Purpose: Adjuvant treatment may improve survival in early-stage squamous cell carcinoma (SCC) of the lung; however, the absolute gain is modest and mainly limited to stage II-IIIA. Current staging methods are imprecise indications of prognosis, but high-risk patients can be identified by gene expression profiling and considered for adjuvant therapy.

Experimental Design: The expression of 29 genes was assessed by reverse transcriptase quantitative PCR in frozen primary tumor specimens obtained from 66 SCC patients who had undergone surgical resection. Expression values were dichotomized using the median as a cutoff value. We used a risk score to develop a gene expression model for the prediction of survival.

Results: The univariate analysis of gene expression in the training cohort identified 10 genes with significant prognostic value: CSF1, EGFR, CA IX, PH4, KIAA0974, ANLN, VEGFC, NTRK1, FN1, and INR1. In the multivariate Cox model, CSF1 (hazard ratio, 3.5; P = 0.005), EGFR (hazard ratio, 2.7; P = 0.02), CA IX (hazard ratio, 0.2; P < 0.0001), and tumor size >4 cm (hazard ratio, 2.7; P = 0.02) emerged as significant markers for survival. The high prognostic value of a risk score based on the expression of the three genes (CSF1, EGFR, and CA IX) was positively validated in a separate cohort of 26 patients in an independent laboratory (P = 0.05).

Conclusions: The three-gene signature is strongly associated with prognosis in early-stage SCC. Positive independent validation suggests its suitability for selecting SCC patients with an increased risk of death who might benefit from adjuvant treatment.

Approximately 40% of stage I and 60% of stage II non–small-cell lung cancer (NSCLC) patients will die within 5 years after curative pulmonary resection, mainly due to the development of distant metastases (1). These patients may well derive a notable benefit from adjuvant chemotherapy. Whereas at present, there are no reliable clinical predictors of relapse after surgery in early-stage NSCLC, transcriptional analysis of primary tumors has identified gene expression profiles strongly related to disease recurrence in adenocarcinoma (28) and, to a lesser extent, in squamous cell carcinoma (SCC; refs. 58).

A meta-analysis (7) of data sets from seven microarray studies (2, 4, 9, 10) identified a 64-gene-expression signature that predicted survival with 85% accuracy; however, only 52 SCCs were included in these analyses (9, 10). A recent study including 51 stage I-III lung SCCs identified a 111-gene signature with a 72% predictive accuracy for disease recurrence (11). However, the lack of commonality of many genes identified between the published prognostic signatures may well reflect the fact that numerous gene expression signatures may be capable of predicting outcome in NSCLC (12). Reverse transcriptase real-time quantitative PCR (RT-QPCR) also allows for accurate and reproducible RNA quantification. The expression pattern of eight genes determined by RT-QPCR correlated with survival in lung adenocarcinoma (13). Furthermore, a five-gene signature, including genes engaged in tumor-macrophage interaction related to tumor progression, was predictive for survival in adenocarcinoma and SCC of the lung (14).

Increasing evidence suggests that SCC and adenocarcinoma of the lung are molecularly two separate entities (15). These two histologic NSCLC subtypes differ in expression levels of 33 receptor tyrosine kinases. For example, adenocarcinomas express higher levels of ERBB2 and ERBB3, whereas EPHB6 levels are higher in SCCs, and neurotrophic tyrosine receptor kinase 2 is absent in adenocarcinomas but present in SCCs (16). In addition, the expression of eight genes involved in DNA repair and metastasis was higher in SCCs than in adenocarcinomas (17).

To construct a reliable prognostic gene signature that could identify lung SCC patients with a high risk of death, we examined, by RT-QPCR, the expression of 29 genes related to disease progression and metastasis. We included genes with high discriminatory power (PH4, CD47 KIAA0794; ref. 18) and those concordantly up- or down-regulated (VEGF C, SERPIN1, Rho GDIβ, STX1A; ref. 12) in microarray studies. EGFR, NTRK1, and INR1 were also included based on the fact that the hazard risk for metastasis development was increased in early-stage NSCLC tumors with overexpression of these receptor tyrosine kinase genes by RT-PCR (16). The rationale for other genes included in the analysis is included in the Supplementary Text S1.

Patients. The study comprised two patient cohorts. The training cohort included 66 patients with lung SCC who underwent curative pulmonary resection between 2000 and 2004 at the hospital of the University of Gdansk, Poland. Stage I-IIIA patients fulfilling the following criteria were enrolled: complete pulmonary resection, availability of snap-frozen tumor sample (tumor specimen containing at least 60% of tumor tissue in the section), and at least a 3-year follow-up for nonrelapsed patients. In all instances, complete mediastinal lymphadenectomy was done and the surgical margins were free of tumor (R0). None of the patients received chemotherapy. The median follow-up was 37 months. The validation cohort comprised 26 stage I-II lung SCC patients who underwent curative pulmonary resection between 1997 and 2004 at the hospital of the University of California San Francisco. None of the patients received chemotherapy. The median follow-up was 56 months. Patient characteristics for both cohorts are summarized in Table 1, and complete details are shown in Supplementary Table S1.

Table 1.

Patient characteristics

CharacteristicsGdansk, Poland training cohort (n = 66)San Francisco, CA, validation cohort (n = 26)
Age, y   
    Median (range) 63 (37-76) 67 (45-65) 
    Mean 62 66 
Sex, n (%)   
    Male 52 (78.8) 17 (65.4) 
    Female 14 (21.2) 9 (34.6) 
Race, n (%)   
    Caucasian 66 (100) 21 (80.8) 
    African-American — 3 (11.5) 
    Other — 2 (7.6) 
Stage, n (%)   
    IA 10 (15.2) 10 (38.5) 
    IB 32 (48.5) 12 (46.2) 
    IIA — 1 (3.8) 
    IIB 22 (33.3) 3 (11.5) 
    IIIA 2 (3) — 
Tumor diameter (cm) 5.9 (1-9.5) 2.5 (1.5-8) 
Predictive accuracy of the three-gene signature (%) 70 70 
CharacteristicsGdansk, Poland training cohort (n = 66)San Francisco, CA, validation cohort (n = 26)
Age, y   
    Median (range) 63 (37-76) 67 (45-65) 
    Mean 62 66 
Sex, n (%)   
    Male 52 (78.8) 17 (65.4) 
    Female 14 (21.2) 9 (34.6) 
Race, n (%)   
    Caucasian 66 (100) 21 (80.8) 
    African-American — 3 (11.5) 
    Other — 2 (7.6) 
Stage, n (%)   
    IA 10 (15.2) 10 (38.5) 
    IB 32 (48.5) 12 (46.2) 
    IIA — 1 (3.8) 
    IIB 22 (33.3) 3 (11.5) 
    IIIA 2 (3) — 
Tumor diameter (cm) 5.9 (1-9.5) 2.5 (1.5-8) 
Predictive accuracy of the three-gene signature (%) 70 70 

Gene expression analysis. Tumor samples were obtained during surgery as blocks of 1 cm3 and snap frozen in liquid nitrogen. Tissues were stored in −80°C until total RNA was extracted with AllPrep kits (Qiagen). Only tumor samples containing >60% of tumor cells on a microscopic section were eligible for further processing. The concentration of RNA was assessed in Nano-drop and the quality of obtained RNA was confirmed on agarose gel. cDNA was synthesized from 1 μg of total RNA using the High-Capacity cDNA Archive Kit (Applied Biosystems). Quantitative RT-PCR reactions of 29 genes (Supplementary Table S2) were run using Applied Biosystems TaqMan Low Density Arrays (microfluidic cards) in an ABI PRISM 7900 HT Sequence Detection System (Applied Biosystems). One channel of a microfluidic card was loaded with a mix of 55 μL TaqMan Universal PCR Master Mix (Applied Biosystems) and 55 μL of a cDNA template corresponding to 100 ng of total RNA.

Relative gene expression values were calculated by the ΔΔCt method (19) using the Sequence Detection System 2.1 software (Applied Biosystems). The ΔΔCt method gives the amount of target gene normalized to an endogenous reference gene and relative to a calibrator sample (reference for all samples). The raw gene expression values were normalized according to the expression of ribosomal 18S RNA. The normalized expression of each gene was calibrated by its expression in an RNA pool (made by pooling equal amounts of total RNA from each sample). Molecular analyses were carried out at the Medical Oncology Service, Catalan Institute of Oncology (Badalona, Spain), for the training cohort and at the University of California (San Francisco, CA) for the validation cohort.

Statistical analysis. We categorized SCC tumors as having good or poor prognosis based on the gene expression profile. The method consisted of five steps: (a) the gene expression was coded as 0 (low) or 1 (high), using the median expression value as a cutoff point; (b) a univariate Cox model with overall survival as the dependent variable was constructed and categorized with gene expression levels as independent variables; (c) the genes that were significant in the univariate analysis were included in a multivariate Cox proportional hazards model for survival; (d) the risk score for each patient was calculated; (e) the curves for overall survival were obtained with the Kaplan-Meier product limit method according to the median value of the risk score for the entire population. Data on patients were analyzed from the date of surgery to the time of relapse or death, or the date on which data were censored. Comparisons were made with the two-sided log-rank test. Patient characteristics were compared between the two risk groups using the Mann-Whitney test. The χ2 and Fisher's exact tests were used to compare categorical variables. The Spearman rho method was used to correlate expression levels of the genes examined. The Bonferroni method was used to adjust for multiple comparisons. All statistical analyses were carried out at a 5% level of significance and with a power of 80%, using the Statistical Package for the Social Sciences, version 13 (SPSS, Inc.).

The overall median survival for the 66 patients in the training cohort was 38.2 months [95% confidence interval (95% CI), 26.6-47 months]. When patients were divided by disease stage, median survival was 41.3 months for stage I patients and 33.6 months (95% CI, 24.9-42.2 months) for stage II patients. Primary tumor size was ≤4 cm in 27 patients and >4 cm in the remaining 39. Thirty-three patients developed distant metastases after surgery, and the remaining 33 patients were free of distant metastasis after a median follow-up of 37 months (range, 24-64 months). Seven of the 33 patients who developed metastasis also had a local relapse.

A strong correlation was observed between expression levels of 10 genes (CSF1, FN1, CA IX, PH4, KIAA0974, ANLN, VEGFC, ISNR, NTRK1, and EGFR); for example, high levels of PH4 were related to low or no expression of CA IX (ρ = −0.33; P = 0.007) and high levels of VEGF-C were related to high levels of CSF-1 (ρ = 0.64; P = 0.0001; Supplementary Table S3). The expression levels of these 10 genes significantly correlated with overall survival in the univariate Cox regression analysis (Table 2; Supplementary Figs. S1-S10). The 10 genes were included in a multivariate Cox regression analysis together with stage, differentiation, primary tumor size (≤4 cm versus >4 cm), sex, and age. The expression levels of three genes (CSF-1, EGFR, CA IX) and tumor size emerged as independent prognostic factors (Table 3).

Table 2.

Univariate analysis of overall survival according to gene expression in the training cohort

Gene expression Median survival (95% CI)P
CSF1  0.002 
    ≤0.90 NR  
    >0.90 27.5 (20.1-34.9)  
FN1  0.002 
    ≤0.57 NR  
    >0.57 31.3 (19.3-43.2)  
CA IX  0.007 
    ≤0.56 29 (20.9-37.2)  
    >0.56 NR  
PH4  0.01 
    ≤0.90 NR  
    >0.90 29.7 (21.9-37.5)  
KIAA0974  0.02 
    ≤0.71 NR  
    >0.71 21.3 (22.3-40.2)  
ANLN  0.02 
    ≤1.09 NR  
    >1.09 31.3 (24.4-38.2)  
ISNR  0.03 
    ≤0.57 NR  
    >0.57 33.2 (19.3-47.1)  
VEGFC  0.03 
    ≤0.67 NR  
    >0.67 33.2 (26.9-39.5)  
NTRK1  0.04 
    ≤0.53 NR  
    >0.53 33.2 (24.9-41.6)  
EGFR  0.05 
    ≤0.76 NR  
    >0.76 33.9 (23.7-44.2)  
SERPINE  0.08 
    ≤0.45 46.9  
    >0.45 33.2 (19.2-46.6)  
Ezrin  0.10 
    ≤0.81 NR  
    >0.81 33.2 (23.1-43.3)  
ARHGDIB  0.11 
    ≤0.76 41.3  
    >0.76 29.7 (19.8-39.5)  
Selectin P ligand  0.12 
    ≤0.71 46.9 (33.1-60.8)  
    >0.71 29.7 (20.2-39.2)  
CXCR4  0.18 
    ≤0.61 41.3  
    >0.61 31.3 (25.6-36.9)  
STX1A  0.18 
    ≤0.72 46.9  
    >0.72 33.6 (27.9-39.3)  
PDPN  0.19 
    ≤0.59 46.9  
    >0.59 33.2 (20.6-45.9)  
HGF  0.21 
    ≤0.51 46.9 (32.9-61.1)  
    >0.51 33.2 (26-40.5)  
Selectin L ligand  0.33 
    ≤0.59 41.3 (30.3-52.3)  
    >0.59 33.2 (23.1-43.4)  
PGK1  0.35 
    ≤0.90 33.6 (20.4-46.8)  
    >0.90 38.2 (28.8-47.5)  
IL8  0.86 
    ≤0.46 36.7  
    >0.46 38.2 (29.1-47.2)  
CD44  0.42 
    ≤1.62 33.6 (21.6-45.6)  
    >1.62 46.9 (31.9-62)  
CD47   
    ≤0.70 41.3 0.36 
    >0.70 33.9 (17-50.9)  
NTRK2  0.68 
    ≤0.32 41.3 (28.3-54.3)  
    >0.32 38.2  
N-cadherin  0.34 
    ≤0.34 46.9  
    >0.34 33.6 (27-40)  
TWIST1  0.69 
    ≤0.77 36.7 (25.8-47.6)  
    >0.77 46.9 (28.7-65.2)  
S100P  0.64 
    ≤0.21 33.6 (26-41.2)  
    >0.21 46.9 (34.9-59.1)  
GPI  0.52 
    ≤0.72 33.9 (22.4-45.5)  
    >0.72 46.9 (25.7-68.2)  
MMP9  0.53 
    ≤0.51 41.3 (29.9-52.7)  
    >0.51 33.6 (21.9-45.3)  
Gene expression Median survival (95% CI)P
CSF1  0.002 
    ≤0.90 NR  
    >0.90 27.5 (20.1-34.9)  
FN1  0.002 
    ≤0.57 NR  
    >0.57 31.3 (19.3-43.2)  
CA IX  0.007 
    ≤0.56 29 (20.9-37.2)  
    >0.56 NR  
PH4  0.01 
    ≤0.90 NR  
    >0.90 29.7 (21.9-37.5)  
KIAA0974  0.02 
    ≤0.71 NR  
    >0.71 21.3 (22.3-40.2)  
ANLN  0.02 
    ≤1.09 NR  
    >1.09 31.3 (24.4-38.2)  
ISNR  0.03 
    ≤0.57 NR  
    >0.57 33.2 (19.3-47.1)  
VEGFC  0.03 
    ≤0.67 NR  
    >0.67 33.2 (26.9-39.5)  
NTRK1  0.04 
    ≤0.53 NR  
    >0.53 33.2 (24.9-41.6)  
EGFR  0.05 
    ≤0.76 NR  
    >0.76 33.9 (23.7-44.2)  
SERPINE  0.08 
    ≤0.45 46.9  
    >0.45 33.2 (19.2-46.6)  
Ezrin  0.10 
    ≤0.81 NR  
    >0.81 33.2 (23.1-43.3)  
ARHGDIB  0.11 
    ≤0.76 41.3  
    >0.76 29.7 (19.8-39.5)  
Selectin P ligand  0.12 
    ≤0.71 46.9 (33.1-60.8)  
    >0.71 29.7 (20.2-39.2)  
CXCR4  0.18 
    ≤0.61 41.3  
    >0.61 31.3 (25.6-36.9)  
STX1A  0.18 
    ≤0.72 46.9  
    >0.72 33.6 (27.9-39.3)  
PDPN  0.19 
    ≤0.59 46.9  
    >0.59 33.2 (20.6-45.9)  
HGF  0.21 
    ≤0.51 46.9 (32.9-61.1)  
    >0.51 33.2 (26-40.5)  
Selectin L ligand  0.33 
    ≤0.59 41.3 (30.3-52.3)  
    >0.59 33.2 (23.1-43.4)  
PGK1  0.35 
    ≤0.90 33.6 (20.4-46.8)  
    >0.90 38.2 (28.8-47.5)  
IL8  0.86 
    ≤0.46 36.7  
    >0.46 38.2 (29.1-47.2)  
CD44  0.42 
    ≤1.62 33.6 (21.6-45.6)  
    >1.62 46.9 (31.9-62)  
CD47   
    ≤0.70 41.3 0.36 
    >0.70 33.9 (17-50.9)  
NTRK2  0.68 
    ≤0.32 41.3 (28.3-54.3)  
    >0.32 38.2  
N-cadherin  0.34 
    ≤0.34 46.9  
    >0.34 33.6 (27-40)  
TWIST1  0.69 
    ≤0.77 36.7 (25.8-47.6)  
    >0.77 46.9 (28.7-65.2)  
S100P  0.64 
    ≤0.21 33.6 (26-41.2)  
    >0.21 46.9 (34.9-59.1)  
GPI  0.52 
    ≤0.72 33.9 (22.4-45.5)  
    >0.72 46.9 (25.7-68.2)  
MMP9  0.53 
    ≤0.51 41.3 (29.9-52.7)  
    >0.51 33.6 (21.9-45.3)  

Abbreviation: NR, not reached.

Table 3.

Multivariate Cox model for overall survival in the training cohort

Hazard ratio (95% CI)P
Tumor size (cm)   
    ≤4 1 (Reference)  
    >4 2.7 (1.1-6.6) 0.02 
CSF1   
    ≤0.90 1 (Reference)  
    >0.90 3.5 (1.5-8.5) 0.005 
EGFR   
    ≤0.76 1 (Reference)  
    >0.76 2.7 (1.2-6.4) 0.02 
CA IX   
    ≤0.56 1 (Reference)  
    >0.56 0.2 (0.07-0.43) <0.0001 
Hazard ratio (95% CI)P
Tumor size (cm)   
    ≤4 1 (Reference)  
    >4 2.7 (1.1-6.6) 0.02 
CSF1   
    ≤0.90 1 (Reference)  
    >0.90 3.5 (1.5-8.5) 0.005 
EGFR   
    ≤0.76 1 (Reference)  
    >0.76 2.7 (1.2-6.4) 0.02 
CA IX   
    ≤0.56 1 (Reference)  
    >0.56 0.2 (0.07-0.43) <0.0001 

A risk score was generated by adding the z scores of the expression levels of each of the three genes multiplied by its corresponding coefficient as follows: risk score = (0.93 × CSF) + (1.1 × EGFR) + (1.4 × CA IX). The risk score was used to classify patients into high (>0) or low (≤0) risk, where high risk meant poor survival. Overall median survival for high-risk patients was 24 months (95% CI, 17.1-30.9 months), whereas it was not reached in low-risk patients (P < 0.00001; Fig. 1). The risk score was able to predict survival with 70% accuracy. The 3-year survival was 62% for low-risk patients and 20% for high-risk patients. When patients were divided into two subgroups according to primary tumor size (≤4 cm versus >4 cm), median survival was not reached for low-risk patients in either subgroup, whereas for high-risk patients, it was 27.5 months for those with tumors ≤4 cm (P = 0.005) and 20 months (95% CI, 10-32.1 months) for those with tumors >4 cm (P = 0.002; Fig. 2). Ninety-three percent of the patients in the low-risk group and 100% of those in the high-risk group were classified correctly.

Fig. 1.

Overall survival of patients in the training cohort according to their risk score.

Fig. 1.

Overall survival of patients in the training cohort according to their risk score.

Close modal
Fig. 2.

Overall survival of patients with tumors ≤4 cm (A) and >4 cm (B) according to their risk score.

Fig. 2.

Overall survival of patients with tumors ≤4 cm (A) and >4 cm (B) according to their risk score.

Close modal

The prognostic value of the risk score was positively validated in an independent cohort of 26 lung SCC patients at the University of California at San Francisco. Median survival in the low-risk group was not reached, whereas it was 60 months (95% CI, 12.8-108.2 months) for the high-risk group (P = 0.05; Fig. 3). The accuracy of the risk score in the validation cohort was 70%.

Fig. 3.

Overall survival of patients in the validation cohort according to their risk score.

Fig. 3.

Overall survival of patients in the validation cohort according to their risk score.

Close modal

As an ancillary analysis, we tested the sensitivity and specificity of the risk score for prediction of metastasis development. Sixty percent of the patients in the training cohort were selected at random for setting the threshold values, and the remaining 40% of the patients formed the validation group. The risk score showed 85% sensitivity and 60% specificity for prediction of metastasis development.

NSCLC is a heterogeneous disease with varying outcome in patients within the same stage. Gene expression profiling by microarrays or RT-QPCR has been used to predict prognosis in NSCLC patients (28). In this first RT-QPCR–based study, including exclusively lung SCC patients, we have evaluated 29 genes on the basis of their potential clinical importance and constructed a prognostic three-gene expression signature (CSF-1, EGFR, and CA IX) for lung SCC. The prognostic value of the gene expression signature was validated in an independent cohort of patients with long follow-up.

The prognostic power of our three-gene signature is comparable with that observed with recent RT-QPCR–based three-gene (20) and five-gene (14) signatures and with a five-microRNA signature (21). The 70% accuracy of our model is similar to the 68% accuracy of a 50-gene expression signature (6) and the 72% accuracy of a 111-gene expression signature (11), both derived from microarrays. However, the 50-gene signature was characterized by relatively low sensitivity (41%; ref. 6), which may limit its potential clinical application. In another microarray-based study (8), the accuracy of a “lung metagene” validated in two independent cohorts of NSCLC patients showed an overall accuracy of 72% and 79%. However, only 14 SCC patients were included in the first validation cohort, and the second validation cohort included only adenocarcinoma patients. A meta-analysis of lung cancer microarray studies, including both adenocarcinomas and SCCs, used extensive statistical methods for conversion of the gene expression values reported in the original studies and found a 64-gene expression signature with 87% accuracy in a subgroup of stage I patients (7).

Numerous gene expression signatures may be capable of predicting outcome in NSCLC, as has been shown in breast cancer (2224). Four gene expression signatures [hypoxia response (25), oncogenic pathways (10), lung metastasis (26), and invasiveness gene (27)], although containing largely nonoverlapping genes, showed similar prognostic value in breast cancer (28). The 186-gene invasiveness gene signature was also associated with prognosis in 60 lung cancer patients, although with relatively low accuracy (27).

The three genes in our model have key functions in tumor invasion and metastasis. CSF-1 (also known as macrophage colony stimulating factor) coincides with dense infiltration by tumor-associated macrophages, related to poor prognosis in NSCLC patients (29). The interaction of lung cancer cells and macrophages promotes tumor invasion and metastasis by means of a self-propagating chemotaxis loop involving reciprocal signaling between carcinoma cells expressing EGFR and macrophages expressing CSF-1 (30). A recently reported 11-cytokine gene signature, including CSF-1, predicted outcome in 75% of lung adenocarcinoma patients (31). Patients whose early-stage tumors contain signatures predicting short survival may benefit from adjuvant treatment, including drugs that specifically target the CSF-1/EGFR chemotaxis loop (32).

High levels of CA IX expression formed part of a favorable prognostic three-gene signature in a microarray analysis of 116 acute myeloid leukemia patients (33). It seems that the overall expression of CA IX decreases with progression and development of metastases, raising the hypothesis that in the later stages of tumor growth, continued CA IX expression is no longer a requirement (34). (The other seven genes identified as prognostic markers in the univariate analysis are discussed in the Supplementary Text S1.).

New gene expression signatures based on functional criteria may provide complementary information that will help to refine a patient's prognosis and inform therapeutic choices. The RT-QPCR assay is convenient in terms of laboratory work load and applicable for large-scale routine use, making it a viable alternative to more complex microarrays. The CSF-1/EGFR/CA IX gene expression signature can identify lung SCC patients with an increased risk of death and may be used to guide treatment strategies. A prospective randomized trial is warranted to validate this gene expression signature as a basis for selecting high-risk patients for adjuvant chemotherapy.

No potential conflicts of interest were disclosed.

Grant support: Spanish Ministry of Health grant FIS05/1621, Red de Centros de Epidemiología y Salud Pública and Red Temática de Investigación Cooperativa de Centros de Cáncer (RD06/0020/0056), La Fundación Badalona Contra el Cáncer, and European Respiratory Society Fellowship 372. These funding entities did not have access to data and were not involved in study design, data collection, decision to publish, or the writing of the manuscript.

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 with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

1
Strauss GM. Adjuvant chemotherapy of lung cancer: methodologic issues and therapeutic advances.
Hematol Oncol Clin North Am
2005
;
19
:
263
–81, vi.
2
Bhattacharjee A, Richards WG, Staunton J, et al. Classification of human lung carcinomas by mRNA expression profiling reveals distinct adenocarcinoma subclasses.
Proc Natl Acad Sci U S A
2001
;
98
:
13790
–5.
3
Garber ME, Troyanskaya OG, Schluens K, et al. Diversity of gene expression in adenocarcinoma of the lung.
Proc Natl Acad Sci U S A
2001
;
98
:
13784
–9.
4
Beer DG, Kardia SL, Huang CC, et al. Gene-expression profiles predict survival of patients with lung adenocarcinoma.
Nat Med
2002
;
8
:
816
–24.
5
Wigle DA, Jurisica I, Radulovich N, et al. Molecular profiling of non-small cell lung cancer and correlation with disease-free survival.
Cancer Res
2002
;
62
:
3005
–8.
6
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
–72.
7
Lu Y, Lemon W, Liu PY, et al. A gene expression signature predicts survival of patients with stage I non-small cell lung cancer.
PLoS Med
2006
;
3
:
e467
.
8
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
–80.
9
Borczuk AC, Shah L, Pearson GD, et al. Molecular signatures in biopsy specimens of lung cancer.
Am J Respir Crit Care Med
2004
;
170
:
167
–74.
10
Bild AH, Yao G, Chang JT, et al. Oncogenic pathway signatures in human cancers as a guide to targeted therapies.
Nature
2006
;
439
:
353
–7.
11
Larsen JE, Pavey SJ, Passmore LH, et al. Expression profiling defines a recurrence signature in lung squamous cell carcinoma.
Carcinogenesis
2007
;
28
:
760
–6.
12
Parmigiani G, Garrett-Mayer ES, Anbazhagan R, Gabrielson E. A cross-study comparison of gene expression studies for the molecular classification of lung cancer.
Clin Cancer Res
2004
;
10
:
2922
–7.
13
Endoh H, Tomida S, Yatabe Y, et al. Prognostic model of pulmonary adenocarcinoma by expression profiling of eight genes as determined by quantitative real-time reverse transcriptase polymerase chain reaction.
J Clin Oncol
2004
;
22
:
811
–9.
14
Chen HY, Yu SL, Chen CH, et al. A five-gene signature and clinical outcome in non-small-cell lung cancer.
N Engl J Med
2007
;
356
:
11
–20.
15
McDoniels-Silvers AL, Stoner GD, Lubet RA, You M. Differential expression of critical cellular genes in human lung adenocarcinomas and squamous cell carcinomas in comparison to normal lung tissues.
Neoplasia
2002
;
4
:
141
–50.
16
Muller-Tidow C, Diederichs S, Bulk E, et al. Identification of metastasis-associated receptor tyrosine kinases in non-small cell lung cancer.
Cancer Res
2005
;
65
:
1778
–82.
17
Rosell R, Skrzypski M, Jassem E, et al. BRCA1: a novel prognostic factor in resected non-small-cell lung cancer.
PLoS ONE
2007
;
2
:
e1129
.
18
Inamura K, Fujiwara T, Hoshida Y, et al. Two subclasses of lung squamous cell carcinoma with different gene expression profiles and prognosis identified by hierarchical clustering and non-negative matrix factorization.
Oncogene
2005
;
24
:
7105
–13.
19
Livak KJ, Schmittgen TD. Analysis of relative gene expression data using real-time quantitative PCR and the 2(−ΔΔC(T)) method.
Methods
2001
;
25
:
402
–8.
20
Lau SK, Boutros PC, Pintilie M, et al. Three-gene prognostic classifier for early-stage non small-cell lung cancer.
J Clin Oncol
2007
;
25
:
5562
–9.
21
Yu SL, Chen HY, Chang GC, et al. MicroRNA signature predicts survival and relapse in lung cancer.
Cancer Cell
2008
;
13
:
48
–57.
22
Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer.
N Engl J Med
2004
;
351
:
2817
–26.
23
Wang Y, Klijn JG, Zhang Y, et al. Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer.
Lancet
2005
;
365
:
671
–9.
24
Bueno-de-Mesquita JM, van Harten WH, Retel VP, et al. Use of 70-gene signature to predict prognosis of patients with node-negative breast cancer: a prospective community-based feasibility study (RASTER).
Lancet Oncol
2007
;
8
:
1079
–87.
25
Chi JT, Wang Z, Nuyten DS, et al. Gene expression programs in response to hypoxia: cell type specificity and prognostic significance in human cancers.
PLoS Med
2006
;
3
:
e47
.
26
Minn AJ, Gupta GP, Siegel PM, et al. Genes that mediate breast cancer metastasis to lung.
Nature
2005
;
436
:
518
–24.
27
Liu R, Wang X, Chen GY, et al. The prognostic role of a gene signature from tumorigenic breast-cancer cells.
N Engl J Med
2007
;
356
:
217
–26.
28
Massague J. Sorting out breast-cancer gene signatures.
N Engl J Med
2007
;
356
:
294
–7.
29
Chen JJ, Lin YC, Yao PL, et al. Tumor-associated macrophages: the double-edged sword in cancer progression.
J Clin Oncol
2005
;
23
:
953
–64.
30
Wyckoff J, Wang W, Lin EY, et al. A paracrine loop between tumor cells and macrophages is required for tumor cell migration in mammary tumors.
Cancer Res
2004
;
64
:
7022
–9.
31
Seike M, Yanaihara N, Bowman ED, et al. Use of a cytokine gene expression signature in lung adenocarcinoma and the surrounding tissue as a prognostic classifier.
J Natl Cancer Inst
2007
;
99
:
1257
–69.
32
Allavena P, Signorelli M, Chieppa M, et al. Anti-inflammatory properties of the novel antitumor agent yondelis (trabectedin): inhibition of macrophage differentiation and cytokine production.
Cancer Res
2005
;
65
:
2964
–71.
33
Greiner J, Schmitt M, Li L, et al. Expression of tumor-associated antigens in acute myeloid leukemia: Implications for specific immunotherapeutic approaches.
Blood
2006
;
108
:
4109
–17.
34
Bui MH, Seligson D, Han KR, et al. Carbonic anhydrase IX is an independent predictor of survival in advanced renal clear cell carcinoma: implications for prognosis and therapy.
Clin Cancer Res
2003
;
9
:
802
–11.

Supplementary data