Behavior Genetics

, Volume 36, Issue 4, pp 553–566

Assortative Mating for Cigarette Smoking and for Alcohol Consumption in Female Australian Twins and their Spouses

Authors

    • Department of PsychiatryWashington University School of Medicine
  • Andrew C. Heath
    • Department of PsychiatryWashington University School of Medicine
  • Julia D. Grant
    • Department of PsychiatryWashington University School of Medicine
  • Michele L. Pergadia
    • Department of PsychiatryWashington University School of Medicine
  • Dixie J. Statham
    • Queensland Institute of Medical Research
  • Kathleen K. Bucholz
    • Department of PsychiatryWashington University School of Medicine
  • Nicholas G. Martin
    • Queensland Institute of Medical Research
  • Pamela A. F. Madden
    • Department of PsychiatryWashington University School of Medicine
Original Paper

DOI: 10.1007/s10519-006-9081-8

Cite this article as:
Agrawal, A., Heath, A.C., Grant, J.D. et al. Behav Genet (2006) 36: 553. doi:10.1007/s10519-006-9081-8
  • 213 Views

Abstract

Background

Non-random mating affects population variation for substance use and dependence. Developmentally, mate selection leading to positive spousal correlations for genetic similarity may result in increased risk for substance use and misuse in offspring. Mate selection varies by cohort and thus, assortative mating in one generation may produce marked changes in rates of substance use in the next. We aim to clarify the mechanisms contributing to spousal similarity for cigarette smoking and alcohol consumption.

Methods

Using data from female twins and their male spouses, we fit univariate and bivariate twin models to examine the contribution of primary assortative mating and reciprocal marital interaction to spousal resemblance for regular cigarette smoking and nicotine dependence, and for regular alcohol use and alcohol dependence.

Results

We found that assortative mating significantly influenced regular smoking, regular alcohol use, nicotine dependence and alcohol dependence. The bivariate models for cigarette smoking and alcohol consumption also highlighted the importance of primary assortative mating on all stages of cigarette smoking and alcohol consumption, with additional evidence for assortative mating across the two stages of alcohol consumption.

Conclusions

Women who regularly used, and subsequently were dependent on cigarettes or alcohol were more likely to marry men with similar behaviors. After mate selection had occurred, one partner’s cigarette or alcohol involvement did not significantly modify the other partner’s involvement with these psychoactive substances.

Keywords

Assortative matingSmokingDrinkingMultiple StagesGenetic

Introduction

Patterns of substance use and substance abuse/dependence tend to aggregate in families (Bierut et al. 1998; Merikangas et al. 1998). Regardless of whether this familial risk is genetic or environmental, one of the factors that may contribute to parent–offspring and sibling covariance is the non-random mating of spouses (Fisher 1918; Eaves 1977; Crow and Felsenstein 1982). This non-random mating has important implications for the study of population variation in risk for substance use disorders, including changes in population variation over time. In particular, compared to random mating, phenotypic assortment (or the tendency for like to marry like) will, for heritable traits, induce a genetic correlation between spouses, which in turn, will increase the genetic variance in the offspring generation (Fisher 1918). With genetically similar parents, their offspring are not only more likely to inherit genetic determinants of substance use and misuse, but also are more likely to be exposed to a family environment which promotes substance use. The importance of assortative mating as a social phenomenon, which contributes to familial resemblance for substance use, has long been noted (see, for example, a recent critique of genomic research on substance use disorders, Merikangas and Risch 2003). What has been overlooked is the potential for changes in patterns or intensity of mate selection to create changes in rates of substance use or abuse/dependence in the next generation through its effects on the distribution of genetic risk.

Vanyukov et al., in their review of studies of assortative mating for the liability to substance abuse, suggest multiple mechanisms that contribute to the correlations between spouses for traits such as alcoholism and drug abuse (Vanyukov et al. 1996). The potential mechanisms that may jointly contribute to spousal correlations include (a) primary phenotypic assortative mating, or selection of a mate based on the observed trait (e.g. cigarette smokers marry other cigarette smokers), (b) social homogamy, or selection of a spouse based on social background factors correlated with substance use (e.g. assortative mating by religious affiliation, with social background factors that are associated with the probability of acceptance or prohibition of substance use varying by religious affiliations) and (c) marital contagion (also called “infection”), where there is a direct effect of one partner’s behavior on the other. This reciprocal exchange may be positive, in which case it constitutes contagion, or negative, in which case, the disorder in one partner may confer a protective influence on the spouse.

These mechanisms of assortative mating may combine to determine spousal similarity for cigarette smoking and alcohol drinking behaviors. For example, an individual with substance dependence may select a partner with a similar genetic and social background and subsequent to this stage of primary assortment, one partner’s substance dependence may influence the spouse’s development of substance use disorders. Of these different possible mechanisms, phenotypic assortative mating, whether based on the “primary” trait of interest or one or more traits that are highly genetically correlated with the primary trait, will have the strongest genetic implications for risk in offspring. First, primary assortative mating for substance use disorders in the parental generation leads to transmission of increased genetic vulnerability to these disorders from parent to offspring. Second, offspring with such genetic vulnerability may be at increased risk for comorbid disorders that are indirectly influenced by these common genetic factors. Third, from a statistical perspective, primary assortative mating leads to changes in the ratio of monozygotic (MZ) and dizygotic (DZ) twin correlations. If there is evidence for primary assortative mating for substance use disorders but this is not included as a parameter in the twin model, heritability estimates may be downwardly biased. Fourth, primary assortative mating may impose some restrictions on genetic association studies for substance use disorders. However, the impact of assortative mating on a polygenic trait is often reduced with increasing number of loci of small variance influencing the trait. Finally, it is important to note that in conjunction with genetic implications, all mechanisms of non-random mating in the parental generation, including primary assortative mating, social homogamy and reciprocal marital interaction, may influence the environment of the offspring. Therefore, such offspring are not only genetically vulnerable to develop substance use disorders but are also exposed to a home environment where both parents may have a lifetime history of these disorders and other comorbid psychopathology.

Positive correlations between spouses for substance abuse/dependence have been noted by multiple studies (Hall et al. 1983a, c; Penick et al. 1987; Maes et al. 1998; Hopfer et al. 2003; Sakai et al. 2004). More specifically, a number of studies have observed spousal similarity for drinking behavior (Rimmer and Winokur 1972; Hall et al. 1983a, b; Schuckit et al. 1994; Kendler et al. 1994; Stallings et al. 1997; Maes et al. 1998; Grant et al. 2003). Using overlapping samples, both Kendler et al. (1994) and Maes et al. (1998) report a modest correlation (0.12–0.27) between spouses for alcohol dependence. Grant and colleagues (2003) have found evidence for moderate primary assortative mating (selection of spouse due to phenotypic similarity) for alcohol dependence in a sample of Australian twins (Grant et al. 2003).

While there is mounting evidence for spousal similarity in smoking behavior (Sackett et al. 1975; Sutton 1980; Price et al. 1981; Ho 1986; Ogden et al. 1997; Cardenas et al. 1997), considerably fewer studies have examined the sources of this similarity. Boomsma et al. showed a modest to moderate correlation between spouses for ever-smoking and current smoking status (Boomsma et al. 1994). In a follow-up to the early work by Boomsma et al., Vink and colleagues report that having a smoking spouse was a significant risk factor for smoking and that this correlation attenuated with increasing age (Vink et al. 2003).

Existing literature on spousal similarity for substance use and abuse has, from a behavioral genetic perspective, two important shortcomings. First, it has failed to investigate the mechanisms that lead to spousal similarity, and hence their implications for genetic risk in the offspring generations, despite the existence of models that can be applied for these purposes (Eaves et al. 1984; Heath and Eaves 1985; Eaves et al. 1989). Second, previous studies have generally failed to consider the multi-stage developmental nature of drug involvement. Over the last decade or so, considerable emphasis has been placed on the modeling of drug abuse/dependence conditional on prior stages of drug use. The causal–common–contingent (CCC) model proposed by Kendler et al. (1999) and the hierarchical bivariate genetic model proposed by Heath et al. (2002), have both aimed at separating the genetic and environmental influences on outcomes such as substance abuse/dependence or smoking persistence from the influences on initiation, which may or may not overlap with abuse/dependence (Kendler et al. 1999; Madden et al. 1999; Heath et al. 2002). Results using both approaches suggest some degree of genetic overlap between smoking initiation and smoking persistence or nicotine dependence and similar results may be expected for alcohol-related behavior as well.

The utilization of such two-stage models for drug involvement has not been extended to studies of assortative mating. Thus, published data on nicotine or alcohol dependence or other cigarette smoking and drinking outcomes have failed to separate the effects of assortment for substance use, and assortment for dependence risk in those who have become users. This has caused some confusion in the broader literature on the genetics of substance use disorders. For example, a recent critique of genomic research on smoking and substance use disorders (Merikangas and Risch 2003) emphasizes the magnitude of spousal similarity for smoking behaviors without addressing whether this high correlation is still observed in couples concordant for initiation. The authors did not consider that the existing literature has failed to demonstrate whether such spousal similarity has relevance for genetic research on substance-related outcomes in those who are users (e.g. regular smokers with nicotine dependence).

In the present manuscript
  1. (1)

    We examine the effects of primary assortative mating and mate interaction on spousal similarity for substance use and related problems, including two individual stages of cigarette smoking behavior: regular smoking and DSM-IV Nicotine Dependence, as well as two stages of alcohol consumption: Regular alcohol use and DSM-IV Alcohol Dependence.

     
  2. (2)

    When evidence for assortative mating for more than one stage of cigarette smoking or alcohol consumption is found, we investigate the effects of assortative mating and mate interaction on the subsequent stage of dependence when conditioned for prior regular consumption.

     

Methods

Sample

The data for this study are drawn from a twin register maintained by the Australian National Health and Medical Research council (Jardine et al. 1984b; Martin et al. 1985; Heath et al. 1995, 2001). The Australian Twin Registry (ATR) is a volunteer panel of same-sex monozygotic (MZ) and dizygotic (DZ) twins and opposite-sex dizygotic (DZOS) twin pairs. Founded in 1978, the ATR consists of two cohorts of twins. Twins born before 1965, who were aged 18 years or older and eligible to complete a mailed questionnaire survey conducted in 1981–1982, are members of the older “1981” cohort (or Cohort 1) (Jardine et al. 1984a) while twins born between 1964 and 1971, who were eligible to complete a mailed questionnaire survey conducted in 1989 constitute the younger “1989” cohort (or Cohort 2) (Heath et al. 2001).

For the present analyses, we utilize data from twins. Interview data on drinking behavior, collected via telephone interviews, was available for 2,087 male and 3,908 female twins from an interview (the Phase I interview) that was administered in 1992–1993 (Heath et al. 1997). While the Phase I interviews included detailed questions on onset and frequency of alcohol use and an assessment of DSM-IIIR alcohol dependence for all Cohort I participants, this interview did not assess smoking or diagnostic criteria for DSM-IV nicotine dependence. Therefore, items assessing tobacco use and nicotine dependence were included in two follow-up interviews. These two interviews were: (a) Phase II interview, which included 339 men and 228 women who were Cohort I participants and were from pairs where at least one twin was diagnosed with DSM-IIIR alcohol dependence in their Phase I interview and (b) Women’s Survey interview, a separate survey of 970 female controls who were Cohort I participants but did not participate in the Phase II interview (female twins from pairs where neither twin had a lifetime history of DSM-IIIR alcohol dependence at Phase I). Interview data regarding smoking behavior were collected from Phase II participants and female controls (Women’s Survey), including items from the Fagerstrom Tolerance Questionnaire (Fagerstrom 1978) and the tobacco dependence section of the Composite International Diagnostic Interview (World Health Organization 1994). Further details for the Women’s Survey and Phase II samples are available in a related publication (Madden et al. 1997).

For 3,843 twins that were interviewed in Phase I, Phase II or Women’s Survey, we had self-report telephone interview data from the spouses (2,897 male and 1,446 female spouses). These interviews included data on cigarette smoking and alcohol consumption, with items that were identical to those used in the twins.

For the analyses presented here, we utilize information regarding cigarette smoking and alcohol consumption behaviors in twin women and their male spouses only (since nicotine dependence data were only available for female pairs where at least one twin had a history of alcohol dependence). In total, for cigarette smoking behavior, we utilized data from 914 female twins (mean age of 40 years) and their male partners while for alcohol consumption we had available to us a substantially larger dataset of 3,179 twin women (mean age of 45 years) with self-report data from 2,897 male partners. As required by the institutional review boards at Washington University School of Medicine, St. Louis, USA and The Queensland Institute of Medical Research, Brisbane, Australia, verbal consent was obtained from all participants prior to the interviews.

Measures

Smoking Behavior

Regular Cigarette Smoking

An individual (twin or male partner) was considered a regular smoker if they had smoked on either 100 or more occasions (lifetime), or had smoked on 20 to a 100 occasions, and as often as one or two days a week (or daily) for a period of 3 weeks or longer at some point in their life. To use a hierarchical modeling approach (Heath et al. 2002), we needed to define regular smoking as an ordinal rather than a binary variable. Non-regular smokers were coded as “0” and the regular smokers were coded as “1” if their age of onset for regular smoking was 17 years or older and “2” if their age of onset was prior to 17 years of age. A twin-cotwin multiple threshold model gave an acceptable fit to the data supporting the hypothesis of an underlying multivariate normal distribution (χ2=11.9 (df = 5) P=0.36, in MZ twins and χ2=8.3 (df = 5) P=0.14, in DZ twins).

Nicotine Dependence
During the interview, participants were asked about their lifetime history of the following 6 DSM-IV nicotine dependence symptoms:
  1. (1)

    Tolerance, coded as a positive symptom if the participant reported having smoked 20 or more cigarettes in a day at the time of their heaviest consumption;

     
  2. (2)

    Withdrawal, coded as positive if participants reported four or more symptoms when they cut down or quit smoking, or reported smoking to alleviate any of these symptoms: irritability or anger, nervousness, restlessness, trouble concentrating, slow heart rate, appetite fluctuations, depression or trouble sleeping;

     
  3. (3)

    Smoking a lot more than intended;

     
  4. (4)

    Spending a lot of time smoking cigarettes (defined in this study as chain-smoking);

     
  5. (5)

    Unsuccessful attempts to quit smoking, either trying or wanting, more than once, to stop or cut back on smoking, or trying, once or more, to quit but finding that they could not;

     
  6. (6)

    Smoking despite persistent physical or psychological problems, such as nervousness, high blood pressure, cough, lung trouble or other serious illness.

     

The DSM-IV symptom of giving up important activities because the participant would not be able to smoke was excluded as it was not assessed in all the interviews. Participants reporting 3 or more dependence symptoms were diagnosed with DSM-IV nicotine dependence (without 12-month clustering of symptoms).

We report estimates for three different measures of nicotine dependence: (i) the traditional (unconditional) measure in which non-regular smokers are coded as zero (NDuncond); (ii) a truncated measure, in which, non-regular smokers are coded as missing in a univariate analysis (NDtrunc); and finally (iii) the estimates obtained under a bivariate two-stage model, (conditional estimate: NDcond), in which we jointly analyze the regular smoking ordinal variable and nicotine dependence, with the latter set to missing for non-regular smokers.

Alcohol Consumption

Regular alcohol use

Participants were asked if they had ever consumed alcoholic beverages once a month for 6 months or longer. Similar to the ordinal regular smoking measure, this variable was trichotomized based on the mean age of onset of regular alcohol consumption so that 0 = non-regular drinkers, 1 = late onset regular drinkers and 2 = early onset (prior to 19 years in men and 21 years in women) regular drinkers. Similar to regular smoking, this ordinal measure was found to possess an underlying multivariate normal liability distribution (χ2=8.8 (df = 5) P=0.12 in MZ twins, and 6.0 (df = 5) P=0.30 in DZ twins).

Alcohol Dependence
The following 7 items were used to assess lifetime DSM-IV alcohol dependence in our sample:
  1. (1)

    Tolerance, measured by items regarding needing to drink a lot more or increase consumption by 50% or more to get the same effect as when the respondent first started to drink regularly;

     
  2. (2)

    Withdrawal, measured by two or more of the following symptoms: having the shakes, sleeplessness, nausea, sweating, increased heart rate and so on, when cutting back on drinking alcohol, and drinking alcohol to alleviate these symptoms. Alternatively, individuals reporting either (a) fits, seizures or convulsions leading to loss consciousness and temporary loss of memory or (b) Delirium tremens (DT), were coded to suffer from alcohol withdrawal;

     
  3. (3)

    More than once, getting unintentionally drunk or drinking for longer durations than intended;

     
  4. (4)

    Persistent desires to cut back (3+ times) or unsuccessful attempts to cut back on drinking;

     
  5. (5)

    Spending a great deal of time recovering from the effects of drinking alcohol;

     
  6. (6)

    Giving up important activities (e.g. sports, work, associating with friends) to drink;

     
  7. (7)

    Continued use despite negative physical (e.g. blackouts, liver disease) or psychological (e.g. depression, paranoia, difficulty thinking) problems.

     

Individuals endorsing a lifetime history of 3 or more symptoms were diagnosed with DSM-IV alcohol dependence (without 12-month clustering of symptoms). Clustering could not be assessed as these items were derived, in the twin sample, from an interview designed to assess DSM-IIIR alcohol dependence (Heath et al. 1997)). For univariate and bivariate analyses, alcohol dependence was conditioned upon prior regular alcohol use (ADtrunc and ADcond, respectively) i.e. non-regular drinkers were coded as missing for their alcohol dependence diagnosis. Results of a more extensive analysis (focusing on ADuncond as a phenotype) incorporating male twins and their female partners, and incorporating twin ratings of spousal alcohol problems, are presented elsewhere (Grant et al. 2003).

Data Analysis

As summary measures of spousal resemblance, we compute female twin-male partner tetrachoric correlations for NDuncond, NDtrunc and ADtrunc using SAS (SAS Institute, 1999). Marital correlations (within and across traits) for regular smoking and NDcond, and for regular alcohol use and ADcond were estimated using a two-stage bivariate normal model (Heath et al. 2002) by fitting models to data using Mx (Neale, 2004).

For univariate genetic model-fitting analyses, incorporating phenotypic assortative mating and reciprocal marital interaction, the total phenotypic variance was partitioned into additive genetic (A), shared environmental (C) and unique environmental (E) influences (Fig. 1). As depicted in Fig. 1, additive genetic influences (A) correspond to those factors (e.g. latent genetic factors that influence nicotine or alcohol metabolism, or the genes that indirectly influence nicotine and alcohol use, such as genes for personality traits of neuroticism or novelty-seeking) that are correlated 1.0 in monozygotiz (MZ) or identical twins and 0.5 in dizygotic (DZ) or fraternal twins, in the absence of primary assortative mating. When one partner selects another based on a certain trait (e.g. smokers marry smokers) this genetic correlation between the twin members of a DZ pair is increased, as described in the following paragraph. Shared environmental influences (C) are correlated 1.0 across members of MZ and DZ twin pairs and refer to environmental factors that both twins experience (e.g. home environment, same school), while unique environmental influences (E) are uncorrelated across members of a twin pair and refer to environmental factors that are specific to each member of the twin pair (e.g. an unshared peer, traumatic event). Therefore, while C contributes to between-family variation, E contributes to within-family (or within-twin pair) differences. Also, from Fig. 1, we note that the covariance between a twin and the spouse of their co-twin is assumed to occur through the co-twin. Similarly, the covariance between the spouses of each twin occurs through the members of the twin pair. In other words, we assume that the spouses depicted in Fig. 1, are genetically unrelated (i.e. not brothers, cousins etc.) and that their genetic relatedness occurs only due to selection of a phenotypically similar mate, who is also the member of an MZ or DZ twin pair.
https://static-content.springer.com/image/art%3A10.1007%2Fs10519-006-9081-8/MediaObjects/10519_2006_9081_Fig1_HTML.gif
Fig. 1

Path diagram depicting a female twin pair (Twin 1 and Twin 2) with their male partners (Spouse1 for twin 1 and Spouse2 for twin2). The circles represent latent genetic (A), shared environmental (C) and unique environmental (E) influences, the ellipses represent the overall liability or vulnerability to regular cigarette smoking while the rectangles represent the observed trait. Two mechanisms of assortative mating are shown: primary assortative mating (μ) between the ellipses, from twin’s vulnerability to regular smoking to spouse’s vulnerability to regular smoking, and reciprocal marital interaction (β) between the rectangles (or observed regular smoking) from twin to spouse. Rc, or the cross-twin within-trait shared environmental correlation is fixed across MZ and DZ twins at 1.0 as is the Rg, or additive genetic correlation between twin members of an MZ pair. The Rg for twin members of a DZ twin pair is no longer 0.5 but is increased to 0.5 (1+μ a2) due to primary assortative mating. The unique environmental factors (E) are uncorrelated

The models utilized for the current analyses also included two sources of spousal similarity:
  1. (i)

    Primary assortment – the selection of a partner based on their phenotype, and reciprocal marital interaction – the influence of one partner’s behavior on their spouse, subsequent to primary assortment. Primary phenotypic assortment was measured with a regression μ(mu) of the partner’s liability on the twin’s liability (ellipses in Fig. 1) for smoking or drinking behavior (this μ component models the initial causes of partner similarity). The μ path results in an increase in the genetic correlation between DZ (dizygotic) twins from 0.5 (i.e. classical twin models) to 0.5 (1+μa2), where a2 is the estimate of heritability (Heath et al. 1984; Heath and Eaves 1985; Eaves et al. 1989), under the simplifying assumption that assortative mating is constant across generations. For this manuscript, where we only address questions regarding the mechanisms of mate selection (ignoring the question of cohort changes), this simplifying assumption is unavoidable.

     
  2. (ii)

    Reciprocal marital interaction – the environmental influence of the twin’s phenotype on the partner’s phenotype was estimated via a reciprocal β(beta) path (this β component models reciprocal mate phenotypic interactions). This level of assortative mating occurs after primary mate selection has occurred and is based on the interactions between the observed traits in the partners (rectangles in Fig. 1). Since we only used data from female twin pairs, we were forced to constrain the reciprocal influence of the female partner on the male partner to the influence of the male partner on the female partner.

     
When there was evidence for primary assortative mating and/or reciprocal marital interactions for more than one stage of cigarette smoking or drinking behavior, we also tested a bivariate model for primary assortative mating and mate interaction. The bivariate model had two unique features:
  1. (a)

    The second variable (nicotine dependence or alcohol dependence) was contingent on the first variable (regular smoking or regular alcohol use) such that individuals who were not regular smokers/drinkers were structurally missing for nicotine/alcohol dependence. Using a triangular decomposition parameterization of the standard bivariate genetic model, this model allows us to partition genetic and environmental influences that are shared across the two stages of smoking/drinking behavior from those factors that are specific to nicotine/alcohol dependence. A necessary assumption for the bivariate genetic model for multiple stage models is that the stage preceding the conditional phenotype must be ordinal, with at least 2 of 3 categories having non-missing values for the conditional phenotype, and possess an underlying liability that is normally distributed. For instance, in the case of regular alcohol use and alcohol dependence, the distal phenotype (alcohol dependence) may be measured in individuals that are classified into 2 of the 3 categories in the preceding stage (late or early regular drinkers, but not non-regular drinkers) (Heath et al. 2002).

     
  2. (b)

    Primary assortment (within and across traits) was considered to be at equilibrium across generations. In other words, we assumed a constant magnitude of primary assortative mating across generations, which implies certain constraints being imposed upon the within-trait and cross-trait covariance for twins, their co-twins and their spouse. The basis of this model, where the covariance across members of a twin pair is constrained by a parameter “ρ” is explained in Eaves et al. (1984).

     

Expectations under bivariate twin models for assortative mating were created using standard tracing rules of path analysis (Neale and Cardon 1992). All models were fitted using Mx (Neale 2004) and age at interview was included as a fixed covariate. Due to computational intensity, only linear effects of age were included. The relative fit of sub-models was examined using the chi-square difference in the −2 log likelihood fit of the sub-models with a full model that freely estimated A, C, E, μ and β.

Results

Sample Characteristics

At the time of their respective interview assessments, the mean age of the female twins was 40.1 (SD 6.5) and on average, their male partners were 3 years older. Female twins and their spouses reported a mean age of marriage to their current spouse as 26.5 years (SD 7.3), with mean age of having first met their spouse at 22.6 years (SD = 7.5). The female twins and their male partners reported a mean of 19.5 years (SD 11.7) of cohabitation as a couple with only 11% reporting fewer than 5 years of cohabitation at the time of the interview.

The prevalence of regular smoking in those who had ever tried cigarettes was 55.6% and 63.1% in the female twins and their male partners respectively. Of these regular smokers, 56.4% of the female twins and 56.6% of their male partners reported a history of nicotine dependence. The mean age of onset for regular smoking was 16 years for twins and their male partners.

Regular alcohol use was endorsed by 80% of the women and 86% of their male partners. Approximately 6% of the female regular drinkers and 25% of their male partners reported a lifetime history alcohol dependence. Mean age of onset of regular alcohol use was 21 years in the women and 19 years in their male partners.

Assortative Mating

In our sample, the spousal correlations were 0.30 [95% C.I. 0.17–0.37] for regular smoking, 0.31 [95% C.I. 0.12–0.38] for unconditional nicotine dependence or NDuncond, 0.10 [95% C.I. 0.001–0.13] for nicotine dependence in couples concordant for regular smoking or NDtrunc. The spousal correlation for NDcond was 0.19 [95% C.I. 0.08–0.25] and the spousal correlation between regular smoking in the female twin and nicotine dependence (NDcond) in their male partner was 0.15 [95%C.I. 0.02–0.30]. For alcohol consumption, spousal correlations were: 0.38 [95% C.I. 0.19–0.43] for regular alcohol use, and 0.15 [95% C.I. 0.07–0.21] for DSM-IV alcohol dependence or ADtrunc and 0.12 [95% C.I. 0.05–0.28] between regular alcohol use in the female twin and alcohol dependence (ADcond) in their male partner.

Regular Cigarette Smoking and DSM-IV Nicotine Dependence

Table 1 includes the standardized parameter estimates, under univariate models, for regular cigarette smoking and for nicotine dependence in regular smokers (NDtrunc) and nicotine dependence, where non-regular smokers were coded as unaffecteds (NDuncond). In the univariate case, for regular smoking, after accounting for significant primary assortative mating (μ = 0.39), we found no evidence for reciprocal marital interaction, (δχ2 = 2.4 (df = 1)). Therefore, while regular cigarette smokers were likely to select other regular smokers as their partners, once mate selection had occurred, there was no influence of one partner’s regular smoking on the other partner. For NDuncond, we were able to drop the reciprocal marital interaction parameter (δχ2 = 0.38 (df = 1), β=0.12: Table 1) but not the parameter that estimated primary assortative mating (δχ2 = 8.6 (df = 1), β = 0.38). For nicotine dependence in regular smokers (NDtrunc), we were able to drop the influence of either reciprocal marital interaction (δχ2 = 3.06 (df = 1), β = 0.19: Model A in Table 1), or primary assortative mating (δχ2 = 0.55 (df = 1), β = 0.31: Model B in Table 1), but not the influence of both sources of assortative mating (δχ2 = 13.57 (df = 2)).
Table 1

Parameter estimates (with 95% C.I.) for univariate models of assortative mating that examine the influence of primary assortative mating and reciprocal marital interaction on regular cigarette smoking, DSM-IV nicotine dependence with non-regular smokers coded as structurally missing (NDtrunc) and DSM-IV nicotine dependence with non-regular smokers coded as unaffecteds (NDuncond)

Squared parameter estimates

Regular smoking

DSM-IV nicotine dependence (NDuncond)

DSM-IV nicotine dependence (NDtrunc) Model A

DSM-IV nicotine dependence (NDtrunc) Model B

Additive genetic (a2)

0.66 [0.20–0.94]

0.73 [0.00–0.91]

0.39 [0.01–0.80]

0.43 [0.01–0.80]

Shared environment (c2)

0.23 [0.01–0.63]

0.07 [0.00–0.42]

0.36 [0.00–0.87]

0.26 [0.00–0.87]

Unique environment (e2)

0.11 [0.05–0.23]

0.20 [0.09–0.36]

0.25 [0.09–0.68]

0.31 [0.09–0.68]

Primary assortative mating (μ)

0.39 [0.26–0.50]

0.38 [0.10–0.49]

0.31 [0.12–0.48]a

Reciprocal marital

0.19 [0.09–0.28]a

Interaction (β)

    

aEither assortative mating parameter could be statistically dropped, but not both i.e. Model A fit as well as Model B

Regular smoking = smoking either 100 or more times, lifetime or smoking between 20 and 100 times, and as often as 1 or 2 days a week (or daily) for a period of 3 weeks or longer at some point in one’s life

NDtrunc = non-regular smokers excluded/missing from diagnosis

NDuncond = non-regular smokers coded as unaffecteds in diagnosis

Parameter estimates under the full bivariate model, including additive genetic, shared environmental and unique environmental influences common to both regular cigarette smoking and nicotine dependence (A1, C1 and E1), as well as effects specific to nicotine dependence (A2, C2 and E2) are shown in Table 2 (Model 1) and the best-fitting model is presented in Fig. 2a. Three primary assortment paths and mate interaction paths (within regular smoking, within nicotine dependence and from regular smoking to nicotine dependence) were estimated in the full model. First, in sub-models 2–4, we attempted to drop the overlap between additive genetic (sub-model 2), shared environmental (sub-model 3) and unique environmental (sub-model 4) factors influencing regular smoking and nicotine dependence, and found that shared environmental factors did not significantly influence the liability to nicotine dependence. Hence, from model 3b, which included A, C and E for regular cigarette smoking and A and E for nicotine dependence (with a large proportion of the additive genetic influence on nicotine dependence overlapping with prior regular smoking), we proceeded to test for the significance of primary assortative mating (sub-model 3b.1, 3b.2) and reciprocal marital interaction (sub-model 3b.1, 3b.3). Our best-fitting model was sub-model 3b.5, in which we could statistically drop all reciprocal marital interaction paths as well as the cross-trait assortative mating path from regular cigarette smoking in the twin to nicotine dependence in their partner. Therefore, similar to the univariate models, for these two stages of cigarette smoking, we found that regular cigarette smokers were more likely to marry other regular cigarette smokers (μ = 0.32). Furthermore, even after accounting for mate selection for regular smoking, women with a history of nicotine dependence were also more likely to select a partner with a history of nicotine dependence (μ = 0.37).
Table 2

Model-fit indices for the bivariate model where nicotine dependence (NDcond) was conditioned for prior regular cigarette smoking

Model#

Model

−2 loglikelihood

Δχ2 (degrees of freedom)

P-value

Akaike’s Information Criteria (AIC)

1

Full

5542.15

 

2

Drop additive genetic (A) across regular smoking and ND

5550.86

8.71 (1)

0.0128

6.71

3

Drop shared environment (C) across regular smoking and ND

5542.18

0.03 (1)

0.8625

−1.97

4

Drop unique environment (E) across regular smoking and ND

5549.60

7.45 (1)

0.0063

5.45

2a

Drop all A on regular smoking

5549.36

7.21 (2)

0.0272

3.21

2b

Drop all A on ND

5558.92

16.77 (2)

0.0002

12.77

3a

Drop all C on regular smoking

5561.06

18.91 (2)

<0.0001

14.91

3b

Drop all C on ND

5542.21

0.06 (2)

0.9704

−3.94

3b.1a

Drop all primary assortment (μ) and marital interaction (β)

5600.81

58.60(6)

<0.0001

46.60

3b.2a

Drop all μ

5558.86

16.65 (3)

0.0008

10.65

3b.3a

Drop all β

5543.61

1.40 (3)

0.7055

−4.60

3b.4a

Drop μ on ND and across regular smoking and ND + all β

5582.52

40.31 (5)

<0.0001

30.31

3b.5a

Drop μ from regular smoking to ND + all β

5545.12

2.91 (4)

0.5730

−5.09

The change in the loglikelihood ratio chi-square was used to select the best-fitting model

aCompare the –2 loglikelihood of this model with –2 loglikelihood of model 3b. Shaded row represents best-fitting model based on sub-model 3b.

https://static-content.springer.com/image/art%3A10.1007%2Fs10519-006-9081-8/MediaObjects/10519_2006_9081_Fig2_HTML.gif
Fig. 2

(a) Parameter estimates (raw and un-standardized) for the bivariate twin model that examines that influence of primary assortative mating within traits (single-headed arrow from twin’s regular cigarette smoking to spouse’s regular cigarette smoking, or twin’s nicotine dependence to spouse’s nicotine dependence). The circles represent latent genetic (A), shared environmental (C) and unique environmental (E) influences, the ellipses represent the overall liability or vulnerability to regular cigarette smoking or nicotine dependence while the rectangles represent the observed trait. The model is simplified by showing variables for only one member of the twin pair, and their spouse. Nicotine dependence here refers to NDcond. (b) Parameter estimates (raw and un-standardized) for the bivariate twin model that examines that influence of primary assortative mating within traits (single-headed arrow from twin’s regular alcohol use to spouse’s regular alcohol use, or twin’s alcoholism to spouse’s alcoholism) and across traits (single-headed arrow from twin’s regular alcohol use to spouse’s alcoholism). The circles represent latent genetic (A), shared environmental (C) and unique environmental (E) influences, the ellipses represent the overall liability or vulnerability to regular alcohol use or alcoholism while the rectangles represent the observed trait. The model is simplified by showing variables for only one member of the twin pair, and their spouse. Alcohol dependence here refers to ADcond

Regular Alcohol Consumption and DSM-IV Alcohol Dependence

Results from the univariate models for alcohol consumption (regular alcohol use and alcohol dependence) are presented in Table 3. Regular alcohol use was influenced by primary assortative mating (δχ2 = 122.2 (df = 2), μ = 0.74) and reciprocal marital interaction (δχ2 = 7.29 (df = 1), μ = −0.25). Similar to the results for nicotine dependence, when DSM-IV alcohol dependence was coded so as to exclude non-regular alcohol drinkers, we found evidence for primary assortative mating (μ = 0.16). However, in contrast to regular alcohol consumption, we did not find evidence for reciprocal marital interaction for alcohol dependence (δχ2 = 0.60 (df = 1)). To further examine the extent of common versus stage-specific assortative mating, we applied the bivariate assortative mating model to alcohol consumption data.
Table 3

Parameter estimates (with 95% C.I.) for univariate models of assortative mating that examine the influence of primary assortative mating and reciprocal marital interaction on regular alcohol use and DSM-IV alcohol dependence with non-regular alcohol users coded as structurally missing

Squared parameter estimate

Regular alcohol use

DSM-IV alcohol dependence (ADtrunc)

Additive genetic (a2)

0.10 [0.00–0.48]

0.51 [0.30–0.68]

Shared environment (c2)

0.41 [0.07–0.56]

0.01 [0.00–0.28]

Unique environment (e2)

0.49 [0.41–0.60]

0.48 [0.32–0.70]

Primary assortative mating (μ)

0.74 [0.55–0.86]

0.16 [0.02–0.29]

Reciprocal Marital Interaction (β)

−0.25 [−0.42–(−0.11)]

Regular alcohol use = ever consumed alcoholic beverages once a month for 6 months or longer

ADtrunc = DSM-IV alcohol dependence with non-regular drinkers excluded/missing from diagnosis

For parameter estimates (including a2, c2, e2, μ and β) from ADuncond (non-regular drinkers coded as zero for alcohol dependence) see Grant et al. 2003

Results of the bivariate model-fitting are detailed in Table 4. Similar to the bivariate model described for cigarette smoking, the full model (Model 1) for alcohol consumption included additive genetic, shared environmental and unique environmental factors that were shared between regular alcohol use and alcohol dependence (A1, C1 and E1) and A2, C2 and E2, which were specific to alcohol dependence. As before, three primary assortative mating and reciprocal interaction paths (within regular alcohol use, within alcohol dependence, from regular alcohol use to alcohol dependence) were incorporated into model 1. In Sub-models 2 (2, 2a, 2b) and 3 (3, 3a, 3b), we examined the statistical significance of the genetic and environmental influences on regular alcohol use and alcohol dependence while estimating all assortment paths. Sub-model 3b, which was the most parsimonious in terms of A, C and E, included genetic, shared and unique environmental influences on regular alcohol use (ACE), genetic and unique environmental influences only on alcohol dependence (AE) and hence allowed for overlap between genetic and non-shared environmental factors across regular alcohol use and alcohol dependence. From Model 3b, we first dropped all primary assortative mating and marital interaction paths (Sub-model 3b.1). Eliminating all sources of spousal similarity resulted in a significant deterioration of model-fit (Table 4). Therefore, in Sub-models 3b.2 and 3b.3, we separately dropped the primary assortative mating and marital interaction paths, respectively. Next, in Model 3b.4, we dropped only the primary assortative mating and reciprocal marital interaction paths across regular alcohol use and alcohol dependence. Model 3b.3 provided a reasonable fit to the data based on the log-likelihood chi-square statistic and the AIC. This model estimated paths for primary assortative mating for regular alcohol use (μ=0.37), alcohol dependence (μ=0.18) as well as a modest cross-trait assortment between regular alcohol use and alcohol dependence (μ=0.13) (Fig. 2b.)
Table 4

Model-fit indices for the bivariate model where alcohol dependence (ADcond) was conditioned for prior regular alcohol use

Model#

Model

−2 loglikelihood

Δχ2 (degrees of freedom)

P-value

Akaike’s Information Criteria (AIC)

1

Full

12518.34

2

Drop additive genetic (A) across regular alcohol use and AD

12526.83

8.49(1)

0.0040

6.49

3

Drop shared environment (C) across regular alcohol use and AD

12518.38

0.04(1)

0.842

−1.96

4

Drop unique environment (E) across regular alcohol use and AD

12533.01

14.67(1)

<0.0001

12.67

2a

Drop all A on regular alcohol use

13298.60

780.3 (2)

<0.0001

776.3

2b

Drop all A on AD

12557.02

38.68 (2)

<0.0001

34.68

3a

Drop all C on regular alcohol use

12533.83

15.49 (2)

0.0004

11.49

3b

Drop all C on AD

12518.46

0.12 (2)

0.9418

−3.88

3b.1a

Drop all primary assortment (μ) and marital interaction (β)

12691.58

73.12(6)

<0.0001

61.12

3b.2a

Drop all μ

12564.81

46.35 (3)

<0.0001

40.53

3b.3a

Drop all β

12526.30

7.84 (3)

0.05

1.84

3b.4a

Drop μ and β across regular alcohol use and AD

12528.45

9.99 (2)

0.0068

5.99

The change in the loglikelihood ratio chi-square was used to select the best-fitting model

aCompare the −2 loglikelihood of this model with −2 loglikelihood of model 3b. Shaded row represents best-fitting model based on sub-model 3b

Discussion

In this study, we sought to examine the contributions of primary assortative mating and marital interaction to spousal resemblance for stages of cigarette smoking and alcohol consumption. Individual differences in regular cigarette smoking and regular alcohol consumption were explained by additive genetic, (Model 2a, Tables 2 and 4) shared environmental (Model 3a, Tables 2 and 4) and unique environmental influences. Only additive genetic (Model 2b, Tables 2 and 4) and unique environmental but not shared environmental factors (Model 3b, Tables 2 and 4) contributed to individual differences in nicotine and alcohol dependence. We also found evidence for substantial genetic overlap between regular use and dependence for both nicotine and alcohol. Results also suggested that there was significant assortative mating for both stages of cigarette smoking (regular cigarette smoking and nicotine dependence) as well as for both stages of alcohol consumption (regular alcohol use and alcohol dependence). Furthermore, when the alcohol-related stages were analyzed jointly, we found evidence for primary assortative mating between female twins and their male partners for regular alcohol consumption (Fig. 2b, path from ellipse labeled Regular drinking Twin 1 to ellipse labeled Regular drinking Spouse 1), alcohol dependence (Fig. 2b, path from ellipse labeled Alcoholism Twin 1 to ellipse labeled Alcoholism Spouse 1) and also between regular alcohol consumption in the female twin and alcohol dependence in their male partner (Fig. 2b, path from ellipse labeled Regular Drinking Twin 1 to ellipse labeled Alcoholism Spouse 1). Therefore, female regular drinkers were more likely to select a partner who was also a regular drinker and, controlling for their lifetime history of alcohol dependence, also more likely to select a regular drinking partner with a history of alcohol dependence. However, regular smokers, after controlling for their own lifetime history of nicotine dependence, were not more likely to select a partner with a history of nicotine dependence (Fig. 2a).

We did not find evidence for reciprocal spousal influences (marital interaction) for regular smoking. As this was a lifetime measure and most female twins and their male partners had initiated regular cigarette smoking prior to their first romantic date with their male partner (only 8–12% began smoking regularly after they had first met their current partner), this result is not surprising. The apparent evidence for reciprocal influences for regular alcohol use was unexpected, since a large proportion of partners (62–76%) also had initiated regular alcohol use before their first date. However, we did note that when regular alcohol consumption and alcohol dependence, conditioned for prior regular alcohol use, were jointly evaluated, there was little evidence for marital interactions for either alcohol-related stage. Therefore, the bivariate model may have provided greater information in accurately identifying the sources of spousal similarity for both regular alcohol use and subsequent alcohol dependence. This was also true for nicotine dependence where we were better able to discriminate the relative importance of primary assortative mating from reciprocal marital interactions in the bivariate model.

Overall, evidence for phenotypic assortative mating has several genetic implications. First, due to assortative mating, DZ twin correlations increase (from 0.5 to 0.5(1+μa2)) in magnitude relative to MZ twin correlations. In the classical twin design where data from spouses or parents of twins is not available, this increased DZ-MZ correlation ratio will be confounded with estimates of shared environmental influences and may downwardly bias estimates of heritability (Fisher 1918; Neale and Cardon 1992; Evans et al., 2002). This occurs because the traditional twin model assumes random mating and therefore, after accounting for genetic similarity (which is fixed to be 100% and 50% in MZ and DZ twin pairs respectively), any additional within-twin pair similarity is due to those environmental influences that members of a twin pair share. A preponderance of twin studies on the etiology of substance use disorders has found that shared environmental factors influence early stages of involvement (use and regular use) with alcohol (Heath et al. 1991; Carmelli et al. 1993; Prescott et al. 1994; Stallings et al. 1999; McGue et al. 2000; Rhee et al. 2003), cigarettes (True et al. 1997; Kendler et al. 1999; Sullivan and Kendler 1999; Koopmans et al. 1999; Madden et al. 2004; Maes et al. 2004) and illicit drugs (van den Bree et al. 1998; Maes et al. 1999; Kendler et al. 2000; McGue et al. 2000; Miles et al. 2001; Rhee et al. 2003). While it is possible that true shared environmental factors influence initiation and regular use of cigarettes and alcohol (e.g. shared peers, familial permissiveness), our results suggest that the genetic consequence of assortative mating, which is especially likely to occur at these early stages of drug involvement, may, in part, contribute to inflated shared environmental estimates for these outcomes.

Second, assortative mating may result in a positive gene–environment correlation (Eaves et al. 1977) (Jencks 1972). Gene–environment correlation refers to the situation where individuals with a certain genetic predisposition, actively or passively, elicit an environment that is correlated to their genetic liability. The environment may be positively (Jencks 1972) or negatively (Cattell 1963) correlated with the genotype. Children receive half their genes from each parent and if spousal similarity is a direct consequence of genetic similarity between spouses, and there is vertical cultural transmission, which is the direct influence of the parents’ sum total of experiences that are passed on to the children, then we would expect the occurrence of passive G–E correlations in subsequent generations. In other words, children who inherit an increased genetic vulnerability to smoke are also more likely to be exposed to an environment that includes parents that smoke. Therefore, if exposure to two smoking parents has a potent environmental effect on an offspring’s smoking, then the offspring of such couples experience the double disadvantage of both increased genetic and increased environmental risk.

Third, the evidence for phenotypic assortative mating raises the possibility of changes in the intensity of assortative mating across birth cohort. This may contribute to changes in rates of substance use or abuse/dependence via effects on the distribution of genetic risk in the next generation, a possibility that has received inadequate attention. Rates of nicotine and alcohol use and dependence have not declined over the last decade or so, despite more stringency being imposed, through legislation, on the acquisition of cigarettes and alcohol (Substance Abuse and Mental Health Services Administration (SAMHSA) 2005). To some extent these increased rates of involvement with cigarettes and alcohol may reflect a combination of increased genetic vulnerability for cigarette and alcohol use and misuse in the offspring, as well as exposure to a home environment where both parents use cigarettes and/or alcohol, as a consequence of assortative mating in prior generations.

Finally, individuals who select their mate for a predisposition to regular cigarette or alcohol use or nicotine or alcohol dependence, may also be assortatively mating for a number of other, highly genetically correlated phenotypes, such as general behavioral disinhibition, conduct disorder, major depression and use and misuse of other psychoactive substances. Maes et al. (1998) showed that marital correlations between individuals with alcoholism and a variety of major psychiatric disorders were elevated (e.g., with GAD r = 0.21; with depression r = 0.12). This outcome of assortative mating has serious implications for other drug use disorders. For instance, as noted by Compton et al. (2004), age of initiation of marijuana use has decreased and rates of cannabis abuse/dependence, in users, have increased by 18% in the last decade (Compton et al. 2004). While part of this increase, especially the increased upper limit on the age of cannabis dependence, may reflect the effect of assortative mating and reciprocal spousal interactions for cannabis use and dependence, it may, in part, also be due to the indirect effects of spousal similarity for general addictive behaviors. The robust correlation between marijuana use disorders and nicotine (Grant et al. 2004) as well as alcohol dependence (Stinson et al. 2005) imply that assortative mating for use and misuse of these legal psychoactive substances may have consequences on other psychopathology, especially illicit drug use and misuse.

The analyses presented here have certain limitations. We did not address the influence of social homogamy (Heath and Eaves 1985; Reynolds et al. 1996, 2000) (e.g. a latent social factor shared solely by mates). We plan to explore the effects of this mechanism of assortment in future research. While our sample is representative for cigarette smoking (Madden et al. 1997) and alcohol consumption (Heath et al. 1997), it does consist of an older cohort of Caucasian twins from Australia and our analyses were also restricted to female twins and their male partners. Follow-up analyses in a younger cohort of twins and examination of both male and female twins and their partners, which is currently unavailable for these phenotypes, will allow us to extend the generalization of these findings in the future. We also did not examine the impact of divorce or duration of marriage on the development of nicotine or alcohol dependence. We found a modest increase in rates of alcohol dependence in women (9.4% vs. 6% in full sample) and men (28.2% vs. 25% in full sample) reporting divorce or separation (11%) at the time of spouse interviews. We hope to study the effect of duration of cohabitation on drinking and smoking outcomes in the future. Additionally, for nicotine dependence, we were unable to discriminate additive genetic influences from shared environmental contributions to risk, reflecting limited power of the Cohort 1 sample for genetic analysis of smoking outcomes (Madden et al. 1997). In conclusion, mate selection appears to play an important role in determining risk of substance use disorders. More research is needed to better characterize these effects and their consistency across generations.

Acknowledgments

Supported by NIH grants DA12854 (PAFM), AA07535, AA11998, AA13221, AA07728, AA10248 (ACH) and DA019951 (MLP), and by a grant from the Alcoholic Beverages Medical Research Foundation (PAFM).

Copyright information

© Springer Science+Business Media, Inc. 2006