Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2013 Jun 1.
Published in final edited form as: Hum Genet. 2012 Apr 11;131(6):791–801. doi: 10.1007/s00439-012-1167-1

The interplay of genes and adolescent development in substance use disorders: Leveraging findings from GWAS meta-analyses to test developmental hypotheses about nicotine consumption

Scott I Vrieze 1, Matt McGue 1, William G Iacono 1
PMCID: PMC3407593  NIHMSID: NIHMS380825  PMID: 22492059

Abstract

The present study evaluated gene by development interaction in cigarettes smoked per day (CPD) in a longitudinal community-representative sample (N=3231) of Caucasian twins measured at ages 14, 17, 20, and 24. Biometric heritability analyses show strong heritabilities and shared environmental influences, as well as cross-age genetic and shared-environmental correlations. SNPs previously associated with CPD according to meta-analysis were summed to create a SNP score. At best, the SNP score accounted for 1% of the variance in CPD. The results suggest developmental moderation with a larger significant SNP score effect on CPD at age 20 and 24, and smaller non-significant effect at age 14 and 17. These results are consistent with the notion that nicotine-specific genetic substance use risk is less important at younger ages, and becomes more important as individuals age into adulthood. In a complementary analysis, the same nicotine-relevant SNP score was unrelated to the frequency of alcohol use at ages 14, 17, 20, or 24. These results indicate that the SNP score is specific to nicotine in this small sample and that increased exposure to nicotine at ages 20 and 24 does not influence the extent of concurrent or later alcohol use. Increased sample sizes and replication or meta-analysis are necessary to confirm these results. The methods and results illustrate the importance and difficulty of considering developmental processes in understanding the interplay of genes and environment.

Introduction

Main effects of individual common variants with associations to complex diseases have been elusive (Manolio, et al., 2009). It appears that individual samples do not possess sufficient power to detect the vast majority of common single nucleotide polymorphisms (SNPs) relevant to a given complex trait of interest. To overcome this hurdle large consortia have coordinated combined analysis over multiple studies with total sample sizes in the tens to hundreds of thousands. These massive samples have produced promising results for a number of complex traits. Perhaps the best example is with human height, where the GIANT consortium has amassed a total sample size of 183,727 Caucasians (Allen, et al., 2010). Consortia have arisen for other traits as well, such as lipid levels (Willer, et al., 2008), body mass index (Speliotes, et al., 2010) and, most important for our purposes, tobacco use (Furberg, et al., 2010; Thorgeirsson, et al., 2010).

The need for huge sample sizes is in part due to a combination of small effects and the large number of tests conducted in a GWAS (e.g., see (Burton, et al., 2009). The problem is compounded when studying developmental or environmental moderators of genetic effects (i.e., GxE or GxD designs). Gene x Environment-Wide Interaction Studies (GEWIS), for example, require at least as many tests as GWAS and much larger sample sizes to ensure adequate power for what is expected to be small GxE effects (e.g., sample sizes four times that for GWAS (Manolio, Bailey-Wilson, & Collins, 2006; Thomas, 2010a, 2010b)). There is presently little reason to believe that GEWIS would be more successful than GWAS have been (Caspi, Hariri, Holmes, Uher, & Moffitt, 2010). Add to this the difficulty of finding existing studies with large sample sizes that possess commensurate phenotypic and environmental measures such that meta-analysis or harmonization is difficult to impossible, and the success of exploratory GEWIS becomes less promising. Unless an a priori GxE hypothesis is extremely strong or the environmental effect is severe and/or uncommon (Caspi, et al., 2010), it is preferable to filter SNPs from genome-wide arrays according to criteria that are likely to result in a subset of SNPs that have promise of environmental interaction (Sullivan, et al., 2009). GxE tests can then commence on the presumably much smaller subset of filtered SNPs.

One simple filter is using only SNPs that have shown a main effect (Sullivan, et al., 2009; Thomas, 2010a, 2010b). SNPs with main effects are more likely than other SNPs to be relevant to disease in the first place and thus also more likely to be differentially relevant conditional on environmental experience. Other potentially useful filtering mechanisms include testing for variance differences conditional on genotype (Pare, Cook, Ridker, & Chasman, 2010) as well as selecting SNPs associated with differences between monozygotic twins (Elashoff, Cantor, & Shain, 1991; Magnus, Berg, Borresen, & Nance, 1981). After filtering, one is left with a subset of SNPs probabilistically enriched for GxE interactions. Even better, filtering could be conducted in one sample (or a meta-sample) and the GxE interaction tested in a separate sample, to avoid capitalization on chance such as when the filtering mechanism and the GxE testing mechanism are non-independent but performed within the same sample.

While there is less literature on gene x development interactions (GxD), they are conceptually similar to GxE interactions and may suffer many of the same pitfalls described above. However, GxD also presents unique challenges. First, as in GxE, there is the need to filter genome-wide arrays for SNPs or other variants that have promise to show a developmental trend (e.g., use SNPs with demonstrated main effect). Second, longitudinal studies are uncommon (relative to cross-sectional studies), expensive, and often span no more than a few assessments over a few years. Obtaining or combining longitudinal sample sizes large enough to detect novel GxD interactions against the background of a million or more common SNPs is probably not achievable at this time. Third and related, using a cross-sectional approach (e.g., two samples that are of different ages) introduces cohort effects that confound any GxD test but, more importantly, they do not allow for the measure of individual change. That is, a cross-sectional approach with two samples of different ages would allow between-group tests but not change at the individual level (Curran & Bauer, 2011). Longitudinal studies are preferred, but they require specialized statistical methods that account for developmental change nested within subject (for a review of methods in longitudinal analysis of behavior, see (Curran & Bauer, 2011; Gibbons, Hedeker, & DuToit, 2010))

Development of Nicotine Use and Addiction

Development and environment are key components of nicotine addiction. As to development, adolescence is a critical time for smoking onset. Most smokers begin smoking in their teenage years and about 90% of smokers express regret about ever starting (Fong, et al., 2004). Smoking-related deaths are top causes of morbidity (Center for Disease Control and Prevention, 2005). Those who have not smoked by age 19 are unlikely to become life-long smokers, and have much lower rates of smoking-related morbidity (Curry, Mermelstein, & Sporer, 2009). On the environmental side, remarkable shifts in public policy have contributed to a stark decrease in smoking among adolescents (Monitoring the Future, 2007; Nelson, et al., 1995). This change has been attributed to increased knowledge about harmful effects of smoking, government informational campaigns, advertisement bans, pack warnings, cigarette taxes, severe restrictions on where tobacco can be used, and more (Cummings, Fong, & Borland, 2009).

Individuals who never use cigarettes will never be addicted to them. This simple fact results in a host of complications for the longitudinal measurement of developmentally-specific genetic influences. One complication is that initiation, regular use, and addiction are all moderately to highly heritable, and are also genetically related (Maes, et al., 2004), indicating that some genetic variants are relevant to both initiation of tobacco use as well as eventual addiction. Indeed, using nicotine is also correlated with alcohol dependence, marijuana dependence, and hard drug dependence (Kendler, Jacobson, Prescott, & Neale, 2003; Kendler, Prescott, Myers, & Neale, 2003), and these correlations are highest in adolescence and subside during the transition to adulthood (Vrieze, Hicks, Iacono, & McGue, submitted). A theoretical account of this pattern of relationships (Iacono, Malone, & McGue, 2008) argues that behavioral disinhibition (e.g., impulsivity, risk-taking) is partly responsible for the observed correlations among cigarette initiation, use and dependence and, more broadly, among substance use disorders. Adolescents initiate nicotine, alcohol, and drug use in part because they tend to act impulsively. Those who act more impulsively will be more likely to experiment with cigarettes and, once regular users, more likely to smoke more often, thus consuming more cigarettes per day.

While theoretically important, the disinhibitory hypothesis in no way excludes other mechanisms of addiction. For example, some individuals self-medicate with alcohol in response to emotional disturbance (Sareen, Bolton, Cox, & Clara, 2006). Individual variation exists in the structure and function of neurotransmitter systems integral to substance use and addiction, such as dopamine, serotonin, and GABA. In cigarette smoking, thanks to large-scale consortia, there are verified genetic influences on biological systems that influence the number of cigarettes smoked per day (Thorgeirsson, et al., 2010). The meta-analysis used to guide SNP selection in the present study was conducted on a discovery sample of 31,266 smokers from the ENGAGE consortium with replication in 45,691 smokers from the Glaxo Smith Kline (Ox-GSK) and Tobacco and Genetics (TAG) consortia. While many SNPs approached genome-wide significance, and are reported in supplementary materials, the authors report three genome-wide significant hits that include SNPs in genes CYP2A6 and CYP2B6 that encode nicotine-metabolizing enzymes (Ray, Lerman, & Tyndale, 2009) as well as in genes that code for nicotinic acetylcholine receptor subunits (CHRNB3 and CHRNA6).

While these genetic association findings (Furberg, et al., 2010; Thorgeirsson, et al., 2010) are clearly important in understanding the genetic and neurological etiology of nicotine use and addiction, they were obtained in a large heterogeneous sample of adults from multiple studies. We hypothesize that the relative impact of genetic risk is moderated by development, such that children and adolescents are driven to smoke less by nicotine-specific genetic risk, but more by impulsivity/disinhibitory processes. As individuals who experiment with substances age and mature, we expect nicotine-specific processes to become more and more important in addiction. Indeed, (Thorgeirsson, et al., 2010) found evidence of this, if only because they did not find considerable overlap between SNPs associated with cigarettes smoked per day and SNPs associated with smoking initiation. In the present study we test the association of the aggregate effect of SNPs identified in (Thorgeirsson, et al., 2010) with CPD at ages 14, 17, 20, and 24, providing a direct assessment of the extent to which the relationship is moderated by age.

Furthermore, the meta-analysis lacked tests of discriminant validity, that is, it did not test whether these nicotine-related hits were also relevant for other substance use disorders, or what the mechanisms of that action may be. In the present study we evaluate the relationship between the nicotine SNPs identified by (Thorgeirsson, et al., 2010) and alcohol use. There are multiple etiological pathways that would result in an association with nicotine SNPs and alcohol use; we discuss two. First, it may be that the SNPs are not etiologically specific for smoking, but rather are relevant for more pervasive behavioral systems that impact general risk for substance use like disinhibition or impulsivity. Second, if a genotype is related to nicotine use then this, by definition, indicates that an individual with the risk genotype will experience increased environmental exposure to nicotine. That is, the genotype causes smoking, but it also causes the environment of nicotine intake. Some have theorized that environmental exposure to drugs and alcohol acts as a “gateway,” in which the experience of drug effects cause further use of other, possibly “harder” drugs (e.g., see (Kandel & Jessor, 2002)). The present study tests whether these known nicotine SNPs are related also to alcohol use. A positive association would tend to support a gateway-type effect of nicotine exposure on alcohol use, especially if a more generalized risk etiology (e.g., disinhibition) can be ruled out. At first blush the method appears paradoxical in that genetic variants, here SNPs, are actually used as proxies for environmental exposure. This approach is called Mendelian randomization (Smith & Ebrahim, 2003, 2005), and has been successfully applied to address environmental effects on metabolic syndrome (Timpson, et al., 2005), drug use (Irons, Iacono, Oetting, & McGue, 2012; Irons, McGue, Iacono, & Oetting, 2007), and other diseases (Schatzkin, et al., 2009).

Method

Sample

Participants (N = 3231, 52% female) were drawn from the Minnesota Twin Family Study (MTFS), a community-representative longitudinal study of Minnesota families. The study design has been extensively described elsewhere (Iacono & McGue, 2002), and is an accelerated-longitudinal design composed of two cohorts. The younger cohort (N = 2084, 50% female) was first assessed at age 11 during the years 1988–2005. The older cohort was first assessed at age 17 (N = 1147, 55% female) during the years 1989–1996. Members of the younger, 11-year-old cohort were invited to participate in follow-up assessments at age 14 and 17, and twins from both cohorts were invited to participate in follow-up assessments at age 20 and 24, for a total of five assessments spanning 13 years. Cohorts were combined for all analyses. Participants received modest honoraria for their assessments and written assent or consent was obtained from all participants, including the parents of minor children. From its inception, the study has been continuously approved by the University of Minnesota IRB.

Actual ages of assessment were near the target ages. Pooling across cohorts in the full sample the mean (SD) ages at assessment were 11.78 (0.43), 14.90 (0.54), 17.84 (0.67), 21.09 (0.81), 24.94 (0.90). Among twins eligible for follow-up assessment, rate of participation was 93.0% at age 14, 87.3% at age 17, 89.4% at age 20, 91.0% at age 24 and 93.6% at age 29.

Analyses were conducted on two samples: 1) the full sample and 2) a subsample of participants selected for current smoking. In the subsample we selected, for each age, only those individuals who smoked at that age. All non-smoker’s CPD values were set to missing, according to the rationale that we do not know how many cigarettes they would smoke, were they in fact a smoker. We selected the subsample of current smokers because the SNPs used in this study were identified by meta-analysis for their association with CPD in current smokers. Since we are developmentally extending the meta-analytic findings, we hewed to their ascertainment method as closely as possible.1 A participant was deemed a smoker if they smoked at least once per month for the 12 months prior to assessment. This threshold was used to maximize the sample size while only keeping those individuals who were currently smoking to some extent. Stricter thresholds, such as averaging 1 cigarette per day, changed the sample size only by 15–50 people, depending on the age of assessment, and had no impact on the overall findings from this report. A similar procedure was followed to identify current drinkers, with the criterion that they drank at least one alcoholic beverage per month in past 12 months.

Measures

Participants at the age-11 and age-14 assessments were assessed with a computerized questionnaire. The exact question was “During the past 12 months, about how many cigarettes did you smoke in a day when you did smoke?” Participants responded on a 7-point scale. Assessments at later ages were conducted in-person by extensively trained research assistants using a modified version of the expanded Substance Abuse Module (SAM; (Robins, Babor, & Cottler, 1987)). The exact question on the SAM was “During the past 12 months, how much have you smoked (tobacco have you used) on a typical day that you smoked (used tobacco)?” Respondents gave an integer number in response. The computerized questionnaire responses were converted to an integer number for concordance with the SAM. CPD measures were log transformed for all analyses (except descriptive analyses) to mitigate the effect of extreme values on parameter estimates. Frequency of drinking alcohol was also measured with a computerized questionnaire (“During the past 12 months, about how many times did you drink alcohol?”) at age 11 and 14 and the SAM (“In the past 12 months, how often on average have you drunk any alcohol (had any alcohol to drink)?”) at later ages. Participants responded on a 10-point scale that ranged from “Less than once a year” to “3 or more times a day.” There was considerably less skew and a much smaller range of responses for the measure of drinking frequency, and values thus were not log-transformed.

Genotyping and Imputation

SNPs were genotyped for all members of all families in the Minnesota Center for Twin and Family Research (N = 7438) on an Illumina 660quad array using DNA derived from whole blood for approximately 90% of the sample and from saliva samples for the remainder. For quality control purposes, each 96-well plate included DNA from two members of a single CEPH family (rotated across plates) and one duplicate sample. Markers were excluded if (see ref (Miller, et al., submitted) for additional details): 1) they had been identified as a poorly genotyped marker by Illumina; 2) had more than one mismatch in duplicated QC samples; 3) had a call rate < 99%; 4) had a MAF < 1%; 5) had more than 2 Mendelian inconsistencies across families; 6) significantly deviated from Hardy-Weinberg equilibrium at p < 1e-7; 7) was an autosomal marker but associated with sex at p < 1e-7; 8) had a significant batch effect at p < 1e-7; or 8) there were more than 2 heterozygous X chromosome calls for males or mitochondrial calls for anyone. In total, 32,153 (5.7%) of the 559,982 SNP markers were eliminated by these screens, with the majority (3.6%) being eliminated because of low MAF. Samples were eliminated if: 1) they had > 5000 no-calls; 2) had a low GenCall score; 3) had extreme heterozygosity or homozygosity; or 4) represented a sample mix-up or we could not confirm known genetic relationships. In total, 160 (2.2%) of the total genotyped sample of 7438 failed one or more of these criteria, with the majority (1.7%) failing because of low call rate. Of the 3672 twins initially enrolled in the study, 3365 provided usable genotypes, of which 3231 were Caucasian and used for the present analysis. Only those individuals with genotypes were used in the present study analyses. Those twins with genotypes versus those without did not differ significantly on measures of nicotine use (CPD, days smoked per month, or DSM-IIIR nicotine dependence symptomatology) or alcohol use, with the exception that individuals with genotypes drank slightly more at age 24 (Cohen’s d = .22, p = .01).

Of the 92 SNPs used in the SNP score, 28 were available on the Illumina 660W quad array used in the full sample. The remaining SNPs were imputed with MaCH (Li, Willer, Sanna, & Abecasis, 2009; Li, Willer, Ding, Scheet, & Abecasis, 2010) based on the August 2010 reference haplotypes, each with satisfactory imputation quality (mean r2 = .92; median = .96; SD = .11; range = [0.52, 1.0]).

SNPs used in the molecular genetic analysis were selected from supplemental materials reported in (Thorgeirsson, et al., 2010). Only SNPs in regions of interest were reported; that is, regions that harbored promising SNPS, including all genome-wide significant SNPs, based on the full meta-sample were reported. All SNPs with associations reported for cigarettes smoked per day were used in the present analysis. The SNP set was winnowed on the basis of linkage disequilibrium in an iterative fashion. The most significant SNP was selected, and all SNPs within r2 > .7 of that SNP were removed from the SNP set. Then the next most significant SNP in the reduced dataset was evaluated, and SNPs with r2 > .7 were removed, and so on. We chose a liberal r2 because our initial SNP set was rather small (602), and we wanted to exclude only the most highly-redundant SNPs. This process retained 92 of the 602 SNPs originally reported in the supplemental materials. For all analyses, these 92 SNPs were summed according to their meta-analytic regression weights to form a genetic risk score for tobacco use. The final 92 SNPs are listed in Supplemental Table 1.

Longitudinal Heritability Analysis

The longitudinal model employed was saturated, in that all across-time CPD covariances were simultaneously estimated. Within-family clustering due to twin status was accounted for with standard twin modeling (Neale & Cardon, 1992). This technique decomposes the observed CPD phenotypic variance-covariance matrix into additive genetic variance (A), shared environmental variance (C), and unshared environmental variance (E). Because so little variance is observed for CPD at age 11, our longitudinal models only included measurements at age 14, 17, 20, and 24. All heritability models were fit with OpenMx 1.0.6 (Boker, et al., 2011) of the R Environment version 2.13.1 (R Development Core Team, 2011). Confidence intervals were computed using non-linear constraints and likelihood ratio tests according to standard practice (Miller & Neale, 1995; Neale & Miller, 1997). All models were fit under full information maximum likelihood to account for missing data. Covariates used in these analyses were year of birth, sex, and cohort status. Covariates were incorporated as fixed effects as described in the next section.

Incorporating the SNP Score in the Longitudinal Design

The genetic SNP score can be incorporated directly into the ACE model described above as a fixed effect. This approach is equivalent to multivariate linear regression except that the model accounts for clustering within families (twin pairs are nested within families) and within subjects over time due to the repeated measurements. Resulting regression coefficient estimates and standard errors are corrected for the non-independent observations. The overall model is a special case of a mixed effects regression (Pinheiro & Bates, 2000), where the variance components are random effects and the SNP score and covariates are fixed effects, with the difference being that we used a variance-covariance matrix structured to account for twin zygosity.

Year of birth, cohort, sex, age of assessment, and the first ten genetic principal components were used as covariates in all SNP score association tests. Genetic principal components account for ancestry effects and were based on a subsample of 10,000 SNPs from sample founders (i.e., unrelated subjects) and computed in Eigenstrat (Price, et al., 2006). Scripts are available upon request.

Mendelian Randomization

Mendelian Randomization is a technique by which one can use genetic variation as a proxy for environmental exposure, and has been the subject of several reviews (Ebrahim & Smith, 2008; Smith, 2011). In the present study we use genetic variation associated with nicotine use as a proxy for the environmental exposure to nicotine. That is, individuals with high-risk genetic variation will be more likely to smoke more cigarettes. If the extent of nicotine use causes later alcohol use (a plausible prediction to be made from the gateway model), then we expect these nicotine-relevant variants to also be associated with alcohol use, even though the strength of the relationship may be attenuated to the extent expected from the less than perfect correlation between alcohol and nicotine use. All alcohol analyses were conducted in the same way as described for nicotine above.

Results

Descriptive Statistics

Descriptive statistics are provided in Table 1 for both the full sample and the subsample of current smokers, including sample sizes, number of twin pairs, and mean (SD) of CPD. In the full sample there are around 1000 or more participants assessed at each of the ages. In the subsample the number of participants is reduced considerably, especially at younger ages. At older ages the samples of current smokers are much larger, comprising 1/2 to 3/4 of the full sample. Our criteria for current smoking was intended to be inclusive, and to capture all individuals who were smoking cigarettes even to only a small extent (e.g., smoking on average one cigarette per month) during the assessment period. Individual trends are reported graphically in Figure 1, along with a running mean and standard deviation (based on overlapping windows of 500 subjects). Very few 11-year-olds smoke to any extent. Some individuals dramatically increase their intake at age 14, but again the vast majority uses no cigarettes (the median at all young ages is zero). By the age 17 assessment the mean CPD increases to around 2 for females and 3 for males. By age 20 and 24 the mean is around 4 for females and 6 for males. As observed in the figure, the standard deviation of CPD increases with the mean, a consequence of the fact that most individuals use zero to one CPD regardless of age. The “dip” at around age 17, visible for males and females, appears to represent the switch from computerized assessment of CPD to in-person interview assessment. Another possibility is that the dip is due to a cohort effect, as the older cohort begins their assessment at age 17, and may have been using cigarettes to a less extent than the younger cohort, resulting in the observed dip. Plotting a running mean only on the younger cohort, excluding the older cohort, yields the same age-17 dip in CPD, and suggests that the decline in CPD is due to the change in assessment format.

Table 1.

Descriptive Statistics.

Sample Measure Age of Assessment
Sex 11 14 17 20 24
Full Sample N M 1041 991 1269 999 1053
F 1043 999 1485 1220 1121
Full MZ pairs M 337 319 409 319 329
F 325 310 465 379 339
Full DZ pairs M 170 160 194 146 161
F 184 175 250 209 183
CPD Mean (SD) M 0.14 (.50) 1.03 (2.75) 3.41 (6.31) 6.12 (8.03) 6.13 (8.18)
F 0.07 (.39) 0.80 (2.74) 2.22 (4.87) 3.68 (6.05) 3.19 (5.61)
Drinking Frequency Mean (SD) M 0.08 (0.34) 0.69 (1.27) 2.14 (1.77) 4.05 (1.66) 4.37 (1.59)
F .03 (.20) 0.65 (1.23) 1.84 (1.70) 3.28 (1.61) 3.67 (1.53)
Subsample of Current Smokers N M 125 367 709 766 793
F 57 283 625 681 545
Full MZ pairs M 27 100 188 213 204
F 6 67 137 143 108
Full DZ pairs M 7 32 77 101 111
F 10 34 84 96 64
CPD Mean (SD) M 1.16 (0.95) 2.79 (3.94) 6.10 (7.40) 7.98 (8.32) 8.13 (8.51)
F 1.21 (1.21) 2.81 (4.56) 5.27 (6.35) 6.60 (6.80) 6.57 (6.53)
Subsample of Current Drinkers N M 12 184 798 916 988
F 3 199 840 1075 1025
Full MZ pairs M 0 37 212 277 299
F 0 34 207 309 289
Full DZ pairs M 0 8 88 129 145
F 0 22 118 170 158
Drinking Frequency Mean (SD) M 2.42 (0.67) 3.01 (1.20) 3.25 (1.26) 4.38 (1.29) 4.62 (1.27)
F 2.33 (0.58) 2.82 (1.13) 3.06 (1.23) 3.67 (1.27) 3.95 (1.28)
Open in a new tab

CPD = Cigarettes smoked per day

Figure 1.

Figure 1

Open in a new tab

Developmental Trajectory of Cigarettes Smoked per Day. The red line is a running mean. The green line is a running standard deviation. The dip at approximately age 17 is when the assessment protocol switched from a computerized questionnaire to an in-person interview.

Twin Heritability Model Results

Variance components from the biometric twin model are reported in Figure 2 as a series of correlation matrices with superimposed heatmaps. The phenotypic correlation matrix is denoted “P.” As the participants age, CPD becomes more stable over time. The correlation between age 14 and age 17 CPD is .55; the correlation between age 20 and age 24 CPD is .78. Note that the A, C, and E matrices are scaled such that the A, C, and E matrix, if summed element-wise, give the P matrix. The “A” matrix denotes that portion of the phenotypic matrix due to the additive effect of genes. It follows the same pattern as the phenotypic matrix, with genetic effects becoming more stable over time. Shared environmental effects are stronger at younger ages and generally taper over time. Unshared environmental effects contribute largely only to variance at assessments, but also contributes increasingly to stability of CPD across assessments, perhaps as a result of the highly addictive nature of cigarette smoking.

Figure 2.

Figure 2

Open in a new tab

Heatmap of Correlation Matrix and Heritability Components of CPD for the Full Sample and Subsample of Current Smokers. The figure is composed of two columns—the full sample is on the left and the subsample of current smokers is on the right. The upper two 4×4 matrices denoted “P” are the total phenotypic correlation matrices for each sample. Below that are the additive genetic (A), shared environmental (C), and unshared environmental (E) components for each sample. The matrices are scaled such that element-wise summation of the A, C, and E matrices gives the phenotypic (P) correlation matrix. This scaling gives the CPD heritabilities and environmental effects along the diagonal of the A, C, and E matrices (e.g., CPD is 31% heritable at age 14 in the full sample).

Including the SNP score in the Model

The SNP score was tested in both the full sample and the subset of current smokers. The score effect was non-significant at each age of assessment for CPD in the full sample. Results are not shown here for lack of space.

The SNP score did, however, show significant effects in the subsample of current smokers, consistent with the ascertainment method used by the meta-analysis from which the SNP score was derived. Results are given in Table 2. The effect was strongest for later ages with regression coefficients of .021 and .014 at ages 20 and 24, accounting for 1.0% of the variance at age 20 and 0.4% at age 24. While the score was not significant for younger ages, the regression coefficients were in the same direction but of a smaller magnitude (.010 and .009 at age 14 and 17). The SNP score at age 14 and 17 accounted for 0.1% to 0.4% of the variance, respectively, suggesting an attenuation of effect instead of the lack of effect. A composite test of the null hypothesis that the regression coefficient was the same at all ages approached statistical significance (Χ2 = 6.2, df =3, p = .10).

Table 2.

SNP Score Effects on CPD and Alcohol Use Frequency by Age. LRT is the likelihood ratio test statistic, distributed as Χ2 on 1 degree of freedom, of the model with the score effect coefficient estimated, versus the model where it is fixed to zero. Measures of nicotine use were log-transformed for this analysis. Coefficients are unstandardized regression weights, with negative values indicating decreased risk for substance use.

Drug Age N Coefficient 95% C.I. LRT p-value Variance Accounted for by Score
Nicotine 14 650 .010 (−.002, .023) 2.64 .10 0.38%
17 1334 .009 (−.003, .022) 2.28 .13 0.08%
20 1447 .021 (.009, .034) 10.86 .001 0.97%
24 1338 .014 (.001, .023) 4.15 .04 0.38%
Alcohol 14 383 −.003 (−.010, .004) 0.60 .44 0.16%
17 1638 .002 (−.002, .006) 0.84 .36 0.10%
20 1991 −.001 (−.004, .003) 0.27 .61 0.03%
24 2013 −.004 (−.007, −.001) 5.57 .02 0.37%
Open in a new tab

In addition to accounting for significant variance at age 20 and 24, the SNP score also accounted for significant covariance (1.0%, p < .05, as tested by 200 bootstrap replications) between CPD at age 20 and 24, indicating that the SNP score is not only associated with CPD at 20 and 24, but also associated with stability in CPD for those individuals smoking between ages 20 and 24.

Also listed in Table 2 is the SNP score relationship with frequency of alcohol use in the subsample of current alcohol users. As is clear from the table, the SNP score was unrelated to frequency of alcohol use at ages 14, 17, and 20. It was, however, significantly associated at age 24, but in the opposite direction as that observed for nicotine use. That is, at age 24 higher SNP scores were associated with greater CPD in smokers but less frequent drinking in drinkers. The SNP score also showed no association with alcohol use in the full sample or subsample of smokers (results not shown).

Discussion

This report is among the first to place SNP main effects identified by meta-analysis in a longitudinal developmental context (see also, (Vrieze, et al., 2011)). This approach can refine the nature of the meta-analytic association, and allow novel tests of existing addiction theory. We obtained three important findings. First, CPD is moderately heritable (30% to 50%) in the full sample and the subsample of current smokers. Genes account for a majority of the stability of smoking across time, especially at later ages. Shared environment accounts for the majority of stability from age 14 to age 17, indicating the potential for significant gene and environmental main effects moderated by development in CPD.

Second, the SNP score was not statistically significantly associated with cigarettes smoked per day at age 14 or 17, but was significantly associated at age 20 and 24. If replicable, this indicates that these genetic variants are less relevant during adolescence, when a large number of individuals begin experimenting with cigarettes and/or regularly using cigarettes. Differences in effect sizes at the different ages approached statistical significance, but requires replication. This suggests the possibility of a GxD effect, and that other mechanisms, such as behavioral disinhibition, play significantly larger roles during adolescence with CPD, consistent with literature cited in the introduction. In addition to significant associations with CPD at age 20 and 24, the longitudinal study allowed us to test whether the score accounted for stability in CPD across age. The score accounted for about 1% of the covariance between age 20 and age 24 smoking, and is thus relevant to the maintenance of CPD over this four year interval.

Third, the SNP score was not significantly related to alcohol use at ages 14, 17, and 20, indicating that these SNPs, even though they increase the likelihood of smoking at ages 20 and 24, play no detectable role in alcohol use. The sensitivity of this analysis is limited by the relatively small genetic effect on nicotine use (r2 < 1%) as well as the small sample size, and requires confirmation in larger samples. While the present results do not rule out the existence of a positive effect of environmental nicotine exposure on alcohol use, such a hypothesized effect is too small to be detected in this sample. There was a significant SNP score effect on alcohol use at age 24, but in the opposite direction as that observed for cigarettes. This result was unexpected and while possibly spurious, nevertheless fails to support the notion that the nicotine SNP score increase risk for alcohol use. One might conjecture that it represents a trend for early adults to specialize in their drug use habits and use available resources (e.g., money, time) for drugs of choice at the expense of other substances. However, in an analysis of the SNP score’s relationship with alcohol use frequency in the subsample of current smokers, there was no significant effect at age 24 (regression coefficient = −0.018, Χ2 = 1.3, df = 1, p = 0.25), suggesting that the effect in general may well be spurious.

The results of this study are preliminary, and require replication and extension, but one can begin to see the value of evaluating genetic effects on substance use in a developmental and environmental context. As sample sizes grow, and investigators begin sharing data at higher rates, tests of developmental and environmental moderation become more powerful and feasible. Mendelian randomization tests offer an elegant quasi-experimental way to test the causal role of an environmental exposure, and will become a mainstay in quasi-experimental research.

The novelty of this study requires a number of cautionary remarks. As noted in the introduction, GxD is conceptually similar to GxE, and likely shares many of the same pitfalls. One major hurdle is filtering candidate SNPs that show potential for GxD, and this is where the current study derives much of its strength. The longitudinal sample used in this report is quite small by GWAS standards, and could not by itself be used profitably to discover reliable GxD interactions against a background of 1 million SNPs. By leveraging results from the meta-analytic literature, which used tens of thousands of subjects, we were able to focus our GxD design only on highly promising SNPs that have high potential for replication.

Even by leveraging meta-analytically identified SNPs, the present sample was not large enough to identify developmental changes in individual SNP effects but, rather, SNPs had to be combined into a single genetic risk score before the effect was large enough to be detectable and testable in this sample. This score appears to be biologically coherent, in that the effects tagged by the SNPs are relevant to either nicotine metabolism or nicotine transmission. Future work may do better by combining scores within explicit biological pathways through the use of pathway databases (Vink, et al., 2009). In this way, genetic scores take on more refined biological meaning and results provide clearer theoretical import. A major challenge to the pathway approach is simply in identifying enough SNPs or other genetic variants within a pathway that have measureable effects on the phenotype. While non-trivial and etiologically informative, it is very unlikely that this SNP score could be utilized in any way in a personalized medicine context. There are many theories as to why no sizable genetic effects have been discovered for complex traits (Manolio, et al., 2009). Among these are the possibility that complex traits arise from polygenic inheritance where large numbers of common SNPs have individually small effects but account for considerable variance in aggregate (e.g., see (Allen, et al., 2010; Visscher, Yang, & Goddard, 2010; Yang, et al., 2010)). Other possibilities include rare variants that may have individually large effects, such as rare nonsense or missense SNPs or copy-number variants such as insertions/deletions, or variants related to gene expression and regulation (e.g., see (Altshuler, et al., 2010; Montgomery, Lappalainen, Gutierrez-Arcelus, & Dermitzakis, 2011). Undoubtedly, future work will be able to combine common tag SNPs from genome-wide arrays with rare and structural variation from sequencing into much more statistically powerful and biologically interpretable risk scores.

A limitation of our approach is that it relies on consortia to produce candidate SNPs for developmental analysis. This ignores the possibility of developmentally-relevant SNPs that result in similar adult status. To use a concrete example, there may exist powerful genetic variants that affect the onset of the pubertal growth spurt, and thus affect the shape of an individual’s growth trajectory, but have little or no influence on their eventual adult height. This SNP would never be identified by meta-analyses of adult height, but is extremely important in understanding how growth culminates in adult height. Just as in the GWAS literature, consortia of longitudinal studies are required, and individual investigators must share their data in order to achieve large enough sample sizes for more meaningful comparisons.

Substance use development is confounded with environmental change during adolescence. As youths age it is arguable that their home environment changes to accommodate an increased expectation of autonomy, and increased needs for independence. In the United States, for example, one can legally drive an automobile at age 16, legally purchase tobacco products at 18, and legally purchase alcohol at 21, all of which are within the time interval under investigation in this study. While a caregiver may restrict automobile or substance use, it is clear that developmental maturation can be confounded with dramatic environmental shifts in substance availability and social norms surrounding use. In fact, this observation can easily explain the current findings. Adolescents at age 14 or 17 may have a genotype that increases their CPD but are unable to satisfy that risk because cigarettes are not readily available to them, thus the SNP score’s effect is moderated not by developmental neurological or biological maturation, but by a developmentally-confounded environment (i.e., development serves as a proxy for environmental change). Add to this the extra complication that individuals with riskier genotypes are more likely to have biological parents with riskier genotypes, who will smoke more than those who do not (i.e., gene-environment correlation (Jaffee & Price, 2007)), and one can appreciate the challenges we face in understanding how genes and environment affect the development of addiction.

Summary

The present study evaluated the developmental moderation of cigarettes smoked per day by a SNP score derived from a meta-analysis of current smokers. The results suggest developmental moderation with an increased effect at age 20 and 24, and decreased effect at age 14 and 17. These results are consistent with the notion that nicotine-specific substance use risk is less important at younger ages, and becomes more important as individuals age into adulthood. The SNP score was unrelated to alcohol use at any age, indicating that the SNP score effects are specific to nicotine, and do not tap more general etiological processes simultaneously relevant to nicotine and alcohol use.

Supplementary Material

1

Acknowledgments

This work was supported by grants R37 DA 05147, R01 DA 13240, and U01 DA 024417 of the National Institute on Drug Abuse; R01 AA 09367 of the National Institute on Alcohol Abuse and Alcoholism; and 5T32 MH 017069 (Vrieze) of the National Institute of Mental Health.

Footnotes

1

Note that we also conducted all analyses with a subsample of individuals who, at any one assessment where either 1) a current smoker or 2) had been a smoker in the past. Current smoker was defined just as in the current smoking sample described in the text. To be a past smoker we required that an individual report smoking on average one cigarette per day for 12 months. The results from this sample did not change in any significant way.

References

  1. Allen HL, Estrada K, Lettre G, Berndt SI, Weedon MN, Rivadeneira F, et al. Hundreds of variants clustered in genomic loci and biological pathways affect human height. Nature. 2010;467(7317):832–838. doi: 10.1038/nature09410. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Altshuler DL, Durbin RM, Abecasis GR, Bentley DR, Chakravarti A, Clark AG, et al. A map of human genome variation from population-scale sequencing. Nature. 2010;467(7319):1061–1073. doi: 10.1038/nature09534. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Boker S, Neale M, Maes H, Wilde M, Spiegel M, Brick T, et al. OpenMx: An Open Source Extended Structural Equation Modeling Framework. Psychometrika. 2011;76(2):306–317. doi: 10.1007/s11336-010-9200-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Burton PR, Hansell AL, Fortier I, Manolio TA, Khoury MJ, Little J, et al. Size matters: just how big is BIG? Quantifying realistic sample size requirements for human genome epidemiology. International Journal of Epidemiology. 2009;38(1):263–273. doi: 10.1093/ije/dyn147. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Caspi A, Hariri AR, Holmes A, Uher R, Moffitt TE. Genetic Sensitivity to the Environment: The Case of the Serotonin Transporter Gene and Its Implications for Studying Complex Diseases and Traits. American Journal of Psychiatry. 2010;167(5):509–527. doi: 10.1176/appi.ajp.2010.09101452. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Center for Disease Control and Prevention. Annual Smoking-attributable mortality, years of potential life lost, and productivity losses. Morbidity and mortality weekly report. 2005;54(25):3. [PubMed] [Google Scholar]
  7. Cummings KM, Fong GT, Borland R. Environmental Influences on Tobacco Use: Evidence from Societal and Community Influences on Tobacco Use and Dependence. Annual Review of Clinical Psychology. 2009;5:433–458. doi: 10.1146/annurev.clinpsy.032408.153607. [DOI] [PubMed] [Google Scholar]
  8. Curran PJ, Bauer DJ. The Disaggregation of Within-Person and Between-Person Effects in Longitudinal Models of Change. Annual Review of Psychology. 2011;62:583–619. doi: 10.1146/annurev.psych.093008.100356. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Curry SJ, Mermelstein RJ, Sporer AK. Therapy for Specific Problems: Youth Tobacco Cessation. Annual Review of Psychology. 2009;60:229–255. doi: 10.1146/annurev.psych.60.110707.163659. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Ebrahim S, Smith GD. Mendelian randomization: can genetic epidemiology help redress the failures of observational epidemiology? Human Genetics. 2008;123(1):15–33. doi: 10.1007/s00439-007-0448-6. [DOI] [PubMed] [Google Scholar]
  11. Elashoff JD, Cantor RM, Shain S. Power and Validity of Methods to Identify Variability Genes. Genetic Epidemiology. 1991;8(6):381–388. doi: 10.1002/gepi.1370080604. [DOI] [PubMed] [Google Scholar]
  12. Fong GT, Hammond D, Laux FL, Zanna MP, Cummings KM, Borland R, et al. The near-universal experience of regret among smokers in four countries: Findings from the International Tobacco Control Policy Evaluation Survey. Nicotine & Tobacco Research. 2004;6:341–351. doi: 10.1080/14622200412331320743. [DOI] [PubMed] [Google Scholar]
  13. Furberg H, Kim Y, Dackor J, Boerwinkle E, Franceschini N, Ardissino D, et al. Genome-wide meta-analyses identify multiple loci associated with smoking behavior. Nature Genetics. 2010;42(5):441–U134. doi: 10.1038/ng.571. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Gibbons RD, Hedeker D, DuToit S. Advances in Analysis of Longitudinal Data. Annual Review of Clinical Psychology. 2010;6:79–107. doi: 10.1146/annurev.clinpsy.032408.153550. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Iacono WG, Malone SM, McGue M. Behavioral disinhibition and the development of early-onset addiction: Common and specific influences. Annual Review of Clinical Psychology. 2008;4:325–348. doi: 10.1146/annurev.clinpsy.4.022007.141157. [DOI] [PubMed] [Google Scholar]
  16. Irons DE, Iacono WG, Oetting WS, McGue M. Developmental trajectory and environmental moderation of the effect of ALDH2 polymporphism on alcohol use. Alcoholism: Clinical and Experimental Research. 2012 doi: 10.1111/j.1530-0277.2012.01809.x. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Irons DE, McGue M, Iacono WG, Oetting WS. Mendelian randomization: A novel test of the gateway hypothesis and models of gene-environment interplay. Development and Psychopathology. 2007;19(4):1181–1195. doi: 10.1017/S0954579407000612. [DOI] [PubMed] [Google Scholar]
  18. Jaffee SR, Price TS. Gene-environment correlations: A review of the evidence and implications for prevention of mental illness. Molecular Psychiatry. 2007;1(1):1–11. doi: 10.1038/sj.mp.4001950. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Kandel DB, Jessor R. The gateway hypothesis revisited. In: Kandel DB, editor. Stages and Pathways of Drug Involvement. New York: Cambridge University Press; 2002. [Google Scholar]
  20. Kendler KS, Jacobson KC, Prescott CA, Neale MC. Specificity of genetic and environmental risk factors for use and abuse/dependence of cannabis, cocaine, hallucinogens, sedatives, stimulants, and opiates in male twins. American Journal of Psychiatry. 2003;160(4):687–695. doi: 10.1176/appi.ajp.160.4.687. [DOI] [PubMed] [Google Scholar]
  21. Kendler KS, Prescott CA, Myers J, Neale MC. The structure of genetic and environmental risk factors for common psychiatric and substance use disorders in men and women. Archives of General Psychiatry. 2003;60(9):929–937. doi: 10.1001/archpsyc.60.9.929. [DOI] [PubMed] [Google Scholar]
  22. Li Y, Willer C, Sanna S, Abecasis G. Genotype Imputation. Annual Review of Genomics and Human Genetics. 2009;10:387–406. doi: 10.1146/annurev.genom.9.081307.164242. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Li Y, Willer CJ, Ding J, Scheet P, Abecasis GR. MaCH: Using Sequence and Genotype Data to Estimate Haplotypes and Unobserved Genotypes. Genetic Epidemiology. 2010;34(8):816–834. doi: 10.1002/gepi.20533. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Maes HH, Sullivan PF, Bulik CM, Neale MC, Prescott CA, Eaves LJ, et al. A twin study of genetic and environmental influences on tobacco initiation, regular tobacco use and nicotine dependence. Psychological Medicine. 2004;34(7):1251–1261. doi: 10.1017/s0033291704002405. [DOI] [PubMed] [Google Scholar]
  25. Magnus P, Berg K, Borresen AL, Nance WE. Apparent Influence of Marker Genotypes on Variation in Serum-Cholesterol in Monozygotic Twins. Clinical Genetics. 1981;19(1):67–70. doi: 10.1111/j.1399-0004.1981.tb00670.x. [DOI] [PubMed] [Google Scholar]
  26. Manolio TA, Bailey-Wilson JE, Collins FS. Opinion - Genes, environment and the value of prospective cohort studies. Nature Reviews Genetics. 2006;7(10):812–820. doi: 10.1038/nrg1919. [DOI] [PubMed] [Google Scholar]
  27. Manolio TA, Collins FS, Cox NJ, Goldstein DB, Hindorff LA, Hunter DJ, et al. Finding the missing heritability of complex diseases. Nature. 2009;461(7265):747–753. doi: 10.1038/nature08494. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Miller MB, Basu S, Cunningham J, Oetting W, Schork NJ, Iacono WG, et al. The Minnesota Center for Twin and Family Research Genome-Wide Association Study. doi: 10.1017/thg.2012.62. submitted. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Miller MB, Neale MC. Using Nonlinear Constraints in Mx to Automate Estimation of Confidence-Intervals for Parameters in Genetic Models. Behavior Genetics. 1995;25(3):279–280. [Google Scholar]
  30. Monitoring the Future. Trends in cigarette smoking and smokeless tobacco. University of Michigan; 2007. [Google Scholar]
  31. Montgomery SB, Lappalainen T, Gutierrez-Arcelus M, Dermitzakis ET. Rare and common regulatory variation in population-scale sequenced human genomes. PLoS Genet. 2011;7(7):e1002144. doi: 10.1371/journal.pgen.1002144. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Neale MC, Cardon LR. Methodology for genetic studies of twins and families. Dortrecht, The Netherlands: Kluwer Academic; 1992. [Google Scholar]
  33. Neale MC, Miller MB. The use of likelihood-based confidence intervals in genetic models. Behavior Genetics. 1997;27:113–120. doi: 10.1023/a:1025681223921. [DOI] [PubMed] [Google Scholar]
  34. Nelson DE, Giovino GA, Shopland DR, Mowery PD, Mills SL, Eriksen MP. Trends in Cigarette-Smoking among Us Adolescents, 1974 through 1991. American Journal of Public Health. 1995;85(1):34–40. doi: 10.2105/ajph.85.1.34. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Pare G, Cook NR, Ridker PM, Chasman DI. On the Use of Variance per Genotype as a Tool to Identify Quantitative Trait Interaction Effects: A Report from the Women’s Genome Health Study. Plos Genetics. 2010;6(6) doi: 10.1371/journal.pgen.1000981. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Pinheiro JC, Bates DM. Mixed-effects models in S and S-PLUS. New York: Springer; 2000. [Google Scholar]
  37. Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA, Reich D. Principal components analysis corrects for stratification in genome-wide association studies. Nature Genetics. 2006;38(8):904–909. doi: 10.1038/ng1847. [DOI] [PubMed] [Google Scholar]
  38. R Development Core Team. R: A language and environment for statistical computing 2011 [Google Scholar]
  39. Ray R, Lerman C, Tyndale RF. Nicotine Dependence Pharmacogenetics: Role of Genetic Variation in Nicotine-Metabolizing Enzymes. Journal of Neurogenetics. 2009;23(3):252–261. doi: 10.1080/01677060802572887. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Robins LN, Babor TF, Cottler LB. Composite International Diagnostic Interview: Expanded Substance Abuse Module. St. Louis: Authors; 1987. [Google Scholar]
  41. Sareen J, Bolton J, Cox B, Clara I. Use of alcohol and drugs to self-medicate anxiety disorders in a nationally representative sample. Journal of Nervous and Mental Disease. 2006;194(11):818–825. doi: 10.1097/01.nmd.0000244481.63148.98. [DOI] [PubMed] [Google Scholar]
  42. Schatzkin A, Abnet CC, Cross AJ, Gunter M, Pfeiffer R, Gail M, et al. Mendelian Randomization: How It Can-and Cannot-Help Confirm Causal Relations between Nutrition and Cancer. Cancer Prevention Research. 2009;2(2):104–113. doi: 10.1158/1940-6207.CAPR-08-0070. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Smith GD. Mendelian randomization for strengthening causal inference in observational studies: Application to gene X environment interactions (vol 5, pg 527, 2010) Perspectives on Psychological Science. 2011;6(3):314–314. doi: 10.1177/1745691610383505. [DOI] [PubMed] [Google Scholar]
  44. Smith GD, Ebrahim S. ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease? International Journal of Epidemiology. 2003;32(1):1–22. doi: 10.1093/ije/dyg070. [DOI] [PubMed] [Google Scholar]
  45. Smith GD, Ebrahim S. What can mendelian randomisation tell us about modifiable behavioural and environmental exposures? British Medical Journal. 2005;330:1076–1079. doi: 10.1136/bmj.330.7499.1076. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Speliotes EK, Willer CJ, Berndt SI, Monda KL, Thorleifsson G, Jackson AU, et al. Association analyses of 249,796 individuals reveal 18 new loci associated with body mass index. Nature Genetics. 2010;42(11):937–U953. doi: 10.1038/ng.686. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Sullivan PF, Cichon S, Craddock N, Daly M, Faraone SV, Gejman PV, et al. A framework for interpreting genome-wide association studies of psychiatric disorders The Psychiatric GWAS Consortium Steering Committee. Molecular Psychiatry. 2009;14(1):10–17. doi: 10.1038/mp.2008.126. [DOI] [PubMed] [Google Scholar]
  48. Thomas DC. Gene-environment-wide association studies: emerging approaches. Nature Reviews Genetics. 2010a;11(4):259–272. doi: 10.1038/nrg2764. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Thomas DC. Methods for Investigating Gene-Environment Interactions in Candidate Pathway and Genome-Wide Association Studies. Annual Review of Public Health. 2010b;31:21–36. doi: 10.1146/annurev.publhealth.012809.103619. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Thorgeirsson TE, Gudbjartsson DF, Surakka I, Vink JM, Amin N, Geller F, et al. Sequence variants at CHRNB3-CHRNA6 and CYP2A6 affect smoking behavior. Nature Genetics. 2010;42(5):448–U135. doi: 10.1038/ng.573. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Timpson NJ, Lawlor DA, Harbord RM, Gaunt TR, Day INM, Palmer LJ, et al. C-reactive protein and its role in metabolic syndrome: mendelian randomisation study. Lancet. 2005;366(9501):1954–1959. doi: 10.1016/S0140-6736(05)67786-0. [DOI] [PubMed] [Google Scholar]
  52. Vink JM, Smit AB, de Geus EJC, Sullivan P, Willemsen G, Hottenga JJ, et al. Genome-wide Association Study of Smoking Initiation and Current Smoking. American Journal of Human Genetics. 2009;84(3):367–379. doi: 10.1016/j.ajhg.2009.02.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Visscher PM, Yang JA, Goddard ME. A Commentary on ‘Common SNPs Explain a Large Proportion of the Heritability for Human Height’ by Yang et al. (2010) Twin Research and Human Genetics. 2010;13(6):517–524. doi: 10.1375/twin.13.6.517. [DOI] [PubMed] [Google Scholar]
  54. Vrieze SI, McGue M, Miller MB, Legrand LN, Schork NJ, Iacono WG. An assessment of the individual and collective effects of variants on height using twins and a developmentally informative study design. PLoS Genet. 2011;7(12):e1002413. doi: 10.1371/journal.pgen.1002413. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Willer CJ, Sanna S, Jackson AU, Scuteri A, Bonnycastle LL, Clarke R, et al. Newly identified loci that influence lipid concentrations and risk of coronary artery disease. Nature Genetics. 2008;40(2):161–169. doi: 10.1038/ng.76. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Yang JA, Benyamin B, McEvoy BP, Gordon S, Henders AK, Nyholt DR, et al. Common SNPs explain a large proportion of the heritability for human height. Nature Genetics. 2010;42(7):565–U131. doi: 10.1038/ng.608. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

1
NIHMS380825-supplement-1.xls (28KB, xls)

RESOURCES