Research on genetic risk has concentrated on laboratory animals, particularly rodents. Most symptoms important for a diagnosis of alcoholism or an alcohol use disorder as currently defined are behavioural. These include failure to control quantity or frequency of drinking, and continuing to drink in the face of consequential medical, legal or social/familial problems. It is clearly difficult to envision completely believable rodent models for such behaviours as getting into trouble at work, a limit to the usefulness of such models (Lovinger & Crabbe 2005). Nonetheless, the tractability of rodent neurobiology, and especially the flowering of mouse genetics and genomics, have made studies of these species attractive. While I will mention other species and attempt to draw parallels with human genetic data where possible, the focus of this review will be on rodent genetic studies. Results from four methods for approaching the neurogenetics of alcoholism will be assessed—inbred strain studies, work with selectively bred lines, candidate gene targeting approaches and gene expression profiling.

I do not believe that there is a rodent model that resembles human alcoholism in all important aspects. The inherent fallacies in attempting such simulacra were eloquently outlined many years ago (McClearn 1979). Rather, I advocate concentrating on specific aspects of the disorder that can be convincingly modelled in rats and mice. These include aspects of the pharmacology of ethyl alcohol (neural sensitivity to the drug, the development of tolerance to its effects with repeated administration, and withdrawal symptoms shared across mammalian species, e.g. seizures or convulsions, autonomic dysregulation). Extensive studies of the neuroadaptations accompanying tolerance development have shown genetic influences to be strong (Kalant et al. 1971; Kalant 1998). Recent studies using the powerful genetic manipulations available in invertebrate species have extended the range of genetic influences thought to be responsible (Scholz et al. 2000; Cowmeadow et al. 2005).

An inbred strain results when brothers and sisters are mated, generation after generation. In each generation, half the allelic genetic variability is lost, until after 20 generations, all same-sex animals have two copies of the same allele for any gene—they are obligatory homozygotes throughout their genome (Falconer & Mackay 1996). However, each inbred strain is genetically unique, and different from all others. The degree of difference depends upon the pedigree history of the strain's derivation, and there are seven major lineages for the more than 100 standard inbred strains that are commercially available (Petkov et al. 2004). If a panel of inbred strains is tested under controlled environmental conditions, the differences in a behaviour (or a neurochemical phenotype, for example) are taken to be of allelic genetic origin. To the extent that they exceed the average individual differences within a strain, they serve as an estimate of the aggregate effect of genetics on the phenotype.

Within an inbred strain, however, there are always individual differences. Most C57BL/6J mice show increasing alcohol drinking over a period of two to three weeks, for example, which was noted in the earliest paper (McClearn & Rodgers 1959). The source of these differences cannot be allelic, and must derive from environmental factors (e.g. social dominance in group-housed animals, food, how well they slept the night before testing). Alternatively, they may be epigenetically based, whether or not they are transmitted to their offspring (Francis et al. 1999, 2003).

What have we learned from inbred strain studies of alcohol-related phenotypes since 1959? When the inbred strain literature on behavioural responses (Phillips & Crabbe 1991) and neurochemical strain differences (Allan & Harris 1991) were first comprehensively reviewed in 1991, only a handful of studies were published and these were nearly all limited to comparisons of the C57BL/6J and DBA/2J strains, previously identified as outliers for alcohol drinking and for the severity of alcohol withdrawal (Kakihana 1979). However, the development of recombinant inbred (RI) strains (Bailey 1971) extended the use of inbred strains to a new purpose. RI strains are derived from re-inbreeding after intercrossing two inbred strains to obtain an F2 generation. The F2 population shows a mosaic pattern of DNA segments on each chromosome due to crossovers during meiosis. After inbreeding to form a new RI strain is complete, the order of genetic segments randomly reshuffled in the genome of the F2 animals is preserved in future generations with the same fidelity as in standard inbreds (Silver 1995). With the advent of high-throughput genomics enabled by the Human Genome Project, genomic markers were rapidly developed. These started with restriction fragment length polymorphisms, soon thereafter succeeded by microsatellite repeat markers, and most recently by single nucleotide polymorphisms, or SNPs. The stability of inbred strains thus allowed genome scientists to build a mouse genome map in parallel with the human genome map.

The mouse map was completed in 2002, but long before that, the genetic marker map was sufficiently dense to allow mapping the location of genes whose alleles were correlated with high or low values on any measured phenotype. Thus, it was possible to study the 26 B×D RI strains then in existence, each derived from crossing (high-drinking strain) C57BL/6J mice with (low-drinking strain) DBA/2J mice. These B×D RI strains differed markedly in their preference for ethanol, and because each possessed a stable mixture of marker alleles that were inherited from either C57BL/6J or DBA/2J progenitors, the location of segments of DNA associated with high and low drinking could be ‘read’ directly from the patterns of RI strain phenotypic differences (Phillips et al. 1998). These locations are initially deemed ‘loci,’ called ‘quantitative trait loci’ (QTL) for two reasons: first, because the trait, drinking, is influenced by many genes, its inheritance pattern is not all or none (Mendelian), but rather graded or quantitative; and second, because any QTL comprises a substantial stretch of chromosomal DNA and includes multiple genes. The stability of inbred strains across generations and laboratories has allowed several groups to combine similar analyses, and several QTL affecting preference drinking have been firmly supported in multiple studies (Rodriguez et al. 1994; Melo et al. 1996; Phillips et al. 1998; Tarantino et al. 1998). However, none of these QTL for drinking has as yet been resolved to the level of a single quantitative trait gene (QTG). This is for several reasons, including (but not limited to) the difficulty of manipulating all the candidate genes in the QTL interval to rule in or out their role; the possibility that multiple genes may reside in a single QTL interval; and the small effect size for any QTL.

QTL mapping has led to at least one QTG. Acute withdrawal from ethanol and pentobarbital has been studied in standard inbreds and the B×D RI strains, and provisional QTL identified (Buck et al. 1997). Subsequent studies winnowed one QTL on chromosome 4 to a few genes (Fehr et al. 2002). Finally, comparisons of purpose-bred mice and standard inbred strains ruled out all but a single gene, Mpdz, as the source of the effects on withdrawal severity (Shirley et al. 2004). This gene encodes a scaffolding protein that participates in neurotransmitter–receptor interactions, and current studies are exploring possible specific partners affected by the Mpdz polymorphism. There are many new methods for mapping QTL and isolating the specific genes of importance. Inbred strains can be surveyed directly for associations between markers and phenotypes (Grupe et al. 2001; Liao et al. 2004), although the statistical mapping power of such analyses has been questioned due to the close pedigree relationships among standard inbred strains, leading to large haplotype similarities. Mapping studies are more difficult, but more powerful, in outbred stocks (Mott et al. 2000; Flint et al. 2005; Valdar et al. 2006).

Another feature of inbred strains that is very useful is their cumulative power. A recent effort has sought to enter phenotypic information for many of the widely used strains into a database. The Mouse Phenome Database (MPD) http://phenome.jax.org/pub-cgi/phenome/mpdcgi?rtn=docs/home contains nearly 1200 phenotypes as of this writing (May 2008) which have been collected on at least eight inbred strains (average of 18 strains/phenotype), and targets 36 strains as high priority. The phenotypes range from behavioural to neurochemical, anatomical and physiological and are derived from 68 projects; some data are available for 598 strains. Owing to the inherent replicability of inbred strain data (Wahlsten et al. 2006), strain mean data can be correlated across phenotypes, and a significant correlation across strain means suggests that some genes affect both phenotypes. The increasing availability of SNP data in the MPD and elsewhere further strengthens the usefulness of this resource.

For example, several studies have explored the negative genetic relationship suggested by McClearn's group between high drinking (McClearn & Rodgers 1959) and low withdrawal (Kakihana 1979). A meta-analysis of strain differences on these two phenotypes found them to be significantly negatively genetically correlated. This correlation was obtained despite differences in the specific preference phenotypes examined (short versus long exposure ethanol preference tests), and acute versus chronic ethanol withdrawal (Metten et al. 1998). A later study surveyed strains for their sensitivity to ethanol's effectiveness in conditioning a taste aversion (CTA) to a paired, novel flavour. Ethanol's efficacy in the CTA paradigm was correlated with high withdrawal and low preference drinking (Broadbent et al. 2002). A recent meta-analysis of dozens of published studies with genetically defined rats and mice has included assessment of panels of inbred mouse strains (Green & Grahame 2008). This analysis confirmed the high preference–low withdrawal–low CTA triad of association, which was also seen in lines selectively bred for drinking (see §3). The authors argue that genetically high drinkers appear to be more sensitive to some aversive properties of ethanol, which limit preference drinking. Interestingly, although there was a strong and consistent association between home cage drinking and intravenous self-administration across genotypes, there was a weaker association between these consummatory responses and the efficacy of ethanol to condition a place preference. The place preference data are complicated by a species difference: whereas mice prefer ethanol-associated locations, rats generally learn to avoid them, for reasons as yet not understood (Green & Grahame 2008). An older study of several inbred mouse strains documented differences in their ability to withhold a rewarded nose poke response for a few seconds. There was a strong correlation between this behavioural analogue of impulsivity and the preference of the strains for alcohol (Logue et al. 1998). A study of impulsive action in multiple mouse strains is finding their response inhibition to a no-go signal in a go/no-go task to be well correlated with the pre-signal impulsive nose-poke data (S. H. Mitchell et al. 2008, personal communication).

Starting in the 1940s, Mardones successfully bred rats for high ethanol preference, proving that the trait was heritable (Mardones & Segovia-Riquelme 1983). The basic experimental design has been used many times since to create other high-preferring rat and mouse lines. High-preferring mice or rats may reach high blood alcohol levels (BALs), but usually only after many months of drinking. More typically, they stop drinking at BALs of approximately 60mg%. In other words, rodents appear to have some internal controls limiting intake which are not shared by susceptible humans. Neurobiological and genetic findings from these selected lines are discussed in a series of recent reviews (Murphy et al. 2002; Bell et al. 2006; Colombo et al. 2006; Quintanilla et al. 2006; Sommer et al. 2006). UChB (versus UChA), P (versus NP), AA (versus ANA), HAD (versus LAD), sP (versus sNP) and msP (versus msNP) rats (and HAP versus LAP mice) were all bred for the same trait using a nearly identical protocol; one could reasonably expect that strong genetic correlates of high alcohol preference should have emerged from these multiple replications.

Much of the early work with the selectively bred rat lines was focused on assessments of neurotransmitter levels, their activity, their interactions with brain receptors and the neuroanatomical differences in these brain systems, in comparisons with naive animals. There is substantial agreement across models that low limbic system levels of serotonin and other serotonergic differences characterize high drinkers (Murphy et al. 2002). Interestingly, dopaminergic differences have not been reported as consistently (Sommer et al. 2006). Differences in endocannabinoid, opioid, GABA, glutamate and other systems remain to be established clearly in multiple models. Somewhat more clarity has evolved from behavioural comparisons of the multiple pairs of selected lines. High drinkers are more sensitive to low-dose alcohol and less sensitive to higher-dose, sedative effects in some studies, but these results have not been generalized consistently across studies. There is good agreement that high drinkers tend to develop tolerance more readily than non-drinkers, and that it persists longer. As noted in analyses of inbred strain panels, the relationship between ethanol-induced CTA and drinking is shared across selections (Phillips et al. 2005; Green & Grahame 2008). Finally, mice were bred from the F2 cross of C57BL/6J×DBA/2J in a short-term selection for high- or low-preference drinking. Mice from the high drinking line were found to be more impulsive in a go/no-go response inhibition task than the low drinkers, but the lines did not differ in a task of impulsive choice, delay discounting (Wilhelm et al. 2007).

Rats have also been bred for their intake of ethanol in a limited access paradigm. HARF rats drink substantial amounts of ethanol in a 20min access period, versus the low-drinking LARF line (Lê et al. 2001). However, these animals have been rarely studied (Shram et al. 2004; Turek et al. 2005). To overcome some limitations of the older selected lines, we have developed a new model for binge drinking. C57BL/6J mice had been shown to self-administer substantial ethanol in a limited access paradigm during the circadian dark (Sharpe et al. 2005). Mice drink and eat mostly during their circadian dark period. In our paradigm, C7BL/6J mice drinking in the dark (DID) readily self-administered approximately 7–8g ethanolkg⁻¹ body weight in 4 hours and became intoxicated (Rhodes et al. 2005). We later showed that inbred strains differed in DID and the response was both reliable and heritable (Rhodes et al. 2007). We are currently selectively breeding mice for this trait: after nine selected generations, average BAL has doubled (because average intake has increased 150%) and the high DID (HDID-1) mice also become intoxicated (Crabbe et al. submitted). These mice should be useful for mechanistic studies of the biological and genetic contributions to excessive drinking. They have not been tested for any other traits. However, the DID phenotype is proving to be a useful way to explore mechanisms underlying binge-like drinking. DID was shown to be reduced by intraperitoneal naltrexone doses that did not affect water or sugar–water intake (Kamdar et al. 2007). Another group has shown that the GABA_B agonist baclofen reduced ethanol DID without affecting water drinking, while two GABA_A agonists, THIP and muscimol, reduced drinking non-specifically (Moore et al. 2007). A CRF₁ receptor antagonist also reduced ethanol DID without affecting sucrose drinking (Sparta et al. 2008). Finally, infusion of urocortin 1 into lateral septum selectively reduced ethanol DID, while CRF reduced intake of both ethanol and water (Ryabinin et al. 2008).

Many other lines have been selectively bred for differential responses to alcohol. These include the long-sleep (LS) and short-sleep (SS) mice, which were selected for differences in the duration of the loss of the righting reflex after high-dose ethanol. These mice have been the subject hundreds of papers, and have been used effectively to map QTL underlying this response (Bennett et al. 2006). A large panel of RI strains has been developed from an LS×SS cross, and these RI strains have also been used to map QTL for ethanol-induced locomotor activation and acute functional tolerance development (Downing et al. 2006; Bennett et al. 2007). The locomotor activating response to ethanol is taken as an analogue of ethanol's euphoric effects in humans. Mice selected for high ethanol-induced activity (FAST) have been compared with low responders (SLOW) in many studies. These findings have been reviewed elsewhere (Palmer & Phillips 2002; Cunningham & Phillips 2003). Selection for differences in the severity of withdrawal after chronic ethanol treatment has been the subject of multiple selective breeding efforts. The Withdrawal Seizure-Prone and -Resistant selected lines remain under active investigation in several laboratories. Early work with these lines has been comprehensively reviewed (Metten & Crabbe 1996) and much of the recent work has focused on the role of neurosteroid systems in withdrawal (Finn et al. 2004; Beckley et al. 2008).

Since the initial review, 34 additional papers have appeared, adding information about alcohol responses for several previously targeted genes, but also introducing 14 new genes to the mix. Two studies used ENU mutagenesis to generate novel mutants, some of which were found to differ in alcohol responses (Pawlak et al. 2005; Hamre et al. 2007). In one study, cerebellar tissue from ethanol-treated mice with or without a protein kinase C (PKC) gamma null mutation was compared in a gene expression profiling analysis. Several genes of potential relevance to the failure of PKC gamma mutants to develop ethanol tolerance were identified (Bowers et al. 2006). With no disrespect intended to the authors of the new studies, the increase in knowledge since 2006 has not revealed any new striking findings (not reviewed here). Gene targeting will remain a very useful tool for confirming the importance of a gene for which there are parallel lines of evidence for its importance. However, it is unlikely to be a sufficient tool for proving gene involvement on an alcohol response.

Gene expression profiling analyses typically discover numerous genes differentially expressed between control and treatment tissues, or between genotypes with high and low sensitivity to alcohol. Increasingly, the resulting lists of genes are being replaced with more sophisticated bioinformatics analyses of the output that seek gene clusters, gene families and/or similarity across multiple studies. For example, data banks containing the whole brain gene expression outputs from three pairs of lines selectively bred for high- or low-preference drinking were explored in conjunction with data from six inbred or isogenic strains of mice. A meta-analysis found nearly 4000 genes consistently differentially expressed in high versus low drinkers (Mulligan et al. 2006). Functional groups were proposed based on analysis of the proteins coded by those genes (the transcriptome). Using congenic strains, which differed from the background strain only in one small piece of DNA from chromosome 9 that contained a QTL for preference drinking first reported in an analysis of B×D RI strains (Phillips et al. 1994), more microarray analyses were performed. Comparison of the datasets reduced the number of candidate genes within the QTL to 16 plus 4 expressed sequence tags (ESTs). These candidates included Scn4b, encoding a sodium channel.

One area of promise is the link between endogenous opioids and alcohol dependence. Substantial data from animal models implicate common genetic contributors to endogenous opioid function and alcohol self-administration or dependence (Town et al. 2000). Numerous gene targeting studies suggest a specific role for the mu opioid receptor in ethanol preference, as its deletion reduces intake in many studies (Crabbe et al. 2006). However, there does not appear to be a straightforward homology with the human receptor. A meta-analysis of many studies seeking to test an association between the specific Asn40Asp polymorphism in the mu opioid receptor, OPRM1, and substance dependence failed to find a significant elevation in relative risk (Arias et al. 2006), even in the several studies of alcohol-dependent subjects. Nonetheless, naltrexone remains an effective therapy for some alcoholics (Heilig & Egli 2006), and a recent study reported an association between response to naltrexone therapy and the Asn40Asp polymorphism in a population of alcoholics (Anton et al. 2008). It should be noted that a smaller study of patients in several Veterans Affairs hospitals saw no similar association (Gelernter et al. 2007).

Animal studies have provided a convincing link between the stress axis and ethanol drinking. Mice lacking functional CRF-1 receptors show normal ethanol preference, but the increase in preference following stress may be attenuated (Sillaber et al. 2002). A similar effect is seen when a CRF-1 receptor antagonist is administered to normal mice (Lowery et al. 2008), and, as noted above, an antagonist also attenuates ethanol DID (Sparta et al. 2008). Also as noted above, intra-septal CRF and urocortin 1 both reduce DID (Ryabinin et al. 2008). Ethanol intake can also be potentiated following induction of physical dependence (Lopez & Becker 2005), and these increases are also blocked by CRF₁ antagonists (Chu et al. 2007; Finn et al. 2007; Richardson et al. 2008) or are absent in the knockout (Chu et al. 2007). A recent study has suggested that a polymorphism in the human CRHR1 gene is associated with heavy drinking in a sample of German 15-year olds in some measures, only if associated with more stressful life events (Blomeyer et al. 2008). It should be noted that two previous studies with adult samples have failed to find significant associations (Dahl et al. 2005; Treutlein et al. 2006).

Finally, I mention the interesting story of alpha synuclein. QTL mapping studies in crosses between P and NP rats identified a region of rat chromosome 4, among others (Bice et al. 1998; Carr et al. 1998). A subsequent gene expression profiling analysis compared tissue from multiple brain areas from inbred P and NP rats, and found many genes and ESTs to be differentially expressed (Liang et al. 2003). Although alpha synuclein had not been mapped in the rat, it was selected as a candidate gene based on its location in the homologous portion of mouse chromosome 6, and it was shown to be expressed more highly in P hippocampus than in NP. Levels of mRNA in the blood of alcohol self-administering cynomolgus monkeys were higher than in controls (Walker & Grant 2006). Subsequent studies reported increased alpha synuclein mRNA (Bonsch et al. 2004b) and protein (Bonsch et al. 2005a) levels in the blood of alcoholics versus that of controls. This group also reported increased levels of homocysteine in plasma of alcoholics, and further that these patients showed a correlated increase in levels of DNA methylation (Bonsch et al. 2004a). Increased plasma homocysteine levels were also correlated with cognitive deficits in early withdrawal (Wilhelm et al. 2006). The DNA hypermethylation is pronounced in the area of the alpha synuclein, but not the presenilin 1, promoter (Bonsch et al. 2005b). Thus, the alpha synuclein story may depend upon epigenetic modifications of the expression of other genes as yet unidentified.

This story, like all others, has its qualifications. A study of 219 multiplex families reported both a SNP and a haplotype association of alpha synuclein polymorphisms with alcohol craving, but not with alcohol dependence (Foroud et al. 2007). However, a subsequent study of two American Indian populations found scattered, weak associations with addiction-related variables, but none survived correction for multiple comparisons (Clarimon et al. 2007). In addition it must be noted that the original difference between P and NP rats was not seen in comparisons of the two pairs of HAD and LAD lines selected for the same phenotype. Nor has any association with the region of mouse chromosome 6 that harbours the gene ever emerged from the many QTL studies in the literature.