tetramerization contact surface in the T1 structures do not overlap with those residues involved in dimerization in the BTB-ZF structures. Of the 24 residues that are found in the T1 tetramer surface, only 6 are common to the BTB-ZF dimer interface (Figure 2). Thus, a unique set of residues has evolved in the T1 domain to mediate tetramerization.

A large expansion of MATH-BTB proteins occurred in C. elegans, where 46 of 178 total BTB proteins belong to this family, whereas other genomes contain many fewer of these proteins (Figure 5). MATH proteins as a whole are largely expanded in C. elegans, with 95 examples present in the Pfam database 44. The MATH domain is thought to be a substrate recognition module in Cul3-based SCF-like complexes 1516.

MATH-BTB proteins differ from most other BTB families in that the BTB domain is found carboxy-terminal to the partner domain. Typically, there are an additional 75 to 100 amino acids following the BTB domain that are likely to be structured and rich in α-helices (Additional data file 1). In contrast to the BTB-ZF proteins, but similar to the BBK proteins, MATH-BTB sequences are highly conserved across the full lengths of the proteins. As a result of this conservation, phylogenetic clustering of the full-length protein sequences can be done with reasonable bootstrap values and shows a clear demarcation between proteins from C. elegans and those from all other species (data not shown). The domain in the C. elegans proteins lacks several BTB signature sequences, such as the 'AH[RK]XVLAA' signature in the B2-A1 region seen in many other long form BTB families (Figure 2). The majority of MATH-BTB proteins from all organisms are predicted to contain the long form elements β1, α1 and β5 (Additional data file 1) and we predict that these BTB domains are dimeric. Indeed, biochemical and biological evidence suggest that BTB-mediated dimerization of the MATH-BTB protein maternal effect lethal (MEL)-26 is required for its function 15100.

Another large expansion is found in Arabidopsis, which contains 21 BTB-NPH3 proteins, or over 25% of the BTB proteins in this genome. BTB-NPH3 proteins are not found in any of the other genomes that we considered, and could represent a plant-specific adaptation of the BTB domain. BTB-NPH3 proteins are involved in phototropism in A. thaliana and are thought to be adaptor proteins that bring together components of a signal transduction pathway initiated by the light-activated serine/threonine kinase NPH1 101102. Heteromerization of BTB-NPH3 proteins have been observed, and the BTB domains of root phototropism (RPT)2 and NPH3 have been shown to interact 101102. In addition, the BTB domain from RPT2 can interact with a region of phototropin 1 that contains light, oxygen and voltage sensing (LOV) protein-protein interaction domains 103. These proteins consist of an amino-terminal BTB domain and an NPH3 domain (Figure 6). The BTB domains in this family are only distantly related to other examples of the fold, and appear to have two leading β-strands in a region preceding the core fold, with an additional β-strand between A1 and A2 (Additional data file 1).

Each of the vertebrate genomes considered here contain genes for two BTB-bZip proteins, named BTB and CNC homology (BACH)1 and BACH2 104105, except for Danio rerio, which has three. These proteins are transcription factors and most closely resemble the BTB-ZF proteins in terms of the BTB sequence and overall protein architecture. The proteins consist of a long form BTB domain, a central region of approximately 400 residues, and a carboxy-terminal basic leucine zipper region (Figure 6). The close similarity of the BTB sequences between the BTB-ZF and BTB-bZip proteins suggest that these domains are likely to be similar in structure. Notably, the long form elements and β5 are predicted, and dimerization residues are similar to the ZF class (data not shown). Accordingly, the BACH proteins have been shown to dimerize and oligomerize in a BTB-dependent manner 63. bZip domains themselves are known to dimerize and, interestingly, the majority of bZip-containing proteins (550 of 738 Pfam bZip_1 domain) contain no other identified domains in the full-length protein 44. Therefore, the domain composition and sequence properties of BTB-bZip proteins are unusual in the context of all bZip proteins, but are compatible with dimeric, and most likely oligomeric, BTB transcription factors.

The Ras homology (Rho)BTB proteins have an unusual architecture, and contain a Rho GTPase domain near the amino terminus, two tandem long form BTB domains, and an approximately 100 residue carboxy-terminal tail with predicted α-helical content (Figure 6). These proteins are highly conserved across their full-lengths, and three examples (RhoBTB1, RhoBTB2/DBC2, RhoBTB3) are found in each of the vertebrates included in this study 106107108. One RhoBTB protein is also present in the insects and in Dictyostelium 107. The first BTB domain of human RhoBTB2 has been shown to interact with Cul3 13 and contains a large 115 residue insertion between A2 and B3, while the second domain is more typical and most closely resembles BTB domains from BBK proteins. The tandem domains are immediately adjacent and may form an intramolecular dimer.

Mutations have been identified in lung cancer patients that do not disrupt the RhoBTB2-Cul3 interaction 13, and these map to regions outside of the predicted Cul3-interacting region (see below). We predict, however, that the Y284D cancer mutation is found in the dimerization interface of the first BTB domain and prevents the proper folding of the domain. This would be analogous to mutants in the dimer interface of PLZF that abrogate function by affecting the folding of the domain 6. The PLZF and BCL6 BTB domains are obligate dimers, and cannot fold as stable monomers (unpublished observation and 43).

Sequence analysis on proteins with the BTB-BACK architecture but no kelch repeats revealed the presence of a conserved carboxy-terminal region of approximately 170 residues. This region in the BTBD1 and BTBD2 proteins has sequence similarity with human protein associated with myc (PAM; NCBI accession number AAC39928), Drosophila highwire (AAF76150) and C. elegans regulator of presynaptic morphology (RPM-1; NP_505267.1) and has been called the 'PHR-like' region (Pfam accession PF08005). It has been shown to interact with topoisomerase 1 109.

Searches with various PHR domain sequences against the Pfam, Prodom and SMART databases identified only automatically generated alignments, and BLAST searches against the PDB did not reveal any significant hits. The domain does not contain extended regions of disorder, and secondary structure predictions suggest that the PHR domain is an all-β fold. Despite the lack of a strongly repeating sequence motif, the PHR may represent a novel type of β-propeller structure, by analogy with the BBK proteins. Using HMM searches, we found from four to seven examples of BTB-BACK-PHR (BBP) proteins in the metazoan genomes, including mammalian BTBD1, BTBD2, BTBD3 and BTBD6. We adopted the name 'PHR domain' for this motif and it has been added to the Pfam database as accession PF08005.

Ankyrin repeats are common protein-protein interaction motifs that are found in proteins of very diverse function, such as transcription regulators, ion transporters and signal transduction proteins 110111. We found examples of BTB-ankyrin proteins in each species that we considered, although, unlike other BTB domain families, these proteins do not fit a single canonical arrangement. For example, some BTB-ankyrin proteins are composed of an amino-terminal BTB domain, a central helical region, 19 ankyrin repeats and a carboxy-terminal FYVE domain (a domain originally found in Fab1, YOTB, Vac1, and EEA1 proteins; Pfam accession PF01363), whereas other examples contain two ankyrin repeats followed by a linker region, two tandem BTB domains, and a 300 residue carboxy-terminal helical region. The three BTB-ankyrin proteins from S. pombe (Btb1p, Btb2p, Btb3p) are components of a SCF-like ubiquitin ligase complex and interact with Pcu3p, a Cul3 homolog 17. Both BTB domains of Btb3p are necessary for this interaction. The BTB sequences from these proteins are only distantly related to other BTB domains, and we thus cannot reliably predict the nature of their interaction surfaces.

A significant number of BTB proteins do not contain other identified sequence motifs (Figure 5). Excluding the Skp1 and ElonginC proteins, 52% of the C. elegans BTB proteins, but only 17% of the human proteins, belong to this family. There may be additional domains in some of these proteins that have yet to be identified.

Several members of the BTB families described here have been found to interact with Cul3-based SCF-like complexes including BTB-ZF 14, BBK 1214112, MATH-BTB 141516, RhoBTB 13, BTB-ankyrin 17, BTB-only 1417 and T1-Kv 16 proteins. The roles of Skp1 and ElonginC as integral components of SCF and VCB complexes, respectively, have long been established 86113. In SCF complexes, F-box proteins such as Cdc4 form precise complexes with Skp1 helices α7 and α8 via their F-box, thus positioning their ligand-binding carboxy-terminal WD40 β-propeller domain such that bound substrate is ubiquinated by the E3 ligase 2526.

Nine of the 49 human BBK proteins have been identified as components of Cul3-based SCF-like complexes 1214 and, in several cases, the BTB domain is necessary and sufficient for interaction with Cul3. We propose that the BBK proteins are structurally analogous to the two-chain Skp1/Fbox or ElonginC/SOCS box complexes 79. In these cases, the central BACK domain would serve to position the carboxy-terminal β-propeller kelch repeats for substrate recognition 114. We expect a similar situation in the BBP proteins, where the PHR domain would act at the substrate recognition module.

BTB domains of 5 of the 46 MATH-BTB proteins from C. elegans have been shown to interact with Cul3. As in the BBK proteins, the MATH-BTB proteins are conserved over much of their entire length, and are likely to be internally rigid. In this scenario, the substrate-recognizing MATH domain is found amino-terminal to the BTB domain, but since the amino and carboxyl termini are very close to each other in the long form BTB domain dimer 522, the MATH domain in these proteins may occupy a similar spatial position relative to the BTB dimer as the BACK-kelch region of BBK proteins.

Some BTB-ZF proteins, including PLZF, have also been shown to bind to Cul3, presumably in a BTB-dependent mode 14. The role of these proteins in Cul3-based SCF-like complexes pose a puzzle, as we do not expect that downstream ZF domains maintain a fixed orientation relative to the BTB domain due to the structurally disordered central region. Further work will be required to understand the structure and function of BTB-ZF proteins in SCF-like complexes.

To aid in understanding the role of the BTB domain in the SCF-like complex, we generated a structural model of a BBK protein dimerized via its BTB domain in a complex with Cul3, Rbx1, E2 and ubiquitin (Figure 7). Three different structures of Skp1 complexes are known 242526, including a Cul1-Skp1 complex 24. We generated a homology model of human Cul3 based on the structure of Cul1, and placed the PLZF BTB dimer by superposing one chain of the dimer with Skp1. Residues in Skp1 that interact with Cul1 are found at positions that do not involve the dimer interface residues in PLZF (Figures 2 and 4). The BTB domain from the BTB-ZF, BBK and MATH-BTB and BTB-bZip families are closely related (Figure 6) and contain mostly the long form of the domain, as discussed above. We predict these to form obligate dimers, similar to those observed in PLZF and BCL6 52255. Proteins from each of these families have been shown to interact with Cul3; therefore, it is reasonable to postulate that these BTB domains drive the dimerization of Cul3 complexes. Indeed, dimerization of adaptor proteins is known to occur 115. The resulting model is similar to the model presented for the ubiquitin-E2-SCF(Cdc4) 26 and E2-SCF^β-TrcP1 complexes 25, except that two ligand-binding kelch/WD40 domains and two E2-ubiquitins localize to the same face of the dimeric complex. In each BBK protein, the BACK domain is between the amino-terminal BTB domain and the carboxy-terminal ligand binding domain, and is likely to be important for positioning the substrate in the complex. A more precise model for a dimeric Cul3-based E3 ligase complex will require the structure of the BACK domain.

Interestingly, some T1-Kv proteins interact with Cul3 16, and an equivalent analysis allows the placement of the T1 tetramer into a model of the SCF-like complex (data not shown), although the tetramerization interface is not fully separate from the putative Cul3 interface (Figure 2). Minor structural adjustments that are not evident from the homology modeling may be required in these cases.

Twenty-five entries comprising nine unique BTB structures were retrieved from the PDB with DALI 116, CE 117 and VAST 118 structure superposition searches. Structural superpositions and sequence alignments were generated with CE, SwissPDBViewer 119 and by manual inspection and adjustments. RMSD values were calculated using SwissPDBViewer, and molecular representations were generated with Pymol 120.

A panel of HMMs describing various families of BTB proteins were trained on structure-guided, manually inspected sequence alignments of BTB domains from the BTB-ZF, BBK, MATH-BTB, T1, Skp1, ElonginC and BTB-NPH3 families. HMMs were matured by iteratively building the results from multiple rounds of sequence search, alignment and training. HMM training and calibration were done with hmmbuild and hmmcalibrate, using default options, from HMMER 2.3.2 34. Family-specific HMMs, including long-form BTB domain HMMs, are available at 35.

All peptides from the translations of all known and predicted transcripts in the genomes of Anopheles gambiae, Apis mellifera, Caenorhabditis elegans, Canis familiaris, Danio rerio, Drosophila melanogaster, Gallus gallus, Homo sapiens, Mus musculus, Pan troglodytes, Rattus norvegicus, Takifugu rubripes and Xenopus tropicalus were retrieved from the latest version of Ensembl 32. Arabidsopsis thaliana, Saccharomyces cerevisiae and Schizosaccharomyces pombe protein sequences were retrieved from Uniprot 46. Dictyostelium discoideum protein sequences were retrieved from Dictybase ('primary features') 121. Proteins containing BTB domains were identified using hmmsearch from the HMMER package 34, with an e-value cutoff of 10, using our panel of HMMs. BTB domains scoring in the e-value range 0.1 to 10 were manually inspected. Peptide sequences, identifiers, names and aliases, domain boundaries of the non-BTB domains (from Pfam annotations 44 included in the Ensembl peptide features) were stored in an Oracle database.

Secondary structure predictions on representative members of each BTB family were completed using the PredictProtein server and the PHD algorithm 122. Scores above 8 over at least 4 consecutive residues were considered valid predictions. Low complexity regions were detected using SEG, at the PredictProtein server. Regions of inherent sequence disorder were detected using the PONDR 123 and DISOPRED 124 servers.

Family-specific HMMs were utilized to generate multiple sequence alignments, which were then merged into larger alignments for clustering. Phylogenetic clustering was attempted with the distance, maximum parsimony and maximum likelihood algorithms in the PAUP*4.0 125, MEGA 2.0 126, Clustal 127 and PHYLIP 3.63 128 software packages. The most probable sequences shown in Figure 2 were retrieved using the hmmemit program from the HMMER package 34. The source code for hmmemit was modified to emit consensus sequences with a probability of 0.4, 0.6 and 0.8 from HMMs for each of the seven families shown in Figure 2.

A model of the ubiquitin-E2-Cul3-Rbx1-BBK complex was generated following the approach used in making the ubiquitin-E2-SCF(Cdc4) model 26. The BBK model was made from a composite of the Skp1 and F-box proteins from the Skp1/Cdc4 26 and Cul1-Rbx1-Skp1-Skp2 complexes 24, in which one chain of the PLZF BTB dimer 22 was substituted for Skp1, and the BACK domain was assumed to adopt the same structure as Skp1 helices α6 and α7 and the F-box and helical linker regions. The Keap1 kelch domain 114 was used to replace the β-propellers of the Cdc4 WD40 domain. Cul1 was replaced by a homology model of Cul3 that was generated using the 3D-PSSM server 129. The E2 enzyme Ubch7 was positioned using a superposition of the RING domains from Rbx1 and c-Cbl from the c-Cbl-Ubch7 complex 130, and the placement of ubiquitin was achieved by superposition of the two E2 enzymes Ubch7 and E2-24 from the structure of the E2-24-ubiquitin complex 131.