The reversible phosphorylation of proteins represents a ubiquitous regulatory mechanism for diverse pathways and systems in eukaryotic cells. The process is controlled by a balance between the antagonistic activities of protein kinases, which catalyse the phosphorylation of serine, threonine or tyrosine residues predominantly (reviewed in 12), and more marginally of other residues, notably histidine 34, and those of protein phosphatases, which cleave the monophosphate esters from the phosphorylated form of the same residues (reviewed in 456). A large range of kinases have been identified, which seem to have arisen by multiple gene duplication events with subsequent selection 7. In contrast the number of different protein phosphatase catalytic subunits is much lower than that of kinases, and phosphatases are in general less discriminating than most kinases in substrate selectivity. This lack of specificity combined with high catalytic efficiency suggest that a 'naked' protein phosphatase activity is potentially toxic 6. The specificity and regulation of many of these enzymes is in fact mediated by accessory proteins (the phosphatase regulatory subunits), a wide variety of which interact with the relatively small repertoire of catalytic subunits (this is not the case for the PTP group, see below). As a consequence, it is speculated that the total number of protein phosphatase holoenzymes involved in regulatory pathways matches, or even exceeds the protein kinase repertoire 8910. There are four broad families of protein phosphatases with distinct evolutionary histories:

1. The PPP group. PPP sequences (

P

P

P

Furthermore, a family of bacterial-like PPP sequences found in eukaryotes (including in P. falciparum) has recently been described 17. Whereas three highly conserved motifs (GDXHG, GDXXDRG and GNH [E/D]) mediating metal coordination in the active centre are considered as the signature of the PPP family, sequences showing no similarities to the known PPP phosphatases beyond the presence of the GDXHG and GDXXDRG motifs were identified in Plants, Plasmodium, Trypanosoma and some fungi. This revealed the existence in eukaryotes of "non-conventional" branches of the PPP family (reviewed in 17).

2. The PPM/PP2C group comprises a highly diverse, evolutionarily recent set of enzymes with Mg²⁺ or Mn²⁺-dependent serine/threonine phosphatase activities. The active forms appear to be highly diverse monomeric polypeptides which in many cases possess regulatory domains in C- or N-terminal extensions. A number of defined motifs and conserved residues relate to binding of activating metal ions, water and phosphate groups, as in the PPP type enzymes, but there is no discernable sequence homology between the two groups, despite remarkable structural similarity 18. A major part of the functions of PP2C (PPM) activities in a variety of species appears to be to modulate stress responses 519. PPM enzymes form part of a superfamily that includes bacterial forms (SpoIIE) and a mitochondrial pyruvate dehydrogenase phosphatase. The PPM family of protein phosphatases is greatly expanded in plants 19.

3. The PTP (Protein Tyrosine Phosphatase) superfamily, which is subdivided into three main families: the tyrosine-specific phosphatases, the dual-specificity PTPs (which include the cdc25-like, the Ccdc14 and the MAPK phosphatase groups 202122), and the low molecular weight phosphatases 23. The tyrosine and dual-specificity phosphatases are involved in signalling, cell growth and differentiation, and in the control of cell cycle progression (for example, cdc25 is a major regulator of cyclin-dependent kinase activity 20, and cdc14 regulates mitosis exit by dephosphorylating CDK targets 22). The enzymes share a common catalytic mechanism mediated by cysteine, arginine and aspartic acid residues. Supplementary domains assist in targeting and substrate specificity 24, in contrast to most other types of phosphatases, which require interaction with regulatory proteins for proper substrate binding.

4. The NIF group (

N

i

f

F

C

p

S

C

p

Malaria remains a major public health problem in tropical and subtropical regions, bearing a huge socio-economic impact on affected countries, most of which are in the developing world. Malaria parasites have a complex life cycle. Infection of human beings by Plasmodium falciparum, the species responsible for the lethal form of human malaria, begins with the bite of an infected Anopheles mosquito, which delivers sporozoites into the bloodstream. These cells establish an infection inside hepatocytes, where they undergo intense multiplication generating several thousand merozoites, a process called exo-erythrocytic schizogony. The merozoites invade erythrocytes, where they also undergo schizogony, the process that is responsible for malaria pathogenesis. Some merozoites, however, arrest the cell cycle and differentiate into male or female gametocytes, which are infective to the mosquito. Once ingested by the insect, the gametocytes develop into gametes and fuse into a zygote. Further development in the mosquito involves a process of sporogony, producing sporozoites that accumulate in the salivary glands and are now ready to infect a new human host (see http://www.malaria.org for information on malaria).

The study of signalling processes (in particular those involving protein phosphorylation/dephosphorylation) in malaria parasites presents considerable interest, both in terms of fundamental biology (how does a eukaryote that is phylogenetically very distant from model organisms regulate growth, proliferation, differentiation and transition between its complex developmental stages?) and in terms of the search for urgently needed novel drug targets 3031. The sequencing of the P. falciparum genome 32 and the availability of an interactive genomic database (PlasmoDB, http://www.plasmodb.org) 33 have dramatically facilitated the identification of potential targets. Probabilistic models of peptide domains sharing an evolutionary history (Hidden Markov Models, HMMs) permit the rapid scanning of a set of conceptual translations from any organism whose genomic sequence is available 34. The plasmodial kinome has thus been characterised and highlighted profound divergences from the kinomes of other eukaryotes 3536. Although a number of studies have been published on individual phosphatases of malaria parasites (see below), a full "phosphatome" analysis has not been reported, while studies of both the kinomes 37 and phosphatomes 38 of other major parasitic unicellular eukaryotes, the trypanosomatids, have recently been published. Here, we use the Pfam collection of HMMs 39 to investigate the phosphatome of P. falciparum in relation to that of members from all major groups of the eukaryotic kingdom 40.

HMM profiles defining the diverse phosphatase catalytic domains (see Methods) were used to scan the predicted proteomes of the following organisms, representing all major groups within the eukaryotic kingdom (with the exception or Cercozoans, for which a representative full genome sequence is not available at present): Homo sapiens (Opisthokonts), Dictyostelium discoideum (Amoebozoa), Arabidopsis thaliana (Plants), P. falciparum (Alveolates), the diatom Thalassiosira pseudonana (Heterokonts), Trypanosma brucei (Discicristates) and Giardia lamblia (Excavates) (see Fig. 1). This allowed the identification of 633 sequences, with a number of sequences per genome ranging from 34 sequences for P. falciparum (the smallest phosphatome in our sample) to 224 sequences for A. thaliana (see Additional files 1, 2, 3, 4, 5). The distribution of the various phosphatase families in each phosphatome is illustrated in Fig. 2. The major expansion of all protein phosphatase types, with the exception of PTPs, in Arabadopsis thaliana is evident. In Homo sapiens the major expansion is in the PTP superfamily. These expansions probably reflect the requirement for flexible and complex intercellular signalling in these multicellular organisms, evidently achieved by distinct evolutionary processes in plants and metazoans. The 34 entries in the P. falciparum phosphatome include 7 sequences of the PPP group clustering with subfamilies of phosphatases whose predicted substrates are distinct from phosphoproteins; see Fig. 3). In the next sections we provide a detailed description of P. falciparum database mining for each of the 4 major phosphatase groups (PPP, PPM, PTP and NIF).

A high-resolution version of Figure 3 is available as a PNG file (see additional file 7)

Four P. falciparum sequences (see Table 1 and Additional file 4) containing domains conforming to the NIF profile were detected. Two of these (PFE0795c [A] and MAL13P1.275 [D]) have features that are consistent with phosphatase activity (in particular the presence of a putative active site DxDx(T/V) motif), while a third sequence (PF10_0124 [C]), although closely related, does not have an intact DxDx(T/V) motif 29, and hence may be catalytically inactive. Both MAL13P1.275 and PF10_0124 possess BRCT domains (the BRCA-C-terminal domain is an evolutionarily conserved phospho-binding domain 71) diagnostic of Fcp1 phosphatases. PFE0795c is a significantly smaller protein, lacking BRCT domains and hence related to SCP type phosphatases 2629. A distinct clade within the phylogenetic tree (Fig. 6) involves NIF type domains associated with the TIM50 sub-unit of the mitochondrial translocase complex. This group includes the P. falciparum sequence PF07_0110 [B]. It is notable that the DxDx(T/V) motif is disrupted in all these sequences, and thus these proteins are unlikely to possess phosphatase activity.

A high-resolution version of this Figure is available as a PNG file (see additional file 10)

CDC25 enzymes form a distinct group of phosphatases that play a major role in cell cycle control 72. These enzymes have little sequence similarity to PTPs, except for the presence of the catalytic CX₅R motif, and appear to have evolved from Rhodanese domains, many of which catalyse sulphur transfer reactions. CDC25s can therefore be identified using a Rhodanese domain HMM profile (PF00581). The cyclin-dependent kinases that mediate cell cycle progression possess conserved threonine and tyrosine residues (T14 and Y15 in human CDK2), whose phosphorylation inactivates the enzyme and causes cell cycle arrest. CDC25 enzymes relieve this block by dephosphorylating these residues. Several P. falciparum CDKs display the conserved threonine and tyrosine residues that are the targets of CDC25 in other systems 2073. A Rhodanese domain HMM search identified three hits in P. falciparum, (Fig. 7), two of which were present on the same polypeptide (PFL0320w). The other sequence (PF13_0027) clustered with human and Dictyostelium CDC25s. We were surprised, however, to notice that this sequence does not contain the CX₅R motif essential for catalytic activity (see Additional files 5 and 6 for an alignment), and may therefore not encode a functional enzyme. Whether or not plasmodial CDKs are regulated by phosphorylation/dephosphorylation of T14 and Y15 remains to be investigated; either this mechanism of cell cycle control does not operate in malaria parasite (there is to date no evidence that the conserved threonine and tyrosine residues are phosphorylated), or the CDC25 functional homologue is too divergent to be detected.

A high-resolution version of this Figure is available as a PNG file (see additional file 11)

Other protein phosphatase groups for which we found no evidence in Plasmodium are the Tyrosine phosphatases (see Fig. 5), the Low Molecular Weight phosphatases 23, the cdc14 phosphatases 22 and the Styx phosphatases. Styx sequences are related to those of PTPs and do recognise phosphotyrosine residues, but are non-catalytic proteins (the catalytic cysteine is replaced by a glycine). Because of their similarity to PTPs, human Styx sequences were picked up in our HMM search and are highlighted on the tree in Fig. 5. However, P. falciparum does not possess obvious Styx homologues. Finally, an HMM search of the P. falciparum database using the PFAM profile for Myotubularin (MTM) family of lipid phosphatases 74 yielded only hits with very low scores, indicating that the parasite does not encode members of this family.

The putative P. falciparum phosphoprotein phosphatase sequences were examined for the presence of signal peptides targeting proteins to various cellular compartments. PlasmoDB records the presence of apicoplast targeting sequences 75, the signal peptides predicted by SignalP 76, and the motif directing proteins to the host erythrocyte 7778; the presence of these motifs on PP sequences is indicated in Table 1. In addition the set of sequences was analysed by the PlasMit algorithm http://gecco.org.chemie.uni-frankfurt.de/plasmit/ for putative mitochondrial targeting. No sequence demonstrated unequivocal (high stringency) mitochondrial targeting with this algorithm, not even the peptide associated with the TIM50 mitochondrial translocase (PF07_0110); however seven sequences yielded a lower score that is still compatible with mitochondrial targeting (in order of probability: PFE0795c > PF07_0110 > PF11_0362 > MAL8P1.109 > PF10_0093 > MAL13P1.44 > PFI1245c). The (presumably inactive, see above) PF10_0124 is borderline in this respect and is classified as non-mitochondrial. It is relevant to repeat here, as pointed out above, that the PP5-like MAL13P1.174 possesses a nuclear localisation signal. It is important to emphasise (i) that the presence or absence of targeting motifs on any sequence is dependent on gene predictions and can vary with database re-annotations, and (ii) that the functionality of such motifs should be verified experimentally.

In addition to the accessory domain instances described earlier, the only sequence containing associated domains with homologues in other organisms is that annotated as erythrocyte membrane-associated antigen (PF10_0177). In the sequence we originally downloaded and used in our analyses, this large polypeptide had an EF-hand domain and a putative acid protease domain in addition to the phosphatase domain. In a recent re-annotation of this locus, however, the open reading frames encoding the phosphatase and protease domains are proposed to be split and expressed as distinct genes. Other domain combinations are discussed above.

In contrast to the catalytic domains of the protein kinase superfamily, the vast majority of which conform to a single HMM profile (Pfam database entry PF00069; Pkinase), the diversity of protein phosphatases is reflected by the presence of 7 distinct Pfam profiles defining catalytic domains with protein phosphatase activity. Please see table 2

The above profiles were downloaded from Pfam 3982, and used unaltered to mine the various genomes for conformant sequences.

Two tyrosine phosphatases (PF00102; Y-phosphatase, PF03162; Y-phosphatase2), and a dual-specificity serine/threonine/tyrosine phosphatase activity (PF00782; DSP) are closely related and are grouped in a single clan, the protein tyrosine-phosphatase superfamily (also referred to as PTP) 82. An additional low molecular weight tyrosine phosphatase [PF01451] with limited sequence similarity, but possessing the characteristic PTP motif, is also listed. Serine-threonine phosphatase activities are found in two distinct groups: a highly conserved group conforming to the Metallophosphatase family (Pfam profilePF00149; note that this family includes a wide range of phosphatase activities in addition to protein phosphatases; the protein phosphatase activities are classified as PPP type) and a structurally unrelated (though catalytically similar) group, the PP2C family (Pfam profile PF00481, PPM).

Catalytic domains were identified by use of the hmmsearch option of HMMER 34 using Hidden Markov Profiles appropriate to the domain of interest using moderately stringent criteria (Expect value [-E] of 10^-3, database record number [-Z] 100000). The initial search used the global model for each domain type, although the local model was subsequently used where appropriate if multiple or fragmented domains were found.

Peptide sequences were aligned under the guidance of an appropriate HMM profile 34 using the hmmalign option of HMMER. Alignment output in ClustalW format was trimmed down to those blocks encompassing match states to the profile and ungapped Fasta formatted sequences extracted from this sub-set of the alignment. (T_coffee seq_reformat option).

MSA of a given sequence set was performed by three independent methods; ClustalW 83, t-coffee 84 and hmmalign 34 guided by the appropriate profile. The alignments used the default settings for each method. Alignments were combined under t-coffee, and quality of alignment assessed.

The eukaryotic kingdom is extremely diverse, and molecular analysis confidently identifies eight major groups within this diversity. For a broad phylogenetic and evolutionary analysis of the protein phosphatases present in P. falciparum the translated gene products (June 2006 versions) were downloaded from completed genome projects of the following species:

Homo sapiens (Opisthokonts)

ftp://ftp.ensembl.org/pub/release-30/human-30.35c/data/fasta/pep/Homo_sapiens.NCBI35.apr.pep.fa.gz

Dictyostelium discoideum (amoebazoa)

http://www.dictybase.org/db/cgi-bin/dictyBase/download/download.pl?area=blast_databases&ID=dicty_primary_protein.gz

Arabidopsis thaliana (viridiplantae)

ftp://ftp.tigr.org/pub/data/a_thaliana/ath1/SEQUENCES/ATH1.pep.gz

Plasmodium falciparum (alveolates)

http://www.plasmodb.org/common/downloads/release-5.5/

Thalassiosira pseudonana (heterokonts)

ftp://ftp.jgi-psf.org/pub/JGI_data/Diatom/thaps1/thaps1Prots.fasta.gz

Trypanosoma brucei (discicristates)

ftp://ftp.sanger.ac.uk/pub/databases/T.brucei_sequences/T.brucei_genome_latest

Giardia lamblia (excavates)

http://gmod.mbl.edu/perl/site/giardia?page=download_tool&file=orfs_aa&type=orfs&noheader=T

At the time of writing, no genomic information was available for any member of the cercozoan group. Translated peptide sets for the six model genomes were combined and subjected to the HMMER hmmsearch option using the above criteria. Identified sequences were retrieved using the blast 85 fastacmd option, and domain conformant subsequences extracted. Appropriate subsequences derived from P. falciparum sequences were added to the dataset. An all against all blastp (-e 0.01) of the sequence set was performed, and Markov clustering of the output performed under control of the Tribe package 86. The inflation parameter (-I) was 1.7, a value which demonstrates a reasonable discrimination without fragmenting clusters to an unusable degree

High quality Multiple Sequence Alignments of the catalytic domains were prepared as described above, and columns displaying low consistency (score < 5) or significant numbers of gaps (> 15%) removed. Alternate neighbour joining phylogenies were visualised using Neighbour-Net, implemented on SplitsTree version 4 87.