| HOME | HELP | CONTACT US | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Clinical Pharmacology |
Medical Oncology Branch, National Cancer Institute, National Institutes of Health, Bethesda, Maryland, USA
Key Words. Targeted agents • Drug resistance • Tyrosine kinase inhibitors • Imatinib • Bcr-Abl • c-KIT • Epidermal growth factor receptor (EGFR) inhibitors • Gefitinib • Erlotinib • Stem cells
Correspondence: Tito Fojo, M.D., Ph.D., Medical Oncology Branch, National Cancer Institute, Building 10, Room 12N226, 9000 Rockville Pike, Bethesda, Maryland 20892, USA. Telephone: 301-402-1357; Fax: 301-402-1608; e-mail: tfojo{at}helix.nih.gov
Received May 15, 2007; accepted for publication November 6, 2007.
Disclosure: No potential conflicts of interest were reported by the author, planners, reviewers, or staff managers of this article.
Among targeted therapies for solid tumors, it can be argued that the most effective therapy to date was conceived nearly 25 years ago and only recently approved. Aromatase inhibitors, the brainchild of Angela Brodie, were finally approved for the treatment of breast cancer in 2004 as a result of, in large part, the perseverance of Dr. Brodie, Dr. Charles Coombes, and a dedicated group of clinical investigators at the Royal Marsden Hospital in London [1–3]. It was around the same time 25 years ago that farnesyl transferase inhibitors were first described, and while the final chapter on this story has yet to be written, the early promise of this targeted therapy has been largely followed by a series of disappointments as it became apparent that not only farnesyltransferase but also geranylgeranyltransferase I could generate biologically active Ras [4–6]. Then, about 15 years ago, Brian Druker, together with Nicholas B. Lydon and other investigators at Ciba-Geigy, began the process of identifying a compound that would inhibit the breakpoint cluster region–Abelson (Bcr-Abl) protein and that ultimately would lead to the approval of imatinib mesylate (Gleevec®; Novartis Pharmaceuticals Corporation, East Hanover, NJ) for chronic myelogenous leukemia (CML). Five years ago, the approval of imatinib, and the identification of the proteasome inhibitor bortezomib (Velcade®; Millennium Pharmaceuticals, Inc., Cambridge, MA) as an agent with activity in multiple myeloma, created an enthusiastic frenzy for new targeted therapies that has yet to abate [7–10].
As we enter the 5th or 15th or 25th anniversary of targeted therapies, evidence has confirmed what we all know but often ignore: solid tumors are very complex. A large group of investigators initially reported the sequence analysis of 13,023 "well annotated" human protein-coding genes in 11 breast and 11 colon cancers and then updated the analysis to include 18,191 genes [11, 12]. They reported that, on average, individual tumors accumulated approximately 90 mutant genes and judged 191 of the genes identified (1.4%!) to be candidate cancer genes (CAN-genes). The average number of mutant CAN-genes was 12 (range, 4–23) for breast cancers and 9 (range, 3–18) for colorectal cancers with a distinct CAN-gene signature found for each cancer. Furthermore, the mutation spectrum was markedly different between breast and colon cancers with all the differences highly significant (p < .0001). According to the authors, "the vast majority of these genes were not known to be genetically altered in tumors and are predicted to affect a wide range of cellular functions, including transcription, adhesion and invasion" and they subsequently concluded that "the genomic landscapes of breast and colorectal cancers are composed of a handful of commonly mutated gene ‘mountains’ and a much larger number of gene ‘hills’ that are mutated at low frequency." The complexity of solid tumors becomes clearer when one considers that only about one fourth of all genes were sequenced, that regulatory regions were not examined, and that neither epigenetic changes nor post-translational modifications nor the role of micro-RNAs were assessed. Indeed, it appears solid tumors are as complex as many had imagined, and more so than most had feared.
Given this complexity, one could argue that, on our current path, it is unlikely that many if any of our "targeted therapies" will cure cancer. It may be that the best one can hope for is some measure of stable disease. Some may argue that this has been the only promise given us by those who, in developing targeted therapies, sought to make cancer a chronic disease. However, we must recognize what "chronic" has translated to until now. In kidney cancer, for example: a 5-month improvement in median survival for sorafenib (Nexavar®; Bayer Pharmaceuticals Corporation, West Haven, CT) when censored for crossover data, a 6-month longer median time to progression for sunitinib (Sutent®; Pfizer, Inc., New York) versus interferon, and a 3.6-month improvement in survival for temsirolimus (CCI-779; Wyeth Pharmaceuticals, Inc., Madison, NJ) [13–15]. We have seen some measure of progress, especially in a disease long regarded as refractory to small molecules, but are a long way—a very long way—from how most of us would define a chronic disease. So what lessons might we learn from our past failures and partial successes? I would argue that we have learned that we need to aim at a target or at targets whose engagement has broad consequences, using a "dirty drug." Consider the following.
The targets of the cytotoxic agents used successfully for so long in the therapy of cancer, whether tubulin, DNA, or the topoisomerases, share one common and important attribute: broad consequences follow their engagement. It is these broad consequences that make these agents toxic, but they also make them effective. It was this knowledge that led many to argue at the most recent dawn of the era of targeted therapy—just a few years ago—that targeted therapies would not only be more effective but also less toxic. While the jury on toxicity is still out, the early evidence suggests that the assumption that greater specificity would result in greater efficacy may have been premature. Indeed, it can be argued that the early evidence suggests that for some, if not most, cases of cancer therapy, dirty may be good, or in any case better than clean. Thus, for example, of the compounds originally developed by Sugen Pharmaceuticals, one of its least specific kinase inhibitors has been approved clinically (sunitinib). Similarly, sorafenib, which was developed as a Raf kinase inhibitor, failed to benefit patients with cancers where Raf was felt crucial and instead found a niche in renal cell carcinoma, where the target is still speculative, but is certainly not Raf, and most likely not even a single target. Indeed, the often stated mechanism of action for sorafenib and sunitinib in kidney cancer—inhibition of vascular endothelial growth factor and angiogenesis—has precious little clinical evidence to support it, and will likely be proven incorrect or at best insufficient [16, 17].
These examples, together with the disappointing results with a myriad of other targeted therapies, are now fueling a drive for agents with broader specificity, including multi-targeted kinase inhibitors, based on the premise that inhibiting more than one target will work best. For example, a recent study examining cell lines, xenotransplants, and primary tumors found multiple receptor tyrosine kinases (RTKs) co-activated in glioblastoma multiforme (GBM) [18]. These observations suggest that redundant "parallel" inputs drive and maintain downstream signaling, and would be expected to limit the efficacy of therapies targeting a single RTK. That concomitant inhibition is required for greater efficacy was demonstrated by the need for combinations of RTK inhibitors, and/or RNA interference, to reduce signaling, cell survival, and anchorage-independent growth. The authors concluded, "effective GBM therapy may require combined regimens targeting multiple RTKs." Given the complexity of solid tumors such as GBM, this makes sense. It always did. Drugs should have broad consequences. However, I would go beyond this to argue that a drug's greater efficacy might eventually reside not only in its ability to target more than one kinase, but also in its ability to more effectively target a given kinase—by avoiding the emergence of resistance. The thesis of this work is that targeted therapies can avoid resistance if they are made less precise, or more dirty, rather than more potent. While such drugs may in fact target multiple kinases, I argue that dirty drugs will be less likely to be dislodged from their binding sites and in turn be more effective. Thus, a hidden attribute of a dirty drug will be its ability to be less susceptible to the development of resistance. Given the short duration of responses in most solid tumors to date, the ability to prevent or delay the emergence of resistance is an important attribute for a drug.
How might they avoid the emergence of resistance and what is the evidence for this thesis? Consider, for example, what we know about microtubule targeting agents (MTAs), compared with what we have learned from the use of imatinib in CML. As investigators who have worked with MTAs looked on enviously, those working with imatinib and CML found and published one resistance-conferring mutation after another. Contrast that with the fact that, despite the widespread use of MTAs for nearly 50 years, mutations in tubulin have never been reported clinically in a tumor refractory to an MTA. Even in cell culture models, resistant cells with mutations at the binding sites for MTAs have infrequently been isolated [19, 20]. In fact, in some cell culture models examining resistance to MTAs, mutations have been found with a much higher frequency in regions far removed from the binding site, and these have been shown to alter the intrinsic stability of the microtubule—not drug binding [20–24]. This alteration in the intrinsic stability in turn confers resistance—a more stable microtubule is more difficult to destabilize and thus resistant to destabilizing agents such as the vinca alkaloids; conversely, a less stable microtubule is more difficult to stabilize and is resistant to microtubule-stabilizing agents such as paclitaxel. But why do mutations occur infrequently and why should mutations at sites other than the binding site be preferred? The most obvious explanation is that MTAs are poisons, not substrates. We must distinguish between a drug that binds as a substrate at a substrate-binding site and a poison that binds elsewhere on a target—even a well-defined site such as those where MTAs bind. Tubulin and the microtubules it forms do not possess the precise binding site found on most kinases, because the MTAs are not its substrates. Even the poisons that trap the topoisomerases in "cleavable complexes" rarely result in mutations that affect binding. The proposal that agents such as the topoisomerase inhibitors and the MTAs target protein interfaces and "take advantage of transient structural and energetic conditions created by the macromolecular complex, which give rise to ‘hot spots’ for drug binding" may also help to explain the fact that mutations that affect drug binding occur rarely in vitro and have never been reported in patients [25]. But an explanation that is likely just as important as the fact that these are poisons not substrate analogues is that Mother Nature developed agents that were "sufficiently dirty" and that oncologists in turn usurped them for a different purpose. While the MTAs effectively target human tubulin, Mother Nature clearly did not develop them with human tubulin as a target. Consequently, although the tubulins are highly conserved in evolution, most MTAs used clinically likely associate imprecisely to their binding sites on human tubulin—at least compared with our kinase inhibitors. Indeed, the dissociation constant for paclitaxel on human tubulin under conditions that one can argue are more relevant to rapidly dividing cancer cells with dynamic microtubules is about 2.5 µM, while that of imatinib for the Abl kinase is tenfold lower at about 100–300 nM [7, 26–28]. Consistent with this is the fact that, although there is one paclitaxel binding site on every molecule of tubulin in a microtubule, a 1:1 stoichiometric binding can be achieved only at high concentrations of paclitaxel, and in fact, at cytotoxic concentrations only a very small number of paclitaxel molecules are bound and this is sufficient to bring about cytotoxicity. For example, all that is needed to reduce the rate or extent of microtubule shortening by 50% is for one paclitaxel molecule to be bound per several hundred tubulin molecules. Indeed, with MTAs, as with topoisomerase poisons, cytotoxicity can be achieved even if binding is not optimal, because a low occupancy rate of the target is sufficient, in part because the targets are "structural" macromolecules [29, 30]. As discussed below, because of this less precise binding of MTAs and the indirect binding of topoisomerase poisons that intercalate in DNA and secondarily engage topoisomerase, mutations that alter drug binding and confer resistance are less likely to emerge. Thus, it might be that the efficacy of MTAs and topoisomerase poisons and their ability as single agents or in combination regimens to cure cancer resides in their inherent imperfections. Given this, one can argue that less perfect drugs might be better. Have we any evidence to argue this?
Consider Bcr-Abl in CML, a target that is essential and that leads to a constitutively active mitogenic signal as one mechanism in the malignant transformation of CML [31]. With our current drug selection approach, how crucial a kinase is to a cancer cell can in part be defined by whether mutations that affect drug binding to the kinase emerge as a mechanism of resistance—and by this criterion, Bcr-Abl is very crucial for CML. But for a cell to be resistant, the mutation must both prevent drug binding and also allow for continued function of the crucial kinase. That some bcr-abl mutations have been identified over and over again in the leukemic cells of patients with disease refractory to imatinib speaks to the fact that the mutations that impair imatinib binding also allow the kinase to retain activity [32]. Because they result in variable kinase activity, different mutations allow for different growth rates—the mutant Bcr-Abl proteins have varying mitogenic activity [33, 34]. But this is where one could argue for a less precise, even dirty, agent. A drug that fits tightly can be easily dislodged without much perturbation of the protein, allowing kinase activity to be retained—and mutations, at least at the binding site, might occur more readily. Is there any evidence for this? Actually, very little, because the question of whether tight-binding drugs can be displaced more readily while sustaining a functional kinase has not been asked; and indeed, to date could probably only be asked with inhibitors of Bcr-Abl and the epidermal growth factor receptor (EGFR). Nilotinib, a second-generation Bcr-Abl inhibitor whose design was based on the imatinib mesylate scaffold, has been studied extensively both in vitro and in patients, and was found in vitro to be to be 20 times more potent than imatinib against unmutated Bcr-Abl [28]. From published data we know, for example, that at concentrations approximately twofold higher than the 90% inhibitory concentration against unmutated Bcr-Abl in cell proliferation assays (2 µM imatinib and 50 nM nilotinib), an approximately 25% higher incidence of mutations is found in the presence of nilotinib compared with imatinib—with a high incidence of mutations observed with both drugs [35]. By contrast, however, to dislodge a dirty drug that fits less tightly, the amino acid change needs to be more drastic and less likely to allow retention of kinase activity—and in turn, be of no survival value. That it would take a more drastic change to disrupt ATP binding and kinase activity is intuitive because the affinity of most kinases for ATP is weak, in the micromolar range [36–41]. ATP binding sites are very accommodating and have low affinities for ATP; because the intracellular concentrations of ATP are millimolar the affinity can be low [42]. We should remember that ATP is a substrate for nearly 800 kinases, and as we are learning, ATP binding sites are quite different—or at least sufficiently different that synthetic chemists have been able to make inhibitors that target one or at most a few kinases with some ease. Indeed, the original concern that inhibitors targeting a given kinase would inhibit many different kinases was incorrect. Because ATP does not bind with great affinity, drugs that compete effectively either directly or indirectly can be easily made. If the drug is one that targets the ATP-binding site, a drug with lower affinity will still be able to compete with ATP—competing ATP is actually not so difficult—and may be more difficult to dislodge. Such a drug may also target other kinases and could cause more toxicity, although the latter is not certain. But it would likely be less susceptible to the development of resistance. So, in fact, the challenge is not to make a very precise agent—this has been done already. The challenge is to make a drug with sufficient affinity to bind the kinase, and disrupt ATP binding, but not so precise that it can be easily dislodged by a simple mutation.
But can less precise drugs really work? Won't we find that tumors can acquire mutations to confer resistance to any drug? Actually, as regards a tumor's adaptability, there is some hope in the emerging story of imatinib resistance—hope that the spectrum of mutations is not limitless. As the data from CML have shown, the clones harboring different CML mutations do not grow at comparable rates either in vitro or in patients [33, 34]. Disease recurrence, as measured by Bcr-Abl transcripts, shows a range of growth that spans a log or more. While this may reflect, at least in part, the ability of the resistant clone to generate the committed progenitors crucial for expansion of the population, it also reflects the differential sensitivity to imatinib of the various mutations and the inherent mitogenic activity of the mutant kinase. Because clonal selection favors the fastest growing cell, one would have expected that one or at most just a very few mutations that optimally prevent imatinib binding while retaining robust kinase activity should emerge in leukemic cells refractory to imatinib—assuming, in a given patient, that the population of leukemic cells had inherently or could acquire a wide spectrum of mutant kinases. But, despite a restricted spectrum of mutations in the original report, to date >70 different point mutations, leading to the substitution of over 50 amino acids in the Abl kinase domain, have been identified in leukemic cells resistant to imatinib, a number that is likely to increase with time [32, 43–45]. What does this plethora of mutations tell us? First, that drug binding was too precise and that inhibiting imatinib binding while retaining kinase activity was easy to accomplish—many different mutations in the binding site, in the P-loop, and in the activation loop could accomplish this. But we can also infer that the spectrum of mutations in the leukemic cells of a single given patient was limited. Had the leukemic population in these patients been composed of cells with a very diverse mutation spectrum to select from, then the most efficient mutations would have been isolated over and over again. That this has not happened suggests that a malignancy may have a limited number of mutations. This limited spectrum of mutations gives one hope that drug combinations or different drugs might have a greater likelihood of success. But success with drug combinations will not come from using imatinib and a drug similar to imatinib that can inhibit the growth of cells harboring mutations that confer imatinib resistance [46]. The second drug must be sufficiently different (or sufficiently dirty) to require a different spectrum of mutations so that, when used in combination with imatinib, the emergence of a mutation that can affect both drugs will not occur—imatinib-type mutations cannot dislodge it. Given the success of the N-ethyl-N-nitrosourea (ENU)-based mutagenesis screen in identifying mutations at the kinase domain, one could argue that potential drug candidates should undergo similar screening and only be advanced clinically as single agents or in combinations if the spectrum of the mutations identified is shown to be different [35]. This concept can be viewed as a modified version of the Goldie-Coldman hypothesis of the 1980s. Goldie and Coldman envisioned that a tumor would be less likely to have the different mutations needed to confer resistance to two or more agents and argued in favor of combination therapy [47]. In the case of CML, one is arguing that, given a finite spectrum of mutations, the likelihood that a single leukemic cell will harbor a single mutation that can confer resistance to more than one drug is reduced. And its likelihood will be further reduced if the mutations that can both confer resistance and retain kinase activity are fewer—a goal that can best be accomplished using drugs with lower affinity, dirty drugs. To be sure, a strategy that uses two specific inhibitors in combination is an attractive alternative that is currently undergoing clinical evaluation. While Goldie and Coldman would predict that the likelihood of success would be greater, for this strategy to succeed a kinase as crucial as Bcr-Abl in CML will need to be the target. The strategy of targeting one crucial kinase with more than one inhibitor is of course different from that of targeting multiple kinases simultaneously, an approach that might be more applicable to the majority of cancers other than CML that lack as crucial a target as Bcr-Abl or have multiple putative targets—although the data in glioblastoma multiforme with multiple kinase inhibitors can be described at best as modestly successful even in an in vitro assay [18]. Nor does the strategy of multiple inhibitors for one crucial kinase address the issue of resistance mechanisms other than mutations that affect drug binding, as has been reported in a gefitinib-sensitive lung cancer cell line that developed resistance to gefitinib as a result of focal amplification of the met proto-oncogene and not EGFR mutations. In this cell line, inhibition of Met signaling restored gefitinib sensitivity [48].
The imatinib story also reminds us that resistance is an inherent property of a malignancy or will almost certainly be acquired. Claims that a drug is not susceptible to the emergence of resistance means that investigators have simply not looked hard enough. That non-small cell lung cancer patients with mutant EGFR experience such wonderful responses to gefitinib (Iressa®; AstraZeneca Pharmaceuticals, Wilmington, DE) and erlotinib (Tarceva®; Genentech, Inc., South San Francisco, CA) only to be faced with an incurable recurrence is clear evidence that, even in a sensitive tumor, resistance will emerge [49–51]; and in a substantial fraction of patients whose tumors harbor EGFR tyrosine kinase domain mutations that render their tumors sensitive to EGFR inhibition, the acquired resistance has been shown to be associated with a second-site EGFR mutation [52]. Similarly in CML, the emergence of resistance to imatinib is evidence that, as with our traditional cytotoxic agents, newer targeted therapies are also vulnerable to this problem. A silver lining to this problem may be the experience with dasatinib and nilotinib (AMN107) clearly showing that, for at least a subset of patients, a newer and possibly better agent can be developed [53–55], albeit, at least in the case of dasatinib, a dirtier agent [56]. Furthermore, we must recognize that mutations in the kinase-binding site have been identified in only 50%–70% percent of patients whose tumors develop resistance to imatinib, suggesting that mechanisms such as drug efflux or influx transporters might account for other cases of drug resistance [57–60]. Although one can take the pessimistic view that a malignancy will eventually find a mechanism of resistance, another way to interpret the observation that not all refractory leukemic cells harbor bcr-abl mutations is to recognize, again, that the spectrum of bcr-abl mutations is limited. Otherwise, the simplest way to become resistant—selecting for the clone with the perfect Bcr-Abl kinase mutation—would have occurred in 100% of imatinib-refractory leukemic cells. Although drug efflux pumps have not been convincingly identified as important clinically, this may be in part because the precise drugs used to date with their high affinities do not need very high intracellular concentrations and can tolerate some pump mediated efflux—an alternate way to view this is that the pumps are unable to reduce the concentrations below those that would be required to impair the binding of these high-affinity compounds. But with a drug that has a lower affinity, "every little bit may count" and even low endogenous levels of an efflux pump may then become important. Attention may need to be given to this in drug development. In this regard one should note that the ENU-based mutagenesis screen identifies mechanisms of resistance resulting from point mutations [35]. Consequently, other mechanisms, such as gene amplification, rearrangements, or epigenetic changes, that can lead to increased expression of a protein will be missed. The final lesson to be gleaned from the experience with imatinib is that acquired mutations as a single mechanism of resistance occur in tumors only if the drug can effect a clinical complete or near complete remission that allows a single clone harboring the mutation that confers resistance to emerge. Specific mutations are less likely to be found following partial responses. This observation has been made in CML, where mutations are usually found in patients who relapse after an initial response to imatinib, as opposed to those with primary resistance [45]. So, for example, the progression of disease invariably seen with sorafenib and sunitinib in patients with kidney cancer is unlikely in most patients to be mediated by a point mutation, but rather reflects a less specific mechanism of resistance that was present and variably expressed in a large fraction of tumor cells at the outset of therapy [13, 17].
Given all this, one can ask whether drug development should be honed to synthesize increasingly precise drugs. As our three-dimensional models become increasingly sophisticated and precise should we strive to synthesize increasingly more perfect drugs? Do we really want the standard of a perfect fit to be a cancer drug and its target and not a glove and a hand? Discussing data derived from the development of tyrosine kinase inhibitors, I have argued that the answer to this should be "no," because a dirty drug might be better because it targets several kinases, and also because it binds a crucial kinase less precisely and is thus more difficult to dislodge. The mutation required to prevent drug binding is then so severe that it is not compatible with kinase activity and thus of no value to the cell in its struggle to adapt. Evidence supporting or refuting this hypothesis should emerge from clinical trials, maybe even from a comparison of dasatinib and nilotinib. Admittedly, a dirty agent may not only be more effective but also more toxic, but such a conclusion should not be reached a priori. Finally, we should recognize that paclitaxel was developed and refined by Mother Nature, our best chemist, over billions of years, while mere humans develop our targeted therapies in pharmaceutical laboratories over a space of a few years. So if we don't get it perfectly right the first time, we must not give up hope.
 |
REFERENCES |
|---|
 Top References |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | CONTACT US | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| THE ONCOLOGIST | STEM CELLS | CME | ALPHAMED PRESS JOURNALS |
