Skip Navigation LinksHome > January 2014 - Volume 9 - Issue 1 > Validation of a Histology-Independent Prognostic Gene Signat...
Journal of Thoracic Oncology:
doi: 10.1097/JTO.0000000000000042
Original Articles

Validation of a Histology-Independent Prognostic Gene Signature for Early-Stage, Non–Small-Cell Lung Cancer Including Stage IA Patients

Der, Sandy D. PhD*; Sykes, Jenna MMath*; Pintilie, Melania MSc*; Zhu, Chang-Qi PhD*; Strumpf, Dan PhD*; Liu, Ni MSc*; Jurisica, Igor PhD*†‡; Shepherd, Frances A. MD; Tsao, Ming-Sound MD*‡‖

Free Access
Article Outline
Collapse Box

Author Information

*Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada; and Departments of Computer Science, Medical Biophysics, §Medicine, and Laboratory Medicine and Pathobiology, University of Toronto, Toronto, Ontario, Canada.

Disclosure: M.-S. Tsao is the M. Qasim Choksi Chair in Lung Cancer Translational Research; F.A. Shepherd is the Scott Taylor Chair in Lung Cancer Research. Dr. I. Jurisica is the Canada Research Chair in Integrative Cancer Informatics. The other authors declare no conflict of interest.

Address for correspondence: Drs. Sandy D. Der and Ming-Sound Tsao, Princess Margaret Cancer Centre, 610 University Avenue, Toronto, Ontario, Canada M5G 2M9. E-mail:;

Collapse Box


Background: Patients with early-stage non–small-cell lung carcinoma (NSCLC) may benefit from treatments based on more accurate prognosis. A 15-gene prognostic classifier for NSCLC was identified from mRNA expression profiling of tumor samples from the NCIC CTG JBR.10 trial. In this study, we assessed its value in an independent set of cases.

Methods: Expression profiling was performed on RNA from frozen, resected tumor tissues corresponding to 181 stage I and II NSCLC cases collected at University Health Network (UHN181). Kaplan–Meier methodology was used to estimate 5-year overall survival probabilities, and the prognostic effect of the classifier was assessed using the log-rank test. A Cox proportional hazards model evaluated the signature’s effect adjusting for clinical prognostic factors.

Results: Expression data of the 15-gene classifier stratified UHN181 cases into high- and low-risk subgroups with significantly different overall survival (hazard ratio [HR] = 1.92; 95% confidence interval [CI], 1.15–3.23; p = 0.012). In a subgroup analysis, this classifier predicted survival in 127 stage I patients (HR = 2.17; 95% CI, 1.12–4.20; p = 0.018) and the smaller subgroup of 48 stage IA patients (HR = 5.61; 95% CI, 1.19–26.45; p = 0.014). The signature was prognostic for both adenocarcinoma and squamous cell carcinoma cases (HR = 1.76, p = 0.058; HR = 4.19, p = 0.045, respectively).

Conclusion: The prognostic accuracy of a 15-gene classifier was validated in an independent cohort of 181 early-stage NSCLC samples including stage IA cases and in different NSCLC histologic subtypes.

Lung cancer is the leading cause of cancer death and non–small-cell lung carcinoma (NSCLC) accounts for approximately 85% of all cases.1 The majority of NSCLC patients are diagnosed in advanced or metastatic stages, which largely are inoperable.2 Early stage I and II NSCLC cases are potentially curable by complete surgical resection,3 and survival can be improved with adjuvant chemotherapy, mainly in stage II4–6 and possibly in stage I patients with 4 cm diameter or higher.7–9 Investigators have continued to seek prognostic markers and markers that are predictive of survival benefit from adjuvant chemotherapy, as it is recognized that these could form the basis for developing personalized approaches to improve the survival of early-stage NSCLC patients.10,11 Currently, stage II patients are treated with adjuvant chemotherapy; although the benefit to the group as a whole has been established,12–14 there are patients with inherently good prognosis who potentially could be spared the morbidity associated with adjuvant chemotherapy. Stage I patients could benefit the most from a strong prognostic marker, as the survival of patients with poor prognosis could potentially be improved by adjuvant chemotherapy.

Our group had developed an mRNA-based classifier comprising 15 genes (Table 4)15 by expression profiling snap-frozen, tumor samples collected in the NCIC CTG JBR.10 trial.14 The accuracy of this classifier as a prognostic marker was established initially by in silico validation in four large public NSCLC expression datasets. In this study, we further tested the performance of this 15-gene classifier in an independent cohort of 181 snap-frozen samples from early-stage NSCLC patients collected at University Health Network (UHN) who did not receive any adjuvant therapy. As a validation study, the same assay platform (Affymetrix) was used here as in the discovery study.15

Table 4
Table 4
Image Tools
Back to Top | Article Outline


NSCLC Tumor Samples and Patients

Snap-frozen tumor samples from NSCLC patients who underwent surgical resection at the UHN during 1996–2005 were retrieved from the UHN tumor bank, using a protocol approved by the UHN Research Ethics Board. Patient tumor stage and histological classification were obtained from electronic patient records. On the basis of the tumor size information, some tumor stages were adjusted accordingly to the 7th edition tumor, node, metastasis (TNM) criteria. Exclusion criteria included patients with stage III or IV disease or those received any form of adjuvant therapy and samples with tumor cellularity of less than 20%. Histology and tumor cellularity were assessed by review of the representative H&E section of frozen tumor samples used for RNA extraction. During this assessment, the tumor histological type was recorded. Cases with significant discrepancy as compared with the original diagnosis were reviewed further to provide the final histological diagnosis. Approximately 210 cases meeting these criteria were identified initially and total RNA was extracted as described previously.15 In brief, frozen tissue fragments were homogenized and total RNA was isolated using the RNAZol (Invitrogen, Carlsbad, CA). Cases with low RNA yield or poor RNA integrity, as defined by RIN scores less than 8.0 by BioAnalyzer (Agilent Technologies, Santa Clara, CA), were excluded. A total of 181 RNA samples were processed by expression profiling using Affymetrix U133 2.0 Plus arrays (Princess Margaret Genomics Centre, Toronto, Ontario, Canada).

Back to Top | Article Outline
Data Analysis

Microarray data were normalized using robust multiarray averaging and deposited with GEO (GSE50081). The expression values for the 15 genes comprising the classifier were extracted using methods described previously (Table 4),15 and prognostic accuracy was tested in the entire set of 181 samples, and by stage and histologic subtype. As outlined previously,15 a risk score was calculated using a weighted sum of four principal components derived from the expression values of the 15 probe sets. This risk score was then dichotomized at the median of −0.1 with high-risk patients being those patients whose risk score was greater than this value and low risk otherwise. Overall survival up to 5 years was used as the endpoint of interest. Patients with more than 5 years follow-up were censored at 5 years as deaths occurring later than 5 years were not likely to be lung cancer related. The survival probabilities were calculated using the Kaplan–Meier method. Differences in survival curves were assessed using the log-rank test. Hazard ratios (HRs) and 95% confidence intervals (CI) were generated using a Cox proportional hazards model. Multivariable analyses adjusted the model for clinical factors such as age, sex, stage, and histology. Schoenfeld residuals for all models were assessed to ensure the proportionality assumption was met. All analyses were performed using the survival package (version 2.36–5) in the open-source software R version 2.12.16 A two-sided p value of 0.05 was used to assess statistical significance.

Back to Top | Article Outline


Patient Cohorts

The primary objective of this study was to validate the prognostic accuracy of the 15-gene classifier in a larger cohort of early-stage NSCLC cases than in our previous study (Table 1).15 Although the JBR.10 cohort included only stage IB and stage II patients, the UHN181 cohort included all stage I and stage II patients, and specifically, 48 stage IA patients. As a result, the ratio of stage I:II in UHN181 (70:30) was higher than in the JBR.10 cohort (55:45). Sex was biased toward males in both UHN181 (54%) and JBR.10 (71%). The most prevalent NSCLC histology in both cohorts was adenocarcinoma, 71 and 52%, respectively, whereas the representation of (SqCC) was 24 and 42%. Sixty-three deaths occurred within 5 years. Of the 118 patients alive, 81 patients (69%) had complete follow-up to 5 years and 15 patients (13%) had follow-up less than 3 years.

Table 1
Table 1
Image Tools
Back to Top | Article Outline
Prognostic Performance of the JBR.10 Signature

Consistent with early-stage NSCLC patients, stage was a prognostic marker in the UHN181 patients although it was not statistically significant in this cohort (Fig. 1A). In comparison, the 15-gene classifier when applied to the entire UHN181 cohort was able to classify patients into low- and high-risk survival groups with statistical significance on both univariable and multivariable analyses (Fig. 1B and Table 2, multivariable-adjusted HR = 1.95; 95% CI, 1.15–3.30; Wald p = 0.013). This result was comparable with the range of HR values observed in our previous validation study15 using four published NSCLC microarray datasets as external validation (HR = 2.26, 2.27, 1.96, and 3.57 for datasets corresponding to Director’s Challenge,17 Netherlands Cancer Institute,18 Duke University,19 and University of Michigan SqCC,20 respectively). The prognostic strength of this classifier was independent of histology (Fig. 2, A–D).

Figure 1
Figure 1
Image Tools
Table 2
Table 2
Image Tools
Figure 2
Figure 2
Image Tools

The 15-gene classifier originally was trained and validated on only stage IB and stage II patients together, and thus its performance for stage IA patients was unknown. When tested in the 48 stage 1A patient subgroup of the UHN181 cohort, the signature classified 27 patients as low-risk group with a 92% 5-year survival rate, and 21 as high-risk with a 5-year survival rate of 61% (HR = 5.61; 95% CI, 1.19–26.45; Wald p = 0.014) (Fig. 2E and Table 3). However, the ability to predict high- or low-risk groups among stage IB patients alone did not achieve statistical significance (HR = 1.43; 95% CI, 0.68–2.99; Wald p = 0.34; Fig. 2F).

Table 3
Table 3
Image Tools
Back to Top | Article Outline


This study was designed to test prospectively in a blinded manner, the prognostic value of the 15-gene signature in a new cohort of banked tumor samples representing complete resections from stage I and II NSCLC patients who received no adjuvant therapy. As a validation study, we chose to use the same type of microarray platform but with the sample processing and hybridization performed in an independent laboratory and more than 5 years after the original study. The results confirmed the prognostic value of the 15-gene signature for stage I and II patients together and showed that its prognostic value was independent of histology. We also have demonstrated its ability to identity high-risk stage I patients. The signature was unable to predict survival in stage IB alone, or combined stage IB and II patients, which we presently can only attribute to cohort variability. However, this study reveals that the signature was able to stratify stage IA patient subgroups with significantly different 5-year survival outcomes (92% for low risk versus 61% for high risk). This may have considerable clinical relevance since at this time stage IA patients are not offered adjuvant chemotherapy.

The 15-gene expression prognostic signature was developed from the microarray analysis of snap-frozen tumor samples collected from patients who participated in the JBR.10 adjuvant chemotherapy trial.14 This was one of the pivotal trials that established adjuvant chemotherapy as beneficial in improving the survival of early-stage NSCLC patients. Among 482 patients who were randomized, 169 patients had snap-frozen tumor samples collected prospectively and banked as part of the trial protocol. Expression microarray profiling was conducted in samples of 133 patients, 62 in the observation arm, and 71 in the adjuvant chemotherapy arm. The prognostic gene signature was identified in the observation (surgery only) cohort, using the Maximizing R Square (MARSA) algorithm.15 The signature was able to classify the 62 patients into 31 low-risk and 31 high-risk individuals, with significantly different survival outcomes (adjusted HR = 18.00; 95% CI, 5.78–56.05; p = 0.001). The prognostic signature validated separately in four independent microarray data sets (a total of 356 stage-IB/II NSCLC patients who had not receive adjuvant treatment), and in an additional 19 JBR.10 observation only patients by real-time quantitative polymerase chain reaction. Most importantly, the signature also seemed to be predictive of improved survival after adjuvant chemotherapy in JBR.10 high-risk patients (HR chemotherapy versus observation, 0.33; 95% CI, 0.17–0.63; p = 0.0005), but not in low-risk patients where chemotherapy seemed to be detrimental (HR = 3.67; 95% CI, 1.22–11.06; p = 0.0133; interaction p = 0.001). However, this could not be validated as microarray datasets from randomized studies with treated and untreated control arms were not available. Nevertheless, Tang et al.21 has recently identified a 12-gene signature that was predictive of responsiveness to chemotherapy in two patient cohorts, the JBR.1015 and the UT Lung SPORE datasets, and made public their dataset. We have determined that the 15-gene signature could also predict response to ACT in the UT Lung SPORE dataset, particularly for stage I patients, although interaction of risk group and ACT on survival was not significant, for stage I alone, or combined with stage II (Supplementary Figure S1, Supplementary Digital Content, This also affirms the concept that there can be multiple classifiers that predict clinical outcomes from genomic datasets despite being inherently different in their gene composition.22 To our best knowledge, no study has yet validated the impact of applying a prognostic and/or predictive classifier to guide treatment decisions for NSCLC patients with regard to receiving adjuvant therapy and evaluate resultant clinical outcomes. Although an accurate classifier that predicts responsiveness to therapy ideally should provide the best guidance to help individualize treatment for patients, it remains possible that a test enabling better prognosis alone and partnered with a specific treatment, whether adjuvant chemotherapy or some other treatment, potentially also could provide clinical utility.

More than 30 prognostic gene signatures have been reported by various investigators,10 but to our knowledge, only one signature23 has become available as a clinical test (Pervenio RS test, Life Technologies, Inc., Grand Island, NY). This 14-gene signature was developed using real-time quantitative polymerase chain reaction directly on RNA isolated from FFPE tumor samples of 361 stage I–IV nonsquamous NSCLC patients resected at the University of California at San Francisco, with validation in two independent cohorts of nonsquamous NSCLC patients: 433 patients with stage I (285 stage IA and 135 stage IB) nonsquamous NSCLC resected in a U.S. institution, and 1006 patients with stage I–III nonsquamous NSCLC resected in multiple institutions in China. In the U.S. validation cohort, the adjusted HR for the signature by Kratz et al. was 2.04 (95% CI, 1.28–3.26; p = 0.0016), which is similar to the HR achieved by the 15-gene signature studied here. However, the Kratz et al.23 signature did not demonstrate ability to predict for adjuvant chemotherapy benefit, as similar HRs were generated in both training cohorts of patients who received or did not receive adjuvant chemotherapy. In contrast, the 15-gene signature demonstrated potential for predicting responsiveness to chemotherapy15 although these aspects have not been validated yet in an independent cohort. Finally, although the signature by Kratz et al. was developed specifically for nonsquamous NSCLC, the 15-gene signature can identify prognostic subgroups for both SqCC and adenocarcinoma patients. Therefore, the 15-gene signature seems to have the most promising potential among all published signatures to identify patients who may or may not benefit from adjuvant chemotherapy among the broad spectrum of early-stage NSCLC cases, independent of stage and histology.

Back to Top | Article Outline


Research funding for this study was provided by Med BioGene Inc. and research grants from the Canadian Institutes of Health Research (CIHR) Proof of Principle Program (grant No. 86173), the Canadian Cancer Society Research Institute (grant No. 020527), partially by the Ontario Ministry of Health and Long Term Care, and the Princess Margaret Hospital Foundation.

Back to Top | Article Outline


1. Jemal A, Siegel R, Xu J, Ward E. Cancer statistics, 2010. CA Cancer J Clin. 2010;60:277–300

2. Goldstraw P, Crowley J, Chansky K, et al.International Association for the Study of Lung Cancer International Staging Committee; Participating Institutions. The IASLC Lung Cancer Staging Project: proposals for the revision of the TNM stage groupings in the forthcoming (seventh) edition of the TNM Classification of malignant tumours. J Thorac Oncol. 2007;2:706–714

3. Kelsey CR, Marks LB, Hollis D, et al. Local recurrence after surgery for early stage lung cancer: an 11-year experience with 975 patients. Cancer. 2009;115:5218–5227

4. Arriagada R, Bergman B, Dunant A, Le Chevalier T, Pignon JP, Vansteenkiste JInternational Adjuvant Lung Cancer Trial Collaborative Group. . Cisplatin-based adjuvant chemotherapy in patients with completely resected non-small-cell lung cancer. N Engl J Med. 2004;350:351–360

5. Douillard JY, Rosell R, De Lena M, et al. Adjuvant vinorelbine plus cisplatin versus observation in patients with completely resected stage IB-IIIA non-small-cell lung cancer (Adjuvant Navelbine International Trialist Association [ANITA]): a randomised controlled trial. Lancet Oncol. 2006;7:719–727

6. Scagliotti GV, Parikh P, von Pawel J, et al. Phase III study comparing cisplatin plus gemcitabine with cisplatin plus pemetrexed in chemotherapy-naive patients with advanced-stage non-small-cell lung cancer. J Clin Oncol. 2008;26:3543–3551

7. Pignon JP, Tribodet H, Scagliotti GV, et al.LACE Collaborative Group. Lung adjuvant cisplatin evaluation: a pooled analysis by the LACE Collaborative Group. J Clin Oncol. 2008;26:3552–3559

8. Strauss GM, Herndon JE 2nd, Maddaus MA, et al. Adjuvant paclitaxel plus carboplatin compared with observation in stage IB non-small-cell lung cancer: CALGB 9633 with the Cancer and Leukemia Group B, Radiation Therapy Oncology Group, and North Central Cancer Treatment Group Study Groups. J Clin Oncol. 2008;26:5043–5051

9. Butts CA, Ding K, Seymour L, et al. Randomized phase III trial of vinorelbine plus cisplatin compared with observation in completely resected stage IB and II non-small-cell lung cancer: updated survival analysis of JBR-10. J Clin Oncol. 2010;28:29–34

10. Zhu CQ, Pintilie M, John T, et al. Understanding prognostic gene expression signatures in lung cancer. Clin Lung Cancer. 2009;10:331–340

11. Subramanian J, Simon R. Gene expression-based prognostic signatures in lung cancer: ready for clinical use? J Natl Cancer Inst. 2010;102:464–474

12. Pisters KM, Evans WK, Azzoli CG, et al.Cancer Care Ontario; American Society of Clinical Oncology. Cancer Care Ontario and American Society of Clinical Oncology adjuvant chemotherapy and adjuvant radiation therapy for stages I-IIIA resectable non small-cell lung cancer guideline. J Clin Oncol. 2007;25:5506–5518

13. Visbal AL, Leighl NB, Feld R, Shepherd FA. Adjuvant chemotherapy for early-stage non-small cell lung cancer. Chest. 2005;128:2933–2943

14. Winton T, Livingston R, Johnson D, et al.National Cancer Institute of Canada Clinical Trials Group; National Cancer Institute of the United States Intergroup JBR.10 Trial Investigators. Vinorelbine plus cisplatin vs. observation in resected non-small-cell lung cancer. N Engl J Med. 2005;352:2589–2597

15. Zhu CQ, Ding K, Strumpf D, et al. Prognostic and predictive gene signature for adjuvant chemotherapy in resected non-small-cell lung cancer. J Clin Oncol. 2010;28:4417–4424

16. R Development Core Team (2010). R: A Language and Environment for Statistical Computing. Vienna, Austria R Foundation for Statistical Computing Available at:

17. Shedden K, Taylor JM, Enkemann SA, et al. Gene expression-based survival prediction in lung adenocarcinoma: a multi-site, blinded validation study. Nat Med. 2008;14:822–827

18. Roepman P, Jassem J, Smit EF, et al. An immune response enriched 72-gene prognostic profile for early-stage non-small-cell lung cancer. Clin Cancer Res. 2009;15:284–290

19. Potti A, Mukherjee S, Petersen R, et al. A genomic strategy to refine prognosis in early-stage non-small-cell lung cancer. N Engl J Med. 2006;355:570–580

20. Raponi M, Zhang Y, Yu J, et al. Gene expression signatures for predicting prognosis of squamous cell and adenocarcinomas of the lung. Cancer Res. 2006;66:7466–7472

21. Tang H, Xiao G, Behrens C, et al. A 12-gene set predicts survival benefits from adjuvant chemotherapy in non-small cell lung cancer patients. Clin Cancer Res. 2013;19:1577–1586

22. Boutros PC, Lau SK, Pintilie M, et al. Prognostic gene signatures for non-small-cell lung cancer. Proc Natl Acad Sci U S A. 2009;106:2824–2828

23. Kratz JR, He J, Van Den Eeden SK, et al. A practical molecular assay to predict survival in resected non-squamous, non-small-cell lung cancer: development and international validation studies. Lancet. 2012;379:823–832

Non–small-cell lung carcinoma; Prognosis; Biomarkers; Adenocarcinoma; Squamous cell carcinoma

Supplemental Digital Content

Back to Top | Article Outline

Copyright © 2014 by the European Lung Cancer Conference and the International Association for the Study of Lung Cancer.


Article Tools



Search for Similar Articles
You may search for similar articles that contain these same keywords or you may modify the keyword list to augment your search.

Other Ways to Connect



Visit on your smartphone. Scan this code (QR reader app required) with your phone and be taken directly to the site.

 For additional oncology content, visit LWW Oncology Journals on Facebook.