Our analysis implicates a reversible drug-tolerant state in the acute response of cancer cell populations to a lethal drug exposure. The collective findings reveal a subpopulation of cancer cells that transiently exhibit a distinct phenotype characterized by the engagement of IGF-1R activity, hypersensitivity to HDAC inhibition, altered chromatin, and an intrinsic ability to tolerate drug exposure, which does not involve drug efflux (Fig. S6C
). Reversible drug tolerance appears to reflect dynamic heterogeneity within a cancer cell population that can be established even following the clonal expansion of single drug-sensitive cells. Such phenotypic heterogeneity has been observed in some clonally-derived normal mammalian cells, such as stem cells (Chang et al., 2008
; Stewart et al., 2006
), and has been implicated in cancer cell fates following drug exposure in culture (Cohen et al., 2008
; Gascoigne and Taylor, 2008
The ability of the drug-tolerant subpopulation to maintain viability following an otherwise lethal drug exposure appears to involve IGF-1R engagement. This was observed in several tested cell line models, suggesting a potentially broad role for IGF-1R signaling in drug tolerance. Notably, IGF-1R activation has been linked to drug resistance and poor prognosis in several cancer settings (reviewed in (Casa et al., 2008
; Pollak, 2008
). Furthermore, several published reports describing cell culture models of acquired resistance to both TKIs and conventional chemotherapy drugs have similarly demonstrated the activation of IGF-1R in drug-resistant derivatives (Buck et al., 2008
; Chakravarti et al., 2002
; Dallas et al., 2009
; Eckstein et al., 2009
Our findings also implicate a distinct chromatin state in the maintenance of the drug-tolerant subpopulation, and the histone demethylase KDM5A was identified as at least one chromatin-modifying enzyme required to establish this state. Notably, reduced methylation of H3K4 has been linked to poor prognosis in cancer patients (Seligson et al., 2005
). It is certainly possible that additional chromatin-modifying enzymes contribute to drug tolerance in various tumor contexts. Indeed, our findings also suggest a role for decreased histone acetylation in this process. Although the regulation of histone demethylase activity is poorly understood (Lan et al., 2008
), our results suggest a role for IGF-1R signaling in modifying KDM5A activity, which at least partly involves the suppression of KDM5A expression.
A transiently maintained drug-tolerant state could provide a mechanism that allows a small subpopulation of tumor cells to withstand an initial onslaught of drug or other stressful stimuli to enable their survival for a period of time until more permanent resistance mechanisms can be established. This is highly reminiscent of the properties of antibiotic-tolerant bacterial subpopulations, also called “persisters”, which similarly exhibit a transient ability to endure potentially lethal stresses (Balaban et al., 2004
; Dhar and McKinney, 2007
). These are slower growing cells, whose survival within a more rapidly proliferating cell population is ensured by the fact that they can readily revert to a non-persister state via epigenetic mechanisms. Consequently, the “burden” of protecting the population from eradication is shared equally among all members of the population (Lewis, 2007
). Our findings suggest that a subpopulation of drug-tolerant cancer cells may behave similarly, and that all of the tumor cells in a population potentially have the ability to stochastically acquire and relinquish this protective phenotype at a low frequency. Our collective findings support a striking analogy between bacterial and cancer cell-derived persisters; thus, both populations reflect dynamic phenotypic heterogeneity of a non-genetic nature, both exhibit intrinsic multi-drug tolerance that does not involve drug efflux, and upon drug withdrawal, both give rise to progeny that are as susceptible to drug exposure as their ancestors. These similarities raise the possibility that a tumor cell population invokes more “primitive” properties associated with microbial populations to ensure survival.
Our findings suggest a more complex “pathway” to stable genetically-conferred resistance to cancer drugs than is implied by the detection of specific drug resistance mutations in tumors. Such mutations are generally thought to arise spontaneously at low frequency in tumor cells prior to drug treatment and are selected during treatment. However, our observations implicate a multi-step “process” (Fig. S15
) mediated by metastable drug-tolerant states associated with chromatin alterations. Importantly, the proposed model is not incompatible with “pre-existing” resistance-conferring mutations. Thus, while drug resistance mutations, such as EGFR
T790M, may be present in rare tumor cells prior to EGFR TKI exposure, they might also arise from reversibly drug-tolerant cells. Significantly, accumulating evidence supports a role for stress-induced mutagenesis as an adaptive mechanism both in bacteria and in cancer cells (Galhardo et al., 2007
), raising the possibility that an increased mutagenesis rate within drug-tolerant cells leads to a greater opportunity for drug resistance mutations to emerge.
The relationship between the reversibly drug-tolerant subpopulation and cancer stem cells is potentially complex. Although DTPs display markers associated with CSCs, their ability to survive lethal drug exposure does not involve drug efflux, a property attributed to at least some drug-resistant CSCs. Moreover, during the transition of DTPs to DTEPs, CSC-specific markers are lost, and yet both cell populations are equally drug-insensitive. Emerging studies of CSCs have clearly revealed their reduced sensitivity to a variety of toxic exposures (Eyler and Rich, 2008
), and recent studies have demonstrated that exposure of mouse tumors to certain chemotherapy drugs can cause tumor regression yielding a population of drug-refractory cells with CSC properties (Kang and Kang, 2007
). Considered with our studies, such findings point to a likely relationship between a reversibly drug-tolerant cancer cell subpopulation and the CSC subpopulation. However, this relationship appears to be complex and certainly merits further exploration.
Reversible drug tolerance may account for accumulating clinical reports demonstrating that cancer patients treated with a variety of anti-cancer drugs can be successfully re-treated with the same drug after a “drug holiday”. The detection of a distinct chromatin state in drug-tolerant cancer cells and consequent hypersensitivity to HDAC inhibitors potentially yields a therapeutic opportunity to prevent the development of stable drug resistance. To test this possibility, we have recently initiated a clinical study to examine the ability of combining a chromatin-modifying agent with erlotinib in NSCLC patients. While the trial is not yet completed, the early clinical data indicate that the inclusion of a chromatin-modifying agent can dramatically improve clinical benefit in a subset of patients demonstrating acquired TKI resistance (S.V.S. and J.S., unpublished observation). When considering that acquired drug resistance may involve multiple distinct molecular mechanisms that arise independently within the same patient, thereby complicating strategies to overcome such resistance with a single rationally-targeted agent, the potential ability to prevent the development of resistance by disrupting a drug-tolerant state is provocative. However, further studies will certainly be required to establish the in vivo significance of the cell culture findings, as well as to determine more precisely the molecular mechanisms underlying reversible drug tolerance.