Oncogene addiction has now been well documented in several experimental cancer models and appears to play an important role in the clinical response to various targeted cancer therapies that have been recently developed. However, a clear molecular mechanism to explain the phenomenon of oncogene addiction has been somewhat elusive. Here, we have tested a specific hypothesis that oncogene addiction is associated with differences in the attenuation rates of multiple competing pro-survival and pro-apoptotic signals emanating from a single oncoprotein source. In three different cell culture models of oncogene dependency, we have observed a similar profile of rapid signal attenuation of several key survival effectors, and a temporally delayed accumulation of at least one pro-apoptotic effector, phospho-p38 MAP kinase. This signaling profile is also observed in a human lung cancer cell line, PC-9, which harbors an activating EGF receptor mutation and is efficiently killed by a pharmacological concentration of gefitinib that exhibits efficacy in a subset of treated patients. These observations suggest that a common signaling cascade associated with coordinated timing of signal attenuation may underlie the apoptotic response to acute oncogene inactivation in the context of several different oncogenes and may also contribute to the clinical response to kinase inhibitors in a subset of cancer patients.
In this model of differential signal attenuation (), we propose that, upon disruption of oncogenic activity, a temporary imbalance in pro-apoptotic and pro-survival signals results from the fact that at least some of the established survival factors, such as ERKs, Akt, and STAT3/5 undergo relatively rapid inactivation, whereas pro-apoptotic signals can persist long enough to drive an apoptotic outcome in the absence of counter-acting survival signals. This model implies that a temporal window exists during which lingering “unchecked” pro-apoptotic signals emanating from an oncoprotein that has recently been inactivated cause the cell to pass a commitment point toward apoptosis. In fact, experimental studies have demonstrated that commitment to an apoptotic outcome can occur within only a few hours following the initiation of an apoptotic signal (Brunet et al., 1998
). Our observation (especially in the PC-9 system) that phospho-AKT levels are rapidly inactivated for a brief period in lung cancer cells treated with the EGF receptor kinase inhibitor gefitinib, and then recover prior to apoptosis (), is consistent with a model wherein a transient loss of pro-survival signaling is sufficient to shift the survival-death balance and to produce an apoptotic outcome. In addition, we have determined that transient transfection of cultured NIH3T3 cells with an activated Ras mutant that is specifically defective for interaction with the PI-3 kinase/Akt survival pathway causes apoptosis in a substantial fraction of transfected cells, whereas activated Ras that retains this interaction does not cause apoptosis (Figure S4
Proposed model of oncogene addiction
While there are potentially multiple mechanisms that could contribute to a transient signaling imbalance following oncogene inactivation, the phosphatase inhibitor studies that we have described point to an important role for cellular phosphatases in determining the “steady-state” levels of phosphorylation of the various well studied downstream effectors of many oncogenic kinases. These finding also implicate phosphatases in determining the temporal kinetics of effector activation and deactivation. While okadaic acid is an admittedly blunt tool for elucidating the precise mechanism by which phosphatases might contribute to differential signal attenuation, other reported studies similarly support a role for phosphatase activity in determining the temporal ordering of signal activation and deactivation. For example, previous studies of an apoptosis model in cultured PC12 neural cells that undergo apoptosis upon withdrawal of nerve growth factor revealed a similar temporally coordinated cascade of ERK1,2 inactivation followed by an accumulation of phospho-p38 prior to apoptosis (Xia et al., 1995
). In addition, Akt has been reported to negatively regulate p38 activity, indirectly through the ASK1 and MKK3/6 kinases (Ichijo et al., 1997
; Kim et al., 2001
). Notably, protein phosphatase 5 (PP5), an okadaic acid-sensitive phosphatase, reportedly dephosphorylates ASK1 and inhibits ASK1 signaling (Morita et al., 2001
), thereby providing another potential mechanism by which phosphatase activity could contribute to the timing of signal attenuation. These findings, together with our findings, suggest that a shared profile of orchestrated attenuation of pro-survival and pro-apoptotic signals may play an important role in a variety of apoptosis settings.
We recognize that we have only examined a relatively small subset of the numerous downstream effectors that have been previously implicated in the response to activating oncoproteins such as Src, EGF receptor, and BCR-ABL. Consequently, it is a near certainty that any differential signal attenuation mechanisms that may be operating in the models we have examined are more complex than the simple model we have described. Moreover, we note that multiple distinct mechanisms of oncogene addiction may be contributing to this phenomenon in the various contexts in which it has been observed. For example, the proliferative arrest and differentiation responses to oncogene inactivation that have been reported in some settings may involve distinct mechanisms; although, it remains possible that an analogous signaling imbalance of a different nature could be contributing to those outcomes.
The proposed model of oncogene addiction differs somewhat from previous descriptions of the phenomenon in that it implies that the apoptotic outcome in response to oncogene inactivation is an active process that requires pro-apoptotic signals derived from the oncoprotein, as opposed to a passive process in which a cell that is dependent on an oncogene-derived survival signal defaults to an apoptotic death upon inactivation of that oncogene. It is well established that several oncoproteins, including Ras and Src, can produce pro-apoptotic outputs in some contexts (Arber, 1999
; Webb et al., 2000
). Moreover, most of the well studied oncoproteins, such as EGF receptor, BCR-ABL, and Ras, promote the activation of numerous downstream effector pathways, and in each case, some of these have been linked to pro-apoptotic outcomes depending on the context. Several potential molecular mechanisms to account for a pro-apoptotic outcome have been reported. Thus, Ras can be pro-apoptotic via an interaction with the effector target, Nore1 (Vos et al., 2003
). Similarly, the EGF receptor can bind directly to the so-called “death ligand”, FAS/CD95 (Reinehr et al., 2003
). In addition, Src can induce apoptosis in B lymphocytes via engagement of the CD20 surface protein (Hofmeister et al., 2000
). Thus, it is reasonable to expect that the state of a cell expressing an activated oncoprotein reflects the net balance of multiple diverse signaling pathways that have been engaged.
Accumulating evidence indicates that many tumor cells are poised on the edge of an apoptotic event and that tumor cells have an increased tendency to undergo apoptosis. In most tumor cells, the balance between pro-survival and pro-apoptotic signals clearly favors a survival outcome, however, it is easy to imagine that a temporary imbalance in those pathways in the initial hours following inactivation of the upstream source of those signals could shift cells toward an apoptotic event, particularly if the survival signals dissipate rapidly. Indeed, we have observed that treatment of PC-9 lung cancer cells for as little as 30 minutes (followed by drug washout) is sufficient to kill approximately half of the cells (Figure S5
). Moreover, the differential signal attenuation model may also explain the emerging concept that tumor cells in which a kinase has undergone mutational activation or is overexpressed are especially sensitive to the killing effects of drugs that target those kinases. Specifically, the model suggests that increased kinase activity is associated with increased pro-apoptotic (as well as pro-survival) output, and that this excessive pro-apoptotic activity contributes to the cell death observed following acute inactivation of the oncogenic kinase. In support of such a conclusion, we have observed that cultured lung cancer cells that are killed by gefitinib treatment can be more efficiently killed in the presence of 10% serum, which contains EGF, the receptor ligand (Figure S6
). This may reflect a role for EGF in driving higher levels of an apoptotic output from mutationally activated EGFR. In addition, this could explain why most normal cells, in which these same kinases are not signaling excessively, are typically not killed by treatment with a drug that targets the kinase. Thus, this differential signal attenuation model raises questions about the appropriateness of the term “addiction” in this setting, which seemingly implies a passive dependency on the oncogenic kinase. As an alternative, we have proposed the term “oncogenic shock” to describe the proposed signaling imbalance that can accompany acute oncogene inactivation, and which ascribes an active pro-apoptotic role to the oncogenic kinase in the consequent cell death response (Sharma et al., 2006
). Notably, with such a mechanism, it may not be relevant whether the oncogenic lesion is an initiating event or a late event in tumorigenesis, but rather, its predominance in the tumor may be critical in determining whether targeting it produces a significant clinical response.
The studies we have performed may also have clinical implications regarding the use of cancer therapies that target oncogenic kinases. For example, co-administration of a drug that promotes cell cycle arrest might suppress apoptosis that would normally be triggered by acute inactivation of an oncogenic kinase, a mechanism that has potentially contributed to the disappointing results observed thus far when EGFR kinase inhibitors are administered together with conventional chemotherapy drugs. These findings also raise the possibility that drugs directed against downstream signaling proteins that shift the balance of kinase-mediated pro-apoptotic and pro-survival signals may also have therapeutic value. Differential signal attenuation may also contribute to the acquisition of drug resistance that frequently develops in patients during treatment with selective kinase inhibitors if genetic or epigenetic changes allow cells to survive during a temporal window between the dissipation of pro-survival and pro-apoptotic signals. These findings also raise the possibility that “on-off cycles”, as opposed to continuous treatment, with targeted kinase inhibitors may be more beneficial to patients by allowing for more “windows of opportunity” for the inhibitors to produce a transient signaling imbalance that leads to apoptotic cell death. Thus, in contrast to traditional chemotherapeutic agents that appear to trigger a DNA damage response, the emergence of small molecules that target specific components of signaling pathways may involve distinct mechanisms of cancer cell death. Further insights into these mechanisms may allow optimal clinical application of these new molecularly targeted agents.