A MicroRNA Signature of Cigarette Smoking and Evidence for a Putative Causal Role of MicroRNAs in Smoking-Related Inflammation and Target Organ Damage

Abstract

Background

Cigarette smoking increases risk for multiple diseases. MicroRNAs (miRNAs) regulate gene expression and may play a role in smoking-induced target organ damage. We sought to describe a miRNA signature of cigarette smoking and relate it to smoking-associated clinical phenotypes, gene expression, and lung inflammatory signaling.

Methods and Results

Expression profiling of 283 miRNAs was conducted on whole blood-derived RNA from 5,023 Framingham Heart Study participants (54.0% women, mean age 55±13 years) using TaqMan assays and high-throughput RT-qPCR. Associations of miRNA expression with smoking status and associations of smoking-related miRNAs with inflammatory biomarkers and pulmonary function were tested with linear mixed-effects models. We identified a six-miRNA signature of smoking. Five of the six smoking-related miRNAs were associated with serum levels of C-reactive protein or interleukin-6; miR-1180 was associated with pulmonary function measures at a marginally significant level. Bioinformatic evaluation of smoking-associated genes coexpressed with the miRNA signature of cigarette smoking revealed enrichment for immune-related pathways. Smoking-associated miRNAs altered expression of select inflammatory mediators in cell culture gain-of-function assays.

Conclusions

We characterized a novel miRNA signature of cigarette smoking. The top miRNAs were associated with systemic inflammatory markers and reduced pulmonary function, correlated with expression of genes involved in immune function, and were sufficient to modulate inflammatory signaling. Our results highlight smoking-associated miRNAs and are consistent with the hypothesis that smoking-associated miRNAs serve as mediators of smoking-induced inflammation and target organ damage. These findings call for further mechanistic studies to explore the diagnostic and therapeutic utility of smoking-related miRNAs.

Introduction

Epidemiological studies have established the profound disease burden attributable to cigarette smoking, including coronary heart disease and stroke, chronic obstructive pulmonary disease, and lung cancer. Despite the steady increase in smoking cessation rates in the United States over the past 40 years, the percentage of smokers and number of tobacco-related deaths per year have risen worldwide, and the burden of the tobacco epidemic has shifted toward low- and middle-income countries.^1–4 Moreover, the biological mechanisms that mediate the associations between cigarette smoking and a diverse set of morbidities remain unclear.

MicroRNAs (miRNAs) are a class of highly-conserved short non-coding RNAs that function as post-transcriptional regulators of gene expression.^{5, 6} Over 2,000 miRNA sequences have been detected in the human transcriptome and are predicted to target more than half of all protein-coding genes.^{7, 8} Because miRNAs exert modest-sized effects on the expression of individual genes and some can target numerous transcripts, miRNAs are purported to coordinate the fine-tuning of gene expression under physiological conditions.^{9, 10} Supporting this theory, essential roles for miRNAs have been discovered in most cellular processes, including cell proliferation, development, differentiation, and apoptosis.^11–14 Moreover, aberrant miRNA expression has emerged as a hallmark of human diseases, starting with cancers but now including cardiovascular disease and inflammatory lung disease.^15–18 Most reported miRNA expression patterns in humans come from miRNA detection in cells or tissue from the target organs of disease, including the global downregulation of miRNA expression in alveolar macrophages and bronchial alveolar epithelium from smokers.^{19, 20}

In recent years, miRNAs were discovered to have reproducible signals and to be stable in blood.^21,22 These results have motivated efforts to profile circulating miRNAs in disease states, with the goal of harnessing their potential as non-invasive diagnostic biomarkers. Thus far, small cohort studies have profiled miRNA expression in induced sputum and plasma from cigarette smokers.^{12, 23} We sought to elucidate a comprehensive, whole blood-based miRNA profile for cigarette smoking that may (1) have predictive value for smoking-associated diseases and (2) shed light on the miRNA-mediated alterations in gene expression theorized to link smoking with smoking-related disease phenotypes. We hypothesized that miRNA profiling of whole blood would capture disrupted gene regulatory networks relevant to the inflammatory and immune responses that link the direct effects of cigarette smoking to target organ damage and smoking-related systemic diseases. To this end, we characterized a whole blood-derived miRNA signature of cigarette smoking in 5,023 Framingham Heart Study participants and evaluated the top miRNAs in relation to smoking-associated clinical phenotypes, smoking-associated gene expression, and inflammatory signaling in lung tissue.

Methods

Study Samples

The Framingham Heart Study (FHS), a community-based prospective study of cardiovascular disease and associated risk factors, was initiated in 1948 and has enrolled three generations of participants.^24–26 On-site evaluations of the Offspring and Third Generation cohort participants, including anthropometric data and biological sample collections, have been conducted every four to eight years. Samples from Offspring cohort Examination 8 (2005–2008; n=2,571) and Third Generation cohort Examination 2 (2008–2011; n=3,245) with complete smoking status data (n=5,816) were eligible for analysis. Non-fasting blood specimens were collected and stored at −80°C until assayed. Total RNA was isolated from peripheral whole blood samples (mean yield=3.50±1.66 µg) in PAXgene tubes (PreAnalytiX, Hombrechtikon, Switzerland) by Asuragen, Inc. (Austin, TX). RNA quality was assessed by total yield (>2 µg), 260/280 ratio (1.7–2.2) (absorbance reading in spectrophotometer at 260 nm and 280 nm; this ratio gives information about the purity of an RNA sample with 1.6-2.0 ratio showing ideal RNA purity), RNA integrity (RIN 7–10), amplification quality (>5 mcg), and cDNA size (median=400 nucleotides). After excluding 8 samples with poor RNA quality, a total of 5,808 samples were available for further analysis (Supplementary Figure 1A).

Clinical Measures

Participant smoking status was stratified as current, former, or never. Current smokers were defined as those who reported smoking on average at least one cigarette per day in the year prior to the FHS clinic examination, former smokers were individuals who reported smoking on average at least one cigarette per day in the past but had not smoked in the past year, and never smokers reported never smoking. Serum C-reactive protein (CRP) was measured in the Offspring cohort at Examination 8 and Third Generation at Examination 2 using a high sensitivity Dade-Behring BN 100 nephelometer.²⁷ Serum interleukin-6 was quantified in Offspring cohort participants at Examination 8 using the Quantikine HS Interleukin-6 Immunoassay kit (R&D Systems, Bio-Techne, Minneapolis, MN).²⁸ Interleukin-6 was not measured in the Third Generation cohort at Examination 2. Intra-assay coefficients of variation for inflammatory marker measurements were <9.2%.²⁹ Forced expiratory volume at one second (FEV1) and forced vital capacity (FVC) were measured on the Offspring cohort at Examination 8 using a water-filled spirometer (Collins Medical, Inc., Braintree, MA) and personal computer software (S&M Instruments, Doylestown, PA) and on the Third Generation at Examination 2 using a dry rolling-seal spirometer and Collins 2000 Pus/SQL software (Collins Medical, Inc., Braintree, MA). In both examinations the highest value among acceptable efforts was used, as per the American Thoracic Society-European Respiratory Society guidelines.30 In addition, the same staff training and quality control activities were used for both of these examinations. Raw values for FEV₁ and FVC were examined as a percentages of the predicted values, with predicted values calculated using standard reference equations derived from healthy never smokers in the NHANES III study.31 The FEV₁/FVC ratio was calculated from the highest acceptable individual measures. A dichotomous airflow obstruction variable was defined as the combination of FEV₁ <lower limit of normal and the FEV₁/FVC ratio <lower limit of normal, with lower limits of normal defined on the basis of published NHANES III data.³¹

Expression Profiling

miRNA profiling was performed using targeted TaqMan miRNA assays (Applied Biosystems, Foster City, CA). A high-throughput RT-qPCR platform was employed, which uses integrated fluidic circuits to perform miniaturized reactions at high efficiency with a batch size of 96 samples versus 96 miRNA targets (BioMark, Fluidigm Corp, San Francisco, CA). Profiling was completed at Boston University, Boston, MA, and the University of Massachusetts, Worcester, MA (laboratory of Dr. Jane Freedman). Reverse transcription and pre-amplification reactions were conducted in two separate pools (A and B), containing primers for 176 and 157 miRNAs, respectively, and carried out according to standard TaqMan protocols. MiRNAs were assayed in sub-panels (A1–2 and B1–2), each of which contained four “housekeeping” RNAs. Excellent reproducibility was demonstrated in preliminary profiling of 24 samples with 96 miRNA assays (mean CV ~3.5%). The analytic pipeline included an additional pilot study in which 754 miRNAs were measured on 455 Offspring cohort samples (Batch 1). The detectable subset of 345 miRNAs was measured in the remaining Offspring cohort samples (n=2,116) and the Third Generation cohort samples (n=3,245) (Batch 2). Of these, 5,729 samples passed quality control measures for miRNA profiling. Removed samples included those with <100 detectable miRNAs in Batch 1 or <20 detectable miRNAs in Batch 2 as well as nine samples recommended for removal by the laboratory technician during miRNA profiling. Samples with missing cell counts (N=705) and missing smoking status (N=1) were excluded, leaving a final sample size of 5,023 (Supplementary Figure 1A). From the set of detected miRNAs, we filtered out miRNAs (n=56) expressed in <500 participants. Outlier miRNA expression levels [threshold cycle (C_T) ≥5 standard deviations from the mean C_T] were treated as missing (a total of 32 outliers for six miRNAs were treated as missing). We further excluded six miRNAs that were identified as snoRNAs or tRNA fragments in the latest version of miRBase.⁷ The final miRNA panel consisted of 283 miRNAs (Supplementary Figure 1B). To normalize miRNA expression data, raw C_T values were adjusted for four technical variables: RNA isolation batch, RNA quality, RNA concentration, and 260/280 ratio. This method was chosen for the following reasons. First, reliable reference or housekeeping RNAs have not been well-characterized in whole blood. Second, exploratory models run on pilot study data using (1) adjustment for the small nuclear RNA RNU48 and (2) global mean normalization with mean C_T values for ~50 experimental miRNAs both resulted in over-fitting. Third, variance analysis demonstrated that these technical variables accounted for the greatest degree of variation in C_T values (40–60%).

Messenger RNA (mRNA) expression profiling of whole blood-derived RNA using the Affymetrix Human Exon Array ST 1.0 was performed as described previously.^{32, 33} Briefly, amplified cDNA was generated using the WT-Ovation Pico RNA Amplification System and processed with the WT-Ovation Exon Module (NuGEN, San Carlos, CA) in an automated genechip array station. Affymetrix Human Exon 1.0 ST microarrays were hybridized, washed, and scanned (Affymetrix, Santa Clara, CA). Microarray data for 17,873 genes were collected using the robust multi-chip average (RMA) method, and 5,626 samples passed mRNA quality controls (Supplementary Figure 1A). Data were normalized with RMA and further adjusted for technical covariates, as described previously.^32,33

Cell Culture, miRNA Transfection, & Cytokine Analysis

To evaluate the impact of smoking-associated miRNAs on the expression of inflammatory mediators in the lung, human lung epithelial A549 (ATCC CCL-185) cells were maintained in Ham’s F12 media (Life Technologies, Thermo Fischer Scientific Corp., Carlsbad, CA) supplemented with 10% FBS (Life Technologies) and 1x penicillin/streptomycin at 37°C and 5% CO2. Cells grown in six-well tissue culture dishes were transfected with human miRNA mimetics (Dharmacon, Inc., Lafayette, CO) of non-targeting (NT) sequence (CN-001000-01-05) or one of three smoking-associated miRNAs—miR-744-3p (MIMAT0004946), miR-1180-3p (MIMAT0005825), or miR-1285-3p (MIMAT0005876)—using DharmaFECT1 to achieve a final concentration of 50 nM. Smoking- associated miRNAs were selected for this functional experiment based on the strength of observed associations with smoking status, inflammatory biomarkers, and/or lung function and the putative involvement of these miRNAs in inflammatory and immune pathways based on gene ontology results from coexpression and target prediction analyses. After 24 hours of transfection, cells were stimulated with 10 ng/mL recombinant human TNF-α (R&D Systems, Bio-Techne, Minneapolis, MN) or vehicle for six hours. A panel of cytokines was measured in the collected cell supernatants using a multiplex bead array (R&D Systems). Cytokine expression was analyzed using a Liquichip100 (Qiagen, Hilden, Germany).

Statistical Methods

Statistical analyses were performed using SAS software version 9.2 (SAS Institute Inc., Cary, NC, USA) or R software version 3.1.1 (R Foundation for Statistical Computing, Vienna, Austria).³⁴ Linear mixed effects models with a random effect term to account for covariance due to familial relationships (the “lmekin” function in R) were used to test the association of miRNA expression with cigarette smoking as a three-level categorical variable (current/former/never). Expression of each miRNA was the dependent variable and smoking status was the exposure. Covariates included age, sex, cohort, four technical covariates (RNA isolation batch [50 levels], RNA quality, RNA concentration, and 260/280 ratio), and complete blood cell (CBC) count variables. CBC counts were imputed from mRNA expression values via Partial Least Square (PLS) prediction, with cross-validated prediction accuracy estimates ranging from 0.25–0.89.

False discovery rate (FDR) and q-values were estimated using the Benjamini-Hochberg method.³⁵ An FDR threshold of 0.10 was applied to identify significant miRNAs in the primary analysis. We then performed pairwise comparisons in (1) current vs. never smokers, (2) current vs. former smokers, and (3) former vs. never smokers. MiRNAs that passed the screening threshold were considered significant in pairwise comparisons at FDR<0.10. To demonstrate the reproducibility of observed associations, we conducted internal validation by splitting our sample into two groups of equal size by family into discovery and validation sets, with ~2,500 individuals in each, and analyzing miRNA associations in each set independently.

We used linear mixed effects models with a random effect term to account for covariance due to familial relationships (the “lmekin” function in R) to assess the association of smoking-related miRNAs with inflammatory markers and pulmonary function measures. MiRNA expression level was treated as the exposure variable. For inflammatory biomarker analyses, the log_e-transformed serum concentration of CRP or interleukin-6 was treated as the outcome. For pulmonary function analyses, the outcome variables were FEV₁, FVC, the ratio of FEV₁/FVC, and airflow obstruction, with raw values of FEV₁ and FVC examined as a percentages of the predicted values as described above.³¹ All models were adjusted for age, sex, cohort, smoking status, imputed CBC counts, and the four technical covariates described above. For associations with clinical phenotypes, the number of tests was modest, so the Bonferroni method was used to correct for multiple testing. The p-value threshold (p<1.4×10⁻³) was calculated by dividing 0.05 by the number of significant smoking-associated miRNAs (6) times the six clinical phenotypes tested [0.05/(6*6)].

For miRNA-mRNA coexpression analysis, pairwise coexpression of all profiled miRNAs and mRNAs was conducted using linear mixed effects models with fix effect terms including age, sex, and technical covariates, and with a random effect term to account for covariance due to genetic relationships in extended pedigrees (the “lmekin” function in R). MiRNA expression was the dependent variable, mRNA expression was the independent variable, and models were adjusted for age, sex, cohort, technical covariates, and imputed CBC counts. Split sample analysis was conducted, with significant coexpressed pairs in both discovery and validation sets (~2,800 samples each) identified from the large number of coexpressed pairs at the conservative threshold of FDR<0.01. The coexpression results were then overlapped with the set of significant miRNAs (FDR<0.10) and significant mRNAs (FDR<0.05) to identify an mRNA-miRNA coexpression signature for cigarette smoking. The more stringent FDR threshold for significant mRNAs than for significant miRNAs reflects the far greater number of mRNAs associated with smoking. To further interrogate the identified mRNA-miRNA correlations, we cross-referenced these pairs with three miRNA target databases: miRDB and miRNAorg (predicted targets) and miRTarBase (experimentally validated targets).

Separate gene ontology (GO) enrichment analyses were conducted on the sets of smoking-associated mRNAs coexpressed with each of our smoking-associated miRNAs based on functional classifications using all biological process terms in the GO Consortium database.³⁶ The enrichment p-value threshold for significance (p<6.1×10⁻⁵) was determined by the Bonferroni method, adjusting for 825 biological process terms in the GO database. GO analysis was also performed on coexpressed genes in which the mRNA was a predicted or validated target of its coexpressed miRNA. Due to the much smaller sample size, the analysis was performed on the aggregate results (rather than separately for each individual miRNA) and used a more liberal threshold (FDR<0.05) to define significant pathways. Network figures were generated using the JAVA-based visual analytic software tool ProteoLens.³⁷

For determining the potential effects of miRNA mimetic transfection on cytokine expression in lung epithelial cells, data were pooled from independent miRNA transfection experiments (n=3), which were analyzed and presented together. Analyses used log_e-transformed data in a two-way ANOVA with random block design across both factors (transfection and stimulation) to account for experiment-to-experiment variability. Dunnett’s test was applied post hoc to compare cytokine concentrations from cells transfected with miRNA mimetics to similarly stimulated cells that were instead transfected with NT control. Statistical analyses were performed using Prism 6.0 (GraphPad Software, Inc., San Diego, CA). A conservative p-value threshold was set at p<0.01.

Study Approval

All participants gave informed consent for participation in this study and collection of biosamples for genetic/genomic analysis. The study protocol was approved by the Boston University Medical Center Institutional Review Board.

Results

Study Sample Characteristics

Out of 5,023 participants (54.0% women, mean age 55±13 years) with data for miRNA profiling, 10% were current cigarette smokers (n=524), 41% were former smokers (n=2,079), and 48% were never smokers (n=2,420) (Table 1). Former smokers (mean age=60 years) were older than current smokers (51 years) or never smokers (52 years). Imputed WBC count was higher in current smokers (mean WBC=7.2) versus former smokers (mean WBC=6.1) and never smokers (mean WBC=5.9). Levels of inflammatory markers were higher and pulmonary function measures were lower in current versus former versus never smokers. Airflow obstruction was highest in current smokers (14.9%), lower in former smokers (6.2%), and lowest in never smokers (2.8%).

Table 1.

Clinical Characteristics

Clinical Characteristic*	Available Sample†	Current Smokers	Former Smokers	Never Smokers	P-value	Total Sample
Sample size (#)	5023	524 (10%)	2079 (41%)	2420 (48%)		5023
Age (years)	5023	51±11	60±13	52±13	<0.0001	55±13
Sex (% women)	5023	52%	55%	54%	0.52	54%
BMI (kg/m2)	5017	27.8±5.9	28.5±5.5	28.0±5.7	0.002	28.2±5.6
SBP (mm Hg)	5023	120±16	124±17	120±17	<0.0001	122±17
DBP (mm Hg)	5021	74±10	74±10	75±10	0.002	74±10
Total cholesterol (mg/dL)	5023	190±37	186±36	187±36	0.02	187±36
HDL-cholesterol (mg/dL)	5018	57±17	59±18	59.2±18	0.01	59±18
Glucose (mg/dL)	5023	101±21	103±22	99±21	<0.0001	101±21
Diabetes (% yes)	5018	6%	12%	7%	<0.0001	9%
BP Med. (% yes)	5023	26%	43%	29%	<0.0001	34%
Lipid Med. (% yes)	5023	24%	36%	22%	<0.0001	28%
Diabetes Med. (% yes)	5023	4%	8%	5%	<0.0001	6%
WBC (billion cells/L)	5023	7.2±1.4	6.1±1.3	5.9±1.3	<0.0001	6.1±1.4
CRP (mg/L)	5014	3.8±6.4	3.0±5.7	2.7±4.9	<0.0001	2.9±5.5
IL-6 (mg/L)	2,123	3.1±2.8	2.8±3.3	2.2±2.3	0.0001	2.6±3.0
FEV1 (pp‡)	4,648	0.92±0.16	0.97±0.16	1.00±0.14	<0.0001	0.98±0.15
FVC (pp‡)	4,648	1.00±0.13	1.02±0.14	1.03±0.13	0.0001	1.02±0.14
FEV1/FVC (pp†)	4,648	0.91±0.10	0.95±0.09	0.97±0.08	<0.0001	0.95±0.09
Airflow Obstruction (% pos)	4,648	15%	6%	3%	<0.0001	5%

^*Clinical characteristics are described as mean ± standard deviation except sample size, cohort, sex, diabetes, and medication use.

^†The maximum available sample size (N=5,023) reflects the number of subjects with miRNA data (N=5,729), excluding those with missing cell counts (N=705) and missing smoking status (N=1). Smaller available samples reflect missing data for the variable of interest.

^‡FEV₁, FVC, and FEV₁/FVC were examined as percentages of predicted values (pp), based on standard reference equations derived from healthy never smokers in the NHANES III study.

MiRNA Associations with Smoking Status

Six miRNAs were associated with three-level smoking status (Table 2). Each was detectable in over 4,400 participant samples with the exception of miR-342-5p, which was measurable in 1,115 samples. Five of these miRNAs were associated with current vs. never smokers, with all but one (miR-342-5p) underexpressed in current smokers. MiR-423-5p was the only marker differentially expressed in former versus never smokers, with decreased expression in former smokers. The distribution of miRNA C_T values by smoking status is shown in Supplementary Figure 2. Split sample analyses demonstrated a consistent pattern for t-statistics of miRNAs considered separately in the discovery and validation sets. For each significant miRNA, the directionality and scale of effect sizes in discovery and validation sets were consistent with the combined results (Supplementary Table 1).

Table 2.

MiRNA Associations with Smoking Status

		Three-level comparison		Pairwise comparison
miRNA	Sample size	P value	q-value*	Groups	Estimated β†	SE	P-value	Fold Change
				C v N	0.31	0.06	1.5×10−6	0.81
miR-1180	4905	8.1×10−6	0.003	C v F	0.24	0.07	2.7×10−4	0.85
				F v N	0.07	0.04	7.2×10−2	0.95
miR-181a-2-3p	4892	5.4×10−5	0.009	C v N	0.27	0.06	1.6×10−5	0.83
				C v F	0.19	0.06	3.3×10−3	0.88
				F v N	0.08	0.04	3.1×10−2	0.95
miR-423-5p	4418	1.3×10−4	0.014	C v N	0.16	0.1	1.0×10−1	0.90
				C v F	−0.09	0.1	3.3×10−1	1.06
				F v N	0.25	0.06	2.7×10−5	0.84
miR-25-5p	4853	1.6×10−4	0.014	C v N	0.36	0.09	4.9×10−5	0.78
				C v F	0.25	0.09	5.8×10−3	0.84
				F v N	0.11	0.05	4.2×10−2	0.93
miR-1285-3p	4636	8.6×10−4	0.057	C v N	0.32	0.08	1.8×10−4	0.80
				C v F	0.25	0.09	3.9×10−3	0.84
				F v N	0.07	0.05	1.9×10−1	0.95
miR-342-5p	1180	1.4 ×10−3	0.075	C v N	−0.64	0.18	4.6×10−4	1.56
				C v F	−0.41	0.19	2.8×10−2	1.33
				F v N	−0.23	0.12	5.5×10−2	1.17

^*We used the Benjamini-Hochberg method to calculate q-values (SAS Proc MULTTEST) and we applied a threshold of FDR<0.10 for the three-level analysis. MiRNAs are sorted by ascending q-value.

^†Estimated β coefficients represent the mean differences in miRNA C_T values between specific smoking groups, where higher C_T values reflect lower miRNA expression.

Abbreviations: C=current; F=former; N=never

MiRNA Associations with Smoking-related Clinical Phenotypes

Four smoking-associated miRNAs showed negative associations with log_e-transformed serum concentration of CRP: miR-1180 (p=2.0×10⁻¹⁹), miR-181a-2-3p (p=6.7 ×10⁻⁶), miR-423-5p (p=2.0×10⁻⁸), and miR-25-5p (p=1.3×10⁻⁷) (Table 3). One miRNA in our smoking signature, miR-1285-3p, was positively associated with the pro-inflammatory biomarker interleukin-6 (p=7.9×10⁻⁶). Analysis of significant miRNAs in relation to pulmonary function measures revealed that miR-1180 was negatively associated with prevalence of airflow obstruction (p=1.8×10⁻³) and positively associated with FEV₁ (p=9.5×10⁻³) at a marginally significant level (Table 4).

Table 3.

MiRNA Associations with Inflammatory Markers

	loge CRP				loge Interleukin-6
miRNA	Samples	Estimated β*	SE	P-value†	Samples	Estimated β*	SE	P-value†
miR-1180	5496	0.07	0.01	2.0×10−9	2434	−0.02	0.01	0.18
miR-181a-2-3p	5484	0.05	0.01	6.7×10−6	2389	−0.02	0.01	0.21
miR-423-5p	4975	0.05	0.01	2.0×10−8	1838	4.3×10−4	0.01	0.97
miR-25-5p	5435	0.04	0.01	1.3×10−7	2383	−0.01	0.01	0.23
miR-1285-3p	5145	0.01	0.01	0.27	2244	−0.05	0.01	7.9×10−6
miR-342-5p	1325	−0.01	0.02	0.40	593	0.04	0.02	0.02

^*Estimated β coefficients represent the mean differences in log concentration of the stated inflammatory marker per one standard deviation difference in miRNA C_T value, where higher C_T values reflect lower miRNA expression.

^†The p-value threshold for significance (p<1.4×10⁻³) was determined by the Bonferroni method [0.05/(6 smoking-associated miRNAs*6 clinical phenotypes tested)]. Significant p-values are shown in bold.

Table 4.

MiRNA Associations with Pulmonary Function Measures

	FEV1				FVC
miRNA	Samples	Estimated β*	SE	P-value†	Samples	Estimated β*	SE	P-value†
miR-1180	4539	−15.1	5.8	9.5×10−3	4539	−17.9	7.2	0.01
miR-181a-2-3p	4530	−1.1	6.1	0.86	4530	−2.3	7.5	0.76
miR-423-5p	4092	−6.4	4.4	0.14	4092	−8.7	5.4	0.11
miR-25-5p	4492	−2.6	4.2	0.53	4492	−8.6	5.3	0.10
miR-1285-3p	4299	4.7	4.6	0.30	4299	1.6	5.7	0.78
miR-342-5p	1100	−1.3	9.0	0.89	1100	5.4	11.1	0.63
	FEV1/FVC				Airflow Obstruction
miRNA	Samples	Estimated β*	SE	P-value†	Samples	Estimated β*	SE	P-value†
miR-1180	4539	−2.8×10−4	8.0×10−4	0.72	4539	0.01	2.7×10−3	1.8×10−3
miR-181a-2-3p	4530	7.2×10−6	8.3×10−4	0.99	4530	1.9×10−3	2.8×10−3	0.50
miR-423-5p	4092	1.6×10−4	5.9×10−4	0.78	4092	4.2×10−3	2.0×10−3	0.03
miR-25-5p	4492	8.3×10−4	5.7×10−4	0.15	4492	−1.3×10−3	1.9×10−3	0.51
miR-1285-3p	4299	5.9×10−4	6.3×10−4	0.35	4299	1.1×10−3	2.1×10−3	0.60
miR-342-5p	1100	−1.8×10−3	1.1×10−3	0.13	1100	0.01	4.1×10−3	0.10

^*Estimated β coefficients represent the mean differences in the stated pulmonary function measure per one standard deviation difference in miRNA C_T value, where higher C_T values reflect lower miRNA expression.

^†The p-value threshold for significance (p<1.4×10⁻³) was determined by the Bonferroni method [0.05/(6 smoking-associated miRNAs*6 clinical phenotypes tested)]. Marginally significant p-values are shown in bold.

MiRNA Correlations with Smoking-Related Genes

Expression levels of approximately 8,200 genes (nearly 50% of those measured) were associated with cigarette smoking at FDR<0.05. The majority of highly differentially expressed genes were upregulated in the context of cigarette smoking. Coexpression analysis identified 4,137 smoking-associated gene transcripts that were coexpressed with at least one of the 345 miRNAs measured and 2,116 smoking-associated gene transcripts that were coexpressed with one or more of the six smoking-associated miRNAs (at FDR<0.01 in the discovery and validation sets). There were 1,666 genes coexpressed with miR-25-5p, 524 with miR-1285-3p, 491 with miR-1180, 446 with miR-181a-2-3p, 275 with miR-423-5p, and none with miR-342-5p. We further identified 116 coexpressed pairs in which the mRNA was a predicted (n=111) or validated (n=5) target of the correlated miRNA. There were 17 predicted targets for miR-1180, 13 predicted for miR-181a-2-3p, 32 predicted for miR-1285-3p, five predicted for miR-25-5p, and five validated and 44 predicted for miR-423-5p (Supplementary Table 2).

GO Analysis

Independent GO enrichment analyses were conducted on the sets of smoking-associated mRNA transcripts that were coexpressed with each of our top miRNAs. Figure 1 shows the network of GO categories for the coexpressed mRNAs. Supplementary Tables 3 and 4 provide detailed information on the GO analysis and the overlapping genes in each category. The expression of miR-1180, our top smoking-associated miRNA, was negatively correlated with the expression of genes involved in monocyte differentiation (enrichment p=3.1×10⁻⁵). MiR-181a-2-3p expression was positively correlated with expression of genes that function in immune cell mediated cytotoxicity (enrichment p=5.7×10⁻⁵). MiR-181a-2-3p was negatively correlated with genes involved in viral response (enrichment p=4.7×10⁻⁶). Coexpressed mRNAs for the remaining miRNAs were enriched for a variety of common pathways, e.g., regulation of transcription and protein processing, which are known to be preferentially targeted by randomly selected sets of miRNAs.³⁸

Figure 1.

Network of Gene Ontology Categories for Smoking-Associated mRNAs Coexpressed with Smoking-Associated miRNAs. Smoking-associated gene transcripts that are coexpressed with one or more smoking-associated miRNAs and represented by one or more GO categories. Only gene transcripts belonging to GO categories meeting the Bonferroni-corrected enrichment p-value threshold of p<10⁻⁵, are included (see Supplementary Tables 3 and 4 for GO results). Blue lines denote negative correlations, while purple lines indicate positive correlations between miRNA and mRNA expression. This figure was drawn by ProteoLens.³⁷

We also performed GO enrichment analysis on the set of miRNA-mRNA coexpressed pairs in which the mRNA was a predicted (n=112) or validated (n=5) target of the correlated miRNA. In the aggregate results, the GO term with the greatest fold enrichment was T cell homeostasis (enrichment p=4.2×10⁻⁴ ; q-value=0.038) (Supplementary Table 5). Two of the four genes in the T cell homeostasis pathway were predicted targets of miR-1180: BAX (coexpression q-value=7.2×10⁻⁸) and P2RX7 (coexpression q-value=6.1×10⁻⁸); expression of miR-1180 was negatively correlated with expression of BAX and positively correlated with expression of P2RX7 (Supplementary Table 2). Other enriched GO terms, such as regulation of gene expression, represent common functional pathways, as described above.

miRNA Effects on Inflammatory Mediators

To determine whether individual miRNAs that are associated with cigarette smoking status and smoking-induced inflammation could be sufficient to modulate the expression of inflammatory mediators, we measured cytokine elaboration by human lung epithelial cells separately transfected with mimetics for miR-1180 and miR-1285-3p A non-targeting (NT) miRNA mimetic served as a negative control. Expression levels of eight cytokines were quantified in cell supernatants: interleukin-6, interleukin-8, granulocyte macrophage colony-stimulating factor (GM-CSF), granulocyte colony-stimulating factor (G-CSF), chemokine C-C motif ligand 2 (CCL2), chemokine C-C motif ligand 20 (CCL20), chemokine C-X-C motif ligand 5 (CXCL5), and chemokine C-X-C motif ligand 6 (CXCL6). Transfection of miR-1180 reduced production of GM-CSF (p=0.01) in stimulated cells (Figure 2). Basal levels of CXCL5 and CXCL6 were increased in response to transfection of both miRNAs, and miR-1285-3p transfection also increased production of CXCL5 (p=0.003) in stimulated cells. Other cytokines tested, including interleukin-6, interleukin-8, G-CSF, CCL20, and CCL2, were not significantly affected by the miRNAs examined.

Figure 2.

Smoking-Related miRNAs Alter Cytokine Expression in Vitro. Multiplex cytokine bead array analysis of A549 cell culture supernatants after transfection with non-targeting (NT), miR-1180, or miR-1285-3p mimetics. Cells were stimulated 24 hours post-transfection with vehicle or 10 ng/mL rhTNF-α for 6 hours. Squares, triangles, and circles represent data collected from three independent experiments, normalized to the cytokine concentrations in the negative control group (NT vehicle). The * indicates p<0.01 (significant p-values shown) when compared to NT control within a given stimulation group as determined by a two-way ANOVA followed by Dunnett’s test, using raw rather than normalized values.

Discussion

Cigarette smoking acutely promotes oxidative stress, as evidenced by increased plasma peroxides and decreased antioxidants, and chronic smoking leads to systemic inflammation characterized by the upregulation of circulating inflammatory cells (primarily neutrophils, lymphocytes, and platelets), inflammatory mediators, and acute-phase proteins.^{39, 40} To test our hypothesis that miRNAs may have predictive value for smoking-associated diseases, we characterized a comprehensive miRNA signature of cigarette smoking using high-throughput RT-qPCR profiling of whole blood RNA from a large, population-based sample.

We identified six miRNAs that were characteristically dysregulated in the context of smoking, five of which were downregulated in smokers versus non-smokers (Table 2). These results are supported by the widespread downregulation of miRNAs found in cigarette smoke-exposed lung tissue from mouse models and human samples.^{12, 20, 41–43} Our whole blood-derived miRNA signature of cigarette smoking is novel, and the results show little overlap with previous studies. This lack of concordance is common across miRNA-smoking studies, and can be explained by diverse tissue sources, non-uniform tobacco exposures, and different miRNA platforms. Two of our top hits —miR-1180 and miR-1285-3p— are recently identified miRNAs^{44, 45} that have been linked to cancer-specific gene regulatory networks; miR-1180 has been shown to inhibit cell apoptosis in human hepatocellular carcinoma both in vivo and in vitro, while miR-1285-3p inhibits the expression of tumor suppressor p53.^{46, 47} MiR-181a-2-3p is involved in the immune response as a positive regulator of B-cell development and T-cell sensitivity.^{48, 49} The remaining three miRNAs have been associated with multiple human cancers, and miR-423-5p and miR-25-5p have been linked to heart disease.^{50, 51} Dysregulation of circulating miR-342-5p has been found in autoimmune conditions; miR-342-5p has been shown to promote inflammatory activation of macrophages in atherosclerotic lesions, consistent with our finding that current cigarette smokers exhibited higher miR-342-5p expression compared to never smokers.^52–54

Whole blood is readily accessible and can be stored for later RNA extraction. This stable form of blood-derived RNA can be isolated with relative ease and high purity, making miRNA results in whole blood particularly amenable to clinical translation. miRNAs detected in whole blood may be present in leukocytes, platelets, erythrocytes, and/or cell-free plasma.⁵⁵ In previous work using the same miRNA detection platform and a small sample of healthy volunteers, several of our coauthors defined the distributions of over 600 circulating miRNAs in whole blood, peripheral blood mononuclear cells (PBMCs), platelets, plasma, and serum.⁵⁶ Of our significant miRNAs, three (miR-181a-2-3p, miR-25-5p, and miR-423-5p) were detectable in whole blood and PBMCs only, while the remaining three (miR-1180, miR-1285-3p, and miR-342-5p) were additionally detectable in platelets.⁵⁶ These data suggest that our miRNA expression patterns are largely derived from leukocytes, and secondarily from platelets. We may also have captured signals from small amounts of cell-free miRNAs, which can be released from or targeted to smoking-relevant tissues. Because miRNA profiling in whole blood cannot distinguish cell type-specific effects, further studies are needed to localize the observed miRNA patterns to specific blood components.

To explore the functional significance of the cigarette smoking-related miRNAs, we evaluated their associations with smoking-related clinical phenotypes: two biomarkers of systemic inflammation and four measures of pulmonary function. Population-based studies have demonstrated that the systemic inflammatory state induced by cigarette smoking is associated with increased levels of acute-phase proteins (CRP, fibrinogen, etc.) and pro-inflammatory cytokines (interleukin-6, TNF-α, etc.) in the peripheral circulation.^57–59 In the Framingham Heart Study, serum CRP and interleukin-6 levels were significantly higher in current cigarette smokers than never smokers.⁶⁰ We investigated the relations of smoking-associated miRNAs to systemic inflammation by relating our significant miRNAs to serum concentrations of CRP and interleukin-6. Four miRNAs were negatively associated with CRP: miR-1180, miR-181a-2-3p, miR-423-5p, and miR-25-5p (Table 3). In support of these findings, mir-181a-2-3p has been shown to downregulate inflammatory factors in monocytes and macrophages.⁶¹ MiR-25-5p is downregulated in human airway smooth muscle cells in response to pro-inflammatory stimuli and is proposed to regulate expression of multiple inflammatory mediators in these cells.⁶² Elevated levels of miR-423-5p were found in surgical lung biopsies of idiopathic pulmonary fibrosis, in which inflammation may play a critical role.^{63, 64} One miRNA, miR-1285-3p, was positively associated with interleukin-6 (Table 3). Additional studies may shed light on the discrepancy between miRNA associations with CRP and with interleukin-6 observed in our cross-sectional analysis.

Cigarette smoking is an established cause of impaired pulmonary function.⁶⁵ Smoking is the major risk factor for chronic obstructive pulmonary disease (COPD), which is defined by irreversible airflow obstruction and is the fourth leading cause of death worldwide.⁶⁶ Smoking-induced COPD is driven by inflammation, as the activation of neutrophils, macrophages, and lymphocytes and associated secretion of inflammatory mediators trigger a chronic inflammatory response in the lung.⁶⁷ Premature senescence of epithelial and immune cells, the latter of which heightens inflammation, has also been implicated in the pathogenesis of smoking-induced COPD.⁶⁸ Nevertheless, the causal mechanisms and molecular players underlying COPD progression in the lungs of smokers have yet to be elucidated. Analysis of our significant miRNAs in relation to pulmonary function measures revealed that miR-1180 was negatively associated with airflow obstruction and positively associated with FEV₁ at a marginally significant level (Table 4). Although our cross-sectional results suggest that miRNAs may serve as biomarkers for smoking-related inflammation and target organ damage, we cannot deduce with certainty the directionality of observed associations.

To better understand the mechanistic connections between dysregulated miRNA expression, systemic inflammation, and lung dysfunction in the context of cigarette smoking, we identified over 1,600 smoking-associated genes that were coexpressed with one or more of our top miRNAs in whole blood. GO analysis of the mRNAs correlated with each miRNA showed enrichment for immune-related pathways among genes coexpressed with miR-1180 and miR-181a-2-3p (Figure 1 and Supplementary Table 3). MiR-1180 expression was negatively correlated with expression of genes involved in differentiation of several immune cell types. MiR-181a-2-3p level was positively correlated with genes for immune cell-mediated cytotoxicity and negatively correlated with viral response genes (Figure 1 and Supplementary Table 3). We further interrogated the regulatory networks that these miRNAs may be involved in by cross-referencing our coexpressed pairs with miRNA target databases. We identified 116 coexpressed pairs in which the mRNA was a predicted or validated target of its correlated miRNA. GO analysis of the predicted targets highlighted T-cell homeostasis, and two of four overlapping genes in this pathway were predicted targets of miR-1180 (Supplementary Tables 2 and 5).

We hypothesize that our miRNA-mRNA coexpressed pairs, especially those involving predicted targets, describe gene regulatory networks primarily localized to leukocytes and platelets. Loss-of-function experiments have demonstrated the broad importance of miRNAs in development, including the developmental programs of B cells, T cells, and other leukocytes.⁶⁹ Moreover, specific “immuno-miRNAs,” including miR-181a-2-3p, have been shown to regulate lymphocyte differentiation and function.⁷⁰ Although platelets are anuclear, they contain cytoplasmic mRNA, carry out protein translation, and express over 280 miRNAs, and platelet miRNA-mRNA coexpression profiles have been correlated with platelet reactivity.^{71, 72} Taken together, these bioinformatic results suggest that smoking-induced dysregulation of the aforementioned miRNAs may play a mediating role in inflammatory responses by modulating miRNA-specific gene regulatory networks in leukocytes and, to a lesser extent, platelets.

Our study has several strengths, including the large community-based cohort of young, middle-aged, and older adults and the ascertainment of miRNAs, inflammatory markers, and pulmonary function measures adhering to rigorous quality control protocols. Our cross-sectional results highlight smoking-associated miRNAs that have the potential to serve as non-invasive biomarkers of smoking exposure or smoking-related diseases. Moreover, our functional evidence suggesting that these miRNAs can mediate lung inflammation combined with our bioinformatic identification of coexpressed predicted gene targets is a first step toward determining disease-relevant miRNA targets that may be funneled into emerging clinical pipelines for RNA interference-based therapeutics.⁷³ Our results highlight select smoking-associated miRNAs and are consistent with the hypothesis that smoking-associated miRNAs function as mediators of cigarette smoking-induced inflammation and downstream target organ damage (Figure 3). Additional functional studies are needed to assess whether these miRNAs are causally associated with smoking-related phenotypes.

Figure 3.

Proposed Roles for Smoking-Associated miRNAs in Mediating Target Organ Damage. Thick arrows indicate established relationships, while thin arrows indicate observed associations or correlations based on epidemiological analyses (Tables 2–4), bioinformatics approaches (Supplementary Tables 2–5), and functional experiment (Figure 2).

We also acknowledge several limitations. Our study participants were largely of European ancestry, and the generalizability of our findings to other races/ethnicities is uncertain. Because we analyzed results across a large number of miRNAs, mRNAs, and miRNA-mRNA coexpressed pairs, our study is susceptible to both false positive and false negative findings despite adjustment for multiple testing. MiRNA profiling was conducted on whole blood and as such may not reflect tissue-specific patterns of expression – even though we adjusted for differential blood cell composition imputed from mRNA expression values. Although we present evidence of robust validation of our cross-sectional results based on analysis of split samples for discovery and validation, our study lacks external replication. We cannot exclude residual confounding or establish causal associations in the cross-sectional results. Our functional study also has limitations: small sample size (n=3 independent experiments); single time-point collection; restriction to one cell type; and reliance on miRNA overexpression, which could target genes that would not be affected under physiological conditions.

The proposed identification of gene regulatory networks involving coexpressed miRNA-mRNA pairs assumes temporal sequence and spatial co-localization, neither of which can be proven by cross-sectional correlations in whole blood. It is possible that different cell types or blood components account for changes in miRNA and mRNA expression and that tissue specificity precludes their joint participation in intracellular regulatory networks. Observed correlations may also indicate common consequences of smoking-induced damage rather than mutual involvement in mediating pathways. We addressed these issues in part by identifying coexpressed pairs in which the mRNA was a predicted target of the correlated miRNA. However, we were underpowered to study miRNA-specific pathway enrichment, and the functional relevance of computational target prediction remains unclear. Because gene repression by miRNAs is generally modest and tunable, whether a target is actually affected depends on the local concentrations of both miRNA and mRNA, such that predicted relationships may be irrelevant in a given tissue.⁷⁴ Furthermore, predicted target analysis only addresses direct targeting relationships outside the context of intact cells, but miRNAs act within complex regulatory networks that cannot be fully captured by first-order correlations.⁷⁵

Because our bioinformatics analysis linked the smoking-associated miRNAs to inflammation, and because miRNAs can target transcripts for cytokines and for signaling intermediates upstream of cytokine expression,^76–80 we tested whether these miRNAs were capable of modulating inflammatory cytokines relevant to cigarette smoking. To our knowledge, none of these miRNAs has previously been connected to pulmonary inflammation. We observed miRNA-specific effects on cytokine elaboration by lung cells for the two miRNAs tested, with miR-1180 reducing production of GM-CSF, and miR-1180 and miR-1285-3p increasing levels of CXCL5 and CXCL6 (Figure 2). Although miR-1285-3p was positively associated with interleukin-6 in the cross-sectional analysis, overexpression of miR-1285-3p in lung cells did not significantly affect interleukin-6 levels. GM-CSF drives alveolar macrophage maturation and innate immune activities in the lung.⁸¹ CXCL5 and CXCL6 are lung epithelium-derived ligands for the CXCR2 receptor on neutrophils; these chemokines recruit their cognate leukocytes to the lungs during pulmonary inflammation.^82,83 Each of these cytokine pathways influences cigarette smoking-induced inflammation and disease,^84–88 so their regulation by miRNA changes could impact the health of cigarette smokers. The cytokine changes suggest that decreased levels of these miRNAs could skew smoking-induced inflammation towards mononuclear cell inflammation and away from neutrophilic infiltration. Future studies are needed to determine the mechanisms responsible for regulating miRNA levels affected by cigarette smoking, to identify the transcripts targeted by these miRNAs that modulate cytokine expression, and to elucidate the net effects of these multiple miRNA changes on integrated inflammatory responses to cigarette smoke. Our results provide evidence in support of the hypothesis that the miRNAs associated with smoking, inflammation, and pulmonary function in an observational study are functionally capable of dictating expression of the cytokines governing inflammation and disease in cigarette smokers.

Clinical Perspective.

Cigarette smoking increases risk for cancer, lung disease, and cardiovascular disease. MicroRNAs (miRNAs) regulate gene expression and may play a role in smoking-induced target organ damage. We sought to identify a miRNA signature of cigarette smoking and relate it to smoking-associated clinical phenotypes, gene expression, and lung inflammatory signaling. By studying miRNA expression levels in approximately 5000 people with cigarette smoking histories, we identified a six-miRNA signature of smoking. The top miRNAs were associated with systemic inflammatory markers and reduced pulmonary function, correlated with expression of genes involved in immune function, and were sufficient to modulate inflammatory signaling. Our results highlight smoking-associated miRNAs and are consistent with the hypothesis that smoking-associated miRNAs serve as mediators of smoking-induced inflammation and target organ damage. These findings call for further mechanistic studies to explore the diagnostic and therapeutic utility of smoking-related miRNAs.