Using the 28 experimentally identified phosphorylated motif sequences in Table 1, we built the sequence logo as shown in Figure 1. In this logo, the fourth position of the EPIYA motif is always tyrosine (Y), which can be phosphorylated. The first and third positions have small variations. The amino acids in the first position are primarily glutamic acid (E), together with asparagine (N). Most residues in the third position are isoleucine (I) and leucine (L), two very similar amino acids. The second and fifth positions have big variations. The second position varies from proline (P), serine (S) to asparagine (N). The fifth position mainly contains alanine (A), glutamic acid (E) and aspartic acid (D). These 28 sequences were used to build and calibrate the HMM by applying Hmmer 2.3.2 (http://hmmer.janelia.org). We then employed the HMM to search the protein non-redundant (NR) database, which contains 9,216,047 protein sequences. The search yielded 107,231 sequences containing at least one copy of EPIYA motif and 3115 sequences with multiple repeats of the EPIYA motif, where the highest number of repeats in a single protein is 29 (see Table 2).

We found that the repeats of EPIYA motif in a protein are highly non-random. As the probability of one protein sequence having a copy of EPIYA motif is 1.13E-02 (104,116/9,216,047), the expected probabilities of one protein sequence containing 24 copies of EPIYA motif are (1.13E-02)²=1.28E-04, (1.13E-02)³=1.44E-06, and (1.13E-02)⁴=1.63E-08, respectively, assuming the combination of the motif in a sequence is random. The observed probabilities of one sequence containing multiple copies of EPIYA motif are much larger than the expected probabilities as shown in Table 2. Hence, the repeats of the EPIYA motif may have been resulted from evolution with biological significance. This is also reflected in Table 1, where most effectors with known EPIYA motif have 2-7 motif repeats. Thus, we suggest that multiple copies of EPIYA motif in the same protein are more likely to be functional than single motif occurrence.

The NR database contains proteins sequences from 27,432 genuses and 121,718 species. Among them the sequences from 2675 genuses and 4646 species contain at least one copy of EPIYA motif, and 368 genuses and 587 species have proteins containing at least two copies of EPIYA motif. The proteins with the EPIYA motif are mainly distributed in lower organisms. As shown in Table 3, the probability of a genus/species containing proteins with the EPIYA motif in archaea, viruses or bacteria is much higher than that in eukaryotes. This indicates that with evolution advanced species mostly lost the EPIYA motif together with its functions for host- pathogen interactions.

We listed top 10 species and genuses with most EPIYA-motif containing proteins for the groups of archaea, viruses, bacteria, protista, fungi, metazoa and viridiplantae (Additional File 1). In archaea, Methanococcus is the genus that includes the most EPIYA-motif containing proteins. In viruses, Potyvirus is the highest in number of EPIYA-motif containing proteins among genuses while Bovine Viral Diarrhea Virus is the highest among species. The top four genuses (and the corresponding species) in bacteria are Helicobacter (Helicobacter pylori), Clostridum (Clostridum botulinum, Clostridum perfringens), Bacillus (Bacillus cereus) and Anaplasma (Anaplasma phagocytophilum). Plasmodium (Plasmodium falciparum) and Tetrahymena (Tetrahymena thermophila) are the top genuses in protista. In fungi and viridiplantae, the corresponding top genuses are Candida (Candida tropicalis) and Oryza (Oryza sativa), respectively. Two well-studied genuses Drosophila (Drosophila melanogaster) and Homo (Homo sapiens) take the top two in metazoa. It should be noted that the data in Additional File 1 are biased, with widely studied species such as Helicobacter pylori having the same gene sequenced many times, while some other species have incomplete proteomes. Nevertheless, this table in Additional File 1 provides some interesting reference for known and putative pathogens with effectors.

Bacterial pathogens can be divided to two types. Some can enter host cells, e.g., Chlamydia and Anaplasma. They are known as intracellular pathogens and most of them have T3SS or T4SS. Some other bacteria are extracellular pathogens with T3SS or T4SS, such as Helicobacter pylori and Campylobacter. As shown in Additional File 1, numerous bacteria containing multiple copies of EPIYA motif belong to pathogens, such as Anaplasma (Anaplasma phagocytophilumn, ranking 4th in the species list) and Chlamydia (Chlamydia trachomatis, ranking 10th in the species list), both of which are intracellular pathogens. Helicobacter (ranking 1st in both genus and species list) and Campylobacter (ranking 10th in the genus list) contain T4SS 18 . Some other T4SS-containing species are not listed in Additional File 1, such as Wolbachia (ranking 13th in the genus list), Escherichia (ranking 18th in the genus list), Mycobacterium (ranking 23rd in the genus list) and Bartonella (ranking 27th in the genus list). Furthermore, we also found that in protista most top 10 species are unicellular parasites that can live in host cells to survive and reproduce by subverting of signalling pathways and inhibiting apoptosis of host cells 19 . However, the pathogens mediators responsible for this modulation are still unknown 20 . Those intracellular protozoan parasites include Plasmodium (ranking 1st in the genus list), Leishmania, Trichomonas, Cryptosporidium and Giardia (correspondingly ranking 58th in the genus list). Table 4 lists most known pathogens including intracellular bacteria (Mycobacteriaceae, Legionellales, Chlamydiales, Rickettsiales and Listeriaceae), extracellular bacteria with T3SS or T4SS (Enterobacteriaceae, Campylobacterales and Rhizobiales) and intracellular protozoan parasites (Apicomplexa and Kinetoplastida), representing 1319 species in total with and without the EPIYA motif. We analyzed the distribution of EPIYA motif in the potential effectors in these pathogens in Figure 2. 310 out of 4646 species with the EPYIA motif belong to such pathogens, and the percentage (310/4646=6.67%) is much higher than that of such pathogens (with and without the EPIYA motif) in all species (1319/121,718=1.08%) (p-value<0.0001, odds ratio=6.54). We also found that the percentage of species with the EPYIA motif belonging to pathogens increases significantly with the increase of EPIYA motif repeats in a protein sequence.

In our search result, there are totally 3115 protein sequences with at least two EPIYA motif repeats. Among them, most are CagA of Helicobacter pylori as this protein has been extensively studied, and 689 out of 3115 are hypothetical proteins whose functions have not been identified. Based on protein functions, the top 40 proteins ranking by occurrence under each protein function type are widely distributed. They not only exist in achaea, viruses and bacteria, but also are found in protista, metazoa and viridiplantae. Besides the known EPIYA-motif containing effectors ankryin and Tarp, they also involve enzymes related to DNA, ATP and tRNA, transcription regulators, tumor suppressors, different types of kinases, zinc-finger proteins, ubiquitin and various metabolic enzymes (Table 5).

Although many of these predicted effectors are false positives and the EPIYA motif may not be functional in them, a significant portion of them is likely to be true effectors. As known effectors, ankyrin and TPR (tetratricopeptide repeat) are related to protein-protein interaction 2122 . Considering the sequence similarity of the above proteins, 44 sequences of ankryin are highly similar among each other and come exclusively from Anaplasma phagocytophilum, Wolbachia endosymbiont and Ehrlichia sp., all of which belong to Rickettsiales. Except the sequences from Haliangium ochraceum (ZP_03879805 and ZP_03880192), other sequences of TPR repeat-containing proteins are also similar and they are from Trichodesmium erythraeum, Stigmatella aurantiaca, Acaryochloris marina, Cyanothece sp. and Microcoleus chthonoplastese. For the 20 hypothetical proteins, YP_034066 and YP_001610012 (Bartonella), YP_153762 and YP_002563468 (Anaplasma), XP_001623017 and XP_001636029 (Nematostella), ZP_01620341 and ZP_01622571 (Lyngbya), XP_001468598 and XP_001686356 (Leishmania) are similar pairs in sequences (with more than 30% sequence identity in each pair), and two proteins in a pair are from the same genus. The EPIYA motif in these proteins is highly conserved during evolution, and it may play similar roles as the motif in CagA.

Among proteins containing at least four copies of EPIYA motif (Additional File 2) with 286 sequences in total, most of them are from bacteria, especially from intracellular bacterial pathogens or extracellular bacterial pathogens with T3SS or T4SS, and some are from protist, e.g., intracellular protozoan parasites. Four out of eight known effectors (CagA, Ankyrin, BepD, and Tarp) are found in these sequences, and thus other proteins from bacteria and protista in Additional File 2 may also be effectors. An interesting observation is that the percentage of protein sequences having the EPIYA motif in archaea is the highest among all groups (see Table 3), but none of these archaeal proteins contain four or more copies of EPIYA motif. Previous studies revealed that CagA sequences with more EPIYA-motif occurrences are more virulent 6 . Since archaea and other organisms have relationships of either mutualism or commensalsim and till now there is no clear evidence for the existence of archea parasites 2324 , it is unlikely that the archaeal proteins containing the EPIYA motif act as pathogen effectors. Compared to other groups, archaea is not well studied, but we can still find some interesting examples, such as Methanobrevibacter smithii (ranking 5th in the species list in Additional File 1). It is the most common commensal archaea in the human gut and plays an important role in digesting polysaccharides, while it may not benefit the host directly. We speculate that EPIYA- motif containing proteins of Methanobrevibacter smithii may have some biological functions in this commensal interaction.

Many functions listed in Table 5 may reflect the fact that these proteins may have multiple functions other than phosphorylation-induced signalling control in the host cell. Some of these proteins may mimic host protein functions. For example, it was suggested that CagA functions as a prokaryotic mimic of the eukaryotic Grb2-associated binder (Gab) adaptor protein 25 . Some of the predicted effectors may mimic singling proteins, such as HPK (histidine protein kinase) listed in Table 5, which is an important part of two-component signal transduction system that recognizes and transmits environmental signals 26 . Some known effectors induce protein expressions with increased expression of RNA polymerase 27 . It is not surprising to see a significant number of proteins in Table 5 are related to protein synthesis, such as RNA polymerase, elongation factor, and helicase. It is noted that CagA itself also contains an RNA polymerase domain based on a BLAST search. These connections also suggest the ancestor proteins of the predicted effectors. Many of the proteins listed in Table 5 are ancient house-keeping genes. The predicted effectors might have evolved from these house-keeping genes by mimicking the host genes. Furthermore, over evolution some of the effectors or their ancestors might have evolved into genes with different functions unrelated to host-pathogen interactions, such as EPIYA-motif containing proteins in archaea and metazoan.

(1) Building HMM based on KK, R4, Tarp and Tir motifs It is known that biological effects of CagA induced by phosphorylation depend on the binding to the SH2 domain. Different combinations of the five amino acids after phosphorylated tyrosine (pY) will bind different SH2 domains and cause different downstream effects. There are two known motifs in CagA that could bind to SH2 domain - EPIYAKVNK and EPIYATIDD(F), which are referred to as KK motif and R4 motif 2829 , respectively. For the Tir protein, we retrieved all sequences in Escherichia coli and Citrobacter rodentium with the pattern of EHIYDEVAA(P) and built a motif (named as Tir motif). We did the same for Tarp protein, yielding the motif ENIYENIYE (named as Tarp motif). We extracted sequences of known KK, R4, Tir and Tarp motifs in proteins CagA, Tir and Tarp (including their variants), and then built HMMs one by one. Figure 3 shows the sequence logo of each motif. Comparing with the sequence logo in Figure 1, the 9-mer motifs are more conserved and specific. This will help reduce the false positive rate in identifying putative effectors, while the downstream SH2 binding partners are also predicted at the same time.

(2) Search results by using HMMs based on KK, R4, Tarp, and Tir motifs Using HMMs based on KK, R4, Tarp and Tir motifs to search the protein sequences containing the EPIYA motif as described above, we found that the results are widespread in many species. In this paper we only focus on the results in bacteria and protista. As shown in Table 6, CagA KK (EPIYAKVNK) motif exists in some known phosphorylation effectors, e.g., Beps (BepD, BepE, BepH) and Tir. CagA R4 (EPIYATIDD) motif exists in Tarp. Both KK and R4 motifs exist in ankyrin. Among 8 proteins for building our EPIYA-motif based HMM, BepF is the only one containing neither KK nor R4 motif. The Tir protein just have one motif -Tir motif in Table 1, while BAF52548 (Tir of E. coli) contains two motifs, i.e., Tir (EHIYDEVAA) motif and EPIYAKIQR, similar to the KK motif. Tarp protein (YP_001654788) of Chlamydia trachomatis contains not only the Tarp motif (ENIYENIYE), but also another motif ENIYESIDD, which is similar to the R4 motif. It reveals that although these proteins are not similar in global sequences (weak similarity exists between Beps sequences), they share the same or similar motifs with significant functional relationships.

Based on the above HMMs of KK, R4, Tir and Tarp motifs, we predicted some new pathogen effectors (Additional File 3) and we assessed them based on the literature. The details of predicted effectors have been listed in Additional File 4, 5, 6, 7.

(1) Bartonella tribocorum: Since BepH contains the EPLYAQVNK (YP_001610013, Y-8) motif (KK motif), we predicted it as a phosphorylation effector like BepD-F secretory proteins.

(2) Lawsonia intracellularis: Lawsonia intracellularis is an obligate intracellular bacterial pathogen, which infects a wide range of animals, mainly pigs, and causes proliferative enteropathy - a type of contagious diseases 3031 . Its symptoms are acute, including diarrhea, loss of appetite and stunting. After an initial close association with the cell membrane of the enterocytes, Lawsonia intracellularis is endocytosed into host cell 32 . Infected host cells are inhibited in maturation, continue to undergo mitosis and proliferation, and at last form hyperplastic crypts, but the mechanism is unknown 33 . The genome sequence of Lawsonia intracellularis indicates that it may possess a type III secretion system, which may assist the bacterium during cell invasion and evasion of the host's immune system and could be a mechanism for inducing cellular proliferation 3435 , but its effectors secreted by T3SS was never reported. Current database contains 20 proteins of Lawsonia intracellularis with the EPIYA motif and all of them are from strain PHE/MN1-00. The maximum sequence identity between any two of these 20 proteins is 22% and most of them are enzymes, e.g. ATP synthase. Among them, in the HMM search result by using the R4 motif, we found that hypothetical protein L10666 (YP_595041) contains two copies of EPIYA motif (EPIYAEIKT Y-149, EPIYAEIKT Y-186), which are similar to the R4 and Tir motifs, respectively. Thus, we speculate that this protein might be the effector of Lawsonia intracellularis to interact with intestinal epithelial cells.

(3) Ehrlichia sp.: It belongs to the same family Ehrilichiaceae as Anaplasma36 . Ankyrin of Ehrlichia sp. and ankyrin of Anaplasma share 89% sequence identity. Ankyrin (T08612) of Ehrlichia sp. contains six copies of 9-mer motifs including the KK and R4 motifs, and thus it is a likely effector of Ehlichia to interact with host.

(4) Wolbachia: Wolbachia belongs to Rickettsiales. Wolbachia is a symbiotic bacterium existing in the sex organ of many insects. Though ankyrin (AAY54257) of Wolbachia and ankryin of Anaplasma share only 15% sequence identity, they contain almost exactly the same motifs. Hypothetical protein WD0942 (NP_966676), which is not similar to ankyrin in sequence, has two motifs, and one of them is EPIYATVPK(Y-318) similar the KK motif. EsorChan1 (AAP34173) contains the motif EPIYDEVYD (Y-77) similar to the Tir motif. Therefore, the above three proteins, especially the first two, are potential effectors of Wolbachia37 .

(5) Pasterurella multocida: As the major pathogen to cause swine infectious atrophic rhinitis, it secretes toxin filamentous hemagglutinin containing six copies of EPIYA motif. Based on the BLAST search results, we found that the filamentous hemagglutinin (AAK61595) of Pasterurella multocida, filamentous hemagglutinin of Bordetella pertussis and Bordetella Parapertussis share ~30% sequence identity [3840]. Filamentous hemagglutinin, the major virulence factor of Bordetella pertussis, not only has adhesion function, but also plays a critical role in immunomodulation. Since filamentous hemagglutinin has the sequences EDIYATINK (Y-2792), which is similar to the KK motif, EHIYADIRD (Y-2550) and ENLYAEISD (Y-2651), both of which are similar to the R4 motif, and EHLYAEINE (Y-2387), which is similar to the Tir motif, we suggest that filamentous hemagglutinin being the effector of Pasterurella multocida and it might be secreted by the TPS (Two-Partner Secretion) system 39 . PfhB2 (NP_244996) has four sequences that are similar to KK, R4 and Tir motifs, and thus it might be another candidate of effector in Pasterurella multocida.

(6) Haemophilus ducreyi: Haemophilus ducreyi is a facultative anaerobic Gram- negative coccobacillus and could cause the sexually transmitted disease chancroid. Large supernatant protein2 (NP_873623) of Haemophilus ducreyi has six copies of EPIYA motif. Its sequence and filamentous hemagglutinin of Bordetella pertussis share 41% sequence identity. Its sequences EPVYADLHF and EPVYADLRF are similar to the R4 motif. Hence, we suggest large supernatant protein2 (NP_873623) is a potential effector of Haemophilus ducreyi and it could be secreted by T4SS 41 . The effector can lead to immunosuppression, inhibition of proliferation, and permanent changes in host cells 424344 .

(7) Haemophilus somnus: Haemophilus somnus can survive in host cells and is the cause of a variety of systemic diseases in cattle, including thrombotic-meningoencephalitis, pneumonia, arthritis, myocarditis, septicemia and other reproductive diseases 4546 . Cysteine protease domain YopT-type (YP_001784809) and filamentous hemagglutinin of Bordetella pertussis share 42% sequence identity. The sequence EPIYATLDK (Y-2933) in YP_001784809 is similar to the KK motif, EHIYEQIGE (Y-2358) similar to the Tarp motif, and EPVYDKVSA (Y-2287) similar to the Tir motif. Thus, YP_001784809 might be the effector of Haemophilus somnus to cause immunosuppression 47 .

(8) Chlamydophila pneumonia: Hypothetical protein CPj0472 (NP_300527) contains three copies of EPIYA motif. EPIYANTPE (Y-647) is similar to the KK motif, EPIYEEIGG (Y-346) is similar to the Tir motif and EPIYDEIPW (Y-681) is similar to the R4 motif. Although we did not find any similar protein through BLAST search, hypothetical protein CPj0472 (NP_300527) is a good candidate for the effector of Chlamydophila pneumonia.

(9) Leishmania major: Leishmania major could parasitize into phagocyte of human or other mammals and is responsible for the disease leishmaniasis, which is a serious zoonosis. Leishmania major have 6 proteins containing at least two copies of EPIYA motif and 4 of them are proteins with unknown functions. Among these 6 proteins, Cytochrome C oxidase subunit VI (XP_001683136) contains two copies of EPIYA motif. One is at position Y-107 with sequence EPLYQPVKK, which is similar to the KK motif. Another one is at position Y-130 with sequence EPLYDVDAA, which is similar to the Tir motif. Hence, XP_001683136 might be an effector. Hypothetical protein (XP_001686159) has three copies of EPIYA motif and the sequences are all EPLYAVTIE, which is similar to KK and R4 motifs. Hypothetical protein (XP_001686160) also has three copies of EPIYA motif and the sequences are all EPLYAVTID, which is similar to the R4 motif. In addition, hypothetical protein XP_001686159 and XP_001686160 share 43% sequence identity. Hypothetical protein (XP_001686356) has 29 copies EPIYA motif (the one with most EPIYA motifs in our data) and all sequences are the same as EPLYAVTLE, which is similar to the R4 motif. Microtubule-associated protein (XP_001687515) contains two copies of EPIYA motif. One is at Y-1543 and another is at Y-1589. The sequences for both of them are ESIYAKDYK, which is similar to the KK motif. Thus, we predict hypothetical protein (XP_001686159), hypothetical protein (XP_001686160), hypothetical protein (XP_001686356) and Microtubule-associated protein (XP_001687515) might also be the effectors of Leishmania major. For another potential effector hypothetical protein (XP_001683914), although it contains two copies of EPIYA motif (ESLYE is at Y-1006 and EHLYD is at Y-1047), they are not similar to KK, R4, Tarp or Tir motif and hence less likely to be an effector than the above five proteins.

(10) Plasmodium falciparum: Plasmodium falciparum can invade human liver cells and RBC to cause dangerous infection malaria. It contains many proteins with the EPIYA motif and 47 proteins with at least two copies of EPYIA motif. Among them, Plasmodium exported protein (XP_001347309) has three copies of motif which are all similar to the Tarp motif. The sequences and the corresponding pY sites are ESIYKNKLK (Y-331), ESIYKNKLK (Y-359), and ESIYKNKLE (Y-387). Thus we predict it as the effector of Plasmodium falciparum. Conserved Plasmodium protein (XP_001347469) has eight copies of EPIYA motif, RNA pseudouridylate synthase (XP_001350676) has nine copies of EPIYA motifs and hypothetical protein (XP_001351018) has three copies of EPIYA motif, but none of them contains any of KK, R4, Tir and Tarp motifs, and therefore is less likely to be the effector than Plasmodium exported protein (XP_001347309).

We applied the subcellular localization prediction for all the predicted bacterial effectors above by using CELLO v.2.5 48 (http://cello.life.nctu.edu.tw). All the associated bacteria are “gram negative”. As a real effector should be secreted from a gram-negative bacterium and then enter a eukaryote host, we perform subcellular localization by using both gram-negative bacterium and eukaryote, respectively. When using gram-negative bacteria as hosting species, 9 out 11 effectors were predicted as extracellular or outer-membrane (Additional File 8). When reapplying the prediction by choosing eukaryotes as the organisms, all 11 effectors are predicted to have nuclear localization (Additional File 8). The above results show that most of our predicted effectors have expected localization attributes as effectors, which provides some supporting evidence for our effector predictions.

1. Protein sequence data: We used the NR (non-redundant) protein database at the National Center for Biotechnology Information (NCBI) in this study. All protein sequences in the FASTA format were downloaded from the NCBI site (ftp://ftp.ncbi.nih.gov/blast/db/FASTA/nr.gz; as of July 6^th 2009; 9,216,047 sequences). We excluded “other” sequences and “unclassified” sequences” in the database (as labelled in http://www.ncbi.nlm.nih.gov/Taxonomy/Browser/wwwtax.cgi?mode=Root).

2. Taxonomy data: The taxonomy data was obtained from the NCBI website (http://www.ncbi.nlm.nih.gov/Taxonomy/txstat.cgi; as of July 6^th 2009).

A hidden Markov model was built by using Hmmer 2.3.2 49 (http://hmmer.janelia.org). We used selected sequences to run the command hmmbuild.exe for building and calibrating the HMM. We then used the HMM to run the command hmmsearch.exe for searching protein sequences. We used a natural cutoff of HMM score such that the last of the all known motifs is retrieved.

We used Perl (release ActivePerl 5.8.8) as the programming language to analyse the data and build the database. We applied SAS 9.0 (http://www.sas.com) as the statistical analysis tool and chose p<0.01 as the significant threshold.

BioEdit 7.0 (http://www.mbio.ncsu.edu/BioEdit/bioedit.html), Lasergene 7 (http://www.dnastar.com/products/lasergene.php), and Blast 50 (http://blast.ncbi.nlm.nih.gov/Blast.cgi) were used to compare and analyse the protein sequences. Sequence logos were constructed using Weblogo 51 .