Circulating microRNA Expression Profiles in Early Stage Non-Small Cell Lung Cancer

Abstract

Circulating micro-RNA (miR) profiles have been proposed as promising diagnostic and prognostic biomarkers for cancer, including lung cancer. We have developed methods to accurately and reproducibly measure microRNA levels in serum and plasma. Here we study paired serum and plasma samples from 220 patients with early stage NSCLC and 220 matched controls. We use qRT-PCR to measure the circulating levels of 30 different miRs that have previously been reported to be differently expressed in lung cancer tissue. Duplicate RNA extractions were performed for 10% of all samples and microRNA measurements were highly correlated among those duplicates. This demonstrates high reproducibility of our assay. The expression of miR-146b, miR-221, let-7a, miR-155, miR-17-5p, miR-27a and miR-106a were significantly reduced in the serum of NSCLC cases while miR-29c was significantly increased. No significant differences were observed in plasma of patients compared to controls. Overall, expression levels in serum did not correlate well with levels in plasma. In secondary analyses, reduced plasma expression of let-7b was modestly associated with worse cancer-specific mortality in all patients and reduced serum expression of miR-223 was modestly associated with cancer-specific mortality in stage IA/B patients. MiR profiles also showed considerable differences comparing African American and European Americans. In summary, we found significant differences in miR expression when comparing cases and controls and find evidence that expression of let-7b is associated with prognosis in NSCLC.

Introduction

Early detection of primary lung cancer is difficult yet important since diagnosis at earlier stages is associated with significantly better survival. Overall, non-small cell lung cancer (NSCLC), which represents about 80% of all lung cancer cases, is a leading cause of cancer death in the developed world. Less than 15% of NSCLC patients live more than 5 years after diagnosis. In comparison, the five-year survival rate is about 33% for cases discovered at stage 2 or earlier. The overall survival of lung cancer patients has not improved over the last few decades. Diagnosis relies heavily on imaging methods, such as chest X-ray, CT- or PET-CT-scan followed by bronchoscopy and biopsy. No blood test is currently available. Also, predictive and prognostic markers – while much in demand – are not yet available for clinical use in lung cancer patients. In experimental laboratories methods based on gene expression profiles in healthy and tumor tissues and studies focusing on copy numbers, or protein levels of EGFR, EGFR mutations, nucleotide excision repair genes and proteins, cell-cycle regulators, and microtubule proteins suggest possible applications in the prognostic and predictive evaluation of individual patients. However, none of these methods have yet matured enough, e.g., through prospective independent validation in randomized trials, to be used in a clinical setting and also do not seem to be useful for primary diagnosis (1).

It is, therefore, of interest that micro-RNA (miR) profiles in tumor tissues are specifically and consistently different from normal tissues (2) and that tumor-specific miR profiles may be reflected in the circulating miR population (3–8). MiRs are short, non-coding single-stranded regulators of gene expression on the level of protein translation and are promising diagnostic and therapeutic targets in deregulated pathways in inflammation, immunity, and cancer (8). In lung cancer, a number of studies have demonstrated that miR expression patterns in cancer tissues are different from the miR profiles found in normal lung tissue (reviewed in (9)). In the present study, our goal was to select 30 miRs that have been reported to be altered in lung tumors and determine of the circulating levels of those miRs were altered when comparing cases and controls.

Previous reports have examined miR expression levels in either serum or plasma of patients. There is no strong agreement as to whether serum or plasma would be superior, therefore we chose to use both serum and plasma. Here we apply quantitative RT-PCR measurements of a panel of different lung cancer-relevant miRs in RNA isolated from both serum and heparinized plasma from a collection of 220 early stage NCSLC patients and 220 matched population controls. The resulting miR profiles are correlated with diagnosis, survival, histology, race, gender, and stage in sera and plasma samples.

Materials and Methods

Study Population

Lung cancer patients were recruited from hospitals in the greater Baltimore metropolitan area from 1998 to 2005. Eligibility criteria for both cases and controls has been previously described (10). Briefly, eligibility criteria included being a United States citizen and residing in Baltimore City or adjacent counties of Maryland or the Maryland Eastern Shore; English-speaking; not currently taking antibiotics or steroid medications; and no known diagnosis of HIV, hepatitis C or hepatitis B. All study participants were interviewed. Blood was obtained by the interviewers and frozen blood components were sent to the Laboratory of Human Carcinogenesis at the National Cancer Institute within 24 hours of collection and stored at −80°C until use. A total of 220 cases with 220 population controls were included (Table 1). For each case and control serum and plasma samples were available. Participants were self-reported African Americans (AA) or European Americans (EA). All cases had histologically confirmed non–small cell lung cancer (NSCLC) and were enrolled within 24 mo of diagnosis. Population controls were recruited via records from the Maryland Department of Motor Vehicles and frequency matched to cases by age and gender. The control population reflects smoking prevalence of the general population of Maryland (11). Ten percent of the samples were randomly selected as duplicates and replicate aliquots of serum/plasma were made for these samples prior to RNA isolation. One plasma sample (a patient sample) was not included because it was coagulated. Thus a total of 967 samples were characterized in this study. Survival time for the was determined by searching the National Death Index (www.cdc.gov/nchs/ndi.htm) through 12/31/2006. The study was approved by the Institutional Review Boards of the National Cancer Institute, the University of Maryland Medical System, the Baltimore VA Medical Center, the Johns Hopkins University School of Medicine, Sina` i Hospital, MedStar Research Institute, and the Research Ethics Committee of Bon Secours Baltimore Health System.

Table 1.

Characteristics of the study population

	Patients	Controls	p-value1
# serum samples	220	220
# plasma samples	220	220
Age at sampling, years			0.864
- Range	37–88	33–86
- Mean (SD)	67.5 (9.0)	67.7 (9.3)
Sex, n (%)			0.924
- Women	114 (52)	112 (51)
- Men	106 (48)	108 (49)
Race, n (%)			0.558
- White	177 (80)	171 (78)
- Black	43 (20)	49 (22)
Smoking, n (%)			<0.001
- Current	102 (46)	17 (8)
- Former	104 (47)	107 (49)
- Never	14 (6)	96 (44)
Tumor Classification, n (%)			NA
- Adenocarcinoma	99 (45)
- Squamous	58 (26)
- NSCLC	35 (16)
- Other	28 (13)
Tumor Staging, n (%)			NA
- Stage I+IA	133 (60)
- Stage IB	47 (21)
- Stage II+IIA	21 (10)
- Stage IIB	19 (9)

¹Student’s t-Test was used for assessing age and Fisher’s exact test was used for sex, race and smoking.

RNA purification

Total RNA purification kit (Norgen Biotek Corp., Ontario, Canada) was used to purify RNA from 100 µL plasma and serum samples according to the instructions of the manufacturer with small modifications: 10 mM dithiothreitol and C. elegans synthetic miR-39, -54, and -238 (IDT, Coralville, Iowa), each at 0.13 pM were added to a volume of kit lysis buffer sufficient for all the samples. This volume was then aliquotted out into 4 mL portions and kept at −20 °C until used. Also, 1 uL of RNAse inhibitor (ABI, 20 U/µL) was added to every elution tube before elution of RNA. Purified RNA was kept at −80 °C before being used for reverse transcription. The low amounts of RNA extractable from 100 µL serum or plasma were not reliably measured by UV-absorbance and thus fixed volumes rather than fixed amounts of RNA were used for the initial RT reaction in accordance with other studies (5).

Heparinase treatment

Heparin is known to inhibit both reverse transcriptase and polymerases (12–14) and appears to be co-purified with the RNA. Heparinase I-treatment of RNA preparations from heparin plasma was therefore carried out before reverse transcription. Twenty µL of purified RNA was transferred to a microtube containing 0.5 µL RNAse inhibitor and 2 µL of heparinase I (Sigma H2519, dissolved in recommended buffer to 1 mg/ml, stored at −20 °C) was added to the solution. After incubation for 1 hr at room tp the solution was used without further treatment for the RT and q-PCR reactions. Prolonged heparinase incubation times did not increase detected miR levels (data not shown).

Reverse transcription

The RT-primer-mix consisted of equal volumes of each of 32 different 5x RT miR-specific stem-loop primers (Applied Biosystems (ABI), Foster City, CA) (supplemental Table S1). Reverse transcription reaction volumes were 10 µL using 1 µL Multiscribe, 3 µL RT-primer-mix, 1 µL 10 X buffer, 0.2 µL 100 mM dNTPs and 0.15 µL RNAse inhibitor, all reagents from ABI. To this was added 4.65 µL RNA purified from sera or plasma. In the latter case the sample had been pretretated with heparinase I as described above. Reverse transcription was performed using a standard protocol (16 °C, 30 min; 42°C, 30 min; 85°C 5 min; hold at 4°C). Reverse transcribed samples were kept at −20°C until used.

Preamplification

Specific target amplification of the cDNA was accomplished using the TaqMan PreAmp master mix and a mix of the TaqMan MicroRNA Assays (ABI) consisting of equal volumes of the 32 different 20x assays diluted with 1x TE buffer to a final concentration of 0.2x. Preamplification mixtures (10 µL) contained cDNA (diluted 1:4 with H₂O), 2.5 µL, mixed with 5 µL 2x TaqMan PreAmp master mix and 2.5 µL of the 0.2x TaqMan miR-assay mix. After 10 min at 95°C the reactions were subjected to 16 cycles of (95°C, 15 s; 60°C, 4 min) and then held at 4°C. Samples were diluted 1:5 with H₂O before next step (q-PCR).

Quantitative PCR

Preamplified samples (diluted 5x with H₂O) and miR TaqMan 20x assays were applied to primed 96.96 dynamic array chips using loading and assay reagents according to the manufacturer (Fluidigm Corp., South San Francisco, CA). All miR-assays were performed in triplicate. In total, each chip contained 96 samples that were mixed with 3x32 different miR assays. After loading the reaction chambers using the integrated fluid circuit HX controller from Fluidigm the real-time PCR including image capture after each cycle was performed in a BioMark MX real-time PCR system (Fluidigm) using single probe (FAM-MGB, reference: ROX) settings and the default hot-start protocol with 40 cycles. Data processing took place using the Fluidigm real-time PCR analysis software (v. 2.1.1). As an overall quality measure data sets that yielded standard deviations of ≥1 CT value for CTs below 25 in more than 2% of the triplicate measurements led to rejection of the whole chip. No individual data with SDs for the triplicates above 1 CT value were included.

Data Handling and Analysis

All samples and data were handled in a blinded fashion except that plasma and serum samples were processed differently and thus this information was known. Each chip array contained a random 50:50 mixture of 96 samples and controls. In the array analyses for CT-values global settings of the software were disabled and CTs were determined for each miR using individually set values based on inspection of the amplification curves. These settings were then used for data from all the chip runs. Input data were pruned for failed reactions, reactions with standard deviations of the triplicate analyses above or equal to 1 CT, and reactions with no signal. In the absence of a well-documented stably expressed endogenous circulating miRs to serve as normalization controls, two C. elegans synthetic miRs (Cel-miR-54 and Cel-miR-238) that were added to the RNA purification lysis buffer were used as normalization controls as in earlier studies (5;7). Subsequently, three different normalization procedures were applied. First, linear regression was used to measure differences in the average ΔCT value for each microRNA between plates, adjusting for case/control status. Plates were then adjusted to have the same average ΔCT values for each miR as a reference plate by adding the plate specific coefficients from the linear regression models to the microRNA ΔCT values. Secondly, median normalization was performed where all expression values are transformed to produce a constant median. Finally, miR-223 was used for normalization where the average CT-values were subtracted from the average miR-223 value for each sample (in this case the Cel-miRs were not used for normalization). Average correlation coefficients of sample duplicates were R = 0.77 for the plate regression method and R = 0.65 and 0.67 for the alternative normalization methods (median-based and miR-223-based normalization methods, respectively). The plate regression method therefore was used for all analyses.

Graphs of miR-expression and statistical comparison of the distribution of values in cases and population controls using unpaired t-tests as well as Spearman rank analysis were accomplished by Graphpad Prism v. 5.00 (Graphpad Software Inc., San Diego, CA). Class comparisons and multivariate predictor analysis (class predictions) of all miRs were performed with BRB-ArrayTools V. 3.8.0 (http://linus.nci.nih.gov/pilot/index.htm). To assess the association of miRNA expression with survival, we categorized each RNA as high or low using median value in cases as the cutoff. WINSTAT 2007 was used for the survival (Kaplan-Meier) analysis.

Results

Measuring micro-RNA in plasma and serum

Thirty microRNAs were selected based on published studies that reported differential expression in lung tumors or altered circulating levels of these microRNAs in lung cancer patients (supplemental Table S1). All 967 samples were analyzed for levels of the 30 different miRs in triplicate by qRT-PCR using TaqMan assays and medium density microfluidic dynamic arrays (96×96) (15). No or very weak and variable signals were obtained from the assays for miR-1, 10a, 17-3p, 128b, 137, 200b, 205, 215, and 499 and these miRs were omitted for all futher analysis. Linear regression models were used to measure and correct for batch effects of the plate to plate variability (supplemental Fig. S1). The reproducibility of the entire procedure from RNA extraction to normalized CT-values was then assessed using data from samples that were processed in duplicate (44 plasma and 44 sera). The duplicate samples were run on separate plates than their corresponding pairs and therefore reflect the true reproducibility measurements. The correlations between first and second analysis for the 21 different miRs that were measurable are shown in supplemental Fig. S2. Correlations are linear with Spearman rank regression coefficients above 0.70 except for 2 miRs (let-7a and miR-155). As expected, the reproducibility is less at lower miR expression values. Previous studies have evaluated miRs-16 and -223 as biological (endogenous) normalizers (5) and both were included and evaluated in the present study. However, miR-16 has also been reported to be upregulated in cancer tissue from stage I NSCLC patients with recurrence after resection (16) and we do find miR-16 differentially expressed between cases and controls. Additionally, miR-223 has been reported to be differently expressed in small cell lung and various other cancers (17;18) in addition to being present at high levels in platelets (19) as is miR-16 (20). For these reasons, we chose not to use miR-16 and miR-223 as normalizers.

Correlation between serum and plasma miR measurements

We found a very low correlation comparing miRNA expression levels in serum to plasma samples from the same individuals. While the overall ranking of the expression levels of miRs in individual samples was similar comparing both plasma and serum (correlation for the set of 21 miRs was R>0.9 in most individuals (data not shown) as reported before (5)) none of the individual miRs showed any appreciable correlation between serum and plasma comparing each miR across all individuals (Fig. 1 and supplemental Fig. S3). All correlation coefficients except one were below 0.2 in the serum-plasma plots. MiR-16, a miR known to increase upon platelet storage (20), was 1.6 fold higher in serum than in plasma (p<0.0001, t-test).

Figure 1.

Serum and plasma correlation of (A) miR-16 and (B) miR-150 expression. Normalized CT-values of miR from plasma (X-axes) and serum (Y-axes) are plotted for all the sample sets.

Low concordance between serum and plasma miR-levels could be due to different quantities of RNA obtained from sera and plasma and variable yields of total RNA from serum and plasma of individual patients. We therefore examined the systematic variation by plotting mean normalized CT-values of the individual miRs in all sera compared with the corresponding values in all plasma samples (supplemental Fig. S4). All mean values were significantly different between serum and plasma. Except for miRs 16, 21, 93, 150, and 486 the CT-values were higher in the plasma samples. Also, the range of values was in most cases larger in plasma samples.

Micro-RNA profiles in cases and controls

Our main goal was to determine if circulating miR expression was different between the 220 NSCLC cases and 220 controls (Table 1). Cases and controls were similar with regards to age, sex and race while cases had increased rates of smoking. The miR expression profiles in sera and plasma were compared between NSCLC samples and controls. Results (supplemental Tables S2 and S3) are summarized in Table 2. Of the 21 assayable miRs we found 7 miRs in serum that had a statistically significant (p<0.05) lower expression in cases (miRs 146b, 221, let7a, 155, 17-5p, 27a, and 106a) while one was significantly increased (miR-29c). Only miR-146b and miR-221, however, had false discovery rates (FDR) below 5% when adjusting for multiple comparison. Despite the significant differences in miR expression levels between cases and controls in serum the expression profiles could not distinguish cases from controls accurately. Using the panel of miRs measured in serum the best predictive model has a performance (accuracy) of only 57–60% (AUC ROC is 0.602) based on a leave-one-out cross-validation method. In plasma no significant associations of case-control status with any miR were found.

Table 2.

Differently expressed (p<0.05) miRs in cases and controls in serum when testing 21 different miRs. FDR, false discovery rate. No significant differences were found in the data set from plasma samples.

Serum miR	Rank	p-value	FDR	Fold-change (case vs. ctrl)	Increased or decreased in cases
146b	1	0.00140	0.02064	0.79173	Decreased
221	2	0.00197	0.02064	0.77816	Decreased
let-7a	3	0.00775	0.05428	0.73502	Decreased
155	4	0.01143	0.06001	0.77129	Decreased
17-5p	5	0.02083	0.08748	0.82339	Decreased
29c	6	0.02800	0.09800	1.17242	Increased
27a	7	0.03742	0.11227	0.87237	Decreased
106a	8	0.04982	0.13078	0.87347	Decreased
Plasma
miR
-	-	NS	-	-

Survival and micro-RNA profiles

Secondary analysis of the data focused on the association of miR-expression with survival. In plasma, reduced expression of miR-let-7b (based on median expression) was significantly associated with worse survival (p=0.025) (Fig. 2A). No miRs expressed in sera were found to be associated with survival.

Figure 2.

Survival analysis. A, Low plasma let-7b expression in patients with poor survival (p=0.025). B, Low serum miR-223 expression in the poor survival group of stage IA/IB patients (p=0.048). The survival analysis uses the median expression as a cut-off between high vs. low expression.

Subgroup analysis: Tumor histology, staging, and micro-RNA profiles

Differences in tumor histology reflect important biological differences which may be reflected in miR expression patterns. Therefore, we compared expression patterns between different histologies and performed stratified analysis on diagnosis and survival for each histology separately (supplemental Table S4). A number of cases did not have detailed histological information and thus for this subgroup analysis only 99 cases with adenocarcinoma histology and 58 cases with squamous histology were included. We found no miRs that were differentially expressed between these two histologies in either serum or plasma (data not shown). Cases with adenocarcinoma histology had reduced miR-146b expression in serum (p=0.04) compared to controls, but miR expression patterns were unable to accurately distinguish cases with adenocarcinoma histology from controls (52–67% accuracy with AUC for the ROC curve being 0.55). Likewise, for cases with squamous cell histologies, miR-221 and miR-let-7a expression were significantly reduced (p=0.04 and p=0.01, respectively) in cases versus controls, although their expression could not accurately distinguish cases from controls (59–72%, depending on the classifier model used and with the AUC of the ROC curve being 0.57). No miRs were differentially expressed in plasma comparing cases and controls in either histology subgroup.

Subgroup analysis: Smoking, gender, stage, and race

Smoking is strongly linked to increased lung cancer rates. We were interested if smoking altered circulating miR levels and therefore may confound associations between lung cancer and circulating miR expression. Self reported smoking status was available for all cases and controls in this cohort. We found that smoking status (current vs. former/never or ever vs. never) was not significantly associated with any miRs in plasma or serum data sets when using all 440 participants after adjusting for cancer versus control status.

We also analyzed differences between circulating levels of miRNAs comparing both different genders and race. No miR was found to be expressed at different levels comparing females and males in either serum or plasma. Approximately 80% of the cases examined were European Americans (EA) and 20% African Americans (AA) (Table 1). We found that 14 different miRs were significantly reduced in AA (p<0.05) compared with EA in plasma, and 10 of these gave FDRs below 0.05 (Table 3). But for serum data, only miR-155 expression was modestly different and was slightly reduced in AA (p<0.05). We also performed a stratified analysis to identify miRs associated with diagnosis in the AA and EA groups separately (supplemental Table S5) but none of the identified differences in these individual groups retained significance after correcting for multiple testing and only models with poor accuracy (50–60%) could be built based on these data.

Table 3.

Differently expressed (p<0.05) miRs in AA and EA in serum and plasma. FDR, false discovery rate.

Serum miR	Rank	p-value	FDR	Fold-change (AA vs. EA)	Increased or decreased in AA
155	1	0.02891	0.607	1.32	Increased
Plasma
miR
486	1	0.00081	0.017	0.64	Decreased
16	2	0.00196	0.0206	0.64	Decreased
221	3	0.00672	0.0368	0.61	Decreased
93	4	0.00701	0.0368	0.65	Decreased
223	5	0.00995	0.0376	0.68	Decreased
24	6	0.01090	0.0376	0.64	Decreased
30a-5p	7	0.01254	0.0376	0.71	Decreased
let-7b	8	0.01791	0.0448	0.68	Decreased
197	9	0.01955	0.0448	0.73	Decreased
let-7c	10	0.02136	0.0448	0.68	Decreased
191	11	0.03263	0.0623	0.69	Decreased
146b	12	0.04578	0.0729	0.72	Decreased
27a	13	0.04834	0.0729	0.73	Decreased
17-5p	14	0.04861	0.0729	0.68	Decreased

Next, correlations with staging (i.e., stage IA,B versus stage IIA,B) were examined (supplemental Table S6). Only the stage I sample group was large enough to allow stratifying on histology. Among consistent findings were that increased miR-16 expression in serum was associated with survival of stage IA/B patients with adenocarcinoma histology (p=0.049, log rank test). Reduced miR-223 in serum stage I A/B patients was also associated with worse survival (p=0.048) (Fig. 2B).

Discussion

The present study is the first large study of circulating miRs in clinically well-described cases and controls that focus on early stage NSCLC and directly compares serum and plasma. We were able to reproducibly and accurately measure miR concentrations in plasma and serum. The main findings of this study are that a set of 7 miRs display 10–20% reduced expression levels in the serum of NSCLC cases compared with case-controls while one (miR-29c) is about 17% increased. Analyses of miRs present in plasma do not show these differences. Many of the reported miR-alterations in lung cancer tissues, in lung cancer of different histology, and in the circulation of lung cancer patients (9;21–25) are not reflected by the circulating miR profiles in our study. We also find in plasma that the expression of 14 miRs is significantly reduced in AA compared to EA samples while one (miR-155) is increased.

Importantly we also find that the miR-expression values for the individual sample sets of paired serum and plasma do not correlate well demonstrating that serum and plasma may be very different in miR expression levels. Plasma exhibited a broader range of CT-values and (for 16 out of 21 miRs) significantly increased expression (mean levels of normalized CT-values). Serum samples originate from clotted whole blood while plasma is obtained by centrifugation of anticoagulated whole blood. Serum and plasma were obtained simultaneously for all participants and all RNA isolations were performed by the same individuals during a short period of time, therefore we expect little confounding of our measurements by variability of blood draw. Serum and plasma samples were processed identically except that the plasma RNA preparation was treated with heparinase in order to avoid inhibition of reverse transcriptase and polymerase in the ensuing reactions. We observed no degradation of the exogenously added synthetic miRs during this step. In preliminary experiments we found comparable yields and a similar quantitative ranking of miRs based on the qRT-PCR signals in EDTA plasma and heparin plasma where only RNA from the latter samples were treated with heparinase. Duplicate samples that were interspersed randomly in the purification batches showed good correlation of the individual miR-expression values and thus do not suggest a method-related preanalytical cause of the differences between plasma and serum samples. Looking at possible biological contributors to the differences it may be noted that during the coagulation process a host of enzymatic activities are released from activated neutrophils and platelets, i.e., there is some degree of cell lysis which could release miR-containing vesicles from various cells including platelets and possibly circulating tumor cells and exosomes. This would not happen in plasma samples where, conversely, such cells are spun down during plasma preparation (exosomes will remain in the plasma due to their small size). The number of circulating tumor cells (estimated to be on the order of 1–10 cells per mL in metastatic disease (about 1 out of 10⁵–10⁷ blood mononuclear cells) and even lower in cases with localized tumors (26;27)) would not, however, be expected to contribute much to the total miR pool in serum even if they all lysed. Rather, part of the exosome population may be trapped in the coagulum of serum samples. This may explain the lower levels found for many of the miRs in sera because it is thought that the bulk of the circulating miRs present in serum or plasma is localized in exosome particles where they are protected against degradation (28;29). Together, these differences between serum and plasma probably account for some of the lack of correlation of miR expression levels between these two sample types. Overall, the results clearly indicate that the outcomes of miR-expression studies using sera and plasma as the starting material for RNA purification cannot readily be compared and that a reliable endogenous miR for normalization of circulating miR-levels is yet to be found.

The 30 miRs that were selected for the present study all have published links to lung cancer and were carefully selected with the hope of identifying those miRNAs most likely to be associated with lung cancer risk. But with this selection process, the majority of the circulating miRs are untested and we can make no claims about the association of these other miRs with lung cancer risk. Another limitations of this study is the fact that serum and plasma samples have been stored at −80 °C for up to 12 years. Therefore, if miRs are not stable in these conditions, it would likely decrease the accuracy of the measurements and decrease the predictive accuracy. MiRs have been shown to be much more stable than mRNA so we do not expect that this is a major issue.

We found miR-146b and miR-221 to be significantly decreased in serum from lung cancer cases irrespectively of stage or histology. MiR-146b was previously reported to be increased in cancerous tissues (21;21) and miR-221 was reported to be either increased in tumor cell lines (30;31) or decreased in lung cancer tissue (23). Also, in sera, miR-29c was distinguished as one of the few miRs with elevated expression in NSCLC cases. This miR targets DNA methyltransferases and thus represses oncogenic DNA methylation (32). The increased expression levels may be speculated to reflect an increased systemic concentration of this miR in response to the presence of cancer processes.

Interestingly, miR-let-7b in plasma emerged as an indicator of survival. Thus, low levels of plasma-miR-let-7b were statistically significantly associated with poor survival. The let-7 family is known to be implicated in suppression of k-Ras and c-Myc oncogenes (33) and these anti-oncogenic miRs are consistently reported to be down-regulated in lung cancer tissues (22;25;34;35). While down-regulated miRs in tissues a priori would be expected to have minimal consequences for the levels of circulating miRs, the lower levels of plasma let-7b found in patients with worse prognosis may reflect a systemic, relative deficiency of this miR predisposing to a more aggressive disease course in individuals with low let-7b if they contract lung cancer. It is noteworthy that let-7b is one of the miRs that are highly expressed in platelets (20) and this may obscure the prognostic value of measuring let-7b expression levels in sera. Accordingly, we do not find the correlation in the serum data.

Prognostic biomarkers for stage I patients may help identify those at high risk of disease progression and provide an opportunity to intervene at an earlier time. Reduced expression of miR-223 in serum was associated with worse survival outcomes in TNM stage I patients. The expression of miR-223 has been shown to be deregulated in a variety of cancers, including small cell lung cancer (17), and can target the cell cycle regulator E2F1 (36) and a microtubule regulatory protein Stathmin1 (37). While our findings require validation in additional cohorts, our data suggests that there is potential for miR-223 as a prognostic biomarker for stage I NSCLC.

Further studies are needed to validate and address the potential diagnostic role of miR profiles in lung cancer, including later stages, and their possible use in stratifying patients with respect to therapeutic and prognostic parameters. Also, more focus on determining the contents of circulating tumor cells or lung-derived exosome miRs (38;39) as opposed to total circulating miR profiles may be a valuable avenue for increasing diagnostic specificity in future studies.

In summary, we have found differences in circulating miRNA levels comparing NSCLC cases and controls. While the differences are significant, our data suggest that circulating miR levels may not provide sufficient predictive accuracy to use as a screening biomarker for lung cancer. The clinical value of circulating miRs in predicting patient survival requires the examination of additional cohorts.