Nicotinic acetylcholine receptor genes on chromosome 15q25.1 are associated with nicotine and opioid dependence severity
- First Online:
- Cite this article as:
- Erlich, P.M., Hoffman, S.N., Rukstalis, M. et al. Hum Genet (2010) 128: 491. doi:10.1007/s00439-010-0876-6
A locus on chromosome 15q25.1 previously implicated in nicotine, alcohol, and cocaine dependence, smoking, and lung cancer encodes subunits of the nicotinic acetylcholine receptor (nAChR) expressed in the mesolimbic system and thought to mediate substance dependence. Opioid dependence severity (ODS), nicotine dependence severity (NDS), smoking status and quantity, and the number of attempts to quit were assessed using questionnaire instruments in 505 subjects who were prescribed opioid medications for chronic pain in outpatient practice sites. Multivariate regression was used to test for genetic association of these phenotypes with 5 SNPs in the nAChR gene cluster on chromosome 15q25.1, adjusting for background variables. A coding variant in CHRNA5 (rs16969968[A]) was significantly associated with 1.4-unit higher ODS (p < 0.00017). A variant in the 3′ untranslated region of CHRNA3 (rs660652[G]) was significantly associated with 1.7-fold higher odds of lifetime smoking (p < 0.0092), 1.1-unit higher NDS (p < 0.0007), 0.7 more pack-years of cigarette smoking (p < 0.0038), and 0.8 more lifetime attempts to quit (p < 0.0084). Our data suggest an association of DNA variants in the nAChR gene cluster on chromosome 15q25.1 with ODS, as well as NDS and related smoking phenotypes. While the association of this locus with NDS and smoking phenotypes is well known, the association with ODS, a dimension of opioid substance dependence, is novel and requires verification in independent studies.
Polymerase chain reaction
Minor allele frequency
Opioid dependence severity
Severity of dependence scale
Nicotine dependence severity
Fagerstrom tolerance scale
Electronic health record
Nicotinic acetylcholine receptor
Cholinergic receptor nicotinic alpha
Single nucleotide polymorphism
Smoking continues to be a major cause of preventable death, disability, illness, and healthcare costs worldwide (Hays and Ebbert 2008). Despite evidence that treatment improves smoking cessation rates, an estimated 20–25% of the adult population in the US still smoke (Centers for Disease Control and Prevention 2009; The Clinical Practice Guideline Treating Tobacco Use and Dependence 2008 Update Panel, Liaisons, and Staff 2008), resulting in 440,000 smoking-related deaths per year and over $85 billion in annual healthcare expenditures (Heitjan et al. 2008). Many of those who report a desire to quit smoking do not seek treatment and are unable to quit smoking on their own. Advances in the effort to reduce the prevalence of smoking may depend, in part, on a better understanding of the molecular components involved in the development and reinforcement of nicotine dependence and personalizing treatment protocols (Johnstone et al. 2002; Li 2006).
The heritability of nicotine dependence estimated from twin studies is 40 to 60% (Sullivan and Kendler 1999; Lessov et al. 2004). Recent genome-wide association and comprehensive candidate-gene studies consistently identified variation on chromosome 15q25.1 as the most significant genome-wide location for lung cancer, nicotine dependence, and smoking (Amos et al. 2008; Hung et al. 2008; Spitz et al. 2008; Thorgeirsson et al. 2008; Wang et al. 2008; Portugal and Gould 2008; Berrettini et al. 2008; Saccone et al. 2007; Li and Burmeister 2009; Lips et al. 2010; Bierut et al. 2008), suggesting a biologically plausible link between this locus, which encodes components of the nicotinic acetylcholine receptor (nAChR), and increased susceptibility to nicotine dependence and consequent lung cancer risk.
Fine-mapping studies found two distinct association signals for smoking located on separate LD clades within the 15q25.1 locus, one of which included rs16969968, a coding polymorphism in CHRNA5 with a functional effect on nAChR activity in vitro. Berattini et al. (2008) found a haplotype encompassing CHRNA5 and CHRNA3 that conferred predisposition to nicotine dependence. However, the exact locations of underlying functional polymorphism(s) in 15q25.1 affecting nicotine dependence and smoking are not fully mapped.
Twin studies suggest common genetic predisposition factors for multiple substance dependence phenotypes (Xian et al. 2008) and this concept has been reflected in current discussions related to addiction onset and course (Robinson and Berridge 2008; Li et al. 2007; Hogarth and Duka 2006). In addition to nicotine dependence, polymorphisms in 15q25.1 are associated with alcohol and cocaine dependence (Grucza et al. 2008; Wang et al. 2009); however, their association with phenotypes of opioid dependence has not been conclusively shown. Given these prior findings and knowledge gaps, we hypothesized that genetic variants of 15q25.1 are associated with NDS, smoking phenotypes, and ODS. We tested these hypotheses in a sample of 505 ambulatory care patients with a history of long-term prescription opioid use.
Subjects and methods
Source population and recruitment
The study’s recruitment and data collection procedures have been described elsewhere (Boscarino et al. 2010). Briefly, the Geisinger Institutional Review Board (IRB) approved the study. Following IRB approval, the electronic health record (EHR) database of Geisinger Clinic was searched to identify individuals with a history of four or more opioid drug prescriptions electronically ordered within a 12-month period. Geisinger Clinic, the ambulatory care division of the Geisinger Health System, is a Pennsylvania not-for-profit corporation operating a multi-specialty group medical practice treating outpatients at primary care clinics, specialty care clinics, community practice sites, and ambulatory surgery centers. All clinics and surgery centers in this healthcare system, which are located in 31 of Pennsylvania’s 67 counties, have used Epic System’s (Epic System Corporation, Verona, WI) EHR since 2001. The clinical directors of 22 primary care and specialty care clinics having the highest number of potential study subjects were contacted for study participation. The directors of nine primary care and three specialty care clinics (including an orthopedics, pain and rheumatoid clinic) agreed to allow study investigators to contact study-eligible subjects for research participation.
Individuals were eligible for the study if they were 18+ years old as of 1 July 2007, had electronic prescriptions for opioid medications four or more times for non-malignant cancer pain at any time from 30 June 2006 through to 1 July 2007, and the majority of opioid drug orders were placed at one of the participating clinics. Individuals were excluded if they had a diagnosis of cancer associated with their medication orders during the study index period, if they were deceased, or if they had previously declined participation in research studies. Because the proportion of non-whites in the sample was <2%, while representative of the area served by the Geisinger System, these individuals were excluded from the analyses to prevent admixture artifacts. We did not restrict eligibility to filled prescriptions in this initial eligibility query because medication use was verified later in the telephone interview.
Interviews were conducted from August 2007 to November 2008. We first mailed an introductory letter to 2,459 eligible individuals to explain the purpose of the research and the study’s confidentiality protocols. Individuals were also notified that they could call to be excluded from further contact. Trained telephone interviewers subsequently attempted to contact 2,373 eligible individuals by telephone; 86 individuals were not contacted because the study quota was filled before they were called. Up to 15 call attempts were made to complete telephone interviews with the study subjects. Telephone contact was made with 1,390 individuals. The remaining 983 were not reachable or not qualified for the following reasons: deceased, institutionalized, not proficient in English, incapable of answering questions, or denied taking pain medications. In the beginning of each interview following an introduction, identification, and explanation of the study by the interviewer, each contacted person was asked if he/she is willing to be interviewed. Those who agreed and completed the interview were sent a consent form accompanied by a buccal swab kit and pre-paid return envelope. Altogether, 685 persons declined to participate. Finally, out of the 705 individuals who completed the interview, 505 (72%) returned the adequately signed consent form and buccal swab. An additional 17 buccal swabs did not produce adequate DNA for genotyping. Up to 5 attempts (by mail and telephone) were made to remind each interviewed participant to return the consent form and DNA sample. Details on data collection follow.
Following verbal consent, a structured diagnostic telephone interview was administered using a computer-assisted telephone interview (CATI) system (WinCati, version 4.2 [Sawtooth Technologies, Northbrook, IL, USA]). For our study, we used an existing diagnostic interview (Kessler and Ustun 2004), modified to assess prescription opioid misuse. The instrument also collected data on other substance use disorders, as well as demographic and socio-economic information. In addition, other questionnaire instruments were used to assess cigarette smoking and tobacco dependence (Fagerstrom 1978; Heatherton et al. 1991).
Interviews required 50 to 60 min to complete and were administered by staff with experience and training in the use of the survey instrument and in emergency mental health referral protocols. A referral list of local drug and alcohol counseling services was available and provided during the telephone interview and by mail, if requested. Onsite phone room managers and investigative staff supervised and monitored the interviewers. The buccal swab kit was mailed to consenting individuals, along with a postage-paid return envelope. Participants were offered an incentive for their interview time and effort and for retuning a buccal swab sample by mail.
Phenotypic measures and potential confounders
Smoking status was defined according to the question “did you smoke 100 cigarettes or more in your life?” administered in the telephone interview.
Pack-years of cigarette smoking was calculated as the product of the reported number of cigarettes smoked per day times the total number of years smoked.
Number of attempts to quit smoking was ascertained from the response to the question: “did you ever try to quit smoking?” and if yes “how many times did you try to quit smoking?”
Nicotine Dependence Severity (NDS): we used the Fagerstrom Tolerance Scale (FTS) to assess NDS in the telephone interview (Fagerstrom 1978). FTS measures smoking behaviors associated with NDS and is scored on a scale of 0–11, where higher scores are associated with greater nicotine dependence and withdrawal symptoms. This score was analyzed both as a continuous variable and a dichotomized variable. For the latter, a score of 7+ was used to define nicotine dependence based on prior studies (Heatherton et al. 1991).
Opioid dependence severity (ODS): we used the Severity of Dependence Scale (SDS) for opioids to assess ODS (Gossop et al. 1995). The SDS is based on 5 items, each scored on a 4-point scale (0–3). The total score is obtained by summing the 5-item ratings. The higher the score on the SDS, the higher the severity of drug dependence as validated in previous studies (Gossop et al. 1995; Kaye and Darke 2002; World Health Organization 2009). The questions comprising the SDS instrument were: (i) Do you think your use of prescription opioids was out of control? (ii) Did the prospect of missing a dose make you anxious or worried? (iii) Did you worry about your use of prescription opioids? (iv) Did you wish you could stop? (v) How difficult did you find it to stop or go without prescription opioids?
Brief pain inventory (BPI): to assess the level of pain among study participants we used the Brief Pain Inventory (BPI) (Cleeland and Ryan 1994; Tan et al. 2004), a widely used pain assessment scale. The BPI was used here to assess the current level of overall pain, the level of pain over the past week, and pain-related functional impairment, i.e., to what extent pain has interfered with the participant’s work or lifestyle over the past week.
Other measures ascertained from EHR: clinic type was defined as the category of clinic in which a participant received the majority of his/her opiate prescriptions (primary care vs. specialty care); and number of opioid prescriptions was the count of all EHR drug orders for opioids received over the past 3 years.
Other measures ascertained in telephone survey: low household income (i.e., classified as low if the total income of the household was ≤$30,000), marital status (i.e., married, separated, divorced, single, or widowed), education level (i.e., K to 8th grade, 9 to 11th grade, high school graduate, some college, or college graduate or higher), and employment status (i.e., employed, not employed or retired).
All statistical tests were performed using SAS version 9.2. (SAS Institute Inc., Cary, NC, USA) except when specified. Multivariate testing for the association of each phenotype with each marker was performed using linear (for continuous traits) or logistic (for binary traits) regression implemented in SAS Proc. GLM or Proc. Logistic, respectively. For each SNP, we first tested an additive intra-locus coding scheme (i.e., subjects assigned 0, 1 or 2 according to the number of copies of the minor allele), followed by dominant/recessive coding schemes (i.e., carriers and homozygotes for the tested allele assigned a value of 1, vs. homozygotes for the alternate allele assigned a value of 0), where the results of the additive analysis were statistically significant. Estimation of linkage disequilibrium and selection of tag-SNPs were performed using the software HaploView and the Tagger subroutine therein (Barrett et al. 2005; de Bakker et al. 2005). Haplotype analysis was performed using the Haplo.GLM subroutine of Haplo.Stats version 1.2 (Schaid et al. 2002) in the R statistical suite version 2.8.1 (R Development Core Team 2008). In this analysis, association tests for 5-SNP haplotypes with NDS and ODS were performed. Rare haplotypes occurring with ≤5 counts in the sample (corresponding to approximately 0.5%) were collapsed.
In models fitted for NDS, smoking phenotypes and ODS, we adjusted for potential confounding by age, sex, clinic type, household income, marital status, education status, and employment status. In models fitted for ODS, we also adjusted for pain level and opioid prescriptions received, given that the ODS in this sample of pain patients is likely to be influenced by the degree of pain and opioid usage. ODS, pack-years, and the number of attempts to quit smoking were log transformed to reduce variable skewness; for ODS, the results of un-transformed analysis are shown, which did not substantially differ from those of the transformed analysis.
For NDS and the various smoking phenotypes (Fig. 2), we show results for the saturated model (adjusted for age, sex, clinic type, low household income, marital status, education status, and employment status) per each phenotype. For ODS (Fig. 3), we show results of five models with stepwise addition of covariates in order to demonstrate that statistical significance of the genetic association increases as more residual confounding is accounted for.
DNA marker selection and genotyping
The approach to marker selection was a combination of agnostic LD tagging with consideration of prior evidence and functional annotation. First, we identified all association signals related to smoking, nicotine dependence, and lung cancer within 15q25.1 in the published literature. Next, we examined the LD structure of the region in the HapMap Caucasian sample (release 23a; http://www.hapmap.org) and, using the algorithm of Gabriel et al. (2002), estimated the boundaries of the LD block in which the majority of these association signals were located. The following settings were used in this estimation: confidence interval for strong LD set to 0.7–0.98; minimum for strong recombination set to 0.9; and minimum fraction of strong LD in informative comparisons set to 0.95. Finally, we searched dbSNP for non-synonymous polymorphisms occurring within this LD block and found one such polymorphism.
We then used the algorithm of de Bakker et al. implemented in the software HaploView v.4.0 (http://www.broad.mit.edu./mpg/haploview) (Barrett et al. 2005; de Bakker et al. 2005) to select five tag SNPs for the target block. Of these five markers, two were force-included (the non-synonymous marker in CHRNA5 [rs16969968] and the most significant signal for lung cancer located in CHRNA3 [rs1051730]). The software selected the remaining three SNPs freely. The resulting set of tag-SNPs spanned 25 kb, included rs16969968, rs660652, rs1051730, rs6495308, and rs12443170 and captured 90% of the common variation (MAF ≥ 10%) in the target block with r2 ≥ 0.75 in HapMap Caucasians.
Marker specifications (NCBI build 36.3)
Physical location (bp)
Nicotine dependence severity
Number of attempts to quit
Smoking quantity (pack-years)
Opioid dependence severity
K to 8th
9 to 11th
Previous studies have suggested a role for 15q25.1 in nicotine, alcohol, and cocaine dependence; the results of this study expand the gamut of substances associated with this locus to include prescription opioids.
With few exceptions, studies that reported an association of 15q25.1 with nicotine dependence and/or smoking placed the main association signal on the CHRNA5 coding SNP rs16969968. Our results for ODS were in line with these studies and placed the main effect on rs16969968[A] as expected. However, our results with NDS and smoking failed to replicate this location and instead placed the association signal on rs660652[G]. We note that our study was designed to detect association with ODS and the sample was selected based on opioid use, which we now know is associated with rs16969968[A]. It is therefore, likely that the main effect of rs16969968[A] on NDS and smoking was masked in our sample, giving rise to an overestimation of a minor effect of rs660652[G], which might not be true in the general population.
In addition to ODS measured using the severity of dependence scale (SDS), we also tested for association of 15q25.1 with lifetime and current opioid dependence (OD) as defined by the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) (American Psychiatric Association 2000), and failed to detect such an association. This lack of detected association with DSM-IV OD may be due to lack of power, since with 185 (36.6%) individuals positive for DSM-IV OD the study had a type 2 error of 17%. More specifically, assuming a minor allele frequency of 0.35, an additive mode of inheritance, a response probability of 0.37 and an effect size of 1.5, our study had 83% power to detect an effect of 15q25.1 on DSM-IV OD if such an effect was truly present. However, the actual type 2 error may have been even greater than 17% if individuals with high opioid dependence were more likely not to participate in the study or to underreport OD symptoms because of stigmatization concerns. However, as we have reported elsewhere (Boscarino et al. 2010), a detailed analysis of study responders and non-responders using the electronic health record found no difference in the number of prescription opioid orders received over a 3-year period between these groups.
Another limitation has to do with the criticism that using the DSM-IV criteria for prescription OD may be inappropriate, since most patients taking these pain medicines experience tolerance and withdrawal as a side effect of this treatment. Thus, the DSM-IV OD criteria for medical use of these medicines may be biased. Due to these and other limitations with DSM-IV (Wu et al. 2010), the American Psychiatric Association is currently revising the criteria for prescription opioid dependence in DSM-V. In any case, our failure to detect association with DSM-IV OD in spite of having detected an association with an endophenotype thereof was somewhat surprising.
Sherva et al. (2010) have examined association of 15q25.1 with multiple substance dependence phenotypes in a sample of illicit drug users. They found nominally significant association of rs16969968 and other 15q25.1 SNPs with opioid dependence defined according to DSM-IV criteria; however, none of these association signals was significant after correction for multiple testing. In addition, as noted above, the focus of our study was on OD among “licit” prescription opioid users, not illicit users. As noted, DSM-IV is currently being revised to address the potential disparity between these phenotypes (Wu 2010). This measurement limitation may account for the lack of significance for DSM-IV OD in our study. Joslyn et al. (2008) described similar findings related to the association of 15q25.1 with alcohol use disorder and its endophenotype—the level of response to alcohol. They suggested that level-of-response phenotypes might be better correlated with biology than the DSM-IV construct of alcohol dependence, and therefore more likely to be useful for detecting genetic associations. DSM-IV criteria are designed to capture clinical endpoints, whereas endophenotypic traits typically capture underlying dimensions that are more etiologically homogeneous and possibly more useful for genetic epidemiological investigation.
Stimulation of neurons in the mesolimbic system by acetylcholine and its exogenous analogs via nicotinic acetylcholine receptors may be central to the development and reinforcement of substance dependence (Janhunen and Ahtee 2007). Multiple genetic association studies have previously demonstrated an association of DNA variation in the nAChR gene cluster on chromosome 15q25.1 with nicotine, alcohol, and cocaine dependence. This study is among the first to show an association of this locus with prescription opioid dependence. Additional studies are needed to evaluate the role of this and other nAChR gene loci in addiction neurobiology, treatment responses, and in public health.
This work was supported by a grant from the Administrative Committee for Research (ACR), Geisinger Clinic, Grant No. TRA-015 to Dr. Boscarino. Preliminary results for this study were presented at the 15th Annual HMO Research Network Conference, Danville, PA, April, 2009.
Conflict of interest
The authors have no conflicts of interest related to this research.