1 Introduction
Although gambling is common to most cultures, only a proportion of individuals develop pathological gambling (PG) resulting in negative psychological, social, and economic consequences for the affected individuals and their social network [Reference Lupi, Martinotti, Acciavatti, Pettorruso, Brunetti and Santacroce1]. Among adults, reported prevalence rates for PG range between 0.02 and 2.0% [Reference Sassen, Kraus and Bühringer2]. In the German population, the prevalence of PG has been estimated to be 0.3% [Reference Meyer, Bischof, Westram, Jeske, de Brito and Glorius3]. Identifying the biological causes of PG may facilitate prevention and treatment.
The precise diagnostic classification of PG is still evolving. In previous decades, PG was classified as an impulse control disorder. However, accumulating research evidence suggests that PG resembles substance-related addictions in many domains [Reference Grant, Potenza, Weinstein and Gorelick4]. As a result, PG (renamed as gambling disorder) is now classified in the 5th edition of the American Psychiatric Association's Diagnostic and Statistical Manual of Mental Disorders as a non-substance-related behavioural addiction [5].
PG is more prevalent in males compared to females [Reference Fattore, Melis, Fadda and Fratta6]. Risk factors are low socioeconomic status [Reference Barnes, Welte, Tidwell M-CO and Hoffman7], immigration [Reference Kastirke, Rumpf, John, Bischof and Meyer8], and high impulsivity combined with emotional instability [Reference Bagby, Vachon, Bulmash, Toneatto, Quilty and Costa9]. PG shows comorbidity with other mental health disorders, including mood and anxiety disorders, personality disorders, alcohol dependence (AD), and substance use [10–Reference Bischof, Meyer, Bischof, Kastirke, John and Rumpf12]. The reasons for this comorbidity are unclear. Twin studies of PG have generated insights into comorbidity with various alcohol use disorders [13–Reference Slutske, Ellingson, Richmond-Rakerd, Zhu and Martin15]. However, the focus of these studies was not to investigate patients with PG but rather to determine correlations between the full continuum of gambling related problems and alcohol use disorders. These studies were performed in the general population [Reference Blanco, Myers and Kendler13, Reference Slutske, Ellingson, Richmond-Rakerd, Zhu and Martin15], and in a male sample from the United States Vietnam Era Twin Registry [Reference Slutske, Eisen, True, Lyons, Goldberg and Tsuang14]. Disordered gambling (DG) was defined as the presence of one or more gambling related problems, including PG. Genetic correlation between DG and alcohol use disorders ranged between 30 and 45%, indicating the strength of the correlation between the genetic liabilities [13–Reference Slutske, Ellingson, Richmond-Rakerd, Zhu and Martin15].
Twin studies have also investigated the genetic mechanisms underlying PG [14,16–Reference Lind, Zhu, Montgomery, Madden, Heath and Martin20]. Research has shown that genes are of equal importance in the aetiology of DG in men and women [Reference Slutske, Zhu, Meier and Martin16, Reference Slutske, Meier, Zhu, Statham, Blaszczynski and Martin17]. Their findings also suggest that genetic factors contribute around 50% to the risk for DG [Reference Slutske, Meier, Zhu, Statham, Blaszczynski and Martin17, Reference Slutske and Richmond-Rakerd19] and that the same genes are involved in the spectrum from less severe DG to PG [Reference Slutske, Eisen, True, Lyons, Goldberg and Tsuang14, Reference Slutske, Meier, Zhu, Statham, Blaszczynski and Martin17]. Although samples sizes were large (up to 6744 individuals [Reference Slutske, Eisen, True, Lyons, Goldberg and Tsuang14]), only 104 (< 2.2% [Reference Eisen, Slutske, Lyons, Lassman, Xian and Toomey18]) or fewer individuals fulfilled the DSM criteria for PG, which reflects the estimated prevalence of PG in the general population. There have been no genome-wide association analyses (GWAS) of PG per se; to date, only a GWAS of a quantitative disordered gambling trait has been reported [Reference Lind, Zhu, Montgomery, Madden, Heath and Martin20].
So far, most molecular genetic studies of PG have applied candidate gene approaches, primarily reporting an involvement of the dopaminergic and serotonergic neurochemical systems [21–Reference Slutske26]. However, multiple neurotransmitter systems have been implicated in PG [27–Reference Fauth-Bühler, Mann and Potenza29].
To our knowledge, the present study is the first case-control GWAS of PG, with all cases being assigned a diagnosis of PG. Due to the small sample size (445 cases and 986 controls), genome-wide significant single marker findings would only be expected for strong effects, such as those reported for the aldehyde dehydrogenase 2 in alcohol consumption [Reference Takeuchi, Isono, Nabika, Katsuya, Sugiyama and Yamaguchi30]. However, pathway analyses and polygenic risk score tests were used to investigate whether the analysis of pathways may uncover hidden effects, and whether polygenic risk score tests may provide information concerning a common genetic background for AD and PG.
Besides limited sample sizes, a further challenge in this research field is the fact that the PG phenotype can be influenced by many factors. Potential confounders include sex, age, and socioeconomic status. As it is not possible to control for all of these possible confounders, in a first step, we controlled for principal components (PC) only. In a second step, the analyses were also controlled for age and sex.
2 Subjects and methods
2.1 Participants
The study was approved by the respective local ethics committees, and all subjects provided written informed consent prior to inclusion.
The sample comprised 445 cases and 986 controls. The PG cases were recruited from inpatient and outpatient treatment centres following presentation for treatment of PG (“Baden-Württemberg Study: assessing psychological, neurobiological, and genetic mechanisms of pathological gambling in Baden-Württemberg”, a federal state of Germany). A smaller part of the sample was recruited via a nationwide telephone survey (‘Pathologisches Glücksspielen und Epidemiologie’ (PAGE), an epidemiological study for pathological gambling in Germany).
All patients were diagnosed according to DSM-III or -IV diagnostic criteria for PG. These criteria were assessed using two different tools. In the Baden-Württemberg study, these criteria were assessed using the South Oaks Gambling Screen (SOGS; cut-off ≥ 5 for PG [Reference Lesieur and Blume31]) based on DSM-III diagnostic criteria. In the Page study, the Composite Diagnostic Interview (CIDI; Version 3.0 of the World Health Organisation) was used to assess the criteria for DSM-IV diagnosis. Comorbid AD was assessed using the German version of the Structural Clinical Interview for DSM-IV (SCID-I; [Reference Wittchen, Zaudig and Fydrich32]), and diagnoses were assigned according to DSM-IV criteria.
For controls, genome-wide genotype data for population-based individuals were obtained from the SHIP-TREND study, which is a longitudinal population-based investigation of individuals from West Pomerania, Germany [Reference Volzke, Alte, Schmidt, Radke, Lorbeer and Friedrich33]. This study investigates the prevalence, incidence, and complex associations of common risk factors, subclinical disorders, and clinical diseases. The sample was randomly stratified for age, sex, and county of residence.
2.2 Genotyping
DNA was extracted from whole blood using chemagic MSMI (PerkinElmer Chemagen Technologie GmbH; Rodgau, Germany) or saliva, collected with Oragene Self-Collection Kit OG-500, and extracted with Oragene prepIT, L2P (DNA Genotek Inc, Ontario, Canada). For all controls, whole blood, extracted with Gentra Puregene Blood Kit (Qiagen; Hilden, Germany) was used. For cases, 147 saliva and 298 blood samples were used. Cases and controls were genotyped using Illumina's HumanOmniExpress (n = 730,525 markers), and HumanOmni2.5 BeadChips (n = 2,450,000 markers), respectively.
2.3 Quality control
Stringent quality control filtering criteria were applied. Detailed information on these criteria is provided in the supplementary text. In brief, single nucleotide polymorphisms (SNPs) with the following characteristics were removed: call rate < 0.98; minor allele frequency < 0.01 in cases or controls; or deviation from Hardy–Weinberg equilibrium of < 10−6 (cases) or < 10−4 (controls). Individuals with the following characteristics were removed: call rates < 0.97; duplicated or cryptic related samples; or outlier status. A consensus SNP set common to both Illumina genotyping platforms (n = 595,867) was used for further analysis.
2.4 Single marker analysis
A logistic regression approach was used for the single marker association tests for autosomal SNPs. PG was used as binary trait. Correction for population stratification was performed using the first five PCs from a principal component analysis across independent autosomal markers (see supplementary text). An additional analysis also included age and sex as covariates.
2.5 Gene-based analysis
A gene-based test was performed using the Versatile Gene-based Association Study programme 2 [Reference Mishra and Macgregor34]. A P-value below α = 2.1 × 10−6 (0.05/23,804) was considered to be significant, as the gene-based test included 23,804 autosomal genes.
2.6 Pathway/gene-set based analyses
The global test was used to determine whether groups of genes were significantly related to the outcome of interest [[Reference Deelen, Uh, Monajemi, van Heemst, Thijssen and Böhringer35]. This was applied to the dataset using three pathway and gene-set databases: the Kyoto Encyclopaedia of Genes and Genomes (KEGG; [Reference Kanehisa and Goto36, 37]); Reactome [Reference Croft, Mundo, Haw, Milacic, Weiser and Wu38, 39]; and Gene ontology (GO; [40]). Details of the procedures used for obtaining gene-sets and mapping SNPs to genes and corresponding pathways, and the methods used to account for possible bias, are described in the supplementary text.
2.7 Polygenic risk scores for DG and AD
Polygenic risk scores were calculated to summarise the genetic effects of markers for DG and AD. These polygenic risk scores were calculated using the method introduced by Purcell et al. [Reference International Schizophrenia Consortium, Purcell, Wray, Stone, Visscher and O’Donovan41]. Marker weights for AD and DG were based on association results obtained in a German GWAS of AD [Reference Frank, Cichon, Treutlein, Ridinger, Mattheisen and Hoffmann42], and summary data of a quantitative gambling trait in an Australian sample reported by Lind et al. [Reference Lind, Zhu, Montgomery, Madden, Heath and Martin20]. A detailed description of the calculation is provided in the supplementary text.
3 Results
Most of the pathological gamblers (n = 338) were recruited from inpatient and outpatient treatment centres within the context of the “Baden-Württemberg Study”. The remaining 107 cases were drawn from the PAGE sample. Following quality control, the sample comprised 1362 individuals: 396 cases and 966 controls. The 396 PG cases (n males = 358, n females = 38) included 280 inpatients, 83 outpatients, and 33 currently non-treated cases. The mean age of these cases was 40.18 years (range 16–75, standard deviation (SD) = 11.00).
For controls, genome-wide genotype data for 966 population-based individuals (n males = 427, n females = 539) were obtained from the SHIP-TREND study. The mean age of the controls was 50.16 years (range 20–81, SD 13.65).
Of the PG cases, 149 had a comorbid diagnosis of AD, and 222 were not alcohol-dependent. For the remaining cases, no information on AD status was available.
A total of 595,867 autosomal SNPs were available for analysis.
The Manhattan plots of the GWAS for the two analyses are shown in Figs. 1 and 2. These analyses use “PC 1 to 5”, and “PC 1 to 5, age and sex” as covariates, respectively. The corresponding Quantile–Quantile plots of the observed vs expected –log 10 P-values of the association analysis are shown in Figs. 3 and 4.
Fig. 1 Manhattan plot of the association P-values for pathological gamblers vs. controls. The horizontal axis represents the genome, which is divided into its autosomes; the vertical axis shows the –log 10 values of the association P-values. The red line shows the genome-wide significance threshold. The blue line shows the threshold for “suggestive” associations. The figure shows results for the analysis including PC 1 to 5.
Fig. 2 Manhattan plot of the association P-values for pathological gamblers vs. controls. The horizontal axis represents the genome, which is divided into its autosomes; the vertical axis shows the –log 10 values of the association P-values. The red line shows the genome-wide significance threshold. The blue line shows the threshold for “suggestive” associations. The figure shows results for the analysis including PC 1 to 5, sex and age.
Fig. 3 Quantile–Quantile plot of association for pathological gamblers vs. controls. The horizontal axis represents the –log 10 of expected association test P-values, and the vertical axis shows the –log 10 of P-values from the P-values. The shaded region shows the 95% confidence bands of expected values under the null hypothesis of no association. Fig. 1 shows results for the analysis that included PC 1 to 5 as covariates. The Lambda was 1 for this analysis.
Fig. 4 Quantile–Quantile plot of association for pathological gamblers vs. controls. The horizontal axis represents the –log 10 of expected association test P-values, and the vertical axis shows the –log 10 of P-values from the P-values. The shaded region shows the 95% confidence bands of expected values under the null hypothesis of no association. The figure shows results for the analysis that included PC 1 to 5, sex and age as covariates. The Lambda was 1.0046 for this analysis.
3.1 Single marker analysis
No SNP achieved genome-wide significance. The top SNPs (16), with P-values of < 5 × 10−5 in the analysis that included PC 1 to 5 only, are listed in Table 1a. The first three top SNPs (P-values of < 10−5) are located on 16p12.3 and 20q13.12. The SNP rs6065904 (P-value = 1.48 × 10−6, odds ratio (OR): 0.53; confidence interval (CI) = [0.41, 0.69]) is located in an intron of phospholipid transfer protein (PLTP). The SNP rs4810479 (P-value = 4.67 × 10−6, OR: 0.57; CI = [0.44, 0.72]) is located nearby and in strong linkage disequilibrium with rs6065904 (r 2 = 0.79) in the upstream region of PLTP. The SNP rs3943418 (P-value = 6.61 × 10−6, OR: 1.71; CI = [1.36, 2.16]) is located in an intron of Xylosyltransferase 1 (XYLT1). All three top SNPs had P-values in the range of 4.6 to 6.9 × 10−5 following the inclusion of age and sex. Top hits corrected for sex and age with P-values of < 5 × 10−5 are shown in Table 1b. Here, the top two SNPs < 10−5 were: (i) rs7591351 (P-value = 5.88 × 10−6, OR: 1.67; CI = [1.34, 2.09], only PC 1 to 5 corrected, 2.94 × 10−4); and (ii) rs6738409 (P-value = 7.39 × 10−6, OR: 0.60; CI = [0.48, 0.75], PC 1 to 5 corrected 1.44 × 10−4). Both SNPs are located in the protein kinase C gene (PRKCE). All further top SNPs < 10−4 are listed in the supplementary tables S1a and b. Please also find a comparison with the SNP top hits of Lind et al. [Reference Lind, Zhu, Montgomery, Madden, Heath and Martin20] in supplementary text.
Table 1a Top SNP findings (P-value < 5 × 10−5) for the logistic regression including PC 1 to 5 as covariates.
CHR: chromosome; OR: odds ratio; L95 and U95: lower and upper 95% confidence intervals; SNP: single nucleotide polymorphism.
Table 1b Top SNP findings (P-value < 5 × 10−5) of the logistic regression including PC 1 to 5, age and sex as covariates.
CHR: chromosome; OR: odds ratio; L95 and u95: lower and upper 95% confidence intervals; SNP: single nucleotide polymorphism.
3.2 Gene-based analysis
In the gene-based analysis, no nominal significant finding survived correction for multiple testing (P-value < 2.1 × 10−6). The top hits (< 10−3) are listed in Tables 2a and 2b. The lowest empirical P-value for the first approach was 3.8 × 10−5 for PCIF1, which encodes PDX1 C-terminal inhibiting factor 1. Nine genes out of 19 shared one signal: for these nine genes, the top SNP was rs6065904. The top hit for the age and sex corrected version was RBM33, with a P-value of 7.6 × 10−5. Here, PCIF1 retained a P-value of 4.88 × 10−4. Descriptions of genes as well as a comparison with results of previous studies are listed in supplementary text and supplementary table S2a and b.
Table 2a Gene-wide associations with P-value < × 10−3 of the gene-based analysis including PC 1 to 5 as covariates.
* *, **, #, ## same SNP drives finding for different genes.
Table 2b Gene-wide associations with P-value < × 10−3 of the gene-based analysis including PC 1 to 5, sex, and age as covariates.
* *,**, ***, #, ## same SNP drives finding for different genes.
3.3 Pathway/gene-set based analyses
In KEGG, 35 out of 257 pathways achieved nominal significance (P-value < 0.05). Of these, 13 had P-values of < 0.01, including three pathways with a corrected P-value of < 0.05 in the analysis that controlled for PC 1 to 5. The SNP- and case-control permutation tests suggested that these three best pathways are reliable, as all achieved P-values of < 0.003. These pathways (hsa05016, Huntington's disease; hsa04152, 5′-adenosine monophosphate-activated protein kinase [AMPK] signalling pathway; and hsa04210, Apoptosis) remained significant after Benjamini-Hochberg correction for the 257 pathways of the KEGG database (see Table 3a). In the sex and age corrected analysis (see Table 3b), no pathway remained significant after correction: 43 had P-values of < 0.05, and 12 had P-values of < 0.01. The top-ranking pathway was AMPK signalling, with a P-value of 5.36 × 10−4. All three previously significant pathways from the first analysis (PC 1 to 5) remained among the top hits P < 0.01 hits. All KEGG pathways with nominally significant P-values of < 0.01 and detailed results are shown in supplementary tables S3a and b as well as in the supplementary text. Tables S3c and d indicate the proportions of genes with overlap between the top pathways. Table S3b shows additional interesting pathways with P-values of < 0.05.
Table 3a KEGG global test results with P-values < 0.01 of the analysis including PC 1 to 5 as covariates.
Values remaining significant after correction are shown in bold.
* Benjamini-Hochberg corrected.
Table 3b KEGG global test results with P-values < 0.01 of the analysis including PC 1 to 5, sex, and age as covariates.
* Previously significant pathways (in the first analysis) shown in bold.
In Reactome and GO, no gene-sets remained significant (i.e. P-value < 0.05) after Benjamini-Hochberg correction for all gene-sets in the databases. Of 1180 pathways, a total of 23 in the analysis correcting for PC 1 to 5, and 18 in the analysis correcting for age and sex, had P-values of < 0.01 (uncorrected) in Reactome. These are listed in supplementary table S4a and b. Of 8474 pathways in GO, a total of 32 (PC 1 to 5 corrected) and 14 (age and sex corrected analysis) had a P-value of < 10−3. These pathways are described in supplementary tables S5a and b as well as in the supplementary text.
3.4 Polygenic risk scores for DG and AD
The association between risk score for AD and PG status was nominally significant (P-value = 0.047, one-sided test, see Table 4a). This value improved after the inclusion of sex and age, yielding a P-value of 0.024 for a one-sided test, see Table 4b. No association was found between risk scores for DG and PG status in either approach (see Tables 5a and 5b). Quartile plots of polygenic risk scores are shown in supplementary figs. S1 and S2.
Table 4a Polygenic risk score results of the AD GWAS by Frank et al. [Reference Frank, Cichon, Treutlein, Ridinger, Mattheisen and Hoffmann42] as applied to the PG data including covariates PC 1 to 5.
AD: alcohol dependence; GWAS: genome-wide association study; PG: pathological gambling.
Table 4b Polygenic risk score results of the AD GWAS by Frank et al. [Reference Frank, Cichon, Treutlein, Ridinger, Mattheisen and Hoffmann42] as applied to the PG data including PC 1 to 5, age and sex as covariates.
AD: alcohol dependence; GWAS: genome-wide association study; PG: pathological gambling.
Table 5a Polygenic risk score results of the GWAS of DG by Lind et al. [Reference Lind, Zhu, Montgomery, Madden, Heath and Martin20] as applied to the PG data including covariates PC 1 to 5.
GWAS: genome-wide association study; PG: pathological gambling; DG: disordered gambling.
Table 5b Polygenic risk score results of the GWAS of DG by Lind et al. [Reference Lind, Zhu, Montgomery, Madden, Heath and Martin20] as applied to the PG data including PC 1 to 5, age and sex as covariates.
GWAS: genome-wide association study; PG: pathological gambling; DG: disordered gambling.
4 Discussion
To our knowledge, the present study represents the first GWAS of PG. No genome-wide significant association was found with any SNP or gene. This was unsurprising, given the small sample size. However, although well below the significance threshold, the three top SNPs (P-values < 10−5) are of potential interest. The SNPs rs6065904 and rs4810479 are located in close proximity to each other within and near the gene PLTP. The product of PLTP is the phospholipid transfer protein. This protein is involved in the phospholipid metabolism, and has been reported to show significantly higher activity in individuals who abuse alcohol [Reference Liinamaa, Hannuksela, Kesäniemi and Savolainen43] and in individuals who drink heavily [Reference Mäkelä, Jauhiainen, Ala-Korpela, Metso, Lehto and Savolainen44]. Meta-analyses have shown strong association between these two SNPs and various lipid metabolism phenotypes, with reported P-values as low as 4 × 10−40 [Reference Chasman, Paré, Mora, Hopewell, Peloso and Clarke45] and 2 × 10−42 [Reference Inouye, Ripatti, Kettunen, Lyytikäinen, Oksala and Laurila46], respectively. The SNP rs3943418 is located in an intron region of XYLT. This gene encodes xylosyltransferase 1, an enzyme that is involved in the proteoglycan metabolism. A potential role in PG is not apparent.
Unsurprisingly, results changed after correction for age and sex. However, the values for the top three hits changed only slightly. Following correction for age and sex, the two top hits were rs7591351 and rs6738409, P-value < 10−5. Both SNPs are in PRKCE, encoding the protein kinase c epsilon. Research has shown that PRKCE influences ethanol and nicotine self-administration in mice, and is associated with alterations in the cholinergic modulation of dopamine release in the nucleus accumbens [Reference Lee and Messing47].
Formal genetic studies in twins have reported a genetic overlap between PG and AD, and between PG and DG [Reference Slutske, Eisen, True, Lyons, Goldberg and Tsuang14, Reference Slutske, Ellingson, Richmond-Rakerd, Zhu and Martin15]. Here, we attempted to demonstrate these overlaps on a molecular level using polygenic score analyses. While nominal significance was found for an overlap between PG and AD, no significant overlap was observed for PG and DG. This discrepancy could be due to the fact that the training sample for AD comprised patients with severe AD [Reference Frank, Cichon, Treutlein, Ridinger, Mattheisen and Hoffmann42], whereas the training sample for DG, which had been recruited from the general population, included only 31 individuals with a PG diagnosis [Reference Lind, Zhu, Montgomery, Madden, Heath and Martin20]. This sample may thus have lacked sufficient statistical power for the detection of an overlap. Nonetheless, given the very small sample sizes, the present finding for PG and AD appears to provide convincing support for a genetic overlap between these two disorders.
The findings of the pathway analyses were promising. The first analysis yielded significant results for Huntington's disease, the AMPK signalling pathway, and apoptosis.
Huntington's disease (HD) is an autosomal dominant inherited degenerative disorder, and is characterised by progressive motor, cognitive, and behavioural deterioration. Research has shown that patients with HD are at increased risk of addiction if they engage in gambling behaviour [Reference Kalkhoven, Sennef, Peeters and van den Bos48]. However, the reasons for this remain contentious. In a simulated gambling task, decision-making deficits- in the form of an increased choice of disadvantageous decks were observed in both HD [Reference Campbell, Stout and Finn49, Reference Busemeyer and Stout50] and PG [Reference Brevers, Koritzky, Bechara and Noël51]. In PG, poorer performance has been linked to motivational processes [Reference Brevers, Koritzky, Bechara and Noël51]. In HD, Campbell et al. [Reference Campbell, Stout and Finn49] attributed a poorer performance to reduced autonomic responsiveness to gambling task losses. In contrast, Busemeyer et al. [Reference Busemeyer and Stout50] concluded that poorer performance in HD was due in part to cognitive processes, but also to altered choice response mechanisms (resulting from recklessness or impulsiveness).
Even though it is still a matter of debate, it is known that symptoms in Huntington are caused by progressive striatal atrophy. The cortico-striatal circuits that are affected in Huntington's disease are also involved in the predisposition to PG and other addictions comprising disinhibition-related symptoms, such as altered impulsivity, sensitivity to reward, and the inability to consider long-term advantage over short-term reward [Reference Kalkhoven, Sennef, Peeters and van den Bos48].
The second top hit of the pathway analyses, and which ranked first in the analysis that included age and sex, was AMPK signalling. AMPK is a sensor of energy status, and acts both as a key regulator of cellular energy homeostasis [Reference Hardie52], and as a central regulator of both lipid and glucose metabolism [Reference Tomita, Tamiya, Ando, Kitamura, Koizumi and Kato53]. In vivo and in vitro studies have shown that AMPK limits anabolic pathways and activates catabolic reactions [Reference Ceni, Mello and Galli54]. AMPK activation is repressed by glucose withdrawal, and is inhibited by chronic ethanol exposure [Reference Ceni, Mello and Galli54].
The third pathway that remained significant after correction is apoptosis. Apoptosis refers to the controlled and regulated process of cell death, which maintains a healthy balance between cellular death and survival [Reference Hassan, Watari, AbuAlmaaty, Ohba and Sakuragi55]. Apoptotic signals guard genomic integrity and are regulated at several levels [Reference Hassan, Watari, AbuAlmaaty, Ohba and Sakuragi55]. It is not clear in what way these far-reaching processes might play a role in PG. Previous research suggests that cocaine abuse alters processes related to apoptosis [Reference Quintero56].
A limitation of the present study is the small sample size, which provided limited power to detect risk alleles. As a result, we could not control for all potential confounders. However, to account for some important potential confounders, we used two approaches: one including PC 1 to 5 only; and one that also included age and sex.
A further limitation is that patients were heterogeneous in terms of assessment instruments and DSM classification. However, there is evidence for a common etiologic structure between the SOGS and the DSM-IV [Reference Slutske, Zhu, Meier and Martin57]. The most likely effect of a heterogeneous patient population in GWAS is a reduction in power. However, this will lead to missing true effects rather than to false positive findings.
Furthermore, polygenic risk score calculations of AD were based on GWAS data from our German GWAS of AD only in order to increase homogeneity and comparability between the samples. Although a larger sample might have yielded more power, this would have been at the cost of greater heterogeneity. Another limitation of the present study was that the results are limited to a population of Caucasian ethnicity. Further studies are required to determine whether, and how, the identified pathways – or the genes that contribute to them – are involved in the aetiology of PG.
In summary, this first GWAS of PG identified pathways and points to genes with possible involvement in the aetiology of PG. The results are consistent with previous formal genetic studies, which showed an overlap between PG and AD on a molecular genetic level. A number of the higher ranked markers, genes, and pathways appear plausible candidates for the PG phenotype and warrant further investigation. Compared to recent GWAS of schizophrenia and other psychiatric disorders, the power of the present sample was low. However, the results are promising, and warrant a future collaborative research effort to uncover the genetic variants that predispose to PG.
Disclosure of interest
The authors declare that they have no competing interest.
Acknowledgements
The study was supported financially by the German Ministry for Work and Social Affairs, Families, Women and Senior Citizens (Ministerium für Arbeit und Sozialordnung, Familien, Frauen und Senioren), Baden-Württemberg, Germany (reference number: 53-5072-7.1).
SHIP is part of the Community Medicine Research net of the University of Greifswald, Germany, which is funded by the Federal Ministry of Education and Research (grants no. 01ZZ9603, 01ZZ0103, and 01ZZ0403), the Ministry of Cultural Affairs, and the Social Ministry of the Federal State of Mecklenburg-West Pomerania. Genome-wide genotyping in SHIP-TREND-0 was supported by the Federal Ministry of Education and Research (grant no. 03ZIK012). SHIP-TREND-0 was also funded by the German Research Foundation (DFG: GR 1912/5-1).
Marcella Rietschel received support from the BMBF 01ZX1311A (e:Med SysAlc), and the BMBF 01ZX1314G (e:Med IntegraMent) of the German Federal Ministry of Education and Research (BMBF) and from the European Community's Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 279227 (CRESTAR). Franziska Degenhardt received support from the BONFOR Programme of the University of Bonn, Germany.
Appendix A Supplementary data
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.eurpsy.2016.04.001.
Comments
No Comments have been published for this article.