Pharmacogenetics- and pharmacogenomics-based ... - Future Medicine

2 downloads 51 Views 405KB Size Report
www.futuremedicine.com. 861 future science group. Pharmacogenetics- & pharmacogenomics-based rational clinical trial designs in oncology REVIEW ...
Review For reprint orders, please contact: [email protected]

Pharmacogenetics- and pharmacogenomics-based rational clinical trial designs in oncology The rapid evolution of molecular technologies that can identify genetic markers and lead to dissecting the inherent variance of individual cancer biology has had a tangible impact on trial designs in oncology. Rational trial designs based on molecular marker expression coupled with drug–marker interactions have started to be adopted, challenging the previous paradigms of morphology-based, single-arm efficacy studies. This review summarizes novel trials being developed based on molecular predictive factor therapeutics and the potential impact these novel trial designs will have on the practice of oncology in future. A variety of clinical trial designs based on tumor and drug–host genetic interactions are discussed and the example of advanced prostate cancer is used to illustrate the changing landscape of clinical trial designs in cancer. KEYWORDS: adaptive clinical trial n genomic signature n molecular profiling n pharmacogenetic n predictive biomarker

Rui Qin1 & Manish Kohli*2 Department of Health Sciences Research, Mayo Clinic, 200 First Street South West, Rochester, MN 55905, USA 2 Department of Oncology, Mayo Clinic, 200 First Street South West, Rochester, MN 55905, USA *Author for correspondence: Tel.: +1 507 284 3903 Fax: +1 507 284 1806 [email protected] 1

Drug development and selection in cancer thera­peutics has traditionally targeted patient and tumor histology characteristics for potential drug testing for efficacy. Over the last decade, a paradigm shift in this practice has taken place because of the tremendous progress achieved in biotech instrumentation and molecular techniques, allowing the relatively easy and rapid identification of target aberrant genes and/or pathways that drive tumor biology and response to treatments. This has led to incorporating molecular pathology-driven tumor profiling and diagnostics into clinical practice. Subtyping the genetic heterogeneity of tumors that is undecipherable with histology alone has provided knowledge of novel molecular targets that are currently being targeted for therapeutic gain, prognosis and prediction of therapeutic responses to specific treatments, the latter encompassing the fields of pharmacogenomics and pharmacogenetics. Pharmacogenomics is defined as the study of the role of inherited and acquired genetic variation in drug response [1] based on multiple if not genome-wide associations; while pharmaco­genetics has traditionally involved mainly variation in drug metabolism, with a focus on an individual candidate gene [2]. Critically molecular prognostic biomarkers provide information about clinical outcome regardless of the therapy, based on expression patterns of a particular gene/pathway, while molecular predictive biomarkers provide information about the effectiveness of the treatment agent used [3]. Since prognostic and predictive biomarkers are not mutually exclusive, the genomic signatures

can indeed be both prog­nostic and/or predictive biomarkers, but their use can impact the designing of clinical trials during the development process, as is discussed in the following text. Several examples of using molecular predictive biomarker-based therapy already exist, such as the treatment of gastrointestinal stromal tumors (GISTs) with imatinib, wherein the presence of mutations in exon 11 in the C‑Kit protooncogene in the tumor increases the response to imatinib treatment (83.5% response rates) compared with mutations in exon 9 (response rate 48%) [4]. Another example is the treatment of non-small-cell lung cancers with first-line tyrosine kinase inhibitors in the presence of gainof-function mutations in EGF genes (EGFR). Therapeutic molecular predictive biomarker targeting has also begun to be used for identifying chemosensitivity and resistance to therapy; that is, as positive or negative predictors of drug efficacy, as in the case of 1p/19q deletion in anaplastic oligodendroglioma, which indicates chemosensitivity to existing chemotherapy cocktails [5]. Meanwhile, the presence of wild-type K‑Ras in metastatic colorectal cancer is associated with response to cetuximab and panitumumab when these agents are added to existing chemotherapy regimens to enhance therapeutic benefits [6,7]. Molecular subtyping of tumors has also provided information on survival beyond traditional clinical characteristics and histo­ pathology, as in the case of the MammaPrint® signature (Agendia, CA, USA) [8] and Oncotype DX (Genomic Health, CA, USA) gene panels [9]. Selected examples of molecular biomarker-based

10.2217/PME.13.78 © 2013 Future Medicine Ltd

Personalized Medicine (2013) 10(8), 859–869

part of

ISSN 1741-0541

859

Review

Qin & Kohli

therapy presently considered as standard of care are highlighted in Table 1. Incorporation of these advances in cancer therapeutics underlies the change in the practice landscape in oncology from empiric and clinico–pathological-based treatments to a more personalized tumor genome-based approach, which also attempts to recognize and reconcile the effects of genome variation and their impact on drug efficacy. The overall goal of this shift is to individualize cancer therapy in a way that enhances the efficacy of the side effect and toxicity profile balance for each patient, guided by the patient’s unique tumor profile. This contrasts with the previous model in cancer therapeutics wherein ‘all comers’ with a distinctive tumor morphology and stage receive the same treatment to test for efficacy, typically in a nonrandomized fashion. This increasing recognition and incorporation of molecular factors with the advent of targeted cancer therapeutics, however, is in its early days of development and brings with it unique challenges and opportunities for evidence-based integration of molecular pathology, cancer pharmacogenomics and pharmacogenetics with novel biomarker-based clinical

trial designs, which will result in furthering the overall goal of personalized cancer care. This article focuses on the types of novel clinical trial designs in cancer medicine that incorporate molecular predictive biomarkers in pharmaco­ genomic and pharmacogenetic principles. Since practice-changing advances in oncology have begun occurring in several tumor types, including advanced colorectal, breast, head and neck, and GIST cancers, we will highlight a practical example of novel clinical trial design used in a single tumor type, by focusing on the recent advances in knowledge about the tumor biology and treatments in advanced castrate-resistant prostate cancer (CRPC), to illustrate specific aspects of the novel trial design approach. CRPC is an ideal tumor type and stage suited to developing and applying novel biomarker-based clinical trial designs in oncology. Multiple treatment options are currently available to treat this stage and, at the same time, this stage harbors tremendous somatic heterogeneity. Genotyping the tumor heterogeneity in individual tumors in order to identify driver mutations that are actually responsible for tumor progression and then devising matching therapeutic strategies with

Table 1. Selected clinically used novel predictive and prognostic molecular targets in oncology and their inhibitors. Tumor type

Target

Therapeutic/predictive or Predictive marker of prognostic marker sensitivity/resistance

Target inhibitor example

Disease stage

Prostate cancer CYP 17

Therapeutic



Abiraterone acetate; TAK‑700 Advanced

Prostate cancer AR

Therapeutic



Enzalutamide

Advanced

Breast cancer

Her‑2/neu

Therapeutic Predictive

Her‑2/neu gene amplification

Trastuzumab, lapatinib, T-DM1; pertuzumab

Adjuvant and advanced setting

Gastric

Her‑2/neu

Therapeutic Predictive

Her‑2/neu gene amplification

Trastuzumab

Advanced

Hepatocellular cancer

VEGFR

Therapeutic



Sorafenib

Advanced

Breast/ovary

PARP

Therapeutic

BRCA mutations

Olparib

Advanced

NSCLC

EML4–ALK Predictive

EML4–ALK translocation

Crizotinib

Advanced

NSCLC

EGFR

Predictive

EGFR mutations

Gefitinib, erlotinib, cetuximab Advanced

Melanoma

BRAF

Predictive



Vemurafenib

Advanced

CNS tumors

1p/19q

Prognostic Predictive

1p/19q deletions

Standard-of-care chemotherapy

Local

Head and neck cancer

EGFR

Predictive



Cetuximab

Advanced

Colorectal

EGFR

Predictive

K‑Ras mutations

Cetuximab, panitumumab

Advanced

GIST

C‑Kit

Predictive

C‑Kit mutations

Imatinib

Adjuvant and advanced setting

AR: Androgen receptor; GIST: Gastrointestinal stromal tumor; NSCLC: Non-small-cell lung cancer; R: Receptor.

860

Personalized Medicine (2013) 10(8)

future science group

Pharmacogenetics- & pharmacogenomics-based rational clinical trial designs in oncology

the multitude of therapeutic options requires the application of adaptive biomarker-based clinical trial designs, which are currently evolving [10].

Novel clinical trial design evolution in oncology Several features distinguish emerging clinical trial designs based on pharmacogenetics or pharmacogenomics from traditional clinical trial designs. First, a prospective clinical trial incorporating pharmacogenetics and pharmacogenomics has the stated purpose to develop, validate or implement a predictive genomic signature (biomarker) associated with a novel therapy or a prognostic biomarker of a tumor type at a specific stage. It is critical to recognize biomarkers that are prognostic from those that are predictive prior to any attempt at incorporating biomarkers into novel therapeutic trial development or use in clinical practice [11]. At the most basic level, the difference is that prognostic biomarkers provide information about patients’ overall disease outcome, independent of any specific therapy, whereas predictive biomarkers provide information on potential therapeutic efficacy of a specific drug or the lack of such an effect (positive or negative predictive biomarker). While a biomarker can be either prognostic or predictive, or both, novel clinical trial designs that incorporate predictive biomarkers are of considerable scientific and clinical interest in personalized oncology medicine. Biomarker development requires several initial considerations before an attempt can be made at incorporation into clinical trial testing designs. Pharmacogenetic and pharmacogenomic biomarker expression must be performed in Clinical Laboratory Improvement Amendment approved laboratories as this affects downstream sensitivity, specificity, precision and reliability of the assay. Following this, biomarker development can be initiated by incorporation into clinical trial design testing for determination of patient eligibility, stratification/randomization of treatments and several other aspects of emerging novel clinical designs, after taking into account considerations that impact the discriminatory power of the biomarker-based approach for delivering therapeutic benefit. This consideration for biomarker development is based on the amount of existing scientific evidence that promotes the potential for further development. Differing strengths of existing evidence for the biomarker being developed and tested prospec­ tively in clinical trials can be broadly categorized into known valid biomarker, probably valid biomarker or exploratory biomarker. This aspect future science group

Review

impacts the type of clinical trial design needed for further testing of the biomarker of interest. An example of a known valid predictive biomarker is the presence of C‑Kit mutation for using imatinib in GIST tumors. This has been widely accepted in the medical and scientific community based on evidence from prospective, randomized, controlled Phase III clinical trials that were designed to test the biomarker; in this case, a predictive biomarker for imatinib efficacy. Alternatively, valid biomarkers can also be derived from meta-ana­lysis of prospective and/or retrospective studies not specifically designed to test the predictive or prognostic impact of the biomarker [12]. A probable valid predictive or prognostic biomarker means that the biomarker of interest has not been extensively validated; however, there is some scientific evidence, albeit less than there is for known and valid biomarkers, that appears to establish an association. An exploratory biomarker, on the other hand, is one that does not meet the criteria for either a known valid biomarker or a probable valid biomarker. The level of evidence is limited to small retrospective studies or pilot studies at most. In order to meet these challenges, clinical trial designs are undergoing a sort of gradual evolution so as to incorporate tumor expression profiles of individual tumors and integrate adaptive approaches for the rational combination of molecular-based therapy in the future, while at the same time, retaining relevant aspects of previous clinico–pathological-based efficacy study designs. For example, genetic variations being evaluated as predictive biomarkers associated with clinical outcomes in these novel clinical trials are usually reduced into an agenomic signature (classifier) through approaches in machine learning, such as regression classification or random forests [13], in order to evaluate the predictive power of the variable, which is then utilized for patient screening, stratification or randomization. The validation of a genomic biomarker as a predictive biomarker is conducted through testing for an interaction between the biomarker and clinical outcomes of treatment. The implementation of a predictive genomic biomarker is through the following treatment assignments: ƒƒ Direct assignment; ƒƒ Fixed randomization; ƒƒ Outcome-adaptive randomization. These considerations are departures from traditional nonbiomarker-based clinical trial designs, www.futuremedicine.com

861

Review

Qin & Kohli

where pharmacogenetic and pharmacogenomic information may be prospectively collected, but ana­lysis is typically of a post hoc nature. Therefore, the potential genomic signature, as a result of the post hoc ana­lysis of pharmacogenetics and pharmacogenomics, is generally not utilized for selecting and treating patients in clinical trials. In addition, in the past, traditional biomarker-based clinical trial designs have been mostly exploratory and focused heavily on prognostic aspects using nonpharmacogenetic clinical characteristics and thereby mitigating the importance of tumor–host–drug interactions. Novel clinical trial designs, on the other hand, which are currently evolving using new therapeutic agents, significantly focus on tumor–host–drug effects for predictive marker development. The differing types of novel designs used to incorporate pharmaco­ genetic and pharmaco­genomic biomarkers are discussed in the following sections. „„ Types of rational clinical trial designs incorporating genomic variations Novel clinical trial designs based on molecular and genomic characteristics include enrichment, biomarker validation designs, adaptive biomarker-based designs and Bayesian adaptive randomization designs. A general assumption for these biomarker-based designs is that the assay methods of pharmacogenetic and pharmacogenomic biomarkers have been well validated, which leads to satisfactory sensitivity, specificity and reliability. Sensitivity and specificity undoubtedly impact each of the biomarker-based clinical trial designs. Table 2 summarizes these with the purpose stated for each type of clinical trial design. The following section provides more detailed explanations of these novel clinical trial designs. Enrichment design

If the mechanism of action of a molecularly targeted therapy is well understood, the genetic variations associated with the oncogene or tumor suppressor gene may be claimed as genomic signature classifiers for selecting potential patients to receive this novel therapy. Enrichment design, also known as targeted design [14], is a perfect choice for such known valid predictive biomarkers for a novel therapy. Unlike traditional clinical trial designs, the genomic biomarker will be used as a screening tool for patient eligibility in the enrichment design. Only patients with positive genomic biomarkers, who are believed to benefit from a novel therapy, will be randomized to experimental therapy and standard control. The 862

Personalized Medicine (2013) 10(8)

other patients with negative genomic biomarker will be given standard control or removed from the study. The enrichment design, when appropriately used with compelling evidence, is more efficient than a conventional randomized design without explicit use of a genomic biomarker for patient selection. The relative efficiency is determined by the prevalence of positive genomic biomarker and the specificity of benefits of experimental therapy [15]. Biomarker strategy design & biomarker by treatment design

If the genomic signature classifier is a probably valid predictive biomarker, then unselected designs such as biomarker-based strategy designs and biomarker by treatment designs may be utilized to validate the genomic signature classifier as a predictive biomarker for a novel therapy [16]. In biomarker-based strategy design, patients are first randomized into marker-based and nonmarker-based strategy groups. Patients within the marker-based strategy group will receive direct assignment of experimental therapy and standard control based on the genomic biomarker; that is, patients with the positive genomic biomarker will receive experimental therapy and those with the negative genomic biomarker will receive standard control. By contrast, in the non-marker-based strategy group, patients will be randomized to receive either experimental therapy or standard control. In biomarker by treatment design, patients are first stratified by genomic biomarker status and then randomized to receive either experimental therapy or standard control within each stratum. The effect of the predictive biomarker may be derived from either the treatment difference between groups or the interaction term of the biomarker and treatment. By comparison, biomarker-based strategy design is not as efficient as the biomarker by treatment design because the former includes patients treated with the same regimen on both the marker-based and non-marker-based strategy groups; that is, some patients receive the same regimen, regardless of which group they are randomized to [17]. Such overlapping of patients receiving the same regimen dilutes the potential effect size, thus decreasing the power of the biomarker-based strategy design. Adaptive accrual design

Alternatively, the adaptive accrual design may be utilized to control patient accrual in the overall population or biomarker-defined subsets for the genomic signature as a probably valid predictive future science group

Pharmacogenetics- & pharmacogenomics-based rational clinical trial designs in oncology

Review

Table 2. Types of rational clinical trial designs in oncology incorporating pharmacogenetics and pharmacogenomics. Statistical design

Predictive Treatment biomarker assignment qualification

Purpose

Enrichment

Known valid

Eligibility screening and fixed randomization

Implementation II/III

Marker strategy

Probably valid

Direct assignment and fixed randomization

Validation, III implementation

Marker by treatment

Probably valid

Stratified fixed randomization

Validation

Adaptive accrual

Probably valid

Fixed randomization and Validation, III subset definition implementation

(Cross-validated) adaptive signature

Exploratory

Fixed randomization and Development, subset definition validation

III

Adaptive threshold

Exploratory

Fixed randomization and Development, subset definition validation

III

Bayesian adaptive randomization

Exploratory

Response-adaptive randomization

biomarker. Adaptive accrual designs refer to two-stage designs where the patient accrual is adaptive to the interim ana­lysis of the treatment effect after the first stage. The genomic signature biomarker is used for defining the target patient population in the adaptive accrual design. The threshold sample-enrichment approach enrolls and randomizes only patients with positive genomic biomarkers in the first stage; as in, the enrichment design, if the interim ana­lysis of treatment effect is statistically significant, this then expands to all patients, regardless of genomic biomarker status in the second stage, otherwise the clinical trial is stopped after the first stage [18]. Another approach seems to reverse the order of first and second stage [19]. In this approach all patients, regardless of genomic biomarker status, are enrolled in the first stage; if the interim ana­lysis of the treatment effect for all patients is not statistically significant, then patient accrual is restricted to those with positive genomic biomarkers only, otherwise the clinical trial is stopped after the first stage. Owing to the sequential testing of the treatment effect within an overall population or genomic biomarkerdefined subset, adjustments for multiplicity are required for strict control for family-wise type I error at a specified level. The straightforward but conservative Bonferroni’s approach, or more refined combination test approach, which make use of correlation between first stage and overall population, can be applied [20]. The adaptive accrual design requires only a moderate level of evidence of predictive biomarker, as interim ana­lysis serves for gathering further evidence. future science group

Phase

III

Development, II implementation

Exploratory biomarker-based clinical trial designs

In some clinical trials attempting to incorporate pharmacogenetics and pharmacogenomics, the genomic signature classifier or biomarker threshold has yet to be developed or must be explored simultaneously. In such situations, both adaptive signature [21] and threshold [22] designs unify the development and validation of a genomic predictive biomarker into a single clinical trial. In the adaptive signature design, a genomic signature is prospectively developed with patients accrued in the first stage and then prospectively used for stratifying patients into biomarker-positive and -negative status, before randomizing them into experimental therapy and standard control groups in the second stage. The therapeutic effect of experimental therapy is examined in the overall patients accrued and in the subgroup with the positive biomarker. The genomic biomarker is deemed to be a predictive biomarker if the treatment effect is observed to be discordant between the overall group and the subgroup of patients with a positive classifier. Cross-validated adaptive signature design is an extension of adaptive signature design that optimizes the efficiency of both biomarker development and validation [23]. In the adaptive threshold design, a biomarker threshold for the quantitative genomic signature must be simultaneously developed to identify a predictive biomarker-defined subgroup. In the first stage, the test for a treatment effect is conducted at a reduced significance level. In the second stage, the test for a treatment effect in the biomarker-defined subgroup is based on the www.futuremedicine.com

863

Review

Qin & Kohli

permutation distribution of maximum log likelihood ratio statistic, with a cut-off value between 0.5 and -1. The genomic predictive biomarkers in adaptive signature and threshold designs can be exploratory, as they are developed and validated in the same clinical trials. Bayesian adaptive randomization designs

These designs use genomic biomarker status to stratify patients and apply response-adaptive randomization to assign currently optimal treatment for each genomic biomarker-defined subgroup of patients. Bayesian logistic regression models with genomic biomarkers, treatment, biomarker and treatment interaction, and possibly other factors, are utilized to incorporate all covariates. Bayesian adaptive randomization designs [24,25] can accommodate multiple genomic biomarkers and also various experimental therapies. As the genomic predictive biomarker is simultaneously developed in the Bayesian adaptive randomization design, the predictive biomarker can also be exploratory, and the assignment of patients is adaptive to cumulative evidence of clinical response in the clinical trial. Although adaptive randomization seems to be attractive to both patients and clinicians, the perceived benefit of adaptive randomization versus fixed randomization became questionable recently after some simulation studies [26,27]. Two Phase II trials of the Bayesian adaptive design in non-small-cell lung cancer and breast cancer are highlighted. The BATTLE trial adopted a Bayesian adaptive design for targeted therapy development in advanced stage non-small-cell lung cancer [28]. Patients enrolled in this trial were undergoing biomarker profiling to determine the status of the following molecular biomarkers: EGFR, K‑Ras/B-raf, VEGF and RXR/CyclinD1, which was later utilized to categorize patients into five distinct groups of biomarker combinations. Patients in each group were randomized to receive one of the four target therapies of erlotinib, sorafenib, vandetanib or erlotonib plus bexarotene. The primary end point of disease control rate at 8 weeks after randomization was evaluated by a Bayesian hierarchical probit model. Instead of equal assignment, patients would be adaptively randomized to the targeted therapy that was deemed most promising based on cumulative clinical data of disease control. Similarly in breast cancer, building on the robust infrastructure of integrating genomics, proteomics, pathology and imaging data in I‑SPY 1 and I‑SPY 2 is a Phase II breast 864

Personalized Medicine (2013) 10(8)

cancer clinical trial in the setting of neo­adjuvant chemotherapy using an adaptive biomarkerbased design [29]. The strategy of I‑SPY 2 was to prospectively develop novel chemotherapy with corresponding biomarker signature(s) in recognition of the fact that locally advanced breast cancer can be a very heterogeneous disease. Patients enrolled in I‑SPY 2 are randomized to receive standard neoadjuvant chemotherapy (weekly paclitaxel plus transtuzumab for HER2+ and weekly paclitaxel for HER2-) plus one of five new drugs. Pathologic complete response will be used to evaluate efficacy of these new drugs for each patient group, defined by combinations of hormone receptor status, human EGF receptor 2 and MammaPrint® status. A Bayesian response-adaptive randomization was adopted to preferentially assign patients into regimens that had shown promise for their biomarker signature(s) with cumulative clinical data. At the same time, novel regimens are selected or dropped after comparison with standard therapy using Bayesian predictive probability. The BATTLE [28] and I‑SPY 2 [29] trials are examples demonstrating the challenges, opportunities and changing landscape of clinical trials in oncology. Both are examples of phase II trials, which used short-term clinical response to identify candidate predictive biomarkers for each therapy being tested in these trials. Although Bayesian adaptive randomization design can be adapted to the long-term end point of overall survival, it is generally not intended for Phase III clinical trial designs. On the other hand, most of the other biomarker-based designs are intended for Phase III clinical trials, as the detection of interaction between biomarkers and treatment needed to validate predictive biomarkers requires a large sample size. The impact of incorporating molecular pathology in prostate cancer (PCa) therapeutics with novel clinical designs for rational drug development in advanced PCa is discussed specifically in the following sections. Novel clinical trial designs in advanced PCa

This shifting repertoire of clinical design approaches was recently demonstrated in advanced PCa using a discontinuation design for a noncytotoxic agent in advanced CRPC, performed in the recently reported Phase II trial of cabozantinib (previously known as XL‑184), an oral inhibitor of MET and VEGF receptor (VEGFR) 2 [30]. However, for this development to take place in PCa, a robust understanding future science group

Pharmacogenetics- & pharmacogenomics-based rational clinical trial designs in oncology

of tumor biology and molecular factor integration with clinical trial design was essential. In the case of PCa, an example of a cancer with a high public health burden in the western world, with an estimated 33,720 deaths in 2011 alone in the USA [31], virtually all PCa-related deaths are known to occur in patients with advanced, metastatic stage disease. The initial treatment for this stage is androgen deprivation therapy (ADT) [32–34]. ADT provides effective control of disease for variable time periods in advanced metastatic hormone-sensitive PCa patients [35–38], but the disease inevitably advances to CRPC. CRPC includes a large group of patients for whom, despite the availability of several standard therapeutic options, overall survival is limited to 20–30 months (median). Recent therapeutic advances in CRPC treatment have included successful targeting of the testosterone–androgen receptor (AR) axis, with several new drugs, including the cytochrome P45017 (CYP17) inhibitor abiraterone acetate [39,40] and AR inhibitors such as enzalutamide [41]. A series of genomic studies have reinforced the importance of the AR gene in PCa, particularly when the disease has progressed to the CRPC phenotype. It has been known for some time that alterations in the AR do not occur in primary PCa, but they are found in 58% of metastatic tumors [42]. Genomic data have demonstrated that the AR pathway is altered in 56% of primary PCa and 100% of metastases. These direct alterations in the AR itself, as well as broader alterations affecting the testosterone–AR axis known to occur frequently during PCa progression, have led to exploitation of the testosterone–AR axis for therapeutic benefit, with a degree of success using traditional clinical trial designs in the recent past. For example, development of CYP17 inhibitors for the treatment of CRPC is based on these traditional clinical trial design templates. Abiraterone is a CYP17 inhibitor (US FDA approved in December 2012) for use as first-line treatment of CRPC after the failure of ADT based on results from randomized Phase III clinical trials [43]. Similarly the novel AR inhibitors such as enzalutamide, a second generation antiandrogen, has also received FDA approval for treating CRPC after the failure of chemotherapy using traditional Phase II and III designs [41]. These new diarylthiohydantoin compounds target AR by binding overexpressed AR in advanced-stage disease with an affinity that is several-fold greater than previously obtained with antiandrogens (bicalutamide and flutamide). In addition, there is disruption of the future science group

Review

nuclear translocation of AR and impairment of DNA binding to androgen response elements and recruitment of coactivators, and, thus these compounds have multifunctional antitumor capabilities. These advances have increased the previously known repertoire of therapeutic choices used for treating CRPC, which included docetaxel chemotherapy [44,45], cabazitaxel chemotherapy [46], Provenge cancer vaccine (Dendreon, WA, USA) [47] and radium‑223. All of these developments have taken place in the context of using traditional trial designs. The use of any of these therapeutic options, including recent novel agents targeting the testosterone–AR axis, at present have been based on clinical characteristics and, to some extent, on understanding of molecular biology associated with this stage of the disease, but it has not taken into account the repertoire of underlying oncogenic molecular alterations in the AR–testosterone axis or other pathways in individual tumors that may specifically drive tumor growth at this stage. It is clear that identifying these unrecognized pharmacogenetic signatures will initially require assessment of the tumor genome by performing biopsies of growing metastatic lesions, which has traditionally not been needed and constitutes a departure from traditional therapeutics in advanced PCa. Genome-based drivers of tumor progression can be identified in these biopsies by applying whole-genome DNA sequencing, which makes it possible to adapt and guide targeted anticancer therapies. Other than the AR–testosterone axis, several other somatic alterations in the PCa genome have also been reported at different stages of cancer progression. Numerous early studies laid a foundation for this approach by assessing copy number gains and losses in tumor DNA using low-resolution comparative genomic hybridization using customized or manufactured arrays. For example, a combined ana­lysis of published PCa comparative genomic hybridization studies, representing a total of 872 individual tumors, revealed multiple regions in the PCa genome that frequently displayed gain or loss [48]. Subsequent reports employing high-resolution copy number ana­lysis (>1 million probes) have provided detailed compendia of focal genomic gains and losses throughout the genomes of localized PCa, as well as metastatic CRPC [49,50]. More recently, exome sequencing has been performed in a limited number of clinical specimens and PCa xenografts representing localized disease, and CRPC have been used to study the spectrum www.futuremedicine.com

865

Review

Qin & Kohli

of somatic alterations in protein coding genes and their putative role in PCa progression [51–53]. However, there has been neither a comprehensive study performed to determine association with response to any standard treatment in CRPC patients, nor has any prospective clinical trial design been undertaken to identify signatures for drug response. The ability to choose between several therapeutic options currently available for treating CRPC stage, unlike in the past, has also increased the need for critical research focus in developing genomic signatures of efficacy and toxicity, which will replace the use of clinical, patient- and/or histopathology-based factors that have so far been the basis for estimating therapeutic benefit or toxicity potential. „„ Pharmacogenomic-based clinical design example in CRPC Targeting the well-observed heterogeneity in CRPC for developing an individualized medicine approach using novel clinical designs has fortunately begun to yield drug development results in 2013 [54]. Cabozantinib (previously known as XL‑184) is an oral inhibitor of MET and VEGFR2, both of which are overexpressed during the emergence of castrate resistance and play a critical role in the development of bone metastasis. In Phase I trials, cabozantinib has shown broad activity against several tumor types with acceptable toxicity profiles and so a Phase II study evaluated efficacy of cabozantinib in metastatic CRPC patients with a good performance status who had progressed on at least one standard treatment for castrate-resistant stage [30]. The Phase II clinical study design with cabozantinib has a key distinguishing feature as it incorporates a Phase II randomized discontinuation study design paradigm [55]. This design attempted to maintain the ability to evaluate tumors for response while minimizing exposure to placebo in tumors with objective regression, while allowing for randomized evaluation where the treatment is to delay progression. In this study, therefore, all patients received an openlabel treatment with 100 mg of oral cabozantinib during a 12‑week lead-in stage, followed by response assessments using RECIST criteria, at which point patients with stable disease were randomly assigned to receive either cabozantinib or placebo. These randomly assigned patients were then observed until they met study-defined progression criteria, at which point treatment assignment was unblinded. Patients were taken off the study if they were receiving cabozantinib or were allowed to restart cabozantinib if on 866

Personalized Medicine (2013) 10(8)

placebo. Patients restarted on cabozantinib after first progression on placebo were observed until subsequent progression. The use of this adaptive design had an 80% power of detection and a hazard ratio of 0.5 for evaluating efficacy with the primary end point being progression-free survival from the point of random assignment after completion of the 12‑week lead-in period in those patients who demonstrated initial stable disease at week 12. By the 12‑week lead-in period, 55 of the total 171 patients enrolled had discontinued the cabozantinib treatments due to either drug toxicity or lack of response. However, the most interesting part of the results of this novel drug design lie elsewhere. After initially enrolling 171 patients for the 12‑week lead-in period and randomizing 122 patients of these to cabozantinib or placebo, the study continuation was suspended on the recommendation of the study oversight committee, as it was felt that unexpectedly positive results on bone scan and decreases in pain observed during the lead-in stage of random assignment among these 122 patients made it unethical to continue randomization of patients to placebo following the leadin period. In this sense, this novel discontinuation trial design was successful in demonstrating drug efficacy and further drug development is currently being pursued in Phase III trials [30]. A future Phase III prospective biomarkerbased clinical trial attempt to develop MET and VEGFR2 overexpression as predictive biomarkers for cabozantinib efficacy will use overall survival as an end point, assuming that MET and VEGFR2 overexpression are considered probable valid predictive biomarkers; biomarker by treatment design would be used. Assuming that MET and VEGFR2 over­ expression patterns being developed for predicting drug efficacy are measured on a continuous scale, these would be categorized as binary and CRPC patients will be stratified using the binary MET and VEGFR2 overexpression patterns before randomizing into cabozantinib and standard treatment arms. In addition to the test of efficacy of cabozantinib versus standard in overall population, such biomarker by treatment design allows formal validation of whether MET and VEGFR2 overexpression can indeed be predictive for cabozantinib efficacy through comparison of cabozantinib efficacy across different biomarker-defined subsets of patients. The two biomarker expression patterns for MET and VEGFR2 overexpression may be reduced into a single signature for a predictive biomarker as in the case of CRPC patients with both MET and future science group

Pharmacogenetics- & pharmacogenomics-based rational clinical trial designs in oncology

VEGFR2 overexpression who alone may benefit from cabozantinib. Noticeably, this type of development and validation can only be carried out in large-scale Phase III clinical trials because there is not only the requirement of overall survival as an end point, but also the requirement of a large sample size in order to detect significant interaction between biomarker signatures and treatment. If, on the other hand, MET and VEGFR2 expression patterns are only acceptable as exploratory predictive biomarkers and the appropriate cut-off points for overexpression are not known, an adaptive threshold design may be more appropriate in order to develop and validate their predictive biomarker intent in a single Phase III clinical trial. CRPC patients in this case will be randomized to receive either cabozantinib or a standard therapy after measuring MET and VEGFR2 expression. A formal test for cabozantinib efficacy will be conducted for overall patients at a reduced statistical significance level and the trial stopped if the two treatment end points are statistically different. Otherwise, a second-stage biomarker-defined cabozantinib efficacy will likely be tested on a subset of patients with either MET or VEGFR2 overexpression. The statistical test for significance will be based on the permutation distribution of the maximized log-likelihood ratio statistic with a cut-off value between 0.5 and -1, assuming that the biomarker overexpression can be recorded as a percentage. If the second-stage biomarkerdefined cabozantinib efficacy is statistically significant, it could be concluded that the genetic biomarker (with estimated cut-off values) of MET and/or VEGFR2 is a predictive biomarker for cabozantinib. In a similar fashion, biomarker-based Phase III clinical trials may be designed for abiraterone acetate or enzalutamide, and other advanced CRPC-stage standard treatments with genetic mutations on the testosterone–AR axis. Since they are most likely exploratory biomarkers (cross-validated), adaptive signature and threshold design may be preferred in many scenarios depending on the level of evidence to support specific biomarkers of interest for an individual therapy.

Conclusion Advancements in molecular technologies and pathology have provided more efficient characterization of tumor biology beyond clinical- and histopathology-based variables. This characterization has ranged from genome-based future science group

Review

associations to a functional and more fundamental explanation of tumor biology within individual tumors. The direct impact of this on clinical trial development has been the advancement of trial designs that are now increasingly being based on the genetic propensity related to cancer biology and the likelihood of drugs that can target specific elements of tumor biology. These novel trial designs in oncology have thus begun to replace traditional efficacy-based trials, in which ‘all comers’ with a similar histomorphological characteristic were included for drug advancement. Taking advantage of the rapidly changing landscape in oncology is critical for successful implementation of novel drug agents that have the ability to enrich therapeutic risk:benefit ratios for individual cancer patients based on the molecular profile of the presenting tumor.

Future perspective Oncology clinical trials will increasingly be devised keeping in mind drug–genome interactions and genetic variations in individual tumors in the future. This will cause a replacement of the traditional single-arm Phase II trial design structure, with novel agents being tested for efficacy based on expression and functional relevance of molecular markers resulting in a greater chance for enhancement of drug efficacy in individual patients. Altering the trial designs in order to accommodate novel genomic signatures and heterogeneity is just one aspect of the developmental process of these novel designs for incorporating biomarkers. Logistic challenges in obtaining tissue for stage-specific biomarker profiling and scientific challenges in accurately identifying genomic and/or proteomic signatures due to inherent variations in expression will also alter clinical practice, with possibly more biopsies being performed prior to initiating drug development. Nonetheless, the goal of practicing personalized and precision medicine in cancer care is likely to be achieved with the implementation of rational clinical trials in future. Financial & competing interests disclosure The authors have no relevant affiliations or financial involve‑ ment with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, con‑ sultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties. No writing assistance was utilized in the production of this manuscript. www.futuremedicine.com

867

Review

Qin & Kohli

Executive summary Background ƒƒ This article focuses on the types of novel clinical trial design in cancer medicine, which incorporate molecular predictive factors and pharmacogenomic principles. Novel clinical trial design evolution in oncology ƒƒ There are several differences between emerging clinical trial designs in oncology, based on pharmacogenetic or molecular characteristics, and traditional efficacy studies. Types of novel clinical trials ƒƒ Novel clinical trial designs based on molecular and genomic characteristics include enrichment, biomarker-based designs, adaptive designs and Bayesian adaptive randomization designs. Current examples of novel trial designs in specific tumor types ƒƒ The shifting repertoire of clinical design approaches has begun to be applied to drug development in several tumors. As an example, in advanced castrate-resistant prostate cancer, a discontinuation design was recently used with cabozantinib (previously known as XL‑184) and was successful in demonstrating efficacy of the agent, while ethically treating all patients enrolled in the study cohort. ƒƒ These novel trial designs in oncology have thus begun to replace traditional efficacy-based trials in which ‘all comers’ with a similar histomorphological characteristic were included for drug advancement. As a result of this, drug development and therapeutics is effectively undergoing substantial transformational changes.

References 1

Weinshilboum RM, Wang L. Pharmacogenetics and pharmacogenomics: development, science, translation. Ann. Rev. Genom. Hum. Genet. 7, 223–245 (2006).

2

Wang L, McLeod HL, Weinshilboum RM. Genomics and drug response. N. Engl. J. Med. 364, 1144–1153 (2011).

3

Oldenhuis CN, Oosting SF, Gietema JA, de Vries EG. Prognostic versus predictive value of biomarkers in oncology. Eur. J. Cancer 44, 946–953 (2008).

4

5

6

7

8

9

Heinrich MC, Corless CL, Demetri GD et al. Kinase mutations and imatinib response in patients with metastatic gastrointestinal stromal tumor. J. Clin. Oncol. 21, 4342–4349 (2003). Cairncross G, Wang M, Shaw E et al. Phase III trial of chemoradiotherapy for anaplastic oligodendroglioma: long-term results of RTOG 9402. J. Clin. Oncol. 31, 337–343 (2012). Kelley RK, Van Bebber SL, Phillips KA, Venook AP. Personalized medicine and oncology practice guidelines: a case study of contemporary biomarkers in colorectal cancer. J. Natl Compr. Canc. Netw. 9, 13–25 (2011). Duffy MJ, O’Donovan N, Crown J, Use of molecular markers for predicting therapy response in cancer patients. Cancer Treat. Rev. 37, 151–159 (2011). van de Vijver MJ, He YD, van’t Veer LJ et al. A gene-expression signature as a predictor of survival in breast cancer. N. Engl. J. Med. 347, 1999–2009 (2002). Albain KS, Barlow WE, Shak S et al. Prognostic and predictive value of the 21-gene recurrence score assay in postmenopausal women with node-positive, oestrogen-receptorpositive breast cancer on chemotherapy: a

868

retrospective ana­lysis of a randomised trial. Lancet Oncol. 11, 55–65 (2010). 10 Kohli M, Tindall DJ. New developments in

the medical management of prostate cancer. Mayo Clin. Proc. 85, 77–86 (2010). 11 Dancey JE, Dobbin KK, Groshen S et al.

Guidelines for the development and incorporation of biomarker studies in early clinical trials of novel agents. Clin. Cancer Res. 16, 1745–1755 (2010). 12 Simon M, Hosking FJ, Marie Y et al. Genetic

risk profiles identify different molecular etiologies for glioma. Clin. Cancer Res. 16, 5252–5259 (2010). 13 Breiman L. Random forests. Mach. Learn. 45,

5–32 (2001). 14 Simon R, Maitournam A. Evaluating the

efficiency of targeted designs for randomized clinical trials. Clin. Cancer Res. 10, 6759–6763 (2004). 15 Maitournam A, Simon R. On the efficiency of

targeted clinical trials. Stat. Med. 24, 329–339 (2005). 16 Sargent DJ, Conley BA, Allegra C, Collette L.

Clinical trial designs for predictive marker validation in cancer treatment trials. J. Clin. Oncol. 23, 2020–2027 (2005). 17 Mandrekar SJ, Sargent DJ. Clinical trial

designs for predictive biomarker validation: one size does not fit all. J. Biopharm. Stat. 19, 530–542 (2009). 18 Liu A, Liu C, Li Q, Yu KF, Yuan VW. A

threshold sample-enrichment approach in a clinical trial with heterogeneous subpopulations. Clin. Trials 7, 537–545 (2010). 19 Wang SJ, O’Neill RT, Hung HM. Approaches

to evaluation of treatment effect in randomized clinical trials with genomic subset. Pharm. Stat. 6, 227–244 (2007).

Personalized Medicine (2013) 10(8)

20 Song Y, Chi GY. A method for testing a

prespecified subgroup in clinical trials. Stat. Med. 26, 3535–3549 (2007). 21 Freidlin B, Simon R. Adaptive signature design:

an adaptive clinical trial design for generating and prospectively testing a gene expression signature for sensitive patients. Clin. Cancer Res. 11, 7872–7878 (2005). 22 Jiang W, Freidlin B, Simon R. Biomarker-adaptive

threshold design: a procedure for evaluating treatment with possible biomarker-defined subset effect. J. Natl Cancer Inst. 99, 1036–1043 (2007). 23 Freidlin B, Jiang W, Simon R. The cross-

validated adaptive signature design. Clin. Cancer Res. 16, 691–698 (2010). 24 Eickhoff JC, Kim K, Beach J, Kolesar JM, Gee

JR. A Bayesian adaptive design with biomarkers for targeted therapies. Clin. Trials 7, 546–556 (2010). 25 Lee JJ, Xuemin G, Suyu L. Bayesian adaptive

randomization designs for targeted agent development. Clin. Trials 7, 584–596 (2010). 26 Korn EL, Freidlin B. Outcome – adaptive

randomization: is it useful? J. Clin. Oncol. 29, 771–776 (2011). 27 Lee JJ, Chen N, Yin G. Worth adapting?

Revisiting the usefulness of outcome-adaptive randomization. Clin. Cancer Res. 18, 4498–4507 (2012). 28 Zhou X, Liu S, Kim ES, Herbst RS, Lee JJ.

Bayesian adaptive design for targeted therapy development in lung cancer – a step toward personalized medicine. Clin. Trials 5, 181–193 (2008). 29 Barker AD, Sigman CC, Kelloff GJ, Hylton

NM, Berry DA, Esserman LJ. I‑SPY 2: an adaptive breast cancer trial design in the setting of neoadjuvant chemotherapy. Clin. Pharmacol. Ther. 86, 97–100 (2009).

future science group

Pharmacogenetics- & pharmacogenomics-based rational clinical trial designs in oncology

30 Siegel R, Ward E, Brawley O, Jemal A.

Cancer statistics, 2011: the impact of eliminating socioeconomic and racial disparities on premature cancer deaths. CA Cancer J. Clin. 61, 212–236 (2011). 31 Smith DC, Smith MR, Sweeney C et al.

Cabozantinib in patients with advanced prostate cancer: results of a Phase II randomized discontinuation trial. J. Clin. Oncol. 31, 412–419 (2013). 32 Huggins C, Hodges CV. Studies on prostatic

cancer: I. The effect of castration, of estrogen and of androgen injection on serum phosphatases in metastatic carcinoma of the prostate. 1941. J. Urol. 168, 9–12 (2002). 33 Huggins C. Prostatic cancer treated by

orchiectomy; the five year results. J. Am. Med. Assoc. 131, 576–581 (1946). 34 Huggins C. Effect of orchiectomy and

irradiation on cancer of the prostate. Ann. Surg. 115, 1192–1200 (1942). 35 Crawford ED, Eisenberger MA, McLeod DG

et al. A controlled trial of leuprolide with and without flutamide in prostatic carcinoma. N. Engl. J. Med. 321, 419–424 (1989). 36 Denis LJ, Keuppens F, Smith PH et al.

Maximal androgen blockade: final ana­lysis of EORTC Phase III trial 30853. EORTC Genito–Urinary Tract Cancer Cooperative Group and the EORTC Data Center. Eur. Urol. 33, 144–151 (1998). 37 Eisenberger MA, Blumenstein BA, Crawford

ED et al. Bilateral orchiectomy with or without flutamide for metastatic prostate cancer. N. Engl. J. Med. 339, 1036–1042 (1998). 38 Prostate Cancer Trialists’ Collaborative

Group. Maximum androgen blockade in advanced prostate cancer: an overview of the randomised trials. Prostate Cancer Trialists’

future science group

Review

Collaborative Group. Lancet 355, 1491–1498 (2000).

resistant prostate cancer. N. Engl. J. Med. 363, 411–422 (2010).

39 Attard G, Reid AH, A’Hern R et al. Selective

48 Sun J, Chang BL, Isaacs SD et al. Cumulative

inhibition of CYP17 with abiraterone acetate is highly active in the treatment of castrationresistant prostate cancer. J. Clin. Oncol. 27, 3742–3748 (2009). 40 de Bono JS, Logothetis CJ, Molina A et al.

Abiraterone and increased survival in metastatic prostate cancer. N. Engl. J. Med. 364, 1995–2005 (2011). 41 Scher HI, Fizazi K, Saad F et al. Increased

survival with enzalutamide in prostate cancer after chemotherapy. N. Engl. J. Med. 367, 1187–1197 (2012). 42 Taylor BS, Schultz N, Hieronymus H et al.

Integrative genomic profiling of human prostate cancer. Cancer Cell 18, 11–22 (2010). 43 Ryan CJ, Smith MR, de Bono JS et al.

Abiraterone in metastatic prostate cancer without previous chemotherapy. N. Engl. J. Med. 368, 138–148 (2013). 44 Petrylak DP, Tangen CM, Hussain MH et al.

Docetaxel and estramustine compared with mitoxantrone and prednisone for advanced refractory prostate cancer. N. Engl. J. Med. 351, 1513–1520 (2004). 45 Tannock IF, de Wit R, Berry WR et al.

Docetaxel plus prednisone or mitoxantrone plus prednisone for advanced prostate cancer. N. Engl. J. Med. 351, 1502–1512 (2004). 46 de Bono JS, Oudard S, Ozguroglu M et al.

Prednisone plus cabazitaxel or mitoxantrone for metastatic castration-resistant prostate cancer progressing after docetaxel treatment: a randomised open-label trial. Lancet 376, 1147–1154 (2010). 47 Kantoff PW, Higano CS, Shore ND et al.

Sipuleucel-T immunotherapy for castration-

www.futuremedicine.com

effect of five genetic variants on prostate cancer risk in multiple study populations. Prostate 68, 1257–1262 (2008). 49 Liu W, Laitinen S, Khan S et al. Copy

number ana­lysis indicates monoclonal origin of lethal metastatic prostate cancer. Natl Med. 15, 559–565 (2009). 50 Mao X, Yu Y, Boyd LK et al. Distinct

genomic alterations in prostate cancers in Chinese and western populations suggest alternative pathways of prostate carcinogenesis. Cancer Res. 70, 5207–5212 (2010). 51 Holcomb IN, Young JM, Coleman IM et al.

Comparative analyses of chromosome alterations in soft-tissue metastases within and across patients with castration-resistant prostate cancer. Cancer Res. 69, 7793–7802 (2009). 52 Ruiz C, Lenkiewicz E, Evers L et al.

Advancing a clinically relevant perspective of the clonal nature of cancer. Proc. Natl Acad. Sci. USA 108, 12054–12059 (2011). 53 Kumar A, White TA, MacKenzie AP et al.

Exome sequencing identifies a spectrum of mutation frequencies in advanced and lethal prostate cancers. Proc. Natl Acad. Sci. USA 108, 17087–17092 (2011). 54 Goodin S, DiPaola RS. Cabozantinib in

prostate cancer: the beginning of a precision paradigm? J. Clin. Oncol. 31, 401–403 (2013). 55 Rosner GL, Stadler W, Ratain MJ.

Randomized discontinuation design: application to cytostatic antineoplastic agents. J. Clin. Oncol. 20, 4478–4484 (2002).

869