Examination of the Involvement of Cholinergic-Associated Genes in Nicotine Behaviors in European and African Americans

Abstract

Introduction

Cigarette smoking is a physiologically harmful habit. Nicotinic acetylcholine receptors (nAChRs) are bound by nicotine and upregulated in response to chronic exposure to nicotine. It is known that upregulation of these receptors is not due to a change in mRNA of these genes, however, more precise details on the process are still uncertain, with several plausible hypotheses describing how nAChRs are upregulated. We have manually curated a set of genes believed to play a role in nicotine-induced nAChR upregulation. Here, we test the hypothesis that these genes are associated with and contribute risk for nicotine dependence (ND) and the number of cigarettes smoked per day (CPD).

Methods

Studies with genotypic data on European and African Americans (EAs and AAs, respectively) were collected and a gene-based test was run to test for an association between each gene and ND and CPD.

Results

Although several novel genes were associated with CPD and ND at P < 0.05 in EAs and AAs, these associations did not survive correction for multiple testing. Previous associations between CHRNA3, CHRNA5, CHRNB4 and CPD in EAs were replicated.

Conclusions

Our hypothesis-driven approach avoided many of the limitations inherent in pathway analyses and provided nominal evidence for association between cholinergic-related genes and nicotine behaviors.

Implications

We evaluated the evidence for association between a manually curated set of genes and nicotine behaviors in European and African Americans. Although no genes were associated after multiple testing correction, this study has several strengths: by manually curating a set of genes we circumvented the limitations inherent in many pathway analyses and tested several genes that had not yet been examined in a human genetic study; gene-based tests are a useful way to test for association with a set of genes; and these genes were collected based on literature review and conversations with experts, highlighting the importance of scientific collaboration.

Introduction

Cigarette smoking is a personally harmful and societally detrimental habit. Despite substantial progress since the first Surgeon General’s report was released 50 years ago, smoking is still the largest cause of preventable disease and death in the United States. An estimated half a million Americans die prematurely from smoking each year, and more than 16 million Americans suffer from smoking-related diseases. The economic costs from smoking and exposure to tobacco smoke are estimated at $300 billion annually, with productivity losses of $150 billion per year.¹

It is widely accepted that nicotine is the major addictive component in tobacco smoke^2,3 and exerts its effect by binding to nicotinic acetylcholine receptors (nAChRs, encoded by the CHRN genes) in the peripheral and central nervous systems.⁴ Accordingly, numerous studies have identified polymorphisms in several CHRN genes associated with nicotine dependence (ND) and cigarettes per day (CPD). The most well-replicated of these associations lies within a cluster of three CHRN genes on chromosome 15q25 (CHRNA3/A5/B4),⁵ although genes for other subunits have been associated with these phenotypes as well. Most notably, CHRNB3/A6 on chromosome 8p11 has been consistently associated with CPD⁶ and ND.^7–13 Other CHRN genes associated with CPD and ND include CHRND/G, CHRNB1, CHRNA10, CHRNA4, and CHRNB2.^{10,11,14–19}

Twin studies have estimated that ND and smoking quantity are roughly 56%–72% and 51%–61% heritable in men and women, respectively,^20–24 yet associations between the CHRN genes and ND and CPD account for a very small proportion of the variance in smoking (roughly 1 CPD).^5,11 Although this is small percentage, 1 additional CPD over a person’s lifetime will accumulate, often to greater than 22000 additional cigarettes in one’s lifetime in the case of the Genetic Epidemiology of Chronic Obstructive Pulmonary Disease²⁵ study (COPDGene, described in the Samples section). In addition, studies have shown an association between higher levels of adolescent ND and heavier smoking trajectory patterns.^26,27 Based on work providing evidence that common single nucleotide polymorphisms (SNPs) explain a large proportion of the heritability in height, Crohn’s disease, bipolar disorder, and type I diabetes,^28,29 we hypothesize that common SNPs also explain a considerable proportion of the variation in smoking quantity and ND. Although it is highly likely rare variants play an equally important role in these behaviors, the purpose of this study was to examine common variants from available datasets. The goal of this study was to utilize set-based approaches to test for association between SNPs within selected genes and nicotine behaviors. We applied SNP-set based methods as a way to increase power to detect associations, and organize associations in a biologically meaningful way.^30,31 Associations can be more easily detected by grouping SNPs into sets, and insights into their combined effects on biological function are more apparent.

Upon chronic exposure to nicotine, nAChRs undergo an upregulation of receptor number^32–34 that is independent of upregulation of CHRN mRNA.³⁵ Although there are many theories describing how this upregulation occurs, including increased receptor trafficking,³⁶ decreased subunit degradation,^37,38 increased nAChR subunit maturation and folding,^39–41 pharmacological chaperoning by nicotine,⁴² and increased translation and second messenger signaling,⁴³ researchers have yet to test several of the proteins suggested to impact upregulation in concert. Studies have demonstrated that nicotine-induced nAChR upregulation leads to nicotine sensitization in brain regions known to play a role in addiction and reward.⁴⁴ Furthermore, nAChR upregulation has been well-documented in rodents exposed to nicotine during prenatal development^45–51 and the long term and widespread consequences of prenatal nicotine exposure have been proposed to play a role in various neurobehavioral and physiological disorders.⁵² We posit that gene products are involved in nAChR upregulation through their interaction with nAChRs, and thus play a role in the development of ND.

We have assembled a list of genes that encode proteins known to interact with nAChRs or are known to play a role in their downstream signaling, by curating the literature, and conversing with experts (listed in the acknowledgements) in the field (Figure 1). The goal of this study was to assess the evidence for association with each gene and CPD and ND in African and European Americans (AAs and EAs, respectively). These analyses were carried out in several studies in which genome-wide genotyping was available and results were later combined as a meta-analysis. Although many studies use smoking quantity as a proxy for ND, research has demonstrated unique genetic effects on CPD, suggesting that a smoking quantity measure may not serve as a simple proxy for the genetic influences on ND.⁵⁴ Thus, while these phenotypes may be correlated, they represent different aspects of smoking behavior and exposure. Based on these findings, both CPD and ND were used as phenotypes in the present study.

Figure 1.

Cellular processes proposed to play a role in nAChR upregulation; the number of genes from our curated list of 96 that play a role in each process is shown in parentheses (several genes were proposed to be involved in more than one cellular process). ER = endoplasmic reticulum. Modified from Melroy-Greif et al.⁵³

Methods

Samples

Four independent samples with genome-wide genotype data and smoking phenotypes were included in the study (Table 1): (1) the Study of Addiction: Genetics and Environment (SAGE),⁵⁵ (2) COPDGene,²⁵ (3) the Multi-Ethnic Study of Atherosclerosis (MESA),⁵⁶ and (4) an in-house sample comprised of unrelated Caucasian subjects from the Colorado Center on Antisocial Drug Dependence and the Genetics of Antisocial Drug Dependence (CADD-GADD).⁵⁷ The SAGE sample is comprised of three substudies recruited via studies on cocaine, alcohol, or ND.⁵⁵ The COPDGene study consists of smokers who have smoked at least 10 pack-years.²⁵ MESA is a population-based study designed to investigate the characteristics of subclinical cardiovascular disease.⁵⁶ Finally, the CADD-GADD sample is a mixture of various samples including probands targeted for drug use, their families and matched control families, as well as community twin samples. Additional details about the CADD-GADD sample recruitment and assessments are provided in Derringer et al.⁵⁷ Study-level principal components (PCs) were supplied along with the SAGE, COPDGene, and CADD-GADD genotype data. PCs were computed in the MESA sample using shellfish (http://www.stats.ox.ac.uk/~davison/software/shellfish/shellfish.php). All genotype data were imputed to Phase 1 of the 1000 Genomes dataset⁵⁸ and cleaned using standard quality control procedures: SNPs with low imputation accuracy (<0.9), low minor allele frequency (MAF) (<0.01), and out of Hardy–Weinberg equilibrium (HWE, P < .001) were excluded.

Table 1.

Study and Sample Characteristics

	Study	N CPD	Agea	Sex (% male)	CPDa	N FTND	Agea	Sex (% male)	FTNDa
EA	COPDGene	6670	62.09±8.84	52.37	25.84±11.44	2568	57.50±7.86	53.35	4.83±2.42
	SAGE	1255	35.74±6.92	37.83	25.90±19.94	1673	37.53±8.65	43.93	2.99±3.29
	CADD-GADD	588	23.44±4.40	73.98	17.96±11.54	NA	NA	NA	NA
	MESA	628	63.03±9.95	56.53	20.59±21.66	NA	NA	NA	NA
AA	COPDGene	3300	54.68±7.21	55.94	21.30±10.41	2567	53.34±6.04	57.78	4.93±2.34
	SAGE	594	39.30±6.81	47.47	23.64±17.39	779	40.03±7.13	49.81	3.92±2.87
	MESA	53	63.64±9.91	49.06	12.77±11.10	NA	NA	NA	NA

NA = not applicable.

^aMean ± standard deviation.

We analyzed two phenotypes when available from each study: quantitative ND symptoms as assessed by the Fagerström Test for Nicotine Dependence (FTND),⁵⁹ and CPD, binned as follows: 0–10, 11–20, 21–30, 31–40, and 41 or more CPD. CPD was only assessed in two of the substudies from SAGE whereas FTND was collected in all substudies. Likewise, FTND was only assessed in current smokers in the COPDGene sample. FTND was not available in the MESA or CADD-GADD samples. Only subjects who had smoked 100 cigarettes or more in their lifetime were included in the analysis.

Analyses

SNPs were annotated to our genes of interest using the hg19 build. SNPs 20kb up- and down-stream of each gene were included, given reports that the majority of genetic variants that influence expression are located within 20kb of a gene.⁶⁰ Only SNPs common across all datasets were evaluated. Among the EAs, this was limited to imputed SNPs from the SAGE, COPDGene, MESA, and CADD-GADD samples. Similarly, for the AA analyses, only SNPs common across the AA SAGE, MESA, and COPDGene samples were included. Age, sex, and the first five PCs from each study were included as covariates in each analysis. Additionally, substudy was used as a covariate in the SAGE analysis.

Joint Association of Genetic Variants

JAG⁶¹ (available at http://ctglab.nl/software/jag/) uses raw data and invokes PLINK (http://pngu.mgh.harvard.edu/~purcell/plink/index.shtml)⁶² to run a genome-wide association on each SNP included in the gene. A multivariate SNP test statistic is calculated by summing the −log₁₀ of each SNP P-value and the empirical P-value for each gene is calculated by summing the −log₁₀(P-value) for each permutation of the phenotype. The empirical P-value (P_emp) represents the number of times the Σ−log₁₀(P-value) exceeds or equals the −log₁₀(P-value) from the genome-wide association. The P_emp based on the Σ−log₁₀(P-value) test statistic tests the hypothesis that, given the linkage disequilibrium (LD) structure of the included SNPs, the multivariate pattern of P-values of all SNPs in a gene is significantly different than what is expected under the null hypothesis of no association. Ten thousand permutations were performed for each test. This analysis was applied to our data in order to test the association between individual genes and CPD and FTND.

Meta-analysis

The P_emp for each gene was combined across studies using the weighted Z-score method.⁶³ False discovery rate (FDR) corrected P-values were generated for each meta-analyzed P-value using the stats package in R.⁶⁴

Replication

Summary P-values for CPD were used from the Tobacco and Genetics (TAG) Consortium⁶⁵ as a replication. These data are freely available at http://www.med.unc.edu/pgc/downloads. MAGMA,⁶⁶ a tool that can perform a gene-based test on summary SNP P-values, was used to test for association between genes trending toward association with CPD in EAs after meta-analysis. Briefly, SNPs in the 1000 Genomes CEU data were annotated to genes using the hg19 build, including SNPs 20kb up- and down-stream of each gene. A gene-based test was run on 17 genes: CHRNA3, CHRNA5, CHRNB4, NCAM1, CHRNE, DNAJA3, IQGAP1, MAPRE1, UNC50, CHRNB1, EPHA4, NRXN1, LRP4, CRELD2, CAMK2A, ITGA7, and APC. The 1000 Genomes CEU data was used as a reference panel for LD as evaluation of the gene test-statistic requires LD estimation between SNPs in the gene. Similar to the JAG analysis, MAGMA uses a SNP-wise model in which the SNP P-values are converted by taking the −log₁₀ of the P-value and combined by computing their sum to obtain a gene-based P-value.

Results

Study and sample characteristics for the COPDGene, SAGE, MESA, and CADD-GADD samples are given in Table 1. Subjects in the COPDGene and MESA samples were the oldest and the CADD-GADD the youngest. EAs in the COPDGene and SAGE had similar smoking patterns, as did AAs in the COPDGene and SAGE samples. The AAs had somewhat lower CPD and higher FTND than their EA counterparts. CPD and FTND were correlated at 0.74 and 0.58 in the SAGE EA and AA samples, respectively. CPD and FTND were less correlated in the COPDGene EA and AA samples (0.37 and 0.35, respectively), likely due to FTND assessment in current smokers only.

Of the 112 total genes in our list (96 genes collected from the literature and 16 nAChR subunit genes), 100 genes had SNPs annotated and were used in the analysis in AAs, and 101 genes had SNPs annotated and were used in the EA analysis.

CPD and ND in EAs

Thirty-three thousand four hundred eighty-three SNPs from our gene set were tested for association with CPD in EAs from the COPDGene, SAGE, MESA, and CADD-GADD samples. The gene P-values were combined using the weighted Z-score method. Although 17 genes were trending at P < .1 (and used in the subsequent replication analysis), only CHRNA3, CHRNA5, and CHRNB4 were associated with CPD with controlling FDR at P < .05 (Table 2).

Table 2.

Results From the JAG Test in EAs

Gene	No. of SNPs	No. of effective SNPsa	COPDGene Pemp	SAGE Pemp	MESA Pemp	CADD-GADD Pemp	Combined Pemp	FDR corrected P
CPD
CHRNA3	102	2	.0001	.128	.209	.434	.0001	.013
CHRNA5	29	1	.0001	.304	.348	.620	.0001	.013
CHRNB4	17	3b	.0001	.639	.742	.212	.0001	.013
NCAM1	402	5c	.006	.154	.721	.337	.004	.193
CHRNE	79	3	.004	.182	.924	.800	.005	.193
DNAJA3	111	2	.004	.527	.900	.472	.007	.258
IQGAP1	246	2	.010	.898	.268	.813	.022	.556
MAPRE1	148	5	.015	.829	.784	.196	.025	.601
UNC50	53	1	.043	.181	.514	.249	.029	.644
CHRNB1	63	3	.038	.216	.878	.531	.039	.729
EPHA4	212	5	.133	.073	.238	.105	.064	.782
NRXN1	2433	22d	.022	.992	.516	.493	.064	.782
LRP4	106	3e	.089	.178	.599	.441	.070	.782
CRELD2	127	1f	.097	.009	.638	.991	.071	.782
CAMK2A	89	5	.178	.059	.034	.174	.075	.782
ITGA7	41	1	.075	.451	.725	.304	.078	.782
APC	231	2	.061	.245	.876	.920	.079	.782
FTND
CHRNA3	102	2	.003	.028	NA	NA	.00042	.042
CHRNA5	29	1	.012	.060	NA	NA	.003	.193
CHRNB4	17	2	.012	.080	NA	NA	.004	.193
CHRNA9	239	4	.009	.221	NA	NA	.008	.280
DNAJA3	111	2	.008	.338	NA	NA	.012	.381
CHRNE	79	3	.218	.004	NA	NA	.017	.482
EPHA4	212	5	.116	.112	NA	NA	.048	.782
CAMK2G	61	2	.008	.794	NA	NA	.056	.782
SVEP1	598	13	.263	.035	NA	NA	.064	.782
CAMK2A	89	5	.368	.013	NA	NA	.067	.782
ERBB2	64	1	.049	.458	NA	NA	.075	.782
RAPSN	65	2	.253	.056	NA	NA	.077	.782
CRELD2	127	1	.237	.073	NA	NA	.081	.782
LRP4	106	3g	.106	.269	NA	NA	.084	.782

The P-value is shown for each study as well as the combined P-value for genes associated with CPD or FTND at P < .1.

^aThe number of signals tested in the gene in each study (unless otherwise noted) as calculated by JAG.

^bOnly two signals calculated in the COPDGene sample.

^cSix signals calculated in the COPDGene sample.

^dTwenty-three signals calculated in the CADD-GADD and MESA samples.

^eTwo signals calculated in the SAGE sample.

^fTwo signals calculated in the MESA sample.

^gTwo signals calculated in SAGE.

Thirty-three thousand four hundred eighty-three SNPs were tested for association with FTND in EAs from the COPDGene and SAGE samples. The gene P-values were combined using the weighted Z-score method. Similar to the results with CPD, while several genes were associated at P < .05, only the association with CHRNA3 survived correction for multiple testing at FDR P < .05 (Table 2).

CPD and ND in AAs

Thirty-eight thousand seven hundred eighty-seven SNPs from our set of genes were tested for association with CPD in AAs from the COPDGene, SAGE, and MESA samples and the gene P-values combined using the weighted Z-score method. Five genes were associated with CPD at P < .1 but none of these associations survived correction for multiple testing at FDR P < .05 (Table 3).

Table 3.

Results From the JAG Test in AAs

Gene	No. of SNPs	No. of effective SNPsa	COPDGene Pemp	SAGE Pemp	MESA Pemp	Combined Pemp	FDR corrected P
CPD
CHRNA9	247	12b	.005	.321	.424	.004	.193
PICK1	29	4c	.067	.082	.043	.040	.729
NTRK2	443	19d	.031	.824	.667	.048	.782
DLG4	7	2	.080	.458	.841	.083	.782
CAMK2B	50	3e	.092	.504	.877	.098	.799
FTND
IQGAP1	400	7	.020	.016	NA	.005	.193
PDIA3	120	5	.011	.701	NA	.021	.556
UNC50	56	2	.019	.649	NA	.031	.653
UBXN2A	271	5	.012	.918	NA	.040	.729
EPHA4	221	10	.028	.791	NA	.055	.782
NLGN1	1668	36	.062	.493	NA	.070	.782
NCAM1	427	10f	.106	.170	NA	.071	.782
FYN	635	15	.085	.406	NA	.084	.782
PPP3CA	408	12	.068	.607	NA	.089	.799
LAMB1	187	10	.156	.111	NA	.093	.799
CACNA2D1	1325	40g	.077	.567	NA	.094	.799

The P-value for each gene is shown for each study as well as the combined P-value for genes associated with CPD or FTND at P < .1.

^aThe number of signals tested in the gene in each study (unless otherwise noted) as calculated by JAG.

^bEleven signals in the MESA data.

^cThree signals in the MESA data.

^dEighteen signals in the MESA data.

^eTwo signals in the MESA data.

^fNine signals in the SAGE data.

^gThirty-nine signals in the SAGE data.

The same set of SNPs was tested for association with FTND and the gene P-values combined in AAs from the SAGE and COPDGene samples. More genes were nominally associated with FTND at P < .1 than CPD, but none survived correction for multiple testing at FDR P < .05 (Table 3).

Replication

Gene-based tests on summary P-values for CPD from the TAG data were run on 17 genes (CHRNA3, CHRNA5, CHRNB4, NCAM1, CHRNE, DNAJA3, IQGAP1, MAPRE1, UNC50, CHRNB1, EPHA4, NRXN1, LRP4, CRELD2, CAMK2A, ITGA7, and APC). No gene was trending toward association at P < .1, with the exception of CHRNA3, CHRNA5, and CHRNB4, which were highly associated at P < 5E⁻¹⁴ (results not shown).

Discussion

Given the highly replicated associations with CHRN genes and smoking behaviors, we examined 96 other genes, known to interact with nAChRs, to test for association in four existing samples for which genome-wide data and phenotypic measures were available. As expected, the CHRN gene cluster on chromosome 15q25 comprised of CHRNA3, CHRNB4, and CHRNA5 showed the strongest associations with nicotine behaviors in EAs in this study, replicating previous associations with these genes and CPD^6,65,67–76 and ND.^{7,10,13,18,77–87} Although there was some overlap between the genes nominally associated with CPD and FTND (CHRNE, DNAJA3, CRELD2, CAMK2A, and EPHA4), no genes were associated after correction for multiple testing at FDR P < .05. Similarly, while primarily driven by the largest sample, COPDGene, two genes were trending toward association in more than one study: CRELD2 and CAMK2A.

The TAG results mirror our own study in that only the gene-wise P-values for CHRNA3/A5/B4 were highly significant. Although no other genes survived correction for multiple testing in our study, we would hypothesize that some of the genes would have been trending toward association with CPD in the TAG data. However, there are some notable differences between the original and replication analyses. First, we limited the analyses to SNPs annotated on all datasets used in our study, while additional signals were likely used in the replication sample, and this may have washed out associated SNPs. Second, there likely exist differences in sample ascertainment and phenotypic variables between the four samples used in our meta-analysis and those in TAG.

Unlike in EAs, no genes, particularly CHRNA3/A5/B4, were associated with CPD or FTND after controlling FDR at P < .05 in AAs. However, two genes showed consistent results across studies and bear future investigation: PICK1 and IQGAP1. None of the cluster genes were associated with CPD in AAs from the SAGE, MESA, or COPDGene samples. This result is perhaps not surprising given previous results with CPD in AAs in the COPDGene sample.⁷⁶ However, a previous study using over 32000 AAs found an association between CHRNA3/A5/B4 with CPD. In that study, only one SNP, residing in the 5′-distal enhancer region of CHRNA5, exceeded genome-wide significance.⁸⁸ It is possible that since our analyses sum individual SNP statistics to get an overall gene P-value this SNP was overshadowed by SNPs of lower significance. In our study, CHRNA5 was nominally associated with FTND in the SAGE sample using an alternative set-based test implemented in PLINK⁶² at P = .094, but this association was not replicated in the COPDGene sample. Although this suggests heterogeneity between the two samples, it aligns with what has been previously seen in the literature. Saccone et al⁸² reported an association between SNPs in CHRNA5 and ND in AAs alone in COGEND, yet no association between CHRNA3, CHRNB4, or CHRNA5 with ND was seen in the COPDGene sample.⁷⁶

Although not significant after correction for multiple testing, several genes previously associated with substance use phenotypes in the literature were nominally associated with nicotine behaviors in the present study and will be discussed briefly here in an attempt to guide future replication efforts. EPHA4, nominally associated with CPD and FTND in EAs, has been associated with opioid dependence in subjects with a history of cocaine dependence.⁸⁹CHRNA9 was nominally associated with FTND in EAs and CPD in AAs and has been previously associated with ND in female Israeli students,⁹⁰ as well as with lung cancer in Caucasians.⁹¹ Recently, rare and common variation in CHRNA9 has been associated with smoking status in EAs and AAs from the Mid-South Tobacco Case-Control study.⁹²

It is interesting there was very little overlap in the genes associated with CPD and FTND between EAs and AAs. These findings might indicate underlying genetic differences in the development of smoking behaviors in EAs and AAs. AAs tend to start smoking later in life^93–96 and have a lower lifetime prevalence of ND compared to EAs.⁹⁷ Additionally, AAs report higher cravings and more pleasurable sensations after smoking,^96,98 lower rates of regular smoking,^{94–97,99–103} higher nicotine intake per cigarette,¹⁰⁴ and slower metabolism rates of nicotine^103–107 compared to EAs. Finally, EAs have higher smoking cessation rates than AAs.⁹⁹ Together with our data, this suggests differences in smoking patterns between EAs and AAs may be partially genetically driven.

Although there was significant overlap in the genes implicated in ND and CPD in EAs, our results show little overlap in the genes associated with CPD and FTND in AAs. Given previous studies demonstrating both unique and common genetic influences on CPD and ND,⁵⁴ we expected to see some overlap between the two phenotypes. This discrepancy in AAs may be due to the fact that FTND was only measured in current smokers in the COPDGene sample and our results could reflect differences in smoking cessation as well. The AA sample was also considerably smaller than the EA sample; thus these analyses had lower power to detect genetic effects and a higher likelihood of false negatives.

While not without limitations, several strengths of the present study help mitigate these limitations. JAG has higher power than several other pathway algorithms. Specifically, using 1694 cases and 2917 controls from the Wellcome Trust Case Control Consortium the gene-based test in JAG had better power than other algorithms to detect known genes associated with Crohn’s Disease.⁶¹ SNP-set based approaches can increase power to detect genetic loci with individually small effects by consolidating SNP associations and can help to prioritize associations based on biological relevance.^30,31 A meta-analysis approach was used as opposed to a mega-analysis so associations with genes across studies could be compared, although differences in age and study ascertainment may have clouded these comparisons. By curating a unique list of genes to test, we avoided many potential limitations when examining already curated gene sets from the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO); for example, when examining gene sets from KEGG or GO, genes that have been well studied are more thoroughly annotated. As of 2011, only about 5000 human genes had been annotated to KEGG pathways.³¹ However, despite being able to curate a unique set of genes, this study was limited to testing SNPs both annotated on the hg19 build and imputed within each sample. In addition, trans effects were ignored by necessity. Similarly, while both JAG and MAGMA take LD within a gene into account when computing the test statistic,^61,66 LD between genes was not taken into account. Finally, gene-based tests do not provide an effect size of the gene so the weighted Z-score method was used to meta-analyze the results. The direction of effect of each individual SNP is thus ignored and the resulting P-value differs from that of a traditional meta-analysis where small effects of opposite direction cancel each other out. To summarize, the direction of effect for each signal in an associated gene would be unknown as well as which signals are driving the association.

In conclusion, we tested several genes believed to be involved in nAChR upregulation with FTND and CPD and performed a meta-analysis over four independent samples. The list of genes tested is by no means an exhaustive list, and simply served to test the hypothesis that SNPs in the identified genes are associated with nicotine behaviors by altering nAChR function and/or expression through protein–protein interactions. Although no genes were associated after correction for multiple testing, except for the reliably replicated CHRNA3/A5/B4 association, it is possible these genes may nevertheless be affecting nAChR upregulation in a manner not directly driven by SNPs; for example, increased transcription factor binding, or enhancer effects. Future studies could assess trans effects in these genes, perform functional bioinformatics, or examine gene-by-gene interactions among this list. This work represents collaboration between neuroscience experts and statistical geneticists and although it did not bear fruit with respect to the current study, future collaborations will continue to be important to identify and characterize novel genetic associations.

Funding

This work was supported by grants from the National Institute on Alcohol Abuse and Alcoholism (R01 AA017889 and T32 AA013525); the National Institute on Drug Abuse (T32 DA017637); and the National Heart, Lung, and Blood Institute of the National Institutes of Health (NIH) (K01 HL125858). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. The authors have no conflicts of interest to report. Funding sources for the datasets used in this study are listed as follows. CADD-GADD: Supported by NIH grants from the National Institute on Drug Abuse (DA011015 and DA012845). COPDGene: Supported by NIH grants (U01HL089897 and U01HL089856) from the National Heart, Lung, and Blood Institute. Use of this data was approved by the COPDGene Ancillary Studies and Publications Committee. MESA: Supported by contracts N01-HC-95159, N01-HC-95160, N01-HC-95161, N01-HC-95162, N01-HC-95163, N01-HC-95164, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168 and N01-HC-95169, from the National Heart, Lung, and Blood Institute, and by grants (UL1-TR-000040, and UL1-TR-001079) from the National Center for Research Resources. These data were obtained through dbGaP. SAGE: Supported by NIH grants from the National Institute on Alcohol abuse and Alcoholism (AA008401); the National Institute on Drug Abuse (DA013423); and the National Cancer Institute (CA089392). These data were obtained through dbGaP.

Declaration of Interests

None declared.