The focus of drug development has moved from cytotoxic compounds identified by screening to therapies that act at specific molecular targets. Although some of these agents may cause visible tumor reduction and may be adequately evaluated by standard oncology drug development methods, other novel therapies may not be amenable to our current practices. Consequently, our concepts of drug development must evolve and address various questions. How should we approach the preclinical development of these agents? Which laboratory experiments are critical prior to studying these new agents in the clinic? What phase I design strategies are appropriate? Agents that target specific molecules may not cause toxicity at effective doses and, therefore, phase I concepts such as maximum tolerated dose (MTD) may not be relevant. What phase II designs suit these agents? Drugs that have a cytostatic effect may be overlooked if tumor response is required as a selection criteria. Phase II studies may not be necessary if the relevant end point is survival, which cannot be fully addressed without a large randomized phase III trial. What can we learn from other areas of medicine? Many nononcology drugs target specific end points and are developed in a pattern quite different from the formula used in oncology. Should we be adopting their methods of drug development?

At the 10th National Cancer Institute-European Organization for Research and Treatment of Cancer Symposium on New Drugs in Cancer Therapy, a workshop discussed these issues. In this commentary, we summarize some of the presentations. Although many new agents are entering or are already in clinical trials, it is imperative that we continue this dialogue with scientists, clinical researchers, cooperative groups, and industry and regulatory authorities because there is not yet an international consensus on how to approach the challenges that lie ahead.

P reclinical A ssessment

The current era of anticancer drug discovery and development is based on the concept that, by identifying small molecules and genes targeting the precise molecular abnormalities that create and drive the malignant phenotype ( 1 ), therapies will emerge that are both more effective and better tolerated than the traditional cytotoxic agents ( 2 , 3 ). Examples of molecularly targeted signal transduction agents include inhibitors of ras farnesylation and receptor kinase inhibitors. Approaches that target the tumor cell environment include metalloprotease and integrin inhibitors as well as antiangiogenic strategies. Such agents present a major challenge to the process of preclinical assessment. Preclinical studies must demonstrate that a candidate agent selected for clinical development acts by the desired molecular mechanism (e.g., inhibition of a specific receptor tyrosine) and thereby produces the desired biologic effect (e.g., inhibition of proliferation or the induction of apoptosis). The agent must also exhibit in vivo antitumor activity and a suitable therapeutic index (i.e., ratio of the dose that brings about an anticancer effect to the dose that brings about a toxic effect).

Traditional end points, such as simple tumor cell killing or growth delay, are not by themselves sufficient for this contemporary need. The opportunity presented is to replace such blunt instruments with more specific end points that are tailor-made for any particular molecular target. Mechanism-based approaches require corresponding mechanism-based end points. Replacing “black box” approaches with tailor-made assessments means that the process of anticancer drug discovery and development becomes less routine and places greater demands on the biologists and pharmacologists involved. In addition, there is pressure both in terms of unmet medical need and commercial concerns to deliver candidates for clinical development faster than ever before ( 4 ). Companies that have already adopted aggressive time scales of 5-7 years from a potential new target to a development candidate are now seeking to shorten these still further. Faster development can be achieved only by exploiting the latest technologic advances, many of which are in their infancy and rapidly changing ( 5 ). These technologies involve 1) methods developed for modern molecular biology and genomics/proteomics to identify and validate new therapeutic targets, 2) robotic high-throughput screening against structurally diverse chemical libraries to discover lead drugs ( 6 , 7 ), 3) combinatorial chemistry for the efficient generation of structural diversity and for rapid production of a library of structural analogues based on an identified lead compound (known as lead explosion) and optimization ( 8 , 9 ), 4) enhanced structural biology and molecular modeling techniques for rational drug design ( 10 ), 5) high-throughput pharmacokinetics assays ( 11 ), and 6) the use of surrogate end points to monitor pharmacodynamics throughout the discovery and development process, particularly involving genomics (e.g., nucleic acid microassays) and proteomics ( 12 ).

Bottlenecks exist. In particular, bottlenecks occur with respect to understanding the biologic function of genes discovered through genomic approaches, the validation of potential drug targets, conversion of lead drugs active in vitro into agents with drug-like properties and pharmacologic activity in the intact animal, and the establishment of appropriate assays to measure specific molecular mechanisms and biologic effects. Therefore, these are the areas that require the greatest attention in terms of technology development at the present time.

Whether a drug discovery program is based on screening, a structure-based approach, or some combination of these approaches, a robust, efficient, and informative test cascade is essential ( 13 ). This should be based on a target profile that defines the essential and desirable properties of the molecule sought (e.g., in terms of potency, selectivity, pharmacokinetics, and therapeutic index). A typical contemporary test cascade consists of a series of hierarchical assays. These assays usually begin with an initial biochemical assay on the recombinant target protein and proceed to in vitro cell tests and in vivo evaluation of pharmacokinetic and pharmacodynamic behavior to a final cancer disease model such as a human tumor xenograft (Fig. 1) . When interesting agents are discovered because of a unique cellular fingerprint or because of an association with molecular target expression, as in the National Cancer Institute's tumor panel ( 14 ), specific assays must also be applied to confirm the mode of action and optimize the lead drug.

Examples of the mode-of-action assays include analysis of the changes in phosphorylation of a particular target protein, the disruption of a protein-protein interaction, or the disruption of an appropriately engineered reporter gene system. A frequent point of failure in drug discovery programs is poor pharmacokinetic behavior when the agent progresses from testing in cell culture to testing in the intact animal ( 15 ). Therefore, it is essential that these agents possess robust, “drug-like” properties (absorption, distribution, metabolism, and excretion) and the potential for sustained administration (e.g., for signal transduction inhibitors). This issue is being addressed by determining approximate pharmacokinetic properties of drug leads by the cocktail or cassette dosing approach that involves administering compounds at low doses in mixtures, followed by high-sensitivity high-performance liquid chromatography-mass to charge ratio-mass spectroscopy analysis ( 16 ). For initial pharmacodynamic end points, the hollow fiber tumor assay provides an “ in vitro-in vivo hybrid” to bridge the divide between cell culture and animal experiments ( 17 ). Surrogate markers of mode of action or biologic effect can be used alongside or instead of the hollow-fiber test. For example, an anti-inflammatory assay could be used to measure effective blockade of the phosphatidylinositol 3-kinase pathway.

The use of such pharmacokinetic and surrogate pharmacodynamic end points provides the advantages of rapid and potentially cost-effective turnaround of in vivo results and a less demanding hurdle during the in vivo lead optimization phase compared with a full-blown disease model. However, suitably optimized compounds will normally require testing in a more challenging biologic system. Human tumor xenografts are often used for this purpose ( 18 ), but these models can be criticized as poorly predictive of the human disease (although it is, in fact, too early to judge this for the newer molecular targets) ( 19 ). These concerns may be especially relevant where the drug effect sought inhibits angiogenesis or metastasis. A case can be made for the construction and use of more complex disease models, such as orthotopic systems ( 20 ) or transgenic animals ( 21 ), as was done for evaluating ras farnesylation inhibitors ( 22 ). In some cases, well-characterized syngeneic models may have advantages over human tumor xenografts. It should be remembered that regressions might not be seen with signal transduction or cell cycle inhibitors, unless of course the agent induces apoptosis.

In all situations, a decision to proceed to the clinic will require some evidence of activity in an in vivo disease model such as a human tumor xenograft. However, where there are real concerns about the clinical relevance of such models, a more controversial position is to look predominantly for evidence of “hitting the target” in preclinical in vivo models and to carry out the critical clinical hypothesis test in humans. Toxicologic evaluation is essential to ensure acceptable safety, to identify organs at risk, and to determine a safe phase I starting dose. However, because excessive preclinical evaluation can cause delay and is often poorly predictive, organizations such as the Cancer Research Organization and the European Organization for Research and Treatment of Cancer favor entering clinical trials after a relatively simple program of rodent-only toxicology testing ( 23 , 24 ).

There are major advantages to be gained if the mode of action and surrogate assays, ideally measuring biochemical effects as close as possible to the precise molecular locus, are capable of translation for use in the early clinical studies because they smooth the transition from preclinical assessment to human investigation. In this regard, noninvasive assays, such as magnetic resonance spectroscopy, magnetic resonance imaging (MRI), and positron emission tomography (PET), are especially attractive ( 25 ).

P hase I T rial D esign

Phase I trial design should incorporate the evaluation of safety and toxicity with a measure of the outcome at the molecular level. The ideal situation for proceeding from the laboratory to a clinical trial after a validated target effect is something that must be coordinated between the laboratory scientists and the clinical researchers before the initiation of the clinical program (Table 1) . As described above, a comprehensive preclinical package is necessary, providing target plasma concentrations that inhibit the response by 50% and methodology for validating target inhibition or modulation in tumors or normal tissue.

Accelerated phase I designs ( 26 ) are becoming widely used for drugs that have conventional cytotoxic activity, and there is a wealth of clinical experience to justify this approach. However, most novel agents (e.g., inhibitors of metalloprotease, protein kinase C, and epidermal growth factor receptor kinase) have a much wider therapeutic ratio, and an MTD may be difficult or impossible to reach. Drugs acting on highly specific targets that may be differentially expressed or activated in cancer cells may result in low normal tissue toxicity; therefore, increasing a dose to normal tissue tolerance may be an irrelevant end point. If preclinical results can provide an end point, such as the free plasma concentration of the drug corresponding to levels in tumors high enough to inhibit the target, then it may be reasonable to increase the dose rapidly, followed by increasing the dose in multiples of the plasma concentration that inhibits the response by 50%. It may also be reasonable to establish the concentration of drug required to inhibit the target and then to use that dose in target-directed trials (i.e., phase II) without attempting to produce toxicity. Phase I design may require fixed dosing or possibly the random assignment of patients to different dose levels with doses chosen from the preclinical studies and based on the minimum dose required to inhibit the target.

The key end point in phase I trials for targeted therapies should evolve from the current idea of a MTD in normal tissue to a more suitable end point of the dose required to maximally inhibit the relevant target. If this cannot be assessed, then all subsequent studies will be carried out blindly. With conventional drugs, clinical researchers assume that a correct dose is being used when responses and dose-limiting toxicity in normal tissues are seen. With molecularly targeted therapies, these end points may not be reached unless there are markers of target activity that are direct or indirect. Clearly, this is not necessary if responses occur in phase I, as with the bryostatin, the partial agonist of protein kinase C ( 27 ), and this recommendation is most relevant for drugs that do not produce a response (e.g., metalloprotease inhibitors). In some of these trials, the rate of change of the relevant markers has been used as the end point ( 1 ). The goal may not be target inhibition but rather subtle target modulation.

Several strategies have been developed to approach the problem of measuring target modulation. These begin to overlap with the current definition of a phase II trial. One approach would be the preoperative administration of drug and subsequent analysis of the drug concentration and target activity in the tumor. Many tumor types are managed with preoperative biopsies (control sample); for these cases, a comparison with non-drug-treated controls is possible. Similarly, tumors that produce effusions or metastases that allow easy retrieval of cells could be used. The prerequisite for a tissue biopsy and a demonstration of target activity would lengthen the time to complete trials and would increase the costs. Nevertheless, all subsequent trials and drug development would benefit from the knowledge that the recommended dose was correct and inhibited the target for which it was designed. Our standard histologic groupings may need to be replaced with more sophisticated stratification of tumors based on molecular genetics tests. Phase II trials of targeted therapy may not use histologic eligibility criteria but rather may enroll patients with tumors from different sites that express similar molecular markers (e.g., p53, c-myc, or Rb gene products)

There are many challenges to successfully incorporating tumor tissue analysis into the design of a clinical trial. Overcoming these challenges requires the commitment of oncologists, pathologists, human subject protection committees, and patients; substantial additional funding; and standardization of tissue sampling to ensure that the tumor is being measured.

Two other less invasive approaches can be used. The first approach uses peripheral leukocytes, which possess many receptors and signaling pathways relevant to novel agents. Demonstrating a “biologically effective dose” in leukocytes can guide the design of dose escalation protocols. This dose is not a “surrogate marker” because it is not replacing the need to assess the tumor target directly in future studies but is nevertheless an important pharmacodynamic measure of biologic activity. Results have been reported with a number of targets and include inhibition of tyrosine phosphorylation ( 29 ), inhibition of tumor necrosis factor-α cleavage enzyme ( 30 ), decreased expression of interleukin 6 ( 31 ), and release of tumor necrosis factor-α release by protein kinase C inhibitors ( 32 ). In some instances where no modulation of the target is observed in these “surrogate” tissues, decisions not to proceed with further testing of the agent may be accurately made without further tumor testing.

The second approach uses newer imaging techniques, including PET and MRI. PET scanning can measure blood flow, thymidine metabolism, and glucose uptake ( 33 ). PET labeling drugs can show tissue concentrations of the drug achieved at tumor sites. An MRI can also show vascular permeability ( 34 ) and blood flow. Doppler techniques, particularly applied to breast cancer tumors, provide another useful way to assess blood flow. Many newer agents directly or indirectly modulate tumor angiogenesis ( 35 ), therefore, inhibition of angiogenesis is a valid biologic end point. These techniques, however, need to be rigorously assessed for reproducibility and validated, which should occur in parallel with the clinical studies (e.g., reliability of repeat assessments). Use of these techniques toward the end of phase I should continue into phase II to allow assessment of long-term treatment.

Issues of scheduling, including long infusions, daily treatments, or long-term intermittent dosing, must be considered, and these issues may require novel pharmacokinetic modeling and sampling. Moreover, issues such as tissue accumulation of drug may be more relevant than plasma peaks and may not be easily measured.

Preclinical models suggest that many of the novel agents may stabilize tumor growth rather than produce tumor regression. This outcome may be valuable to patients and may be worthwhile when applied to adjuvant therapy aimed at dividing cells. Long-term dosing may be necessary to observe and maintain stabilization or response. Because many patients in phase I trials receive therapy for only a brief time, often only 1 or 2 months, the information required for evidence of biologic activity and phase II dosing may not be available from the initial dose studies. Overlap and continuous progression to phase II should be in the trial design so that fitter patients can be treated for a longer period.

It is not known whether molecularly targeted therapies will be active alone, combined with cytotoxic agents, or combined with other targeted therapies. Each of these permutations cannot be fully explored with the current preclinical and phase I design without delaying the answers well into the next century. Communication between laboratory researchers and clinical investigators is necessary to design batteries of tests that will allow early clinical trials of these therapies, based on preclinical data. Certainly, results of preclinical studies have predicted some of our current clinical combinations (such as 5-fluorouracil-folinic acid and gemcitabine-cisplatin) and may be a model for synergistic interactions that may not be predicted a priori. Novel formulations that contain two or more agents may be desirable and may allow the initiation of early trials without the difficulties inherent in coordinating the simultaneous use of a number of experimental therapies owned by different companies.

P hase II and B eyond : A ssessment of E fficacy

After phase I trials are completed, the main goal of subsequent clinical research is to define as accurately and efficiently as possible the efficacy of the new agent in a variety of malignancies. Efficacy in this context is defined as increased survival, cure rates, or quality of life. Traditionally, phase II trials have been used to screen new agents for evidence of biologic activity in multiple types of tumors before proceeding into large phase III trials in which efficacy may be convincingly demonstrated. Objective tumor regression, defined arbitrarily as a 50% decrease in the sum of the products (using the longest diameter and the corresponding perpendicular measurement) that is observed in a minimum number of patients, has been found to be a useful tool for selecting or rejecting agents for subsequent evaluation. Although not all of the agents that are selected on the basis of response results in phase II trials have been subsequently shown to improve survival, sufficient numbers have done so to validate response as a reliable phase II end point.

If stasis of tumor growth rather than tumor regression is the anticipated clinical outcome, screening agents for efficacy with standard phase II response end points may be unrewarding. Two general approaches are possible: 1) to identify surrogate end points for efficacy in phase II besides tumor response or 2) to move the new agent directly from phase I into randomized studies.

There has been much interest in the first of these suggestions: the identification of alternative end points to response that may be used to appropriately select or reject a new drug in a phase II design. However, it is important to note that any proposed alternative end point must be viewed as being experimental because, until now, only response has been used to identify agents that have an impact on survival. Possible candidate end points are as follows: time to progression, changes in tumor markers, measures of target inhibition, PET scanning, proportion of patients with early disease progression, and assessment of clinical benefit.

Although median time to disease progression has the advantage of being a well-described and standardized end point in randomized trials, it is difficult to interpret when there is no control group and when patient numbers are limited, as is the case in a phase II setting.

Tumor markers also share the advantage of being well described and easily measured. The problem with using markers as an end point to select active agents is that the use of markers alone to identify active agents has never been validated. Rustin et al. ( 36 ) suggested that, in ovarian cancer, the tumor marker CA 125 may substitute for the rate of response in phase II trials, but these data were generated in instances where the drugs also produced tumor regression. For cytostatic agents in general, it is not known whether a tumor marker will decrease with treatment and, if it does not, whether the agent can still cause tumor stasis. Another report ( 37 ) has suggested that a change in the rate of increase in a relevant marker may be a useful end point for identifying an active agent. In studies of a metalloproteinase inhibitor, patients were observed to show a decline in the pretreatment rate of increase of the serum tumor marker, an effect that also seemed to demonstrate a dose-response relationship. Randomized trials based on these data are proceeding for patients with ovarian and pancreatic cancers, and the results of these studies may contribute to validation of this end point. Furthermore, before a tumor marker is accepted as a surrogate for drug activity, it must be demonstrated that the drug under study cannot alter levels of the marker through changes unrelated to tumor burden. Such a change would be the secretion of the marker from the cell, which has been observed for some agents ( 38 ).

Measure of target inhibition is appealing from a scientific point of view. However, because many novel agents target molecules that are only hypothetically linked to tumor growth, the inhibition of these target molecules has not yet been shown to be clinically effective. Measures of target inhibition (preferably in tumor tissue) are a more relevant end point for phase I trials where optimal dosing is the goal.

PET scanning is a nuclear medicine technique that uses radiopharmaceuticals such as 2-[ 18 F]fluoro-2-deoxy- d -glucose to detect differences in tissue function and metabolism. PET scanning can predict responses early when there is corresponding evidence of clinical regression ( 39 ). Will it, however, assess changes in tumors that result from effective cytostatic therapy when a response is not documented? This is possible, but remains to be shown. If PET scanning can only anticipate tumor regression, then this costly test is not really an alternative to response.

Progression rate (i.e., the proportion of patients whose disease progressed in the first 8 weeks of therapy and thus have a best response rating of progression) is something that the National Cancer Institute of Canada Clinical Trials Groups is exploring as a means of identifying active agents in phase II trials. In a review of the group's phase II data, agents identified as “ interesting”—because they surpassed a minimum response rate (the usual phase II decision rule)—were the same agents that would have been selected because they produced less than a maximum rate of progressive disease. That is, ignoring the response rate completely would have yielded the same list of active agents. Although interesting, this idea needs to be further explored with larger datasets and with progressive disease rate criteria for accepting and/or rejecting a new drug. Retrospective and historic controls cannot be used for comparison because there is evidence that, with changes in supportive care management, there may be marked improvement in the no-treatment groups over time.

Clinical benefit response can be defined in several ways, depending on the type of cancer and the symptoms. Improvement of pain has been successfully used as an end point in the mitoxantrone-prednisone study of patients with prostate cancer ( 40 ) and in the gemcitabine trial of patients with pancreatic cancer ( 41 ). Quality-of-life tools may be used to quantify the response and may be of more value in randomized studies of new agents.

An alternative approach to developing agents unlikely to cause tumor regression would be to forego the traditional phase II screening process and proceed directly to randomized, phase III trials. Preclinical data supporting such a development plan must be very strong to merit the investment of time and resources that this approach entails. The selected tumor type must be appropriate; i.e., there must be some scientific basis for anticipating that the novel agent would be effective in a particular disease because of the frequency of the targeted abnormality. Rules should be in place to terminate trials early for ineffective therapies or if cumulative toxic effects are observed, which phase I trials seldom reveal. Finally, some suggestion from phase I data that the new drug is inhibiting the activity of the target or is causing biologic and/or toxic changes related to effective doses would be useful.

At this time, therefore, despite the interest in defining alternative phase II end points that could identify those cytostatic drugs most likely to improve survival, no end point, except response, can be considered validated. Identification of valid alternative end points through their use in the selection of drugs that are later shown to be effective in phase III trials will take time. This time will be well spent, however, if we can avoid the need to subject every novel cytostatic agent to evaluation in large randomized trials.

L essons F rom N ononcology D rugs

Traditionally, we have treated oncologic therapies as unique, with an uncompromising set of standards and guidelines. However, should these rigid standards be applied to agents that may not deserve the reputation of cytotoxic compounds and may require prolonged administration? Are there lessons that we can learn from fields outside oncology that can be applied to the evaluation of our novel agents?

The historic goal of systemic anticancer therapy has been cure. Obtaining a complete response has always been considered a necessary first step, and partial responses have been used as evidence, albeit less compelling, of activity in relatively resistant solid tumors. As new target-based therapies enter clinical trials, it is important to reconsider the goals of treatment. Cure is still a lofty and ideal goal, when feasible. However, for incurable tumors, prolonged stabilization of disease with minimal or no toxicity would be viewed as a major success. Furthermore, it could also be beneficial to slow progression of disease, if such an effect could be accomplished with modest toxicity.

This cytostatic paradigm, defined as slowing disease progression, has many parallels in therapeutics outside oncology. Three examples are tacrine for Alzheimer's disease, riluzole for amyotrophic lateral sclerosis, and prednisolone for rheumatoid arthritis.

Early clinical evaluation of tacrine in Alzheimer's disease suggested reversible dose-dependent improvement ( 42 ). Subsequently, Davis et al. ( 43 ) conducted a 632-patient multicenter trial by use of an enrichment design. The first stage (6 weeks) of the study included random assignment of patients to one of three sequences of treatments, with each treatment lasting 2 weeks. The tacrine dose for each sequence was as follows: 1) tacrine at 0 (placebo), 40, or 80 mg/day; or 2) tacrine at 40, 80, or 0 mg/day; or 3) tacrine at 40, 0, or 80 mg/day. Only those patients (n = 215) with evidence of benefit from tacrine were continued in the study. In the second stage, there was a placebo washout period, and then patients were randomly assigned to receive a placebo or to receive their best dose (40 or 80 mg/day) from the first stage. In the final stage, all patients received the dose of tacrine that gave their best response.

The concept of enrichment before randomization is important because it allows exclusion for patients who are noncompliant, who may have low tolerance for toxicity, or who are nonresponsive to the drug. The primary end point of this study was change in the Alzheimer's Disease Assessment Scale during the 6-week placebo-controlled randomization period. Tacrine was demonstrated to be statistically significantly better than placebo because there was a slower decline in the Alzheimer's Disease Assessment Scale for patients in the tacrine group (two-sided P <.001). Thus, tacrine was not demonstrated to consistently reverse the disease process, only to slow its progression.

Amyotrophic lateral sclerosis has historically been a progressively fatal neuromuscular disease without effective treatment. Riluzole was selected for evaluation in this disease because of its effects on glutamatergic transmission, hypothesized to be involved in the pathophysiology of amyotrophic lateral sclerosis ( 44 ). Bensimon et al. ( 45 ) conducted a double-blind, placebo-controlled trial in seven French centers that enrolled 155 patients. Enrollment was enhanced by the accrual of patients with limb and bulbar disease, with subsequent stratification. The primary end points were survival and change in functional status (limb and bulbar function, symptoms, and clinical examination). The major finding was that deterioration of muscle strength was statistically significantly slower in the riluzole-treated patients (two-sided P = .028). Like the tacrine study for Alzheimer's disease, riluzole did not cause disease regression but did inhibit progression.

The use of corticosteroids in rheumatoid arthritis has been controversial for years ( 46 ). The Arthritis and Rheumatism Council Low-Dose Glucocorticoid Study Group ( 47 ) evaluated prednisolone in rheumatoid arthritis in a 13-center (all in the U.K.) double-blind, placebo-controlled study of 128 patients. Stratification was by center. More important, physicians were free to prescribe any other treatment for the patient except systemic corticosteroids. The primary end points used radiographic criteria. The study duration was 2 years. The study demonstrated decreased radiographic progression in the prednisolone group, based on either the Larsen index (two-sided P = .004) or the percentage of patients with new erosions (two-sided P = .007). Again, there was no remission of the disease, only a reduction in radiologic progression.

These three positive trials of drugs for neurologic and rheumatologic indications illustrate important principles that can potentially be applied to the development of cytostatic antineoplastic agents. The most important point is that phase II trials can be well-controlled randomized trials, ideally using an oral agent and a placebo (double-blind) control. A potential end point of such studies is delay of disease progression, which may require a variety of statistical tests. Survival analysis (time to disease progression) is often used in adjuvant trials but has not hitherto been used in phase II testing. Alternatively, trial designs incorporating two or more active treatment cohorts, as well as a placebo, and analyzing change in measurable disease over a fixed period may be more efficient in phase II testing.

The tacrine study illustrates the potential value of enrichment designs, which can exclude poor-risk patients before patients are randomly assigned to treatment. A particularly useful design for oncology trials may be the randomized discontinuation trial, where all of the patients can receive active treatment before being randomly assigned to treatment ( 48 ). In this design, all patients initially receive the study drug. At the end of an initial evaluation period (2-4 months), patients with substantial disease progression or toxicity are removed from the trial, and the remaining patients are randomly assigned to continue treatment or to discontinue treatment, ideally with a double-blind placebo control.

Clearly, novel phase II trial designs will be required, which may greatly hasten the critical pathways for development of these exciting new agents. This will require a major change in the thinking of clinicians, sponsors, and regulators.

C onclusion

Only a few years ago, it seemed that oncology had come to a relative standstill with a paucity of new agents. We now have a large number of exciting molecular-based therapies in development and entering the clinical arena. Many of these agents target signal transduction molecules and the cancer cell environment. An international consensus is needed to define acceptable end points for targeted agents, to avoid missing activity or novel outcomes that may not fit our current paradigms. Targeted therapies may surprise us and may result in outcomes not predicted by laboratory experiments or covered by new guidelines. This will require ongoing, open discussion, creative approaches, and timely reviews of our drug development programs so that novel therapies are not overlooked.

Table 1.

Ideal requirements for human trials of targeted agents

Trial
 
Requirements
 
End points
 
Preclinical Scientific rationale Good manufacturing practice formulation 
 Initial formulation 
 Evidence of antitumor activity Evidence of target effect 
  Xenograft/other response 
 Application to human tumors Direct measure of target 
  Indirect measure of target 
  Pharmacokinetic assay 
  Preclinical toxicology 
  Safe starting dose 
Phase I Good manufacturing practice formulation Pharmacokinetics  
  Direct suppression of target 
 Pharmacokinetic assay Indirect suppression of target 
 Measures of target effect Toxicity 
  Maximum tolerated dose classical toxicity 
 Safe starting dose  Minimal effective dose 
Phase II Recommended dose level Evidence of tumor response 
  Evidence of molecular effect 
 Test for target effect Direct/indirect target measures 
 Toxicity profile Other measures of effectiveness 
 Evidence of target effect  Change in tumor markers 
   Time to disease progression 
Randomized phase II/IIII Evidence of target effect Improved time to disease progression 
 Confirmed safety Decrease in metastases  
 Effective dose Tumor response 
 Expanded toxicity profile Survival benefit 
  Improved quality of life 
  Economic feasibility 
Trial
 
Requirements
 
End points
 
Preclinical Scientific rationale Good manufacturing practice formulation 
 Initial formulation 
 Evidence of antitumor activity Evidence of target effect 
  Xenograft/other response 
 Application to human tumors Direct measure of target 
  Indirect measure of target 
  Pharmacokinetic assay 
  Preclinical toxicology 
  Safe starting dose 
Phase I Good manufacturing practice formulation Pharmacokinetics  
  Direct suppression of target 
 Pharmacokinetic assay Indirect suppression of target 
 Measures of target effect Toxicity 
  Maximum tolerated dose classical toxicity 
 Safe starting dose  Minimal effective dose 
Phase II Recommended dose level Evidence of tumor response 
  Evidence of molecular effect 
 Test for target effect Direct/indirect target measures 
 Toxicity profile Other measures of effectiveness 
 Evidence of target effect  Change in tumor markers 
   Time to disease progression 
Randomized phase II/IIII Evidence of target effect Improved time to disease progression 
 Confirmed safety Decrease in metastases  
 Effective dose Tumor response 
 Expanded toxicity profile Survival benefit 
  Improved quality of life 
  Economic feasibility 

Fig. 1.

Flow chart schema for the development of targeted therapies. Cascade starts at testing and continues to clinical development.

Fig. 1.

Flow chart schema for the development of targeted therapies. Cascade starts at testing and continues to clinical development.

We thank Dr. M. Krul, Dr. H. Pinedo, and the other organizers of the 10th NCI/EORTC Symposium on New Drugs.

References

(1)
Hunter T. Oncoprotein networds.
Cell
 
1997
;
88
:
333
-46.
(2)
Gibbs JB, Oliff A. Pharmaceutical research in molecular oncology.
Cell
 
1994
;
79
:
193
-8.
(3)
Kerr DJ, Walkman P, editors. New molecular targets for cancer chemotherapy. Boca Raton (FL): CRC Press; 1994.
(4)
Drews J. Genomic sciences and the medicine of tomorrow.
Nat Biotechnol
 
1996
;
14
:
1516
-8.
(5)
Workman P. Cell proliferation, cell cycle and apoptosis targets for cancer drug discovery: strategies, strengths, and pitfalls. In: Thomas NS, editor. Apoptosis and cell cycle control: basic mechanisms and implications for treating malignant disease. Oxford (U.K.): Bios; 1996. p. 205-32.
(6)
Bevan P, Ryder H, Shaw I. Identifying small-molecule lead compounds: the screening approach to drug discovery.
Trends Biotechnol
 
1995
;
13
:
115
-21.
(7)
Workman P. Towards intelligent drug screening in the post-genome era?
Anticancer Drug Des
 
1997
;
12
:
525
-31.
(8)
Terrett NK, Gardener M, Gordon DW, Koblecki RJ, Steele J. Combinatorial synthesis—the design of compound libraries and their application to drug discovery.
Tetrahedron
 
1995
;
51
;
8135
-73.
(9)
Hogan JC Jr. Combinatorial chemistry in drug discovery.
Nat Biotechnol
 
1997
;
15
:
328
-30.
(10)
Blundell TL. Structure-based drug design.
Nature
 
1996
;
384(6604 Suppl)
:
23
-6.
(11)
Rodrigues AD. Preclinical drug metabolism in the age of high-throughput screening: an industrial perspective.
Pharm Res
 
1997
;
14
:
1504
-10.
(12)
Kleyn PW, Vesell ES. Genetic variation as a guide to drug development.
Science
 
1998
;
281
:
1820
-1.
(13)
Workman P. New drug discovery and development. In: Cassidy J, editor. Oxford handbook of oncology. Oxford (U.K.): Oxford University Press. In press 1999.
(14)
Monks A, Scudiero DA, Johnson GS, Paull KD, Sausville E. The NCI anti-cancer drug screen: a smart screen to identify effectors of novel targets.
Anticancer Drug Des
 
1997
;
12
:
533
-41.
(15)
Workman P. Pharmacokinetics and cancer: successes, failures and future prospects.
Cancer Surv
 
1993
;
17
;
1
-26.
(16)
Olah TV, McLoughlin DA, Gilbert JD. The simultaneous determination of mixtures of drug candidates by liquid chromatography/atmospheric pressure chemical ionization mass spectrometry as an in vivo drug screening procedure.
Rapid Commun Mass Spectrom
 
1997
;
11
:
17
-23.
(17)
Hollingshead MG, Alley MC, Camalier RF, Abbott BJ, Mayo JG, Malspeis L, et al. In vivo cultivation of tumor cells in hollow fibers.
Life Sci
 
1995
;
57
:
131
-41.
(18)
Boven E, Winograd B, editors. The nude mouse in oncology research. Boca Raton (FL): CRC Press; 1991.
(19)
Gura T. Systems for identifying new drugs are often faulty [ news].
Science
 
1997
;
278
:
1041
-2.
(20)
Georges RN, Mukhopadhyay T, Zhang Y, Yen N, Roth JA. Prevention of orthotopic human lung cancer growth by intratracheal installation of a retroviral antisense K-ras construct.
Cancer Res
 
1993
;
53
:
1743
-6.
(21)
Adams JM, Cory S. Transgenic models of tumor development.
Science
 
1991
;
254
:
1161
-7.
(22)
Kohl NE, Omer CA, Conner MW, Anthony NJ, Davide JP, deSolms SJ, et al. Inhibition of farnesyltransferase induces regression of mammary and salivary carcinomas in ras transgenic mice.
Nat Med
 
1995
;
1
:
792
-7.
(23)
Burtles SS, Newell DR, Henrar RE, Connors TA. Revisions of general guidelines for the preclinical toxicology of new cytostatic anticancer agents in Europe. The Cancer Research Campaign (CRC) Phase I/II Clinical Trials Committee and the European Organization for Research and Treatment of Cancer (EORTC) New Drug Development Office.
Eur J Cancer
 
1995
;
31A
:
408
-10.
(24)
Burtles SS, Jodrel DI, Newell DR. Evaluation of ‘ rodent-only' preclinical toxicology for phase I trial of new cancer treatment. The Cancer Research Campaign (CRC) experience [abstract].
Proc Am Assoc Cancer Res
 
1998
;
39
:abstract
2476
.
(25)
Workman P. Bottlenecks in anticancer drug discovery and development: in vivo pharmacokinetic and pharmacodynamic issues and the potential role of PET. In: Komar D, editor. PET for drug development and evaluation. Dordrecht (The Netherlands): Kluwer; 1995. p. 277-85.
(26)
Simon R, Freidlin B, Rubinstein L, Arbuck SG, Collins J, Christian MC. Accelerated titration designs for phase I clinical trials in oncology.
J Natl Cancer Inst
 
1997
;
89
:
1138
-47.
(27)
Philip PA, Rea D, Thavasu P, Carmichael J, Stuart NS, Rockett H, et al. Phase I study of bryostain 1: assessment of interleukin 6 and tumor necrosis factor alpha induction in vivo. The Cancer Research Campaign Phase I Committee.
J Natl Cancer Inst
 
1993
;
85
:
1812
-8.
(28)
Nemunaitis J, Poole C, Primrose J, Rosemurgy A, Malfetano J, Brown P, et al. Combined analysis of studies of the effects of the matrix metalloproteinase inhibitor marimastat on serum tumor markers in advanced cancer: selection of a biologically active and tolerable dose for longer-term studies.
Clin Cancer Res
 
1998
;
4
:
1101
-9.
(29)
Ferry DR, Smith A, Malkhandi J, Fyfe DW, deTakats PG, Anderson D, et al. Phase I clinical trial of the flavonoid quercetin: pharmacokinetics and evidence for an in vivo tyrosine kinase inhibition.
Clin Cancer Res
 
1996
;
2
:
659
-68.
(30)
Solomon KA, Covington MB, DeCicco CP, Newton RC. The rate of pro-TNF-alpha following inhibition of metalloprotease-dependent processing to soluble TNF-alpha in human monocytes.
J Immunol
 
1997
;
159
:
4524
-31.
(31)
Mouawao R, Merle R, Antoine S, Gozy EC, Weil M, Khayat D, et al. Relationship between functional IL6 receptor, intracytoplasmic IL6 expression and serum IL6 level in metastatic malignant melanoma.
Proc Am Assoc Cancer Res
 
1998
;
39
(abstract
1040
).
(32)
McDonald AC, Propper D, King D, Champian K, Graf P, Man A, et al. Phase I and pharmacokinetic study of CGP 41251, an inhibitor of protein kinase C [ abstract].
Proc ASCO
 
1997
;
16
:abstract
742
.
(33)
Price P. Is there a future for PET scanning in oncology? [ editorial].
Eur J Nucl Med
 
1997
;
24
:
587
-9.
(34)
Hayton P, Brady M, Tarassenko L, Moore N. Analysis of dynamic MR breast images using a model of contrast enhancement.
Med Image Nal
 
1996
;
1
:
207
-24.
(35)
Kerbel RS, Viloria-Petit A, Okada F, Rak J. Establishing a link between oncogenes and tumor angiogenesis.
Mol Med
 
1998
;
4
:
286
-95.
(36)
Rustin GS, Nelstrop AE, Bolis G, Gore M, ten Bokkel Huinuik W, Spaczynski M. Use of CA 125 to assess activity of topotecan versus paclitaxel in relapsed ovarian carcinoma measure by response and early progression [abstract].
Proc ASCO
 
1997
;
16
:abstract
1256
.
(37)
Rasmussen H, Rugg T, Brown P, Baullet M, Millar A. A 371 patient meta-analysis of studies of marimasat in patients with advanced cancer [ abstract].
Proc ASCO
 
1997
;
16
:abstract
1538
.
(38)
Thalmann GN, Sikes RA, Chang SM, Johnston DA, von Eschehbach AC, Chung LW. Suramin-induced decrease in prostate-specific antigen expression with no effect on tumor growth in the LNCaP model of human prostate cancer.
J Natl Cancer Inst
 
1996
;
88
:
794
-801.
(39)
Timothy AR, Cook GJ. PET scanning in clinical oncology [ editorial].
Ann Oncol
 
1998
;
9
:
353
-5.
(40)
Moore MJ, Osoba D, Murphy K, Tannock IF, Armitage A, Findlay B, et al. Use of palliative end points to evaluate the effects of mitoxantrone and low-dose prednisone in patients with hormonally resistant prostate cancer.
J Clin Oncol
 
1994
;
12
:
689
-94.
(41)
Burris H, Moore MJ, Cripps C, Andersen J, Portenoy RK, Storniolo AM, et al. Improvement in survival and clinical benefit with gemcitabine as first-line therapy for patients with advanced pancreas cancer: a randomized trial.
J Clin Oncol
 
1997
;
15
:
2403
-13.
(42)
Summers WK, Majovski LV, Marsh GM, Tachiki K, Kling A. Oral tetrahydroaminoacridine in long-term treatment of senile dementia, Alzheimer type.
N Engl J Med
 
1986
;
315
:
1241
-5.
(43)
Davis KL, Thal LJ, Gamzu ER, Davis CS, Woolson RF, Gracon SI, et al. A double-blind, placebo-controlled multicenter study of tacrine for Alzheimer's disease. The Tacrine Collaborative Study Group.
N Engl J Med
 
1992
;
327
:
1253
-9.
(44)
Rowland LP. Ten central themes in a decade of ALS research.
Adv Neurol
 
1991
;
56
:
3
-23.
(45)
Bensimon G, Lacomblez L, Meininger V. A controlled trial of riluzole in amyotrophic lateral sclerosis. ALS/Riluzole Study Group.
N Engl J Med
 
1994
;
330
:
585
-91.
(46)
Byron MA, Mowat AG. Corticosteroid prescribing in rheumatoid arthritis—the fiction and the fact.
Br J Rheumatol
 
1985
;
1024
:
164
-6.
(47)
Kirwan JR. The effect of glucocorticoids on joint destruction in rheumatoid arthritis. The Arthritis and Rheumatism Council Low-Dose Glucocorticoid Study Group.
N Engl J Med
 
1995
;
333
:
142
-6.
(48)
Kopec JA, Abrahamowicz M, Esdaile JM. Randomized discontinuation trials: utility and efficiency.
J Clin Epidemiol
 
1993
;
46
:
959
-71.