Biomarkers and Oncology Drug Development

Elizabeth Mansfield, PhD, US Food and Drug Administration (FDA), Silver Spring, Maryland, USA, discussed the use of biomarkers and biomarker tests for prediction, prognosis, and patient selection in targeted drug development trials. Once identified a test must be developed that measures the biomarker in a useful way. Some biomarkers are both prognostic and predictive, but trials must be designed to demonstrate the marker's value for one or the other. Selective biomarkers are neither prognostic nor predictive but are used to select a treatment population.

In 2011, the FDA published the Companion Diagnostic Draft Guidance, announcing that if a companion diagnostic test is necessary to find a drug to be safe and effective, the FDA must approve at least one instance of that test. Table 1 shows examples of pairs of diagnostic tests and biomarkers. Test selection and development must take into consideration appropriate measurement characteristics, be analytically validated, be uniformly used in registration trials, and be available for approval together with the drug.

Preselection is difficult, because all or most included patients are marker-positive and different tests with different performances are likely to have been used. External or non-Clinical Trial Assay tests may bias an enrolled population. In addition, no negative markers are available to validate the test or drug performance.

To gain approval, patients with acceptable benefit-to-risk profiles should be selected and any serious safety signals should be removed or avoided, the population should be narrowed to those who are most likely to benefit, and the biomarker test must work and be available at drug approval.

Challenges in Targeted Agent Development

Patient selection is key in developing targeted agents that offer real patient benefits. Selecting the right patients for a drug involves understanding the tumor biology and using the appropriate diagnostics, endpoints, and trial design. Gwendolyn A. Fyfe, MD, San Francisco, California, USA, used human epidermal growth factor receptor (HER) 2, epidermal growth factor receptor (EGFR [HER1]), and VEGF as examples to illustrate the challenges in developing targeted cancer therapies.

Although development of trastuzumab seemed difficult at the time, a great deal was known about HER2, including the fact that breast cancer patients with HER2 overexpression had shortened median survival. To target HER2, a well-characterized therapeutic and a diagnostic for selecting patients were needed. HER2 was unusual, in that there was almost a dichotomous response to trastuzumab, with dramatic growth inhibition in overexpressing cell lines and no effect in normal cell lines. When only marker-positive patients were treated, longer survival resulted.

EGFR was more difficult. It was assumed that overexpression should be targeted, but the value of EGFR overexpression as a prognostic marker remains unclear. Survival curves from erlotinib-versus-placebo trials are not as straightforward as those observed in trastuzumab studies. Simulations show that clinically significant separation of Kaplan-Meier curves in a randomized trial requires an effect in at least 30% to 50% of patients. The current poor understanding of EGFR biology makes it difficult to select patients who might benefit from erlotinib.

Choosing the right endpoint is important, as illustrated with anti-VEGF therapy development. Early bevacizumab single-agent trials reported varying response and survival rates. With a drug that targets tumor infrastructure but not the tumor itself, the response rate can be dangerously misleading. When bevacizumab was combined with chemotherapy for colorectal cancer, both responders and nonresponders had a survival benefit.

Selecting the right patients for the right drug is key in the development of drugs that offer real benefits. Targeted therapy generally takes longer and is more expensive than standard drug development, but may identify important new therapeutics.

Hits and Misses in Targeted Therapy

Karen A. Gelmon, MD, University of British Columbia, Vancouver, British Columbia, Canada, discussed the current confusion about targeted therapy among clinicians. The hope has been that agents that target the fundamental molecular changes of malignancy would yield treatments that reduce normal tissue toxicity and increase survival and cure rates.

Successful targeted agents include imatinib for chronic myelogenous leukemia and gastrointestinal stromal tumors and trastuzumab for HER2-positive breast cancer. Other agents looked promising but did not deliver anticipated results, including sunitinib for renal cell carcinoma, bevacizumab for breast cancer, and iniparib for triple-negative breast cancer.

Targeted therapies have been successful when the abnormal target is a critical driver of the malignancy, is associated with poorer outcomes and can be successfully targeted without significant toxicity, and when the mechanism is known. So far, every effective targeted agent, with the exception of VEGF, has a response-prediction biomarker, but finding it or proving it has been challenging.

In the past, anticancer drug development focused on patient response. Current and future development depends on determining the molecular profile of the tumor and finding the most appropriate drug to target that profile. Researchers are looking at predictive, intermediate endpoints and other molecular biomarkers to find the right biomarker for the right drug for the right patient. This must be done early, or there will be increasing numbers of negative Phase 3 trials.

Molecular profiling shows that each tumor may be associated with many mutations. Defining driver versus passenger mutations is a challenge that requires functional studies, which require a good link between preclinical and clinical work. The intratumor heterogeneity model [Yap TA et al. Sci Transl Med 2012] proposes that ubiquitous driver events may provide more tractable biomarkers and targets than heterogeneous events that may lead to drug resistance and treatment failure.

Personalizing anticancer therapy requires a process that includes patient referral, surrogate tissue or archived tumor analysis, patient allocation to trials that is based on molecular characteristics, patient monitoring, and reanalysis of the tumor and other tissues for resistance mechanisms upon disease progression [Yap TA et al. Nat Rev Cancer 2009].

Dr. Gelmon concluded that the oncology community needs to understand both the promise and the limitations of the targets. They also need to be alert to new toxicities of novel targeted agents, and not rush to assume benefit or lack of benefit prior to Phase 3 testing and mechanistic studies to define the target.