The Future of Predictive Testing

Over the past 10 years, there has been an abundance of disease-based cDNA expression studies motivated in part by drug discovery programs in search of novel genes as potential therapeutic targets. The focus has been in the area of oncology, specifically hematological (clonal) malignancies such as non-Hodgkin's lymphoma (follicular and chronic lymphocytic).^1-4 The emphasis, however, has recently expanded to include solid tumors such as bladder, colon, ovary, prostate, breast, lung, and brain.^5-12

Figure 1. Utilizing Aureon Laboratories´ Magic™ software, the cellular tissue components contained within an image of non-small cell lung (adeno) carcinoma are segmented and classified. The figure illustrates the postprocessing, color-annotated identification of all cellular elements including epithelial (BLUE) and stromal nuclei (GREEN), fibro-connective tissue (PINK), and Inflammatory cells (YELLOW). Individual features are then quantified including the degree of (epithelial) tumor cellularity and cell-to-cell relationships.

The accumulation of the genomic information has made it rather apparent that many of the identified genes, while potentially therapeutically promising, are also informative with respect to subclassifying clinical pathologic diagnoses. Furthermore, by expanding studies to include clinically well-annotated patient samples with appropriate follow-up, some of these gene signatures have even been successful in predicting outcome.

These efforts have proven critical in the success of several biological and drug-targeted therapeutic discovery programs, where the target molecule dictates the likelihood of response; examples include Herceptin (Her2/neu) for breast cancer,¹³ Gleevec (BCR-ABL) for chronic myelogenous leukemia,⁴ and (c-kit) gastrointestinal stromal tumors,¹⁴ and Avastin (VEGF) for advanced colon and lung cancer.^15,16 These agents are predicted to work in a population of patients whose tumor growth and/or survival rate are driven by either the target or target-dependent processes.¹⁷

To address the changing landscape of therapeutic drug discovery, the FDA is gradually putting pressure on the biopharmaceutical industry to develop companion diagnostics in parallel with the targeted therapies so that the industry develops "targeted designs" for clinical trials.¹⁸ These types of trials would be designed to:

• reduce the number of enrolled patients

• improve upon the optimization strategies for defining the pharmacologic dose of the targeted therapy

• minimize the number of unnecessary adverse drug events

• reach the targeted end point.

It is anticipated that over the next few years such companion diagnostics will be required to select patients, initiate treatment protocols, and potentially stratify patients for selective therapies.

The challenge, however, will be in the commercialization and clinical application of these complex diagnostics for routine use in medical practice. General oversight for such predictive diagnostics will likely reside in specialized reference laboratories such as Genomic Health, US Labs, LabCorp, and Aureon Laboratories. Currently, these groups are regulated by CLIA/CLEP licensing, and some have CAP (College of American Pathologists) certification.

A number of mechanisms have been established to ensure appropriate utilization of human specimens, patient confidentiality, and laboratory techniques, including proficiency tests and Standard Operating Procedures for the assay, personnel, and result reporting. In addition, the FDA has recently published draft guidelines for the development and use of multiplex tests for heritable DNA markers, mutations, and expression patterns (see http://www.fda.gov/cdrh/oivd/guidance/1220). The regulatory bodies that monitor these types of tests, however, have very little experience in this area; therefore, further work is necessary to ensure that the organizations responsible for these complex tests adhere to certain guidelines.

Testing obstacles

Although the overwhelming data is provocative, prevailing technical and study design issues continue to slow down the progress of bringing these types of tests and their results to the clinic. Challenges such as gene-expression reproducibility, low-expression genes, expression averaging across array platforms, agreed-upon robust standards, specimen sampling error, and the small number of analyzed patients are significant and have yet to be entirely resolved. Furthermore, novel mathematical approaches to analyze the data are necessary to render a more meaningful and individualized conclusion.

Technical advances, including high-quality expression arrays from Affymetrix, correlative testing with quantitative PCR, and specimen manipulation (i.e., laser capture microdissection and RNA amplification strategies), have been useful but not sufficient enough to address the inherent heterogeneity that exists within the disease process being studied. In addition, there are still only a handful of studies in which gene expression patterns have been translated into either a clinical diagnostic or prognostic assay. Two recent examples include a diagnostic kit for the epidermal growth factor receptor (EGFR) in colon cancer and other solid tumor types¹⁹ and a prognostic assay for the ZAP-70 tyrosine kinase in chronic lymphocytic leukemia (CLL).^20,21 Both are protein-based, FDA-approved, immuno-histochemical assays that have been recently included in clinical diagnostic laboratories.

The latest development in the predictive expression arena has been the release of a 21-gene panel (Oncotype DX, Genomic Health, Inc.) for predicting recurrence of breast cancer after hormonal adjuvant therapy.²² This assay was initially derived from cDNA expression arrays performed on fresh frozen tissue samples and then reformatted as an RT-PCR test, which utilizes formalin-fixed, paraffin-embedded (FFPE) tissue specimens.

As discussed, one of the major drawbacks in using data derived from tissue expression profiling has been the implicit loss of tissue heterogeneity once the sample has been processed. Laser capture microdissection systems have aided in creating defined "cellular" signatures²³ ; however, the in situ cellular relationships are minimized as a result of the microdissection process. Furthermore, the majority of these assays are complex, requiring special hardware platforms and technical sophistication with respect to specimen handling, especially when RNA is involved.

Industry growth

There has been a renewed effort by the pharmaceutical industry to employ more sensitive and comprehensive pharmaco-efficacious methods to better understand a patient's response to therapy, including safety and toxicity concerns. One example is the association of specific point mutations in the EGFR gene with respect to IRESSA (AstraZeneca) response in patients diagnosed with non-small cell lung cancer (NSCLC).²⁶ Such analysis helps to illustrate the benefits of the industry's efforts; however, the low incidence of EGFR mutations and the mixed results from several clinical trials suggest that additional factors are clearly involved in the response pathway and that further studies are needed.

When it comes to predictors of molecular toxicity, there are a number of examples in which genetic polymorphisms in drug metabolizing enzymes have impacted upon clinical outcome.²⁷ One example includes the presence of the UGT1A1 (uridine diphosphate-glucuronosyltransferase) variant, which when present in patients treated with Irinotecan results in a higher chance of grade IV toxicity.²⁸ This example illustrates how molecular phenotypic data can have an impact on the development of appropriate treatment algorithms. Of course, the final goal needs to be the creation of more highly integrative patient management systems that are tailored to the specific disease process at hand.

In regard to the IRESSA example above, a key question facing AstraZeneca—and all biopharmaceutical companies today—is "What are the best approaches and technologies currently available to evaluate response, given the limited amount and type of tissue specimens available from clinical trials?" Ultimately, this information could potentially assist in the design of subsequent clinical trials, as well as impact on preclinical and development projects of similar class drugs.

One approach would be to use FFPE tissue samples from the de novo primary tumor specimen of patients who received IRESSA and where outcome data is available; in this case, factors present in the primary tumor sample could potentially be used to assess and/or predict a clinical objective response to the drug. Similar approaches are currently underway in prostate cancer whereby specific biomarker expression patterns have been identified in the FFPE prostatectomy specimen and linked to a therapeutic response.²⁹ Although there are limitations in the types of assays that can be performed with FFPE, the samples in general are very attractive for clinical trial management because they can be easily procured and stored for long periods of time.

Ongoing evaluation

The following pilot study was designed to utilize the FFPE samples obtained from a patient cohort with NSCLC enrolled in an expanded access program for IRESSA. The goal was to evaluate whether IRESSA response, as outlined below, was associated with EGFR mutation status in the patient's primary tumor sample. The objective response was defined as per WHO criteria and included: complete response (CR), partial response (PR), minor response (MR), stable disease (SD), and progressive disease (PD). The EGFR mutation status was evaluated from DNA extracted from each of the patient's tumor samples. Then, the mutation data was incorporated along with routine clinical variables such as gender, smoking history, and ethnicity to create a unique profile of each patient.

The specific characteristics of the EGFR analyses, including binary evaluation, e.g., presence/absence of mutation, and quantitative attributes (e.g., number and type of exon involved (18–24)) and type of mutation (missense/deletion), were all considered patient features and evaluated as part of the integrative "systems pathology" model. A similar "systems approach" has been reported in prostate cancer in which elements of the patient's clinical history, biomarker data, and bio-imaging analytics from the primary tumor specimen were placed into a mathematical model to predict PSA recurrence.³⁰ It is anticipated that once the IRESSA pilot study has been completed, a model predicting outcome based in part on EGFR mutation status will be developed.

The pilot study included an initial assessment of the tissue sample for overall quality using standard staining practices with Hematoxylin and Eosin (H&E). Digitized images were obtained for morphometric analysis, which included characterization of the tumor sample with respect to overall cellular composition, degree of differentiation as reflected by cytoarchitectural changes, angiogenic profiles, inflammatory composition, and even tumor grading characteristics (see Figure 1 for an example). The digitized H&E section was segmented by a proprietary software program that utilizes computational algorithms to modify what is traditionally a subjective assessment.

The clinical trial tissue specimens were then analyzed using a series of antibodies as biomarkers that were selected based on their reported scientific association with NSCLC tumor biology and/or EGFR pathway functionality. Antigens were evaluated with immuno-fluorescent labeling strategies utilizing spectral imaging in a multiplexed format. This conserved the overall use of the valuable tissue sample while maximizing on the type and quality of the data that was generated. Since the output was a fluorescent signature of the tumor sample linked to the histomorphology, a quantitative assessment of proteins and their cellular distribution was possible. The results were recorded from the tissue sample as quantitative continuous variables and then entered into the predictive model.

Previous strategies using standard Immunohistochemistry for assessing EGFR status in tissue samples and associating this with response have been inconclusive,³¹ thus prompting the need for incorporating more quantitative approaches to evaluate tumor antigens in clinical trial materials. In addition, recent publications have further confirmed the critical need for quantifying tumor antigens from tissue sections, especially when developing predictive models.³²

The final step in the development of the IRESSA response pilot study was to define the specific primary and secondary objectives, which were based on the clinical outcome of the trial. Even though the pilot study is underway, previously described models using similar strategies have proven very successful in predicting clinical outcome.³⁰

Summary

In Summary, traditional cancer diagnostics begin with a histologic assessment of the patient's tumor sample (i.e., morphology, tumor grade, staging); then, additional assays are utilized, including Immunohistochemistry, FISH, Cytogenetics, Flow Cytometry, and Real-Time PCR, to confirm a diagnosis. What has been lacking from this approach is the ability to make an accurate assessment of the patient's predictive response to specific therapies, especially in a clinical trial setting. Given that the future of molecular medicine is "smart drug" therapeutics and "targeted" clinical trial designs, there is a critical and immediate need for integrative and accurate personalized predictive analyses.

Michael J. Donovan is senior vice president, research collaborations, with Aureon Biosciences Corporation, 28 Wells Ave., Yonkers, NY 10701, michael.donovan@aureon.com

Carlos Cordon-Cardo is on the board of directors at Aureon Biosciences Corporation and the director of molecular pathology at Memorial Sloan-Kettering Cancer Center, 1275 York Ave., New York, NY 10021, (212) 639-7746.

References

1. S. Dave, G. Wright, B. Tan et al., "Prediction of Survival in Follicular Lymphoma Based on Molecular Features of Tumor-Infiltrating Immune Cells," New England Journal of Medicine, 351, 2159–2169 (2004).

2. M. Iwashima, B.A. Irving, N.S. van Oers, A.C. Chan, A. Weiss, "Sequential Interactions of the TCR with Two Distinct Cytoplasmic Tyrosine Kinases," Science, 263, 1136–1139 (1994).

3. L. Stoudt, "Molecular Diagnosis of the Hematologic Cancers," New England Journal of Medicine, 348, 1777–1785 (2003).

4. J.E. Cortes, S.M. O'Brien, F. Giles, R.H. Alvarez, M. Talpaz, H.M. Kantarjian, "Investigational Strategies in Chronic Myelogenous Leukemia," Hematology/Oncology Clinics of North America, 18, 619–639 (2004).

5. S. Sanchez-Carbayo, N. Socci, E. Charytonowicz et al., "Molecular Profiling of Bladder Cancer Using cDNA Microarrays: Defining Histogenesis and Biological Phenotypes," Cancer Research, 62, 6973–6980 (2002).

6. S. Nishizuka, S.T. Chen, F. Gwadry et al., "Diagnostic Markers that Distinguish Colon and Ovarian Adenocarcinomas: Identification by Genomic, Proteomic, and Tissue Array Profiling," Cancer Research, 63, 5243–5250 (2003).

7. L. Hartmann, K. Lu, G. Linette et al., "Gene Expression Profiles Predict Early Relapse in Ovarian Cancer After Platinum-Paclitaxel Chemotherapy," Clinical Cancer Research, 11, 2149–2155 (2005).

8. G. Glinsky, A. Glinsky, A. Stephenson, R. Hoffman, W. Gerald, "Gene Expression Profiling Predicts Clinical Outcome of Prostate Cancer," Journal of Clinical Investigation, 113, 913–923 (2004).

9. P. Meltzer, "Gene Expression Profiling in Breast Cancer Research," Breast Disease, 19, 23–27 (2004).

10. S. Loi, C. Desmedt, F. Cardoso, M. Piccard, C. Sotiriou, "Breast Cancer Gene Expression Profiling: Clinical Trial and Practice Implications," Pharmacogenomics Journal, 6, 49–58 (2005).

11. J. Heighway, T. Knapp, L. Boyce et al., "Expression Profiling of Primary Non-Small Cell Lung Cancer for Target Identification," Oncogene, 21, 7749–7763 (2002).

12. J. Nigro, A. Misra, L. Zhang et al., "Integrated Array-Comparative Genomic Hybridization and Expression Array Profiles Identify Clinically Relevant Molecular Subtypes of Glioblastoma," Cancer Research, 65, 678–686 (2005).

13. J. Ross, J. Fletcher, G. Linette et al., "The HER-2/neu Gene and Protein in Breast Cancer 2003: Biomarker and Target of Therapy," The Oncologist, 8, 307–325 (2003).

14. R. Logrono, D.V. Jones, S. Faruqi, M.S. Bhutani, "Recent Advances in Cell Biology, Diagnosis, and Therapy of Gastro-intestinal Stromal Tumor (GIST)," Cancer Biology & Therapy, 3, 251–258 (2004).

15. J. Sparano, B. Giantonio, P. O'Dwyer, R. Comis, "Evaluating Antiangiogenesis Agents in the Clinic: The Eastern Cooperative Oncology Group Portfolio of Clinical Trials," Clinical Cancer Research, 10, 1206–1211 (2004).

16. R. Herbst, D. Johnson, E. Mininberg, "Phase I/II Trial Evaluating the Anti-Vascular Endothelial Growth Factor Monoclonal Antibody Bevacizumab in Combination With the HER-1/Epidermal Growth Factor Receptor Tyrosine Kinase Inhibitor Erlotinib for Patients With Recurrent Non-Small-Cell Lung Cancer," Journal of Clinical Oncology (2005).

17. J. Baselga, "Skin as a Surrogate Tissue for Pharmacodynamic End Points," Clinical Cancer Research, 9, 2389–2390 (2003).

18. R. Simon and A. Maitournam, "Evaluating the Efficiency of Targeted Designs for Randomized Clinical Trials," Clinical Cancer Research, 10, 6759–6763 (2004).

19. M. Resnick, J. Routhier, T. Konkin, E. Sabo, V. Pricolo, "Epidermal Growth Factor Receptor, c-MET, B-Catenin, and p53 Expression as Prognostic Indicators in Stage II Colon Cancer," Clinical Cancer Research, 10, 3069–3075 (2004).

20. M. Crespo, F. Bosch, N. Villamor et al., "ZAP-70 Expression as a Surrogate for Immunoglobulin-Variable-Region Mutations in Chronic Lymphocytic Leukemia," New England Journal of Medicine, 348, 1764–1775 (2003).

21. L. Rassenti, L. Huynh, T. Toy et al., "ZAP-70 Compared With Immunoglobulin Heavy-Chain Gene Mutation Status as a Predictor of Disease Progression in Chronic Lymphocytic Leukemia," New England Journal of Medicine, 351, 893–901 (2004).

22. S. Paik, S. Shak, G. Tang et al., "A Multigene Assay to Predict Recurrence of Tamoxifen-Treated, Node-Negative Breast Cancer," New England Journal of Medicine, 351, 2817–2826 (2004).

23. V. Luzzi, M. Mahadevappa, R. Raja, J. Warrington, M. Watson, "Accurate and Reproducible Gene Expression Profiles from Laser Capture Microdissection, Transcript Amplification, and High Density Oligonucleotide Microarray Analysis," Journal of Molecular Diagnostics, 5, 9–14 (2003).

24. W. Pao, V. Miller, M. Zakowski et al., "EGF Receptor Gene Mutations are Common in Lung Cancers from 'Never Smokers' and are Associated with Sensitivity of Tumors to Gefitinib and Erlotinib," PNAS, 101, 13306–13311 (2004).

25. S.W. Han, T.Y. Kim, P.G. Hwang et al., "Predictive and Prognostic Impact of Epidermal Growth Factor Receptor Mutation in Non-Small-Cell Lung Cancer Patients Treated with Gefitinib," Journal of Clinical Oncology, 23, 2493–2501 (2005).

26. H. Cortes-Funes, C. Gomez, R. Rosell et al., "Epidermal Growth Factor Receptor Activating Mutations in Spanish Gefitinib-Treated Non-Small-Cell Lung Cancer Patients," Annals of Oncology, 16, 1081–1086 (2005).

27. H.L. McLeod and J. Yu, "Cancer Pharmacogenomics: SNPs, Chips, and the Individual Patient," Cancer Investigation, 21, 630–640 (2003).

28. W. Lee, A.C. Lockhart, R. Kim, M. Rothenberg, "Cancer Pharmacogenomics: Powerful Tools in Cancer Chemotherapy and Drug Development," The Oncologist, 10, 104–111 (2004).

29. M.R. Smith and J.B. Nelson, "Future Therapies in Hormone-Refractory Prostate Cancer," Urology, 65 (5 Suppl), 9–16 (2005).

30. C. Cordon-Cardo, A. Kotsianti, M.J. Donovan et al., "Improved Prediction of PSA Recurrence Through Systems Pathology," Journal of Clinical Oncology, ASCO Meeting Abstract, 4591 (2004).

31. H.S. Parra, R. Cavina, F. Latteri et al., "Analysis of Epidermal Growth Factor Receptor Expression as a Predictive Factor for Response to Gefitinib (IRESSA, ZD1839) in Non-Small-Cell Lung Cancer," British Journal of Cancer, 91, 208–212 (2005).

32. M. Rubin, T. Bismar, O. Andre et al., "Decreased Alpha-Methylacyl CoA Racemase Expression in Localized Prostate Cancer is Associated with an Increased Rate of Biochemical Recurrence and Cancer-Specific Death," Cancer Epidemiology Biomarkers & Prevention, 14, 1424–1432 (2005).