Contributions of Circulating microRNAs for Early Detection of Lung Cancer

Simple Summary Blood-based cancer biomarkers are a minimally invasive means to achieve improved assessment of risk and/or earlier detection of lung cancer. Using serum samples from individuals that went on to develop lung cancer within one year following blood draw and from matched controls, we investigated 30 microRNAs for individual and combined utility to discriminate lung cancer cases from controls. We additionally assessed the contributions of top-performing microRNA candidates for improving on the performance of a previously validated four-protein marker panel for lung cancer detection. The results of this study indicate complementary performance of combining circulating microRNA and the four-protein-marker panel in assays for early detection of lung cancer. Abstract There is unmet need to develop circulating biomarkers that would enable earlier interception of lung cancer when more effective treatment options are available. Here, a set of 30 miRNAs, selected from a review of the published literature were assessed for their predictive performance in identifying lung cancer cases in the pre-diagnostic setting. The 30 miRNAs were assayed using sera collected from 102 individuals diagnosed with lung cancer within one year following blood draw and 212 controls matched for age, sex, and smoking status. The additive performance of top-performing miRNA candidates in combination with a previously validated four-protein marker panel (4MP) consisting of the precursor form of surfactant protein B (Pro-SFTPB), cancer antigen 125 (CA125), carcinoembryonic antigen (CEA) and cytokeratin-19 fragment (CYFRA21-1) was additionally assessed. Of the 30 miRNAs evaluated, five (miR-320a-3p, miR-210-3p, miR-92a-3p, miR-21-5p, and miR-140-3p) were statistically significantly (Wilcoxon rank sum test p < 0.05) elevated in case sera compared to controls, with individual AUCs ranging from 0.57–0.62. Compared to the 4MP alone, the combination of 3-miRNAs + 4MP improved sensitivity at 95% specificity by 19.1% ((95% CI of difference 0.0–28.6); two-sided p: 0.006). Our findings demonstrate utility for miRNAs for early detection of lung cancer in combination with a four-protein marker panel.


Introduction
Lung cancer is the leading (18%) cause of cancer mortality worldwide, accounting for an estimated 1.8 million deaths [1]. Five-year relative survival rates among lung cancer subtypes are closely linked with tumor stage at time of diagnosis, with an estimated overall survival rate of 59.8% when diagnosis is made at the localized stage, 32.9% at the regional stage, and 6.3% at the distant stage, according to recent United States National Cancer Institute (NCI) Surveillance, Epidemiology, and End Results (SEER) Program data [2]. NCI National Lung Cancer Screening Trial (NLST) data indicate that a 20% reduction in lung cancer mortality is achievable in high-risk populations through individual screening with low-dose computed tomography (LDCT) [3]. Similar findings have since been reported based on results from the Dutch-Belgian Nederlands-Leuvens Longkanker Screenings Onderzoek (NELSON) lung cancer screening trial [4]. Nevertheless, translation of LDCT to the general population has proven challenging, and the implementation of imaging-based lung cancer screening to reduce lung cancer disease burden is likely to remain low in the near term. Liquid biopsies offer an ideal minimally invasive means of individualized risk assessment to address the need to improve lung cancer screening, either by identifying subjects at increased risk of lung cancer that would benefit from screening or for complementing LDCT-based screening, notably for assessment of indeterminate imaging findings [5].
MicroRNAs (miRNAs) are~20-22-nucleotide non-coding RNAs that participate in regulation of gene expression, typically by binding to target miRNA transcripts and thus repressing miRNA translation. Dysregulation of the miRNA landscape was observed in multiple disease states, including cancer [6]. In lung cancers, altered miRNA expression has been found and characterized in tumor tissues as well as body fluids, suggesting utility for miRNAs as potential biomarkers in minimally invasive, liquid biopsy assays for assessing risk or detecting and monitoring disease status [7].
Here, using a cohort of sera (Table 1) collected up to one year prior to diagnosis of lung cancer (N = 102) and controls matched for age, sex, and smoking status (N = 212), we validated a panel of 30 miRNAs (Table 2) for predictive performance in distinguishing lung cancer cases from controls. The panel was developed based on review of previous reports of differential expression in lung cancer clinical specimens [8][9][10][11][12][13]. We further assessed the contributions of miRNAs for improving upon the predictive performance of a previously validated four-protein marker panel (4MP) consisting of cancer antigen 125 (CA125), carcinoembryonic antigen (CEA), cytokeratin-19 fragment (CYFRA 21-1), and the precursor form of surfactant protein B (Pro-SFTPB) for distinguishing lung cancer cases from controls compared to the 4MP alone [14,15].

Study Population
Serum samples for this nested case-control validation study were selected from the Carotene and Retinol Efficacy Trial (CARET) [139]. CARET was a randomized, doubleblind, placebo-controlled trial conducted to assess the safety and cancer prevention efficacy of daily supplementation with β-carotene plus retinyl palmitate in 18,314 persons at high risk for lung cancer. Participants were enrolled at six United States study centers from 1985 to 1994 and were followed for cancer and mortality outcomes through 2005. Cancer status was confirmed based on review of clinical records and pathology reports. All CARET participants provided informed consent at recruitment and throughout follow-up, and the study was conducted with oversight from the respective institutional review boards at each of the study centers. CARET pre-diagnostic serum samples collected up to one year prior to diagnosis from 102 ever-smoker patients with lung cancer and 212 smoking-status-, age-, and sex-matched controls ( Table 1) were used to test the individual performance of a panel of 30 candidate miRNAs ( Table 2). Samples used for miRNA testing were aliquots of CARET specimens previously used for developing a four protein biomarker panel (4MP) for assessing lung cancer risk based on cancer antigen 125 (CA125), carcinoembryonic antigen (CEA), cytokeratin-19 fragment (CYFRA 21-1), and the precursor form of surfactant protein B (Pro-SFTPB) [14,15].

miRNA Isolation
Expression analysis of 30 total micro RNAs were performed using validation cohort of pre-diagnostic CARET serum samples (N = 102) collected up to one year prior to diagnosis of lung cancer and corresponding controls (N = 212) matched for age, sex, and smoking status. Total RNAs were isolated from 200 µL of serum using Norgen silicon-carbide resin Total RNA Purification Kit 17250 (Norgen Biotek, Thorold, ON, Canada) and eluted with 25 µL of elution solution. For normalization of sample-to-sample variation in the RNA isolation step, 10 fmol of Caenorhabditis elegans cel-miR-39 and cel-miR-54 (mirVana miRNA mimic, Applied Biosystems, Foster City, CA, USA); in a total volume of 5 µL were added to each denatured sample after mixing the serum sample with lysis buffer. Total RNA concentrations were quantified in each sample via a NanoDrop ND-1000 spectrophotometer.

TaqMan Assay
RNA was reverse transcribed using the TaqMan miRNA Reverse Kit (Applied Biosystems) in a 15 µL multiplex RT reaction. Relative quantification of targets was performed by qRT-PCR in reaction buffer containing SYBR Green dye (SsoAdvanced Universal SYBR SuperMix 1725270, Bio-Rad Laboratories, Hercules, CA, USA). The intercalation of SYBR Green into the PCR products was monitored in real time using a CFX384 Touch Real-time PCR Detection System (Bio-Rad Laboratories). Ct values were normalized to miR-16-5p.

Statistical Analyses
Receiver operating characteristic (ROC) curve analysis was performed to evaluate the predicative performance of biomarkers for distinguishing lung cancer cases from matched controls. A combination rule for miRNA was developed using logistic regression models. The combined model of the miRNA panel plus the protein 4MP was developed by fitting a logistic regression with the miRNA panel score and the 4MP scores as two separate predictors. The estimated AUC of the proposed 3-marker miRNA panel and miRNA panel plus 4MP were derived by using the empirical ROC estimator of the linear combination corresponding to the respective models. One thousand bootstrap iterations with replacement were implemented to estimate the confidence intervals and p values. Statistical significance was considered at p < 0.05.

Candidate miRNA Selection
A panel of 30 candidate miRNAs was developed for testing based on review of literature reporting of differential expression in lung cancer patient biofluids or tumor tissues relative to control specimens. Priority was given to candidates identified across multiple independent studies. Table 2 summarizes the test panel candidates, with observed up-or downregulation relative to controls, and the number of studies in lung cancer that meet the criteria of: (i) inclusion of clinical samples and (ii) relevant controls, (iii) a validation component, and (iv) significance p ≤ 0.05 for the indicated miRNA. Relevant references for previous reports of dysregulated miRNA expression in clinical samples are also summarized in Table 2.

Predictive Performance of miRNAs for Early Detection of Lung Cancer
Assay of the panel of 30 miRNAs was performed using 102 case serum samples collected one-year preceding diagnosis of lung cancer and 212 control serum samples matched on the basis of age, sex, and smoking history ( Table 1).
Four of the 30 miRNAs were below the limit of quantitation. Five miRNAs (miR-320a-3p, miR-210-3p, miR-92a-3p, miR-21-5p, and miR-140-3p) were statistically significantly (Wilcoxon's rank sum p < 0.05) elevated in cases compared to matched controls with individual AUCs ranging from 0.57-0.62 ( Figure 1A and Table 3; Tables S1 and S2). The performance of individual miRNA candidates was associated with stage at time of diagnosis, (Table S1). Stratification of cases according to lung cancer histological subtype revealed the predictive performance of individual miRNA candidates was conserved ( Figure 1B).

Candidate miRNA Selection
A panel of 30 candidate miRNAs was developed for testing based on review of lit ature reporting of differential expression in lung cancer patient biofluids or tumor tissu relative to control specimens. Priority was given to candidates identified across multi independent studies. Table 2 summarizes the test panel candidates, with observed updownregulation relative to controls, and the number of studies in lung cancer that m the criteria of: (i) inclusion of clinical samples and (ii) relevant controls, (iii) a validati component, and (iv) significance p ≤ 0.05 for the indicated miRNA. Relevant references previous reports of dysregulated miRNA expression in clinical samples are also summ rized in Table 2.

Predictive Performance of miRNAs for Early Detection of Lung Cancer
Assay of the panel of 30 miRNAs was performed using 102 case serum samples c lected one-year preceding diagnosis of lung cancer and 212 control serum samp matched on the basis of age, sex, and smoking history ( Table 1).
Four of the 30 miRNAs were below the limit of quantitation. Five miRNAs (m 320a-3p, miR-210-3p, miR-92a-3p, miR-21-5p, and miR-140-3p) were statistically sign cantly (Wilcoxon's rank sum p < 0.05) elevated in cases compared to matched contr with individual AUCs ranging from 0.57-0.62 ( Figure 1A and Table 3; Tables S1 and S The performance of individual miRNA candidates was associated with stage at time diagnosis, (Table S1). Stratification of cases according to lung cancer histological subty revealed the predictive performance of individual miRNA candidates was conserved (F ure 1B).   We next focused on those miRNAs that were statistically significantly elevated in case sera and, using logistic regression models, developed a combination rule. The combination of miR-21-5p + miR-320a-3p + miR-210-3p was identified as the highest performing panel with a resultant AUC of 0.63 (95% CI: 0.56-0.69) and 22.6% sensitivity at 95% specificity ( Figure 2).
Using logistic regression models, we further determined whether the combination of miRNAs plus our previously validated 4MP would yield improved classification performance in differentiating lung cancer cases from controls when compared to the 4MP alone. A combined panel of the three-marker miRNA panel plus the 4MP yielded an AUC of 0.81 (95% CI: 0.73-0.89) (Figure 3A,B). In comparison to the 4MP alone, the combined miRNA panel plus 4MP resulted in statistically significantly improved sensitivity (19.1% (95% CI of difference 0.0-28.6); 2-sided p: 0.006) at 95% specificity (Table 4). case sera and, using logistic regression models, developed a combination r bination of miR-21-5p + miR-320a-3p + miR-210-3p was identified as the hig ing panel with a resultant AUC of 0.63 (95% CI: 0.56-0.69) and 22.6% sens specificity ( Figure 2).  Using logistic regression models, we further determined whether the combination o miRNAs plus our previously validated 4MP would yield improved classification perfor mance in differentiating lung cancer cases from controls when compared to the 4MP alone. A combined panel of the three-marker miRNA panel plus the 4MP yielded an AUC of 0.81 (95% CI: 0.73-0.89) ( Figure 3A,B). In comparison to the 4MP alone, the combined miRNA panel plus 4MP resulted in statistically significantly improved sensitivity (19.1% (95% CI of difference 0.0-28.6); 2-sided p: 0.006) at 95% specificity (Table 4).

Discussion
Earlier interception of lung cancer is directly linked with a more favorable outcome Blood-based lung cancer biomarkers offer utility for assessment of individual risk, detec tion of asymptomatic disease and stratification of equivocal findings from PET or CT im aging. Such blood-based assays would integrate directly into existing clinical workflow and are especially ideal for implementation in a primary care setting. Here, we evaluated 30 miRNA candidates and identified a marker panel comprising three miRNAs for iden

Discussion
Earlier interception of lung cancer is directly linked with a more favorable outcome. Blood-based lung cancer biomarkers offer utility for assessment of individual risk, detection of asymptomatic disease and stratification of equivocal findings from PET or CT imaging. Such blood-based assays would integrate directly into existing clinical workflows and are especially ideal for implementation in a primary care setting. Here, we evaluated 30 miRNA candidates and identified a marker panel comprising three miRNAs for identifying individuals at high risk of developing or harboring disease. Importantly, the combination of miRNAs with a previously validated 4MP protein panel yielded significantly improved sensitivity at the highest specificity thresholds. When applied in a screening context, such performance would allow for identification of additional cases with a minimal false-positive burden.
We note some further considerations with respect to this study. Serum specimens were sourced from Carotene and Retinol Efficacy Trial (CARET) cohort samples for which enrolment was restricted to individuals at high-risk for developing lung cancer, based on smoking-history and/or asbestos exposure and age between 45 and 69 years. Accordingly, the predictive performance of the combined model outside of this age range or for lowerrisk populations could not be determined. The performance of the combined miRNA plus 4MP among different ethnic and racial groups was likewise not assessed. Our study did not include an external cohort for further validation of the combined biomarker panels.
Indeterminate lung nodules identified by imaging present a challenging diagnostic scenario. Both the 4MP protein panel and miRNA candidates have been previously shown to contribute towards assessing indeterminate pulmonary nodules and offer improved performance compared to nodule size alone in predicting likelihood of cancer, supporting the observed association between the miRNA plus 4MP panel markers and lung cancer [10,15]. Furthermore, miRNA is known to regulate protein expression and mechanistic link between dysregulated miRNA and proteins in the context of cancer pathophysiology has been previously established [140][141][142][143][144].

Conclusions
A panel of miRNA plus an existing four-protein marker panel yielded notable improvements in sensitivity at high specificity compared to the protein panel alone. Our findings demonstrate the merits of including miRNAs when developing biomarker panels for lung cancer detection or risk assessment and suggest important implications for improving lung cancer screening and detection that warrant additional study.   Informed Consent Statement: All human subjects research reported herein was conducted consistent with National Institutes of Health Informed Consent Exemption category 4 guidance and involved retrospective study of existing publicly available, deidentified banked serum specimens obtained from the CARET Biorepository.
Data Availability Statement: Relevant data supporting the findings of this study are available within the Article and Supplementary Materials or are available from the authors upon reasonable request.

Conflicts of Interest:
The authors declare no conflict of interest.