Association of serotonin system-related genes with homicidal behavior and criminal aggression in a prison population of Pakistani Origin

The serotonin transporter (SLC6A4), 5-HT2A (HTR2A) and 5-HT2B (HTR2B) recepter genes, express proteins that are important regulators of serotonin reuptake and signaling, and thereby may contribute to the pathogenesis of aggressive criminal behavior. 370 sentenced murderers in Pakistani prisons and 359 men without any history of violence or criminal delinquency were genotyped for six candidate polymorphisms in SLC6A4, HTR2A and HTR2B genes. An association of higher expressing L/L and LA/LA variants of the 5-HTTLPR polymorphism was observed with homicidal behavior (bi-allelic: OR = 1.29, p = 0.016, tri-allelic: OR = 1.32, p = 0.015) and in the murderer group only with response to verbal abuse (OR = 2.11, p = 0.015), but not with other measures of self-reported aggression. L/L and LA/LA genotypes of the 5-HTTLPR polymorphism were associated with higher aggression scores on STAX1 scale of aggression compared to lower expressing genotypes (S/S, S/LG, LG/LG) in prison inmates. No associations were apparent for other serotonergic gene polymorphisms analyzed. Using the Braineac and GTEx databases, we demonstrated significant eQTL based functional effects for rs25531 in HTTLPR and other serotonergic polymorphisms analyzed in different brain regions and peripheral tissues. In conclusion, these findings implicate SLC6A4* HTTLPR as a major genetic determinant associated with criminal aggression. Future studies are needed to replicate this finding and establish the biologic intermediate phenotypes mediating this relationship.

concentration in the cerebrospinal fluid, was found in murderers and arsonists and predicts recidivism, and other studies report severity of aggressive behaviors correlate with neuroendocrine evidence of low serotonin function 16 . Moreover, 5-HT 2A antagonists may reduce aggression indicating a role for serotonin 17 . Therefore, genes of serotonin system are candidate genetic determinants of antisocial behavior including aggression and criminal aggression [18][19][20] . Surprisingly few studies have examined the role of serotonergic gene variants in aggression or criminal aggression including homicide [21][22][23] .
The serotonin transporter gene (SLC6A4 or 5-HTT) has several polymorphisms associated with differential expression of serotonin transporter. 5-HTTLPR (serotonin-transporter-linked polymorphic region), was originally reported as a bi-allelic 44 bp insertion/deletion polymorphism in 5′ regulatory region of SLC6A4, with a long (L) 5-HTTLPR allele that has a relatively higher gene expression of serotonin transporter than the short (S) allele [24][25][26] . Additionally, rs25531 in 5-HTTLPR differentiates the L allele into L A and L G , where L A mediates a relatively higher transcriptional efficiency of SLC6A4 as compared with alleles L G and S 27,28 . The 5-HTTLPR polymorphism has been associated with a range of psychopathological conditions including aggression and violent behavior, but with inconsistent results and no clear conclusions as to what allelic combination confers the increased risk 29 .
Serotonin Transporter Intronic VNTR Enhancer (STin2) is a functional 17 bp VNTR polymorphism of human SLC6A4 gene, and consists of 9, 10 or 12 repeat alleles 30,31 , and the 12 repeat allele has greater gene expression of SLC6A4 32 . Similarly, rs1042173 is another functional polymorphism, located in 3′ noncoding region of the SLC6A4 gene, and may influence the mRNA expression of serotonin transporter 33,34 . Studies of both of these SLC6A4 polymorphisms report inconsistent results for their association with aggression 35,36 and different psychiatric disorders 37,38 .
The HTR2A receptor is one of the most abundantly expressed serotonin receptors in the brain 39 , and is located mainly on post-synaptic neurons 40 . The rs6311 single nucleotide polymorphism in the HTR2A promoter affects expression 41 . Most studies investigating rs6311 failed to find any significant association with psychiatric traits such as aggression 42 , suicidal behavior 43 or schizophrenia 44 although there are exceptions 45 . Similarly, rs17440378 is an intronic SNP in HTR2B gene, another important serotonin receptor in the brain, and was recently reported to be a significant genetic determinant of aggression in cannabis-exposed subjects from a Caucasian GWAS study with functional annotations 46 .
Because the most extreme phenotype may be the most informative in terms of genetic associations, the present study investigated the association with homicidal behavior and self-reported aggression of six different polymorphisms in these three serotonergic system-related genes in sentenced prisoners of Pakistani origin.

Results
The basic characteristics of the study sample set (including demographic details and parameters pertinent to measures of self-reported aggression) are as described previously 47 and summarized in Supplementary Table S1. Only male prison inmates were included in this study. The mean age of the prison inmates was 36.4 ± 11.8 years, most had a middle class social status (≈72%), a high school education (≈59%) and 56% had been married. The prison inmates and controls were matched demographically [all males and no statistical differences with respect to age (p = 0.177), district of origin (p = 0.456), socio-economic status (p = 0.677) and education (p = 0.296).
The primary information (polymorphism, region, type, observed minor allele frequencies, HWE statistics and percentage of samples typed) for serotonergic genes polymorphisms are described in Supplementary Table S2. The allele and genotype frequencies for all studied serotonergic genes polymorphisms did not deviate from Hardy-Weinberg equilibrium in the control group (Supplementary Table S2).
A comparative analysis of allelic and genotypic frequencies in prison inmates and controls for serotonergic gene polymorphisms was done (Table 1). The higher expressing L allele and L A /L A genotype of bi-allelic 5-HTTLPR polymorphism and tri-allelic locus of rs25531 in 5-HTTLPR, respectively, were more frequent in prison inmates compared with controls [L vs. S: OR = 1.30, p = 0.017 and L A /L A : OR = 1.88, p = 0.023 (Table 1)], however, the difference became insignificant after correction for multiple testing. No significant group differences were observed for the allele and genotype distributions of the other serotonergic gene polymorphisms in prison inmates and controls.
The association between polymorphisms in serotonergic genes and history of homicidal behavior was also tested assuming different genetic models (Table 2). There was an association of the bi-allelic and tri-allelic (rs25531) 5-HTTLPR polymorphisms with homicide (OR = 1.29, p = 0.016 and OR = 1.32, p = 0.015) in a logadditive model. However, no association was apparent for other polymorphisms of serotonin-related genes and homicidal behavior under all genetic models analyzed. In addition, pair-wise linkage disequilibrium (LD) and haplotype plot structure analysis for serotonergic polymorphisms demonstrated no significant D́ measures between each pair of serotonergic loci analyzed except for bi-allelic and tri-allelic 5-HTTLPR polymorphisms in SLC6A4 gene (Supplementary Figure S1). The lack of a broader association of serotonergic polymorphisms with criminal aggression was confirmed by a combined genotype analysis considering all the serotonergic polymorphisms analyzed in SLC6A4, HTR2A and HTR2B genes. However, when haplotype association analysis was performed considering the strong LD between bi-allelic and tri-allelic 5-HTTLPR polymorphisms in SLC6A4 gene only according to LD block structure, the L-L A haplotype of bi-allelic and tri-allelic 5-HTTLPR polymorphisms was found to be associated with 1.35 fold increased risk of homicidal behavior (OR = 1.35; p = 0.011) (Supplementary Table S3).
The serotonergic gene polymorphisms were also analyzed by stratification of the prison inmates' data into subgroups manifesting self-reported aggression and histories such as childhood history of abuse, history of parental aggression, parental marital problems, any minor psychiatric problems and substance use disorder (Table 3 and  Supplementary Tables S4 A-E). The L/L genotype of 5-HTTLPR was more prevalent in prison inmates reporting more susceptibility to provocation by verbal abuse (OR = 2.11, p = 0.015) ( Table 3). Also, tri-allelic rs25531 in 5-HTTLPR polymorphism was associated with history of any minor psychiatric problems (OR = 0.26, p = 0.045) and HTR2B rs17440378 polymorphism with parental marital problems (OR = 1.76, p = 0.037) (Supplementary Tables S4D and E). However, no associations were found for other serotonergic polymorphisms in the stratified data analyses. The prison inmate group was further analyzed for potential moderation of genetic susceptibility of criminal behavior by environmental risk factors including current age, age at time of committing murder, region, socio-economic, educational and marital status, and ethnic subgroup or caste. However, no effect of these factors was apparent for any of the serotonergic polymorphisms analyzed (Supplementary Tables S5 A-F).
State anger score was also determined for prison inmates sentenced for murder (using STAXI subscale, fifteen items) and was correlated with different genotypes of serotonin system genes ( Table 4). Prison inmates carrying L and L A alleles of 5-HTTLPR and tri-allelic rs25531 5-HTTLPR polymorphisms, had higher aggression scores compared with inmates with other (S and L G ) alleles (p = 0.0005 for both polymorphisms). However, mean aggression scores did not differ significantly between genotypes for other serotonergic gene polymorphisms.
A gene-gene interaction analysis was also run using the MDR method for the five polymorphisms of three serotonergic system genes analyzed including the tri-allelic rs25531 in 5-HTTLPR, 3′UTR rs1042173 and STin2 polymorphisms from the SLC6A4 gene, rs6311 and rs17440378 SNPs from the HTR2A and HTR2B genes, respectively, with murderer/ normal control status as response. Permutation tests failed to validate any significant 2nd, 3rd or 4th order interactions among these five serotonergic system genetic polymorphisms (all p > 0.15), suggesting no significant G x G interaction effects for SLC6A4, HTR2A and HTR2B genes in the present study. Table 1. Allele and genotype distribution for 5-HTTLPR, rs25531 in 5-HTTLPR, STin2, 3ʹUTR (SLC6A4 gene), rs6311 (HTR2A gene) and rs17440378 (HTR2B gene) polymorphisms and their association with homicide history. OR, odds ratio; CI, confidence interval; n (%), frequency. 1 Bonferroni correction for multiple testing was applied (p-value threshold 0.0083). Statistically significant p-values (< 0.01) and associated OR values are highlighted in bold.
Using the GTEx database, we observed that no eQTL data was available for the tri-allelic rs25531 in 5-HTTLPR polymorphism while rs1042173, rs6311 and rs17440378 SNPs were significantly associated with expression of SLC6A4, HTR2A and HTR2B genes, respectively, in some peripheral human tissues but not in different regions of the brain except for few suggestive associations that included HTR2A*rs6311 in caudate/basal ganglia region (p = 0.05) and HTR2B*rs17440378 in cerebellum (p = 2.9 × 10 -3 ) as well as in spinal cord/cervical c-1 (p = 0.03) regions (Supplementary Tables S6 to S8).

Discussion
In this study, we investigated genetic association of common polymorphic variants in three serotonin system genes (SLC6A4, HTR2A and HTR2B) with homicidal behavior and self-reported aggression in Pakistani prison inmates sentenced for violent murder(s). We found: (1) A genetic association of the bi-allelic and tri-allelic 5-HTTLPR polymorphisms in the SLC6A4 gene with homicidal behavior and criminal aggression but not for other polymorphisms of the three serotonin system genes analyzed; (2) Other measures of self-reported aggression and histories (including childhood history of abuse, history of parental aggression, parental marital problems, and substance abuse/dependence) and environmental risk factors (including current age, age at time Table 2. Association of the studied polymorphisms in serotonergic system genes with criminal delinquency assuming different genetic models. OR, odds ratio; CI, confidence interval; n (%), frequency. 1 Bonferroni correction for multiple testing was applied (p-value threshold 0.0166). Statistically significant p-values (< 0.0166) and associated OR values are highlighted in bold.

Polymorphisms
Model Genotypes Prison inmates, n (%) Normal controls, n (%) OR (95% CI) www.nature.com/scientificreports/ of committing murder, region, socio-economic, educational and marital status and ethnic subgroup) had no significant role in moderating association of serotonin system polymorphisms and homicidal behavior/criminal aggression in the present sample set (except for described associations); (3) L and L A alleles of the 5-HTTLPR polymorphism had a dose dependent relationship with greater aggression score in sentenced murderers; (4) no significant G × G interaction effects for SLC6A4, HTR2A and HTR2B genes in the MDR analysis following the hypothesis that these genes may have a synergistic effect in the development of homicidal behavior/criminal aggression since it is a complex trait, and (5) tri-allelic 5-HTTLPR and other serotonergic polymorphisms as functional eQTLs influencing gene expression in different regions of normal human brain and peripheral tissues as determined by the Braineac and GTEx databases. Table 3. Association of 5-HTTLPRpolymorphisms with self-reported aggression, emotional reactivity, related psychopathologies and childhood adversities in sentenced murderers. OR, odds ratio; CI, confidence interval. Boldface indicates p < 0.05 was considered as statistically significant.
Response to measure 5-HTTLPR genotypes (recessive model)    www.nature.com/scientificreports/ We also genotyped our samples for the newer tri-allelic variant of 5-HTTLPR polymorphism where L allele is further divided into L A (higher transcriptional efficiency of serotonin transporter gene) and L G (lower transcriptional efficiency of serotonin transporter gene comparable to the S allele) variants based on rs25531 SNP within 5-HTTLPR polymorphism. This tri-allelic variant of 5-HTTLPR polymorphism is a better indicator of expression differences due to the 5-HTTLPR polymorphism and thereby superior study power 48 . Therefore, the main hypothesis was based on tri-allelic HTLPR variant in the present study and the statistical analysis reported above for bi-allelic HTLPR polymorphism was included only for comparison with other reports as most previous studies have only analyzed bi-allelic version of 5-HTTLPR polymorphism.

Self-reported aggression (Lifetime)
Efforts were made to analyze those polymorphisms of serotonin system genes that were reported to have functional relevance to serotonin homeostasis. A critical review of the influence of 5-HTTLPR genotypes on SLC6A4 promoter efficiency and expression 49 , reported that apart from a couple of studies 50,51 , most studies find that the longer repeat (L and L A ) variants functionally mediate higher basal and induced transcriptional activity of SLC6A4 promoter resulting in greater gene expression, higher levels of serotonin transporter and an increased serotonin reuptake activity as compared to L G or S variants [24][25][26][27][28][52][53][54][55][56][57][58][59][60][61] . In further support of these findings, less in vivo serotonin transporter mRNA levels are associated with S allele of 5-HTTLPR polymorphism [62][63][64] . STin2 VNTR polymorphism mediates transcriptional regulation of SLC6A4 gene expression as shown in transgenic mice and in vitro functional data from embryonic stem cells 32,65 , where the 12 repeat allele (12R) displayed higher enhancer-like activity and greater expression of serotonin transporter mRNA as compared to 10 repeat allele (10R) 54,66 . However, data regarding the functional relevance of SLC6A4 3′UTR (rs1042173) variant 34,67 and HTR2A rs6311 polymorphism 68,69 are somewhat conflicting.
With the availability of modern genetic variant-gene expression databases such as the Brainneac and GTEx, it is possible to evaluate the link between genetic polymorphisms and altered gene expression which may help in more precise understating of connection between GWAS/CGAS reported genetic polymorphisms and molecular pathogenesis of different diseases. Therefore, as an alternative validation strategy, we performed a follow-up functional annotation analysis of different serotonergic system polymorphisms analyzed. We identified a significant eQTL pair for SLC6A4*rs25531 in hippocampus region of normal human brain providing further evidence of its functionality and candidacy as potential genetic determinant of pathological aggression.
We also observed that rs1042173, rs6311 and rs17440378 SNPs were also significant eQTLs of SLC6A4, HTR2A and HTR2B genes, respectively, modulating their expression in some human brain and peripheral tissues, although no association with criminal aggression was apparent for these polymorphisms in our study. It is important to note here that differences in methylation and transcriptome regulation patterns in different brain regions may result in distinct gene expression profiles 70 . Moreover, previously it has been reported that risk effects of HTR2B rs17440378 polymorphism on aggressive behavior appear to be cannabis use dependent rather than driven by aggression alone or aggression under the influence of other drugs 46 , which may explain no association observed with criminal aggression in the present study using violent murders sample set and a similar explanation may be extendable to other serotonergic polymorphisms analyzed for which no genetic association was observed in the present study. Overall, a functional regulatory role for rs25531 in SLC6A4 gene is suggested that may influence the gene expression and risk of criminal behavior. These data also stress the need of direct genetic variant mediated functional studies in target tissues to confirm any potential effects of serotonergic system polymorphisms on relevant protein function and/or regulation.
Some previous studies have examined serotonergic polymorphisms in prison inmates, however, studies examining extreme violence phenotype such as sentenced murderers are scarce especially from Indo-Pak subcontinent. Previously, Cherepkova and colleagues reported significant over-presentation of 5-HTTLPR and STin2 risk alleles and haplotype in convicted subjects and MMA fighters (including those convicted of grave crimes or murder) as compared to control group 71 . In agreement with our results, the L allele was more frequent in aggressive adolescent prisoners of Korean origin as compared to normal non impulsive population controls in an earlier study 72 . L/L genotype of 5-HTTLPR was found to be associated with aggressive and violent behavior 73 , borderline personality disorders 74 and suicidal behavior 75 . Low heart rate and 5-HTTLPR L/L genotype were associated with higher arrest rates for violence in an incarcerated sample of American men 76 . Other psychopathologies in a violent population, such as depression, were also reported to be associated with both the L allele and L/L genotype of 5-HTTLPR polymorphism in Russian criminal offenders 21 .
Some studies report different results such as an association of the lower expressing serotonin transporter S allele or S/S genotype of 5-HTTLPR polymorphism with extreme criminal behavior in Chinese male prisoners 77 and violent behavior in adult German criminal offenders with history of childhood ADHD 78 . Other studies also reported association of S allele or S/S genotype with antisocial alcoholism 79 , concussion history and personality traits in rugby players 80 , suicidal ideation in acute coronary syndrome patients 81 , higher impulsivity, hostility and neuroticism in anxiety phenotype 82 , hyperactivity-impulsivity in children moderated by peer problems earlier in childhood 83 and increased exposure to life stressors moderated by ADHD symptoms early in life 84 . Some meta-analyses also describe association of low expression allele/genotype of 5-HTTLPR polymorphism with anti-social behavior 85 , violent suicide attempts 86 and bipolar disorder 87 .
Like current study, fewer studies analyzed the tri-allelic variant of rs25531 in5-HTTLPR polymorphism and differentiated between L A and L G alleles while determining association with anti-social behaviors and aggression. The S and L G (low SLC6A4 expression) variants of tri-allelic rs25531 in5-HTTLPR polymorphism were found associated with higher scores on aggression, and total behavior problems in Mongolian children 88 and higher neuroticism in healthy Chinese men 89 .A few studies also reported no association between 5-HTTLPR polymorphism and anti-social phenotypes including antisocial alcoholic behavior 90 and aggressive traits 91 .
The lack of agreement regarding the association of 5-HTTLPR polymorphism with aggressive behavior may be explained by several reasons, including but not limited to: small sample sizes; differences in measures or types of homicidal/criminal behavior and aggression (for example reactive aggression vs. pro-active or pre-meditated www.nature.com/scientificreports/ aggression); differences in genetic background or in allele/genotype frequencies based on ethnic origin; and differences in study design and robustness of data analysis (for example demonstrated conformance to HWE). Also, analysis of limited number of polymorphisms in serotonergic genes has its own limitations as expression of a particular gene can be modulated by numerous such genetic variants which may or may not exhibit linkage disequilibrium. Moreover, which allele of the 5-HTTLPR or tri-allelic rs25531 in 5-HTTLPR polymorphism confers the risk of homicidal behavior and aggression is not clear 29 . Two hypotheses have been proposed in this regard; the L and L A variants of 5-HTTLPR polymorphism (mediating a higher functional expression of serotonin transporter gene) result in greater serotonin reuptake and reduced availability of synaptic serotonin, which may increase susceptibility to proactive aggression and anti-social phenotypes 29,85,92 . In contrast, S and L G alleles of 5-HTTLPR polymorphism (that mediate relatively lower transcriptional efficiency of serotonin transporter gene), due to lower serotonin uptake and higher synaptic serotonin, may result in heightened psychotic or paranoid features and thereby a predisposition towards paranoid reactive aggression phenotypes 29,92 . Genome Wide Association Studies (GWASs) of aggression and criminal aggression are few. Despite some interesting leads highlighting novel loci and pathways in nervous, neuroendocrine and immune systems, GWASs of aggression and related phenotypes (antisocial behavior, anger, conduct problems and callous, uncaring and unemotional aggressive traits) failed to replicate any findings [93][94][95] . A GWAS study of violent prisoners in Finland (n = 5,983), the most relevant phenotype to the current study, found suggestive association with a polymorphism (rs11649622) in the CDH13 gene that codes for a neural adhesion protein 96 . One of the largest aggression GWAS studies (n = 18,988) investigating aggressive behavior in children (the EAGLE consortium) reported significant association with the rs11126630 polymorphism located near the excitatory synapse development gene LRRTM4, and using a subsequent "classical candidate gene association analysis" also with AVPR1A (arginine-vasopressin receptor 1A) gene 97 . Other GWASs report associations with: (a) HTR2B serotonin receptor in cannabis-related aggression 98 ; (b) 13 different SNPs associated with violence in schizophrenia 99 ; and (c) dopaminergic, glutamatergic and neuroendocrine polygenic risk score (PRS) associated with callous-unemotional traits 100 . We also analyzed the GWAS reported HTR2B rs17440378 SNP in our homicidal prison inmate sample set but failed to detect any significant association with criminal aggression. Finally, a meta-analysis of GWAS data from five different population studies did not find any significant markers for anti-social behavior 101 . As in many other conditions, GWASs of aggression and criminal behavior largely failed to detect associations for candidate genes, including serotonin neurotransmission, that are mostly being investigated in candidate gene association studies 94 . Perhaps with increasing sample sizes and more homogenous and in-depth definition of the aggression and criminal aggression, we may find such associations, like other complex traits for which GWAS has been a successful approach 95 .
Study limitations include small sample size due the rare population being studied because we sought a group with an extremely severe phenotype, limited time allowed for interview to obtain all clinical phenotypic data, and analysis of a limited number of serotonergic polymorphisms in a limited number of genes instead of conducting an unbiased GWAS study. Another limitation of our study is the lack of any direct functional assessment of the effects of analyzed serotonin system polymorphisms on relevant protein function, including regulation. Inclusion of information about epigenetic marks and expression profile would have enriched the data set. Therefore, replication studies are needed in larger well characterized samples using a GWAS approach, genome-wide DNA methylation and RNAseq data on the transcriptome combined with additional data on developmental history that would permit estimation of G × G and G × E interactions in relation to extreme aggressive behavior and related endophenotypes.
In conclusion, the L and L A variants of 5-HTTLPR VNTR polymorphism with demonstrated eQTL based functional significance are associated with homicidal behavior and self-reported aggression in this rare sample set of Pakistani prison inmates sentenced for murder(s). Analysis of genetic risk factors in extreme criminal violence may increase understanding of the neurobiological basis of less severe aggression. Further studies should also determine whether these findings extend to less severe aggression and related anti-social endophenotypes.

Subjects and methods
Ethical approvals, consent and study subjects. The details of consent, subject recruitment and study sample set have been described previously 47 . Briefly, the present study was conducted after obtaining permission and approval from higher police authorities and Institutional Ethics Committee at University of the Punjab, Lahore and all experiments were performed in accordance with latest version of Declaration of Helsinki guidelines. All the study subjects gave written informed consent after the nature of the research procedures had been fully explained to them.
This study comprised 729 subjects, including 370 men in three major district prisons of Punjab, Pakistan that were sentenced for first-degree murder(s), and 359 age and ethnicity matched control men randomly selected from the general population and found to be without any personal or family history of criminal aggressive behavior. Only those prison inmates who had been sentenced for first-degree homicide(s) by a court of law and/ or self-confession of the said crime with alleged/convicted/condemned status, were enrolled in the study. A random sampling procedure was ensured by randomly selecting murderers from different barracks within a prison and recruiting murderers from three different major district prisons (and their sub-jails) in Punjab province of Pakistan. It should be noted that prison inmates, as well as normal controls, who had a DSM-IV diagnosis or family history of any major psychiatric disorder, were excluded from the study. Prison inmates were interviewed to record self-reported aggression and histories based on State Trait Anger Expression Inventory (STAXI), with a cut-off value of > 20 as high aggression score, followed by collection of blood samples for genetic analysis. www.nature.com/scientificreports/ Genotyping of SLC6A4, HTR2A and HTR2B polymorphisms. A modification of standard phenolchloroform extraction method was used for the isolation of genomic DNA from blood samples 102 . All the subjects were genotyped for fourSLC6A4 and one HTR2A polymorphisms as described in Supplementary Table S9. Briefly, specific primers were used to amplify the target region containing the polymorphism of interest in a standard PCR reaction. Amplified DNA fragments were resolved on 2% agarose gels and genotypes were directly scored for polymorphisms with PCR based genotyping approach (5-HTTLPR and STin2 polymorphisms). The HTR2B rs17440378 SNP was genotyped using a custom iPLEX SNP genotyping assay and the MassARRAY system according to the standard protocol 103 . The remaining polymorphisms (tri-allelic rs25531 in 5-HTTLPR and, 3′UTR rs1042173 in SLC6A4 and rs6311 inHTR2A) were genotyped using a PCR-RFLP based method where PCR amplification products were subjected to restriction digestion by relevant restriction enzyme and restriction digestion products were run on 3% agarose gels to determine respective genotypes in each sample. The representative gel electropherograms for each of PCR or PCR-RFLP based genotyping assays have been provided in the Supplementary Figures S3 to S6. The genotypes of all polymorphisms analyzed were scored by two independent researchers to minimize bias.

Statistical analysis.
The study data were coded and analyzed using the Statistical Package for Social Sciences (SPSS) version 20 for Windows and the SNPstats program 104 . Hardy-Weinberg equilibrium (HWE) analysis was performed in the control group by means of a chi-square test using the SNPstats program. Minor allele frequencies in the present sample for polymorphisms analyzed were all 5% or above. Allele and genotype frequencies were compared between prison inmates and controls using chi-square tests. Odds ratios (ORs) with 95% confidence intervals (CIs) and associated p-value were calculated as a measure of association. SNPstats was also used to determine genetic associations assuming the dominant, recessive and log-additive models by unconditional logistic regression analysis. Combined genotype and haplotype frequencies for SLC6A4, HTR2A and HTR2B polymorphisms were also analyzed and compared. The pairwise linkage disequilibrium analysis for serotonergic gene polymorphisms was conducted using the Haploview program 105 . Bonferroni correction for multiple testing was performed when determining genotype, combined genotype and haplotype associations with murderer phenotype and aggression scores. A p-value (two tailed) of less than 0.05 was considered significant unless otherwise stated. Gene-gene interactions were also tested using the Multifactor Dimensionality Reduction (MDR) method 106 . Briefly, this method aims to reduce the number of cells in 2nd and higher order interactions by grouping them based on their level of risk. In case such as this study where there is a binary outcome, multi-loci genotypes are grouped into low and high risk genotypes, then each is separately tested against other genotype combinations using a Wald test. Significant Wald statistics are then validated using permutation tests. We used the "mbmdr" library in R 107 , with a liberal p = 0.10 cutoff for the risk classification selecting the low and high risk genotype combinations, and p = 0.05 for the permutation test. Five polymorphisms were entered into the analysis: the tri-allelic rs25531 in 5-HTTLPR, 3′UTR rs1042173 and STin2 polymorphisms from the SLC6A4 gene, rs6311 SNP from the HTR2A gene, and rs17440378 polymorphism from the HTR2B gene, and 2nd, 3rd and 4th order interactions were tested. Only subjects with data for all five of the above polymorphisms were included in the analysis (370 sentenced murderers and 359 normal controls).

Expression quantitative trait loci (e-QTL) analysis.
To evaluate the association between SNP based genetic markers of serotonin system genes and their relative gene expression, cis-eQTL data were extracted using two modern databases including the Braineac database from the UK Brain Expression Consortium (UKBEC) available at www. brain eac. org, and Genotype-Tissue Expression (GTEx) project dataset version 8 (https:// www. gtexp ortal. org/). The Braineac dataset harbors eQTL data of 10 brain regions from 134 postmortem neuropathologically normal individuals of European origin. While the GTEx database is a comprehensive public resource that contains genotypic and tissue specific gene expression data of samples collected from 54 non-diseased tissue types (including brain tissue) across almost 1000 individuals of primarily Caucasian ancestry.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. www.nature.com/scientificreports/