Molecular phylogeny of the kelch-repeat superfamily reveals an expansion of BTB/kelch proteins in animals

Background The kelch motif is an ancient and evolutionarily-widespread sequence motif of 44–56 amino acids in length. It occurs as five to seven repeats that form a β-propeller tertiary structure. Over 28 kelch-repeat proteins have been sequenced and functionally characterised from diverse organisms spanning from viruses, plants and fungi to mammals and it is evident from expressed sequence tag, domain and genome databases that many additional hypothetical proteins contain kelch-repeats. In general, kelch-repeat β-propellers are involved in protein-protein interactions, however the modest sequence identity between kelch motifs, the diversity of domain architectures, and the partial information on this protein family in any single species, all present difficulties to developing a coherent view of the kelch-repeat domain and the kelch-repeat protein superfamily. To understand the complexity of this superfamily of proteins, we have analysed by bioinformatics the complement of kelch-repeat proteins encoded in the human genome and have made comparisons to the kelch-repeat proteins encoded in other sequenced genomes. Results We identified 71 kelch-repeat proteins encoded in the human genome, whereas 5 or 8 members were identified in yeasts and around 18 in C. elegans, D. melanogaster and A. gambiae. Multiple domain architectures were identified in each organism, including previously unrecognised forms. The vast majority of kelch-repeat domains are predicted to form six-bladed β-propellers. The most prevalent domain architecture in the metazoan animal genomes studied was the BTB/kelch domain organisation and we uncovered 3 subgroups of human BTB/kelch proteins. Sequence analysis of the kelch-repeat domains of the most robustly-related subgroups identified differences in β-propeller organisation that could provide direction for experimental study of protein-binding characteristics. Conclusion The kelch-repeat superfamily constitutes a distinct and evolutionarily-widespread family of β-propeller domain-containing proteins. Expansion of the family during the evolution of multicellular animals is mainly accounted for by a major expansion of the BTB/kelch domain architecture. BTB/kelch proteins constitute 72 % of the kelch-repeat superfamily of H. sapiens and form three subgroups, one of which appears the most-conserved during evolution. Distinctions in propeller blade organisation between subgroups 1 and 2 were identified that could provide new direction for biochemical and functional studies of novel kelch-repeat proteins.


Background
The kelch motif is an ancient and evolutionarily-widespread sequence motif of 44-56 amino acids in length. Eight residues within the motif are highly-conserved and constitute a consensus sequence [1][2][3] (Fig. 1A). Kelch motifs occur as groups of five to seven repeats and have been identified in proteins of otherwise distinct molecular architecture, termed the kelch-repeat superfamily. Currently, over 28 kelch-repeat proteins have been sequenced and functionally characterised in diverse organisms including viruses, plants, fungi and mammals [3]. Several kelch repeat-containing proteins have been recognised in Bacteria and Archaea (GenBank NP_713516, NP_639451 and Pfam01344 species distribution link), revealing the universal nature of the repeats. On the basis of the crystal structure determined for a single kelch-repeat protein, Hypomyces rosellus galactose oxidase (PDB 1GOF), the sets of repeated kelch motifs are predicated to form a β-propeller structure [2,4].
The β-propeller is a widespread and universal protein domain fold, as revealed from crystal structures of diverse proteins [5][6][7]. Many unrelated primary sequences can adopt the stereotypical topology of a β-propeller; however, several sequence repeat motifs have been identified in relation to known crystal structures that are predictors of propeller domains. These include the WD motif, the regulator of chromosome condensation 1 (RCC1) motif and the tachylectin-2 repeat, as well as the kelch motif [3,[5][6][7][8][9]. Each "blade" of a β-propeller structure is formed from a four-stranded antiparallel twisted β-sheet fold, in which each β-sheet packs onto the adjoining sheets through hydrophobic contacts (Fig. 1B). Four to eight of these β-sheet modules are radially arranged around a central axis to form the β-propeller domain, with the twist of the β-sheets within each module producing the propellerblade appearance (Fig. 1C). Intra-and inter-blade loops of varying lengths protrude above, below, or at the sides of the β-sheets and contribute variability to the binding properties of individual β-propellers [3,5,6] (Fig. 1B). The whole structure is closed and stabilised by interactions between the first and last blades. In the four-bladed β-propellers of hemopexin and collagenase this is achieved by disulphide bonding between the first and last blades [10,11]. In the other examples for which there is crystal structural information, the last blade of the β-propeller is assembled as a composite from sequences at the aminoand carboxy-terminal ends of the domain, that are held together by hydrogen bonds [3,5,6,9] (example of galactose oxidase shown in Fig. 1C).
Kelch-repeat β-propellers undergo a variety of binding interactions with other molecules. Several kelch proteins, including D. melanogaster kelch and mammalian IPP, bind and cross-link F-actin through the β-propeller domain [12,13]. In contrast, the kelch repeats of other proteins, such as gigaxonin, Keap1, or recombinationactivating gene 2, (RAG-2), have unique binding partner proteins [14][15][16] and the β-propeller of fungal galactose oxidase corresponds to the catalytic domain of the enzyme [2,4]. The kelch-repeat β-propeller is thus considered to form a general protein-protein interaction module that associates with diverse and specific binding partners [3]. Several kelch-repeat proteins contain other protein domains, of particular note being the Broad-Complex, Tramtrack, and Bric-a-Brac/Poxvirus and Zincfinger (BTB/ POZ) domain (CDD6184, Pfam 00651, SMART0225) [1,3,17]. The BTB/POZ domain was first identified as a dimerisation domain of transcription factors and also mediates homodimerisation of kelch and other BTB/kelch domain proteins [10,17,18].
As summarised above, the current view of the kelch-repeat superfamily has been compiled from studies of individual kelch-repeat proteins from diverse organisms. The extent and complexity of the kelch-repeat family within an individual species is unknown. An overview of this domain family would be of value in order to make a rational assignment of structural subgroups and to develop a coherent perspective on the evolution of this protein domain between modern organisms. A clear view of kelch-repeat proteins could also open up new possibilities for sequence analysis, prediction and experimental analysis of structure/function relationships, and a greater understanding of the specific properties of kelch-repeat proteins within the large β-propeller fold group. In modern life sciences, such molecular information provides a fundamental context for well-targeted biochemical and biological studies of individual proteins. Building on the availability of complete genome sequence information for H. sapiens, D. melanogaster, A. gambiae, C. elegans, S. cerevisiae, S. pombe and A. thaliana we have addressed these questions through database mining and bioinformatic sequence analyses.
the β-strands [2,3,6]. Therefore further searches were made with the kelch-repeats of all 28 known superfamily members, as described in the Methods. These searches identified 18 additional kelch-repeat proteins encoded in the human genome. Cross referencing all 75 entries against GenBank identified 9 of the entries as partial sequences and/or duplicate entries for the same protein or hypothetical ORF, and two of the entries as non-kelch containing proteins. We also cross-referenced the search results to the domain entries for kelch in the Pfam [19] and SMART [20] domain databases. Many entries were listed in both SMART and Pfam, however a number of the proteins we had identified were not listed in these databases (indicated in Table 1), even though when we The kelch motif and the beta-propeller fold Figure 1 The kelch motif and the β-propeller fold. A, Consensus sequence of the kelch motif. Compiled from [2] and [3]. The sequence motif is shown in relation to the four β-sheets of a propeller blade structure, as determined for the kelch motifs of fungal galactose oxidase [4]. In the consensus, G= glycine, Y = tyrosine, W = tryptophan, s = small residue; l = large residue; h = hydrophobic residue. B, Structure of a kelch-repeat propeller blade. A single blade from the crystal structure of fungal galactose oxidase (1GOF) is shown. β sheets 1-4 are colored as in panel A. The N-and C-termini join to adjacent blades. As indicated by the mapping of the consensus amino acids onto the blade, the most highly-conserved residues are located in the βsheets. In various examples of β-propeller proteins, the intra-and inter-blade loops have variable sequences and contribute to protein-protein interactions [3]. C, Structure of the kelch-repeat β-propeller domain of fungal galactose oxidase (1GOF). βsheets 1-4 in each blade are colored as in panel A. As indicated, this β-propeller contains seven blades. The β-4 strand of blade 7 is derived from the amino-terminus of the domain and thus closes the circular structure.  searched these polypeptides against SMART or Pfam, kelch motifs were clearly identified. Furthermore, the numbers of kelch repeat proteins assigned to H. sapiens in the Species tree or Taxbreak links of Pfam and SMART were over-estimated because of inclusion of incomplete ORFs and multiple entries for the same polypeptide. We also carried out additional searches of GenBank with single kelch motifs from the 28 known kelch-repeat proteins that were distinctly longer than the CDD kelch-motif consensus, in order to search more extensively for proteins containing more divergent repeats. From these multiple evaluations and with exclusion of partial sequences (as described in the Methods), we identified at least 71 kelchrepeat proteins encoded in the human genome (Table 1).
To determine the number of repeated kelch motifs in each protein or hypothetical protein, BLASTP searches were made with each sequence against the Conserved Domain Database (CDD) and Pfam, together with manual identi-fication of kelch motifs. The number of kelch motifs identified varied from two to seven. Four blades is the minumum number that has been documented from crystal structures of β-propeller domains [5,6,8,9]. Thus, it appeared unlikely that entries encoding two or three kelch motifs corresponded to complete ORFs and these were excluded from further analysis (entries NP-689579, XP_209285, XP_058629). On this basis, 12.7 % (9/71) of sequences were predicted to contain five-bladed β-propellers, 84.5 % (60/71) to be six-bladed, and 2.8 % (2/71) to contain seven-bladed β-propellers (Table 1). To our knowledge, only one seven-bladed kelch repeat protein has been identified previously, fungal galactose oxidase [2,4].
In galactose oxidase, the single kelch-repeat protein for which there is crystal structure information, the propeller is circularised by formation of a composite seventh blade, with the β-one to β-three strands provided from the most  C-terminal sequence repeat and the β-four strand provided by sequence amino-terminal to the first full sequence repeat, a mechanism referred to as "N-terminal β-strand closure" [3,4], (Fig. 1C). We examined the the human kelch-repeat proteins by secondary structure prediction of β-sheets and by manual analysis of the sequence repeats, and found that for 77.5 % (55/71) of the proteins the β-propeller structure was predicted to be closed by a C-terminal β-strand. For five sequences, no clear prediction could be made (Table 1).

Chromosomal localisation of Human kelch-repeat proteins
The encoding sequences for human kelch-repeat proteins are dispersed throughout the genome, being located on all chromosomes except chromosome 21 and the Y chromosome (Table 1). Several instances of genes in physical proximity were noticed, for example NP_006460 and NP_067646 at 1q31.3 and NP_569713 and NP_060114 at 3q27.3 (Table 1). However, in the majority of cases these did not correspond to the most closely-related protein sequences as would be expected for recently-duplicated genes. One exception was NP_055130 and NP-751943 which were located at 14q21.3 and which were the most closely-related to each other (46 % identity).
Overall, there was no evidence for physical grouping of kelch-protein encoding sequences within the human genome. In contrast, genes encoding the numerous F-box/ kelch proteins of A. thaliana are clustered such that some of the most highly-related sequences are encoded from physically close genomic locations [21,22].

Domain Architecture of Human Kelch-repeat proteins
Twenty-eight kelch-repeat proteins from various organisms were previously grouped into 5 structural categories according to the positioning of the kelch repeats within the polypeptide sequence and the presence of other conserved structural domains [3]. To evaluate the complexity of domain architectures within a single organism, each human kelch-repeat protein sequence was re-analysed by searching against CDD, SMART and Pfam and then subgrouped according to domain architecture.
Strikingly, 72 % (51/71) of the human kelch-repeat proteins contained a BTB/POZ domain. In all but one of the proteins, the BTB domain was amino-terminal to the kelch domain (Table 1). This hypothetical protein, LZTR-1, contained two tandem BTB domains. Four (5.6%) kelch-repeat proteins contained a single additional conserved domain. Muskelin was the only kelch-repeat protein identified in the human genome to contain a discoidin domain (CDD 7753, Pfam 00231, SMART 00231, also known as F5/F8 type C domain) (Prag, Collett and Adams, in preparation). The discoidin domain acts as a protein-protein interaction domain in a number of extracellular and intracellular proteins and, in clotting factors V and VIII, mediates phospholipid binding [23]. Another kelch-protein, XP_048774, contained a F-box domain (CDD9197, Pfam 00646). The F-box is a domain of about forty residues, first identified in cyclin A, that interacts with Skp1 to anchor proteins to the ubiquitinligase assembly for ubiquitination and targeting to proteosome-mediated degradation [24]. The combination of Fbox and kelch-repeat domains has previously been described in A. thaliana, where at least 67 F-box/kelch proteins and hypothetical proteins are encoded in the genome [21,22]. Several of these function in lightdependent regulation of the circadian clock, but the function of many others is obscure [21,22,25]. To our knowledge, this is the first recognition of a F-box/kelch protein in an animal genome. One predicted kelch-repeat protein, NP_055608, contained a leucine carboxyl methyltansferase (LCM) domain (CDD9631, Pfam 04072) with 34% identity to the LCM domain of protein phosphatase 2 leucine carboxyl methyltransferase [26]. Recombination-activating gene-2 (RAG-2) contains a plant homeodomain (PHD) finger domain (Pfam00628) at the carboxy-terminus [16].
Six kelch-repeat proteins (11 %) were very large, multidomain proteins (Table 1). Attractin/mahogany (that are splice variants from a single gene; 27-29) and MEGF8 are each over 1000 amino acids long and contained a CUBdomain, kelch repeats, a C-type lectin domain and EGFlike domains. Diverse functions have been attributed to attractin and mahogany that include a role in T-cell interactions (attractin, the secreted splice variant) [27] and obesity regulation in mice (mahogany, the transmembrane splice variant) [28,29]. Host cell factor-1 and -2 (HCF-1 and HCF-2) are also large proteins which contain amino-terminal kelch-repeats, two fibronectin type III domains and, in the case of HCF-1, a series of unique HCF repeats. These proteins function as transcriptional coactivators of herpes simplex virus immediate early gene expression [30,31].
We identified three hypothetical kelch-repeat proteins as containing unrelated unique sequences that did not correspond to recognised structural domains, positioned either amino-or carboxy-terminal to the kelch repeats (Table 1).
Rab9 effector p40 [32] and six other kelch-repeat proteins were short polypeptides, from 350-442 amino acids in length, that consisted almost entirely of kelch repeats ( Table 1). Five of these proteins or hypothetical proteins, including p40, contained six sequence repeats and thus are predicted to form six-bladed β-propellers. Two hypothetical proteins, NP_060673 and XP_114323, consisted of putative seven-bladed β-propellers. Together, these structural distinctions form the basis for the novel categorisation of human kelch-repeat proteins that is presented here (Table 1).

Structural relationships of human BTB/kelch proteins
The unexpectedly large number of BTB/kelch proteins encoded in the human genome prompted us to study this group in more detail, with the aim of identifying structural subgroups that might also represent functional subsets. The 38 full-length sequences that contained single BTB domains and predicted six-bladed β-propellers were aligned according to sequence similarity in CLUSTALW and viewed as neighbourhood-joining trees. Alignment of the full-length sequences revealed three subgroups of approximately equal size, which we termed subgroups 1 to 3 ( Fig. 2A). When the same analysis was performed with the kelch domains alone, the same grouping was apparent for subgroup 1 and a substantial proportion of subgroup 2, termed subgroup 2A (Fig. 2B). In an alignment of the BTB domains only, subgroups 1 and 2 were maintained for the majority of sequences (Fig. 2C).
Unrooted trees produced by a separate method for alignment based on maximum parsimony analysis of sequences, PROTPARS, did not support subgroup 3 but consistently demonstrated the relationship of the sequences in subgroups 1 and 2A (data not shown). We focused on these robustly-related kelch-repeat sequences in subgroups 1 and 2A, for a closer analysis of the kelchrepeat domains.
CLUSTALW multiple sequence alignment of the kelchrepeat domains from each of subgroups 1 and 2A demonstrated distinctive features in terms of repeat organisation. In both subgroups ( Fig. 3 and Fig. 4), the intrablade loop between β-strands 2 and 3 (the 2-3 loop, Fig. 5A) and the interblade 4-1 loop were major sources of variation within the repeats with regard to their length and primary structure. In the context of an intact β-propeller domain, the 1-2 and 3-4 loops protrude above one face of the βsheets and the 2-3 loop protrudes from the opposite face (Fig. 5A). The 4-1 loop lies either on the same face as the  2-3 loop, or may be positioned more closely to the βsheet core of the propeller (Fig. 5). In subgroup 1, the longest 2-3 loops were found in repeats 1, 5 and 6, with shorter loops in blades 2, 3 and 4. The longest 4-1 loop were that between repeats 5 and 6 ( Fig. 3). In the context of a β-propeller, this suggests that the side of the propeller formed by repeats 5, 6 and 1 may be particularly involved in protein interactions (see Fig. 1C). In subgroup 2A, the longest 2-3 loops were those in repeats 1 and 2, repeats 4 and 5 had intermediate 2-3 loops and repeats 3 and 6 contained the shortest 2-3 loops. The longest 4-1 loops were those between repeats 1 and 2, and repeats 3 and 4 ( Fig. 4). This suggests that there is a different organisation of binding sites in subgroup 2A β-propellers, with perhaps two binding faces formed by repeats 1 and 2, and repeats 4 and 5. At the level of individual sequences, there were also specific examples of variation from the standard repeat organisation that could be of functional importance for individual proteins. For example, NP_695002 in subgroup 2A has an unusually long and highly charged 3-4 loop in repeat 1 and XP_ 040383 has a long 3-4 loop in repeat 4 (Fig. 4).

Relationships between human BTB/kelch proteins
We also found that the consensus sequences for the fold were distinctive between the two subgroups. The 50 % identity consensus sequence from each subgroup was realigned against the kelch-repeat unit to derive mean 50 % identity consensus sequences for subgroup 1 and subgroup 2A. These motifs were mapped against the known blade structure of galactose oxidase (Fig. 5). The consensus motifs included both amino acids of importance to the fold (located within the β-strands) and certain amino acids within loops, that would be predicted to contribute to binding interactions. Of note, the average length of the motif was shorter in subgroup 1 than subgroup 2A. The subgroup 2A consensus is predicted to contain a longer 2-3 loop. The consensus motifs were distinct in the positioning of highly-conserved charged residues within the loop regions (Fig. 5). The conservation of these charged residues was most pronounced in subgroup 1, where these positions were conserved in the motif to the 70 % identity threshold level (data not shown). These distinctions in loop characteristics are also suggestive of different modalities of protein-protein interactions for the β-propellers of subgroups 1 and 2A. With regard to previouslycharacterised protein-binding properties, we observed that the BTB/kelch proteins that bind to actin were split between subgroups 1 and 3; thus this function does not have a simple relationship to the primary structure ( Fig.  2A).

Kelch-repeat Proteins encoded in invertebrate genomes
We wished to compare the evolutionary development of kelch-repeat proteins between humans and modern invertebrates, and so repeated the analysis of kelch-repeat pro-teins and their structural subgroups encoded in the genomes of D. melanogaster, A. gambiae and C. elegans [33][34][35]. We identified 18 kelch-repeat proteins encoded in the Drosophila and Anopheles genomes ( Table 2). Seventeen of these were orthologues conserved between the two species (the average identity between orthologous genes of D. melanogaster and A. gambiae is 56 % [36]) and one was unique to each species. Thus, an Actinfilin homologue was identified in A. gambiae but not in D. melanogaster and the D. melanogaster genome contained a homologue of NP_116164 which was not present in A. gambiae ( Table 2). Only three kelch-repeat proteins were previously characterised in D. melanogaster, namely Kelch [1,37], Muskelin [38] and Drosophila host cell factor [39]. Two others, diablo and scruin-like at the midline (SLIM-1), have been recognised as kelch-repeat proteins [33].
Within the group of 19 proteins and hypothetical proteins, 95 % contained six kelch-repeats. Only one protein with five kelch-repeats was identified in either D. melanogaster or A. gambiae, which corresponded to an orthologue of the human F-box/kelch protein, XP_048774 ( Table 2). 56 % of the kelch-repeat proteins of D. melanogaster and A. gambiae were BTB/kelch proteins. Both D. melanogaster and A. gambiae contained one discoidin/ kelch protein orthologous to muskelin, one F-box/kelch protein, three kelch and multidomain proteins, one kelch and unique protein, and two propeller-only proteins. Thus, all of the 19 kelch-repeat proteins identified had homologues in the human genome and the BTB/kelch domain architecture was the most prevalent (Table 2).
We identified 16 kelch-repeat proteins encoded within the C. elegans genome (Table 3). Of these proteins, only kel-1, spe-26 and CeHCF have been functionally characterised. Kel-1 is an intracellular protein involved in the regulation of feeding behavior during larval development [40]. Spe-26 contributes to the cellular organisation of spermatocytes and mutations are associated with sterility [41]. CeHCF might be involved in the regulation of cell proliferation [42][43][44]. 43.7 % (7/16) of the proteins had the BTB/kelch domain architecture, two were homologues of HCF and attractin with similar multidomain architectures, two contained unique sequences outside of the kelch repeats and two were propeller-only proteins, both of which were predicted to form six-bladed β-propellers.
A single F-box/kelch protein was identified, but no muskelin-like protein was found (  helical fold that may constitute a protein interaction site [45]. The RING domain is a zinc-finger fold that mediates protein-protein interactions [46].

Kelch-repeat Proteins encoded in yeast genomes
Several kelch-repeat proteins have been studied functionally in budding and fission yeast but none of these correspond to BTB/kelch proteins [3,47,48]. We investigated whether the prevalance of the BTB/kelch domain architecture we had identified in multicellular animals extended Relationships of consensus kelch-motifs to propeller blade structure   [1], and is expressed as two isoforms, of 689 amino acids or 1477 amino acids [37]. A. gambiae EAA12172 apparently corresponds to the long form of kelch. * Kelch-repeat proteins not listed in either Pfam or SMART species trees. With the exception of NP_060786, all the most closely-related human BTB/kelch proteins are members of human BTB/kelch subgroup 1.   to yeast, by analysing the complement of kelch-repeat proteins encoded in the S. pombe and S. cerevisiae genomes [49,50]. We found that each genome encoded a small number of kelch-repeat proteins (five in S. pombe, eight in S. cerevisiae), none of which corresponded to a BTB/kelch protein (Table 4). Proteins and hypothetical proteins consisting of an amino-terminal kelch β-propeller and an extended coiled-coil region [3,47,48] and a protein corresponding to a putative leucine carboxyl methyltransferase [26] were common to S. pombe and S. cerevisiae. The other encoded kelch-repeat proteins were non-homologous (Table 4). Muskelin-like 1 protein and Ral-2p were identified in S. pombe but not S. cerevisiae [38,51]. Two proteins with distantly-related kelch repeats, Gpb1/Krh1 and Gpb2/Krh2, have been characterised functionally as G protein-coupled receptor-binding proteins in S. cerevisiae [52,53]. Homologous proteins were not identified in S. pombe in the context of our study. Thus, the BTB/kelch domain architecture was not identified in these yeasts.

Restriction of BTB/kelch proteins to metazoan animals and poxviruses
Because the BTB/kelch domain architecture appeared prevalent in animals but was not identified in yeast we were interested to consider if any other organisms might contain kelch-repeat proteins with this domain architecture. A number of BTB/kelch proteins have been reported as hypothetical open reading frames (ORFs) in the poxvirus family of animal viruses [54]. The Conserved Domain Architecture Retrieval Tool (CDART) database at NCBI lists 333 entries for BTB/kelch proteins, all of which originate from vertebrates, insects, C. elegans or poxviruses. To date, the BTB domain has only been identified in eucaryotes (Pfam 00651 species tree). In addition to reviewing the SMART and Pfam species trees for categorisation of the BTB/kelch domain architecture, we conducted our own BLASTP and TBLASTX searches of the A. thaliana genome database [55] with the CDD kelch motif consensus (this search tool identified 44 BTB/kelch proteins from the human genome and is thus very effective in uncovering these proteins) and identified 72 protein sequences, the majority of which were F-box/kelch proteins, some of which were serine-threonine phosphatase/ kelch proteins, [21,22,56], and none of which were BTB/ kelch proteins. Searches with the BTB domains of several human or invertebrate kelch-repeat proteins also did not identify BTB/kelch proteins in A. thaliana. BLAST genomes searches of the databases of complete or partiallysequenced eucaryotic animal and plant genomes at NCBI (Entrez/genome_tree, [57]), that included the fullysequenced genomes of the Apicomplexium Plasmodium falciparum [58], the Microsporidium Encephalitozoon cuniculi [59], the plant Oryza sativa (rice; [60]) and the fungus Neurospora crassa [61] identified many predicted kelch-repeat-containing proteins, but no ORFs that had the BTB/kelch domain architecture. Results for selected domain architectures in five eucaryotic organisms are presented in Fig. 6. We did, however, note in Apicomplexia species, two proteins with K Tetra /kelch domain architecture (NP_705330 and EAA22466). The K tetra domain (Pfam 02214) is a distant structural relative of the BTB/ POZ domain [62]. Overall, these results provide a significant indication that protein-encoding sequences for the BTB/kelch domain architecture have become expanded during the evolution of multicellular animals, compared to Apicomplexia, fungi, plants and other eucaryotes.

Discussion
Our analysis of the molecular organisation and phylogeny of kelch-repeat proteins provides a new view on the evolution of the kelch-repeat domain and the proteins that share this domain. Several features distinguish the kelchmotif and its domain superfamily from other sequence motifs that fold as β-propeller structures. First, the consensus sequence of a kelch motif differs from that of other βpropeller motifs, such as the WD motif, RCC1 motif, tachylectin-2 repeat or YWTD motif [3,5,7]. Secondly, the majority of the β-propeller domains that have been crystallised are seven-or eight-bladed β-propeller structures [3,5,7,8]. A geometric preference for assembly of seven-bladed β-propellers over six or eight bladed-forms has been demonstrated by mathematical modelling [63] and indeed the majority of WD motif proteins, a considerably more numerous domain family in metazoa than the kelch-repeat proteins, are predicted to form sevenbladed β-propellers [7,64]. Our comprehensive evaluation of kelch-repeat domains in the proteomes of multiple eucaryotic organisms clearly delineates a preponderance of predicted six-bladed β-propellers in the kelch-repeat superfamily. In addition, WD proteins have been identified as proteins of eucaryotic animals and plants [5,6,64], whereas kelch-repeats are universally represented in eucaryotes, bacteria and viruses.
The data also present new insights into the evolution of domain organisation in kelch-repeat proteins. Kelchrepeat domains are found in proteins from all forms of life, yet the diverse domain architectures of kelch-repeat proteins and the significant differences in the relative representation of domain architecture groups encoded in the genomes of animals, fungi and plants attest to considerable functional diversification and biological specialisation of the superfamily. The prevalence of the Fbox/kelch domain architecture in plants has been recognised [21,22]. We report clear evidence for the prevalence of the BTB/kelch domain architecture in the kelch-repeat proteins of metazoan animals from C. elegans to human (Fig. 6). The expansion of this subgroup appears animalspecific, because although kelch repeats are found in all eucaryotes and the BTB domain is present in fungi and animals, the combination of the BTB/kelch domain organisation was only identified in animals, where the subgroup appears to have expanded as a gene family in conjunction with multicellularity. Other domain architectures, such as F-box/kelch or discoidin domain/ kelch, are present in a wider range of eucaryotic organisms, suggestive of an earlier evolutionary origin.
The other organisms in which BTB/kelch proteins have been identified are members of the animal poxvirus family. In these viruses, which are thought to have acquired these genes by horizontal transfer from animals, BTB/ kelch-encoding ORFs are found with the variable long terminal regions (LTR) of the genomes which determine the species-trophism and strain-specific properties of each virus. [54]. Cowpox virus, which has the broadest host range, contains six BTB/kelch-encoding sequences, whereas in variola virus (smallpox virus), that has a single host, all the kelch/BTB sequences are disrupted by mutations that truncate the ORFs [54,65]. Therefore, it has been suggested that the BTB/kelch proteins encoded in poxviruses could detemine the host range and/or pathogenicity of a virus [54]. Alternatively, the proteins might act to modulate host immune response to virus infection or might be necessary for in vivo host/virus interactions, for example subversion of host cell functions such as cytoskeletal organisation or formation of cell signaling protein complexes, that would affect the ability of the virus to persist within the host. The three BTB/kelch proteins of vaccina virus are not necessary for viral replication in culture and further targeted experiments are needed to establish the function of BTB/kelch proteins in cowpox and other poxviruses [66]. A greater understanding of the normal functions of animal BTB/kelch proteins could help reveal the roles of poxvirus BTB/kelch proteins.
In addition, the large numbers of BTB/kelch proteins identified in metazoan animal genomes, particularly in human, present an interesting enigma. Why do animals need so many BTB/kelch proteins? Only eight out of the 51 human BTB/kelch proteins are of known function (Table 1). Although there are multiple examples of BTB/ kelch proteins that bind actin [3], multiple sequence alignment of the human BTB/kelch proteins showed that the known actin-binding kelch-repeat proteins did not belong to a specific sequence subgroup ( Fig. 2A). This result is not unexpected, given the low sequence constraints for assembly of a β-propeller and the preference for a binding "supersite" on one propeller face [67].
Our studies of the two robust BTB/kelch subgroups indicate distinctions in specific secondary or tertiary structural features of these β-propellers, with regard to loop and βpropeller organisation, that could be important for the functions of these proteins. Although, for the reasons stated above, it is unlikely that the these features specify a common binding partner, we suggest that the differences in positioning of the major loops in the kelch-repeats of subgroups 1 and 2A could provide important indications of the location and characteristics of putative proteinbinding sites. The studies identified general distinctions between the two subgroups (the positioning of the longest 2-3 and 4-1 loops within the domain and the mean length of 2-3 loops) and also unusual features of single repeats in specific proteins. This information could be important for the analysis of proteins within these groups, to accelerate rational design of mutational and functional studies. It would also be of obvious interest to have crystal structure information for a representative β-propeller from each BTB/kelch subgroup and for the other domain architecture groups. Interestingly, the BTB/kelch proteins of insects and C. elegans were all, with one exception, most closely-related to the subgroup 1 human BTB/kelch proteins (Tables 2 and 3, Fig. 2). Thus, subgroup 1 may represent the BTB/kelch proteins of earliest origin from which others have diversified during evolution, making these proteins particularly significant for further study.
The majority of eucaryotic kelch-repeat proteins are predicted to be intracellular proteins. Many proteins in the βpropeller fold family function as enzymatic, scaffolding, or transducer components of protein complexes [5,6,8].
We speculate that many of the currently uncharacterised BTB/kelch-repeat proteins of animals will turn out to be components of multiprotein complexes with roles in reception of extracellular cues, cell signaling and transport, or cell organisation.

Conclusions
The kelch-repeat superfamily constitutes a distinct and evolutionarily-widespread family of β-propeller domaincontaining proteins. Expansion of the family during the evolution of multicellular animals is mainly accounted for by a major expansion of the BTB/kelch domain architecture. BTB/kelch proteins constitute 72 % of the kelchrepeat superfamily of H. sapiens and form three subgroups, one of which appears the most-conserved during evolution. Distinctions in propeller blade organisation between subgroups 1 and 2, with respect to loop length and charge distribution and the positioning of major loops within the β-propeller domain were identified that could provide new direction for biochemical and functional studies of novel kelch-repeat proteins.

BLAST Database searches and identification of Domain Architectures
The whole human genome http://www.ncbi.nlm.nih.gov/ genome/seq/HsBlast.html was searched with the 46 residue consensus sequence for the Kelch motif (PRSGAGVVVVGGKIYVIGGFDGSQSLSSVEVYDPETNT-WEKLPSMP; CDD543) by the basic local alignment search tools BLASTP and PSI-BLAST [67,68] at standard parameters, with a default expectation value of 10 and reporting up to 500 protein sequences. The database was also searched with the kelch repeats from human host cell factor-1 (aa1-aa357; [30]), mouse leucine-zipper-like transcriptional regulator (aa49-aa369; [69]), rat Actinfilin (aa331-aa616; [70]), human Rag2 (aa1-aa348; [16]), human muskelin (aa 249-636; [71]), the complete sequences of all the previously-characterised kelch-repeat proteins [3] and examples of single kelch-repeats that differ in length or sequence characteristics from the kelch motif consensus (repeat 3 of mouse muskelin, repeats 3, 4, 5 of RAG-2, repeat 1 of NP_695002, repeat 4 of NP_040383, repeat 1 of NP_071324). The listings of kelch proteins in Pfam 9.0 [19] (Wellcome Trust-Sanger Institute) and SMART [20] (EMBL) domain databases were also analysed and any apparent additional human kelch-repeat proteins searched against GenBank. From these searches, we found that the species listings in Pfam and SMART did not include all the kelch-repeat proteins identified in our searches and contained multiple redundant entries including partial ORFs for the same protein. Each identified protein sequence was reciprocally searched against the non-redundant protein database of GenBank and Conserved Domain Database [72] (which includes the Pfam and SMART databases) to confirm the identification of kelch repeats and to identify additional domains. Novel hypothetical ORFs within the human genome were used to query the database of human expressed sequence tags (ESTs) to determine whether corresponding translation products could be identified. Hits that appeared ambiguous as putative incomplete open reading frames in the initial search were excluded from further analysis, but were also searched against the mouse and rat genome databases at NCBI. The same search methods were used to identify kelch-repeat proteins in the genome databases of D. melanogaster [33], A. gambiae [34], C. elegans [35], S. cerevisiae [50], S. pombe [49] and A. thaliana [55] at NCBI. A. gambiae proteins were not listed in the Pfam or SMART species trees. All these searches were carried out in the period March 18 to April 20, 2003. A second set of searches were carried out between 14 th June and 22 nd July, 2003. Each identified sequence from these species was used in BLAST searches of the human genome to identify the closest human homologue (ie, an ORF with significant identity to the entire protein sequence, not only to the kelch repeats). Previously unknown proteins identified in these searches were also used to query the human genome database to determine if additional human kelch-repeat proteins could be identified. Other eucaryotic genomes were searched through the BLAST Genomes interface at NCBI [57]. At the time of search, this database included complete or partial genome sequence information for 48 eucaryotic genomes (12 complete), 222 eubacterial genomes and 18 archaebacterial genomes.
Entries in the human proteome which appeared to be partial sequences included GenBank XP_054631, which was 91 % identical to the C3IP1 Kelch-protein (Accession NP_067646). Comparison of the two sequences at the nucleotide level showed that a single nucleotide deletion (ACAGTGG→ACATGG) in the XP_054631 entry generated an additional start-codon, in the equivalent position to valine100 in C3IP1. Additionally, a single mutation in XP_054631 (GTCCGACG→GTCTGACG) generates a stopcodon and consequently a truncated open reading frame. Thus, the XP_054631 and NP_067646 entries appeared to be derived from the same gene and only NP_067646 is included in Table 1. The entry XP_209285 encoded a open reading frame of 128 amino acids, containing 3 kelch-repeats with highest sequence identity to rat actinfilin, NP_663704, length 640 amino acids [70]. Additional searches with rat actinfilin did not identify a full-length human sequence. The entry NP_689579 encoded a 263 residue polypeptide containing 2 kelchrepeats. We identified orthologues in both Mus musculus and Rattus norvegicus of 303 amino acids and 238 amino acids, respectively. Alignments of all three orthologues indicate a highly conserved segment identified in all three orthologues, a segment only present in human, and a segment only present in mouse. These discrepancies may arise from alternate predictions of intron-exon boundaries during genome annotation. All these entries were excluded from Table 1 and from the further sequence analyses at this time.
Prediction of β-propeller closure mechanisms β-propeller closure mechanisms for the human kelchrepeat proteins were predicted by use of Jpred secondary structure prediction, to identify beta-sheet secondary structure either amino-or carboxy-terminal to the kelch motifs in each polypeptide sequence http://www.comp bio.dundee.ac.uk/~www-jpred. Each sequence was also examined manually for the presence of a tryptophan residue amino-terminal to the first motif or carboxy-terminal to the last motif, as a second predictor of amino-terminal or carboxy-terminal β-propeller closure mechanisms, respectively, [3].

Multiple sequence alignment and identification of human BTB/kelch protein subgroups
All the full-length human BTB/kelch protein sequences predicted to form six-bladed propellers (38 sequences) were aligned by CLUSTALW [73] multiple sequence alignment at EMBnet http://www.ch.EMBnet.org, European Bioinformatics Institute http://www.ebi.ac.uk and at the San Diego Supercomputer Center Biology Workbench 3.2 http://workbench.sdsc.edu, with similar results. Align-ments, bootstrap analysis and neighborhood-joining trees were carried out at default parameters from the full-length sequences, from the kelch-repeats alone, or from the BTB domains alone, at the Biology Workbench and are presented as neighbourhood-joining trees (rooted dendrograms prepared by reference to the unrooted trees) prepared in the Phylip Drawgram 3.5c software at Biology Workbench [74], with annotation in Adobe Illustrator. These analyses identified two major robustly-related subgroups within the kelch/BTB proteins. These groupings were substantiated by a second method, PROTPARS (Felsenstein, J. 1993. PHYLIP (Phylogeny Inference Package) version 3.5c, distributed by the author, Department of Genetics, University of Washington, Seattle) for maximum parsimony analysis of sequence relationships, also carried out at Biology Workbench. CLUSTALW multiple sequence alignments of the kelch-repeats from the two subgroups consistently identified by CLUSTALW and PROTPARS were used to derive a 50 % identity consensus sequence for each subgroup and are presented in Boxshade 3.2. The 50 % identity consensus sequence derived from each set of six kelch motifs were re-aligned with each other to prepare a averaged single motif consensus at 50 % identity for each subgroup in CLUSTALW. These averaged consensus motifs were mapped against the tertiary structure of a kelch-repeat β-propeller blade using the structure of fungal galactose oxidase (PDB 1GOF) as a template and the Swiss-PDB viewer software http:// www.expasy.org/spdbv/. cell adhesion and guidance proteins, is secreted by activated