Bioinformatic and biological avenues for understanding alcohol use disorder

Abstract

Alcohol Use Disorder (AUD) is a multifarious psychiatric condition resulting from complex relationships between genetics, gene expression, neuroadaptations, and environmental influences. Understanding these complex relationships is essential to uncovering the mechanisms involved in the development and progression of AUD, with the ultimate goal of devising effective behavioral and therapeutic interventions. Technical advances in the fields of omics-based research and bioinformatics have yielded insights into gene interactions, biological networks, and cellular responses across humans and animal models. This review highlights several of the newly developed sequencing methodologies and resultant discoveries in neuroscience, as well as the importance of a multi-faceted and integrative approach for determining causal factors in AUD.

Introduction

Alcohol Use Disorder (AUD) is influenced by both genetic and environmental factors, affecting a series of molecular systems throughout different tissues and cell types. The central nervous system (CNS) is remarkably heterogeneous, consisting of an unknown number of specialized cells that govern behavior. A key challenge in biomedicine is to uncover the identity and function of individual cells, classify cellular populations, and discern their cooperative molecular roles in forming continuously adapting biological networks. The growth and development of high-throughput biological assays, combined with an increasing suite of bioinformatics tools and strategies, are important tools to address these challenges. Furthermore, the rapid proliferation of large-scale datasets is helping to outline the molecular processes that define CNS homeostasis and the pathophysiology of disease.

Since the initial sequencing of the human genome (Lander et al., 2001; Venter et al., 2001), omics-based sciences have received considerable attention, spurring international scientific initiatives and facilitating genome-wide comparisons across diverse human populations and model organisms. Transcriptional regulation of gene expression from DNA is coordinately regulated, dictating the cellular milieu available for shaping physiological processes. Using RNA deep-sequencing (RNA-Seq), the orchestrated arrangement of expressed genes can be systematically collected and quantified from multiple biological samples spanning the entire transcriptome, within a single experiment (Mortazavi, Williams, McCue, Schaeffer, & Wold, 2008; Wang, Gerstein, & Snyder, 2009). RNA transcripts can be assembled de novo or mapped to a reference genome, eliminating the bias for known and protein-coding cDNAs inherent in earlier DNA microarray analyses. Unbiased profiling of the transcriptome provides a comprehensive overview of the affected biological systems across both the coding and non-coding domains. Sequencing studies consistently show that less than 2% of the human genome encodes for proteins; however, a large percentage of the human genome (~80%) is biochemically active (ENCODE Project Consortium, 2012; ENCODE Project Consortium et al., 2007). For example, long non-coding RNAs (lncRNAs), once thought of as transcriptional noise, are gaining recognition as a class of biomolecules with key regulatory functions in transcription, protein localization, subcellular organization, and RNA processing (Wilusz, Sunwoo, & Spector, 2009).

Transcript data can also be incorporated into bioinformatics analyses to discover novel protein functions (Marcotte, 2000). In some instances, transcript levels may actually be a stronger predictor of phenotypic traits than protein levels (Ghazalpour et al., 2011). RNA-Seq and other sequencing-based approaches have seen a precipitous drop in associated laboratory costs, coupled with widespread practices to efficiently 1) interrogate cellular taxonomy, 2) catalog a host of transcripts (mRNAs, non-coding RNAs, and small RNAs), 3) determine the transcriptional structure of genes, novel non-coding RNAs, patterns of alternative splicing, and post-transcriptional modifications, and 4) quantify expression levels in relation to experimental or disease-associated perturbations. A number of scientific initiatives are creating publicly accessible resources, contributing to the widespread use of RNA-Seq for testing hypotheses, and answering biological questions. The Genotype-Tissue Expression (GTEx) project is one of these initiatives (GTEx Consortium, 2013), which can be incorporated into genome-wide association studies (GWAS) to determine the potential functional effects of genetic variants. By profiling thousands of transcriptomes from healthy human donor tissues and cell lines, the GTEx project has established the shared and tissue-specific influences on gene expression due to nearby and distant autosomal genetic loci (Ongen et al., 2017). The genetically emergent expression profiles demonstrate a distinct clustering of brain regions, which are separable from non-brain tissues (GTEx Consortium, 2017). Taking advantage of these publicly available human genotype and tissue-specific gene expression datasets will permit a deeper understanding of the way genetic variants may influence the development of brainbehavior relationships inherent to psychiatric disorders.

Utilizing human postmortem brain tissue

Addiction to alcohol and other substances of abuse is due, at least in part, to persistent changes in gene expression. Establishing associations between human alcoholic subjects and experimental models is essential for studying AUD in a laboratory setting. To accomplish this, high-throughput and integrated systems approaches are needed to characterize the functional state of the brain disease. RNA-Seq of human postmortem hippocampus tissue shows overlapping gene expression signatures related to plasticity between chronic alcohol abuse and cocaine addiction (Zhou, Yuan, Mash, & Goldman, 2011). Molecular networks related to chronic and excessive alcohol drinking behavior exhibit system-wide disruptions in inter-gene connectivities compared with matched non-alcoholic control subjects, suggesting pervasive molecular adaptations in neuronal function. Coherent changes in neuronal gene expression are accompanied by several genes with unknown biological function, including a number of non-coding RNAs. The interrelationship among gene pairs within a gene network is informative for predicting deeply conserved biological processes (Costanzo et al., 2010; Kachroo et al., 2015). Thus, examining gene-gene interactions disrupted by human disease may provide new biological insights for singular genes and rewired systems (see Table 1).

Table 1.

Short list of publicly accessible resources.

Name	Description	Link
International Mouse Phenotyping Consortium	A consortium that aims to discover and ascribe biological function to each gene by testing each mutant mouse line	http://www.mousephenotype.org/
The GTex Project	Repository for human genotype, gene expression, and clinical data	https://gtexportal.org/home/
ENCODE	Database for functional elements of the human genome	https://www.encodeproject.org/
The Library of Integrated Network-Based Cellular Signatures (LINCS)	A database of cellular, tissue, and organismal response signatures to various drugs and molecular factors	http://lincsportal.ccs.miami.edu/dcic-portal/
Connectivity Map database (CMap)	Gene expression profiles from thousands of small-molecule compounds and genetic reagents tested in 5 human cell lines	https://www.broadinstitute.org/connectivity-map-cmap
CMap and LINCS Unified Environment (CLUE)	Cloud-based infrastructure for integrating LINCS and CMap analyses	https://clue.io/

Epigenomic control of gene expression networks, mediated by histone H3 lysine 4 trimethylation (H3K4me3) positioned within promoter regions near active transcription start sites, is characterized in greater detail through separate biological systems for each substance of abuse (Farris, Harris, & Ponomarev, 2015). H3K4me3 and other chromatin markers may overlap single nucleotide polymorphisms (SNPs) that regulate transcriptional properties within gene promoter regions in particular cell types and tissues. Individual and shared-risk variants associated with distinct phenotypic traits are frequently localized within the boundaries of epigenetic marks, influencing the binding of transcription factors to downstream properties among specialized cell types (Brown et al., 2017; Trynka et al., 2013). SNP heritability contributes to common patterns of variation in human gene expression networks across neuropsychiatric disorders (Gandal et al., 2018), with open chromatin regions bearing a larger per-SNP based effect for scoring some polygenic traits (Salvatore et al., 2018). Genetic variants associated with alcohol dependence are enriched within specific gene co-expression networks correlated with lifetime alcohol consumption (Farris, Arasappan, Hunicke-Smith, Harris, & Mayfield, 2015). Combining gene network studies in human alcoholics with epigenetic and single variant analyses will be vital to uncover the shared contributions of genetic and environmental factors capable of driving the molecular adaptations in AUD.

Cross-species conservation for studying AUD

Despite millions of years of evolution, an expansive set of bioinformatics analyses indicates that humans and other species share genomic structure (Gerstein et al., 2014). Greater than 90% of the mouse genome shows evidence for conserved synteny with the human genome, with a nearly identical set of protein-coding homologs (Mouse Genome Sequencing Consortium et al., 2002). Although the expression patterns of human and mouse protein-coding transcripts can differ (Yue et al., 2014), the strong degree of evolutionary conservation supports the utility of mice, and other animal models, for the study of complex traits. Resources such as the International Mouse Phenotyping Consortium extensively characterize genetically engineered mutant mice throughout the entire genome of conserved genes (Dickinson et al., 2016). Over 100 null mouse mutants for individual candidate genes have been studied for alcohol-related behavioral phenotypes, with ~72% of mutants having a known effect in a common model of voluntary alcohol consumption (Crabbe, Phillips, Harris, Arends, & Koob, 2006; Mayfield, Arends, Harris, & Blednov, 2016). The broad array of genes currently implicated is expected, given the polygenic mode of inheritance and the spectrum of molecular systems involved in AUD and other comorbid behaviors. Additionally, similar to findings from human GWAS, animal behavioral phenotypes can vary widely with respect to the null allele carried across mixed genetic backgrounds (Sittig et al., 2016). Complementing single gene-based procedures with systems-oriented approaches strengthens the ability to establish causal determinants of neurobiological substrates in behavior.

Addiction-related alterations in gene expression are time-, tissue-, and substance-dependent (Piechota et al., 2010). Chronic intermittent exposure to ethanol in an animal model of dependence results in temporal changes in gene co-expression networks that may correspond to early epochs of multi-cellular CNS plasticity in the development of AUD (Osterndorff-Kahanek et al., 2015; Smith et al., 2016). Recurring bouts of chronic ethanol exposure, a hallmark of dependence and disease progression, concurrently shift the dynamics between coding and non-coding interaction networks. Akin to protein-coding ensembles, the expression of small non-coding RNA molecules, for example microRNAs (miRNAs), is discretely modulated in rodent models of chronic ethanol exposure (Osterndorff-Kahanek et al., 2018; Tapocik et al., 2013). The transition to excessive ethanol drinking behavior can be altered by directly manipulating the expression of distinct miRNAs (i.e., miR-206 and miR-30a-5p) that converge on the same downstream target, brain-derived neurotrophic factor (Bdnf) (Darcq et al., 2015; Tapocik et al., 2014). Redundancy is a known property of biological networks, imparting a series of checks and balances to ensure integrity of the system as a whole. Paired profiling of mouse prefrontal cortex for protein-coding genes and miRNAs following chronic ethanol consumption demonstrates a reciprocal relationship among multiple biological networks and pathways, analogous to certain features observed in human molecular networks (Nunez et al., 2013). Studying the intricate wiring of biomolecule expression across different model organisms and behavioral paradigms further assists in defining the subsystems involved in the acquisition, maintenance, and recurrence of alcohol dependence (Fig. 1).

Fig. 1.

Schematic representation for the human medical diagnosis of alcohol use disorder (AUD) in relation to different animal models used in biomedical research. An AUD diagnosis encapsulates many different phenotypic traits (e.g., ‘craving’ and ‘withdrawal’). Genome-wide studies of model organisms that address latent features of AUD can help decipher coordinated biological activity of interrelated modules found in humans (colored blocks along y-axis).

RNA-seq of cellular populations

Honing the resolution of gene expression networks in CNS cell-types is important for understanding the physiological processes that mediate allostasis. Neurogenomic studies of brain tissue have primarily used whole tissue homogenates, which include neurons, microglia, astrocytes, and oligodendrocytes. Chronic ethanol exposure is linked to activation of the immune system and neuroinflammation (Crews, Lawrimore, Walter, & Coleman, 2017), facilitated by the various cell types in the brain. Behavioral studies also support the role of multiple neuroimmune genes, ascertained from transcriptome profiling of CNS tissue from different species, in the regulation of ethanol consumption (Blednov et al., 2012). Microglia account for a small proportion of the total number of cells in the CNS, but are the primary cells responsible for immune-driven responses. Inhibition of microglial activation by the broad-spectrum antibiotic minocycline decreases ethanol self-administration in mice (Agrawal, Hewetson, George, Syapin, & Bergeson, 2011). RNA-Seq analysis of purified microglial cells from ethanol-consuming mice demonstrates that they possess a unique gene network that is largely undetected in mixed cellular populations (McCarthy, Farris, Blednov, Harris, & Mayfield, 2018). Ethanol consumption induces a constellation of toll-like receptors, Tgf-beta signaling, and Nf-kappa-b signaling genes within microglia. Network connectivity identified the protease factor, sialic acid binding Ig-like lectin H (Siglech), as a putative hub gene with a suspected role in ethanol-induced neuroimmune signaling. These findings complement in vitro studies demonstrating that microglia depletion blunts induction of the pro-inflammatory gene, tumor necrosis factor a (TNF a), and enhances expression of anti-inflammatory genes, interleukin-4 (IL4), and interleukin-10 (IL10) following acute ethanol withdrawal (Walter & Crews, 2017). Further research using microglia-depleted mice is needed to determine the specific in vivo role of microglia during ethanol consumption (Elmore et al., 2014). In addition, in vivo studies are needed to validate the potential roles of Siglech and TGF-β in the ethanol-induced immune responses identified with RNA-Seq.

Pro-inflammatory microglial cells can give rise to reactive macroglial cells known as astrocytes, leading to downstream changes in neuronal morphology and function (Liddelow et al., 2017). Astrocytes participate in a variety of inter- and intra-cellular signaling cascades throughout different regions of the CNS. Excitatory neuronal populations releasing glutamate may evoke the propagation of calcium waves (Cornell-Bell, Finkbeiner, Cooper, & Smith, 1990; Dani, Chernjavsky, & Smith, 1992), prompting a form of bidirectional communication among neighboring cells. Selective stimulation of astrocytes using designer receptors exclusively activated by designer drugs (DREADDS), intended to modify glutamate transmission, attenuates cocaineseeking behavior (Scofield et al., 2015) and ethanol self-administration during periods of abstinence (Bull et al., 2014). Isolation of adult astrocytes from ethanol-consuming mice for RNA-Seq analysis indicates that a distinct set of calcium-related signaling events involving calmodulin binding is altered after voluntary binge-like drinking, which are separate from ethanol-responsive genes found in a heterogeneous cellular suspension (Erickson, Farris, Blednov, Mayfield, & Harris, 2018). Concurrent with the calcium-dependent responses, there was an upregulation of extracellular matrix (ECM) genes in astrocytes after chronic ethanol exposure, consistent with reports of ECM regulation of behavior and synaptic plasticity in the drug-addicted brain (Lasek, 2016). The simultaneous assessment of transcriptome-wide changes in defined cellular populations, such as microglia and astrocytes, underscores the specialized cellular landscapes that are generated by chronic ethanol exposure.

Single-cell sequencing

Creating a complete inventory of the intrinsic machinery for individual cellular phenotypes is a daunting endeavor facing the scientific community. By harnessing many of the advancements in single-cell RNA sequencing (scRNA-Seq), global research efforts such as the Human Cell Atlas are diligently pursuing facets of this goal (Regev et al., 2017). The inherent architecture of the CNS is characterized by a vast assemblage of functionally heterogeneous cell types configured in their local microenvironments (Lovatt et al., 2014). Applying scRNA-Seq to adult and fetal brain samples helps distinguish the signatures of major CNS cell types (Darmanis et al., 2015). Individual nuclei can be obtained from human postmortem brain tissue, permitting analysis of low levels of RNA, and reducing artifacts that may arise from cell dispersion procedures (Krishnaswami et al., 2016). Sequencing 3227 neuronal nuclei from different human postmortem cortical regions uncovered intra-subtype differences between individual neurons (Lake et al., 2016). Neuronal variation may be due to the pronounced somatic mosaicism present within neuronal genomes (Lodato et al., 2015; McConnell et al., 2013), which can gradually accumulate throughout the lifespan of individual cells (Bae et al., 2018; Lodato et al., 2018). Such long-lived postmitotic neuronal populations may be at higher risk for harboring somatic mutations relevant for driving neuropsychiatric disorders (McConnell et al., 2017).

Mouse studies have corroborated the evidence found in humans for somatic mutations in neurons, which differ from several other cell lineages (Hazen et al., 2016). The mechanisms responsible for introducing aberrant single nucleotide variants are unknown; however, depending on the genomic location, such variants may affect alternative splicing, protein function, and transcriptional regulation of gene expression. Comparison of human and mouse dopaminergic neuronal subtypes indicates a strong preservation across species, suggesting the potential for cell replacement therapies (La Manno et al., 2016). Under the control of Cre recombinase, transgenic mice can be used in conjunction with scRNA-Seq to map the molecular terrain of known neuronal cell types, as well as to discover new cell types and cellular markers (Tasic et al., 2016). Importantly, single neuronal expression profiles are strongly associated with their respective morphological and physiological properties (Fuzik et al., 2016; Tasic et al., 2016). Established patterns of gene expression are reproducible large-scale fingerprints for varying activity among neuronal systems (Tyssowski et al., 2018). The scalability of scRNA-Seq and the beginning of the use of tissue-based sequencing may establish a molecular-based atlas of the mammalian brain and identify new, improved cellular markers (Jaitin et al., 2014; Vanlandewijck et al., 2018). Mapping the molecular composition of neuronal ensembles with single-cell sequencing using animal behavioral models of differing disease stages will be essential for recognizing causal factors and relevant molecular approaches for disease intervention.

Medication discovery and repositioning

Leveraging genetics in concert with transcriptome measurements offers a viable approach for treating neuropsychiatric disorders (Breen et al., 2016; So et al., 2017). Transcriptome-based signatures have proven valuable for charting the shared relationships and novel mechanisms of action for many chemical compounds (Lamb et al., 2006). Armed with these experimental datasets, bioinformatics tools can predict and validate large-scale screens that target the druggable genome (Jeong, Moon, Song, & Yoon, 2017; Liu et al., 2017; Sawada, Iwata, Tabei, Yamato, & Yamanishi, 2018). The Library of Integrated Network-Based Cellular Signatures (LINCS) is an example of one scientific resource that provides a set of representative cellular response signatures to chemical and genetic perturbations. The LINCS database includes profiles of gene transcription, cell proliferation, protein binding, and other phenotypes across cell types and pharmacological agents (Subramanian et al., 2017; Vempati et al., 2014). Currently, the dataset contains genome-wide gene expression profiles for more than 40,000 perturbagens and approximately 1200 cells (Koleti et al., 2018). The Connectivity Map database (CMap), a library containing over 1.5 million gene expression profiles from thousands of small-molecule compounds and genetic reagents, tested in multiple cell types, can be queried for perturbations that give a related gene expression response for a given pharmacological agent. To facilitate these analyses, a cloud-based computational infrastructure, known as the CMap and LINCS Unified Environment (CLUE), has been built with user-friendly web applications and software tools to enable researchers to access, analyze, and integrate public data alongside their own experiments. Together, these methods can accelerate the prioritization of novel compounds for further functional and behavioral evaluation, including use of animal models of alcohol drinking behavior (Ferguson et al., 2017). Interrogating the druggability of the genetic and genomic states rooted in disease will guide development of future drug designs and foster drug repositioning to broaden the scope of available treatments.

Conclusions

Individually tailoring specific therapies and empowering prevention is the fundamental goal of precision medicine. Bioinformatics bridges different areas of systems biology, melding aspects of both clinical and basic research. Consolidating information across the fields of genetics, transcriptomics, and neuroscience has already begun to shed light on relevant pathways in disease (Network and Pathway Analysis Subgroup of Psychiatric Genomics Consortium, 2015), critical developmental circuits (Parikshak et al., 2013; Willsey et al., 2013), and resting-state brain activity (Wang et al., 2015). Human and animal genetics studies, as well as other approaches, have started to embrace open-source resources for understanding the neurobiology of alcohol and substance abuse. For example, a recent GWAS for alcohol consumption demonstrated the potential functional role for significant SNPs in tissue-dependent gene expression (Clarke et al., 2017). Utilizing an assortment of burgeoning databases and repositories obtained from human and animal subjects can answer questions, as well as fuel new, unexplored questions in neuroscience. High-throughput technologies and bioinformatics techniques are constantly evolving, allowing us to delve deeper into the neurobiology of AUD and find improved avenues for treatment. The expansion and implementation of these cross-disciplinary approaches will continue to be beneficial for interpreting the molecular anatomy underlying AUD and other polygenic, comorbid disorders (Salvatore et al., 2018).