Quantitative analysis reveals how EGFR activation and downregulation are coupled in normal but not in cancer cells

Abstract

Ubiquitination of the epidermal growth factor receptor (EGFR) that occurs when Cbl and Grb2 bind to three phosphotyrosine residues (pY1045, pY1068 and pY1086) on the receptor displays a sharp threshold effect as a function of EGF concentration. Here we use a simple modelling approach together with experiments to show that the establishment of the threshold requires both the multiplicity of binding sites and cooperative binding of Cbl and Grb2 to the EGFR. While the threshold is remarkably robust, a more sophisticated model predicted that it could be modulated as a function of EGFR levels on the cell surface. We confirmed experimentally that the system has evolved to perform optimally at physiological levels of EGFR. As a consequence, this system displays an intrinsic weakness that causes--at the supraphysiological levels of receptor and/or ligand associated with cancer--uncoupling of the mechanisms leading to signalling through phosphorylation and attenuation through ubiquitination.

Figure 1. EGFR phosphorylation dose–response curves.

(a) Schematic representation of EGFR ubiquitination (Ub) and total EGFR Tyr phosphorylation (pY tot). x_T represents the half-maximal EGF concentration for EGFR ubiquitination (that is, the ubiquitination threshold). EGF concentration is expressed in a.u. (b) Schematic representation of Tyr phosphorylation sites in the cytoplasmic tail of EGFR-WT (left) and of the add-back mutants (1045+, 1068/86+, 3Y+), some of which were used in c–f. The intracellular domain including the kinase domain (red) and C-terminal tail (black line) of the EGFR are shown. The position of the nine phosphorylatable Tyr residues is shown. Tyr residues involved in Cbl/Grb2 binding and subsequent EGFR ubiquitination are indicated in blue, while the other Tyr residues are depicted in black. Phosphorylation of Y1045 (c) or Y1068 (d) in EGFR-WT or the indicated add-back mutants. Experimental points are taken from ref. . Phosphorylation is plotted, for each condition, as normalized to the maximum pY value obtained in that condition (pY/pY_MAX). (e) Experimentally determined dose–response phosphorylation curves for EGFR-WT and EGFR-3Y+. Experimental points are taken from ref. . EGFR phosphorylation is expressed, for each condition, as normalized to the maximum value obtained in that condition (pY/pY_MAX). (f) Experimental ratio of total Tyr phosphorylation (pY tot) of EGFR-WT and EGFR-3Y+ after 2 min of stimulation with the indicated concentrations of EGF. Data are derived from densitometry analysis of IBs from three independent experiments±s.d. (see Supplementary Fig. 1 for representative IB). In (c–f) EGFR-WT and mutants were expressed in NR6 fibroblasts, which are devoid of endogenous EGFR.

Figure 2. Modelling EGFR phosphorylation: the free-kinase regime accounts for experimental phosphorylation curves.

(a) Schematic representation of the enzymatic reactions that lead to EGFR phosphorylation in an active dimer. Top, Michaelis–Menten reaction whereby the tyrosine kinase (TK) domain of an EGFR binds reversibly to the C-terminal tail of a partner EGFR with the binding/unbinding rate constants k_on and k_off, respectively. After adding one phosphate group, the TK dissociates from the substrate with rate constant k_cat. Bottom, the simplified reaction scheme for Tyr phosphorylation is characterized by k_KIN. (b) Two limiting regimes (free and saturated) can be identified for the reaction catalysed by kinases, depending on the stability of the complex between the catalytic subunit of EGFR and its Tyr substrate. Curves represent the average of 10⁵ runs of the stochastic Gillespie algorithm applied to the standard Michaelis–Menten reaction scheme depicted in a (see Supplementary Note 1 for details). (c) Wiring diagram of the MPM. Top, phosphorylation of individual Tyr residues (blue circles) occurs independently of the phosphorylation of the other Tyr residues. This results in a branched wiring diagram, in which each phosphorylation event occurs with the same probability. Bottom, EGFR molecules (red circles) with the same total number of phosphoryl groups (blue circles) are grouped together to generate a linear chain of increasingly phosphorylated EGFRs. Only the three Tyr residues relevant for EGFR ubiquitination are shown (that is, Y1045, Y1068, Y1086). and are the rates of addition and subtraction of one phosphoryl group from an EGFR molecule that carries α-phosphorylated Tyr residues. (d) A comparison of EGFR-WT and -3Y+ phosphorylation computed by the MPM in the free-kinase/free-phosphatase regime as a function of EGF (model, solid lines) or determined experimentally (exp, dashed lines: data taken from Fig. 1e). EGFR phosphorylation was normalized to the maximum pY value (pY/pYmax). (e) A comparison of the ratio of total pY of EGFR-WT and EGFR-3Y+, as a function of EGF concentration (at 2 min), computed by the MPM (dashed line) or determined experimentally (solid line, data taken from Fig. 1f, see also Supplementary Fig. 1).

Figure 3. Cbl is limiting for EGFR ubiquitination.

(a) Reactions involved in EGFR ubiquitination. (1) Cbl binds to pY1045 irrespective of whether it is free or in a complex with Grb2 (Grb2 is thus represented with a dotted line); (2) Grb2 binds to pY1068 (or pY1086, not represented in the scheme) irrespective of whether it is free or in a complex with Cbl (represented with a dotted line); (3) Cbl and Grb2 bind to each other; (4) if the Cbl/Grb2 complex binds to EGFR via reaction (1), the binding described by (2) occurs via a first-order reaction (Supplementary Note 1, Eq. (8)). Similarly, if the initial binding takes place via reaction (2), the ensuing reaction (1) will be first order. Reaction (3) can also become first order, if it is preceded by reaction (1) for Cbl only and by reaction (2) for Grb2 only (not shown). According to the cooperative hypothesis, first-order reactions are favoured (thick arrows). (b) Quantitation of Grb2 and Cbl molecules in HeLa cells. Panels 1, 2, 3: increasing amounts of HeLa cell lysate were subjected to IB and IP as indicated and compared with increasing amounts of in vitro purified Grb2 (1) or Cbl (2, 3) proteins, as described in Methods. Bottom: table indicating the amount of critical players involved in the EGFR ubiquitination reaction in HeLa cells. The number of surface EGFR molecules was measured by ¹²⁵I-EGF saturation binding (see Methods). Data are expressed as number of surface EGFRs per cell. Average results, calculated from at least three independent experiments±s.e.m., are shown. (c) Top, HeLa cells, transfected with empty vector (Vector), WT Cbl (overexpression, OE) or Cbl70Z mutant, were treated with EGF at the indicated concentrations for 2 min. Lysates were IP and IB as shown (see also Supplementary Fig. 3a). Bottom, quantitation of the effect of Cbl OE or Cbl70Z expression on EGFR ubiquitination by densitometry analysis of IBs, as shown in the upper panel, from three independent experiments. EGFR ubiquitination is expressed, for each condition, as normalized to the maximum value obtained in the empty vector control (Ub/Ub_WT).

Figure 4. The MPM-B reproduces the EGFR phosphorylation and ubiquitination dose–response curves.

(a) Comparison of experimental (dashed lines) and modelled (solid lines) phosphorylation (pY) and ubiquitination (Ub) dose–response curves for EGFR-WT (left) and EGFR-Y1045+ (right). Note that modelled curves are for EGFR-3Y+, since it behaves as EGFR-WT when normalized to the max (Fig. 1e). Experimental data are expressed for each condition as normalized to the maximum value for that condition; simulations are normalized to optimized maxima (see Supplementary Table 1). Inset shows the ratio of ubiquitination to phosphorylation as a function of EGF concentration for experimental and modelled data. Simulations were performed with MPM-B in the left panel, and a modified version of MPM-B where we set to zero all reactions leading to phosphorylation of site Y1068, in the right panel. In both panels, data were obtained from densitometry analysis of IBs shown in Supplementary Fig. 1 and ref. . Average results, calculated from at least three independent experiments, are shown. Error bars in the plots represent the s.e.m. (b) Comparison of experimental (dashed lines) and modelled (solid lines) EGFR ubiquitination dose–response curves under conditions of Cbl modulation. Experimental data were taken from Fig. 3c. Cbl OE and downregulation (Cbl 70Z) were modelled as a twofold increase and decrease in the MPM-B, respectively. Data are expressed, for each condition, as normalized to the maximum value in the empty vector control (Ub/Ub_WT, left) or to their own maximum (Ub/Ub_MAX, right). Error bars in the plots represent the s.e.m. calculated form at least three independent experiments.

Figure 5. Cooperativity is required to generate the EGFR ubiquitination threshold.

(a) Effect of simulations with a cooperative versus noncooperative MPM-B model on the EGFR-3Y+ ubiquitination threshold. In the cooperative model, the presence of enforced proximity of Cbl, Grb2 and EGFR (cooperative, red line) increases the steepness of ubiquitination as compared with phosphorylation (black solid line). In the noncooperative regime, phosphorylation and ubiquitination share the same steepness. Simulations of the noncooperative model (blue dashed line) were performed with MPM-B, except that the parameters kb45^*, kb68^* and kbcg^* (see Fig. 3a for symbols' definitions) were reduced by setting fLOC=1, as explained in Supplementary Note 2, such that the conditions kb45*>>kb45 × [Cbl], kb68*>>kb68 × [Grb2] and kbcg*>>kbcg × [Grb2] are no longer verified. (b) The distribution of singly (1-B, black lines) and doubly (2-B, red lines) bound EGFR-Cbl complexes changes in the presence (cooperative, solid lines) or absence (noncooperative, dashed lines) of an enforced-proximity mechanism. Simulations were performed as in a.

Figure 6. The EAM reproduces the EGFR phosphorylation and ubiquitination curves.

(a) Wiring diagram of the EAM for an EGFR with three phosphorylatable Tyr. To reduce the complexity of the model, we did not introduce dimers explicitly; however, with a change of variables, we followed them as individual EGFR molecules that can either be monomeric or dimeric. Thus, EGFRs in a dimer can either be competent for phosphorylation and, therefore, activation or not, depending on whether the partner EGFR is bound to EGF. Each chemical species represents a single moiety of EGFR that can be either a monomer (circles and ovals) or one of the two members of a dimer (squares). EGFR moieties are characterized by different attributes: (i) presence (black) or absence (grey) of ligand; (ii) closed (ovals) or extended (circles) conformation; (iii) number of pYs (circled P). EGFR moieties in dimers may be inactive (can only be dephosphorylated, blue) or active (can be either phosphorylated or dephosphorylated, red). Opening/closing and phosphorylation/dephosphorylation follow first-order kinetics, while EGF binding (EGF is not explicitly shown) and the transition from monomer to dimer are second-order reactions. The kinetics of this latter reaction is proportional to the concentration of all species taking part in dimerization. (b) Comparison of experimental (dashed lines) and modelled (solid lines) phosphorylation and ubiquitination dose–response curves for EGFR-WT. Note that modelled curves are for EGFR-3Y+, since it behaves as EGFR-WT when normalized to the max (Fig. 1e). Inset shows the ratio of ubiquitination to phosphorylation as a function of EGF concentration for experimental and modelled data. Experimental data are the same as those shown in Fig. 4a left, and have been normalized for the maximum value for that condition. Simulations are normalized to optimized maxima (see Supplementary Table 1). (c,d) Sensitivity analysis. We varied each parameter by 1% and computed the sensitivity coefficient σ for the Hill coefficient n_H (c) and for both the phosphorylation half-maximum level pY_0.5 and the ubiquitination threshold x_T (d), for EGFR phosphorylation (blue) and ubiquitination (red). See Supplementary Note 3.

Figure 7. Downmodulation of EGFR levels shifts the ubiquitination dose–response curve.

(a) We varied each parameter of the EAM 10-fold and computed the sensitivity coefficient S for the EGFR ubiquitination threshold position x_T. Only parameters whose variation resulted in a sensitivity coefficient of at least 0.1 are reported. (b) EGFR knockdown (KD) in HeLa cells was achieved by transfection with an anti-EGFR siRNA oligo. Control cells were transfected with mismatched oligo. Cells were then treated for 2 min with EGF as indicated. IP and IB were as shown. Quantitation of EGFR ubiquitination is shown in c. (c) Comparison of model predictions and experimental assessment of the ubiquitination threshold in EGFR-KD (KD) and control (WT) HeLa cells. Simulations were performed with EAM (in the simulation of the EGFR-KD the decrease in EGFR levels was assumed to be 4.2-fold, as determined in ¹²⁵I-EGF saturation binding assays). EGFR ubiquitination is expressed, for each condition, as normalized to the maximum value obtained in that condition (Ub/Ub_MAX). Experimental data are reported as mean±s.d. from at least three independent experiments.

Figure 8. Advanced model for EGFR Ub and pY as a function of EGF concentration and EGFR number.

(a) Left, relative EGFR phosphorylation (pY/EGFR_T, black lines) and ubiquitination (Ub/EGFR_T, red lines) levels, as given by the EAM, for the indicated EGF concentrations. The grey area represents the physiological range of EGFR levels. Dashed lines indicate the maximal phosphorylation and ubiquitination. Data are normalized to the maximum phosphorylation/ubiquitination at 100 ng ml⁻¹ EGF. Red squares and black circles represent experimental measurements of EGFR Ub and pY, respectively, obtained by the ELISA-based assay in NIH-EGFR clones with increasing numbers of EGFRs (from lowest to highest: NIH-EGFR^phy , physiological EGFR; NIH-EGFR^m-ov, medium overexpression; NIH-EGFR^h-ov, high overexpression). Right, representative immunofluorescence images of EGFR surface levels in the indicated NIH-EGFR clones. Scale bar, 18 μm. Bottom, number of surface EGFRs per cell, as measured by ¹²⁵I-EGF saturation binding. (b) Left, relative EGFR ubiquitination (left, Ub/EGFR_T, red line) and phosphorylation (right, pY/EGFR_T, black line), as given by the EAM, for the indicated EGF concentration, normalized to the maximum. Squares represent experimental measurements of EGFR ubiquitination (left) or phosphorylation (right), obtained by the ELISA-based assay in the indicated NIH-EGFR clones (red) or cell lines with increasing EGFR number (black, see also Supplementary Fig. 9). (c) Relative EGFR ubiquitination levels (Ub/EGFR_T), as given by the EAM, at 100 ng ml⁻¹ of EGF under control (red line) or Cbl overexpressing (red dashed line) conditions (modelled as a × 100 increase in line with data in Fig. 4b). Data are normalized to the maximum ubiquitination in each condition. Red squares and circles represent experimental measurements of EGFR ubiquitination, obtained by the ELISA-based assay, in the NIH-EGFR clones (as in a) infected with a lentiviral vector driving inducible Cbl overexpression (Cbl^oe, red circles), or empty vector (Vector, red squares), treated with doxocycline. Right, verification of Cbl overexpression in the indicated NIH-EGFR clones infected with empty vector (−) or vector driving inducible Cbl overexpression (+); tubulin, loading control. Densitometry analysis revealed an ∼80–100-fold Cbl overexpression of Cbl. Experimental data in a–c are reported as mean±s.d. from at least three independent experiments.