REVERSE phenotyping—Can the phenotype following constitutive Tph2 gene inactivation in mice be transferred to children and adolescents with and without adhd?

Abstract Introduction Experimental models of neuropsychiatric disorders, for example, ADHD, are used to mimic specific phenotypic traits of a complex human disorder. However, it remains unresolved to what extent the animal phenotype reflects the specific human trait. The null mutant mouse of the serotonin‐synthesizing tryptophan hydroxylase‐2 (Tph2 ‐/‐) gene has been proposed as experimental model for ADHD with high face validity for impulsive, aggressive, and anxious behaviors. To validate this ADHD‐like model, we examined the Tph2‐/‐ phenotype in humans when considering allelic variation of TPH2 function (“reverse phenotyping”). Methods 58 participants (6 females, 8–18 years) were examined, of whom 32 were diagnosed with ADHD. All participants were phenotyped for impulsivity, aggression, and anxiety using questionnaires, behavioral tests, and MRI scanning while performing the 4‐choice serial reaction time task. Additionally, participants were genotyped for the TPH2 G‐703T (rs4570625) polymorphism. To analyze the relation between TPH2 G‐703T variants and the impulsive/aggressive/anxious phenotype, mediation analyses were performed using behavioral and MRI data as potential mediators. Results We found that the relation between TPH2 G‐703T and aggression as part of the reverse Tph2‐/‐ phenotype was mediated by structure and function of the right middle and inferior frontal gyrus. Conclusion At the example of trait aggression, our results support the assumption that the Tph2 null mutant mouse reflects the TPH2 G‐703T‐dependent phenotype in humans. Additionally, we conclude that “reverse phenotyping” is a promising method to validate experimental models and human findings for refined analysis of disease mechanisms.

Translational studies have gained importance in neuropsychiatric research, improving our understanding of mechanisms and pathways to comorbidities. In the case of the serotonin-synthesizing tryptophan hydroxylase-2 (TPH2) gene, the null mutant mice (Tph2 -/-) have been discussed as an experimental model for ADHD. Functional variants of the human TPH2 gene have been associated with ADHD in numerous studies (Manor et al., 2008;McKinney et al., 2008;Sheehan et al., 2005;Walitza et al., 2005), and Tph2 -/mice are characterized by increased impulsivity (Lesch & Merschdorf, 2000;Sachs et al., 2013), which is a core symptom of ADHD. However, as the Tph2 -/mouse also demonstrates increased aggression and decreased anxiety-like behavior (Lesch et al., 2012;Waider et al., 2019), it is not known to what extent the Tph2 -/mouse reflects ADHD symptomatology. To investigate whether the alterations following constitutive Tph2 gene inactivation in mice can also be found in humans via the way the TPH2 influences human (impulsive, aggressive, and anxious) behavior, we employ the "reverse phenotyping" approach "where phenotypes are refined based on genetic marker data" (Schulze & McMahon, 2004).
Based on the assumption that the animal phenotype reflects ADHDassociated symptoms, we examined children and adolescents with and without ADHD. Thus, in this study, the reverse Tph2 -/phenotype refers to the examination of the animal phenotype (i.e., altered impulsivity, aggression, and anxiety) in humans.
Altered impulsivity, aggression, and anxiety also correspond to phenotypes frequently observed in patients with ADHD, especially in the context of the abovementioned comorbidities. In humans, the influence of TPH2 is often addressed using either the tryptophan depletion approach or by stratification for the functional TPH2 G-703T (rs4570625) polymorphism. In patients with ADHD and typically developing children (TDC), tryptophan deficiency has been associated with increased impulsivity (Stoltenberg et al., 2012), heightened aggression (Duke et al., 2013), and altered anxiety (Lowry et al., 2008).
For example, carriers of the TPH2 T allele (T + ) presented increased risk of impulse control disorders (Gizer et al., 2009), disturbed affective behavior (Canli et al., 2008), and reduced response inhibition (Stoltenberg et al., 2012). Studies using tryptophan depletion on aggression in TDC and ADHD Polier et al., 2013) showed that depleted patients reacted strongest in comparison with nondepleted patients and TDC. In socially anxious T + patients, an overactive presynaptic serotonergic neurotransmission in the amygdala and anterior cingulate was described (Furmark et al., 2016), and in a sample of patients with ADHD with and without comorbid anxiety, anxiety was related to a reduction in impulsivity and response inhibition deficits suggesting that a "mildly elevated trait anxiety confers a protective influence by reducing the degree of impairment seen in ADHD" (Ruf et al., 2016).
In this fMRI study, we examined children with and without ADHD. Participants were genotyped for TPH2 G-703T and phenotyped according to trait impulsivity, aggression, and anxiety. In addition, structural and functional MRI was acquired using the 4-choice serial reaction time task (4-CSRTT) . Waiting impulsivity as measured via the 4-CSRTT is one form of impulsivity, defined as the tendency to premature responding, that is, to respond before target onset. Thus, it involves the aspects of response inhibition and top-down control, mediated by motivational aspects and reward processing (Robinson et al., 2009;Voon, 2014). The underlying brain network includes the ventromedial and dorsolateral prefrontal cortex as well as the anterior cingulate cortex representing motor or response inhibition (Mechelmans et al., 2017), the reward-related nucleus accumbens (Neufang et al., 2016), and hippocampus and amygdala responsible for reward-based learning (Dalley et al., 2011).
We used the dimensional approach as suggested by the rDoC (https://www.nimh.nih.gov/resea rch/resea rch-funde d-by-nimh/ rdoc/index.shtml), to address neuropsychiatric impairments in terms humans. Additionally, we conclude that "reverse phenotyping" is a promising method to validate experimental models and human findings for refined analysis of disease mechanisms.
Thus, all subjects were examined as one sample using regressionbased statistical analyses.
The aim of the study was to examine how TPH2 G-703T was related to trait impulsivity, aggression, and anxiety (i.e., the reverse Tph2 -/phenotype) in children and adolescent with and without ADHD. Additionally, the question was if brain structure and function mediated this relation. Therefore, TPH2 G-703T variant group comparisons were performed in a first step, to see whether TPH2 G-703T directly influences phenotype variables. In a second step, mediation analyses were performed to reveal whether the influence of TPH2 G-703T on phenotype variables indeed was direct or rather indirect via neural parameters.
Based on the earlier reported findings, we assumed that TPH2 G-703T modulated all three phenotype traits: considering the protective effect of anxiety on ADHD, we hypothesized a positive effect on anxiety (i.e., GG > T + ). In contrast, the influence of aggression and impulsivity was supposed to be negative, that is, T + presented higher impulsive and aggressive trait scores compared with GG. However, we did not expect to find significant differences between TPH2 G-703T variants on impulsivity, aggression, and anxiety (direct effect), but assumed that the relation between TPH2 G-703T and impulsivity, aggression, and anxiety was mediated by neural structure and/ or function (indirect effects), for example, of the prefrontal cortex and the amygdala.

| Subjects
We examined 59 subjects (6 females), comprising 32 patients with ADHD (3 females) and 26 TDC (3 females). One ADHD patient did not finish MRI scanning and was excluded from statistical analyses. Remaining 58 subjects were aged from 8 to 18 years. M = 13.5 ± 2.2 years showed normal physical maturation (Tanner stages: M = 3.1 ± 1.2), (Marshall & Tanner, 1969, 1970 and intelligence was screened via the "Culture Fair Intelligence Test" (Weiß, 2006) (M = 105.0 ± 14.4, range: 80-153). Healthy participants were recruited within the CRC-TRR-58, and patients with ADHD were recruited from in/outpatient clinics of the Department of Child and Adolescent Psychiatry Würzburg, within the research network ESCAlife. All patients were diagnosed with ADHD according to the DSM-V by trained clinicians (Table S1). Eighteen patients were medicated with methylphenidate (9 retard, daily dosage: M = 30±10.8 mg), and unmedicated patients were either medicationnaïve (n = 8) or stopped medication use for more than a year (n = 6).

| Reverse Tph2 -/-Phenotyping
The reverse Tph2 -/phenotype was determined via the (a) "hyperactivity/impulsivity" scale of the "Diagnostic System for Mental Disorders in Children and Adolescents according to ICD-10 and DSM-V" (Döpfner & Görtz-Dorten, 2017), (b) the questionnaire for aggressive behavior in children and adolescents (Goertz-Dorten & Döpfner, 2010), and (c) the trait scale of German version of the State-Trait Anxiety Inventory for Children (Unnewehr et al., 1992).
All three instruments were administered at the examination day, and completion was supervised by an examiner.

| Genotyping
Blood sampling and genotyping were performed in all subjects.
Genomic DNA was extracted from whole-blood samples according to a standard desalting protocol. Genotyping procedures were performed using PCR and gel electrophoresis. Genotyping for the TPH2 G-703T (rs4570625) polymorphism was performed according to published protocols (Hahn et al., 2013). TPH2 G-703T distribution (TT = 0, 0%; GT = 25, 43.1%; GG = 33, 56.9%; p(Exact) = .7845) did not deviate significantly from the expected numbers calculated according to the Hardy-Weinberg equilibrium using the program DeFinetti provided (https://wpcalc.com/en/equil ibriu m-hardy -weinb erg/). Based on the findings showing that TPH2 expression is decreased in carriers of the T allele (Lin et al., 2007), we defined two groups, subjects homozygous for the TPH2 G allele (GG, n = 33) and carriers of at least one T allele (T + , n = 25).

| Experimental paradigm
The 4-CSRTT examines waiting impulsivity defined as the capability to inhibit a response to earn a reward. A trial begins with a short presentation of 4 boxes, which starts the waiting period. After a certain cue-target interval, the target appears as a green dot, located in one of the four boxes. Correct and quick responding was remunerated by a monetary reward. Premature responses were defined as reactions before target onset in the anticipation of reward. The task consisted of one block outside the scanner (2.5 min) and five blocks within the MR scanner (14 min) (Neufang et al., 2016).
All participants were trained outside the scanner and prior to the actual task. After training sessions, a first baseline block was conducted to determine the individual mean reaction time window (rt, M rt ± 2SD). The baseline block consisted of 20 trials and did not include a reward. The individual rt windows were used for reward definition in all consecutive blocks, which were performed in the MR scanner. Reward was provided in terms of a win of 10 euro cents, when the answer was correct and within the individual rt window, or a win of 1 euro when subjects answered correctly and were faster than their individual rt window. Subjects encountered a loss of 1 euro when response times for correct responses were longer than the rt window. Incorrect answers were neither rewarded nor punished.
The experiment in the scanner included 5 further blocks, 4 blocks with the opportunity to earn a reward and a second baseline block without reward, with each having 20 trials. The 4 reward blocks had a hierarchical structure to increase the tendency for premature responses. In detail, in the 1st block in the scanner, the presentation duration of the target was 64 ms, and the cue-target interval was 2 s and only green, that is, correct targets were presented. In block 3, target presentation duration was decreased to 32 ms. In block 4, presentation duration remained 32 ms; in addition, cue-target interval varied from 2 s to 6.5 s. Finally in block 5, in addition to short presentation duration and varying cue-target intervals, distractor targets were presented in terms of blue or yellow dots (for further detail see (Neufang et al., 2016)). Total task duration was 14 min in the MR scanner and 2.5 min for the first baseline block.

| Data acquisition
Scanning was performed on a 3 Tesla TIM Trio Scanner (Siemens).
Whole-brain T2*-weighted BOLD images were recorded with a gra-

| fMRI processing
Data processing was performed using the Statistical Parametric Mapping Software Package (SPM12, Wellcome Department of Imaging Neuroscience, London, UK, Wellcome Trust Centre for Neuroimaging; http:// www.fil.ion.ucl.ac.uk/spm/). Data preprocessing in the native space included the steps of temporal and spatial alignment: All images were slice time-corrected, realigned to the first functional image, and unwarped. Images were then spatially normalized into a standard stereotactic space (Montreal Neurological Institute), resampled to an isotropic voxel size of 2 × 2×2 mm 3 , and spatially smoothed with a Gaussian kernel of 8mm full width at half maximum. Statistical analysis on the individual first level (singlesubject level) was based on the general linear model (GLM) approach.
Model specification included the definition of experimental condition, in our case "cue," "target," and "reward." Reward trials were subdivided into "win" and "loss" trials. Break periods were defined as "rest." In addition to the experimental conditions, nuisance regressors were specified, that is, "error trials" and "realignment parameters" (i.e., six regressors containing movement in three spatial and three rotational axes), to correct for error variance and movement artifacts. For each condition, onset times were determined from log files with onsets of the cue condition were determined at the time when the cue picture was presented. Onset times of target trials were defined in terms of the appearance of the target picture, and onset times of reward trials (win and loss) were the time points when the reward feedback picture appeared on the screen. The onsets of error trials were defined as the target onsets of incorrect trials. On the single subjects, three contrasts of interest were calculated, "cuerest" to identify cue-specific brain activation, "target-rest" to isolate target-induced brain activation, and "reward" in terms of "win-loss" to identify brain activation associated with the receipt of monetary reward. Resulting contrast images entered statistical group analysis. For mediation analyses, beta values of brain areas which were significantly related to the reverse Tph2 -/phenotype were extracted.

| sMRI processing
Structural MRI data were analyzed using the FreeSurfer (versions 5.3) software (Fischl et al., 2002). Analysis and quality-control protocols of the ENIGMA consortium (http://enigma.ini.usc.edu/proto cols /imaging-protocols) were applied including the recon-all -all stream and the segmentations of 68 (34 left and 34 right) cortical gray matter regions based on the Desikan-Killiany atlas (Desikan et al., 2006). In this study, we focused on regions associated with waiting impulsivity, that is, the inferior frontal gyrus (IFG, triangular/opercular/orbital part), the middle frontal gyrus (MFG), the anterior cingulate, hippocampus, amygdala, and nucleus accumbens.
Regional volume scores were corrected for global brain volume in terms of % of intracranial volume.

| Mediation analyses
In line with earlier publications (e.g., (Bi et al., 2017)), we performed mediation analyses to address the complex mechanism how TPH2 G-703T was related to the reverse Tph2 -/phenotype and which roles brain structure and activation played. We used the Macro PROCESS for SPSS developed by Hayes (Hayes, 2017) (http://www.proce ssmac ro.org/downl oad.html), which has been used in numerous earlier scientific reports, for example, (Kneer et al., 2020;Mertens et al., 2018). The simple mediation model includes two consequent variables; thus, two linear models are required and defined as follows: with i M and i Y being regression constants; e M and e Y , error estimates of M and Y; and a, b, and c', the regression coefficients given to the antecedent variables in the model (Figure S1 X, Y, M1, a1, b1 and c').
When using control variables (e.g., in our case diagnosis and age), the simple mediation model changes as follows: with y' representing the direct effect of X on Y, αβ defining the indirect effect of X on Y as mediated via M, and the sum of y' + αβ the total effect of X on Y.
In this study, we defined three models, one for anxiety, one for impulsivity, and one for aggression, with TPH2 G-703T as independent factor (X) and the reverse Tph2 -/phenotype as dependent variables (Y) and four mediator variables. Mediator variables were the two impulsivity/aggression/anxiety-specific activation scores (Table 2) and the corresponding regional volume (i.e., impulsivityassociated brain activation in the right IFG pars triangularis (M1) and the right MFG (M2), and the volumes of IFG tri (M3) and MFG (M4); aggression-related brain activation in the same regions and their structural volumes; and activation correlated with anxiety in the right IFG pars opercularis and the MFG, and the regional volumes of the IFG op and the MFG] ( Figure S1). All three models were tested twice varying between parallel (PROCESS, model 4) and serial mediation effects (PROCESS, model 6). In the case of serial mediation, the total effect of c is dissected into five different indirect effects: Statistical significance of direct effects was reached when p < .05, FDR-corrected for 15 comparisons (for a detailed description of direct effects, Table 3). Indirect effects were tested for significance using bootstrapping (no of bootstrap samples: 10,000). In this case, the test for significance was one-sided assuming that indirect effects (i.e., αβ as the indirect effect of X on Y, mediated via M) were > 0. Therefore, bootstrapped confidence intervals of 95% were defined. If confidence intervals did not include 0, the indirect effect was considered as significant.

| Statistical analysis
1. In a first step, group comparisons were performed on phenotype scores using two-sample tests with the independent factor TPH2 G-703T (GG versus. T + ) and phenotype-related trait dimensions (impulsivity, aggression, and anxiety) as dependent variables.
2. To address the interaction between TPH2 G-703T genotype, the reverse Tph2 -/phenotype, and the waiting impulsivity network, mediation analyses were defined. To do so, waiting impulsivity parameters which were influenced by the reverse Tph2 -/phenotype had to be identified in a pre-analysis using multiple regressions (2a) with phenotype scores as independent factors, (a) behavioral performance (# premature responses, accuracy, rt), (b) regional brain volumes, and (c) neural activation (contrast images for cue, target, and reward) as dependent variables.
Age and diagnostic group (ADHD versus. TDC) were included as nuisance variables. (2b) Finally, mediation analyses were performed We used PROCESS model 4 (assuming mediators operate in parallel) and PROCESS model 6 (assuming that mediators operate in serial) as suggested by Hayes (Hayes, 2017) ( Figure S1).

Task condition
Regressor k x y z Z region

TA B L E 2
Multiple regressions of the reversed Tph2 -/phenotype on brain activation using multiple regression models (independent factors: impulsivity, aggression, anxiety)

| Analysis step 1
Group comparisons did not reveal any significant difference between TPH2 G-703T variants; however, in absolute values T + were TA B L E 3 Mediation model with the independent factor X = TPH2 G-703T, the dependent variable Y = aggression, potentially mediating variables M1 = aggression-related activation in the right IFG tri , M2 = aggression-associated activation in the MFG, M3 = right IFG tri volume, and M4 = right MFG volume, as well as nuisance variables: age and diagnostic group more impulsive compared with GG (uncorrected p < .05), and aggression scores were higher (Table 1).

| Analysis step 2a
Multiple regressions revealed (a) on the behavioral level, a positive influence of impulsivity on rt (Table S2). (b) On brain structure, no significant correlations were found (Table S3). (c) In brain activation, the reversed Tph2 -/phenotype correlated predominantly with activation in the reward condition: Whereas impulsivity and aggression correlated with overlapping regions in the triangular part of the right IFG and the right MFG, anxiety correlated with an isolated region in right MFG. In addition, correlation with aggression was negative, and correlations with impulsivity and anxiety were positive (Table 2, Figure 1). During cue processing, impulsivity and aggression positively correlated with activation in the opercular part of the IFG (impulsivity: left IFG, aggression: right IFG), the same was found for anxiety during target processing (right IFG) ( Table 2).

| Analysis step 2b
Mediation analyses using (a) behavioral, (b) structural, and (c) functional parameters alone did not reveal a significant interaction with TPH2 G-703T and the impulsive/aggressive/anxious phenotype.
TA B L E 1 A priori group differences between TPH2 G-703T polymorphisms using two-sample t tests (GG versus T + ) However, when combining structural and functional parameters, significant serial mediations were found: The relation between the TPH2 G-703T genotype and the aggressive phenotype was mediated by brain activation in the right IFG (triangular part) and right MFG volume (Table 3, ind 4), and right IFG volume (Table 3,

| D ISCUSS I ON
In this study, we addressed the mechanisms underlying the reversed Tph2 -/phenotype in children and adolescents with and without ADHD. We showed that the influence of TPH2 G-703T on aggression (as part of the reversed Tph2 -/phenotype) was mediated by brain activation and regional volume of the triangular part of the right IFG and the right MFG. We did not find similar associations for impulsivity and anxiety. Thus, for aggression we were able to demonstrate a similar genotype/phenotype relation in mice and humans.
F I G U R E 1 represents brain regions, associated with the reversed Tph2 -/phenotype. On the left, brain activation is plotted on a representative brain surface. On the right, scatterplot represents correlations Agg Imp Anx F I G U R E 2 represents the significant mediation model for aggression as part of the reversed Tph2 -/phenotype with the independent factor X = TPH2 G-703T, the dependent variable Y = aggression, and mediating variables M1 = aggression-related activation in the right IFG tri , M2 = aggression-associated activation in the MFG, M3 = right IFG tri volume, and M4 = right MFG volume. On the left side, significant T+-specific indirect effect is presented, that is, in T-allele carriers, the TPH2 G-703T genotype is related to trait aggression not directly but via right IFG activation and right MFG volume. On the right side, GG-specific indirect effect is depicted, that is, in GG homozygotes, and the TPH2 G-703T genotype is related to trait aggression not directly but via right IFG activation, MFG activation and right MFG volume

| Aggression and prefrontal structures
Aggression in humans is associated with an imbalance within a cortico-limbic network (Rosell & Siever, 2015), that is, a deficient regulation of the amygdala via prefrontal areas (Coccaro et al., 2011).
Fronto-amygdala pathways have been involved in the control of aggressive impulses (Bufkin & Luttrell, 2005); reduced volumes in the prefrontal cortex and amygdala were associated with antisocial traits (Gregory et al., 2012), MFG activation counteracted with aggressive reactions (Achterberg et al., 2016), and IFG-amygdala connectivity was reduced in aggressive individuals (Bogerts et al., 2018).
Specifically, findings suggested right-lateralized alterations for both volume reductions in the prefrontal cortex (Cha et al., 2015) and IFGamygdala connectivity (Gilam et al., 2017). In line with these findings, we found that prefrontal regions were lateralized in the right hemisphere and that predominantly the IFG and MFG were involved.
However, we did not find any correlation between aggression and amygdala. This finding could not be explained by diagnostic group, as (a) all analyses were corrected for group and (b) there were no a priori group differences between ADHD and TDC in our sample. In addition, post hoc group-specific multiple regressions did not reveal significant relations between aggression and amygdala volume neither for ADHD nor for TDC (ADHD: beta aggression = 0.06, p = .792; TDC: beta aggression = 0.04, p = .858). Likewise, the lack of relation between aggression and amygdala volume was not confounded by group-specific brain development, as multiple regressions were corrected for age, and post hoc group-specific age correlations did not reveal a significant correlation (ADHD: R rAMY_vol*age = −0.07, p = .722; TDC: R rAMY_vol*age = −0.02, p = .921). Thus, especially the role of the amygdala should be further investigated in future studies including greater sample sizes.

| Aggression and TPH2 G-703T
Aggression has been shown to vary between TPH2 G-703T variants in numerous studies. For example, Yoon et al. (2012) linked TPH2 G-703T to anger-related personality traits, finding that GG had a higher anger control scores compared with T + and that the correlation between aggression and orbitofrontal cortex volume differed between genotypes, suggesting that the orbitofrontal cortex is "an intermediate phenotype that bridges serotonin synthesis and anger-related traits. The mechanism underlying the effect of the TPH2 gene on OFC abnormality, however, may be complex and may involve several processes related to anger expression." (Yoon et al., 2012) More generally, Wolf et al. (2018) showed serotonergic modulation of aggression in a pharmacological fMRI study. Participants were playing a video game during fMRI scanning after a single dose of selective 5-HT reuptake inhibitor (escitalopram). They found that medication reduced neural response in the right IFG and ACC to violent but not nonviolent actions and underlined the validity of serotonin in the modeling of aggressive behavior (Rosell & Siever, 2015;Wolf et al., 2018). Similar effects have also been found in mice: After injections of 5-HT receptor agonists in the prefrontal cortex, animals showed reduced aggressive behavior in terms of attack bites and lateral threats (Centenaro et al., 2008). Likewise, we found that the mediation effect differed between TPH2 G-703T variants: In contrast to GG, in T + IFG volume was not a significant mediator. Thus, our results expand on these findings, (a) suggesting IFG as further region increasing susceptibility to serotonergic imbalance and (b) showing that brain function in addition to structure plays a role in serotonergic modulation.

| Relation between impulsivity, aggression, and anxiety
At first sight, examining impulsivity, aggression, and anxiety in humans may seem counter-intuitive (Cohn et al., 2016). However, they represent a cohesive behavioral program for reacting in stressful and dangerous situations: the fight or flight response, with fight representing the "impulsive/aggressive" way to react and flight the rather "anxious" manner to respond. Thus, impulsivity, aggression, and anxiety represent internal emotional states and are natural adaptive consequences of stress to cope with the stressor (Lesch, 2005). Additionally, in the context of ADHD symptoms aggression and anxiety have been shown to closely interact (Schatz & Rostain, 2006), even though the direction of relations is still being discussed. The exacerbation hypothesis proposes that the presence of anxiety increases the risk of aggression arguing that anxiety increases the emotional response to a stressful situation, thus, lowers the threshold for the need to react, and inhibits the fine-tuned behavioral reaction. The attenuation hypothesis assumes that the presence of anxiety protects against aggression as anxiety inhibits process inducing rather freezing behavior than aggressive one. A very recent study, in return, showed that the relation between aggression and anxiety was direct (neither mediated nor moderated by further variables), that is, aggression was negatively related to anxiety and did not interact with ADHD symptoms at all (Murray et al., 2018). Overlapping function between impulsivity and aggression and inverse relation between aggression and anxiety could be found in our fMRI findings as well: a priori correlations revealed a positive relation between impulsivity and aggression, whereas it was negative between aggression and anxiety. Likewise, impulsivity-and aggression-related regions were in overlapping areas, whereas anxiety was associated with an isolated region in the MFG. Finally, while the more anxious subjects were, the stronger they activated the MFG, it was reverse in aggression (i.e., the more aggressive subjects were, the weaker MFG and IFG activation).

| TPH2 G-703T (rs4570625) polymorphism and ADHD
Even though numerous papers showed that this polymorphism plays a significant role in the context of ADHD via reduced serotonin in T-allele carriers, these findings are not undisputed as they were not replicated in recent studies. For example, the TPH2 G-703T (rs4570625) polymorphism did not show up as a risk gene in large GWAS study (Demontis et al., 2019) and a recent meta-analysis (Ottenhof et al., 2018) did not show an association with ADHD. In addition, the common noncoding G-703T polymorphisms have not been conclusively shown to alter expression of TPH2, and thus, serotonin function: Scheuch et al. (2007) reported that it is more likely the rs11178997 of the human TPH2 that significantly reduces promoter activity, not rs4570625 (Scheuch et al., 2007). Similar, Heinrich et al. (2017) showed that methylation in the TPH2, probe cg14791008 was associated with reward-based reaction times and ADHD symptoms in ADHD patients (Heinrich et al., 2017). However, a genome-wide link between genes, cortical surface area, and ADHD has been shown in a recent paper by Gutiérrez et al. (2002). They found for ADHD significant negative genetic correlations between surface area of the MFG and IFG and numerous chromosomes (see Grasby et al., Table S1), for example, between the triangular part of the IFG and chromosome 11q23.1. Chromosome 11q23.1 has been linked to serotonergic 5-hydroxytryptamine A-receptor (Gutiérrez et al., 2002;Weiss et al., 1995), which also have both been reported to be relevant in ADHD (Hou et al., 2018;Li et al., 2014;Oades et al., 2008)  to address multiple behavioral/cognitive dimensions like in the reversed phenotyping approach, when studying TPH2 effects on brain structure and function of the normally and pathologically developing brain.

| LI M ITATI O N S
In summary, we examined the reversed Tph2 -/phenotype was present in humans. We were able to detect a mechanism for aggression but not for impulsivity and anxiety. One argument is that the reversed Tph2 -/phenotype is a phenotype of a complete knockout mouse, which is not reflected by T + . Analogies between heterozygote knockout mice would have been more realistic, and therefore, it is plausible that we did not find the full phenotype. Another explanation might be that the waiting impulsivity network was too complex.
Impulsivity tasks stimulating, for example, only the fronto-striatal loop such as the GoNogo task might have shown similar PFC results, however, leaving more statistical variance to only one further region instead of multiple, enhancing the probability of significant results also for the second region (Cohen, 2013). However, in our study, we chose the 4-CSRTT as the current study is part of a translational project and the 4-CSRTT is a task of which versions for both humans and animals have been designed and published by the same laboratory (Bari et al., 2008;Voon et al., 2014), offering a unique chance to compare behavioral performance between two species.
Finally, the approach of mediation analyses has statistical limitations. One major aspect is that mediations promise to predict causal relations. However, this is only the case, if one pre-assumption is fulfilled, the no confounding or ignorability assumption (Lee et al., 2019). This means that all major variables are implemented in the model. Thus, potential confounders might also have influenced our findings, for example, the bivariate differentiation between genotypes. We know that epigenetic markers modulate the impact of genotypes (Schuebel et al., 2016); thus, further genetic markers such as methylation (Gottschalk & Domschke, 2016) of TPH2 G-703T (rs4570625) would have been highly advantageous. Furthermore, we examine the same variables for three different phenotype dimensions. Thus, insignificant results on anxiety and impulsivity may reflect that additional variables were lacking, which was not the case regarding aggression. Finally, we cannot exclude that "unmeasured confounding may still introduce bias even if known confounders have been adjusted for" page 698, (Lee et al., 2019).

| CON CLUS ION
In this study, we used the "reverse phenotyping" approach as a potential methodological way to combine findings from animal and human literature. At the example of aggression, we were able to show that the animal phenotype in humans was based on a similar gene X (brain X)-phenotype interaction as described in animal findings. In our view, this is a promising approach for a direct way to examine experimental models in psychiatric populations. Future studies need to confirm how the found relationships may account for ADHD-associated symptoms and covarying comorbid disorders in the context of ongoing development throughout early and later adulthood. DFG and BMBF had no further role in the interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.

CO N FLI C T O F I NTE R E S T
None of the authors declare any conflict of interest. SN, AA, and AR performed statistical analyses and manuscript drafting. All participants finalized the submitted manuscript.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1002/brb3.2054.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon request.