Eukaryotes have devoted approximately two percent of their genome to kinases, highlighting the importance of protein kinase function. Protein kinases are involved in numerous cellular processes, and aberrant kinase activity has been directly implicated in the etiology of a wide spectrum of human pathologies. In recent years, several kinase-directed drugs, including Gleevec^®, Iressa^®, Herceptin^® and Avastin^®, were approved to treat human diseases [1]. Currently, more than 50 protein kinase drug candidates are in clinical trials to treat diseases including cancer, chronic inflammation, metabolic disorders, and neurodegenerative disease. The importance of protein phosphorylation in global regulation of cellular processes is apparent from estimates that at least one third of all proteins are phosphorylated [2]. For the vast majority of these proteins, however, the protein kinase(s) responsible for their phosphorylation is not known. In addition, the function of many protein kinases is completely unknown or has been poorly characterized. Despite their central role in health and disease, the identification of protein kinase substrates remains a significant challenge. Techniques that advance our knowledge about the substrates of specific kinases will certainly aid in our understanding of this biologically essential class of protein enzymes.

A growing number of methods to identify substrates of protein kinases are available. Most commonly, proteins from cells are isolated either from gels, by immunoprecipitation, or by metal affinity chromatography, and the phosphorylated sequences are determined by mass spectrometry [3–5]. Kinase-substrate relationships have also been determined through the use of enzyme inhibitors, functional knockouts and analogue sensitive kinase alleles combined with mass spectrometry. There are limitations inherent to these approaches, however, such as functional redundancy, poor characterization of enzyme-inhibitor specificity, and lack of sensitivity due to under-representation of low abundance proteins in these screens. In addition, mass spectrometry-based approaches often require considerable amounts of time and a high level of technical expertise to complete assays, operate instrumentation, and perform data analysis. The use of phospho-specific antibodies against consensus phosphorylation sites has been helpful in addressing some of these issues, but the consensus sequence information or high quality phospho-specific antibodies are not available for many discovery efforts.

In vitro based platforms using purified components offer solutions to some of the limitations of mass spectrometry-based approaches for protein kinase substrate identification. For example, the use of peptide arrays in substrate screens has been valuable for defining consensus phosphorylation sites for many protein kinases. However, it is very difficult to predict which cellular proteins are in fact substrates of protein kinases based solely on short primary sequence information, and using this data to identify bona fide kinase substrates is often problematic. Experiments have demonstrated that the ability of a kinase to modify a protein at a specific site is influenced by its structural context, including post-translational modifications such as glycosylation [6–8], localized sequential "priming" phosphorylations [9], inhibitory sequences [10], secondary structure and solvent accessibility [11]. For these reasons, using arrays of full-length but denatured proteins to screen for kinase substrates, as recently described by Feilner et al using arrays of denatured Arabidopsis proteins [12], has significant limitations, Solution-based assays using purified proteins that have been expressed, purified, and demonstrate functional activity can overcome some of these limitations but can be costly to achieve on a large scale.

Proof-of-principal for the use of arrays of native or functional proteins for kinase substrate profiling was described by MacBeath and Schreiber [13]. Although this study utilized arrays of only a few proteins, it demonstrated some of the advantages that functional protein arrays offer over existing assays. For example, arrays containing thousands of proteins can be prepared with similar amounts of protein to rapidly establish kinase-substrate relationships with limited concentration-based artefacts. Second, the identity of the substrates is known immediately following the assay since full-length proteins are used and the position of each protein on the array is known. Third, the probability that a phosphorylated structured protein revealed in an array experiment is indeed a cellular target of the kinase should be much higher than candidate substrates identified from experiments based on peptides. Lastly, because arrays may contain up to thousands of proteins, target selectivity is quickly addressed with minimal amounts of reagent. Recently, a seminal study by Ptacek et al. profiled the activity for 87 purified yeast protein kinases against a proteome array containing over 4000 proteins expressed and purified from S. cerevisiae [14]. The approach successfully identified thousands (4,192) of phosphorylation events mapping to 1,325 different substrate proteins. The integration of these results with multiple data types has also led to new proposals regarding yeast protein kinase interaction networks. More recently, Boyle et al. used high-content human protein microarrays to identify the actin-regulatory protein cortactin, a protein that is upregulated in several cancers, as a novel substrate of the Abl and Abl-related gene (Arg) nonreceptor tyrosine kinases [15]. These investigators went on to show that Abl-family kinases target cortactin as an effector of cytoskeletal rearrangements in response to PDGF.

In the current study, a detailed characterization of the protein microarray kinase-substrate identification assay is presented. Arrays of human proteins are used, and critical experimental parameters are investigated to address optimal assay performance. We show that the kinase-substrate phosphorylation microarray assay reconstitutes enzyme-substrate interactions observed in solution-based assays using a set of solution-validated protein kinase-substrate pairs. Results are also presented that show that these arrays can be used to quickly identify novel substrates of human protein kinases on functional human protein arrays containing thousands of highly purified proteins and that pooling-deconvolution strategies can be employed to rapidly identify specific kinase-substrate relationships.

Posttranslational modification of proteins is one of the principle regulatory mechanisms in eukaryotes. In particular, protein phosphorylation has been demonstrated to be crucial in the proper regulation of nearly all cellular processes, including metabolism, cell organization and differentiation, responses to environmental stimuli and cell-cell interactions. Nevertheless, a thorough understanding of the full range of phosphoproteins modified by a specific protein kinase is often lacking. Proteome microarrays have been described in which sets of proteins, or nearly every protein in the case of yeast, were expressed, purified, and deposited on a surface in an addressable format [22–24]. Using such a proteome-scale microarray, Ptacek et al created a map of the yeast phosphorylome by incubating the arrays with nearly every yeast kinase and identifying thousands of phosphorylation events [14]. These studies revealed a plethora of potentially novel biological functions as well as new regulatory interactions for a spectrum of kinases. More recently, a human protein array was used to identify a novel substrate for Arg and Abl kinases, the targets of the anti-cancer drug Gleevec [15]. In the present study, we performed a detailed characterization of the ProtoArray^® technology as employed to define kinase-substrate interactions, and have identified a number of key features that can impact kinase-substrate discovery efforts.

The first task was to benchmark performance of the kinase-substrate application on protein arrays relative to standard solution-based assays. This was accomplished by defining a test set of kinase-substrate pairs and determining the correlation of substrate phosphorylation on the array relative to a solution assay. Twenty out of 24 (83%) kinase-substrate pairs were reconstituted on arrays, demonstrating that substrate phosphorylation on protein arrays is highly consistent with solution-based methods, even under the relatively generic conditions employed in these assays. Several possibilities exist for the inability to observe phosphorylation of four protein on the arrays including (1) binding of the protein to the array surface in such as way as to preferentially mask a phosphorylation site; (2) denaturing of the protein on the array that abolishes a kinase interaction domain; (3) altered kinetics of the phosphorylation reaction on the array surface; or (4) a combination of several of these factors. Nevertheless, this "false-negative" rate must be considered in light of the advantages of the array format including speed, low material requirements, and the ability to survey thousands of purified proteins in a single experiment.

In a second set of experiments, we investigated the effect of assay conditions – specifically, the presence or absence of BSA in the assay buffer – on the results obtained with the arrays. Interestingly, 75% of the proteins phosphorylated in solution were observed to be phosphorylated on the arrays either in the presence or absence of BSA; however, the substrates identified under the different conditions did not completely overlap. A subset of protein substrates were observed only in the presence of BSA, while others were only seen to be phosphorylated on the array in the absence of BSA. In the former case, BSA may be stimulating substrate phosphorylation through a 'crowding' effect by increasing the effective local concentration of protein kinase at the array surface. In the latter case, BSA may be interfering with substrate phosphorylation through either binding to protein kinases, and/or by being a substrate for the kinase itself, thereby lowering the amount of available ³³Pγ-ATP and/or kinase in the assay below the threshold required to give adequate phosphorylation of the substrate. Assays in which BSA was included in the buffer gave rise to a higher mean Z-Score across the test set of 24 kinase-substrate pairs, with 17 of the 24 pairs resulting in higher Z-Scores when BSA was present in the assay buffer. Based on this result, BSA-containing buffers are recommended if a single assay condition is to be employed. Further investigation will be required to determine the exact mechanism of the effect of BSA in this assay, and carrying out assays under both assay conditions should be considered in order to maximize the probability of observing substrate phosphorylation.

The study investigating the yeast phosphorylome reported that approximately two-thirds of previously annotated phosphorylation events were not observed in their protein array experiments. One possible explanation for this observation is that the amount of protein spotted on the array was too low to be phosphorylated by kinase. To explore this issue, we measured the relative amount of protein that must be spotted on the arrays for significant phosphorylation to be observed, and determined that the median LOD for the test set of kinase-substrate interactors was approximately 2 ng/μL, which is equivalent to approximately 2 pg of spotted protein. The median level of proteins on the yeast and human proteome arrays are approximately four-fold greater than this amount, suggesting that the majority of proteins on the arrays are present in quantities sufficient to observe phosphorylation. It should also be noted that this amount of spotted protein is approximately 1000-fold less protein than is required to observe phosphorylation in a solution-based assay.

Another factor that could influence the identification of proteins on the array as substrates for exogenous kinases could be masking of potential phosphorylation sites as a result of phosphorylation occurring during protein production by endogenous kinases. Using both anti-phosphoamino acid antibodies and phosphoamino acid stains such as ProQ Diamond, we have shown that many yeast and human proteins expressed from either yeast or insect cells and spotted on protein arrays are indeed phosphorylated (data not shown). As shown in Figure 6, CK2 phosphorylated several proteins only after dephosphorylation of proteins on the array with lambda protein phosphatase. For some proteins, the removal of the phosphate groups through enzymatic treatment with phosphatase was required in order to observe substrate phosphorylation on the array. It should also be pointed out that phosphorylation by some kinases may require that substrates are pre-phosphorylated or "primed" on specific residues.

We have provided further evidence that high content human protein arrays can be used to make novel insights into the biology of human protein kinases. In one example, 23 substrates were identified and validated for CamK2, 11 of which could be assigned to specific functions or pathways. The two protein substrates exhibiting the strongest signals were Doublecortin (DCX) transcript variants 4 and 2. Doublecortin, which has not previously been reported as a CAMK2 subsrate, is important for neurite outgrowth in the developing brain, and acts by stabilizing microtubules [25]. The observation that that calmodulin-kinase 2 phosphorylates doublecortin suggests a mechanism by which CamK2 regulates neuronal migration by influencing microtubule stability, and is consistent with the well established role for CamK2 in neuroplasticity [26, 27].

We have also demonstrated for the first time that a pooling-deconvolution strategy can be applied to protein kinase substrate identification on protein arrays. In these experiments, four protein substrates were identified (FLJ22795, SH3YL1, CRKL, ABI1) that were uniquely phosphorylated on arrays treated with the Abl protein kinase. Signaling through Abl is critical to regulation of several cellular functions including organization of the actin cytoskeleton [28]. CRKL is a known substrate of Abl [29], and ABI1 is a known Abl interacting protein that has been shown to facilitate phosphorylation of Mena, a protein involved in cell adhesion and motility [30]. The protein SH3YL1 is homologous to the yeast protein Ysc84, which localizes to the cortical cytoskeleton in yeast and is involved in coupling endocytosis to the actin network [31]. This type of data argues that bona fide interactors can quickly be identified from screens on protein arrays. However, it is likely that a subset of proteins identified from such screens will not validate in vivo as protein kinase substrates, particularly if kinase and substrate occupy separate subcellular compartments. In addition, various factors in the in vitro array assay, such as kinase and/or substrate concentration, cofactors or lack thereof, or ionic conditions, may lead to inappropriate phosphorylation events. Finally, it is likely that many membrane-associated proteins, especially those which have regions that span the membrane, do not exhibit a native conformation on the array due to the lack of lipid, and thus may be phosphorylated in regions that are normally not accessible to a kinase. Assimilation of protein microarray data with orthogonal data types such as protein expression, localization, and interaction networks will most certainly enrich inventories of in vitro kinase-substrate pairs and expand our understanding of protein kinases function in cellular processes. We believe such integrated datasets will provide novel insights to intracellular phosphoprotein signaling which could ultimately foster important new efforts for drug discovery and development.