Plants contain hundreds of protein kinases that are believed to provide cellular signal transduction services, but the identities of the proteins they are targeting are largely unknown. Using an Arabidopsis MAPK (mitogen-activated protein kinase) (MPK6) as a model, Sörensson et al. describe in this issue of the Biochemical Journal how arrayed combinatorial peptide scanning offers an efficient route to discovery of new potential kinase substrates.
Protein kinases are ubiquitous modifiers of proteins both in prokaryotic and eukaryotic cells, where they catalyse addition of monophosphate groups to the side chains of, most commonly, specific serine, threonine and/or tyrosine residues in the target protein backbone. Such modifications can potentially change the target protein's catalytic activity, its affinity for interaction partners, its location within the cell, its ability to be modified further by other effectors and even its susceptibility to proteolysis. These altered properties are essentially reversible, since the cell also possesses a large suite of phosphoprotein phosphatases dedicated to the removal of phosphate groups. The performance of a large fraction of the eukaryotic proteome, a subpopulation sometimes referred to as the phosphoproteome, is thus dependent on the balance between kinase-driven phosphorylation and phosphatase-catalysed dephosphorylation events.
Plant genomes are particularly rich in genes encoding protein kinases, with over 1000 representatives of this protein superfamily identified in the genome of the model species, Arabidopsis thaliana , and large-scale proteomics surveys have confirmed the presence of large numbers of phosphoproteins in Arabidopsis tissues . However, although these surveys provide an initial glimpse of the populations of proteins that are being modified by phosphorylation in different organs and subcellular compartments, they do not reveal the identity of the kinases responsible for generating these phosphoproteins, information that is crucial to understanding how the major signalling networks are organized and regulated. Indeed, only a handful of such target–kinase relationships have been identified to date in plants.
The plant protein kinase superfamily consists of different classes of kinases that are distinguished primarily on the basis of their amino acid sequence similarity, but which are also presumed to possess distinct functional differences. Among these kinase classes, the MAPKs (mitogen-activated protein kinases) are notable for their evolutionary conservation across eukaryotic taxa, for functioning within hierarchical cascades and for their strict preference for phosphorylating target motifs consisting of serine or threonine residues positioned immediately upstream of a proline residue (-S/TP-). In plants, these proline-directed kinases (named MPKs) have been found to be involved in a remarkably wide spectrum of biological activities, ranging from responses to environmental stress and pathogen challenge to control of cytokinesis . However, that involvement has most often been inferred from the phenotypes displayed by gain-of-function or loss-of-function MPK mutants, and in only a few cases has a discrete MPK–target relationship been biochemically characterized. Given that Arabidopsis possesses 20 MPKs, and that the great majority of Arabidopsis proteins are predicted to possess at least one -S/TP- target motif, it is clear that new strategies will be needed if we are to make significant progress in defining MPK-based signalling networks.
In this issue of the Biochemical Journal, Sörensson et al.  describe one such strategy that allowed them to predict, and then validate, functional connections between a series of novel protein targets and one of the most widely studied plant MPKs, MPK6. In the course of their earlier characterization of two paralogous MPK6 targets, AtPHOS32 and AtPHOS34 , this group had noted that the amino acid sequence immediately surrounding the serine residue phosphorylated by MPK6 was widely conserved in homologous proteins from other species. They hypothesized that this sequence conservation might reflect the existence of a local surface topography that identified potential MPK6 targets. To test this idea, they designed a series of synthetic peptides whose combinatorial sequences were centred on the -SP- motif that typifies many MAPK substrates. By systematically varying the identity of the amino acid occupying each of the nine positions either upstream or downstream of this motif, a so-called scanning array of peptides was generated whose members could be assayed in vitro as substrates for recombinant MPK6. Since Arabidopsis MPK6 has a closely related paralogue, MPK3, whose biological activities overlap extensively with those of MPK6, the peptides were tested as substrates for both MPK6 and MPK3, as well as for MPK4, a representative of a different group of plant MPKs.
The activity assays revealed that over 40 of the 162 combinatorial peptides surveyed could be phosphorylated in vitro by MPK6, which enabled the authors to define a local sequence preference landscape for this kinase, namely L/P-P/X-S-P-R/K. Many of those peptides phosphorylated by MPK6 also served as substrates for the MPK3 paralogue, as anticipated, but not for the more distantly related MPK4. By searching the Arabidopsis proteome for proteins whose structure contained the MPK6-preferred sequence, a set of 46 proteins could be identified as candidate MPK6 substrates. Although little functional information existed concerning the post-translational modification of many of the proteins in the candidate list, prior detection in vivo of phosphopeptides derived from several others in unbiased phosphoproteomic surveys added credibility to the prediction that at least some of the candidates represented genuine MPK substrates. Further validation was provided by testing the ability of MPK6 to phosphorylate full-length recombinant forms of five of the candidate proteins in vitro. All five proved to be substrates for MPK6, as well as for MPK3, whereas they were distinctly poorer substrates for MPK4.
The target sequence preference derived from the scanning peptide array thus appears to have been remarkably successful in its ability to predict potential MPK3/MPK6 substrates, at least as assessed by in vitro assays. In addition, the preferred target sequence consistently displays a bias in favour of MPK3/MPK6 kinase activity as opposed to MPK4 activity, indicating that, even within the very restricted vicinity of the canonical -SP- motif, protein topology can help select for different MPKs as signalling partners. The demonstration that manipulation of expression of one of the in vitro-validated MPK6 substrates generated a stomatal development phenotype very reminiscent of that observed earlier in mpk3mpk6 double mutants  is particularly convincing evidence of the ability of this prediction strategy to generate biologically relevant outcomes. Differentiation and spatial organization of stomatal openings in plant surfaces is controlled by an elaborate network of signalling and transcriptional regulators that has been shown previously to include MPK6 .
Earlier studies aimed at identifying possible substrates for plant MPKs have relied upon in vitro kinase reactions conducted on large-scale arrays of immobilized recombinant proteins. Feilner et al.  examined the ability of recombinant MPK3 and MPK6 to phosphorylate substrates within a ~1500-member array of recombinant Arabidopsis proteins, and found that MPK3 and MPK6 could each phosphorylate 39–48 proteins on the array. Notably, 26 of these were substrates for both kinases, consistent with the view that MPK3 and MPK6 share considerable functional redundancy in vivo. The study by Feilner et al.  also validated most of the observed array phosphorylation events by using classical in vitro kinase assays, but it is interesting that none of the proteins found to be phosphorylated in their study appears in the list of predicted MPK6 substrates described by Sörensson et al. . There is a similar intriguing paucity of overlap between the Sörensson candidate list and the proteins found to be phosphorylated by recombinant MPK6 on a larger (2158 element) protein microarray used by Popescu et al. . Of the 184 putative MPK6 targets detected in that ambitious survey, only one also appears as a predicted MPK6 substrate in the study by Sörensson et al. . Since the sites of phosphorylation of MPK6 target proteins in the Popescu et al.  and Feilner et al.  protein array studies remain unknown, it is unclear whether the local sequence preference computed for MPK6 by Sörensson et al.  is somehow biased in favour of a certain type of protein substrate which was not adequately represented on either the Feilner or Popescu arrays, neither of which sampled the full Arabidopsis proteome. Once full proteome chips become available, or other truly comprehensive phosphoproteome sampling strategies are developed, this issue can perhaps be resolved.
The search for kinase substrates is not, of course, restricted to plant systems. The kinases controlling the status of the yeast phosphoproteome have been extensively characterized  and the potential of human kinases and phosphatases to serve as ‘druggable targets’ has attracted intense interest in the medical research community. At least one commercial service draws on the large body of knowledge around human kinase–substrate interactions to provide a bioinformatics-based prediction of the most probable amino acid sequence surrounding the phosphorylated residue in the protein targets of any given human kinase . These algorithms also appear to have a broader applicability across taxa; when applied to the yeast kinome, they enabled a notable degree of success in predicting yeast kinase target sites . When the human kinase-derived Kinexus algorithms were applied to Arabidopsis MPK6 and the Arabidopsis proteome, the resulting target site prediction, -P/L-D/L/P-SP-X-, shares some key elements with the prediction by Sörensson et al. , -L/P-L/X-SP-R/K-, most notably the preference for either proline or leucine in the −2 position relative to the canonical SP phosphosite. However, the experimentally determined MPK6 preference for a basic residue in the +2 position reported by Sörensson et al.  was not captured by the Kinexus model. Whether this reflects a fundamental difference between human and plant MAPKs, or simply reflects the bias of the algorithms, is unclear at this point. It is noteworthy that the somewhat larger target site motif of -PXSP- embedded in both the Sörensson et al.  and Kinexus predictions has been previously touted as a ‘high-stringency’ site for MAPK-directed phosphorylation, but examination of the limited set of plant MPK target sites previously identified shows that both -PXS/TP- and -S/TP- sites are represented with similar frequency. Again, as discussed above, we will need more data in order to understand the significance of these interesting patterns.
Finally, it appears that the target site prediction generated from the scanning peptide array of Sörensson et al.  has the power to distinguish preferred MPK3/MPK6 target sites from those preferred by MPK4, but can this distinction be generalized to related MPKs, or does it simply represent a unique situation? It would certainly be informative to extend the analysis to MPKs related to MPK3/MPK6, such as MPK10 (Group A), and to MPK4 relatives in Group B, such as MPK11 and MPK5. The Group D MPKs would be another interesting clade to examine, since this set of MPKs diverged from its sister clades early in plant MPK evolution , and contains unusual forms whose biological roles remain largely mysterious. If the predictive power of the peptide scanning array strategy can be effectively extended to these other clades, it will represent a major addition to the resources currently available for identifying biological targets and roles for the many plant MPKs about which virtually nothing is presently known. Wider use of this technique would then make it possible to begin mapping in detail the complex signalling and response networks into which these pervasive kinases are integrated in plant biology.