Serum miR-223: A Validated Biomarker for Detection of Early-Stage Non–Small Cell Lung Cancer

In the absence of screening, lung cancer is generally diagnosed in advanced stage, and is characterized by 5-year survival ranging a low 9%–18%, depending on international differences in cancer biology, stage at diagnosis, access to high-quality care, and data collection practice (1, 2). Since the National Lung Screening Trial (NLST) demonstrated 20% lung cancer mortality reduction following screening with low-dose CT (3), the latter is the screening tool recommended by the leading cancer organizations (4, 5). Beyond the NLST enrolment criteria of age and tobacco smoking exposure (3), additional risk factors (e.g., asbestos exposure, family history, cancer history, and chronic obstructive pulmonary disease) have recently been proposed as possible indications to CT screening (5). Given the vast population of smokers and nonsmokers at risk, and the low rates of lung cancer detection (typically 1%–3% of screenees; ref. 4), a novel, noninvasive tool is needed for refining screening selection criteria, aiming to limit unnecessary CT screens and medical/surgical procedures related to false-positive CT findings (6, 7). miRNAs have been suggested as a new class of tumor biomarkers, because the levels of some circulating miRNAs are altered in various human tumors, including non–small cell lung cancer (NSCLC) which represents approximately 85% of lung cancers (8, 9). miRNAs are small noncoding RNAs of up to 24 nucleotides in length that have important functions in different biological processes, including cell proliferation, differentiation, and apoptosis, by post-transcriptionally regulating the expression of human protein-coding genes (10, 11). miRNA presence and stability in biofluids are strong points supporting their application as lung cancer biomarkers (7, 12–15).

After critical review of the vast literature on circulating miRNA biomarkers, we found 20 different signatures reported as highly performing for detection of early-stage NSCLC. These signatures, predominantly assessed by qRT-PCR, comprise from three to 32 miRNAs, show widespread inconsistencies across studies, and are difficult to replicate due to scarcely detailed materials and methods (16–18). Besides NSCLC heterogeneity, multiple preanalytical and analytical variables are crucial issues that may partly explain the inconsistency of NSCLC signatures (17, 19, 20). Our review identified eight miRNAs that over the last 5 years were adequately reported as highly performing biomarkers of early (stage I–II) NSCLC: miR-223-3p, miR-20a-5p, miR-448, miR-145-5p, miR-628-3p, miR-29c-3p, miR-210-3p, and miR-1244 (17). However, for possible translation to clinical application in the context of lung cancer screening, a simple and consistently reproducible signature of NSCLC, including very few miRNAs, would be desirable. For this purpose, we focused on two of the eight mentioned biomarkers, namely miR-223-3p (hereafter miR-223) and miR-29c that were suggested to be, respectively, highly sensitive and highly specific for NSCLC detection (17).

The objective of this study is two-fold. First, to assess the reproducibility of serum miR-223 and miR-29c measurements using droplet digital PCR (ddPCR). Second, to externally validate the efficacy of these two miRNA biomarkers, individually and in combination, for detecting stage I–II NSCLC.

The work flow of the study is summarized in Fig. 1. In preliminary experiments, conditions for analysis by ddPCR were set-up for miR-223 and miR-29c. We assessed the technical feasibility of measurement of these molecules in the serum with ddPCR, a method less affected by PCR inhibitors compared with qRT-PCR, not requiring internal/external normalization (21–23) and directly yielding absolute measurement of molecule copies/μL. We applied stringent protocols in the preanalytical and analytical phases of miRNA measurement, and ascertained the repeatability, precision, and stability of determinations, adopting recommended standardized criteria for reporting the methodology (19). After optimizing ddPCR assay conditions, we determined the serum level of miR-223 and miR-29c in a cohort of sequentially enrolled subjects (75 early-stage NSCLC cases and 111 cancer-free controls). These measurements were used to assess the efficacy of serum miR-223 and miR-29c in discriminating NSCLCs from controls, by ROC curve analysis, in a training set and in an independent, blind validation set.

This study initially included 212 individuals, ethnically Italian, of both genders, age ≥ 45 years [mean age, 65.4 ± 8.0 (SD) years; 137 male; 75 female]. The cohort comprised 92 consecutive patients with clinical stage I–II NSCLC candidate to surgical resection and 120 healthy volunteer controls without history of cancer. Controls were recruited in the Varese Province population in collaboration with three general practitioners. At enrollment, control subjects had not undergone CT screening; however, they were reported by their doctors to be in good health, without symptoms. To evaluate the impact of smoking on serum miRNA levels, in the control group we arbitrarily enrolled a proportion of smokers similar to that in the NSCLC group; we did not strive to strictly match the two groups by age and gender, as these variables were shown to have no impact on miRNA signatures (24–28). The study was conducted at the Ospedale di Circolo and University of Insubria (Varese, Italy) between October 2014 and March 2018. For each participant, 5 mL peripheral blood was drawn and serum was collected as described below. All samples from cancer cases were obtained before initiating any anticancer therapy. NSCLC histotype and stage were confirmed pathologically after surgery (lung lobectomy and mediastinal lymph node staging in all patients), based on seventh edition of tumor–node–metastasis classification (29). Exclusion criteria were: age < 45 years; lung cancer histology other than adenocarcinoma or squamous cell carcinoma; pathologic stage of disease > II; and serum sample showing hemolysis. After exclusions, the cohort of this study was composed of 186 subjects, comprising 75 stage I–II NSCLC cases and 111 healthy controls. The first 40 NSCLCs and 56 controls formed the training set. After sample size computing, the validation set included the subsequently enrolled 35 NSCLCs and 55 controls. Demographic and clinico-pathologic data of patients and controls in the training set and in the validation set are shown in Table 1.

Written informed consent to collect samples and to participate in the study was obtained from all patients and controls. This study was conducted in accordance with the declaration of Helsinki and was approved by the Varese University Hospital Ethical Committee (approval no. 0037527 of September 30, 2014).

Peripheral blood samples were drawn with 21G needle in sterile vacuum tubes with clot activator (BD Vacutainer) early in the morning and were left at room temperature for minimum 40 minutes to maximum 90 minutes. Then the sera were separated by centrifugation at 800 × g for 8 minutes at room temperature, subdivided in 500 μL aliquots in 1.5 mL RNase-free plastic tubes and stored at –80°C until use. Because of the low concentration of miRNA molecules in serum samples, we always operated with constant volumes in all subsequent procedural steps.

Total RNA was extracted from serum samples with the miRNeasy Serum/Plasma kit (Qiagen), using 200 μL of serum and following the manufacturer's instructions. Spike-in molecule mix (Exiqon) was added before starting the process, to check for loss of material during the whole procedure. A total of 1 μg of MS2 phage carrier RNA (Sigma Aldrich) was also added to each serum sample to improve efficiency of miRNA extraction. RNA was eluted from the column with 14 μL of nuclease-free water and stored at –80°C.

For quantitative detection of miRNAs by qRT-PCR (exclusively for evaluation of hemolysis) or ddPCR, purified serum miRNA was first converted to cDNA by reverse transcription using miRCURY LNA Universal RT microRNA PCR System (Exiqon). The reverse transcription reaction was set-up in a total volume of 10 μL, consisting of 4.5 μL of nuclease-free water, 2 μL of 5× reaction buffer, 0.5 μL of UniSp6/cel-miR-39-3p RNA spike-in to evaluate the efficiency of the reverse transcription step, 1 μL of enzyme mix, and 2 μL of RNA template.

Because hemolysis can affect the level of some miRNAs (30, 31), the 212 samples initially considered were checked for hemolysis prior to further analysis. Although miR-223 and miR-29c were selected also by their reported scarce sensitivity to hemolysis, we excluded all hemolyzed sera to avoid this confounding factor. Accordingly, all samples assayed in the miRNA selection phase and in the testing phase were hemolysis free.

To assess hemolysis, for each sample we quantified by qRT-PCR the serum level of miR-451a and miR-23a-3p, two miRNAs, respectively, localized inside and outside red blood cells, and calculated the difference between the respective Cqs. This method has been previously described and validated as a sensitive procedure to evaluate the risk of hemolysis (13, 32). Briefly, 4 μL of cDNA template diluted 1:40 was used in a 10 μL reaction, adding 5 μL of ExiLENT SYBR Green PCR Master Mix (Exiqon) and primers. Triplicate reactions were performed for each sample and target; we used the manufacturer's instruction for cycling conditions [95°C for 10 minutes, followed by 40 cycles of 95°C for 10 seconds, and 60°C for 1 minute (1.6°C/second ramp rate)]. Samples were considered hemolyzed when their ΔCq was >5.

After preliminary tests described in Supplementary Data S1, optimal conditions for ddPCR analysis of miR-223 and miR-29c were established (Supplementary Figs. S1 and S2). The ddPCR reaction was prepared in a 20 μL volume, by adding 10 μL of 2× EvaGreen Supermix (Bio-Rad), the appropriate volume of cDNA after 20-fold dilution, the desired miRCURY LNA PCR primer set at the appropriate dilution, and nuclease-free water up to 20 μL. Each 20 μL ddPCR reaction was loaded onto an 8-channel Droplet Generation Cartridge (Bio-Rad), 70 μL of Droplet generation oil for EvaGreen (Bio-Rad) were added into the appropriate oil wells and the cartridge was placed into the QX200 Droplet Generator (Bio-Rad). The resulting emulsified reactions were transferred to a 96-well plate (Bio-Rad) with a multichannel pipette (Rainin) and the plate was sealed with Pierceable Foil (Bio-Rad). PCR amplification was performed in a T100 Thermal Cycler (Bio-Rad). Cycling conditions were: 95°C for 5 minutes, followed by 40 cycles of 95°C for 30 seconds and 60°C for 1 minute, followed by signal stabilization steps (4°C for 5 minutes and 90°C for 5 minutes). The ramp rate was 2°C/second. All samples were run in duplicates. After PCR, plates were placed into the QX200 Droplet Reader (Bio-Rad) for analysis. Measurements of miRNAs were expressed in copies/μL of PCR reaction volume.

Categorical variables were described using proportions and compared using χ² test. Continuous data were expressed as mean ± SD, or median value and interquartile range (IQR), and were compared by Student t test or by Mann–Whitney U test, as requested by data distribution. To assess the miRNA efficacy in discriminating NSCLC from control subjects, the ROC curve was constructed, the AUC with 95% CI was calculated, and the cutoff value of biomarker concentration for the best sensitivity and specificity combination was estimated using Youden Index, in a training set and in a validation set. Sample size of the validation set was calculated on the basis of miR-223 average differences between controls and NSCLCs in the training set, assuming type I error of 5% and type II of 20%.

For the panel combining two miRNAs, the performance in detecting NSCLC was analyzed using a logistic regression model based on the training set. A P < 0.05 was considered statistically significant. Statistical analysis was performed using SPSS 11.7 Software (SPSS Inc.) and MedCalc version 10.0.3 (MedCalc Software). Graphs of figures were generated using GraphPad Prism 8.0 (GraphPad Software, Inc.).

Preliminary experiments were conducted on samples from patients with NSCLC and controls to evaluate precision and stability of miR-223 and miR29c measurements. Results are reported in Supplementary Fig. S3 and Supplementary Tables S1 and S2. Intra- and interassay coefficients of variation (CV) for miR-223 were small (2.94% and 6.60%, respectively) and similar to those of spike-ins (Supplementary Table S1); miR-29c showed higher CVs (8.24% and 18.97%, respectively). Altogether, the repeatability of measurements was acceptable for the purpose of our study.

The long-term stability of miR-223 and miR-29c was tested in 23 serum samples, by two measurements carried out, respectively, at time 0 and after 7 months. Paired samples t test indicated that miR-223 and miR-29c determinations were stable over time (Supplementary Table S2).

In the training set, gender and smoking status distribution were similar in patients and in controls; the latter group however was younger (64.1 ± 7.9 years vs. 68.0 ± 7.7; P = 0.02; Table 1). In all serum samples, miR-223 and miR-29c were successfully assayed by ddPCR, except for one miR-29c determination in the control group. Both miRNAs showed significantly higher median copy number in the serum of patients with NSCLC than in control sera (miR-223, P < 0.001; miR-29c, P < 0.05; Fig. 2), the differences between smoker and nonsmoker controls being nonsignificant (miR-223, P = 0.155; miR-29c, P = 0.212). For miR-223 the ROC curve analysis yielded an AUC value of 0.753 (95% CI, 0.655–0.836), sensitivity 72.5% (95% CI, 56.1–85.4), and specificity 78.6% (95% CI, 65.6–88.4) in distinguishing patients with NSCLC from controls (Fig. 3). The estimated cutoff value of miR-223 concentration for the best sensitivity and specificity combination was 530 copies/μL. For miR-29c the AUC was 0.632 (95% CI, 0.527–0.729), sensitivity 52.5% (95% CI, 36.1–68.5), specificity 74.6% (95% CI, 61.0–85.3), and the estimated cutoff value was 15.7 copies/μL (Fig. 3).

The diagnostic performance of miR-223 was similar for adenocarcinoma and squamous cell carcinoma cases, the respective AUCs being 0.768 (95% CI, 0.697–0.829) and 0.805 (95% CI, 0.725–0.869). The miR-29c AUC was also similar for the two histologic subtypes [respectively, 0.651 (95% CI, 0.574–0.723) and 0.649 (95% CI, 0.550–0.724)].

The logistic regression for the combination of miR-223 with miR-29c was defined as Ln p/(1−p) = –1.339 + 0.00147 × (miR-223) + 0.00722 × (miR-29 c). The model yielded an ROC curve with AUC of 0.750 (95% CI, 0.651–0.833; Fig. 3), equal to that of miR-223 alone (0.753).

Because miR-29c showed a modest AUC (0.632) and the two miRNA combination did not increase the discriminating power of miR-223 alone, we focused on validating only miR-223 as biomarker of stage I–II NSCLC.

For the validation set, the estimated sample size was 32 cases and 32 controls. The serum samples of 35 patients with stage I–II NSCLC and 55 cancer-free controls were used to form the independent, blind validation set (Table 1), which was encrypted by a person not involved in the analyses. The cohort inclusion criteria were consistent throughout the study and the sequential assignment of the population, first to the training set and then to the validation set, resulted in homogeneity between the two sets, as shown by lack of significant differences in demographics and clinico-pathologic data (Table 1).

In the validation set, the miR-223 ROC curve analysis yielded an AUC of 0.808 (95% CI, 0.712–0.884), sensitivity 74.3% (95% CI, 56.7–87.5), and specificity 78.2% (95% CI, 65.0–88.2) in distinguishing patients with stage I–II NSCLC from controls (Fig. 4). The estimated best cutoff value was 550 copies/μL, which was not different from that in the training set (530 copies/μL), in line with the small CV of miR-223 determinations (Supplementary Table S1A).

Aberrant expression of miRNAs in serum/plasma samples of patients with lung cancer may be exploited for early diagnosis of the disease (6, 7, 14, 25, 33). However, clinical transferability of miRNA determinations for the purpose of lung cancer diagnosis has been greatly limited by difficult replication of circulating miRNA findings and by incoherent NSCLC signatures across studies (17–19). Scarce reproducibility of circulating miRNA findings in early-stage NSCLC likely has multiple explanations. First, miRNA quantifications are commonly performed by qRT-PCR. This method is not ideal for accurate miRNA measurement in biofluids due to several reasons, including the presence of PCR inhibitors in the samples and limited precision of qRT-PCR determinations. The fact that qRT-PCR is a relative quantification method is a concern, as a major controversy surrounds the reference(s) to be selected for data normalization (16, 19–21). Second, miRNA biomarkers of NSCLC are sensitive to stage of the disease (12, 34–36). Third, the control cohort composition is a relevant issue. Reviewing the literature, we observed that the “healthy control” group composition was inconsistent across studies, especially with respect to smoking habit (14, 17, 37). Cigarette smoking has been suggested as a possible confounding factor (38–40); however, the data available in the literature are insufficient to clarify the impact of smoking on the expression of miRNAs in biofluids. To provide an adequate control group in our study, we arbitrarily enrolled among controls a proportion of smokers similar to that of the NSCLC group. As the controls did not undergo lung cancer screening by CT, we cannot exclude potential asymptomatic lung cancers among them. At the time of writing this article, 2 years after ending enrollment and blood sampling, the General Practitioners collaborating in cohort recruitment performed a review (96% complete) of the controls' medical files, which showed no lung cancer case emerged during follow-up, neither with symptoms nor incidentally.

Notably, worldwide results of CT screening in asymptomatic smokers (41), and our past experience with CXR screening in the Varese Province (42), showed about 1% prevalence of lung cancer at baseline screening. Such low potential occurrence of lung cancer cases in our control cohort would only modestly affect the results of this study; if any, the impact would possibly be a slightly underestimated sensitivity of miR-223 biomarker.

Difficult replication of findings of miRNA biomarker studies is witnessed by the fact that none of the numerous circulating miRNA signatures until now proposed for screening of NSCLC has been consistently reproduced by other authors (17, 18). Moreover, the signatures until now published are impractical to translate to clinical application, because they include numerous miRNAs and are time consuming and/or complex to determine (17, 21). A simple miRNA-based signature, realistically translatable to a population screening setting is needed. This study validates miR-223 as early-stage NSCLC biomarker, robustly measurable with ddPCR. For miR-223 we found an AUC of 0.753 in the training set and 0.808 in the validation set, in line with the AUC values reported in Chinese cohorts of stage I–II NSCLCs using the qRT-PCR technique (Table 2). Of note, we observed that miR-223 AUC was similar for adenocarcinoma and squamous cell carcinoma, confirming the findings of Yang and colleagues (43). Demonstration of replicability of the miR-223 discriminating performance in multiple independent studies of ethnically different cohorts, using different technologies for miRNA determination, validates serum miR-223 as an effective biomarker of early-stage NSCLC and suggests its possible use for refining risk stratification of individual candidates to chest CT screening.

Several studies previously reported significantly increased circulating level of miR-223, as part of complex miRNA panels proposed for detection of early-stage NSCLC (6, 33, 37, 43–45). For miR-223, important functions have been suggested in the mechanisms that regulate tumor proliferation and invasion. While the source of overexpressed miR-223 in the circulation remains obscure (release from tumor, blood cells, or other cell types as a response to cancer), resection of lung cancer was shown to be followed by significant reduction of circulating miR-223 level (33). Separate investigations indicate opposing roles for miR-223; it was observed to promote tumor progression in lung cancer A549 cells (46), while in Lewis lung carcinoma cells the transfection of miR-223 was shown to inhibit migration and invasion (47). Overexpressed miR-223 was also reported in lung cancer tissue compared with peritumoral tissue (48), in gastric cancer compared with normal gastric mucosa (49), in T-cell acute lymphoblastic leukemia (50), and in prostate cancer cell lines (51).

Concerning serum miR-29c, the AUC value of 0.632 obtained in our training set was lower than that recorded for miR-29c in a Chinese cohort of stage I NSCLC by Zhu and colleagues (0.727; ref. 25). Moreover, combining miR-29c with miR-223 did not improve the discriminating power of the latter. Accordingly, miR-29c did not prove to be a valuable circulating biomarker of NSCLC.

Our study has limitations. Validation of miR-223 was performed in a relatively small cohort, 90 subjects. Because miRNA determinations require technically demanding laboratory steps, our findings should be further verified in a different institution laboratory setting. The miR-223 cutoff value developed in our training set (530 copies/μL) cannot be generalized; cutoff value should be determined in each laboratory that provides independent miRNA measurements.

A strength of this study, performed exclusively in patients with stage I–II NSCLC, is the careful selection from the literature of miR-223 and miR-29c, two candidate biomarkers of early-stage NSCLC that are scarcely influenced by hemolysis (17). Methodologic strengths are the exclusion of hemolyzed samples and the adoption of ddPCR for quantitating miRNA serum levels, a technique featuring superior precision and sensitivity compared with qRT-PCR (21–23). To allow for reproducibility of findings, we reported our miRNA determination method according to standardized criteria (19), and addressed the dMIQE criteria (52). Another strength is the robustness of miR-223 and miR-29c quantitative assays, documented by limited variability of determinations with ddPCR in our laboratory, and by stability over time (Supplementary Data S1).

Using ddPCR technology, we externally validated miR-223 as robustly measurable serum biomarker of early-stage NSCLC showing an AUC of 0.753 in the training set and 0.808 in the validation set. These findings warrant further confirmation by independent studies. The determination of serum miR-223 concentration is relatively simple and has a modest cost that would be acceptable for refining lung cancer risk stratification, independent of smoking history and age.

No potential conflicts of interest were disclosed.

Conception and design: L. Dominioni, A. Poli, D.M. Noonan, P. Campomenosi

Development of methodology: P. D'Antona, R. Cinquetti, E. Gini, D.M. Noonan, P. Campomenosi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. D'Antona, M. Cattoni, L. Dominioni, A. Imperatori, N. Rotolo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Dominioni, A. Poli, F. Moretti, D.M. Noonan, P. Campomenosi

Writing, review, and/or revision of the manuscript: L. Dominioni, A. Poli, D.M. Noonan, P. Campomenosi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. D'Antona, M. Cattoni, E. Daffrè, A. Imperatori

Study supervision: L. Dominioni, A. Poli, P. Campomenosi

This work was supported by grants from Fondazione Comunitaria del Varesotto (to L. Dominioni); Associazione PREDICA Onlus (to L. Dominioni); PRIN 2010NECHBX_003 (to D.M. Noonan); and AIRC (Associazione Italiana per la Ricerca sul Cancro IG15895, to D.M. Noonan). This work was co-funded by a donation from Mr. Piero Francesco Macchi and Mrs. Carlotta Biasini (to L. Dominioni). P. D'Antona and M. Cattoni are PhD students of the “Biotechnology, Biosciences and Surgical Technology” course at Università degli Studi dell'Insubria. The authors gratefully acknowledge the collaboration of the following general practitioners of the Varese Province: Drs. Maurizio Damiani, MD; Dario Sinapi, MD; and Laura Zanzi, MD. This work is in memory of Luigi (Gigi) Monteverde.

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.