Our study delineates the responses of the system-wide cellular phosphorylation network upon systematic inactivation of individual kinases or phosphatases. Because the phosphorylation network is one of the main cellular backbones for the processing of information and the implementation of cellular responses, it is highly dynamic. Our measured behavior is only a single snapshot of a large number of possible outcomes, which were constrained by the growth and experimental conditions that we chose.
The first surprising observation that we made was that 7550 phosphopeptides were consistently identified but did not show a substantial amount of regulation. This may be due to, first, our cutoffs being conservative; thus, many putative regulatory events may not have been reproducible or strong enough to be deemed substantial. Second, 22% of the kinase and phosphatase mutants could not be analyzed, mainly because the corresponding genes are essential for cellular viability. Perhaps their essentiality is at least partly due to a generally higher impact on the phosphoproteome, as indicated recently (10
), or because their substrates need to be phosphorylated constitutively. Third, in yeast, a large number of paralogous kinase isoforms exist (for example, Tpk1, Tpk2, and Tpk3). Given this, it is reasonable to expect some overlap or redundancy in substrates, which could lead to a considerable number of phosphorylation sites that would appear unregulated as long as only one of the paralogous duplicates was deleted. Fourth, the yeast populations that we analyzed consisted in a strict sense of many mixed subpopulations (for example, cells in different cell cycle states), and it can be assumed that an identical phosphorylation site can become phosphorylated by different kinases during the cell cycle. Therefore, analyzing deletions of single kinases or phosphatases would only manifest in slight, if any, regulation for such sites; for example, a cell cycle phase–specific regulation is masked by all cells that are not in that particular phase at any given time point. Fifth, we also analyzed whether the regulated and nonregulated phosphopeptides fell into different protein abundance classes (for example, the nonregulated are of low abundance and therefore regulation is more difficult to observe), but this was not the case. Overall, it is likely that all five possible explanations contribute to the observed result.
Another finding of this study was the unexpectedly strong dominance of indirect effects (as opposed to direct molecular target effects), which were often without a resulting strong cellular phenotype. To some extent, this observation fits with a view of signaling networks having to be highly flexible and redundant to respond to an ever-changing environment while maintaining stable cellular states (44
). This constrains the architecture of the system, as described by the “law of requisite variety” (45
), a fundamental law in systems control theory. It states that stable systems have to encode a number of control states that is higher than or equal to the number of states to be controlled. Considering that for each cell the space of “environmental states” is enormous, consequently, also the cellular “control variable space” must have an equal or greater size. The combinatorial possibilities of the phosphoproteome seem to ideally fulfill this demand (44
An alternative explanation for this observation might also be found in the theory of Neutral Evolution (47
). It is possible that only a small number of the observed phosphorylation events are actually relevant for the function and survival of the cell, whereas most phosphorylation events would simply have no effect, or at least have no negative effect, on the cell. As a result, such phosphorylation sites would not be counterselected during evolution. The data generated in this study do not, by themselves, support or refute this hypothesis. Finally, the low correlation between phenotype and the degree of change in the phosphoproteome may have been affected by the growth conditions chosen here, the lack of sensitivity of the phenotypic assays, or the possibility that the phosphoproteomics data were not sampled deeply enough to find such correlations.
In addition to revealing insights into the architecture of cellular signaling, our data set also describes the proteome-wide functional states of yeast cells; this might be useful for determining diagnostic markers for stress conditions, functional states of key pathways, or the activity of a given kinase or phosphatase. These markers could be used in conjunction with targeted proteomics approaches to not only study basic biological processes but also determine how a given pharmacological intervention would affect the cellular signaling network.
With targeted proteomics methods, not only can the cellular information flux under many conditions be observed, at high throughput, but this approach also enables us to understand for all phosphorylation sites whether the observed change is a “true” regulation event or simply as a result of a change in protein abundance (48
) because both the phosphopeptide and several proteotypic peptides corresponding to the protein could be relatively or absolutely quantified, thus determining the phosphorylation site occupancy and regulation. Overall, our data provide global starting points, and constraints, toward understanding the complexity of phosphorylation regulation in yeast and other organisms. In the future, the results should be complemented by similar data for specific cellular conditions, time courses, or small-molecule interventions, thereby sharpening—step by step—our view of the events in the phosphorylation network. The ensuing insights in general design rules and motifs in cellular information processing will be essential for our ability to develop kinase-based drugs in an informed way.