Research Article

Alterations in Bronchial AirwaymiRNAExpressionfor Lung Cancer DetectionAna B. Pavel1,2, Joshua D. Campbell1,2, Gang Liu2, David Elashoff3,Steven Dubinett3, Kate Smith4, Duncan Whitney4, Marc E. Lenburg1,2, andAvrum Spira1,2, the AEGIS Study Team

Abstract

We have previously shown that gene expression alterations innormal-appearing bronchial epithelial cells can serve as a lungcancer detection biomarker in smokers. Given that miRNAsregulate airway gene expression responses to smoking, we eval-uated whether miRNA expression is also altered in the bronchialepithelium of smokers with lung cancer. Using epithelial brush-ings from the mainstem bronchus of patients undergoing bron-choscopy for suspected lung cancer (as part of the AEGIS-1/2clinical trials), we profiled miRNA expression via small-RNAsequencing from 347 current and former smokers for which geneexpression data were also available. Patients were followed forone year postbronchoscopy until a final diagnosis of lung cancer(n ¼ 194) or benign disease (n ¼ 153) was made. Followingremoval of 6 low-quality samples, we used 138 patients (AEGIS-1)as a discovery set to identify four miRNAs (miR-146a-5p,

miR-324-5p, miR-223-3p, and miR-223-5p) that were down-regulated in the bronchial airway of lung cancer patients (ANOVAP < 0.002, FDR < 0.2). The expression of these miRNAs is signi-ficantly more negatively correlated with the expression of theirmRNA targets than with the expression of other nontarget genes(K-S P < 0.05). Furthermore, these mRNA targets are enrichedamong genes whose expression is elevated in cancer patients (GSEAFDR<0.001). Finally,we found that the additionofmiR-146a-5p toan existing mRNA biomarker for lung cancer significantly improvesits performance (AUC) in the 203 samples (AEGIS-1/2) serving anindependent test set (DeLong P < 0.05). Our findings suggest thatthere are miRNAs whose expression is altered in the cytologicallynormal bronchial epithelium of smokers with lung cancer, and thatthey may regulate cancer-associated gene expression differences.Cancer Prev Res; 1–9. �2017 AACR.

IntroductionLung cancer remains the leading cause of cancer-related

death in the United States and the world due, in large part,to our inability to detect the disease at its earliest and curablestage. Once a pulmonary lesion is identified, physicians mustdecide between CT surveillance versus airway/lung biopsy.When biopsy is required, the approach can include bronchos-copy, transthoracic needle biopsy (TTNB), or surgical lungbiopsy (SLB). The choice among these procedures is deter-mined on the basis of considerations such as lesion size andlocation, the presence of adenopathy, the risk associated withthe procedure, and local expertise. Although bronchoscopy isrelatively safe (less than 1% of procedures complicated bypneumothorax; ref. 1), this procedure is limited by its sensi-

tivity (from 34% to 88%), depending on the location and sizeof the lesion (2). Even with newer bronchoscopic guidancetechniques, the sensitivity for the detection of lung cancer isbelow 70% for peripheral lesions (3).

A nondiagnostic bronchoscopy in this setting leads to a clinicaldilemma as to which of these patients should undergo furtherinvasive diagnostic testing (TTNBor SLB). To facilitate this clinicaldecision, we recently developed and validated a gene expression–based classifier that distinguishes between smokers with andwithout lung cancer using mRNA isolated from cytologicallynormal cells in the mainstem bronchus (4, 5). We demonstratedthat this biomarker can improve the diagnostic sensitivity ofbronchoscopy for lung cancer detection.

The ability to identify gene expression changes associated withcancer status in the normal appearing airway supports the idea ofan airway molecular field of injury spanning the respiratory tract(6). In this current study, we extend the field of injury concept tomiRNAs. miRNAs are a class of small, noncoding RNAs thatrepress gene expression and protein translation of their targetsby complementary binding to the 30 UTR of RNA transcripts. Inaddition, comparedwithmRNAs,miRNAs are thought to bemorestablemolecules,making themmore easilymeasured in degradedtissues (7). Previous studies have shown that smoking alters theexpression of miRNAs in the bronchial airway epithelium (8, 9).Wehypothesize that similar tomRNA, theremight also bemiRNAexpression changes associated with the presence of lung cancer inbronchial epithelium from themainstem bronchus that may playa role in regulating cancer-associated gene expression differencesand that integrating miRNA with gene expression could improvelung cancer detection.

1The Graduate Program in Bioinformatics, Boston University, Boston, Massa-chusetts. 2Section of Computational Biomedicine, Boston University School ofMedicine, Boston, Massachusetts. 3University of California Los Angeles, LosAngeles, California. 4Veracyte, South San Francisco, California.

Note: Supplementary data for this article are available at Cancer PreventionResearch Online (http://cancerprevres.aacrjournals.org/).

M.E. Lenburg and A. Spira are the co-senior authors of this article.

Corresponding Authors: Avrum Spira, Boston University School of Medicine,72 East Concord Street, E601, Boston, MA 02118. Phone: 617-414-6980; E-mail:aspira@bu.edu; Marc E. Lenburg, mlenburg@bu.edu; and Ana B. Pavel,anapavel@bu.edu

doi: 10.1158/1940-6207.CAPR-17-0098

�2017 American Association for Cancer Research.

CancerPreventionResearch

www.aacrjournals.org OF1

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

Materials and MethodsSelection of patients

As previously described, over 1,000 current and formersmokers undergoing bronchoscopy for suspected lung cancerwere enrolled in the Airway Epithelial Gene Expression in theDiagnosis of Lung Cancer (AEGIS) trials, two independent,prospective, multicenter, observational studies (registered asNCT01309087 and NCT00746759; refs. 4, 5). Exclusion criteriafor patients enrolled in AEGIS trials were age less than 21 years,no history of smoking (defined as having smoked <100 cigar-ettes), and a concurrent cancer diagnosis or history of lungcancer. All study protocols were approved by the InstitutionalReview Board at each medical center, and written informedconsent was obtained from all patients prior to enrollment.Patients were followed prospectively for up to one year post-bronchoscopy until a final diagnosis was obtained.

In this study, we profiled miRNA expression via small RNAsequencing for 347 AEGIS patients. In choosing patients toinclude in our study, we were limited by patients with a benigndiagnosis and matched them approximately 1:1 with patientsdiagnosed with lung cancer. Moreover, we attempted to balancethe cases and controls for smoking status, cumulative smokeexposure (pack-years), gender, and age. For all of the samplesselected for small RNA sequencing, gene expression profiling ofthe large RNA fraction had been performed previously usingAffymetrix Human Gene 1.0 ST arrays (4, 5) and was availablefor data integration.

We assigned 138 (� 40%) samples from AEGIS-1 to be used asa discovery set (Table 1); these samples were drawn exclusivelyfrom the training set previously used to develop the gene expres-sion classifier (4, 5). The remaining 203 samples comprise our testset (Table 1) and consist exclusively of samples from the AEGIS-1(n¼ 133) andAEGIS-2 (n¼ 70) test sets thatwere previously usedto validate the gene expression classifier (5).

High-throughput sequencing of small RNAOn the basis of our previous work on the effect of multiplexing

on miRNA expression quantitation (10), we sequenced 347samples in three batches by multiplexing 12 samples per laneon an Illumina HiSeq 2000. A total of 200 ng of total RNA fromeach sample was used for library preparation. The TruSeq SmallRNA Sample Prep Kit (Illumina)was used for thefirst batch, whilethe NEBNext Multiplex Small RNA Library Prep Set (Illumina)was used for the second and third batches. RNA adapters wereligated to 30 and 50 ends of the RNA, and the adapter-ligated RNAwas reverse transcribed into single-stranded cDNA. The RNA 30

adapter was designed to targetmiRNAs and other small RNAs thathave a 30 hydroxyl group resulting from enzymatic cleavage byDicer or other RNA processing enzymes. The cDNA was thenamplified by PCR, using a common primer and a primer contain-ing one of 12 index sequences. The introduction of the six-baseindex tag at the PCR step allowed multiplexed sequencing ofdifferent samples in a single lane of a flowcell. A 0.5% PhiX spike-inwas also added in all lanes for quality control. Eachmultiplexedlibrary was hybridized to one lane of the four 8-laneHigh-Outputsingle-read flow cells on a cBot Cluster Generation System (Illu-mina) using TruSeq Single-Read Cluster Kit (Illumina). Theclustered flowcell was loaded onto a HiSeq 2000 sequencer fora multiplexed sequencing run, which consists of a standard 36-cycle sequencing read with the addition of a 7-cycle index read.

miRNA alignment and quality controlTo estimate miRNA expression, we used a small RNA sequenc-

ing pipeline described previously (10). Briefly, the 30 adaptersequence was trimmed using the FASTX toolkit. Reads longerthan 15 nt were aligned to hg19 using Bowtie v0.12.7 (11)allowing up to onemismatch and alignment to up to 10 genomiclocations. miRNA expression was quantified by counting thenumber of reads aligning to mature miRNA loci (miRBase v20)using Bedtools v2.9.0 (12, 13).miRNA countswithin each samplewere normalized to log2 RPM values by adding a pseudocount ofone to each miRNA, dividing by the total number of reads that

Table 1. Patient demographics

Discovery setn ¼ 138

Test setn ¼ 203

Cancer status (n)a

Lung cancer 88 103Benign disease 50 100

Gender (n)Females 62 84Males 76 119

Age (SD; n) 59 (11; 138) 59 (10; 203)Smoking status (n)Current 46 88Former 92 115

Cumulative smoke exposure -pack-yr. (SD; n)

36 (24; 137) 37 (29; 199)

Race (n)White 109 149Black 24 46Unknown 5 8

Lesion size (n)<3 cm 52 71�3 cm 58 91Infiltrate 15 31Unknown 13 10

Histology (n)NSCLC 72 79NSCLC stageI 11 16II 3 5III 15 19IV 29 26Not specified 14 13

NSCLC subtypeAdenocarcinoma 31 34Squamous 27 25Large cell 2 4Not specified 12 16

SCLC 16 21SCLC stageLimited 4 8Extensive 8 12Not specified 4 1

Uncertain histology 0 3Diagnosis of benign disease (n)Resolution or stability 11 26Alternative diagnosis 39 74Type of alternative diagnosisSarcoidosis 9 17Inflammation 3 2Fibrosis 1 1Infection 8 14Other 18 40

NOTE: n indicates number of patients with available clinical data.Abbreviations: NSCLC, non–small cell lung cancer; SCLC, small-cell lung cancer.aP < 0.05.

Pavel et al.

Cancer Prev Res; 2017 Cancer Prevention ResearchOF2

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

aligned to all miRNA loci within that sample, multiplying by 1�106, and then applying a log2 transformation (10). The log2 RPMexpression values follow a normal distribution by an Anderson–Darling test (P ¼ 2.2 � 10�16; ref. 14).

Next, we examined the distribution of read lengths present ineach sample to ensure that the sequences we observed were of theproper length for miRNA. The read length distribution ought tofollow a normal distribution with a mean of 22 bases. We filteredout samples whose distribution had an abundance of reads wellbelow or above the mean of 22 bases (with less than one millionreads aligned to 22 read length), indicating that the sample wasnot properly sequenced, the adapters were improperly trimmed,or the samplewas of poor quality. Six such sampleswere removed,leaving 341 samples included in the downstream analysis. Inaddition, we removed miRNA loci with a low number of alignedreads (less than 20 on average). A total of 463miRNA loci passedour filter and were included in the analysis. Finally, we appliedComBat (15) to normalize the miRNA expression in the threedifferent batches. Large-scale variability inmiRNA expression wasexamined by principal components analysis. No outlier sampleswere detected using the first two principal components, and therewere no apparent global differences in miRNA expressionbetween samples from AEGIS-1 and AEGIS-2 (SupplementaryFig. S1).

Data availabilityRaw FASTQ files as well as the normalized miRNA expression

data are available onGene ExpressionOmnibus (GEO) under theGEO accession number GSE93284. We used mRNA data fromWhitney and colleagues and Silvestri and colleagues (GSE66499;refs. 4, 5).

Differential expression analysisTo identify smoking-associated miRNAs, while correcting for

covariates, we applied an F test (anovaR function; ref. 16) betweena multiple linear regression (lm R function), with miRNA expres-sion as the response variable, and smoking status, age, gender,cancer status, and pack-years as independent variables, andanother multiple linear regression that did not include the smok-ing status as an independent variable.

Similarly, to identify miRNAs with cancer-associated expres-sion patterns in the discovery cohort, while correcting for covari-ates, we applied an F test between a multiple linear regression,with miRNA expression as the response variable, and cancerstatus, age, gender, smoking status, andpack-years as independentvariables, and another multiple linear regression that did notinclude the cancer status as an independent variable.

The P values were adjusted for FDR using Benjamini–HochbergFDR (17) and are denoted with q-value.

Identifying miRNA–mRNA relationshipsWe analyzed the correlations between the differentially

expressedmiRNAs and their targets as predicted in the TargetScandatabase (18). We included the conserved targets as defined inTargetScan 5 and 6 (8mer� 0.8; 7mer-m8� 1.3; 7mer-1A� 1.6).The probability of conserved targeting (19) has the advantage ofidentifying targeting interactions that are not only more likely tobe effective but also those that aremore likely to be consequential.Correlation coefficients were calculated using Pearson product-moment coefficient. For each miRNA, we compared the resultingdistribution of correlation coefficients with the distribution of

correlation coefficients between the miRNA and all the genesthat have not been predicted to be targeted by it in TargetScan,using the Kolmogorov–Smirnov test. Next, we tested whetherthe negatively correlated targets (correlation FDR <0.1) of eachdifferentially expressed miRNA were enriched among the geneswhose expression is associated with cancer status by geneset enrichment analysis (GSEA; ref. 20). For this enrichmentanalysis, genes were ranked by the t statistic of a multiple linearregression, with miRNA expression as the response variable, andcancer status, age, gender, smoking status, and pack-years asindependent variables.

Incorporating miRNA expression into the mRNA classifierFirst, we calculated the prediction score of the mRNA classifier

(4, 5). Then, for each cancer-associated miRNA, we integrated themRNA classifier score with the miRNA's expression using logisticregression (glmnet R package). The coefficients of the logisticregression, corresponding to the intercept (a0 ¼ 1.8480041),weight of the classifier score (a1 ¼ 4.3879703), and weight ofthe miRNA's expression (a2 ¼ -0.3724577), were determined inthe discovery set, and the performance of the fully specifiedmodelwas evaluated in the independent test set samples. Classificationperformance was assessed using the area under the receiveroperating characteristic curve (ROC AUC). The statistical signif-icance of the AUC improvement was computed by DeLong test(21) from the pROC R package (22).

Figure 1.

Enrichment of known smoking-related miRNAs by GSEA. A set of 23 previouslydescribedmiRNAs that are expressed at lower levels in bronchial airway samplesfrom current smokers are significantly enriched among the miRNAs mostrepressed among current smokers in the current dataset (q < 0.001). The red toblue bar shows all 463 miRNAs ranked from most induced in smokers to mostrepressed (as shown in the distribution of t statistics at the bottom), while thevertical black lines show the position, within this ranked list, of the 23 miRNAspreviously found by microarray to have decreased expression in the bronchialairway of smokers. The green line is the running enrichment score, which has asignificantly negativeminimum, indicating that the previously reported miRNAsare among the miRNA most repressed among current smokers in the currentsmall RNA sequencing dataset.

Bronchial miRNA Expression in Lung Cancer

www.aacrjournals.org Cancer Prev Res; 2017 OF3

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

ResultsPatient population

miRNA expression was profiled via small RNA sequencing for347 patients (194 cancer-positive and 153 cancer-negative sub-jects) participating in the AEGIS-1 and AEGIS-2 trials. Of the 347miRNA samples, 341 passed the sequencing quality control filter(10). The characteristics of thediscovery set (138 samples) and thetest set (203 samples) are shown in Table 1. Except for cancerstatus, the other clinical variables are not significantly differentbetween the training and test sets. We also found significantassociations between cancer status and age and lesion size in thediscovery set and with pack-years and lesion size in the test set(Supplementary Table S1).

Identifying smoking-associated miRNAs in airway epitheliumPrevious work has shown that cigarette smoke creates a molec-

ular field of injury throughout the airway, and specifically thatmiRNA expression is alteredwith tobacco smoke exposure (9, 23–28).We therefore used the ability to detect miRNAs with smokingstatus–associated expression as apositive control for the quality ofthe miRNA expression data.

A set of 28 miRNAs was previously identified as modulatorsof smoking-related gene expression changes in airway epithe-lium (9), with most of them (n ¼ 23) being downregulatedin current smokers compared with never smokers. We foundthat the miRNAs previously identified as being repressed by

smoking were significantly enriched among the miRNAs thatwere most downregulated in current smokers from AEGIS(GSEA q < 0.001), as shown in Fig. 1.

In addition, using our data, we identified significantly differ-entially expressed miRNAs between current and former smokersby linear regression. We found 135 smoking-associated miRNAsby P < 0.05 (Supplementary Table S2). The top 30 differentiallyexpressed miRNAs in the discovery set (q < 0.01) are shown inSupplementary Fig. S2. Among these, we found miRNAs whoseexpression has been previously associated with smoking, such asmiR-218, miR-365, miR-30a, and miR-99a (9).

We also evaluated the relationship between bronchial miRNAexpression and other potentially relevant clinical variables, suchas gender, age, and pack-years (Supplementary Tables S3–S5).We found that in addition to smoking status, gender is alsoassociated with miRNA expression (85 differentially expressedmiRNAs, P < 0.05).

Cancer-associated miRNA alterations in the bronchial airwayepithelium

Using the discovery set (n ¼ 138), we identified 42 miRNAsthat showed differential expression between patients with andwithout cancer by linear regression at a liberal P value threshold ofP < 0.05 (Supplementary Table S6). Of these, four miRNA iso-forms showed evidence of differential expression at FDR<0.2(P < 0.002). These four are: miR-146a-5p, miR-324-5p, miR-223-3p, and miR-223-5p. The expression profiles of these four

Figure 2.

miRNAs significantly differentiallyexpressed in bronchial epitheliumbetween patients with and withoutlung cancer.A, Expression of hsa-miR-146a-5p (P ¼ 0.0008, q ¼ 0.125). B,Expression of hsa-miR-324-5p (P ¼0.0007, q ¼ 0.125). C, Expression ofhsa-miR-223-3p (P ¼ 0.0007, q ¼0.125). D, Expression of hsa-miR-223-5p (P ¼ 0.0016, q ¼ 0.184).

Pavel et al.

Cancer Prev Res; 2017 Cancer Prevention ResearchOF4

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

miRNAs are shown in Fig. 2. Each of these miRNA has previ-ously been reported to have tumor suppressor–like activity(29–32). Consistent with the potential for these miRNAs tofunction as tumor suppressors, we find that the four differen-tially expressed miRNA isoforms are downregulated in thebronchial airway of patients with lung cancer.

Cancer-associatedmiRNAs as potential regulators of the airwaygene expression alterations

miRNAs often lead to the degradation of the mRNAs to whichthey bind. Therefore, we sought to determine whether the expres-sion of these miRNAs was negatively correlated with the expres-sion of their gene targets. We found that the distribution of thecorrelation coefficients of each cancer-associated miRNA and itspredicted mRNA targets (binding site predicted targets fromTargetScan) is significantly more negative than the distributionof correlation coefficients for nontarget genes (P < 10�9 for eachmiRNA; Fig. 3).

To begin to understand the potential biological impact of thecancer-associated expression of these miRNAs, we investigatedwhether the expression of their gene targets is associated withcancer. From the predicted targets (TargetScan), we identified thegenes whose expression is significantly negatively correlated

(correlation q < 0.1) with the cognate miRNA. The negativelycorrelated predicted targets of each of the four miRNAs weresignificantly enriched among the genes whose expressionincreased in the airway epithelium of patients with cancerrelative to those with a benign diagnosis (GSEA q < 0.001; Fig. 4).

In addition, the set of genes predicted to be regulated by thesefour miRNAs (n¼ 254 in total; TargetScan binding site predictedtargets and negatively correlated miRNA – mRNA expression) isenriched by DAVID (33) for cancer-associated pathways, such assignaling pathways regulating pluripotency of stem cells (P ¼0.001), pathways in cancer (P ¼ 0.007), the TGFb signalingpathway (P¼ 0.035), and the Ras signaling pathway (P¼ 0.043).

miRNA expression adds to mRNA in the detection of lungcancer

We next sought to assess whether bronchial miRNA expres-sion could add to the performance of an mRNA biomarker forlung cancer we previously identified (4). Using the training setsamples, we used logistic regression to build five cancer pre-diction models: one model contained the mRNA biomarkerscore alone, the other four models contained the mRNA bio-marker score in combination with one of the four miRNAs weidentified as having significant cancer-associated expression.

Figure 3.

miRNAs with cancer-associatedexpression are negatively correlatedwith their predicted targets. Thedistribution of miRNA–mRNAcorrelations for each miRNA and itspredicted targets is shown witha solid line. The null distribution ofmiRNA–mRNA correlations for eachmiRNA and all nontargets is shownwith a dashed line. The differencebetween the two distributions wastested using the Kolmogorov–Smirnov test.

Bronchial miRNA Expression in Lung Cancer

www.aacrjournals.org Cancer Prev Res; 2017 OF5

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

Figure 4.

The negatively correlated and predicted gene targets of the four differentially expressed miRNAs are enriched among genes that are expressed more highly in thebronchial airway of patients with cancer: the distribution of gene sets consisting of negatively correlated and predicted targets of miR-146a-5p (A; 50 genes),miR-324-5p (B; 43 genes), miR-223-3p (C; 89 genes), and miR-223-5p (D; 72 genes) was examined in a list of genes ranked in the discovery set (n ¼ 138) by theassociation of their expression levels with cancer status in bronchial airway samples using GSEA. All of these gene sets are significantly enriched amongthe genesmost induced in the bronchial airway of patientswith lung cancer (GSEA q <0.001 for each). The red to blue bar shows all genes ranked frommost inducedin the bronchial airway of patients with cancer to most repressed (as shown in the distribution of t statistics at the bottom), while the set of vertical blacklines in each panel shows the position of the predicted gene targets of each of the miRNAs whose expression is significantly negatively correlated with that miRNA.The green line is the running enrichment score, which has a significantly positive maximum in each panel, indicating that these genes are enriched amongthe genes most induced in patients with cancer.

Pavel et al.

Cancer Prev Res; 2017 Cancer Prevention ResearchOF6

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

Next, we compared the ROC curve AUC of the mRNA bio-marker alone with the four miRNA-containing models using atest set (Table 1; Supplementary Table S1) comprised of AEGIS-1 and AEGIS-2 samples that are independent of the AEGIS-1samples used to identify the four miRNAs with cancer-associ-ated expression and independent of the samples used to devel-op the mRNA biomarker. We found that adding miR-146a-5pto the mRNA biomarker significantly improved the AUC in thetest set, from 0.66 to 0.71 (P ¼ 0.025). The AUC of biomarkersincorporating either miR-324-5p or either of the two isoformsof miR-223 was not significantly different than the AUC of themRNA biomarker alone (P > 0.25) in the test set. The perfor-mance metrics of each miRNA combined with the mRNAbiomarker are provided in Supplementary Table S7.

DiscussionWehavepreviously identifiedbronchial airway gene expression

differences between patients with and without lung cancer andshown that they can be used as a biomarker with clinical utility inthe setting of patients with inconclusive results following bron-choscopy for suspect lung cancer (4–6). In this study,wewished todeterminewhethermiRNA expressionmight also be altered in thenormal-appearing epithelium of the mainstem bronchus, wheth-er these miRNA expression differences might play a role inregulating the observed gene expression differences, and whetherlung cancer–associatedmiRNAsmight have the potential to aid inthe detection of disease.

We identified four miRNA isoforms (miR-146a-5p, miR-324-5p,miR-223-3p, andmiR-223-5p) that have altered expression inthe airway epithelium of patients with lung cancer. That all fourmiRNAs have decreased expression in the bronchial airway oflung cancer patients is consistent with prior studies that havefound miRNAs with cancer-specific expression, mostly down-regulated, in tumors compared with normal tissue (34). Intrigu-ingly, all four of the miRNAs we identified have previously beenimplicated in tumor-suppressive pathways. Specifically, miR-146a has been previously shown to inhibit cell growth,migration,and EGFR signaling (29, 30, 35), while inducing apoptosis.Furthermore, miR-146a/b expression levels have been shown tobe significantly elevated during senescence (a cellular programthat irreversibly arrests the proliferation of damaged cells; ref. 36).miR-223 has been shown to function as a tumor suppressor in theLewis lung carcinoma cell line by targeting insulin-like growthfactor-1 receptor and cyclin-dependent kinase-2 (32), and miR-324 has been associated with nasopharyngeal cancer (31).Although miRNA expression differences have been well docu-mented in tumors, our results are the first to demonstrate alteredexpression of not just these cancer-related miRNAs, but anymiRNA in the bronchial airway of lung cancer patients.

We found that the expression of mRNAs that are predictedtargets of these miRNAs is significantly negatively correlated,suggesting that the expression of downstream genes is inducedas a consequence of the cancer-dependent loss of miRNA expres-sion. Moreover, predicted targets with negatively correlatedexpression profiles are enriched for genes involved in processesimportant for cancer, such as the pluripotency of stem cells, TGFb,and Ras signaling pathways. Among the 50 significantly nega-tively correlated predicted targets of miR-146a-5p, we foundAPPL1. The protein encoded by APPL1 gene binds to many otherproteins, including PIK3CA, RAB5A,DCC,AKT2, and adiponectin

receptors, aswell as proteins of theNuRD/MeCP1 complex,whichare involved in cell proliferation and cross-talk between adipo-nectin and insulin signaling pathways (37, 38). Interestingly, wealso observed a significantly negative correlation between miR-146a-5p and PIK3CA, suggesting that miR-146a-5p might mod-ulate the PI3K/AKT pathway. In addition to the important role ofPI3K/AKT pathway in cell death/survival, increased PI3K activityhas been observed in lung cancer (39) and has been shown tooccur early and potentially be reversible in the airway of smokerswith premalignancy (39, 40). The anticorrelation of these differ-entially expressed bronchial miRNAs with cancer-associatedmRNA targets suggests their role as lung cancer-associated reg-ulators of gene expression and that they could potentially serve asbiomarkers of disease.

We assessed each differentially expressed miRNA's ability toenhance the performance of an mRNA-based lung cancer bio-marker and found that miR-146a-5p significantly improves per-formance. One possible explanation for why miR-223-3p andmiR-223-5p did not improve biomarker performance is that oneof their targets (SNCA) is already a component of the mRNAclassifier; thus, miR-223 expression might be substantially redun-dant with SNCA expression levels. If this hypothesis is correct, itwould suggest that miR-146a adds to the biomarker's perfor-mance because the mRNA biomarker does not already capturemiR-146a–related expression information.

In this study, we demonstrate for the first time the presenceof an miRNA field of injury in the bronchial airway for lungcancer. We identify miRNAs that are known to play a role incancer-related processes, and importantly, we demonstrate thata multi `omics data integration approach may improve lungcancer detection.

Disclosure of Potential Conflicts of InterestK. Smith is the clinical trial manager at Veracyte, Inc. D. Whitney is the

vice president (Discovery Research) at Veracyte, Inc. M.E. Lenburg is aconsultant for Veracyte. A.E. Spira is a consultant/advisory board memberfor Veracyte, Inc. No potential conflicts of interest were disclosed by theother authors.

Authors' ContributionsConception and design: A.B. Pavel, M.E. Lenburg, A. SpiraDevelopment of methodology: A.B. Pavel, G. Liu, M.E. Lenburg, A. SpiraAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): G. Liu, K. Smith, D. Whitney, M.E. LenburgAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): A.B. Pavel, J.D. Campbell, D. ElashoffWriting, review, and/or revision of the manuscript: A.B. Pavel, J.D. Campbell,G. Liu, D. Elashoff, S. Dubinett, D. Whitney, M.E. Lenburg, A. SpiraAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): G. LiuStudy supervision: M.E. Lenburg, A. SpiraOther (reviewed manuscript): K. Smith

AcknowledgmentsBoston University owns patents related to the subject matter of this man-

uscript. We would also like to thank Jacob Kantrowitz and Jessica Vick for theirhelp in revising the early draft of the manuscript.

The AEGIS Study Team

Beth Israel Deaconess Medical Center, Boston, MAPrincipal investigator: Armin Ernst and Gaetane Michaud; coinvestigators:

AdnanMajid, Sidharta PenaGangadharan, Andres Sosa, RenelleMyers,Michael

Bronchial miRNA Expression in Lung Cancer

www.aacrjournals.org Cancer Prev Res; 2017 OF7

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

Kent, Malcolm DeCamp, Dilip Nataraj, Samaan Rafeq, David Berkowitz, SalehAlazemi, Robert Garland; study coordinators: Arthur Dea, Paula Mulkern,Christina Carbone.

Cleveland Clinic, Cleveland, OHPrincipal investigator: Tom Gildea; coinvestigators: Francisco Almeida,

Joseph Cicenia, Mike Machuzak; study coordinators: Meredith Seeley.

Columbia University, New York, NYPrincipal investigator: Charles Powell; coinvestigators: William Bulman,

Joshua Sonett, Rebecca Toonkel; study coordinators: Kivilcim Sungur-Stasik.

Georgia Clinical Research, Austell, GAPrincipal investigator: Stuart Simon; coinvestigators: Chad Case, Alexan-

der Gluzman, Charles Hartley, Steven Harris, Aristidis Iatridis, JermaineJackson, C. Coy Lassiter, Brion Lock, Kathryn McMinn, Chad Miller, SriramParamesh, Craig Patterson, Brett Sandifer, Samuel Szumstein, James Wal-dron, Christy Wilson, Paul Zolty, Stephen Strazay, Ashley Waddell; StudyCoordinators: Betsy Rambo, Monica Haughton, Stacy Beasley, Penny Mur-ray, Debra Yeager.

Indiana University, Indianapolis, INPrincipal investigator: Francis Sheski; coinvestigators: PraveenMathur; study

coordinators: MaryAnn Caldwell, Annette Hempfling.

Jamaica Hospital Medical Center, Jamaica, NYPrincipal investigator: Craig Thurm; coinvestigators: Aradhana Agarwal,

Akash Ferdaus; study coordinators: Kelly Cervellione.

Louisiana State University, New Orleans, LAPrincipal investigator: Stephen Kantrow; coinvestigators: SusanGunn,David

Welsh, Jennifer Ramsey, Jaime Palomino, Richard Tejedor; study coordinators:Connie Romaine.

Medical University of South Carolina, Charleston, SCPrincipal investigator: Gerard Silvestri; coinvestigators: Nicholas James Pas-

tis, Nichole Tripician Tanner, Peter Doelken, John Terrill Huggins: studycoordinators: Jeffrey Waltz, Katherine Taylor, Kalon Eways.

National Jewish Health, Denver, COPrincipal investigator: Ali Musani; coinvestigators: David Hsia, Joseph Sea-

man, Justin Thomas: study coordinators: Phillip Lopez, Jami Henriksen.

New York University, New York, NYPrincipal investigator: William Rom; coinvestigators: Eric Leibert, Derrick

Raptis, James Tsay, Robert Lee, Eric Bonura; study coordinators: Katie Schliess-man, Ellen Eylers.

North Florida/South Georgia Veterans Health System,Gainesville, FL

Principal investigator: Peruvemba Sriram; study coordinators: Ana Thomas,Katherine Herring, Carmen Lowell.

Overlake Hospital, Bellevue, WAPrincipal investigator: AmyMarkezich; coinvestigators: JamesCopeland, Eric

Gottesman, Todd Freudenberger, William Watts: study coordinators: TinaFortney.

Pulmonary Associates, P.A., Phoenix, ArizonaPrincipal investigator:MarkGotfried; coinvestigators: RobertComp,Andreas

Kyprianou, James Ross, Ronald Servi; study coordinators: Li Yi Fu, SherryHarker.

Pulmonary and Allergy Associates, P.A., Summit, NJPrincipal investigator: Robert Sussman; coinvestigators: Donatella Graffino,

Mark Zimmerman, Robert Restifo, Vincent Donnabella, Federico Cerrone, JohnOppenheimer, Robert Capone, Jaime Cancel, Edward Dimitry, Matthew

Epstein, Sue Fessler, Erwin Oei, Frederic Scoopo, Chirag Shah; study coordi-nators: Virginia Hala, Kathy Izzo, Marissa Reinton-Lim, Hazel Scherb, MaryAnnConstantino.

St. Elizabeth's Medical Center, Brighton, MAPrincipal investigator: Samaan Rafeq and Armin Ernst; coinvestigators: Ali

Ashraf, Antonio DeGorordo Arzamende, Deirdre Keogh, Ryan Chua, Ali Kho-dabandeh; study coordinators: Arthur Dea, Paula Mulkern.

St. James's Hospital, Trinity College, Dublin, IrelandPrincipal investigator: Joe Keane; study coordinators: Jennifer Winkles, Eliot

Woodward.

Temple University, Philadelphia, PAPrincipal Investigator: John Travaline; coinvestigators: Peter Bercz, Wissam

Chatila, Brian Civic, Francis Cordova, Gerard Criner, Gilbert D'Alonzo, VictorKim, Samuel Krachman, Albert Mamary, Nathaniel Marchetti, Aditi Satti, KartikShenoy, Irene Permet Swift, Maria-Elena Vega Sanchez, SheilaWeaver, NicholasPanetta, Parag Desai, Fred Kueppers, Namrata Patel, Kathleeen Brennan, AlexSwift, David Ciccolella, Fred Jaffe, Jamie Lee Garfield; study coordinators: CarlaGrabianowski, Carolina Aguiar.

University of Alabama, Birmingham, ALPrincipal investigator: Mark Dransfield; study coordinator: Sherry Tidwell.

University of British Columbia, Vancouver, BC, CanadaPrincipal investigator: Stephen Lam; coinvestigators: Annette McWilliams;

study coordinators: Sharon Gee.

University of California- Davis, Sacramento, CAPrincipal investigator: Richart Harper; coinvestigators: Ken Yoneda, Jason

Adams, Katherine Cayetano, Andrew Chan, Heba Ismail, Charles Poon, Rokh-sara Rafii, Christian Sebat, Yasmeen Shaw, Matthew Sisitki, Will Tseng; studycoordinators: Maya Juarez, Kaitlyn Kirk, Claire O'Brien.

University of Missouri, Columbia, MOPrincipal investigator: Vamsi Guntur; coinvestigators: Normand Caron,

Harjyot Sohal, Casey Stahlheber, Danish Thameem, Shilpa Patel, OusamaDabbagh, Rajiv Dhand, Rachel Kingree, Yuji Oba, Jason Goodin; study coordi-nators: Marta Fuemmeler, Angie Vick, Michel O'Donnell.

University of Pennsylvania, Philadelphia, PAPrincipal investigator: Anil Vachani; coinvestigators: Andrew Haas Colin

Gillespie,Daniel Sterman,; study coordinators: KristinaMaletteri, KarenDengel.

University of Virginia, Charlottesville, VAPrincipal investigator: George Verghese; coinvestigators: Cynthia Brown,

Elizabeth Gay, Borna Mehrad, Manojkumar Patel, Mark Robbins, C. EdwardRose, Max Weder, Kyle Enfield; study coordinators: Peggy Doherty.

University of Wisconsin, Madison, WIPrincipal investigator: Scott Ferguson; coinvestigators: Mark Regan, Jennifer

Bierach; study coordinators: Michele Wolff.

Vanderbilt University, Nashville, TNPrincipal investigator: Pierre Massion; coinvestigators: Alison Miller; study

coordinators: Gabe Garcia, Anna Ostrander, Wendy Cooper, Willie Hudson.

Virginia Commonwealth University, Richmond, VAPrincipal investigator: Wes Shepherd; coinvestigators: Hans Lee, Rajiv Mal-

hotra, Ashutosh Sachdeva; study coordinators: ChristineDeWilde, Anna Priday.

William Jennings Bryan Dorn VAMCPrincipal investigator: Brian Smith and Andrea Mass; study coordinators:

Justin Reynolds, Andrea Peterson, Isaac Holmes.

Pavel et al.

Cancer Prev Res; 2017 Cancer Prevention ResearchOF8

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

Yale University, New Haven, CTPrincipal investigator: Gaetane Michaud; coinvestigators: Daniel Boffa,

Frank Detterbeck, Kelsey Johnson, Anthony Kim, Jonathan Puchalski, LynnTanoue, Kyle Bramley; study coordinators: Christina Carbone.

Grant SupportThis researchwas supported by grants from theNIHEarlyDetection Research

Network (5U01CA152751 to A. Spira and M.E. Lenburg), the Department ofDefense (DOD W81XWH-11-2-0161 to A. Spira and M.E. Lenburg), and the

Boston University Coulter Award (0-057-281-A594-5 to A. Spira and M.E.Lenburg). This study was sponsored by Veracyte, Inc.

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.

Received March 31, 2017; revised July 8, 2017; accepted August 28, 2017;published OnlineFirst September 6, 2017.

References1. Tukey MH, Wiener RS. Population-based estimates of transbronchial lung

biopsy utilization and complications. Respir Med 2012;106:1559–65.2. Rivera MP, Mehta AC, Wahidi MM. Establishing the diagnosis of lung

cancer: Diagnosis and management of lung cancer, 3rd ed: AmericanCollege of Chest Physicians evidence-based clinical practice guidelines.Chest 2013;143:e142S–65S.

3. WangMemoli JS, Nietert PJ, Silvestri GA. Meta-analysis of guided bronchos-copy for the evaluation of the pulmonary nodule. Chest 2012;142:385–93.

4. Whitney DH, Elashoff MR, Porta-Smith K, Gower AC, Vachani A, FergusonJS, et al. Derivation of a bronchial genomic classifier for lung cancer in aprospective study of patients undergoing diagnostic bronchoscopy. BMCMed Genomics 2015;8:18.

5. Silvestri GA, Vachani A,WhitneyD, ElashoffM, SmithKP, Ferguson JS, et al.A bronchial genomic classifier for the diagnostic evaluation of lung cancer.N Engl J Med 2015;373:243–51.

6. Spira A, Beane JE, Shah V, Steiling K, Liu G, Schembri F, et al. Airwayepithelial gene expression in the diagnostic evaluation of smokers withsuspect lung cancer. Nat Med 2007;13:361–6.

7. Etheridge A, Lee I,Hood L,GalasD,WangK. ExtracellularmicroRNA: a newsource of biomarkers. Mutat Res 2011;717:85–90.

8. Perdomo C, Campbell JD, Gerrein J, Tellez CS, Garrison CB, Walser TC,et al.MicroRNA4423 is a primate-specific regulator of airway epithelial celldifferentiation and lung carcinogenesis. Proc Natl Acad Sci U S A 2013;110:18946–51.

9. Schembri F, Sridhar S, Perdomo C, Gustafsond AM, Zhangd X, Ergun A,et al. MicroRNAs as modulators of smoking-induced gene expressionchanges in human airway epithelium. Proc Natl Acad Sci U S A 2009;106:2319–24.

10. Campbell JD, Liu G, Luo L, Xiao J, Gerrein J, Juan-Guardela B, et al.Assessment of microRNA differential expression and detection in multi-plexed small RNA sequencing data. RNA 2015;21:164–71.

11. Langmead B, Trapnell C, Pop M, Salzberg SL. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome.Genome Biol 2009;10:R25.

12. Griffiths-Jones S. The microRNA Registry. Nucleic Acids Res 2004;32:D109–11.

13. Quinlan AR, Hall IM. BEDTools: a flexible suite of utilities for comparinggenomic features. Bioinformatics 2010;26:841–2.

14. Thode HC. Testing for normality. New York, NY: CRC Press; 2002.15. Johnson WE, Li C, Rabinovic A. Adjusting batch effects in microarray

expression data using empirical Bayes methods. Biostatistics 2007;8:118–27.

16. Chambers J. Linear models. Pacific Grove, CA: Wadsworth & Brooks/Cole;1992.

17. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practicaland powerful approach to multiple testing. J R Stat Soc Series B Methodol1995;57:289–300.

18. Lewis BP, Burge CB, Bartel DP. Conserved seed pairing, often flanked byadenosines, indicates that thousands of human genes are microRNAtargets. Cell 2005;120:15–20.

19. Friedman RC, Farh KK-H, Burge CB, Bartel DP. Most mammalian mRNAsare conserved targets of microRNAs. Genome Res 2008;19:92–105.

20. SubramanianA, TamayoP,Mootha VK,Mukherjee S, Ebert BL,GilletteMA,et al. Gene set enrichment analysis: a knowledge-based approach forinterpreting genome-wide expression profiles. Proc Natl Acad Sci U S A2005;102:15545–50.

21. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the Areas undertwo or more correlated receiver operating characteristic curves: a nonpara-metric approach. Biometrics 1988;44:837.

22. Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez JC, et al. pROC:an open-source package for R and Sþ to analyze and compare ROC curves.BMC Bioinformatics 2011;12:77.

23. Powell CA, Klares S, O'Connor G, Brody JS. Loss of heterozygosity inepithelial cells obtained by bronchial brushing: clinical utility in lungcancer. Clin Cancer Res 1999;5:2025–34.

24. Wistuba II, Lam S, Behrens C, Virmani AK, Fong KM, LeRiche J, et al.Molecular damage in the bronchial epithelium of current and formersmokers. J Natl Cancer Inst 1997;89:1366–73.

25. Franklin WA, Gazdar AF, Haney J, Wistuba II, La Rosa FG, Kennedy T, et al.Widely dispersed p53 mutation in respiratory epithelium. A novel mech-anism for field carcinogenesis. J Clin Invest 1997;100:2133–7.

26. Guo M, House MG, Hooker C, Han Y, Heath E, Gabrielson E, et al.Promoter hypermethylation of resected bronchial margins: a field defectof changes? Clin Cancer Res 2004;10:5131–6.

27. Miyazu YM. Telomerase expression in noncancerous bronchial epithelia isa possible marker of early development of lung cancer. Cancer Res 2005;65:9623–7.

28. Spira A, Beane J, Shah V, Liu G, Schembri F, Yang X, et al. Effects of cigarettesmoke on the human airway epithelial cell transcriptome. Proc Natl AcadSci U S A 2004;101:10143–8.

29. Chen G, Umelo IA, Lv S, Teugels E, Fostier K, Kronenberger P, et al. miR-146a inhibits cell growth, cell migration and induces apoptosis in non-small cell lung cancer cells. PLoS One 2013;8:e60317.

30. Labbaye C, Testa U. The emerging role of MIR-146A in the control ofhematopoiesis, immune function and cancer. J Hematol Oncol 2012;5:13.

31. Li G, Liu Y, Su Z, Ren S, Zhu G, Tian Y, et al. MicroRNA-324-3p regulatesnasopharyngeal carcinoma radioresistance by directly targeting WNT2B.Eur J Cancer 2013;49:2596–607.

32. NianW, Ao X,WuY, Huang Y, Shao J,Wang Y, et al.miR-223 functions as apotent tumor suppressor of the Lewis lung carcinoma cell line by targetinginsulin-like growth factor-1 receptor and cyclin-dependent kinase 2. OncolLett 2013;6:359–66.

33. Huang DW, Sherman BT, Lempicki RA. Systematic and integrative analysisof large gene lists using DAVID bioinformatics resources. Nat Protoc2009;4:44–57.

34. 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.

35. Kumaraswamy E, Wendt KL, Augustine LA, Stecklein SR, Sibala EC, LiD, et al. BRCA1 regulation of epidermal growth factor receptor(EGFR) expression in human breast cancer cells involves micro-RNA-146a and is critical for its tumor suppressor function. Oncogene2015;34:4333–46.

36. Bhaumik D, Scott GK, Schokrpur S, Patil CK, Orjalo AV, Rodier F, et al.MicroRNAs miR-146a/b negatively modulate the senescence-associatedinflammatory mediators IL-6 and IL-8. Aging 2009;1:402–11.

37. Mao X, Kikani CK, Riojas RA, Langlais P, Wang L, Ramos FJ, et al. APPL1binds to adiponectin receptors and mediates adiponectin signalling andfunction. Nat Cell Biol 2006;8:516–23.

38. Buechler C, Wanninger J, Neumeier M. Adiponectin receptor bindingproteins – recent advances in elucidating adiponectin signalling pathways.FEBS Lett 2010;584:4280–6.

39. Singh B, Reddy PG, Goberdhan A, Walsh C, Dao S, Ngai I, et al. p53regulates cell survival by inhibiting PIK3CA in squamous cell carcinomas.Genes Dev 2002;16:984–93.

40. Gustafson AM, Soldi R, Anderlind C, Scholand MB, Qian J, Zhang X, et al.Airway PI3K pathway activation is an early and reversible event in lungcancer development. Sci Transl Med 2010;2:26ra25.

www.aacrjournals.org Cancer Prev Res; 2017 OF9

Bronchial miRNA Expression in Lung Cancer

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098

Published OnlineFirst September 6, 2017.Cancer Prev Res   Ana B. Pavel, Joshua D. Campbell, Gang Liu, et al.   Cancer DetectionAlterations in Bronchial Airway miRNA Expression for Lung

  Updated version

  10.1158/1940-6207.CAPR-17-0098doi:

Access the most recent version of this article at:

  Material

Supplementary

  -17-0098.DC1

http://cancerpreventionresearch.aacrjournals.org/content/suppl/2017/09/02/1940-6207.CAPRAccess the most recent supplemental material at:

   

   

   

  E-mail alerts related to this article or journal.Sign up to receive free email-alerts

  Subscriptions

Reprints and

  .pubs@aacr.orgDepartment at

To order reprints of this article or to subscribe to the journal, contact the AACR Publications

  Permissions

  Rightslink site. (CCC)Click on "Request Permissions" which will take you to the Copyright Clearance Center's

.17-0098http://cancerpreventionresearch.aacrjournals.org/content/early/2017/10/12/1940-6207.CAPR-To request permission to re-use all or part of this article, use this link

for Cancer Research. on June 19, 2020. © 2017 American Associationcancerpreventionresearch.aacrjournals.org Downloaded from

Published OnlineFirst September 6, 2017; DOI: 10.1158/1940-6207.CAPR-17-0098