Confirmation of prior evidence of genetic susceptibility to alcoholism in a genome wide association study of comorbid alcoholism and bipolar disorder

Introduction

Alcoholism and unipolar affective disorders are strongly comorbid (Davis et al., 2008, Pottenger et al., 1978, Helzer and Pryzbeck, 1988, Regier et al., 1990, McElroy et al., 2001, Kessler et al., 1997). Several studies have shown that up to 70% of cases of alcoholism also have major depressive disorder (Davis et al., 2008, Pottenger et al., 1978). This comorbid unipolar affective disorder has been repeatedly shown to be both primary and secondary to alcoholism (Gurling et al., 1984, Merikangas and Gelernter, 1990, Merikangas et al., 1985, Merikangas et al., 1994). Comorbidity between bipolar disorder (BPD) and alcoholism is well established from epidemiological, twin, family and linkage studies (O’Sullivan et al., 1988, Gurling et al., 1984, Merikangas and Gelernter, 1990, Preisig et al., 2001, Sonne and Brady, 2002, Berrettini et al., 1997, Winokur et al., 1996). These studies show that about 50% of individuals with bipolar disorder also manifest alcohol abuse or dependence and altogether about 75% have a substance use disorder.

The early twin and adoption studies can be interpreted as suggesting that there is a higher heritability for alcoholism combined with criminality than for other subtypes of alcoholism (Kaij, 1960, Gurling et al., 1984, Gurling and Cook, 1999). Men with alcohol dependence have much higher rates of dis-social personality disorder and criminality than women with alcoholism (Cloninger et al., 1978, Cloninger et al., 1986). Two important adoption studies from Sweden found replicated evidence for genetic effects associated with criminality and male alcoholism, known as type two alcoholism, which had an early age of onset and strong genetic transmission (Sigvardsson et al., 1982). Women with alcohol dependence have higher rates of depression and anxiety than men and the heritability may be higher (Klerman et al., 1996, Merikangas et al., 1985, Merikangas et al., 1994, Gurling and Cook, 1999, Kuo et al., 2006a, Bierut et al., 1999). The relationship between anxiety disorders and alcoholism seems to be very similar to the relationship between alcoholism and depression (Mullan et al., 1986, Merikangas et al., 1994).

Alcoholism is strongly recurrent in both the first and second degree relatives of alcoholic probands (Cotton, 1979). This is consistent with there being both genetic and cultural effects on risk. Genetic effects in alcoholism were first established using classical twin and adoption studies (Kaij, 1960, Bohman, 1978, Sigvardsson et al., 1996, Sigvardsson et al., 1982, Ball, 2005). Heritability estimates for human diseases have inherent limitations but one twin study of alcohol dependence estimated heritability to be between 50% and 61% (Hansell et al., 2008). The largest twin study, of 9,000 male pairs, found that alcoholism was 53% “heritable” (Kendler et al., 1997). However, twin studies do not always show substantial heritabilities or increased concordance for alcoholism in monozygotic (MZ) compared to dizygotic (DZ) female twins (Gurling et al., 1984, Pickens et al., 1991, Prescott et al., 2007, Prescott et al., 2005).

Genetic linkage studies on large samples of families have been carried out by several research groups in European and Native American populations (Cook et al., 1996, Guerrini et al., 2005, Kendler et al., 2006, Kuo et al., 2006b, Hesselbrock et al., 2001, Ehlers et al., 2004). The most important have been those from the USA Consortium on the Genetics of Alcoholism (COGA) (Foroud et al., 2000). For example a lod above 3.00 for linkage with alcohol dependence syndrome (ADS) was found on chromosome 3. Another lod above 3.00 with ADS was found on chromosome 6 (Cantor and Lanning, 1999). Lods above 3.00 were found with alcoholism on chromosome 11 and 4 in Native American families (Long et al., 1998). Some of these positive findings were replicated in a linkage study in Ireland (Prescott et al., 2006). In the USA COGA study a combined ADS and conduct disorder (Dick et al., 2004) phenotype showed positive lod scores over a region of chromosome 2 that includes the Tachykinin A Receptor 1 (TACR1) locus. The TACR1 locus was later reported to be associated with bipolar disorder and attention deficit hyperactivity disorder (ADHD) (Perlis et al., 2008, Ferreira et al., 2008, Yan et al., 2009). Lod scores greater than 3.0 were found on chromosomes 7, 2, 3, 5, 9, and 14 when antisocial families were selected to achieve greater homogeneity (Jacobson et al., 2008). Replication of the linkage to alcoholism on chromosome 1 originally found by the USA COGA study (Dick et al., 2002) has been achieved in Finland (Lappalainen et al., 2004) with further support from a linkage study in the UK (Guerrini et al., 2005). This locus appears to be showing linkage to a combined depression and alcoholism phenotype (Nurnberger et al., 2001).

Many small and medium scale case control genetic association studies have already shown that genetic effects on susceptibility to alcoholism are partly derived from neurotransmitter receptor variants (Kohnke, 2008), from alcoholism metabolism genes (Dick et al., 2007, Kuo et al., 2008, Luo et al., 2005) and from other central nervous system (CNS) related genes. Most striking in alcoholism is the protective effect of the alcohol-metabolising aldehyde dehydrogenase 2 (ALDH2) variants in Japanese and Chinese Asian populations (Yoshida, 1992, Agarwal and Goedde, 1987). The possible involvement of the alcohol dehydrogenase (ADH) gene cluster has been reinforced by linkage studies repeatedly implicating chromosome 4 (Luczak et al., 2006, Edenberg et al., 2006) and by several allelic assocation studies (Kuo et al., 2008, Edenberg et al., 2006, Ball, 2005). The GABA receptor alpha2 subunit (GABRA2) gene is the most replicated gene increasing risk of alcoholism (reviewed in Enoch (2010) and recently examined in Bierut et al., 2010). The Bierut study was a GWAS of alcohol dependence syndrome using 1,897 alcohol dependent individuals drawn from the COGA study, the Family Study of Cocaine Dependence (FSCD) study and the Collaborative Genetic Study of Nicotine Dependence (COGEND) study datasets in which GABRA2 was also evaluated independently. In this study five SNPs at GABRA2 were nominally associated (uncorrected P < 0.05), with odds ratios between 1.11 and 1.16 (Bierut et al., 2010). Another GWAS of probands drawn from the COGA study reported association with alcohol dependence and a cluster of genes on chromosome 11. This study also implicated rs35164 (just downstream of CDH11) in early onset alcoholics and found evidence to support associations of alcohol dependence with carboxypeptidase E (CPE), deoxyribonuclease II beta (DNASE2B), solute carrier family 10 (sodium/bile acid cotransporter family), member 2 (SLC10A2), ADP-ribosylation-like factor 6 interacting protein 5 (ARL6IP5), inhibitor of DNA binding 4, dominant negative helix-loop-helix protein (ID4), GATA binding protein 4 (GATA4), spectrin repeat containing, nuclear envelope 1 (SYNE1), and adenylate cyclase 3 (ADCY3) (Edenberg et al., 2010). Other genes possibly mediating risk of alcoholism are discussed in Edenberg et al 2010. In a further GWAS analysis of probands from the COGA study using event-related brain oscillations as an endophentype, association between the serotonin receptor 7 (HTR7) and the endophenotype as well as with alochol dependence alone was reported (Zlojutro et al., 2010).

A GWAS from Germany with 487 alcohol dependent males and 1358 controls followed up analysis of 140 SNPs in a further 1024 male patients and 996 age-matched controls. The SNPs in the follow up study were made up of 121 SNPs that showed a nominal significance of P<10⁻⁴ in the original GWAS and 19 SNPs in the human homologues of rat genes whose brain expression was altered after chronic alcohol exposure (Treutlein et al., 2009). In the data from the combined study two intergenic SNPs located on chromosome 2q35 met genome-wide significance (rs7590720, P = 9.72 × 10⁻⁹; rs1344694, P = 1.69 × 10⁻⁸) in a region which has been implicated in linkage studies for alcohol phenotypes. Nine further SNPs were located in genes, including CDH13 and ADH1C, which have been reported to be associated with alcohol dependence.

An Australian and Dutch pooling-based GWAS of nicotine dependence (ND; 1273 cases and 1113 controls) and alcohol dependence (AD; 1224 cases and 1162 controls) revealed three SNPs for comorbid AD/ND (rs7530302 near MARK1 on chromosome 1 (P = 1.90 × 10⁻⁹), rs1784300 near DDX6 on chromosome 11 (P = 2.60 × 10⁻⁹) and rs12882384 in KIAA1409 on chromosome 14 (P = 4.86 × 10⁻⁸). None of the SNPs achieved genome wide significance in a Australian/Dutch meta-analysis in the same paper, but a gene network diagram based on the top results revealed in this study overrepresentation of genes coding for ion-channels and cell adhesion molecules {Lind, 2010 #91}.

Additionally a genome-wide association of pooled DNA from alcoholics was compared to quantitative trait loci implicated in mouse addiction phenotypes. This convergent strategy identified brain-expressed susceptibility genes involved in cell adhesion, enzyme activity, protein and transcriptional processes, neuroreceptors, ion channel and transport processes and cell structure. The set of proteins implicated was described as proving a “connectivity constellation” for susceptibility to addiction (Liu et al., 2006, Uhl et al., 2008a, Liu et al., 2005).

Family, adoption, twin, linkage and allelic association studies have contributed to the evidence for genetic susceptibility to bipolar affective disorder with markers at several genes showing replicated association with bipolar disorder (Baum et al., 2008, Craddock et al., 2008, Moskvina et al., 2009, Ferreira et al., 2008, Askland, 2006, Askland et al., 2009). Recently, Johnson and colleagues found several overlapping clusters of single nucleotide polymorphisms (SNPs) when comparing the genome-wide findings of four studies on bipolar disorder and one on substance dependence, including alcoholism and illicit drug use (Johnson et al., 2009). There have been no previous genome-wide association studies of bipolar alcoholism. This study aims to identify genetic regions associated with bipolar disorder that are comorbid with alcoholism and also to distinguish between genes involved in alcoholism and those involved in bipolar affective disorder.

Materials and methods

Results

The main aim of our study was to detect those SNPs and genes previously associated with alcoholism that could be replicated in our subgroup of alcoholic bipolar research subjects. Demographic breakdown of the groups was as follows: i) BPALC male 80 (56%), female 63 (44%). ii) NABPD, male 113 (31%), female 250 (69%). iii) Controls male 217 (43%), female 293 (57%).

There were 3799 SNPs nominally associated with alcoholic bipolar subjects with significance p <0.01. 2136 of these SNPs were within 50kb of a gene and 1491 were within the gene. The nominally significant associations are reported in Table 1 and Supplementary Table 1. Twelve SNPs attained p-values of <1.0 × 10⁻⁴ but no markers attained conventional genome-wide significance levels. Table 1 lists the tests of association with SNPs associated with bipolar alcoholism and in the non alcoholic bipolar group (NABPD) and this enables us to differentiate genetic effects on bipolar alcoholism from genetic effects on pure bipolar disorder (ie NABPD). Supplementary Table 1 reports the full genome wide findings, arranged by BPALC p-value, where SNPs had p<0.01.

As described in the methods section, in order to confirm the involvement of genes previously implicated in alcoholism and addictions phenotypes, we employed gene wise permutation tests on all SNPs within and next to a previously implicated gene and corrected for multiple tests . The results from these tests are shown in Table 2. Replications with gene wise significance at p<0.05 for association with the bipolar alcoholic group included the cadherin 11 (CDH11), collagen type XI alpha 2 (COL11A2), neuromedin U receptor 2 (NMUR2), exportin7 (XP07) and semaphorin associated protein 5A (SEMA5A) genes. Only the gene wise association with CDH11 remained significant after Bonferroni correction for the genes tested (n=38). Also shown in Table 2 are the tests of association with SNPs in the alcoholism associated genes in the bipolar research subjects after the alcoholic bipolar subjects have been removed, denoted once again as the NABPD group.

Lastly, genes which were implicated by SNP associations and replication, but which did not achieve gene-wise significance in our sample are also detailed in Table 2. The most replicated alcoholism genes ALDH, ADH and GABRA2 were not implicated at SNP or gene wise level in the BPALC phenotype.

We also note that the remaining SNPs which we found were nominally associated with BPALC and which confirmed previous reported gene associations with addiction phenotypes in the “connectivity constellation” failed to show gene-wise significance in our study (data not shown). These connectivity genes were: the neural cell adhesion molecule (NRCAM) gene, brain-specific angiogenesis inhibitor 3 (BAI3), contactin associated protein-like 2 (CNTNAP2), CUB and Sushi multiple domains 1 (CSMD1), catenin (cadherin-associated protein), alpha 2 and 3 (CTNNA2 & 3), Down syndrome cell adhesion molecule (DSCAM), protein tyrosine phosphatase, receptor type, D (PTPRD), sarcoglycan zeta (SGCZ), brain-specific angiogenesis inhibitor 3 (BAI3), astrotactin 2 (ASTN2), ankyrin repeat and sterile alpha motif domain containing 1B (ANKS1B), contactin 4 and 6 (CNTN4 & 6) (Uhl et al., 2008b). The SNPs associated with these genes are listed in Supplementary Table 1.

The top-ranked (by genome wide significance values) alcoholic bipolar SNP associations and the genes they represent are listed in Table 1 and continued in Supplementary Table 1. Selected findings are noted below. The most significant SNP (p=4.54×10⁻⁶) was rs8062326, a single SNP placed midway between the synaptotagmin 17 (SYT17) and Inositol 1,4,5-triphosphate receptor-interacting protein-like 2 (ITPRIPL2) genes. One of two SNPs intronic to the collagen, type I, alpha 2 (COL1A2, highest ranked SNP rs369982, p = 2.06 × 10⁻⁵) gene was ranked second most associated and when all SNPs in this gene were analysed the gene-wise test was significant at p = 5 × 10⁻⁴. Ranked 29th, the first of forty five SNPs nominally associated with BPALC were located in and around cadherin 11 (CDH11, highest ranked SNP rs429065, p = 1.0 × 10⁻⁴, gene-wise p = 6 × 10⁻⁴). rs429065 is located 22 Kb downstream of CDH11 (Figure 1). Eleven intronic SNPs at CDH11 were found to be associated with BPALC of which two were also nominally associated with NABPD. Two of the associated SNPs from a genome-wide study of pooled DNA from the COGA sample (Johnson et al., 2006) were also nominally associated in our sample. These were intronic SNP rs35200 (p = 1.9 × 10⁻²) and rs35164 (p = 1.4 × 10⁻⁴) which is 5,484 base pairs downstream from CDH11. rs25164 was also implicated with early age of onset alcoholism in the GWAS by Edenberg and colleagues (Edenberg et al., 2010). Excluding the CDH11 SNPs replicated above, no strongly associated SNPs from previous GWAS replicated in this study. Eighteen SNPs within the introns of cadherin 12 (CDH12, Table 3) and 10 of 14 SNPs genotyped within the introns of cadherin 13 (CDH13, Table 2) were nominally associated with BPALC. In total eighty four SNPs in or near cadherin-family genes were nominally associated with BPALC.

Table 1 and Supplementary Table 1 also report the top-ranked BPALC-associated SNPs which were also nominally significant in NABPD genetic associations. Most significant and ranked 2^nd genome wide was a SNP intronic to musashi homolog 2 (MSI2, highest ranked SNP rs1024820, p = 1.24 × 10⁻⁵). However these SNPs were not in LD with the MSI2 gene, nor was the gene associated with bipolar alcoholism using gene-wise analysis. There were six BPALC associated SNPs in or within 25 Kb of tetraspanin 8 (TSPAN8, highest ranked SNP rs1705236, p = 1.24 × 10-4; gene wise p non significant) three of which were also associated with NABPD.

We found six nominally BPALC-associated SNPs near or within the GABA-A beta-3 receptor (GABRB3, highest ranked SNP p = 2.6 × 10⁻⁴, uncorrected gene-wise significance p = 2 × 10⁻³). The tachykinin receptor 1 (TACR1, also known as neurokinin-1 receptor, NK1R) was nominally associated with BPALC with three SNPs, one of which, rs3771829 (p = 3 × 10⁻³) was also associated with NABPD (p = 1.7 × 10⁻²). There were eight SNPs nominally associated with BPALC near the gene for insulin-like growth factor 1 (IGF1, highest ranked SNP: rs12426318, 154 kb downstream, p = 1.8 × 10⁻³). Neither TACR1 nor IGF1 survived gene wise testing for either phenotype.

Lastly we performed gene wise testing on the remaining genes implicated by clustered (7 or more) nominal SNP associations with the BPALC phenotype, but which had no prior literature association with alcoholism or related phenotypes. These represent novel BPALC candidate genes (Table 3). Three genes survived Bonferroni correction: KIAA1772, DC2 protein, and zinc finger and BTB domain containing 16 (ZBTB16).

Discussion

This genome-wide study comparing individuals with bipolar disorder with and without alcoholism diagnoses has replicated five previous associations with alcoholism and addiction phenotypes by gene wise analysis that appear to be independent of bipolar disorder. Our most significant SNP associations with BPALC were in or near genes involved in cell adhesion, neurotransmitter pathways, enzymatic activity, cellular messengers, connective tissue, and cell regulation but none of these reached genome wide significance (p<10⁻⁸) and can therefore the associations can only be described as suggestive. The presence of association between these SNPs and the BPALC cases but not in the NABPD cases suggests genetic effects on alcoholism independent of bipolar affective disorder.

First we discuss SNP associations then gene wise associations. Although we found a SNP strongly associated with BPALC at Synaptotagmin 17 (SYT17, ranked 1^st genome-wide) we could not be certain that it was actually this gene that was implicated because it was equidistant between SYT17 and Inositol 1,4,5-triphosphate receptor-interacting protein-like 2 (ITPRIPL2). SYT17 is expressed abundantly in the frontal and temporal lobes, hippocampus, hypothalamus, amygdala, substantia nigra, and pituitary. Synaptotagmins are considered to be important in the docking and fusion of synaptic vesicles and plasma membrane in neurotransmitter release and are thus worthy candidate genes (Fukuda et al., 1994). ITPRIPL2 is also expressed in the brain but little is known about its function.

The simultaneous presence of nominal SNP association between the markers in both the BPALC and NABPD groups suggests putative genetic effects on alcoholism through an affective disorder mechanism. One such gene, MSI2 (the gene nearest the highest-ranked SNP in the NABPD phenotype) encodes a brain-expressed protein which may play a role in post transcriptional gene regulation. Another such gene, tetraspanin 8 (TSPAN8) is independently associated with bipolar disorder, and encodes a protein which mediates signal transduction events that play a role in the regulation of cell development, activation, growth and motility. Our most significant TSPAN8 SNP was rs1705236 and this was found to be the most well associated SNP in a recent large whole genome association study of bipolar disorder, which notably included the UCL sample used in this study (Sklar et al., 2008). TSPAN8 was nominally associated with NABPD at uncorrected gene wise level (p=3.9 × 10⁻³) but not for the BPALC phenotype. This further reinforces the role of TSPAN8 as a bipolar disorder candidate gene.

Genes implicated by gene-wise significance in the present study include those that are functionally involved in second-messenger systems, ion and neurotransmitter function, neuronal adhesion, differentiation and architecture. These are considered part of the “connectivity constellation” of genes (Uhl et al., 2008b). We applied a Bonferroni correction to the gene wise association test presented in table 2 in an attempt to correct for the multiple tests performed. And indeed it is encouraging to see that the result for CDH11 survives this correction. However, this correction is almost certainly overly conservative given the prior evidence for association at each of the loci tested in table 2. The cadherin (CDH) gene family are a useful example of connectivity genes. These genes encode a large group of transmembrane proteins that mediate calcium-dependent cell-cell adhesion and the generation of synaptic complexity in the developing brain {Benson, 1998 #93}. Cadherins have been implicated in mnemonic processes, addictions and bipolar disorder (Johnson et al., 2009, Uhl et al., 2008b). We discuss the cadherins of relevance to the present paper here.

CDH11 (Tables 1, 2) was implicated in a genome-wide study of pooled COGA DNA (Johnson et al., 2006), a recent GWA of alcoholism from COGA families (Edenberg et al., 2010),and was implicated in our study with positive corrected gene-wise significance tests and high genome-wide ranking. Twenty two CDH11 SNPs in the UCL sample were nominally implicated in BPALC and seven were associated with NABPD in the UCL sample. Despite evidence that our strongest CDH11 SNP, rs429065 may be nominally associated with bipolar disorder in the Wellcome Trust Case Control Consortium (p=1.2 × 10-2) and STEP-BD samples (p = 2.7 × 10⁻² (WTCCC, 2007; Sklar et al., 2008). CDH11 was however not associated with NABPD at the level of the gene

CDH12 (gene wise p = 3.5 × 10⁻³) is found near a linkage peak for alcohol dependence symptom count (Hansell et al., 2010), was found over-expressed in both post-mortem brain studies of individuals with a history of alcohol misuse or dependence (Sokolov et al., 2003) and violent suicide victims (Thalmeier et al., 2008), and is now included on a custom SNP array chip for genotyping of candidate addiction genes (Hodgkinson et al., 2008).

CDH13 was not implicated in the present study but was recently associated with bipolar disorder and addictions in a study of GWA data for both disorders (Johnson et al., 2009) and in a GWA of alcohol dependence (Treutlein et al., 2009). CDH11 is implicated, and CDH12 tentatively implicated, in alcoholism in bipolar subjects in the present study. Thus the cadherin family remain robust candidate genes through replication and offering plausible pathways for alterations in synaptic neuroplasticity in individuals with alcohol dependence (Sokolov et al., 2003).

In the present study the remaining connectivity genes implicated by gene wise testing include: i) a member of the semaphorin protein family, Semaphorin associated protein 5A (SEMA5A). This gene is involved in axonal guidance during neural development (Adams et al., 1996). SEMA5A was previously associated with both bipolar disorder and substance dependence (Johnson et al., 2009) and successful smoking cessation (Uhl et al., 2008c). ii) neuromedin U receptor 2 (NMUR2) is a brain expressed protein found in the hypothalamus which was implicated in the pooled COGA genome-wide association data (Johnson et al., 2006). iii) Collagen type XI alpha 2 (COL11A2) has been implicated in sensorineural deafness but has not been reported as being associated with addictions or affective disorder phenotypes. iv) Exportin 7 (XPO7) mediates the nuclear export of proteins with broad substrate specificity and was implicated in the pooled COGA genome-wide association data (Johnson et al., 2006).

Three genes KIAA1772, oligosaccharyltransferase complex subunit (OSTC), and zinc finger and BTB domain containing 16 (ZBTB16) with novel gene wide association results remain unconfirmed candidate genes implicated by multiple SNP clustering and corrected significance values and as such they are worthy of further investigation.

Important negative findings include the genes of the Gamma-aminobutyric acid (GABA) system which encode receptors for the major inhibitory transmitter of the nervous system. GABA receptors are of strong interest in addictions, alcoholism, and affective disorders (Agrawal et al., 2008, Bierut et al., 2010). GABA receptor alpha 2 (GABRA2) is the most replicated gene in alcoholism (Bierut et al., 2010). GABRB3 polymorphisms while less robustly implicated in alcoholism than GABRA2 have been associated with mood-related alcohol expectancy (Young et al., 2004) and are more common in the children of alcoholics than children of non alcoholics (Namkoong et al., 2008). In the present study GABRB3 showed gene wise association with BPALC but did not survive Bonferroni correction. None of the GABA receptor genes including GABRA2 were implicated at corrected gene wise level in the present study. This may reflect the small study size, genetic heterogeneity, or possibly that alcoholism in bipolar disorder is mediated by different pathways.

The relationship between bipolar and unipolar affective disorder and alcohol dependence is bidirectional (Sonne and Brady, 2002) and there is little room for doubt that the two disorders have reciprocal effects on each other. Alcohol dependence may also change expression and function of the same neurotransmitters as those involved in bipolar disorder, thereby “prompting” the symptoms of unipolar and bipolar affective disorder (Tohen et al., 1998).

The present study has benefitted from a well-characterized, ancestrally homogenous and ancestrally-matched control sample. The main drawback was the small sample size of bipolar alcoholics for a genome-wide study.

Although low significance values are needed for convincing GWAS results the fact that genes previously implicated in the genetics of alcoholism were found by us also to be associated with bipolar-alcoholism after gene wise association tests suggests that our results are robust. The “connectivity cluster” genes are strongly represented in this study. We also report tentative new alcoholism gene associations in a bipolar-alcoholic sample which are in need of further investigation in additional alcoholic bipolar and alcoholism samples.

Supplementary Material