Non–Small-Cell Lung Cancer mRNA Expression Signature Predicting Response to Adjuvant Chemotherapy

Abstract

Despite undergoing an apparent complete resection of non– small-cell lung cancer (NSCLC) with curative intent, 33% of patients with pathologic stage IA and 77% with stage IIIA disease die within 5 years of diagnosis, most because of metastatic disease present at the time of surgical resection. Several randomized trials showed that adjuvant chemotherapy (ACT) directed against this microscopic metastatic disease improves survival of patients with resected NSCLC. However, the effect of ACT on prolonging overall and disease-free survival is modest, with 4% to 15% improvement in 5-year survival, and often ACT is associated with serious adverse effects. Therefore, prospectively identifying the subgroup(s) of patients who will most likely benefit from any or a specific type of ACT would be of substantial clinical benefit. Currently, a patient’s TNM stage is the main clinical variable that provides prognostic information to suggest which patients need ACT. However, the TNM information (or the specific tumor histopathologic subtype) does not predict which patients within a TNM-stage category will derive survival benefit from ACT. A prognostic biomarker signature (whether it is derived from studies of the tumor or other patient materials, such as blood) separates a population with respect to the outcome of interest, irrespective of treatment, and can be used to estimate disease-related patient trajectories. Subramanian and Simon, in a review of 16 studies on tumor gene expression–based prognostic signatures in lung cancer, suggested guidelines for the design, analysis, and evaluation of prognostic signature studies. By contrast, a predictive biomarker signature separates a population with respect to the outcome of interest in response to a particular treatment and can be used to predict the usefulness of a given treatment in a specific patient. Because predictive signatures directly address the question of which patients are more likely to benefit—or not—from a specific treatment, they have more direct impact on the clinical decisions concerning treatment selection. In interactions with individual patients, although it is great to be able to tell a patient that he or she has a tumor with a profile that augurs for a good prognosis and that no further treatment is needed, it does little good to describe molecular studies of the patient’s tumor showing that the patient’s prognosis is poor, and that there is no specific treatment approach. By contrast, it is obviously more satisfying to the physician and patient to be able to indicate that, although a patient has a tumor with a poor prognosis, the tumor’s characteristics suggest that it is most likely to benefit from a specific type of treatment. Thus, prognostic and predictive biomarker signatures need to be integrated into clinical care. Individual biomarkers have been investigated as predictive markers for ACT response. The International Adjuvant Lung Cancer Trial (IALT Bio) study analyzed 761 NSCLC tumors from a large clinical trial and showed that patients with completely resected NSCLCs (by clinical pathologic stage) whose tumors did not express ERCC1 (ERCC1-negative tumors) benefited from cisplatin-based adjuvant chemotherapy, whereas patients with ERCC1-positive tumors did not. Other potentially predictive tumor biomarkers, including RRM1, p53, and RAS, gave inconsistent results. It is in this context that Zhu et al report that an NSCLC tumor– derived 15-gene expression signature is both prognostic for survival of untreated patients and predictive for survival after ACT. The gene expression signature was first derived using microarray analyses of frozen tumor tissue from 62 patients who were in the observation group of the National Cancer Institute of Canada Clinical Trials Group JBR.10 randomized controlled clinical trial of adjuvant vinorelbine plus cisplatin versus observation. The authors identified the minimum number of genes required to provide mRNA expression levels that enabled the classification of these patients into good and poor prognosis groups. The authors found this signature to be prognostic independent of clinically available variables (histology, stage, age, and sex); in addition, they found it to be prognostic in four other published NSCLC microarray data sets (resected tumors without adjuvant therapy) and in other observed patients enrolled onto the JBR.10 trial. They were also able to test the same samples by the more quantitative methodology for mRNA expression patterns, quantitative real-time polymerase chain reaction, which confirmed the microarray findings. However, the real surprise came when Zhu et al tested the signature on patients enrolled onto the JBR.10 trial who received ACT. They found that patients whose tumors were predicted to have a poor prognosis but who received ACT exhibited significantly (and dramatically) better survival than the observed patients (those who did not receive ACT) whose tumors had a poor prognosis signature. By contrast, if a patient’s tumor showed a good prognosis signature and ACT was administered, the patient did significantly worse than patients with a good signature who were just observed. Thus, the tumor biomarker information benefited patients in several ways: for patients whose tumors showed a good prognosis, it not only suggested that the patients did not need ACT, but that it might in fact be harmful. For patients with tumors with poor prognoses, the tumor biomarker JOURNAL OF CLINICAL ONCOLOGY E D I T O R I A L S