Family Transmission and Heritability of Externalizing Disorders:  A Twin-Family Study FREE

Antisocial behavior and substance dependence disorders are most notablycharacterized by the exceedingly high financial and human costs they exacton society. Although institutions such as the criminal justice system andtreatment programs have been developed to cope with the consequences of thesedisorders, effective intervention will also require a better understandingof their etiology, particularly the nature of familial resemblance. Such anunderstanding requires answers to fundamental questions, such as "Is intergenerationaltransmission attributable to a general vulnerability to all disorders in thisspectrum or is familial transmission primarily due to disorder-specific vulnerabilities?"and "Are the origins of familial resemblance largely genetic or environmentalin nature?"

Family studies consistently show that the presence of a substance dependenceor antisocial behavior disorder in first-degree relatives greatly increasesa person's risk of eventually developing the disorder.¹ Althoughthere is evidence of a general familial risk factor, family studies^2- 6 tendto show that the familial aggregation of different disorders is relativelymore specific than general. Adoption studies^7- 11 aregenerally consistent with family studies in suggesting specific genetic andenvironmental risk factors rather than a general addictive tendency or patternof deviant behavior. In addition, the familial aggregation of substance dependenceand antisocial behavior disorders tends to be the same in men and women, althoughwomen may require greater familial loading before expressing the disorder.^6,10- 14

Twin studies, however, tend to show substantial genetic overlap amongdisinhibitory syndromes, including conduct disorder and alcohol dependence,^15- 17 child and adult antisocialbehavior,^15- 16,18 adultantisocial behavior and alcohol and drug dependence,^15- 16,19 anddependence among different classes of illicit drugs.^20- 21 Twinstudies^15- 16 also tend to showthe same pattern of genetic and environmental effect for men and women. Althoughtwin studies^15- 21 consistentlyreveal genetic overlap among these disorders, they also find evidence of geneticand environmental risk factors that are specific to each disorder. Analyses^22- 24 that examine thepatterns of comorbidity in large epidemiologic samples consistently identifya broad dimension of risk underlying antisocial personality disorder (ASPD),conduct disorder, alcohol dependence, and drug dependence. This general vulnerabilityfactor is typically referred to as externalizing (EXT),^{15- 16,22- 24} andfindings from recent twin studies^15- 16,25 indicatethat the source of the comorbidity in these externalizing disorders can belargely attributed to common genetic factors.

Rare are twin-family studies that integrate the 2 literatures,^26- 28 allowing for thesimultaneous estimation of general and specific transmission and the geneticand environmental contributions to these effects. Although family studiestend to show specificity of transmission and twin studies tend to show generality,a recent critique²⁹ of family studies concludedthat analyses typical of family studies do not provide valid tests of alternativecomorbidity models. As a remedy, the authors recommended a model-fitting approach,which we use in the present investigation. The present study is also uniquein that, to our knowledge, it is the first twin-family study that incorporatesthe assessment of multiple externalizing disorders in both biological parentsand their twin offspring. Such a study provides an opportunity to clarifywhether familial resemblance (both parental and sibling) is due to generalvs disorder-specific vulnerabilities while also estimating the genetic andenvironmental contributions to these general and specific risk factors.

Participants were members of 542 families participating in the MinnesotaTwin Family Study (MTFS), a longitudinal-epidemiologic study investigatingthe development of substance abuse and related disorders. A comprehensivedescription of the goals, design, and sample characteristics of the MTFS isavailable elsewhere.^30- 31 Familieswere identified using public birth records of twins born in Minnesota betweenJanuary 1, 1972, and December 31, 1979, and were recruited the year the twinsturned 17 years old. For this investigation, each family consisted of thebiological mother and father and their same-sex twin offspring (N = 2168).All but 1 biological mother was also the rearing mother. At the time of theassessment, 464 biological fathers (85.6%) resided with their twin offspring,56 biological fathers (10.3%) had joint custody or visitation rights but didnot currently live with their twin offspring, and 22 biological fathers (4.0%)had little or no contact with their twin offspring but did participate inthe assessment. Of the 542 families, 357 included a monozygotic (MZ) twinpair (169 male and 188 female pairs) and 185 included a dizygotic (DZ) twinpair (87 male and 98 female pairs). Zygosity was determined by the agreementof 3 estimates: parental responses to a standard zygosity questionnaire, MTFSstaff evaluation of physical similarity, and comparison of ponderal and cephalicindexes and fingerprint ridge counts. A serologic analysis was conducted ifthe 3 estimates did not agree. Twins ranged in age from 16.6 to 18.5 years(mean ± SD age, 16.9 ± 0.6 years), mothers from 33 to 59 years(mean ± SD age, 43.9 ± 4.8 years), and fathers from 32 to 66years (mean ± SD age, 46.2 ± 5.4 years). Consistent with thedemographics of Minnesota during the years the twins were born, 98% of thefamilies were white. All twins and parents gave informed assent or consentas appropriate, and all study protocols were reviewed by an internal reviewboard.

Externalizing disorders were assessed in person at the Department ofPsychology, University of Minnesota, via structured clinical interviews administeredby trained interviewers with either a bachelor's or a master's degree in psychology.Lifetime symptoms of DSM-III-R³² alcoholand drug dependence were assessed using the Substance Abuse Module of theComposite International Diagnostic Interview.³³ Drugassessment covered amphetamines, cannabis, cocaine, hallucinogens, inhalants,opioids, phencyclidine, and sedatives. The drug class for which a participantreported the most symptoms was used as their number of drug dependence symptoms.The DSM-III-R symptoms of conduct disorder and adultantisocial behavior (the adult criteria for ASPD) were assessed using a structuredinterview designed by MTFS staff.³⁴ The twinoffspring were also assessed for adult antisocial behavior despite most notyet reaching age 18 years, as is required for a diagnosis of ASPD. This wasdone for 2 reasons. First, despite most twins being younger than 18 years,they reported engaging in a substantial amount of adult antisocial behavior,with 5.7% of male twins and 1.8% of female twins reporting 4 or more adultsymptoms (the number necessary for a diagnosis of ASPD in DSM-III-R), which is consistent with previous prevalence rates of ASPDin adult community samples.³⁵ Second, a previouswork by our group¹⁶ has shown that adult antisocialbehavior has the highest loading on a general externalizing factor and sois the best single measure of the vulnerability to externalizing disorders.

Mothers also reported on the presence of conduct disorder and substancedependence symptoms in their children using the parent version of the DiagnosticInterview for Children and Adolescents.³⁶ Asymptom was considered present if it was reported by either the mother ora twin. Previous studies³⁷ using the MTFS samplehave shown that each informant contributes unique and valid information andthat this best-estimate approach provides the most comprehensive assessmentof psychiatric symptoms. All interview data were reviewed (referring to audiotapesof the interview when necessary) in a case conference consisting of at least2 clinical psychology graduate students with extensive training in descriptivepsychopathology and differential diagnosis. Consensus between the diagnosticiansregarding the presence or absence of symptoms was reached before assigningsymptoms. The consensus process yielded κ reliabilities ranging from0.81 for conduct disorder to 0.95 for adult antisocial behavior.

Structural equation modeling was used to investigate the transmissionof externalizing disorders from parents to their offspring. In particular,we sought to determine the extent to which a general vulnerability to allexternalizing disorders could account for familial resemblance and to identifyany disorder-specific vulnerability factors (ie, effects that contributedto familial resemblance that were independent of or unaccounted for by a generaltransmission effect). A general vulnerability to externalizing disorders wasconceptualized as the latent phenotype EXT and was operationalized as thecovariance among symptoms of alcohol and drug dependence, conduct disorder,and adult antisocial behavior, that is, the vulnerability factors common toall the disorders. For each member of a family, the 4 diagnostic variableswere allowed to load on an EXT variable. General transmission effects wereoperationalized as the latent correlations between the parent and offspringEXT phenotypes. The specific transmission effects of a given disorder wereestimated by allowing the residuals (ie, the variance unaccounted for by EXT)of the parental symptom count variables to covary with the residuals of thetwins' symptom count variables. Heritability estimates for the general andspecific vulnerabilities were estimated by allowing the correlation betweenmembers of the twin pair to vary across zygosity (with greater similarityin MZ twins being indicative of a genetic effect) and by doubling the parent-to-childeffect (because children receive 50% of their genetic material from each parent).Because MZ twins would be expected to be more similar than DZ twins, for theinitial model, the latent correlation between twin A and twin B EXT was allowedto vary across zygosity.

Owing to positive skew in the distributions of the symptom count variablesthat is typical of a community sample, we used a normalizing (Blom) transformationthat has been shown to optimize model selection when analyzing psychiatricsymptom count data.³⁸ Symptom counts were usedprimarily to increase statistical power because the number of persons meetingfull criteria for psychiatric diagnoses in a community sample such as theMTFS is lower than that in clinically referred samples. Model fitting wasconducted in Mplus 2³⁹ using a robust maximum-likelihoodestimator, which is appropriate when analyzing nonnormal variables. The fitof models was evaluated using the mean-adjusted (Satorra-Bentler) χ² fit statistic,⁴⁰ the Bayesian informationcriterion (BIC) (χ²−df ln N),⁴¹ and the root-mean-square error of approximation (RMSEA).⁴² The mean adjusted χ² provides an overallestimate of model fit for nonnormal data. The BIC is a function of a model's χ² value and df, penalizing model fit for theretention of unnecessary parameters. Models that provide the best fit whileretaining the fewest parameters yield lower (more negative) BIC values. TheRMSEA provides an estimate of discrepancy in model fit per 1 df: values less than 0.08 indicate a good fit to the data, and valuesless than 0.05 indicate a very good fit to the data.⁴²

Before model fitting, we examined sex differences by testing the nullhypothesis that the correlation matrix for families with a male twin pair(n = 256) did not differ from the correlation matrix for families with a femaletwin pair (n = 286). Because there was no evidence to reject the null hypothesisthat the correlation matrices were the same (χ²₁₂₀ =144.27, P = .08, BIC = −611.2, RMSEA = 0.015),the sample was collapsed across the sex of the twin pair.

Results of the model fitting are provided in Table 1. An initial model that allowed for only general transmissionof externalizing disorders from parents to offspring provided an adequatefit to the data (χ²₂₅₅ = 491.97, BIC = −1113.3,RMSEA = 0.059). Next, we tested for any disorder-specific transmission effectsby allowing the residual variance of each parental disorder to covary withthe residual variance of the corresponding disorder in the offspring. A significanteffect would indicate that parents and offspring were more similar than wouldbe expected given only a general transmission of the disorders. To maximizeour ability to detect specific effects, we conducted 1 df tests separately for each paternal and maternal disorder. Resultsof these tests are given in Table 1 under "Specific transmission effects" and are delineated from thegeneral transmission–only model by a reduction of 1 df (from 255 to 254 df; the loss of 1 df is necessary to estimate the additional parameter) andsubsequent changes in model fit indexes. A specific effect was retained inthe model if it yielded a lower BIC value than the initial general transmission–onlymodel. We did not detect any form of disorder-specific transmission from parentsto offspring (Table 1).

We also examined whether siblings were more similar than would be expectedgiven their general vulnerability to all the externalizing disorders (ie,their similarity on the latent EXT phenotype) by allowing the residual varianceof each disorder to correlate across members of the twin pair. The same 1 df tests were conducted for each disorder. Cross-twin specificeffects were detected for conduct disorder, alcohol dependence, and drug dependence(Table 1). This finding indicatesthat there are familial vulnerability factors to these disorders that areindependent of the general vulnerability but that are not due to parentalexternalizing disorders.

Figure 1 displays parameterestimates for the best-fitting model, which entailed general transmissiononly from parents to offspring (mother and father EXT to child EXT), generaltransmission across siblings (twin A EXT to twin B EXT), and specific transmissioneffects across siblings for conduct disorder, alcohol dependence, and drugdependence (ie, the residual variances for each disorder were allowed to correlateacross twin A and twin B).

Maternal and paternal general transmission effects were estimated asthe latent correlation between parent and child EXT. The latent correlationsbetween mother and father EXT and child EXT were r =0.24 (95% confidence interval [CI], 0.14-0.34) and r =0.35 (95% CI, 0.25-0.44), respectively. Equating the parental effects yieldeda lower BIC value (ΔBIC = −4.0), indicating that there was nota statistically significant difference in the strength of maternal and paternaltransmission (Table 1). The resultinglatent correlation between parent and child EXT was moderate and representsthe general transmission of externalizing disorders from parents to offspring(r = 0.30; 95% CI, 0.22-0.38) (Figure 1).

The magnitude of the specific cross-twin effects, which represent riskfactors for a particular disorder independent of the general vulnerability,did not differ across zygosity (ie, allowing the parameters to differ acrosszygosity did not improve the fit of the model as indexed by the BIC). Therefore,parameter estimates for the entire sample are presented in the Figure 1. Disorder-specific effects were moderate for conduct disorder(r = 0.34; 95% CI, 0.25-0.42) and small for alcohol(r = 0.13; 95% CI, 0.06-0.21) and drug dependence(r = 0.19; 95% CI, 0.08-0.31). The fact that thedisorder-specific effects did not differ across zygosity suggests that theyare primarily due to environmental (rather than genetic) factors.

To ensure that the parental transmission effects did not differ by thesex of the offspring, we conducted the same tests of specific and generaltransmission effects separately for male and female twins. The results wereconsistent with those given in Table1, that is, we did not detect any specific transmission from parentsto offspring for either male or female twins. In addition, the general transmissioneffects could be equated for male and female twins without a loss in modelfit.

The design of a twin-family study allows 2 ways to estimate the heritabilityof the general vulnerability to externalizing disorders. Comparing the similarityof MZ and DZ twins yields an estimate of the broad-sense heritability (h²),⁴³ whichincludes additive and nonadditive genetic effects. Because nonadditive effectsinvolve the interaction of genetic effects, MZ twins are necessary to estimatethe broad-sense heritability. Examining the similarity of parents and offspringyields an estimate of the narrow-sense heritability (a²),⁴⁴ which includes additive geneticeffects only and is the extent to which a trait "breeds true" from one generationto the next.^43- 44

The latent correlation between mother and father EXT revealed substantialassortative mating (the tendency for like to mate with like) for the generalvulnerability to externalizing disorders (r = 0.51;95% CI, 0.41-0.61). Assortative mating tends to increase the similarity ofDZ twins, thereby decreasing the heritability estimate when comparing thesimilarity of MZ and DZ twins.⁴⁴ Despite thisfinding, there was a significant difference in the latent correlation betweenthe twins of MZ and DZ pairs for EXT, as constraining the latent correlationbetween twin A and twin B EXT across zygosity yielded an increase in the BICvalue (ΔBIC = 28.4). The latent correlation between twin A and twinB EXT was r = 0.80 (95% CI, 0.72-0.88) for MZ twinsand r = 0.31 (95% CI, 0.08-0.53) for DZ twins. Becausethe MZ correlation was greater than twice that of the DZ correlation, an h² of 0.80 provides a reasonable estimate ofthe broad-sense heritability for the general vulnerability to externalizingdisorders.^43- 44 In addition, becausethe MZ correlation is more than twice that of the DZ correlation, there isno evidence of shared environment effects on the general vulnerability toexternalizing disorders, including cultural or environmental transmissionfrom parents to offspring. In the absence of shared environmental effects,the heritability can also be estimated by doubling the parent-child effects,as children receive 50% of their genetic material from each parent.^43- 44 Using the equated parent-to-childeffect, this approach yielded a narrow-sense heritability estimate of a² = (2 × 0.30) = 0.60 for the generalvulnerability to externalizing disorders.

Last, to ensure that the latent EXT phenotype was measuring the sameconstruct for mothers, fathers, and adolescent twins, we tested a model thatequated the factor loadings across the different family members. This modelyielded a lower BIC value while maintaining a good overall fit to data (χ²₂₆₁ = 386.12, BIC = −1256.9, RMSEA = 0.042), indicatingthat the latent EXT phenotype measures an equivalent construct across familymembers; that is, the general vulnerability to the 4 externalizing disordershas the same meaning for mothers, fathers, and their offspring. Equating thefactor loadings did not change any results regarding the generality and specificityof transmission or the parent-child associations and heritability estimatesfor the general vulnerability to externalizing disorders.

This investigation sought to answer basic questions regarding the familialresemblance of 4 externalizing disorders: conduct disorder, adult antisocialbehavior, alcohol dependence, and drug dependence. Specifically, we soughtto determine the extent to which familial resemblance for externalizing disorderscould be attributed to general transmission, that is, risk factors commonto all the disorders, and the extent to which family resemblance was due torisk factors that were specific to each disorder, that is, factors that increasedrisk for only 1 disorder.

Regarding parents and offspring, our results suggest that general transmissionand not disorder-specific transmission can account for familial resemblance.This result was obtained despite an analytic strategy that maximized the likelihoodof detecting specific effects (a 1 df test for eachmaternal and paternal disorder) after controlling for the general transmissioneffect. These results indicate that rather than increased risk of a particulardisorder, what parents pass on to the next generation is a vulnerability toa spectrum of disorders, with each disorder representing a different expressionof this general vulnerability.

The general vulnerability to externalizing disorders was estimated tobe highly heritable (broad-sense h² =0.80), which is consistent with findings from twin studies^16,25 thathave explicitly modeled the genetic and environmental components of varianceof this general vulnerability. The slightly lower estimate for the narrow-senseheritability (a² = 0.60) suggests thecontribution of certain nonadditive genetic factors and underscores the importanceof including MZ twins when attempting to detect genetic effects.⁴⁵ Theheritability of the general vulnerability is greater than that reported forindividual disorders in the spectrum (h² =0.35-0.60),^{17- 21,46- 48} whichsuggests that the common variance among externalizing disorders may be themost appropriate target for molecular genetic research.

Our results, however, are not incompatible with disorder-specific riskfactors, and, in fact, such effects were detected for conduct disorder, alcoholdependence, and drug dependence but for sibling rather than parental resemblance.The fact that these specific effects were detected for horizontal (ie, sibling-to-sibling)but not vertical (ie, parent-to-child) transmission suggests that this distinctionmay be important when examining familial resemblance among first-degree relatives,as it may reconcile evidence for the generality and specificity of familialtransmission. Also, these disorder-specific effects did not differ in magnitudefor MZ and DZ twins, suggesting that they are due more to environmental thangenetic factors and that they are present after accounting for the presenceof parental externalizing disorders. This suggests that there are environmentalfactors (eg, peers,⁴⁹ neighborhoods,⁵⁰ and sibling cooperation⁵¹)that act independently of parental externalizing disorders and increase therisk of and sibling resemblance for conduct disorder, alcohol dependence,and drug dependence in adolescence.

Although we detected substantial assortative mating for the generalexternalizing vulnerability, this did not seem to affect the heritabilityestimates. Assortative mating tends to increase similarity between DZ twins,thereby decreasing the heritability estimate. However, because the heritabilityestimate cannot exceed the correlation between MZ twins reared together,⁴⁴ assortative mating will not decrease the heritabilityestimate if, as in the present case, the DZ twin correlation remains lessthan half the MZ correlation.

Finally, there was no evidence to suggest that the family transmissionof externalizing disorders differed for men and women. This is also consistentwith twin and adoption studies^{10,12- 13,15- 16} thatdo not detect sex differences in the genetic architecture and structural relationsamong externalizing disorders, despite mean level differences in externalizingsymptoms.

The present work should be interpreted in the context of its strengthsand limitations. First, we did not explicitly model and estimate componentsof genetic and environmental variance. Although multivariate twin-family modelsexist,⁵² their analytic complexity and requisitesample sizes make such models prohibitive, and, as yet, only bivariate modelshave been reported in the literature.^26- 28 Theadditional requirements of modeling the general vulnerability as a latentphenotype while also estimating the genetic and environmental components ofvariance for the general and specific transmission effects would be problematicin terms of model identification and power to detect effects; such an approachis also unlikely to yield results of any substantive difference from the currentmodel.

Regarding its strengths, this study is one of few that entailed thein-person assessment of both biological parents and a twin pair for multipledisorders. Also, because families were recruited from the community ratherthan through a clinically referred proband, the results are likely to generalizeto the population. However, future investigations will be needed to determinewhether these results also generalize to other racial and ethnic groups. Inaddition, as with any cross-sectional study, the results are limited to aparticular developmental period, in this case late adolescence. Future investigationsare needed to determine whether the same relations hold at both earlier (childhoodand middle adolescence) and later (adulthood) developmental epochs.

In conclusion, results of the present investigation indicate that mostfamilial resemblance for externalizing disorders is due to a highly heritablegeneral vulnerability. The mechanism, then, for most genetic factors thatcontribute to externalizing disorders is most likely through a broad behavioralsystem or psychopathologic process characterized by behavioral undercontrolthat underlies all the disorders in the externalizing spectrum.^{15- 16,25,53} Becausethis general vulnerability contributes to the risk of a host of disorders,a greater understanding of this broad behavioral system (its genetic architecture,developmental course, and affective and cognitive mechanisms) is most likelyto have the greatest impact on public health and the remediation of the deleteriouseffects of externalizing disorders.

Correspondence: Brian M. Hicks, MA, Department of Psychology, Universityof Minnesota, 75 E River Rd, Minneapolis, MN 55455 (hicks013@umn.edu).

Submitted for publication November 10, 2003; final revision receivedMarch 2, 2004; accepted March 16, 2004.

This research was supported in part by grants AA00175, AA09367, DA05147,and MH65137 from the US Public Health Service, Washington, DC; the Eva O.Miller Fellowship from the University of Minnesota (Mr Hicks); and traininggrant MH17069 from the National Institute of Mental Health, Bethesda, Md (MrHicks).

This work was completed by Mr Hicks in partial fulfillment of the requirementsfor the degree of Doctor of Philosophy at the University of Minnesota underthe supervision of Drs Krueger and Patrick.