Tyrosine phosphoproteomics of fibroblast growth factor signaling: a role for insulin receptor substrate-4.

Signal transduction by receptor tyrosine kinases is initiated by recruitment of a variety of signaling proteins to tyrosine-phosphorylated motifs in the activated receptors. Several signaling pathways are thus activated in parallel, the combination of which decides the cellular response. Here, we present a dual strategy for extensive mapping of tyrosine-phosphorylated proteins and probing of signal-dependent protein interactions of a signaling cascade. The approach relies on labeling of cells with "heavy" and "light" isotopic forms of Arg to distinguish two cell populations. First, tyrosine-phosphorylated proteins from stimulated ("heavy"-labeled) and control samples ("normal"-labeled) are isolated and subjected to high sensitivity Fourier transform ion cyclotron resonance mass spectrometry analysis. Next, phosphopeptides corresponding to tyrosine phosphorylation sites identified during the tyrosine phosphoproteomic analysis are used as baits to isolate phosphospecific protein binding partners, which are subsequently identified by mass spectrometry. We used this approach to identify 28 components of the signaling cascade induced by stimulation with the basic fibroblast growth factor. Insulin receptor substrate-4 was identified as a novel candidate in fibroblast growth factor receptor signaling, and we defined phosphorylation-dependent interactions with other components, such as adaptor protein Grb2, of the signaling cascade. Finally, we present evidence for a complex containing insulin receptor substrate-4 and ShcA in signaling by the fibroblast growth factor receptor.

of the intrinsic tyrosine kinase in membrane-spanning receptors of the FGF receptor (FGFR) family. As receptors become activated, they autophosphorylate on several residues outside the kinase domain, thus allowing recruitment of cytosolic substrates and initiation of multiple downstream signaling pathways in parallel. The FGFR has been shown to initiate signaling through direct binding of substrates such as phospholipase C␥ (PLC␥) (1,2), FGFR substrate-2 (FRS2) (3,4), and ShcA (5). The end point of FGFR activation is defined in a cell-specific manner by the integration of these signaling pathways. A thorough understanding of cellular signaling processes requires the identification of the components of the entire signaling network followed by studies of the functional role of individual proteins. Large scale analysis of post-translationally modified (e.g. phosphorylated) proteins by mass spectrometry (MS) has gained momentum with the development of high resolution and high sensitivity hybrid mass spectrometers (e.g. quadrupole time-offlight (qTOF) instruments) and now constitutes a powerful functional proteomics platform for the analysis of protein networks of signaling processes (6).
Since signaling proteins are not all "turned off" in the quiescent cell, and the stoichiometry of activation may vary according to the stimulus applied, a major challenge in MS-based cell signaling analysis has been to develop techniques that allow comparison of the level of post-translational modifications between proteins from stimulated and unstimulated cell samples. Stable isotope labeling of proteins or peptides in combination with MS analysis allows direct comparison of protein levels between differentially labeled samples without the use of twodimensional gels. Stable isotope tags have been introduced into proteins in vivo through metabolic labeling (7)(8)(9) and in vitro into proteins or peptides via chemical reactions using isotopecoded affinity tag (ICAT) or similar reagents (10 -13). Our laboratory devised a whole cell [ 13 C]Arg metabolic labeling strategy (stable isotope labeling by amino acids in cell culture (SILAC)) that renders proteins from two cell populations (cultured in medium with "heavy" [ 13 C]Arg or "normal" [ 12 C]Arg, respectively) distinguishable from each other by an offset in mass (9). MS analysis discriminates between peptides from the two cell populations by the ϩ6.0-Da mass shift of the "heavy" Arg-containing peptides, and the isotopic ratio of [ 13 C]Arg/ [ 12 C]Arg peptides measures the relative protein level in the two samples. SILAC has recently been used in combination with affinity purification of proteins that associates with the Src homology 2 (SH2) domain of the signaling adaptor protein Grb2 to identify components of the signaling complex induced by epidermal growth factor (EGF) stimulation of HeLa cells (14).
Intracellular signaling events depend on tyrosine phosphorylation to modulate enzyme activity of target proteins and promote signal-dependent protein-protein interaction. Tyr(P) in particular sequence contexts binds phosphotyrosine binding (PTB) or SH2 domains and hereby recruits specific proteins to a signaling complex (15). Moreover, tight regulation by tyrosine phosphatases causes a very low level of tyrosine phosphorylation in inactive cells (16) and provides the foundation for the central role of this particular post-translational modification in cell signaling events. Thus, maps of tyrosine-phosphorylated proteins provide an extensive view of the components of signaling networks. We have previously applied qTOF MS to investigate the tyrosine phosphorylation in signaling by the EGF receptor (17) and the FGFR (18). In these studies, Tyr(P) immunoprecipitates from "activated" and control cells were resolved side-by-side by SDS-PAGE and stained for visualization of proteins, to allow detection of bands specific to the "activated" sample. These bands were excised from the gel and proteins were identified by qTOF MS/MS. However, this approach is impeded when several proteins are identified in one band, because it is not clear which one is phosphorylated. Moreover, the low level and substoichiometric phosphorylation of many signaling proteins require a large quantity of starting material to obtain sufficient protein for visualization on a gel. These limitations are becoming particularly apparent as the sensitivity of MS analysis equals or even exceeds conventional gel staining techniques. To circumvent these issues, we combined SILAC with immunoprecipitation (IP) of Tyr(P) proteins from pooled lysates of basic fibroblast growth factor (FGF2)stimulated and untreated cells to identify components of the FGFR-1 signaling cascade. Furthermore, MS analysis was performed on the newly released linear ion trap Fourier transform (LTQ-FT) mass spectrometer (19,20), which identified twice as many proteins as analysis performed on the qTOF instrument. 28 proteins were enriched in the phosphotyrosine IP as a consequence of FGF2 stimulation, and we propose these as signaling components of the FGFR-1 signaling pathway. We identified several proteins novel to FGFR signaling such as the insulin receptor substrate-4 (IRS-4). We used SILAC in combination with phosphopeptide affinity purification (21) to identify Grb2 as a FGF2 signal-dependent binding partner of IRS-4. Furthermore, IRS-4 efficiently co-immunoprecipitated with the docking protein ShcA, which was also identified in our phosphoproteomic analysis of FGF2-induced signaling, after FGF2 stimulation. Thus, we provide evidence for an IRS-4⅐ShcA⅐Grb2 signaling complex involved in FGFR-1-mediated signal transduction.
Cell Culture, IP, and Transient Transfections-Cells were cultured in serum-free medium for 24 h before treatment with 100 ng/ml human recombinant FGF2 (Invitrogen) for 5 min. Lysates were prepared using 1% Nonidet P-40 in PBS (pH 7.4) supplemented with 1 mM sodium orthovanadate and Complete protease inhibitors (Roche Applied Science). For SILAC analysis of phosphorylated proteins, ϳ5 ϫ 10 7 cells were harvested in 6 ml of lysis buffer. Tyr(P) IP, SDS-PAGE, and protein band visualization were performed as previously described (18). The lanes of Tyr(P) IP proteins were excised in 10 fractions (ϳ30 -200 kDa) for analysis by MS.
Nanoflow Liquid Chromatography Tandem Mass Spectrometry (Nano-LC-MS/MS)-Reverse-phase nano-LC-MS/MS was performed using an Agilent 1100 nanoflow LC system (Agilent Technologies Inc.) comprising a solvent degasser, a nanoflow pump, and a thermostated -autosampler. The LC system was coupled to a QSTAR Pulsar hybrid qTOF mass spectrometer (AB-MDS Sciex) or a 7-tesla LTQ-FT instrument (Thermo Electron), using a modified nanoelectrospray ion source (Proxeon Biosystems) interface. Binding and chromatographic separation of peptides was achieved in a 20-cm fused silica emitter (75-m inner diameter; Proxeon Biosystems) packed in-house with a methanol slurry of reverse-phase ReproSil-Pur C18-AQ 3-m resin (Dr. Maisch GmbH) and mounted in the nanoelectrospray ion source.
Briefly, the tryptic peptide mixtures were autosampled at a flow rate of 500 nl/min onto the packed column and then eluted at a flow rate of 250 nl/min. The peptides were separated using a linear gradient of 1.6 -24% acetonitrile in 0.5% acetic acid over 40 min and ionized by an applied voltage of 2.4 kV to the emitter.
Both mass spectrometers were operated in data-dependent acquisition mode to automatically switch between MS and MS/MS. For QSTAR analysis, survey MS spectra were acquired for 1 s, and the three most intense ions (doubly, triply, or quadruply charged) were isolated and sequentially fragmented for 1.5 s by low energy collision-induced dissociation. All MS and MS/MS spectra were acquired with the Q2pulsing function switched on and optimized for enhanced transmission of ions in the MS (m/z 300 -1300) and MS/MS (m/z 85-1300) mass ranges. LTQ-FT acquisition was designed to utilize the sequencing power of the ion trap with regard to sensitivity and speed, while taking advantage of the ultra-high mass accuracy and dynamic range of the Fourier transform ion cyclotron resonance (FT) detector. The FT-MS was used for acquisition of MS survey scans with a resolution of 100,000 at m/z 400 (945-ms detection time), whereas the linear ion trap simultaneously produced MS/MS spectra of the five most intense ions in each survey scan. Automatic gain control was used to accumulate ions for FT-MS survey scans (target value 10,000,000) and MS/MS acquisition (target value 2,000).
Database Searches: Peptide Identification-All MS/MS spectra files from each LC run were centroided and merged to a single file, which was searched using the Mascot Search Engine (Matrix Science) against the mammalian NCBI nonredundant database with carbamidomethyl cysteine as a fixed modification and oxidized methionine, phosphorylated tyrosine, and [ 13 C]Arg as variable modifications. Tryptic constraints were applied allowing up to two missed cleavages. Peptide searches were performed with an initial tolerance on mass measurement of 0.2 Da in MS mode and 0.2 Da in MS/MS mode for QSTAR data and 10 ppm in MS mode and 0.8 Da in MS/MS mode for LTQ-FT data.
Determination of Protein Ratios-[ 13 C]Arg/[ 12 C]Arg peptide ratios were measured for all identified Arg-containing peptides essentially as described (14), applying MSQuant (CEBI), which is open source software we have made available on the World Wide Web at msquant-.sourceforge.net/. The software handles both QSTAR and LTQ-FT raw files. Briefly, the "heavy" ([ 13 C]Arg) to "normal" ([ 12 C]Arg) peptide ratios are determined from centroided MS peaks and averaged over consecutive MS cycles for the duration of the respective LC-MS peaks in the total ion chromatogram. A protein's ratio is calculated as the mean of the ratios of Arg-containing peptides used to identify it. To avoid false positives, all protein ratios and S.D. values hereof were normalized by a factor that was equal to the average protein ratio of all proteins with ratios below 1.3.
Peptide Affinity Purification-Biotin-linked peptides were synthe-sized on a solid-phase peptide synthesizer (Intavis AG) using Rink amide resin (Merck). The synthesis and procedure for pull-downs were essentially as recently described (21). In short, a phosphorylated (Tyr(P) 921 ) and nonphosphorylated (control) version of the peptide corresponding to amino acids 914 -929 of IRS-4 ( 914 IRS-4 929 ) were synthesized. An ϳ250-nmol peptide was coupled to immobilized streptavidin (Ultralink; Pierce) and subsequently incubated with ϳ10 mg of precleared cell lysate from [ 13 C]Arg-labeled (Tyr(P) 921 peptide) or nonlabeled (control peptide) cell cultures for a minimum of 3 h at 4°C. Peptide-protein complexes were washed three times in lysis buffer, and bound proteins were resuspended in SDS-PAGE sample buffer. Elution fractions from the two peptide affinity purifications were combined, resolved by SDS-PAGE, and excised in 10 fractions for nano-LC-MS/MS analysis.

RESULTS
Tyrosine Phosphoproteomic Investigation of FGF2-induced Signaling-FGFR-1 signaling relies on the reversible tyrosine phosphorylation of downstream effector proteins. In the present study, we set out to identify proteins that are tyrosinephosphorylated or tightly associated with a tyrosine-phosphorylated protein after FGFR-1 activation by FGF2 treatment. The TREX293 cell line was stably transfected with cDNA encoding human FGFR-1, and the transfectant (TREX/FGFR-1) responded to treatment with FGF2 by strong phosphorylation of the FGFR-1.
Complete incorporation of [ 13 C]Arg in TREX/FGFR-1 cells was achieved by day 6 of culturing in [ 13 C]Arg labeling me-dium, and cells appeared morphologically identical to cells grown in "normal" ([ 12 C]Arg) medium. For FGF2 stimulation experiments, cells were cultured for 6 days in parallel in [ 13 C]Arg or [ 12 C]Arg medium, the last 24 h with serum depletion, before treatment of the [ 13 C]Arg-labeled cells with FGF2 for 5 min. The [ 13 C]Arg and [ 12 C]Arg lysates were pooled for IP with anti-Tyr(P) antibodies. Identification of the extracted proteins was achieved by a GeLC-MS/MS approach, which includes one-dimensional SDS-PAGE protein separation, in situ trypsin digestion, and peptide sequencing by liquid chromatography in combination with MS analysis. The isotopic ratio of Arg-containing peptides ([ 13 C]Arg peptide intensity/[ 12 C]Arg peptide intensity) was determined as a measure of the quantity of a protein originating from the FGF2-stimulated compared with the unstimulated sample in the Tyr(P) IP. Proteins with [ 13 C]Arg/[ 12 C]Arg peptide ratios Ͼ1 constitute components of the FGF2-induced cell signaling event (see Fig. 1a for an overview of the procedure). Four independent experiments were performed; three experiments were analyzed using qTOF MS, and one experiment was analyzed using the LTQ-FT MS. In total, the three qTOF runs combined identified more than 400 unique proteins, of which 19 had relative ratios higher than 1.3. However, in the single LTQ-FT analysis, we were able to identify and quantify a total of 5024 verified peptides identifying 860 unique proteins with a summed Mascot peptide score FIG. 1. Strategies for SILAC-based cell signaling investigations. One cell population is grown in medium containing "normal" arginine ([ 12 C]Arg), whereas another cell population is grown in medium containing "heavy" [ 13 C]Arg. Proteins from the two cell populations are distinguished during MS analysis by the ϩ6-Da mass shift of Arg-containing peptides from the "heavy" population. a, a tyrosine phosphoproteomic strategy was used to identify proteins involved in signaling initiated by FGF2 stimulation. Cells labeled with [ 13 C]Arg were stimulated with FGF2 for 5 min, whereas cells cultured in normal medium were left untreated. After lysis, proteins from stimulated and unstimulated cells were mixed 1:1 and immunoprecipitated using anti-Tyr(P) antibodies. The eluted proteins were processed and analyzed by MS. Proteins with [ 13 C]Arg/[ 12 C]Arg peptide ratios Ͼ1.3 are enriched in the tyrosinephosphorylated protein fraction after FGF2 treatment and were identified as components of the FGF2-stimulated signaling pathway. b, the function of tyrosine phosphorylation is often to promote signal dependent protein-protein interactions. Affinity purification using phosphorylated and nonphosphorylated versions of a peptide corresponding to a known phosphorylation motif as "bait," can therefore be used to identify direct and phosphorylation-dependent protein-protein associations. We used phosphorylated (Tyr(P) 921 ) and nonphosphorylated Ͼ35 (95% significance score threshold). Iterative calibration algorithms were used to achieve a final average absolute mass accuracy better than 1.7 ppm of the precursor ions analyzed by FT-MS, which is equivalent to a mass deviation of 0.0025 Da for a typical tryptic peptide with mass of 1,500 Da. The LTQ-FT MS/MS data set provided unique identification of nine proteins with [ 13 C]Arg/[ 12 C]Arg ratios above 1.3, in addition to verification of 18 of the 19 proteins determined from the combined qTOF analyses. For example, the docking protein ShcA was only identified from the LTQ-FT analysis. The WD repeat protein 6 (WDR6) was the only signaling component, which was identified only during the qTOF MS analyses. It was identified in each of the three individual qTOF analyses, and the reason for its absence in the LTQ-FT MS experiment is not clear. Fig.  2a depicts the number of "specific" proteins (i.e. enriched in the IP after FGF2 stimulation) versus the total number of identified proteins in the two types of MS analysis. The high sensitivity of the LTQ-FT MS instrument led to the identification of twice as many proteins as the qTOF instrument; however, only 4% of these proteins were "specific" (i.e. were significantly enriched in the Tyr(P) IP after FGF2 stimulation). Keeping in mind that MS analysis has reached sub-"gel staining" sensitivity, this finding emphasizes the need for quantitative labeling protocols such as SILAC or ICAT (10, 23) for comparative proteomics studies to distinguish true from false positives.
In each experiment, all quantifiable peptides from each protein were analyzed for isotopic ratio (Fig. 2b). For inclusion of a protein on the list of FGF2-signaling proteins, we routinely applied the following criteria. We required that the mean isotopic ratios of the proteins were greater than 1.3 and larger than 1 by twice the S.D. value (Fig. 2c). Protein ratios varied in some cases between experiments (especially for proteins with high enrichment ratios such as PLC␥). Most likely, this was due to small changes in the basal (unstimulated) activity of the FGFR-1 and changes in the signal-to-noise ratio between experiments (9). Tables I and II present FGF2-induced signaling proteins with their average [ 13 C]Arg/[ 12 C]Arg peptide ratios (Table I), and tyrosine phosphorylation sites identified during our MS analysis (Table II).
Phosphorylated (Tyr(P) 921 ) IRS-4 Peptide Binds Grb2, PLC␥, and Phosphatidylinositol 3-Kinase (PI3K) p85 Subunit-IRS-4 was identified in HEK293 cells as a prominent phosphorylated protein after treatment of cells with insulin and insulin-like growth factor-1 (24). Our MS analysis revealed that FGF2 signaling can also induce tyrosine phosphorylation of IRS-4, and Tyr 921 was identified as a phosphorylation site (Table II). To gain insight on the function of this particular phosphorylation site, we synthesized a biotin-linked peptide derived from IRS-4 ( 914 IRS-4 929 ) in a phosphorylated (Tyr(P) 921 ) and nonphosphorylated version to screen for signaling-dependent binding partners according to the strategy described in Ref. 21. The phosphorylated and nonphosphorylated peptides were used as bait and incubated with lysates of [ 13 C]Arg-and [ 12 C]Arglabeled cells, respectively. Peptides and direct binding partners were purified on streptavidin-coated beads, and the eluates were combined for one-dimensional SDS-PAGE and nano-LC-MS/MS analysis (see Fig. 1b for an overview of the procedure). Using this strategy, we identified the small adaptor protein Grb2 and PLC␥ with [ 13 C]Arg/[ 12 C]Arg binding ratios Ͼ10 and PI3K p85 subunit with ratio of ϳ5. Background binders were revealed by their ratios close to 1 (Fig. 3). These results indicate that FGF2-mediated phosphorylation of IRS-4 can induce specific interactions with Grb2, PLC␥, and PI3K. In accordance with our results, the sequence surroundings of Tyr(P) 921 (Tyr(P)-Val-Asn-Met) create consensus binding sites for the SH2 domains of Grb2 (Tyr(P)-X-Asn-X), PLC␥ (Tyr(P)-Val-X- The proteins with a 1:1 ratio either contain an unchanged FGF2-independent basal level of tyrosine phosphorylation or originate from background. The LTQ-FT MS analysis identified 30% more signaling proteins than the qTOF analysis and verified all but one signaling protein (indicated by an asterisk) identified in the three qTOF analysis runs combined. b, the panels show peptide doublets from selected proteins. The ratios reflect varying degrees of enrichment in the tyrosine-phosphorylated protein fraction after FGF2 stimulation. The peptide doublet in the bottom right panel shows a 1:1 peptide ratio, which indicates either that the corresponding protein was immunopurified nonspecifically or that its phosphorylation status was not changed by FGF2 stimulation. The top panel spectra are from the LTQ-FT MS analysis, and bottom panel spectra are from the QSTAR analysis. c, for each protein, two or more peptide ratios were measured. The mean peptide ratio (mean) and standard deviation (s.d.) were calculated. The values presented here are data for PLC␥ from four experiments. X), and the p85 subunit (Tyr(P)-X-X-Met) as determined from screening of degenerate phosphopeptide libraries (25). The Tyr(P) 921 motif is one of two sequences in IRS-4 that conform to the consensus sequence for Grb2 SH2 domain binding, and it is the highest scoring using the Scansite prediction engine (26).
IRS-4 Interacts with Grb2 upon FGF2 Stimulation-Whereas peptide affinity purification is a powerful method for screening for direct binding partners, it is still limited by the bait peptide sequence being out of proper protein context. To determine whether FGF2 treatment of TREX/FGFR-1 cells induces interaction between full-length IRS-4 and Grb2 in cell culture, we performed IP of IRS-4 from cell lysates and analyzed the eluates by Western blotting against Grb2. As expected, Grb2 was recruited by IRS-4 upon FGF2 stimulation (Fig. 4, lanes 1 and 2).
FRS2 Inhibits Phosphorylation and Grb2 Binding of IRS-4 -FRS2 is a prominent 90-kDa tyrosine-phosphorylated protein in FGF signaling (27). It contains a PTB domain (28) that interacts directly and constitutively with the FGFR-1 (3,4). FRS2 is essential for FGF-mediated neurite outgrowth and MAPK signaling in PC12 cells (29). Our phosphoproteomic analysis did not identify FRS2 in our samples, and we thus presume that TREX293 cells express low level, if any, FRS2. A distant relative of IRS-4, Grb2-associated protein 1 (GAB1), is recruited to the FGFR via a FRS2⅐Grb2 complex (30). To test if the FGFR recruits IRS-4 in a similar manner, we ectopically expressed FRS2 in TREX/FGFR-1 cells and subsequently immunoprecipitated IRS-4 to evaluate phosphorylation and Grb2 binding. If FRS2 is an intermediary in FGFR recruitment of IRS-4, we would expect overexpression of FRS2 to increase the phosphorylation and Grb2 association of IRS-4. However, whereas overexpression of FRS2 led to a high increase in FGF2-induced MAPK phosphorylation, it did not augment IRS-4 signaling. In fact, it led to a decrease in Grb2 binding and IRS-4 phosphorylation (Fig. 4). Thus, it appears that the FRS2 is not involved in engaging IRS-4 to FGFR-1 signaling.
ShcA Co-immunoprecipitates with IRS-4 -The SH2 and PTB domain-containing docking protein ShcA was identified in our phosphoproteomic LTQ-FT analysis with a high enrichment ratio in the FGF2-stimulated cells (Table I). Western blot analysis revealed that ShcA was indeed phosphorylated in TREX/FGFR-1 cells after treatment with FGF2 (Fig. 5). Moreover, ShcA co-immunoprecipitated efficiently with IRS-4 ( Fig.  5; compare the Western blots of ShcA from the IRS-4 IP with the total lysate), and FGF2 stimulation led to an increase in ShcA associated with IRS-4 (Fig. 5). The ShcA/IRS-4 association observed in unstimulated cells may be due to a basal activation of the FGFR-1 or alternatively due to interactions independent of FGFR-1 activation. The binding is not due to nonspecific interactions of ShcA with immunoglobulins or the protein G-Sepharose matrix, since IP with control immunoglobulins did not lead to ShcA co-precipitation (Fig. 5). DISCUSSION FGFs and FGFRs are critical during embryonic development with prominent roles in CNS (31) and limb development (32). Furthermore, FGFs are potent angiogens and neurotrophic factors, and deregulated FGF/FGFR signaling is associated with a variety of malignancies. The pleiotropic effects are underscored by cell culture studies in which FGF stimulation,  depending on cell context, induces proliferation, differentiation, migration (33), or survival (31). The cellular outcome is decided by the interplay of effector molecules that are recruited by the receptor upon activation. In the present study, we used SILAC, a whole cell isotopic labeling method for comparative MS analysis (8,9), in combination with Tyr(P) IP to investigate the FGF2-induced signaling cascade. The analysis identified proteins that are modified by tyrosine phosphorylation or tightly associated with a phosphorylated protein after FGF2 stimulation of TREX/FGFR-1 cells. We used complementary MS instruments for sample analysis to compare the performance of the newly released LTQ-FT mass spectrometer with qTOF MS. The three qTOF experiments combined resulted in identification of a total of 444 unique proteins with a Mascot protein score higher than 35 and an average absolute mass accuracy of 16.1 ppm. In contrast, the single LTQ-FT analysis identified 860 unique proteins with an average absolute mass accuracy of 1.7 ppm. As a consequence, we identified 27 FGF2 signaling proteins in the LTQ-FT MS analysis encompassing all but one of the 19 proteins identified in the three qTOF experiments. Database search peptide assignments were of increased confidence in the LTQ-FT experiments due to the superior mass accuracy. In total, FGF2 stimulation led to an increased level of 28 proteins in the Tyr(P) IP, which provide a broad view of the FGF2-activated signaling network in a specific cell context (Tables I and II). The specificity of our quantitative proteomics assay is underscored by the Ͼ800 proteins that were nonspecifically immunopurified, either due to basal FGF2-independent phosphorylation or unspecific binding to the phosphotyrosine antibodies or Sepharose matrix. A Link to Actin Cytoskeletal Dynamics-The Rho GTPases Rho, Rac, and Cdc42 are key regulators of actin cytoskeletal dynamics involved in cell motility. Rac and Cdc42 Rho GTPases control actin polymerization involved in lamellipodial and filopodial cell protrusions, whereas Rho regulates the assembly of contractile actin/myosin filaments (34). Rho GTPases are switched "on" in the GTP-bound state, which is promoted by guanine nucleotide exchange factors. The p21-activated kinase (PAK)-interacting exchange factors (PIX␣ and PIX␤) are guanine nucleotide exchange factors for the Cdc42 and Rac GTPases (35). PIX proteins bind PAK (36), a Rac/Cdc42 effector protein involved in actin cytoskeletal dynamics and cell motility, and are important for localization of PAK to focal complexes (35). The GIT1 and GIT2 (also known as Cat-1 and Cat-2, respectively) are GTPase-activating proteins for the Arf family of small GTPases (37,38). GIT proteins bind PIX (37) and associate with paxillin (39 -41), a focal adhesion complex scaffold protein. Paxillin, GIT, PIX, and PAK co-localize in the cell, and reports of mutual functional dependence and synergy have led to the suggestion that they exist as a multiprotein signaling module (42). The complex is recruited to the membrane by Rac and participates in actin cytoskeletal remodeling, lamellipodia formation, and cell migration (42,43). In addition to identifying paxillin, GIT1, GIT2, ␤PIX, and ␣PIX in the  , lanes 1 and 2), and these events were partially abrogated by transient expression of FRS2 (top panels, lane 3). FGF2 stimulation led to an increase in MAPK phosphorylation that was greatly increased upon FRS2 expression (bottom panels).
fraction of Tyr(P) proteins after FGF2 stimulation, we also mapped tyrosine phosphorylation sites Tyr 118 and Tyr 88 in paxillin (Table II). Phosphorylation of Tyr 118 in paxillin is mediated by the nonreceptor tyrosine kinase focal adhesion kinase (FAK) and creates a binding site for the SH2 domain of adaptor protein Crk (44,45), which is important for paxillinmediated lamellipodia formation (46,47). FAK itself also plays a role in lammelipodia formation by the paxillin/GIT/PIX/PAK module (48), possibly by creating binding sites for Crk on paxillin (47). In accordance, we detected FAK activation, as indicated by Tyr 576 phosphorylation in the catalytic domain (49), downstream of FGF2 stimulation (Table II). In conclusion, our study implies a role for the FAK⅐paxillin⅐GIT⅐PIX⅐PAK complex in FGF2 signaling, providing a link to actin cytoskeletal rearrangements with functional relevance to cell migration and very likely neurite outgrowth, where roles for PAK and ␤PIX have been established in FGF-induced neurite extension in PC12 cells (50).
A Role for IRS-4 in FGFR-1 Signaling-FGF2 stimulation also led to phosphorylation of IRS-4, a docking protein of the IRS family (24). IRS-4 displays limited overall sequence iden-tity (Ͻ30%) with IRS-1 and -2, although high sequence conservation is found in the canonical IRS PTB and pleckstrin homology domains. IRS-4 is phosphorylated in HEK293 cells after stimulation with insulin and insulin-like growth factor-1 and co-immunoprecipitates with Grb2 and PI3K (24,51). We found that FGFR-1 signaling leads to tyrosine phosphorylation of the IRS-4 in TREX/FGFR-1 cells, and the MS analysis immediately yielded one tyrosine phosphorylation site, Tyr 921 . Peptide affinity purification using isotopically labeled cells (SILAC) and subsequent MS analysis revealed highly specific interaction of the phosphorylated (Tyr(P) 921 ) 914 IRS-4 929 peptide with Grb2, PLC␥, and PI3K p85. Grb2 links receptor tyrosine kinases to Ras signaling (52); however, the FGFR-1 itself does not contain Grb2 binding sites and thus relies on docking proteins for recruitment of this adaptor. Co-IP from FGF2-stimulated cells confirmed IRS-4/Grb2 interaction and thus identifies IRS-4 as a novel docking protein in FGFR-1 signaling that relays Grb2 binding.
To our knowledge, there have been no previous reports on IRS proteins in FGF signaling, but activation of the receptor tyrosine kinases for brain-derived neurotrophic factor (53) and EGF (54,55) can, under some conditions, lead to phosphorylation of IRS-1 and -2. It was proposed that consensus PTB domain binding motifs (-Asn-Pro-X-Tyr(P)-) in the respective receptor tyrosine kinases serve as binding sites for the IRS proteins, and it has been shown that an EGF receptor depleted of -Asn-Pro-X-Tyr(P)-motifs did not phosphorylate IRS1 and IRS2 (54). However, the FGFR-1 does not contain the IRS PTB domain binding consensus sequence, so another mechanism must exist for association of this kinase with IRS-4. The docking protein FRS2 is a prominent effector protein in FGFR signaling, and it contains a PTB domain that binds to an unorthodox nonphosphorylated sequence in the juxtamembrane domain of the FGFR (4). Since overexpression of the FRS2 diminished IRS-4 phosphorylation and Grb2 binding in our set-up, one possibility is that the two docking proteins compete for the same binding motif in the FGFR-1. However, our attempts to co-immunoprecipitate the FGFR with IRS-4 revealed very weak binding (data not shown), and biophysical studies have predicted that the IRS PTB domain will not recognize the FRS2 binding sequence in the FGFR-1 (56,57). On the other hand, GAB1, a distant relative of the IRS family of docking proteins (58), activates PI3K in FGFR signaling after being recruited into a ternary complex consisting of GAB1⅐Grb2⅐FRS2. GAB1 binds the N-terminal SH3 domain of Grb2, and Grb2 simultaneously couples to phosphorylated FRS2 via its SH2 domain, after FGF2 stimulation (29,30). It is possible that IRS-4 is involved in FGFR-1 signaling in a similar manner to GAB1. However, our MS analysis did not reveal any FRS2 in the TREX/FGFR-1 fraction of tyrosine-phosphorylated proteins, indicating that TREX293 cells express low level, if any, FRS2. This correlates with the lack of detection of endogenous FRS2 by immunoblotting in previous FGFR signaling studies using 293 cells (28,59). Ectopic FRS2 expression in TREX/FGFR-1 cells induced a marked increase in FGF2-mediated Erk activation; however, IRS-4 tyrosine phosphorylation and its association with Grb2 were diminished. This finding argues against a role of FRS2 in recruiting IRS-4 to the FGFR-1 and suggests a mechanism distinct from FGFR-1-mediated GAB1 activation.
IRS-4 and ShcA Interact in FGFR-1 Signaling-We found the SH2 and PTB domain-containing docking protein ShcA highly enriched in the phosphorylated fraction of proteins after FGF2 treatment. There are reports of ShcA phosphorylation in FGFR signaling (27,28), but its place in this signaling cascade is incompletely resolved. It has been shown that ShcA binds to an FGFR-1-derived phosphopeptide (a motif containing Tyr(P) 728 ) (5). However, a mutant Y728F FGFR-1 was still capable of activating ShcA (60), so additional recruitment mechanisms may apply. The fact that ShcA and IRS-4 had similarly high enrichment ratios in the FGF2-stimulated pool of phosphorylated proteins led us to speculate that the two proteins may be part of the same signaling branch from the FGFR-1. Thus, ShcA could be involved in recruiting IRS-4 to FGFR-1 signaling. In accordance with this hypothesis, we observed co-IP of ShcA with IRS-4, and the interaction was increased in FGF2-stimulated cells. This is somewhat surprising, since other studies have failed to detect ShcA interactions with IRS-1 and IRS-2 proteins by co-IP (61,62). However, one study using the yeast two-hybrid system showed direct interaction between IRS-1 and ShcA, and the authors confirmed their observations by affinity purification experiments (63). Binding involves the ShcA PTB domain, but although interaction was highly increased after phosphorylation of IRS-1, a significant degree of binding was detected between the proteins in the unphosphorylated state (63). The discrepancy between the IP studies and the yeast two-hybrid study may be ascribed to low affinity of ShcA/IRS-1 interaction. However, ShcA was efficiently co-immunoprecipitated with IRS-4 (Fig. 5b), which indicates that the IRS-4⅐ShcA complex is more stable than IRS-1/ShcA association. It is of particular interest that tyrosines and their immediate sequence surroundings in the region of IRS-1 responsible for ShcA binding (amino acids 583-661) (63) are well conserved in IRS-4 (Tyr 700 , Tyr 717 , and Tyr 743 corresponding to Tyr 608 , Tyr 628 , and Tyr 658 in IRS-1), whereas the remaining part of the region exhibits low conservation. In conclusion, our analysis reveals an IRS-4⅐ShcA signaling complex that promotes FGFR-1 signaling independently of FRS2. ShcA and IRS-4 possibly synergize in promoting association with the FGFR-1. ShcA may directly bind the FGFR-1 via interaction with phosphorylated Tyr 728 as previously described (5), whereas the membrane phospholipid-binding pleckstrin homology domain of IRS-4 may target the complex to the juxtamembrane area in proximity of the FGFR-1 cytoplasmic region.
The function of the IRS-4⅐ShcA signaling complex was not studied in detail, but our results indicate that downstream signaling may proceed via the adaptor protein Grb2. Although FRS2, which contains four potential Grb2 binding sites, has been shown to be important for FGFR-mediated Grb2-Ras activation in PC12 cells and mouse embryonic fibroblasts (29,59), there is also evidence for FRS2-independent FGFR signaling. In a murine brain endothelial cell line (IBEC) it has been shown that FGF-induced differentiation involves Ras signaling and ShcA activation but not FRS2 signaling (64). In TREX/ FGFR-1 cells, MAPK phosphorylation was weak after FGF2 stimulation but was significantly increased after transfection with FRS2. This indicates that MAPK activation may not be the primary end point of IRS-4⅐ShcA signaling. Grb2 may consequently serve other purposes in the complex, one possibility being that it strengthens the IRS-4/ShcA interaction in a similar manner to what was described for the FRS2/GAB1 interaction (see above) (30).
FGFR-1 Signaling Attenuation-Signaling cascades are regulated by activation as well as deactivation steps. In our phosphoproteomic investigation, we identified Cbl-b, SHIP2, and ODIN, all of which are proteins that may be involved in FGFR-1 signal attenuation and termination. Cbl-b is a ubiquitin ligase, which mediates ubiquitination of activated receptor tyrosine kinases (65). SHIP2 is an SH2 domain-containing inositol phosphatase that has been shown to negatively regulate insulin signaling, presumably by reducing the level of phosphatidylinositol 3,4,5-trisphosphate in the cell membrane (65). Recently, ODIN (ankyrin repeat and SAM domain-containing protein 1) has been shown to attenuate EGF and platelet-derived growth factor signaling; however, the mechanism remains to be established (66).
Proteins Novel in FGFR-1 Signaling-We identified a number of proteins that are novel in FGFR-1 signaling. These proteins include the IRS-4, GIT proteins, ODIN, and SHIP2, which were described above and PSTPIP 2 (Pro-Ser/Thr phosphatase-interacting protein 2). Moreover, our study revealed involvement of annexins (VII and XI) in the signaling cascade. Annexins are Ca 2ϩ -binding proteins that are targeted to membranes by associating with phospholipids. They are involved in membrane fusion and trafficking events such as exo-and endocytosis and are considered regulators of Ca 2ϩ homeostasis (67). Furthermore, annexin VII is considered to be an atypical G protein in which Ca 2ϩ binding stimulates GTPase activity, and it may serve as a Ca 2ϩ /GTP sensor in exocytotic events (68). Annexin VII and XI have been identified as tyrosinephosphorylated proteins in cells stimulated with platelet-derived growth factor and EGF growth factor (69); however, the functional context of the phosphorylation is speculative. Tyrosine phosphorylation of other members of the annexin family has been shown to modulate the phospholipid binding characteristics of annexins and may be involved in the internalization and sorting process of insulin and EGF receptors (70,71).
Finally, we identified two uncharacterized proteins downstream of FGF2 stimulation: liver-specific bHLH-Zip transcription factor (LISCH7) and WDR6 and the scarcely characterized NS1-associated protein 1/HnRNP Q. Bioinformatic resources (SMART (available on the World Wide Web at smart.emblheidelberg.de) (72) and TMHMM and SignalP (www.cbs.dtu.dk) (73)) predict LISCH7 to be a transmembrane protein with an extracellular immunoglobulin-like domain. WDR6 belongs to the WD40 repeat superfamily. A WD40 repeat consists of four ␤-strands, and clusters of at least four WD40 repeats make up ␤-propeller structures that constitute platforms for protein interactions (74). The WDR6 protein remains functionally uncharacterized, but interestingly, it was recently found constitutively associated with IRS-4 immunopurified from HEK293 cells (75). LISCH7 (FGF2/control ratio 7.25) and WDR6 (FGF2/control ratio 3.28) were both highly enriched in the Tyr(P) IP from FGF2-stimulated cells, which serve as an additional indication of a biological role for these proteins in FGF signaling. NS1-associated protein 1 is an RNA-binding protein (76), which was previously identified in signaling initiated by overexpression of the FGFR-1 where it was phosphorylated in the RNA binding domain (18).
Perspectives-In the present study, we introduce a powerful ion trap Fourier transform MS instrument, LTQ-FT MS, for functional proteomics investigations. The instrument provides attomole sensitivity, and the peptide ions can be measured with low to sub-ppm mass accuracy, which dramatically improves confidence in proteins identified via database searching (19). These features make this instrument ideally suited for proteomic signal transduction analysis, where the low level and substoichiometric post-translational modifications of the proteins involved can hamper detection by MS. The sensitivity of the LTQ-FT MS is illustrated by the fact that we used only ϳ5 ϫ 10 7 cells (confluent cultures from two T180 culture flasks) in our tyrosine phosphoproteomic experiment to identify 27 signaling proteins. In order to fully benefit from high sensitivity MS in signaling proteomics, one must employ a quantitative labeling approach such as SILAC to distinguish background proteins (in our case Ͼ800) from "true" signaling proteins (in our case 28) (see Fig. 2a). This is pertinent, considering that the sensitivity of the MS analysis has surpassed conventional gel staining techniques such as silver staining. The sensitivity and selectivity of the method may soon allow studies with primary cell cultures.
Receptor tyrosine kinases initiate multiple intersecting signaling pathways with overlapping downstream targets (such as transcriptional regulation), and the signaling cascade is regulated by the cellular wiring (i.e. expression of components of the signaling cascade). Cytosolic signaling proteins are limited in number, and most are promiscuous by serving as more or less common effectors to different receptors. Thus, specificity in signaling is achieved through the formation of signaling networks, shaped by interactions between individual signaling proteins, and regulated by expression levels and accessibility (77). Due to this complexity of signaling, many signaling cascades show redundancy, thus complicating the "classical" approach of characterizing a pathway of one protein at a time. We believe the dual MS-based protocols for tyrosine phosphoproteomic analysis presented here provide a powerful discovery tool for unbiased mapping of entire signaling cascades by providing information on individual proteins and the complexes of which they are components. Thus, they complement "classical" biochemistry by providing leads for characterization and by providing an overview of signaling proteins, yielding information on functionality and putative points of signaling redundancy.