Whole Genome Sequencing of Pedigrees With High Density of Substance Use and Psychiatric Disorders: A Meeting Report

ABSTRACT

The National Institute of Drug Abuse convened a panel of scientists with expertise in substance use disorders (SUD) and genetic methodologies primarily to determine the feasibility of performing whole genome sequencing utilizing existing pedigree collections with a high density of SUD and psychiatric disorders. A major focus was on determining if there had been any successes in identifying genetic variants for complex traits in family‐based designs. Such information could provide assurance that whole genome sequencing might provide significant pay‐offs particularly in the pursuit of rare variants and copy number variants. An important goal was to discuss and evaluate optimal strategies for studying genetic variants in human samples. Specific topics were (a) to consider whether a smaller number of cases typically available in family studies versus the larger number available in biobanks can reveal unique information; (b) to identify potential gaps in information available in biobank data that might be supplemented with family data; (c) to consider the optimal SUD phenotypic definitions (e.g., quantity of use, problem‐oriented) and data collection instruments (self‐report or clinician administered) that are both practical and efficient to collect, and likely to provide important insights concerning prevention, intervention, and medication development. Conclusions reached by the panel included optimism about the successes that have occurred in the existing family studies ascertained to include densely affected pedigrees. Evaluation of methodologies led, overall, to a panel consensus that steps should be taken to utilize biobank collection in conjunction with family‐based investigations for optimal variant discovery.

Short abstract

NIDA convened a panel with expertise in substance use disorders and genetic methodologies to discuss whole genome sequencing utilizing existing pedigree collections with a high density of substance use and psychiatric disorders. The recommendation was to utilize biobank collections with family‐based investigations for optimal variant discovery.

1. Introduction

On April 25, 2023, the National Institute on Drug Abuse (NIDA) convened a panel of scientists with expertise in genetics and substance use disorders (SUD) to discuss the utility and implications of performing whole genome sequencing (WGS) in pedigrees with a high density of substance use and psychiatric disorders. The meeting was initiated with presentations by investigators with pedigree data from densely affected pedigrees with SUD. These included presentations by Howard Edenberg, PhD, from the Collaborative Study on the Genetics of Alcoholism (COGA), “A Family‐based Study for Gene Discovery and Understanding,” Shirley Hill, PhD, “The Pittsburgh Multiplex for AUD family study: Results of Family‐Based Association Analyses,” and Thorgeir Thorgeirsson, PhD, “Families with Multiple cases of Psychiatric Disease and SUDs: Whole Genome Sequencing and Rare High‐Impact Variants.”

Other presentations focused on methodological issues with presentations by Aiden Corvin, MB, PhD, TCD “Family‐Based whole Genome Sequencing: Lessons Learned to Date,” Jennifer Below, PhD, “Distant Relatedness‐based Identification of Variant Effects (DRIVE),” and David Goldman, MD, “Recurrent CNVs in Families and Founder Populations.” Additionally, presentations were given by Suzanne Leal, PhD, “The Power to Detect Associations for Complex Traits Using Family‐based Data,” Scott Vrieze, PhD, “Rare Variants and Complex Behavior in Pedigrees: Recruiting Participants Based on Phenotype or Genotype,” Laura Almasy, PhD, “Genetic Underpinnings of Comorbid Polysubstance Misuse and Major Depression in Mexican American People.” A Discussion session included Nancy J. Cox, PhD, Mark Daly, PhD, Ben Neale, PhD, and Huda Zoghbi, MD.

Among the specific questions that the group was asked to consider was whether there had been any successes in identifying genetic variants using family‐based designs, and if so, what pedigrees already exist that could be tapped for further analyses. Additionally, because the existence of comorbid psychiatric conditions in some of the existing pedigrees may influence identification of variants, the group considered how this may influence detection. Another important question for consideration was how extended pedigree data may influence what can be discovered. Should we consider focusing on having complete trios (two parents and an offspring) for all families sequenced or possibly fewer families but some with extended pedigree members across multiple generations? Finally, the issue of genome wide association studies (GWAS) versus family‐based analyses as a method for identifying genetic variants was discussed. The pros and cons of each strategy was also a planned consideration.

A major focus of our discussion was determining if there have been any successes in identifying genetic variants for complex traits in family‐based designs. Presentations by members of the panel provided examples for discussion. A related issue was discussion of how to define the phenotype of interest. Both quantity of consumption and problem‐oriented diagnoses have been used to describe phenotypic variation in SUD. Problem‐oriented diagnoses tend to be binary, classifying individuals on the basis of presence or absence of SUD although symptom counts have also been used. Several studies have found genetic distinctions between level of consumption and criteria‐based disorder phenotypes [1, 2].

Some pedigree collections are unselected for presence of other psychiatric disorders and contain considerable comorbidity with commonly occurring comorbid conditions such as major depressive disorder and anxiety disorders while others have selected families to reduce comorbidity. Currently, it is unknown whether the genetic underpinnings of narrow phenotypes without comorbidity differ from broader phenotypes that include not only a SUD but also presence of one or more comorbid psychiatric conditions.

Several questions were posed for discussion. Are family‐based analyses better for identifying rare variants and copy number variants (CNVs) than GWAS? This question was expanded to consideration of appropriate methods for genetic and genomic discovery, their limitations and their potential for translation and potential usefulness with respect to treatment. A search for mechanisms associated with addiction including those having to do with craving was considered an area where significant gaps in knowledge appear to exist.

The group also considered whether there had been successes where comorbid psychiatric disorders are present. If pedigrees with a high density of substance use and psychiatric disorders exist, what is the likelihood that genetic variants will be discovered using short or long‐read sequencing approaches? What is the sample size needed to detect rare variants and do these differ when assessing family data or cases and controls?

The purpose of this report is to summarize the presentations and discussions from this conference focusing on these and other questions. If it appears that family‐based methods have been successfully used to uncover important genetic variations in existing densely affected family collections, what is the feasibility of utilizing these collections? Among the relevant questions for future research is the extent to which data across samples from pedigree studies can be meaningfully merged. Harmonization would require similarity in ascertainment schema, phenotypic assessment, and ethnic diversity. These issues are relevant for studies of samples from individuals as well as families. Merging of data across studies with varying ascertainment schema can be expected to increase heterogeneity that may result in reduced power to detect genetic variants. Some support was offered for utilizing both pedigree data and large biobank data sets together. However, a major question was how to integrate large‐scale studies from biobank data with family designs.

Also addressed were the ethical issues involved in sharing of data. To what extent does the issue of privacy and confidentiality of participant information come into play when considering the wider distribution of data needed to accomplish the goal of larger scale family‐based analyses? Does risk for re‐identification increase with family data, a concern that is especially relevant to family members who did not participate in the original data collection and, therefore, have not signed consents to share their data?

1.1. Status of Variant Discovery Using Family‐Based Methods

1.1.1. Studies Targeting Alcohol Use Disorder and SUD

The following is not an exhaustive list of successes using family‐based methods but rather an overview of findings presented at the meeting illustrating successful variant finding using modest sample sizes but targeting families that could be classified as having members with extreme phenotypes based on their membership in densely affected families with multiple cases. Two studies, the COGA [3, 4] and the Pittsburgh Multiplex Study are discussed as examples of family‐based studies in which ascertainment strategies were used to increase the likelihood that extreme phenotypes would be found.

The COGA has been in existence for 34 years producing over 600 publications (https://cogastudy.org) based on study of densely affected families and comparison control families. The initial design was selection of a proband in treatment for alcoholism as defined by DSM‐III‐R or DSM‐IV, which in turn led the investigators to members of the probands' families. All adult family members willing to participate were assessed using the Semi‐Structured Assessment for the Genetics of Alcoholism (SSAGA), a structured psychiatric interview. In order to achieve a densely affected set of families, investigators chose those that had a minimum of two individuals with problematic alcohol use defined by the Diagnostic and Statistical Manual (DSM) criteria. Since that time, gene finding efforts to identify variants associated with alcoholism, now known as alcohol use disorder (AUD), have been expanded to include follow‐up of third generation offspring. Data collection has been multimodal throughout. This has included extensive genetic inquiry, development of a large database of electrophysiological recordings, clinical interview data including familial environmental factors and age of onset to begin use of substances and including longitudinal assessment of the third generation [4, 5, 6]. This data collection would appear to be ideal for WGS given the numerous genetic findings reported from this effort. These genetic findings are in different domains including SUD and its age of onset, comorbid conditions, electrophysiological phenotypes, the influence of family environment, and developmental differences [4, 5, 6].

The Pittsburgh‐based multiplex family study was initiated in 1986 drawing on individuals in treatment for alcoholism who reported a minimum of two affected male siblings. Early reports focused on cognitive aspects of individuals with AUD in contrast to their unaffected siblings using event‐related potentials (ERP) [7]. Personality variation was reported showing greater similarity in affected pairs of adult siblings than discordant pairs [8]. These early studies pointed to the need to assess younger members of families with AUD segregating within them to potentially uncover endophenotypes and genetic variants before the onset of heavy alcohol and drug use that would complicate interpretation of findings [9]. Subsequently, a study of children was initiated in 1990 that provided longitudinal follow‐up including clinical interviews, ERP assessments and collection of MRI scans. Follow‐up of offspring has revealed that one component of the ERP, the P300 potential, predicts SUD outcome by age 22 [10]. Clinical characteristics of the third generation followed through childhood, adolescence and young adulthood has revealed differences in those with membership in these multiplex families identified through a pair of male AUD probands [11]. Similarly, brain morphology differences using MRI has been reported in a series of publications [12, 13, 14]. Inclusion of environmental effects, whether prenatal or postnatal or cross‐generational, has been a focus of some of these studies that have also included evaluation using epigenetic methodology [13, 14, 15].

In recognition of the fact that etiological factors, both genetic and environmental, may vary by gender, a second Pittsburgh family collection was begun with ascertainment of families through a minimum of two AUD affected adult female siblings. As was the case with the family collection identified through two adult male siblings with AUD, all family members identified through two female siblings who were willing to participate were assessed with structured psychiatric interviews that assessed all DSM Axis I diagnoses using current state of the art diagnostic interviews, namely the Diagnostic Interview Schedule (DIS), a forerunner of the SSAGA.

Genetic variation has been assessed using linkage and family‐based association analyses along with exome sequencing in a subset of individuals from the family studies. Results of several family‐based analyses have been reported which have revealed the utility of these densely affected families for gene finding. A few of these were discussed including ASTN1, ACN9, KIAA040, and CHRM2. One of the genes of special interest to the panel was the ACN9 gene, a gene involved in glucogenesis, the process whereby glycogen stored in the liver is converted to glucose for energy utilization. The relevance to AUDs was discussed based on craving for food, which is triggered by low glucose and possibly involved in craving for alcohol.

In summary, the Pittsburgh study is rich with endophenotypic assessments including over 3000 ERP recordings across three generations, MRI data for child and young members of the 3rd generation, and repeated assessment of environmental variables in the longitudinally followed children. Genetic variation has been assessed across the three generations for individuals from the two pedigree studies revealing promising genes for follow‐up.

Taken together, the multitude of findings from the COGA and Pittsburgh Multiplex Family Study suggest that densely affected families selected for presence of AUDs and related SUD have the potential to reveal variants of potential consequence for better understanding of the etiology of these disorders. Whether these families can reveal reproducible rare variants has not yet been tested but there are promising signs. Exome sequencing in search of rare variants within exomes of five nicotinic receptor genes that utilized COGA samples found increased DSM IV alcohol dependence and cocaine dependence symptoms in association with a missense variant in CHRNB3 [16]. A recent report from the Pittsburgh Multiplex Family Study using exome sequencing of one proband per family versus a much larger control data set has revealed a number of suggestive rare variants for future follow‐up including a significant finding for a rare variant in a nicotinic receptor CHRNA9 [17].

1.2. Other Family‐Based Analyses With Identification of Rare Variants

Family‐based analyses with successful identification of rare variants using WGS were reported by Aiden Corvin, MB, PhD, and Thorgeir Thorgersson, PhD. Successful identification of recurrent CNVs was reported by David Goldman, MD.

Studies of genetic variation in schizophrenia have been assessed using GWAS methodology but collectively the 270 common variants found explain only a small proportion of risk [18]. Using the all ancestry GWAS results, the proportion of variance in liability for all SNPs was 0.073 and even less when genome‐wide SNPs were considered (0.024). The Schizophrenia Exome Sequencing Meta‐analysis (SCHEMA) study has confirmed association with ultra‐rare variants at 10 genes [19]. This and other studies indicate an excess burden of rare protein truncating variants in schizophrenia cases compared to controls, evident across diverse human ancestries [20], and suggesting that there are many more ultra‐rare variants to be found. Corvin and colleagues [21] conducted a family study using WGS of 35 individuals across 6 pedigrees that revealed deleterious missense variants in three genes (ATP2B2, SLC25A28, GSK3A) that co‐segregated with disease status in three of the six pedigrees.

Significant overlap in risk for many psychiatric illnesses with SUD [22, 23] has been observed in large data sets using GWAS methodology. Overlap between SUD and physical illness based on the same genetic variant has also been shown in studies using GWAS methodology [24]. As noted in the presentation by Dr. Thorgeirsson that focused on families significant comorbidity between psychiatric disorders and SUD is often observed. This was demonstrated in their study of body mass index (BMI) and nicotine addiction [25] that suggested a common biological basis for both, a result seen by others who also found differences in nicotine metabolism using a Mendelian Randomization approach [26]. Enrichment of SUD among relatives diagnosed with psychiatric diseases such as schizophrenia and depression may reflect overlapping genetic risk. For example, the polygenic risk score (PRS) for schizophrenia correlates significantly with risk for other psychiatric disorders [22], including SUD [27]. Also, overlap has been seen for one rare variant, a truncating mutation in RBM12 for psychosis and SUDs [28].

Because comorbidity between various SUDs has been shown to have common genetic underpinnings addiction has been discussed as a unitary genetic vulnerability [29], suggesting that the overlap is biologically based. Other investigations in unselected groups of participants have been used to identify variants associated with more than one SUD [30] or with other psychiatric disorders [31]. If comorbidity across all psychiatric conditions including SUD can be demonstrated to have common genetic underpinnings, removal of comorbid cases in the selection of cases for study may be considered as counterproductive. However, alternative scientific possibilities exist. SUD can be an attempt to self‐medicate depressive or other psychiatric symptoms [32] though not necessarily indicative of a common genetic diathesis. Similarly, chronic use of alcohol can lead to symptoms of depression though not necessarily MDD. Increased use of both nicotine and caffeine has been observed in individuals with schizophrenia and bipolar disorder [33] though it is unlikely that use of both nicotine and caffeine among individuals with a psychiatric disorder indicates a common genetic diathesis for all conditions. Moreover, SUD within families is influenced by behavioral modeling of family members [34] so that apparent biologically based comorbidity for SUD may reflect an indirect effect and not evidence of a molecular genetic basis for the overlap.

Accordingly, some investigators have viewed comorbidity as a potentially complicating factor in identification of rare variants specifically associated with genetic risk for particular SUDs. As a result, various strategies have been used to study cases where there is minimal comorbidity to uncover mechanisms unique to SUD [17]. It is noteworthy that a large‐scale analysis of UK biobank data investigating comorbidity between various psychiatric disorders and between psychiatric disorders and physical disorders found that the statistical strategy used to investigate comorbidity had a major effect on whether the disorders could be considered as having overlapping genetic risk [35]. Specifically, disease‐disease networks identified by pairwise methods offered differing solutions than those identified using a Bayesian direct multimodal map (BDMM) technique, the latter providing evidence that while some disease‐disease associations reflect genetic overlap implicating common molecular mechanisms, many of the comorbidities thought to reflect an underlying molecular mechanism were indirect effects. The underlying true morbidities are not known at study initiation. Consequently, investigators have taken varying approaches to inclusion of comorbidities as they exist within the population utilized while others have chosen a highly selected ascertainment schema to address the molecular basis of SUD.

2. Design Considerations

As reviewed previously, consideration of the population under study is an important aspect of what may be revealed by genomic inquiry. Comparison of available power between case–control and family‐based studies have shown equal power in some situations but not others. It has been shown that the family structure available for testing is influential in the power that can be achieved [36]. Specifically, 200 cases and 200 controls were compared with (a) 200 trios with an affected offspring and two parents, (b) 200 discordant sibling pairs (one affected and one unaffected) without parents. Results showed discordant sib‐pairs were 50% less powerful than case–control designs. For common diseases, the case–control comparison was slightly more powerful than the trio design and designs with three discordant sibs. Importantly, for rare diseases the calculations showed that the trio design was more powerful than case–control designs. These calculations suggest that the search for rare variants can be achieved with good power utilizing both family‐based and case–control designs [36].

Amplifying our discussion of design considerations with implications for statistical analyses and power to detect rare variants were presentations by Suzanne Leal, PhD, and Jennifer Below, PhD, Laura Almasy, PhD, and Scott Vrieze, PhD. These presentations discussed the challenges involved in identifying the most useful pedigrees that could reveal rare variants with high impact, while noting that novel methods have been utilized that can account for comorbid disorders, utilize distant relatives with genetic sharing, model background polygenic risk, and include ethnically diverse samples.

2.1. The Power to Detect Associations for Complex Traits

Among the topics considered was the power to detect associations for complex traits such as SUD using family data. The presentation by Suzanne Leal, PhD, included discussion of the power to detect associations between rare variants and complex traits from studies of single case ascertainment rather than families to provide background on what can be expected with family data. Highlights of her work include a meta‐analysis of whole exome sequencing (WES) data from GWAS collections from 22,004 individuals that included complex anthropomorphic phenotypes including height, BMI based on height and weight, and based on the waist/hip ratio (WHR). These phenotypes were analyzed for single nucleotide variants (SNVs) and gene‐based associations [37]. Results were tested for replication in up to 16,418 individuals from 10 cohorts. The authors concluded that there was a dearth of findings from their analyses and suggested even larger sample sizes may be needed. However, it was also noted that to combine the effects of rare variants and reduce the burden of multiple testing, candidate gene sets based on previously published GWAS loci did enable the team to identify functional genes in two loci associated with height.

2.2. Distant Relatedness‐Based Identification of Variant Effects (DRIVE)

Presentation by Jennifer Below, PhD, “Distant Relatedness‐based Identification of Variant Effects (DRIVE)” offered a perspective on how individuals beyond the nuclear family can be utilized to uncover genetic variants associated with complex traits. Her work has included studying whether genetic contributors to traits within families also contribute to risk in the general population. Often variants in genes implicated in clinically ascertained family studies of specific traits show incomplete penetrance in population samples leading some to dispute the evidence of their effects as has been the case of the cardiac trait, Type 5 Long QT Syndrome (LQT5) [38]. In other cases, variants in genes implicated in clinically ascertained family studies do not explain a large proportion of the population level trait variance as in the case of developmental stuttering [39]. The absence of overlapping genomic regions in family studies and GWAS investigations suggest that for some traits the genetic architecture uncovered in families that are highly enriched with cases may be distinct from that seen in the general population. This observation also suggests that common and rare variants may act together to elevate risk. Because common and rare variants most likely act in concert in most complex disorders, studying families along with population‐based samples within the same investigation appears to be a potentially valuable plan. Recently developed approaches such as PRS informed linkage analysis in families with high prevalence of SUD and enrichment of SUD networks of individuals who share identical copies of a gene (e.g., the DRIVE) approach [40] may improve power for mapping rare causal variants with incomplete penetrance.

2.3. Genetic Underpinnings of Comorbid Polysubstance Misuse and Major Depression in Mexican American People

The presentation by Laura Almasy, PhD, “Genetic Underpinnings of Comorbid Polysubstance Misuse and Major Depression in Mexican American People” illustrates how substance use and psychiatric disorders, particularly depression, become intertwined often complicating description of the phenotype of interest. Using univariate and bivariate analysis of depression and cannabis use from data collected from 1284 Mexican‐American individuals from 75 extended pedigrees, the investigators found significant univariate peaks for depression and cannabis use, and a pleiotropic peak influencing both depression and cannabis use on Chromosome 11q23, a locus previously linked to each of these phenotypes [41]. These results highlight the importance of deep phenotyping in family‐based studies to collect information not only on the SUD phenotype of interest but also comorbid conditions that may influence results.

2.4. Rare Variants and Complex Behavior in Pedigrees: Selecting Samples on Phenotype or Pedigrees

The presentation by Scott Vrieze, PhD, raised the issue of sample selection and whether it is more useful to collect a large number of individual samples or focus on pedigrees. In his research, an emphasis has been placed on participating in large scale sequencing efforts with a recent study including 3.4 million individuals from four major global clines [42]. In this presentation, Dr. Vrieze suggested that a pedigree sampling design may not increase the power for rare variant discovery for complex behaviors such as addiction. As an alternative, he suggested utilizing population‐based samples with GWAS methodology, noting that such efforts have revealed thousands of loci associated with a range of addiction measures. These loci could serve as a starting point for rare variant discovery. Within these loci, rare putative loss of function variants may be found that could provide strong evidence for an association with addiction within existing databases (e.g., UK Biobank, FinnGen, All of Us). He noted that a current limitation of biobank data is its reliance on self‐report measures. Consequently, a strategy was proposed in which a genotype call‐back would be implemented in which carriers would be contacted to enlist their family and to conduct in‐depth assessment. Such an approach attempts to leverage the relative strengths of large scale biobank data and family studies as a cost‐effective alternative to recruitment, sequencing and detailed phenotyping of the extremely large number of individuals required for a naïve genome‐wide rare variant association discovery.

Following the presentations of results of these studies, a group discussion provided suggestions for how future efforts could be coalesced to provide rare variants that might implicate specific mechanisms related to addiction that are reproducible. The following is organized around the major questions that emerged from the discussion.

3. Rare Variant Discovery—Implications for Etiology and Treatment

In a discussion led by Nancy Cox, PhD, Mark Daly, PhD, Benjamin Neale, PhD, and Huda Zoghbi, MD, several issues were raised. While it was generally agreed that rare variants may not be useful in prediction of which individuals will develop an SUD, they are especially useful for understanding mechanism. Focus on coding regions using WES was viewed as a means for identifying the relevant genes much more quickly and efficiently. However, WGS would provide an opportunity to investigate noncoding regions and mitochondria. The panel concluded that due to the existence of so many impressive pedigrees that have already been collected with a high density of SUD, uncovering additional variants within these families currently is a knowledge gap that could be addressed. Also, it was noted that creative use of biobank data has the potential to reveal rare variants. The goal of these studies should be to identify rare variants associated with the extreme phenotype, which may represent as few as 1% of the population. One suggestion for doing this would be to identify children in high risk families with genetic variants associated with disorders that are prodromal to development of AUD and SUD. Another suggestion was to focus on genes of known importance to metabolism. A glucose metabolism gene, ACN9, has been revealed in studies of AUD [43, 44]. The interaction between metabolism and mitochondria functioning was also noted with a suggestion that sequencing should routinely include the mitochondria. A challenge for completing such studies is the sample size that can be achieved. It is important to pursue variants for which we have confidence based on their discovery in well powered samples. The discussion emphasized the need for recognition of the complementary nature of sequencing families as a component of an overall approach in which these results are then cross‐referenced to large‐scale efforts.

4. How to Measure the Phenotype

The genetic factors associated with quantity of alcohol consumption or quantity of drug use may be quite different than the underlying genetic factors that are associated with problem‐oriented definitions of AUD and SUD. Use of a problem checklist, the Alcohol Use Disorders Identification Test (AUDIT), has often been the method of assessment for large scale GWAS studies of AUD that utilize data from existing resources such as the Million Veterans Study, the UK Biobank study and respondents in the 23 and Me databases [2, 30, 45, 46, 47]. As first noted by Caetano [48], the development of the AUDIT by the study originators sought to provide a quick and efficient interview method, though limited in breadth in comparison to psychiatric interviews available at the time, such as the (World Health Organization DIS or Composite International Diagnostic Interview [CIDI]). The AUDIT though widely used is not a clinician‐administered inquiry and was not developed to be based on DSM criteria though it taps some of same constructs. A further limitation is that it is designed to assess problems in the past year. Although the results of analyses of the AUDIT are limited to past year substance use, and other psychiatric comorbidity is largely unknown, a number of variants have been identified [1, 2]. Similarly, the AUDIT‐C has been used to assess quantity of alcohol consumed. A problem‐oriented definition of AUD and SUD that takes a lifetime perspective may provide greater phenotypic depth than symptom checklists but both would appear to provide greater depth than consumption measures alone. While there is a moderate correlation between quantity of alcohol consumed and problems resulting from alcohol use the association is clearly not perfect with each having unique capability as predictors of future behavior or consequences of use [1, 2, 49]. Recent reports have demonstrated that the genetic underpinnings of alcohol consumption and AUD as measured with the AUDIT show overlap but also unique variants related to each were identified [31, 50].

Cross‐inheritance has been analyzed using the National Epidemiological Survey on Alcohol and Related Conditions (NESARC) interview data combined with exome sequencing to address transmission in four SUDs: AUD, nicotine use disorder (NUD), cannabis use disorder (CUD), and nonmedical prescription opioid use disorder (NMPOUD) [51]. Polygenic scores (PGS) were calculated within the European Ancestry (EA) portion of the study that included 12,505 individuals for the four SUDs and a control group comprised of “super‐normal controls.” The study performed in‐person interviews with the AUDADIS‐5 instrument that provides DSM V diagnoses. The AUDADIS‐5 offers simplicity of administration but is not directly comparable to widely used instruments such as the DIS or CIDI or the SSAGA developed for the COGA study. The SSAGA was developed in collaboration with investigators at Washington University where the DIS and CIDI interview methods were developed, allowing inclusion of items in the SSAGA from well‐validated items from the DIS [52]. Utilization of the SSAGA with extensive item commonality with the DIS and CIDI can increase the potential for harmonization of data across studies.

Results of NESARC, a nationally representative sample that specifically examined types of disorders, relating them to exome sequencing [53], point to the importance of examining cross‐inheritance for improved understanding of etiology that can provide clues concerning medication development. Results of the analyses revealed both shared and nonshared genetic contributions to the pathogenesis of SUDs.

An additional consideration is whether defining the phenotype based on alcohol or drug use behaviors alone captures the full picture of why some individuals escape addiction following use of these substances and others do not. The Research Domain Criteria (RDoC) approach has been considered as an alternative to definitions that rely on either quantity of use or problems resulting from use. Goldman and colleagues examined three functional domains thought to be important in the addiction cycle: Negative Emotionality, Executive Functioning, and Incentive Salience, identified through factor analysis in the deeply phenotyped NIAAA sample of AUD and non‐AUD patients [53]. These three clinical addiction factors showed a parallel with the three main RDoC domains and major personality structures suggesting the similarity in cross‐generation inheritance of trait‐like conditions that predispose individuals to development of psychiatric disorders and SUD. As noted by Dr. Goldman, use of these RDoC traits for investigating the genetic underpinnings of SUD though offering a more detailed view of the phenotype has limitations because the trait‐like conditions may be, in fact, state conditions due to the consequence of exposure in addition to being causally related to the development of the disorders.

4.1. The Value of Deep Phenotyping and Measures of Exposure

As emphasized in the presentation by Dr. Goldman, an important question for genomic studies of AUD or SUD, whether family‐based or population‐based, is the value of deep phenotyping and inclusion of measures of alcohol and drug exposure. As the conference revealed, the COGA sample, NIAAA sample, and the Pittsburgh Multiplex Family study have deep phenotyping based on in‐person comprehensive psychiatric interviews, psychophysiology including EEG and ERP data and environmental exposure data. The value of the EEG and ERP phenotypes include previous demonstrations of their usefulness in gene‐finding efforts and their relationship to neuronal excitability in the brain [54, 55, 56] and the capacity of ERP variants to predict SUD outcome in youth [10]. Whether all of this deep phenotyping will ultimately prove to be useful in uncovering genetic loci that could not otherwise be revealed is open to debate. However, if the data have already been collected, it is hard to argue that we are not better off using it. Conversely, those who argue that millions of data points are necessary to achieve power to detect important variation might conclude otherwise. However, ascertainment strategies that favor more extreme phenotypes can alter the imbalance in available cases seen in family‐based and population‐based samples to improve power. This has been the strategy used by both the COGA and Pittsburgh studies.

Also discussed by Dr. Goldman were the advantages and disadvantages of family studies and founder populations. Loci detectable in family studies can be expected to be alleles of large effect relative to the measured phenotype. Unfortunately, signals uncovered in GWAS investigations may not be detectable in family‐based studies if the smaller number of individuals collected from the family studies results in lower power. However, data from families are likely to show significance when following up on a GWAS signal from a larger study, such as those based on biobank samples, and allow dissection of the discovered variant obtained from a larger sample that may be more heterogeneous. Nevertheless, family‐based designs offer greater opportunities for controlling environmental exposures and genetic background. Within families and founder populations sufficient instances of rare and uncommon variants can be collected to enable detection of within family associations with good power. An additional feature of family studies not often recognized is the potential to overcome the problem of within‐gene genetic heterogeneity such that families representative of different functional loci at the same gene can be combined to generate significant LOD scores. Notably in large case/control studies there are also potentially powerful ways (SKAT‐O) to co‐analyze multiple putatively functional loci. On the whole, families and founder populations continue to offer benefits for circumventing genetic heterogeneity and for homing in on functional consequences of rare and uncommon variants detected by sequencing.

Discussion of this situation prompted concern about how to incorporate biobank data, which often utilizes quantity of consumption measures exclusively rather than problem oriented definitions into their data sets. Although ICD codes are present in some datasets, assessment is performed by individuals of varying levels of behavioral and psychiatric training such as general practitioners in health care settings. Whether this variation in methodology between large‐scale biobank data and deep phenotyping available from smaller family‐based collections can be successfully merged across types of datasets is currently unknown.

5. Diagnostic Algorithms for SUD and Psychiatric Disorders

Some discussion of the limitations of DSM nomenclature was raised. Comments were offered concerning the binary nature of diagnoses made using DSM. Genetic analyses can be improved with quantitative phenotypes that allow severity to be included in the phenotypic characterization. At least one investigation has taken this approach [47]. Suggestions were offered concerning the value of moving toward RDoC [57] definitions that might be able to characterize the phenotype more precisely. A related issue with respect to individual phenotypes is the families from which the individual emanates and the extent of comorbidity within families included within the study sample. Some studies have not restricted the presence of other comorbidity, preferring to study all available pedigrees irrespective of the loading for other psychiatric disorders. This approach assures a more representative sample of the population but may limit gene finding where genes unique to each disorder may have epistatic effects within the same individual carrying both disorders. An approach consistent with epidemiological methods that strives for a representative sample differs from laboratory‐based methodology where, for example, animal studies are designed to reduce as much heterogeneity as possible limiting each experiment to one species, one strain, and even including only animals from the same vendor. Findings discovered under tightly controlled conditions are then examined in other species. Clearly, there are advantages and disadvantage of both approaches from the standpoint of studying complex disorders in human samples.

Some of the work presented emphasized inclusion of families with multiple psychiatric disorders (Dr. Thorgeirsson). The COGA study has pursued a similar strategy including families with an array of psychopathology but insuring density of SUD cases. Unlike the COGA study, the Pittsburgh Multiplex Family Study took an approach more similar to laboratory‐based experimentation where purity of type is the goal. Families densely affected with bipolar disorder, recurrent major depressive disorder or schizophrenia were excluded. Because the study was focused initially on AUD, probands were required to have developed AUD prior to developing SUD by at least 1 year. Because of differences in ascertainment, harmonization across families for new sequencing may need to be considered.

6. Considerations of Mechanism

The group was in agreement that risk for developing a SUD is best characterized as a polygenic risk. The issue then becomes how to define what is meant by susceptibility or proneness to addiction. The panel discussed the close overlap between susceptibility to obesity and to AUDs. This has been shown in the tendency of individuals who have previously undergone bariatric surgery to later turn to heavy use of alcohol [58, 59]. This observation led to the suggestion that possibly we should consider expanding the concept of addiction to focus on reward systems biology [60]. Individuals with multiplex AUD background show genetic influences [61] for CHRM2, DRD2, and FTO, variants previously seen in population‐based samples that have searched for genes conferring risk for increased BMI [62]. Importantly, although most studies of BMI focus on adults, child/adolescent cases from multiplex for AUD families were found to differ significantly in BMI from controls [61] by having greater BMI and showing genetic variation associated with risk for AUD. Results of these studies underscore the value of focusing on metabolism as one important direction in the search for rare variants influencing addiction. Broadening the search to include phenotypes associated with reward system biology may prove to elucidate mechanisms for both SUD and obesity that can have valuable treatment potential.

7. Genes Versus Genes × Environment

Large scale investigations while well powered for finding common genetic variation in the case of GWAS studies are often less able to provide the environmental variation that may be needed to detect important gene × environment interactions. Harmonization of many smaller investigations with in‐depth environmental measures may provide the opportunity for understanding important gene by environment interactions. Panel members offered several examples of how individuals with the same genetic variant differed in SUD outcome suggesting the importance of environmental factors. Discovery of the relevant environmental factors is a challenging part of rare variant inquiry in which the variant may be impactful only in the presence of the particular environmental variation. The panel agreed that childhood adversity has now been demonstrated to influence outcome for a variety of physical and mental health outcomes and should be a good starting point for better understanding of gene by environment interactions. Also, discussed was the importance of unique environmental effects. Longitudinal data that collect the child's report of negative life events rather than presuming the importance of a particular environmental factor may provide an important window on the unique environment of children [63].

8. Strategies for Uncovering Rare Variants and Other Genetic Variation

8.1. WGS Versus WES

The primary purpose of the meeting was to discuss the utility and implications of performing WGS in pedigrees with a high density of SUD. The utility of such an effort was viewed as dependent on the availability of pedigrees. Results of presentations from the meeting suggest that such pedigrees do exist, for example, in the Pittsburgh and COGA studies. Additionally, the added value of performing WGS in comparison to WES was discussed. While WES can provide information on the protein coding regions of a large number of genes, noncoding regions where a significant amount of information for directing the activity of genes is located is not included. Other considerations include the reliability and validity of each approach. Comparison of WES and WGS from an analysis of the same samples has revealed some differences reported in 2015 [64]. Although the number of SNVs and small insertions/deletions (indels) was similar, the distribution of coverage depth, genotype quality and minor read ratio were more uniform for WGS than WES. Additionally, Sanger sequencing performed on these samples allowed comparison of false positive reads using the two methods, which found a higher false positive rate for identified SNVs for WES (78%) versus WGS (17%). Although improvements in sequencing have been made since the time of that publication, the direct comparison using both methods provides useful information. A more recent analysis [65] has suggested that previous comparisons usually did not take into account short read mappability limitations. Accordingly, commonly noted limitations of WES may be due to mappability limitations of some regions rather than sequence composition. Accordingly, WES findings offer results for further follow‐up. However, utilizing WGS in carefully selected family cohorts appears to offer new opportunities for variant discovery. Due to the higher cost of WGS relative to WES over the past several years, WGS has not as often been utilized. However, the cost of WGS today has become more feasible with long read WGS now less than $1000.

8.2. Analysis of Pedigree Members Versus Large‐Scale Unrelated Individuals—Reproducibility and Power Issues

Utilizing members of pedigrees for genetic analysis offers many advantages for statistical control and feasibility. These include (1) elimination of population stratification due to greater ethnic similarity within families than in population samples; (2) similarity of within family environmental effects; (3) a feasible size for investigator initiation of laboratory‐based measures; (4) capability for tracking and performing longitudinal inquiry that can increase power through repeated measures analyses; (5) capability of re‐assessment of diagnoses to improve phenotypic accuracy; (6) capability for identifying variants of large effect when pedigrees are ascertained for familial severity of the disorder.

Large‐scale analyses of unrelated individuals offer many advantages. These include (1) previous demonstrations that samples from a few hundred thousand [46] to a few million [42] can reveal multiple variants with impressive p‐values [30, 42, 46, 66]; (2) greater ancestral diversity of individuals included for improved representativeness; (3) potential to uncover new variants with broadening of ancestral diversity; (4) potential to uncover shared and unique genetic variants across multiple ancestries [50]; (5) potential to improve fine mapping due to diverse patterns of linkage disequilibrium.

Clearly, there are advantages and disadvantages to both approaches. Importantly, both strategies have the potential to introduce bias in the results obtained. As noted by Kaplan and colleagues [67] among these are sampling errors, measurement errors, multiple comparison errors, and errors due to systematic exclusion of information. Errors due to systematic exclusion of information appears to be a serious issue in investigations that involve electronic health records (EHR) because it is estimated that only 40% of individuals ever get their information into an EHR [67]. Another concern is appreciation of the importance of effect size in evaluating the biological significance of findings. Results from large‐scale studies can be misinterpreted if statistical significance is viewed as the degree of biological importance of a finding. As noted by Kaplan, at a constant p level, effect size declines as a function of sample size so that at a p value of 0.05 obtained from a sample of 150 subjects can reveal an effect size of 0.32 while one with 15,000 subjects can reveal one with only an effect size of 0.03.

Members of the panel expressed a concern that discussion of the most useful methods of uncovering genetic variation not lapse into former discussions in the literature that focused on GWAS versus family study approaches. Rather the panel suggested that a focus on the potential for utilizing family‐based studies in combination with large‐scale exome sequencing from biobanks might provide the most useful information. This complementary approach may provide more actionable data because of the ability to track rare variants within families with the opportunity to test the rigor and reproducibility of these variants in large‐scale efforts where the power to detect variants associated with outcome can potentially be uncovered.

9. NIDA/NIH Funding Priorities

The panel noted that currently there is a large gap in research on family‐based designs due to greater emphasis on large‐scale GWAS efforts that utilize biobank data. Improved statistical probabilities brought about by sample size require further evaluation for their biological significance. Importantly, effect size is a critical issue and this can be achieved with both smaller and larger samples [67]. Moreover, even small effect sizes can lead to major biological understanding. Detection of signals of high biological significance may be uncovered by family studies in which extreme phenotypes are present due to study design. These include (1) ascertainment through multiple affected relatives; (2) inclusion of cases with early onset of the disorder; (3) utilizing families with multiple generations of a disease. Examples of this can be seen in observations of families with members with Tourette's syndrome in Costa Rica [68] and in observations of multiple cases of breast cancer observed in Utah that then led to identification of BRCA1 and BRCA2 [69].

Given these considerations, the panel suggested that because of the currently large gap in the potential to uncover variants with significant impact that may be driving multigenerational SUD within families, especially those with a high density of cases, there is an opportunity for utilizing family studies as a complementary strategy to biobank exploration. These family studies would be most useful when multimodal observations are available, preferably with longitudinal follow‐up. Availability of psychiatric diagnoses beyond addiction phenotypes in the family studies that could be harmonized with biobank data would be an important aspect of classifying families as well as individuals.

10. Conclusions and Future Directions

The panel was convened to determine if it is currently feasible to conduct WGS in existing pedigrees. Genome wide association studies with increasingly larger sample sizes have identified variants with impressive statistical significance. These GWAS studies provide information on common variants associated with substance use and psychiatric disorders. However, rare variant discovery requires new strategies. The primary value in conducting studies in pedigrees is to reveal rare variants that may escape attention in large‐scale case–control initiatives. The panel was in agreement that discovery of rare variants will not take us forward in identifying cases for intervention at the individual level, but are quite important in understanding the mechanisms involved in specific SUD. Greater understanding of the mechanisms can have implications for treatment and medication development.

The panel concluded that there had been multiple successes in identifying variants associated with substance use and psychiatric disorders based on family studies. Although strategies for recruitment of participants vary from those selecting participants unrestricted by presence of comorbidity to those that select through a focused strategy removing cases where there is clear evidence of other psychopathology, nevertheless significant findings are present sometimes for the same gene as was presented for the ACN9 gene. Further work may be needed using both strategies for comparison and possible identification of differences.

The panel agreed that large‐scale biobank studies are needed to complement family studies. A number of issues were discussed regarding the limitations in the use of biobank data. Among these are the way in which the phenotype is identified. All agreed that problem oriented diagnoses are preferable to diagnoses based on quantity of consumption alone. Ideally, future studies should incorporate both. Related to future plans for large scale studies, it was recommended that where possible quantitative traits and utilization of RDoC criteria be utilized as well. Most studies to date have relied on DSM diagnoses that are binary based on presence or absence of a disorder.

While family data provides the opportunity for laboratory‐based methodologies that can amplify diagnostic rigor such as that seen with EEG and ERP measures, families are often recruited based on the referred nature of individuals within them. In other words, the proband that leads investigators to the family is identified through a treatment encounter. Greater utilization of biobank data that offsets limitations of referred individuals is the existence of new statistical methods that utilize distant relatedness in biobank data to identify Mendelian disease. Use of PRS analysis has also broadened capabilities in use of large‐scale biobank data.

The single most important conclusion reached by the panel was that both family studies and biobank studies have unique and important characteristics that recommend them. The panel encouraged the use of combined strategies in which both family data and biobank data are combined to leverage the strengths of each.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.