This study was designed to identify TGF-β signaling pathway-related serum microRNAs (miRNA) as predictors of survival in advanced non–small cell lung cancer (NSCLC). Serum samples from 391 patients with advanced NSCLC were collected before treatment. Global miRNA microarray expression profiling based on sera from four patients with good survival (>24 months) and four patients with poor survival (<6 months) was used to identify 140 highly expressed serum miRNAs, among which 35 miRNAs had binding sites within the 3′-untranslated regions of a panel of 11 genes in the TGF-β signaling pathway and were assayed by quantitative RT-PCR for their associations with survival in a training (n = 192) and testing set (n = 191). Out of the 35 miRNAs, survival analysis using Cox regression model identified 17 miRNAs significantly associated with 2-year patient survival. MiR-16 exhibited the most statistically significant association: high expression of miR-16 was associated with a significantly better survival [adjusted hazard ratio (HR) = 0.4, 95% confidence interval (CI): 0.3–0.5]. A combined 17-miRNA risk score was created that was able to identify patients at the highest risk of death. Those with a high-risk score had a 2.5-fold increased risk of death compared with those with a low risk score (95% CI: 1.8–3.4; P = 1.1 × 10−7). This increase in risk of death was corresponding to a 7.8-month decrease in median survival time (P = 9.5 × 10−14). Our results suggest that serum miRNAs could serve as predictors of survival for advanced NSCLC. Cancer Res; 73(15); 4801–9. ©2013 AACR.

Non–small cell lung cancer (NSCLC) accounts for 85% of all lung cancers, and its overall 5-year survival rate is only 17% (1). Approximately 65% to 75% of patients with NSCLC possess unresectable advanced or metastatic disease at diagnosis (2). The overall response rate for platinum-based therapy, which is the standard of care for these patients, ranges between 17% and 32% (3), with a median survival time (MST) of only 7 to 9 months (4). The current challenge posed by the rather short survival time in patients with advanced NSCLC demands the discovery of new biomarkers that enable prediction of survival time and categorization of different prognostic groups, thereby enabling health care providers to personalize treatment and helping patients to manage expectations effectively.

Alterations in microRNA (miRNA) expression have been known to affect multiple cellular processes, including cell differentiation, cell-cycle progression, and apoptosis (5–7). It has been clearly shown that miRNAs are of potential clinical relevance and possess enormous promise to serve as novel biomarkers in cancer diagnosis and prognosis (8–10). Stable existence of cell-free miRNA in plasma/serum has been repeatedly shown in numerous studies (11–13), and alterations in miRNA profiles or expression levels are potentially indicative of physiologic and pathologic change. A number of studies have shown the potential of miRNAs in plasma/serum as biomarkers for specific cancer diagnosis and prognosis (14). In this study, we aimed to identify potential serum miRNA biomarkers for predicting patient survival in advanced-stage NSCLC. We show that expression levels of selected serum miRNAs are significantly associated with patient survival.

To focus on serum miRNAs associated with the cancer signaling network, we used a pathway-based miRNA profiling approach, specifically focusing on the TGF-β pathway. This pathway plays critical roles in control of cell proliferation, differentiation, apoptosis, adhesion, invasion, and the cellular microenvironment (15). TGF-β signaling has tumor suppressor effects during the early phase of tumor progression and, in contrast, assists in tumor progression by promoting tumor cell invasion, dissemination, and immune evasion in late-stage disease (16).

Patient population and clinical data collection

This study was conducted in 391 Caucasian patients with newly diagnosed, histologically confirmed NSCLC recruited at the University of Texas MD Anderson Cancer Center between January 2002 and January 2009, which is part of an ongoing lung cancer study initiated in 1991. None of the patients had undergone treatment before study enrollment. All study participants signed informed consent and underwent a 45-minute in-person interview by trained MD Anderson staff with a structured questionnaire that elicited information on demographic characteristics, medical history and smoking history. Immediately after each interview, a 40-mL peripheral blood sample was drawn. About 10 mL was drawn into a gold top serum-separating tube, processed for serum extraction within 2 hours, and then transferred into liquid nitrogen tanks for long-term storage. Clinical and follow-up data were abstracted from medical charts. Serum samples were selected retrospectively at the time of analysis according to the following requirements: (i) the patient had been diagnosed with stage III or IV NSCLC, (ii) serum was available in sufficient volume for RNA isolation, and (iii) demographic, clinical, and follow-up data for the patient were available. There were a total of 1,060 advanced NSCLC Caucasian patients recruited during the study period. We excluded 12 patients without sufficient serum volume, one patient with more than 10 years of serum storage, 17 patients with serum samples collected after treatment started, 201 newly recruited patients with demographic and clinical data yet to be input into database at the time of sample selection, and 22 patients lost to follow-up before 2 years. Among the remaining 807 patients, 484 patients had their serum samples stored in Cryo vials and the other 323 serum samples were stored in MAPI straws (Cryo Bio System). To minimize the effect of serum storage condition on miRNA expression, we only used the 484 samples stored in Cryo vials. After excluding 29 patients with hemolyzed serum samples upon visual determination and 64 patients without sufficient serum RNA concentration after isolation, 391 patients were included in our final dataset. Comparisons of the patients included in this study to the overall study population of 1,060 showed no differences by basic demographic and clinical data distribution including age, sex, smoking status, disease stage, histology, grade, performance status and treatment regimens, except that patients included in this study had a slightly high death rate. This would be expected given that the majority of excluded patients were recently recruited patients who were still alive and did not have sufficient follow-up time. Study approval was obtained from the MD Anderson Institutional Review Board.

The initial global miRNA screening included 8 Caucasian patients, 4 with good survival (survival time >24 months) and 4 with poor survival (survival time <6 months), matched by age, sex, histologic stage, and smoking status (Fig. 1). Assignment of the 383 samples to the training and testing populations was carried out by sorting the samples by enrollment date and then distributing in an alternating manner (Table 1).

Figure 1.

Schematic of study design.

Figure 1.

Schematic of study design.

Close modal
Table 1.

Selected characteristics of NSCLC patient population recruited at the University of Texas MD Anderson Cancer Center between 2002 and 2009

CharacteristicsStudy population (N = 391)Training set (N = 192)Testing set (N = 191)
Vital status, n (%) 
 Alive 57 (14.6) 33 (17.2) 24 (12.6) 
 Dead 334 (85.4) 159 (82.8) 167 (87.4) 
Mean age, y (SD) 62.05 (11.3) 63.41 (11.3) 60.9 (11.3) 
Sex 
 Male 203 (51.9) 107 (55.7) 92 (48.2) 
 Female 188 (48.1) 85 (44.3) 99 (51.8) 
Smoking status, n (%) 
 Never 80 (20.5) 34 (17.7) 45 (23.6) 
 Former 165 (42.2) 91 (47.4) 70 (36.7) 
 Current 146 (37.3) 67 (34.9) 76 (39.8) 
TNM classification, n (%) 
 IIIA 58 (14.8) 31 (16.2) 27 (14.1) 
 IIIB (dry) 63 (16.1) 31 (16.2) 30 (15.7) 
 IIIB (wet) 28 (7.2) 12 (6.3) 14 (7.3) 
 IV 242 (61.9) 118 (61.5) 120 (62.8) 
Histology 
 Adenocarcinoma 214 (54.7) 97 (50.5) 114 (59.7) 
 Squamous 87 (22.3) 44 (22.9) 39 (20.4) 
 Other 90 (23.0) 51 (26.6) 38 (19.9) 
Performance status, n (%) 
 0 82 (21.0) 43 (22.4) 39 (20.4) 
 1 186 (47.6) 95 (49.5) 85 (44.5) 
 2–4 76 (19.4) 34 (17.7) 40 (20.9) 
 Unknown 47 (12.0) 20 (10.4) 27 (14.1) 
Histology grade, n (%) 
 Well differentiated 20 (5.1) 8 (4.2) 12 (6.3) 
 Moderately differentiated 60 (15.4) 26 (13.5) 34 (17.8) 
 Poorly differentiated 141 (36.1) 70 (36.5) 68 (35.6) 
 Undifferentiated 5 (1.3) 5 (2.6) 0 (0.0) 
 Unknown 165 (42.2) 83 (43.2) 77 (40.3) 
Treatment regimena, n (%) 
 Chemotherapy 100 (25.6) 43 (22.4) 52 (27.2) 
 Chemoradiotherapya 114 (29.2) 61 (31.8) 50 (26.2) 
 Chemotherapy + Surgery 18 (4.6) 8 (4.2) 10 (5.2) 
 Chemoradiotherapya+ Surgery 17 (4.4) 10 (5.2) 7 (3.7) 
 Surgery 15 (3.8) 3 (1.6) 12 (6.3) 
 Radiation 53 (13.6) 30 (15.6) 23 (12) 
 Radiation + surgery 5 (1.3) 1 (0.5) 4 (2.1) 
 No treatment 69 (17.7) 36 (18.8) 33 (17.3) 
MST, mo 10.3 10.3 10.0 
MFT, mo (range) 10.3 (1.0–85.7) 10.4 (1.0–76.2) 10.0 (1.6–85.7) 
CharacteristicsStudy population (N = 391)Training set (N = 192)Testing set (N = 191)
Vital status, n (%) 
 Alive 57 (14.6) 33 (17.2) 24 (12.6) 
 Dead 334 (85.4) 159 (82.8) 167 (87.4) 
Mean age, y (SD) 62.05 (11.3) 63.41 (11.3) 60.9 (11.3) 
Sex 
 Male 203 (51.9) 107 (55.7) 92 (48.2) 
 Female 188 (48.1) 85 (44.3) 99 (51.8) 
Smoking status, n (%) 
 Never 80 (20.5) 34 (17.7) 45 (23.6) 
 Former 165 (42.2) 91 (47.4) 70 (36.7) 
 Current 146 (37.3) 67 (34.9) 76 (39.8) 
TNM classification, n (%) 
 IIIA 58 (14.8) 31 (16.2) 27 (14.1) 
 IIIB (dry) 63 (16.1) 31 (16.2) 30 (15.7) 
 IIIB (wet) 28 (7.2) 12 (6.3) 14 (7.3) 
 IV 242 (61.9) 118 (61.5) 120 (62.8) 
Histology 
 Adenocarcinoma 214 (54.7) 97 (50.5) 114 (59.7) 
 Squamous 87 (22.3) 44 (22.9) 39 (20.4) 
 Other 90 (23.0) 51 (26.6) 38 (19.9) 
Performance status, n (%) 
 0 82 (21.0) 43 (22.4) 39 (20.4) 
 1 186 (47.6) 95 (49.5) 85 (44.5) 
 2–4 76 (19.4) 34 (17.7) 40 (20.9) 
 Unknown 47 (12.0) 20 (10.4) 27 (14.1) 
Histology grade, n (%) 
 Well differentiated 20 (5.1) 8 (4.2) 12 (6.3) 
 Moderately differentiated 60 (15.4) 26 (13.5) 34 (17.8) 
 Poorly differentiated 141 (36.1) 70 (36.5) 68 (35.6) 
 Undifferentiated 5 (1.3) 5 (2.6) 0 (0.0) 
 Unknown 165 (42.2) 83 (43.2) 77 (40.3) 
Treatment regimena, n (%) 
 Chemotherapy 100 (25.6) 43 (22.4) 52 (27.2) 
 Chemoradiotherapya 114 (29.2) 61 (31.8) 50 (26.2) 
 Chemotherapy + Surgery 18 (4.6) 8 (4.2) 10 (5.2) 
 Chemoradiotherapya+ Surgery 17 (4.4) 10 (5.2) 7 (3.7) 
 Surgery 15 (3.8) 3 (1.6) 12 (6.3) 
 Radiation 53 (13.6) 30 (15.6) 23 (12) 
 Radiation + surgery 5 (1.3) 1 (0.5) 4 (2.1) 
 No treatment 69 (17.7) 36 (18.8) 33 (17.3) 
MST, mo 10.3 10.3 10.0 
MFT, mo (range) 10.3 (1.0–85.7) 10.4 (1.0–76.2) 10.0 (1.6–85.7) 

Abbreviations: MST, median survival time; MFT, median follow-up time.

aCombined chemotherapy and radiation concurrently or consequentially.

Total RNA isolation

Total RNA was isolated from serum samples using the miRNeasy Mini Kit (Qiagen) following the manufacturer's protocol. For whole-genome miRNA profiling, 700 μL serum was used for RNA isolation, and for qRT-PCR assay, the serum amount was reduced to 400 μL. Synthetic cel-miR-39 and cel-miR-54 were added to each sample as internal controls for evaluation of successful extraction. RNAs were eluted twice with 30 μL of water. The total concentration of small RNA molecules was quantified by NanoDrop ND-1000 spectrometer (Thermo Scientific). To minimize any potential inaccuracies generated by the Nanodrop measurements, we carefully kept consistent the amount of serum samples and spike-in miRNAs prepared from the same batch for isolation—throughout all experiments. In addition, the raw Ct value for each miRNA was normalized to the raw Ct value for spike-in miRNAs obtained from each individual sample to eliminate potential variations introduced from the isolation process and RNA quantification. To eliminate potential hemolysis-introduced miRNA bias, 30 serum samples were randomly selected and tested for free hemoglobin levels on VITROS Fusion 5.1 Chemistry System (Ortho Clinical Diagnostics) at the University of Texas MD Anderson Cancer Center Core Laboratory by standard protocol.

Selection of TGF-β signaling pathway related serum miRNAs

Whole-genome serum miRNA profiling was carried out on the 8 serum samples using the TaqMan Human MicroRNA Microarray Card Set v3.0 (Applied Biosystems). In brief, purified serum RNA samples were reverse-transcribed using Megaplex RT Primers followed by a pre-amplification step using Megaplex PreAmp Primers for maximum sensitivity of detection. For final quantification, the TaqMan Universal Master Mix II was added, and each sample was loaded on the MicroRNA Microarray Cards. Quantitative PCR was conducted on the 7900HT Fast Real-Time PCR System (Applied Biosystems). The expression threshold for each miRNA detector was automatically determined. A miRNA candidate was considered highly expressed when more than 75% of samples in both study groups generated detectable Ct values of less than 35.

To identify miRNA candidates associated with the TGF-β signaling pathway, a total of 23 candidate genes were selected using Gene Ontology (http://www.geneontology.org) and the National Center for Biotechnology Information PubMed (http://www.ncbi.nlm.nih.gov) databases as previously described (17). miRNAs potentially binding these candidate genes were predicted by TargetScanHuman v5.2 (www.targetscan.org; ref. 18), and miRNAs associated with the binding sites broadly conserved among vertebrates with higher probability of preferential scores (19) were considered for further analysis.

Quantitative real-time PCR assay

Selected miRNAs were tested using individual TaqMan MiRNA Assays (Applied Biosystems) to quantitatively measure the difference in their expression levels between patient groups with varied survival times. Each assay was tested in duplicates. Data points that generated duplicated Ct values with over one cycle variance were excluded from analysis. The mean Ct value obtained from each sample was normalized to the averaged expression of spike-in miRNAs and then subjected to analysis with the 2−ΔΔCt method (20).

Statistical analysis

Statistical analyses were conducted using Intercooled STATA software, version 10 (College Station). Chi-square (χ2) analysis was used to evaluate differences in patient characteristics. Survival time was calculated from the date of study enrollment to the date of death or last patient follow-up. The median time between diagnosis of NSCLC and study enrollment (date of blood draw) was 24 days. Rank sum test was used to assess differences in median miRNA expression levels between patients with stage III and stage IV NSCLC. The associations between 2-year survival and serum miRNA expression levels were estimated as hazard ratios (HR) and 95% confidence intervals (CI) for each miRNA using the Cox proportional hazards model, adjusted for age, sex, smoking status, clinical stage, and treatment regimen. Patients with survival times more than 24 months were censored at 24 months in Cox regression analysis. In all statistical analyses, a P value less than or equal to 0.05 was considered significant. The combined 17-miRNA risk score for each patient was derived by linear combination of the product of reference-normalized expression level of each miRNA by its Cox regression corresponding coefficient (21). All patients were dichotomized by the median risk score, and individuals with a risk score higher or lower than the median were classified as high or low risk groups, respectively. The optimized statistical method, C-index was used to evaluate predictive accuracy of 2-year survival (22).

Patient characteristics

A total of 391 Caucasian patients with advanced NSCLC were included in the analysis that covered a broad range of follow-up and survival times (Table 1). For the initial global miRNA screening using 8 samples, survival times in the poor- and good-survival groups ranged from 2.47 to 3.82 months and from 24.61 to 65.23 months, respectively. Overall, there was no significant difference between the training set and testing set with respect to vital status, sex, smoking status, performance status, TNM stage, or histology distribution (Table 1). A slight difference was seen between the two populations in terms of age, which however, was adjusted in subsequent analysis. At the time of analysis, 326 (85.1%) patients had died and 57 (14.9%) were still alive, all of which had survived for more than 2 years.

Whole-genome serum miRNA profiling

In the combined pool of 754 human miRNAs, excluding endogenous controls and low-quality data points, 166 and 171 serum miRNAs were identified to be highly expressed in the poor- and good-survival groups, respectively, with 140 serum miRNAs were highly expressed in both groups. miRNA target prediction using the TargetScan database identified 53 potential binding sites for 65 miRNAs in the 3′ untranslated regions (UTR) of 11 TGF-β pathway genes (Supplementary Material: Table S1). Cross referencing of this list of miRNAs with the 140 highly expressed serum miRNAs identified 35 candidate miRNAs for further analysis (Fig. 1). These 35 miRNA candidates included members in families associated with cancer progression processes, such as the let-7 family (23), the miR-17∼92 cluster (24), and the miR-200 family (Supplementary Material: Table S2; refs. 25, 26). In addition, it was noteworthy that one of the candidates, miR-30d, was one of the 4 biomarkers previously identified to be predictive of survival in early-stage NSCLC (27).

Association of serum miRNA expression with NSCLC survival

The 35 candidate miRNAs were assessed for association with survival duration. In the training set, expression levels of 25 serum miRNAs were significantly associated with 2-year survival using the median expression levels as analysis cutoff (data not shown). When the same cutoffs were applied to the testing set, 17 of 25 miRNAs remained significant with the same direction of effect (Table 2). The results showed that higher expression levels (above the median value) of these 17 serum miRNAs were significantly associated with higher probability of 2-year survival in both the training and testing sets using the same cutoff points. miR-16 presented the most significant result: in the combined dataset analysis, the 2-year survival rate in the high-expression group reached 37.9%, whereas only 3 of 174 patients (1.7%) in the low-expression group survived more than 2 years, with an adjusted HR of 0.4 (95% CI: 0.3–0.5), and a P value of 1.7 × 10−10 (Fig. 2A). The MST advantage in the high-expression group compared with the low-expression group ranged from 2.8 months to 11.1 months among these 17 serum miRNAs.

Figure 2.

Kaplan–Meier 2-year survival curves for patients with advanced NSCLC grouped by low (solid line) and high (dashed line) expression of serum miR-16 (A), and high (solid line) and low (dashed line) risk scores (B). N = number of patients with an event (death) at 2 years/total number of patients in the dataset. MST, median survival time.

Figure 2.

Kaplan–Meier 2-year survival curves for patients with advanced NSCLC grouped by low (solid line) and high (dashed line) expression of serum miR-16 (A), and high (solid line) and low (dashed line) risk scores (B). N = number of patients with an event (death) at 2 years/total number of patients in the dataset. MST, median survival time.

Close modal
Table 2.

Serum miRNA expression associated with 2-year survival in patients with advanced NSCLC from the study population recruited at the University of Texas MD Anderson Cancer Center between 2002 and 2009

Training setTesting setCombined set
2-year survival2-year survival2-year survival
Expression levelaN (No) n%N (Yes) n%Adjusted HRb(95% CI)PMSTN (No) n%N (Yes) n%Adjusted HRb(95% CI)PMSTN (No) n%N (Yes) n%Adjusted HRb (95% CI)PMST
miR-16 <0.71 83 (97.7) 2 (2.4) 1 (reference)  7.9 88 (98.9) 1 (1.1) 1 (reference)  9.0 171 (98.3) 3 (1.7) 1 (reference)  8.5 
 >0.71 51 (60.7) 33 (39.3) 0.44 (0.29–0.68) 1.9 × 10−4 16.7 54 (63.5) 31 (36.5) 0.37 (0.25–0.55) 1.3 × 10−6 19.0 105 (62.1) 64 (37.9) 0.39 (0.29–0.52) 1.7 × 10−10 18.0 
miR-30e-3p <0.58 86 (100) 0 (0) 1 (reference)  8.3 89 (98.9) 1 (1.1) 1 (reference)  9.2 175 (99.4) 1 (0.6) 1 (reference)  8.7 
 >0.58 48 (56.5) 37 (43.5) 0.45 (0.29–0.69) 3.1 × 10−4 20.4 54 (62.8) 32 (37.2) 0.38 (0.25–0.58) 5.6 × 10−6 19.0 102 (59.7) 69 (40.4) 0.4 (0.30–0.54) 7.8 × 10−10 19.8 
miR-106a <0.68 80 (92.0) 7 (8.1) 1 (reference)  7.6 83 (98.8) 1 (1.2) 1 (reference)  8.6 163 (95.3) 8 (4.7) 1 (reference)  8.5 
 >0.68 56 (64.4) 31 (35.6) 0.57 (0.39–0.85) 5.7 × 10−3 15.7 60 (65.2) 32 (34.8) 0.40 (0.28–0.59) 2.8 × 10−6 16.0 116 (64.8) 63 (35.2) 0.47 (0.36–0.61) 2.0 × 10−8 16.0 
miR-148b <0.72 78 (96.3) 3 (3.7) 1 (reference)  7.8 78 (96.3) 3 (3.7) 1 (reference)  8.8 156 (96.3) 6 (3.7) 1 (reference)  8.3 
 >0.72 51 (63.0) 30 (37.0) 0.49 (0.33–0.74) 5.9 × 10−4 16.4 57 (67.1) 28 (32.9) 0.51 (0.35–0.75) 5.3 × 10−4 14.4 108 (65.1) 58 (34.9) 0.5 (0.38–0.65) 3.4 × 10−7 15.5 
miR-26a <0.68 78 (89.7) 9 (10.3) 1 (reference)  8.5 77 (96.3) 3 (3.8) 1 (reference)  9.0 155 (92.8) 12 (7.2) 1 (reference)  8.7 
 >0.68 57 (66.3) 29 (33.7) 0.65 (0.44–0.95) 2.8 × 10−2 15.7 65 (68.4) 30 (31.6) 0.45 (0.30–0.66) 5.8 × 10−5 15.5 122 (67.4) 59 (32.6) 0.51 (0.39–0.67) 7.5 × 10−7 15.7 
miR-27a <0.67 81 (93.1) 6 (6.9) 1 (reference)  8.6 79 (97.5) 2 (2.5) 1 (reference)  9.0 160 (95.2) 8 (4.8) 1 (reference)  8.7 
 >0.67 55 (64.0) 31 (36.1) 0.59 (0.4–0.87) 7.3 × 10−3 16.4 64 (67.4) 31 (32.6) 0.47 (0.32–0.71) 2.3 × 10−4 16.0 119 (65.8) 62 (34.3) 0.51 (0.39–0.67) 9.3 × 10−7 16.2 
miR-106b <0.70 81 (93.1) 6 (6.9) 1 (reference)  7.2 85 (95.5) 4 (4.5) 1 (reference)  8.9 166 (94.3) 10 (5.7) 1 (reference)  8.3 
 >0.70 55 (64.0) 31 (36.1) 0.52 (0.35–0.77) 1.1 × 10−3 16.4 58 (66.7) 29 (33.3) 0.5 (0.34–0.72) 2.8 × 10−4 15.5 113 (65.3) 60 (34.7) 0.53 (0.41–0.68) 9.4 × 10−7 16.0 
miR-301a <1.16 71 (92.2) 6 (7.8) 1 (reference)  7.6 73 (93.6) 5 (6.4) 1 (reference)  8.3 144 (92.9) 11 (7.1) 1 (reference)  7.9 
 >1.16 50 (65.8) 26 (34.2) 0.53 (0.36–0.8) 2.1 × 10−3 14.4 51 (66.2) 26 (33.8) 0.54 (0.35–0.82) 3.7 × 10−3 16.0 101 (66.0) 52 (34.0) 0.53 (0.40–0.70) 6.7 × 10−6 15.8 
miR-27b <0.72 79 (91.9) 7 (8.1) 1 (reference)  7.1 75 (93.8) 5 (6.3) 1 (reference)  9.0 154 (92.8) 12 (7.2) 1 (reference)  8.2 
 >0.72 56 (65.9) 29 (34.1) 0.53 (0.36–0.79) 1.6 × 10−3 14.8 66 (74.2) 23 (25.8) 0.66 (0.46–0.93) 1.9 × 10−2 11.1 122 (70.1) 52 (29.9) 0.58 (0.45–0.75) 2.2 × 10−5 13.2 
miR-152 <0.68 77 (91.7) 7 (8.3) 1 (reference)  7.8 82 (92.1) 7 (7.9) 1 (reference)  8.6 159 (91.9) 14 (8.1) 1 (reference)  8.4 
 >0.68 55 (66.3) 28 (33.7) 0.6 (0.41–0.87) 7.2 × 10−3 14.4 56 (70) 24 (30.0) 0.60 (0.42–0.86) 5.3 × 10−3 12.7 111 (68.1) 52 (31.9) 0.59 (0.46–0.75) 3.3 × 10−5 14.1 
miR-20a <0.67 79 (92.9) 6 (7.1) 1 (reference)  7.8 76 (92.7) 6 (7.3) 1 (reference)  9.0 155 (92.8) 12 (7.2) 1 (reference)  8.5 
 >0.67 55 (65.5) 29 (34.5) 0.63 (0.42–0.95) 2.7 × 10−2 14.3 66 (71.0) 27 (29.0) 0.58 (0.41–0.83) 2.5 × 10−3 12.9 121 (68.4) 56 (31.6) 0.62 (0.48–0.79) 1.5 × 10−4 14.1 
miR-30a-5p <0.61 75 (89.3) 9 (10.7) 1 (reference)  8.7 76 (90.5) 8 (9.5) 1 (reference)  8.6 151 (89.9) 17 (10.1) 1 (reference)  8.7 
 >0.61 55 (65.5) 29 (34.5) 0.69 (0.48–1.01) 5.5 × 10−2 9.6 66 (72.5) 25 (27.5) 0.58 (0.41–0.83) 2.6 × 10−3 11.6 121 (69.1) 54 (30.9) 0.63 (0.49–0.81) 3.2 × 10−4 11.5 
miR-148a <0.74 79 (92.9) 6 (7.1) 1 (reference)  8.2 80 (90.9) 8 (9.1) 1 (reference)  9.6 159 (91.9) 14 (8.1) 1 (reference)  8.8 
 >0.74 55 (64.7) 30 (35.3) 0.57 (0.39–0.84) 3.8 × 10−3 14.3 59 (70.2) 25 (29.8) 0.72 (0.49–1.04) 8.3 × 10−2 12.9 114 (67.5) 55 (32.5) 0.63 (0.49–0.81) 4.1 × 10−4 14.1 
miR-145 <0.62 77 (91.7) 7 (8.3) 1 (reference)  8.2 73 (90.1) 8 (9.9) 1 (reference)  9.6 150 (90.9) 15 (9.1) 1 (reference)  8.8 
 >0.62 54 (65.1) 29 (34.9) 0.62 (0.42–0.92) 1.7 × 10−2 15.7 67 (72.0) 26 (28.0) 0.70 (0.49–0.99) 4.3 × 10−2 12.4 121 (68.8) 55 (31.3) 0.64 (0.49–0.82) 4.4 × 10−4 12.7 
miR-92a <0.63 77 (88.5) 10 (11.5) 1 (reference)  8.6 73 (91.3) 7 (8.8) 1 (reference)  8.9 150 (89.8) 17 (10.2) 1 (reference)  8.7 
 >0.63 59 (67.8) 28 (32.2) 0.70 (0.48–1.04) 7.6 × 10−2 14.1 70 (72.9) 26 (27.1) 0.66 (0.47–0.92) 1.6 × 10−2 11.1 129 (70.5) 54 (29.5) 0.67 (0.53–0.86) 1.6 × 10−3 12.4 
let-7c <0.66 74 (89.2) 9 (10.8) 1 (reference)  8.4 66 (88.0) 9 (12.0) 1 (reference)  9.2 140 (88.6) 18 (11.4) 1 (reference)  8.6 
 >0.66 56 (68.3) 26 (31.7) 0.65 (0.44–0.94) 2.4 × 10−2 14.3 68 (73.9) 24 (26.1) 0.72 (0.50–1.04) 7.7 × 10−2 11.6 124 (71.3) 50 (28.7) 0.68 (0.53–0.87) 2.6 × 10−3 13.4 
let-7b <0.60 78 (90.7) 8 (9.3) 1 (reference)  8.6 89 (92.7) 7 (7.3) 1 (reference)  9.3 167 (91.8) 15 (8.2) 1 (reference)  9.0 
 >0.60 55 (64.7) 30 (35.3) 0.65 (0.44–0.97) 3.5 × 10−2 11.3 53 (67.1) 26 (32.9) 0.67 (0.46–0.98) 3.9 × 10−2 13.7 108 (65.9) 56 (34.2) 0.69 (0.53–0.89) 4.5 × 10−3 12.7 
Training setTesting setCombined set
2-year survival2-year survival2-year survival
Expression levelaN (No) n%N (Yes) n%Adjusted HRb(95% CI)PMSTN (No) n%N (Yes) n%Adjusted HRb(95% CI)PMSTN (No) n%N (Yes) n%Adjusted HRb (95% CI)PMST
miR-16 <0.71 83 (97.7) 2 (2.4) 1 (reference)  7.9 88 (98.9) 1 (1.1) 1 (reference)  9.0 171 (98.3) 3 (1.7) 1 (reference)  8.5 
 >0.71 51 (60.7) 33 (39.3) 0.44 (0.29–0.68) 1.9 × 10−4 16.7 54 (63.5) 31 (36.5) 0.37 (0.25–0.55) 1.3 × 10−6 19.0 105 (62.1) 64 (37.9) 0.39 (0.29–0.52) 1.7 × 10−10 18.0 
miR-30e-3p <0.58 86 (100) 0 (0) 1 (reference)  8.3 89 (98.9) 1 (1.1) 1 (reference)  9.2 175 (99.4) 1 (0.6) 1 (reference)  8.7 
 >0.58 48 (56.5) 37 (43.5) 0.45 (0.29–0.69) 3.1 × 10−4 20.4 54 (62.8) 32 (37.2) 0.38 (0.25–0.58) 5.6 × 10−6 19.0 102 (59.7) 69 (40.4) 0.4 (0.30–0.54) 7.8 × 10−10 19.8 
miR-106a <0.68 80 (92.0) 7 (8.1) 1 (reference)  7.6 83 (98.8) 1 (1.2) 1 (reference)  8.6 163 (95.3) 8 (4.7) 1 (reference)  8.5 
 >0.68 56 (64.4) 31 (35.6) 0.57 (0.39–0.85) 5.7 × 10−3 15.7 60 (65.2) 32 (34.8) 0.40 (0.28–0.59) 2.8 × 10−6 16.0 116 (64.8) 63 (35.2) 0.47 (0.36–0.61) 2.0 × 10−8 16.0 
miR-148b <0.72 78 (96.3) 3 (3.7) 1 (reference)  7.8 78 (96.3) 3 (3.7) 1 (reference)  8.8 156 (96.3) 6 (3.7) 1 (reference)  8.3 
 >0.72 51 (63.0) 30 (37.0) 0.49 (0.33–0.74) 5.9 × 10−4 16.4 57 (67.1) 28 (32.9) 0.51 (0.35–0.75) 5.3 × 10−4 14.4 108 (65.1) 58 (34.9) 0.5 (0.38–0.65) 3.4 × 10−7 15.5 
miR-26a <0.68 78 (89.7) 9 (10.3) 1 (reference)  8.5 77 (96.3) 3 (3.8) 1 (reference)  9.0 155 (92.8) 12 (7.2) 1 (reference)  8.7 
 >0.68 57 (66.3) 29 (33.7) 0.65 (0.44–0.95) 2.8 × 10−2 15.7 65 (68.4) 30 (31.6) 0.45 (0.30–0.66) 5.8 × 10−5 15.5 122 (67.4) 59 (32.6) 0.51 (0.39–0.67) 7.5 × 10−7 15.7 
miR-27a <0.67 81 (93.1) 6 (6.9) 1 (reference)  8.6 79 (97.5) 2 (2.5) 1 (reference)  9.0 160 (95.2) 8 (4.8) 1 (reference)  8.7 
 >0.67 55 (64.0) 31 (36.1) 0.59 (0.4–0.87) 7.3 × 10−3 16.4 64 (67.4) 31 (32.6) 0.47 (0.32–0.71) 2.3 × 10−4 16.0 119 (65.8) 62 (34.3) 0.51 (0.39–0.67) 9.3 × 10−7 16.2 
miR-106b <0.70 81 (93.1) 6 (6.9) 1 (reference)  7.2 85 (95.5) 4 (4.5) 1 (reference)  8.9 166 (94.3) 10 (5.7) 1 (reference)  8.3 
 >0.70 55 (64.0) 31 (36.1) 0.52 (0.35–0.77) 1.1 × 10−3 16.4 58 (66.7) 29 (33.3) 0.5 (0.34–0.72) 2.8 × 10−4 15.5 113 (65.3) 60 (34.7) 0.53 (0.41–0.68) 9.4 × 10−7 16.0 
miR-301a <1.16 71 (92.2) 6 (7.8) 1 (reference)  7.6 73 (93.6) 5 (6.4) 1 (reference)  8.3 144 (92.9) 11 (7.1) 1 (reference)  7.9 
 >1.16 50 (65.8) 26 (34.2) 0.53 (0.36–0.8) 2.1 × 10−3 14.4 51 (66.2) 26 (33.8) 0.54 (0.35–0.82) 3.7 × 10−3 16.0 101 (66.0) 52 (34.0) 0.53 (0.40–0.70) 6.7 × 10−6 15.8 
miR-27b <0.72 79 (91.9) 7 (8.1) 1 (reference)  7.1 75 (93.8) 5 (6.3) 1 (reference)  9.0 154 (92.8) 12 (7.2) 1 (reference)  8.2 
 >0.72 56 (65.9) 29 (34.1) 0.53 (0.36–0.79) 1.6 × 10−3 14.8 66 (74.2) 23 (25.8) 0.66 (0.46–0.93) 1.9 × 10−2 11.1 122 (70.1) 52 (29.9) 0.58 (0.45–0.75) 2.2 × 10−5 13.2 
miR-152 <0.68 77 (91.7) 7 (8.3) 1 (reference)  7.8 82 (92.1) 7 (7.9) 1 (reference)  8.6 159 (91.9) 14 (8.1) 1 (reference)  8.4 
 >0.68 55 (66.3) 28 (33.7) 0.6 (0.41–0.87) 7.2 × 10−3 14.4 56 (70) 24 (30.0) 0.60 (0.42–0.86) 5.3 × 10−3 12.7 111 (68.1) 52 (31.9) 0.59 (0.46–0.75) 3.3 × 10−5 14.1 
miR-20a <0.67 79 (92.9) 6 (7.1) 1 (reference)  7.8 76 (92.7) 6 (7.3) 1 (reference)  9.0 155 (92.8) 12 (7.2) 1 (reference)  8.5 
 >0.67 55 (65.5) 29 (34.5) 0.63 (0.42–0.95) 2.7 × 10−2 14.3 66 (71.0) 27 (29.0) 0.58 (0.41–0.83) 2.5 × 10−3 12.9 121 (68.4) 56 (31.6) 0.62 (0.48–0.79) 1.5 × 10−4 14.1 
miR-30a-5p <0.61 75 (89.3) 9 (10.7) 1 (reference)  8.7 76 (90.5) 8 (9.5) 1 (reference)  8.6 151 (89.9) 17 (10.1) 1 (reference)  8.7 
 >0.61 55 (65.5) 29 (34.5) 0.69 (0.48–1.01) 5.5 × 10−2 9.6 66 (72.5) 25 (27.5) 0.58 (0.41–0.83) 2.6 × 10−3 11.6 121 (69.1) 54 (30.9) 0.63 (0.49–0.81) 3.2 × 10−4 11.5 
miR-148a <0.74 79 (92.9) 6 (7.1) 1 (reference)  8.2 80 (90.9) 8 (9.1) 1 (reference)  9.6 159 (91.9) 14 (8.1) 1 (reference)  8.8 
 >0.74 55 (64.7) 30 (35.3) 0.57 (0.39–0.84) 3.8 × 10−3 14.3 59 (70.2) 25 (29.8) 0.72 (0.49–1.04) 8.3 × 10−2 12.9 114 (67.5) 55 (32.5) 0.63 (0.49–0.81) 4.1 × 10−4 14.1 
miR-145 <0.62 77 (91.7) 7 (8.3) 1 (reference)  8.2 73 (90.1) 8 (9.9) 1 (reference)  9.6 150 (90.9) 15 (9.1) 1 (reference)  8.8 
 >0.62 54 (65.1) 29 (34.9) 0.62 (0.42–0.92) 1.7 × 10−2 15.7 67 (72.0) 26 (28.0) 0.70 (0.49–0.99) 4.3 × 10−2 12.4 121 (68.8) 55 (31.3) 0.64 (0.49–0.82) 4.4 × 10−4 12.7 
miR-92a <0.63 77 (88.5) 10 (11.5) 1 (reference)  8.6 73 (91.3) 7 (8.8) 1 (reference)  8.9 150 (89.8) 17 (10.2) 1 (reference)  8.7 
 >0.63 59 (67.8) 28 (32.2) 0.70 (0.48–1.04) 7.6 × 10−2 14.1 70 (72.9) 26 (27.1) 0.66 (0.47–0.92) 1.6 × 10−2 11.1 129 (70.5) 54 (29.5) 0.67 (0.53–0.86) 1.6 × 10−3 12.4 
let-7c <0.66 74 (89.2) 9 (10.8) 1 (reference)  8.4 66 (88.0) 9 (12.0) 1 (reference)  9.2 140 (88.6) 18 (11.4) 1 (reference)  8.6 
 >0.66 56 (68.3) 26 (31.7) 0.65 (0.44–0.94) 2.4 × 10−2 14.3 68 (73.9) 24 (26.1) 0.72 (0.50–1.04) 7.7 × 10−2 11.6 124 (71.3) 50 (28.7) 0.68 (0.53–0.87) 2.6 × 10−3 13.4 
let-7b <0.60 78 (90.7) 8 (9.3) 1 (reference)  8.6 89 (92.7) 7 (7.3) 1 (reference)  9.3 167 (91.8) 15 (8.2) 1 (reference)  9.0 
 >0.60 55 (64.7) 30 (35.3) 0.65 (0.44–0.97) 3.5 × 10−2 11.3 53 (67.1) 26 (32.9) 0.67 (0.46–0.98) 3.9 × 10−2 13.7 108 (65.9) 56 (34.2) 0.69 (0.53–0.89) 4.5 × 10−3 12.7 

aExpression levels were calculated using the 2−ΔΔCt method. ΔCt was defined as the raw Ct value for a specific miRNA (or spike-in control miRNAs) in an individual sample minus the mean raw Ct value of this miRNA (or spike-in control miRNAs) in all the tested samples. ΔΔCt was defined as the difference between ΔCt of the specific miRNA and ΔCt of the spike-in control miRNAs.

bAdjusted for age, sex, smoking status, clinical stage, and treatment regimen.

Association of serum miRNA expression with NSCLC stage

As clinical stage is an established prognostic factor for NSCLC, serum miRNA expression was compared between patients with stage III and stage IV NSCLC. Among the 35 miRNAs analyzed, the expression levels of 11 miRNAs were significantly different between stage III and stage IV (Supplementary Material: Table S3). Among these 11 miRNAs, 9 were overlapping with the significant miRNAs associated with survival (Table 2). We included stage as an adjustment factor in our analysis of 2-year survival.

Prediction of NSCLC survival by 17-miRNA risk score

The combined effects of these single serum miRNAs significantly associated with survival were investigated by calculating a risk score using the linear combination of expression levels of these 17 miRNAs for each individual on the training data set (Table 3). When the same cutoff was applied to the testing set, similar to the training set, patients with high-risk scores exhibited significantly increased risk of death compared with those with low-risk scores (HR = 2.71; 95% CI: 1.65–4.45; Table 3). The corresponding risk of death in the combined dataset was 2.46 (95% CI, 1.77–3.44). Kaplan–Meier 2-year survival curve showed that patients with the high-risk scores had a MST of 7.8 months compared with 15.6 months in the low risk group (Fig. 2B).

Table 3.

Serum miRNA 17-marker risk score associated with 2-year survival in patients with advanced NSCLC from the study population recruited at the University of Texas MD Anderson Cancer Center between 2002 and 2009

Risk score2-Year Death, n%2-Year Survival, n%Adjusted HRa (95% CI)PMST
Training Set Low 37 (64.9) 20 (35.1) 1 (reference) 0.002 14.3 
 High 56 (100.0) 0 (0.0) 2.18 (1.32–3.59)  7.8 
Testing Set Low 42 (62.7) 25 (37.3) 1 (reference) 8.0 × 10−5 17.4 
 High 52 (100.0) 0 (0.0) 2.71 (1.65–4.45)  7.6 
Combined Set Low 79 (63.7) 45 (36.3) 1 (reference) 1.1 × 10−7 15.6 
 High 108 (100.0) 0 (0.0) 2.46 (1.77–3.44)  7.8 
Risk score2-Year Death, n%2-Year Survival, n%Adjusted HRa (95% CI)PMST
Training Set Low 37 (64.9) 20 (35.1) 1 (reference) 0.002 14.3 
 High 56 (100.0) 0 (0.0) 2.18 (1.32–3.59)  7.8 
Testing Set Low 42 (62.7) 25 (37.3) 1 (reference) 8.0 × 10−5 17.4 
 High 52 (100.0) 0 (0.0) 2.71 (1.65–4.45)  7.6 
Combined Set Low 79 (63.7) 45 (36.3) 1 (reference) 1.1 × 10−7 15.6 
 High 108 (100.0) 0 (0.0) 2.46 (1.77–3.44)  7.8 

aAdjusted for age, sex, smoking status, clinical stage, and treatment regimen.

The prognosis values of 2-year survival were expressed as C-index with 100 times of bootstrap in the training, testing, and combined datasets (Table 4). In the combined dataset, the C-index was 0.57 (95% CI = 0.54–0.60) for the baseline model (age, gender, and smoking status), 0.64 (95% CI = 0.61–0.66) with addition of disease stage, and 0.69 (95% CI = 0.64–0.72) with further addition of the 17-miRNA risk score (Table 4).

Table 4.

C-index for prognosis models in prediction of 2-year survival among NSCLC study population recruited at the University of Texas MD Anderson Cancer Center between 2002 and 2009

ModelTraining set (95% CI)Testing set (95% CI)Combined set (95% CI)
Baselinea 0.59 (0.55–0.64) 0.56 (0.51–0.61) 0.57 (0.54–0.60) 
Baseline + Stage 0.67 (0.62–0.71) 0.61 (0.57–0.66) 0.64 (0.61–0.66) 
17-miRNA risk score only 0.63 (0.60–0.67) 0.64 (0.58–0.68) 0.64 (0.60–0.66) 
Baseline + Stage + 17-miRNA risk score 0.70 (0.65–0.74) 0.68 (0.63–0.73) 0.69 (0.64–0.72) 
ModelTraining set (95% CI)Testing set (95% CI)Combined set (95% CI)
Baselinea 0.59 (0.55–0.64) 0.56 (0.51–0.61) 0.57 (0.54–0.60) 
Baseline + Stage 0.67 (0.62–0.71) 0.61 (0.57–0.66) 0.64 (0.61–0.66) 
17-miRNA risk score only 0.63 (0.60–0.67) 0.64 (0.58–0.68) 0.64 (0.60–0.66) 
Baseline + Stage + 17-miRNA risk score 0.70 (0.65–0.74) 0.68 (0.63–0.73) 0.69 (0.64–0.72) 

aBaseline model included age, gender, and smoking status.

It is well recognized that patients with cancer of similar clinical characteristics vary considerably in prognosis, treatment response, and survival. Factors currently known to indicate poor prognosis of patients with NSCLC include the presence of pulmonary symptoms, tumor size more than 3 cm, nonsquamous histology, metastases to multiple lymph nodes, and vascular invasion (28). Since these factors are largely qualitative and some of them require invasive procedures to assess, there has been an increasing demand for the discovery of noninvasive biomarkers that accurately predict survival in patients with advanced NSCLC. This would assist in determining personalized treatment or follow-up regimens, and prospectively identify subgroups with unfavorable prognosis to better guide end-of-life management.

In recent years, endogenous circulating miRNAs have attracted extensive research interest because they remain stable and are well protected even under harsh conditions. A number of studies have investigated roles of circulating miRNAs in lung cancer, and most of them focused on identifying miRNA markers for disease screening and early detection (29–32). Only a few recent studies have looked into associations of circulating miRNA expression with NSCLC survival and clinical outcomes (27, 33–36). Hu and colleagues (27) identified a serum miRNA signature for prediction of overall survival in patients with early-stage NSCLC based on genome-wide profiling using RNA-seq in 30 pairs of good- and poor-survival patients, and subsequent qRT-PCR validation. Four serum miRNAs (miR-486, miR-30d, miR-1, and miR-499) were reported to be associated with overall survival of early-stage NSCLC. Several studies using a candidate gene approach have shown that high expression of miR-21 was a risk factor for poor chemosensitivity (36), poor survival (34), and may serve as an indicator of postoperative recurrence (35). A recent study detected cell-free miRNA profiles in pleural effusions from patients with stage IV NSCLC and identified a 5-miRNA (miR-134, miR-151, miR-345, miR-93, and miR-100) risk score as a prognosis marker for overall survival (33). In addition, several studies on associations between circulating miRNA levels and lung cancer stages have been reported (37–39). A study on plasma miRNA indicated that plasma miR-30e-3p was significantly higher in patients with NSCLC with nonresectable tumor compared with those with resectable tumor; but in patients with only resectable tumors, high level of plasma miR-30e-3p was associated with a higher rate of disease-free survival (30). Overall, the results of these studies have been inconsistent, due to small sample size and heterogeneous patient population.

Our current study was one of the largest in NSCLC and we focused on patients with stage III and IV NSCLC. We adopted the pathway-based approach and focused our investigation on miRNAs related with genes in the TGF-β signaling pathway. Our study was designed to include an internal validation by distributing the entire population into a training set and a testing set, from where consistent results were obtained. For survival analysis, we identified serum miRNAs that predicted 2-year survival using Cox regression model (Table 2). Notable differences in survival times were identified for the patients. Having the ability to better predict prognosis—survival benefits greater than 9 months to a year—could be critical in treatment selection and end-of-life management for patients with advanced stage who are typically given a dismal prognosis. miR-16 expression showed the most significant association with 2-year survival in our study. A recent study on resected stage I and III NSCLC tumors revealed the association of high-level miR-16 expression with poor prognosis including disease-free survival and overall survival (40); Our current study was the first to report serum miR-16 in association with advanced stage NSCLC survival. We also showed an MST advantage of 11.1 months in patients with high serum levels of miR-30e-3p in contrast to those with low levels. These results were consistent with previously reported findings that plasma miR-30e-3p was associated with favorable disease-free survival (30). In addition, our results highlighted the significance of the let-7 family, one of the most studied tumor suppressor miRNA families involved in lung cancer, with consistent upregulation observed in tissues for those with a favorable survival (10, 23, 41). Several other tissue-based investigations have also suggested miR-20 and miR-106a (41) as lung cancer prediction factors, which was consistent with our results in serum.

In addition to individual miRNA analysis, we also evaluated the collective prognosis value of multiple miRNAs by creating a risk score analysis for the 17 identified TGF-β pathway-enriched miRNAs that were significantly associated with 2-year survival. When the patient population was stratified by this combined risk score, a notable, highly significant MST advantage of 7.8 months was observed between the low risk group and the high risk group. To quantitatively present the prognosis power of the 17-miRNA risk score, the C-index for survival prediction of the study population was increased to 0.69 when the risk score was additionally included into the model, compared with 0.64 when only age, gender, and smoking status and stage were included. Additional biomarkers are needed to further improve the prediction efficiency for clinical applications.

There were a few limitations in our study. First, the cellular pathways are redundant and are often coregulated. It is possible that our results based on the TGF-β pathway and the miRNA regulation of this pathway could have instead identified effects of serum miRNA expression on other cellular pathways. In addition, a majority of miRNAs have a suite of potential binding targets raising the possibility of pleiotropic effects. Second, we could not determine the miRNA releasing mechanisms and cells of origin. Our results indicated that at least some of the significant serum miRNAs could be derived from tumor cells as they are associated with both disease stage progression and patient survival. However, it is also possible that serum miRNAs come from other origins such as inflammatory cells, endothelium cells, or blood cells. For example, previous studies on circulating miRNAs have suggested the effect of hemolysis on specific miRNA (e.g., miR-16) expression, which may limit the potential use of these miRNAs in the clinical setting (42). To address this issue, we measured hemoglobin levels in 30 randomly selected serum samples (10 each from the low, medium, and high survival groups). In all the samples tested, the hemoglobin levels were below 15 mg/dL, lower than the previously reported level of 25 mg/dL that started to affect the measurement of serum miRNA (42), suggesting the confounding effect of hemolysis in this current study is minimal. Third, we employed TargetScan to focus on prediction of the canonical 3′UTR miRNA binding; however, noncanonical miRNA binding sites in the coding and other regions are also capable of regulating gene expression (43–49). Future studies that include noncanonical miRNA binding sites would complement our current analysis. Finally, although we divided our sample set into a training and testing set, potential systematic biases still could be present in the sample set because the same overall cohort was used. Validation of our reported miRNA candidates in an independent external patient cohort is warranted.

In conclusion, our study was built on a large population of 391 advanced-stage NSCLC cases, which were distributed into 3 datasets for qualitative whole-genome serum miRNA profiling using a small-scale discovery set, followed by individual qRT-PCR assay evaluation on a training set and a testing set. Seventeen serum miRNAs were able to predict 2-year survival in an analysis using the continuous model, and a risk score was constructed on the basis of these 17 miRNAs, which would potentially predict 2-year survival for late-stage NSCLC patients. If validated in an independent population, these serum miRNAs are promising to be used as noninvasive prognostic biomarkers in clinical applications, hence benefiting disease and treatment management for patients with advanced-stage NSCLC.

No potential conflicts of interest were disclosed.

Conception and design: Y. Wang, J. Gu, X. Wu

Development of methodology: Y. Wang, J. Gu, M.A.T. Hildebrandt, X. Wu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Wang, J.A. Roth, X. Wu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Wang, M.A.T. Hildebrandt, Y. Ye, X. Wu

Writing, review, and/or revision of the manuscript: Y. Wang, J. Gu, J.A. Roth, M.A.T. Hildebrandt, S.M. Lippman, Y. Ye, J.D. Minna, X. Wu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.M. Lippman, X. Wu

Study supervision: J. Gu, J.A. Roth, X. Wu

This work was supported in part by National Cancer Institute (R01 CA111646, P50 CA070907, and R01 CA55769). Additional funding was provided by MD Anderson Research Trust and MD Anderson institutional support for the Center for Translational and Public Health Genomics.

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.

1.
American Cancer Society
: 
Cancer facts and figures
.
Atlanta, GA
,
American Cancer Society
2012
.
2.
Esteban
E
,
Casillas
M
,
Cassinello
A
. 
Pemetrexed in first-line treatment of non–small cell lung cancer
.
Cancer Treat Rev
2009
;
35
:
364
73
.
3.
Gebbia
V
. 
Does an optimal therapeutic sequence exist in advanced non–small cell lung cancer?
Expert Opin Pharmacother
2008
;
9
:
1321
37
.
4.
Schiller
JH
,
Harrington
D
,
Belani
CP
,
Langer
C
,
Sandler
A
,
Krook
J
, et al
Comparison of four chemotherapy regimens for advanced non-small-cell lung cancer
.
New Engl J Med
2002
;
346
:
92
8
.
5.
Bartel
DP
. 
MicroRNAs: Genomics, biogenesis, mechanism, and function
.
Cell
2004
;
116
:
281
97
.
6.
Esquela-Kerscher
A
,
Slack
FJ
. 
Oncomirs–microRNAs with a role in cancer
.
Nat Rev Cancer
2006
;
6
:
259
69
.
7.
Miska
EA
. 
How microRNAs control cell division, differentiation and death
.
Curr Opin Genet Dev
2005
;
15
:
563
8
.
8.
Calin
G
,
Ferracin
M
,
Cimmino
A
, et al
A microRNA signature associated with prognosis and progression in chronic lymphocytic leukemia
.
N Engl J Med
2005
;
353
:
1793
801
.
9.
Lu
J
,
Getz
G
,
Miska
EA
,
Alvarez-Saavedra
E
,
Lamb
J
,
Peck
D
, et al
MicroRNA expression profiles classify human cancers
.
Nature
2005
;
435
:
834
8
.
10.
Yanaihara
N
,
Caplen
N
,
Bowman
E
,
Seike
M
,
Kumamoto
K
,
Yi
M
, et al
Unique microRNA molecular profiles in lung cancer diagnosis and prognosis
.
Cancer Cell
2006
;
9
:
189
98
.
11.
Chen
X
,
Ba
Y
,
Ma
LJ
,
Cai
X
,
Yin
Y
,
Wang
KH
, et al
Characterization of microRNAs in serum: a novel class of biomarkers for diagnosis of cancer and other diseases
.
Cell Res
2008
;
18
:
997
1006
.
12.
Lawrie
CH
,
Gal
S
,
Dunlop
HM
,
Pushkaran
B
,
Liggins
AP
,
Pulford
K
, et al
Detection of elevated levels of tumour-associated microRNAs in serum of patients with diffuse large B-cell lymphoma
.
Br J Haematol
2008
;
141
:
672
5
.
13.
Mitchell
P
,
Parkin
RK
,
Kroh
EM
,
Fritz
BR
,
Wyman
SK
,
Pogosova-Agadjanyan
EL
, et al
Circulating microRNAs as stable bloodbased markers for cancer detection
.
Proc Natl Acad Sci USA
2008
;
105
:
10513
18
.
14.
Brase
JC
,
Wuttig
D
,
Kuner
R
,
Sultmann
H
. 
Serum microRNAs as non-invasive biomarkers for cancer
.
Mol Cancer
2010
;
9
:
306
.
15.
Massague
J
. 
TGF beta in cancer
.
Cell
2008
;
134
:
215
30
.
16.
Meulmeester
E
,
ten Dijke
P
. 
The dynamic roles of TGF-beta in cancer
.
J Pathol
2011
;
223
:
205
18
.
17.
Lin
MB
,
Stewart
DJ
,
Spitz
MR
,
Hildebrandt
MAT
,
Lu
C
,
Lin
J
, et al
Genetic variations in the transforming growth factor-beta pathway as predictors of survival in advanced non–small cell lung cancer
.
Carcinogenesis
2011
;
32
:
1050
6
.
18.
Lewis
BP
,
Burge
CB
,
Bartel
DP
. 
Conserved seed pairing, often flanked by adenosines, indicates that thousands of human genes are microRNA targets
.
Cell
2005
;
120
:
15
20
.
19.
Friedman
RC
,
Farh
KKH
,
Burge
CB
,
Bartel
DP
. 
Most mammalian mRNAs are conserved targets of microRNAs
.
Genome Res
2009
;
19
:
92
105
.
20.
Livak
K
,
Schmittgen
TD
. 
Analysis of relative gene expression data using real-time quantitative PCR and the 2(-DeltaDelta C(T)) method
.
Methods
2001
;
25
:
402
8
.
21.
Liu
HS
,
Zhu
L
,
Liu
BY
,
Yang
L
,
Meng
XX
,
Zhang
W
, et al
Genome-wide microRNA profiles identify miR-378 as a serum biomarker for early detection of gastric cancer
.
Cancer Lett
2012
;
316
:
196
203
.
22.
Zhou
XH
,
Obuchowski
N
,
McClish
D
. 
Thoughts on the ZOM/BWC controversy—replies
.
Acad Radiol
2003
;
10
:
1179
80
.
23.
Takamizawa
J
,
Konishi
H
,
Yanagisawa
K
,
Tomida
S
,
Osada
H
,
Endoh
H
, et al
Reduced expression of the let-7 microRNAs in human lung cancers in association with shortened postoperative survival
.
Cancer Res
2004
;
64
:
3753
6
.
24.
Hayashita
Y
,
Osada
H
,
Tatematsu
Y
,
Yamada
H
,
Yanagisawa
K
,
Tomida
S
, et al
A polycistronic microRNA cluster, miR-17-92, is overexpressed in human lung cancers and enhances cell proliferation
.
Cancer Res
2005
;
65
:
9628
32
.
25.
Lin
QF
,
Mao
WD
,
Shu
YQ
,
Lin
F
,
Liu
SP
,
Shen
H
, et al
A cluster of specified microRNAs in peripheral blood as biomarkers for metastatic non-small-cell lung cancer by stem-loop RT-PCR
.
J Cancer Res Clin
2012
;
138
:
85
93
.
26.
Roybal
JD
,
Zang
Y
,
Ahn
YH
,
Yang
YN
,
Gibbons
DL
,
Baird
BN
, et al
miR-200 inhibits lung adenocarcinoma cell invasion and metastasis by targeting Flt1/VEGFR1
.
Mol Cancer Res
2011
;
9
:
25
35
.
27.
Hu
ZB
,
Chen
X
,
Zhao
Y
,
Tian
T
,
Jin
GF
,
Shu
YQ
, et al
Serum microRNA signatures identified in a genome-wide serum microRNA expression profiling predict survival of non-small-cell lung cancer
.
J Clin Oncol
2010
;
28
:
1721
6
.
28.
National Cancer Institute
. 
Non-small cell lung cancer treatment (PDQ)-Health professional version
. National Cancer Institute (http://www.cancer.gov/cancertopics/pdq/treatment/non-small-cell-lung/healthprofessional).
Accessed May 2013
.
29.
Foss
KM
,
Sima
C
,
Ugolini
D
,
Neri
M
,
Allen
KE
,
Weiss
GJ
. 
miR-1254 and miR-574-5p serum-based microRNA biomarkers for early-stage non–small cell lung cancer
.
J Thorac Oncol
2011
;
6
:
482
8
.
30.
Silva
J
,
Garcia
V
,
Zaballos
A
,
Provencio
M
,
Lombardia
L
,
Almonacid
L
, et al
Vesicle-related microRNAs in plasma of nonsmall cell lung cancer patients and correlation with survival
.
Eur Respir J
2011
;
37
:
617
23
.
31.
Zhang
CN
,
Wang
C
,
Chen
X
,
Yang
CH
,
Li
K
,
Wang
JJ
, et al
Expression profile of microRNAs in serum: a fingerprint for esophageal squamous cell carcinoma
.
Clin Chem
2010
;
56
:
1871
9
.
32.
Zhang
TF
,
Wang
QM
,
Zhao
D
,
Cui
YL
,
Cao
BR
,
Guo
LP
, et al
The oncogenetic role of microRNA-31 as a potential biomarker in oesophageal squamous cell carcinoma
.
Clin Sci
2011
;
121
:
437
47
.
33.
Wang
TT
,
Lv
MM
,
Shen
SN
,
Zhou
S
,
Wang
P
,
Chen
YQ
, et al
Cell-free microRNA expression profiles in malignant effusion associated with patient survival in non–small cell lung cancer
.
PLoS ONE
2012
;
7
:
e43268
.
34.
Wang
ZX
,
Bian
HB
,
Wang
JR
,
Cheng
ZX
,
Wang
KM
,
De
W
. 
Prognostic significance of serum miRNA-21 expression in human non–small cell lung cancer
.
J Surg Oncol
2011
;
104
:
847
51
.
35.
Le
HB
,
Zhu
WY
,
Chen
DD
,
He
JY
,
Huang
YY
,
Liu
XG
, et al
Evaluation of dynamic change of serum miR-21 and miR-24 in pre- and post-operative lung carcinoma patients
.
Med Oncol
2012
;
29
:
3190
7
.
36.
Wei
J
,
Liu
LK
,
Gao
W
,
Zhu
CJ
,
Liu
YQ
,
Cheng
T
, et al
Reduction of plasma microRNA-21 is associated with chemotherapeutic response in patients with non–small cell lung cancer
.
Chinese J Cancer Res
2011
;
23
:
123
8
.
37.
Zheng
DL
,
Haddadin
S
,
Wang
Y
,
Gu
LQ
,
Perry
MC
,
Freter
CE
, et al
Plasma microRNAs as novel biomarkers for early detection of lung cancer
.
Int J Clin Exp Pathol
2011
;
4
:
575
86
.
38.
Ma
YX
,
Tian
ZN
,
Zhang
W
. 
Circulating miR-125b is a novel biomarker for screening non-small-cell lung cancer and predicts poor prognosis
.
J Cancer Res Clin
2012
;
138
:
2045
50
.
39.
Wu
J
,
Yang
T
,
Li
X
,
Yang
Q
,
Liu
R
,
Huang
J
, et al
Alteration of serum miR-206 and miR-133b is associated with lung carcinogenesis induced by 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone
.
Toxicol Appl Pharmacol
2013
;
18
:
002
.
40.
Navarro
A
,
Diaz
T
,
Gallardo
E
,
Vinolas
N
,
Marrades
RM
,
Gel
B
, et al
Prognostic implications of miR-16 expression levels in resected non-small-cell lung cancer
.
J Surg Oncol
2011
;
103
:
411
5
.
41.
Castillejo
A
,
Rothman
N
,
Murta-Nascimento
C
,
Malats
N
,
Garcia-Closas
M
,
Gomez-Martinez
A
, et al
TGFB1 and TGFBR1 polymorphic variants in relationship to bladder cancer risk and prognosis
.
Int J Cancer
2009
;
124
:
608
13
.
42.
McDonald
JS
,
Milosevic
D
,
Reddi
HV
,
Grebe
SK
,
Algeciras-Schimnich
A
. 
Analysis of circulating microRNA: preanalytical and analytical challenges
.
Clin Chem
2011
;
57
:
833
40
.
43.
Novitskiy
SV
,
Pickup
MW
,
Gorska
AE
,
Owens
P
,
Chytil
A
,
Aakre
M
, et al
TGF-beta receptor II loss promotes mammary carcinoma progression by Th17-dependent mechanisms
.
Cancer Discov
2011
;
1
:
430
41
.
44.
Raynal
S
,
Nocentini
S
,
Croisy
A
,
Lawrence
DA
,
Jullien
P
. 
Transforming growth factor-beta 1 enhances the lethal effects of DNA-damaging agents in a human lung-cancer cell line
.
Int J Cancer
1997
;
72
:
356
61
.
45.
Didiano
D
,
Hobert
O
. 
Perfect seed pairing is not a generally reliable predictor for miRNA-target interactions
.
Nat Struct Mol Biol
2006
;
13
:
849
51
.
46.
Jeon
HS
,
Jen
J
. 
TGF-beta signaling and the role of inhibitory smads in non–small cell lung cancer
.
J Thorac Oncol
2010
;
5
:
417
9
.
47.
Masui
T
,
Wakefield
LM
,
Lechner
JF
,
Laveck
MA
,
Sporn
MB
,
Harris
CC
. 
Type-Beta transforming growth-factor is the primary differentiation-inducing serum factor for normal human bronchial epithelial-cells
.
Proc Natl Acad Sci USA
1986
;
83
:
2438
42
.
48.
Masui
T
,
Wakefield
LM
,
Lechner
JF
,
Laveck
MA
,
Gerwin
BI
,
Sporn
MB
, et al
Type-Beta transforming growth-factor induction of squamous differentiation of normal human bronchial epithelial-cells is antagonized by epinephrine and cholera-toxin
.
J Cell Biochem
1986
;
10C
:
118
.
49.
Liu
R
,
Zhang
CN
,
Hu
ZB
,
Li
G
,
Wang
C
,
Yang
CH
, et al
A five-microRNA signature identified from genome-wide serum microRNA expression profiling serves as a fingerprint for gastric cancer diagnosis
.
Eur J Cancer
2011
;
47
:
784
91
.