MAOA variants differ in oscillatory EEG & ECG activities in response to aggression-inducing stimuli

Among the genetic variations in the monoamine oxidase A (MAOA) gene, upstream variable number tandem repeats (uVNTRs) of the promoter have been associated with individual differences in human physiology and aggressive behaviour. However, the evidence for a molecular or neural link between MAOA uVNTRs and aggression remains ambiguous. Additionally, the use of inconsistent promoter constructs in previous studies has added to the confusion. Therefore, it is necessary to demonstrate the genetic function of MAOA uVNTR and its effects on multiple aspects of aggression. Here, we identified three MAOA alleles in Koreans: the predominant 3.5R and 4.5R alleles, as well as the rare 2.5R allele. There was a minor difference in transcriptional efficiency between the 3.5R and 4.5R alleles, with the greatest value for the 2.5R allele, in contrast to existing research. Psychological indices of aggression did not differ among MAOA genotypes. However, our electroencephalogram and electrocardiogram results obtained under aggression-related stimulation revealed oscillatory changes as novel phenotypes that vary with the MAOA genotype. In particular, we observed prominent changes in frontal γ power and heart rate in 4.5R carriers of men. Our findings provide genetic insights into MAOA function and offer a neurobiological basis for various socio-emotional mechanisms in healthy individuals.

www.nature.com/scientificreports www.nature.com/scientificreports/ efficiency of monoamine oxidase A was inconsistent and depended on the transfected constructs 35 . These findings suggest that factors such as post-translational factors or the modulation of protein levels and activity, and possibly compensatory mechanisms between the different neurotransmitter systems regulated by MAOA, may work together in modulating the role of MAOA uVNTR genotype on aggression.
While studies using positron emission tomography (PET) and [ 11 C]clorgyline did not find an association between MAOA genotype and brain MAOA activity in healthy men 32,33 , studies using functional magnetic resonance imaging (fMRI) support an effect of the MAOA genotype on aggression-related functional changes in the nervous system. The MAOA genotype has been associated with the level or distribution of neural activity in cortical regions of the brain, such as the dorsal anterior cingulate cortex or the ventromedial prefrontal cortex, in response to Go/No-Go tasks or programmes including social exclusion 36,37 . In addition, heathy L-MAOA carriers showed prominent structural and functional changes in the corticolimbic network, which plays a crucial role in regulating emotional states during the recall of negative events 17 . Recent work showed that L-MAOA carriers are more susceptible to pharmacological regulation by serotonin, with changes in the corticolimbic network, indicating a possible link between the serotoninergic system and MAOA genotype 38 . The effects of MAOA genotype on the dopaminergic system remain controversial. For example, the dopamine levels in response to a violent video was not dependent on MAOA genotype 39 . As previously indicated, the conflicting results obtained by different studies could be due to the broad role of MAOA in modulating different neurotransmitter systems, the heterogeneity in measuring neural function and human behaviour, the in vitro system used to measure the effect of uVNTR on MAOA expression, and to the effect of gene × environmental interactions [40][41][42][43] . The results of such studies in women are also perplexing. It remains unclear whether MAOA is an X-inactivated gene, making it challenging to evaluate enzyme expression or activity in heterozygous genotypes 44 .
Beyond the conflicting results, historical evidence points towards an association of MAOA genotype to neural function in aggressive behaviour. Thus, in the present study, we investigated the effects of the MAOA genotype on multiple aspects of aggression, including neural and cardiac activities. In addition, given the fact that allelic distribution varies across ethnic/racial groups, we characterized the MAOA genotype distribution in the Korean population for the first time 22 . Transcriptional efficiency was assessed using in vitro reporter gene assays based on full MAOA promoter sequences to more accurately test the transcriptional contribution of the uVNTR. We also investigated the functional relationship between MAOA genotypes and aggression by analysing electroencephalogram (EEG)-and electrocardiogram (ECG)-based neurobiological responses to commonly encountered aggression stimuli in both men and women. In particular, EEG and ECG are useful tools for studying differences in neural activity because they are capable of measuring real-time responses and are highly accessible, with fine temporal resolution across the spectrum range of EEG and the direct autonomic reactivity range of ECG. Moreover, many EEG and ECG studies have shown that specific forms of oscillatory activity are associated with antisocial behaviour or aggressive tendencies [45][46][47][48][49][50][51][52][53][54][55][56] . In particular, frontal EEG asymmetry, which is a representative functional feature of aggression 45,46 , is affected by dopamine regulation 57 . Nevertheless, few studies of the relationship between MAOA genotypes and aggression have used EEG and ECG approaches in a pool of healthy adults.

Results
Genetic variation of MAOA in the Korean population. To characterize genetic variation of MAOA in the Korean population, we identified MAOA uVNTR genotypes and allelic frequencies. Although many studies have followed the original method, which classifies alleles as 2R, 3R, 4R, and 5R 22,23,58 , some groups have suggested a new classification method for the sequence repeats, namely, 2.5R, 3.5R, 4.5R, and 5.5R [59][60][61][62] . The difference lies in the first 15 bp half-repeat sequence (−1141/−1127 bp), which is next to the repeated 30 bp sequence (−1262/−1142 bp) 63 ; the redefined classification includes this sequence (Fig. 1a). Seven different MAOA genotypes were detected in our population, including two genotypes in men and five genotypes in women ( Table 1). The polymerase chain reaction (PCR) products of the seven observed types of MAOA uVNTRs are shown in Figs 1b and S7a. Among the three MAOA uVNTR alleles detected, the 3.5R allele was the most prevalent ( Table 1). The 2.5R allele appeared only in heterozygous genotypes, such as 2.5R/3.5R and 2.5R/4.5R in women, and at a very low frequency. The genotype distribution in our study was not in Hardy-Weinberg equilibrium (χ 2 = 15.59, df = 4, χ 2 /df = 3.90, p (χ 2 > 15.59) = 0.0036).
Functional analysis of the promoter activity of MAOA uVNTR alleles. To characterize MAOA uVNTR function in vitro, the transcriptional efficiency of the observed alleles was assessed in SH-SY5Y and JAR cells using reporter gene constructs containing the MAOA promoter region fused to the firefly luciferase gene. For each promoter construct, we used a full 1.47 kb upstream fragment from −1470 to +34 bp 63 , which encompasses three different MAOA uVNTR alleles (2.5R, 3.5R, and 4.5R) containing transcription-regulatory regions of MAOA (Fig. 2a). Reporter gene expression levels gradually decreased with each allele in the order of 2.5R, 3.5R, and 4.5R in both SH-SY5Y and JAR cells (Fig. 2b,c). Statistically significant differences were detected between the negative control condition and each allele (F, 65.41; p < 0.0001 for 2.5R; p, 0.0002 for 3.5R; and p, 0.0261 for 4.5R in SH-SY5Y cells; F, 20.14; p < 0.0001 for 2.5R; p, 0.0001 for 3.5R; p, 0.0239 for 4.5R in JAR cells). In a comparison among three alleles, 3.5R showed significantly higher MAOA transcriptional efficiency than 4.5R (F, 65.41, p, 0.0365), and the 2.5R allele showed the highest value, with statistical significance in SH-SY5Y cells (F, 65.41, p < 0.0001). There was also a significant difference in MAOA transcriptional efficiency between the 2.5R and 4.5R alleles in JAR cells (F, 20.14, p, 0.0003). Since the transcriptional efficiency decreased as the MAOA uVNTR sequence increased in length, we performed additional reporter gene assays with nine alleles spanning 2R-6R to determine the length-dependent effect of MAOA uVNTR on transcription. The PCR products of the promoters carrying each of the MAOA uVNTR allele-fused constructs, along with the corresponding observed MAOA genotypes, are shown (Figs S1a,b and S7b,c). Although the range of absolute values differed, the pattern of transcriptional efficiency for the 2.5R, 3.5R, and 4.5R alleles was preserved (F, 191.8; p < 0.0001 between 2.5R www.nature.com/scientificreports www.nature.com/scientificreports/ and 3.5R; p < 0.0001 between 2.5R and 4.5R; and p < 0.0351 between 3.5R and 4.5R in SH-SY5Y cells; F, 57.74; p < 0.0001 between 2.5R and 3.5R; p, 0.0345 between 2.5R and 4.5R; and p, ns between 3.5R and 4.5R in JAR cells); in addition, the 5.5R allele, which was found in other ethnic/racial groups 20,22 , showed a relatively higher value than those of the 3.5R and 4.5R alleles (Fig. S1c-f). However, no length-dependent effect of the MAOA uVNTR on transcription was observed. Nevertheless, our results showed higher transcriptional efficiency for the shorter uVNTR variants, 2.5R and 3.5R, that was more evident in SH-SY5Y cell line as compared to JAR. Due to the low frequency of the 2.5R allele in the population, we focused the next experiments on subject carrying the 3.5R and 4.5R alleles.
Effects of MAOA genotypes on questionnaire scores. To test the relationship between MAOA genotypes and psychological aggression and to compare hemi-and homozygous 3.5R and 4.5R carriers, the scores of three self-report questionnaires (the Buss-Durkee Hostility Inventory (BDHI), the Buss-Perry Aggression  www.nature.com/scientificreports www.nature.com/scientificreports/ Questionnaire (BPAQ), and the Peer Conflict Scale (PCS)) were analysed. No significant differences in the mean scores of either the 3.5R/Y and 4.5R/Y genotypes in men or the 3.5R/3.5R and 4.5R/4.5R genotypes in women were found on any of the questionnaires (Fig. S2). Similarly, no significant differences in the mean BPAQ and PCS subcategory scores were observed between genotypes. Therefore, the effect of the MAOA genotype on psychological aggression was negligible in our study population.
Effects of MAOA genotypes on EEG response. We analysed EEG-based neurobiological responses to aggression-inducing stimuli to examine associations between MAOA allelic variation and neural or functional differences. When aggression-inducing stimuli were applied through the F8 channel, the power spectral density was displayed as a heat map according to the frequency interval (Fig. 3a). We assessed significant difference between the genotypes during S only for the EEG values with significant differences between NS and S in the analysis. Mean square error (MSE) analysis, which indicates the oscillatory variability of the time (neutral stimulus (NS) and stimulus (S)) and frequency (α, β, γ, θ, and δ waves) intervals, revealed that the MSE value of the γ wave in the F8 channel during S was significantly larger than that during NS in both men and women, except for the 3.5R/3.5R group (t, 2.425, p, 0.0416 for 3.5R/Y; t, 6.125, p < 0.0001 for 4.5R/Y; t, 3.692, p, 0.018 for 4.5R/4.5R; and t, 3.711, p, 0.0017 for heterozygous group) (Fig. 3b,c). Moreover, the MSE value of the γ wave in the F8 channel also differed significantly between MAOA genotypes during S in men; specifically, the MSE values of the 4.5R/Y group were larger than those of the 3.5R/Y group (t, 2.375, p, 0.0408). However, for the relative power value of the γ wave in the F8 channel during S, the values did not differ significantly by MAOA genotype both in men and women (Fig. 3d,e). In addition, for the other four frequency intervals in the F8 channel and the five frequency intervals in the Fp2 channel, the MSE values during S did not show significant differences between genotypes (Figs. S3 and S4). For the relative power value of the five frequency intervals in the Fp2 and F8 channels, significant differences between genotypes were not observed as well (Figs. S3 and S4). Therefore, when aggression-inducing stimuli were applied, the oscillatory change in the γ wave in the F8 channel was greater in hemizygous 4.5R carriers than in hemizygous 3.5R carriers.
Effects of MAOA genotype on ECG responses. We analysed the heart rates (HRs) from the ECG results to investigate autonomic reactivity related to aggression according to the MAOA genotype. Since low HR in the resting state is a marker for aggression 48,64-67 , we assessed significant differences between the genotypes during NS only for the average HR values, with significant differences between NS and S in the analysis. The average HR was significantly lower for 4.5R/Y than for 3.5R/Y during NS (t, 2.604, p, 0.0234), and the values were significantly higher in both genotypes during NS than during S in men (Fig. 4a). However, the average HR was not significantly different either between genotypes or between scenes (NS vs. S) in women (Fig. 4b). MAOA promoter constructs carrying the 2.5R, 3.5R, or 4.5R alleles fused to firefly luciferase were co-transfected with a plasmid containing Renilla luciferase into SH-SY5Y cells and JAR cells. Empty pcDNA vector was used as a negative control (−). Data are presented as the ratio of firefly to Renilla luciferase activity (mean ± SEM) from one of three independent experiments with 4-8 wells. *, **, and *** above the horizontal lines indicate statistically significant differences among MAOA uVNTR alleles at p < 0.05, 0.01, and 0.001, respectively, by one-way ANOVA with Tukey's post hoc test; * and ** above the bar graph indicate statistically significant differences between the negative control group and MAOA uVNTR alleles at p < 0.05 and 0.01, respectively, by one-way ANOVA with Tukey's post hoc test.
www.nature.com/scientificreports www.nature.com/scientificreports/ Since ΔHR and HR variance indicate the degree of change in autonomic reactivity, we assessed significant differences between the genotypes during S only for the values, with significant differences between NS and S in the analysis. ΔHR was significantly higher during S than during NS in all genotypes (t, 4.975 p < 0.0001 for 3.5R/Y; t, 6.929 p < 0.0001 for 4.5R/Y; t, 8.333, p < 0.0001 for 3.5R/3.5R; t, 3.716, p, 0.0020 for 4.5R/4.5R; and t, 7.537, p < 0.0001 in the heterozygous group) (Fig. 4c,d). In particular, ΔHR was significantly higher in the 4.5R/Y group than in the 3.5R/Y group during S in men (t, 2.364, p, 0.0426) (Fig. 4c). However, no such difference in ΔHR between genotypes was observed in women (Fig. 4d). In addition, the variance of HR was not significantly different either between genotypes or between scenes both in men and women (Fig. 4e,f). Taken together, the ΔHR by aggression-inducing stimuli was greater in hemizygous 4.5R carriers than in hemizygous 3.5R carriers, while average HR by neutral stimuli was lower in hemizygous 4.5R carriers than in hemizygous 3.5R carriers in men.

Discussion
MAOA has long been a gene of particular interest for studying the association between genetic variation and behavioural or neurobiological characteristics associated with aggression 16,17,[24][25][26][36][37][38]68 . However, despite numerous publications on this matter, studies are still needed to investigate whether some relationships are present in all racial groups 69 . Our research found that the frequencies of the common 3.5R and 4.5R alleles in Koreans, 63.8% and 34.8%, respectively, differed from the observed frequencies of 35-39% and 59-63% for 3R and 4R, respectively, in Caucasians (assumed to be equivalent to 3.5R and 4.5R in our classification) 22,60,70 . The allele frequency observed here was similar to those previously reported for Chinese (54.5-64.7% and 34.8-44.2% for 3R and 4R, respectively) 35,69,71 and Japanese (62% and 38% for 3R and 4R, respectively) 72 individuals. Although not based on sequence analyses, these data provide a basis for the representative genetic distribution of MAOA uVNTR in Asians. We verified the allele frequency difference in MAOA uVNTR by race and geographic location 22,70,73,74 . Previous studies showed that 2R was rare in Asians 35,71,72 and that 5R was found only in Caucasians 22,60 . In our study, we observed 2.5R at a very low frequency, which corresponds to 2R in previous studies.
The molecular basis of the difference in aggression between MAOA genotypes has largely been pursued in terms of expression levels using in vitro systems. These studies began with Sabol's initial finding in vitro, which provided evidence for classifying MAOA genotypes as H-MAOA and L-MAOA 22 . The results for the most frequent alleles, which demonstrated lower promoter activities for 3R than for 4R, were replicated by others 20,23 , but these studies also showed inconsistent results for the relative promoter activity of the 5R allele, which were not All values are represented as the means ± SEM. The sample size for EEG measurement was 84, including 36 men (21 for 3.5R/Y, 15 for 4.5R/Y) and 48 women (23 for 3.5R/3.5R, 5 for 4.5R/4.5R, and 20 for the heterozygous group). *, **, and *** above the horizontal lines indicate statistically significant differences between MAOA genotypes at p < 0.05, 0.01, and 0.001, respectively, by repeated measures of two-way ANOVA with Bonferroni post hoc test; * and ** above the graph bars indicate statistically significant differences between NS and S at p < 0.05 and 0.01, respectively, by repeated measures of two-way ANOVA with Bonferroni post hoc test.
www.nature.com/scientificreports www.nature.com/scientificreports/ noted due to its rare frequency 20,22 . We found that the differences in promoter activity among the three alleles were inconsistent with previous findings 20,22,23 ; in our results, the difference between the 3.5R and 4.5R alleles was minor and dependent on the cell line used, with the 2.5R allele showing the highest transcriptional efficiency. Although our results are consistent with one previous report, this result was not discussed further due to the use of glioblastoma cell lines rather than neuroblastoma 35  . Additionally, the GR/AR response element (−1350/−1335 bp) and its upstream region in the MAOA promoter affect basal luciferase activity but do not respond to glucocorticoids or androgens 78 . Our results show that, using the full promoter under in vitro conditions, the promoter activity is higher in shorter uVNTR variants, although there is no a length dependent effect for variants ranging from 2.5R up to 6R. This finding suggests that the MAOA uVNTR may not be the only cis-regulatory element modulating MAOA transcriptional efficiency. For instance, epigenetic modification of the genome, such as DNA methylation, can be proposed as a potential mechanism that contributes to MAOA expression 32,33 . Indeed, MAOA promoter methylation in white blood cells has been associated with brain MAOA activity measured by PET. Future studies of MAOA promoter methylation and transcriptional efficiency will provide more reliable results and mechanisms related to the role of the MAOA uVNTR in transcription. (e,f) represent variance in HR during NS and S between MAOA genotypes in men and women, respectively. All values are represented as the means ± SEM. The ECG sample size was 70, including 30 men (18 for 3.5R/Y, 12 for 4.5R/Y) and 40 women, (20 for 3.5R/3.5R, 3 for 4.5R/4.5R, and 17 for the heterozygous group). * and ** above the horizontal line indicate statistically significant differences between MAOA genotypes at p < 0.05 and 0.01, respectively, by repeated measures of two-way ANOVA with Bonferroni post hoc test; *, **, and *** above the graph bar indicate statistically significant differences between NS and S at p < 0.05, 0.01, and 0.001, respectively, by repeated measures of two-way ANOVA with Bonferroni post hoc test.
www.nature.com/scientificreports www.nature.com/scientificreports/ Beyond the ongoing studies to decipher the molecular mechanisms modulating MAOA levels, many studies have demonstrated functional phenotypes associated with H-MAOA and L-MAOA carriers 17,[36][37][38] . We also discovered that EEG and ECG responses related to aggression differed with respect to the MAOA genotype. We found that hemizygous 4.5R carriers showed greater oscillatory variability in the right frontal γ wave in response to aggression-inducing stimuli. The EEG signal in the frontal region has mainly been studied with respect to the asymmetrical activity pattern of the α wave, which is related to specific types of aggression, such as positive or negative affect 45,46,49 . However, no effects were observed for the α wave of the right hemisphere in our results. Therefore, we can raise the possibility of other candidates. In early investigation, enhanced cortical δ wave activity was associated with violence or antisociality 50,80 . Studies of the γ wave in relation to aggression are relatively rare, but some reports suggested that the frontal γ wave shows increased activity following unpleasant stimuli 81 and is also involved in impulse control in relation to addiction 82 . Thus, the greater γ responses of 4.5R carriers in our results might be associated with impulsive tendencies. Even though many neuroimaging studies have also shown that neural activity in the prefrontal/frontal cortex or corticolimbic neural circuits are related to aggressive behaviours, such as impulsivity [83][84][85][86] , this evidence is insufficient to causally link individual brain areas to specific EEG frequencies, including the γ wave. Moreover, variations in the γ wave might be induced during visual processing. Growing evidence indicates that this oscillatory variation can be driven by micro-saccadic movement or cross-frequency coupling of θ waves caused by this movement 87,88 . First, transient increases in γ power due to micro-saccadic movement generally occur in the 200-300 ms range 88 . However, the latency and duration of this effect have differed among some MEG studies [89][90][91] . Second, there is neural communication related to induced γ oscillation in visual cortical areas 87 . However, whether γ synchronization occurs between the visual cortex and the prefrontal/frontal cortex is unclear. Future research on γ dynamics caused by visual cues should be supplemented with eye-tracking systems or correction methods, such as averaging trials. Taken together, our results suggest that the level of oscillatory change in the frontal γ wave under aggression-inducing stimuli can be another functional phenotype that varies with the MAOA genotype in relation to aggression.
Furthermore, we discovered differences in ECG responses to aggression-inducing stimuli between MAOA genotypes. Previous studies proposed that a low HR during resting state is among the best-replicated physiological characteristics of aggression 48,[64][65][66][67] , which may also indicate a fearless temperament 92 . We found a lower average HR under NS in hemizygous 4.5R carriers than in 3.5R carriers in men. Therefore, the physiological phenotypes of HR in 4.5R carrier will often be closely interrelated aggressive tendencies. Meanwhile, in our results, the average HR decreased in response to stimuli in men. Comparisons of average HR in response to emotional stimuli in adults with more and less aggressive tendencies or with and without a history of violent behaviour have revealed inconsistencies across studies 65,80,93,94 . Since HR variabilities are regulated by interactions between the vagal (which reduces HR) and sympathetic (which increases HR) systems 95 , these values have been used to measure emotional regulation capacity during psychological processes 80,96,97 . Then we remain focused on interpreting HR variabilities rather than average HR in the analyses under aggression stimulation. The correlation between HR variabilities and antisocial behaviour is not yet clear 98 ; the possible connection between 4.5R carriers with larger HR variabilities under aggression stimuli and higher aggression is open to debate. Nevertheless, the different HR variability dynamics in MAOA-deficient mice support the feasibility of differences in HR variabilities in humans according to MAOA uVNTR genotype 99 . Therefore, this result implies that HR variabilities in response to aggression stimuli may be a neurobiological phenotype associated with genetic variations in MAOA. Furthermore, the use of human cardiomyocytes which contain MAOA 100,101 has potential for studying enzyme expression and activity in the heart according to MAOA genotype indirectly. Although MAOA studies in human cardiomyocytes have still focused on the role of myocardial regulation of ROS 102,103 , there is evidence that MAOA is involved in the catabolism of norepinephrine in mouse cardiomyocytes 104 . This system can be suitable for predicting intracellular mechanism of MAOA in heart which corresponds to the phenotypes of HR between MAOA genotypes.
As the MAOA gene can be expected to follow the principles of X-linked gene expression, research in women is likely to have multiple outcomes with regard to aggression. Our EEG and ECG results in women with homozygous MAOA genotypes were inconsistent with those in men, as well as statistical significance was not obtained for women because of the low number of homozygous 4.5R carriers. Moreover, our findings showed no significant neurobiological response in women with heterozygous MAOA genotypes, similar to other studies 105,106 . This difference highlights the need for research at the cellular level regarding whether X-inactivation occurs in heterozygous MAOA genotypes to better understand genetic effects on aggression in women.
In this study, we identified new functional phenotypes of oscillatory EEG and ECG activities induced by aggression-related stimuli that vary with MAOA genotypes: greater reactivity of the γ signals and greater HR variabilities in 4.5R carriers than in 3.5R carriers. This result suggests that MAOA does have some genetic effects on neural mechanisms and may extend to emotional control mechanisms related to aggression. We also attempted to provide a basis for the neural mechanism indirectly through transcriptional expression according to MAOA uVNTR alleles via a reporter gene assay in vitro. Our study calls for further investigation into the contribution of uVNTR and additional mechanisms to MAOA transcription. Future studies should address two points. First, we must clarify whether MAOA uVNTR modulates the expression and activity of the enzyme, as well as neurotransmitter levels, neural activity and behaviour. Second, further studies are required to understand the differences in oscillatory activity from cognitive and emotional perspectives. Addressing these points will facilitate the interpretation of various human socio-emotional behaviours, such as depression 107 , anxiety 108 , and prosociality 109 , as well as cognitive abilities such as memory 110,111 , with regards to genetic variations in MAOA.
www.nature.com/scientificreports www.nature.com/scientificreports/ Methods Study subjects. In the present study, 524 college and high school students (260 men and 264 women, age: 18.0 ± 1.4) in Korea were recruited and provided oral epithelial cells for genotyping. All subjects had no record of pathological behaviour and no criminal record. Of the total sample, 334 participants (169 men and 165 women) completed three self-report questionnaires. Of the participants who completed the questionnaires, 94 participants were randomly selected, and their neurobiological signals were measured. Since, the difference in brain wiring between right-handed and left-handed individuals remains unclear [112][113][114][115] , to avoid bias, we decided to select only right-handed individuals as our subjects. We collected valid data from 84 participants (36 men and 48 women) using EEG measurements and from 70 participants (30 men and 40 women) using ECG measurements. EEG samples with poor electrode attachment or subject movement were excluded from the analysis. ECG samples that were not consistent with PQRST waveforms, i.e., the series of the electrical events of heart rhythms (i.e., the P wave represents atrial depolarization, the QRS complex represents ventricular depolarization and contraction, and the T wave represents repolarization of ventricles) 116 were also excluded from the analysis. This study was approved by the Institutional Review Board (IRB) of DGIST [Number: DGIST_170614-HR-009-04]. All participants provided written informed consent after hearing a detailed explanation of the experimental procedures, in accordance with the Declaration of Helsinki. ance with the manufacturer's instructions. In addition, all PCR products were purified using a PureLink ™ Quick Gel Extraction and PCR Purification Combo Kit (Invitrogen, USA). The purified amplicons were split into two aliquots; one was submitted to an automated Sanger' sequencing service (Cosmogenetech, Korea), and the other was separated via 3% agarose gel electrophoresis and visualized to confirm all genotypes twice, especially for heterozygous genotypes. The alleles of the MAOA uVNTR-repeated region were genetically identified based on a redefined genetic characterization method that include copies of the 1R sequence and the first half of 0.5R 22,59-62 . We followed the naming convention of 2.5R, 3.5R, and 4.5R in this study, which is assumed to be equivalent to 2R, 3R, and 4R in other studies 22,23,58 . Construct generation. DNA fragments containing the 2R, 2.5R, 3R, 3.5R, 4R, 4.5R, 5R, 5.5R, and 6R alleles of the MAOA gene promoter were generated by DNA synthesis (Cosmogenetech, Korea) based on the MAOA DNA sequence 63 (GenBank accession number m89636). The alleles were PCR-amplified using two primers containing SacI (−1470, 5′-ACTGGCCGGTACCTGAGCTCCTGCAGCGAGCG-3′) and HindIII (+35, 5′-CCGGATTGCCAAGCTTGCTTTGGCTGAC-3′) restriction sites. The 1442-1577 bp DNA fragments containing the MAOA gene promoter were digested with restriction enzymes and ligated into the pGL4. 10 [luc] vector (Promega, USA). All constructs were verified by sequencing (Cosmogenetech, Korea).

Identification of genotypes.
Cell lines, transfection, and reporter gene assays. We used the brain-derived SH-SY5Y human neuroblastoma cell line and the JAR human placental choriocarcinoma cell line that have been previously used to the study the in vitro functional characteristics of MAOA 22 . SH-SY5Y cells exhibit prominent features of catecholaminergic neurons, in which MAOA is predominantly found 117 , and are used as an in vitro model of the dopaminergic system, which expresses tyrosine hydroxylase and dopamine-beta-hydroxylase, as well as the dopamine transporter 118,119 . SH-SY5Y cells are also driven towards adrenergic phenotypes under certain growth conditions 120,121 . JAR cells are derived from placental trophoblasts, in which MAOA is expressed 122,123 , and are used to study the serotonergic system, especially the serotonin transporter 124,125 . Firefly DNA and control TK DNA (pGL4.74, Promega, USA) were co-transfected into SH-SY5Y (ATCC, CRL2266) and JAR (ATCC, HTB144) cells using the SF Cell Line 4D-Nucleofector ® X Kit (Lonza, Switzerland) using a 4D-Nucleofector system (Lonza, Switzerland) according to the manufacturer's instructions. pcDNA and control TK DNA were co-transfected as negative controls. For each transfection, 2 × 10 6 cells were resuspended in 100 µL of Nucleofector solution, 1 µg of firefly DNA and 1 µg of control DNA was added in a single Nucleocuvette TM , and the electroporation settings recommended for the cell line by 4D-Nucleofector system were applied (Pulse code: CM137). Transfected cells were plated at a density of 1 × 10 4 cells/well in 96-well plates with culture medium and were cultured at 37 °C and 5% CO 2 for 48 h and then lysed with 50 µL of PLB solution in each well following the instructions of the Dual-Luciferase Reporter Assay System (Promega, USA). Next, 100 µL of LAR II solution was added to 20 µL of transferred PLB lysate in a luminometer plate, and firefly luciferase activity was measured. After reading, 100 µL of STOP & Glo reagent was added, and Renilla luciferase activity was measured. Luminescence was measured on a Spark 10 M (Tecan, Switzerland), and the results are shown as the ratio of firefly luciferase to Renilla luciferase. Tests were repeated in 6-12 wells using the transfected cells, and two or three independent experiments were performed.
Self-report questionnaires. The following three self-report questionnaires on aggression were used in this study: K-BDHI 126 , adapted from the Buss-Durkee Hostility Inventory (BDHI) 127 ; K-BPAQ 128 , adapted from the Buss-Perry Aggression Questionnaire (BPAQ) 129 , which includes plan failure scenarios 130,131 ; and K-PCS 132 , adapted from the Peer Conflict Scale (PCS) 133 . The peer-to-peer conflict scales for adolescents were classified into four factors: overt/relational and proactive/reactive. All questionnaires were validated in the Korean population 134-136 . www.nature.com/scientificreports www.nature.com/scientificreports/ Stimuli and procedure. Experiments were conducted in a soundproof room maintained at 25 °C and constant humidity, with one subject and one experimenter in pairs. Neurobiological signals were measured while the subject watched a 118-s-long video that consisted of 'fixation cross (4 s) -black -neutral scene (5 s) -black -scene 1 (39 s) -black -scene 2 (32 s) -black -scene 3 (11 s) -end' (Fig. S5). We used a fixation cross set against a white background, as well as four black scenes between the other scenes, as the baseline. We used a scene of an empty school classroom, which is a common sight in the daily lives of the subjects, as the neutral stimulus (NS). Three scenes (scene 1, scene 2, and scene 3) related to aggression were used as the stimulus (S). Scene 1 showed an abusive argument between a driver and bossy passenger on a bus, as verbal abuse is reported to be related to verbal aggression 137 . Scene 2 showed a chat screen of peer conflict, in which one student is bullied and burdened with all of the assignment. Bullying has been reported to trigger proactive and reactive aggression in adolescents 138,139 . Scene 3 showed a person scratching a blackboard with their nails, which has been reported to lead to irritability 140,141 . Data acquisition, processing, and analysis. Neurobiological signals were measured using an MP36 (BIOPAC, CA, USA) with a 4-channel system (sampling rate = 1000 Hz, bandpass filter = 0.5-100 Hz), which is useful for quick and simple measurements in large populations (sampling rate = 1000 Hz, bandpass filter = 0.5-100 Hz). In studies analysing the neurobiological responses of aggression using EEG, frontal α asymmetry is a well-known signal pattern for aggression 45,46 . Reportedly, relatively greater left frontal α asymmetry at resting state is an indicator of approach motivation, while greater right asymmetry is an indicator of avoidance motivation 45,142 . Similar results have also been obtained for the β signal 49 . Our subjects, Koreans, tend to show indirect emotional expressions such as avoidance of irritability compared with Westerners in several cross-cultural studies 143 . Additionally, studies of EEG responses to neutral, happy, and fear-related sounds in adult Korean men showed differences in α, β, and γ power values between emotional stimuli, especially in the right frontal region (Fp2 and F8 channels), compared with the left frontal region (F3, Fp1, F7) 144 . Therefore, we decided to analyse individual differences in the EEG signals concentrated on the right frontal region, Fp2 and F8, rather than analysing the asymmetric difference between the left and right frontal regions. Moreover, in control experiments, Fp2 and F8 showed prominent responses to the stimuli, while the differences in responses from P7 and P8 on stimulation were insignificant (Fig. S6). These two channels are not only accessible but are also related to aggression 46 . In our study, EEG was recorded through two electrodes on Fp2 (right side of prefrontal cortex) and F8 (right side of dorsolateral prefrontal cortex), with ground and reference electrodes on the neck, in accordance with the BIOPAC EEG electrode placement guidelines and the International 10-20 EEG system [145][146][147] . For EOG (Electrooculogram), an active electrode was placed above the left eye, with a ground electrode on the mid-forehead and a reference electrode on the left cheek bone for removal of artefacts from eye-movement or gross muscle movement. These artefacts were removed by independent component analysis using BIOPAC AcqKnowledge program version 4.2. For ECG, an active electrode was place on the right wrist, with a ground electrode on the right foot and a reference electrode on the left wrist, following Einthoven's triangle. For data analysis, we targeted all frequency intervals, as many studies have shown that α, β, γ, and θ signals in the frontal regions are related to aggression 45,46,49,82,148 . For the signals from the Fp2 and F8 channels of each subject, discrete Fourier transforms were performed using the fast Fourier transform algorithm in MATLAB. Power spectral density was calculated in the area of each frequency interval (0.5 and 4 (δ), 4 and 8 (θ), 8 and 13 (α), 13 and 30 (β), and 30 and 90 (γ)) by the trapezoidal method. The mean value of the power spectral density of each frequency interval during neutral stimulus or stimulus was normalized to the mean value at baseline and was then expressed as the relative power value. We performed a 2000-point fast Fourier transform by dividing the data into 0.25 s intervals to trace dynamic changes in EEG over time. The values from each participant were normalized to area of the whole frequency domain (0-90 Hz). The mean values from all participants were then visualized by heat-map analysis. Apart from mean values, we calculated the power spectral density of each participant 149 . The average squared difference between the power spectral density for all 0.25 s intervals and measured values for each 0.25 s interval during NS or S at frequency intervals were presented as MSE. The ECG signal was corrected by detrending using the method of least squares, and R-R intervals (unit: ms) were obtained by calculating the local maximum value per heartbeat period. These were converted to HR values (unit: BPM) by 60000/(R-R intervals) 150 . The standard deviation of HR was calculated as the variance of HR, and the difference between the maximum and minimum values was calculated as ΔHR. The level of HR variabilities was then analysed using the variance of HR and ΔHR. Statistical analysis. All statistical analyses were performed using Prism 8.0. One-way ANOVA with Tukey's post hoc test was used to identify statistically significant differences in transcriptional efficiency values between allelic groups or between allelic groups and the negative control group. Since MAOA is located on the X chromosome, the results from men and women were analysed separately. To identify significant differences in the scores of questionnaires and their subcategories between genotypes, we used two-way ANOVA with the Bonferroni post hoc test. EEG results were analysed by separating the five frequency intervals (α, β, γ, θ, and δ). Repeated measures of two-way ANOVA with the Bonferroni post hoc test was used to determine statistically significant differences in EEG and ECG values between NS and S for each genotype. To determine significant differences in EEG and ECG values between genotypes during NS or S, we also used repeated measures of two-way ANOVA with the Bonferroni post hoc test. Statistical significance was set at p < 0.05.