C-type lectin-like domains in Fugu rubripes

Known fish C-type lectins

A number of fish CTLDcp sequences have been reported separately in the literature and public sequence databases. The best-studied and most distinct set are serum antifreeze proteins (AFPs) from several cold-water-living species [7, 16, 17]. These sequences consist mostly of just a CTLD, and were classified as group VII members based on domain architecture. A three-dimensional structure of the sea raven antifreeze protein has been determined experimentally [18].

Apart from AFPs, several other soluble bony-fish CTLDcps have been described: 5 isoforms of Salmo salar serum lectin (SSL) [19], three collectins from different Cyprinidae carp family species [20], skin mucus protein AJL-2 [21] and two C-type lectins (eCL-1 and eCL-2) from gills of Japanese eel [22], two lectins from rainbow trout liver [23], a carp lectin [24], goldfish lectin OL-1 (GI: 26000685, unpublished), and a liver lectin from Gillichthys mirabilis (long-jawed mudsucker), annotated as "mannose receptor C" [25].

Known membrane-bound CTLDcps from bony fishes include a polycystic kidney disease protein 1 (PKD1) orthologue from Fugu [26], a rainbow trout Kupffer cell receptor homologue [27], and a set of putative killer cell receptors (KLR) identified recently [28]. Although predicted coding sequences for CTLDcps from winter flounder (GI:28394504, unpublished) and medaka fish [29] do not contain a recognizable transmembrane (TM) domain, based on CTLD sequence and, in the case of the medaka CTLDcp, domain structure, they should be assigned to group II, as the absence of TM regions may be a result of incomplete prediction.

The only known CTLDcp sequence from cartilaginous fishes is a tetranectin homologue from reef shark cartilage [30].

Fugu genome sequence

The Fugu rubripes genome, available since 2002 [31], is the second vertebrate genome sequenced. It is 8 times smaller than the human genome and is proving to be an effective instrument in analyzing the human genome because of its compactness, low content of repetitive elements and the relatively large evolutionary distance between fish and mammals, which is estimated to be about 430 Myr [32].

Currently three versions of the Fugu rubripes genome assembly are publicly available. The second version of the assembly (v.2), constructed from 4.1 million sequencing reads (5.4 X sequence coverage), was reported in the original publication announcing the completion of the Fugu rubripes genome sequencing [31]. The third version (v.3) was released in August 2002, has slightly better coverage (5.7X) and improved scaffold contiguity. Sequence data for all three assembly versions can be downloaded from the Joint Genome Institute web site [33]. The JGI site and the EnsEMBL web site [34] are the two main portals to the Fugu rubripes genome annotation. Although EnsEMBL and JGI annotations and genome browsers are different, they share the same gene and transcript structure predictions created by the EnsEMBL pipeline.

Several analyses of the draft Fugu genome sequence targeting different protein families have been published recently [35–39], which showed its usefulness for evolutionary and functional studies as well as gene discovery. Here we present an analysis of the presence of the CTLD superfamily in the draft assembly of the Fugu rubripes genome.

Comparison of assembly versions 2 and 3

Protein database searches

Due to almost complete lack of cDNA or EST sequences for Fugu rubripes, most of the EnsEMBL gene structure predictions are based on homology with known protein sequences from other organisms, mostly mammals. We expected a significant fraction of CTLDcps to be conserved between fish and human, and, therefore, to be predicted correctly by EnsEMBL in the Fugu genome. So our first approach to detecting Fugu CTLDcps was to search a sequence database of predicted Fugu proteins with a hidden Markov model (HMM) for the CTLD. This search returned 69 significant matches. Some of the identified genes had a description assigned to them, apparently derived from the description of the sequence they were found to be homologous to. These descriptions, however, could not be used as a reliable basis for assigning orthology and paralogy relationships. For example, a sequence, which we later identified as an Endo180 orthologue (SINFRUG00000058766 in v.2 assembly annotation) is described as "80 KDA SECRETORY PHOSPHOLIPASE A2 RECEPTOR PRECURSOR PLA2", while another gene, which we designated as an aggrecan orthologue (SINFRUG00000069597 in v.2 annotation) was annotated as "ADRENOLEUKODYSTROPHY PROTEIN (ALDP)". Therefore, we reviewed domain architecture and sequence similarity matches for each of the sequences found to verify phylogenetic relationships.

Homology detection

The results of Inparanoid [40] comparison (see Methods) of all human to all Fugu CTLDcps were used to initially cluster the set of Fugu proteins and detect approximate orthology/paralogy links. Inparanoid has an important advantage over phylogenetic tree reconstruction software, as it does not require a multiple alignment of sequences but creates a distance matrix of the local pairwise alignments. This method assigned putative human orthologues to 25 Fugu proteins. Orthology relationships for the other 44 sequences from the set were established by individual analyses.

Revision of CTLDcp gene structure predictions

While analyzing phylogenetic relationships predicted by Inparanoid, we discovered several systematic and sporadic mistakes in the EnsEMBL gene predictions. The most widespread mistake was a failure to include exons encoding TM domains into gene structure prediction. Consequently, almost all EnsEMBL-predicted Fugu CTLDcps were soluble proteins, whereas very few human CTLDcps are. Simple comparison with the GenScan [41] features overlapping the CTLD-encoding genes showed that absence of TM domains is a result of coding sequence (CDS) mis-prediction rather than a fundamental difference in Fugu CTLDcps. GenScan predictions, in turn, could not be used as a basis for our analysis because they sometimes contain regions that are absent from human or mouse orthologues, and often merge neighboring genes. Another general problem was observed with proteins that had a previously unknown domain architecture (see below). In such cases individual domains were split into separate gene models.

In addition to these systematic problems, there were multiple sporadic ones. For example, our analysis of the Fugu genome shows that, similarly to the human and mouse genomes, the selectin cluster is well conserved and contains all three selectin genes in tandem (SELE, SELL, SELP), located on scaffolds 1045 (32046–41921) and 166 (83937–93826) in the v.2 and v.3 Fugu genome annotations, respectively. However, the EnsEMBL annotation contains a prediction of two overlapping genes (v.2: SINFRUG00000085188 and SINFRUG00000085187; v.3: SINFRUG00000123102 and SINFRUG00000123101), one of which is located in the intron of the other (Figure 1).

To solve these problems, we had to manually revise the predicted structure for all genes encoding proteins detected by the protein-level searches, and correct them using supporting evidence available in the EnsEMBL database, as well as additional evidence generated by us. The latter included similarity features produced by genome-wide GeneWise and BLAST searches with CTLD profiles and sequences, transmembrane domain predictions, and similarity matches to the complete sequence of supposed human or mouse orthologues.

As the final stage of the CTLDcp identification process, we performed a set of DNA-level comparisons to ensure that the CTLD-containing loci that are not covered by EnsEMBL-predicted genes, or for which transcript predictions are wrong and, thus, not detectable by protein database searches, were not omitted from the analysis. This "quality control" step led to identification of an additional set of 25 well conserved CTLDcps, which had both new and known domain architectures, as well as additional individual CTLDs, which were merged with neighboring CTLDcp loci if appropriate.

Groups of Fugu CTLDcps

After all these searches, we had identified a set of 94 Fugu rubripes loci encoding CTLDcps (Table 1), which in total contain 173 individual CTLDs, including PTR/Link-type CTLDs [42]. Fugu CTLDcps were named according to their human orthologues, established on the basis of domain composition and sequence similarities. Where more than one homologue was present in Fugu, a name was produced by adding a suffix of the form "-FXX", where XX is a sequential number of the paralogue, to the name of the closest human homologue. Predicted CTLDcps that do not have homologues among the known CTLDcps have identifiers of the form ANZ000. A few of these novel genes were orthologous to loci in other vertebrate genomes supported by expression data, but otherwise are un-characterized, and were assigned descriptive names (CBCP, Bimlec, SEEC, CETM, NLSLH).

We have clustered Fugu CTLDcps using the classification scheme for human CTLDcps based on domain composition; this comprises 14 groups [15]. Link/PTR-domain-containing CTLDcps, apart from hyalectans, were placed into a separate group. Among the Fugu CTLDcps that did not have mammalian homologues we detected a distinct group of soluble dual-CTLD sequences, which we have called F1 (Figure 2). The remainder of the Fugu-specific CTLDcps were assigned to the U (Unclassified) group. Gene structure prediction for members of the U group is the lowest in quality, due to lack of supporting evidence apart from similarity to CTLD sequence profiles and GenScan predictions.

All but two groups of human CTLDcps have detectable representatives in the Fugu genome (Figure 3, Table 1). We did not detect any orthologues for groups V (NK cell receptors) and VII (lithostathine/Reg family). The member repertoire for most of the other groups is very well conserved between Fugu and human. However, groups II and III, which include some of the best-studied mammalian CTLDcps, have a significantly different member composition in Fugu. In summary:

Novel CTLDcps conserved between Fugu and mammals

Discovering novel superfamily members in existing database sequences is one of the most important and exciting outcomes of a systematic computer-based study such as this. We predicted putative Fugu orthologues for several uncharacterized mammalian CTLDcps (Bimlec, MGC3279, KIAA0534, CETM, SEEC, CBCP, NLSLH) that are well conserved between Fugu and mammals. Most of the predictions were supported by mammalian cDNA sequences from public databases, but for two of them (NLSLH and CBCP) no full-length cDNA from any organism was found in DBs. The high level of genomic sequence conservation over evolutionary time from fish to human, as in the case of NLSLH, and the presence of partial cDNA and EST sequences from rodents and human, as in the case of CBCP, were strongly suggestive that the predictions are correct. The novel CTLDcps that could be attributed to one of the 14 known groups have been discussed in the preceding sections for the corresponding groups; those that do not fit into the existing classification are described below.

A large (~2100 aa) proteoglycan (CBCP), containing a set of chondroitin sulphate proteoglycan (CSPG) repeats [52], which are homologous to the NG2 ectodomain [53], a calcium-binding Calx-β domain [54] and a CTLD, is a novel member of a protein family which had not been reported previously to have members containing CTLDs; examples of this family also include the human MCSP/CSPG4 [55] and mouse FRAS1 [56] genes. The prediction was supported by three overlapping but incomplete cDNA sequences from human and mouse, high levels of conservation between human and Fugu (~50% identity), and the compact structure of the predicted Fugu gene. CBCP has been placed in a new CTLD group, XVII; its domain structure is shown in Figure 2. We have cloned a full-length cDNA of mouse CBCP confirming the domain structure predicted in this study (A.N. Zelensky, in preparation). The CTLD of CBCP lacks Ca-binding residues, and its long loop region is short, resembling that of the group V CTLDs.

Another protein with a novel domain organization, whose prediction is strongly supported by available cDNAs, is SEEC (S CP, E GF, E GF, C TLD-containing protein) (see Figure 2), which is well conserved between human and Fugu. Although not described in a publication, a full-length human SEEC cDNA (AK074773) was sequenced in the NEDO high-throughput sequencing project. The predicted Fugu SEEC is 63% identical to the human sequence. The sperm-coating glycoprotein (SCP) domain, which is present in a broad set of organisms from yeast and plants to mammals, but whose function is unknown [57], is rarely observed in combination with other domains in proteins; in only one other known protein (from sea urchin) is it found together with an EGF domain [58], and SEEC is the first example of a CTLD-SCP combination. The potential Ca/carbohydrate-binding motif (QPD) characteristic of galactose specificity is present in the CTLD. SEEC has been placed in a new CTLD group XVI.

A predicted protein named "novel L-selectin homologue" (NLSLH) because its CTLD is most similar to selectin CTLDs is duplicated in Fugu (NLSLH and NLSLH-F1) but only moderately conserved (32% identity) between Fugu and human. The putative human orthologue is located on Ch1q25.1 about 18 Mb further from the centromere than the selectin cluster and is supported only by EST sequences (AA912157, AA889574), but not cDNAs. No conserved domains except for the CTLD could be detected in the human and Fugu NLSLH loci so, if the predictions are correct, NLSLH is a soluble single-CTLD-containing protein. Carbohydrate-binding motifs are not present in the NLSLH and NLSLH-F1 CTLDs.

Finally, a type I transmembrane protein Bimlec, whose prediction is supported by a full-length human cDNA, was placed in a new group XV.

Dating the CTLDcp duplications

We found 12 groups of unlinked Fugu-specific CTLDcp paralogues (Table 1), and attempted to determine the duplication dates using two approaches: (1) based on the estimation of the number of synonymous nucleotide substitutions (Ks) in the coding sequences and (2) based on the molecular clock hypothesis.

For all but two pairs of duplicated genes, Ks values estimated with four different methods (see Methods) were between 1.5–2.5, which indicates a complete saturation of the synonymous sites (Figure 4(A)). Ks values so high cannot provide an accurate estimation of the duplication age, but we can conclude with confidence that the CTLDcp gene duplications are at least 150 Myr old, which is the time required for complete saturation of silent sites assuming a mutation rate of 2.5 substitutions/silent site/billion years in fish [59]. If, however, Ks values presented in Figure 4(A) and the silent mutation rate are close to correct, the corresponding duplication timeframe is predicted to be 300–500 Myr.

In order to date the duplications based on molecular clock measurements, we aligned duplicated Fugu CTLD sequences with their vertebrate orthologues present in GenBank, and built linearized phylogenetic trees based on the alignments. As human and mouse sequences were invariably available, the divergence time between these two species [96 Myr; 60] was used to calibrate the clock, together with the divergence time between Actinopterygii and Sarcopterygii [430 Myr; 32]. Symmetrical tree topology ((H, M) (F, F1)), expected for a Actinopterygian-specific duplication, was revealed by at least one phylogeny reconstruction method we used for the following six homologue groups (data not shown): brevican, neurocan, MManR, SRCL, tetranectin and HAPLN3, with duplications dated 369, 284, 397, 377, 360 and 312 Myr, respectively. A typical tree with symmetrical topology is shown for MManR in Figure 4(B). The other six alignments (aggrecan, versican, layilin, attractin, C1qRP, CRTL1) produced trees with topologies suggesting a duplication predating the split between Actinopterygian and Sarcopterygian. The portion of symmetrical topologies (50%) in the CTLD set is similar to the ratio reported by Taylor and coworkers in fish: 15 of 27 (55 %) [61], and 25 of 53 (45%) [62] for bigger and more heterogeneous gene collections.

Draft assembly limitations

A systematic study based on draft-quality whole-genome data for an organism like Fugu rubripes has some limitations, as the genomic sequence is incomplete, fragmented and sometimes misassembled, and the expressed sequence information is scarce. On the other hand, many of the genomes that are currently being sequenced will be released and remain for sometime in the same state as the Fugu genome data are now. Indeed, more than a year after the initial release [31] very few improvements to the Fugu genomic data [v.3 assembly and EST sequencing project; [63]] have been published. Therefore, it is essential to extract useful biological information from draft-quality whole-genome sequences. Our study is such an attempt.

We have mentioned four limitations of the draft-state assembly – incompleteness, fragmentation, misassembly and lack of expression information. While the last might appear the biggest problem, we found that ab initio predictions combined with manual curation and interspecies comparison have proven to be very accurate (e.g. see PKD1L example), thanks to the compactness of the Fugu genome, smaller ratio between intron and intergenic region sizes compared with mammalian genes, wealth of data for comparative analysis etc. We do not expect that sequencing the remaining 5% of the Fugu genome, which is mostly heterochromatic regions, will lead to discovery of many new CTLDcps. From the comparison of the Fugu CTLDcp repertoire discovered by us and found in other fish species independently, the only surprising omission in our results is a MBP orthologue. MBP sequences have been found in several other fish species. Their absence in Fugu may represent a bona fide gene loss. As to the fragmentation, only a few of the CTLDcps are split between scaffolds, namely versican and some MManR paralogues (Figure 1). All of the fragmented genes are big, and in most cases the fragments can be combined easily to reveal the full sequence. Finally, misassembly signs were observed in several CTLDcp loci while comparing two versions of the assembly. These showed as presence of repeated regions in the v.2 assembly, which disappeared in the v.3 assembly.

Two groups identified in higher vertebrates are not detectable in Fugu

We could not detect CTLDcp representatives for groups V (NK cell receptors) and VII (lithostathine) in the Fugu genome. CTLDs in the members of these groups have lost their carbohydrate-binding activities, and perform functions that have, apparently, evolved after evolutionary separation of tetrapods from fish, or which are mediated by other proteins in fish. For example, group VII members are secreted into the digestive tract – a system that is very flexible evolutionally. Group V is probably one the youngest and most rapidly evolving sets of CTLDcps; its component members vary significantly even between rodents and human, a phenomenon connected to the co-evolution with the acquired immune system proteins that group V CTLDcps interact with.

Our conclusion on the absence of group V CTLDcps in the Fugu genome is at odds with the conclusions of two studies describing group V CTLDcp evolution in chordates. A recent paper describes possible CD94 homologues (cichlid killer cell lectin receptor, cKLR) in bony fishes Paralabidochromis chilotes and Oreochromis niloticus, which are encoded by a large multi-gene family with at least 10 members [28]. Another recent work described sequencing of a CD94 homologue in a tunicate [64].

The decision by Sato et al. [28] to assign putative fish killer cell receptors to group V rather than to group II was based on several considerations, including gene structure, absence of canonical Ca²⁺/carbohydrate-binding residues, and phylogenetic analysis based on the CTLD alignment. The latter consideration is mentioned as the most important one. However, as the authors themselves note, bootstrap values for placing cKLR on the group V branch, are "low to moderate". Indeed, we found that in phylogenetic trees built using different methods (maximum parsimony, distance estimation method with PAM matrix followed by neighbor-joining tree reconstruction, maximum likelihood) from the ClustalW alignments of cKLR sequences with group V and group II CTLD sequences from Fugu, mouse and human, cKLR placement is unstable. As shown in Figure 5, on a tree built by the neighbor-joining method we found cKLR on the branch containing the Fugu-specific subset of group II CTLDcps (DC-SIGN-F1 – DC-SIGN-F8), most of which do contain residues required for Ca²⁺/carbohydrate binding. On a tree built by the maximum parsimony method, we found cKLR on a separate branch equally related to group II and group V sequences (not shown). Also, a BLAST search with the complete cKLR sequence (GI 31789959) in the non-redundant NCBI protein database returns members of the ASGR subgroup of group II as top matches. Therefore, we judge that sufficient support for assignment of cKLR to group V is lacking and the question of the presence of the NK-cell receptor family in fishes is still open.

As to the putative CD94 homologue from tunicates, it is indeed more similar to CD94 than to any other CTLDcp. However, the low level of sequence homology and the lack of evidence for existence of group V CTLDcps in more advanced taxa does not allow a confident statement that the sequence from tunicates is a CD94 orthologue, rather than a result of convergent evolution.

Expansion of the innate immunity CTLDcp groups in Fugu

Unlike pairwise unlinked duplications (see below), tandem duplications and other gene family expansions are limited to two groups, namely the DC-SIGN subgroup of group II and MManR. In mammals, members of these subgroups play an important role in innate immune responses. In particular, DC-SIGN is actively studied due to its ability to bind and internalize a broad range of bacterial and viral pathogens, including HIV-1 and Mycobacterium tuberculosis (reviewed in [65]), while MManR is also implicated in binding and phagocytosis of a wide range of microorganisms [66]. Expansion of these groups, most notably the DC-SIGN subgroup, in Fugu may reflect a larger role for innate immunity in host defense in lower vertebrates. Interestingly, multi-copy clusters comprising at least 10 genes encoding close cKLR homologues were identified in another cichlid fish species Oreochromis niloticus [28], which suggests another parallel between the expanded DC-SIGN subgroup in puffer fish and cKLRs of cichlids.

There are no extra members, however, in the Fugu collectin group – another CTLD group directly involved in innate immunity in mammals. Moreover, the mannose binding protein (MBP), which is the best-studied mammalian collectin involved in lectin complement activation pathway, was not detected by us. The absence of MBP orthologues in Fugu is rather puzzling, as MBP sequences have been found in several other fish species (Danio rerio, Cyprinus carpio and Carassus auratus; [20]), and are well conserved within the Cyprinidae carp family. The collectin family is also present and expanded in the Urochordate Ciona intestinalis with nine collectin genes identified in the draft genome sequence [67], although it is not clear whether one of these nine genes is an MBP orthologue. Given the role of MBPs in complement activation in mammals, and their presence and level of conservation in the carp family, it is possible that the Fugu MBP orthologue does exist but is not covered by the draft genome sequence. Complement-activating C-type lectins from lower organisms have been identified but not completely sequenced [68]; they have multiple CTLDs as in CPL-III from the protochordate Clavelina picta [69] or lack the collagen domain and show more similarity to other CTLDcps such as the glucose-binding lectin (GBL) from another tunicate, Halocynthia roretzi [70].

Fugu dual CTLD molecules – a missing link between vertebrate and invertebrate CTLDs?

Previous whole-genome studies of the CTLD superfamily in two invertebrates [13, 14] failed to identify any groups of CTLDcps common to both invertebrates and vertebrates. A group of predicted dual CTLD-containing proteins in Fugu (F1) may be the first vertebrate group that has detectable homologues in invertebrates. Alternatively, it is possible that none of the Fugu F1 group members are in fact orthologous to the invertebrate sequences, as sequence similarities are only moderate (~30% ID) and the domain architecture is simple and could have evolved independently in different lineages. However, several observations suggest that at least F1 members from Fugu and zebrafish and SCARF proteins from Girardia tigrina evolved from the same predecessor. First, similarity levels between fish sequences and between fish and planarian sequences are about the same. It is unlikely that the fish sequences are unrelated, which implies that F1 members are evolving quickly, and only major structural features of these molecules are under selective pressure [49]. Second, the CTLDs of planarian and, in all cases at least one CTLD of the fish dual-CTLDcps, contain residues characteristic of Ca/carbohydrate binding. In vertebrates, ability to bind carbohydrates is associated with the oldest CTLDcps groups, and is considered to be an ancestral feature of the CTLD. Indeed, both vertebrate CTLDcp groups that we failed to find in Fugu (V and VII) have lost sugar-binding properties. This is also the case for the antifreeze proteins from fish and snake venom CTLDcps, which have only been found in the corresponding clades. Third, similar domain organization (two CTLDs, no transmembrane domain) is also observed in two other known groups of invertebrate CTLDcps: immulectins from various insect species [71, 72] and nine proteins from C. elegans, classified as group D1 by Drickamer and Dodd [13]. Despite identical domain structure, none of these proteins shows statistically significant homology to the fish F1 group members or their putative homologues from planarian or Drosophila. Altogether, this indicates that domain structure alone cannot establish an evolutionary link between the fish and invertebrate sequences. Hence, the suggestive link between the F1 group fish members and the planarian and Drosophila proteins is even more interesting.

CTLDcp classification update

The existing classification of CTLDcps is generally accepted and popularly used in studies of the superfamily and recently has been updated [15]. The classification divides CTLDcps into monophyletic groups of proteins with identical overall domain architecture based on a combination of structural and phylogenetic information. Although two previous large-scale studies [13, 14] showed it to be inapplicable for description of invertebrate CTLDcps, our analysis of the puffer fish genome indicates that it is sufficient to describe the superfamily in all vertebrates, with only minor modifications and some extensions.

Our newly discovered CTLDcps, with a few exceptions, do not fit into the existing classification because of their unique domain architecture. We propose several new groups to accommodate the novel CTLDcps which have been found in both higher and lower vertebrates and are supported by cDNA sequences:

• XV – Bimlec (type I transmembrane protein), which in phylogenetic trees is not placed on the same branch as group VIII sequences, has a distinct exon-intron structure of the CTLD region and a neck not similar to the neck region of the group VIII sequences;

• XVI – SEEC, based on unique domain architecture;

• XVII – CBCP, based on unique domain architecture;

Additional groups may be required for the sequences not supported by sufficient expression data (NLSLH) and other sequences from the "unclassified" group whose presence in higher vertebrates is not clear. Also, clade-specific groups, such as fish antifreeze proteins (AFP), dual-CTLD sequences (group F1) predicted by us and so far identified only in fish, or snake venom CTLDcps which lack orthologues in other vertebrates, are required.

It has been suggested previously [19, 48] that AFPs belong to group VII based on their domain architecture and exon-intron structure. However, our phylogenetic analysis of an alignment of CTLD sequences from all known human and mouse CTLDcps and 26 different fish CTLD-containing protein sequences identified by searching the NCBI protein database with BLAST, indicates that they constitute a phylogenetically distinct group including all known soluble fish CTLD-containing proteins, except Cyprinidae collectins. As to the exon-intron structure, introns in the group XII (EMBP) CTLDs are at exactly the same positions as in group VII and AFP-like CTLDs, which suggests that all three groups are closely related but does not allow classification of the fish AFP-like sequences to either of the mammalian groups. Interestingly, just like most of the AFPs, mammalian EMBPs contain an atypical WIGL motif with a glycine in the fourth position, a substitution not observed in any other mammalian CTLD we analyzed. Taken together, these observations indicate that in a broader evolutionary perspective the differences between some of the groups including CTLDcps with a very similar domain architecture (VII, XII and AFP; II and V) become less distinct, which makes classification of the "intermediate" or "ancestral" sequences, equally related to more than one group, problematic.

Selective duplication of the Fugu CTLDcp-encoding genes and the whole-genome duplication hypothesis

The hypothesis that whole-genome duplications were one of the main driving forces in vertebrate evolution, providing genetic material for increased diversity and progressive development [73], and that there were two rounds of whole-genome duplication in vertebrate phylogeny (the 2R hypothesis) [73, 74], is actively debated [75, 76]. A more recent whole-genome duplication is suggested for the Actinopterygian branch [61]. Ray-finned fish are the most diverse group of vertebrates, and based on the initial observation that each of the four human HOX gene clusters has two homologues in zebrafish [77] it was suggested that they have undergone an additional round of a whole-genome duplication after the divergence from Sarcopterygian about 430 Myr ago [61]. Analysis of the genome duplication in fish can give a picture of a duplicated genome after 300–400 Myr of evolution and fill the gap between the now generally accepted recent tetraploidizations in plants [78] and yeast [79] and the alleged more ancient duplication(s) of the ancestral vertebrate genome.

Although many fish genes are indeed duplicated [61, 77, 80–82], it is not clear whether the copies were created by a complete genome duplication (autopolyploidy), merge of different genomes (allopolyploidy), regional duplication, or simply a series of tandem duplications. Attempts to show that ancient tetraploidization (has not) occurred usually involve: (i) searching for an excess of paralogue groups where the number of members is double the number of alleged duplications (i.e. 2 in case of Actinopterygian duplication, and 4 in case of vertebrate duplication, the "one to four rule") [74, 76]; (ii) showing that a statistically significant number of duplications took place at approximately the same time by molecular clock estimation or synonymous substitution counting [83, 84]; (iii) using phylogenetic methods to assess the relation between duplication and speciation events [61]; and (iv) showing that duplicated genes are arranged in paralogous blocks on chromosomes (paralogons) [62, 85, 86]. We used these approaches to analyze the nature of the observed CTLDcp duplications in Fugu.

Our results clearly show that tandem gene copying is a mechanism of CTLD family evolution and led to generation of three gene clusters: DC-SIGN-F2 – DC-SIGN-F5 (4 genes), CRTL1-F1 and CRTL1-F2, and AFPL-F1 and AFPL-F2. Members of the two latter clusters are nearly identical and may be an assembly artifact. Twelve other duplicated genes are not linked in the current assembly and have sequences much more diverged than tandem duplicates. Of the 12 genes only MManR, which has 3 paralogues, is present in more than two copies. We consider this is important evidence in favor of a whole-genome duplication, as sporadic duplications cannot explain such a strong bias towards two-member paralogue groups. Unfortunately, the results of duplication time estimations are less conclusive as they give only a broad timeframe for the possible duplication events of about 300–400 Myr. As in the case of some other fish gene families reported previously [61, 62, 87, 88], molecular phylogeny reconstruction performed by us often indicates that duplications occurred before the divergence between fish and tetrapods. However, this could be an artifact of the method caused by different selection pressures on duplicates. Unfortunately, there is practically no overlap between vertebrate and invertebrate CTLD families, so we could not use invertebrate sequences to refine phylogenetic analysis. To conclude: phylogenetic relationships between CTLD paralogues and estimated duplication time distribution indicate that there was a burst in duplication activity in the Fugu genome 300–400 Myr ago. While we cannot determine definitively the nature of the duplications (tandem, regional or whole-genome), a pronounced bias in the number of two-member paralogue groups strongly suggests that there was a single large-scale or whole-genome duplication event in fish.

Another interesting observation is that CTLDcp genes were either duplicated, or retained after a large-scale duplication, in a pronounced selective manner. One group (I) is duplicated completely, while in other groups only partial duplications are found. Interestingly, group I (lecticans), which in tetrapods contain four large (>2000 amino acids) proteins, very similar to each other in sequence and domain structure, is a good candidate for demonstrating the 2 R hypothesis. If the four lecticans arose as a result of the alleged two rounds of the whole-genome duplication early in vertebrate history, the fact that the family was also completely duplicated in fish and retained after the duplication appears very non-random and implies some functional explanation. In the human genome, all four genes encoding lecticans are located on different chromosomes (1, 5, 15 and 19), but it is not clear whether they are linked in Fugu.

Another group that conforms to the 2 R hypothesis is group VI, which in tetrapods has four members with almost identical domain structure in mammals (Pla2R, MManR, DEC-205 and Endo180). Though in the Fugu genome we identified 7 group VI sequences, some of which are fragmented (Figure 1), phylogenetic analysis shows that only one member of the family (MManR) was quadruplicated, while others are present in a single copy. Both molecular clock and Ks-based methods date the MManR duplications at approximately the same time as other CTLDcp gene duplications. Phylogenetic trees, built on alignment of the overlapping portions (Figure 1) of the complete sequences and three largest fragments (fMManR-F1, fMManR-F2, fMManR-F3) have symmetrical structure, with fMManR-F1, fMManR-F2 and fMManR-F3 forming a separate branch (Figure 4(B)). A whole-genome duplication, generating fMManR and fMManR-F1, followed by tandem duplications of fMManR-F1, producing fMManR-F2 and fMManR-F3, can explain this topology.