PERSPECTIVES
`
`O P I N I O N
`
`Why is cancer drug discovery
`so difficult?
`
`Alexander Kamb, Susan Wee and Christoph Lengauer
`Abstract | Thirty-five years after the ‘war on cancer’ was declared, the discovery
`of anticancer drugs remains a highly challenging endeavour. Here, we consider
`the factors responsible, such as tumour heterogeneity, and suggest strategies to
`improve the chances of short-term success in the development of novel
`anticancer drugs.
`
`Oncology has one of the poorest records for
`investigational drugs in clinical develop-
`ment, with success rates that are more than
`three times lower than for cardiovascular
`disease1,2 (FIG. 1). The widespread, relentless
`and lethal nature of cancer persists, with
`only incremental overall improvements
`in treatment outcomes, despite billions of
`dollars of public and private investment.
`The few notable successes, such as imatinib
`(Gleevec; Novartis) in the treatment of
`chronic myeloid leukaemia (CML), are,
`so far, exceptions.
`Acknowledging that drug discovery
`is difficult in general, here we discuss the
`specific obstacles that the anticancer-drug
`hunter must confront. We cover the two
`most popular therapeutic modalities: low-
`molecular-mass drugs and unconjugated
`biologicals. Although optimal drug-discovery
`programmes aim to integrate the different
`stages of the research and development
`process into a single coherent operation,
`we break the subject into three different
`elements: targets, drugs and patients.
`After assessing the impediments to oncology
`drug discovery, we recommend specific
`strategies to combat this disease.
`
`Targets: essential versus non-essential
`For anticancer drug targets, the most
`fundamental distinction is between those
`that have essential functions and those that
`have non-essential functions. In this context,
`essential means that at least one vital cell type
`in the human body depends on the target for
`survival. Inhibitors of essential functions are
`
`likely to have narrow therapeutic windows,
`owing to the requirement for their targets in
`normal cells. By contrast, drugs that target
`non-essential proteins (the large majority
`of the proteome) should be well tolerated,
`but their efficacy might be limited unless
`the proper tumour type can be defined. In
`mammals especially, discrimination between
`essential and non-essential genes is not always
`easy. Mouse knockout mutants provide a
`convenient test for genetic dispensability.
`Such experiments reveal that the majority of
`proto-oncogenes — as defined by their muta-
`tion, amplification or activation in tumours
`— are essential. So, drugs that target activated
`oncogenes run the risk of having serious
`side effects, although this test for viability
`is admittedly stringent because it assesses
`requirements for gene function throughout
`development, and not only in the adult.
`Intentional inhibition of essential func-
`tions to kill cancer cells results in on-target
`or on-mechanism toxicity in normal
`cells, and clinicians rely on differences in
`dose–response and drug distribution within
`tumours and normal tissues to provide a
`therapeutic window. Even though normal
`tissue is remarkably robust, there are thresh-
`olds below which survival is not possible.
`For instance, humans cannot survive a 90%
`tissue loss in most organs. However, it is
`evident that if 10% of tumour cells continue
`to proliferate in the face of anticancer treat-
`ment, the therapeutic regime will have little
`effect on ultimate disease outcome. This
`disparity is one of the stark challenges of
`cancer therapy.
`
`There are nearly 200 drugs approved for
`cancer, with hundreds more in development,
`and they represent a wide assortment of
`mechanisms and modalities (TABLE 1). For
`many decades, drug discovery focused on
`agents that block essential functions and kill
`dividing cells — the traditional cytotoxics.
`These drugs include compounds with pleio-
`tropic effects, such as DNA-modifying agents
`(for example, cisplatin), as well as drugs
`that interfere very precisely with defined
`physiological processes such as microtubule
`polymerization (for example, taxol), metabo-
`lite synthesis (for example, methotrexate)
`and chromosome topology (for example,
`irinotecan). An exception to the historical
`focus on targeting essential functions are anti-
`hormonal therapies such as oestrogen-
`receptor modulators (for example, tamoxifen)
`and aromatase inhibitors (for example, letra-
`zole). Anti-hormonals target activities that are
`classified as non-essential because these func-
`tions relate to the proliferation of specialized
`but dispensable normal tissues, such as breast
`epithelium, for example.
`In the past few years, various novel, tar-
`geted agents have burst onto the scene. Some
`of these agents bind to proteins that are essen-
`tial in all cells, and therefore are not easily
`distinguished from traditional cytotoxics. The
`latest agents include those that target cell divi-
`sion in new ways (for example, aurora-kinase
`inhibitors and cyclin-dependent-kinase
`inhibitors), as well as other processes such as
`protein turnover (for example, bortezomib
`(Velcade; Millennium) and chromatin
`modification (for example, histone-deacety-
`lase inhibitors)3–6. These new drugs might
`reasonably be called ‘neocytotoxics’. Although
`these drugs often stem from sophisticated,
`target-driven screening and medicinal-
`chemistry efforts, it is not immediately clear
`what advantages they offer compared with
`traditional cytotoxics. Nonetheless, they con-
`tinue to attract interest based on the possibili-
`ties of interfering with different biochemical
`mechanisms and using new chemical types.
`In some cases, broader therapeutic windows
`might be ultimately achieved by refining the
`paralogue selectivity of the compounds, such
`that they avoid inhibiting essential cellular
`functions and target only the members of a
`protein family that are prominent in tumours.
`
`NATURE REVIEWS | DRUG DISCOVERY
`
` VOLUME 6 | FEBRUARY 2007 | 115
`
`Abraxis EX2080
`Cipla Ltd. v. Abraxis Bioscience, LLC
`IPR2018-00162; IPR2018-00163; IPR2018-00164
`
`

`

`Response rate
`Treatment-related death rate
`
`8 7 6 5 4 3 2 1
`
`b
`
`Rate (%)
`
`P E R S P E C T I V E S
`
`20
`
`15
`
`10
`
`5
`
`a
`
`Percentage of success
`
`0
`
`1991–1994
`
`1999–2002
`
`Oncology
`
`All
`
`0
`1995–1998
`Cardiovascular CNS Infectious
`Period
`disease
`Figure 1 | The challenge of anticancer drug development. a | Historically, oncology compounds
`have had a significantly lower success rate in clinical development than compounds in other areas,
`such as cardiovascular disease. The rates shown are the success rates from first-in-human to registra-
`tion for ten large pharmaceutical companies in the United States and Europe for the period
`1991–2000. Data taken from REF. 1. b | Low response rates in Phase I oncology trials. Trends in
`response and treatment-related (toxic) death rates for studies initially submitted to Meetings of the
`American Society of Clinical Oncology 1991–2002 (REF. 2). The contribution of trial-level data to the
`period average was weighted by the number of enrollees. Error bars indicate standard error. Part a
`reproduced, with permission, from REF. 1 © (2004) Macmillan Magazine Ltd. Part b reproduced, with
`permission, from REF. 2 © (2004) American Medical Association.
`
`(also known as MS4A1), a B-cell marker, to
`bind and destroy non-Hodgkin’s lymphoma
`cells. This effect improves two-year survival
`from about 85% with chemotherapy alone,
`to 95% (REF. 12). Trastuzumab is effective in
`breast tumours that overexpress HER2/neu
`protein through DNA amplification. The
`subset of patients that express HER2/neu
`(about a third of those with node-positive
`breast cancer) experience a 50% reduction
`in disease recurrence over a period of 20
`months13. This is a substantial improvement,
`but it also illustrates that some breast cancers
`might have pre-existing resistance to a useful
`drug, whereas others might acquire resistance
`and progress.
`
`Drugs: selective versus multi-targeted
`A good target is useless without a corre-
`spondingly good cognate drug. Apart from
`favourable pharmaceutical properties, the
`goal of many drug-optimization efforts is a
`molecule that inhibits its target in a selective
`or carefully crafted way. Many traditional
`cytotoxic drugs, although demonized with
`epithets such as ‘slash and burn’ agents,
`make highly specific interactions with
`their molecular targets. Methotrexate, for
`instance, binds at picomolar concentrations
`to dihydrofolate reductase (DHFR) and has
`a multi-log-fold preference for DHFR over
`its secondary target, thymidylate synthase.
`From the chemistry and pharmacology
`perspectives, these are excellent compounds.
`The collateral damage they produce is
`strictly on-mechanism, a result of the
`biological roles of their targets.
`Toxicity can also be off-target, derived
`from the inhibition of unintended or
`unknown functions. In most cases, clinical
`effects are probably a blend of on-target and
`off-target activities. As clinicians escalate
`dose, the possibility for off-target effects
`increases, adding to toxicity, but possibly
`to efficacy as well. With more physiological
`functions compromised by the drug, side
`effects inevitably arise, but the increased
`stress in the tumours might offset the burden
`of toxicity to a point. Off-target side effects
`are common for small-molecule drugs and
`are likely to have a larger role for inhibitors
`that target sites that are conserved among a
`family of proteins such as kinases. One of the
`clear distinguishing features of biologicals,
`including antibodies, is the reduced likeli-
`hood of unknown off-target interactions.
`Proteins generally make highly selective con-
`tacts with their targets, and clinical failures
`due to unpredicted off-target toxic effects
`ought to be minimal. Rather, efficacy and
`on-target toxicity are the principal concerns.
`
`Other recently developed small-molecule
`drugs inhibit elements in key signalling
`pathways, mostly kinases, which might
`not be essential in normal adult cells. In
`this way, they offer an approach to cancer
`therapy that is, in principle, distinct from
`the traditional and neocytotoxics. Imatinib
`is the prototype for this class of new cancer
`drugs. The target it was designed to hit,
`Ableson kinase (ABL), is activated in CML
`cells by a chromosomal translocation, which
`creates a unique dependency on this specific
`protein. As judged by the mouse knockout
`phenotype, animals have limited require-
`ment for ABL activity. This is presumably
`one of the reasons that imatinib is highly
`effective in CML and is well tolerated as
`chronic therapy7. Both of these features are
`rare for cancer drugs.
`It is important to note that many of the
`familiar protein kinases have essential func-
`tions. Proteins such as AKT, mammalian
`target of rapamycin (mTOR; also known as
`FRAP1), and extracellular signal-regulated
`kinase (ERK; also known as MAPK1) are
`embedded in crucial survival and prolifera-
`tion pathways. This raises the possibility that
`inhibitors of these components might behave
`much like cytotoxic drugs, with narrow
`therapeutic windows. By contrast, kinases
`such as ABL, epidermal growth factor recep-
`tor (EGFR), and HER2/neu (also known as
`ERBB2) lie upstream in signalling networks.
`Therefore, they are less likely to be required
`in all cell types and less apt to generate broad
`cytotoxicity. Consistent with this view, side
`
`effects for small-molecule drugs that target
`ABL and EGFR generally conform with the
`genetic and expression data — myelosup-
`pression and skin rash, respectively. By
`contrast, mitogen-activated protein kinase
`(MAPK)/ERK kinase (MEK; also known
`as MAP2K1) and mTOR inhibitors have a
`broader toxicity profile at their maximum
`tolerated doses (MTDs), along with modest
`efficacy at these doses8,9.
`Bevacizumab (Avastin; Genentech),
`a monoclonal antibody against vascular
`endothelial growth factor (VEGF), targets
`activities — vascular genesis and physiology
`— which some predicted would be essential
`in normal adults. However, the drug has
`acceptable side effects and, depending on
`the perspective, its performance in the clinic
`has been either phenomenal or mildly disap-
`pointing10. In several studies where the anti-
`body is administered in combination with
`chemotherapeutics, bevacizumab extends
`survival by a few months in patients that
`have historically been refractory to all new
`treatments. From an efficacy standpoint, it
`seems to behave as a useful chemotherapeu-
`tic and is active in a broad array of cancers,
`with statistically significant, yet limited,
`effectiveness in end-stage tumours11.
`Rituximab (Rituxan; Genentech) and tras-
`tuzumab (Herceptin; Genentech) are mono-
`clonal antibodies that bind non-essential
`proteins that have restricted expression
`in adult tissues. The basis for the efficacy
`of these drugs is still a matter of debate.
`Rituximab relies on expression of CD20
`
`116 | FEBRUARY 2007 | VOLUME 6
`
` www.nature.com/reviews/drugdisc
`
`Abraxis EX2080
`Cipla Ltd. v. Abraxis Bioscience, LLC
`IPR2018-00162; IPR2018-00163; IPR2018-00164
`
`

`

`Activation of apoptosis
`pathways
`ABL (Gleevec;
`Interference with signal
`Novartis)
`transduction, response
`Cathepsin K
`Inhibition of tumour spread
`Telomerase
`Induction of senescence
`VEGF (Avastin;
`Interference with blood supply
`Genentech/Roche)
`of tumour
`Antibody-directed cytotoxicity CD20 (Rituxan;
`Biogen Idec/
`Genentech)
`Microtubules (Taxol)
`
`Signalling
`
`Invasion/metastasis
`Immortalization Senescence
`Host
`Angiogenesis
`
`Tumour-associated
`membrane proteins
`
`Traditional
`cytotoxics
`
`Replication/
`cytokinesis
`Metabolism
`
`Neocytotoxics
`
`Protein turnover
`
`Interference with DNA
`synthesis, cell division
`Reduction of essential
`metabolite
`Inhibition of acceleration of
`protein degradation
`
`Table 1 | Therapeutic mechanisms of action of anticancer drugs
`Therapeutic
`Targeted
`Mechanism of action
`target or
`process
`of therapeutics
`modality
`Transformation
`
`Apoptosis
`
`Target example
`(drug)
`
`BCL2
`
`Thymidylate synthase
`(5-FU)
`Proteasome
`(Velcade; Millennium
`Pharmaceuticals)
`HDAC interactions
`
`ATPase/chaperone
`superfamily
`
`Epigenetics
`
`Stress response
`
`Remodelling chromatin, DNA
`methylation
`Interference with cellular stress
`buffering
`
`ABL, Ableson kinase; BCL2; B-cell lymphoma 2; HDAC, histone deacetylase; VEGF, vascular endothelial growth factor.
`
`The concept of a multi-targeted agent, or
`‘dirty drug’, is widely discussed in oncology14.
`Many clinical-stage kinase inhibitors are
`fairly non-selective15. In our experience,
`clinicians tend to prefer multi-targeted
`drugs because they seek to maximize the
`chance for clinical antitumour activity.
`They are experts at managing attendant
`toxicities. Scientists, on the other hand,
`prefer specific drugs because their effects
`are more predictable. The less selective
`the compound, the more unreliable the
`conclusion about the root cause of its
`activity. Because cancer models are notori-
`ously problematic, it is risky to advance
`compounds on the basis of suppression
`of tumour growth in disease models16.
`Often, only the correlation between phar-
`macodynamic effect and pharmacological
`exposure provides some comfort that the
`observed effects are on-target. Indeed, the
`mechanism of the antitumour activity of
`imatinib was only confirmed in the clinic
`when drug-resistant tumours emerged with
`mutations in the catalytic domain of the
`oncogenic fusion protein BCR–ABL17.
`Recent clinical results for two multi-
`targeted kinase inhibitors illustrate the
`pros and cons associated with such drugs.
`Sunitinib (Sutent; Pfizer) and sorafenib
`
`(Nexavar; Bayer/Onyx) interfere with
`several kinases including VEGF receptors
`(VEGFRs) and platelet-derived growth factor
`receptors (PDGFRs). Sunitinib received
`FDA approval on the basis of its activity
`against gastrointestinal stromal tumour
`(GIST) and renal cell carcinoma (RCC), and
`sorafenib has produced signs of efficacy in
`RCC. Although reasonable hypotheses for
`clinical activity in both settings have been
`advanced, it remains unclear which kinases
`are involved in the responses. In the case of
`RCC, the idea that VEGFR has an important
`role in the therapeutic effect of the drug is
`strengthened by the observation that beva-
`cizumab has some activity in RCC patients
`as a single agent, an effect that cannot easily
`be attributed to antibody-dependent cellular
`cytotoxicity18. But differences among the
`therapeutic agents in patient progression
`and survival rates are mysterious19. A clear-
`cut benefit of having multiple activities in
`a single drug is demonstrated by imatinib
`which, despite being one of the most selec-
`tive kinase inhibitors known, interferes with
`the functions of c-KIT and PDGFR as well
`as ABL. Imatinib is effective against GIST
`and hypereosinophilia, due to its inhibition
`of c-KIT and (presumably) PDGFR,
`respectively20,21.
`
`P E R S P E C T I V E S
`
`Patient selection: drug response as a QTL
`Exceptional heterogeneity and adaptability
`are cardinal features of cancer. Pathologists
`have classified tumours into dozens of his-
`tological subtypes, and have further graded
`them to reflect the degree of progression of a
`particular subtype. This classification scheme
`only begins to capture the variability among
`cells within a tumour and among different
`tumours. At the molecular level, it is likely
`that no two cancers are identical. The range
`of the disease is probably wider than for any
`other therapeutic area. Superimposed on
`the germline differences that distinguish
`individual people at the genomic level is epi-
`genetic variation that is derived from the cell
`types from which the tumours originate, as
`well as further genetic and epigenetic changes
`that accumulate as these deviant somatic cells
`evolve in the body.
`The physical manifestation of tumour
`heterogeneity is reflected in observed differ-
`ences in drug responses, and is the probable
`cause of acquired resistance. Variants in a
`population of tumour cells might have a
`selective advantage under conditions that
`are imposed by cancer therapy and could
`produce clones of drug-resistant cells22.
`Tumour heterogeneity is also a logical
`explanation for pre-existing drug resist-
`ance in cancer patients. Parameters that
`are presumably related to drug sensitivity
`and tumour aggressiveness display a wide
`range of variation among malignancies
`(TABLE 2). Notwithstanding imatinib and
`occasional idiosyncratic ‘Lazarus responses’
`in other therapeutic settings, it seems that
`drug response is, in general, a continuous
`variable. Even a drug that is highly selective
`for a particular target encounters numerous
`mechanisms that might affect the sensitivity
`of a tumour. The anticancer activity of a
`drug might depend on the dozens of cellular
`efflux pumps, proliferation rate, checkpoint
`apparatus, repair processes and apoptotic
`machinery, to mention only a few possibili-
`ties. In aggregate, small differences in any of
`these mechanisms could produce a signifi-
`cant effect. Studies in cell lines and primary
`malignant cells support the view that drug
`response is a quantitative trait, much like
`height in the human population23. Most
`drugs that are tested against a panel of cell
`lines or primary tumour specimens display
`unimodal, continuous variation of activity24
`(FIG. 2). The extreme bimodal antiprolifera-
`tion in dose–responses that are observed in
`some panels of cell lines and tumour types
`with certain therapeutics represent excep-
`tional cases. Oncologists generally speak of
`‘responsive’ or ‘sensitive’ tumours, but this
`
`NATURE REVIEWS | DRUG DISCOVERY
`
` VOLUME 6 | FEBRUARY 2007 | 117
`
`Abraxis EX2080
`Cipla Ltd. v. Abraxis Bioscience, LLC
`IPR2018-00162; IPR2018-00163; IPR2018-00164
`
`

`

`organisms, the possibility of unravelling the
`genetic basis of a continuously distributed
`drug response in tumours seems slight in
`the near term.
`It is well known that clinical toxicity
`and efficacy are difficult to predict from
`preclinical experiments or theory. Efficacy,
`especially for small-molecule drugs, is
`nearly always dose-related, so clinicians
`push cancer drugs to the MTD in clinical
`development. This strategy deals with
`heterogeneity in an empirical, practical way.
`A Phase I trial design that includes multiple
`cancer types ensures a broad sampling
`of clinical heterogeneity. Occurrences of
`drug activity — for instance, with histone-
`deacetylase inhibitors in cutaneous T-cell
`lymphoma — can be followed up in subse-
`quent focused studies26. Clinicians might
`further balance the chance of increased
`efficacy against the chance of increased
`toxicity by combining two or more drugs.
`This approach only makes sense if the
`combination maintains or widens the
`therapeutic window. To provide an advan-
`tage, a drug combination must enhance
`the effect on tumour cells without an
`equivalent increase in toxicity. Once again,
`the trade-off between efficacy and toxicity
`is resolved in the clinic, although it is some-
`times guided by preclinical experiments.
`It remains to be seen how the industry and
`the regulatory agencies will react to strate-
`gies that hinge on advancing pairs of agents
`that are expected to have significant activity
`only in combination, and not as single
`agents27. Barring breakthroughs in patient-
`selection methods or a windfall of imat-
`inibs, the ‘all comers’ clinical-development
`approach remains a valid — if frustrating
`and expensive — route to drug approval.
`
`Imatinib: new standard or outlier?
`The outstanding — even revolutionary
`— qualities of imatinib tend to obscure some
`clinical features that raise concerns about
`generality. Imatinib works substantially
`better in the earlier chronic phase of CML
`compared with the later blast-crisis phase
`(97% versus 49% haematological response)7
`(FIG. 3). In addition, many of the responders
`in blast crisis relapse within months. This
`behaviour raises the possibility that targeted
`agents might run into some of the same
`obstacles that have plagued cytotoxics: low
`responsiveness of advanced cancers and
`acquired resistance.
`The emerging story of erlotinib, another
`clinically successful signal-transduction
`inhibitor, also raises this concern. Erlotinib
`(Tarceva; OSI/Genentech), which targets
`
`the EGFR kinase, prolongs survival of
`non-small-cell lung cancer patients28.
`Early clinical studies with this inhibitor and
`its cousin, gefitinib, also an EGFR inhibitor,
`indicated that it might work far better in
`patients whose tumours have activating
`EGFR mutations29. However, subsequent
`studies have revealed that responses linked
`to these mutations are not durable, and that
`tumours with such mutations seem to be
`generally more sensitive to chemotherapeu-
`tics, not only to erlotinib/gefitinib30 (FIG. 4).
`Therefore, the link to EGFR inhibition
`might be misleading.
`It is perhaps too early to draw firm
`conclusions, but the current data indicate
`that imatinib, rather than being a new
`paradigm, might be an exception. The
`clinical successes of imatinib, erlotinib and
`trastuzumab inspired the idea that genetic
`mutations or amplifications signified a
`dependency of the tumour on a particular
`protein. Drugs that targeted a mutant or
`overexpressed protein were considered
`likely to produce impressive single-agent
`responses. However, evidence is accumulat-
`ing that, at least in the advanced and more
`lethal stages of cancer, tumours might have
`already abrogated this dependency, or can
`easily do so. Blast-crisis CML is a case in
`point. The chronic-phase tumours are over-
`whelmingly sensitive, whereas the late-stage
`blast-crisis cells are often resistant. Those
`tumours that respond often subsequently
`develop variations in the target that result
`in resistance or circumvent inhibition of the
`target. Although a substantial fraction of
`this acquired resistance can be reversed by
`drugs such as the second-generation BCR–
`ABL inhibitor dasatinib (Sprycel; Bristol-
`Myers Squibb), which has broader activity
`against clones of BCR–ABL mutants that
`are insensitive to imatinib, remissions are
`transitory31. It is unlikely that a single BCR–
`ABL inhibitor can simultaneously possess
`sufficient breadth of activity to inhibit all
`mutant enzymes that arise, and also have
`adequate selectivity to be safe. Whether
`these properties can be mimicked by
`particular combinations of drugs, as in the
`current standard of care for HIV infection,
`is an open question. Yet even if such broad
`coverage can be achieved, the addiction
`of a tumour to BCR–ABL might diminish
`as the malignancy evolves. Experiments in
`mouse genetic models of cancer support
`this view32. Mice that are engineered with
`controllable oncogenes reveal that, in sev-
`eral situations, tumours arise that initially
`depend on the oncogene, but gradually lose
`this reliance.
`
`P E R S P E C T I V E S
`
`Table 2 | Variation in tumour biology
`Parameter
`Range
`Cell-cycle period
`30–60 hours
`Apoptotic index
`0.1–4.0%
`Proliferative index
`1–70%
`S-phase fraction
`0.01–0.40
`Data from REF. 22.
`
`is misleading, a myth perpetuated by the
`arbitrary classification of clinical outcomes
`into categories such as ‘stable disease’ and
`‘partial response’. In this light, responsive
`tumours are those with a sensitivity to a drug
`that is sufficiently shifted from the MTD to
`generate a clinical response.
`Quantitative traits have proved difficult
`to dissect, perhaps because their underlying
`genetic determinants are multifactorial
`(polygenic) and nonlinear25. Therefore, we
`might expect quantitative trait locus (QTL)
`interactions to pose similar analytical
`obstacles in somatic cells. Whatever germ-
`line differences exist between two patients’
`tumours are likely to be augmented during
`malignant growth. Therefore, the number
`of germline QTLs (that is, those present in
`the normal genome) plus somatic QTLs
`(that is, those that arise in malignant cells
`during growth) that potentially contrib-
`ute to drug response is staggering. This
`becomes especially apparent if we expand
`the definition of somatic QTLs to include
`heritable changes other than alterations
`of the DNA sequence; that is, epigenetic
`changes in the broadest sense. Based on
`experiments in highly tractable model
`
`90
`
`70
`
`50
`
`30
`
`p = 0.0002
`
`p = 0.3573
`
`p = 0.0039
`
`Mean OD (% untreated control)
`
`10
`
`WT
`PM
`ITD
`Figure 2 | Highly variable anticancer drug
`response might be a quantitative trait. Dot
`plot of the cytotoxic response (mean of triplicate
`experiments) of individual samples to CEP-701 at
`the 100-nM dose level, grouped by FLT3 mutation
`status. Bars show the group mean ± SEM. P values
`are from Student’s t test. ITD, internal tandem
`duplication; OD, optical density; PM, point muta-
`tion; WT, wild type. Figure reproduced, with
`permission, from REF. 24 © (2004) American
`Society of Hematology.
`
`118 | FEBRUARY 2007 | VOLUME 6
`
` www.nature.com/reviews/drugdisc
`
`Abraxis EX2080
`Cipla Ltd. v. Abraxis Bioscience, LLC
`IPR2018-00162; IPR2018-00163; IPR2018-00164
`
`

`

`P E R S P E C T I V E S
`
`With respect to patient stratification
`— the area where the most significant
`advances will ultimately be achieved —
`we believe that the challenges of quantita-
`tive traits are sufficiently great to preclude
`significant progress in the cases of continu-
`ous response distributions. Rather, it is
`advisable to concentrate on drugs for which
`there is evidence in cell lines or tumours
`of bimodal sensitivity. We consider the
`analogy between simple Mendelian traits
`and normal height variation in a popula-
`tion on the one hand, and CML and the
`bulk of epithelial cancers on the other,
`as instructive. The underlying basis for
`dwarfism is accessible, while the molecular
`origins of quantitative differences in height
`remain obscure. When clear qualitative
`differences in tumour response to drugs
`exist, the underlying molecular cause might
`reveal itself, as for CML and the BCR–ABL
`translocation.
`We emphasize that our proposals are
`not a prescription for basic cancer research,
`which can afford to take a long-term
`view. Rather, we seek feasible, short-term
`solutions to the problems of cancer drug
`discovery. Although immensely challeng-
`ing, the impediments to predictable cancer
`therapy are not insuperable, nor are the
`molecular underpinnings of drug response
`and cell survival unknowable. Ultimately,
`cancer must yield to a systematic and
`sustained assault.
`
`15
`
`20
`
`0
`
`5
`
`10
`Months
`Chemotherapy, wild type (n = 99)
`Chemotherapy, mutant (n = 14)
`Erlotinib + chemotherapy, wild type (n = 99)
`Erlotinib + chemotherapy, mutant (n = 15)
`
`1.2
`
`1.0
`
`0.8
`
`0.6
`
`0.4
`
`0.2
`
`0
`
`Survival rate
`
`Figure 4 | Target-based stratification might
`prove inadequate. Kaplan-Meier curves by
`treatment received and epidermal growth factor
`receptor (EGFR) mutation status30. P = 0.958 for
`erlotinib plus chemotherapy versus chemother-
`apy alone among patients with EGFR-mutant
`tumours (dashed lines) and P = 0.294 for erlotinib
`plus chemotherapy versus chemotherapy alone
`among patients with wild-type tumours (solid
`lines). All P values refer to log-rank tests.
`Adapted, with permission, from REF. 30 © (2005)
`American Society of Clinical Oncology.
`
`which all the key ingredients for success,
`some controllable but many not, come
`together: first, a proliferative disorder
`with dependency on a single target that is
`non-essential in most normal cells; second,
`a relatively selective kinase inhibitor; and
`third, a clear way to select patients who will
`respond to the drug.
`
`Prescriptions
`If hope for the emergence of further drugs
`like imatinib is misplaced, at least in the
`short term, what angle should scientists
`working in oncology drug discovery take?
`We suggest that a merger of thoughtful
`innovation with practical experimental
`plans is most realistic. With regard to target
`selection, a balance between essential and
`non-essential functions seems prudent. It
`is probable that continued efforts to inhibit
`essential proteins might only produce
`incremental benefits to patients. Exclusive
`focus on non-essential targets, however,
`will produce more failures through lack
`of efficacy, but successful drugs will have
`wider therapeutic windows. The identifica-
`tion of non-essential targets that tumours
`have come to rely on requires better
`computational and experimental models
`— no simple task. The possibility of find-
`ing so-called synthetic-lethal drug targets,
`which are only essential in cancer cells
`that carry mutations in specific tumour-
`suppressor genes or oncogenes, is attractive
`in theory33. However, the search for such
`genes — if they exist — might be frustrated
`by tumour heterogeneity and awkward
`tools for somatic-cell genetics. In the mean-
`time, it seems sensible to use cancer models
`in conservative ways; that is, to study
`cell-autonomous functions with robust
`phenotypes, or use the models purely to test
`the pharmaceutical properties of candidate
`drugs. Non-autonomous functions that
`involve multiple cell types are, in general,
`too complex to model in a reliable way.
`Tests of compelling biological rationales
`should be reserved for the clinic.
`With regard to drug development,
`we favour the philosophy of high chemical
`selectivity. With few clearly defined
`interactions, preclinical and clinical data
`are more easily interpreted, and the
`possibility of observing bimodal responses
`that are amenable to genetic analysis
`is higher. If single-agent activity is not
`observed in experimental models, a rapid
`search for combination partners should be
`undertaken. When synergies are identified,
`immediate assessment of potential toxicities
`is compulsory.
`
`Percentage of patients surviving
`
`100
`
`80
`
`60
`
`40
`
`20
`
`CP
`
`AP
`
`MyBC
`
`0
`
`0
`
`6
`
`36
`42
`48 54 60
`30
`24
`18
`12
`Months after start of imatinib
`Figure 3 | The value of early detection, the
`right drug and the right patient population.
`Overall survival after the start of imatinib for
`patients in chronic phase (CP), accelerated phase
`(AP) and myeloid blast crisis (MyBC)7. In CP, the
`estimated overall survival rate was 88% at 30
`months. Median overall survival for patients in AP
`was 44 months (range 0.4–50 months), with a
`survival rate of 62% at 30 months. Median overall
`survival for patients in MyBC was 6 months (range
`0.1–52 months), with survival rates of 27% at 12
`months and 17% at 24 months. Figure repro-
`duced, with permission, from REF. 7 © (2005)
`Wiley Interscience.
`
`Therefore, experience in the clinic and
`with mouse models indicates that metastatic
`cancers of epithelial origin — such