The Contribution of Rare and Common Variants in 30 Genes to Risk Nicotine Dependence

Abstract

Genetic and functional studies have revealed that both common and rare variants of several nicotinic acetylcholine receptor (nAChR) subunits are associated with nicotine dependence (ND). In this study, we identified variants in 30 candidate genes including nicotinic receptors in 200 sib pairs selected from the Mid-South Tobacco Family (MSTF) population with equal numbers of African Americans (AAs) and European Americans (EAs). We selected 135 of the rare and common variants and genotyped them in the Mid-South Tobacco Case-Control (MSTCC) population, which consists of 3088 AAs and 1430 EAs. None of the genotyped common variants showed significant association with smoking status (smokers vs. non-smokers), Fagerström Test for Nicotine Dependence (FTND) scores, or indexed cigarettes per day (CPD) after Bonferroni correction. Rare variants in NRXN1, CHRNA9, CHRNA2, NTRK2, GABBR2, GRIN3A, DNM1, NRXN2, NRXN3, and ARRB2 were significantly associated with smoking status in the MSTCC AA sample, with Weighted Sum Statistic (WSS) P values ranging from 2.42 × 10⁻³ to 1.31 × 10⁻⁴ after 10⁶ phenotype rearrangements. We also observed a significant excess of rare nonsynonymous variants exclusive to EA smokers in NRXN1, CHRNA9, TAS2R38, GRIN3A, DBH, ANKK1/DRD2, NRXN3, and CDH13 with WSS P values between 3.5 × 10⁻⁵ and 1 × 10⁻⁶. Variants rs142807401 (A432T) and rs139982841 (A452V) in CHRNA9 and variants V132L, V389L, rs34755188 (R480H), and rs75981117 (N549S) in GRIN3A are of particular interest because they are found in both the AA and EA samples. A significant aggregate contribution of rare and common coding variants in CHRNA9 to the risk for ND (SKAT-C: P= 0.0012) was detected by applying the combined sum test in MSTCC EAs. Together, our results indicate that rare variants alone or combined with common variants in a subset of 30 biological candidate genes contribute substantially to the risk of ND.

INTRODUCTION

In recent years, candidate gene and genome-wide association studies (GWAS) have identified several common genetic variants associated with the risk of nicotine dependence (ND). These genes include the nicotinic acetylcholine receptor (nAChR) subunit genes CHRNA5, CHRNA3, and CHRNB4 (clustered on human chromosome 15q) and the CHRNA6 and CHRNB3 genes (clustered on chromosome 8p).^1–3 Examples of findings involving genes other than nicotinic receptors are the nicotine metabolism gene CYP2A6,² the dopamine receptor gene DRD2 and its closely linked gene ANKK1,^{4, 5} the dopamine hydroxylase gene DBH,⁶ the brain-derived neurotrophic factor gene BDNF,^{6, 7} and the synaptic maintenance gene NRXN1.^{8, 9} However, the variants of these susceptibility genes can explain only a small to modest part of the estimated heritability for ND; e.g., alleles of the CHRNA5-CHRNA3-CHRNB4 nAChR gene cluster explain < 1% of the variance in the amount smoked.¹⁰ On the other hand, there is increasing evidence that both common and rare or low-frequency genetic variants are playing a significant role in the involvement of each susceptibility gene for ND and other complex human diseases.^11–13

Several studies have revealed that rare variants of nAChR subunits are associated with ND both genetically and functionally. Wessel et al.¹⁴ investigated the contribution of common and rare variants in 11 nAChR genes to Fagerström Test for Nicotine Dependence (FTND) scores in 448 European-American (EA) smokers who participated in a smoking cessation trial. Significant association was found for common and rare variants of CHRNA5 and CHRNB2, as well as for rare variants of CHRNA4. Xie et al.¹⁵ followed up on the CHRNA4 finding by sequencing exon 5, where most of the rare nonsynonymous variants were detected, in 1,000 ND cases and 1,000 non-ND control subjects with equal numbers of EAs and African Americans (AAs), and reported that functional rare variants within CHRNA4 might reduce ND risk. Recently, Haller et al.¹⁶ detected protective effects of rare missense variants at conserved residues in CHRNB4 and examined functional effects of the three major association signal contributors (T375I and T91I in CHRNB4 and R37H in CHRNA3) in vitro, the minor alleles of which increased cellular response to nicotine. However, like the other two studies, Haller et al.¹⁶ limited their sequencing targets to nAChR subunits.

To address whether genes other than nAChR subunit genes having common variants associated with ND also contain rare ND susceptibility variants, this study was conducted with the goal of determining both the individual and the cumulative effects of rare and common variants in genes/regions implicated in ND candidate gene studies and/or GWAS through pooled sequencing of a subset of our Mid-South Tobacco Family (MSTF) samples followed by conducting validation in an independent case-control sample. Additionally, we implemented a three-step strategy to identify association signals of rare and common variants within the same genomic region. First, we evaluated each common variant individually with a univariate statistic; i.e., logistic and linear regression models. Second, rare variants were grouped by genomic regions and analysed using burden tests, i.e., the Weighted Sum Statistic (WSS);¹⁷ third, we tested for combined effects of rare and common variants with a unified statistical test that allows both types of variants to contribute fully to the overall test statistic.¹⁸

MATERIALS AND METHODS

Subjects

Four hundred subjects (200 sib pairs) were selected for variant discovery from the MSTF population based on ethnic group (AAs or EAs), smoking status (smokers or non-smokers), and FTND scores (light smokers: FTND < 4 or heavy smokers: FTND 4). The reasons for us to choose participants from our family study as discovery samples for deep-sequencing analysis were based on the following two main factors. First, recent studies have shown that rare variants are enriched in family data. If one family member has a rare allele, half of the siblings are expected to carry it, and hence, variants that are rare in the general population could be very commonly present in certain families.¹⁹ Second, family-based designs are advantageous for their robustness to population stratification. Participants in this family-based study were recruited between 1999 and 2004 primarily from the Mid-South states within the USA. More detailed descriptions of demographic and clinical data for these participants can be found in Supplementary Table 1 and previous publications from our group.^{9, 20–22}

Subjects used for variant validation and analysis were recruited from the same geographical area during 2005–2011 as part of the Mid-South Tobacco Case-Control (MSTCC) study under the same recruitment criteria used for the MSTF sample except the subjects were required to be biologically unrelated to each other. Written informed consent was obtained from all participants under the aegis of a human research protocol approved by the IRB of the University of Virginia and University of Mississippi Medical Center. Questionnaires assessing various smoking-related behaviours and other characteristics of interest were administered to participants. Individuals exhibiting substance dependence or abuse other than for alcohol were excluded. The MSTCC sample included 3,088 unrelated AAs (1,454 smokers and 1,634 non-smokers) and 1,430 unrelated EAs (758 smokers and 672 non-smokers). All smokers had smoked at least 100 cigarettes in their lifetimes, while non-smokers were required to have smoked 1–99 cigarettes in their lifetimes, but had no tobacco use in the past year. The ND of each smoker was assessed by the FTND, a commonly used measure, as well as indexed cigarettes per day (CPD) based on a 0 to 3 scale (0: 1–10 CPD, 1: 11–20 CPD, 2: 21–30 CPD and 3: > 30 CPD). Detailed characteristics of the MSTCC AA and EA samples are summarized in Table 1.

Table 1.

Demographic and Phenotypic Characteristics of MSTCC AA and EA Samples

Characteristic	AA (N = 3,088)		EA (N = 1,430)
	Smokers	Non-smokers	Smokers	Non-smokers
Sample size	1,454	1,634	758	672
Female (%)	681 (46.8)	962 (58.9)	380 (50.1)	451 (67.1)
Age, years (SD)	43.6 (12.5)	42.1 (14.2)	41.6 (12.2)	45.1 (14.9)
Indexed CPD (SD)	1.9 (0.4)	N/A	1.9 (0.5)	N/A
FTND Score (SD)	8.6 (1.2)	N/A	8.0 (1.9)	N/A

Notes:

1) SD = standard deviation; N/A = not applicable.

2) Indexed CPD and FTND scores are for smokers only.

3) Indexed CPD: 0 (1–10 CPD), 1 (11–20 CPD), 2 (21–30 CPD), 3 (>30 CPD).

4) FTND Score: possible range 0–10.

Sequencing and Genotyping

We used a customized capture panel of 30 targeted genes, which included nAChR subunit genes and several neurotransmitter receptor and metabolism genes. Almost all of these genes have been reported by our or other research groups to be associated with at least one ND measure in either AA or EA samples. Please refer to Table 2 for the detailed gene list and related references. The coding regions, UTR regions, and flanking sequences of these genes were covered by the Agilent Sure Select Capture panel (250 kb). We divided the 400 samples from the MSTF study into eight pools based on ethnic group, smoking status, and FTND scores to conduct high-throughput sequencing (50 samples/pool).²³ The concentration of each DNA sample was first measured using the QuantiT^™ dsDNA assay kit (Life Technologies, Carlsbad, CA) and then 50 DNA samples were pooled in equimolar amounts, as suggested by the manufacturers. Each pooled DNA sample was subjected to library preparation, targeted capture, and high-throughput sequencing (72 bp paired-end) according to the protocols suggested by the manufacturers. Base quality recalibration and alignment were performed using the Burrows-Wheeler Aligner (BWA)²⁴ referencing hg19. We used Syzygy¹¹ to call variants from the pooled targeted resequencing data.

Table 2.

Biological Information on Rare and Common Variants of 30 Candidate Genes

Gene	(SNP type)/amino acid change	dbSNP ID	Chr.	Hg19 position	Allele 1/Allele 2	Allele 1 freq. in AA (%)	Allele 1 freq. in EA (%)	PhyloP score	SIFT prediction	Polyphen prediction	Ref.
CHRNB2	E34G	rs200223952	1	154541974	G/A	0	0.07	4.33	Tolerated	Benign	20
	Y178*	–		154543833	A/C	0.02	0	0.08	Premature stop codon
NRXN1	T274P	rs77665267	2	50280522	G/T	0.03	0.07	5.01	Damaging	Benign	9
	R206L	–		50280725	A/C	0.11	0.04	4.46	Damaging	Probably damaging
	(Intron)	rs10208208		50593914	T/G	14.82	2.32	0.05
	(Intron)	rs10490227		50659515	T/C	23.63	13.73	−0.43
	(Intron)	rs6721498		50713012	G/A	49.37	52.02	−1.17
	Y367*	–		50765614	T/G	0.02	0	−0.12	Premature stop codon
	S62*	–		50850606	T/G	0.02	0	6.04	Premature stop codon
	(Intron)	rs2193225		51079482	C/T	21.54	50.36	−1.19
CHRNA1†	E436D	rs61737716	2	175613317	A/C	0	0.04	0.65	Tolerated	Probably damaging
DRD3†	G9S	rs6280	3	113890815	T/C	26.30	63.00	0.11	Tolerated	Benign	60
	(Intron)	rs7638876		113894300	T/C	19.20	62.20	−2.06
	(Intergenic)	rs9825563		113900220	G/A	48.98	32.71	−0.30
CHRNA9	A312T	rs56210055	4	40356031	A/G	7.19	0.85	6.20	Damaging	Possibly damaging
	A315V	rs55633891		40356041	T/C	15.07	12.55	4.48	Damaging	Benign
	A432T	rs142807401		40356391	A/G	0.06	0.07	4.45	Tolerated	Benign
	A452V	rs139982841		40356452	T/C	0.14	0.04	6.02	Damaging	Probably damaging
DRD1	(Intergenic)	rs265975	5	174862195	C/T	35.36	60.71	−0.31			61
	(3′ UTR)	rs686		174868700	A/G	43.08	63.57	−0.26
	R226W	–		174869427	A/G	0.02	0	2.92	Damaging	Probably damaging
	(5′ UTR)	rs4532		174870150	C/T	11.44	33.39	−0.80
DDC	(Intron)	rs1451371	7	50553051	C/T	30.62	47.20	0.72			21
	(Intron)	rs3735273		50596864	T/C	36.10	20.96	−0.41
	E61D	rs11575292		50611601	A/C	1.38	0.18	0.39	Tolerated	Probably damaging
	(Intron)	rs921451		50623285	C/T	22.22	30.50	−0.14
TAS2R38	R274C	rs114288846	7	141673087	A/G	1.93	0.11	0.48	Damaging	Probably damaging	62
	V262A	rs1726866		141672705	A/G	32.88	49.82	0.88	Tolerated	Benign
	W135G	rs139843932		141672670	C/A	0.78	0.04	2.68	Damaging	Probably damaging
CHRNA2	S488*	–	8	27320497	T/G	0.02	0.07	1.74	Premature stop codon
	R121L	–		27324833	A/C	0.02	0	1.50	Damaging	Possibly damaging
	T22I	rs2472553		27328511	A/G	16.62	13.32	−0.33	Tolerated	Benign
CHRNB3	(Intergenic)	rs10958725	8	42524584	G/T	30.63	74.85	0.35			63
	(Intergenic)	rs10958726		42535909	T/G	39.79	75.02	−0.13
	(Intergenic)	rs4736835		42547033	C/T	34.85	74.75	−1.27
	(Intergenic)	rs6474412		42550498	T/C	34.78	74.54	−1.64
	(5′ UTR)	rs4950		42552633	A/G	27.16	73.95	−0.40
	(Intron)	rs13280604		42559586	A/G	27.40	73.94	0.43
	(Intron)	rs6474415		42562938	A/G	23.03	73.66	−0.69
	H410Y	–		42587678	T/C	0.05	0	2.31	Damaging	Possibly damaging
	K451E	rs35327613		42591735	G/A	4.91	0.25	1.60	Tolerated	Benign
NTRK2	L140F	rs150692457	9	87322819	C/G	0.35	0	0.64	Damaging	Probably damaging	64
	(Intron)	rs1187272		87404086	A/G	37.19	66.53	2.11
	C623*	–		87563481	A/C	0.02	0	−0.14	Premature stop codon
GABBR2	P742Q	–	9	101068407	T/G	0.03	0	5.15	Damaging	Probably damaging	65
	G671C	-		101068621	A/C	0.02	0	5.07	Damaging	Probably damaging
	(Intron)	rs2491397		101205162	T/C	44.61	51.63	0.70
	(Intron)	rs2184026		101304348	T/C	6.31	22.78	−0.78
	A120A	rs3750344		101340316	C/T	26.07	18.20	0.33
GRIN3A	(Intron)	rs11788456	9	104348150	G/A	45.09	44.93	0.20			53
	(Intron)	rs17189632		104368002	A/T	36.72	43.59	0.11
	N549S	rs75981117		104433048	C/T	0.11	0.51	3.22	Damaging	Possibly damaging
	R480H	rs34755188		104433255	T/C	0.33	1.88	4.16	Damaging	Probably damaging
	V389L	–		104449017	A/C	0.02	0.04	4.27	Damaging	Possibly damaging
	V132L	–		104499868	A/C	0.02	0.04	3.84	Damaging	Probably damaging
DNM1	L16M	rs61757224	9	130965795	A/C	0.05	0.19	1.07	Damaging	Probably damaging	66
	S126*	–		130981002	A/C	0.05	0	5.99	Premature stop codon
	R228L	–		130982360	T/G	0.03	0	6.05	Damaging	Probably damaging
	Y231*	–		130982464	A/C	0.02	0	2.36	Premature stop codon
	F336F	rs3003609		130984755	T/C	11.29	54.62	−0.03
DBH	(Intergenic)	rs3025343	9	136478355	A/G	2.03	10.37	0.63			2
	I340T	rs182974707		136509437	C/T	0.06	0.04	1.84	Tolerated	Possibly damaging
	A362V	rs75215331		136513028	T/C	0.06	0.07	5.39	Damaging	Probably damaging
	Y389*	–		136513110	A/C	0.06	0	1.76	Premature stop codon
	T395P	–		136513126	C/A	0.02	0	4.42	Damaging	Probably damaging
	G482R	rs41316996		136521654	A/G	0.06	0.32	1.89	Damaging	Probably damaging
	R549C	rs6271		136522274	T/C	1.58	6.39	1.68	Damaging	Probably damaging
CHAT	(5′ UTR)	rs1880676	10	50824117	A/G	4.91	23.21	2.03			67
	A120T	rs3810950		50824619	A/G	4.89	23.19	0.88	Tolerated	Benign
	E188G	rs75011234		50827946	G/A	0.33	0.36	1.64	Damaging	Probably damaging
	L243F	rs8178990		50830171	T/C	1.14	4.98	2.26	Damaging	Probably damaging
	G284S	rs146236256		50833616	A/G	0	0.04	5.92	Damaging	Probably damaging
	P299L	rs868749		50833662	T/C	0.02	0.04	6.01	Damaging	Probably damaging
LOC100188947†	(Intron)	rs1329650	10	93348120	T/G	9.50	26.86	−2.00			1
	(Intron)	rs1028936		93349797	C/A	8.10	18.32	−0.33
CHRNA10	R421C	rs2231548	11	3687429	A/G	1.22	0.07	2.36	Damaging	Probably damaging
	R351W	rs139793380		3687639	A/G	0.10	0.04	1.65	Damaging	Probably damaging
	L348R	rs147150654		3687647	C/A	0.21	0	3.01	Damaging	Probably damaging
	V248L	rs2231542		3688615	A/C	0.08	0	1.19	Damaging	Possibly damaging
	W86G	–		3690532	C/A	0.02	0.04	4.55	Damaging	Probably damaging
	E85G	rs77958837		3690534	C/T	0.02	0	2.80	Tolerated	Benign
	T77N	rs55719530		3690558	T/G	1.67	1.92	3.86	Damaging	Probably damaging
BDNF	V74M	rs6265	11	27679916	T/C	3.13	14.36	3.63	Damaging	Possibly damaging	7
	(Intron)	rs6484320		27703188	T/A	7.48	18.11	1.01
	E6K	rs66866077		27720937	T/C	1.27	5.84	0	Damaging	Benign
	(Intron)	rs2030324		27726915	G/A	47.29	49.27	0.12
	(Intron)	rs7934165		27731983	A/G	47.22	49.12	1.07
NRXN2	T1371P	–	11	64390287	G/T	0.51	0.76	4.09	Damaging	Probably damaging
	V53G	–		64410118	C/A	0.03	0	2.81	Damaging	Possibly damaging
	E267G	–		64457927	C/T	0.02	0	2.37	Damaging	Possibly damaging
ARRB1†	H198P	–	11	74989678	G/T	0.02	0.04	4.33	Damaging	Possibly damaging	68
ANKK1	C52W	rs111789052	11	113258762	G/C	6.09	0.40	2.25	Damaging	Probably damaging	5
	R122H	rs35877321		113264382	A/G	0.22	1.27	0.70	Damaging	Possibly damaging
	R185Q	rs115800217		113265724	A/G	10.26	0.94	0.62	Tolerated	Probably damaging
	R237*	rs56047699		113266815	T/C	0.02	0.04	1.36	Premature stop codon
	S313*	-		113268045	A/C	0.19	0.11	−0.43	Premature stop codon
	G318R	rs11604671		113268059	A/G	10.92	42.70	−0.01	Tolerated	Benign
	P351S	rs186633697		113269742	T/C	0.21	0	0.03	Tolerated	Benign
	E376K	rs56299709		113269817	A/G	1.24	0.04	3.29	Tolerated	Probably damaging
	R445C	rs78229381		113270024	T/C	6.16	0.58	0.66	Damaging	Probably damaging
	E458G	rs184645039		113270064	G/A	0.54	0.11	4.23	Damaging	Probably damaging
	H490R	rs2734849		113270160	G/A	16.71	43.24	−1.12	Tolerated	Benign
	E587*	rs113005509		113270450	T/G	2.28	0.14	0.84	Premature stop codon
	Q657*	rs202222056		113270660	T/C	0.49	0	1.08	Premature stop codon
	R734C	–		113270891	T/C	0.03	0.11	0.06	Damaging	Probably damaging
DRD2	E181*	–	11	113286325	A/C	0.02	0.04	3.68	Premature stop codon		5
	(Intron)	rs2075654		113289066	T/C	4.25	19.73	0.49
	(Intron)	rs2075652		113294898	A/G	4.87	1.12	−0.01
	(Intron)	rs4586205		113307129	T/G	35.61	71.86	−0.88
NRXN3	Y234*	rs199840331	14	79181259	A/C	0.02	0	−0.33	Premature stop codon
	G696*	–		79433576	T/G	0.16	0.04	6.33	Premature stop codon
	T99P	–		79933611	C/A	0.05	0.04	5.18	Damaging	Possibly damaging
CHRNA5	(Intron)	rs588765	15	78865425	T/C	29.46	38.84	−0.27			69
	V134I	rs2229961		78880752	A/G	0.40	0.95	5.99	Damaging	Probably damaging
	K167R	rs80087508		78882233	G/A	1.87	0.11	5.01	Damaging	Probably damaging
	D398N	rs16969968		78882925	A/G	6.01	29.51	3.19	Tolerated	Benign
CHRNA3	(3′ UTR)	rs578776	15	78888400	G/A	46.33	65.19	0.09			69
	H217Y	rs72650603		78894335	A/G	0.05	0.22	6.42	Damaging	Probably damaging
	Y215Y	rs1051730		78894339	A/G	12.81	30.20	2.54
	(Intron)	rs6495308		78907656	C/T	29.74	29.26	−1.56
	R37H	rs8192475		78911230	T/C	1.04	4.40	3.28	Damaging	Possibly damaging
CHRNB4	R497C	–	15	78917483	A/G	0.05	0	−1.82	Damaging	Probably damaging	69
	F462V	–		78917588	C/A	0.02	0	4.75	Damaging	Probably damaging
	R349C	rs56235003		78921602	A/G	0.10	0.61	1.40	Damaging	Probably damaging
	P145A	–		78922214	C/G	0.02	0	5.76	Damaging	Probably damaging
	S140G	rs56218866		78922229	C/T	4.25	0.83	2.22	Tolerated	Possibly damaging
	T91I	rs12914008		78923505	A/G	0.73	3.58	1.72	Tolerated	Benign
	N41S	rs75495090		78927863	C/T	1.40	0.22	4.40	Damaging	Probably damaging
CDH13	N39S	rs72807847	16	82892037	G/A	3.18	0.83	1.25	Tolerated	Benign
	V464I	rs200591230		83711918	A/G	0.02	0.07	3.39	Damaging	Probably damaging
ARRB2	T84P	–	17	4619841	C/A	0.02	0	4.29	Damaging	Probably damaging	68
	H281Q	–		4622686	A/C	0.03	0	−0.47	Damaging	Possibly damaging
CHRNA4†	(3′ UTR)	rs2236196	20	61977556	A/G	35.26	73.67	−0.24			20
	P457L	rs201739273		61981180	A/G	0.03	0	0.49	Damaging	Possibly damaging
	(Intron)	rs2273504		61988061	A/G	15.84	17.85	−0.53

Notes:

1)^† = none or only one rare variant validated in this gene, so burden rare variant analysis was not applicable; - = not reported in dbSNP database by 2/17/2014; Chr. = chromosome; Freq. = frequency; Ref. = reference.

2) SNP positions are based on human genome reference assembly build 37.1 (hg19).

3) PhyloP score is basewise vertebrate conservation score.

Together, about 62 GB (868 million reads) of raw sequencing data was obtained from deep-sequencing analysis of the eight pooled DNA samples, with an average of 108 million reads per pooled sample. After appropriate quality control and data filtering, more than 80% of the raw sequencing data was mapped to hg19. A total of 147 million reads were mapped to the targeted regions, which were 100% covered with a median coverage of 106× for each individual DNA sample. Minor allele frequencies (MAF) were calculated for 25 common variants within coding regions and compared with our previous genotyping results based on the TaqMan^® assay for individual DNA samples, which revealed that the MAF correlations between the results of the two methods are 0.97 for AA samples and 0.90 for EA samples.²³

After removing intronic and synonymous variants, we identified 430 putative functional variants with a minimum read of more than 500 and an MAF of more than 0.75% from our deep-sequencing analysis of pooled DNA samples. Next, based on their SIFT²⁵ and PolyPhen²⁶ scores and MAF rankings, we selected 130 variants, which included 118 rare and 12 common variants, for further validation using independent MSTCC samples. An additional 62 common variants were chosen from the literature on association studies of the 30 genes for validation, based on the fact that they had been reported to be nominally or significantly associated with different ND measures (for a detailed list of these reports, please see Table 2). Selection of the 130 rare and common variants was based on the SIFT²⁵ and PolyPhen²⁶ predictions with the following criteria: 1) all premature stop codons; 2) damaging variants presented in either smoker or non-smoker samples; and 3) damaging and benign variants with an MAF ratio > 1.5 between the smoker and non-smoker samples with the goal of increasing the likelihood of detecting significant single nucleotide polymorphisms (SNPs) from the two groups. These SNPs were genotyped on the TaqMan^® OpenArray^® genotyping system (Life Technologies, Carlsbad, CA) for the case control samples. All experiments related to deep sequencing and genotyping validation were performed in the Laboratory of Neurogenetics at the NIAAA, NIH.

Data analysis

We arbitrarily used a 5% MAF threshold to define rare and common variants for all samples. Conservation status was determined by the basewise vertebrate conservation PhyloP score.²⁷ A site was defined as conserved when its PhyloP score was ≥ 2, corresponding to a P value of 0.01. Both SIFT²⁵ and PolyPhen²⁶ were used to predict the effect of nonsynonymous variants on protein structure and function. SIFT yields two predictions: tolerated and damaging, and PolyPhen offers three: benign, possibly damaging, and probably damaging. Because all samples were recruited from the same geographical region of Mississippi following exactly the same inclusion and exclusion criteria, significant population stratification was not detected in smokers vs. non-smokers in either AAs or EAs based on principal component analysis of 49 and 51 common variants included in this study, respectively, for each ethnic group (Supplementary Figure 2) and other genotyping results on the same samples (data not shown).

For common variants, we performed individual SNP-based association analysis with smoking status using logistic regression models and with FTND and indexed CPD using linear regression models as implemented in PLINK.²⁸ Additive, dominant, and recessive genetic models were tested for each SNP, adjusted for sex and age in the AA and EA samples separately. All common variants were in Hardy-Weinberg equilibrium within population.

As reported that grouping rare variants together would increase statistical power for association analysis, we used the WSS pooling method ¹⁷ to test for association of rare variants with smoking status. This method is applicable to genomic regions with at least two rare nonsynonymous variants. In most cases, one genomic region contained a single gene, the exceptions being the ANKK1/DRD2 and CHRNA5/A3/B4 gene clusters. The WSS method can accommodate only binary response variables because of its intrinsic characteristics.¹⁷ In WSS, rare variant counts within the same genomic region for each individual are accumulated rather than collapsed, as implemented in the Cohort Allelic Sums Test (CAST).²⁹ This method puts greater weight on alleles with lower frequencies in controls, which have a higher tendency to be functional both biologically and statistically. Scores for all subjects are then ordered, and the WSS is computed as the sum of ranks for all cases. Variants over-represented in cases will have larger WSS values. Then 10⁶ permutations were performed to determine P values for each genomic region. Limited by computational burden, 10⁸ permutations were implemented only when 10⁶ phenotype rearrangements were insufficient to acquire an exact P value.

After obtaining association results for common and rare variants separately, we evaluated the cumulative effects of both rare and common variants on smoking status using the combined sum tests (i.e., SKAT-C and Burden-C) and adaptive sum tests (i.e., SKAT-A and Burden-A) with age and sex controlled.¹⁸ Smoking status was used as the sole response variable for the following two reasons: 1) to keep analysis results consistent with rare variant analysis; and 2) the other two phenotypes (FTND and indexed CPD) are available for smokers only, use of which means excluding around half of the samples and rare variants presented only in non-smoker samples. The combined sum tests choose the weight parameter in such a manner that rare and common variants contribute equally to the overall test statistic. In contrast, the adaptive sum tests are more powerful if the overall effect sizes of rare and common variants are very different, for example, when a trait is associated only with rare or common variants in the region. Because the relative contribution of rare and common variants to ND risk is unknown, we used both tests to estimate their combined effects. Burden and variance-component (e.g., SKAT) tests are two major types of group-wise association tests proposed for rare variant analysis, which in this case were extended to accommodate combined analysis of rare and common variants by adjusting the weighting scheme. Only genomic regions with at least one rare and one common variant can be analysed by this approach.

To determine the effect directions of significant results obtained from the above group-wise tests, we performed case control-based association analysis for each rare variant using PLINK.²⁸ Then rare variants were separated into two groups based on their estimated odds ratios (OR): if OR > 1, the rare variant was predicted to increase smoking risk; if OR < 1, the rare variant was considered to be protective. However, limited by low frequencies of the rare variants and our moderate sample size used in this study, the OR was not available for every rare variant, which happened mostly for rare variants with fewer copies of the minor allele. In this case, we assigned the variant to the risk group if more minor alleles were present in smokers; otherwise, to the protective group. For collapsing methods, such as the WSS test, the statistical power decreases dramatically as the proportion of functional variants excluded from the analysis increases.³⁰ Also, because most of the genes or genomic regions investigated in this study have only 2 to 4 rare variants, splitting them on the basis of their effect directions would provide little information about association with the phenotype of interest given our sample sizes.³¹

As a result, we only performed effect direction specific combined and adaptive sum tests, not WSS, as described above to further characterize cumulative variant effect directions. Even though we put rare and common variants with the same effect direction together, some of the groups still had limited number of variants. For groups with one rare variant and one common variant, SKAT-C and Burden-C tests are equivalent, so do SKAT-A and Burden-A tests; if only rare or common variants exist in a group, SKAT-C will provide the same results as SKAT-A, which also applies to Burden-C and Burden-A; in cases of only one rare or common variant, all four tests are equivalent to logistic regression analysis.

Bonferroni corrections were used to select significant association results for all analyses. Uncorrected P values are presented throughout the manuscript.

RESULTS

Description of variants and their functionality prediction

There existed 135 out of the 192 variants selected for validation in the MSTCC samples based on genotyping results, which include 33 novel variants (25%; without rs numbers in the dbSNP database as searched on 2/17/2014) in 30 candidate genes (Table 2). As shown in Figure 1A, 58% of these variants (n = 78) are missense; 11% (n = 15) are nonsense–premature stop codons; and 2% (n = 3) are synonymous; the remaining 29% (n = 39) are from intronic, intergenic, or untranslated regions. Of the 93 non synonymous variants, 79 (85%) were predicted to be damaging by PolyPhen, SIFT, or both. The prediction concordance rate between SIFT and PolyPhen programs was 51% (69/135); 14 of 69 were predicted as tolerated by SIFT and benign by PolyPhen; the remaining 55 were predicted to be damaging by SIFT and possibly or probably damaging by PolyPhen. All 33 novel variants were non synonymous; they will be mentioned as amino acid change throughout the manuscript. Additionally, 55%of the coding variants were located at conserved sites (53/96; PhyloP score ≥ 2)²⁷ compared with only 5% of non-coding variants (2/39). The proportion of conserved sites is significantly different among the coding and non-coding variants (Fisher’s Exact p = 1.59×10⁻⁸).

Figure 1.

Descriptive statistics of the 135 validated variants. (A) Proportions of different variant types. Almost 70% of the validated variants lead to amino acid changes. All novel-identified variants (without rs# in dbSNP database by 2/17/2014) are missense. (B) The MAF distribution of variants for the AA and EA samples. The four categories are singleton-only one copy of a rare allele identified in the AA and EA samples, MAF < 1%, MAF < 5%, and common variants. The AA sample has more singletons and low-frequency variants (1% < MAF < 5%) and fewer common variants than the EA sample.

Of the validated variants, 67% are rare (91/135; MAF < 5%) in AAs, EAs, or both (Table 2), many appearing only once in all individuals (17/91 = 19% are singletons) and 7 appearing once only in both the AA and EA samples. Among the 44 common variants, 77% (34/44) belong to non-coding regions compared with 5% (5/91) of the 91 rare variants (Fisher’s exact test P = 8.82 × 10⁻¹⁸), which is consistent with data from exome sequencing studies that non-synonymous coding variants are significantly skewed toward low frequencies.³² Figure 1B compares the MAF distributions in the AA and EA samples for different MAF groups, revealing a higher percentage of singletons and rare variants with MAF between 1% and 5%, and a lower proportion of common variants in the AA sample relative to the EA sample.

Association analysis results for common variants

There are 24 SNPs across 12 genes (DRD3, CHRNA9, DRD1, DDC, CHRNB3, NTRK2, GABBR2, BDNF, ANKK1, DRD2, CHRNA3, and CHRNA4) and one genomic region (LOC100188947) that show nominally significant association (P< 0.05) with smoking status, FTND, or indexed CPD in the AA sample (Supplementary Table 2). Of them, rs1051730 in CHRNA3 has the lowest P value, 0.0016 (OR = 2.45; 95% confidence interval [CI] = 1.41, 4.26), which is nominally associated with smoking statusunder the recessive model. Twenty-one SNPs of 8 genes (NRXN1, CHRNA9, TAS2R38, CHRNB3, NTRK2, DBH, CHAT, BDNF, and CHRNA3) and one genomic region (LOC100188947) are nominally associated with the three phenotypes in the EA sample. Both rs1726866 of TAS2R38 and rs2030324 of BDNF have the smallest P value, 0.0017, in the EA sample. The SNP rs1726866 shows nominal damaging effects toward FTND (beta= 0.30; 95% CI = 0.11, 0.49) under the additive model, while rs2030324 nominally protects against FTND (beta= −0.51; 95% CI = −0.83, 0.19) under the recessive model.

The SNPs rs55633891 in CHRNA9, 5 SNPs (rs10958725, rs10958726, rs4736835, rs6474412, and rs13280604) in CHRNB3, rs1187272 in NTRK2, rs1329650 in LOC100188947, and rs6484320 in BDNF show nominally significant associations in both the AA and EA samples (Supplementary Table 2). However, none of these SNPs survives Bonferroni correction (threshold of significance for AAs = 1.13 × 10⁻⁴ for 49 variants, 3 genetic models, and 3 phenotypes; for EAs = 1.09 × 10⁻⁴for 51 variants, 3 genetic models, and 3 phenotypes). Of note, some variants have MAF > 5% in only one sample, which were not called common variants based on our definition, but we performed individual variant analysis for these SNPs.

Association analysis results for rare variants

By using the WSS method, 10 genes (NRXN1, CHRNA9, CHRNA2, NTRK2, GABBR2, GRIN3A, DNM1, NRXN2, NRXN3, and ARRB2) are significantly associated with smoking status in the AA sample (Table 3), with P values ranging from 1.31 × 10⁻⁴ for CHRNA2 to 2.42 × 10⁻³ for GRIN3A based on 10⁶ permutations. The family-wise error rate (FWER) for 19 genomic regions or genes tested in AAs, which contain at least two nonsynonymous rare variants, is 2.63 × 10⁻³ (0.05/19). There are 7 genes (NRXN1, CHRNA9, TAS2R38, GRIN3A, DBH, NRXN3, and CDH13) and 1 gene cluster (ANKK1/DRD2) showing significant associations, at P values between 1 × 10⁻⁶ (DBH and NRXN3) and 3.5 × 10⁻⁵ (CDH13) in the EA sample based on 10⁶ or 10⁸ permutations (i.e., permuting subjects’ smoker/non-smoker status for 10⁶ or 10⁸ times; see Table 3). With 11 genes tested for EAs, the FWER threshold is 4.55 × 10⁻³ (0.05/11). TAS2R38 (P= 2 × 10⁻⁶), NRXN3 (P = 1 × 10⁻⁶), and CDH13 (P= 3.5 × 10⁻⁵) are the three genes that required 10⁸ permutations in order to obtain a reliable P value.

Table 3.

Significant Rare Variant Association Results Using Weighted Sum Statistic (WSS) in AA and EA Samples

Gene	AA Sample		EA Sample
	SNPs	Permuted p value	SNPs	Permuted p value
NRXN1	rs77665267 (p.T274P) - (p.R206L) - (p.Y367*) - (p.S62*)	2.28×10−4	rs77665267 (p.T274P) - (p.R206L)	2×10−6
CHRNA9	rs142807401 (p.A432T) rs139982841 (p.A452V)	3.81×10−4	rs56210055 (p.A312T) rs142807401 (p.A432T) rs139982841 (p.A452V)	8×10−6
TAS2R38	rs139843932 (p.W135G) rs114288846 (p.R274C)	0.5346	rs139843932 (p.W135G) rs114288846 (p.R274C)	2×10−6†
CHRNA2	- (p.S488*) - (p.R121L)	1.31×10−4	N/A	N/A
NTRK2	rs150692457 (p.L140F) - (p.C623*)	4.25×10−4	N/A	N/A
GABBR2	- (p.P742Q) - (p.G671C)	1.58×10−4	N/A	N/A
GRIN3A	rs75981117 (p.N549S) rs34755188 (p.R480H) - (p.V389L) - (p.V132L)	2.42×10−3	rs75981117 (p.N549S) rs34755188 (p.R480H) - (p.V389L) - (p.V132L)	8×10−6
DNM1	rs61757224 (p.L16M) - (p.S126*) - (p.R228L) - (p.Y231*)	3.53×10−4	N/A	N/A
DBH	rs182974707 (p.I340T) rs75215331 (p.A362V) - (p.Y389*) - (p.T395P) rs41316996 (p.G482R) rs6271 (p.R549C)	0.2427	rs182974707 (p.I340T) rs75215331 (p.A362V) rs41316996 (p.G482R)	1×10−6
NRXN2	- (p.T1371P) - (p.V53G) - (p.E267G)	1.49×10−3	N/A	N/A
ANKK1/DRD2	rs35877321 (p.R122H) rs56047699 (p.R237*) - (p.S313*) rs186633697 (p.P351S) rs56299709 (p.E376K) rs184645039 (p.E458G) rs113005509 (p.E587*) rs202222056 (p.Q657*) - (p.R734C) - (p.E181*)	0.8114	rs111789052 (p.C52W) rs35877321 (p.R122H) rs115800217 (p.R185Q) rs56047699 (p.R237*) - (p.S313*) rs56299709 (p.E376K) rs78229381 (p.R445C) rs184645039 (p.E458G) rs113005509 (p.E587*) - (p.R734C) - (p.E181*)	6×10−6
NRXN3	rs199840331 (p.Y234*) - (p.G696*) - (p.T99P)	2.17×10−4	- (p.G696*) - (p.T99P)	1×10−6†
CDH13	rs72807847 (p.N39S) rs200591230 (p.V464I)	0.5231	rs72807847 (p.N39S) rs200591230 (p.V464I)	3.5×10−5†
ARRB2	- (p.T84P) - (p.H281Q)	1.32×10−4	N/A	N/A

^†P value based on 10⁸ permutations.

Notes:

1) Permuted p value = value based on 10⁶ permutations; - = not reported in dbSNP database by 2/17/2014; N/A = not applicable; i.e., without two rare nonsynonymous variants in gene or region.

2) SNPs included in both AA and EA rare variant analysis are underlined.

3) Significant association p values after correction for multiple testing(p< 2.63×10⁻³ for AA sample and p< 4.55×10⁻³ for EA sample) are given in bold. See “Materials and Methods” for details.

The genes NRXN1, CHRNA9, GRIN3A, and NRXN3 have significantly larger WSS values in both AAs and EAs. NRXN1 has two nonsynonymous substitutions (R206L andrs77665267) and two premature stop codons (S62* andY367*) in the AA sample (P= 2.28 × 10⁻⁴), while only R206L and rs77665267 were detected in the EA sample (P= 2 × 10⁻⁶). The two nonsynonymous variants (rs142807401 and rs139982841) of CHRNA9 are found in both the AA (P= 3.81 × 10⁻⁴) and EA (P= 8 × 10⁻⁶) samples, as are the four SNPs (V132L, V389L, rs34755188, andrs75981117) of GRIN3A (P = 2.42 × 10⁻⁴ in AAs; P= 8 × 10⁻⁶ in EAs). For NRXN3, there are two premature stop codons (rs199840331 and G696*) and one nonsynonymous variant (T99P) included in the analysis for AA subjects (P = 2.17 × 10⁻⁴) and one premature stop codon (G696*) and one nonsynonymous variant (T99P) included in the analysis for EA subjects (P = 1 × 10⁻⁶).

Association analysis results for rare and common variants

CHRNA9, with two rare variants (rs142807401 and rs139982841) and two common variants (rs56210055 and rs55633891), and DRD1, with one rare variant (R226W) and three common variants (rs265975, rs686, and rs4532), are nominally associated with smoking status after correcting for sex and age in the AA sample (Table 4). The P values are 0.0495 for CHRNA9 using Burden-A method and 0.0458 using Burden-C, and 0.0430 using Burden-A for DRD1. All four variants of CHRNA9 result in amino acid changes, among which rs56210055 has an MAF of 7.19% in AAs, but only 0.85% in EAs. So in the EA sample, with three rare variants (rs56210055, rs142807401, and rs139982841) and one common variant (rs55633891), CHRNA9 shows significant association, with P values of 0.0012, 0.0032, 0.0036, and 0.0080 using SKAT-C, Burden-C, SKAT-A, and Burden-A, respectively (Table 4). The first three P values survive multiple testing correction for 12 genes, which have at least one rare and one common variant and were eligible to be included in this analysis in the EA sample (0.05/12 = 0.0042). Both rare and common variants of CHRNA9 contribute to the risk for ND in EAs and possibly in AAs.

Table 4.

Significant Combined and Adaptive Sum Test Results of Cumulative Rare- and Common-Variant Effects on Smoking Status in AA and EA Samples

Gene	AA Sample					EA Sample
	Rare variant(s)	Common variant(s)	Effect direction	P value† Separated	Pooled	Rare variant(s)	Common variant(s)	Effect direction	P-value†
									Separated	Pooled
NRXN1	rs77665267 (p.T274P) -(p.S62*)	rs10208208 rs6721498	↑	0.5553 0.3023 0.5354 0.2996	0.5537 0.9368 0.5398 0.9275	rs77665267 (p.T274P) rs10208208	rs10490227	↑	0.1016 0.0527 0.0738 0.0362	0.1062 0.0772 0.0683 0.0505
	- (p.R206L) - (p.Y367*)	rs10490227 rs2193225	↓	0.4244 0.1991 0.5232 0.3547		- (p.R206L)	rs6721498 rs2193225	↓	0.4648 0.2849 0.4506 0.4596
CHRNA9	rs139982841 (p.A452V)		↑	0.2381 0.2381 0.2381 0.2381	0.0706 0.0766 0.1448 0.0495	rs142807401 (p.A432T)	rs55633891 (p.A315V)	↑	0.0143 0.0143 0.0246 0.0246	0.0012 0.0032 0.0036 0.0080
	rs142807401 (p.A432T)	rs56210055 (p.A312T) rs55633891 (p.A315V)	↓	0.0381 0.0143 0.0942 0.0353		rs56210055 (p.A312T) rs139982841 (p.A452V)		↓	0.0119 0.0072 0.0119 0.0072
DRD1	- (p.R226W)	rs265975 rs4532	↑	0.0572 0.0393 0.0549 0.0372	0.1121 0.0458 0.1224 0.0430	N/A
		rs686	↓	0.8101 0.8101 0.8101 0.8101
ANKK1/DRD2	rs56299709 (p.E376K) - (p.R734C) - (p.E181*) rs2075654	rs111789052 (p.C52W) rs115800217 (p.R185Q) rs11604671 (p.G328R) rs78229381 (p.R445C) rs2734849 (p.H490R)	↑	0.4445 0.1179 0.3915 0.1370	0.3915 0.0950 0.4850 0.1361	rs111789052 (p.C52W) rs35877321 (p.R122H) rs56299709 (p.E376K) rs78229381 (p.R445C) rs184645039 (p.E458G) - (p.R734C)	rs2075654 rs4586205	↑	0.3828 0.1708 0.2649 0.1303	0.5566 0.8627 0.4306 0.8619
	rs35877321 (p.R122H) rs56047699 (p.R237*) - (p.S313*) rs186633697 (p.P351S) rs184645039 (p.E458G) rs113005509 (p.E587*) rs202222056 (p.Q657*) rs2075652	rs4586205	↓	0.4965 0.0371 0.5029 0.0273		rs115800217 (p.R185Q) rs56047699 (p.R237*) - (p.S313*) rs113005509 (p.E587*) - (p.E181*) rs2075652	rs11604671 (p.G328R) rs2734849 (p.H490R)	↓	0.7714 0.3605 0.8413 0.2840
CHRNA5/ A3/B4	rs72650603 (p.H217Y) rs8192475 (p.R37H) - (p.R497C) - (p.F462V) rs56235003 (p.R349C) - (p.P145A) rs12914008 (p.T91I) rs75495090 (p.N41S)	rs16969968 (p.D398N) rs1051730 (p.Y215Y)	↑	0.0901 0.0323 0.0701 0.0596	0.2406 0.4546 0.1398 0.4909	rs2229961 (p.V134I) rs72650603 (p.H217Y) rs8192475 (p.R37H)	rs588765 rs16969968 (p.D398N) rs1051730 (p.Y215Y)	↑	0.7389 0.2600 0.6989 0.3322	0.7570 0.9566 0.7185 0.9078
	rs2229961 (p.V134I) rs80087508 (p.K167R) rs56218866 (p.S140G)	rs588765 rs578776 rs6495308	↓		0.5737 0.4017 0.4497 0.4983	rs80087508 (p.K167R) rs56235003 (p.R349C) rs56218866 (p.S140G) rs12914008 (p.T91I) rs75495090 (p.N41S)	rs578776 rs6495308	↓	0.6270 0.3729 0.5879 0.5716

Notes:

1) P values for each gene or region were obtained by four statistical methods; i.e., SKAT-C, Burden-C, SKAT-A, and Burden-A;

^†= p values from top to bottom for each gene or region were obtained in the abovementioned order.

2) Only genes or regions with at least one rare and one common variant were eligible for the pooled analysis; N/A = not applicable.

3) “↑” = variants increase smoking risk estimated from individual variant-based odds ratios (if available) or minor allele counts in Cases and Controls, “↓” = variants decrease smoking risk; effect direction specific tests were applied with p values listed under “Separated”.

4) SNP rs numbers are based on dbSNP database (accessed on 2/17/2014).

5) SNPs included in both AA and EA samples for this analysis are underlined.

6) Nominal significant associations (p< 0.05) for both “Pooled” and “Separated” analyses are given in bold, including p values, SNP, and gene names. See the section of “Materials and Methods” for details.

Nominally significant associations were also detected in effect-direction separated analysis for NRXN1, CHRNA9, DRD1, ANKK1/DRD2, and CHRNA5/A3/B4 (Table 4). Two rare variants (rs77665267 and rs10208208) and one common variant (rs10490227) of NRXN1 in EAs show a P value of 0.0362 using the Burden-A method, indicating a possible combined risk effect of the three variants. The common variant, rs10490227 did not show any significant association with smoking status in individual SNP-based analysis; however, it is nominally associated with FTND (Supplementary Table 2). For CHRNA9, its nominal association in AAs seems to be driven mainly by one rare variant (rs142807401) and two common variants (rs56210055 and rs55633891) with decreased probability of smoking. SNPs rs142807401 and rs55633891 have opposite effects in the EA sample, which suggests population-specific effects or is simply caused by the rough assignment of effect directions as described in Materials and Methods. Three of the four variants in DRD1, which increase smoking risk, result in a nominal association in the AA sample (Burden-C P= 0.0393 and Burden-A P= 0.0372).

Burden-C and Burden-A methods worked as expected for the effect-direction separated analysis according to their theoretical designs and assumptions. BesidesNRXN1, CHRNA9, and DRD1, these two methods discovered nominal associations between the two genomic regions (ANKK1/DRD2 and CHRNA5/A3/B4) that contain the most variants in this study and smoking status in the AA samples as well. Eight rare variants and one common variant in ANKK1/DRD2 together decrease smoking risk, while eight rare variants and two common variants in CHRNA5/A3/B4 display the opposite effect (Table 4).

For groups with rare variants only, the combined and adaptive sum tests revealed nominal associations between TAS2R38, GRIN3A, DNM1, DBH, and smoking status, respectively, in either AAs or EAs (Supplementary Table 4). This can be seen as a confirmation of the association signals detected by the WSS method. Non-significant association results for rare variant analysis and rare and common variant combined analysis are presented in Supplementary Tables 3 and 4.

DISCUSSION

Although none of the 44 common variants showed significant association with any of the three nicotine phenotypes (smoking status, FTND, and indexed CPD) after Bonferroni correction in this study, rare variants in 10 genes (NRXN1, CHRNA9, CHRNA2, NTRK2, GABBR2, GRIN3A, DNM1, NRXN2, NRXN3, and ARRB2) in the AA sample and 7 genes (NRXN1, CHRNA9, TAS2R38, GRIN3A, DBH, NRXN3, and CDH13) plus 1 gene cluster (ANKK1/DRD2) in the EA sample are significantly associated with smoking status using the WSS method. Further, we also detected a significant cumulative effect of both rare and common variants in CHRNA9 that contribute to smoking status with age and sex controlled in the EA sample when applying both the combined and the adaptive sum test.

Among the common variants that are nominally associated with any of the three ND measures, SNP rs1051730 is of great interest. This SNP has the smallest common variant association P value in the AA sample, which has been reported as the most significant genome-wide association in meta-analyses of subjects of European ancestry (P= 2.75 × 10⁻⁷³).^{2, 3, 6, 33} Another wasrs16969968, the most robust genetic finding on chromosome 15q25 in subjects of European ancestry, with a P value of 5.57 × 10^{−72.2, 3, 6, 33} Although we did not find significant associations for these two SNPs in our EA sample, which is likely attributable to the small sample size (758 smokers vs. 672 non-smokers), the nominally significant association presented for the AA sample is of interest, providing an independent replication of the association of this SNP with smoking in our independent samples.

HapMap data show that rs1051730 and rs16969968 are in strong linkage disequilibrium in European and Asian populations but not in AAs (r²= 0.40).³⁴ In a meta-analysis of AA samples, Chen et al.³⁴ found that rs16969968 is more strongly associated with heavy smoking (P= 0.0011) than is rs1051730 (P= 0.011). In our AA sample, however, only rs1051730 is nominally associated with smoking status (P= 0.0016; OR= 2.45; 95% CI = 1.41, 4.26) under the recessive model even though the correlation coefficient between rs1051730 and rs16969968 is 0.42; this is consistent with the HapMap data. As a coding synonymous variant, rs1051730 is expected to have less functional significance than rs16969968, a missense mutation (aspartate to asparagine). So while the functional significance of rs16969968 has been demonstrated in vitro³⁵ and to some extent via α5 knockout mouse models that show a role for the gene,³⁶ the functional relevance of rs1051730 is undetermined. Based on our study result, we suspect that rs1051730 is in linkage disequilibrium (LD) with another functional missense variant with a large effect but low MAF, other than rs16969968, in our AA sample; or it changes CHRNA3 expression in a significant way.

For rare variants, although we have 10 and 8 genomic regions significantly associated with smoking status in the AA and EA samples, respectively, the two ethnic samples provide replication for each other only for four genes that overlapped across the samples: NRXN1, CHRNA9, GRIN3A, and NRXN3. Among the four genes, CHRNA9 and GRIN3A have rare nonsynonymous variants that are seen in both populations, which could be of importance in an evolutionary functional context because of the implication that they are ancient. Because CHRNA9 is also significantly and nominally associated with smoking status for rare and common variant combined analysis in both the EA and AA sample, it will be discussed first.

CHRNA9, which codes for nAChR α9, is located on chromosome 4p15.1-p14 and contains five exons and four introns.³⁷ The protein is composed of 479 amino acids (UniProtKB/Swiss-Prot ID: Q9UGM1; RefSeq ID: NP_060051) and contains two highly conserved domains, which are the neurotransmitter-gated ion-channel ligand binding domain (aa 31–236) and the neurotransmitter-gated ion-channel transmembrane region (aa 244–457).³⁸ The nAChR α9 can form homo- or hetero-oligomericcation-selective channels in conjunction with nAChR α10³⁹ and is usually expressed in the cochlea, keratinocytes, pituitary gland, B-cells, and T-cells.^39–41 Both α9 and α10 nAChR subunits also are coexpressed in dorsal root ganglion neurons.⁴²

The four variants in CHRNA9 that contribute to the association signals are rs56210055 (p.A312T), rs55633891 (p.A315V), rs142807401 (p.A432T), and rs139982841 (p.A452V). All have PhyloP Scores >4 (Table 2). Both ala³¹² and ala³¹⁵ lie within a transmembrane region composed of 22 amino acids (aa 302–323), whereas ala⁴³² and ala⁴⁵² are located within the cytoplasmic region (aa 324–457). The rs139982841 variant has also been identified in lung cancer tissues in the catalogue of somatic mutations in cancer (COSM587183).

Other researchers have reported nominally significant association of CHRNA9 (rs4861065) with ND in a female Israeli sample⁴³ and of CHRNA9 (rs766988 and rs4861065) with response inhibition, as well as of CHRNA9 (rs4861065) with selective attention in a subset of the same sample, in which neurocognitive functions are putatively implicated in ND susceptibility.⁴⁴ Chikova et al.⁴⁵ revealed that rs56159866 and rs6819385 in CHRNA9 are associated with an increased risk of lung cancer, while three SNPs, rs55998310, rs56291234, and rs182073550 (single nucleotide deletion) protect against lung cancer.

All these SNPs are either synonymous variations or within intronic or UTR regions, and therefore lack any obvious direct functional effect but may affect protein production at the transcriptional and/or translational levels or simply manifest association through linkage disequilibrium with other functional variants. In contrast, the four variants we reported in this study all cause amino acid changes, among which rs56210055 (p.A312T) and rs55633891 (p.A315V) may affect nAChR stability or the permeability of the ion channel, while rs142807401 (p.A432T) and rs139982841 (p.A452V) may influence downstream signalling characteristics based on the amino acid locations they affect. Based on the effect direction specific analysis results shown in Table 4, these four variants may have a mixture of risk and protective effects in affecting smoking risk. Thus, future functional studies are warranted for these four SNPs inCHRNA9.

GRIN3A is localized on chromosome 9q34 and consists of nine exons,⁴⁶ which code for glutamate receptor ionotropic NMDA 3A (GluN3A). The deduced protein contains 1115 amino acids (UniProtKB/Swiss-Prot ID: Q8TCU5; RefSeq ID: NP_597702.2) and shows 92.7% identity to rat NMDA receptor 3A.⁴⁶ Functional NMDA receptors are heterotetramers composed of two ζ subunits (GluN1) and two ε subunits (GluN2A, GluN2B, GluN2C, or GluN2D) or third subunits (GluN3A or GluN3B), which serve critical functions in neuronal development, functioning, and degeneration of the mammalian central nervous system.⁴⁷ GluN3A suppresses NMDA receptor functions in a dominant-negative way.^{48, 49} GluN3A-containing NMDA receptors display reduced Ca²⁺ permeability and low sensitivity to Mg²⁺ blockade.^{50, 51} The transcript of GRIN3A was detected by in situ hybridization in human fetal spinal cord and forebrain.⁵²

All four substituted amino acids, val¹³², val³⁸⁹, arg⁴⁸⁰, andasn⁵⁴⁹, are located in the extracellular region of GluN3A and are conserved, with PhyloP scores > 3 (Table 2). We have previously reported common variants of GRIN3A significantly associated with different ND measures in the MSTF population.⁵³ Different variants within GRIN3A have also been associated with Alzheimer’s disease⁵⁴ and schizophrenia.⁵⁵ The recent work by Takata et al.⁵⁵ identified disease association of a missense variant in GRIN3A (p.R480G, rs149729514; P = 0.00042; OR = 1.58) in a Japanese schizophrenia case-control cohort. This association was supported by their meta-analysis with independent Han-Chinese case-control and family samples (combined P= 3.3 × 10⁻⁵). However, as the authors suggested, the GRIN3A R480G variant was not detected in AA and EA populations, and thus it seems to be Asian specific.

In this study, instead of finding the glycine substitution at residue 480, we identified a histidine substitution at the same position of GluN3A in both AAs and EAs. The ingenious connection between the two studies confers great functional importance for this residue not only in ND, but also in other psychiatric disorders. Another variant, rs75981117 (p.N549S), is an N-linked glycosylation site on GluN3A, which could be important for both the structure and function of the protein. SNPs rs75981117 (p.N549S), rs34755188 (p.R480H), and V389L together show a nominal protective effect against smoking risk in AAs (Supplementary Table 4). The functional importance of the four variants may show in ND-related mouse models, as Marco et al.⁵⁶ recently discovered that overexpression of GluN3A in mouse striatum mimicked the synapse loss observed in Huntington’s disease mouse models, whereas genetic deletion of GluN3A prevented synapse degeneration, ameliorated motor and cognitive decline, and reduced striatal atrophy and neuronal loss in the YAC128 Huntington’s disease mouse model.

Because of space limitations, we cannot elaborate on the potential functional importance of the rare variants we identified in NRXN1, CHRNA2, TAS2R38, NTRK2, GABBR2, DNM1, DBH, NRXN2, ANKK1/DRD2, NRXN3, and CDH13 here. To interpret the results of this study more appropriately, five main limitations need to be considered. First, rare variants are usually population specific, or even sample specific, which, on one hand, makes replication very difficult and on the other hand, reveals that the rare variants identified in this study are just a starting point. Association studies of these biological candidate genes in other populations and samples are thus warranted. Second, we limited our search to biological candidate genes, which makes these findings not surprising at the gene level. If we are to uncover new genes, more comprehensive and hypothesis-free analyses, particularly genome-wide sequencing analyses of rare variants, are needed. Third, although none of the 44 common variants showed significant association with any of the three nicotine phenotypes after Bonferroni correction, this does not mean common variants in general are not important in affecting smoking risk. The primary reason for our failure to identify significant association of these common variants with ND measures is more likely related to our sample size, especially for EAs, with a sample size of only 1430. Another reason may be the selection of these common variants from our previous studies, 30 and 7 of which showed nominal or significant associations in preceding analysis of MSTF and MSTCC samples, respectively (see Supplementary Table 2). Nineteen out of the 30 common variants chosen based on previous MSTF study results were found nominally associated with at least one of the three ND measures (i.e., smoking status, FTND, and indexed CPD) in either AA or EA case control samples; however, all 7 common variants selected from one meta-analysis study on CHRNB3 including MSTCC samples showed nominal significance in this study composed solely of MSTCC subjects. Such analysis result difference is likely caused by sample difference – family and case control samples. Although both samples were recruited from the same geographical locations, they were recruited at different time periods with the family samples recruited from 1999–2004 and the case control samples recruited from 2005–2011. This difference is also consistent with regression to the mean for two samples. Fourth, it is hard to dissect the contribution of each rare variant and the relative contributions of rare vs. common variants, hampered by our sample size and the statistical methods we applied. Five, although our subjects were recruited from the same geographical area and the two ancestry-based groups; i.e., AA and EA, are well separated according to our previous reports using common variants,^{57, 58} we still could not completely rule out the possibility of some hidden distributional differentiation of rare and low-frequency variants in our samples, considering the insights provided by the 1000 Genomes Project analyses⁵⁹ and currently lack of the genome-wide profiles of these variants.

We used one type of burden test; i.e., WSS,¹⁷ to accumulate counts of rare variants in separate genomic regions and then examined their overrepresentation in cases vs. controls. The burden test is a compromise between extremely low allele frequency and limited statistical power, which enables detection of pooled rare variant effects but is incapable of disentangling individual effects of rare variants. For combined analysis of rare and common variants, we implemented the combined and adaptive sum tests;¹⁸ the former assumes equal contribution of rare and common variants, and the latter presumes rare variants have different effects than common variants. Without knowing the relative contribution of rare and common variants to any trait of interest, we highly encourage applying both tests to analyze the same dataset as used in this study. We also performed effect direction-specific analyses to examine the combined effect directions of rare and common variants. Because of the limited number of rare variants available for each gene or genomic region and the expected substantial power loss of burden tests when functional variants are excluded, this analytical strategy was applied only to the combined and adaptive sum tests. Nominal association results provided evidence for combined-effect direction speculation of the variant groups; however, no significant association was discovered. This strategy will be more effective with a larger number and more accurate classification of rare variants.

This study demonstrates for the first time the contribution of common and, particularly, rare variants within a subset of biological candidate genes besides nAChR subunit genes, to the risk for ND. Our findings about these variants, especially rs56210055 (p.A312T), rs55633891 (p.A315V), rs142807401 (p.A432T), and rs139982841 (p.A452V) in CHRNA9 and V132L, V389L, rs34755188 (p.R480H), and rs75981117 (p.N549S) in GRIN3A are interesting and encouraging and deserve further study using both in vitro and in vivo approaches.

Supplementary Material

2

Figure S1. Density plot of estimated minor allele frequency (MAF) distributions from pooled-sequencing analysis of the 200 sib pairs. Many more rare variants were discovered in AAs than in EAs.

Figure S2. Scatter plots of principal components 1 and 2 for smokers and non-smokers based on principal component analysis of 49 and 51 common variants available for AAs and EAs, respectively. Smokers and non-smokers are uniformly mixed within the two ethnic groups. Significant population stratification was not detected between smokers and non-smokers.