The S. mansoni phylome reconstructed in this work was derived from the comparative analysis of all proteins encoded in the parasite genome (predicted proteome) and their homologs in 12 other eukaryotic proteomes whose genomes were completely sequenced (Table 1). The set of selected species is particularly rich in metazoans (11 species), including ten invertebrates, one tunicate, and one vertebrate. One choanoflagellate, Monosiga brevicollis, was included as outgroup of the phylogenetic reconstruction. The metazoan species selected represent important evolutionary innovations, e.g. the origin of the third germ layer, the development of organs, systems, complex patterns of communication, and the emergence of the adaptive immune system, making this dataset set especially suitable for addressing the evolutionary innovations in S. mansoni in the context of metazoan evolution.

1 - Code assigned to each species in the S. mansoni phylome. 2 - Taxonomic identifier at NCBI (TaxID). 3 - Number of proteins analyzed per species. 4 - Database from which the protein data were retrieved.

To perform the phylogenetic analyses, we applied an automated pipeline similar to the one used for the human phylome project 39 . This pipeline is illustrated here (Figure 1). The resulting alignments, phylogenies, and orthology predictions can be accessed at PhylomeDB 47 (http://phylomedb.org).

Using this phylogenomic approach, we analyzed 11,763 S. mansoni proteins and obtained 7,964 phylogenetic trees covering 70% of the parasite’s proteome. This coverage is remarkably similar to that of other phylome data of newly sequenced genomes such as that of the pea aphid Acyrthosiphon pisum (67%) 38 .

The absence of trees for the remaining 3,490 proteins is either due to a possible high degree of divergence between the S. mansoni proteins and their homologs in the other selected species, an indication of the uniqueness of the parasite’s proteome, or it reflects the presence of errors in gene models. Out of the 7,964 phylogenetic trees, 3.038 (38%) correspond to trees with “seed” proteins with a completely unknown function and without any GO 48 assignment in SchistoDB 17 .

In order to create a complete list of orthology and paralogy relationships among S. mansoni proteins and those encoded in the other eukaryotic proteomes included in this work, we analyzed the parasite’s phylome using a species-overlap algorithm as previously described 39 . The comprehensive catalogue of phylogeny-based orthology and paralogy relationships among S. mansoni and other species was made publicly available at PhylomeDB 47 .

Owing to the increasing rate at which new fully sequenced genomes are released, the accumulation of genomic and proteomic data has been much higher than the rates at which genes or proteins are experimentally characterized. Aiming at producing a high confidence set of functional predictions for S. mansoni proteins, we used the evolutionary relationships as inferred from phylogenetic trees to obtain subsets of one-to-one (single homolog in S. mansoni and in other species) homology relationships among S. mansoni proteins and the homologs from other species included in the present study (Figure 2).

By using such phylogeny-based approach, we transferred 10,175 functional annotations (GO terms 48 ) to 3,451 S. mansoni proteins, from which 790 (7% of the parasite’s proteome) were previously annotated as “hypothetical protein”, corresponding to proteins whose function had not been predicted or experimentally tested before (Additional file 1 Table S1). The transfer was performed from each ortholog with known function in the selected taxa to the S. mansoni “seed” protein. For the other proteins that already had any functional prediction, the annotation was confirmed or improved. Consequently, a “seed” protein could receive more than one functional description. In these cases, all functional annotations were maintained allowing the user to choose the closest related transferred functional annotation, those that came from model organisms, or even to create a consensus based on all of them.

To validate the applied methodology, we retrieved reviewed S. mansoni proteins from UniProt 49 , including experimentally confirmed ones, to evaluate the annotation transferred by the phylogenomic approach. The functional annotations performed by PhylomeDB correspond to known functions in the aforementioned database (Additional file 1 Table S2). Even though the BLAST search may detect distant homologs with additional domains, our subsequent phylogenetic reconstruction and our selection of orthologs will select those orthologs that are likely to have similar domain architecture. This is an additional reason why an orthology-based annotation is preferred over sequence similarity searches, since orthologs as compared to paralogs have a higher tendency to share a similar domain architecture 50 .

Although less reliable than those based on one-to-one orthology relationships, annotation transfer based on more complex subsets (one-to-many, many-to-one, or many-to-many) may provide important hints to predict the biological function of S. mansoni proteins. However, in these cases, one or more genes are co-orthologous to a set of genes in another genome due to lineage-specific duplication(s) that can be associated with functional shifts, affecting the reliability of the functional transfer 38 51 . An example of a one-to-one transfer from a Drosophila melanogaster protein to a S. mansoni protein comes from the phylogenetic reconstruction of the Phy000V14T_SCHMA (Smp_170950) protein, potentially related to the glycine cleavage system, and its homologs in the selected species (Figure 3). The analysis of this tree resulted in six transfers of functional annotation from homologous proteins to the S. mansoni “seed” protein. The GO terms in all six functional annotations are related to aminomethyltransferase activity and glycine catabolic process providing further support for the annotation transfer. In this example, to illustrate a case of a one-to-one transfer, we chose the functional annotation transferred from Drosophila melanogaster once, according to the information available in UniProt 49 , it is one of the orthologs with known function and experimental validation. Tags for homologous sequences with experimental validation are not available in PhylomeDB 47 . However, links to UniProt 49 and other databases are provided.

To explore the benefits offered by comparative genomics in order to improve functional annotation of genes and gene products, it is also necessary to consider the limitations involved in this approach. Although it is generally accepted that functional annotation through orthology, rather than just homology relationship, constitutes one of the most promising annotation approaches, these surveys are designed to provide predictions regarding the likely protein function, but it does not substitute experimental confirmation 36 52 . Functional diversity is often associated with significant divergence at the sequence level, but high levels of identity do not ensure that two or more proteins perform the same function, since subtle changes in active sites are able to completely change the protein function 53 .

As we previously mentioned, evolutionary analysis involving fully sequenced genomes/proteomes remains a challenge. Although the tools here applied were not originally designed for large scale phylogenetic analysis, we adapted them to work on a large scale, since we strongly believe that a system-wide perspective on evolutionary processes can greatly improve the understanding on how genomes came to be and what evolutionary process took them there. Functional prediction as described in the present work could be used as a starting point for future projects, prioritizing the selection of certain genes or proteins for new experimental studies.

An additional advantage of the phylogeny-based approach is that it readily provides a collection of gene evolutionary histories that can be mined for particular events. Since gene duplication is considered one of the main mechanisms for functional innovation and diversification 54 , we explored the S. mansoni phylome to identify protein families that have been specifically expanded in this lineage, since its diversification from the other sequenced metazoans. We used the above-mentioned species-overlap algorithm that identifies duplication nodes and also provides clues of the relative dating of the duplication event 39 55 .

Such analysis revealed that in 3,051 reconstructed phylogenetic trees there is at least one paralog connected to the “seed” protein through a duplication node (Additional file 1 Table S3). Among these, 211 phylogenies show lineage-specific duplications in the three Schistosoma species in comparison with the other taxa. These expansions are small-to-moderate in size, resulting in a total of two to ten paralogs, and include some of the most significant expansions as discussed below.

The inclusion of S. haematobium and S. japonicum proteomes gave us a high resolution within Schistosoma genus and allowed us to make comparisons across this taxon. In general, the expansions observed in S. mansoni can also be observed in the other two Schistosoma species, although with variable number of paralogs in each species. As previously observed by evolutionary relationships, cytogenetic data, and syntenic analyses, the present study shows that S. mansoni is more closely related to S. haematobium than to the S. japonicum 56 57 58 59 . Moreover, 170 evolutionary trees have only S. mansoni and S. haematobium proteins, while only six phylogenies have solely S. mansoni and S. japonicum proteins. Meanwhile, most of the homologous proteins shared by S. mansoni and S. haematobium are annotated as “hypothetical protein” and do not have any predicted function or significant hits with known proteins in public databases as UniProt 49 , Pfam 60 , or non-redundant (nr) NCBI database (ftp://ftp.ncbi.nih.gov/blast/db).

A small number of phylogenetic trees (1,45%) had only sequences of S. mansoni. These could be the result of very recent duplication events of proteins that are specific to this species. However, many of these genes were not found in the genetic map of S. mansoni 16 18 and they do not contain protein domains traceable at Pfam 60 . BLAST searches against the non-redundant (nr) NCBI database detected a few non-Schistosoma proteins as significant hits that were annotated as hypothetical in all cases. For these reasons we rather believe that these sequences correspond to spurious predictions. Further analyses will be conducted in the future in order to confirm or refute this hypothesis.

Among the most significant protein expansions in S. mansoni we identified tetraspanins, fucosyltransferases, venom allergen-like proteins (SmVAL), tegumental-allergen-like proteins (SmTAL), leishmanolysins, and elastases, which were previously proposed as drug targets, once they can be related to morphological or physiological specificities of this parasite 5 20 61 62 63 64 65 . In these cases, the protein family membership ranged from 6 to 23 paralogs encoded in the parasite’s genome.

Tetraspanins are small proteins with four transmembrane domains involved in the coordination of intra and intercellular processes, such as signal transduction, cell proliferation, adhesion, and migration, cell fusion and host-parasite interactions 66 67 . The function of schistosome tetraspanins are not completely understood, but cell-cell interactions and maintenance of cell membrane integrity might be performed by these proteins as well as they can be receptors for host ligands, acting on immune evasion 61 . The suppression of two tetraspanin genes (Sm-tsp-1 and Sm-tsp-2) by RNA interference in mice also suggests that these proteins play important structural roles in the parasite’s tegument, being a good target for anti-schistosomal vaccine 68 . Figure 4 illustrates an example of tetraspanin lineage-specific duplications. In this case, the number of homologs in the three Schistosoma species varies from six to eight. Tree topology shows distinct well-supported clades suggesting that structural and/or functional variants might be present. Three proteins in this dataset have experimental evidence: Phy0048JNS_SCHHA (Q26499), Phy0048WJL_SCHMA (P19331), and Phy0005UU9_DROME (O46101) 49 69 70 .

Venom allergen-like proteins (SmVAL), also called sperm-coating protein-like (SCP-like), are structurally related proteins members of the SCP/TAPS family. In Platyhelminthes, these proteins have been linked as potential modulators of immune function and components of sexual development 71 . Although the specific function of each SmVAL family member is unknown, there is evidence suggesting potential roles in larval penetration, host immune response modulation, and adult worm development 63 71 . Furthermore, analyses of SmVAL transcripts demonstrated that the corresponding genes are upregulated in infective stages of the parasite, highlighting SmVAL proteins as candidates for novel vaccine strategies 71 72 .

Fucosyltransferases are enzymes that catalyses the fucose transfer from the donor guanosine-diphosphate fucose to different acceptor molecules such as oligosaccharides, glycoproteins, and glycolipids 73 . In schistosomes, fucosyltransferases are involved in producing immunomodulatory epitopes during infection, granuloma formation, egg/endothelium interactions, and were previously highlighted as anti-schistosomal candidates 63 74 .

Tegumental-Allergen-Like proteins (SmTALs) are members of a protein family present in parasitic Platyhelminthes 64 75 . These proteins are located inside the tegument and have different life-cycle expression patterns 64 . The tegumental protein Sm22.6 is considered the main target for human IgE in S. mansoni and human IgE response against this protein is associated with the development of age-dependent partial immunity to S. mansoni infections in endemic areas 64 76 .

Leishmanolysin, (also called invadolysin and SmPepM8), is a major surface protease member of the metallopeptidase M8 family. This protein can perform activities in schistosomes similar to those performed in Leishmania where these proteins are involved in different types of processes like degradation of the extracellular matrix and inhibition or perturbations of host cell interactions 63 72 . In turn, elastases are serine proteases that in schistosomes play a pivotal role in the penetration by cercariae of host skin to initiate infection. Recent studies have also revealed that these proteases can be employed by schistosomes to overcome or evade the host immune response 77 78 . Members of S. mansoni peptidase families such as leishmanolysins, cercarial elastases, and cathepsin D proteins were subjected to a detailed study in respect to their domain architectures, functional properties, and evolutionary relationships as described elsewhere 65 .

Another specific feature of schistosomes is related to their tegument. Distinct from nematodes, which have a cuticle covering and protecting the organism body, schistosomes are covered by a living syncytium bounded by a complex multilaminate surface, which undergoes several adaptations soon after infection is initiated 79 80 81 . The external double membrane plays a crucial role in host-parasite interactions, being responsible for diverse mechanisms of survival 19 82 83 . The development of a tegument, highly specialized and resistant to immune damage, was accompanied by evolutionary adaptations, for example, the expansions of other protein families encoding annexins, cadherins, and innexins.

Annexins are widely distributed in eukaryotes performing a broad range of important biological processes related to tegument membrane 84 85 86 . In schistosomes, annexins appear to be involved in parasite’s stability protecting against immune attack by the host as well as against structural breakdown 85 86 . Cadherins are adhesion molecules that mediate Ca²⁺-dependent cell-cell adhesion and whose duplication events happened probably in parallel to the advent of a third germ layer in flatworms 5 87 . Innexins are components of gap-junction proteins, the intercellular channels that allow for the exchange of ions and other small signal molecules 88 89 . In C. elegans, innexins have been implicated in different processes like electrical coupling between pharyngeal muscles, calcium propagation in the gut, gap junction-mediated oocyte, and sensory neuron identity 89 .

In summary, we identified that approximately 45% of the S. mansoni predicted proteins that were covered by this phylogenomic analysis have, at least, one paralog encoded in the parasite genome that might have arisen by gene duplication events that occurred after its divergence from other selected taxa (Additional file 1 Table S3). In other eukaryotic genomes this value ranges from 30 and 65% 90 , whereas in C. elegans this value is equal to 49% 91 .

Altogether, the present results indicate that besides the exploitation of host endocrine and immune signals, the parasite genome exhibit multiple events of gene duplication which may be, at least partially, an adaptive response related to the parasitic lifestyle. These expansions probably reflect the intriguing complexity of evolutionary events that happened over time, resulting in important characteristics in schistosome’s biology with consequences to the disease it causes. Taking into account the host environment and the selective forces that it imposes to a parasite, the phylogeny of host(s) and parasite(s) are probably closely related, once this coevolution will be responsible for the continuity or elimination of such an interaction. Nonetheless, previous empirical experiments involving schistosomes and the intermediate host provide further support to suggest the potential for host-schistosome coevolution 92 .

In this context, it is important to analyze the evolutionary history of protein families during screening for potential targets for drug and vaccine development. Incorporating the evolutionary perspective in drug development studies can improve our understanding regarding drug resistance and effectiveness, as well as to guide new strategies of drug discovery. Gene duplication events as well as adaptive evolution should be considered during this process, since an anti-parasitic drug could bind a single protein or in all proteins encoded by a multi-gene family 93 . As a consequence, therapies which target a subset of genes that arose by duplication may not be effective at low doses. To solve this problem, the drug's effectiveness can be increased when a single-copy gene is targeted and its function is inactivated causing complete perturbation of a vital pathway 93 94 .

Predicted proteomes from 13 fully sequenced eukaryotic genomes were downloaded from JGI Genome Projects, SchistoDB, Quest For Orthologs, WormBase, BeetleBASE, and FlyBase (Table 1). The taxon sampling was selected according to the availability of the predicted proteomes and based on the phylogenetic position of each species. The comprehensive taxa selected cover important evolutionary innovations making this dataset set especially suitable for addressing the evolutionary innovations in schistosomes in the context of metazoan evolution. Model organisms were also included to provide functional annotations that could be potentially transferred to S. mansoni homologous proteins.

To reconstruct the complete collection of phylogenetic trees for all S. mansoni proteins and their homologs in other 12 fully sequenced organisms (Table 1), we used a similar automated pipeline to that described earlier for the human proteome 39 (Figure 1). A local database was created containing data from the S. mansoni proteome and those of 12 other completely sequenced genomes/proteomes. Alternative splicing products and identical sequences from the S. mansoni proteome were filtered out. For each protein encoded in the S. mansoni genome (“seed”), a Smith-Waterman search 96 (E-value ≤ 10^-5) was performed against the above mentioned database to retrieve proteins with significant sequence similarity. Sequences that aligned with a continuous region longer than 50% of the query sequence were selected and aligned using MUSCLE 3.6 40 , MAFFT 41 , DIALIGN-TX 42 , and M-Coffee 43 with default parameters. Positions in the alignment containing a high number of gaps were eliminated using trimAl 44 , with a consistency cutoff of 0.1667 and a gap score cutoff of 0.1. Neighbor-joining trees were derived from the trimmed alignments using scoredist distances as implemented in BioNJ 45 and maximum likelihood trees were obtained as implemented in PhyML using the NJ tree as a starting point 46 . For each “seed” protein phylogenetic reconstruction, we tested four different evolutionary models (JTT, WAG, BLOSUM62, VT, LG, CpREV, and DCMut). In all cases a discrete gamma-distribution model with four rate categories plus invariant positions was assumed, the gamma parameter and the fraction of invariant positions were estimated from the data. Tree support values were computed by approximate likelihood ratio test (aLTR) as implemented in PhyML 46 97 . The evolutionary model best fitting the data was determined by comparing the likelihood of the used models according to the Akaike Information Criterion (AIC) 98 .

To derive orthology and paralogy relationships among S. mansoni proteins and those encoded in the other genomes included in this study we used a species-overlap algorithm as described in 39 and as implemented in ETE (Environment for Tree Exploration) 99 . This algorithm uses the level of species overlap between the two daughter partitions of a given node to define it as duplication or a speciation event. The analysis starts at the protein used to generate the tree (“seed” protein) and runs through the internal nodes of the tree until it reaches the root. All the trees were rooted at the midpoint. If the two partitions share any species (if there is species overlap), the node is defined as a duplication node and the proteins are considered paralogous ones. Otherwise (if there is no overlap) the node is defined as a speciation node leading to orthologous proteins. Once all the nodes have been classified, the algorithm establishes the orthology and paralogy relationships between the “seed” protein and other proteins included in the tree according to the original definition of these terms 39 100 . A previous study has shown that the species-overlap algorithm produces reliable orthology predictions with higher sensibility than a strict reconciliation method 101 .

Based on the list of orthology and paralogy relationship we performed the transfer of functional annotation from each ortholog with known function to the S. mansoni “seed” proteins. To produce a confident set of functional predictions for S. mansoni proteins, we classified the list of orthologs in different subsets of orthology relationships (one-to-one, one-to-many, many-to-one, and many-to-many) between the S. mansoni proteins and the other proteins included in this phylome data. If no duplication has occurred since the speciation, the two genes form a one-to-one relationship. If subsequent duplications have occurred, other types of orthology relationships (one-to-many or many-to-many) were assigned 51 . One example of this classification is provided (Figure 2).

To further analyze such large data set, we built the SchistoPhylomeSQL a Structured Query Language relational database using MySQL as a database management system. This local database integrates information from PhylomeDB (http://phylomedb.org) and SchistoDB (http://www.schistodb.net). Access to the database was obtained using DbVisualizer version 7.0.5 (http://www.dbvis.com), a graphical user interface that allows developing and accessing database management system (DBMSs) in different operating systems. The SchistoPhylomeSQL database was the main resource for data mining in this work. Perl scripts and SQL queries were implemented to parse the text files and load them to the database.

Using ETE 99 , we analyzed the S. mansoni phylome data to identify protein families that were specifically expanded in the S. mansoni lineage since its diversification from the other metazoans (Additional file 1 Table S3). The duplication events defined by the species-overlap algorithm that only comprised paralogs from S. mansoni were considered lineage-specific duplications. In cases where the information extracted from more than one phylogenetic tree contained the same paralogous proteins, changing only the “seed” protein position, the data was filtered to obtain a non-redundant list of in-paralogs.