Research ArticleEGFR Signaling

EGFR-activated Src family kinases maintain GAB1-SHP2 complexes distal from EGFR

See allHide authors and affiliations

Science Signaling  12 May 2015:
Vol. 8, Issue 376, pp. ra46
DOI: 10.1126/scisignal.2005697


Complexes of signaling proteins that are nucleated upon activation of receptor tyrosine kinases are dynamic macromolecular assemblies held together by interactions, such as the recognition of phosphotyrosines by Src homology 2 (SH2) domains. We predicted that reversible binding and phosphatase activity enable dynamic regulation of these protein complexes, which could affect signal transduction. We explored how dynamics in the interactions among the epidermal growth factor (EGF) receptor (EGFR), GRB2-associated binder protein 1 (GAB1), and SH2 domain–containing phosphatase 2 (SHP2) affected EGFR signaling output, specifically SHP2 binding to tyrosine-phosphorylated GAB1, which relieves the autoinhibition of SHP2. Among the effects of activated SHP2 is increased extracellular signal–regulated kinase (ERK) activity. We found that in H1666 lung adenocarcinoma cells, EGFR-activated Src family kinases (SFKs) counteracted repeated GAB1 dephosphorylation events and maintained the association of SHP2 with phosphorylated GAB1 at a cytosolic site distal from EGFR. A computational model predicted that an experimentally verified delay in SFK inactivation after EGFR inactivation, combined with an amplification of GAB1 phosphorylation in cells with proteins in a specific range of concentrations, enabled GAB1 phosphorylation and GAB1-SHP2 complexes to persist longer than EGFR phosphorylation persisted in response to EGF. This SFK-dependent mechanism was specific to EGFR and did not occur in response to activation of the receptor tyrosine kinase c-MET. Thus, our results quantitatively describe a regulatory mechanism used by some receptor tyrosine kinases to remotely control the duration of a signal by regulating the persistence of a signaling protein complex.


In receptor-mediated cell signaling, outside-in information transfer occurs because ligand-receptor binding in the extracellular compartment promotes intermolecular binding events in the cell interior mediated by phosphotyrosine–Src homology 2 (SH2) domain and other types of protein-protein interactions. Static textbook representations of this process belie the reversibility and relatively weak nature of phosphotyrosine–SH2 domain interactions (and other relevant protein-protein interactions) (1), and that phosphate groups on tyrosines can be removed by protein tyrosine phosphatases (PTPs) on times scales that are small compared to the overall time scale for signal transduction (1, 2). These issues, coupled with sometimes receptor- or cell context–dependent details of how specific downstream protein-protein interactions are regulated, create complexities that are typically absent in schematic representations of signaling pathways but which can have a substantial effect on signal transduction. Here, we explored these issues to understand the ability of the epidermal growth factor (EGF) receptor (EGFR) to drive and maintain the association of SH2 domain–containing phosphatase 2 (SHP2) with the adaptor protein GRB2-associated binder protein 1 (GAB1), a binding event that promotes SHP2 activity (3).

SHP2 regulates signaling through numerous pathways, such as stimulating the activity of extracellular signal–regulated kinase (ERK) (3). SHP2 is basally autoinhibited by an intramolecular interaction between its N-terminal SH2 domain and its PTP domain that limits substrate access to the PTP domain (3). Phosphotyrosine engagement of the SH2 domains of SHP2 relieves this inhibition and activates the phosphatase (3). Downstream of EGFR, SHP2 can be activated by binding to phosphorylated GAB1, which can complex with EGFR indirectly by binding with the EGFR adaptor GRB2 and be phosphorylated by receptor tyrosine kinases (RTKs) including EGFR (3, 4). SHP2 binds to phosphorylated Tyr627 and Tyr659 in GAB1, although binding of the N-terminal SH2 domain of SHP2 to phosphorylated Tyr627 in GAB1 is thought to be the dominant event in promoting SHP2 activity (3). Compared to EGF, hepatocyte growth factor (HGF) promotes more sustained phosphorylation of GAB1 and ERK (5, 6), as well as more substantial redistribution of GAB1 to the cell periphery (7). Thus, differences in ERK activation by different RTKs may involve spatiotemporal differences in SHP2 engagement by GAB1.

Although GAB1-SHP2 complexes can be observed for 30 min or more in response to RTK activation (8), the complexes are unlikely to exist in a stable form for this amount of time because SH2 domain–containing proteins generally dissociate from phosphotyrosines within seconds after initial complex formation (9, 10). Because phosphorylated EGFR tyrosines can be dephosphorylated with relatively small time scales (2), it seems likely that similarly rapid regulation of GAB1 tyrosines could occur. However, the kinetics of GAB1 dephosphorylation have not been quantified. If GAB1 dephosphorylation occurs during the time scale of overall GAB1-SHP2 complex persistence, rephosphorylation of GAB1 by a tyrosine kinase could enable the persistence of GAB1-SHP2 complexes. Moreover, if RTKs were the only kinases that could play this role, GAB1-SHP2 complexes might exist mainly as membrane-associated species in complex with RTKs. This possibility is suggested, for example, by typical representations of complexes containing phosphorylated EGFR, phosphorylated GAB1, SHP2, and GRB2, an adaptor containing SH2 and SH3 domains that can bind EGFR at Tyr1068 (11) and GAB1 (12). In contrast, the ability of a cytosolic kinase to phosphorylate GAB1 could extend the effective persistence time and length scales of GAB1-SHP2 complexes distal from a signal-initiating RTK.

Here, we identified a mechanism in adenocarcinoma cells in which EGFR enhanced the persistence of GAB1-SHP2 complexes distal from the receptor through many cycles of GAB1 dephosphorylation by activating Src family kinases (SFKs), a substantial fraction of which were present in the cytosol. To interpret our data, we constructed a kinetic model composed of 386 reactions and characterized by parameters taken from the literature or fit to our data. To best recapitulate our data, the model required that there be a time lag between EGFR inactivation and SFK inactivation, consistent with the time lag required for EGFR dephosphorylation, an effect we validated experimentally. For certain combinations of protein concentrations, the model also predicted that SFKs could effectively amplify EGFR activity to maintain a larger amount of phosphorylated GAB1 than phosphorylated EGFR, an effect which could contribute to, but is not required for, the enhanced persistence of GAB1-SHP2 complexes relative to phosphorylated EGFR we observed. In response to HGF, GAB1-SHP2 complexes formed in an SFK-independent manner and remained in complex with c-MET (the receptor for HGF), suggesting that the mechanism identified downstream of EGFR may not be generic. Thus, we quantitatively described a “remote control” mechanism in which membrane-associated receptors activated cytosolic kinases to promote the persistence of functional protein complexes held together by phosphotyrosine–SH2 domain interactions.


SHP2 associated with GAB1 longer than with EGFR in response to EGF, and GAB1-SHP2 complex maintenance required kinase activity to counteract GAB1 dephosphorylation

To understand the dynamics of SHP2-containing protein complex assembly in response to EGFR activation, we probed the phosphorylation of EGFR at Tyr1068 and GAB1 at Tyr627 in SHP2 immunoprecipitates and whole-cell lysates from EGF-treated H1666 lung adenocarcinoma cells, a cell line in which SHP2 promotes the phosphorylation of ERK (13). EGFR and GAB1 both associated with SHP2 in response to EGF, but EGFR association diminished more quickly than GAB1 association, an effect that was most visible after 120 min of EGF treatment (Fig. 1A). Phosphorylated EGFR was detectable in the lysates of cells treated with EGF for 120 min (Fig. 1A), although total EGFR abundance was reduced by that time. Overall, these data suggest a change in stoichiometry of the SHP2 complex over time and prompt the question of how GAB1-SHP2 complexes were maintained as the amounts of phosphorylated and total EGFR dropped. Similar trends were observed in HeLa cells (fig. S1).

Fig. 1 Time scales of protein dephosphorylation and signaling complex disassembly.

(A) H1666 cells were treated with EGF as indicated, and SHP2 immunoprecipitates (IP) or whole-cell lysates were analyzed by Western blotting (WB) with antibodies against the indicated proteins. (B and C) H1666 cells were treated with EGF and then with gefitinib as indicated. Whole-cell lysates and SHP2 IPs were analyzed by Western blotting with antibodies against the indicated proteins. Blot signals for phosphorylated GAB1 (pGAB1) and phosphorylated EGFR (pEGFR) were quantified and normalized to ERK signal (whole-cell lysates) or SHP2 signal (SHP2 IP). Normalized pGAB1 and pEGFR signals were divided by their respective maxima (A) or their values at the time gefitinib was added (B and C). (D and E) H1666 cells were treated with EGF (15 min), with or without pervanadate (15 min), and then with gefitinib (5 min). Whole-cell lysates and SHP2 IPs were analyzed by Western blotting with antibodies against the indicated proteins. Densitometry data are represented as means ± SEM. *P < 0.05 for comparisons of normalized pGAB1 signals to pEGFR signals at a given time point. All blot images are representative of three sets of biological replicates.

To determine whether these findings were related to a potential difference in EGFR and GAB1 dephosphorylation rates, we measured the dephosphorylation of EGFR and GAB1 in response to treatment with the EGFR inhibitor gefitinib in cells that had been first treated with EGF. Gefitinib induced maximal dephosphorylation of EGFR for the gefitinib concentration used within about 1 min (Fig. 1B), similar to the rate we previously observed in HeLa cells (2). GAB1 dephosphorylation proceeded slightly more slowly, but the dephosphorylation time scale was on the order of 2 to 3 min (Fig. 1B), substantially smaller than the time scale with which SHP2 appears to remain in complex with GAB1 even without EGFR in the complex. GAB1 remained associated with SHP2 for several minutes after gefitinib treatment, at times when EGFR was no longer present in the immunoprecipitated complex (Fig. 1C). PTP inhibition with pervanadate induced phosphorylation of EGFR and GAB1 and EGFR-SHP2 and GAB1-SHP2 associations that were insensitive to EGFR inhibition (Fig. 1, D and E). The apparent decreased abundance and electrophoretic mobility of GAB1 with pervanadate treatment have been reported in cells displaying increased GAB1 phosphorylation (3, 14).

The dynamics quantified in Fig. 1 suggest that many rounds of GAB1 rephosphorylation are needed to counteract the effects of PTPs to maintain GAB1-SHP2 complexes over hours of EGF-mediated signaling. The association of GAB1 with SHP2 in the absence of EGFR, with GAB1 undergoing many rounds of dephosphorylation during the maintenance of the complex, may suggest that kinases other than EGFR are responsible for rephosphorylating GAB1.

SFKs were required for EGFR-initiated GAB1 phosphorylation and maintenance of GAB1-SHP2 association

In COS7 cells, SFKs mediate much of the total tyrosine phosphorylation of GAB1 in response to EGF (3, 15). We thus explored the possibility that SFKs maintained the phosphorylation of Tyr627 in GAB1 and GAB1-SHP2 association in response to EGF in H1666 cells. We used H1666 cells with stable knockdown of endogenous EGFR and EGFRY845F reconstitution to avoid potential confounding effects due to the ability of Src to promote EGFR kinase activity through phosphorylation of Tyr845 (16). Pretreating cells with the SFK inhibitor PP2 resulted in a modest decrease in EGF-induced EGFR phosphorylation that was not statistically significant (fig. S2A), but larger, significant decreases in GAB1 phosphorylation and GAB1 binding to SHP2 (Fig. 2, A and B). Although some phosphorylated EGFR immunoprecipitated with SHP2 in PP2-treated cells, the amount was reduced relative to that found in untreated cells (Fig. 2B and fig. S2B). We interpret this result, which we discuss in further detail later, to indicate that some fraction of SHP2 associated with EGFR in a GAB1-independent manner, perhaps through direct SHP2-EGFR interaction. Treating EGF-treated cells subsequently with PP2 also greatly reduced GAB1-SHP2 association (Fig. 2C). In 293T cells expressing constitutively active SrcY527F, GAB1 was constitutively associated with SHP2 and constitutively phosphorylated in the absence of EGF and in the presence of gefitinib (Fig. 2D and fig. S2C). Given that specific SFKs can compensate for one another (17) and that PP2 inhibits multiple SFKs, the data in Fig. 2 suggest that one or more SFKs, but not necessarily Src itself, participate in the phosphorylation of Tyr627 in GAB1 and GAB1-SHP2 association in response to EGFR activation.

Fig. 2 Requirement of SFKs for EGF-initiated GAB1 phosphorylation and GAB1-SHP2 binding.

(A to C) H1666 cells with knockdown of endogenous EGFR and reconstitution with EGFRY845F were treated with dimethyl sulfoxide (DMSO) or PP2 and then with EGF as indicated and lysed (A and B), or were treated with EGF and then with PP2 as indicated and lysed (C). SHP2 immunoprecipitates (IP) or whole-cell lysates were analyzed by Western blotting (WB). Blot signals for phosphorylated GAB1 (pGAB1) were quantified and normalized to ERK signal (whole-cell lysates) or SHP2 signal (SHP2 IP). For (A) and (B), normalized pGAB1 signals were divided by the maximum signal from DMSO-treated cells. For (C), normalized (Norm.) pGAB1 signals were divided by the signal from cells treated with DMSO for 1 min. (D) 293T cells transfected with a vector encoding GAB1 or pcDNA3 empty vector (EV) and a vector encoding SRCY527F or p3xFlag EV were treated with EGF and then with gefitinib before lysis. SHP2 IPs were analyzed by Western blotting with antibodies against the indicated proteins. Densitometry data are represented as means ± SEM. *P < 0.05 when comparing normalized pGAB1 signals from DMSO-treated cells to signals from PP2-treated cells at a given time point. All blot images are representative of three sets of biological replicates.

GAB1-SHP2 complexes were present mainly in the cytosol

The fact that GAB1-SHP2 complexes lacking EGFR were present suggests, but does not guarantee, that GAB1-SHP2 complexes may have been present in the cytosol. When subcellular fractionation was performed, essentially all SHP2 and most GAB1 were found in the cytosolic fraction (Fig. 3A). Conversely, virtually all phosphorylated EGFR was found in membrane fractions. The vast majority (87 ± 2%) of transferrin receptor, a membrane marker, and a similar amount of phosphorylated EGFR were found in the aggregate membrane fractions. Thus, small amounts of both receptors were present as contaminants in nonpelleted fractions, similar to previous publications. However, given that all SHP2 was in the cytosolic fraction, the vast majority of phosphorylated EGFR was clearly separated from SHP2 complexes. The endosome marker Rab5 was associated more with the first of the two membrane fractions than the second, but was also present in the soluble or cytosolic fraction, as has been observed elsewhere using higher centrifugal forces than we were able to achieve (18). The presence of cytosolic Rab5 may result from Rab5 binding to an endogenous Rab5 inhibitor (19). The endoplasmic reticulum marker Grp94 was present in the first of the two membrane fractions and in the soluble fraction. Because Grp94 is soluble within the lumen of the endoplasmic reticulum, Grp94 was likely observed in the soluble fraction owing to endoplasmic reticulum rupture during lysis.

Fig. 3 Subcellular localization of GAB1-SHP2 complexes.

(A and B) Membrane and cytosolic fractions were prepared from H1666 cells treated with or without EGF. Whole-cell lysates or SHP2 immunoprecipitates (IP) were analyzed by Western blotting (WB) with antibodies against the indicated proteins. “M1” and “M2” indicate membrane fractions prepared from 9300 rpm and 100,000g spins, respectively, and “C” indicates the cytosolic fraction. (C and D) H1666 cells were treated with DMSO or GDC-0941 and then with EGF and lysed. Whole-cell lysates or SHP2 IPs were analyzed by Western blotting with antibodies against the indicated proteins. All blot images are representative of three sets of biological replicates.

When the cytosolic fraction was compared against the membrane fraction containing the greater amount of phosphorylated GAB1 of the two membrane fractions, EGF-induced GAB1-SHP2 complexes were found in the cytosolic fraction and SHP2 complexes containing only EGFR were found in the membrane fraction (Fig. 3B). Although some reports suggest that SFKs are mainly located in the membrane because of SFK myristoylation (20), we found SFKs in membrane and cytosolic fractions, as well as an EGF-inducible band for SFKs phosphorylated at Tyr418 in both membrane and cytosolic fractions (arrow in Fig. 3A). On the basis of the apparent molecular weight of the band and confirmation by RNA interference that the most prominent bands in SFK and phosphorylated SFK blots were Src (fig. S3A), the EGF-inducible bands in question may be Lyn, which is present in lung adenocarcinoma cell lines (21). GRB2 abundance was increased in membrane fractions in response to EGF (Fig. 3A), suggesting that phosphorylated EGFR in membrane fractions retained the ability to form complexes with adaptor proteins.

Because the phosphatidylinositol 3-kinase (PI3K)–dependent recruitment of GAB1 to the plasma membrane through the GAB1 pleckstrin homology (PH) domain is essential for the function of GAB1 in some cell types (7), we also probed the effect of PI3K inhibition on GAB1-SHP2 complexes, even though the largest amount of GAB1 was cytosolic. In H1666 cells treated with the PI3K inhibitor GDC-0941, EGF-induced phosphorylation of Ser473 in AKT was reduced, but GAB1-SHP2 association was not impaired (Fig. 3, C and D). Because the phosphorylated lipids that bind the PH domain of GAB1 are generally less abundant in endosomes than in the plasma membrane, the endocytic process could promote the presence of GAB1-SHP2 complexes in the cytosol. When EGFR endocytosis was inhibited, however, protein compartmentalization was unchanged (fig. S3, B and C).

EGF and HGF promoted different dynamics of GAB1-SHP2 complex persistence

In some cells, GAB1 phosphorylation and GAB1-SHP2 association are sustained longer in response to HGF than EGF. These effects may explain why HGF promotes more sustained ERK phosphorylation (5, 6), but the mechanistic details by which c-MET and EGFR differentially affect the duration of GAB1-SHP2 association have not been fully explored. In H1666 cells, HGF promoted a lower but more protracted GAB1-SHP2 association than did EGF, and c-MET remained in complex with SHP2 longer than did EGFR (Fig. 4A). These effects may have occurred because c-MET phosphorylation was more persistent and c-MET was more stable than EGFR (fig. S4A). Although SFKs can be activated by c-MET (22), SFK inhibition did not significantly reduce GAB1-SHP2 association, suggesting that a kinase other than an SFK (possibly c-MET) regulates the phosphorylation of GAB1 in response to HGF (Fig. 4B and fig. S4B). The effect of PP2 on EGF-mediated GAB1-SHP2 association in Fig. 4B was probably smaller than that in Fig. 2 because of the larger EGF concentration used in Fig. 4B. Similar to observations from cells treated with EGF and gefitinib, the GAB1-SHP2 association lasted longer than the c-MET–SHP2 association in cells treated with HGF and subsequently with the c-MET inhibitor PHA665752 (fig. S4B).

Fig. 4 Sustained association of GAB1 and SHP2 downstream of c-MET.

(A) H1666 cells were treated with EGF or HGF as indicated, and SHP2 immunoprecipitates (IP) were analyzed by Western blotting (WB). Blot signals for c-MET, phosphorylated EGFR (pEGFR), and phosphorylated GAB1 (pGAB1) were normalized by SHP2 signal. Normalized pGAB1 signals were divided by the maximum signal for EGF treatment, and normalized c-MET and pEGFR signals were divided by their respective maxima. (B) Cells were treated with DMSO or PP2 and then HGF or EGF before being lysed. SHP2 IPs were analyzed by Western blotting using antibodies against the indicated proteins. Blot signals for pGAB1 were normalized by SHP2 signal. Normalized pGAB1 signals were divided by their respective maxima. Densitometry data are represented as means ± SEM. *P < 0.05 for comparisons of EGF to HGF treatment. All blot images are representative of three sets of biological replicates.

A delay in SFK inactivation after EGFR inactivation was predicted to enable GAB1 phosphorylation and GAB1-SHP2 complexes to last longer than EGFR phosphorylation

To quantitatively explore the relationships between EGFR and GAB1 phosphorylation, SFK activity, and GAB1-SHP2 binding, we developed a computational mechanistic model of GAB1-SHP2 complex dynamics that included the protein phosphorylation and dephosphorylation processes and reversible protein binding for the complexes (Fig. 5A, fig. S5A, and Materials and Methods). Most parameters were taken from the literature (table S1). Four key rate constants were determined through fitting. The rate constant for EGFR dephosphorylation (kdp) was fit to the decrease in the phosphorylation of EGFR throughout the course of a 5-min gefitinib treatment (fig. S5B). The rate constants for GAB1 phosphorylation (kG1p) and dephosphorylation (kG1dp) were fit to the response of the phosphorylation of Tyr627 in GAB1 and GAB1-SHP2 association to both long and short EGF treatments (Fig. 1A and fig. S5C) and treatment with EGF followed by gefitinib (Fig. 1, B and C). The rate constant for EGFR degradation (kdeg), a process needed to allow the phosphorylation of EGFR and GAB1 to decrease with time (fig. S5D) because we assume a constant activity of PTPs, was fit to the dynamics of total EGFR abundance in response to EGF (Fig. 1A).

Fig. 5 Model topology and validation.

(A) The model topology includes the depicted processes leading to EGFR and SFK phosphorylation and activation, respectively. (B and C) Model predictions for the association of phosphorylated EGFR (pEGFR) or phosphorylated GAB1 (pGAB1) with SHP2 (lines) were compared to experimental data (points) from (B) an experiment where H1666 cells were treated with EGF and then gefitinib, and (C) an experiment where cells were treated with EGF alone as indicated. (D) SHP2 was immunoprecipitated from H1666 cells treated with EGF as indicated. Percent SHP2-bound GAB1 was determined for experimental data by first calculating the percent of GAB1 unbound with SHP2, which was calculated by dividing the normalized GAB1 signal in the SHP2 immunoprecipitate supernatant by that of the GAB1 signal in whole-cell lysates. This value was subtracted from 100% to calculate the percentage of SHP2-bound GAB1, which was then compared to the model prediction for percentage of SHP2-bound GAB1 in response to 15-min EGF treatment. (E) Model predictions were made for normalized amounts of pEGFR, pGAB1, and active SFK, or for pGAB1 with a model topology where SFKs inactivate instantly after gefitinib addition. (F) The sensitivity of model predictions for time-integrated GAB1-SHP2 association during a 5-min treatment with 1 μM gefitinib [preceded by a 15-min treatment with EGF (10 ng/ml)] to 10-fold changes in the model parameters was calculated. (G) H1666 cells were treated with EGF and then with gefitinib as indicated. Whole-cell lysates were analyzed by Western blotting with antibodies against the indicated proteins, and blot signals for pEGFR, pGAB1, and the lower–molecular weight, EGF-induced phosphorylated SFK (pSFK) band indicated by the arrow were quantified and normalized to the ERK signal. Densitometry data are represented as means ± SEM. Blot images are representative of three sets of biological replicates. (H) Model error was calculated for ranges of kdp and kG1dp. Red circle indicates error minimum.

The previously mentioned GAB1-independent mechanism of SHP2-EGFR association is not accounted for in our model topology. However, this omission should not affect model conclusions because parameter fits did not rely on EGFR-SHP2 association data. Moreover, because data suggested that EGFR and SHP2 dissociated immediately after gefitinib treatment (Fig. 1C), and because GAB1-dependent and GAB1-independent EGFR-SHP2 complexes dissociated with similar kinetics over the course of a 120-min EGF treatment (fig. S2B), model predictions for relative abundances of EGFR-SHP2 and GAB1-SHP2 complexes would be unchanged even if model fits did account for EGFR-SHP2 association. About half of EGFR-SHP2 complexes depended on SFK activity and thus were likely GAB1-dependent (Fig. 2B). Therefore, the capacity for GAB1 to recruit SHP2 to EGFR is an important mechanism of EGFR-SHP2 complex formation and a valid inclusion for our model topology.

The parameterized model recapitulated differences in the rates of EGFR-SHP2 and GAB1-SHP2 dissociation after gefitinib treatment (Fig. 5B). Although both EGFR-SHP2 and GAB1-SHP2 complexes were predicted to return to basal amounts after 1 to 2 min of gefitinib treatment, EGFR-SHP2 complexes fell to 50% of their peak concentration about five times faster than did GAB1-SHP2 complexes. The model also accurately predicted differences in the relative amounts of EGFR-SHP2 and GAB1-SHP2 complexes over a 120-min EGF treatment time course (Fig. 5C). The model further predicted that 22% of GAB1 was bound with SHP2 15 min after treatment with EGF, which is consistent with an experimental measurement of 26 ± 5% (Fig. 5D and fig. S5E). Finally, the model predicted that even at peak EGFR phosphorylation, only 1.5% of SHP2 existed in complex with EGFR (fig. S5F), which was qualitatively consistent with our experimental findings that SHP2 and GAB1-SHP2 complexes existed almost exclusively in the cytosol (Fig. 3, A and B).

The fitted rate constant for GAB1 dephosphorylation (kG1dp) was similar in magnitude to that for EGFR dephosphorylation (kdp) (table S1), with each implying a dephosphorylation time scale of ~0.1 min, despite a GAB1 dephosphorylation rate, and therefore a GAB1-SHP2 dissociation rate, that was smaller than the rate of EGFR dephosphorylation after gefitinib treatment of cells (Figs. 1B and 5E). This apparent contradiction was explained most substantially by a time lag for the inactivation of SFKs after EGFR inactivation, which was determined by a parameter sensitivity analysis that identified parameters for SFK activation and deactivation as two of the most important parameters controlling GAB1-SHP2 complex persistence other than dephosphorylation of GAB1 (kG1dp) and of EGFR (kdp) (Fig. 5F). The predicted time lag arose because EGFR phosphorylation, which was assumed to drive SFK activity, was not maximally reduced instantaneously after EGFR inactivation because of the finite, nonzero (though rapid) kinetics of EGFR dephosphorylation (Fig. 5E). The finite time scale for the SFK deactivation process also contributed to the time lag between EGFR inactivation and SFK inactivation. The effects of the time lag between EGFR inactivation and SFK inactivation were apparent, for example, in model predictions showing that the GAB1 phosphorylation rate remains relatively high compared to the EGFR phosphorylation rate after gefitinib treatment (fig. S5G). If in our model SFKs were assumed to deactivate instantaneously upon EGFR inactivation, the predicted rates of EGFR and GAB1 dephosphorylation in response to gefitinib treatment were nearly identical, with GAB1 dephosphorylation being slightly faster because of a difference in dephosphorylation rate constants (Fig. 5E). Consistent with the model (Fig. 5E), dephosphorylation at the lower–molecular weight SFK Tyr418 band proceeded gradually after gefitinib treatment with kinetics that mirrored EGFR dephosphorylation (Fig. 5G). In interpreting the results of Fig. 5G and comparing them against the explanation proposed above, it is important to bear in mind that SFK phosphorylation at Tyr418 regulates SFK activity and that our model assumes that the EGFR kinase inactivates immediately upon gefitinib addition to cells. With those points in mind, Fig. 5G can be interpreted as showing that after EGFR kinase inactivation at 0 min, SFK activity gradually decreased over a time period of ~2 min.

Plotting the model error for a range of kG1dp and kdp values provided additional insight (Fig. 5H). Although the error minimum was achieved for kG1dp and kdp of similar magnitudes, a relatively low model error was still achieved if one rate constant was increased and the other was decreased by up to an order of magnitude. Model error substantially increased when either rate constant was changed by more than an order of magnitude from its best-fit value. Thus, there was some capacity to explain the data by speeding up one dephosphorylation process and slowing the other, but both rate processes needed to proceed fairly rapidly to explain the data reasonably.

An amplification of phosphorylated GAB1 relative to phosphorylated EGFR was predicted by the model for some combinations of protein concentrations

Beyond the time lag in SFK inactivation relative to EGFR inactivation, the difference in cellular EGFR and GAB1 dephosphorylation rates was predicted by the base model to be augmented or increased by an SFK-mediated amplification process that produced a larger concentration of phosphorylated GAB1 than phosphorylated EGFR (Fig. 6A). The amplification occurred as a combined result of a smaller time scale for the GAB1 phosphorylation step (~3 s) than for EGFR phosphorylation (~4.5 s) and as a result of the relative slowness of other steps leading to EGFR phosphorylation, including ligand binding and EGFR dimerization (Fig. 6B). When parameter values were adjusted to remove the amplification process from the model, the difference between EGFR and GAB1 dephosphorylation rates was reduced but not eliminated because of the time lag for SFK inactivation (Fig. 6C).

Fig. 6 Model predictions for an SFK-mediated amplification of GAB1 phosphorylation that is possible for certain protein abundance profiles.

(A) Model predictions were made for the number of phosphorylated EGFR (pEGFR) and phosphorylated GAB1 (pGAB1) per cell in response to EGF (10 ng/ml). (B) Model predictions were made for the number of pEGFR or pGAB1 per cell in response to EGF (10 ng/ml) for the base model parameters. Model parameters were adjusted to make the time scales for EGFR and GAB1 phosphorylation equivalent (=τphosph), or model parameters were adjusted such that the EGFR and GAB1 phosphorylation time scales are equivalent and all forward and reverse rate constants for EGF binding, adenosine 5′-triphosphate (ATP) binding, and EGFR dimerization are increased or decreased by an order of magnitude to speed up the processes that occur before EGFR phosphorylation (=τphosph + fast phosph.). (C) Model predictions were made for normalized pEGFR or pGAB1 per cell for response to 1 μM gefitinib [after 15-min treatment with EGF (10 ng/ml)] for the base model topology or for a model topology in which SFK-mediated amplification of GAB1 phosphorylation does not occur (no amp.). (D) A metric for the degree of amplified EGFR signal, the ratio of phosphorylated GAB1 to phosphorylated EGFR (pGAB1/pEGFR), and model error were calculated for 10-fold combinatorial variations in GAB1, SHP2, and SFK concentrations. (E) Model predictions were made for pGAB1/pEGFR for 300 random parameter sets, in which each parameter was randomly varied up to an order of magnitude above or below its base value. (F) The cellular expression of EGFR and GAB1 was adjusted to be consistent with values reported by Kulak et al. (23) for HeLa cells, and model predictions were made for the number of pEGFR or pGAB1 per cell in response to EGF (10 ng/ml).

To explore the robustness of model predictions for the amplification process, two types of analysis were used. To check if the assumption of equivalent GAB1, SFK, and SHP2 abundance was responsible for the prediction of the SFK-mediated amplification mechanism, the model was refit and errors were calculated for 10-fold combinatorial variations in GAB1, SHP2, and SFK concentrations. The smallest errors were associated with ratios of cellular concentrations of phosphorylated GAB1 to phosphorylated EGFR greater than unity (Fig. 6D). In a second robustness analysis, new parameter sets were generated by randomly perturbing each model parameter up to an order of magnitude above or below its base value. For 78% (234 of 300) of parameter sets, the ratio of phosphorylated GAB1 to phosphorylated EGFR was greater than unity (Fig. 6E). Therefore, the model prediction of amplification of EGFR signal by SFKs appeared robust to perturbations of about an order of magnitude to model parameters.

Although absolute protein abundance is rarely known for cell lines, mass spectrometry analysis of HeLa cells by Kulak et al. (23) has revealed that GAB1 is 60-fold less abundant than EGFR, a difference in GAB1 and EGFR abundance that greatly exceeds the differences considered in Fig. 6 (D and E). With so much less GAB1 than EGFR, it seems unlikely that an amplification of phosphorylation from EGFR to GAB1 would be possible. Indeed, when protein concentrations were adjusted to reflect the relative abundances of GAB1 and EGFR reported by Kulak et al., no amplification was predicted in response to EGF treatment during a 120-min time course (Fig. 6F). Thus, the possibility that amplification of phosphorylation from EGFR to GAB1 can occur depends highly on the abundance of EGFR relative to GAB1. In further support of this notion, when the calculations in Fig. 6 (D and E) were redone using the protein abundances suggested by Kulak et al., the best-fit model did not produce amplification for any combination of protein concentrations in the calculations analogous to those in Fig. 6D, and only 4 of 300 trials displayed amplification in the calculations analogous to those in Fig. 6E.

The model was applied and extended to predict the relevance of additional processes that may potentially regulate system dynamics

The model was also applied to explore the predicted effects of changes to the topology motivated by several previously documented findings. Because SHP2 can promote SFK activity by preventing the normal localization of c-Src kinase (CSK) to paxillin and PAG/CSK-binding protein (8, 24), the model was updated to allow for active SHP2, in addition to EGFR, to activate SFKs, which created a positive feedback loop. Relative to the base model (fig. S6A), the model allowing SHP2 activity to enhance SFK activity resulted in more robust GAB1 phosphorylation (fig. S6B), which was reflected by the need to reduce the model-fitted rate constant for GAB1 phosphorylation by several orders of magnitude to best explain our experimental data. This modification did not, however, substantially alter the minimum model error relative to the base model (fig. S6, A and B). The incorporation of a distinct reaction for GAB1 dephosphorylation by SHP2 (25) also altered best-fit rate constants without substantially altering the ability of the model to recapitulate the experimental data (fig. S6C). Thus, although the model alterations considered here change certain quantitative rate processes, they do not fundamentally alter the model’s ability to explain the overall rates of GAB1 phosphorylation and GAB1-SHP2 complex formation.


We described a mechanism in which EGFR drives the persistence of cytosolic GAB1-SHP2 complexes through many cycles of GAB1-SHP2 dissociation and GAB1 dephosphorylation through the ability of SFKs, activated by EGFR, to rephosphorylate GAB1 repeatedly distal from EGFR (Fig. 7A). This may allow EGFR to control signaling processes through SHP2 at intracellular locations where it may not otherwise be able to do so because of EGFR confinement to membrane compartments. For protein concentrations that permit an amplification of tyrosine phosphorylation from EGFR to GAB1, this mechanism may also allow a relatively small amount of phosphorylated EGFR to exert substantial control over signaling through more abundant phosphorylated GAB1.

Fig. 7 Schematic of EGFR-mediated GAB1-SHP2 complex maintenance.

(A) Basally, SHP2 activity is suppressed through an intramolecular interaction between the N-terminal SH2 domain and the catalytic domain. When SHP2 SH2 domains engage tyrosine phosphorylated GAB1, the SHP2 intramolecular tethering is relieved, and SHP2 activity increases. EGFR activation appears to promote this process primarily through the intermediary SFKs, which counteract many rounds of GAB1 tyrosine dephosphorylation to enable GAB1-SHP2 complexes to persist distally from the EGFR. With a lower abundance than their cytosolic counterparts, GAB1-SHP2 complexes may also exist at the membrane in complex with EGFR, and some SHP2 may bind EGFR directly in a GAB1-independent fashion. (B) Many cycles of GAB1 phosphorylation/dephosphorylation and SHP2 binding/unbinding occur per minute in response to EGFR activation.

The mechanism identified clearly involves many rounds of SHP2 binding and unbinding from phosphorylated GAB1 and many rounds of GAB1 phosphorylation and dephosphorylation over the typical time scale for EGFR-mediated signal transduction, and the quantitative model results can be used to estimate the frequencies of these events. On the basis of characteristic GAB1 phosphorylation and dephosphorylation times (see Materials and Methods), model parameters suggested that when phosphorylation and dephosphorylation reaction rates are maximized, Tyr627 in GAB1 undergoes about six cycles of dephosphorylation and rephosphorylation per minute in response to EGF (Fig. 7B). Model parameters also suggested that SHP2 cycles between GAB1-bound and GAB1-unbound states ~14 times during the time that GAB1 is phosphorylated in each GAB1 cycle, or ~100 times per minute at maximal rates. Thus, the mechanism in Fig. 7A involves a highly dynamic process of GAB1 phosphotyrosine “futile cycling” (26) and maintenance by SFKs.

The specific location within the cell where SFKs drive repeated rounds of GAB1 phosphorylation has important implications for how far within the cell GAB1-SHP2 complexes persist. Although the results with PI3K inhibition in Fig. 3 argued against any need for GAB1 to be membrane-bound to be phosphorylated by SFKs, certain evidence argues for SFK activity mainly at the membrane. Indeed, SFKs are typically thought of as membrane-bound species due in part to N-terminal myristoylation. Myristoylation itself may promote SFK activity (20), and some studies suggest a requirement for SFKs to move to the membrane to become activated (27). Still, in some cellular settings, substantial fractions of total and active forms of SFKs have been found in the cytosol (28, 29), as found here in H1666 cells. The potential importance of this detail can be further understood through order-of-magnitude estimates of specific process time scales. Assuming a cell radius of 10 μm (30) and a diffusivity of 0.94 μm2/s, based on the diffusivity of tubulin and an adjustment due to estimated hydrodynamic radii of tubulin and the GAB1-SHP2 complex (31, 32), the characteristic time for a GAB1-SHP2 complex to diffuse from the plasma membrane to the cell center is ~18 s. Comparison of that time scale against a characteristic time for GAB1 dephosphorylation of ~6 s suggests that GAB1-SHP2 complexes formed exclusively at the plasma membrane may not persist over the entire cellular length scale. The ability of SFKs to drive the phosphorylation of GAB1 in the cytosol could overcome this limitation, potentially increasing the ability of SHP2 to regulate signaling processes, for example, by regulating tyrosine phosphorylation of paxillin in focal adhesions (8).

Although the results with dynamin inhibition (fig. S3) suggested that EGFR endocytosis was not required for the presence of GAB1-SHP2 complexes in the cytosolic fraction, there are ways that EGFR endocytosis could influence system dynamics in ways we did not explicitly consider. For example, the clustering of phosphorylated EGFR in endocytic pits enhances the tendency for GRB2 to rebind EGFR near the cell periphery (33). This effect could potentially skew the distribution of GRB2-GAB1-SHP2 complexes toward EGFR-containing endocytic membranes within the cytosol. EGFR endocytic trafficking would also move the most upstream processes in this signaling pathway into the cell interior, which may overcome diffusional limitations that could otherwise limit GAB1-SHP2 access to some intracellular locations. Indeed, the ability of endocytosis-impaired and constitutively active EGFR mutants to sequester SHP2 at the cell periphery appears to antagonize the ability of SHP2 to participate fully in the activation of ERK (13). The extent to which this functional impairment of SHP2 activity involves a perturbation to the ability of GAB1-SHP2 complexes to exist distal from EGFR mutants remains to be determined.

Our measurements of complex persistence relied on coimmunoprecipitation. This approach has been used in numerous analyses of SHP2 complexes, but such data are subject to potentially confounding issues including protein complex dilution during lysis, washing, and immunoprecipitation and potential loss of complex stability during incubations. As described previously, our data suggest that, in addition to GAB1-SHP2 complexes, EGFR-GRB2 complexes were also preserved in our analysis. Thus, key phosphotyrosine–SH2 domain linkages were at least qualitatively preserved in our protocol. To the extent that the confounding issues mentioned above affected the quantitative abundance of complexes we measured, the effects would have been uniformly present because of synchronization of lysis times among samples, making the issues unlikely to have affected qualitative trends. So, although quantitative model predictions and inferences regarding absolute rates of protein dephosphorylation or complex abundance may have been affected by such issues in the data used to train the model, the key qualitative experimental and modeling inferences are unlikely to have been affected. In the future, it would clearly be useful to develop more direct approaches to visualize intracellular GAB1-SHP2 complexes in living cells.


Cell culture

H1666 cells were maintained in ACL4 supplemented with 10% fetal bovine serum (FBS) (34), and 293T and HeLa cells were maintained in Dulbecco’s modified Eagle’s medium supplemented with 10% FBS. Cell lines were from the American Type Culture Collection. Before the experiments, cells were serum-starved by switching them to medium containing 0.1% FBS for 16 to 18 hours.

Stable short hairpin RNA and expression constructs

The pLKO vector containing a short hairpin sequence targeting the 3′ untranslated region of human EGFR (5′-AGAATGTGGAATACCTAAGG-3′) was provided by D. Haber (Harvard Medical School). Lentivirus was produced by calcium phosphate–mediated transfection of 293FT cells (Invitrogen) with vector and the packaging plasmids pCMV-VSVG, pMDL-gp-RRE, and pRSV-Rev (M. Farquhar, University of California, San Diego). Virus was harvested 48 and 72 hours after transfection and used to infect target cells, which were selected in puromycin. EGFR complementary DNA (cDNA) encoding EGFRY845F (S. Parsons, University of Virginia) was inserted in the pMSCV vector. Retrovirus was produced by calcium phosphate–mediated transfection of amphotropic Phoenix cells (G. Nolan, Stanford University) with vector. Virus was harvested 24, 30, and 48 hours after transfection and used to infect target cells, which were selected in hygromycin. Constructs were validated by sequencing. EGFR knockdown was validated by Western blot.

Transient expression constructs and RNA interference

The p3xFlag-CMV-7.1 vector containing SrcY527F cDNA was provided by T. Miller (Stony Brook University). The pcDNA3 vector containing hemagglutinin-tagged GAB1 was provided by T. Hirano (Osaka University). Cells were plated in six-well plates in medium lacking antibiotics and were transfected the following day with 1 μg of SrcY527F DNA or p3xFlag-CMV-7.1 empty vector DNA and 1 μg of GAB1 DNA or pcDNA3 empty vector DNA using 6 μl of Lipofectamine 2000 (Invitrogen). Cells were switched to serum-free medium 4 hours later, and then treated and lysed the next day before proceeding to immunoblotting and immunoprecipitation.

Subcellular fractionation

Serum-starved cells were treated with or without EGF, washed, and collected in hypotonic buffer [10 mM tris-HCl (pH 7.4), 1 mM MgCl2, 1 mM EDTA] supplemented with 1 mM phenylmethylsulfonyl fluoride, additional protease inhibitors, and phosphatase inhibitors. Crude lysates were generated with a Dounce homogenizer and spun at 3000 and then 9300 rpm in a benchtop centrifuge, for 5 min at each speed, with pelleted fractions from each spin reserved. The supernatants remaining were ultracentrifuged at 100,000g for 60 min, and pelleted and supernatant fractions were reserved. Membrane pellets from the 9300 rpm and 100,000g spins were washed with phosphate-buffered saline, resuspended in hypotonic buffer, and centrifuged again. After additional washes, membrane pellets were resuspended in lysis buffer to solubilize proteins before immunoblotting or immunoprecipitation.


Immunoblotting was performed as described previously (13).


Cell lysates were prepared per the immunoblotting protocol. Total protein (500 μg) was incubated at 4°C overnight with agarose beads conjugated to SHP2 or control antibody. Beads were washed three times with lysis buffer, resuspended in LDS sample buffer, and boiled before immunoblotting.

Antibodies and other reagents

EGFR antibody (Ab-12) was from Thermo Fisher Scientific. ERK (#4695), phosphorylated GAB1 Tyr627 (#3233), phosphorylated c-MET Tyr1234/Tyr1235 (#3126), transferrin receptor (#13113), Rab5 (#3547), and c-MET (#8198) antibodies were from Cell Signaling Technology. Phosphorylated SFK Tyr418 antibody (44660G) was from Life Technologies. SHP2 (sc-280), GAB1 (sc-9049), SFK (sc-8056), and GRB2 (sc-255) antibodies were from Santa Cruz Biotechnology. Actin (MAB 1501) antibody was from Millipore. Phosphorylated EGFR Tyr1068 (#1727) antibody was from Epitomics. Grp94 antibody (ADI-SPA-850) was from Enzo Life Sciences. Infrared dye–conjugated secondary antibodies were from Rockland Immunochemicals and Life Technologies, respectively. Gefitinib (LC Laboratories), GDC-0941 (LC Laboratories), PHA665752 (Santa Cruz Biotechnology), Dynasore (EMD Millipore), and PP2 (Sigma) were reconstituted in DMSO and used at concentrations of 1, 0.5, 1, 80, and 10 μM, respectively. Incubation times of 5, 30, 60, and 30 min were used for gefitinib, GDC-0941, Dynasore, and PP2, respectively, unless otherwise noted. For most experiments, recombinant human EGF (PeproTech) was used at 10 ng/ml. Unless otherwise noted, EGF incubation time was 15 min. To compare the effects of EGF against recombinant human HGF (PeproTech), we used equivalent dissociation constant–normalized concentrations of the ligands [EGF (38 ng/ml) or HGF (50 ng/ml)] (35, 36). Pervanadate was prepared as previously described (2) and used at 100 μM with an incubation time of 15 min.


Statistical analyses were performed with a paired two-tailed Student’s t test.

Model development and implementation

The model consists of a set of coupled ordinary differential equations to describe the rate processes leading to EGFR and GAB1 phosphorylation, SFK activation, and EGFR-GRB2, GRB2-GAB1, and GAB1-SHP2 associations. The model topology leading from EGF binding to EGFR phosphorylation is based in part on a previously published model (2). Essential processes and parameters are summarized in Fig. 5A and table S1, respectively. In total, the model includes 386 reactions, 101 species, and 29 parameters.

EGF binding at the plasma membrane was modeled as a reversible process characterized by association (37) and dissociation (38) rate constants. EGF was modeled at a constant concentration of 10 ng/ml. Association and dissociation rate constants for ATP and gefitinib with EGFR were previously calculated (2). ATP was assumed to be at a constant cellular concentration of 1 mM (39). Gefitinib, when included, was modeled at a constant cellular concentration of 1 μM.

The EGFR dimerization rate constant was calculated as described previously (2) assuming 6 × 105 EGFR proteins per H1666 cell, which was estimated by a Western blot–based comparison of total EGFR abundance in H1666 cells relative to PC9 cells, for which we previously determined EGFR abundance at the membrane using 125I-EGF. Dimer uncoupling rate constants in the presence or absence of EGF have been described previously (40, 41). All dimer species were assumed to be symmetric, with the exception of allowing for asymmetric EGF binding.

EGFR phosphorylation was modeled as a process that occurs between ATP-bound EGFR dimers where both receptors are simultaneously phosphorylated at a representative tyrosine that can bind GRB2, with distinct rate constants for phosphorylation occurring in the presence or absence of EGF (42). GAB1 phosphorylation at a representative tyrosine that can bind SHP2 was modeled as a process catalyzed by an active SFK. Because our experimental data suggest that SFKs are the primary mediator of GAB1 phosphorylation in H1666 cells, we did not include the possibility of EGFR phosphorylating GAB1. EGFR and GAB1 dephosphorylation were modeled as zero-order with respect to PTPs, which precludes a need to consider distinct PTP species.

EGFR degradation was modeled as permissible for any GRB2-bound EGFR species because GRB2 mediates the interaction of the CBL ubiquitin ligase with EGFR, which plays a primary role in ligand-mediated EGFR degradation (43). Any proteins bound to EGFR targeted for degradation were assumed to become instantaneously unbound from that EGFR species.

GRB2 was modeled as being able to bind phosphorylated EGFR using experimentally derived rate constants for association and dissociation (10). GAB1 binding to GRB2 was modeled as an SH3 domain–mediated interaction using previously described rate constants for association and dissociation (44). SHP2 binding to phosphorylated GAB1 was modeled using the same association and dissociation rate constants as for GRB2 binding EGFR, assuming similar phosphotyrosine–SH2 domain–mediated interactions for GRB2-EGFR and GAB1-SHP2 binding events.

The steps leading to SFK activation are complex, including separate steps for the dephosphorylation of the C-terminal negative regulatory tyrosine (Tyr530 in Src), phosphorylation of the positive regulatory tyrosine (Tyr418 in Src), and binding of SFK SH2 domains to proteins including EGFR (17). We approximated this as a first-order rate process in which phosphorylated EGFR activates SFK, using a previously derived kS,a for EGF-bound EGFR (42), similar to topologies used in previous models (45). Nevertheless, even with this simplification, our model fit could accurately recapitulate our experimental data, indicating that our approach allowed for sufficiently rapid SFK activation to reproduce the GAB1 phosphorylation kinetics observed experimentally. SFK inactivation was modeled with CSK serving as the reaction enzyme, using a previously derived kS,i for CSK (46, 47).

The parameters kG1p, kG1dp, kdp, and kdeg were determined by fitting the model to data gathered from H1666 cells, including data for changes in phosphorylation of Tyr627 in GAB1 and GAB1-SHP2 association in response to EGF or gefitinib (kG1p and kG1dp), changes in phosphorylation of EGFR Tyr1068 in response to gefitinib (kdp), and decreases in total EGFR abundance in response to EGF (kdeg). Parameter fitting was undertaken using simulated annealing to minimize the total error (difference between model output and experimental data), with error computed as the sum of the squares of the differences between model outputs and the experimental values. The best-fit parameter results are included in table S1.

Model sensitivity to changes in parameters was computed by increasing and decreasing parameter values by a factor of 10. Sensitivity was measured by summing the integrated differences between the original model and the two perturbed outputs over time. To compare differences among parameter perturbations, sensitivities were reported as percentages of the maximum perturbed parameter.

Calculations reflect 6 × 105 EGFR proteins per H1666 cell. For the base case model, GRB2, GAB1, SHP2, SFK, and CSK were also assumed to be present at concentrations of 6 × 105 species per cell, based on ranges of previously reported intracellular protein concentrations (44).

Codes were generated and compiled using the Systems Biology Toolbox 2 package for MATLAB (48). The simulannealbnd function in the Global Optimization Toolbox was used to fit the unknown rate constants to experimental data.

Using model parameters and output, various process time scales (τ) were estimated as follows: τGAB1,phos = [(kG1p)(aSFKmax)]−1, where aSFKmax is the maximum concentration of active SFKs possible in response to EGF (10 ng/ml); τGAB1,dephosphorylation = kG1dp−1, τEGFR,phosphorylation = kcatE−1, τEGFR,dephosphorylation = kdp−1; τSHP2,binding = [(kS2,f)(pGAB1max)]−1, where pGAB1max is the maximum concentration of phosphorylated GAB1 possible in response to EGF (10 ng/ml); τSHP2,dissociation = kS2,r−1 and τdiffusion,i = r2/6Di, where r is the cell radius, and Di is the diffusivity of species i.


Fig. S1. Differential rates of EGFR-SHP2 and GAB1-SHP2 complex disassembly in HeLa cells.

Fig. S2. SFK-mediated GAB1 phosphorylation and GAB1-SHP2 binding.

Fig. S3. Effect of EGFR endocytosis on GAB1 and SHP2 cellular distribution.

Fig. S4. Sustained HGF-mediated association of GAB1 and SHP2.

Fig. S5. Model fitting to experimental data.

Fig. S6. Effect of variations in GAB1 phosphorylation (kG1p) and dephosphorylation (kG1dp) rate constants on model error.

Table S1. Model parameters.

File S1. Representative model code for running an EGF and gefitinib treatment simulation.


Acknowledgments: M. Marks, M. Chou, Y. Argon, and L. Holzman provided valuable technical assistance and reagents. M. Frame, M. Lemmon, and B. Neel assisted with valuable scientific discussions. D. Haber, M. Farquhar, G. Nolan, S. Parsons, T. Miller, and T. Hirano provided reagents. A. Walsh and C. Monast provided technical assistance. Funding: This work was supported in part by the National Science Foundation (CBET 1450751). C.M.F. was supported in part by the University of Pennsylvania Training Program in Cancer Pharmacology (R25 CA101871-07) and by a fellowship from the Ashton Foundation. Author contributions: C.M.F., J.M.B., and M.J.L. designed the experiments. C.M.F. and J.M.B. performed the experiments. C.M.F. and M.J.L. designed the computational model, and C.M.F. implemented the model. C.M.F., J.M.B., and M.J.L. wrote the manuscript. Competing interests: The authors declare that they have no competing interests.
View Abstract

Stay Connected to Science Signaling

Navigate This Article