Several mass spectrometry (MS) studies focused on the largescale elucidation of potential substrates for ATM and ATR kinases have now revealed the striking complexity of the DDR. In a pioneering study in which phosphospecific pS/pT-Q antibodies were used to enrich for phosphoproteins prior to SILAC-based MS quantification (Table 1), Elledge and colleagues identified over 900 phosphorylation sites on 700 proteins that were upregulated at least 4-fold upon induction of double-strand breaks using ionizing radiation (IR) (Matsuoka et al., 2007). These phosphorylation sites were found on proteins that perform a remarkable diversity of functions, including previously unidentified sites on known DDR proteins such as ATM itself, FANCD2, and BRCA1, as well as proteins that previously were not associated with the DDR, including RNA-splicing factors. By superimposing this list of proteins on known protein-protein interactions, the authors have generated multiple DDR modules that interact to form a larger network. Owing to the remarkable number of new DDR responsive SQ/TQ substrates identified in this study, it is not surprising that many of the modules identified were previously unknown DDR network components; indeed, some might account for previously enigmatic cellular processes observed in response to DNA damage. Most intriguing of these observations was the suggestion that the DDR network intimately intersects with multiple components of the insulin signaling network, including the PI3K-AKT pathway. It is tempting to speculate that the DDR co-opts this network to inhibit or delay apoptosis in response to DNA damage, thereby providing the necessary time for the DNA repair pathways to respond. Understanding the interactions between these two networks might provide clues regarding how cells decide between cell survival and apoptosis following DNA damage.

As is the case with all large-scale studies, despite the generation of massive amounts of data, many fundamental questions remain. For instance, what is the functional significance of the multiple phosphorylation sites identified on ATM and other DDR proteins? Are responsive phosphorylation sites ATM or ATR substrates, or might other PI3K-like protein kinase (PIKK) family members, known to respond to DNA damage (e.g., DNA-PK and SMG-1), be responsible (Matsuoka et al., 2007)? It is worth noting that the study performed by Elledge and coworkers provides only a glimpse into the complexity of the phosphorylation events involved in the DDR. This point is highlighted by a recent study in which phospho-motif specific antibodies were used to enrich for peptides containing pS/pT-Q, with the goal of identifying substrates of SQ/TQ-directed kinases (e.g., ATR) activated by ultraviolet (UV) irradiation, which causes single-strand breaks (Stokes et al., 2007). Despite the identification of over 200 UV-responsive phosphorylation sites with the proper SQ/TQ motif, only 46% overlapped with those identified in the IR study. This lack of overlap between the two data sets could result from biochemical differences in the DDR network in response to double-strand (IR) versus single-strand (UV) breaks or could be caused by experimental variation between the two studies, including cell line and phosphospecific-versus-phosphomotif antibody differences.

The phosphorylation sites identified in both data sets are of greater interest, as they might be associated with common pathways that respond to multiple DNA insults, or they could represent an activation cascade of multiple members of the PIKK family in response to specific DNA-damage insults. For instance, in response to double-strand breaks, ATM and DNA-PK activation is followed by subsequent ATR activation via both ATM-dependent and -independent mechanisms (Matsuoka et al., 2007). To unravel the kinase dependency of selected phosphorylation sites, biochemical perturbation was combined with quantitative MS to identify common nodes and potential PIKK-specific substrate candidates from these two studies. Specifically, to identify ATM-dependent phosphorylation events, cells were treated with IR in the presence of KU-55933, a specific ATM inhibitor; to identify ATM-dependent phosphorylation events, ATR-deficient Seckel cells were treated with UV. Approximately 70% of the sites that were diminished in the UV-irradiated Seckel cells were also upregulated in the general cellular response to IR and, therefore, likely represent ATR-dependent sites phosphorylated in response to either UV- or IR-induced DNA damage, possibly through the sequential activation of ATR by ATM in the latter case. Interestingly, two ATR-dependent sites, one on EYA3, a phosphatase involved in organogenesis, and one on SMC1, a member of the cohesin complex, are also ATM-dependent. This finding suggests that these proteins might be shared nodes in the cellular response to different forms of DNA damage, activated by both ATM and ATR. Although SMC1 was previously implicated as a substrate for both ATM and ATR kinases, with kinase dependency linked to the extent of DNA damage (Wakeman and Xu, 2006), EYA3 is a newly defined common node whose role in regulating the DDR remains undetermined. In addition to these common nodes, the splicing factor SFRS14 was found to be ATR dependent in the UV study and ATM independent in the IR study. SFRS14 is, therefore, likely to be an ATR-specific substrate involved in both UV- and IR-activated DDR in an ATM-independent fashion.

Owing to its unbiased nature of data acquisition, quantitative (MS)-based phosphoproteomic approaches have been particularly informative in the discovery of new components in signaling networks downstream of ErbB family members. For instance, temporal analysis of EGFR signaling dynamics in HeLa cells using a combination of SCX with TiO₂ and SILAC-based MS quantification (Table 1) identified 6,600 phosphorylation sites, of which 90% were previously unidentified (Olsen et al., 2006). Of these phosphorylation sites, just over 1,000 were modulated at least 2-fold upon EGFR activation. Applying clustering approaches to this data set, multiple signaling profiles were obtained where the biological function, such as signal initiation or negative regulation, of each cluster was inferred from the presence of prominent phosphorylated proteins of known function. As is often the case for discovery based proteomic approaches, the majority of proteins identified in this study were previously not associated with growth factor activation, including a series of phosphorylation sites on transcription factors and coregulators such as DAFT-1 and WBRS9. Unfortunately, despite the vast amount of data generated in this study, minimal mechanistic insight was revealed regarding the functional role of the regulated phosphorylation sites. This lack of biological information further emphasizes the importance of using additional biological perturbations (e.g., EGFR kinase inhibitors or stimulation with alternative ligands for the EGF family) and phenotypic measurements to accompany and amplify the existing phosphoproteomic data set.

Statistical modeling approaches designed to correlate signaling network nodes and cancer phenotypes have been developed to address these limitations, and have been applied to a large scale phosphotyrosine data set of temporal EGFR, ErbB2, and ErbB3 signaling activated by EGF or heregulin stimulation of human mammary epithelial cells (Wolf-Yadlin et al., 2006). This study, in which phosphopeptide enrichment was performed using pY pan-specific antibodies and phosphorylation site changes were quantified with iTRAQ-MS (Table 1), determined that cellular phenotypic differences can be attributed to specific changes in downstream signaling networks in response to growth factor stimulation. For instance, following EGF stimulation, cells engineered to stably express elevated ErbB2 levels are highly migratory compared to parental cells. This phenotype was linked to phosphorylation changes associated with decreased cell-cell adhesion and a subset of the known cell migration signaling network, including FAK and p130cas. Partial least-squares regression (PLSR) (reviewed in Janes and Lauffenburger, 2006) identified phosphorylation sites in the ErbB signaling network that were highly correlated with cell migration or proliferation, lending functional consequence to the phosphoproteomic data set. Derivation of such network-phenotype relationships represents a rich source of potential targets for drug discovery. One such example is the successful use of a small molecule inhibitor of Annexin II, a target that strongly correlates with migration, for the inhibition of migration and invasion in cancer cells (Falsey et al., 2006). Further statistical modeling demonstrated that nine phosphorylation sites on six proteins involved in receptor endocytosis and phosphoinositide 3-kinase (PI3K)-mediated pathways could fully recapitulate the predictive capability of the PLSR model (Kumar et al., 2007). These nine nodes correspond to phosphorylation “biomarkers” that might be able to predict the proliferative or migratory potential of EGFR- and ErbB2-driven tumors.

The need for integration is illustrated by two recent phosphoproteomic studies of RTK coactivation in glioblastoma (Huang et al., 2007; Stommel et al., 2007). Each study utilized a different approach (large-scale quantitative MS versus RTK microarrays), yet both demonstrated that simultaneous coactivation of multiple RTKs contributes to EGFR TKI resistance in glioblastoma. Although both studies found that inhibition of multiple RTKs is required to overcome this resistance, the microarray studies were extended to multiple cell lines and xenografts (e.g., more breadth), whereas the MS-based study provided a quantitative analysis of the signaling networks associated with RTK activation (e.g., more depth). In the future, one can envisage a workflow in which discovery MS is employed to initially determine the signaling profiles characteristic of EGFR TKI resistance. Representative signature nodes from these profiles can then be used in a protein microarray format for a targeted screen of a large number of human gliomas. Patients with the appropriate profile can subsequently be considered as candidates for treatment with selected small molecule kinase inhibitors. Conversely, information gleaned from higher throughput targeted screening can isolate clinical conditions that could benefit from further mechanistic insight by discovery approaches. For instance, if a targeted screen uncovered a subset of patients that demonstrated clinical hypersensitivity to TKIs, discovery-based MS could then be used to probe the potential mechanisms of action.

Additional mechanistic insight will be achieved by merging integrated phosphoproteomic data sets together with complementary computational modeling approaches as a means of simplifying data sets and accessing nonintuitive signaling targets for future therapeutic development. One promising example is the use of deterministic methods based on ordinary differential equations (ODE). For instance, an EGFR mechanistic model containing 148 processes with 103 variable species has been described (Hornberg et al., 2005). Inclusion of quantitative EGFR phosphoproteomic data into ODE-based models will provide additional information for model construction and refinement. Furthermore, in silico perturbation of network components in these expanded models (e.g., sensitivity analysis to determine nodes for which the outcome of the system is most susceptible) (Savageau, 1971) will reveal critical nodes embedded in the complex signaling web which might serve as logical points for further experimental interrogation and potential therapeutic intervention.