To identify the kelch-repeat proteins encoded in the human genome, BLAST and PSI-BLAST searches of the human genome predicted protein database were carried out with the kelch-motif consensus (CDD543, Pfam01344, SMART 00612) as a query sequence. This search identified 57 kelch-repeat proteins and hypothetical proteins. We noted that several of the known human kelch-repeat proteins 3 were not identified by this method, probably because there are relatively few consensus residues in each kelch motif, none of which is completely invariant across all examples of the motif, and also because of variation in the lengths of the loops between the β-strands 236. 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 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 kelch-repeat 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 identification 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 5689. 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 24.

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" 34, (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).

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 2122.

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 ubiquitin-ligase assembly for ubiquitination and targeting to proteosome-mediated degradation 24. The combination of F-box 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 2122. Several of these function in light-dependent regulation of the circadian clock, but the function of many others is obscure 212225. 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 CUB-domain, kelch repeats, a C-type lectin domain and EGF-like 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) 2829. 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 3031.

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).

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 kelch-repeat domains.

CLUSTALW multiple sequence alignment of the kelch-repeat 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 β-sheets 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 β-sheet 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).

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 previously-characterised 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).

We wished to compare the evolutionary development of kelch-repeat proteins between humans and modern invertebrates, and so repeated the analysis of kelch-repeat proteins and their structural subgroups encoded in the genomes of D. melanogaster, A. gambiae and C. elegans 333435. 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 137, 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 424344. 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 (Table 3), 34. Instead, two hypothetical proteins with distinctive domain architectures were identified : NP_506605 which also contained a cyclin carboxy-terminal domain (CDD 7965, Pfam 02984, SMART 00385) and NP_506602, that contained a RING domain (CDD 8941, Pfam 00097, SMART 00184). The cyclin carboxy-terminal domain forms an α-helical fold that may constitute a protein interaction site 45. The RING domain is a zinc-finger fold that mediates protein-protein interactions 46.

Several kelch-repeat proteins have been studied functionally in budding and fission yeast but none of these correspond to BTB/kelch proteins 34748. We investigated whether the prevalance of the BTB/kelch domain architecture we had identified in multicellular animals extended to yeast, by analysing the complement of kelch-repeat proteins encoded in the S. pombe and S. cerevisiae genomes 4950. 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 34748 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 3851. 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 5253. 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.

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, 212256, 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 partially-sequenced eucaryotic animal and plant genomes at NCBI (Entrez/genome_tree, 57), that included the fully-sequenced 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.

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 (PRSGAGVVVVGGKIYVIGGFDGSQSLSSVEVYDPETNTWEKLPSMP; CDD543) by the basic local alignment search tools BLASTP and PSI-BLAST 6768 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 kelch-repeats. 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.

β-propeller closure mechanisms for the human kelch-repeat 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.compbio.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.

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. Alignments, 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/.