Ligand-dependent responses of the ErbB signaling network: experimental and modeling analyses

Marc R Birtwistle^1,2,a, Mariko Hatakeyama^3,a, Noriko Yumoto³, Babatunde A Ogunnaike¹, Jan B Hoek² & Boris N Kholodenko²

^aThese authors contributed equally to this work

Article highlights

Synopsis

The ErbB receptors, ErbB1 (EGFR), ErbB2 (HER2/NEU), ErbB3, and ErbB4, are widely expressed throughout the human body and control cell fate decisions, such as proliferation, apoptosis, and differentiation (Yarden and Sliwkowski, 2001). Given the wide physiological expression of ErbB receptors and their importance in controlling cell fate decisions, it is not surprising that deregulation of ErbB signaling is implicated in the progression of multiple human cancers (Yarden and Sliwkowski, 2001). Because the ErbB signaling network is a complex, dynamic system with multiple feedbacks, it is difficult to understand its overall signaling behavior through traditional, qualitative methods. In the current work, we combine traditional experimental methods with quantitative modeling to understand ErbB signaling in MCF-7 breast cancer cells. We specifically focus on the short-term (30 min) dynamic behavior of all four ErbB receptors and the key downstream intermediates ERK and Akt in response to stimulation with the ligands epidermal growth factor (EGF) and heregulin (HRG) (Figure 1). Based on the schematic in Figure 1, we build an ordinary differential equation model consisting of 117 species, 235 parameters, and 96 net reactions to represent the dynamics of ligand-dependent ErbB network activation. We use this model to generate hypotheses about how the cell dynamically controls ERK and Akt activation in a ligand-dependent fashion, and experimentally corroborate the predictions that EGF-induced ERK activity is much more sensitive than HRG-induced ERK activity to the MEK inhibitor U0126 (Figure 7A and B), and that the phosphoinositol-3 kinase (PI-3K) inhibitor wortmannin has a large affect on post- but not pre-peak EGF-induced ERK signaling.

Figure 1

Simplified schematic representation of the ErbB signaling model. ErbB receptor ligands (EGF and HRG) activate different ErbB receptor dimer combinations, leading to recruitment of various adapter proteins (Grb2, Shc, and Gab1) and enzymes (PTP1-B, SOS, and RasGAP). These membrane recruitment steps eventually lead to the activation of ERK and Akt.

Full figure and legend (24K)Figures & Tables index

Figure 7

U0126 Titration of 5 min ERK activation. (A) Immunoblots are representative of two independent experiments. U0126 concentrations are as follows: (0): control; (1): 10 M; (2): 5 M; (3): 1 M; (4): 500 nM; (5): 250 nM; (6): 125 nM; (7): 62 nM; and (8): 0 nM. (B) Comparison of the experimental U0126 titration and model predictions. Error bars represent the range of the data. ERK activity for each ligand is normalized to its own control, and not a single reference point.

Full figure and legend (125K)Figures & Tables index

Gaining insight into the behavior of such a large model for the purposes of hypothesis generation is a difficult task, and in this work, we relied mainly on parametric sensitivity analysis to accomplish this. Naturally, the sensitivity analysis results revealed many obvious modes of regulation, such as control of ERK activation by MEK, and control of Akt activation by PI-3K. However, these results also revealed non-obvious, ligand-dependent modes of regulation: (i) ERK activation is robust to parametric perturbations at high ligand doses, while Akt activation is not. This difference in robustness to parameter variation may be due to the involvement of ERK in multiple negative feedback loops, and the absence of such feedback from Akt. It is well known from control theory that negative feedback endows a system with robustness to disturbances (Ogunnaike and Ray, 1994; Freeman, 2000). (ii) PI-3K abundance is a dominant factor for the post-peak (declining portion of the ERK response curve) but not pre-peak (rising portion of the ERK response curve) EGF-induced ERK activation. This phenomenon is due to degradation of active ErbB1 homodimers. At short times, there are still many 1-1 homodimers that can signal to ERK; as time progresses, however, these homodimers are degraded. Therefore, at longer times, signaling to ERK must rely on alternative mechanisms—in this case, the PI-3K-Grb2-associated binder 1 (Gab1) pathway.

The results of the sensitivity analysis allowed us to gain general understanding of the ligand-dependent control of ERK and Akt activity. While this general understanding is valuable by itself, understanding how known, cancer-correlated network abnormalities affect signaling is of primary and immediate interest. We used the model to investigate how such a network abnormality, ErbB2 overexpression, which occurs in approximately 25% of breast cancer, would affect the dynamics of ERK and Akt activation. Simulation results suggested that it should have a large effect on signaling at high EGF doses by transforming normally transient signals into more sustained signals. Significantly, this model prediction quantitatively agrees with the data of Wolf-Yadlin et al (2006) and Kumar et al (2007), which shows that at 10 and 30 min after EGF stimulation, ERK activation is between 1.15- and 2-fold higher for human mammary epithelial cells overexpressing ErbB2. This effect of ErbB2 overexpression is a direct consequence of ligand-induced ErbB1 homodimer internalization and degradation. EGF transmits signals only through ErbB1 homodimers and ErbB1-ErbB2 heterodimers, and ErbB1 homodimers undergo preferential ligand-induced internalization and degradation relative to other ErbB dimers (Baulida et al, 1996). As ErbB2 overexpression shifts the ErbB dimer distribution toward more ErbB1-ErbB2 heterodimers rather than ErbB1 homodimers, it therefore sustains signaling due to decreased internalization and degradation.

Our computational analyses of the ErbB signaling network generated a number of hypotheses regarding how MCF-7 cells control ERK and Akt activities in a dynamic and ligand-dependent fashion. These hypotheses can and should be tested against the experiments, and their results will help us and others refine and improve the model. In this study, we performed two experiments to test our hypotheses. To test the hypothesis that HRG-induced ERK activation should be robust to the MEK inhibitor U0126, while EGF-induced ERK activation should not, we performed a U0126 titration on 5-min ERK activity (Figure 7A and B). These results confirmed the prediction that HRG-stimulated ERK activation was more robust than EGF-stimulated ERK activation to U0126. We experimentally investigated the effect of the PI-3K inhibitor wortmannin on ERK signaling, and the results corroborated our prediction that PI-3K is a dominant factor for post-peak but not pre-peak EGF-induced ERK signaling. However, from the wortmannin experiments we also discovered limitations of our current model, which suggested directions for future model improvements, such as incorporation of the tyrosine phosphatase SHP2.

Building a quantitative model of the ErbB signaling network that incorporates stimulation of all four ErbB receptors with two ligands simultaneously was a difficult endeavor. This endeavor was worthwhile, however, as we generated many hypotheses that were only realizable using model-based analysis. Experimental investigation of all these hypotheses is beyond the scope of a single study; however, over time the combined results of many studies can be used to test these hypotheses. When conflicts arise, our current understanding and model of the ErbB signaling network will be revised accordingly.

MORE ARTICLES LIKE THIS

These links to content published by NPG are automatically generated.