Glutamatergic GRIN2B and Polyaminergic ODC1 Genes in Suicide Attempts: Associations and Gene--Environment Interactions with Childhood/Adolescent Physical Assault

ORIGINAL ARTICLE Glutamatergic GRIN2B and polyaminergic ODC1 genes in suicide attempts: associations and gene--environment interactions with childhood/adolescent physical assault

M Sokolowski, YJ Ben-Efraim, J Wasserman and D Wasserman

The complex etiology of suicidal behavior has frequently been investigated in relation to monoaminergic neurotransmission, but other neurosystems have shown alterations as well, involving excitatory glutamatergic and inhibitory g-aminobutyric acid (GABA) molecular components, together with the modulating polyamines. Sufficiently powered and family-based association studies of glutamatergic and GABAergic genes with suicidal behavior are nonexistent, but several studies have been reported for polyamines. We therefore conducted, for the first time ever, an extensive family-based study of 113 candidate single- nucleotide polymorphisms (SNPs) located in 24 glutamatergic and GABA genes, in addition to interrelated polyaminergic genes, on the outcome of severe suicide attempts (SAs). The family-based analysis (n ¼ 660 trios) was supplemented with gene-- environment interaction (G Â E), case--control (n ¼ 519 controls) and subgroup analyses. The main observations were the previously unreported association and linkage of SNPs rs2268115 and rs220557 in GRIN2B, as well as of SNPs rs1049500 and rs2302614 in ODC1 (Po10À2). Furthermore, GRIN2B haplotypic associations were observed, in particular with a four-SNP AGGC haplotype (rs1805247--rs1806201--rs1805482--rs2268115; Po10À5), and a third SNP rs7559979 in ODC1 showed G Â E with serious childhood/adolescent physical assault (Po10À4). SA subjects were characterized by transdiagnostic trait anger and past year alcohol--drug use disorders, but not by alcohol--drug use at SA, depression, anxiety or psychosis diagnoses. We also discuss a first ever confirmatory observation of SNP rs6526342 (polyaminergic SAT1) in SA, originally identified in completed suicides. The results suggest that specific genetic variants in a subset of glutamatergic (GRIN2B) and polyaminergic (ODC1) neurosystem genes may be of importance in certain suicidal subjects.

Molecular Psychiatry (2013) 18, 985--992; doi:10.1038/mp.2012.112; published online 31 July 2012 Keywords: alcohol--drugs; anger; childhood/adolescent physical assault; glutamate GRIN2B; polyamine ODC1; suicide attempt

INTRODUCTION during childhood/adolescence.10 Such exposures may be of The etiology of suicidality is complex, involving multiple genetic particular relevance for suicidality and suicide-related psycho- and environmental influences throughout life, as well as their pathologies among subjects with certain genetic variants in stress- interactions in between, as conceptualized in a stress--vulnerability responsive genes, through gene--environment interactions (G Â E) model.1,2 Suicidality is a heterogeneous phenomenon, co-occur- with stressful life events (SLEs) as typified by the serotonin ring with multiple psychiatric diagnoses and conditions. Further- transporter.11 more, the suicidal process encompasses a spectrum of suicidality In addition to the monoaminergic neurotransmitters, alterations ranging from suicidal ideation, suicide attempt (SA) behaviors of in other types of brain synaptic neurotransmission processes are various characteristics, to the most severe behavioral outcome of indicated, for example, the widely engaged excitatory glutama- completed suicides.2--4 tergic system. Glutamate receptors are divided into two types, The heritability of suicidality has been observed to be in the metabotropic and ionotropic, with the latter subdivided into range of 30--55% on the genome-wide level by using twin N-methyl-D-aspartate (NMDA) and non-NMDA receptors (NMDARs studies,5 but the possible contributions of specific genes and their and non-NMDARs). The classical paradigm has assigned a special different allelic variants is still in the process of ongoing study.3,6,7 role to the ionotropic NMDARs in long-term changes of synaptic Most attention has historically been focused on selected genetic plasticity,12 that is, memory and learning, and subsequent variants in serotonergic genes, but a polygenetic perspective on cognitive and affective pathology. Altered function of the suicidality is also further being pursued by the study of novel glutamatergic system is suggested in the etiologies of alcohol candidate genes in other neurosystems.1,3,6--8 Novel candidate misuse and dependency,13 depressive and anxiety disorders14--16 genes of particular interest for suicidal behavior can be suggested and psychotic schizophrenia/bipolar disorders,17 which are all by genome-wide association studies or by physically measured commonly (but not only) found in suicidality. Interestingly, intermediate phenotypes.9 In addition to genetic factors, environ- intermediate phenotypes of completed suicides have been shown mental factors are of established importance in the suicidal as alterations in gene expression and binding of non-NMDARs18--21 process; for example, the exposure to sexual and physical traumas and NMDARs.18 Reduced NMDAR densities were observed

National Centre for Suicide Research and Prevention of Mental Ill-Health (NASP), Karolinska Institute (KI), Stockholm, Sweden. Correspondence: Dr M Sokolowski, National Centre for Suicide Research and Prevention of Mental Ill-Health (NASP), Karolinska Institute, S-171 77, Stockholm, Sweden. E-mail: [email protected] Received 14 February 2012; revised 20 June 2012; accepted 26 June 2012; published online 31 July 2012 GRIN2B, ODC1 and physical assault in suicide attempts M Sokolowski et al 986 post-mortem in the frontal cortex22 and zinc had reduced potency Secondary outcomes to inhibit NMDARs in the post-mortem hippocampus from suicide Nine SA-related covariates, also described previously,45--48 were used for 23 victims. Increased expression of NMDAR agonist quinolinic acid exploratory post-hoc analysis as secondary outcomes: (1) the use of a in the anterior cingulate cortex was observed post-mortem in the violent SA method (n ¼ 104; ICD-10 codes X70--77, 79--83); (2) SA 24 brains of depressed suicide completers. Furthermore, the performed while intoxicated by alcohol or drugs (n ¼ 106; 90.6% alcohol NMDAR antagonist ketamine was shown to rapidly alleviate per se); (3) past year dependence/harmful use of alcohol or substance suicide ideation in subjects with treatment-resistant major dependence except tobacco (hereby referred to as ‘past year alcohol--drug 25 depressive disorder. use disorders’; n ¼ 116 with 63.7% alcohol per se);46 (4 and 5) elevated Other neurosystems interrelated with glutamate have also been anger/aggression as assessed by median cutoffs on the Trait Anger Scale49 shown to be altered in suicidal behaviors. These include the major (n ¼ 324) and the Past Feelings and Acts of Violence Scale50 (n ¼ 329); inhibitory g-aminobutyric acid (GABA)-ergic system, either with (6) current depressive symptoms as assessed by a Beck Depression Inventory reference to benzodiazepine binding attributed to the GABA-A (BDI) score of X17 (n ¼ 298); (7) lifetime depression diagnoses category 26--28 receptor or as directly for the GABA-A receptor and other (ICD-10 codes mild/moderate/severe/recurrent depression F32--F33 and 18,29--33 GABAergic genes. Furthermore, the observations on poly- dysthymia F34.1; n ¼ 142); (8) lifetime anxiety diagnoses category (ICD-10 34--36 amines in suicidal behavior is of interest in relation to codes F40--45, that is, panic, phobias, general anxiety, somatization, glutamate as well, as intracellular and extracellular polyamines obsessive--compulsive and posttraumatic stress disorder (PTSD); n ¼ 176); exert direct modulatory functions on NMDARs and non- and (9) lifetime schizophrenia or psychosis-related diagnoses category 37,38 NMDARs, respectively, for example, subunit-specific modula- (ICD-10 codes schizophrenia F20, delusional F22, psychotic F23, schizoaffec- tory mechanisms of GRIN2B-containing receptor complexes. tive F25, bipolar F31 and psychotic depression F33.3; n ¼ 88). Sufficiently powered and family-based association studies of glutamatergic or GABAergic genes in relation to suicidal behaviors Stressful life events are nonexistent. Studies of two glutamatergic genes (GRIA3 46 and GRIK2) were in relation to treatment-emergent suicidal Environmental exposures were assessed as previously described. ideation.39,40 However, several studies have associated polyami- Questions originally obtained from the Life Events section of the European nergic single-nucleotide polymorphisms (SNPs) explicitly with Parasuicide Study Interview Schedule version 5.133 and the PTSD section K suicidal behaviors.34,41--43 To gain better insight, we here of the Composite International Diagnostic Interview (CIDI) v2.1.34 were conducted a family-based association study of 113 SNPs located used to compile a broad lifetime exposure SLE checklist, with the across all NMDAR genes, as well as other candidate glutamatergic, cumulative exposure to at least 5 out of any of 19 SLEs used as a 46 GABAergic and polyaminergic neurosystem genes (Supplemen- dichotomous exposure variable (that is, median cutoff). One of the tary Tables S1 and S2), testing three succeeding hypotheses: (1) questions from CIDI PTSD-trauma questionnaire concerning serious direct, possible sexually dimorphic44 association and inheritance physical assault or attack (CIDI.22.7: ‘Were you ever seriously physically of certain SNP-allele(s) with SA will point to specific candidate attacked or assaulted?’) was also used in G Â E analysis as previously 46 gene(s) of interest for further downstream analysis; (2) these described, enabling additional analysis in relation to the age of exposure candidate gene(s) will show augmented association by haplotypes; (that is, exposure under or over 18 years of age, hereby referred to as and (3) these candidate gene(s) will contain SNPs associating ‘childhood/adolescent’ or ‘adulthood’, respectively). with SA through G Â E(s) involving exposure to physical assault (in childhood/adolescence or adulthood, or both) or lifetime, DNA purification, candidate SNP/gene selection and genotyping cumulative SLEs. In addition, we also wished to study covariates DNA was extracted as previously described.45 We selected 120 SNPs in 24 characterizing any genetic associations in SA subjects. glutamate, GABA and polyamine genes with hypothesized involvement in suicidal behavior-related psychopathology, with particular focus on GABRA1, all NMDAR subunits and major polyamine-metabolizing enzyme SUBJECTS AND METHODS genes (Supplementary Table S1). Selection of SNPs was guided by their Research subjects and the SA main outcome putative potential to exert functional effects (indicated by in silico prediction tools or their locations in gene flanks and exons), by their Nuclear family trios (all complete with both biological parents and one SA tagging potential (number of SNPs tagged) and/or by certain previous offspring per trio; n ¼ 660) and a sample of unrelated, nonsuicidal healthy studies of these SNPs. Following SNP genotyping as previously described47 controls (n ¼ 519) were collected in the Ukraine in 2001--2006, as 45 by use of an Illumina (San Diego, CA, USA) BeadStation 500GX (GoldenGate previously described. Various aspects of the sample recruitment, Assay), quality control procedures were applied. A total of 113 SNPs selection criteria, demographics, ancestry and International Classification (Supplementary Table S2) were available for analysis, fulfilling the criteria of Diseases--10th edition (ICD-10) psychiatric diagnoses have been 45--48 for Hardy--Weinbergequilibrium (P40.001 in healthy controls), minor allele described previously. Of the offspring, 100% (n ¼ 660) were recruited frequency (MAF, 40.01), call rate (490%; median 99.9%, lower quartile from emergency care because of a severe SA as defined by a score of X2 45--48 99.8%), absence of Mendelian inheritance errors, 100% intra-assay and an on the Medical Damage Rating Scale as previously described, estimated 99.4% 1-year inter-assay reproducibility, respectively. and this SA being well ascertained for each trio offspring45--48 represented the main outcome measure of the study. The SA offspring comprised 51.1% males (n ¼ 337) and 48.9% females (n ¼ 323), with mean ages of Statistical analyses 24.6 (s.d.±7.3) and 23.8 (s.d.±7.1) years, and had X3 Ukrainian or Data were stored and organized for analyses using FileMaker Pro Russian grandparents among 94.4% (n ¼ 318) and 93.2% (n ¼ 301) SA (FileMaker, Claris Corporation, Santa Clara, CA, USA). Sporadic missing subjects, respectively. Unrelated, nonsuicidal controls (n ¼ 519) had no data were imputed using the software BEAGLE v3.3.51 Family-based psychiatric diagnoses, as was previously described,45--48 and comprised genetic analyses and association tests were performed using the software 45.9% males (n ¼ 238) and 54.1% females (n ¼ 281), with mean ages of Haploview52 v4.2, FBAT v2.0.3 (www.biostat.harvard.edu/~fbat/ 35.7 (s.d.±16.1) and 34.6 (s.d.±14.8) years, and had X3 Ukrainian or default.html), software package GENASSOC53 as implemented in Stata Russian grandparents among 94.1% (n ¼ 224) and 93.6% (n ¼ 263), v9.2 (StataCorp, College Station, TX, USA), and PBAT v3.6 (www.biostat. respectively. The collection of research subjects followed the code of harvard.edu/~clange/default.htm) as implemented in SNP & Variation Suite ethics of the World Medical Association (Declaration of Helsinki), and a 7 (Golden Helix), used mainly as previously described.45--48 Family-based written consent was obtained. The study was approved by the Research G Â E analyses were conducted using the software fbat-i,54 run in Ethics Committee at the Karolinska Institute (Dnr 97--188) and by the R-environment v2.13.1 (www.r-project.org) and software package Ministry of Health in Ukraine. GENASSOC,53 both used as previously described.46 Exploratory post-hoc

Molecular Psychiatry (2013), 985 --992 & 2013 Macmillan Publishers Limited GRIN2B, ODC1 and physical assault in suicide attempts M Sokolowski et al 987 Table 1. Signiﬁcant association and linkage of SNPs in GRIN2B and ODC1 genes with SA

MAF (minor/ major Minor:major transmitted Overtransmitted Case--control OR SNP ID alleles) Gene Sex N alleles (±95% CI: P-value) allele OR (95% CI) FDR Q-value (95% CI)

rs1049500 0.07 (A/G) ODC1 f 323 29 A:59 G (1.4 Â 10À3) Major (G) 2.03 (1.30--3.17) 1.0 Â 10À1 (2.0 Â 10À2) 1.93 (1.26--2.97) rs2302614 0.07 (C/A) ODC1 f 323 29 C:59 A (1.4 Â 10À3) Major (A) 2.03 (1.30--3.17) 1.0 Â 10À1 (2.0 Â 10À2) 1.93 (1.26--2.97) rs2268115 0.46 (C/A) GRIN2B mf 660 364 C:291 A (4.3 Â 10À3) Minor (C) 1.25 (1.07--1.46) 1.4 Â 10À1 (3.8 Â 10À2) 1.27 (1.10--1.49) rs2268115 0.47 (C/A) GRIN2B m 337 189 C:139 A (5.8 Â 10À3) Minor (C) 1.36 (1.09--1.69) 1.4 Â 10À1 (4.3 Â 10À2) 1.35 (1.09--1.69) rs220557 0.35 (C/A) GRIN2B mf 660 319 C:255 A (7.6 Â 10À3) Minor (C) 1.25 (1.06--1.47) 1.4 Â 10À1 (5.5 Â 10À2) 1.47 (1.25--1.73) Abbreviations: 95% CI, 95% confidence interval; f, analysis in subgroup of female SA offspring; FDR, false discovery rate; m, analysis in subgroup of male SA offspring; MAF, minor allele frequency (in the parents) of the investigated sample; mf, analysis in the total sample of male and female SA offspring combined; N, number of trios; OR, odds ratio; SA, suicide attempt; SNP, single-nucleotide polymorphism. Significant association and linkage of four (out of 113) SNPs, located in 2 (out of 24) investigated genes. The 95% confidence intervals refer to the number of transmissions (column 6), the OR of cases vs matched pseudocontrols (column 8) or the age-adjusted OR of logistic regression (column 10); the latter (column 10) refers to a confirmatory, post-hoc re-analysis of the family-based findings by using a less powered case--control design, comparing SA offspring with unrelated, healthy and nonsuicidal control subjects (n ¼ 519). Uncorrected, two-tailed P-values are shown, as well as the FDR Q-values in relation to 185 family-based tests (and in parenthesis in relation to the study-wide n ¼ 1394 tests). analysis using subgroups as covariates was done using the software Association of a GRIN2B ‘AGGC’ risk haplotype with SAs package GENASSOC,53 mainly as previously described,55 using likelihood All possible haplotypes of adjacent SNPs were first analyzed; that ratio tests to determine subgroup ‘predominance’ (nominal PLRTp0.05). is, for GRIN2B all haplotypes between 2 and 11 adjacent SNPs in Case--control confirmatory re-analyses were performed using logistic length (‘sliding window’ analysis). For GRIN2B, the top-significant command in Stata v9.2 with one-tailed confidence intervals, adjusted for results showed overtransmission for six haplotypes age, and G Â E interaction on the additive scale of measurement was done (all Po4.0 Â 10À4) of 4--7 SNPs in length, all containing the minor as previously described.56 Haplotype phase was resolved using the software risk C allele of SNP rs2268115. We then additionally tested all PHASE,57 v.2.1. Analyses were conducted under an additive model of possible sub-haplotypes of these seven SNPs. Table 2 shows the inheritance, unless otherwise stated. top-significant four-SNP haplotype, ‘AGGC’ (P ¼ 8.4 Â 10À6). A Uncorrected (nominal), two-tailed P-values are shown. The false discovery similar analysis of all ODC1 sub-haplotypes did not reveal any rate (FDR) was investigated by calculating positive FDR analogs (Q-values) of similar benefit of haplotypes over single-SNP analysis (data not P-values with software QVALUE (www.genomics.princeton.edu/storeylab/ shown). qvalue/),58 aiming at less than one false discovery among the top significant findingsoftheinitialanalysis(n ¼ 185 tests, referred to as ‘screening’ Q- AGÂ E between ODC1 SNP rs7559979 and childhood/adolescent value). In addition, the study-wide FDR for a total of n ¼ 1394 statistical tests physical assault on SA outcome performed is also described. The exploratory analysis of subgroups as An analysis of G Â Es did not reveal significance (that is, secondary outcomes was not included in the FDR analysis. Statistical power PG Eo0.01) for the GRIN2B or ODC1 SNPs or AGGC haplotype, was calculated by using the software QUANTO v.1.2.4 (http://hydra.usc. Â 59 which had shown direct genetic associations (Tables 1 and 2). edu/gxe/) for analysis in the total sample. SNP odds ratios (ORs) with 80% However, one significant G Â E was revealed between a third power (using a ¼ 0.05, log-additive model, n ¼ 660 trios, population ODC1 SNP rs7559979 and childhood/adolescent physical assault risk ¼ 0.01) were in the range between 1.25 and 2.25 (0.494MAF40.013), À5 (Figure 2c; ORG Â E ¼ 2.32 (1.51; 3.56), P ¼ 9.6 Â 10 ; study-wide and G Â E ORs with 80% power (additionally using environmental exposure FDR Q ¼ 1.1 Â 10À2). The G Â E was of lower magnitude with frequencies 18.3--51.6%, marginal environmental effects 0.46--1.94 and lifetime physical assault exposure (Figure 2b; ORG Â E ¼ 1.51 (1.03; assuming no main genetic effect) were in the range between 1.65 and 2.21)) and was not observed with SLE-high or adulthood physical 2.60 (0.474MAF40.06), respectively. assault (Figures 2a and d) as marginal exposures, respectively. Furthermore, the G Â E with childhood/adolescent physical assault remained significant among individuals who were not co-exposed RESULTS to adulthood physical assault or SLE-high (Figures 2e and f), Association and linkage of SNPs in the GRIN2B and ODC1 genes respectively. The G Â E was confirmed in a less powered case-- with SAs control re-analysis, mainly when using a recessive coding (age- Family-based association analyses of 113 selected SNPs in 24 adjusted ORG Â E ¼ 3.11 (1.29; 7.55)). The case--control design also glutamate, GABA and polyamine genes (see Supplementary enabled statistical analysis on the additive (rather than multi- Tables S1 and S2) were conducted in the whole sample (n ¼ 660 plicative) interaction scale, further confirming the G Â E interaction trios) on the main outcome of SAs, as well as in male and females (attributable proportion because of interaction, AP ¼ 0.63 (0.30; separately for SNPs showing Po0.25 in the whole sample (185 0.96)). analyses in total). Table 1 shows that the top five association signals had a screening FDR of Qo0.15 (that is, o1 false positive Past year alcohol--drug use disorders and high anger characterized among them) in the trio sample. Some findings of potential the associated SA subjects interest in other genes were also observed at higher FDR The post-hoc covariate analysis of nine subgroups showed that thresholds (QX0.15; data not shown and see Discussion). We both GRIN2B rs2268115 and rs220557 SNP associations were could rule out that associations had been induced by poor predominant in past year alcohol--drug use disorders (PLRTp0.05), genotyping quality, sporadic missing data, population stratifica- in addition to the use of violent SA method among females tion or by the family-based design. The top five association signals (PLRTp0.05), whereby the combined analysis is shown (Table 3). (Table 1) thus mainly suggested two candidate genes of interest, Furthermore, a high anger subgroup showed maintained/enriched that is, ODC1 (through SNPs rs1049500 and rs2302614; pairwise significances for rs2268115/rs220557, respectively (data not r2 ¼ 1.00, Figure 1) and GRIN2B (through SNPs rs2268115 and shown). For the GRIN2B AGGC haplotype, none of the subgroups rs220557; pairwise r2 ¼ 0.18, Figure 1). were directly predominant per se; however, high anger showed

& 2013 Macmillan Publishers Limited Molecular Psychiatry (2013), 985 --992 GRIN2B, ODC1 and physical assault in suicide attempts M Sokolowski et al 988

13700k 13750k 13800k 13850k 13900k 13950k 14000k 14050k 14100k 14150k

10580 10582 10584 10586 10588

Figure 1. Schematics of the GRIN2B (left panel) and ODC1 (lower-right panel) NCBI (National Center for Biotechnology Information) reference genes are shown, with the corresponding matrices of the pairwise linkage disequilibrium (LD) values for the investigated single-nucleotide polymorphisms (SNPs) in each gene, respectively. The upper-most gene schematics depict the full gene reference transcripts, with thin-angled horizontal lines representing introns and vertical lines representing exons. The scaled bars indicate kilobase positions at chromosomes 12 and 2 (GRCh37 assembly), respectively. Numbers in the squares of the two matrices represent r2values between each corresponding SNP in the total sample (also depicted graphically with shades of grey, from white (r2 ¼ 0) to black (r2 ¼ 1)). The black triangle in the GRIN2B matrix depicts the studied AGGC haplotype.

Table 2. Association of a GRIN2B four-SNP ‘AGGC’ haplotype with SA

MAF (minor:major AGGC haplotype AGGC haplotype analysis SNPs alleles) Position SNP alleles AGGC:other Case-- control haplotype(s) transmitted OR re-analysis: OR Freq (±95% CI: P-value) (95% CI) Q-value (95% CI)

rs1805247 0.11 (G:A) 13 715 975 A rs1806201 0.32 (A:G) 13 717 508 G 0.16 215:132 1.63 6.7 Â 10À3 1.25 (1.02--1.54) (±18: 8.4 Â 10À6) (1.31--2.02) rs1805482 0.26 (A:G) 13 764 774 G rs2268115 0.46 (C:A) 13 869 725 C Abbreviations: 95% CI, 95% confidence interval; MAF, minor allele frequency (in the parents) of the investigated sample; position refers to chromosome 12 (GRCh37 assembly); OR, odds ratio; SA, suicide attempt; SNP, single-nucleotide polymorphism. A GRIN2B four-SNP ‘AGGC’ haplotype showed augmented association and linkage. The 95% confidence intervals for the number of transmissions and for the OR of cases vs matched pseudocontrols. Case--control re-analysis was a confirmatory, post-hoc re-analysis of the family-based findings by using a less powered case--control design, with the age- adjusted OR of logistic regression. Uncorrected, two-tailed P-values are shown, as well as study-wide Q-values (n ¼ 1394 tests).

enriched significance (Table 3, columns 3--5 vs 6--8), and a high subgroups showed either predominance or enriched significance anger subgroup delimited by further covariates showed predo- (data not shown). minance (PLRTp0.05; Table 3, columns 9 and 10). For the ODC1 SNP rs1049500, high anger showed predominance in both male and female SAs (PLRTp0.05; Table 3), as well as predominance for DISCUSSION the delimited high anger subgroup (PLRT ¼ 0.0027), thus not supporting the putative female sex specificity observed on SA Here we report the first ever observed association and linkage of outcome alone (Table 1). For the ODC1 rs7559979 Â childhood/ multiple genetic variants (Tables 1 and 2) in glutamatergic GRIN2B adolescent physical assault G Â E, predominance of high anger and polyaminergic ODC1 genes on SAs. Furthermore, we also was observed (PLRTp0.05; Table 3) as well. None of the other report for the first time that a SNP in ODC1 moderated the

Molecular Psychiatry (2013), 985 --992 & 2013 Macmillan Publishers Limited GRIN2B, ODC1 and physical assault in suicide attempts M Sokolowski et al 989 a 2.5 b 2.5 previously reported associations concerning various suicidal behavior-related4 neuropsychological measures. Carriers of the 2 2 AA genotype of rs1805247 were shown to have reduction in the 1.5 1.5 * NMDA-related intracortical facilitation and long-term potentiation plasticity,65 whereas carriers of the CC (or GG) genotype of 1 1 rs1806201 have been shown to have altered cognitive functions, for example, less use of a win-stay strategy in the Iowa gambling 0.5 0.5 66 Transmission ratio Transmission ratio test (increase in risky behaviors), lowered orbitofrontal response 0 0 inhibition by increased false-alarm rate (impaired impulse control Not SLE-high SLE-high Unexposed PA-Life and conflict processing) and reduced event-related potentials.67 c 2.5 d 2.5 All these described GRIN2B SNP associations and their allelic directions are concordant with the GRIN2B AGGC risk haplotype 2 2 *** observed herein, as the first, second and last SNPs (and alleles) 1.5 1.5 included in the haplotype were rs1805247 (A), rs1806201 (G) and 1 1 rs2268115 (C), respectively. ODC1 encodes an anabolic rate-limiting enzyme of the 0.5 0.5 polyamine biosynthesis pathway, catalyzing ornithine to putres- Transmission ratio Transmission ratio 0 0 cine, with putative consequences for modulation of glutamate Unexposed PA-U18 Unexposed PA-O18 receptors.37,38 The cellular polyamine stress response, in part e 2.5 f 10 connected to HPA activation, is marked by an increase in ODC 68 2 8 activity. Although we found no direct precedent in the literature ** *** concerning ODC1 SNP associations, two major modulators of ODC1 1.5 6 (that is, OAZ1 and OAZ2) were recently shown to have increased 1 4 gene expression in multiple brain regions of completed suicides.35 0.5 2 Interestingly, SNPs in the SAT1 gene encoding for a polyamine rate- Transmission ratio Transmission ratio limiting catabolic enzyme have shown associations with completed 0 0 suicides,42,43 SAs34 and altered SAT1 expression in suicide Unexposed PA-U18 Unexposed PA-U18 35,41--43,69 Not PA-O18 Not SLE-high brains. We could also observe associations of SAT1 SNPs, if considering a more liberal screening threshold of Qo0.25 (for Figure 2. The gene--environment interaction (G Â E) analyses bet- example, at nominal P-value of p0.05; data not shown), that is, for ween ODC1 single-nucleotide polymorphism (SNP) rs7559979 and rs3764885 (G-allele overtransmission; P ¼ 0.031) and rs6526342 (C- different, partially overlapping environmental exposures. (a) Life- time, cumulative occurrence of at least 5 out of 19 different allele overtransmission, in females P ¼ 0.033 and overall P ¼ 0.049). Thus, the previously reported SAT1 rs6526342 C-allele associations stressful life events (SLEs; SLE-high) as exposure (n ¼ 339 exposed), 42,43 (b) lifetime physical assault (PA) as exposure (n ¼ 185 exposed), (c) in completed suicides was here confirmed for the first time childhood/adolescent physical assault (PA-U18) as exposure (n ¼ 121 ever in SAs as well. Although SAT1 rs6526342 was demonstrated to exposed), (d) adulthood physical assault (PA-O18) as exposure affect transcription,42,43 we have no corresponding molecular-level (n ¼ 121 exposed), (e) PA-U18 among not PA-O18 exposed, as ex- data for our associated SNPs, but it is noteworthy that three SNPs posure (n ¼ 73, 60% of PA-U18 exposed), and (f) PA-U18 among not in the GRIN2B AGGC haplotype are exonic (rs1805247, rs1805482 SLE-high exposed, as exposure (n ¼ 21, 17% of PA-U18 exposed), and rs1806201) and for ODC1, rs1049500 is exonic and rs2302614 respectively. Black squares (’) depict the ratio of minor G-alleles vs maps to the 50-untranslated region. At the liberal Q 0.25 FDR major A-alleles transmitted (that is, minor allele transmission ratio), o whereas black ovals (K) depict transmission ratio of major A-alleles level, we however also observed overtransmission of alleles in the vs minor G-alleles transmitted (that is, major allele transmission ra- glutamate transporter SLC6A1 (C-allele of rs1728818), GRIN3A (A- tio) in each exposure-stratum, respectively. Error bars represent the allele of rs10989591), GRIN2D in overall (C-allele of rs276717) and standard errors of the ratios. G Â E nominal P-values: *Po0.05; females (C-allele of rs220557), GABRG1 in females (T-allele of **Po0.01; ***Po0.001. rs12374299 and G-allele of rs1497567) and GABRA1 in males (A- allele of rs4428455) and females (A-allele of rs13156895), respectively. Although 25% of the findings at this level are association between childhood/adolescent physical assault and SAs statistical false discoveries according to FDR, the results might still (Figure 2). Exploratory follow-up analyses indicated that all suggest the importance of SNPs in additional glutamatergic and observed associated SA subjects shared the characteristic of high GABAergic signaling genes for suicidality, perhaps reflecting trait anger, and that GRIN2B SNP associations were also character- previously demonstrated gene expression alterations,18,29--33 ized by past year alcohol--drug use disorders and a violent SA hypotheses that remain to be tested further. method in females. Finally, we conclude that none of the initial SA Our characterization of the GRIN2B and ODC1 associations by associations (Table 1) were robustly specific for a certain sex after subgroups (Table 3) suggested high trait anger, past year alcohol-- accounting for these covariates. In a broader perspective, GRIN2B drug use disorders and/or use of violent SA method (in females) as and ODC1 functions might interact with and influence neurofunc- predominant, in contrast to depression, anxiety or psychosis tions implicated in suicidal behaviors,4 for example, the stress- diagnoses. A scale originally designed to predict violent acts50 did responsive hypothalamic--pituitary--adrenal (HPA) axis.7,46 Although not characterize our genetic SA associations as well as trait anger we cannot make any causal interpretations from our data, one may scale, despite overall interscale correlation among all SAs envision, for example, that the altered NMDAR functioning (r2 ¼ 0.44, Po0.001). Neither did concurrent depressive symptoms observed in suicidality,23--25 the role of NMDARs in HPA axis activa- (BDI) explain the genetic SA associations, despite overall interscale tion60,61 involving GRIN2B62 and the HPA dysregulation commonly correlation (with anger r2 ¼ 0.19, Po0.001). Although we cannot observed in suicidal behavior at the level of cortisol3,4,7,63 might be exclude some nonnegligible background role of these two self- related to the associations reported herein. report measures in high anger, diagnoses of depression, anxiety or The importance of GRIN2B gene integrity for human cognitive psychosis were not related to high anger (interscale correlations function was recently substantiated by the observation of with anger r2o0.1, P40.05). Trait anger has been described by, for deleterious de novo mutations in mental retardation.64 Our results example, attention biases toward hostile stimuli, negative reason- with GRIN2B SNPs and AGGC haplotype show congruence with ing, rumination70 as well as by amygdala--orbitofrontal functional

& 2013 Macmillan Publishers Limited Molecular Psychiatry (2013), 985 --992 GRIN2B, ODC1 and physical assault in suicide attempts M Sokolowski et al 990 Table 3. Past year alcohol/drugs and high anger were predominant among the associated SA subjects

Results, within subgroup Results, not in subgroup Difference, subgroup vs not

Marker (gene) Subgroup(s)Sex Minor:major transmitted alleles Minor:major transmitted Interaction ORa LRT N (P-value) OR (95% CI) N alleles (P-value) OR (95% CI) (95% CI) P-value

rs2268115 (GRIN2B) Past year alcohol- 146b 98 C:46 A (1.5 Â 10À5) 2.13 (1.50--3.02) 495 258 C:230 A (2.1 Â 10À1) 1.12 (0.94--1.34) 1.90 (1.28--2.81) 1.1 Â 10À3 drugm&f Violent SA methodf rs220557 (GRIN2B) Past year alcohol- 146b 88 C:48 A (6.0 Â 10À4) 1.83 (1.29--2.61) 495 222 C:202 A (3.3 Â 10À1) 1.10 (0.91--1.33) 1.67 (1.12--2.49) 1.1 Â 10À2 drugm&f Violent SA methodf c À2 AGGC (GRIN2B) High angerm&f 324 102 AGGC:57 1.79 (1.29--2.47) 323 109 AGGC: 73 1.49 (1.11--2.01) 1.59 (1.00--2.51) 4.5 Â 10 other (4.0 Â 10À4) other (7.6 Â 10À3) À2 À1 À2 rs1049500 (ODC1) High angerm&f 324 35 A:58 G (1.7 Â 10 ) 1.66 (1.09--2.52) 323 36 A:31 G (5.4 Â 10 ) 0.86 (0.53--1.39) 1.92 (1.02--3.64) 4.3 Â 10 rs7559979 73 (+E) 46 G:18 A (5.0 Â 10À4) 3.63d (1.99--6.63) 45 (+E) 24 G:18 A (3.5 Â 10À1) 1.42d (0.74--2.74) 2.56e (1.05--6.24) 3.9 Â 10À2 (ODC1) Â physical High angerm&f 249 (ÀE) 100 G:142 A 278 (ÀE) 124 G:132 A assault (6.9 Â 10À3) (6.2 Â 10À1)

Abbreviations: 95% CI, 95% conﬁdence interval; E+ and EÀ,GÂ E analysis among those exposed vs not exposed, respectively, to serious childhood/adolescent physical assault at p18 years of age; f, analysis in subgroup of female SA offspring; High anger; Trait Anger Scale (TAS) score of X20; LRT, likelihood ratio test; m, analysis in subgroup of male SA offspring; m&f, analysis in the total sample of male and female SA offspring combined; N, number of trios within the indicated subgroup(s); OR, odds ratio; Past year alcohol/drug, past year dependence/harmful use of alcohol or substance dependency disorders; SA, suicide attempt; Violent SA method, SA by using a violent method (for example, hanging, drowning, shooting, jumping or crashing). aRatio of ORs between subjects in subgroup vs not in subgroup, respectively. b Combined subjects of past year alcohol-drugm&f (nm ¼ 83, nf ¼ 33) and violent SA methodf (additional nf ¼ 30). cResults of using a delimited high anger subgroup (n ¼ 270, 83% of all high anger subjects), as interaction ORs 41 were observed for certain other subgroups, that is, delimited by excluding violent SA method males and females intoxicated by alcohol or drugs during SA, respectively. dG Â E ORs between exposed vs not exposed, respectively, to childhood/adolescent physical assault, for high and low anger subjects. eRatio of G Â E ORs between subgroups of high and low anger, that is, three-way interaction OR in the full model. The most predominant subgroups observed (column 2) for the associations among all SAs (column 1), with results shown for SA subjects within the subgroup (columns 3--5), for SA subjects not in the subgroup (columns 6--8), and for the difference in between these two types of SA subjects (columns 9 and 10), respectively. A total of 9 subgroups were explored (see Materials and methods) for the original G and gene--environment interaction (G Â E) associations of the main outcome SA (that is, see Tables 1 and 2, Figure 2 and text). Uncorrected, two-tailed representing the transmission distortion P-values are shown, in addition to the LRT P-values.

connectivity,71 and has been observed in a range of depression 8.9% of the general population, with a main effect on SA risk and anxiety diagnoses but is not classified as a main factor of any (OR ¼ 1.7, Po0.01), attributing for 5.9% of population SA cases. diagnosis (that is, ‘transdiagnostic’),70 representing one possible Joint risk exposure to GG genotype and childhood/adolescent explanation for the lack of predominance of either depression or physical assault was estimated to occur only among 1.5% of the anxiety diagnosis categories. Our negative result for psychotic general population (OR ¼ 2.8, Po0.01), but nevertheless attributed diagnoses might be because of relatively low statistical power. for 2.6% of population SA cases because of the G Â E. Impulsive aggression/anger coupled with serotonergic hypofunc- Furthermore, the 12.5% GG carriers unexposed to childhood/ tion is a long-known feature of suicidal behaviors,4,72 being further adolescent physical assault tended toward a lowered SA risk interrelated with alcohol--drug use disorders and NMDAR func- (OR ¼ 0.7, P ¼ 0.06), attributing for À3.7% of population SA cases. tion.73,74 It is interesting that past year (but not SA-concomitant) This indicates a putative future potential for G Â Es in, for example, alcohol--drug use disorders were implicated herein for GRIN2B guiding the allocation of preventive resources against childhood/ SNPs, as this gene was recently attributed a key role for ethanol adolescent physical assault and/or to subjects with increased action in the HPA axis regulatory bed nucleus of the stria terminalis genetic risk. of mice.75 Furthermore, our ODC1 G Â E showed SLE-type specificity The association results reported herein suggest, for the first for childhood/adolescent physical assault rather than other SLE- time ever, a putative involvement of specific GRIN2B and ODC1 type lifetime exposures (Figure 2), perhaps reflecting the genetic variants in the suicidal process, in part by G Â E with importance of particular childhood traumas for suicidality marked childhood/adolescent physical assault. Future use of prospective by impulsive anger/aggression.4,10,72 In comparison, G Â Es invol- and experimental--functional approaches, involving detailed phe- ving childhood physical abuse in particular and anger/aggression notyping and genotyping of subjects, might reveal neurobiologi- have been reported for HPA genes FKPB5 and CRHR1.46,76 We cal mechanisms involving GRIN2B and ODC1 in the underlying finally note that the physical assault measure used by us is a psychopathology of suicidal behaviors, with novel opportunities single-item trauma question from the CIDI diagnosis PTSD section, for improvement of prevention and intervention in clinical and differing from the more commonly used physical abuse/neglect population-wide settings. multi-item scores of the childhood trauma questionnaire. In summary, it seems plausible that both the GRIN2B and ODC1 associations might reflect a similar pattern of SAs characterized by CONFLICT OF INTEREST high anger and/or alcohol--drug use disorders, together with other The authors declare no conflict of interest. interrelated aspects thereof. Estimation from the case--control data77 of attributable fractions allowed theoretical quantification of the putative impact of the ACKNOWLEDGEMENTS ODC1 G Â E on SAs in the general population. Exposure to We thank all interviewers at the Human Ecological Health organization/Odessa childhood/adolescent physical assault as sole risk factor (that is, National Mechnikov University, Odessa, Ukraine; Professor Vsevolod Rozanov for among the 86% nonrisk AG/AA carriers) was estimated to occur in coordination of the material collection in Ukraine; Dr Vladymyr Bogatov and

Molecular Psychiatry (2013), 985 --992 & 2013 Macmillan Publishers Limited GRIN2B, ODC1 and physical assault in suicide attempts M Sokolowski et al 991 laboratory technician Lars Holmberg for technical assistance; Dr Tatyana Reytarova aspartate antagonist in patients with treatment-resistant major depressive for logistic assistance; and all those who have given their consent to participate as disorder. J Clin Psychiatry 2010; 71: 1605--1611. research subjects in this study. We thank the SNP&SEQ Technology Platform at 26 Manchon M, Kopp N, Rouzioux JJ, Lecestre D, Deluermoz S, Miachon S. Uppsala University for their excellent quality of services and expertise in the geno- Benzodiazepine receptor and neurotransmitter studies in the brain of suicides. typing conducted. The study was funded by the Marianne and Marcus Wallenberg Life Sci 1987; 41: 2623--2630. Foundation, the Knut and Alice Wallenberg Foundation and the American 27 Pandey GN, Conley RR, Pandey SC, Goel S, Roberts RC, Tamminga CA et al. Foundation for Suicide Prevention. Benzodiazepine receptors in the post-mortem brain of suicide victims and schizophrenic subjects. Psychiatry Res 1997; 71: 137--149. 28 Marazziti D, Dell’Osso B, Baroni S, Masala I, Di Nasso E, Giannaccini G et al. Decreased density of peripheral benzodiazepine receptors in psychiatric patients REFERENCES after a suicide attempt. Life Sci 2005; 77: 3268--3275. 1 Mann JJ. Neurobiology of suicidal behaviour. Nat Rev Neurosci 2003; 4: 819--828. 29 Sequeira A, Klempan T, Canetti L, Ffrench-Mullen J, Benkelfat C, Rouleau GA et al. 2 Wasserman D, Sokolowski M, Wasserman J, Rujescu D. Neurobiology and the Patterns of gene expression in the limbic system of suicides with and without genetics of suicide. In: Wasserman D, Wasserman C (eds). Oxford Textbook of major depression. Mol Psychiatry 2007; 12: 640--655. Suicidology---The Five Continents Perspective. Oxford University Press: Oxford, UK, 30 Klempan TA, Sequeira A, Canetti L, Lalovic A, Ernst C, Ffrench-Mullen J et al. 2009, pp 165--182. Altered expression of genes involved in ATP biosynthesis and GABAergic 3 Wasserman D, Terenius L, Wasserman J, Sokolowski M. The 2009 Nobel neurotransmission in the ventral prefrontal cortex of suicides with and without conference on the role of genetics in promoting suicide prevention and the major depression. Mol Psychiatry 2009; 14: 175--189. mental health of the population. Mol Psychiatry 2010; 15: 12--17. 31 Merali Z, Du L, Hrdina P, Palkovits M, Faludi G, Poulter MO et al. Dysregulation in 4 Mann JJ, Arango V, Avenevoli S, Brent D, Champagne F, Clayton P et al. Candidate the suicide brain: mRNA expression of corticotropin-releasing hormone receptors endophenotypes for genetic studies of suicidal behavior. Biol Psychiatry 2009; 65: and GABA(A) receptor subunits in frontal cortical brain region. J Neurosci 2004; 24: 556--563. 1478--1485. 5 Voracek M, Loibl LM. Genetics of suicide: a systematic review of twin studies. Wien 32 Poulter MO, Du L, Weaver ICG, Palkovits M, Faludi G, Merali Z et al. GABAA Klin Wochenschr 2007; 119: 463--475. receptor promoter hypermethylation in suicide brain: implications for the 6 Rujescu D, Giegling I. The genetics of neurosystems in mental ill-health and involvement of epigenetic processes. Biol Psychiatry 2008; 64: 645--652. suicidality: beyond serotonin. Eur Psychiatry 2010; 25: 272--274. 33 Poulter MO, Du L, Zhurov V, Palkovits M, Faludi G, Merali Z et al. Altered 7 Currier D, Mann JJ. Stress, genes and the biology of suicidal behavior. Psychiatr organization of GABA(A) receptor mRNA expression in the depressed suicide Clin North Am 2008; 31: 247--269. brain. Front Mol Neurosci 2010; 3:3. 8 Ernst C, Mechawar N, Turecki G. Suicide neurobiology. Prog Neurobiol 2009; 89: 34 Fiori LM, Wanner B, Jomphe V, Croteau J, Vitaro F, Tremblay RE et al. Association 315--333. of polyaminergic loci with anxiety, mood disorders, and attempted suicide. PLoS 9 Meyer-Lindenberg A, Weinberger DR. Intermediate phenotypes and genetic One 2010; 5: e15146. mechanisms of psychiatric disorders. Nat Rev Neurosci 2006; 7: 818--827. 35 Fiori LM, Bureau A, Labbe A, Croteau J, Noe¨lS,Me´rette C et al. Global gene 10 Brodsky BS, Stanley B. Adverse childhood experiences and suicidal behavior. expression profiling of the polyamine system in suicide completers. Int J Psychiatr Clin North Am 2008; 31: 223--235. Neuropsychopharmacol 2011; 14: 595--605. 11 Caspi A, Hariri AR, Holmes A, Uher R, Moffitt TE. Genetic sensitivity to the 36 Chen GG, Fiori LM, Moquin L, Gratton A, Mamer O, Mechawar N et al. Evidence of environment: the case of the serotonin transporter gene and its implications for altered polyamine concentrations in cerebral cortex of suicide completers. studying complex diseases and traits. Am J Psych 2010; 167: 509--527. Neuropsychopharmacology 2010; 35: 1477--1484. 12 Rao VR, Finkbeiner S. NMDA and AMPA receptors: old channels, new tricks. Trends 37 Strømgaard K, Mellor I. AMPA receptor ligands: synthetic and pharmacological Neurosci 2007; 30: 284--291. studies of polyamines and polyamine toxins. Med Res Rev 2004; 24: 589--620. 13 McCool BA, Christian DT, Diaz MR, La¨ck AK. Glutamate plasticity in the 38 Mony L, Kew JNC, Gunthorpe MJ, Paoletti P. Allosteric modulators of NR2B- drunken amygdala: the making of an anxious synapse. Int Rev Neurobiol 2010; containing NMDA receptors: molecular mechanisms and therapeutic potential. Br 91: 205--233. J Pharmacol 2009; 157: 1301--1317. 14 Gao S-F, Bao A-M. Corticotropin-releasing hormone, glutamate, and c-aminobu- 39 Laje G, Paddock S, Manji H, Rush AJ, Wilson AF, Charney D et al. Genetic markers tyric acid in depression. Neuroscientist 2011; 17: 124--144. of suicidal ideation emerging during citalopram treatment of major depression. 15 Hashimoto K. Emerging role of glutamate in the pathophysiology of major Am J Psychiatry 2007; 164: 1530--1538. depressive disorder. Brain Res Rev 2009; 61: 105--123. 40 Brent D, Melhem N, Turecki G. Pharmacogenomics of suicidal events. 16 Riaza Bermudo-Soriano C, Perez-Rodriguez MM, Vaquero-Lorenzo C, Baca-Garcia Pharmacogenomics 2011; 11: 793--807. E. New perspectives in glutamate and anxiety. Pharmacol Biochem Behav 2011; 41 Fiori LM, Turecki G. Genetic and epigenetic influences on expression of spermine 100: 752--774. synthase and spermine oxidase in suicide completers. Neuropsychopharmacology 17 Cherlyn SYT, Woon PS, Liu JJ, Ong WY, Tsai GC, Sim K. Genetic association studies 2010; 13:725--736. of glutamate, GABA and related genes in schizophrenia and bipolar disorder: a 42 Fiori LM, Mechawar N, Turecki G. Identification and characterization of decade of advance. Neurosci Biobehav Rev 2010; 34: 958--977. spermidine/spermine N1-acetyltransferase promoter variants in suicide comple- 18 Sequeira A, Mamdani F, Ernst C, Vawter MP, Bunney WE, Lebel V et al. Global brain ters. Biol Psychiatry 2009; 66: 460--467. gene expression analysis links glutamatergic and GABAergic alterations to suicide 43 Sequeira A, Gwadry FG, Ffrench-Mullen JMH, Canetti L, Gingras Y, Casero RA et al. and major depression. PLoS One 2009; 4: e6585. Implication of SSAT by gene expression and genetic variation in suicide and major 19 Freed WJ, Dillon-Carter O, Kleinman JE. Properties of [3H]AMPA binding in depression. Arch Gen Psychiatry 2006; 63: 35--48. postmortem human brain from psychotic subjects and controls: increases in 44 Ober C, Loisel DA, Gilad Y. Sex-specific genetic architecture of human disease. Nat caudate nucleus associated with suicide. Exp Neurol 1993; 121: 48--56. Rev Genet 2008; 9: 911--922. 20 Noga JT, Hyde TM, Herman MM, Spurney CF, Bigelow LB, Weinberger DR et al. 45 Wasserman D, Geijer T, Rozanov V, Wasserman J. Suicide attempt and basic Glutamate receptors in the postmortem striatum of schizophrenic, suicide, and mechanisms in neural conduction: relationships to the SCN8A and VAMP4 genes. control brains. Synapse 1997; 27: 168--176. Am J Med Genet B Semin Med Genet 2005; 133B: 116--119. 21 Kalkman HO. Circumstantial evidence for a role of glutamine-synthetase in 46 Ben-Efraim YJ, Wasserman D, Wasserman J, Sokolowski M. Gene-environment suicide. Med Hypotheses 2011; 76: 905--907. interactions between CRHR1 variants and physical assault in suicide attempts. 22 Nowak G, Ordway GA, Paul IA. Alterations in the N-methyl-D-aspartate (NMDA) Genes Brain Behav 2011; 10: 663--672. receptor complex in the frontal cortex of suicide victims. Brain Res 1995; 675: 47 Wasserman D, Wasserman J, Rozanov V, Sokolowski M. Depression in suicidal 157--164. males: genetic risk variants in the CRHR1 gene. Genes Brain Behav 2009; 8: 72--79. 23 Nowak G, Szewczyk B, Sadlik K, Piekoszewski W, Trela F, Florek E et al. Reduced 48 Wasserman D, Sokolowski M, Rozanov V, Wasserman J. The CRHR1 gene: a marker for potency of zinc to interact with NMDA receptors in hippocampal tissue of suicide suicidality in depressed males exposed to low stress. Genes Brain Behav 2008; 7: 14--19. victims. Pol J Pharmacol 2003; 55: 455--459. 49 Spielberger C. State-Trait Anger Expression Inventory. Psychological Assessment 24 Steiner J, Walter M, Gos T, Guillemin GJ, Bernstein H-G, Sarnyai Z et al. Severe Resources: Orlando, FL, 1988. depression is associated with increased microglial quinolinic acid in subregions of 50 Plutchik R, van Praag HM. A self-report measure of violence risk, II. Compr the anterior cingulate gyrus: evidence for an immune-modulated glutamatergic Psychiatry 1990; 31: 450--456. neurotransmission? J Neuroinflammation 2011; 8:94. 51 Browning BL, Browning SR. A unified approach to genotype imputation and 25 DiazGranados N, Ibrahim LA, Brutsche NE, Ameli R, Henter ID, Luckenbaugh DA haplotype-phase inference for large data sets of trios and unrelated individuals. et al. Rapid resolution of suicidal ideation after a single infusion of an N-methyl-D- Am J Hum Genet 2009; 84: 210--223.

& 2013 Macmillan Publishers Limited Molecular Psychiatry (2013), 985 --992 GRIN2B, ODC1 and physical assault in suicide attempts M Sokolowski et al 992 52 Barrett JC, Fry B, Maller J, Daly MJ. Haploview: analysis and visualization of LD and 65 Mori F, Ribolsi M, Kusayanagi H, Siracusano A, Mantovani V, Marasco E et al. haplotype maps. Bioinformatics 2005; 21: 263--265. Genetic variants of the NMDA receptor influence cortical excitability and plasticity 53 Cordell HJ, Clayton DG. A unified stepwise regression procedure for evaluating in humans. J Neurophysiol 2011; 106: 1637--1643. the relative effects of polymorphisms within a gene using case/control or 66 Ness V, Arning L, Niesert HE, Stuettgen MC, Epplen JT, Beste C. Variations in the family data: application to HLA in type 1 diabetes. Am J Hum Genet 2002; 70: GRIN2B gene are associated with risky decision-making. Neuropharmacology 124--141. 2011; 61: 950--956. 54 Hoffmann TJ, Lange C, Vansteelandt S, Laird NM. Gene-environment interaction 67 Beste C, Baune BT, Domschke K, Falkenstein M, Konrad C. Dissociable influences tests for dichotomous traits in trios and sibships. Genet Epidemiol 2009; 33: of NR2B-receptor related neural transmission on functions of distinct associative 691--699. basal ganglia circuits. Neuroimage 2010; 52: 309--315. 55 Zandi PP, Belmonte PL, Willour VL, Goes FS, Badner JA, Simpson SG et al. 68 Gilad GM, Gilad VH. Overview of the brain polyamine-stress-response: regulation, Association study of Wnt signaling pathway genes in bipolar disorder. Arch Gen development, and modulation by lithium and role in cell survival. Cell Mol Psychiatry 2008; 65: 785--793. Neurobiol 2003; 23: 637--649. 56 Zou GY. On the estimation of additive interaction by use of the four-by-two table 69 Guipponi M, Deutsch S, Kohler K, Perroud N, Le Gal F, Vessaz M et al. Genetic and and beyond. Am J Epidemiol 2008; 168: 212--224. epigenetic analysis of SSAT gene dysregulation in suicidal behavior. Am J Med 57 Stephens M, Smith NJ, Donnelly P. A new statistical method for Genet B Semin Med Genet 2009; 150B: 799--807. haplotype reconstruction from population data. Am J Hum Genet 2001; 68: 70 Owen JM. Transdiagnostic cognitive processes in high trait anger. Clin Psychol Rev 978--989. 2011; 31: 193--202. 58 Storey JD, Tibshirani R. Statistical significance for genomewide studies. Proc Natl 71 Fulwiler CE, King JA, Zhang N. Amygdala-orbitofrontal resting-state functional Acad Sci USA 2003; 100: 9440--9445. connectivity is associated with trait anger. NeuroReport 2012; 23: 606--610. 59 Gauderman WJ. Sample size requirements for matched case-control studies of 72 Turecki G. Dissecting the suicide phenotype: the role of impulsive-aggressive gene-environment interaction. Stat Med 2002; 21: 35--50. behaviours. J Psychiatr Neurosci 2005; 30: 398--408. 60 Kuzmiski JB, Marty V, Baimoukhametova DV, Bains JS. Stress-induced priming of 73 Heinz AJ, Beck A, Meyer-Lindenberg A, Sterzer P, Heinz A. Cognitive and glutamate synapses unmasks associative short-term plasticity. Nat Neurosci 2010; neurobiological mechanisms of alcohol-related aggression. Nat Rev Neurosci 2011; 13: 1257--1264. 12: 400--413. 61 Herman JP, Mueller NK, Figueiredo H. Role of GABA and glutamate circuitry in 74 Ron D, Wang J. The NMDA receptor and alcohol addiction. In: Van Dongen A (ed). hypothalamo-pituitary-adrenocortical stress integration. Ann NY Acad Sci 2004; Biology of the NMDA Receptor. CRC Press: Boca Raton (FL), 2009 [cited 28 January 1018: 35--45. 2012]. 62 Ziegler DR, Cullinan WE, Herman JP. Organization and regulation of paraven- 75 Wills TA, Klug JR, Silberman Y, Baucum AJ, Weitlauf C, Colbran RJ et al. GluN2B tricular nucleus glutamate signaling systems: N-methyl-D-aspartate receptors. subunit deletion reveals key role in acute and chronic ethanol sensitivity of J Comp Neurol 2005; 484: 43--56. glutamate synapses in bed nucleus of the stria terminalis. Proc Natl Acad Sci USA 63 Mann JJ, Currier D. A review of prospective studies of biologic predictors of 2012; 109: E278--E287. suicidal behavior in mood disorders. Arch Suicide Res 2007; 11:3--16. 76 Bevilacqua L, Carli V, Sarchiapone M, George DK, Goldman D, Roy A et al. 64 Endele S, Rosenberger G, Geider K, Popp B, Tamer C, Stefanova I et al. Interaction between FKBP5 and childhood trauma and risk of aggressive Mutations in GRIN2A and GRIN2B encoding regulatory subunits of NMDA behavior. Arch Gen Psych 2012; 69: 62--70. receptors cause variable neurodevelopmental phenotypes. Nat Genet 2010; 42: 77 Botto LD, Khoury MJ. Commentary: facing the challenge of gene-environment 1021--1026. interaction: the two-by-four table and beyond. Am J Epidemiol 2001; 153: 1016--1020.

Supplementary Information accompanies the paper on the Molecular Psychiatry website (http://www.nature.com/mp)

Molecular Psychiatry (2013), 985 --992 & 2013 Macmillan Publishers Limited