identifying gene networks underlying the neurobiology of … · identifying gene networks...

Identifying gene NetworksUnderlying the Neurobiology of ethanol and Alcoholism

C u r r e n t R e v i e w s

306 Alcohol research: C u r r e n t R e v i e w s

ALcohoL reSeArch:

Aaron r. Wolen, Ph.D., and michael F. miles, m.D., Ph.D.

Aaron r. Wolen, Ph.D., was a doctoral candidate in theDepartment of Human andMolecular Genetics, andmichael F. miles, m.D., Ph.D.,is a professor in the Departmentsof Pharmacology and Toxicologyand Neurology, VirginiaCommonwealth University,Richmond, Virginia.

For complex disorders such as alcoholism, identifying the genes linked to these diseasesand their specific roles is difficult. Traditional genetic approaches, such as geneticassociation studies (including genome-wide association studies) and analyses ofquantitative trait loci (QTLs) in both humans and laboratory animals already have helpedidentify some candidate genes. However, because of technical obstacles, such as thesmall impact of any individual gene, these approaches only have limited effectiveness inidentifying specific genes that contribute to complex diseases. The emerging field ofsystems biology, which allows for analyses of entire gene networks, may helpresearchers better elucidate the genetic basis of alcoholism, both in humans and inanimal models. Such networks can be identified using approaches such as high-throughput molecular profiling (e.g., through microarray-based gene expressionanalyses) or strategies referred to as genetical genomics, such as the mapping ofexpression QTLs (eQTLs). Characterization of gene networks can shed light on thebiological pathways underlying complex traits and provide the functional context foridentifying those genes that contribute to disease development. KEY WORDS: Alcoholism;alcohol use disorders (AUDs); genetics; genetic basis of alcoholism; genetictechnology; genetic association studies; quantitative trait loci (QTLs); genetic mapping;gene networks; genomes; genetical genomics; human studies; animal models

The multiple genetic, environmental,and behavioral factors that play arole in the development of alcohol

use disorders (AUDs) make it difficultto identify individual genes linked tothese disorders. Nevertheless, somegenetic risk factors (i.e., specific variants)associated with AUDs have been iden-tified within many genes, some of whichcode for proteins involved in knownbiological pathways. Despite this progress,it has been exceedingly difficult todetermine which genes may be themost relevant to developing therapeuticinterventions for alcoholism. The majorobstacles in treatment development arethat gene–disease associations revealvery little about the underlying biologyand that any implicated gene variantexplains only a tiny proportion of anindividual’s overall risk for an AUD.Recent work focusing on the study ofalcohol-related gene networks is helping

to shed light on the molecular factorsaffecting alcoholism and other complexdiseases. This article will provide anoverview of approaches used to identifyor construct gene networks and describehow systems biology approaches are help-ing to better understand complex traitssuch as behavioral responses to beveragealcohol (i.e., ethanol) and alcoholism.

Traditional Approaches toDissecting complex Traits

The predominant experimental strategyused by contemporary geneticists toidentify the genetic factors involved in complex traits, such as behavioralresponses to alcohol, essentially is anexpansion of the gene mapping approachproposed by Botstein and colleagues(1980) over 30 years ago. For thisapproach, investigators scan their samples

for genetic variations (i.e., polymor-phisms) that segregate with the trait—that is, which are found in sampleswith the trait more often than wouldbe expected by chance and thereforemight contribute to the developmentof that trait. In recent human studies,this approach typically has been appliedin genome-wide association studies(GWASs) of large, population-basedsamples that comprise both case subjects(i.e., individuals expressing the trait, orphenotype, under investigation) andunaffected control subjects. Hundredsof complex diseases and traits, includingsusceptibility to AUDs, have been analyzed using GWASs, resulting in the identification of several importantlinks between genetic variants and thesediseases (Bierut et al. 2010). Overall,however, the success of this approachhas been mixed, and greater progresshas been hindered by insufficient

gene Networks Underlying Neurobiology 307

sample sizes, stratified populations, theinvolvement of rare gene variants (i.e.,alleles) that each only have a small effectsize, and heterogenous phenotypic con-structs (i.e., using different criteria todistinguish cases from controls).1

A similar forward-genetics approachthat most often is used for studyinganimal models of complex traits iscalled quantitative trait locus (QTL)mapping. A quantitative trait is a phe-notype that is determined by severalgenes, each of which has a variablecontribution to the trait. The locationsof the involved genes on the chromo-somes are referred to as QTLs. QTLmapping studies typically are conductedusing inbred strains of mice and theirvarious derivatives. For example, theC57BL/6J (B6) and DBA2/J (D2)inbred mice frequently are used inalcohol research because they clearlydiffer in various responses to alcohol,including development of functionaltolerance (Grieve and Littleton 1979),locomotor activation (Phillips et al.1998), and sensitivity to withdrawalsymptoms (Metten and Crabbe 1994).Because the environmental conditionsin these experiments can be controlled,any differences observed between themouse strains in these phenotypes mostlikely can be attributed to genetic dif-ferences. QTL mapping studies thenseek to detect the polymorphisms under-lying the complex traits of interest byscanning for alleles that co-vary withthe traits.

Similar experiments also can be con-ducted with special derivatives of inbredstrains known as recombinant inbred(RI) mice. These animals are derivedby cross-breeding two or more distinctparental strains (which often divergewidely for the trait of interest), followedby inbreeding of the offspring for severalgenerations (Bailey 1971). Given thecorrect breeding strategy, this method

results in a panel of RI mouse strainsthat differ in the degree to which theyexhibit a certain phenotype of interest.At the same time, each of the strainseffectively is isogenic, meaning that forall genes, the genome carries two iden-tical alleles (i.e., is homozygous). As aresult, when two animals from the sameRI mouse strain are bred, their off-spring will have the exact same geneticmakeup (i.e., genotype) as the parents.This makes it possible to directly inte-grate results generated from disparateexperiments, in different laboratories,and at different times if they all use ani-mals from the same RI mouse strain.This feature of RI mouse panels, andinbred animals in general, is particularlyvaluable for QTL mapping because theexpense and time involved with geno-typing or sequencing a strain only isincurred once.

The molecular and genetic resourcesoutlined above have greatly increasedthe power and resolution of QTLmapping for various behaviors or othertraits of interest. Yet despite theseadvances, the DNA regions identifiedas QTLs typically still are relatively largeand may contain several genes; accord-ingly, few genes have been validated ascontributing to quantitative traits (i.e.,being quantitative trait genes [QTGs]).This difficulty is attributable largely to the lack of sufficient recombinationevents in existing mouse panels toreduce the size of DNA segments thattypically are inherited together (i.e.,haplotype block size) for fine mappingand to the generally small effect size forany single QTG.

genomic Approaches to Disease Dissection

Because of the technical obstaclesimpeding their more effective use, bothGWASs and QTL mapping studies todate have identified a deluge of disease-associated genetic loci but only fewactual causal genes. Moreover, even themost successful studies have failed toplace the disease-associated genes in anykind of biological context that would

serve to explain the underlying func-tional biology. Without elucidating thecomplex interactions of the molecularphenotypes that stand between geneticvariation and disease, it will be difficultor impossible to develop new and effec-tive approaches to treating such diseases.

The emerging field of systems biologyis tackling this immense challenge bystudying networks of genes, proteins,metabolites, and other biomarkers thatrepresent models of genuine biologicalpathways. Studying complex diseasesin terms of gene networks rather thanindividual genes or genomic loci shouldaid in uncovering disease genes. Withthis approach, the effects of multiplegenes in the network are combined,producing a stronger signal and reducingthe number of statistical tests of associ-ation that must be performed.

These benefits effectively weredemonstrated in two recent humanassociation studies that modified thetypical GWASs strategy by seeking associations only within groups offunctionally related genes, rather thanacross the entire genome. The first ofthese studies (Ruano et al. 2010) dis-covered that cognitive ability, a complexphenotype with a large genetic compo-nent, was significantly linked to genesencoding molecules called G-proteinsthat consist of three different subunits(i.e., heterotrimeric G-proteins). Thesecond study (Reimers et al. 2011)found that genes related to signalingpathways involving the neurotransmittersglutamate and γ-aminobutyric acid(GABA) signaling collectively con-tribute to alcohol dependence.

Network-based approaches to thedissection of complex diseases also canbe applied to animal models, yieldingexperimental results that are more gen-eralizable to humans because the path-ways represented by these networks are more evolutionarily conserved thanindividual genes. This should encouragegreater collaboration between researchersstudying a common disease in differentspecies. In fact, the biology underlyinggene networks is so complex that anyhope of deriving novel therapeutics maybe entirely contingent on the extent to

1 This is an issue faced by GWASs researchers when classifyingsamples as cases or controls. If cases are limited to only individ-uals who have been diagnosed with an AuD, it becomes difficultto enlist a sufficient number of participants. Moreover, many ofthe control subjects could very well be undiagnosed alcoholics or people who meet some but not all of the diagnostic criteria for an AuD. As a result, the control group could be polluted withnear-cases, diluting any detectable group differences.

which scientists with diverse areas ofexpertise are willing to share and inte-grate datasets and make the process ofinterpretation a collaborative one.

Using High-Throughput MolecularProfiling to Define DiseaseAs the human and mouse genomes werebeing assembled using the cutting-edge,high-throughput DNA sequencers that made these endeavors possible,new technologies began to emergethat, for the first time, allowed near-comprehensive profiling of other cellularcomponents. The term profiling refersto the measurement of different typesof biological molecules, such as DNAto identify polymorphisms, messengerRNA (mRNA) to determine transcriptabundance, proteins to identify certainchemical modifications that occur afterthe initial protein synthesis, and metabo-lites to evaluate biochemical processes inthe cells. Platforms for high-throughputapproaches for all these types of molec-ular profiling have become increasinglycommonplace. Concurrently, methodsfor analyzing data produced by thesetechnologies constantly are evolving,yielding results that are simultaneouslymore sensitive and more specific. As aresult, researchers are better able toappreciate systems-level changes associ-ated with disease.

Of these various high-throughputprofiling techniques, microarray-basedgene expression platforms have featuredmost prominently in biomedical researchto date. Through an unbiased profilingof the transcriptome—that is, a mea-surement of all mRNA molecules pro-duced within a cell or tissue sample—microarray expression studies allowresearchers to identify patterns of geneexpression associated with a disease. Insome cases, such patterns can betterdefine a complex phenotype by identi-fying disease subtypes. For example,microarray analysis of breast cancertumors identified gene expression sig-natures that predict patient prognosisand therefore help physicians tailortreatment regimens (van’t Veer et al.2002). From a basic research perspec-

tive, microarray expression profiles canhelp tease apart the complex interactionsthat underlie the development of a dis-ease by implicating a subset of geneswhose regulation is altered with thedisease. With this information, it maybecome feasible to reconstruct the under-lying biological pathways and enhanceunderstanding of disease etiology.

Genomic approaches have beenapplied to alcoholism directly bystudying postmortem human brain tissue isolated from alcoholics andmatched control subjects using geneexpression microarrays. This has revealednovel information about changes inthe brain’s transcriptome that are asso-ciated with chronic ethanol consumption.One of the findings was a significantderegulation of genes encoding proteinsthat synthesize and maintain myelin,the substance that forms a sheath sur-rounding the long extensions (i.e., axons)of nerve cells and that is essential foreffective nerve signal transmission(Lewohl et al. 2000; Mayfield et al.2002). However, the nature of thesestudies makes it impossible to determinewhether such gene expression devia-tions actually are risk factors that con-tribute to AUDs or simply representmolecular consequences of excessivealcohol consumption that are unrelatedto the behaviors constituting alcoholism.

Animal models can assist greatly inthis analysis by allowing for experi-ments that are far more detailed andinformative but too invasive to ever be performed with humans. Althoughanimal models could never replicate a phenotype as complex as alcoholism,they can mimic certain facets of thetrait, which then can be associated withspecific expression signatures usinggene expression microarrays. For example,a genetic predisposition for alcoholismmay entail a stronger-than-averagepreference for alcoholic beverages. Thisparticular facet of alcoholism is capturedby rodent models that selectively werebred to maximize a penchant for oraversion to ethanol, such as the aptlynamed high-alcohol preference (HAP)and low-alcohol preference (LAP) mice(Grahame et al. 1999). In order to

identify genes that may alter the per-ceived desirability of ethanol, geneexpression microarrays were used tocompare the brain transcriptomes ofHAP and LAP mice, as well as of sev-eral other inbred mouse strains thatdrastically differ in ethanol preference(Mulligan et al. 2006). This importantstudy identified a diverse array ofmolecular pathways associated withdifferences in ethanol preference. Someof the genes that had the largest effectsizes were related to neuronal functionand to the maintenance of the cells’normal internal conditions (i.e., cellu-lar homeostasis).

Another important facet of a geneticpredisposition to alcoholism is a com-paratively blunted sensitivity to theeffects of ethanol. Studies have shownthat people who initially are less sensi-tive to acute ethanol exposure are morelikely to have a family history of alco-holism and are at greater risk for devel-oping an AUD (Schuckit 1984, 1994).As mentioned earlier, the B6 and D2inbred mice frequently are used ingenetic studies of ethanol sensitivity.For this reason, Kerns and colleagues(2005) used microarray expressionstudies to dissect the effect of acuteethanol exposure on the brain’s tran-scriptome using the B6 and D2 inbredmouse strains. The investigators analyzedthree brain regions involved in a brainsystem called the mesocorticolimbicreward pathway, which is involved inmediating the rewarding properties of alcohol and other drugs. For eachregion analyzed, the study identified aspecific set of genes (i.e., a gene module)whose expression was altered in responseto acute ethanol exposure. These genemodules contained greater-than-expected numbers of genes involved inseveral signaling pathways (i.e., retinoicacid signaling, neuropeptide expression,and glucocorticoid signaling). Moreover,similar to the microarray studies ofpostmortem human alcoholic brains(Lewohl et al. 2000; Mayfield et al.2002), several genes involved in myeli-nation robustly were altered by alcoholexposure, particularly in the prefrontalcortex (Kerns et al. 2005).



In examining the responses to acuteor chronic alcohol exposure in rodentbrains, these and numerous othergenomic studies have enhanced theunderstanding of the “ethanol tran-scriptome” and provided a more com-prehensive picture of the genes andmolecular pathways that contribute to specific facets of AUDs than what is possible with studies of postmortemhuman brains (Daniels and Buck2002; Mulligan et al. 2011; Rimondiniet al. 2002; Saito et al. 2004; Treadwelland Singh 2004). Moreover, thesestudies effectively have demonstratedhow gene expression microarrays canhelp close the information gap thatexists between DNA variation andcomplex diseases. However, prioritizingthe long lists of genes produced bycomparative microarray studies con-ducted in either species has provenexceedingly difficult. As the costs asso-ciated with validating a given gene’srole in driving a complex trait are con-siderable, an effective strategy for pri-oritizing candidate genes is crucial.Investigators therefore have used moresystems-level approaches that combinegenetic, genomic, and pharmacologicalmethods to better delineate gene net-works causally related to ethanolbehaviors. Networks allow us to inferrelationships between genes and deter-mine which are most important.

The gene Network As a modern genetic map

The previous section mentioned sev-eral studies that used gene-expressionmicroarrays to define lists of genesresponding to ethanol or otherwise relevant to AUDs. Although these studieshave provided important biologicalinsights, the question of how such listscan be used to further advance under-standing of a complex disease is noteasily answered. Network-basedapproaches can greatly improve the interpretability of differential gene-expression results by providinginformation about the relationshipsbetween genes.

Networks are systems of intercon-nected components. For example, theWorld Wide Web is a global networkof computers sharing documents con-nected by hyperlinks; road maps arerenderings of city networks connectedby highways; social networks are groupsof people connected through friend-ships; cellular signaling pathways aregroups of proteins connected throughmolecular interactions; et cetera.Placing such complex systems within a network framework makes it possibleto formally analyze the relationshipsthat constitute these systems. Genenetworks typically are visualized asmathematical graphs—that is, a collec-tion of vertices and edges, where genesare represented by nodes and the linesconnecting the nodes indicate that somerelationship exists between the genes.

Many published network analyses of gene groups use information aboutpre-existing biological relationships,which may be derived from sourcessuch as literature co-citation analysis(i.e., genes mentioned together in a scientific abstract), protein–proteininteraction databases, or gene ontologygroupings. Some commercial tools are available for such studies, such asIngenuity Pathway Analysis (IngenuitySystems, Redwood City, CA). However,although these sources provide categoriesfor interpreting the genomic data, theyalso force such interpretation into themold of pre-existing information,thereby partially defeating the goal ofgenomic studies.

Genomic data collected with high-throughput molecular profiling pre-sents the opportunity to derive novelgene–gene interactions. The maturityof gene expression microarrays relativeto similar technologies designed tomeasure other molecular phenotypeson a genomic scale has meant thatgene networks primarily are renderedas gene coexpression networks. In thecontext of gene coexpression networks,links between nodes typically indicatethat the expression levels for two genesare strongly correlated with one anotheracross whatever conditions an experi-ment entails (e.g., across tissues, time

points, treatments, or individuals).Each link in a gene network essentiallyrepresents a testable hypothesis thatcan be validated through follow-upmolecular experiments. Indeed, coex-pression networks have been used toidentify protein interactions that arenovel (Scott et al. 2005) and conservedacross species (Stuart et al. 2003).

Various novel and innovative methodsexist for generating gene coexpressionnetworks. Although a comprehensivereview of these methods is beyond thescope of this article, a few select methodsare described in more detail in the side-bar. In their simplest form, however,gene coexpression networks can beconstructed by calculating Pearson cor-relations between all gene pairs andapplying a cutoff threshold to determinewhich genes should be connected. Thesimplicity of this approach makes it anappealing choice for conducting a firstround of analyses.

Wolen and colleagues (in press) haveattempted to better define the meso-corticolimbic reward pathway’s tran-scriptional response to acute ethanolexposure by expanding the originalB6/D2 study (Kerns et al. 2005) toinclude members of the BXD family of recombinant inbred mouse strains.The naturally occurring DNA poly-morphisms that distinguish each BXDstrain cause heritable changes in geneexpression, making it possible to iden-tify genetically coregulated transcriptsacross the BXD family. Microarrayexpression data from the prefrontalcortex of BXD family animals wereused to look for evidence of coregula-tion among the 307 ethanol-responsivegenes identified in the original B6/D2study (see figure 1). The analysis iden-tified several groups of intercorrelatedgene modules, indicating this gene setis comprised of several gene networks(figure 1).

A variety of calculations can be usedto gauge the relative importance of aparticular gene to the network as awhole (Horvath and Dong 2008). Thesimplest measurement of node impor-tance is determined by the degree of“connectivity”—that is, the number of

other genes the node is connected to inthe network. However, a gene’s “position”in the network also is an importantconsideration. For example, a gene thatserved as the sole connection betweentwo otherwise independent gene networkswould rank fairly low on a priority scalebased on connectivity alone, despitebeing an important channel of inter-module communication. A measurementof “betweenness centrality” (Girvanand Newman, 2002) can highlightsuch a gene by determining the fre-quency with which a node is includedin the shortest paths between all possi-ble node combinations.

Figure 2 highlights six subnetworkstaken from the larger coexpression net-work depicted in figure 1. The network

comprising genes known as Mbp, Mobp,Mal, and Plp1 is of particular interest,because these genes all play a role in theformation and stabilization of myelin.In addition, a parallel analysis was con-ducted using microarray data only fromthe prefrontal cortex of the same BXDstrains after they had received an injectionof 1.8 g/kg ethanol into their abdomi-nal cavity. This analysis revealed thatthe myelin gene network persisted butunderwent minor topographical mod-ifications. Most notably, additionalconnections were detected betweenPlp1 and two additional genes calledMog and Lpar1. The absence of Mog inthe original network probably was anartifact of the method used to formthese networks, and the gene likely

should have been included. Plp1 andLpar1, in contrast, were effectivelyunrelated at baseline and only showedevidence of coregulation after ethanoltreatment, suggesting this is a genuinemolecular response to ethanol (see figure 2B).

The Lpar1 gene encodes a receptorfor lysophosphatidic acid (LPA), a sig-naling molecule containing phosphateand lipid components (i.e., a phospho-lipid). Regulation of Lpar1 is criticalfor proper nerve-cell formation (i.e.,neurogenesis), including in a brain regioncalled the hippocampus in adults(Matas-Rico et al. 2008). In addition,Lpar1 regulates the breakdown of themyelin sheath (i.e., demyelination)that occurs after nerve injury (Inoue


constructing gene Networks

Various methods exist for generatinggene networks. As mentioned inthe main text, the simplest method

for constructing gene coexpressionnetworks involves calculating Pearsoncorrelations for all pair-wise genes andapplying a hard threshold to determinewhich genes should be connected.The robustness of these networks, initially referred to as “relevance net-works,” can be assessed through anapproach called permutation testing(Butte et al. 2000). A more rigorousmethod for constructing gene coex-pression networks utilizes a graph the-oretical approach to identify denselyintercorrelated gene modules calledparacliques (Baldwin et al. 2005).Paracliques represent gene–gene inter-action networks with extensive, butnot perfect, strong expression correla-tions between all genes in the network(http://grappa.eecs.utk.edu/grappa/root). Paracliques can contain mem-bers with missing links. Therefore,paracliques provide an attractive com-promise by augmenting coexpression

with genes whose correlational rela-tionships to a network are strong, butpermissibly imperfect with a propor-tion of the network. This proportion,called the proportional glom factor, isa user-defined parameter.

A potential limitation of both rele-vance networks and paracliques is thatthey rely on hard thresholds to classifythe relationship between genes as eitherconnected or unconnected. Thedichotomy imposed by this approachmay be artificially limiting these net-works, causing biologically meaningfulrelationships to be overlooked (Carteret al. 2004). For example, the absenceof the Mog gene from the myelin net-work described in the main articleand depicted in figure 2A followingethanol treatment is symptomatic ofthis limitation. An approach calledweighted gene coexpression networkanalysis (WGCNA) is an increasinglypopular method that avoids thesepotential pitfalls by utilizing a “soft-thresholding” approach to generatenetworks that conform to a scale-free

topology (Zhang and Horvath 2005).Scale-free networks follow the powerdistribution they are named for, com-prising many nodes that have sparseconnections and a few that are highlyinterconnected. In addition to providingan accurate model for metabolic net-works (Jeong et al. 2000), neural net-works of the roundworm Caenorhabditiselegans (Watts and Strogatz 1998),and the World Wide Web (Albert andJeong 1999), the scale-free topologyalso typifies gene coexpression networks(van Noort et al. 2004). Some researchersrecently have used WGCNA to definecorrelated gene modules associatedwith blood alcohol levels using the“drinking-in-the-dark” paradigm ofexcessive ethanol consumption in B6mice (Mulligan et al. 2011). WGCNAalso can be implemented as a freelyavailable package (Langfelder andHorvath 2008) for the R StatisticalEnvironment and provides an excellentset of tutorials (available at genetics.ucla.edu/labs/horvath/CoexpressionNetwork/Rpackages/WGCNA).

Gene Networks Underlying Neurobiology 311

Lpar1 regulates the breakdown of themyelin sheath (i.e., demyelination)that occurs after nerve injury (Inoue et al. 2004). The fact that Lpar1 isbrought into this network by ethanolexposure suggests the intriguing possi-bility that this gene may play a role inthe loss of white matter2 commonlyobserved in long-term alcoholic patients(Kril and Halliday 1999). This exampleillustrates how studying ethanol-inducedchanges in gene-network topology canproduce testable hypotheses relevant to the neurobiology of alcoholism.Obviously, alterations in the gene net-work occurring after acute ethanolexposure might not always be relevantto alterations in brain structure andfunction (i.e., neural plasticity) or toxiceffects that occur with chronic exposure,such as in alcoholism. Therefore, find-ings regarding networks relevant to oneethanol behavioral phenotype should

be considered “specific” to that pheno-type unless other genetic, pharmaco-logical, or behavioral data suggestslinks to other aspects of ethanol’sactions in animal models or humans.More generally, this example demon-strates how systems-level methods, likegene coexpression analysis, can helpgreatly expand the information contentof gene expression microarray studiesby filling in information about thegene–gene relationships.

Bridging the Gap BetweenGenomics and Gene Mapping

Genetical GenomicsAnother important early advancementtoward a more systems-level approachto identifying disease-associated genes

was the application of gene mappingmethods to high-throughput moleculardata in order to identify causal linksbetween molecular phenotypes andgenomic regions. Like classical physio-logical or behavioral phenotypes, geneticfactors influencing high-throughputmeasures of transcript, protein, andmetabolite abundance can be identi-fied by QTL mapping. To date, suchanalyses mostly have been applied togene-expression microarrays, mappinggene expression QTLs (eQTLs). Thislargely is related to technical constraints,because whole-proteome expressionprofiling currently cannot be done withthe same degree of sensitivity, coverage,and throughput as mRNA profiling.

The strategy of performing geneticlinkage analysis on genome-widemolecular profiles was formalized andtermed “genetical genomics” by Jansenand Nap (2001). This proposal pri-marily focused on gene-expressionmicroarrays and posited that mappingeQTLs would enable researchers toconstruct robust gene networks as wellas link these networks to metabolic orother phenotypes. The investigatorsalso suggested that eQTL mapping couldgreatly aid in the identification of can-didate genes underlying classical QTLsfor disease traits. The first study tocarry out QTL analysis across genome-wide gene expression microarrays was conducted using an experimentalcross between two strains of the yeastSaccharomyces cerevisiae (Brem et al.2002). Subsequently, several investiga-tions applied the approach to mam-malian systems (Schadt et al. 2003;York et al. 2005), including brain geneexpression (Chesler et al. 2003, 2005).

These early genetical genomics studiesalso characterized the two major classesof eQTLs, labeled cis and trans eQTLs,which differ with respect to the posi-tion of the eQTL relative to the genewhose expression is altered (figure 3).

Figure 1 Correlation heatmap depicting patterns of co-expression among genes previouslyidentified as being regulated by acute ethanol (Kerns et al. 2005). Each coloredsquare represents the Pearson correlation (r) between a pair of genes, calculatedusing microarray expression data of prefrontal cortex tissue collected from B6, D2,and 27 BXD mouse strains. The blue and red colors indicate the strength and direc-tion of the gene–gene correlation. Hierarchical clustering was applied to group genesbased on the similarity of their expression profile across this dataset. In doing so,modules of co-expressed genes are revealed as cohesive blocks along the diagonal.

2 The term “white matter” refers to brain structures made up primarily of nerve fibers that are enclosed by the myelin sheathsand therefore have a whitish appearance. Conversely, the term“gray matter” refers to brain structures composed mainly of thebodies of nerve cells, which have no myelin sheath, resulting in a grayish appearance.

A cis eQTL is located at the same siteof the genome as the gene under study.In contrast, a trans eQTL can be locatedelsewhere in the genome, away fromthe gene whose expression is altered. A good example of how a trans eQTLcould manifest involves transcriptionfactors (TFs). These are proteins thatbind with regulatory DNA regions thatare located in front of a gene. Onlywhen the TF binds to the correspondingDNA sequence can the first step in the process of gene expression—transcription—begin. The interactionbetween the TF and the DNA involvesa certain part of the TF called the TFDNA–binding domain that allows theTF to recognize and bind with specificregulatory DNA sequences. Throughthis mechanism, certain TFs only mayactivate the transcription of specificsets of genes. Accordingly, a polymor-phism at the DNA-binding domain ofa certain TF can affect the TF’s ability

to recognize and bind its recognitionsites, causing altered expression of allgenes regulated by this TF. In otherwords, the abundance of all transcriptsfrom those genes would co-vary withthe TF polymorphism. Such a casemight be recognized by a clustering oftrans eQTLs at the site of the causalpolymorphism, sometimes referred toas a regulatory hotspot. Theidentification of trans eQTL clusterscan be a powerful approach for identi-fying key regulators underlying a com-plex trait of interest. Figure 4 depicts the eQTL results for

the same list of 307 ethanol-responsivegenes identified in the B6/D2 studythat earlier was used to construct coex-pression networks. This analysisrevealed that these coexpression networksshare common eQTLs that drive thiscoordinated expression. Furthermore,the strongest eQTLs underlying manyof these genes mapped to one end of

chromosome 7, forming a trans eQTLcluster. These findings provide prelimi-nary evidence that acute ethanol–responsive genes comprise a handful ofgene coexpression networks in the pre-frontal cortex and that a key regulatorof these networks resides on chromo-some 7. A more extensive analysis of thistype has recently been completed (Wolenet al., in press). The genes comprising trans eQTL

clusters often have biological functionsthat have been conserved among species,suggesting that these hotspots mayhave a biological relevance. Accordingly,the search for trans eQTLs may allowresearchers to identify biological func-tions associated with complex traitsthrough defining quantitative traitgene networks (QTGNs). Mozhui andcolleagues (2008) have, for example,dissected a trans eQTL cluster on distalmouse chromosome 1 and identified a candidate gene (Fmn2) that they

312 Alcohol Research: C u r r e n t R e v i e w s

Figure 2 A) Gene networks that are regulated by acute alcohol exposure were identified in the same prefrontal cortex dataset used in Figure 1. Genenetworks were generated by applying a hard threshold of 0.75 to the gene correlation matrix. The inset box contains a cognate of the myelinnetwork (red) that was generated with expression data from the same strains following ethanol treatment. The ethanol-induced modificationsof this network include the addition of a novel connection between Plp1 and Lpar1. B) Scatterplots illustrating the correlation between Plp1and Lpar1 at baseline and following ethanol treatment. The effective absence of any correlation between these genes at baseline suggeststhat this relationship is driven by ethanol exposure.

Ethanol and Brain Gene Networks 313

propose has a major influence on theexpression of linked gene networks.Moreover, a diverse group of pheno-typic QTLs seemed to be located inthe same region, including severalrelated to ethanol.

Genetical Genomics Studies to Identify Gene VariantsIncreasing Disease RiskThe integration of eQTL and classicalQTL data enables identification of keymarkers of disease-causing variants.The effectiveness of this approach wasdemonstrated by a genetical genomicsanalysis of liver expression data from a population of mice placed on a high-fat diet (Schadt et al. 2003). The pur-pose of this diet was to model an obe-sity-like phenotype, which was mea-sured using fat-pad mass (FPM). QTL

mapping for FPM revealed a signifi-cant QTL on chromosome 2 that alsoharbored over 400 eQTLs. By scan-ning this region for cis eQTL-linkedgenes that also were strongly correlatedwith FPM, the researchers were able toidentify two novel obesity candidategenes. Saba and colleagues (2006) used a

similar approach to identify candidategenes for alcohol preference and acutefunctional tolerance to alcohol. Thislarge-scale study included rodent strainsselectively bred for ethanol phenotypes(i.e., HAP and LAP mice) as well as asubset of the BXD family of recombi-nant inbred mice. Applying microarrayexpression profiles using mRNAsobtained from the entire brain, theinvestigators identified independentlists of genes whose expression differed

between the HAP and LAP strains andbetween the BXD strains with highand low levels of acute functional tol-erance. The genetic regulation of thesegene lists then was mapped using BXDexpression and genotypic data. High-priority candidate genes were high-lighted by screening for differentiallyexpressed genes with cis eQTLs thatoverlapped previously mapped behavioralQTLs for either alcohol preference(Belknap and Atkins 2001) or acutefunctional tolerance (Kirstein et al. 2002).The rationale for prioritizing candi-

date QTGs on the basis of their havingcis eQTLs located at the same sites asclassical QTLs is based on the hypothesisthat the variability of a complex phe-notype is linked to a particular locusbecause the causal gene is being pro-duced in variable quantities through a

Figure 3 Illustration of the concept of cis and trans expression quantitative trait loci (eQTLs). A) The left-most gene (red) codes for a transcriptionfactor (TF) protein that activates the transcription of genes A (green) and B (blue) by binding to their respective promoters (gray). In thewildtype, or “normal,” scenario all genes are transcribed at their full potential, as indicated by the bar graph on the right. B) A variant (i.e.,polymorphism) (triangle) in gene A’s promoter region hinders TF binding, causing a reduction in the rate at which gene A is transcribed,while gene B is unaffected. Thus, gene A is being regulated by a cis eQTL because its level of expression is associated with a nearby poly-morphism located on the same chromosome. C) A polymorphism in the TF gene’s DNA binding region (hexagon)—the region of the TFprotein that binds to gene promoters—hinders binding with all downstream promoters, regardless of whether the regulated gene is locatednear the TF gene, like gene A, or located on an entirely different chromosome, like gene B. In fact, all genes regulated by this TF would belinked to a trans eQTL at the site of this TF polymorphism.

cis-acting polymorphism. This hypothesisis somewhat of an oversimplificationand leaves out several importantcaveats. Nevertheless, increasing evidence supports the importance ofgene-expression variability in regulatingcomplex traits. In fact, recent evidenceindicates that SNPs associated with avariety of complex traits are more likelyto contain cis eQTLs than normallywould be expected (Nicolae et al. 2010).This indicates that the importance ofexpression variability in complex traitregulation is not limited to geneticmodel systems and that it may be pos-sible for GWASs and QTL mappingstudies to improve their track recordby incorporating expression data.

Dissecting Complex Diseases Through IntegratingGenomic ApproachesThe discussion above has illustrated howtraditional QTL mapping and GWASsapproaches can benefit from systems-biological approaches by filling in criticalinformation about the molecular phenotypes that stand between DNAvariation and complex disease (figure5). The incorporation of data fromhigh-throughput molecular profilingtechnologies, such as gene expressionmicroarrays, can better define a diseaseby identifying groups of genes thatrespond to or covary with disease-associated traits. Network analysis ofdisease-associated genes allows investi-gators to move beyond static gene lists,partially reconstruct the underlyingmolecular pathways, and prioritizegenes based on their importance to thelarger network. Applying QTL mappingto each gene’s expression trait makes itpossible to identify the genomic regionsthat regulate each gene’s expression and uncover the existence of regulatoryhotspots that exert enormous influenceover a gene network. The series of studiesdiscussed below has demonstrated howeffective these methods are for dissectingcomplex traits, particularly when theyare integrated. Zhu and colleagues (2004) followed

up the genetical genomics analysis of

liver expression data from mice on ahigh-fat diet that was mentioned ear-lier (Schadt et al. 2003) by generatinggene networks from the same microar-ray gene-expression dataset. This analysisincluded two distinct approaches tonetwork construction: The first strategyformed networks on the basis of thecovariation among gene-expressiontraits—that is, genes whose expressionchanged in the same manner were con-sidered linked. In the second strategy,gene-network interactions were deter-mined on the basis of the similarity oftheir eQTL profiles. Thus, the networkswere constructed once without andonce with the benefit of genotypic andeQTL data. Because links within genenetworks represent causal relationships,analyses of these links can help researchers

predict how a system will respond tothe perturbation of a specific gene.Zhu and colleagues (2004) tested thishypothesis by measuring differentialexpression in response to a pharmaco-logical substance that interfered withthe function of a central gene in bothpredicted networks. They found thatthe eQTL profile network was asignificantly better predictor of whichgenes would be affected by the pharma-cological perturbation than the networkconstructed with expression data alone.Therefore, integrating both gene-expression and genotypic informationinto network construction greatlyenhances the predictive value of genenetworks.The ultimate goal of systems-level

analyses of complex diseases is to uncover

314 Alcohol Research: C u r r e n t R e v i e w s

Figure 4 Genome–genome plot of peak expression quantitative trait loci (eQTLs) for the samedataset used in figure 1. Each point indicates the chromosomal position of a gene versus the position of its peak eQTL. Point color is used to communicate the strength of association between a gene and its eQTL, measured by logarithm of the odds (LOD)score. A LOD score is a ratio that measures that probability that a gene is linked togenetic markers, versus the probability that it is not. Thus, the higher a LOD score, themore likely a gene’s expression level genuinely is regulated by an eQTL. Points plottedalong the diagonal likely represent cis eQTLs, which also tend to have stronger LODscores. Perpendicular tick-marks along both axes show the distribution of data points.Along the x-axis the dense clustering of tick-marks toward the proximal tip of chromo-some 7 indicates the presence of a trans eQTL cluster, suggesting this region may harbor an important regulator of the gene co-expression modules seen in figure 1.

Gene Networks Underlying Neurobiology 315

information necessary to establish acorrelation between disease phenotype,mRNA abundance, and the underlyingDNA polymorphism or causal genenetwork. Schadt and colleagues (2005) demonstrated this approach of integrating genotypic data, gene-expression data, and disease endophe-notypes,3 using the same liver expres-sion dataset mentioned above, and anovel network construction techniquetermed likelihood-based causality modelselection (LCMS). The investigatorsfirst identified all QTLs associated witha classical phenotype and then win-nowed the list of potentially associatedgene-expression traits on the basis oftheir correlation or eQTL overlap withthe phenotype of interest. Candidategenes then were ranked by applying

the LCMS technique, which uses theeQTL data to establish causal relation-ships between DNA loci and tran-scripts as well as between transcriptsand phenotypes and finally identifies a model that best fits the data.By ranking genes according to their

performance in these models, theinvestigators identified several novelobesity candidate genes as well asuncovered additional support for theinvolvement of a gene called Hsd11b1that previously had been implicated inobesity risk (Rask et al. 2002). Becausethis gene seemed to be relevant to thephenotype they were investigating, theresearchers then sought to reconstructthe gene network in which Hsd11b1participates by performing the LCMSprocedure with Hsd11b1 as the trait

of interest. The resulting network wasable to successfully predict genes thatwould be affected by inhibition ofHsd11b1. A similar approach has beenused by other investigators to identifytranscriptional networks associatedwith ethanol sensitivity behavior in the fruit fly Drosophila melanogaster(Morozova et al. 2011). This progres-sion from phenotype to gene networkto candidate gene and back to a genenetwork is a striking example of thepromise that combining geneticalgenomics and gene-network analysisprovides for understanding complextraits such as alcoholism.As previously mentioned, such net-

work-based techniques also have beenapplied to provide novel insight intothe functional consequences associatedwith ethanol exposure in the mesocor-ticolimbic reward pathway. Preliminaryresults have identified several gene-coexpression networks that are robustlyaltered by ethanol in a tightly coordi-nated fashion (Wolen et al., in press).These studies used the BXD panel, sothat the genetic regulation of ethanol-induced expression changes and behav-ioral responses also could be examined.Similar to results shown in figure 1,this analysis has revealed that ethanol-responsive gene networks are regulatedby a small number of loci that largelyare specific to a given network. At leastone of these loci overlaps a previouslymapped QTL for loss of righting reflex,a measure of acute ethanol sensitivity(Bennett et al. 2002; Markel et al.1997). This work suggests that focus-ing on identifying gene networks bothgreatly reduces the complexity of whole-genome expression data and provides a wealth of hypotheses regarding bothfunctional implications and regulatorymechanisms relevant to ethanol’s action.

Figure 5 Diagram of how the genomic approaches discussed here can be integrated to identifygene networks and candidate genes for complex traits such as alcoholism. Theinformation flow indicates how gene networks, expression quantitative trait locus(eQTL) and behavioral QTL analyses can be used together to identify candidategenes as potential targets for intervention. Note that resulting networks or candidategenes are entirely derived from experiments rather than relying on prior knowledge.In some cases, use of biomedical literature on gene–gene interactions can be usedto augment such experimentally-derived networks.

3 An endophenotype is a heritable trait or characteristic that is thought to be an intermediate phenotype between a geneticpredisposition and a clinical disorder; for example, certain neuro-biological characteristics have been noted in people with alcoholismand may be used as endophenotypes to identify people at risk foralcoholism. Endophenotypes are thought to be useful for geneidentification under the assumption that they are simpler andcloser to the genetic underpinnings of the disorder.

expression profiles associated with ethanolor alcoholism has provided modernneuroscience with a wealth of molecularinformation regarding ethanol’s effectson the body. At the same time, alcoholresearchers must make sense of a plethoraof weak genetic signals and large listsof genes whose expression changes inresponse to alcohol. Newer approaches,such as exome4 sequencing studies andcertain approaches to analyzing geneexpression (e.g., RNA-Seq analyses),promise added clarity but also maydeliver even more confusing data. Bycombining behavioral, genetic, andgenomic studies through geneticalgenomics and gene-network analysisdesigns, researchers may be able toconstruct gene networks rich in func-tional relations to ethanol behaviors.Additional refinements in ethanol-related gene-network structures andcausal relation of such networks toaspects of ethanol-induced behaviorswill provide a new generation of candi-date genes for therapeutic interventionin alcoholism. ■

Acknowledgements

The authors would like to thank Drs.Elissa Chesler, Michael Langston, andRobert Williams for discussions andcollaborations that contributed to thismanuscript, as well the anonymousreviewers for their helpful feedback.This work was supported by NationalInstitute on Alcohol Abuse andAlcoholism (NIAAA) grants P20AA–017828, U01 AA–016667 andU01 AA–016662 to MFM andNational Institute of Mental Health(NIMH) training grant MH–20030supporting ARW.

Financial Disclosure

The authors declare that they have nocompeting financial interests.

referencesALBERT, R., AND JEoNG, H. Internet: Diameter of the World-Wide Web. Nature 401:130–131, 1999.

BAILEy, D.W. Recombinant-inbred strains: An aid to find-ing identity, linkage, and function of histocompatibilityand other genes. Transplantation 11(3):325–327,1971. PMID: 5558564

BALDWIN, N.E.; CHESLER, E.J.; KIRoV, S.; ET AL. Computational,integrative, and comparative methods for the elucidationof genetic coexpression networks. Journal of Biomedicine& Biotechnology 2005(2):172–180, 2005. PMID: 16046823

BELKNAP, J.K., AND ATKINS, A.L. The replicability of QTLs for murine alcohol preference drinking behavior acrosseight independent studies. Mammalian Genome12(12):893–899, 2001. PMID: 11707775

BENNETT, B.; BEESoN, M.; GoRDoN, L.; AND JoHNSoN, T.E.Reciprocal congenics defining individual quantitativetrait loci for sedative/hypnotic sensitivity to ethanol.Alcoholism: Clinical and Experimental Research26(2):149–157, 2002. PMID: 11964553

BIERuT, L.J.; AGRAWAL, A.; BuCHoLz, K.K.; ET AL. A genome-wideassociation study of alcohol dependence. Proceedingsof the National Academy of Sciences of the UnitedStates of America 107(11):5082–5087, 2010. PMID:20202923

BoTSTEIN, D.; WHITE, R.L.; SKoLNICK, M.; AND DAVIS, R.W.Construction of a genetic linkage map in man usingrestriction fragment length polymorphisms. AmericanJournal of Human Genetics 32(3):314–331, 1980.PMID: 6247908

BoTToMLy, D.; WALTER, N.A.; HuNTER, J.E.; ET AL. Evaluating geneexpression in C57BL/6J and DBA/2J mouse striatum usingRNA-Seq and microarrays. PLoS One 6(3):e17820, 2011.PMID: 21455293

BREM, R.B.; yVERT, G.; CLINToN, R.; AND KRuGLyAK, L. Geneticdissection of transcriptional regulation in budding yeast.Science 296(5568):752–755, 2002. PMID : 11923494

BuTTE, A.J.; TAMAyo, P.; SLoNIM, D.; ET AL. Discoveringfunctional relationships between RNA expression and chemotherapeutic susceptibility using relevancenetworks. Proceedings of the National Academy ofSciences of the United States of America 97(22):12182–12186, 2000. PMID: 11027309

CARTER, S.L.; BRECHBüHLER, C.M.; GRIFFIN, M.; AND BoND, A.T.Gene co-expression network topology provides a frame-work for molecular characterization of cellular state.Bioinformatics 20(14): 2242–2250, 2004. PMID:15130938

CHEN, H., AND SHARP, B.M. Content-rich biological networkconstructed by mining PubMed abstracts. BMCBioinformatics 5:147, 2004. PMID: 15473905

CHESLER, E.J.; Lu, L.; SHou, S.; ET AL. Complex trait analysisof gene expression uncovers polygenic and pleiotropicnetworks that modulate nervous system function. NatureGenetics 37(3):233– 242, 2005. PMID: 15711545

CHESLER, E.J.; Lu, L.; WANG, J.; ET AL. WebQTL: Rapidexploratory analysis of gene expression and genetic networks for brain and behavior. Nature Neuroscience7(5):485–486, 2004. PMID : 15114364

CHESLER, E.J.; WANG, J.; Lu, L.; ET AL. Genetic correlates ofgene expression in recombinant inbred strains: A rela-tional model system to explore neurobehavioral pheno-types. Neuroinformatics 1(4):343–357, 2003. PMID:15043220

DANIELS, G.M., AND BuCK, K.J. Expression profiling identi-fies strain-specific changes associated with ethanolwithdrawal in mice. Genes, Brain, and Behavior1(1):35–45, 2002. PMID: 12886948

DEFRIES, J.C.; WILSoN, J.R.; ERWIN, V.G.; AND PETERSEN, D.R.LS × SS recombinant inbred strains of mice: Initial char-acterization. Alcoholism: Clinical and ExperimentalResearch 13(2):196– 200, 1989. PMID: 2658655

EISEN, M.B.; SPELLMAN, P.T.; BRoWN, P.o.; AND BoTSTEIN, D.Cluster analysis and display of genome-wide expressionpatterns. Proceedings of the National Academy ofSciences of the United States of America 95(25):14863–14868, 1998. PMID: 9843981

GIRVAN, M., AND NEWMAN, M.E. Community structure insocial and biological networks. Proceedings of theNational Academy of Sciences of the United States ofAmerica 99(12):7821–7826, 2002. PMID: 12060727

GRAHAME, N.J.; LI, T.K.; AND LuMENG, L. Limited accessalcohol drinking in high- and low-alcohol preferring select-ed lines of mice. Alcoholism: Clinical and ExperimentalResearch 23(6):1015– 1022, 1999. PMID: 10397285

GRIEVE, S.J., AND LITTLEToN, J.M. Age and strain differencesin the rate of development of functional tolerance toethanol by mice. Journal of Pharmacy and Pharmacology31(10):696–700, 1979. PMID: 41043

HoRVATH, S., AND DoNG, J. Geometric interpretation ofgene coexpression network analysis. PLoS ComputationalBiology 4(8):e1000117, 2008. PMID: 18704157

INouE, M.; RASHID, M.H.; FuJITA, R.; ET AL. Initiation of neu-ropathic pain requires lysophosphatidic acid receptorsignaling. Nature Medicine 10(7):712–718, 2004.PMID: 15195086

JANSEN, R.C., AND NAP, J.P. Genetical genomics: Theadded value from segregation. Trends in Genetics17(7):388–391, 2001. PMID: 11418218

JEoNG, H.; ToMBoR, B.; ALBERT, R.; ET AL. The large-scaleorganization of metabolic networks. Nature 407(6804):651–654, 2000. PMID: 11034217

KERNS, R.T.; RAVINDRANATHAN, A.; HASSAN, S.; ET AL. Ethanol-responsive brain region expression networks: Implicationsfor behavioral responses to acute ethanol in DBA/2J ver-sus C57BL/6J mice. Journal of Neuroscience 25(9):2255–2266, 2005. PMID: 15745951

KIRSTEIN, S.L.; DAVIDSoN, K.L.; EHRINGER, M.A.; ET AL.Quantitative trait loci affecting initial sensitivity andacute functional tolerance to ethanol-induced ataxia andbrain cAMP signaling in BXD recombinant inbred mice.Journal of Pharmacology and Experimental Therapeutics302(3):1238–1245, 2002. PMID: 12183685

KRIL, J.J., AND HALLIDAy, G.M. Brain shrinkage in alcoholics:A decade on and what have we learned? Progress inNeurobiology 58(4):381– 387, 1999. PMID: 10368034

LANGFELDER, P., AND HoRVATH, S. WGCNA: An R package forweighted correlation network analysis. BMC Bioinformatics9:559, 2008. PMID: 19114008


4 The exome is the part of the genome formed by the coding por-tions of genes (i.e., exons) that provide the genetic blueprint usedin the synthesis of proteins and other functional gene products. Itis the functionally most relevant part of the genome and, there-fore, most likely to contribute to the phenotype of an organism.


LEWoHL, J.M.; WANG, L.; MILES, M.F.; ET AL. Gene expres-sion in human alcoholism: Microarray analysis of frontalcortex. Alcoholism: Clinical and Experimental Research24(12):1873–1882, 2000. PMID: 11141048

MARKEL, P.D.; BENNETT, B.; BEESoN, M.; ET AL. Confirmation ofquantitative trait loci for ethanol sensitivity in long-sleep andshort-sleep mice. Genome Research 7(2):92–99, 1997.PMID: 9049627

MATAS-RICo, E.; GARCIA-DIAz, B.; LLEBREz-zAyAS, P.; ET AL.Deletion of lysophosphatidic acid receptor LPA1 reducesneurogenesis in the mouse dentate gyrus. Molecularand Cellular Neurosciences 39(3):342–355, 2008.PMID: 18708146

MAyFIELD, R.D.; LEWoHL, J.M.; DoDD, P.R.; ET AL. Patterns ofgene expression are altered in the frontal and motorcortices of human alcoholics. Journal of Neurochemistry81(4):802–813, 2002. PMID: 12065639

METTEN, P., AND CRABBE, J.C. Common genetic determi-nants of severity of acute withdrawal from ethanol, pen-tobarbital and diazepam in inbred mice. BehaviouralPharmacology 5(4 and 5):533–547, 1994. PMID:11224305

MoRozoVA, T.V.; MACKAy, T.F.; AND ANHoLT, R.R. Transcriptionalnetworks for alcohol sensitivity in Drosophila melanogaster.Genetics 187(4): 1193–1205, 2011. PMID: 21270389

MozHuI, K.; CIoBANu, D.C.; SCHIKoRSKI, T.; ET AL. Dissectionof a QTL hotspot on mouse distal chromosome 1 thatmodulates neurobehavioral phenotypes and geneexpression. PLoS Genetics 4(11): e1000260, 2008.PMID: 19008955

MuLLIGAN, M.K.; PoNoMAREV, I.; HITzEMANN, R.J.; ET AL.Toward understanding the genetics of alcohol drinkingthrough transcriptome meta-analysis. Proceedings ofthe National Academy of Sciences of the United Statesof America 103(16):6368–6373, 2006. PMID: 16618939

MuLLIGAN, M.K.; RHoDES, J.S.; CRABBE, J.C.; ET AL. Molecularprofiles of drinking alcohol to intoxication in C57BL/6Jmice. Alcoholism: Clinical and Experimental Research35(4):659–670, 2011. PMID: 21223303

NICoLAE, D.L.; GAMAzoN, E.; zHANG, W.; ET AL. Trait-associat-ed SNPs are more likely to be eQTLs: Annotation toenhance discovery from GWASs. PLoS Genetics 6(4):e1000888, 2010. PMID: 20369019

PHILLIPS, T.J.; BRoWN, K.J.; BuRCKHART-KASCH, S.; ET AL.Alcohol preference and sensitivity are markedly reducedin mice lacking dopamine D2 receptors. NatureNeuroscience 1(7):610–615, 1998. PMID: 10196569

RASK, E.; WALKER, B.R.; SöDERBERG, S.; ET AL. Tissue-specificchanges in peripheral cortisol metabolism in obesewomen: Increased adipose 11beta-hydroxysteroid dehydrogenase type 1 activity. Journal of ClinicalEndocrinology and Metabolism 87(7):3330–3336,2002. PMID: 12107245

REIMERS, M. A.; RILEy, B. P.; KALSI, G.; ET AL. Pathway basedanalysis of genotypes in relation to alcohol dependence.Pharmacogenomics Journal, 2011, April 5 [Epub ahead ofprint]. PMID: 21468025

RIMoNDINI, R.; ARLINDE, C.; SoMMER, W.; AND HEILIG, M. Long-lasting increase in voluntary ethanol consumption andtranscriptional regulation in the rat brain after intermittentexposure to alcohol. FASEB Journal 16(1):27–35, 2002.PMID: 11772933

RuANo, D.; ABECASIS, G.R.; GLASER, B.; ET AL. Functionalgene group analysis reveals a role of synaptic het-erotrimeric G proteins in cognitive ability. AmericanJournal of Human Genetics 86(2):113–125, 2010.PMID: 20060087

SABA, L.; BHAVE, S.V.; GRAHAME, N.; ET AL. Candidate genesand their regulatory elements: Alcohol preference andtolerance. Mammalian Genome 17(6):669–688, 2006.PMID: 16783646

SAITo, M.; SzAKALL, I.; ToTH, R.; ET AL. Mouse striatal tran-scriptome analysis: Effects of oral self-administration ofalcohol. Alcohol 32(3):223–241, 2004. PMID: 15282116

SCHADT, E.E.; LAMB, J.; yANG, X.; ET AL. An integrativegenomics approach to infer causal associationsbetween gene expression and disease. Nature Genetics37(7):710–717, 2005. PMID: 15965475

SCHADT, E.E.; MoNKS, S.A.; DRAKE, T.A.; ET AL. Genetics ofgene expression surveyed in maize, mouse and man.Nature 422(6929):297–302, 2003. PMID: 12646919

SCHuCKIT, M. A. Subjective responses to alcohol in sonsof alcoholics and control subjects. Archives of GeneralPsychiatry 41(9):879–884, 1984. PMID: 6466047

SCHuCKIT, M.A. Low level of response to alcohol as a predic-tor of future alcoholism. American Journal of Psychiatry151(2):184–189, 1994. PMID: 8296886

SCoTT, R.E.; WHITE-GRINDLEy, E.; RuLEy, H.E.; ET AL. P2P-Rexpression is genetically coregulated with componentsof the translation machinery and with PuM2, a transla-tional repressor that associates with the P2P-R mRNA.Journal of Cellular Physiology 204(1):99–105, 2005. PMID:15617101

STuART, J.M.; SEGAL, E.; KoLLER, D.; AND KIM, S.K. A gene-coexpression network for global discovery of conserved

genetic modules. Science 302 (5643):249–255, 2003.PMID: 12934013

TABAKoFF, B.; SABA, L.; PRINTz, M.; ET AL. Genetical genomicdeterminants of alcohol consumption in rats andhumans. BMC Biology 7:70, 2009. PMID: 19874574

TREADWELL, J.A., AND SINGH, S.M. Microarray analysis ofmouse brain gene expression following acute ethanoltreatment. Neurochemical Research 29(2):357–369,2004. PMID: 15002731

VAN’T VEER, L.J.; DAI, H.; VAN DE VIJVER, M.J.; ET AL. Geneexpression profiling predicts clinical outcome of breastcancer. Nature 415(6871): 530–536, 2002. PMID:14654526

VAN NooRT, V.; SNEL, B.; AND HuyNEN, M.A. The yeast coex-pression network has a small-world, scale-free architec-ture and can be explained by a simple model. EMBOReports 5(3):280–284, 2004. PMID: 14968131

WATTS, D.J., AND STRoGATz, S. H. Collective dynamics of“small-world” networks. Nature 393(6684): 440–442,1998. PMID: 9623998

WICKHAM, H. ggplot2: Elegant Graphics for Data Analysis.New york: Springer, 2009.

WILLIAMS, R.W.; BENNETT, B.; Lu, L.; ET AL. Genetic structureof the LXS panel of recombinant inbred mouse strains: Apowerful resource for complex trait analysis. MammalianGenome 15(8):637– 647, 2004. PMID: 15457343

WILLIAMS, R.W.; Gu, J.; QI, S.; AND Lu, L. The genetic structureof recombinant inbred mice: High-resolution consensusmaps for complex trait analysis. Genome Biology2(11):RESEARCH0046, 2001. PMID: 11737945

WoLEN, A.R.; PHILLIPS, C.A.; Langston, M.A.; et al. Geneticdissection of acute ethanol respoonsive gene networksin prefrontal cortex; Functional and mechanistic implica-tions. PLoS, in press.

yoRK, T.P.; MILES, M.F.; KENDLER, K.S.; ET AL. Epistatic andenvironmental control of genome-wide gene expression.Twin Research and Human Genetics 8(1):5–15, 2005.PMID: 15836804

zHANG, B., AND HoRVATH, S. A general framework forweighted gene co-expression network analysis.Statistical Applications in Genetics and MolecularBiology 4:Article17, 2005. PMID: 16646834

zHu, J.; LuM, P.y.; LAMB, J.; ET AL. An integrative genomicsapproach to the reconstruction of gene networks in seg-regating populations. Cytogenetic and GenomeResearch 105(2–4):363–374, 2004. PMID: 15237224

identifying gene networks underlying the neurobiology of … · identifying gene networks...

Documents