Statistical Genetics of Substance Use (SU) and
Substance Use Disorders
Kenneth S. Kendler, MD Virginia Institute of Psychiatric and Behavioral
Genetics
Virginia Commonwealth University
Advanced Genetic Epidemiology Statistical Workshop
Oct 22, 2012
Paradigm 1- Basic Genetic Epidemiology - What Have We
Learned?
• Genetic factors play a substantial role in the etiology of Substance Use Disorders (AD).
• Heritability – the proportion of individual differences in a particular disorder or trait in a particular population that results from genetic differences between individuals.
• Heritability estimates typically in the range of 50-60%
• How does this compare to other psychiatric and biomedical disorders?
Heritability Psychiatric Disorders Other Important Familial Traits
~zero LanguageReligion
20-40% Anxiety disorders,Depression, Bulimia,Personality Disorders
Myocardial Infarction,Normative Personality, Breast Cancer, Hip Fracture
40-60% Alcohol DependenceDrug Dependence
Blood Pressure, AsthmaPlasma cholesterol, Prostate Cancer, Adult-onset diabetes
60-80% SchizophreniaBipolar Illness
Weight, Bone Mineral Density
80-100% Autism Height, Total Brain Volume
Heritability Of Psychiatric Disorders
How Consistent are the Estimates of Heritability of AD Across Space and
Time?
• Heritability is not a characteristic of a disorder – rather it is a feature of a disorder in a specific population at a specific time.
• We will look quickly at twin studies of AD and other SUDs.
Genetic & environmental proportions of variance in alcoholism estimated from studies of male twins
Clinical sample
cotwin followupPopulation registry
archival diagnosis
Volunteer registry
personal interviewFigure 2a, Prescott,
Maes & Kendler, 2005
Summary Slide
• Based on published meta-analyses or ones we did ourselves (with Joe Bienvenu) – pretty large CIs.
• Main results of non-alcohol SUDs from two studies – VATSPSUD and Vietnam Era twin study. Some reports from the Australian and Norwegian registiries.
How Consistent are the Estimates of Heritability of AD Across Time?
• Swedish Temperance Board Registration Data – 8,935 pairs of male-male twins born 1902-1949.
How Consistent are the Estimates of Heritability of AD Across Space and
Time?
• Swedish Temperance Board Registration Data – 8,935 pairs of male-male twins born 1902-1949.
• Complete birth cohort.
• Sweden underwent several dramatic changes.
• Income increased 6-fold
• Government experimented with changes in governmental control of access to alcohol.
How Consistent are the Estimates of Heritability of AD Across Space and
Time?
• In 1917, Sweden adopted a nationwide alcohol rationing system that strictly limited the amount of alcohol that an individual was permitted to purchase. An individual's official limit varied according to sex, age, and financial situations, and was, for men older than 25 years, usually between 1 and 3 L of hard liquor per month.
Interactions between gender, culture and genes – the role of social factors to constrain behavior.
• Tobacco consumption and year of birth in Swedish twins.
• Study done with Nancy Pedersen on the SATSA sample.
• Study males and females separately
0
0.2
0.4
0.6
0.8
1910-1924 1925-1939 1940-1958
Birth Cohort
Female Presence Female Heritability
Male Presence Male Heritability
Prevalence And Heritability OfRegular Tobacco Use
Three Birth Cohorts Of Men And Women In Sweden
Heritability
How Consistent are the Estimates of Heritability of SUDs Across Space
and Time?
• So, to the best of our knowledge, the heritability of AD is relatively robust – across multiple European populations living in Australia, North American and Europe and across a half century of Swedish history that saw dramatic changes in that country.
• But a quite different picture is seen for regular smoking behavior with a large gene x cohort interaction in women.
• Know much less about results for other psychoactive substance abuse and dependence.
Paradigm 2- Advance Genetic Epidemiology
• Many questions relevant to SUDs
• Begin with question of multivariate models –
• What is the relationship between the genetic and environmental risk factors for SUDs and for psychiatric disorders?
Paradigm 2- Advance Genetic Epidemiology – Multivariate Models
• Examine this question in 2,111 personally interviewed young adult members of the Norwegian Institute of Public Health Twin Panel. Statistical analyses were performed with the Mx and Mplus programs.
Panic Disorder
Major Depression
Agoraphobia
Somatoform
Disorder
Specific Phobia
Generalized
Anxiety Disorder
Dysthymia
Schizoid PD
Schizotypal PD
Avoidant PD
Dependent PD
Social PhobiaAxis I
Internalizing
Axis II
Internalizing
Axis I
Externalizing
Axis II
Externalizing
Antisocial PD Histrionic PD
Narcissistic PD
Obsessive –
Compulsive PD
Borderline PD
Paranoid PD
Eating Disorders
Drug Abuse /
Dependence
Conduct Disorder
Alcohol Abuse /
Dependence
.63
.71
.72
.80
.56
.65
.44.88
.81
.81
.67
.56
.56
.87
.73
.51
.66
.84
.88
.35
.44
.95
.37
.48
.61
.45
.49
.16 .36
.38
.23.28
Factor 1 Factor 2
Factor 3 Factor 4
.36
Paradigm 2- Advanced Genetic Epidemiology – Multivariate Models
• Replicating earlier results from our Virginia twin analyses and from the Minnesota group, SUDs are genetically part of the externalizing group of disorders.
Paradigm 2- Advanced Genetic Epidemiology – Multivariate Models
• Let’s drill down deeper into the relationship between AD and SUD to directly address the question of the specificity or non-specificity of genetic risk factors for AD.
Illicit SubstanceGenetic Factor
AlcoholDependence
CannabisDependence
CocaineDependence
NicotineDependence
AA A A
EE E E
E1
.68.82
.35.20 .31 .68
.37.27 .48 .18
.53.47 .28 .48
CaffeineDependence
A
.56
E
.14
.80
Licit SubstanceGenetic Factor
.77 .52.15
.82
AlcoholDependence
CannabisDependence
CocaineDependence
NicotineDependence
CaffeineDependence
Paradigm 2- Advance Genetic Epidemiology – Multivariate Models
• Similar to prior analyses from this sample, these results suggest that ~ 70% of heritabiity for AD is shared (this time with other drugs of abuse) and 30% unique to AD.
• For cocaine dependence, for example, 85% of total heritability is shared with other drugs and 15% is unique.
• In general, pretty clear that non-specific genetic effects outweigh specific effects.
Paradigm 2- Advance Genetic Epidemiology – Development
• Genes and environment act through time.
• Focus on alcohol intake in 1796 members of male-male pairs from the Virginia Adult Twin Study of Psychiatric and Substance Use Disorders.
• Assessed retrospectively using a life-history calendar.
NICOTINE
Paradigm 2- Advanced Genetic Epidemiology – Development
• One more developmental question –
• Do we see differential developmental changes in the impact of specific genetic risk factors for AD versus non-specific risk factors for externalizing disorders.
• Again ~ 1700 males from VATSPSUD
0
0.05
0.1
0.15
0.2
0.25
12-14 15-17 18-21 22-25 26-29 30-33
Age
Reg
ress
ion
Coe
ffic
ient
s P
redi
ctin
g A
lcoh
ol In
take Genetic Risk for
AlcoholDependence
Genetic Risk forExternalizingDisorders
Paradigm 2- Advanced Genetic Epidemiology
• Twin-family designs – ask a new set of questions.
Paradigm 2- Advanced Genetic Epidemiology
• How to capture the conditionality of genetic influences on SUDs.
• No initiation, no chance to express genetic risk.
• How to model?
• CCC model – causal, contingent, common pathway.
Paradigm 2- Advanced Genetic Epidemiology – Gene x Environment
Interaction
• Definition – the impact of genetic risk factors on disease risk is dependent on the history of environmental exposures. OR
• – the impact of environment risk factors on disease risk is dependent on genotype.
• Probably no area of psychiatric genetics research that is more controversial and artifact prone.
• A range of conceptual and statistical issues - Buyer beware!
Gene x Environment Interaction
Just show one classical example – in type I (adult non-ASPD) alcoholism from Cloninger’s Swedish adoption studies.
Risk only high in subjects at high genetic risk and exposed to high risk environment.
Paradigm 2- Advanced Genetic Epidemiology – Gene x Environment
Interaction
• Again ~ 1700 males from VATSPSUD
• Asked – would the heritability of alcohol consumption in adolescence be modified by key environmental risk factors
– Alcohol Availability
– Peer Deviance
– Prosocial Behaviors
Alcohol Availability 12-14
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
Genetic Risk AD
Dri
nks
/mo
(z)
Alc Avl=+1 sd
Alc Avl=mean
Alc Avl=-1 sd
Peer Group Deviance 12-14
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
Genetic Risk ExtD
Dri
nks
/mo
(z)
PGD=+1 sd
PGD=mean
PGD=-1 sd
Lack of Prosocial Activities 12-14
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
Genetic Risk ExtD
Dri
nks
/mo
(z)
LPSA=+1 sd
LPSA=mean
LPSA=-1 sd
Paradigm 2- Advance Genetic Epidemiology – Gene x Environment
Interaction
• Many other interesting G x E findings for alcohol use.
• A few other examples many from the work of my colleague Danielle Dick.
• One general theme – Genetic effects on alcohol use are more pronounced when social constraints are minimized and/or when the environment permits easy access to alcohol and/or encourage its use.
Gene-Environment Interaction Alcohol Use
• Marital Status (Heath et al., 1989)
• Religiosity (Koopmans et al., 1999)
• Urban/rural residency (Rose et al., 2001)
• Neighborhood characteristics (Dick et al., 2001)
• Parenting/Peers (Dick et al., 2006, 2007)
Adoption Studies
• Let’s take a short detour in to adoption studies.
• Obvious point about scientific inference
• More secure about any finding when you can reach it by multiple routes, esp when they have different potential biases.
Sample
• Follow-up in 9 public data bases (1961-2009) in Sweden of adoptees and their biological and adoptive relatives.
• Identified 18,115 adoptees born 1950-1993; 78,079 biological parents and siblings; 51,208 adoptive parents and siblings.
• DA recorded in medical, legal or pharmacy registry records.
Sources of Data
• 1. The Swedish Hospital Discharge Register included all hospitalizations (including for DA) for all people in Sweden from 1964-2009. Every record has the main discharge diagnosis and eight secondary diagnoses.
• 2. The Swedish Prescribed Drug Register included all prescriptions in Sweden picked up by patients from May 1st 2005 through 2009. It is complete, as all prescriptions are registered at the National Board of Health and Welfare.
• 3. The Swedish mortality register contained all causes of death and time of death from 1961-2009.
Sources of Data
• 4. The National Censuses provided information on education and marital status in 1960, 1970, 1980, and 1990.
• 5. The Total Population Registry included annual data on education and marital status from 1990-2009.
• 6. The Multi-Generation Register provided information on family relationships from 1932 to 2009 including all adoptions and adoptive and biological parents and siblings. Biological siblings reared with the adoptee were excluded.
• 7. The Outpatient Care Register included information from all outpatient clinics in Sweden from 2001-2009.
Sources of Data
• 8. The Primary Health Care Register included outpatient care data on diagnoses and time for diagnoses 2001-2007 for 1 million patients from Stockholm and middle Sweden.
• 9 The Swedish Crime Register included national complete data on all convictions, including those for DA, from 1973-2007.
Definition of Drug Abuse
• We identified DA in the Swedish hospital discharge, mortality, primary care, and outpatient care registers by the following ICD codes: ICD8: Drug dependence (304); ICD9: Drug psychoses (292), and Drug dependence (304); ICD10: Mental and behavioral disorders due to psychoactive substance use (F10-F19), except those due to alcohol (F10) or tobacco (F17).
Definition of Drug Abuse
• DA was also identified in the Crime Register by codes 5011, 5012 which reflect crimes related to DA. Crimes related only to alcohol abuse or to trafficking in or possession of drugs of abuse were excluded.
• DA was identified in individuals in the Prescribed Drug Register who had retrieved (on average) more than 4 daily doses a day for 12 months from either of Hypnotics and Sedatives (Anatomical Therapeutic Chemical (ATC) Classification System N05C and N05BA) or Opioids (ATC: N02A). Cancer patients were excluded.
• The 820 unique cases of DA in our cohort came from the following registries: Discharge– 527, Crime – 313, Outpatient – 264, Prescribed Drug – 118, and Primary Health Care – 8. No unique cases of DA were identified through the mortality register.
Odds Ratios and 95% Confidence Intervals for the Registration of Drug Abuse Between the Five
Registers Used in this Study
Hospital Discharge
Outpatient Primary Health Care
Drug Prescription
Crime 32.9 (32.2-33.4)
65.2 (63.9-66.5) 47.4 (41.8-53.7) 5.6 (5.3-5.9)
Hospital Discharge 118.0 (115.7-120.4)
69.8 (61.8-78.7) 20.9 (20.2-21.7)
Outpatient 94.4 (83.5-106.8)
29.6 (28.5-30.8)
Primary Health Care
37.9 (30.9-46.4)
• In this study, we could perform both adoption designs – Affected parent → adopted away offspring– Affected adoptee → biological and adoptive
relatives
• Design # 1 – – Risk for DA was significantly elevated in
adopted away offspring of biological parents with DA (OR=2.09, 95% CIs 1.66-2.62).
• Design # 2 – Risk for DA was significantly elevated in
biological full and half-siblings of adoptees with DA (OR=1.84, 1.28-2.64 and OR=1.41, 1.19-1.67, respectively).
– Risk for DA was significantly elevated in adoptive siblings of adoptees with DA (OR=1.95, 1.43-2.65).
• Next, sought to create empirical indices of genetic and environmental risk for DA in the adoptees.
• Used a multiple regression approach from which we derived an index.
• Genetic risk for DA in the adoptee was indexed by a range of features including– biological parental low educational attainment
and divorce– a parental or sibling history of
• DA, • criminal activity • treatment for psychiatric or alcohol problems.
• Environmental risk for DA was predicted by a diverse set of characteristics including – adoptive parental history of
• divorce, • premature death, • criminal activity, • hospitalization for medical or alcohol problems,
– adoptive sibling history of • DA• hospitalization for psychiatric, alcohol or medical
problems.
• Examined individually in logistic regression, where the genetic and environmental risk indices were divided into ten deciles, both risk scores were strongly predictive of DA.
• OR per decile = 1.13 for genetic risk (1.1310 =3.39)
• OR per decile = 1.10 for environmental risk (1.1010 =2.59,)
• The correlation between risk scores (a measure of assortative placement) was small (+0.11) but significant (p<0.001).
• Examined individually, DA in the adoptee was also significantly predicted by male sex, a younger AFCAP and a later birth year.
• Our key dependent variable, DA, is dichotomous. We initially used logistic regression and modelled DA as a function of the genetic risk score, the environmental risk score, sex of the adoptee and AFCAP. However a key a priori goal of these analyses was to determine if genetic and environmental risk factors interacted in the etiology of DA. We have previously argued that the scale of raw probabilities, rather than the logistic scale, is more appropriate for such analyses 16. Therefore, for our analyses of gene x environment interaction, we used PROC GENMOD in SAS 17 with the identity link and specified the variance to be binomial. We specified the effects of the explanatory variables (and the interaction term) to be additive on the scale of probabilities. All p values are reported two-tailed
• DA is an etiologically complex syndrome strongly influenced by a diverse set of genetic risk factors reflecting a specific liability to DA and a vulnerability to other externalizing disorders and by a range of environmental factors reflecting marital instability, and psychopathology and criminal behavior in the adoptive home. Adverse environmental effects on DA are more pathogenic in individuals with high levels of genetic risk.
Gene- Environment Correlation
• Better termed genetic control of sensitivity to the environment.
• Time is too limited to give details.
• My sense is that this process is of substantial importance in mediating the impact of genetic risk factors on SUDs.
• That is, in part, genes impact on risk for SUDs by increasing the chances that individuals seek out high risk environments which expose them to substances of abuse and encourage them in their use and misuse.
Paradigm 2- Advance Genetic Epidemiology
• Integrated etiologic models.
• To just get a start looking at causal pathways.
Genetic Risk
Alcoholism
Genetic Risk
Ext DisordersBirth Year
Low Church
Attendance
Alcohol
Household Use
Parental
Alcohol Attitude
Childhood Phys
Sexual Abuse
.37 .08
ADHD
Conduct
Disorder 15-17
Alcohol Use 15-17
NeuroticismSensation
Seeking
Low Parental
Monitoring 15-17
Peer Group
Deviance 15-17
Alcohol
Availability 15-17
Early Onset
Anxiety Disorder
.24 .07 .23.23 .14.13.32
.27 .09 .06 .13 .30.08
-.06.08
.12.12
.09.16
.18 .17 .27 .18 .08 .12.34 .17 .27
.13-.10
.19 .07
.14 .08 .08.15
.14 .13
.09
Symptoms of Alcohol
Use Disorders
.06
.15.30
.05-.06
-.06.26
.10.06
.08.12
.07.24
.21
.30.07
.22 .10.06
.23.37
.07 .08.06 -.08.08
.06
.08-.06
.06 .09 .06
Genetic Risk
Alcoholism
Genetic Risk
Ext DisordersBirth Year
Low Church
Attendance
Alcohol
Household Use
Parental
Alcohol Attitude
Childhood Phys
Sexual Abuse
.37
ADHD
Conduct
Disorder 15-17
Alcohol Use 15-17
NeuroticismSensation
Seeking
Low Parental
Monitoring 15-17
Peer Group
Deviance 15-17
Alcohol
Availability 15-17
Early Onset
Anxiety Disorder
.23
.12
Symptoms of Alcohol
Use Disorders
.30.05
-.06 .26.10
.06.07
.24
.21
.30 .22 .10.37
.07 .08
Genetic Risk
Alcoholism
Genetic Risk
Ext DisordersBirth Year
Low Church
Attendance
Alcohol
Household Use
Parental
Alcohol Attitude
Childhood Phys
Sexual Abuse
ADHD
Conduct
Disorder 15-17
Alcohol Use 15-17
NeuroticismSensation
Seeking
Low Parental
Monitoring 15-17
Peer Group
Deviance 15-17
Alcohol
Availability 15-17
Early Onset
Anxiety Disorder
.13 .30
.12.09
.16
.27 .12.34 .17
.14 .08.15
.14 .13
.09
Symptoms of Alcohol
Use Disorders
.30 .26.08
.23 .08.08
.06.06
Paradigm 2- Advance Genetic Epidemiology – Integrative
Developmental Model
• Evidence for two etiologic pathways characterized by genetic and temperamental factors and by psychosocial adversity.
Paradigm 2- Advance Genetic Epidemiology – Multivariate Model
for DSM-IV Criteria for AD
• Attempted to distinguish two hypotheses.
• 1. Each of the seven AD criteria index the same set of risk genes so that the diagnosis of AD is genetically homogeneous.
• 2. The DSM-IV syndrome of AD is genetically heterogeneous, arising from multiple sets of risk genes that are each reflected by a distinct set of diagnostic criteria.
– Rodent studies suggest relatively distinct set of risk genes for different alcohol-related traits.
Paradigm 2- Advance Genetic Epidemiology – Multivariate Model
for DSM-IV Criteria for AD
• Long arduous task of complex model fitting.
• 7,548 personally interviewed male and female twins from the Virginia Adult Twin Study of Psychiatric and Substance Use Disorders
• Had to take account of the fact that lots of people did not meet our screening criteria and skipped out of the alcohol section.
• This is the best fit model --
Excessive Quantity / Frequency
Perception of Alcohol Problem
Tolerance WithdrawalLoss of Control
Desire to Quit
Preoccu-
pation
Activities Given Up
Continued Use Despite
Problems
A1 A2 A3
.29.22.67 .26.51.42 .24.30.54 .54.30.33 .33.43.30 .27.50.19 .28.51.50 .44.48.33 .47.38.29
E1 E2
.17 .36 .51 .22 .67 .01 .58 .19 .43 .45 .33 .45 .63 .28 .39 .58 .28 .47 .41 .34 .36 .30 .45 .42 .34 .29 .59
ES1 ES2 ES3 ES4 ES5 ES6 ES7 ES8 ES9
Paradigm 2- Advance Genetic Epidemiology – Multivariate Model
for DSM-IV Criteria for AD
• This is the best fit model –
• Robustly supported second hypothesis – evidence for three genetic factors, which we tentatively called:
– heavy use and tolerance
– loss of control with alcohol associated social dysfunction
– withdrawal and continued use despite known problems.
8 Major Conclusions
• 1. SUDs are substantially heritable and heritability estimates for AD appear to be relatively stable across time and space. For smoking, we have evidence of potentially strong gene x cohort effects.
• 2. Roughly 2/3rds of genetic risk factors for AD and other SUDs are not-disorder specific but are shared with other forms of substance abuse and with other externalizing disorders generally.
• 3. In early adolescence, siblings resemblance for alcohol, nicotine and cannabis consumption is entirely due to environmental factors. With increasing age, we see an increasing degree of genetic influence.
• 4. For at least AD, we do not have strong evidence from GE models for parent-offspring environmental transmission.
• 5. Genes for SUDs appear to be rather substantially moderated by environmental exposures, especially those which either relax social constraints and/or permit easy access but also those that reflect adverse family environments.
Conclusions
• 6. G-E correlation is probably an important etiologic factor in SUDs. Genes can impact on SUDs via outside the skin pathways.
• 7. I presented one very rough integrated etiologic model for AD – showing how genetic/termpermental and environmental adversity pathways might inter-relate in the etiology.
• 8. DSM-IV criteria for AD appear, from a genetic perspective, to be etiologically complex reflecting multiple dimensions of genetic risk. Would we see the same for other SUDs?
Key Collaborators
• Mike Neale PhD
• Danielle Dick PhD
• Carol Prescott PhD
• Hermine Maes PhD
• Lindon Eaves PhD
• Charles Gardner PhD
• Steve Aggen PhD
• John Myers MA
• Ted Reichborn-Kjennerud MD
• Nathan Gillispie
• Jan Sundquist
• Kristina Sundquist
Support
• NIAAA including our Alcohol Research Center at VCU
• NIDA
• NIMH
• Virginia Commonwealth University’s generous support for the Virginia Institute for Psychiatric and Behavioral Genetics
• No conflicts of interest