The Brisbane Longitudinal Twin Study: Pathways to Cannabis Use, Abuse, and Dependence project-current status, preliminary results, and future directions.

Blood has been collected for DNA and various haematological and immunological measures. Where possible, any singleton siblings of the twins whose age was within 5 years of the twins are also recruited and tested in an identical protocol. For all studies, written, informed consent was obtained from a parent or guardian and ethics approval was obtained from the Human Research Ethics Committee at the Queensland Institute of Medical Research.

This section summarizes the statistical techniques for analyzing the 19UP Online Survey, CATI, SRQ and genome wide association data. Once phenotypic and diagnostic data become available our intention is to begin running psychometric modelling and biometrical genetic analyses before turning to preliminary tests of genome wide association. The project’s initial aim was to replicate previous findings based on psychometric modeling (latent factor, latent class and factor mixture modelling) of cannabis use and CUD phenotypes using isomorphic twin data from Virginia and Minnesota ( Gillespie, Kendler, & Neale, 2011 ; Gillespie, et al., 2007a ; Gillespie et al., 2009a ; Gillespie, Neale, & Kendler, 2009b ; Gillespie, et al., 2012 ; Gillespie, et al., 2007b ; Iacono, et al., 1999 ). Employing the same model fitting strategies in these reports, our aim is to derive and test empirical measures of cannabis use and CUD using the Australian BATS data. When combined with data from North America, tests of measurement invariance can be performed to identify sex, age and cultural differences in the point prevalence of cannabis use and CUD as well as other illicit drug use disorders (DUD). The next step will be to fit biometrical genetic models to the CUD, DUD and correlated phenotypic data in order to identify phenotypes with largest genetic variance. Multivariate twin model fitting to the twin data will then enable us to identify the sources of genetic and environmental covariance between correlated and comorbid DUD and psychiatric phenotypes. Finally, the empirically best fitting CUD and DUD phenotypes will then be used as part of preliminary tests of genome wide association involving combined SNP and genomic data available from the BLTS. Assuming between 7.8% and 15.1% of women and men will meet criteria for a lifetime diagnosis of cannabis abuse or dependence, then among the 2,100 subjects with complete phenotypic and genotypic information, between 164 and 321 individuals will meet traditional diagnostic criteria for a CUD. Analyses using the DSM-IV abuse and dependence diagnoses will be conducted in two ways: (1) all remaining subjects will be used as controls, and the GWAS analyses will make use of the traditional DSM diagnostic categories as well as an empirically derived quantitative CUD measure based on item responses; (2) only those who used cannabis but did not develop CUD will be used as controls. These complementary analyses will disentangle a potential confound arising because CUDs are conditional upon use, and the fact that for some individuals, liability to manifest a CUD remains a latent prospect because of lack of exposure opportunities. In other words, it will distinguish between those factors that influence both initiation and progression to substance use disorders from those that influence its use and progression only. Additional GWAS analyses will be based on individual maximum likelihood latent phenotypic and latent genetic factor scores calculated for the entire sample. The success of this proposed method for estimating factor scores that maximizes genetic variance has already been demonstrated by researchers at Virginia Commonwealth University with mood and anxiety disorders and related phenotypes ( Hettema et al., 2008 ). Additional analyses will model the association both at the factor level and at the level of the individual DSM-IV/V items (while accounting for the covariation among them). This approach can nest the test of association within the multivariate models that best fit the cannabis use and DSM-IV/V item level data. This will permit genome-wide tests of associations based on the best fitting CUD phenotypes that may include two or more latent phenotypes to best explain the covariation.

Data for this project have been collected from Australian twins and their non-twin siblings as part of the ongoing Brisbane Longitudinal Twin Study (BLTS) at the Queensland Institute of Medical Research (QIMR). The BLTS began in 1992 when twins were recruited from primary and secondary schools in the greater Brisbane area via media appeals and by word of mouth. Twins were ascertained along with family members as part of a study examining the development of melanocytic naevi at ages twelve and fourteen, and of cognition at age sixteen at the Queensland Institute of Medical Research (QIMR). Currently the BLTS sample comprises both adolescent and young adult twins (3,408 individuals) and their non-twin siblings (1,572), constituting 1,703 families. This includes both monozygotic (MZ) and dizygotic (DZ) twin pairs, including opposite-sex DZ twin pairs, along with singleton siblings of twins, and the twins’ parents. The proportions of twins by sex and zygosity in the sample closely mirror population expectations further strengthening our confidence in its representativeness. The BLTS recruits approximately 100 new twins per year and is now a longitudinal collection of psychiatric phenotypes, environmental and psychological risk factors, as well as neurobiological correlates and endophenotypes for psychiatric disorders that are summarized in Figure 1 . The BLTS is the core resource on which this project is based. Data collection for the project began in 2009 and is funded by the US National Institutes of Health (NIH) US National Institute on Drug Abuse (NIDA) to explore the genetic and environmental pathways to cannabis use, abuse and dependence on 1,000 BLTS twins and their siblings. These data are being supplemented with an additional 1,100 twins with funding from the Australian National Health and Medical Research Council along with seed funding from an NHMRC Australia grant to the co-author IH. Data collection is expected to be completed by mid 2013. When the current project began, the average age of the BLTS sample was 22.8 years (SD=4.3). Based on current sample size and factoring sample increases of 100 families per year, the project’s sample will include a minimum of 2,100 twins and non-twin siblings who will be 22 years or older by 2013. This age is close enough to the ages when initiation of cannabis and other drug use occurs but old enough to have passed through the maximum period of risk for onset of CUD and other DUDs ( Gillespie et al., 2009b ). The 2010 National Drug Strategy Household Survey found that the mean age of initiation of cannabis use in Australians is 18.5 years ( Australian Institute of Health and Welfare, 2010 ). Typical response rates across the BLTS projects since 1992 range from 73-85% ( Wright & Martin, 2004 ).

Since 1992, the BLTS data collection has been ongoing with twins at varying ages and stages of participation. A key strength of this project will be the ability to tie the CUD and DUD data with a phenotypically-rich, longitudinal collection of environmental and psychological risk factors, psychiatric phenotypes and diagnoses as well as neurobiological correlates and endophenotypes such as brain imaging (see Figure 1 ). In addition to measures of melanocytic naevi at ages twelve and fourteen (see Zhu et al., 2007 ), BLTS twins and siblings have been measured with the Junior Eysenck Personality Questionnaire ( Eysenck, 1965 , 1972 ) at 12, 14 and 16 years along with tests of cognition at 16 (see Wright & Martin, 2004 ). The proportions of twins by sex and zygosity in our sample closely mirror population expectations, further strengthening our confidence in its representativeness. In addition to these initial target variables, personality (most pertinently, Neuroticism) is assessed. Since 1999 we have also been collecting common symptoms of anxiety, depression and somatic distress using the Somatic and Psychological Health Report (SPHERE) ( Hickie et al., 2001a ; Hickie et al., 2001b ) at ages 12, 14, 16. The SPHERE is a 34-item self-report questionnaire incorporating items from other self-report and diagnostic instruments (Schedule of Fatigue and Anergia, SOFA ( Hickie et al., 1996 ); and General Health Questionnaire, GHQ-30) and criteria for somatisation disorder from DSM-III-R ( American Psychiatric Association, 1987 ). The SPHERE was developed and validated in two samples of primary care attenders (N=1593) ( Hickie, et al., 2001a ; Hickie, et al., 1996 ). Further data on psychometric properties have been derived from other smaller specialist samples (breast cancer, chronic pain study, older patients, post-infective patients, specialist psychiatry practice) and other general practice samples ( Bennett et al., 2004 ; Butow et al., 2005 ; Clover et al., 2009 ; Wilhelm et al., 2008 ). Adolescent-type anxiety and depression both show substantial heritability (~40%) ( Hansell et al., 2012 ). For twins at age 12 we have recently introduced the novel Verbal and Spatial Reasoning Test for Children (VESPARCH) ( Mellanby & Langdon, 2010 ) to assess verbal and spatial abilities. At 14, the twins complete a binocular rivalry task which gives an estimate of the interhemispheric attentional switching rate, a putative endophenotype for bipolar disorder ( Miller et al., 2010 ). At 16, a battery of cognitive tasks is administered including verbal and performance IQ (Multidimensional Aptitude Battery ( Jackson, 1984 ) and Wechsler Adult Intelligence Scale—Revised ( Wechsler, 1981 ), relational processing, perceptual speed, choice reaction time, working memory (delayed response task) as well as reading and language tests and academic achievement. EEG is also recorded while at rest, and during an n-back task so that event-related potentials can be recorded. At age 21, the twins return to participate in an NIH and NHMRC funded brain imaging study that includes structural and functional (at rest, and during an n-back task) MRI, as well as diffusion tensor imaging (DTI - HARDI). To date, these data (N=560 so far, budgeted for 1100 with current funding, and new funding is being applied for to expand this) are collected on the 4T imaging facility at the Wesley Hospital, Brisbane, Australia. Processing of these images is in collaboration with the University of Queensland and UCLA ( Blokland et al., 2008 ; de Zubicaray et al., 2008 ). With seed funding from the Brain and Mind Research Institute at the University of Sydney we have extended the same brain imaging protocol back to twins aged 16 and 12 years. To date, 80 twins in each age group have been collected as proof of principle. Since 2007 we have also been collecting SPHERE data at age 20+ in those undergoing MRI scanning. Currently, we are systematically collecting SPHERE data on all subjects who have participated in any phase of the BLTS. Since mid 2010 we have begun obtaining, under the NIH/NIDA Pathways to Cannabis Use, Abuse and Dependence project DSM-IV and DSM-V item level data on cannabis abuse and dependence, diagnostic data for Nicotine and Alcohol, as well as pilot epidemiological data for ecstasy and methamphetamine use. Using the same protocol described below we have also begun obtaining DSM-IV item-level data on mood, anxiety and fatigue with funding from the Australian National Health and Medical Research Council (NHMRC). Between 2009 to 2011, data collection for the NIH/NIDA and NHMRC projects began via online survey and computer assisted telephone interview (CATI) using Version 1 of the protocol. Beginning July 1st 2012, the online survey and CATI instruments were merged into an entirely online protocol (Version 2). Under both protocols described in detail below, ascertainment began with adult twins and non-twin singleton siblings from the BLTS sample in order to obtain data from individuals who had passed through the age of maximum risk for the onset of cannabis use (typically 16-18 years) and cannabis related problems (typically 19-21 years). Flowcharts of Versions 1 and 2 are illustrated in Figures 2 and 3 respectively.

In order to make data collection more cost-efficient the 19UP Online Survey and CATI were merged into an entirely online series of self-report questionnaires (SRQs) and put release July 2012. All CATI questions were converted and transferred to an online protocol which was divided into a series of 3 smaller SRQs: SRQ-1 includes all NIH/NIDA funded drug phenotypes; SRQ-2 includes a range of heritable phenotypes including migraine and headaches, inattention, hair loss and joint flexibility; SRQ-3 assesses relationships, romantic preferences, personality & internet use. In terms of risk to participants, the QIMR HREC and VCU IRB approved procedures in the CATI protocol were operationalized and re-approved for the online automated protocol. Subject ascertainment for participation in the Version 2 protocol again began with adult twins and non-twin singleton siblings in order obtain data from individuals who had passed through the age of maximum risk for the onset of cannabis use. Subjects are sent an approach letter describing the project's aims and protocols and inviting them to participate in a series of 3 online SRQs. For non-responders follow-up telephone calls are then made every 2-4 weeks to monitor progress, note reasons for refusal to participate, answer questions and offer provision to complete interview over telephone if required. The SRQ-1 included identical demographic, occupational and general health questions followed by measures of licit and illicit drug use as well as psychiatric criteria for diagnosing mental health and drug use disorders as described in Version 1 above. At the beginning of SRQ-2 and SRQ-3 subjects are required to reconfirm their informed consent. SRQ-2 assesses a range of heritable phenotypes including height, weight, hair and skin color, hair texture, hair greying and balding, handedness, sunscreen use, moliness and melanoma, acne, herpes, age of death of close family members, asthma, wheezing and prevalence of atopic conditions, use of eyeglasses, travel sickness, migraines and recurrent headaches, pain tolerance, physical activity, joint flexibility, fibroids, polycystic ovarian syndrome, endometriosis, menstruation and menopause, anorexia and bulimia nervosa, and sleep patterns. SRQ-3 assesses relationship status, romantic relationships and romantic preferences, personality and internet use. It should be noted that SRQ-2 and SRQ-3 are funded from different sources than SRQ-1.

Prior research has demonstrated that genetic and environmental factors play critical etiologic roles in the pathways to cannabis use disorder (CUD) ( Gillespie et al., 2011 ; Gillespie et al., 2007a ; Gillespie et al., 2012 ; Gillespie et al., 2007b ). Although the role of environmental risk factors is increasingly better understood ( Gillespie, et al., 2012 ) very little is known about the putative genes involved because to date there have only been two published tests of genome wide association for a CUD phenotype ( Agrawal et al., 2011 ; Verweij et al., 2012 ). This is despite the fact that cannabis use and CUD are major public health issues that have long attracted public concern and controversy in many developed countries ( Schlosser, 2004 ). Improved phenotypic measurement along with the identification of quantitative trait loci responsible for variation in CUD are still required to fill gaps in our knowledge, to develop targeted treatments as well as to provide an empirical basis for addressing policy issues and public concerns about the putative effects of cannabis use. The US National Institutes of Health (NIH) and National Institute on Drug Abuse (NIDA) 'Pathways to Cannabis Use, Abuse and Dependence' project was funded to address these empirical needs; by beginning to elucidate the genetic and environmental risk factors in the pathways to CUD. This article describes in detail the twin sample, methods and procedures involved in the data collection along with preliminary findings and plans for future research and collaboration.

For BLTS same sex twin pairs, zygosity has already been determined by typing nine independent DNA microsatellite polymorphisms plus the X/Y amelogenin marker for sex-determination by polymerase chain reaction (ABI Profiler system) and for most this has been (or will be) confirmed by GWAS.

Data collected from this population-based cohort of Brisbane twins using the (Version 1) 19UP Online Survey and CATI interviews clearly suggests that major mental disorders (notably anxiety and depressive disorders) as well as licit and illicit drug use are prevalent among young adults. We have identified high lifetime rates of alcohol, nicotine and cannabis use as well as cocaine, amphetamine-type stimulants and Ecstasy among males and females alike. We have also shown in our pilot data that for Ecstasy users, their knowledge of drug content and safety is lacking and this underscores the need for better epidemiological data to address an emerging public health concern. The project has the capacity to examine longitudinally the relationships between ages of onset of various substance misuse disorders and a range of other specific mental disorders. Central to the project’s focus, our data suggest that among lifetime cannabis users, the psychiatric criteria for CUD including craving are also prevalent (with the exception of legal problems). With data collection to be completed towards late 2013, our next step will be to model and derive an empirical CUD phenotype based on the DSM-IV and DSM-V psychiatric criteria. Currently, the sample is underpowered to determine the best fitting empirical model to explain sources of symptoms covariation. However, the patterns of twin pair correlations suggest there is likely to be significant familial aggregation for these items (including craving) that can be explained by varying combinations of genetic and shared-environmental risks. Increased samples will enable us to fit better powered biometrical genetical analyses ( Neale & Cardon, 1992 ) in the near future. As soon as empirical CUD phenotypes can be derived these can be used as part tests of genome wide association to identify QTLs for CUD. For multivariate association, we will also use methods ( Medland & Neale, 2010 ) that allow GWA tests at both the factor level and at the level of the individual DSM-IV / V items while accounting for the covariation among them. Using this approach we can nest tests of association within the multivariate models that best fit the cannabis use and item level abuse and dependence data. In other words, this will permit genome wide tests of associations based on the best fitting CUD phenotypes that may include two or more latent phenotypes to best explain the covariation between the DSM-IV item level data. The same approach, which can be applied to the remaining licit and illicit drug data, can also be used to identify pleiotropic effects across drug classes. Advances that raise our understanding of CUDs and identify QTLs responsible for drug use disorders will have an important impact on society and public health. There remains a strong empirical need for psychometrically well-defined CUD phenotypes, a comprehensive model to explain the etiology of cannabis initiation and liability to CUDs, as well as association with sufficient resolution to identify the responsible QTLs. The identification of QTL responsible for CU and CUD will begin to fill gaps in our knowledge, open the way to developing better, targeted treatments and to provide an empirical basis for addressing policy issues and public concerns about the effects of cannabis use.

All twins and siblings with 19UP Online Survey and CATI data have been or will be genotyped using the Illumina 610k SNP array. As of May 2012, ~2,639 (74% of the sample) have been genotyped using the Illumina 610k SNP array. Extensive quality control has already been performed using PLINK ( Purcell et al., 2007 ). This has included tests of Hardy Weinberg Equilibrium, analysis of missing genotype rates, inbreeding, identity by state, identity by descent statistics for individuals and pairs of individuals (to ensure that reported relationships are accurate and that distant relatives in the dataset are properly accounted for), non-Mendelian transmission in family data (when available), sex checks based on X chromosome SNPs, and tests of non-random genotyping failure. The data were then imputed to contain 2,428,106 SNPs. Imputation boosts the power of many chips towards levels obtained from hypothetical "complete" arrays containing all HapMap SNPs ( Howie et al., 2009 ; Spencer et al., 2009 ). Moreover, imputation, which is easily implemented in the software program PLINK ( Purcell, et al., 2007 ), combines information across multiple reference panels which will mean that genome-wide association study (GWAS) data obtained from different arrays such as the MFTS data ( Iacono, et al., 1999 ) can be merged for future meta-analyses. A subset of the BLTS subjects have participated in the Brisbane Systems Genetics Study funded by the Australian NHMRC grants to Drs P. Visscher and A. McRae. This is a family-based study aimed at elucidating the genetic factors affecting gene expression methylation and the role of gene regulation in mediating endophenotypes and complex diseases. To date, genome wide expression has been assessed on 870 individuals using the Illumina HumanHT-12 v3.0 413 Beadchip. We are currently in the process of assessing methylation status on about 630 of the same subjects at ~485,000 CpG sites across the genome was assessed using the Infinium HM450 & HM27 BeadChips

The United Kingdom and Ireland were traditionally the principal countries of origin for the majority of immigrants to Australia, reflecting the colonial history of the country. Since World War II (1939-1945), Australia’s population has become more ethnically diverse as people have emigrated from a wider range of countries. The proportion of residents born in other countries increased from 10% in 1947 to 24% in 2000. In 1947, 81% of new arrivals came principally from the United Kingdom and Ireland, and to a lesser extent from New Zealand, Canada, South Africa, and the United States. Although only 39% of new arrivals in 2000 came from these major English-speaking countries, people of European descent still constitute 91% of Australia’s population. Most claim British or Irish heritage, there are also Italian, Dutch, Greek, German, and other European groups. Moreover, twins and siblings from the BLTS who were recruited in South East Queensland are largely Anglo-Saxon or Anglo-Celtic background. The BLTS sample will reflect the population structure of Australia at the time this twin cohort were first recruited, and the few minority individuals who will be included in this cohort will be of predominantly Asian ancestry. The remainder of the sample will be of European ancestry. The under-representation of minorities is scientifically justified given the relative lack of genetic epidemiological data on cannabis use and related disorders.