Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2020 Apr 1.
Published in final edited form as: Genet Epidemiol. 2019 Feb 11;43(3):318–329. doi: 10.1002/gepi.22192

Assessing pleiotropy and mediation in genetic loci associated with chronic obstructive pulmonary disease

Margaret M Parker 1, Sharon M Lutz 2, Brian D Hobbs 1,3, Robert Busch 1,3, MerryLynn N McDonald 4, Peter J Castaldi 1,5, Terri H Beaty 6, John E Hokanson 7, Edwin K Silverman 1,3, Michael H Cho 1,3
PMCID: PMC6416067  NIHMSID: NIHMS1008736  PMID: 30740764

Abstract

Genetic association studies have increasingly recognized variant effects on multiple phenotypes. Chronic obstructive pulmonary disease (COPD) is a heterogeneous disease with environmental and genetic causes. Multiple genetic variants have been associated with COPD, many of which show significant associations to additional phenotypes. However, it is unknown if these associations represent biological pleiotropy or if they exist through correlation of related phenotypes (“mediated pleiotropy”). Using 6,670 subjects from the COPDGene study, we describe the association of known COPD susceptibility loci with other COPD-related phenotypes and distinguish if these act directly on the phenotypes (i.e. biological pleiotropy) or if association is due to correlation (i.e. mediated pleiotropy). We identified additional associated phenotypes for 13 of 25 known COPD loci. Tests for pleiotropy between genotype and associated outcomes were significant for all loci. In cases of significant pleiotropy, we performed mediation analysis to test if SNPs had a direct association to phenotype. Most loci showed a mediated effect through the hypothesized causal pathway. However, many loci also had direct associations, suggesting causal explanations (i.e. emphysema leading to reduced lung function) are incomplete. Our results highlight the high degree of pleiotropy in complex disease-associated loci and provide novel insights into the mechanisms underlying COPD.

Keywords: Pleiotropy, Mediation, COPD, Emphysema, Causal Inference, GWAS

BACKGROUND

Chronic obstructive pulmonary disease (COPD) is one of the leading causes of respiratory mortality worldwide (Soriano et al., 2017). COPD is diagnosed by a single test of post-bronchodilator airflow limitation (measured by spirometry), but the disease is highly heterogeneous. Clinicians have noted for decades that clinical manifestations of COPD including emphysema, chronic bronchitis, exercise capacity, and hypoxemia vary widely among patients with similar spirometric measurements (Burrows B, Fletcher CM, Heard BE, Jones NL, 1966). More recently, analysis of computed tomography (CT) images has led to the identification of measures, such as quantitative emphysema, airway wall thickening, and functional small airways disease, that are associated with COPD outcomes independent of spirometry (Galbán et al., 2012; Han et al., 2011; Nakano et al., 2005). Importantly, many of these CT measurements reflect physiological processes which lead to the phenomenon of airflow limitation (Hogg, 2004).

COPD is caused by both environmental (predominantly cigarette smoking) and genetic factors. The heritability of COPD case/control status has been estimated to be between 35–60% (Ingebrigtsen et al., 2010; Zhou et al., 2015), and evidence suggests that the components of COPD (i.e. emphysema and airway disease) may be independently heritable (Patel et al., 2008). To assess the genetic component of COPD, large genome-wide association studies (GWAS) studies have been performed, identifying more than 20 loci associated with COPD case/control status (Cho et al., 2014; Hobbs et al., 2017). Many of these loci also show significant associations with other COPD-related phenotypes (e.g. cigarette smoking (Consortium, 2010; Thorgeirsson et al., 2010), emphysema (Cho et al., 2015a)), a phenomenon known as genetic pleiotropy. There are multiple types of pleiotropy (Paaby & Rockman, 2013; Solovieff, Cotsapas, Lee, Purcell, & Smoller, 2013) and different types have different implications for the biological mechanism underlying disease. As defined in Solovieff et al (Solovieff et al., 2013), “biological pleiotropy” refers to a single locus that has independent effects on multiple phenotypes (i.e. locus → phenotype 1 and locus → phenotype 2). Alternatively, “mediated pleiotropy” refers to locus that has an effect on a phenotype only through its correlation to another phenotype (i.e. locus is associated with phenotype 2 because the causal mechanism is locus → phenotype 1→ phenotype 2). Although biologically interesting, mediated pleiotropy implies that the locus-to-outcome association is only observed because it is preceded by a true association on the causal pathway, and therefore only biological pleiotropy is considered “genuine” pleiotropy.

The aims of this manuscript are two-fold: 1) describe the association of known COPD loci with other COPD- related phenotypes; and 2) distinguish if these loci act directly on a set of COPD-related traits (i.e. biological pleiotropy) or if the association is predominately due to correlation (i.e. mediated pleiotropy), with particular attention to the hypothesized causal pathways of emphysema and airway measurements on lung function (Burrows B, Fletcher CM, Heard BE, Jones NL, 1966; Galbán et al., 2012; Patel et al., 2008). To achieve this, we used 6,670 extensively phenotyped Non-Hispanic White subjects of the COPDGene study2, a study of smokers with and without COPD.

METHODS

Study Design

The complete study protocol for COPDGene had been described in detail elsewhere (Cho et al., 2010; Regan et al., 2011), but briefly, we enrolled 10,192 self-identified Non-Hispanic Whites and African Americans between the ages of 45 and 80 years with a minimum of 10 pack-years lifetime smoking history. Participants were extensively phenotyped, including pre- and post- bronchodilator spirometry and computed tomography (CT) of the chest. The known-COPD loci we tested were originally identified in European-derived populations, and therefore, this analysis includes only Non-Hispanic White participants of COPDGene (n=6,670). Their clinical characteristics are presented in Table 1.

Table 1.

Complete list of 11 tested phenotypes and covariates, study population mean and standard deviation and measurement method.

Phenotypes N Mean (SD) Description
Age 6670 62.09 (8.84)
Sex (% Male) 6670 52.40%
Pack-years of lifetime smoking 6670 47.31 (26.02) 1 pack-year = 1 pack of cigarettes smoked daily for 1 year
Current smokers 6670 39.20%
FEV1 6651 2.21 (0.95) Lung function: Forced expiratory volume in 1 second
FEV1/FVC 6651 0.64 (0.17) Lung function: FEV1 / Forced vital capacity
Emphysema 6298 0.97 (1.60) Quantitative emphysema, using log of −950 Hounsfield Units
Cigarettes per day 6670 25.84 (11.44) Cigarettes smoked per day, averaged over lifetime
SGRQ 6669 26.59 (22.65) Symptoms: St George’s Respiratory Questionnaire
BDR (% FEV1 change) 6590 6.32 (9.50) Pre/Post-bronchodilator FEV1 % change
BMI 6670 28.68 (6.06) Body Mass Index
Emphysema ratio 6192 1.81 (7.5) Ratio of emphysema in upper third to lower third from inspiratory chest CT
Total lung capacity (%P) 6304 98.09 (16.22) Total lung capacity, percent of predicted value
Chronic Bronchitis 6670 20.94% Chronic cough and phlegm for at least three months per year for at least two years
Pi10 6163 3.67 (0.14) Airway thickness: square root of wall area for bronchus with 10mm lumen perimeter
Pi15 6163 5.14 (0.19) Airway thickness: square root of wall area for bronchus with 15mm lumen perimeter
Open in a new tab

FEV1 = forced expiratory volume in 1 second, FVC = forced vital capacity, SGRQ= St. George’s Respiratory Questionnaire, BMI = body mass index, BDR = bronchodilator response, Pi10, Pi15 = standardized airway wall thickness (square root of wall area) for an airway of size 10mm and 15mm, CT= computed tomography, %P = percent predicted.

Genetic loci for assessment

All study participants were genotyped using the Illumina OmniExpress BeadChip imputed versus the 1000 Genomes phase 1 version 3 reference panel (Siva, 2008). We focused on loci that have been previously identified in large GWAS of COPD or emphysema as genome-wide significant (P < 5 ×10−8), including 22 COPD loci identified from a recent case-control analysis (Hobbs et al., 2017), rs626750 in DLC1 (Cho et al., 2015a) (associated with emphysema), and rs75200691 in MMP12 (Cho et al., 2014) (associated with severe COPD) and rs28929474 in SERPINA1 (D L DeMeo, 2004; Foreman et al., 2017) (a known Mendelian locus for COPD). All loci had <1% missing values, were in Hardy-Weinberg equilibrium in control subjects (P > 1 ×10−6) and had a population frequency > 1% in Non-Hispanic Whites.

The primary confounder of single nucleotide polymorphism (SNP) to phenotype associations is genetic ancestry. To control for this, we estimated principal components (PCs) that summarize genetic ancestry to include as covariates in all analyses. PCs were estimated in the Eigenstrat program (AL Price, NJ Patterson, RM Plenge, ME Weinblatt, & NA Shadick, 2006; Alkes L Price, Zaitlen, Reich, & Patterson, 2010), using 32,033 uncorrelated SNPs with minor allele frequency > 0.05 as input, as previously described (Cho et al., 2014).

Phenotypes for assessment

Phenotypes collected in COPDGene include measures of spirometry, quantitative imaging, biomarkers, and questionnaire data. From these, we selected a subset to analyze based on: 1) minimal missingness (<10%); 2) strength of association with known COPD loci (in the top 5 associated phenotypes for at least one of the 25 SNPs); and 3) Pearson correlation < 0.7 with any other phenotype (calculated after transformations for non-normality). This resulted in 11 phenotypes for analysis, representing a broad range of clinically important metrics of COPD. These include: 1) the forced expiratory volume in one second over the forced vital capacity (FEV1/FVC); 2) emphysema (measured as the percentage of low attenuation area less than – 950 Hounsfield Units (HU)); 3) average cigarettes smoked per day: 4) the St. George’s respiratory questionnaire score (SGRQ) that summarizes health-related quality of life; 5) bronchodilator responsiveness (measured as the absolute difference in FEV1 before versus after bronchodilation); 6) body mass index (BMI); 7) the ratio of emphysema at −950 H (HU) in the upper third versus lower third of the lung; 8) total lung capacity as assessed on inspiratory chest CT; 9) chronic bronchitis (based on chronic cough and phlegm for at least three months per year for at least two years); 10) standardized airway wall thickness for an airway of size 10mm internal perimeter (pi10) and; 11) standardized airway wall thickness for an airway of size 15 mm internal perimeter (pi15). A detailed list of the 11 analyzed phenotypes and how they were collected is presented in Table 1.

Statistical analysis

We tested each of the 25 SNPs for association with each of the 11 phenotypes using linear or logistic regression. All outcomes were adjusted for age, sex, and principal components of genetic ancestry. Spirometry (FEV1/FVC), imaging (emphysema, Pi10, Pi15, emphysema ratio, and total lung capacity), and symptom-related phenotypes (chronic bronchitis, SGRQ and bronchodilator responsiveness) were also adjusted for pack-years of smoking; Imaging, spirometry, and chronic bronchitis were also adjusted for current smoking status; bronchodilator response was also adjusted for baseline FEV1; and imaging phenotypes were also adjusted for scanner type. Non-normally distributed phenotypes were either log transformed (BMI, emphysema) or inverse normal transformed (cigarettes smoked per day, emphysema ratio). We defined significant associations using a Bonferroni corrected P of 0.002 (0.05/25).

Tests for Pleiotropy

SNPs with greater than 1 associated phenotype were tested for pleiotropy via the R package pleiotropy (Lutz, Fingerlin, Hokanson, & Lange, 2016), which formally tests whether multiple phenotypes are associated with a single locus of interest using a permutation-based framework. For each SNP associated with multiple outcomes, we test for pleiotropy using the “pleiotropySNP” function with 10,000 permutations and included covariates defined in Supplemental Table 1. The method permutes the 2 (or more) phenotypes of interest and compares observed P-values of association to a set of permuted P-values. We used the “cutoff” approach to evaluate this comparison, as defined in Lutz et al. (Lutz et al., 2016). A multiple-comparison corrected P-value of P < 0.0038 was considered significant pleiotropy.

Test for Mediation

The pleiotropy method tests whether a SNP is associated with multiple outcomes. However, it cannot distinguish between biological and mediated pleiotropy, and significant results may be due to correlation between phenotypes. Therefore, we performed mediation analysis using the causal inference approach to mediation analysis (Pearl, 2001; Vanderweele & Vansteelandt, 2009), which allows us to asses if a SNP has a direct effect on the outcome of interest or if the SNP only acts on the phenotype of interest through a secondary phenotype (i.e. a mediated effect). Differentiating between these two scenarios the necessary to understand the biological mechanism through which a locus affects multiple phenotypes.

We used the R package Mediation (Imai, Keele, & Tingley, 2010), which uses a three-step procedure to estimate the average causal mediation effects (ACME) and the average causal direct effect (ADE). The first step is to estimate the distribution of the mediator for each genotype, using a model of the mediator as a function of genotype (e.g. emphysema ~ genotype + PC1 + PC2 +PC3 + PC4 + PC5). The next step is to estimate the phenotype’s potential outcomes for each of the genotype and mediator values. This is achieved by fitting a model of the phenotype as a function of the genotype, mediator and covariates (e.g. FEV1/FVC ~ emphysema + genotype + covariates). The final step combines the fitted models with the “mediate()” function, providing estimates and p-values of direct and indirect effects of genotype on outcome. All outcomes and mediators in our analysis were continuous measurements, so linear regressions were used to estimate inputs to the mediate function. The first five PCs summarizing genetic ancestry were included as covariates of the SNP to mediator relationship, as we expect the primary confounder of this relationship to be genetic ancestry. Confounders defined in Supplemental Table 1 were included as covariates of the mediator to outcome relationships.

Mediation analysis requires that the causal model (i.e. the temporal sequence of phenotypes) is pre-specified. We based causal models on current knowledge of COPD pathophysiology in order to determine the mediator and the outcome in each mediation analysis. In cases where associations included both emphysema and FEV1/FVC, emphysema was assumed to be the mediator and FEV1/FVC was assumed to be the outcome (Burrows B, Fletcher CM, Heard BE, Jones NL, 1966; Galbán et al., 2012). In cases where cigarette smoking was associated with a SNP, it was assumed to be the mediator when using other outcomes (e.g. FEV1/FVC) (Consortium, 2010; Thorgeirsson et al., 2010). In the case where a SNP was associated with BMI and emphysema, emphysema was assumed to be the mediator and BMI was assumed to be the outcome (Coxson et al., 2004). In all other cases FEV1/FVC was assumed to be the mediator for outcomes.

RESULTS

We analyzed 11 clinically important COPD phenotypes for their association with 25 genetic loci known to affect disease susceptibility. To distinguish independent signals, we chose to assess phenotypes that were not highly correlated (Pearson correlation < 0.7) (Supplementary Figure 1). Of the 25 previously identified COPD loci, 13 were significantly associated with more than one phenotype at P < 0.002 in our dataset (Figure 1, Table 2). The outcome associated with the greatest number of known COPD susceptibility loci was FEV1/FVC, followed by emphysema and cigarette smoking (Supplementary Figure 2). Additional associations that did not reach significance across multiple phenotypes are presented in Supplementary Table 2, including an association of the SERPINA1 z-allele variant (rs28929474) and emphysema.

Figure 1.

Figure 1.

Open in a new tab

Association of 25 COPD-related SNPs with 11 tested phenotypes. Association was assessed using linear or logistic regression controlling for age, sex, 5 principal components of genetic ancestry, and additional relevant covariates (delineated in the Methods section).

Table 2.

COPD loci associated with more than one phenotype at P < 0.002 (n=13) in linear or logistic regression. All tests for pleiotropy of SNP and associated outcomes were significant (P < 0.00038).

SNP Chr Gene Phenotypes associated at p < 0.002 (effect direction) Pleiotropy P-value
rs10429950 1 TGFB2 FEV1/FVC (−), emphysema (+) 0.0013
rs1529672 3 RARB FEV1/FVC (+), emphysema (−) 0.0007
rs2047409 4 TET2 FEV1/FVC (+), emphysema (−) 0.0009
rs6837671 4 FAM13A FEV1/FVC (+), SGRQ (−) 0.0008
rs13141641 4 HHIP FEV1/FVC (−), emphysema (+), emphysema ratio (+), SGRQ (+) 0.0001
rs2076295 6 DSP FEV1/FVC (+), emphysema (−) 0.0004
rs2070600 6 AGER FEV1/FVC (+), emphysema (−), total lung capacity (−) <0.0001
rs75200691 8 DLC1 emphysema (−), BMI (+) 0.0004
rs626750 11 MMP12 FEV1/FVC (−), SGRQ (+) 0.0004
rs754388 14 RIN3 FEV1/FVC (−), emphysema (+), emphysema ratio (+) 0.0001
rs1441358 15 THSD4 FEV1/FVC (+), emphysema ratio (−) 0.0024
rs17486278 15 CHRNA5 FEV1/FVC (+), emphysema (−), emphysema ratio (−), cigarettes per day (−) <0.0001
rs12459249 19 CYP2A6 emphysema (−), total lung capacity (−), cigarettes per day (−) 0.0001
Open in a new tab

Tests of Pleiotropy

We tested for pleiotropy in the 13 loci associated with more than one phenotype via the R package pleiotropy (Lutz et al., 2016), and all tests were significant at P < 0.0038 (0.05/13 tests) (Table 2). Many of the loci associated with multiple phenotypes were associated with both FEV1/FVC and emphysema (TGFB2, RARB, TET2, HHIP, DSP, AGER, RIN3, and CHRNA5). In all cases, the direction of effect across phenotypes was consistent with expectations (i.e. loci associated with a decrease in FEV1/FVC were also associated with an increase in emphysema). We also found loci associated with both SGRQ scores and FEV1/FVC (HHIP, FAM13A, MMP12), and loci associated with both total lung capacity and emphysema (AGER, CYP2A6). We identified 2 loci where one of the identified phenotypes was cigarette smoking (CHRNA5, CYP2A6), both of which have been previously described as smoking-related loci (Bierut, 2009; Consortium, 2010). Lastly, we found an emphysema-associated locus near DLC1 was also associated with BMI.

Mediation Analyses

To distinguish biological pleiotropy from mediated pleiotropy, we performed mediation analysis (Salinas, Wang, & DeWan, 2017). An example causal diagram used to conceptualize the COPD disease process is presented in Figure 2. All loci showed a mediated effect through the hypothesized causal pathways consistent with mediated pleiotropy, with one exception, a locus previously associated with cigarette smoking, CYP2A6 (Table 3). This locus’s effect on total lung capacity was not significantly mediated through cigarettes per day (P = 0.1). However, there was a significant mediated effect of this locus (CYP2A6) through smoking of on the related trait of emphysema (P < 0.0001). Both CYP2A6 and the other smoking-associated locus, CHRNA5, showed evidence of direct effects on COPD-related outcomes (i.e. not entirely mediated through cigarette smoking); Assuming no unmeasured confounding, these findings suggest these genetic variants also contribute to disease independent of smoking.

Figure 2.

Figure 2.

Open in a new tab

Example directed acyclic graph (DAG) used for mediation analysis. Genetic ancestry was summarized by the first 5 principal components.

Table 3.

Mediated and direct effects estimated from mediation analysis for loci associated with great than 1 phenotype. Causal models used in mediation analysis were pre-specified based on current knowledge of COPD pathophysiology. The first five PCs summarizing genetic ancestry were included as confounders of the SNP to mediator relationship. Confounders defined in Supplemental Table 1 were included as covariates of the mediator to outcome relationships. Estimates that are significant after correction for multiple comparisons (P < 0.00227) are indicated by an asterisk(*).

SNP Gene Mediator Outcome Direct effect Direct effect P Mediated effect Mediated effect P
rs10429950 TGFB2 emphysema FEV1/FVC −0.0030 0.1703 −0.0098 0.0002*
rs1529672 RARB emphysema FEV1/FVC 0.0067 0.0091 0.0106 0.0003*
rs2047409 TET2 emphysema FEV1/FVC 0.0045 0.0242 0.0087 0.0001*
rs6837671 FAM13A FEV1/FVC SGRQ −0.2840 0.3847 −0.9940 <1×10−5*
rs13141641 HHIP emphysema FEV1/FVC −0.0032 0.1100 −0.0122 <1×10−5*
rs13141641 HHIP FEV1/FVC SGRQ 0.0507 0.0130 0.0993 <1×10−5*
rs13141641 HHIP emphysema ratio FEV1/FVC −0.0087 0.0011* −0.0064 <1×10−5*
rs2076295 DSP emphysema FEV1/FVC 0.0067 0.0020* 0.0073 0.0020*
rs2070600 AGER emphysema FEV1/FVC 0.0039 0.4100 0.0312 <1×10−5*
rs2070600 AGER FEV1/FVC total lung capacity −1.7390 0.0049 −1.9900 <1×10−5*
rs75200691 DLC1 emphysema BMI 0.0057 0.0160 0.0032 0.0004*
rs626750 MMP12 FEV1/FVC SGRQ 0.6600 0.1100 1.2390 <1×10−5*
rs754388 RIN3 emphysema FEV1/FVC −0.0058 0.0220 −0.0101 0.0020*
rs754388 RIN3 emphysema ratio FEV1/FVC −0.0122 0.0028 −0.0036 0.0003*
rs1441358 THSD4 emphysema ratio FEV1/FVC 0.0064 0.0217 0.0027 0.0005*
rs17486278 CHRNA5 emphysema FEV1/FVC 0.0057 0.0044 0.0125 <1×10−5*
rs17486278 CHRNA5 cig per day FEV1/FVC 0.0168 <1×10−5* 0.0024 <1×10−5*
rs17486278 CHRNA5 cig per day emphysema −0.1404 <1×10−5* −0.0184 <1×10−5*
rs17486278 CHRNA5 cig per day emphysema ratio −0.1235 <1×10−5* −0.0156 <1×10−5*
rs17486278 CHRNA5 emphysema ratio FEV1/FVC 0.0105 <1×10−4* 0.0059 <1×10−5*
rs12459249 CYP2A6 cig per day emphysema 0.0962 0.0011* 0.0136 2×10−5*
rs12459249 CYP2A6 cig per day total lung capacity 1.4021 4×10−5* 0.0320 0.1000
Open in a new tab

Additionally, mediation analysis revealed the AGER locus had both a mediated and direct effect on total lung capacity (P < 0.0001 and P = 0.004, respectively), suggesting this SNP influences both phenotypes (emphysema and total lung capacity). As expected, the association of FAM13A and MMP12 with St. Georges Respiratory Questionnaire score was mediated though FEV1/FVC without a direct effect (P = 0.33 and P = 0.13, respectively). Lastly, the association of DLC1 with emphysema was mediated through BMI but the locus also had a direct effect on emphysema (P = 0.001).

Unsurprisingly, many of the loci (n=8) showed associations with both FEV1/FVC and emphysema, two of the most important components of COPD. Although there are strong correlations clinically between these phenotypes (Thomsen et al., 2015), it remains unclear what genetic mechanisms underlie these relationships. We found variable evidence of direct effects at these loci. The TGFB2, HHIP and AGER loci showed a significant mediated effect on FEV1/FVC through emphysema, however there was not a significant direct effect, suggesting that SNPs affect spirometric ratio only through their association with emphysema. In contrast, we found evidence for both direct and mediated effects for RARB, TET2, DSP, RIN3, and CHRNA5 on FEV1/FVC.

Parametric response mapping (PRM) phenotypes

To further assess the relationship of genotypes to the pathophysiologic processes that result in airflow limitation (emphysema and small airways disease), we incorporated a recently developed measurement, parametric response mapping (PRM). This measurement matches inspiratory and expiratory lung scans to produce measurements of both small airway disease and emphysema (Galbán et al., 2012). These more specific measurements strongly associate with COPD severity (Pompe et al., 2015, 2017) and are hypothesized to better capture components of disease pathophysiology. As such, we performed mediation analysis using these measurements as mediators on the 8 loci associated with both FEV1/FVC and emphysema. Causal models used either PRM defined emphysema (PRMemph) or PRM defined airways disease (PRMair) as the mediator and FEV1/FVC as the outcome.

In most cases, the results of this analysis agreed with the conclusions of the mediation analyses using FEV1/FVC and the standard measure of emphysema (i.e. for the TGFB2 and AGER loci, there is no direct effect of the SNP on FEV1/FVC and for the RARB, DSP, and RIN3 loci there is both a direct and mediated effect of SNP on FEV1/FVC) (Supplemental Table 3). However, for the TET2 locus, our data support the opposite conclusion (no direct effect on FEV1/FVC). Finally, for the HHIP and CHRNA5 loci conclusions about direct effects on FEV1/FVC depend on the mediators chosen (either PRMemph or PRMair). For example, all of the association of HHIP on FEV1/FVC is through PRMemph when it is included as a mediator, however this is not true when PRMair is included in the model instead. This suggests that the PRM measurements may capture different aspects of the disease than either emphysema or FEV1/FVC alone.

Sensitivity Analyses

There are several assumptions to mediation analysis. First, one must account for a SNP-by-mediator interaction on the outcome of interest (should it exist). We considered the possibility of a SNP-by-mediator interaction in all tested models by including an interaction term in the mediation analysis. However, we found no evidence of a significant interaction using this method (Supplemental Figure 3). Additionally, we used the Imai method (Imai, Keele, & Tingley, 2010) to assess how sensitive our results were to unmeasured confounding of the mediator-to-outcome relationship. We did not find any evidence that results were unduly influenced by unmeasured confounding (Supplemental Figure 4). Lastly, we used the Umediation package (Lutz et al., 2017) to test how sensitive our results were to unmeasured confounding of the SNP-mediator relationship and the SNP-outcome relationship. We found limited evidence that an unknown confounder affected our results (Supplemental Figure 5).

DISCUSSION

Large genome-wide association studies have established multiple loci associated with COPD (Hobbs et al., 2017) and related phenotypes. However, a comprehensive assessment of pleiotropy and causal modeling of the relationships between genetic variants and these phenotypes has not been adequately described. Identifying these relationships may give insights into the biological mechanisms of disease and the utility of specific phenotypes in COPD as mediators of disease outcome. Using the well-phenotyped COPDGene study population, we identified multiple additional associated phenotypes for 13 known loci, all of which showed significance in a formal test of pleiotropy. To distinguish between biological pleiotropy and mediated pleiotropy, we performed mediation analysis. Nearly all loci showed a mediated effect through the hypothesized causal pathway, consistent with mediated pleiotropy. However, many (although not all) loci also had direct associations, indicating that causal explanations using existing phenotypes (or endophenotypes) are incomplete.

Cigarette smoking is the most important environmental risk factor for COPD. While the association of the 15q25.1 locus (which includes genes coding for nicotinic acetylcholine receptors) with nicotine dependence is indisputable, evidence for its independent association with airflow obstruction is variable. Our results suggest that genotype at rs17486278 in CHRNA5 has an independent effect on FEV1/FVC that is not completely mediated through cigarette smoking (p <1×10−5). There are 2 possible explanations for this finding: 1) the smoking measurement used (average cigarettes smoked per day) does not accurately capture the pathological effects of smoking and the observed association of rs17486278 to FEV1/FVC is a result of residual confounding; or 2) independent of smoking, genotype at rs17486278 affects airflow limitation. Previous studies have presented contradictory evidence of these explanations (Obeidat et al., 2018; Siedlinski et al., 2013; Vanderweele et al., 2012; Wilk et al., 2012; Zhang et al., 2017). While there is evidence of association between CHRNA5 and lung function in never-smokers (Wilk et al., 2012), suggesting an independent effect on lung function, recent analyses have found no association of this region with lung function or lung cancer in never-smokers (Obeidat et al., 2018; Wang, Broderick, Matakidou, Eisen, & Houlston, 2011). Likewise, previous meditation analyses in this region have found both a direct effect (Siedlinski et al., 2013) and no direct effect (Vanderweele et al., 2012) of CHRNA5 genotype on airflow obstruction. It remains unclear if this locus acts on COPD in addition to the effects of smoking.

While whether or not there is a direct effect of genetic variants in the 15q25 locus is still controversial, there is some biological plausibility to the independent association of this locus. In the GTEX project, rs17486278 is a lung eQTL for both CHRNA5 and RP11–650212.2. CHRNA5 is expressed in lung tissue, airway smooth muscle, and bronchial epithelial cells (Wilk et al., 2000). Additionally, a CHRNA5 variant was found to be associated with bronchial hyper-responsiveness in children (Torjussen et al., 2012). A previous study that silenced CHRNA5 in bronchial epithelial cells found reduced expression of adhesion molecules and increased cell motility (Krais et al., 2011) that may influence the repair and remodeling processes that are important to COPD development. 15q25 also harbors several other genes, including PSMA4 (Sakornsakolpat et al., 2018) and IREB2. DeMeo et al found differential expression of IREB2 mRNA and protein in lung tissue samples from COPD cases as compared to healthy controls (DeMeo et al., 2009), and IREB2 plays a key role in iron homeostasis and COPD (Cloonan et al., 2016), suggesting it may be the underlying COPD susceptibility gene.

Our results also highlight the important relationship between lung function and imaging measurements, particularly emphysema. Lung parenchymal destruction leads to reduced lung elastic recoil and airflow limitation (Burrows B, Fletcher CM, Heard BE, Jones NL, 1966). Because the pathophysiology of emphysema is a specific disease process, emphysema might be expected to act as an endophenotype. We found variable evidence of this conclusion. A few loci appeared to be associated exclusively with emphysema and not with lung function (e.g. DLC1 and SERPINA1). By contrast, other loci had effects that were entirely mediated through measured emphysema (e.g. TGFB2, HHIP and AGER). Lastly, there were loci that showed evidence of both a mediated effect through emphysema and a direct effect of genotype on airflow limitation. This suggests that different loci could act through different biological mechanisms, and highlights the need for better phenotypes and biomarkers in COPD to dissect causal pathways.

We used mediation analysis to help elucidate the biological mechanism through which a SNP affects multiple phenotypes. This method relies on important assumptions, including that the causal pathway is correctly specified, that there is no unmeasured confounding, and that the mediator is accurately and precisely measured. We specified causal models a priori based on pathophysiological knowledge of COPD. We believe we accounted for known confounders, however it is possible that results are due to an unknown or unmeasured confounder. To assess the robustness of our results to this possibility, we performed 2 sensitivity analyses to test these assumptions (Imai, Keele, & Yamamoto, 2010; Lutz et al., 2017, Supplement). Based on these analyses, we do not believe unmeasured confounders unduly influence our results. However, strong unmeasured confounding could have significant effect, particularly on our estimates of causal mediated effects. Additionally, this manuscript focuses on the effect of a genetic locus on an outcome through a single mediator. We acknowledge the possibility of multiple mediators given that COPD is a complex disease with multiple underlying mechanisms. Specifically, at the CHRNA5 and CYP2A loci, genotype may act through both smoking and emphysema on outcome (FEV1/FVC and total lung capacity, respectively). Future work should test more complex causal models to better understand the biological mechanisms underlying COPD.

We chose the COPDGene study sample for its extensive phenotyping, collected using standardized protocols, including imaging measurements that are hypothesized to better assess disease processes. However, there are a few limitations of this dataset. Smoking behaviors (including the average number of cigarettes smoked per day measurement used in our analysis) are based on self-report. Participants may not have reported changes in smoking behaviors over their lifetime, potentially resulting in residual confounding that could bias reported results. Moreover, available measures of smoking behaviors (cigarettes smoked per day and current smoking status) may not completely capture the toxic effects of tobacco inhalation and recent work suggests objective biomarkers of smoking (e.g. exhaled carbon monoxide or cotinine) can more accurately capture smoking behavior (Obeidat et al., 2018). These measures were not available in the COPDGene study. Additionally, because the COPDGene cohort was ascertained based on case/control status, analyzing secondary quantitative phenotypes can be biased due to this ascertainment condition. Associations with these secondary phenotypes and genetic regions of interest have been identified in previous studies where secondary analyses correcting for ascertainment was applied (Cho et al., 2015b). However, these methods were not applied in our data. And lastly, the reduction in sample size as compared to previously published large GWASs (Hobbs et al., 2017) affected our ability to detect possibly significant phenotype associations and pleiotropy.

We wish to distinguish our mediation analysis from the Mendelian randomization method (Smith et al., 2005; Smith & Ebrahim, 2005), which aims to establish causality of an exposure on an outcome through instrumental variable analysis (e.g. where SNP is the instrument for the exposure of interest). While the methods have similar experimental designs and causal models, mediation analysis tests the mechanism by which an exposure is related to an outcome (direct relationship and/or indirect relationship) via the inclusion of a mediator variable. Unlike Mendelian randomization, we do not assume that all of the effect of the exposure (i.e. the SNP) is through the mediator, and we do not assume the exposure cannot be pleiotropic. In fact, the latter is exactly the hypothesis we wish to test.

We did not find substantial evidence for pleiotropy for non-respiratory related phenotypes in COPDGene. However, larger studies have found association with other phenotypes (such as pulmonary fibrosis, blood cell counts, and anthropomorphic measures). The smaller sample size as compared to previously published large GWASs (Hobbs et al., 2017) may have affected our ability to detect possibly significant phenotype associations and pleiotropy. Thus, pleiotropy and mediation by other phenotypes may require further study.

Identifying the genetic causes of COPD and its related phenotypes is essential to the understanding of disease etiology and to the identification of disease subtypes. Our work provides a systematic assessment of pleiotropy and mediation in a complex disease, one of the most common diseases and causes of death in the U.S. and worldwide, and novel insights into the biological mechanisms underlying COPD. We have identified genetic loci affecting multiple COPD-related phenotypes and used causal inference to show that 1) there is a high degree of pleiotropy in COPD phenotypes, and 2) while most of these cases appear to be results of mediated pleiotropy, in many cases, we find evidence of direct effects.

Supplementary Material

Supp info

Acknowledgments

FUNDING

Funding: This study was supported by the COPDGene project with funding by NIH U01 HL089856 and U01 HL089897 and by the COPD Foundation through contributions made by AstraZeneca, Boehringer-Ingelheim, Novartis, Pfizer, GSK, Siemens and Sunovion. The project was supported by the Parker B. Francis Research Opportunity Award (BDH,MNM) and NIH grants T32HL007427 (MMP), K01HL125858 (SML), K08 HL136928 (BDH), R01 HL124233 (PJC) and R01 HL126596 (PJC),R00 HL121087 (MNM), and R01 HL113264 (EKS, MHC).

Footnotes

CONFLICTS OF INTEREST

PJC reports person fees from GlaxoSmithKline. EKS reports grants and other support from COPD Foundation, grants and personal fees from GlaxoSmithKline, personal fees from Merck and personal fees and other expense payments from Novartis, outside the submitted work. COPDGene is supported by the COPD Foundation through contributions made by AstraZeneca, Boehringer-Ingelheim, Novartis, Pfizer, GSK, Siemens and Sunovion.

References:

  1. Bierut LJ (2009). Nicotine dependence and genetic variation in the nicotinic receptors. Drug and Alcohol Dependence. 10.1016/j.drugalcdep.2009.06.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Burrows B, Fletcher CM, Heard BE, Jones NL, W. J (1966). The emphysematous and bronchial types of chronic airways obstruction. A clinicopathological study of patients in London and Chicago. Lancet, 1(7442), 830–5. [DOI] [PubMed] [Google Scholar]
  3. Cho MH, Boutaoui N, Klanderman BJ, Sylvia JS, Ziniti JP, Hersh CP, … Silverman EK (2010). Variants in FAM13A are associated with chronic obstructive pulmonary disease. Nature Genetics, 42(3), 200–202. 10.1038/ng.535 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Cho MH, Castaldi PJ, Hersh CP, Hobbs BD, Barr RG, Tal-Singer R, … Silverman EK (2015a). A genome-wide association study of emphysema and airway quantitative imaging phenotypes. American Journal of Respiratory and Critical Care Medicine, 192(5), 559–569. 10.1164/rccm.201501-0148OC [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Cho MH, Castaldi PJ, Hersh CP, Hobbs BD, Barr RG, Tal-Singer R, … Silverman EK (2015b). A genome-wide association study of emphysema and airway quantitative imaging phenotypes. American Journal of Respiratory and Critical Care Medicine, 192(5), 559–569. 10.1164/rccm.201501-0148OC [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Cho MH, McDonald M-LN, Zhou X, Mattheisen M, Castaldi PJ, Hersh CP, … Beaty TH (2014). Risk loci for chronic obstructive pulmonary disease: a genome-wide association study and meta-analysis. The Lancet Respiratory Medicine, 2(3), 214–225. 10.1016/S2213-2600(14)70002-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Cloonan SM, Glass K, Laucho-Contreras ME, Bhashyam AR, Cervo M, Pabón MA, … Choi AMK (2016). Mitochondrial iron chelation ameliorates cigarette smoke-induced bronchitis and emphysema in mice. Nature Medicine, 22(2), 163–174. 10.1038/nm.4021 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Consortium, T. T. and G. (2010). Genome-wide meta-analyses identify multiple loci associated with smoking behavior. Nature Genetics, 42(5), 441–7. 10.1038/ng.571 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Coxson HO, Chan IHT, Mayo JR, Hlynsky J, Nakano Y, & Birmingham CL (2004). Early emphysema in patients with anorexia nervosa. American Journal of Respiratory and Critical Care Medicine, 170(7), 748–752. 10.1164/rccm.200405-651OC [DOI] [PubMed] [Google Scholar]
  10. DeMeo DL (2004). 1:Antitrypsin deficiency 2: Genetic aspects of 1-antitrypsin deficiency: phenotypes and genetic modifiers of emphysema risk. Thorax, 59(3), 259–264. 10.1136/thx.2003.006502 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. DeMeo DL, Mariani T, Bhattacharya S, Srisuma S, Lange C, Litonjua A, … Silverman EK (2009). Integration of Genomic and Genetic Approaches Implicates IREB2 as a COPD Susceptibility Gene. American Journal of Human Genetics, 85(4), 493–502. 10.1016/j.ajhg.2009.09.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Foreman MG, Wilson C, DeMeo DL, Hersh CP, Beaty TH, Cho MH, … Silverman EK (2017). Alpha-1 antitrypsin PiMZ genotype is associated with chronic obstructive pulmonary disease in two racial groups. Annals of the American Thoracic Society, 14(8), 1280–1287. 10.1513/AnnalsATS.201611-838OC [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Galbán CJ, Han MK, Boes JL, Chughtai KA, Meyer CR, Johnson TD, … Ross BD (2012). Computed tomography-based biomarker provides unique signature for diagnosis of COPD phenotypes and disease progression. Nature Medicine, 18(11), 1711–1715. 10.1038/nm.2971 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Han MK, Kazerooni EA, Lynch DA, Liu LX, Murray S, Curtis JL, … Washko GR (2011). Chronic Obstructive Pulmonary Disease Exacerbations in the COPDGene Study: Associated Radiologic Phenotypes. Radiology, 261(1), 274–282. 10.1148/radiol.11110173 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Hobbs BD, Jong K. de, Lamontagne M, Bossé Y, Shrine N, Artigas MS, … Cho MH (2017). Genetic loci associated with chronic obstructive pulmonary disease overlap with loci for lung function and pulmonary fibrosis. Nature Genetics, 49(3). 10.1038/ng.3752 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Hogg JC (2004). Pathophysiology of airflow limitation in chronic obstructive pulmonary disease. The Lancet, 364(9435), 709–721. 10.1016/S0140-6736(04)16900-6 [DOI] [PubMed] [Google Scholar]
  17. Imai K, Keele L, & Tingley D (2010). A general approach to causal mediation analysis. Psychological Methods, 15(4), 309–334. 10.1037/a0020761 [DOI] [PubMed] [Google Scholar]
  18. Imai K, Keele L, & Yamamoto T (2010). Identification, Inference and Sensitivity Analysis for Causal Mediation Effects. Statistical Science, 25(1), 51–71. 10.1214/10-STS321 [DOI] [Google Scholar]
  19. Ingebrigtsen T, Thomsen SF, Vestbo J, Van Der Sluis S, Kyvik KO, Silverman EK, … Backer V (2010). Genetic influences on chronic obstructive pulmonary disease - A twin study. Respiratory Medicine, 104(12), 1890–1895. 10.1016/j.rmed.2010.05.004 [DOI] [PubMed] [Google Scholar]
  20. Krais AM, Hautefeuille AH, Cros MP, Krutovskikh V, Tournier JM, Birembaut P, … Hainaut PL (2011). CHRNA5 as negative regulator of nicotine signaling in normal and cancer bronchial cells: Effects on motility, migration and p63 expression. Carcinogenesis, 32(9), 1388–1395. 10.1093/carcin/bgr090 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Lutz SM, Fingerlin TE, Hokanson JE, & Lange C (2016). A general approach to testing for pleiotropy with rare and common variants. Genetic Epidemiology, 41(January), 163–170. 10.1002/gepi.22011 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Lutz SM, Thwing A, Schmiege S, Kroehl M, Baker CD, Starling AP, … Ghosh D (2017). Examining the role of unmeasured confounding in mediation analysis with genetic and genomic applications. BMC Bioinformatics, 18(1). 10.1186/s12859-017-1749-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Nakano Y, Wong JC, De Jong PA, Buzatu L, Nagao T, Coxson HO, … Paré PD (2005). The prediction of small airway dimensions using computed tomography. American Journal of Respiratory and Critical Care Medicine, 171(2), 142–146. 10.1164/rccm.200407-874OC [DOI] [PubMed] [Google Scholar]
  24. Obeidat M, Zhou G, Li X, Hansel NN, Rafaels N, Mathias R, … Sin DD (2018). The genetics of smoking in individuals with chronic obstructive pulmonary disease. Respir Res, 19(1), 59 10.1186/s12931-018-0762-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Paaby AB, & Rockman MV (2013). Pleiotropy: What do you mean? Reply to Zhang and Wagner. Trends in Genetics. 10.1016/j.tig.2013.05.003 [DOI] [PubMed] [Google Scholar]
  26. Patel BD, Coxson HO, Pillai SG, Agust?? AGN, Calverley PMA, Donner CF, … Lomas DA (2008). Airway wall thickening and emphysema show independent familial aggregation in chronic obstructive pulmonary disease. American Journal of Respiratory and Critical Care Medicine, 178(5), 500–505. 10.1164/rccm.200801-059OC [DOI] [PubMed] [Google Scholar]
  27. Pearl J (2001). Direct and Indirect Effects. Proceedings of the Seventeenth Conference on Uncertainy in Arifical Intelligence, 411–420. [Google Scholar]
  28. Pompe E, Galbán CJ, Ross BD, Koenderman L, ten Hacken NH, Postma DS, … Mohamed Hoesein FA (2017). Parametric response mapping on chest computed tomography associates with clinical and functional parameters in chronic obstructive pulmonary disease. Respiratory Medicine, 123, 48–55. 10.1016/j.rmed.2016.11.021 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Pompe E, Van Rikxoort EM, Schmidt M, R??haak J, Gallardo Estrella L, Vliegenthart R, … Mohamed Hoesein FAA (2015). Parametric response mapping adds value to current computed tomography biomarkers in diagnosing chronic obstructive pulmonary disease. American Journal of Respiratory and Critical Care Medicine. 10.1164/rccm.201411-2105LE [DOI] [PubMed] [Google Scholar]
  30. Price AL, Patterson NJ, Plenge RM, Weinblatt ME, & Shadick NA. (2006). Principal components analysis corrects for stratification in genome-wide association studies. Nat. Genet, 38(8), 904–909. 10.1038/ng1847 [DOI] [PubMed] [Google Scholar]
  31. Price AL, Zaitlen NA, Reich D, & Patterson N (2010). New approaches to population stratification in genome-wide association studies. Nature Reviews. Genetics, 11(7), 459–63. 10.1038/nrg2813 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Regan EA, Hokanson JE, Murphy JR, Lynch DA, Beaty TH, Curran-everett D, … Investigators C (2011). Genetic epidemiology of COPD (COPDgene) study design. Epidemiology, 7(1), 1–10. 10.3109/15412550903499522.Genetic [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Sakornsakolpat P, Morrow JD, Castaldi PJ, Hersh CP, Bossé Y, Silverman EK, … Cho MH (2018). Integrative genomics identifies new genes associated with severe COPD and emphysema. Respiratory Research. 10.1186/s12931-018-0744-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Salinas YD, Wang Z, & DeWan AT (2017). Statistical Analysis of Multiple Phenotypes in Genetic Epidemiological Studies:From Cross-Phenotype Associations to Pleiotropy. American Journal of Epidemiology, (March). 10.1093/aje/kwx296 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Siedlinski M, Tingley D, Lipman PJ, Cho MH, Litonjua AA, Sparrow D, … Silverman EK (2013). Dissecting direct and indirect genetic effects on chronic obstructive pulmonary disease (COPD) susceptibility. Human Genetics, 132(4), 431–441. 10.1007/s00439-012-1262-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Siva N (2008). 1000 Genomes project. Nature Biotechnology, 26(3), 256–256. 10.1038/nbt0308-256b [DOI] [PubMed] [Google Scholar]
  37. Smith GD, & Ebrahim S (2005). What can mendelian randomisation tell us about modifiable behavioural and environmental exposures? BMJ : British Medical Journal, 330(7499), 1076–1079. 10.1136/bmj.330.7499.1076 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Smith GD, Ebrahim S, Lewis S, Hansell AL, Palmer LJ, & Burton PR (2005). Genetic epidemiology and public health: Hope, hype, and future prospects. Lancet, 366(9495), 1484–1498. 10.1016/S0140-6736(05)67601-5 [DOI] [PubMed] [Google Scholar]
  39. Solovieff N, Cotsapas C, Lee PH, Purcell SM, & Smoller JW (2013). Pleiotropy in complex traits: challenges and strategies. Nature Reviews Genetics, 14(7), 483–495. 10.1038/nrg3461 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Soriano JB, Abajobir AA, Abate KH, Abera SF, Agrawal A, Ahmed MB, … Vos T (2017). Global, regional, and national deaths, prevalence, disability-adjusted life years, and years lived with disability for chronic obstructive pulmonary disease and asthma, 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. The Lancet Respiratory Medicine, 5(9), 691–706. 10.1016/S2213-2600(17)30293-X [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Thomsen L, Shaker S, Dirksen A, Pedersen J, Tal-Singer R, Bakke P, & Vestbo J (2015). Correlation Between Emphysema and Lung Function in Healthy Smokers and Smokers With COPD. Chronic Obstructive Pulmonary Diseases: Journal of the COPD Foundation, 2(3), 204–213. 10.15326/jcopdf.2.3.2014.0154 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Thorgeirsson TE, Gudbjartsson DF, Surakka I, Vink JM, Amin N, Geller F, … Stefansson K (2010). Sequence variants at CHRNB3-CHRNA6 and CYP2A6 affect smoking behavior. Nature Genetics, 42(5), 448–453. 10.1038/ng.573 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Torjussen T, Lodrup Carlsen K, Munthe-Kaas M, Mowinckel P, Carlsen K, Helms P, … Zhu G (2012). Alpha-nicotinic acetylcholine receptor and tobacco smoke exposure: Effects on bronchial hyperresponsiveness in children. Pediatric Allergy and Immunology. Retrieved from http://ovidsp.ovid.com/ovidweb.cgi?T=JS&PAGE=reference&D=emed10&NEWS=N&AN=2012064775 [DOI] [PubMed] [Google Scholar]
  44. Vanderweele TJ, Asomaning K, Tchetgen Tchetgen EJ, Han Y, Spitz MR, Shete S, … Lin X (2012). Genetic variants on 15q25.1, smoking, and lung cancer: An assessment of mediation and interaction. American Journal of Epidemiology, 175(10), 1013–1020. 10.1093/aje/kwr467 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Vanderweele TJ, & Vansteelandt S (2009). Conceptual issues concerning mediation, interventions and composition. Statistics and Its Interface, 2(4), 457–468. 10.4310/SII.2009.v2.n4.a7 [DOI] [Google Scholar]
  46. Wang Y, Broderick P, Matakidou A, Eisen T, & Houlston RS (2011). Chromosome 15q25 (CHRNA3-CHRNA5) variation impacts indirectly on lung cancer risk. PLoS ONE, 6(4). 10.1371/journal.pone.0019085 [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Wilk JB, Djousse L, Arnett DK, Rich SS, Province M. a., Hunt SC, … Myers RH (2000). Evidence for major genes influencing pulmonary function in the NHLBI Family Heart Study. Genetic Epidemiology, 19(July 1999), 81–94. [DOI] [PubMed] [Google Scholar]
  48. Wilk JB, Shrine NRG, Loehr LR, Zhao JH, Manichaikul A, Lopez LM, … Stricker BH (2012). Genome-wide association studies identify CHRNA5/3 and HTR4 in the development of airflow obstruction. American Journal of Respiratory and Critical Care Medicine, 186(7), 622–632. 10.1164/rccm.201202-0366OC [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Zhang J, Peng S, Cheng H, Nomura Y, Di Narzo AF, & Hao K (2017). Genetic Pleiotropy between Nicotine Dependence and Respiratory Outcomes. Scientific Reports, 7(1). 10.1038/s41598-017-16964-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Zhou JJ, Cho MH, Lange C, Lutz S, Silverman EK, & Laird NM (2015). Integrating Multiple Correlated Phenotypes for Genetic Association Analysis by Maximizing Heritability. Human Heredity, 79(2), 93–104. 10.1159/000381641 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supp info
NIHMS1008736-supplement-Supp_info.docx (877.8KB, docx)

RESOURCES