AMultigeneAssay Is Prognostic of Survival in Patients with Early-Stage Lung Adenocarcinoma

Purpose: Clinical staging does not adequately risk stratify patients with early stage non ^ small cell lung cancer.We sought to generate a real-time PCR (RT-PCR)^ based prognostic model in patients with early stage lung adenocarcinoma, the dominant histology of lung cancer in the United States. Experimental Design:We studied gene expression of 61candidate genes in 107 patients with completely surgically resected lung adenocarcinoma using RT-PCR.We used crossvalidation methods to select and validate a prognostic model based on the expression of a limited number of genes. A risk score was generated based on model coefficients, and survival of patients with highand low-risk scores were analyzed. Results:We generated a four-gene model based on expression ofWNT3a, ERBB3, LCK , and RND3. Risk score predicted mortality better than clinical stage or tumor size (adjusted hazard ratio, 6.7; 95% confidence interval,1.6-28.9; P = 0.001). Among 70 patients with stage I disease, 5-year overall survival was 87% among patients with low-risk scores, and 38% among patients withhigh-risk scores (P = 0.0002). Among all patients, 5-year overall survival was 62% and 41%, respectively (P = 0.0054). Disease-free survival was also significantly different among lowand high-risk score patients. Conclusions:This multigene assay predicts overall and disease-free survival significantly better than clinical stage and tumor size in patients with early stage lung adenocarcinoma and performs especially well in patients with stage I disease. Prospective clinical trials are needed to determine whether high-risk patients with stage I disease benefit from adjuvant chemotherapy. Lung cancer is the most common cause of death in the United States and worldwide (1). Despite recent advances, long-term survival remains poor, with no >15% of patients still alive at 5 years after diagnosis. Despite improved understanding of the molecular biology of lung cancer, treatment decisions continue to be guided largely by clinical stage, designated as stages I through IV. Even in the most curable subset of non–small cell lung cancer (NSCLC), stage I disease, 5-year survival averages no >70% (2, 3). Although recent randomized clinical trials have shown improved survival with the use of postoperative adjuvant chemotherapy in stages II and III NSCLC, they have failed to detect a benefit for adjuvant chemotherapy for patients with stage I disease (4, 5). However, ongoing efforts to identify prognostic genomic biomarkers for risk stratification and predictive markers of therapeutic benefit offer considerable promise to alter current decision making paradigms. A number of prognostic biomarkers have been proposed for NSCLC, but none has yet been successfully translated into clinical application (6, 7). Although several genome-wide expression microarray-based prognostic models of lung cancer have been reported, such array-based technology may be suboptimal for clinical use due to the need for specialized laboratory facilities and complex statistical analyses (8–12). Prognostic models based on gene expression of a limited number of genes using quantitative real-time PCR (RT-PCR) may be more clinically practical. RT-PCR, considered the gold standard for measurement of gene expression, is highly reproducible and is relatively simple to analyze (13). Two recently reported RT-PCR–based prognostic models of lung cancer were able to risk stratify patients with resected NSCLC, although it is unclear whether they predicted outcome better than clinical factors, such as tumor size, and whether prognostic information differed by NSCLC histology (14, 15). We set out to generate a prognostic risk score for survival in patients with completely resected lung adenocarcinoma based on genes previously identified in microarray models of prognosis in NSCLC. As different histologic subtypes of NSCLC Imaging, Diagnosis, Prognosis Authors’Affiliations: Thoracic Oncology Program, Department of Surgery and Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California; Catalan Institute of Oncology, Hospital GermansTrias I Pujol, Barcelona, Spain; Medical University of Gdansk, Gdansk, Poland; and Divison of Hematology/Oncology, University of California Davis Cancer Center, Sacramento, California Received 3/3/08; revised 4/22/08; accepted 5/8/08. Grant support: Private foundation (anonymous donor). The costs of publication of this article were defrayed in part by the payment of page charges.This article must therefore be hereby marked advertisement in accordance with18 U.S.C. Section1734 solely to indicate this fact. Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/). Requests for reprints: Dan J. Raz, University of California, San Francisco, 513 Parnassus Avenue S-321, San Francisco, CA 94131. Phone: 415-476-1239; Fax: 415-353-9530; E-mail: Dan.raz@ucsf.edu. F2008 American Association for Cancer Research. doi:10.1158/1078-0432.CCR-08-0544 www.aacrjournals.org Clin Cancer Res 2008;14(17) September1, 2008 5565 Research. on May 2, 2017. © 2008 American Association for Cancer clincancerres.aacrjournals.org Downloaded from are known to have different patterns of gene expression, we chose to restrict our study to patients with adenocarcinoma, the dominant histology of NSCLC in the United States. We planned to separately analyze patients with stage I disease, as we hypothesized that risk among these patients would be stratified best by molecular modeling compared with more advanced stage disease. Moreover, patients with stage I disease who are at high risk for recurrence and death may benefit from adjuvant therapies. Finally, we also sought to compare our model to the five-gene prognostic model recently reported by Chen and colleagues (14). Materials andMethods Patients. Between January 1997 and June 2004, 120 patients who had undergone complete surgical resection of lung adenocarcinoma without preoperative chemotherapy or radiation treatment at the University of California San Francisco, and had fresh-frozen tissue banked for genomic analysis, were entered into the study. Eligible patients had undergone surgical resection with curative intent and had adequate mediastinal lymph node staging. Thirteen patients were excluded due to insufficient banked tissue, inadequate RNA quality, or weak expression of housekeeping genes. We did not perform a formal power calculation for this study as our sample size was limited by the availability of banked tissue at our institution and our inclusion criteria, the available sample size was on par with prior molecular marker prognostic studies. Information on clinical variables and patient follow-up were obtained from a prospectively maintained database including all subjects with banked tissue in the study. The primary end point was overall survival (OS). Vital status and date of death was determined by querying the Social Security Death Index using the subject’s social security number (available online at Social Security Death Index). Disease-free survival was defined as the time from surgery until radiographic evidence of recurrent disease or time until the last documented physician follow-up visit in the absence of recurrent disease. Patients consented to tissue specimen collection prospectively, and the study was approved by the University of California, San Francisco, institutional review board (CHR# H8714-28880-01). Gene selection. Of 217 genes identified from previously published microarray and PCR-based studies of prognosis in early stage lung cancer (8, 10, 12, 16–18), 76 cancer-related genes were identified by study investigators or by literature review. Fifteen genes were excluded due to nonfunctioning primers or very weak expression in tumor tissue by RT-PCR in a pilot study. The remaining 61 genes are listed in Supplementary table. Sample preparation and analysis. All tissue was frozen in liquid nitrogen at the time of the operation. Tissue was macrodissected into 1 cm sections that was ground in liquid nitrogen. RNA was extracted using a TriZol (Invitrogen) extraction protocol. Taqman RT-PCR was done on cDNA in 384-well plates using Prism 7900HT machine (Applied Biosystems). The expression of each gene was assayed in triplicate. Samples were compared with commercially available pooled normal lung RNA (Clontech), and normalized to 18S ribosomal rRNA (Applied Biosystems). Detailed methods on sample preparation and analysis are described in Supplementary methods. Statistical Analysis. The salient features of our data structure are (a) a right censored survival end point (death) with modest event numbers (47—after subsetting to exclude missings); (b) a multitude (61) of predictors as constituted by the (log) gene expression values obtained from real-time quantitative PCR (y-y Cts described elsewhere); and (c) select clinical and demographic covariates. In view of these features and dimensions, the primary data analytic tool we used was L1 penalized Cox proportional hazards regression (19). This methodology extends the simultaneous coefficient shrinkage and predictor selection that is inherent in L1 penalization (20), where it has proven highly effective (21, 22), to the survival data setting. Earlier extensions, largely motivated by microarray gene expression applications, were either computationally prohibitive (23) or reliant on approximation (24). The tradeoff between model fit, as captured via the Cox partial likelihood, and model complexity [number of included predictors (genes)], as captured via the L1 penalty, is governed by a tuning parameter that weights the contribution of the penalty. Determination of the value of this parameter makes recourse to crossvalidation, as is detailed elsewhere (19, 24). Briefly, the tuning parameter value that achieves optimal fit, as evaluated over unseen data furnished by the crossvalidatio

[1]  David E. Misek,et al.  Gene-expression profiles predict survival of patients with lung adenocarcinoma , 2002, Nature Medicine.

[2]  Yi Zhang,et al.  Gene expression signatures for predicting prognosis of squamous cell and adenocarcinomas of the lung. , 2006, Cancer research.

[3]  R. Tibshirani,et al.  Least angle regression , 2004, math/0406456.

[4]  L. Dominioni,et al.  Stage I nonsmall cell lung carcinoma , 2000, Cancer.

[5]  M. Tyers,et al.  Molecular profiling of non-small cell lung cancer and correlation with disease-free survival. , 2002, Cancer research.

[6]  K. Torgersen,et al.  Lck is required for stromal cell-derived factor 1 alpha (CXCL12)-induced lymphoid cell chemotaxis. , 2002, Blood.

[7]  L. Leslie,et al.  A decade of progress. , 1969, Minnesota medicine.

[8]  D. Botstein,et al.  Diversity of gene expression in adenocarcinoma of the lung , 2001, Proceedings of the National Academy of Sciences of the United States of America.

[9]  L. Tanoue,et al.  Three-Gene Prognostic Classifier for Early-Stage Non–Small-Cell Lung Cancer , 2009 .

[10]  Bengt Bergman,et al.  Cisplatin-based adjuvant chemotherapy in patients with completely resected non-small-cell lung cancer. , 2004, The New England journal of medicine.

[11]  E. Lander,et al.  Classification of human lung carcinomas by mRNA expression profiling reveals distinct adenocarcinoma subclasses , 2001, Proceedings of the National Academy of Sciences of the United States of America.

[12]  D. Harpole,et al.  Stage I nonsmall cell lung cancer. A multivariate analysis of treatment methods and patterns of recurrence , 1995, Cancer.

[13]  L. Liotta,et al.  Laser capture microdissection and microarray expression analysis of lung adenocarcinoma reveals tobacco smoking- and prognosis-related molecular profiles. , 2002, Cancer research.

[14]  Jiang Gui,et al.  Penalized Cox regression analysis in the high-dimensional and low-sample size settings, with applications to microarray gene expression data , 2005, Bioinform..

[15]  Alexandre Pelzer,et al.  Small G‐protein RhoE is underexpressed in prostate cancer and induces cell cycle arrest and apoptosis , 2005, The Prostate.

[16]  M. D. den Bakker,et al.  A critical appraisal of prognostic and predictive factors for common lung cancers , 2006, Histopathology.

[17]  Lu Zhang,et al.  Large scale real-time PCR validation on gene expression measurements from two commercial long-oligonucleotide microarrays , 2006, BMC Genomics.

[18]  S. Mukherjee,et al.  A genomic strategy to refine prognosis in early-stage non-small-cell lung cancer. , 2006, The New England journal of medicine.

[19]  N. Hanna,et al.  Vinorelbine Plus Cisplatin vs Observation in Resected Non–Small-Cell Lung Cancer , 2006 .

[20]  F. Harrell,et al.  Prognostic/Clinical Prediction Models: Multivariable Prognostic Models: Issues in Developing Models, Evaluating Assumptions and Adequacy, and Measuring and Reducing Errors , 2005 .

[21]  R. Tibshirani Regression Shrinkage and Selection via the Lasso , 1996 .

[22]  Biao He,et al.  Wnt signaling in lung cancer. , 2005, Cancer letters.

[23]  M. You,et al.  Differential expression of critical cellular genes in human lung adenocarcinomas and squamous cell carcinomas in comparison to normal lung tissues. , 2002, Neoplasia.

[24]  A. Jemal,et al.  Cancer Statistics, 2006 , 2006, CA: a cancer journal for clinicians.

[25]  O. Stoeltzing,et al.  Multiple molecular marker testing (p53, C-Ki-ras, c-erbB-2) improves estimation of prognosis in potentially curative resected non-small cell lung cancer , 2000, British Journal of Cancer.

[26]  Michael Thomas,et al.  Identification of metastasis-associated receptor tyrosine kinases in non-small cell lung cancer. , 2005, Cancer research.