Temporal Perturbation of Tyrosine Phosphoproteome Dynamics Reveals the System-wide Regulatory Networks *S

Signal transduction systems are known to widely regulate complex biological events such as cell proliferation and differentiation. Because phosphotyrosine-dependent networks play a key role in transmitting signals, a comprehensive and fine description of their dynamic behavior can lead us to systematically analyze the regulatory mechanisms that result in each biological effect. Here we established a mass spectrometry-based framework for analyzing tyrosine phosphoproteome dynamics through temporal network perturbation. A highly time-resolved description of the epidermal growth factor-dependent signaling pathways in human A431 cells revealed a global view of their multiphase network activation, comprising a spike signal transmission within 1 min of ligand stimulation followed by the prolonged activation of multiple Src-related molecules. Temporal perturbation of Src family kinases with the corresponding inhibitor PP2 in the prolonged activation phase led to the down-regulation of the molecules related to cell adhesion and receptor degradation, whereas the canonical cascades as well as the epidermal growth factor receptor relatively maintained their activities. Our methodology provides a system-wide view of the regulatory network clusters involved in signal transduction that is essential to refine the literature-based network structures for a systems biology analysis.

It is well known that the signal transduction system within a cell leads to the determination of diverse cell fates, such as proliferation, differentiation, or apoptosis (1,2). To date, the extensive analyses that have been performed on cell signaling under a variety of experimental conditions have revealed many of the key molecules and related events that result in various biological effects. Regarding the epidermal growth factor (EGF) 1 receptor pathway, which is one of the most intensively studied signaling systems, the related literature has especially been accumulated and even integrated into a large scale interaction map in a recent study (3). In addition to a series of biochemical analyses of each signaling molecule, recent approaches using protein microarrays and mass spectrometry have also added a global view of ErbB signaling initiation in vitro (4,5). Such an integrative view of this biological system, however, cannot provide a fundamental theory on how the entire signaling network is regulated without analyzing their dynamic behaviors at the system level.
Recent technological advances in mass spectrometrybased proteomics have enabled us to make a large scale identification of signaling molecules through the enrichment of phosphorylated proteins or peptides (6 -11). In combination with protein/peptide labeling strategies based on stable isotope labeling by amino acids in cell culture (SILAC) (12) or by using iTRAQ (isobaric tags for relative and absolute quantitation) reagents (13), their time course activation profiles have also been generated through the relative quantitation of labeled peptides in mass spectra (14 -17).
In this study, we established an integrated framework for analyzing the dynamic behavior of the SILAC-encoded tyrosine phosphoproteome through temporal network perturbation. To acquire highly time-resolved dynamic data in a high throughput manner, we developed a method for the efficient enrichment of tyrosine-phosphorylated signaling molecules to greatly shorten the time for sample preparation and subsequent LC-MS/MS analysis (see Fig. 1a). After relative quantitation based on the SILAC-encoded data is automatically performed using the AYUMS algorithm (18), the multiple quantitative data for different time points are integrated into fine activation profiles through data normalization (see Fig.  1b). A precise and highly time-resolved description of the network dynamics provides a theoretical basis for estimating the temporal perturbation effect in relation to specific signaling molecules, allowing us to obtain a system-level view of the regulatory relationships in transducing signals.
Here we applied our methodology to the signaling networks that work upon EGF stimulation in human A431 cells as a model system. A time-resolved description of their tyrosine phosphoproteome dynamics revealed the property of multiphase network activation, comprising a spike signal transmission within 1 min of ligand stimulation followed by the prolonged activation of multiple Src-related molecules. Temporal perturbation of Src family kinases with PP2 in the prolonged activation phase enabled us to clearly distinguish between sensitive and robust pathways in response to this treatment, providing a system-level view of the Src function in the EGF signaling of A431 cells.
Sample Preparation for Mass Spectrometric Analysis-Immunoprecipitation of tyrosine-phosphorylated protein complexes was performed as described previously (14). After washing the immunoprecipitates quickly with a lysis buffer containing no detergents, elution of the target molecules was conducted with 25 mM phenyl phosphate (19). The eluted protein complex (ϳ1 g) was digested with trypsin directly in the solution, desalted with ZipTip (C 18 ; Millipore), and finally concentrated to a volume of as little as 20 l to inject into the nano-LC system.
Mass Spectrometry and Protein Identification-After applying the peptide mixture to a C 18 column (800-m inner diameter ϫ 3 mm long), reversed-phase separation of the captured peptides was done on a column (150-m inner diameter ϫ 75 mm long) filled with HiQ-Sil C 18 (3-m particles, 120-Å pores; KYA Technologies) using a direct nanoflow LC system (Dina, KYA Technologies). The peptides were eluted with a linear 5-65% gradient of acetonitrile containing 0.1% formic acid over 120 min at a flow rate of 200 nl/min and sprayed into a quadrupole time-of-flight tandem mass spectrometer (Q-Tof-2, Micromass) (20,21). Acquisition of MS/MS spectra was performed using the following parameters: dynamic exclusion time, 120 s; duty cycle, 2 s; mass tolerance, 0.1 Da. The MS/MS signals were then converted to text files by MassLynx (version 3.5, Micromass) and processed against the RefSeq (National Center for Biotechnology Information (NCBI)) human protein database (39,397 sequences as of July 3, 2006) (22) using the Mascot algorithm (version 2.1.02, Matrix Science) with the following parameters: variable modifications, oxidation (Met), N-acetylation, pyroglutamination (Gln), phosphorylation (Ser, Thr, and Tyr), and stable isotopes of Arg 6 and Arg 10 (Arg); maximum missed cleavages, 3; peptide mass tolerance, 500 ppm; MS/MS tolerance, 0.5 Da. Protein identification was based on the criterion of having at least one MS/MS data with Mascot scores that exceeded the thresholds (p Ͻ 0.05). A randomized decoy database created by a Mascot Perl program estimated a false discovery rate at 0.93% for all the identified peptides.
Protein Quantification and Data Normalization-For each LC-MS/MS experiment, activation changes regarding all the peptides with Mascot scores (м20) at 1 and t min after EGF stimulation were automatically calculated as relative ratios to the state before stimulation using the AYUMS algorithm (version 1.00) (18) where the value t corresponded to either 0.5, 2, 5, 10, 15, 20, 25, or 30 min. Relative quantification by AYUMS was performed based on their peak intensities of SILAC-encoded peptides. For each protein identified, the relative quantitation value for each peptide data was all normalized to the average value of -fold activation for 1 min of ligand stimulation. The average ratio and its standard deviation were then calculated for each stimulation time based on the normalized quantitation values, yielding time series activation profiles. We assessed the quantification variability based on all the -fold changes at 1 min regarding the proteins with the quantitative data on at least three peptides per sample. The average relative standard deviation was 24.8%, leading to more than a 95% statistical confidence in the selected threshold of 1.5-fold.
Network Perturbation Analysis-Serum-starved A431 cells labeled with either L-arginine or L-[U-13 C 6 15 N 4 ]arginine were treated with 100 ng/ml EGF as described above. After stimulation for 1 min, Src family kinase inhibitor PP2 (Calbiochem) was added to the Arg 10-labeled cells at a final concentration of 5 M and incubated for another 1, 4, 9, and 14 min, respectively. The carrier solution for PP2 was added to the Arg 0-labeled cells and incubated for the same interval as a control. The cells were then lysed, mixed equally, and subjected to nano-LC-MS/MS analysis in the same manner as described above. For this analysis, we used a new mass spectrometer (Q-STAR Elite, Applied Biosystems) and set the parameters to 60 s for dynamic exclusion, Smart IDA (information-dependent acquisition) mode for duty cycle, and 100 ppm for mass tolerance. The perturbation effect with PP2 was represented as a relative ratio of Arg 10 to Arg 0 for each protein using the MSQuant program (version 1.4.3) (23).

Identification of Tyrosine-phosphorylated Molecules in
EGF-stimulated Human A431 Cells-Because tyrosine phosphorylation is known to play a key role in early EGF signaling, we tried describing the time-resolved dynamics of tyrosinephosphorylated proteins for 30 min after EGF stimulation using the SILAC technology (Fig. 1a). To make a comprehensive identification of the target signaling molecules, tyrosinephosphorylated protein complexes were enriched by affinity selection with anti-Tyr(P) antibodies. The peptide mixture generated from the protein complexes through tryptic digestion was then subjected to nano-LC-MS/MS analysis. Our tyrosine phosphoproteome analysis of EGF-stimulated A431 cells generated 28,248 MS/MS spectra in total. A database search against the RefSeq human protein database (22) led to the identification of 3,730 peptides, which were ultimately assigned to 790 unique peptides (supplemental Table 1). Among a total of 136 proteins identified, 77 proteins were identified by more than two peptides, whereas 59 proteins were supported by single peptide identification (supplemental Table 1).

Generation of Activation Profiles through Comprehensive Quantification and Data Normalization of Time Series -Fold
Changes-Regarding the proteins identified based on the arginine-containing peptide(s), we performed automatic relative quantitation of the SILAC-encoded MS data using the AYUMS algorithm (18). To normalize the data sets for different time intervals of EGF stimulation, the activation change value for each peptide data was all normalized to the average -fold of each protein for 1 min of ligand stimulation. The average ratio and its standard deviation were then calculated for each stimulation time based on the normalized -fold values to generate time series activation profiles (Fig. 1b).
Among the 60 proteins quantified, 56 molecules showed more than 1.5-fold changes upon EGF stimulation at each maximum activation or down-regulation point (supplemental Table 2). Functional classification of these EGF-induced proteins indicated the prominence of the molecules related to cell adhesion and cytoskeletal rearrangement as well as canonical signaling molecules (supplemental Table 2), reflecting the previous report regarding EGF-induced regulation of cell-cell contacts in human A431 cells (24).
The time course profiles for the 19 signaling molecules measured at all the time points are shown in Table I and Fig. 2. As shown in Fig. 2a, spike activation of the EGF receptor was observed within 1 min of ligand stimulation followed by sustained signaling with the -fold change lower than the first transmission. The other signaling molecules were also subjected to the fast activation upon EGF stimulation, but their dynamic behaviors in the prolonged activation phase depended on molecules (Fig. 2, a-d). Notably the effectors related to cell adhesion, such as the integrin and catenin families, mostly displayed sustained activation profiles (Fig. 2, b and c). As some of these adhesion molecules as well as the EGF receptor are classified as Src substrates in the public database (25), our time-resolved data strongly suggested the widespread cooperation of Src family kinases with the maintenance of the prolonged activation phase of EGF signaling in human A431 cells.

Network Perturbation of the EGF-dependent Prolonged Activation Phase with the Src Family
Inhibitor PP2-Next to analyze the network-wide role of Src family kinases in the prolonged activation phase of EGF-stimulated A431 cells, we tried to make a quantitative description of the regulatory effect on the tyrosine phosphoproteome by network perturbation with the Src family inhibitor PP2. We performed time course PP2 treatment from 1 min after the EGF input was added when the phase of the EGF signaling shifted from the spike activation to the prolonged activation. The perturbation effect was described as a relative ratio to the activation state with no PP2 treatment (supplemental Table 3).
The result indicated widespread down-regulation of the molecules related to cell adhesion and receptor degradation as well as some other EGF receptor effectors (Vav2, Shp2, and PLC␥1), whereas the canonical cascades (Shc, Grb2, and ERK2) as well as the EGF receptor relatively maintained their activities (Table II). Our perturbation analysis revealed many signaling effectors that were not described as c-Src interactors in a series of previous studies (Table II) prolonged activation was prominently observed regarding cell adhesion and structural molecules as well as the EGF receptor. As some of the cell-cell contact molecules, such as caveolin 1, catenin ␦1, and junction plakoglobin, as well as the EGF receptor are considered to be Src substrates in the context of signal transduction (25), our time-resolved descrip-tion of signaling dynamics indicated a probability that Src family kinases play an influential role in regulating the prolonged activation phase of EGF signaling in A431 cells. The temporal perturbation of Src family kinases with the corresponding inhibitor PP2 demonstrated the strong correlation of Src family kinases with cell migration-associated pathways,  whereas it had only a slight impact on the EGF receptor activity. We also observed a strong down-regulation of the receptor degradation pathways through the inhibition of Src family kinases, reflecting the previous evidence of the effect of c-Src on EGF receptor endocytosis (26). Thus, our network perturbation strategy based on a highly time-resolved description of tyrosine phosphoproteome dynamics provided a network-wide view of the pathway clusters controlled by Src family kinases in A431 cells. Based on some previously described network models (27)(28)(29)(30)(31)(32)(33), our methodology enables us to refine such general network structures into a "cell typespecific" architecture. We expect that mathematical analysis based on these cell type-specific network models will lead us to the efficient identification of potential drug targets for specific disease conditions and also enable us to theoretically estimate the effect of the corresponding drugs on a networkwide scale prior to clinical application. Acknowledgments-We thank K. Kudo for technical support in operating the nanoflow LC system. We are also thankful to A. Saito and T. Adachi for processing proteomics data. * This work was supported by a grant-in-aid for Scientific Research on Priority Areas from the Ministry of Education, Culture, Sports, Science and Technology of Japan. The costs of publication of this  (25) are indicated in underlined italics. b In every case, PP2 treatment was performed after 1 min of EGF stimulation. For each time period of EGF stimulation, the perturbation effect on each signaling molecule is described as a ratio relative to the activation state with no PP2 treatment. c ND, not determined.
article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.