The immunoglobulin superfamily in Drosophila melanogaster and Caenorhabditis elegans and the evolution of complexity

The anatomy and physiology of an organism is determined primarily by the protein repertoire encoded in its genes and the expression patterns of these genes. This means that determining the protein repertoires of organisms makes a significant contribution to an understanding of the molecular basis of their anatomy and physiology and of why they differ between organisms.

In this paper, we describe the determination of the immunoglobulin superfamily (IgSF) repertoire in the fly Drosophila melanogaster and compare it with that found in the nematode Caenorhabditis elegans. IgSF proteins are well known for their roles in cell-cell recognition and communication–both crucial processes during embryonal development. A comparison of the functions and the size of this superfamily in the two organisms should give some idea of the nature of the changes in protein repertoires that underlie the increases in physiological complexity in the fly, for example, a more elaborate nervous system.

The IgSF repertoire in C. elegans was initially investigated by Hutter et al. (Hutter et al.,2000) and by Teichmann and Chothia(Teichmann and Chothia, 2000). As we show below, refinements of the genome sequence and protein predictions carried out since then have revealed additional members of the IgSF. Another smaller superfamily whose members are involved in cell adhesion processes, the cadherins, has been described previously for both the worm and fly(Hill et al., 2001).

We first describe the determination of the IgSF repertoire in Drosophila and of the new IgSF sequences in C. elegans. We then analyse the IgSF proteins common to both organisms and specific to each,in terms of their homologies and functions. In the conclusion, we discuss the implications of our results for an understanding of the role of this superfamily during the metazoan evolution and as a framework for further experimental investigation.

The complete set of predicted protein sequences of D. melanogasterwas obtained from The Berkeley-Drosophila-Genome Project (The Berkeley Drosophila Genome Project, Sequencing Consortium, 2000). They were copied from the website at http://www.fruitfly.org/sequence/release3download.shtml. The predicted worm proteins were obtained from WormBase(Stein et al., 2001; C. elegans Sequencing Consortium, 1998) and from the website at http://www.wormbase.org/downloads.html. We also made some use of the predicted protein sequences of the genomes of Anopheles gambiae(http://www.ensembl.org/Anopheles_gambiae/)and Caenorhabditisbriggsae(http://www.ensembl.org/Caenorhabditis_briggsae/).

The names used here for the predicted proteins are the identifiers given in FlyBase and WormBase except for those proteins with names given by experimentalists who previously determined their sequences and, in most cases,their function. These specific names start with a capital letter to denote that they refer to proteins; small letters refer to genes.

A schematic overview of the procedures used to analyse these sequences is shown in Fig. 1 and described in detail below.

Domains in the sequences from fly and worm resources described above were identified using hidden Markov models (HMMs)(Krogh et al., 1994; Eddy, 1998; Karplus et al., 1998), which are probably the most sensitive automatic sequence comparison method currently available (Park et al., 1998; Madera and Gough, 2002). They are sequence profiles that, built from multiple sequence alignments, represent a family of sequences. The database SUPERFAMILY contains a library of HMMs that represent the sequences of domains in proteins of known structure(Gough et al., 2001; Gough and Chothia, 2002). These domains are whole small proteins or the regions of large proteins that are known to be involved in recombination. They are described on the Structural Classification of Proteins (SCOP) Database(Murzin et al., 1995; Lo Conte et al., 2002) where they are classified in terms of their evolutionary and structural relationships. The sequences of SCOP domains are made available through the ASTRAL database (Brenner et al.,2000; Chandonia et al.,2002) and these are used to seed the HMMs in SUPERFAMILY.

Previous to the work described here, the SUPERFAMILY HMMs were matched to the protein sequences predicted from the available genome sequences including those of Drosophila and C. elegans. The results of these matches are available from the public SUPERFAMILY database(Gough et al., 2001; Gough and Chothia, 2002). We extracted from SUPERFAMILY all Drosophila and C. eleganssequences that are matched by HMMs for IgSF domains with an expectation value score (E-value) of less than 0.01. The E-value is a theoretical value for the expected error rate. Large-scale tests show that these theoretical expectations are very close to the observed error rates. In our case, an E-value threshold of 0.01 corresponds to 1% error in the structural assignment(Gough et al., 2001).

HMM matches close to the E-value threshold were inspected by eye and judged for their correctness. In some cases they were also checked by using SMART(Schultz et al., 2000) to make domain assignments. As a result, three sequences matched with only marginally significant scores by SUPERFAMILY were rejected.

Unassigned regions of roughly 100 residues length with IgSF domains on both sides were inspected for the pattern of key residues that is a characteristic of the immunoglobulin superfamily (Chothia et al., 1988; Harpaz and Chothia, 1994). Several additional IgSF domains were detected by this procedure.

The proteins identified as containing one or more IgSF domains were examined for other features and domains, using six servers.

1. The SUPERFAMILY database: the sequences matched by IgSF HMMs were examined further to see if they are also matched by HMMs for other types of domains.

2. The Pfam database (Bateman et al.,2002): Pfam includes HMMs for protein domains of unknown structure. The IgSF proteins were submitted to this server to see if there were any additional matches.

3. The SMART (Schultz et al.,2000) server was used to check and extend the results of the SUPERFAMILY and Pfam HMM matches.

4. The SignalP server (Nielsen et al.,1999) was used, with the default options for eukaryotes, to identify signal sequences.

5. The TMHMM server (Krogh et al.,2001) was used, with default options, to identify transmembrane helices.

6. The Predictor programme (Eisenhaber et al., 1999) was used to identify GPI anchors.

These predictions were edited manually and compared with information from the literature (see below).

The IgSF proteins are either soluble or they are attached to the membrane by a transmembrane helix or a GPI anchor. For ten proteins, the GPI Predictor(Eisenhaber et al., 1999)found sites for attachment of GPI anchors. For proteins with a transmembrane helix, the IgSF domains are always in the extracellular region. After the immunoglobulin superfamily itself, the next most abundant superfamily in IgSF proteins are fibronectin type III domains, followed by the ligand-binding domain of the LDL receptor, BPTI-like domains and protein-kinase like domains. Domains from 21 superfamilies are found in both organisms, six and 10 domain superfamilies are specific to the fly and the worm, respectively.

In the analyses of metazoan genome sequences, a significant fraction of the predictions made for large proteins are incomplete, particularly at their N and/or C termini (Teichmann and Chothia,2000; Hill et al.,2001). Some of these errors can be detected if there are already experimental determinations of the predicted sequences, or of close homologues, and corrected by matching the experimental sequences to the genome using the GENEWISE procedure (see below).

To detect whether predicted protein sequences are incomplete they were matched against three sets of experimental sequences

1. Experimentally determined IgSF proteins in the public databases. The IgSF proteins were matched to sequences in the NRDB90 sequence database(Holm and Sander, 1998) using FASTA (Pearson and Lipman,1988) with an E-value threshold of 0.001 and a sequence identity higher than 50%. For 36 IgSF proteins, we found matches in NRDB90 that were identical in sequence but at least 30 amino acids longer than the predicted sequence.

2. A library of some 9000 full-length Drosophila cDNAs(http://www.fruitfly.org/sequence/dlcDNA.shtml). For 28 IgSF proteins we found cDNAs hits that were identical in sequence but at least 30 amino acids longer than the original predicted sequence (see Tables 1, 2, 3). In these cases, it is very likely that the cDNAs represent the complete version of the gene or a longer splice variant.

3. The Drosophila IgSF sequences were matched against those predicted for the Anopheles gambiae genome(http://www.fruitfly.org/sequence/dlcDNA.shtml)using Smith-Waterman alignments (Smith and Waterman, 1981).

Predicted IgSF proteins that had matched experimental versions of their sequences in NRDB, or close sequence homologues in Anopheles that are greater in length by at least 30 amino acids were checked using the GENEWISE program (Birney and Durbin,2000). GENEWISE, using an HMM algorithm, tries to identify the exons in DNA that are homologous to the query protein. Because this method relies on the similarity of the two sequences, homologues with a sequence identity of more than 50% are usually required for a significant match. The homologous protein was compared with the chromosomal region containing the Drosophila gene and with up to 30 kb of surrounding DNA at either end of the gene. In eight cases (see Tables 1 and 3), the sequence found by GENEWISE was longer than both the original sequence and any matching cDNAs. Some C. elegans gene predictions were revised in a similar manner using homologues from Caenorhabditis briggsae. Details are described below.

In addition to these improvements in the sequences of the current FlyBase release number 3(http://www.fruitfly.org/sequence/dlMfasta.shtml),there are 13 cases of genes predicted by the previous release, number 2, that are shorter or absent in the current release. These sequences are indicated in Tables 1 to 3.

IgSF proteins in C. elegans were described previously(Hutter et al., 2000; Teichmann and Chothia, 2000). In Teichmann and Chothia (Teichmann and Chothia, 2000), 64 proteins were identified. Since then, new predictions based on revised genome sequences have been released(http://www.wormbase.org/downloads.html). These were analysed using procedures similar to those described above for Drosophila proteins. This resulted in a new total of 80 IgSF proteins in C. elegans. Of these 80, 53 are identical or nearly identical to those found in the previous work, eight are revised versions of old predictions and 19 are new (Tables 2 and 3). For the revised versions,the respective homologue in C. briggsae was examined and taken in one case (SSSD1.1) to improve the gene prediction using GENEWISE(Birney and Durbin, 2000).

In discussing the IgSF proteins we find that it is useful to divide them into six classes. These classes are based on broad functional similarities,although within each class the proteins also have common features in terms of domain architecture. Proteins that share a particular domain architecture belong largely, but not always, to the same cluster of closely related IgSF proteins. Details of these relationships are described in Tables 1 to 3 and the text below.

These are proteins that span the cell membrane via a transmembrane helix or are attached to the cell surface by a GPI anchor. They have an extracellular region that is exclusively, or almost exclusively, composed of IgSF and fibronectin type III (FnIII) domains, and cytoplasmic domains that are not kinases or phosphatases. Experimentally characterised proteins in this class are mainly cell-adhesion molecules that play important roles in development.

These are proteins that span the cell membrane via a transmembrane helix. They have an extracellular region that is exclusively, or almost exclusively,composed of IgSF and FnIII domains, and cytoplasmic domains that are kinases or phosphatases. All experimentally characterised proteins in this class are cell-surface receptors that bind various factors.

These are proteins that span the cell membrane via a transmembrane helix or are attached to the cell surface by a GPI anchor. They have an extracellular region that is composed of IgSF domains and a variety of different domains. Experimentally characterised proteins in this class act as signalling molecules during neural development.

These proteins have a variety of different domain architectures that can consist of just IgSF domains but can also include other domains, some of which are unusual. They act as intercellular messengers: secreted by one cell and interacting with cell surface receptors on other cells. Three different groups of proteins fall into this class: (1) proteins for which it has been shown experimentally that they are secreted; (2) proteins that have a signal sequence but no transmembrane helix or GPI anchor predicted; and (3) proteins that do not have a signal sequence, transmembrane helix or GPI anchor predicted but show sequence similarity to a proteins from (1) or (2) according to the E-value threshold described below.

These proteins are usually rather long with more than ten IgSF domains in a row and sometimes other domains. They act in the extracellular space in cell-adhesion and cell-cell recognition processes, and thus do not have transmembrane domains or GPI anchors.

These proteins are usually rather long with more than ten IgSF domains in a row, sometimes in combination with FnIII domains in a characteristic pattern. Some muscle proteins also have kinase domains. Experimentally characterised proteins in this class are all involved in muscle function.

All proteins were grouped into these six classes if (1) experimental work demonstrated functions characteristic to one class, (2) features in domain architecture clearly pointed towards affiliation to one class, and/or (3) the protein showed sequence similarity to a protein member of a specific class according to the E-value threshold described below. The few proteins for which none of the criteria (1), (2) or (3) apply were grouped into a `bin' class called `proteins of unknown cellular localisation'.

The final set of IgSF protein sequences in the two organisms have a variety of domain architectures. Figures 2 and 3 illustrate the variety of these domain architectures we found in the IgSF repertoire of fly and worm in terms of the number and kind of different domains observed in the proteins. The number of domains per protein varies from one in small signalling proteins to 68 in fly Titin. There are a few very long proteins that are in the muscle and extracellular matrix proteins classes.

In the following sections we describe and compare the IgSF proteins. To discover the relationships described below for IgSF proteins in C. elegans and Drosophila, we considered a combination of E-values for the matching sequence pairs or, for closely related proteins, sequence identities, match lengths and domain architectures. For proteins that are closely related to known structures or are very short, we also used key residue analysis (Chothia et al.,1988; Harpaz and Chothia,1994). But before presenting this it is useful to discuss the different levels of sequence similarities that exist in these proteins and their relation to function.

By definition, all the proteins considered here contain at least one IgSF domain and are therefore homologous in at least that region. However,relationships at this basic level are not very informative. What is of more use are relationships that imply some functional annotation. We tried,therefore, to identify by sequence comparisons clusters of closely related IgSF proteins whose members are likely to have been produced by relatively recent gene duplication events and to have similar functions. To do this we first determined the extent to which indications of affiliation to one of the six functional classes can be detected from comparison of sequences. We took the 58 Drosophila IgSF proteins whose function has been experimentally characterised and allocated them to one of the six functional classes described in the last section. The 58 proteins were then matched to each other using the Smith-Waterman algorithm(Smith and Waterman, 1981). The scores in terms of E-value and sequence identity made by each of the matched pairs were examined.

For protein pairs whose sequence identities are greater that 40%, their close relationship is obvious. But for those where it is smaller than 40%, a statistical measure such as the E-value is much more reliable for inference of homology than sequence identity (Brenner et al., 1998). For those pairs that have E-values lower than 10^–20 we plot the results shown in Fig. 4. Matches that occur between proteins in the same functional class and those that occur between proteins in different classes are distinguished. It clearly shows that most of matches with an E-value lower than 10^–35 are between proteins within the same functional classes. The exceptions, where proteins of different functional classes match with E-values lower than 10^–35, arise from two clusters. The Beat proteins cluster has 14 members of which four are cell-surface class I proteins and ten are secreted proteins. Lachesin and Amalgam are two closely related proteins the first of which is a cell surface class I protein and the second is in the secreted proteins class.

We then examined protein pairs whose match scores have E-values larger than 10^–35 and sequence identities of less that 40%. When the cut-off parameters were slightly loosened (E-value cut-off of 10^–30 or sequence identity cut-off of 30%), only very few more matches between proteins of the same functional classes appeared. When the cut-off parameters were further loosened, we only found matches between proteins of different functional classes.

Thus, the matches made between the 58 Drosophila proteins suggest that sequences with identities of 40% or greater or E-values below 10^–35 belong to the same functional class. Note that the match region covered more than 50% of the length of both proteins. (It should be noted that not all proteins within a functional class match each other with a score less that 10^–35. This means that only positive results are significant; a negative one just means a function cannot be implied by sequence comparisons.)

All the IgSF proteins meeting these conditions were then grouped into clusters of closely related, homologous proteins using a single linkage algorithm: a protein qualifies as a member of a cluster if it matches at least one of the other cluster members within the above mentioned thresholds. All clusters were inspected by eye to ensure accuracy, and a few clusters were split into separate clusters based on domain architectures and inter-domain connections of subgroups of proteins within the cluster, as described below. We used these clusters to assign uncharacterised proteins that were homologous to characterised proteins to the six functional classes.

The calculations described above identified 142 IgSF proteins in Drosophila and 80 proteins in C. elegans. We have ignored different splice variants. Those proteins known to have splice variants are represented by the longest sequence known to us. The two sets of proteins were compared in terms of their domain architectures, sequence similarities(percent identity and E-value), key residues and inter-domain connecting regions. Similarities between Drosophila and C. elegansproteins detected by these criteria would imply their presence in their common ancestor. Lack of evidence would suggest either the evolution of the protein beyond the criteria described above subsequent to their divergence or,possibly, its loss in one of the two organisms since their divergence. In Table 1, we list the 106 proteins in Drosophila that appear to be not closely related to those in C. elegans (see below). In Table 2, we list the 45 proteins in C. elegans that appear to be not closely related to those in Drosophila. In Table 3, we list the 36 Drosophila proteins and the 35 from C. elegans that are closely related to each other according to the criteria described above.

Drosophila and Anopheles gambiae (mosquito) diverged from their common ancestor some 250 million years ago. Of the 142 Drosophila proteins, 128 have a clear orthologue in Anopheles: i.e. the Drosophila and Anopheleshomologues match each other with scores better than those they made to any other protein. A similar situation applies to C. elegans: C. elegans and C. briggsae diverged some 40 million years ago. Here, eight IgSF proteins in C. elegans lack an orthologue in C. briggsae. The existence of clear orthologues is good evidence that the matching proteins are not pseudo-genes. The absence of a match, however, does not necessarily mean that the sequence is a pseudo-gene. This may arise from incomplete predictions, the loss of the protein in Anopheles or C. briggsae, or its recent formation in Drosophila or C. elegans.

Prior to this work, 58 Drosophila and 22 C. elegansproteins had been identified by experimental work and assigned a function. All but 25 of the other 84 Drosophila and the 58 C. elegans IgSF proteins have been assigned to one of the six functional classes defined above. Those not classified, 12 in Drosophila and 13 in C. elegans, are placed in a class termed `proteins of unknown cellular localisation' (see Tables 1 and 2).

The assignments to these functional classes have been made on the basis of sequence homology and/or the presence or absence of signal sequences and transmembrane helices. The problem with using the latter features is that the prediction of long protein sequences often misses out N-terminal and C-terminal regions (Teichmann and Chothia,2000; Hill et al.,2001). Thus, we might expect that, in some cases, proteins currently placed in the secreted proteins class, because they have a signal sequence but no transmembrane helix or GPI anchor site, will be transferred to a cell surface class by subsequent discovery of a C-terminal region with one of these features. Similar revisions could well transfer proteins currently in the unknown class to the secreted or cell surface classes.

Table 4 summarises the distribution of the proteins, and clusters of closely related proteins,between the different functional classes. In both organisms, the two largest functional classes are the cell surface class I proteins (82 and 30 in fly and worm, respectively) and the secreted proteins class (22 and 12 proteins) many of whose members have important roles during development. These proteins form three-quarters of the Drosophila IgSF repertoire and half of that in C. elegans. The average size of the two clusters in Drosophila is larger than in C. elegans. The other four functional classes have similar numbers of fly and worm proteins. As mentioned above, these numbers are likely to be modified when more accurate data become available, but any such changes are unlikely to change the general result.

The IgSF repertoire in Drosophila comprises 142 proteins. Of these, 89 belong to one of 18 clusters that contain two or more closely related proteins that have totally or largely been produced by gene duplication. This means that half the repertoire in the fly, i.e. 89-18=71 proteins, have been produced by gene duplication. Some proteins have been duplicated only once, some several times. In some instances the duplications have been followed by the loss or gain of domains. The six largest clusters are Defective Proboscis extension Response (DPR) proteins (23 members), the Beat proteins (14), the Three-IgSF-Cluster (8), Sidestep (6), Kekkons (6) and Wrapper/Klingon (5) clusters. Another six clusters have only two or three members (see Tables 1 and 3).

Many members of the large clusters have been previously identified: 20 proteins in the DPR cluster (Nakamura et al., 2002), all 14 Beat proteins(Fambrough and Goodman, 1996),Sidestep on its own (Sink et al.,2001), three Kekkons(Musacchio and Perrimon,1996), and Wrapper and Klingon(Butler et al., 1997; Noordermeer et al., 1998). Except for the cluster of Wrapper/Klingon, all these larger clusters are in the set of Drosophila-specific proteins that do not have C. elegans orthologues. This is an example of the lineage-specific expansions of protein families described by Aravind et al.(Aravind et al., 2000).

These two clusters had been identified and their functions determined prior to this work (Fambrough and Goodman,1996; Nakamura et al.,2002; Pipes et al.,2001). Although some of the Beat proteins have only marginal or no sequence matches, key residue analysis shows they are all related to each other. Note that some Beat proteins are attached to the cell membrane whilst others are secreted.

It proved to be difficult to reconstruct all the relationships between Dpr1 to Dpr20 described by Nakamura et al.(Nakamura et al., 2002). In some cases, the relationships are very remote and could only be shown by key residue analysis. For some of the sequences, the gene predictions were improved using the GENEWISE procedure (see above) and the Dpr-1 homologue as the query sequence (see above and Table 1). Dpr-12 has been mentioned in the work by Nakamura et al., but it could not be found in the set of predicted proteins. Owing to its small size (56 amino acids: the size of half an Ig domain), it has been disregarded in this analysis. CG31114-PA, CG14469-PA, CG15380-PA and CG15183-PA are predicted proteins that also belong to the same cluster, but were not mentioned previously.

We were able to identify three novel Dscam-like proteins (CG18630-PA in proposed fusion with CG7060-PA, CG32387-PA and CG31190-PA). Dscam is the Drosophila homologue of the human Down's syndrome cell adhesion molecule (DSCAM), which is required for axon guidance(Schmucker et al., 2000). The Dscam-like proteins hence represent interesting experimental targets.

This protein has been described recently as Drosophila homologue of the human Contactin (Falk et al.,2002). In fact, it makes a somewhat better match to Axonin, as was also found previously for its worm orthologue C33F10.5A(Teichmann and Chothia, 2000). The differences between Axonin and Contactin are subtle, but can be important when looking at the detailed functions of the proteins: For example, Contactin is known to display heterophilic but no homophilic binding activities(Falk et al., 2002), while both were observed for Axonin (Kunz et al., 2002). Both proteins interact with members of the L1 family,e.g. NrCAM, and are involved in axon guidance.

These two proteins match the N-terminal and C-terminal halves of CG31970-PA. They are also adjacent on the chromosome. We propose a fusion of the two predictions to give one protein.

The IgSF repertoire in C. elegans comprises 80 proteins. Of these 25 belong to one of seven clusters of two or more homologous worm proteins. This means that 25-7=18 proteins have been produced by gene duplication. This is only one quarter of the C. elegans repertoire; as we have just seen the proportion in Drosophila is one-half. The two largest clusters are the Zig proteins (eight members) and PVR-like kinases (five members). The other four have only two members (see Tables 2 and 3). Only 22 out of the 80 C. elegans protein have been identified by experiments.

Only Zig-2, Zig-3 and Zig-4 have sequence matches with E-values smaller than 10^–35. The membership of the other sequences in this family is based on their similar domain architecture, functional roles and manual inspection of the sequence alignments (see Aurelio et al., 2003).

The SSSD1.1 sequence in Wormbase has 623 amino acid residues. Using the homologous C. briggsae sequence and the GENEWISE procedure, we were able to identify additional exons, which increase the length of the predicted protein to 744 residues. SSSD1.1 is probably the C. elegansorthologue of Turtle (see Table 3).

Table 3 lists the proteins in the 26 clusters of closely related IgSF proteins that this work indicates as having homologues in Drosophila and C. elegans. These contain in all 36 proteins from Drosophila and 35 from C. elegans, i.e. a quarter of those in the first organism and just under half of those in the second.

Previous work had proposed putative orthologues for the Drosophilaproteins DPTP9 (K04D7.4), Lar (C09D8.1), PTP6 (F56D1.4), ImpL2 (C14F5.2,F42F12.2, Y48A6A.1), Kirre (K02E10.8, now SYG-1), Neuroglian (C18F3.2/3) and Klingon/Wrapper (F41D9.3b). Details of these, and the relationships found in this work are described in Table 3.

The cell surface class I has been mentioned above as the largest class in both organisms and as one of the two classes with large expansions in the fly. This is also true for the subset of those proteins common to both organisms: Drosophila has 21 while C. elegans has 12 proteins in the 11 clusters of the cell surface class I. There is only one cluster in this functional class, Neuroglian, where there are more members in the worm than in the fly (two and one, respectively). The clusters in the other functional classes have similar contributions from the two organisms with one exception. The exception is the PVR cluster of kinases, which has one member from Drosophila but five from C. elegans. An expansion of the cluster of kinases in C. elegans has been reported before(Rubin et al., 2000).

In both organisms, the number of proteins in the two largest functional classes, the cell surface class I and secreted proteins class, is higher for the organism-specific proteins than in the shared set described above: in the worm, 13 proteins are in these two functional classes and have a Drosophila homologue, while 25 proteins in these two classes are worm-specific. In the fly, this relationship is even stronger: 25 cell surface class I and secreted proteins have homologues in C. elegans, whereas more than three times as many or 82 proteins in these classes are fly specific. That means that, in addition to the expansion of fly proteins that have homologues in the worm, both organisms also developed a large set of organism-specific proteins, with again a larger expansion in the fly. Proteins of these classes play major roles in cell adhesion processes, and are most likely to contribute to the formation of fly specific characteristics.

We have deposited information on each of the IgSF proteins described in this analysis in an interactive, supplementary database that can be found at http://www.mrc-lmb.cam.ac.uk/genomes/FlyGee/. The information includes: alternative protein identifiers or experimental names, sequence homologies, structural annotation in terms of domains,transmembrane helices and signal sequences, the amino acid sequence and extensions of the gene predictions using NRDB90 or cDNA data, or references to literature. The database can be queried using keywords or protein identifiers. Each hit can include several sequences that all represent or point to the same protein: the predicted protein, other sequences such as a matching cDNA sequence, or the sequence found using GENEWISE, an experimentally determined sequence and/or the gene prediction from the previous release of the fly genome.

We have identified 142 IgSF proteins in Drosophila, described their domain architecture, and obtained an indication of the type of function that many of the novel proteins are involved in. We have also extended the work that was previously carried out on IgSF proteins in C. elegans. These results should be of use in the experimental characterisation of these proteins. Experiments, in turn, will refine or correct results reported here.

Some 26 clusters of closely related IgSF proteins are common to the two organisms and members of these clusters were present prior to the divergence of worm and fly. However, three-quarters of the Drosophila repertoire and half the C. elegans repertoire have emerged since their divergence. This means that a significant fraction of pathways involving the IgSF proteins in the much simpler organism, C. elegans, are not a subset of those in Drosophila but different. We also pointed to the particular expansion of two functional classes, many of whose members are involved in cell adhesion processes that play important roles during development. Relative to C. elegans, the greater size of the Drosophila IgSF repertoire, and the particular nature of many of its proteins, must be one of the contributing factors responsible for, for example, the formation of a more complex cellular structure in Drosophila.

The larger number of IgSF proteins in Drosophila contrasts with a smaller total number of genes: the current counts are 13,639 genes in Drosophila and 19,537 genes in C. elegans(Clamp et al., 2003). Some superfamilies in an organism expanded to improve its adaptation to its environment but without substantial increase in physiological complexity. Such changes in the protein repertoire could be called `conservative protein family expansions'. One example is the large expansion of two chemoreceptor families in the worm as compared with the fly(Robertson, 1998). Expansion of other superfamilies can lead to the evolution of organisms of higher complexity. This process could be called `progressive protein family expansions'. One example are the expansions of signal transduction domain superfamilies in the metazoan worm as compared with the unicellular baker's yeast (Chervitz et al., 1998). Another example, described here, is the expansion of the IgSF superfamily in Drosophila compared with that of C. elegans.

The general validation of this simple distinction between conservative and progressive protein family expansions will require a fuller investigation of the relationship between the size and function of protein superfamilies in organisms of different complexity.

C.V. has a pre-doctoral fellowship from the Boehringer-Ingelheim Fonds. We thank Lincoln Stein, Keith Bradnam, Leyla Bayraktaroglu, Aubrey de Grey, Don Gilbert, Marc Champagne, Agnes Southgate, Birgit Eisenhaber, Bernard de Bono and Julian Gough for their help at various stages of the project.