To identify all protein kinase genes in the three trypanosomatid genomes, we searched GeneDB 29 for all genes bearing Pfam protein kinase domains, as well as by BLAST using representatives of all major protein kinase gene families, including aPKs. All ePKs were examined for the presence of the 11 characterized subdomains, and specifically for the presence of the key lysine in subdomain 2 and aspartic acids in subdomains 6 and 7. The genomic analysis revealed 179, 156, and 171 ePKs and 17, 20, 19 aPKs in L. major, T. brucei and T. cruzi respectively. These numbers suggest that phosphorylation is an important mechanism for cellular regulation in all three trypanosomatids and are considerably larger than that described for another intracellular parasite that transits diverse environments; Plasmodium falciparum. P. falciparum possesses 65 ePKs and 20 ePK-related sequences, designated FIKK 303132. The latter have not yet been shown to have protein kinase activity 30. The activation of many ePKs requires phosphorylation in the activation loop between subdomains 7 and 8. These kinases are typically marked by an RD motif within subdomain 6 33. In T. brucei, 130 of the 156 ePKs are RD kinases, further supporting the concept that phosphorylation networks are complex and important in these organisms.

We examined the relationship of the trypanosomatid ePKs to the groups and families of kinase domains of human, worm, fly, and yeast using the available datasets 34. Most ePKs had a highly significant BLAST score against at least one member of the 4-kinome dataset. For example, 58% of the T. cruzi ePKs had an E-value of at least 10^-40, and 77% had a score of at least 10^-30 against a member of this dataset (Figure 1). Based on BLAST E-values, as well as phylogenetic inference (see below), assignments to ePK groups and families were made (Table 1, Additional file 1). We generated phylogenetic trees of the kinase domains of the entire T. brucei ePK kinome, seeding the tree with human and yeast PKs to facilitate classification (Figure 2 shows the MRBAYES tree). The trees were generally consistent with the BLAST assignments (the few that did not match are marked by an asterisk in the tree). Of note are several unique kinases that are on long branches originating near the center of the tree, indicating their high divergence from other ePKs.

In general, the TriTryp kinomes are closely related. COGS are clusters of orthologous genes, as revealed by analysis of mutual BLASTP hits across the genomes. The majority (68%) of the ePK genes reside in COGS that contain members from each of the three species (Additional file 1). Conversely, only a small number of genes appear to be unique to a species (20 in L. major, 11 in T. cruzi, and 3 in T. brucei, Additional file 1). As with other genes in these organisms, members of these COGs are generally syntenic among the three species, furthering the concept that the molecules are orthologous. Figure 3 compares the representation in the major groups and families of human protein kinases with those of L. major and Table 1 shows the representation of groups and families among the TriTryps, with further details including systematic gene names provided in Additional file 1.

A key difference between host and parasite kinomes is the complete lack of ePKs that map to the tyrosine kinase (TK) and tyrosine kinase-like (TKL) groups in the trypanosomatids. Representatives of the former in humans include receptor protein kinases such as the insulin receptor and cytosolic kinases such as src. The latter group contains ePKs such as RAF1 and TGFβR2. We also found no evidence of the receptor guanylyl cyclase (RGC) group of proteins, which are structurally related to protein kinases. These groups of ePKs are also absent in malaria parasites, which further lack the STE group of kinases. Interestingly, it has been reported recently that the genome of the unicellular protist Entamoeba histolytica encodes TKs with SH2 domains, TKLs, and a large family of putative receptor serine-threonine ePKs 18. As E. histolytica also possesses genes encoding putative 7 transmembrane receptors and heterotrimeric G proteins, which the trypanosomatids lack 5, the mechanisms regulating cell signaling appear to be very different among the parasitic protozoa.

As noted above, phosphorylation on tyrosine is well documented in trypanosomatids. We propose that this activity is likely to be due to the action of atypical tyrosine kinases such as Wee1 and dual-specificity kinases that can phosphorylate serine, threonine, and tyrosine. Multiple members of the dual specificity kinase families (DYRKs, CLKs, and STE7) are present in the trypanosomatid genomes. Although Wee1 is functionally a tyrosine kinase, it most closely resembles serine/threonine kinases such as Chk1 and cAMP-dependent kinases in structure and primary amino acid sequence 35. In yeast and higher eukaryotes Wee1 phosphorylates a conserved tyrosine residue in the ATP binding pocket of CDK1 (cdc2), inactivating the protein kinase. This mechanism is likely to be conserved in the three trypanosomatids, since there are two Wee1 family members in L. major and T. cruzi and one in T. brucei. In addition, CRK3, the putative functional CDK1 homologue in trypanosomatids 363738, contains a conserved tyrosine residue in the same subdomain as the human CDK1 regulatory tyrosine 3940. In T. brucei, 18 other CMGC members also have this tyrosine within subdomain 1 (Additional file 2), reiterating the potential for widespread regulation of protein kinase activity via tyrosine phosphorylation. The potential conservation in regulatory mechanisms for CDK activity between yeast, mammals and trypanosomatids may not extend to all protozoa, as the putative P. falciparum Wee1 lacks a key active site residue suggesting it may not be active 31 and dual-specificity ePKs appear to be absent. In addition, no tyrosine phosphorylation has been demonstrated to date in that species. The existence of other unusual protein tyrosine kinases in trypanosomatids is an intriguing possibility given the large number of protein kinases in the trypanosomatid kinomes that cannot be easily placed into typical ePK groups or families (see below).

Most mammalian receptor kinases belong to the tyrosine kinase group, a group which is lacking in trypanosomatids. However, in plants, most receptor kinases are serine/threonine kinases. Bearing this in mind, we searched the T. brucei genome for genes bearing the protein kinase Pfam domain plus the annotation of a transmembrane domain. Ten candidates fit the criteria (see Additional file 3), these were spread among a variety of ePK groups, with a somewhat higher representation among the STE kinases. At this juncture, there is no evidence that any domain of these molecules is displayed on the parasite surface, where it might respond to host or parasite derived ligands. Alternatively, if surface-localized, the kinase could phosphorylate host or parasite molecules to modify their environment. We note with interest previous reports of an ectokinase with a substrate profile characteristic of CK1 in Leishmania 7778. Intriguingly, one of the L. major CK1 genes identified in this analysis encodes a protein with a predicted signal anchor sequence (LmjF17.1780). Assessing whether any parasite protein kinases interface with the host environment is an important arena for future experimental studies.

Approximately 8% of the ePKs of each species are predicted to be catalytically inactive, based on the presence of mutations in essential residues (K in subdomain 2 and D in subdomains 6 and 7). Most of these possess an orthologue in at least one other trypanosomatids. Of the 13 T. brucei ePKs predicted to be catalytically inactive, 11 are mutated to a predicted non-catalytic form in each of the three species. Genome-wide, the level of amino acid sequence identity among COG members averages 61 +/- 7% between T. brucei and T. cruzi 79, with a similar level of identity for a sampling of ePKs (60% +/- 7%). The ePKs predicted to be inactive show a lower level of identity at 44% +/- 8%. Hence, while conserved, these sequences are somewhat more divergent across species.

We also estimated the synonymous (Ks) and nonsynonymous nucleotide (Ka) nucleotide substitution rates in T. brucei versus T. cruzi genes encoding ePKs predicted to be catalytically active or inactive. The Ka/Ks ratio (sometimes designated as dN/dS) can reflect the selective constraints on a gene. Ka/Ks = 1 is expected for genes evolving neutrally. Ka/Ks < 1 is thought to indicate selection to remove amino acid replacements. In the rare cases where the Ka/Ks > 1, selection for amino acid divergence is usually invoked. For a random subset of 90 active ePKs, Ka = 0.336, Ks = 4.822 and Ka/Ks = 0.077. For the inactive ePKs, these figures were 0.535, 9.639, and 0.110, respectively. These data indicate that in both sets synonymous mutations are highly preferred. Nonetheless, the Ka and Ka/Ks were significantly different in the active versus inactive datasets (p = 0.0003 and p = 0.0045, Mann-Whitney U-test, two tailed). There was no statistically significant difference in the calculated Ks scores between these two datasets (p = 0.3). These findings suggest that the encoded proteins continue to play a significant functional role within the organisms, although the predicted lack of catalytic activity indicates this role is likely to be via a distinct mechanism, such as regulation via protein-protein interaction. Indeed, a recent analysis has shown that inactive protein kinases are not an exception in metazoa and that a few have evolved novel functions, some of which might be involved in processes that enhance the complexity of regulatory phosphorylation networks 80.

A characteristic of human ePKs is the presence of accessory domains. Indeed, over 50% of human protein kinases have additional Pfam domains, and more are found when criteria are relaxed 21. We examined all of the trypanosomatid ePKs for significant matches to additional Pfam domains (Table 2). In the case of L. major, only 25 ePKs possessed additional Pfam domains that met the default cutoff. Three additional ePKs, which had orthologues in T. brucei or T. cruzi that had a significant Pfam domain, were found to possess partial motifs, and others had domains of lower significance. The accessory domains were generally conserved among members of a COG. Notably four of the five most common Pfam domains on human ePKs are absent in the trypanosomatid kinome: Ig, fn3, SH2, and SH3. Ig and fn3 domains are generally extracellular domains that interact with ligands, so their absence may not be surprising given the paucity (or absence) of receptor ePKs in trypanosomatids. SH2 domains interact with phosphotyrosine, and their absence in the trypanosomatid genomes could suggest a co-evolution with dedicated tyrosine kinases. SH3 domains bind to proline rich sequences.

Several unusual domain combinations are found in trypanosomatid protein kinases (Table 2, examples shown in Figure 8). A search of eight eukaryotic kinomes using the KinG Kinases in Genomes Resource 81 revealed that some accessory domains found in trypanosomatids that were not associated with ePK catalytic domains in other species. For example, LmjF35.4000, which is a Leishmania-specific gene, contains a unqiue ePK catalytic domain, along with three TPR motifs (which are present on some plant receptor kinases) along with a domain associated with ubiquitin transferase (HECT), a domain not seen in the sampled genomes. A T. cruzi ePK, Tc00.1047053511727.210, has a TPR motif and HAMP domain (found on various signaling proteins including histidine kinases), both of which are recognizable on the T. brucei orthologue, but not on the L. major orthologue. Others domains are associated with ePKs of different classification. For example, the very large STE kinase LmjF15.1200, has an unusual juxtaposition of two domains related to cyclic nucleotide binding and a PAS domain, associated with signal sensing. In other species, the PAS domain is found on CAMK and TKL kinases, while the cNMP domain is restricted to AGC kinases. The L. major domain structure is likely conserved in T. cruzi, although the coding region is interrupted by a contig break. No orthologue is present in T. brucei. Some other accessory domains are found on similar groups of kinases (RWD, PX). Despite the paucity of identified domains, the trypanosomatid ePKs are generally considerably larger than the 250 aa kinase domain. For example, half of T. cruzi ePKs are larger than 64 kDa, and 38 are larger than 100 kDa.

The parasites possess a complement of atypical protein kinases, including representatives of all of the more well-characterized families: RIO, alpha, PIKK and PDK (Figure 9 and Additional file 4), although no functional analyses have been carried out to date on any representative aPK from trypanosomatids. The RIO family of atypical kinases is related to ePKs, but RIO proteins lack the sequences known to be involved in peptide binding in ePKs 82. Nonetheless, the catalytic residues are present. Trypanosomatids possess two RIO proteins, which are clearly assigned to the RIO1 and RIO2 subfamilies. In other organisms both RIO1 and RIO2 are required for ribosomal biogenesis, and RIO is involved in cell cycle progression 26. Interestingly, the similarity between human and trypanosomatid RIO2 extends into the N-terminus, where the structure of the human enzyme shows a winged helix-turn-helix motif 82. Such helix-turn-helix motifs are often found on DNA binding proteins such as transcription factors, a class of proteins which are rare in trypanosomatids.

The alpha kinases, so named because they phosphorylate their substrates within alpha helices, show a small amount of sequence similarity to ePKs, with conservation of the catalytic residues in subdomains 2, 6, and 7 27. One set of the alpha kinases in trypanosomatids is comprised of small molecules, being little more than the 241 aa alpha domain. This type of alpha kinase is present in all three species. However, L. major possesses two additional alpha kinase genes, which are very large (>1000 amino acids). Interestingly, three of the four L. major alpha kinases are found on a 12 kb segment on chromosome 36. None of the alpha kinases appear to be fused to an ion channel, as is the case for certain vertebrate alpha kinases 27.

The PIKK kinases represent a particularly interesting family in which the protein kinase domain structurally resembles that of phosphatidylinositol 3-kinases 25. In addition to the kinase domain, these proteins also have FAT and FATC motifs which are not found in the lipid kinases. The similarity between the trypanosomatid PIKK kinases and those in the 4-kinome dataset is highly significant, with E-values of 10^-90 or better. PIKK kinases are quite large in general and those in T. brucei are no exception, ranging in predicted size from 271 to 468 kDa. The parasites possess clear homologues to the specific PIKK kinases involved in genome surveillance: ATM and ATR 83. They also have four kinases that belong to the FRAP family (this family includes FRAP and mTOR). TOR (target of the immunosuppressive agent rapamycin) modulates translation and cell cycle in response to nutrient and growth signals 84. Multiple drugs targeting mTOR are in trials for the treatment of various cancers 84.

The trypanosomatids contain 3 genes encoding putative pyruvate dehydrogenase kinases (PDK). In mammals, the activity of mitochondrial pyruvate dehydrogenase is tightly regulated by multi-site serine phosphorylation of the E1α subunit 85. Despite their exclusive phosphorylation of serine residues, the PDKs lack the domains characteristic of ePKs. Rather, these kinases have two distinct domains. A C-terminal domain that shares structural conservation with the GHKL ATPase/kinase superfamily (including members of the histidine kinase family) and an N-terminal domain that resembles a histidine phosphoryl transfer domain of bacterial two component systems 86. The presence of an active pyruvate dehydrogenase in trypanosomatids with an E1α subunit suggests that regulation of activity by phosphorylation is likely to be conserved in these species.

All ePKs were retrieved from GeneDB 29 through a combination of searches with the protein kinase Pfam domain, BLAST analysis using diverse ePKs, and examination of COGS. L. major (version 5.0, Feb 2005); T. brucei (version 4.0, Feb 2005) and T. cruzi (version 3.0, July 2004) were the final datasets used in this study. In the case of T. cruzi, in which the genome strain is a hybrid, the two presumed alleles were identified through analysis of COGs 79, and counted as one gene, even though up to 7% sequence allelic sequence divergence occurs in this strain 5. All ePKs were examined for the presence of the 11-subdomain structure, and the presence of lysine in subdomain 2 and aspartic acid in subdomains 6 and 7, which are required for catalysis 2489. Those lacking these residues were categorized as catalytically inactive. Similarly, a few ePKs lacked any sequence resembling subdomain 1, which functions in ATP binding, and were also categorized as catalytically inactive. Atypical PKs were identified by BLAST analysis using representatives of each group of aPKs from other species as queries.

Each ePK was analyzed for the presence of additional domains by hidden Markov model analysis of the Pfam database 90. The default cutoffs were used.

All predicted ePKs from each species were analyzed by BLAST analysis against the 4-kinome dataset comprised of all human, Saccharomyces cerevisiae, Drosophila melanogaster, and Caenorhabditis elegans protein kinase domains 34. Since phylogenetic inference indicated that members of trypanosomatid COGs were more closely related to each other than to ePKs of the 4-kinome dataset, all COG members were classified congruently (the sole exception we found was two kinases that shared extensive homology outside of the kinase domain, LmjF15.1200 and Tc00.1047053505977.13). Assignments required an E-value difference of 5 logs or more between groups or families of ePKs for one of the trypanosomatid orthologues. When all trypanosomatid orthologues had E-values poorer that 10^-16, or had similar E-values for different groups of ePKs, those ePKs were designated "unique".

Due to the relatively low degree of sequence conservation at the nucleotide level within some of these families, phylogenetic inference was carried out on amino acid alignment, rather than attempting alignment at the nucleotide level. Kinase domains were identified by analysis of alignments with the Pfam protein kinase domain, and extended manually as needed. Insertions larger than 25 amino acids were identified and removed prior to subsequent analyses. The SAM (Sequence Alignment and Modeling System using Hidden Markov Model (HMM) software was used to build HMMs representing the kinase domains of the gene families discussed in this paper 91. These trained models were then used to identify residues capable of discriminating between the various domain families. In addition, by aligning the sequences of the kinase domains to these models, we created multiple sequence alignments of these gene families. These alignments were then visually inspected to verify that all subdomains were appropriately aligned, as well as to allow removal of both gene-specific insertions (in addition to those previously removed) and deletions and phylogenetically uninformative residues. These edited HMM-generated alignments were used as the starting point for phylogenetic reconstruction of these domain families.

Phylogenetic analysis of the kinase domains of these proteins was carried out using a variety of techniques. As a preliminary step in the phylogenetic investigation of our dataset, we used the Neighbour-joining approach as defined by Saitou and Nei as implemented in ClustalX 92. Due to the relatively high degree of divergence that might be expected within our dataset we used the correct for multiple substitutions option in our analysis. Bootstrapping was carried out on the dataset with 1,000 replicates.

These aligned amino acid sequences were also subjected to parsimony analysis using PAUP*, version 4.0b8 9394. Given the relatively large number of taxa in this dataset, the use of an exhaustive search was not possible. In its place, an heuristic search strategy was employed to attempt to find the best tree by reducing the set of trees examined and just calculating the score for likely trees. It should be noted that this method is not guaranteed to identify the most parsimonious trees from the sequences. We carried out 100 random stepwise addition sequences of taxa, each with TBR swapping and MAXTREES set to 10,000. Parsimony bootstrapping on the dataset was performed with 1,000 replicates and the same settings, except that only 10 random stepwise addition sequences were used per bootstrap replicate.

The same datasets were analysed using a Maximum Likelihood approach, as implemented in the web available PHYML application 9596. Analysis using the WAG amino acid substitution model inferred the starting tree. The proportion of invariable sites and the gamma distribution parameter were estimated by maximizing the likelihood of the phylogeny. The number of substitution rate categories was set at four for these analyses. Non-parametric bootstrap analysis was then carried out on the original data set with 500 replicates (the upper limit available for this web service). Majority-rule consensus trees were created for each of the three methodologies outlined above.

Finally, MRBAYES was used to carry out a Bayesian analysis of our data 97. We used the WAG amino acid substitution model (based on nuclear genes and globular protein sequences, respectively), with a gamma rate distribution estimated from the data set to infer the phylogeny of our dataset. Starting from random trees, four parallel Markov chains were run to sample trees using the Markov Chain Monte Carlo (MCMC) principle. In general, 1,000,000 generations were run; after the burn-in phase, every 100^th tree was saved.

Analysis of the synonymous/non synonymous substitution rates required construction of pairwise, codon aligned, sequence alignments. These were generated using the PAL2NAL web server 98, which converted full-length amino acid alignments and the corresponding DNA sequences into a codon-based DNA alignment. The amino acid sequence alignments were obtained using the Water programme from the EMBOSS package using default parameters 99. Estimation of Ka/Ks (dN/dS) ratios was then carried out by maximum likelihood using the pairwise codon-based substitution model in Codeml, which is part of the Phylogenetic Analysis by Maximum Likelihood (PAML) suite of programs 100.