To date, 10 genes encoding members of the classical MKP family have been isolated and characterized from mammalian genomes (see Table 1). They all share some common features, including an extended active-site motif with significant sequence similarity to the corresponding region of the VH-1 protein tyrosine phosphatase that was isolated from vaccinia virus [4]. In addition, their amino termini contain two short regions that are homologous to sequences that flank the Cdc25 phosphatase catalytic site [5] and a cluster of basic amino-acid residues that play an important role in binding to the MAP kinases [6]. The MKPs can be further subdivided into groups depending on their substrate specificity for the MAP kinases, subcellular localization (nuclear, cytoplasmic or both) and the structural organization of their genes.

A gene closely related to the MKP family is STYX. The protein encoded by STYX contains many of the hallmarks of the MKPs but lacks a critically important catalytically active cysteine residue, having a glycine instead [7]. Interestingly, alteration of this glycine residue to cysteine renders the STYX protein catalytically active, suggesting that other critical residues have been retained. STYX appears to function as a phosphotyrosine/phosphoserine/phosphothreonine-binding protein analogous to the SH2 and PTB domains. It is unclear whether it plays any role in signaling through the MAP kinase pathways, however.

On the basis of structures predicted from genomic sequence, the MKPs can readily be divided into the following subgroups: subgroup I, DUSP1, DUSP2, DUSP4 and DUSP5; subgroup II, DUSP6, DUSP7, DUSP9 and DUSP10; and subgroup III, DUSP8 and DUSP16 (Figure 1, Table 1). Division into these three subgroups is also supported by phylogenetic analysis (Figure 2), and to a considerable extent by substrate preference.

The genes of subgroup I consist of four exons [8,9,10,11], and the exact positions of the introns are highly conserved. The first exon encodes the amino terminus of each of the three proteins and ends within the second of the Cdc25 homology (CH2) domains. The active-site motif of all four proteins is encoded within exon 4, and the length of exon 3 is identical for all these DUSPs, suggesting that they arose from a common ancestral gene.

Subgroup II genes (DUSP6,7 and 9, and provisionally 10) consist of three exons [12]. As with subgroup I, exon 1 encodes the amino terminus and the 3'-most exon encodes the active site. Unlike subgroup I, however, the central portion of the protein is encoded by a single exon. DUSP10 has an intron/exon structure consistent with a placement within subgroup II, but it shares some features with members of subgroup III, such as a similar MAP kinase docking-site motif and similar substrate specificity. In addition, it has an extended amino-terminal region that is not present in any of the other DUSPs. Assignment of DUSP10 to subgroup II or III or to its own subgroup may need re-evaluation in the future.

The genes of subgroup III (DUSP8 and DUSP16) consist of six exons [13,14,15]. The lengths of exons 2, 3, 4 and 5 are identical in the two genes. Unlike the two previous subgroups, however, exon 1 encodes only the first of the CH2 domains. Exon 2 ends within the same region of the second CH2 domain as does exon 1 of subgroups I and II, however. The central portion of the protein is encoded by two exons, as in subgroup I, rather than the single exon of subgroup II. This suggests that subgroup II genes may have diverged earlier and that subgroup I and III genes may be more closely related to each other. Exon 5 encodes the catalytic site of these DUSPs but not the carboxyl terminus; the sixth exon encodes the large extended carboxyl terminus, with the intron between the catalytic domain and the carboxyl terminus of the protein being diagnostic of this subgroup.

Overall, the high conservation of some intron/exon boundaries within all three subgroups of DUSPs indicates a common ancestral gene for each subgroup. The similarities of substrate specificity among subgroup members suggest that these ancestral genes are themselves likely to have shown some substrate preference.

Dual specificity phosphatases have been found in a wide variety of phylogenetically distinct eukaryotes, including yeasts, worms, flies and plants, as well as mammals. Although budding yeast does not encode any predicted protein tyrosine kinases, it has 16 dual specificity and low-molecular-weight phosphatases and five phosphatase-like STYX genes. A direct role has been demonstrated for the budding yeast dual specificity phosphatase gene MSG5 in the regulation of the MAP kinase Fus3 [16]. Six MKPs have been noted in the Drosophila melanogaster genome sequence [17]. Of these, two have been characterized: puckered has been shown to encode a MKP that inactivates basket, the Drosophila equivalent of mammalian JNK [18]; in addition, a Drosophila DUSP6-like gene has recently been identified [19]. The sequence of the Caenorhabditis elegans genome includes 26 predicted DSPs, including some MKPs [20]. One of these, LIP-1, has been suggested to inactivate MAP kinase, mediating inhibition by the ligand Notch of a signaling pathway involving the GTPase Ras, during C. elegans vulval development [21]. Moreover, at least two MKPs have been identified in Arabidopsis thaliana, one of which is required for relief from genotoxic stress [22].

All the DUSPs share strong amino-acid sequence identity over their catalytic domains (37%-50%), with members of certain subgroups sharing up to 75% identity (for example, DUSP6 with DUSP7 and DUSPQ or DUSP8 with DUSP16). The catalytic domain invariably contains the highly conserved consensus sequence DX₂₆(V/L)X(V/I)HCXAG(I/V)SRSXT(I/V)XXAY(L/I)M, in the single-letter amino-acid code where X is any amino acid. The three amino acids indicated in bold have been shown to be absolutely essential for catalysis. The cysteine is required for the nucleophilic attack of the phosphorus of the substrate and the formation of the thiol-phosphate intermediate; the conserved arginine binds the phosphate group of the phosphotyrosine or phospho-threonine, enabling transition-state stabilization; whereas the aspartate enhances catalysis by protonating the leaving group oxygen [23].

Determination of the crystal structure of the catalytic domain of DUSP6 [24] has revealed that this domain adopts a shallow cleft conformation, similar to that of VH1-related phosphatase (VHR) [25]. Within this cleft, the phosphatase is able to accommodate both phosphotyrosine and phosphothreonine side chains. In the absence of substrate, DUSP6 exists in a low-activity state, with key residues (Arg299 and Asp262 in DUSP6) disengaged from interaction. Upon binding of the substrate ERK2, however, the active-site residues are rearranged, adopting a catalytically active conformation [24] and enabling the dephosphorylation of both Thr183 and Tyr185 in ERK2 (Figure 3b). In particular, Asp262 (the conserved aspartate in the consensus sequence) is located in a loop, 5.5 Å away from the nucleophilic cysteine and arginine in the active site. This suggests that the loop must undergo conformational rearrangement for activation. Upon binding of the substrate ERK2, closure of the Asp262 loop over the active site occurs, positioning the aspartate residue for catalysis. Thus DUSP6 adopts a catalytically active conformation [24].

The amino terminus is much less conserved in the DUSPs than the catalytic domain. Nevertheless, all DUSPs contain at their amino terminus two conserved regions that show similarity to the Ccd25 phosphatase, designated CH2 domains. In Cdc25, these domains flank the catalytic site of the enzyme, whereas in the DUSPs they are found upstream of the active site [5]. They have been suggested to play a role in substrate binding, but this has yet to be formally demonstrated.

All DUSPs have, near to their amino termini, a MAP kinase docking site, which consists of a cluster of positively charged amino acids [6]. It has been proposed that the number of consecutive positively charged residues in this docking site of MKPs may play a role in determining binding specificity and therefore catalytic activity (Figure 4) [14]. The corresponding docking site on the MAP kinases has been shown to consist of negatively charged residues (known as the common docking domain, CD), suggesting that electrostatic interactions are critical for the binding of MAP kinases and MKPs. Consistent with this suggestion, an altered form of ERK2 (D319N) corresponding to the Drosophila sevenmaker mutation is highly compromised in its ability to bind to, and activate, DUSP6 [26,27]. The low-molecular-weight DSPs do not have the MAP kinase docking site present in MKPs, but VHR (a low-molecular-weight DSP) has been shown to be capable of inactivating ERK in cells [28,29], suggesting that it may have a distinct mechanism of interaction.

Further regions that are potentially important for docking to and activation by MAPKs have now been defined in certain subsets of MKPs. The FXFP motif, and variants of this sequence (present in DUSP1, DUSP4, DUSP6, DUSP7 and DUSP9) has been shown to mediate ERK binding [30,31,32]. Another such region is the D-domain (or D-box), which comprises a cluster of basic residues amino-terminal to an L/I-X-L/I motif [30,32,33]. The consensus for this site was derived from alignment of the sequence of the delta domain of c-Jun, the docking site for JNK, with similar motifs found in other transcription factors that are known targets of the MAP kinases. This domain is present in all MKPs identified to date [34] and is thought to mediate binding to both JNK and ERK MAP kinases [30,32]. Given the diversity of docking sites, the prevailing view is that the affinity of the MKPs for their substrates may be governed by the number, type and accessibility of docking sites.

The subgroup III DUSPs, DUSP8 and DUSP16, have an extended carboxyl terminus containing PEST sequences (abundant in proline, glutamate, serine and threonine residues) that are frequently found in rapidly degraded proteins [35]. Removal of the PEST sequences from these proteins can result in their stabilization ([36] and our unpublished observations), consistent with a role for the domain in rapid turnover.

In 1998, Arkinstall's group [27] demonstrated that binding of DUSP6 to purified ERK2 enhances the catalytic activity of the phosphatase. Since then, several papers have attempted to elucidate the mechanism of catalytic activation of DUSP6 [37,38] and to define the regions of interaction between DUSP6 and ERK2. The derivation of the solution structure of the ERK2-binding (EB) domain of DUSP6 [38] has provided considerable insight into this issue (Figure 3a). The EB domain of DUSP6 contains a positively charged cluster of arginine residues that binds to a highly negatively charged region of ERK2. This binding causes a conformational change in DUSP6, affecting the association of its EB domain with its catalytic domain, which results in the re-positioning of the conserved aspartate residue (Asp262 in DUSP6). These conformational effects, in conjunction with the catalytic domain binding to phosphorylated ERK2, allosterically trigger the rearrangement of DUSP6 active-site residues, resulting in a high-activity state of the phosphatase (Figure 5). This mechanism of activation of DUSP6 by its substrate may be generally applicable to the DUSPs; recent data support a similar model for both DUSP4 and DUSP9 [27,39], and binding of DUSP1 to p38 results in the catalytic activation of the phosphatase [40].

An alternative regulatory mechanism is exemplified by the binding of DUSP1 to ERK1, which results in phosphorylation of the phosphatase [41]. This phosphorylation does not directly affect the phosphatase activity of DUSP1, but instead leads to its stabilization, by reducing the rate of proteosome-mediated degradation of this rather labile protein [41].

Assessment of substrate specificity for the DUSPs has proved problematic because many of the assays used do not reflect the situation in vivo. From data published so far, however, it seems that many of the DUSPs have preferred substrates. For example, after transfection into mammalian cells, DUSP6 completely inactivates ERK1 and ERK2, even at low concentrations, but not JNK or p38 kinases [49,54]; similarly, DUSP9 also displays some specificity towards ERK [45]. In contrast, DUSP8 and DUSP16 appear highly specific for the inactivation of JNK and p38 kinases but fail to inhibit ERK activation [14,15,54]; similarly, DUSP10 displays specificity for p38 and JNK kinases [46,52]. These 'specificities' should be treated with caution, as they may be specific to a certain cell type or physiological status [26,55,56]. Some of the specificity may reside in the docking sites present on the various MKPs, but it must be emphasized that access to substrate may be an equally important factor in regulating specificity in vivo.