Phosphoproteomic fingerprinting of epidermal growth factor signaling and anticancer drug action in human tumor cells

Recent breakthroughs in cancer therapeutics (1–3) have drawn attention to the importance of signaling proteins as targets for rational drug development (4–6). Compared with traditional nonselective cancer treatments, drugs that disrupt cancer-specific signaling networks benefit from a higher ratio of efficacy to toxicity, resulting in improved patient acceptability (7–9). Enzymes and receptors represent the most common drug targets identified to date, but upstream proteins of this kind are often coexpressed in normal tissues and may be responsible for unforeseen toxicities (10, 11). Distinct combinatorial patterns of downstream signaling amplification in the cancer cell, as might be detectable by identifying the modified substrates of rogue receptors or enzymes (12–16), could reveal new anticancer targets and/or interventions (17, 18).

Tyrosine phosphorylation is the most potent signaling mechanism regulating mammalian cell proliferation. Inducing as it does a profound hydrophilic extrusion of local peptide conformation (19), tyrosine phosphorylation is readily detectable by specific antibodies (20). Generic phosphotyrosine antibodies may be used to map and/or isolate critical signaling proteins and substrates that are inducibly modified by tyrosine kinases (21–26). However, the practical utility of this approach has thus far been limited by technical shortcomings, including the sensitivity and specificity of affinity purification, the possibility of secondary binding or dimerization of heterologous molecules, and false-positive artifacts induced by abundantly expressed nonphosphorylated proteins.

Representing as it does the complex end product of a multistep biological process, cancer has been a central focus for discovery-based experimental platforms of this kind (27–29). Conventional proteomic comparisons have revealed important differences in expression between normal and preinvasive tissues (30), between normal and malignant primary tumors (31–34), and between different malignant primary tumors (35). Moreover, such studies have also cast reasonable doubts on the significance of reported cancer-specific alterations in mRNA expression (30). Global proteomic comparisons of this kind remain of little routine use at present, however, reflecting both restrictions of dynamic range and an inability to prioritize importance to the numerous changes in expression detected.

We previously developed a phosphoantibody-based technology for detecting functionally important protein modifications (36) and have since applied this approach to the combinatorial detection of such events in human tumors of varying phenotype (37–39). Here, we extend this approach by combining generic phosphotyrosine antibodies with two-dimensional PAGE to capture and separate, respectively, both upstream (effector) and downstream (substrate) signaling proteins in human cancer cells.

Monoclonal peroxidase-conjugated phosphotyrosine antibody (PY20) was purchased from Transduction Laboratories (Lexington, KY) and conjugated protein A-Sepharose was purchased from Zymed (San Francisco, CA). Epidermal growth factor (EGF), transforming growth factor-α (TGFα), platelet-derived growth factor (PDGF), phenylphosphate, thiourea, ethanolamine, dimethylpimelimidate (DMP), Triton X-100, Igepal, EDTA, DTT, and diaminobenzidine (DAB) were purchased from Sigma Chemical Co. (St. Louis, MO). Enhanced chemiluminescence (ECL) reagent and immobilized pH gradient (IPG) electrophoretic strips and buffers were purchased from Amersham Biosciences (Uppsala, Sweden), SyproRuby protein gel and blotting stains were from Molecular Probes (Eugene, OR), Bradford's reagent and prestained molecular weight markers were from Bio-Rad (Hercules, CA), protease inhibitors were from Roche (Mannheim, Germany), and polyvinylidene difluoride membranes were from Millipore (Bedford, MA). The lyophilized EGF receptor (EGFR) tyrosine kinase inhibitor, gefitinib (Iressa, ZD1839; a gift from AstraZeneca, UK), was dissolved in DMSO, stored at −20°C as a 10 mm stock, and diluted immediately before use.

Human epidermal carcinoma A431 and mouse fibroblast Swiss 3T3 cell lines were obtained from the American Type Culture Collection (Rockville, MD) and were cultured in Dulbecco's MEM supplemented with 10% FCS (Invitrogen Life Technologies, Carlsbad, CA), 2 mm glutamine, 100 units/ml penicillin, and 292 mg/ml streptomycin. Cells were incubated at 37°C in a humidified atmosphere containing 5% CO₂ and were routinely serum starved for 16 h before growth factor stimulation experiments. Unless otherwise stated, cells were stimulated with 50 ng/ml EGF, TGFα, or PDGF for 30 min; where specified, A431 cells were treated with 100 nm gefitinib for the indicated durations before or after EGF stimulation. For protein extraction, cells grown in 150-mm tissue culture dishes were rinsed with ice-cold PBS before lysis in 1 ml nondenaturing buffer [50 mm Tris-HCl (pH 7.5), 0.5% Triton X-100, 0.5% Igepal, 150 mm NaCl, 1 mm EDTA, 50 mm NaF, 1 mm Na₃VO₄ plus protease inhibitors]. Protein lysates were clarified by centrifugation at 4°C using an Eppendorf microcentrifuge at 14,000 rpm for 10 min. Estimation of protein concentration was performed using a bicinchoninic acid assay kit (Pierce Biotechnology, Rockford, IL). About 50 and 100 mg of proteins from A431 and Swiss 3T3 cells, respectively, were routinely used for 4G10 affinity purification experiments. For control experiments requiring protein dephosphorylation, EGF-stimulated cells were lysed in nondenaturing lysis buffer with 1 mm Na₃VO₄ omitted. DTT was added to the lysates to a final concentration of 5 mm before incubation at 37°C for 1 h. For two-dimensional PAGE analysis, cell lysates or affinity-purified proteins were added with 3 volumes of 10% trichloroacetic acid (TCA) in ice-cold acetone added with 20 mm DTT and incubated at −20°C overnight. The next day, the protein precipitate was collected by centrifugation at 14,000 rpm at 4°C for 30 min. The pellet was washed in ice-cold acetone with 20 mm DTT and centrifuged again for 20 min at 14,000 rpm at 4°C. The pellet was air-dried and resuspended in either standard rehydration buffer containing 8 m urea, 2% 3-[(3-cholamidopropyl)dimethylamino]-1-propanesulfonate, 20 mm DTT, and 0.5% IPG buffer or modified rehydration buffer (standard buffer plus 2 m thiourea).

Antibody was bound to protein A-Sepharose at room temperature for 3 h. The Sepharose beads were washed twice with 10 bed volumes of 0.2 m sodium borate (pH 9) and then resuspended in 10 bed volumes of the same solution; DMP was added to a final concentration of 20 mm and mixed for 30 min at room temperature. The reaction was terminated by washing the beads with 0.2 m ethanolamine (pH 8); the beads were then resuspended in 10 volumes of ethanolamine and incubated for 2 h at room temperature to block excess DMP. After discarding the ethanolamine, the beads were washed with 10 volumes of 0.2 m glycine (pH 3) and then stored at 4°C in PBS containing 0.01% NaN₃. A 10 ml aliquot was removed at each step before DMP addition, before ethanolamine incubation, and after glycine wash. The protein samples prepared were analyzed by one-dimensional PAGE followed by gel staining with Coomassie blue to ensure that the coupling of 4G10 to protein A-Sepharose was efficient. For chromatographic purification, protein lysates were first diluted to 1–2 mg/ml with lysis buffer; 1 mg 4G10 antibody/100 mg total protein was used to batch purify tyrosine-phosphorylated proteins from stimulated and unstimulated cells; incubation was carried out for 3 h at 4°C. For control experiments, protein A-Sepharose beads without 4G10 were incubated with the lysates under the same conditions. Following flow-through of experimental lysates, 4G10-Sepharose beads were washed once with 10 bed volumes of lysis buffer and 20 bed volumes of wash buffer [10 mm Tris-HCl (pH 7.5), 25 mm NaCl, 1 mm Na₃VO₄]. Bound proteins were eluted thrice from the beads, each time using 1 bed volume of wash buffer containing 150 mm phenylphosphate and incubating at 37°C for 10 min. The eluates were pooled, concentrated to 1 ml using a Centriprep centrifuge at 3000 × g, and further concentrated to 20–50 ml using a Microcon centrifuge at 12,500 × g. Protein yield was estimated using Bradford's reagent. Concentrate was precipitated with TCA/acetone and processed as described above for two-dimensional PAGE. Regeneration of the 4G10-Sepharose conjugate was achieved by washing the beads with 20 bed volumes of 1.5 m NaCl followed by 20 bed volumes of PBS. The 4G10 conjugate was then stored in PBS/0.01% NaN₃ at 4°C for future use.

After overnight rehydration of IPG strips at room temperature with protein samples, isoelectric focusing (IEF) was carried out using an IPGPhor unit (Amersham Biosciences) at 500 V for 250 Vh, 1000 V for 500 Vh, and 8000 V for 8000 Vh. Following IEF, the strips were equilibrated with buffer [50 mm Tris-HCl (pH 8.8), 6 m urea, 30% glycerol, 65 mm DTT, 2% SDS] for 30 min at room temperature. Two-dimensional PAGE was performed using a minigel apparatus at 15 mA/gel. Proteins were gel stained using SyproRuby fluorescent reagent for up to 48 h, and the gels were scanned using a Typhoon 8600 laser imager (Amersham Biosciences). Spot detection and determination of densitometric intensity were carried out using Image Master 2D Elite software and spot excision was performed by an automated spot picker (Amersham Biosciences).

In some experiments, two-dimensional PAGE gel proteins were transferred to polyvinylidene difluoride membranes using a Western blotting apparatus (Bio-Rad). Blots were stained for protein using SyproRuby; this assay was noted to be much less sensitive than use of the same reagent for gel staining. The blot was then scanned using a Typhoon 8600 imager. Following imaging, the blot was incubated overnight at 4°C in blocking buffer (PBS, 1% BSA, 0.1% Tween 20). The blot was then probed with phosphotyrosine antibody (PY20) conjugated with horseradish peroxidase (PY20H) at 1:1000 dilution for 1 h at room temperature. After washing thrice with PBS containing 0.1% Tween 20, the blot was developed using ECL. Following further washing to remove ECL reagents, the blot was incubated with another peroxidase substrate (DAB) and was developed colorimetrically. It was shown in our laboratory that prior staining with SyproRuby did not affect subsequent immunodetection results.

Protein spots from two-dimensional gels were digested with trypsin for 16 h at 37°C. The resulting peptides were extracted from the gel with 10% (v/v) acetonitrile, 1% (v/v) trifluoroacetic acid solution and purified using a Poros R2 GELoader column. For matrix-assisted laser desorption/ionization (MALDI) analysis, peptides were eluted on the target plate with matrix (8 mg/ml α-cyano-4-hydroxycinnamic acid in 70% v/v acetonitrile, 1% trifluoroacetic acid) and allowed to air-dry. MALDI mass spectrometry (MS) was performed using a Micromass TofSpec 2E time-of-flight (TOF) MS. A nitrogen laser was used to irradiate the sample. The spectra were acquired in the reflectron mode in the mass range 750–3500 a.m.u.; a near-point calibration that yields a mass accuracy < 100 ppm was applied. Peptide masses were searched in Homo sapiens or Mus musculus databases from Swiss-Prot and TrEMBL with PeptIdent. For electrospray ionization (ESI) analysis, digested peptides were examined by ESI-TOF MS/MS using a Micromass Q-TOF MS equipped with a nanospray source using either flow injection coupled to a Waters CapLC or manually acquired using borosilicate capillaries. After peptide selection, MS/MS data were collected over the m/z 50–2000 with variable collision energy settings. Data were used to interrogate either the National Center for Biotechnology Information database using the Mascot search engine or the Swiss-Prot database using the ProteinLynx package.

Two-dimensional gel analysis of nonpurified lysates from EGF-stimulated A431 cells reveals hundreds of protein species (Fig. 1, left panel). Attempts to superimpose the protein expression map generated on its antiphosphotyrosine immunoblotting profile (Fig. 1, right panel) confirmed the technical difficulties in unambiguously identifying which protein spots are tyrosine phosphorylated.

To achieve more sensitive visualization of tyrosine-phosphorylated proteins, we then used antiphosphotyrosine affinity chromatography to enrich protein lysates for tyrosine phosphoproteins. EGFR-overexpressing A431 human epidermoid carcinoma cells were used as a source of inducible tyrosine phosphoproteins. Cells were stimulated with 50 ng/ml EGF for 30 min before lysis. For chromatography, varying amounts of 4G10 phosphotyrosine antibody were coupled covalently to a constant volume of protein A-Sepharose (maximum IgG binding capacity: 8 mg/ml of beads), which was duly incubated with protein lysates derived from EGF-stimulated cells; these experiments established that 1 μg 4G10/100 μg protein from ligand-activated A431 cells is optimal for affinity purification (Table 1). Because reutilization of the column was associated with an ∼50% loss of binding efficiency (data not shown), only fresh columns were used for experiments; similarly, as further experiments revealed that most tyrosine-phosphorylated proteins from these cells lie within pH 4–7, IEF strips corresponding to this detection range were routinely used.

Initial attempts to isolate 4G10-purified proteins from EGF-stimulated cell lysates were hampered by poor sensitivity: only a few faint protein clusters around 80 kDa in molecular weight and pI 5–6 were detected by protein stain and/or antiphosphotyrosine immunoblot (Fig. 2, upper panels). Pertinent to this, it was noted that a sizable pellet remained after resuspension of the acetone precipitate before rehydration of the IEF strip, suggesting inefficient solubilization. Because thiourea may enhance the solubility of membrane proteins (40), 2 m thiourea was added to the standard rehydration buffer. This modification resulted in the appearance of several additional protein clusters exhibiting strong tyrosine phosphorylation as determined by correlative antiphosphotyrosine immunoblotting (Fig. 2, lower panels).

To authenticate the characterization of affinity-purified A431 cell proteins as tyrosine phosphorylated, we excluded false-positive identification of nonphosphorylated heterologous proteins (e.g., such as those that bind nonspecifically to 4G10 antibodies) using two quality control strategies: (a) confirmation that the protein spots identified are EGF inducible rather than constitutive and (b) demonstration that omission of tyrosine phosphatase inhibitors from the lysis buffer eliminates such proteins. As shown in Fig. 3A, quiescent A431 cells (top panels) express fewer and less intense tyrosine-phosphorylated proteins than do EGF-stimulated samples (bottom panels) as indicated by affinity-purified protein abundance (SyproRuby; left panels) and antiphosphotyrosine immunoblotting (IB:PY20; right panels). In those instances where phosphotyrosine signals were detected in the absence of corresponding protein spots, the responsible species were assumed to be either low in abundance or coprecipitated with phosphorylated binding partners. Conversely, a few protein-stained spots did not correspond to regions of significant signal as determined by antiphosphotyrosine immunoblotting, thus failing to confirm these proteins as part of the EGFR signaling pathway. 4G10-enriched proteins were then prepared from EGF-stimulated A431 cell lysates that were subjected to dephosphorylation (see “Materials and Methods”): unlike control EGF-stimulated cell lysates, no proteins from the dephosphorylated lysate could be isolated by 4G10 affinity purification, confirming that 4G10 protein capture is indeed phosphotyrosine dependent (Fig. 3B).

As an initial step toward mapping the EGF-inducible tyrosine phosphoproteome of A431 cells, EGF-stimulated cell lysates comprising ∼50 mg protein were affinity purified using 4G10 and then eluted and resolved using two-dimensional PAGE. SyproRuby gel spots detected by Image Master 2D Elite software were excised and then subjected to peptide mass fingerprinting (PMF) by MALDI-TOF and/or tandem MS/MS using ESI-Q-TOF (Fig. 4A). An abbreviated map (gel protein stain) of the EGF-inducible phosphoproteome visualized is shown in Fig. 4B. Twenty proteins, most of which are tyrosine phosphorylated, were unambiguously identified from 30 spots analyzed. At least five major clustered groups of signaling proteins were evident—signaling enzymes (EGFR, phospholipase Cγ, and protein phosphatase 2A), cytoskeletal proteins (actin and tubulin), adaptor proteins [Shc, p85 subunit of phosphatidylinositol 3′-kinase (PI3K), and Grb2], focal adhesion proteins (ezrin, cortactin, γ-catenin, p120^Cas, and integrins), and molecular chaperones (Hsp90, Hsp70, Hsc71, Grp75, and Grp78). The tyrosine phosphoproteins identified are listed in Table 2. Of these phosphoproteins, actin and Hsc71 were subsequently identified as major immunodetectable phosphoproteomic constituents in a panel of archival human breast cancer tissues (Lim et al., submitted for publication).

To test further the reproducibility and discrimination of the system, we then determined the phosphoproteome induced by varying ligands in two different cell lines. EGF and TGFα both act via the EGFR tyrosine kinase; as predicted, indistinguishable phosphoproteomes were visualized in EGF- and TGFα-treated A431 cells (Fig. 5A). The EGF-inducible tyrosine phosphoproteome of A431 human cancer cells was then compared with that of Swiss 3T3 mouse fibroblasts stimulated with the mesenchymal cell mitogen PDGF. As shown in Fig. 5B, four protein groups from PDGF gels resembled those in EGF gels (K, L, S and U), whereas a further four protein spots appeared distinct although similarly located (G 1–4); these latter spots were excised from the respective gels and submitted for MALDI analysis and/or ESI tandem MS/MS. Spots that looked similar in both gels were confirmed to be the same protein, whereas the suspected variant phosphoprotein cluster represented γ-catenin (plakoglobin) in EGF-stimulated A431 cells and PI3K (p85 subunit isoforms thereof) in PDGF-stimulated 3T3 cells. Of note, the prominent identification of actin and p85 is consistent with the potency of PDGF in stimulating actin reorganization via SH2-dependent p85 (41) and other phosphotyrosine-dependent PDGF receptor signaling pathways (42, 43).

Growing cancer cells are more susceptible than growth-arrested cells to the cytotoxic effects of cycle-specific anticancer drugs, a point of clinical relevance (37). Pertinent to this, we have noted that A431 cells exhibit substantial basal EGFR activity despite the absence of added ligand (data not shown), consistent with autocrine growth. To explore this observation further, A431 cells were subjected to EGFR inhibition using gefitinib before cell lysis and 4G10 purification. Preligand or postligand treatment of A431 cells with gefitinib (100 nm, 1 h) is equally effective in abrogating tyrosine phosphorylation in both control and stimulated cells (Fig. 6); control experiments involving PDGF-treated cells confirm that gefitinib does not inhibit binding of 4G10 to phosphotyrosine (data not shown).

Despite decades of research, the goal of rational anticancer drug development remains elusive; with few exceptions, breakthroughs in cancer therapeutics continue to be made on a trial-and-error basis. Looking to the future, however, this approach appears cost-ineffective and commercially unsustainable. By the same token, multiplex gene expression profiling of human tumors using DNA microarray has yielded certain advances in clinical phenotype prediction, but the applicability of this technology to drug development remains unproven. Newer approaches to the function-specific characterization of tumor biology and drug action therefore seem essential.

Relevant to this challenge, the present study is the first to combine immunoaffinity chromatography and two-dimensional PAGE for the purpose of fingerprinting the signal transduction pathway downstream of a ligand-activated receptor tyrosine kinase. Although conceptually simple, this achievement has proven technically far more difficult than the execution of a simple two-dimensional antiphosphotyrosine immunoblot. Other groups have already reported strategies using the latter approach: for example, Soskic et al. (21) identified time-dependent changes in the PDGF signaling pathway using phosphotyrosine and phosphoserine immunoblotting to localize whole-proteome two-dimensional protein spots via blot overlay followed by in-gel digestion of target protein spots. Another published approach has involved the use of antiphosphotyrosine immunoprecipitation (22) or chromatography (44) to identify phosphorylated substrates of receptor tyrosine kinases using either one-dimensional PAGE followed by MALDI and ESI-MS or two-dimensional PAGE followed by MALDI and/or immunoblotting (23–25). Newer MS-based strategies involving phosphopeptide enrichment (14) and chemical modification of phosphorylated sites (13) are also under development.

Compared with these methodologies, the approach described in the present study benefits from technical simplicity combined with direct identification, yielding greater specificity of the signaling data produced. Central to the technical success of this approach has been the incorporation of thiourea into the dissolution buffer, thus improving the sensitivity of assay via greater recovery of low-abundance precipitated phosphoproteins. By permitting a direct technique to confirm the identity of purified proteins in this way, our approach has provided reliable new information as to the pI and molecular size of the signaling isoforms identified. To ensure correct identification of phosphoproteins, we sought to exclude confounding by heterologous binding of nonphosphorylated protein species: to this end, we performed affinity purification of phosphoproteins from EGF-stimulated A431 cell lysates mixed with 1% SDS, an anionic detergent able to dissociate protein complexes. These control experiments were only partly successful because the detergent also disrupted the binding of antiphosphotyrosine to the phosphoprotein, thus reducing the sensitivity and utility of this quality control.

The definition of the EGFR phosphoproteome summarized in Table 2 should prove relevant to human tumor biology. Of the 20 putative tyrosine phosphoproteins identified here by MS analysis of EGF-dependent cancer cell signaling, many are concordant with observations reported previously. For example, heat shock induces tyrosine phosphorylation of stress proteins such as Hsp70 (45), which are also up-regulated following ligand stimulation of receptor tyrosine kinases in breast cancer (46, 47); because heat shock proteins are directly implicated in apoptotic defects underlying cancer chemoresistance (48), reported enhancements of pharmacological tumor cell kill by tyrosine kinase inhibitors could reflect accelerated oncoprotein degradation via interaction with such chaperones (49).

Our identification of actin as one of the most abundant species in the EGFR phosphotyroproteome is intriguing. Previous investigators have confirmed the tyrosine phosphorylation of actin and its abrogation by the SH2-containing phosphatase SHP-1 (50). We initially identified actin via overlay of PY20 immunoblots onto protein-stained gels; correlative experiments using 4G10 phosphotyrosine antibodies also detected actin in breast cancer tissues. To test whether the latter finding was specific for tyrosine phosphorylation, these same tissues were duly processed in the presence or absence of tyrosine phosphatase inhibitor and subjected to in vitro dephosphorylation; the resulting two-dimensional PAGE data confirmed that actin-associated phosphotyrosine decreased in the lysates subjected to dephosphorylation, thus confirming its phosphorylated status (data not shown). Given the established importance of actin in neoplastic cell transformation (51) and metastasis (52), our identification of this molecule as a tyrosine-phosphorylated species in ligand-activated cancer cells raises the prospect of using actin-interactive drugs to inhibit human cancer growth or spread (53). Of note, these findings suggest that the catalytically inert downstream substrates of tyrosine phosphorylation cascades may prove to be as important in determining tumor cell phenotype and novel drug targets as the upstream enzymes and receptors that have thus far dominated the attention of researchers in this field.

Although consistent with established paradigms of autocrine ligand production, the demonstrated ability of gefitinib to abrogate basal (i.e., as well as ligand-inducible) A431 cell tyrosine phosphorylation raises additional possibilities of potential relevance to rational drug development. The first is that this basal activity accrues from spontaneous (ligand-independent) dimerization of EGFRs expressed at high level in this cell line; in this case, immunotherapeutic strategies aimed at neutralizing receptor-ligand interaction would not be expected to work. The second possibility is that another tyrosine kinase signaling cascade contributes to the basal level of phosphorylation in A431 cells, implying that gefitinib might inhibit kinases other than EGFR. While only a speculation, this possibility could account for clinically unexplained anomalies relevant to gefitinib such as the lack of response correlation with tumor EGFR expression levels and the high frequency of bone pain relief reported in patients with metastatic disease.

In common with other available proteomic approaches, a limitation of the present strategy is that it does not yet have the capability to quantify relative amounts of different tyrosine-phosphorylated species with accuracy. In addition, further developmental work will be needed to scale up this approach into an efficient and cost-effective platform for routine clinical and/or research use. Nonetheless, by focusing attention on a critical subset of molecules involved in cell signaling, the use of phosphoproteomics as described here promises to usher in a new era of functional molecular pathology and rational drug discovery.

We thank Dr. Kok Long Ang for helpful discussions, Dr. Graeme Guy and Prof. Malcolm Paterson for manuscript review, and the National Cancer Centre and National Medical Research Council of Singapore for support.