Predictive and Prognostic Factors for Non–Small Cell Lung Cancer—Potholes in the Road to the Promised Land
Adi F. Gazdar and Joan H. Schiller
Affiliations of authors: Hamon Center for Therapeutic Oncology Research, Department of Pathology (AFG), and Division of Hematology and Oncology, Department of Internal Medicine and Simmons Comprehensive Cancer Center (JHS), University of Texas Southwestern Medical Center, Dallas, TX
Surgery offers the best chance of long-term cures for early-stage non–small cell lung cancer, the most common form of lung cancer. However, between one-third and one-half of early-stage tumors will relapse after curative intent resection. Adjuvant chemotherapy (ACT) offers a survival advantage for up to 15% of early-stage resected cancers and has become the standard of care. Thus, having the ability to select the minority of patients who are likely to benefit from ACT would be enormously beneficial to the majority of early-stage patients who undergo resection who could avoid experiencing the morbidity associated with ACT. This strategy would also create economic benefits in addition to the health benefits for patients. In this issue of the Journal, Chen et al. (1) present a microarray-based gene expression signature that has both prognostic and predictive values for early-stage non–small cell lung cancer.
Many markers have been tested for their predictive and prognostic value in lung cancer, and some have entered clinical practice. These include pathological typing that may be used to include or exclude certain forms of therapy—for example, adenocarcinoma histology is a strong predictor of outcome of pemetrexed therapy in advanced patients, and serious hemorrhagic complications may occur after bevacizumab therapy in patients with squamous histologies. Lung cancers are remarkably heterogeneous at the molecular level, and at least nine “driver” oncogene mutations have been described in lung adenocarcinomas (2). These mutations may lead to oncogene addiction, and effective therapies have been approved by the US Food and Drug Administration (FDA) for epidermal growth factor receptor mutations and ALK translocations, and many other individualized forms of therapy are currently in clinical trials (2). However, nearly half of lung adenocarcinomas and most of the squamous cell carcinomas lack known driver mutations.
The completion of the human genome project in the year 2003 was a bonanza for medical research because of the development and clinical applications of genomic-based assays including increasingly dense microarray platforms for global analyses of gene expression, copy number variation, DNA methylation and microRNA. In particular, gene expression studies have grabbed the limelight, with at least 16 prognostic or predictive gene signatures having been described for lung cancer (1,3), as well as many for breast and the other major cancers. Although these studies offer promise for personalized medicine applications (4,5), there are also many pitfalls. Given that tens of thousands of expressed genes are examined simultaneously, it is not surprising that few of the reported gene signatures have multiple genes in common. The biological relevance (if any) of most signature genes is unknown. For continuous variables, patients are usually divided arbitrarily, often on the basis of the median value. Major mistakes in the methods of analysis and even potential fraud have been detected in some published studies, and others cannot be fully interpreted or verified. Independent reviewers could not replicate the data for one highly publicized study for the individualized selection of chemotherapy for breast cancer patients (6). The study was inadequately documented, the role of biostatisticians was minimal, and major analytical mistakes were detected. After a lengthy battle of words between critics and investigators and investigations by the National Cancer Institute and the Institute of Medicine, the study was suspended, several published reports in leading journals were withdrawn, some investigators were reassigned or resigned, and lawsuits were filed on the behalf of patients who claimed they had received wrongfully identified nonstandard therapies.
The above mentioned problems have led to a certain amount of disillusionment and skepticism regarding signature studies, and they stimulated the FDA to declare that “in vitro diagnostic multivariate index assays are also medical devices subject to FDA review” (6). Certain guidelines for the planning, execution, and transparent reporting of such assays have also been instituted (6,7). Baggerly and Coombes (6,7) identified five essential requirements for authors and journals to follow, and they should be fulfilled before starting clinical trials based on “omic” signatures to guide treatment. The requirements are that authors and journals give readers access to 1) the raw data, 2) the code used to derive the results from the raw data, 3) evidence of provenance of the raw data, 4) written descriptions of any nonscriptable analysis steps, and 5) prespecified analysis plans.
Why is there a need for yet another gene expression–based prognostic signature study in lung cancer? A critical appraisal of the published studies of gene expression signatures for lung cancer found serious problems in methodology and analysis in most of them (3). Subramanian and Simon (3) also pointed out that the major clinical usefulness of a prognostic signature for lung cancer is to identify early-stage patients who will benefit from ACT and those who could be spared. The study reported by Chen et al. (1) differs from most other published studies in several aspects. They appear to have satisfied the requirements of Baggerley and Coombes (6), including presenting a detailed description of their non-scriptable analysis steps in Sweave format in the Supplementary Materials (available online). By comparing the expression patterns of nonmalignant and malignant breast tissues, they had previously defined a malignancy-risk gene signature that was associated with cancer risk in nonmalignant breast tissue and was a prognostic factor for breast cancer (8). Whereas most microarray-based signatures largely consist of genes whose relationship to cancer is peripheral or unknown, the malignancy-risk gene signature was markedly enhanced for genes related to cell proliferation (8). As both breast and lung cancers are characterized by early loss of normal growth control mechanisms, they applied the malignancy-risk signature to lung cancers by using three publicly available datasets. Chen et al. (1) used one large dataset to test their hypothesis and the other two to validate their findings by generating an overall malignancy-risk score. Surprisingly, the malignancy-risk gene signature had both prognostic and predictive value: Of patients who did not receive ACT, those with a low–malignancy-risk score had increased overall survival compared with those with a high-malignancy score. Results from the test and validation sets were in agreement, even though the studies used different expression platforms to generate their data. As the authors suggest, this may be the first gene expression signature that has potential clinical value in two major forms of cancer. Another positive aspect is that the first author is a member of the Biostatistics Department, indicating that the authors recognized the importance of this discipline to such studies.
Although the study by Chen et al. (1) suggests that clinically relevant subgroups can be identified by a gene expression–based signature, it cannot be “translated” until it has been independently validated in a prospectively conducted trial. The road to the promised land of personalized therapy is a long one and littered with potholes. Hopefully, Chen et al. have successfully avoided most of them.
1. Chen D-S, Hsu Y-L, Fulp WJ, et al. Prognostic and predictive value of a malignancy-risk gene signature in early-stage non–small cell lung cancer. J Natl Cancer Inst. 2011;103(24):1859-1870.
2. Pao W, Girard N. New driver mutations in non-small-cell lung cancer. Lancet Oncol. 2011;12(2):175-180.
3. Subramanian J, Simon R. Gene expression-based prognostic signatures in lung cancer: ready for clinical use? J Natl Cancer Inst. 2010;102(7):464-474.
4. Xie Y, Minna JD. Predicting the future for people with lung cancer. Nat Med. 2008;14(8):812-813.
5. Xie Y, Minna JD. Non-small-cell lung cancer mRNA expression signature predicting response to adjuvant chemotherapy. J Clin Oncol. 2010;28(29):4404-4407.
6. Baggerly KA, Coombes KR. What information should be required to support clinical “omics” publications? Clin Chem. 2011.
7. Baggerly K. Disclose all data in publications. Nature 2010;467(7314):401.
8. Chen DT, Nasir A, Culhane A, et al. Proliferative genes dominate malignancy-risk gene signature in histologically-normal breast tissue. Breast Cancer Res Treat. 2010;119(2):335-346.
Gregory D. Pawelski