Sequence data is available for many nematode species primarily because of the recent sampling of nematode transcriptomes using ESTs 2122. In this study, a total of 130,357 contig-level EST consensus sequences, assembled from 262,497 ESTs from 29 nematode species, were translated into putative primary sequences of nematode proteins (Table 1). In addition, the complete gene-sets of 84,408 proteins from five genome sequencing projects (3 Caenorhabditis species, B. malayi, and Ancylostoma caninum) were added. Hence, a total of 214,159 polypeptides/proteins from 32 nematode species in four nematode clades were used for the subsequent analysis (Table 1). The complete dataset is available online for retrieval 26.

Protein families were built using MCL clustering 27 with the Markov cluster algorithm (MCL), which would not suffer greatly from potential problems caused by multi-domain proteins, promiscuous domains, or fragmented sequences. In total, the 214,159 nematode coding sequences were clustered into 54,036 protein families. Protein families conserved across multiple species suggest conserved features and essential functions, therefore a total of 5,326 multi-species families (112,271 sequences), with members from at least three different nematode species, were chosen for further evaluation. Of these, 1,939 protein families (36%) did not share homology to protein family models in Pfam-A (Figure 1).

A protein family built by MCL can include multiple EST contigs originating from a single gene. To reduce this redundancy, we first clustered EST contigs into EST clusters, each containing a group of contigs likely representing the same gene 28293031. Then a step-wise approach was implemented for each MCL family to: i) generate a multiple alignment from all members, ii) build a Hidden Markov model (HMM) from the multiple alignment, iii) calculate a matching score for each member of the family based on the HMM, and iv) retain only the single EST contig from each EST cluster assigned with the best matching score as the sole representative of the gene in the protein family. Finally, an additional filtering step required each valid family to have at least 10% (in length) of its full alignment contributed simultaneously by sequences from 3 or more species. All of the above led to the identification of 1,593 multi-species non-Pfam Nematode protein Families (NFs) with a total of 13,963 coding sequences (Figure 1).

The NFs were further categorized by sequence similarities and taxonomic origins of their members. Comparison to the NR-noNema database (all protein sequences in the non-redundant NR database except those from nematodes), at a BLAST e-value cutoff of 1.0e^-03, identified 835 NFs (8,764 proteins) containing homology in non-nematode species although they were derived from nematodes (Figure 1) (see Additional file 1). Approximately 90% of these NFs shared primary sequence similarities to arthropod proteins, among which 26 families (NFa; 212 sequences) were found to be homologous only to sequences from arthropods but not to any proteins from non-nematode and non-arthropod species at a BLAST e-value cutoff of 1.0e^-03 (see Additional file 2). Molecular features conserved in the sequences of both nematodes and arthropods were evident in these families, such as small insertions/deletions [see Additional file 3]. Both Nematoda and Arthropoda belong to Ecdysozoa, sharing the common pattern of growth-by-molting 3233, therefore these protein families likely reflect the evolutionary conservation between these organisms at the molecular level. In addition, macrocyclic lactones, such as avermection and milbemycin, have been successfully used as endectocides to treat both the nematode endoparasites and arthropod ectoparasites simultaneously 34. Hence, the 26 NFs that were conserved only among nematodes and arthropods could be potential targets for the development of novel endectocides. Interestingly, five of the 26 families were mapped to canonical KEGG metabolic enzymes 353637 as various subunits of the electron transfer Complex I [see Additional file 2].

The remaining 758 NF families (5,199 sequences) did not contain members with sequence similarities to any non-nematode proteins with the BLAST e-value cutoff of 1.0e^-03 (Figure 1). With no obvious homology to either non-nematode proteins or Pfam-A entries, they became candidate novel protein families specific (or restricted) to nematodes. Their conservation among multiple nematode species, especially of those spanning all the four nematode clades (Table 2) (phylogeny based on 38) included in this study (see below), suggests that they may have emerged in early nematode ancestors after they diverged from other animals, and they may include the molecular determinants archetypical to the phylum Nematoda. Although their nematode-specificity implies only limited knowledge currently available, close investigation will likely reveal conserved functions essential to many nematodes, and the interference with their functions will likely cause severe damaging effects in nematode parasites in a novel, safe, and broad fashion.

In addition, by comparing to a database containing all the currently available sequences from free-living nematodes (at a BLAST e-value cutoff of 1.0e^-05), these nematode-restricted NFs were further divided into 296 NFp (putatively specific to parasitic nematodes) [see Additional file 4] and 462 NFn (conserved across both free-living and parasitic nematodes) [see Additional file 5]. The NFp and NFn groups contained 1,514 and 3,685 proteins, respectively, averaging at 5 and 8 members per family with different family size distributions [see Additional files 1 and 6], suggesting differences between the two groups. In addition, while the majority in the NFp group (90%) contained members from only a single nematode clade, a similar number of NFn families were found to span 1, 2, or 3 nematode clades respectively (~31% of all NFn for each) (Figure 2). These results indicate that proteins within the NFn families are conserved in more evolutionarily divergent nematode species and are thus likely involved in essential nematode function across the phylum Nematoda; on the other hand, the NFp families tend to be restricted to smaller evolutionary niches and are most likely related to the specific patterns of parasitism that were hypothesized to emerge independently, at multiple times, during nematode evolution 38.

RNA interference (RNAi) has become an efficient high-throughput approach for rapidly determining gene functions via transcript knockdown in many organisms, and especially in C. elegans 39404142. However, applying RNAi in parasitic nematodes possesses significant challenges. For example, their obligate parasitic life cycles, with movement into and out of the host, make both the delivery of double-stranded RNA and the assessment of phenotype difficult. Although successes have been demonstrated in several parasitic nematode species (reviewed in 2143), these methodologies are far from established for large-scale investigation.

Gene functions derived from RNAi experiments in C. elegans can be further extrapolated, to an extent, to orthologous genes in other nematodes 21. The NFp families did not have members from the free-living C. elegans. A total of 356 of the 462 NFn families had C. elegans members, most of them (321) had RNAi results available. Among them, 85 families contained C. elegans genes associated with non-wild type RNAi phenotypes, including 62 with strongly deleterious effects (Emb, Ste, Stp, Lva, Lvl, and Gro) [see Additional file 5]. Such RNAi results could shed light on the putative functions of their counterparts in other nematodes included in the same protein families. For example, NF_0208_1522 contained two members from each of the three clade V free-living Caenorhabditis species, as well as four from animal parasites (AC02485 and OS00413 from clade V and SS02646 and PT01276 from clade IVa), and one from the clade IVb plant parasite M. incognita (MI03217). The inclusion of the two C. elegans insulin-like genes, ins-17 and ins-18, suggested that this family represented a group of conserved nematode proteins likely regulating the growth and lifespan as demonstrated by RNAi in C. elegans.

Furthermore, the distribution of these RNAi results among NFn families showed that families conserved in nematodes spanning a broader evolutionary distance, especially those with members from all the four nematode clades included in this study, were much more likely to have observable phenotypes with RNAi knockdown in C. elegans (Figure 3). This suggested that these multi-clade NFn families, which might have emerged in the early common ancestors of Nematoda and remained to be conserved in many nematode species since then, could be the most essential genes required for nematode survival.

Based on sequence similarities, members of the NFs were mapped to the Kyoto Encyclopedia of Genes and Genomes (KEGG), which offers curated information about genes and proteins, as well as molecular wirings of interactions and reaction networks especially in the canonical metabolic pathways 3637. As expected for these novel families, none of the NFp members could be mapped, and the limited assignments for the NFn families were always derived from their C. elegans members that were previously annotated by KEGG [see Additional file 5]. In addition, for each of the eight NFn families mapped through KEGG, a same KEGG Orthology (KO) entry was always assigned consistently to all of its family members meeting the mapping criteria (Table 3), confirming that the family members grouped by MCL were indeed homologous proteins.

Unexpectedly, all the eight KEGG entries assigned to NFn proteins, such as the various subunits of electron-transfer complexes (Table 3), were canonical enzymes with extensive knowledge available, including sequences of orthologous groups from many non-nematode species. It was intriguing because the proteins included in these nematode families, especially the C. elegans members that had been previously annotated in KEGG, had to contain a fair amount of sequence homology to be recognized as the canonical enzymes, yet they were found without similarities to any non-nematode proteins by our discovery pipeline. Close examination showed that this conflict was caused by a slightly looser requirement of homology during the KEGG mapping. Therefore, the putative annotation assigned to these nematode proteins represented the relatively low levels of sequence similarities that were still able to reveal their functions with confidence.

More interestingly, we were able to identify unique sequence features of these nematode proteins, such as nematode-specific insertions and deletions, in all the eight NFn families with KEGG annotations. Such nematode-specific features may have prevented their homology from being identified in our initial screening. For example, members of NFn family NF_0313_0956 were mapped to KO: K03951 as the NADH dehydrogenase (ubiquinone) 1 alpha subcomplex 7. Indeed, these nematode sequences could be forcibly aligned with the group of proteins from non-nematode organisms that were assigned to the same KEGG entry, after allowing two fragments of nematode-specific insertions (Figure 4). The lack of a homologous 3D model of this enzyme made it impossible to investigate the impact on its structure caused by these insertions, but they likely created additional loops in the nematode proteins that may introduce novel functional features specific to Nematoda. These results demonstrated the mechanism of directed diversification of existing protein folds in these proteins during nematode evolution.

Energy generation mechanisms are extremely complicated in nematodes. Free-living nematodes, such as C. elegans, rely on mammalian-type aerobic electron transfer for the generation of ATP. However, this oxygen-based energy generation mechanism is thought to be unlikely for many parasites because of the low levels of oxygen in their parasitic environments and the lack of an efficient circulatory system and respiratory organs in nematodes. Instead, an anaerobic energy generation independent of oxygen has been suggested. Studies of the clade III intestinal parasite Ascaris suum have revealed that a developmental switch around stage L3, wherein an anaerobic pathway in adults, named the malate dismutation pathway or the PEPCK-succinate pathway, replaces the mammalian-type aerobic energy generation found in embryos and larvae 44454647. Our previous investigation of the adult transcriptome from another clade III parasite Dirofilaria immitis has suggested a similar mechanism 28.

With KEGG mapping, we identified a total of six components of the well-defined energy-generating electron transfer complexes among NFn families, each with relatively weak yet clear homology to the canonical enzymes. Based on this, and the finding that five NFa families conserved in only nematodes and arthropods were also mapped to the same pathway [see Additional file 4], we propose that the early common ancestors of nematodes may have obtained a series of novel features in their energy generation to collectively and cooperatively accommodate the severe challenges imposed by the different life styles found in complex parasitism, and that those NFa families may have represented an intermediate evolutionary path, which would have emerged in the common ancestors of Ecdysozoa, that leads to unique features specific to Nematoda. This phyla-specific energy generation mechanism, significantly distinct from the canonical pathway of oxidative phosphorylation used by mammalian hosts, offers a prime target for the development of next generation parasite control strategies with potentially high specificity and minimal toxicity.

To offer further characterization, NF families were scanned against the InterPro database 48 for generic sequence features [see Additional files 4 and 5]. Not surprisingly, even with both KEGG and InterPro mappings, we were able to obtain information for only 30 NFp and 124 NFn families, leaving the majority of the NF families (~80%, 606/758) completely un-annotated.

Protein structure diverges more slowly than primary sequence 49, therefore fold similarity and structure-based alignments were used for the detection of distant homology of the 606 NF families with no KEGG or InterPro annotation. Firstly, we generated predictions of basic structural information for each, including secondary structure, domain architecture, and flexible/dynamic regions. These predictions are integrated and displayed within a customized genome browser 50 for easy navigation 26. Secondly, structural homology to previously defined protein folds in Protein Data Bank (PDB) 51 or Pfam 1920 were searched for using a newly improved version of the meta-predictor, Meta-BASIC 52, which combines sequence profile, secondary structure, and prediction of the burial states of individual amino acid with various scoring systems and meta profile alignment algorithms. The putative matches with a confident 3D-Jury cutoff score of 50, which corresponds to a false positive rate of less than 5% 49, are available online via graphic display 26. Of the 3,926 sequences from the 606 families, we were able to identify putative homology for 56 polypeptides from 9 families to known protein folds in PDB, and 14 in 11 families to those in Pfam [see Additional files 4 and 5]. Close investigation of such distant homology can help to elucidate potential function (as described below).

An example is family NF_0103_0974 with a domain of 136 amino acids conserved in all the five members. Several structure prediction methods included in Meta-BASIC, such as the homology modeling tool FFAS3 53 and threading algorithms 3D-PSSM 54 and INUB 55, all assigned this conserved domain as a match to the PDB entry 1buqa, which was classified as the structure signature for a group of nuclear transport factor 2 (NTF2) like proteins. Further structural modeling using Modeller 56 showed that the nematode domain contained all the major components of this fold. The NTF2-like superfamily contains members with diverse functions, including enzymes such as enscytalone dehydratase, delta-5-3-ketosteroid isomerase, and limonene-1,2-epoxide hydrolase, and non-enzymatic homologues such as NTF2 57. Even though none of these functions could be clearly assigned to NF_0103_0974, the presence of a cysteine cluster might suggest the existence of zinc binding site in these nematode proteins (Figure 5).

Proteins secreted or expressed at surfaces are essential components of the cellular regulatory networks that ensure proper interactions with the environment for survival. Thus far, all the commercially available anthelmintics have a gain-of-function mode of action targeting channels and receptors associated with membranes 58. In addition, nematode antigens are believed to be most effective when secreted from glands 59 or expressed on exposed surfaces such as the intestinal lumen in hookworm 60, where they come into contact with and are therefore targeted by effector molecules from the host immune system. Among the NFn and NFp families, there were 45% and 27%, respectively, having signal peptide for secretion predicted in their sequences, and 26% and 21%, respectively, containing members predicted to have both signal peptide and transmembrane domains. With the caution that some of these predictions might be putative targeting signals for transport to intracellular compartments such as mitochondria or peroxisomes, we were able to identify 149 NF families, from the total of 758, as candidates for expression at accessible surfaces [see Additional files 4 and 5].

The intestine has been our focus in other studies 61, because it is one of the major organs in nematodes creating a key surface at the intestinal apical membrane to interact with the environment. The easy accessibility of the nematode intestine has made it an attractive target for immune or chemical control of parasitic species 626364656667. Comparative studies among intestinal transcriptomes from the free-living C. elegans and parasites A. suum and H. contortus identified a group of 241 protein families (IntFam-241) expressed in the intestine of all three nematodes. This group was further proposed to represent an ancient group of intestinal proteins responsible for the core intestinal functions in many nematode species 61. There were 12 NFn families from this study overlapping with the IntFam-241. Majority of them (11/12) spanned three or more nematode clades, and eleven had predictions of either signal peptides or transmembrane sequences [see Additional files 4 and 5]. In addition, all of the 12 NFn families had C. elegans members with RNAi information available, and all but two of them had observable RNAi phenotypes [see Additional file 5 and Additional file 6], suggesting that they warrant further investigation.

Detailed information on genetic materials and cDNA library construction are available online 2326. ESTs were processed and clustered as described earlier 28293031. EST contig sequences were translated individually by Prot4EST, a 6-tier translation pipeline combining both similarity-based methods and de novo predictions 7172. Only one translation was accepted to represent each EST contig, during which false translation was likely reduced by retaining preferably the longest open reading frame with strong supporting evidence, if available, in the form of similarities to known or predicted proteins. The gene-sets from the genomic sequencing projects were: C. elegans (Wormbase v158; 23,162 proteins), C. briggsae (downloaded June, 2006; 19,723 proteins), C. remanei (preliminary set, October, 2005; 25,775 proteins), B. malayi (11,750 proteins), and A. caninum (preliminary set; 4,038 proteins).

WU-BLASTP (wordmask = seg postsw) was used to query the translated sequences against protein databases, and WU-TBLASTN (wordmask = seg lcmask) for searching against nucleotide databases 73. Databases used for sequence comparisons were: i) NR-noNema, containing all sequences from the non-redundant protein database NR except those from nematodes (downloaded 06/06/2007), ii) NR-noNema-noArthropoda, NR sequences with those originated from nematode and arthropod species removed (downloaded 06/06/2007), iii) Free-living, all the 71,496 protein sequences from the free living species C. elegans, C. briggsae, C. remanei, P. pacificus, and Z. punctata.

An all-against-all WU-BLASTP was performed on the total of 214,159 translated sequences from the 32 nematode species. Raw BLAST results were fed to a C-language implementation of Markov cluster algorithm 74, a fast and scalable unsupervised clustering algorithm based on simulation of flow in graphs 27. MCL simulates flow in a protein similarity graph, assigning complete protein sequences into families based on density and strength between them. It makes no attempt to decompose the sequences into their component domains, but rather produces protein clusters that correlate well with the overall domain architecture. The tightness of MCL clustering is determined by a user-defined parameter, the inflation factor. A larger inflation factor leads to a higher granularity of clustering, resulting in the generation of protein families based more likely on local domains and less likely on global similarities. To avoid such high granularity, we validated the clustering with three inflation factors, 1.6, 1.8, and 2.0, respectively. With them, the numbers of protein families differed by only 5% (data not shown). After manual inspection, the inflation factor of 1.8 was chosen to generate 54,036 protein families for further screening and downstream analysis. HMM screening against Pfam-A entries (v20; 8,296 entries) 1920 were performed using hmmpfam in HMMER v2.1 at a p-value cut-off 1.0e^-05. To remove sequence redundancy, an automatic screening pipeline was implemented to: i) generate multiple alignments for each family with MUSCLE v3.52 75, ii) build a HMM from the multiple alignment using hmmbuild from HMMER v2.1, iii) calculate a matching score for each member of the family based on the HMM using hmmsearch from HMMER v2.1, and iv) retain only one EST contig assigned with the best matching score for an EST cluster, each of which can be approximated as the collection of EST contigs originated from a single genomic locus.

A hidden Markov modeling-based algorithm, Phobius 76, was used with default settings. Each query sequence was further annotated as SP for containing signal peptide, TM for containing transmembrane region, or intracellular, based on raw Phobius outputs.

The E-value cut-off of 1.0e^-10 reported by WU-BLASTP against the Genes Database Release 43.0 from Kyoto Encyclopedia of Genes and Genomes (KEGG) was used for pathway mappings. For each query, the top match and all the matches within a range of 30% of the top BLAST score, if meeting the cut-off, were accepted for valid KEGG associations 353637. Default parameters for InterProScan v16.1 77 were used to search against the InterPro database 48.

The following structural information was predicted for the NF family members: secondary structure prediction by PsiPred 78, domain architecture with SSEP-Domain 79, detection of flexible and dynamic regions using DISOPRED2 80. To investigate distant homology, query sequences were submitted to Meta-BASIC 52 against data sets of meta-profiles derived from PDB 51 and Pfam 1920. For the proteins identified as putative matches by Meta-BABIC, potential globular regions were identified with GlobPlot 81, and were subsequently submitted to the Structure Prediction Meta Server 82 for additional analyses. Secondary structure prediction was performed with PsiPred and ProfSec via the meta server, collected models were screened with 3D-Jury 83, a consensus fold recognition prediction method, for final predictions. Homology models were further obtained with Modeller version 6.2 56 with additional refinement of structural alignments performed according to the Verify3D results as previously described 84.