Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Ketone body 3-hydroxybutyrate as a biomarker of aggression

Abstract

Human aggression is a complex behaviour, the biological underpinnings of which remain poorly known. To gain insights into aggression biology, we studied relationships with aggression of 11 low-molecular-weight metabolites (amino acids, ketone bodies), processed using 1H nuclear magnetic resonance spectroscopy. We used a discovery sample of young adults and an independent adult replication sample. We studied 725 young adults from a population-based Finnish twin cohort born 1983–1987, with aggression levels rated in adolescence (ages 12, 14, 17) by multiple raters and blood plasma samples at age 22. Linear regression models specified metabolites as the response variable and aggression ratings as predictor variables, and included several potential confounders. All metabolites showed low correlations with aggression, with only one—3-hydroxybutyrate, a ketone body produced during fasting—showing significant (negative) associations with aggression. Effect sizes for different raters were generally similar in magnitude, while teacher-rated (age 12) and self-rated (age 14) aggression were both significant predictors of 3-hydroxybutyrate in multi-rater models. In an independent replication sample of 960 adults from the Netherlands Twin Register, higher aggression (self-rated) was also related to lower levels of 3-hydroxybutyrate. These exploratory epidemiologic results warrant further studies on the role of ketone metabolism in aggression.

Introduction

Aggression is a significant behavioural problem affecting children and adults, with consequences not only for individuals, but families, schools, and communities as well1. It is also a complex and heterogenous behaviour that has been examined from multiple perspectives, including epidemiologic2, genetic3, epigenetic4,5, and biomarkers6, to name a few. With the heritability of aggression in both children and adults found to be about 50%3, aggression has a strong biological basis. Yet, much variation in aggressive behaviour remains to be explained. Some of the mechanisms, pathways, and molecules involved in the biological-level aspects of the behaviour are known; however, in both adults and children, there remain biological pathways to discover and biomarkers yet to identify6,7,8. With increased understanding of the biology of aggression, the opportunity to improve our interventions as well as pharmacological treatments arises. Thus, there is a need to generate new hypotheses and approaches to enhance our understanding of aggression and its associated problems.

One such suggested approach is using proton nuclear magnetic resonance (NMR) spectroscopy to find new biomarkers, because it allows for a hypothesis-free way to examine phenotypic associations9,10,11. NMR is not a new technology, but applying the approach to behavioural/psychological problems, including aggression, is gaining interest6. The NMR methodology is generally able to produce information on the levels of lipids, amino acids, organic acids, and other small organic molecules. Associations between lipid concentrations and aggression have previously been noted6,12,13,14,15, while other low-molecular-weight molecules might hold more promise for novel findings. Amino acids and ketone bodies (low-molecular-weight molecules categorized in NMR) have been associated with emotional and behavioural problems other than aggression16,17,18,19,20, and some studies have shown amino acids and similar molecules associated with anger and psychopathy21, schizophrenia-related violence15, and childhood aggression22. Accordingly, these areas offered a reasonable place to focus our exploratory investigation.

Thus, in this study we combine an NMR approach with the aggression phenotype in efforts to generate new insights into our biological understanding of aggression. We conducted exploratory analyses on the association between aggression and amino acid and ketone body biomarkers, seeking to explore association(s) found in more depth and with an attempt to replicate them in an independent sample. The discovery dataset includes longitudinal aggression measurements in adolescence and biomarker data in young adulthood, while the replication dataset includes biomarker and aggression data in young adults.

Materials and methods

Discovery dataset (FinnTwin12)

The FinnTwin12 study is a longitudinal population-based cohort of Finnish twins born in 1983–1987 that was initiated to track behavioural development and health habits from mid-childhood onward23,24. Twins and their families were identified through the Finnish Central Population Registry and initial enrolment of the twins occurred between the ages of 11 and 12. The baseline response rate was 87% (N = 5600 twins) and remained high (between 85 and 90%) for all data collection waves. Questionnaire data collection waves occurred around ages 11/12, 14, 17, and 22, and were collected from parents (age 11/12), teachers (age 11/12 and 14), and twins themselves (ages 14, 17 and 22).

Additionally, a subset of twins (from 1035 families) was more intensively studied using additional assessments (semi-structured psychiatric interview and questionnaires) in young adulthood (mean age = 22.4, SD = 0.70; n = 1347, response rate 73.0%), as well as a plasma sample, anthropometrics and neuropsychological tests for twins tested in person in young adulthood (n = 779). Previous investigation has shown that these more intensively studied twins do not differ from the rest of the cohort regarding behavioural or emotional problems25.

A blood sample was collected after an overnight fast with the request that the participant abstain from alcohol and tobacco since the night before the sample. Plasma was extracted immediately and stored at − 80 °C. Samples were processed in one batch in the autumn of 2010 using an automated high-throughput serum NMR metabolomics platform (i.e., the Brainshake/Nightingale platform)24,26,27. This platform has demonstrated good quality control related to sample preparation and experimental factors28,29. The metabolites utilized here were the amino acids (alanine, glutamine, histidine, isoleucine, leucine, phenylalanine, tyrosine, valine) and ketone bodies (acetate, acetoacetate, 3-hydroxybutyrate). All metabolite data were available with units mmol/l. Pregnant women (n = 53) and individuals on cholesterol medication (n = 1) were excluded from analysis; those excluded were no different in their aggression scores compared to those included. Thus, the final sample for study was 725 twin individuals.

Ethical approval for all waves of data collection was obtained from the ethical committee of the Helsinki and Uusimaa University Hospital District and Indiana University’s Institutional Review Board. All data collection and sampling was performed in compliance with the ethical guidelines. At ages 12 and 14, parents provided consent for the twins, while twins themselves provided written consent for age 17 and 22 collection waves.

Aggression measurement

Behavioural and emotional development in FinnTwin12 were measured at ages 12, 14, and 17 (from parents, teachers, twins themselves, and co-twins rated their twin as well) using the modified Multidimensional Peer Nomination Inventory (MPNI)30. The MPNI is a 37-item questionnaire producing multiple subscales including aggression (6 items; 4 direct aggression items, 2 indirect aggression items). Each question about the child’s behaviour (e.g., “Does the child call people names when angry with them?”) has four response choices: ‘not observed in child’, ‘observed sometimes, but not consistently’, ‘definitely observed, but is not prominent’, and ‘clearly observed’. Response choices are scored 0–3, and aggression scores for individual raters (e.g., teacher rating at age 12) are formed by taking the mean of all items in the aggression subscale (no missing values were allowed). A general aggression score was created (using all 6 items). We use the four different types of raters at three different ages (seven aggression ratings in total) because the literature indicates that different raters observe different aspects of children’s behaviour in the different settings available to them31,32, and by using multiple ratings we can assess the robustness of the biomarker associations across settings, observers, and adolescence. Additionally, a composite aggression score (called ‘Aggression Combined’) was used for initial analyses in order to minimize multiple testing and to combine information from different informants, who tend to rate aggression slightly differently31,33. It was formed by calculating the mean of the aggression subscale scores for the parent rating at age 12 and the teacher ratings at ages 12 and 14 combined.

Covariates

Based on careful literature review, several covariates, measured at the time of sample collection, were included in the regression model because they were related to both biomarker levels and aggression, and they were available in both the discovery and the replication samples. Body mass index (BMI) was calculated as kg/m2 (mean: 23.3, standard deviation (SD): 3.94). Leisure-time physical activity was expressed as metabolic equivalent of task (MET) hours/day based on structured questions considering the intensity, frequency, and duration of activities34. Smoking status was categorized as current, former, never/experimenter. Alcohol consumption frequency was categorized as none, once a month or less, 2–4 times per month, 2–3 times per week, 4 or more times per week. BMI, physical activity, alcohol use and smoking have been shown to be related to both metabolite levels26,35,36,37 and aggression38,39,40. Finally, self-rated general health was dichotomized to either poor health (poor + fair) or good health (good + very good). This measure was included to roughly capture somatic health problems (e.g., diabetes) in these young adults that could affect their metabolite levels41, and which could also be related to aggression42.

Replication dataset (NTR)

The Netherlands Twin Register (NTR), established in 1987, collects data on twins and their family members. Data collection is nationwide via questionnaires mailed to parents of young twins (ages 1–12) and to twins themselves at age 14 and beyond. Twins recruited as adolescents and adults and their families are invited to take part in data collection by survey every 2–4 years43. Multiple survey waves have included the Achenbach System of Empirically Based Assessment (ASEBA) Adult Self-Report (ASR)44, which includes questions that allow for the calculation of an aggression score. A subgroup of adult participants was invited into the NTR Biobank project45,46. As part of this project, they were visited at home where samples were collected between 7:00 and 10:00 after overnight fasting. Participants were asked not to smoke before sampling. Fertile women took part from day 2–4 of their menstrual cycle, or in the pill-free week. During the visit, body composition measurements were taken and information regarding physical health and lifestyle behaviour (e.g., smoking and drinking patterns, exercise, medication use) was collected.

Metabolite data, processed using the same Brainshake/Nightingale platform as FinnTwin1227, were assessed in 5204 participants (age range: 18–87 years). As in FinnTwin12, we excluded women who were pregnant (n = 19); those excluded were no different in their aggression scores compared to included individuals. Additionally, all individuals over the age of 30 years at the time of sampling (n = 3843) were excluded to make the FinnTwin12 and NTR samples similar in age composition and because changes in body chemistry occur that accompany, for example, menopause and the ageing process for both sexes47. The two data collection timepoints (wave 8 and 10) that obtained aggression scores (ASR questionnaire) and were closest to the time of blood sampling were analysed. ASR aggression scores at wave 8 were available from 819 participants (mean age: 28.1 years, range 18–38 years), and ASR aggression scores from wave 10 from 665 participants (mean age: 32.7 years, range 21–39 years). The aggression score used for the analyses was the score closest to the time of blood sample collection. Of note, aggression scores were generally available closer to the time of sampling for the NTR dataset (mean 3.4 years; range 0–10 years) than the FinnTwin12. There were 150 (non-pregnant, age 30 or younger at time of blood sampling) participants who had aggression score values collected before their blood sample, and these scores were excluded to make the dataset temporally uniform (blood samples first, aggression scores later; this is the opposite of the temporality of the FinnTwin12 sample). Their aggression scores were no different than those of included individuals. Thus, the final sample for study was 960 individuals. For replication, only the metabolite 3-hydroxybutyrate (mmol/l) was examined.

NTR projects were approved by the Central Ethics Committee on Research Involving Human Subjects of the VU University Medical Centre, Amsterdam, an Institutional Review Board certified by the U.S. Office of Human Research Protections (IRB number IRB00002991 under Federal-wide Assurance FWA00017598; IRB/institute code NTR 03-180). All data collection and sampling were performed in compliance with the ethical guidelines, and all participants provided written informed consent.

Aggression measurement

Aggression in the Adult NTR was analysed from surveys 8 and 10, which both included the ASR, part of the ASEBA for assessing problems and strengths in children and adults, with multiple subscales including aggression44. Each aggression question has three response choices which were summed over items, if the number of missing items was not larger than three. For participants with no more than three missing values, missing items were imputed. If participants had completed the ASR at both surveys 8 and 10, we took their aggression score closest to the blood sampling (but always after blood sampling).

Covariates

We aimed to mirror variables and categories as close to the FinnTwin12 dataset as possible. BMI was calculated as kg/m2 (from the biobank home visit) (mean: 23.0, SD: 3.43). Leisure-time physical activity (from the same survey wave that the aggression score was taken) was calculated as MET minutes. Smoking status was categorized as current, former, never (from the biobank home visit). Alcohol consumption frequency (from the same survey wave as the aggression score) was categorized as none, once a month or less, 2–4 times per month, 2–3 times per week, 4 or more times per week. Self-rated general health (from the same survey wave that the aggression score was taken) was dichotomized to either poor health (poor + fair) or good health (reasonable + good + very good). Since the NTR metabolite data were processed in different batches, there was a batch variable available for adjustment.

Analysis

For descriptive statistics (supplemental), we calculated means and standard deviations (SD) and ranges for the aggression scores for the different raters that were used in the analyses in FinnTwin12 (as well as in the NTR replication dataset).

Exploratory analyses in FinnTwin12 involving the available low-molecular-weight metabolites (n = 11; 8 amino acids and 3 ketone bodies) utilised the composite aggression score (Aggression Combined). Spearman correlations were computed between the untransformed metabolites, and between the metabolites and the Aggression Combined variable. Thereafter, due to their non-normal distribution, all metabolites were rank transformed (generally the best solution for most of the metabolites). Then, linear regression was performed with individual metabolites as the response variable and Aggression Combined as the main predictor variable, adjusted for age, sex, and BMI. To obtain robust standard errors (i.e., corrected for familial relatedness), the cluster option in Stata was specified48. Beta coefficients and 95% confidence intervals (CI) were reported and those reaching significance (p < 0.05) were investigated further. After exploratory analyses were performed, we proceeded to investigate significant findings further in an independent sample of young adults.

To further investigate 3-hydroxybutyrate (the only biomarker that reached a nominal p < 0.05 adjusted for multiple testing for 9 independent variables obtained from Matrix Spectral Decomposition (matSpD: https://gump.qimr.edu.au/general/daleN/matSpD/), which estimated the equivalent number of independent variables in our correlation matrix of 11 metabolites), additional linear regression modelling was employed to establish consistency of the association. In addition to the primary Aggression Combined variable, we examined the separate general aggression ratings from different raters as well (parents at age 12, teacher(s) at age 12, teacher(s) at age 14, self ratings at ages 14 and 17, and the twin’s co-twin at ages 14 and 17). Initial models designated 3-hydroxybutyrate as the response variable and the different general aggression ratings as the main predictor variable, adjusted for age, sex, BMI, and familial relatedness. Further models additionally adjusted for leisure-time physical activity (METs), smoking status, alcohol consumption frequency, and self-rated general health. Sex interactions were also investigated. Supplemental post hoc analysis investigated whether including multiple general aggression ratings in the same model could clarify the results, since a previous investigation of ours indicated that multiple raters in one model improved aggression information (different raters observe in different settings and can provide unique information31) and improved prediction of a later personality disorder33.

To obtain further evidence to support or refute the association found in our discovery sample, we analysed the NTR dataset to test replication of our results. The initial Spearman correlation of the untransformed 3-hydroxybutyrate and the aggression rating was computed. Then, the 3-hydroxybutyrate variable was rank transformed, as in the FinnTwin12 data. Linear regression models were run with 3-hydroxybutyrate as the response variable and the aggression rating as the predictor variable, adjusted for age, sex, BMI, batch and familial relatedness. Additionally, fully adjusted models further adjusted for leisure-time physical activity (METs), smoking status, alcohol consumption frequency, and self-rated general health. Sex interactions were also investigated to study if the aggression–3-hydroxybutyrate relationship differed in males and females49.

Finally, additional analyses were run to verify the maximum likelihood estimates of the association examined in the fully adjusted frequentist models above with a Bayesian approach, a powerful tool for incorporating information from previous studies, for controlling confounding, for interpretation of study results and evaluation of hypotheses about exposure–disease relationships50. We fitted several Bayesian random-intercept (2-level or mixed) regression models separately on the FinnTwin12 and NTR data with 3-hydroxybutyrate as the response variable and the aggression rating as the predictor variable (we used ‘Aggression Combined’ for FinnTwin12). We adjusted for age, sex, BMI, MET, smoking status, alcohol consumption frequency, and self-rated general health (batch was also included for NTR), and ran Bayesian models with and without an interaction of sex and aggression included, as well as stratified on sex. The clustering of the twins was accounted for by setting each twin pair a random intercept.

The Bayesian model parameters to estimate included the overall intercept, the regression coefficients and the random intercepts along with their variance components. For the overall intercept and the regression coefficients, we set a weakly informative (default) prior (normal (0,10,000)). For the variance of the error term, we set a non-informative (default) prior (inverse-gamma prior with shape and scale parameters equal to 0.1). For the random intercepts, we set a hierarchical prior normal (0, sigma), with the hyperparameter sigma set a uniform prior with bounds at the 95% confidence interval of the maximum likelihood estimate of the variance of the random intercepts (i.e., 95% CI for the variance from the frequentist random-intercept model). In fact, there appeared a convergence problem when using a non-informative (default) prior for the variance of the random intercepts. To overcome this problem, we incorporated information on the probable range of the variance via the prior distribution. Gibbs sampling was used to draw samples from the posterior distribution (20,000 iterations, including a burn-in of 7000 iterations). To confirm the convergence of Markov chain Monte Carlo (MCMC) chains, we performed a visual evaluation (trace, density and autocorrelation plots). From these models, we reported the estimated regression coefficient for aggression as the posterior expectation (mean) and 95% credible intervals assessed from the posterior distribution. Lastly, to assess whether the use of informative priors affects the results, we performed a sensitivity analysis by incorporating information on the association obtained from the Bayesian sex-stratified analysis of the FinnTwin12 cohort to the priors for the aggression coefficients in the NTR sex-stratified models. The coefficients were assigned the normal distribution with the mean set to the posterior mean and the variance set to the squared standard deviation obtained from the corresponding Bayesian sex-stratified analysis on the FinnTwin12 cohort.

All analyses were performed using Stata version 15.1 or 16.0 (Stata Corporation, College Station, TX, USA).

Results

Discovery (FinnTwin12) results

Descriptive means (SD) and ranges of the aggression scores from the different raters used in the analyses are provided in Supplemental Table 1. Spearman correlations between untransformed metabolites indicated generally low to moderate associations, except the high correlations between 3-hydroxybutyrate and acetoacetate (r = 0.70) and the branched-chain amino acids (leucine, isoleucine, and valine) with each other (r range = 0.62–0.73) (Table 1). Spearman correlations between metabolites and the Aggression Combined variable were low, ranging from − 0.13 to 0.14.

Table 1 FinnTwin12 Spearman correlations of untransformed metabolites and aggression (n=526).

Next, linear regression models with rank-transformed metabolites as the response variable and Aggression Combined as the main predictor variable, also adjusted for age, sex, BMI, and familial relatedness, revealed that only 3-hydroxybutyrate was significantly associated with aggression, including after adjustment for multiple testing (Fig. 1; Supplemental Table 2). The association was negative, with lower 3-hydroxybutyrate levels being associated with higher aggression levels.

Figure 1
figure1

FinnTwin12 regression coefficients (and 95% confidence intervals) of Aggression Combined predictor variable in linear regression models with metabolites as response variables, adjusted for age, sex, body mass index, and familial relatedness. ace acetate, acace acetoacetate, bohbut 3-hydroxybutyrate, ala alanine, gln glutamine, his histidine, ile isoleucine, leu leucine, phe phenylalanine, tyr tyrosine, val valine.

In linear regression models further exploring 3-hydroxybutyrate’s association with aggression, the basic model (3-hydroxybutyrate as the response variable, and aggression ratings from different raters as main predictor variables in separate models, adjusted for age, sex, BMI, and familial relatedness) results indicated that nearly all aggression ratings demonstrated the trend of negative association of 3-hydroxybutyrate with aggression, with teacher ratings at age 12 and self ratings at age 14 showing the strongest effect sizes (and also being statistically significant) (Fig. 2; Supplemental Table 3). In fully adjusted models (basic model plus MET, smoking status, alcohol consumption frequency, and self-rated general health), the negative associations remained, with only small attenuation to the effect sizes (Table 2). No sex interactions were detected (p-value range = 0.10–0.84; Supplemental Table 4).

Figure 2
figure2

FinnTwin12 regression coefficients (and 95% confidence intervals) of aggression ratings from different raters as (separate) predictor variables in linear regression models with rank-transformed 3-hydroxybutyrate as the response variable, adjusted for age, sex, body mass index, and familial relatedness. AggC aggression combined, P12 parent rating age 12, T12 teacher rating age 12, T14 teacher rating age 14, S14 self rating at age 14, TW14 co-twin rating at age 14, S17 self rating at age 17, TW17 co-twin rating at age 17.

Table 2 FinnTwin12 multivariable linear regression models with 3-hydroxybutyrate as the response variable, regression coefficients of aggression ratings from different raters as predictor variables in separate models, adjusted for age, sex, BMI, MET, smoking status, alcohol consumption frequency, self-rated general health, and familial relatedness.

In supplemental analyses, because the aggression scores from the different raters were only weakly to moderately correlated (r range: 0.14–0.47), multiple aggression ratings were added into the same model as predictor variables, adjusted for age, sex, BMI, and familial relatedness. The tested combinations included the three aggression scores that made up the Aggression Combined variable (parent age 12, teacher age 12, teacher age 14), both age 12 ratings, all three age 14 ratings, and both age 17 ratings, as well as the two ratings that were consistently significant (teacher age 12 and self age 14). Interestingly, in the model including teacher age 12 and self age 14 ratings, both were statistically significantly negatively associated with 3-hydroxybutyrate (Supplemental Table 5).

Replication (NTR) results

In the NTR replication dataset, the Spearman correlation between the aggression score and untransformed 3-hydroxybutyrate was − 0.07 (p = 0.045). The initial linear regression model with rank-transformed 3-hydroxybutyrate as the response variable and aggression as the main predictor variable, adjusted for age, sex, BMI, batch and familial relatedness, showed the same negative association trend (p = 0.138) as the FinnTwin12 models (Table 3). There was no sex interaction detected in the initial model (p = 0.246), however, in the fully adjusted model there was (p = 0.015). Regarding the fully adjusted models, the negative association was present for males only; it became slightly positive in direction for females, albeit with wide confidence intervals. Table 3 displays results for the sexes pooled and males and females modelled separately.

Table 3 NTR multivariable linear regression models (initial and fully adjusteda) with 3-hydroxybutyrate as the response variable, regression coefficients of aggression rating as predictor variable, with sexes pooled and separated.

Bayesian modelling

For all the fitted Bayesian models, the convergence checks revealed no autocorrelation and demonstrated the MCMC achieved the target distribution. In FinnTwin12, the initial non-informative priors model without the sex interaction term resulted in an aggression rating variable posterior mean of − 84.0 (95% credible interval: − 91.0, − 76.9), while the model with the sex × aggression interaction term indicated a posterior mean of − 86.6 (95% credible interval: − 100.4, − 72.5) for the interaction term. For the males-only model, the aggression posterior mean was − 37.9 (95% credible interval: − 43.8, − 32.0). For the females-only model, the aggression posterior mean was − 119.5 (95% credible interval: − 127.8, − 111.1). In NTR, the initial non-informative priors model without the sex interaction term resulted in an aggression rating variable posterior mean of 10.6 (95% credible interval: − 15.3, − 5.9), while the model with the sex × aggression interaction term indicated a posterior mean of 83.5 (95% credible interval: 72.5, 94.4) for the interaction term. For the males-only model, the aggression posterior mean was − 80.4 (95% credible interval: − 86.3, − 74.6). For the females-only model, the aggression posterior mean was 14.2 (95% credible interval: 9.9, 18.5). Finally, in the sensitivity analysis of NTR sex-stratified models using (FinnTwin12-based) informative priors for the sex-specific regression coefficients for aggression, the results remained similar (data not shown).

Discussion

In this study, we initially tested 11 potential NMR-processed low-molecular-weight biomarkers for association with a composite aggression score, and 3-hydroxybutyrate, a ketone body, was the only nominally statistically significant biomarker (negatively associated with aggression). Upon further analysis, in both basic and fully adjusted models, nearly all aggression ratings (from different raters and ages) showed a consistent pattern of negative association, with three being statistically significant. When including both teacher (age 12) and self (age 14) aggression ratings in the same model, both were significantly associated with 3-hydroxybutyrate, showing that an external rater (teacher) and an adolescent’s self-rating contribute uniquely to this association. Importantly, we were able to assess this biomarker–aggression association in a large and independent sample, the NTR. In both basic and fully adjusted NTR (frequentist) models, the negative association of 3-hydroxybutyrate and aggression was found, but statistical significance was not reached until the sexes were separated in the fully adjusted model (as indicated by the significant sex interaction). Finally, Bayesian modelling indicated a sex interaction in both FinnTwin12 and NTR cohorts, and the negative association—low 3-hydroxybutyrate levels with high aggression levels—remained in sex-stratified models (with non-informative priors) for both sexes in the FinnTwin12 cohort and for males only in the NTR cohort.

The association of 3-hydroxybutyrate with aggression is unlikely to be spurious because it was seen in two independent samples, and was consistent across raters and ages. Because we were not testing any specific hypothesis, and thus were not guided solely by statistical significance (p-values), our focus was on identifying repeated patterns of associations, which we saw here. Additionally, although we were not addressing causality in these models, we saw stability in the trend regardless of temporality (biomarker measured before or after aggression). Sex differences were evident, but these will need to be clarified by future investigations. Furthermore, the trend in FinnTwin12 was nearly uniform across raters and ages. This suggests that the association is related to generalized aggression (liability to behave aggressively in different contexts), which is a more serious and important type of aggression because it leads to the poorest outcomes51. Our model including multiple raters of aggression in the same model also supports this notion of generalized aggression, with ratings from both the teachers and adolescents themselves uniquely contributing to the association of aggression with 3-hydroxybutyrate. The teen him/herself provides their perspective about their behavior across all environments, while teachers have a unique perspective. Compared to parents for example, teachers regularly observe the child amongst other children of the same age, making for more valid predictions of the student’s liability to behave aggressively33,51. Combining both of these types of observers provides unique information, helping to clarify the breadth of this association of aggression with 3-hydroxybutyrate.

To our knowledge, this is the first time ketone bodies, and 3-hydroxybutyrate specifically, have been associated with aggression. In a urinary metabolite study with aggression in the NTR, no significant associations were found, but several near-significant results pointed to energy metabolism as a potential group of metabolites warranting further investigation22. Additionally, there are two recent studies showing a link between 3-hydroxybutyrate and depression, one in mice that was negatively correlated18 and one in humans that was positively correlated19. We know from previous studies that depression and aggression often co-occur25,52, and the concept of a ‘p factor’ that potentially underlies all psychopathology has been gaining traction53. However, our finding of a negative aggression–3-hydroxybutyrate relationship is in contrast with earlier reported positive depression–3-hydroxybutyrate relationships and warrants studying these traits separately in relation to metabolites.

In evaluating the biological plausibility of this association, we can consider generally what we know about this compound and how it might relate to aggression. During periods of exercise or fasting (low availability of glucose as an energy source), ketone bodies become involved in an alternative energy pathway41. 3-Hydroxybutyrate, in particular, has been shown to be not only involved in the alternative energy pathway, but it also has important cell signalling and regulatory functions, including pathways involved in reducing oxidative stress and inflammation (both linked to aggression7,54) and epigenetic modifications19,55. We also know that while 3-hydroxybutyrate is generally produced in the liver, it can cross the blood–brain barrier and act in the brain56,57, where we know that aggression is also regulated58,59. Finally, although there appear to be complex biological mechanisms at play, ketogenic diets (which increase ketone body levels) have been popular treatments in neurologic disorders, such as epilepsy, for decades60. The field of psychiatry has recently become more interested in the potential benefits of this diet, including in attention-deficit disorders and autism61, both of which can co-occur with aggression, however, aggression and the ketogenic diet has yet to be considered. Intriguingly, researchers have shown that a ketogenic diet with beta-hydroxybutyrate for male Drosophila with aggression after traumatic brain injury markedly reduced the levels of aggression62.

This study suggesting a new biomarker of aggression has important strengths in its large sample size, multiple ratings of aggression across stages of development, and the direction of association replicated in an independent sample. There are also limitations. First, we did not have aggression measured at the same time as the biological sample, in either dataset. However, it is known that aggression is a relatively stable trait, especially in aggressive males51. Although aggression levels in childhood and adolescence are generally higher than in adulthood, aggression levels are relatively stable in their rank order among similar aged peers63,64. There is longitudinal continuity in aggression from adolescence to adulthood33,51,65, which is the development period covered in the present study. Additionally, a fairly recent study with similar temporal data collection limitations, using childhood aggression measures (as developmental trajectories) and biological samples in young adulthood, was able to identify cytokines (a measure of inflammation) as a novel group of biomarkers for aggression7. Regarding biomarker stability, while we do see high heritability in the energy metabolism biomarkers (53.2% for 3-hydroxybutyrate)66, it would also be important in future studies to have repeated biological samples to confirm the stability of this biomarker over time, concurrent with the stability of aggression measures. However, repeated blood samples from children may be challenging to obtain, and thus this work might need to be done in young adults first, or using less invasive samples such as urine or saliva, which might produce different results than plasma (the recent urine biomarker study on NTR children did not show any significant findings with 3-hydroxybutyric acid and aggression)22.

In conclusion, this exploratory study using NMR-processed metabolites in plasma samples has revealed a potential new biomarker of aggression: 3-hydroxybutyrate, a ketone body related to an alternative energy pathway. We observed a negative association—high aggression levels with low ketone bodies levels—in our Finnish dataset, which was suggestively replicated in our independent Dutch dataset. There is biological plausibility for associating this ketone body with aggression, as well as the ability to potentially influence the biological levels of 3-hydroxybutyrate via diet modification. Thus, it is well-warranted to further investigate this epidemiological association at the biological level, hopefully shedding more light on the complexities and heterogeneity of the aggression phenotype.

Data availability

The datasets generated and/or analysed during the current study are not publicly available due to participant confidentiality, but are available from the corresponding author on reasonable request.

References

  1. 1.

    Fergusson, D. M., Horwood, L. J. & Ridder, E. M. Show me the child at seven: The consequences of conduct problems in childhood for psychosocial functioning in adulthood. J. Child Psychol. Psychiatry 46, 837–849. https://doi.org/10.1111/j.1469-7610.2004.00387.x (2005).

    Article  PubMed  Google Scholar 

  2. 2.

    Farrington, D. P. & Loeber, R. Epidemiology of juvenile violence. Child Adolesc. Psychiatr. Clin. N. Am. 9, 733–748 (2000).

    CAS  Article  Google Scholar 

  3. 3.

    Odintsova, V. V. et al. Genomics of human aggression: Current state of genome-wide studies and an automated systematic review tool. Psychiatr. Genet. 29, 170–190. https://doi.org/10.1097/YPG.0000000000000239 (2019).

    CAS  Article  PubMed  Google Scholar 

  4. 4.

    van Dongen, J. et al. Epigenome-wide association study of aggressive behavior. Twin Res. Hum. Genet. 18, 686–698. https://doi.org/10.1017/thg.2015.74 (2015).

    Article  PubMed  Google Scholar 

  5. 5.

    Cecil, C. A. M. et al. DRD4 methylation as a potential biomarker for physical aggression: An epigenome-wide, cross-tissue investigation. Am. J. Med. Genet. B Neuropsychiatr. Genet. 177, 746–764. https://doi.org/10.1002/ajmg.b.32689 (2018).

    CAS  Article  PubMed  Google Scholar 

  6. 6.

    Hagenbeek, F. et al. Discovery of biochemical biomarkers for aggression: A role for metabolomics in psychiatry. Am. J. Med. Genet. B Neuropsychiatr. Genet. 171, 719–732. https://doi.org/10.1002/ajmg.b.32435 (2016).

    Article  PubMed  Google Scholar 

  7. 7.

    Provencal, N., Suderman, M. J., Vitaro, F., Szyf, M. & Tremblay, R. E. Childhood chronic physical aggression associates with adult cytokine levels in plasma. PLoS ONE 8, e69481. https://doi.org/10.1371/journal.pone.0069481 (2013).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  8. 8.

    Kaplan, J. R. et al. Demonstration of an association among dietary cholesterol, central serotonergic activity, and social behavior in monkeys. Psychosom. Med. 56, 479–484. https://doi.org/10.1097/00006842-199411000-00001 (1994).

    CAS  Article  PubMed  Google Scholar 

  9. 9.

    Patti, G. J., Yanes, O. & Siuzdak, G. Innovation: Metabolomics: The apogee of the omics trilogy. Nat. Rev. Mol. Cell Biol. 13, 263–269. https://doi.org/10.1038/nrm3314 (2012).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  10. 10.

    Emwas, A. H. et al. NMR spectroscopy for metabolomics research. Metabolites https://doi.org/10.3390/metabo9070123 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  11. 11.

    Gebregiworgis, T. & Powers, R. Application of NMR metabolomics to search for human disease biomarkers. Combin. Chem. High Throughput Screen. 15, 595–610. https://doi.org/10.2174/138620712802650522 (2012).

    CAS  Article  Google Scholar 

  12. 12.

    Repo-Tiihonen, E., Halonen, P., Tiihonen, J. & Virkkunen, M. Total serum cholesterol level, violent criminal offences, suicidal behavior, mortality and the appearance of conduct disorder in Finnish male criminal offenders with antisocial personality disorder. Eur. Arch. Psychiatry Clin. Neurosci. 252, 8–11 (2002).

    Article  Google Scholar 

  13. 13.

    Hagenbeek, F. et al. Adult aggressive behavior in humans and biomarkers: A focus on lipids and methylation. J. Pediatr. Neonat. Individ. Med. 7, e070204 (2018).

    Google Scholar 

  14. 14.

    Ramsay, H. et al. Cognition, psychosis risk and metabolic measures in two adolescent birth cohorts—CORRIGENDUM. Psychol. Med. 48, 2628. https://doi.org/10.1017/S0033291718002192 (2018).

    Article  PubMed  Google Scholar 

  15. 15.

    Chen, X. et al. Dysregulation of amino acids and lipids metabolism in schizophrenia with violence. BMC Psychiatry 20, 97. https://doi.org/10.1186/s12888-020-02499-y (2020).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  16. 16.

    Coppola, A. et al. Branched-chain amino acids alter neurobehavioral function in rats. Am. J. Physiol. Endocrinol. Metab. 304, E405–E413. https://doi.org/10.1152/ajpendo.00373.2012 (2013).

    CAS  Article  PubMed  Google Scholar 

  17. 17.

    St-Jean, A. et al. Branched-chain and aromatic amino acids in relation to behavioral problems among young Inuit from Nunavik, Canada: A cohort study. Pediatr. Res. 82, 416–422. https://doi.org/10.1038/pr.2017.115 (2017).

    CAS  Article  PubMed  Google Scholar 

  18. 18.

    Chen, L., Miao, Z. & Xu, X. beta-hydroxybutyrate alleviates depressive behaviors in mice possibly by increasing the histone3-lysine9-beta-hydroxybutyrylation. Biochem. Biophys. Res. Commun. 490, 117–122. https://doi.org/10.1016/j.bbrc.2017.05.184 (2017).

    CAS  Article  PubMed  Google Scholar 

  19. 19.

    Setoyama, D. et al. Plasma metabolites predict severity of depression and suicidal ideation in psychiatric patients—A multicenter pilot analysis. PLoS ONE 11, e0165267. https://doi.org/10.1371/journal.pone.0165267 (2016).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  20. 20.

    Yang, J. et al. Landscapes of bacterial and metabolic signatures and their interaction in major depressive disorders. Sci. Adv. https://doi.org/10.1126/sciadv.aba8555 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  21. 21.

    Gulsun, M. et al. Possible relationship between amino acids, aggression and psychopathy. Int. J. Psychiatry Clin. Pract. 20, 91–100. https://doi.org/10.3109/13651501.2016.1144771 (2016).

    CAS  Article  PubMed  Google Scholar 

  22. 22.

    Hagenbeek, F. et al. Urinary amine and organic acid metabolites evaluated as markers for childhood aggression: The ACTION biomarker study. Front. Psychiatry 11, 165. https://doi.org/10.3389/fpsyt.2020.00165 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  23. 23.

    Kaprio, J. The Finnish Twin Cohort Study: An update. Twin Res. Hum. Genet. 16, 157–162. https://doi.org/10.1017/thg.2012.142 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  24. 24.

    Rose, R. J. et al. FinnTwin12 cohort: An updated review. Twin Res. Hum. Genet. 22, 302–311. https://doi.org/10.1017/thg.2019.83 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  25. 25.

    Bartels, M. et al. Childhood aggression and the co-occurrence of behavioural and emotional problems: Results across ages 3–16 years from multiple raters in six cohorts in the EU-ACTION project. Eur. Child Adolesc. Psychiatry 27, 1105–1121. https://doi.org/10.1007/s00787-018-1169-1 (2018).

    MathSciNet  Article  PubMed  PubMed Central  Google Scholar 

  26. 26.

    Bogl, L. H. et al. Abdominal obesity and circulating metabolites: A twin study approach. Metabolism 65, 111–121. https://doi.org/10.1016/j.metabol.2015.10.027 (2016).

    CAS  Article  PubMed  Google Scholar 

  27. 27.

    Soininen, P., Kangas, A. J., Wurtz, P., Suna, T. & Ala-Korpela, M. Quantitative serum nuclear magnetic resonance metabolomics in cardiovascular epidemiology and genetics. Circ. Cardiovasc. Genet. 8, 192–206. https://doi.org/10.1161/CIRCGENETICS.114.000216 (2015).

    CAS  Article  PubMed  Google Scholar 

  28. 28.

    Holmes, M. V. et al. Lipids, lipoproteins, and metabolites and risk of myocardial infarction and stroke. J. Am. Coll. Cardiol. 71, 620–632. https://doi.org/10.1016/j.jacc.2017.12.006 (2018).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  29. 29.

    Kettunen, J. et al. Genome-wide study for circulating metabolites identifies 62 loci and reveals novel systemic effects of LPA. Nat. Commun. 7, 11122. https://doi.org/10.1038/ncomms11122 (2016).

    ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

  30. 30.

    Pulkkinen, L., Kaprio, J. & Rose, R. J. Peers, teachers and parents as assessors of the behavioural and emotional problems of twins and their adjustment: The Multidimensional Peer Nomination Inventory. Twin Res. 2, 274–285. https://doi.org/10.1375/136905299320565762 (1999).

    CAS  Article  PubMed  Google Scholar 

  31. 31.

    Achenbach, T. M., McConaughy, S. H. & Howell, C. T. Child/adolescent behavioral and emotional problems: Implications of cross-informant correlations for situational specificity. Psychol. Bull. 101, 213–232. https://doi.org/10.1037/0033-2909.101.2.213 (1987).

    CAS  Article  PubMed  Google Scholar 

  32. 32.

    De Los Reyes, A. & Kazdin, A. E. Informant discrepancies in the assessment of childhood psychopathology: A critical review, theoretical framework, and recommendations for further study. Psychol. Bull. 131, 483–509. https://doi.org/10.1037/0033-2909.131.4.483 (2005).

    Article  PubMed  Google Scholar 

  33. 33.

    Whipp, A. M. et al. Early adolescent aggression predicts antisocial personality disorder in young adults: A population-based study. Eur. Child Adolesc. Psychiatry 28, 341–350. https://doi.org/10.1007/s00787-018-1198-9 (2019).

    Article  PubMed  Google Scholar 

  34. 34.

    Aaltonen, S. et al. Motor development and physical activity: A longitudinal discordant twin-pair study. Med. Sci. Sports Exerc. 47, 2111–2118. https://doi.org/10.1249/MSS.0000000000000650 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

  35. 35.

    Kujala, U. M. et al. Associations of aerobic fitness and maximal muscular strength with metabolites in young men. JAMA Netw. Open 2, e198265. https://doi.org/10.1001/jamanetworkopen.2019.8265 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  36. 36.

    Wurtz, P. et al. Metabolic profiling of alcohol consumption in 9778 young adults. Int. J. Epidemiol. 45, 1493–1506. https://doi.org/10.1093/ije/dyw175 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  37. 37.

    Xu, T. et al. Effects of smoking and smoking cessation on human serum metabolite profile: Results from the KORA cohort study. BMC Med. 11, 60. https://doi.org/10.1186/1741-7015-11-60 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  38. 38.

    Derks, I. P. M. et al. Testing bidirectional associations between childhood aggression and BMI: Results from three cohorts. Obesity (Silver Spring) 27, 822–829. https://doi.org/10.1002/oby.22419 (2019).

    Article  Google Scholar 

  39. 39.

    Pels, F. & Kleinert, J. Does exercise reduce aggressive feelings? An experiment examining the influence of movement type and social task conditions on testiness and anger reduction. Percept. Mot. Skills 122, 971–987. https://doi.org/10.1177/0031512516647802 (2016).

    Article  PubMed  Google Scholar 

  40. 40.

    Grant, J. D. et al. The role of conduct disorder in the relationship between alcohol, nicotine and cannabis use disorders. Psychol. Med. 45, 3505–3515. https://doi.org/10.1017/S0033291715001518 (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  41. 41.

    Laffel, L. Ketone bodies: A review of physiology, pathophysiology and application of monitoring to diabetes. Diabetes Metab. Res. Rev. 15, 412–426 (1999).

    CAS  Article  Google Scholar 

  42. 42.

    Turgay, A. Aggression and disruptive behavior disorders in children and adolescents. Expert Rev. Neurother. 4, 623–632. https://doi.org/10.1586/14737175.4.4.623 (2004).

    Article  PubMed  Google Scholar 

  43. 43.

    Ligthart, L. et al. The Netherlands Twin Register: Longitudinal research based on twin and twin-family designs. Twin Res. Hum. Genet. 22, 623–636. https://doi.org/10.1017/thg.2019.93 (2019).

    Article  PubMed  Google Scholar 

  44. 44.

    Achenbach, T. M., Ivanova, M. Y. & Rescorla, L. A. Empirically based assessment and taxonomy of psychopathology for ages 1(1/2)-90+ years: Developmental, multi-informant, and multicultural findings. Compr. Psychiatry 79, 4–18. https://doi.org/10.1016/j.comppsych.2017.03.006 (2017).

    Article  PubMed  Google Scholar 

  45. 45.

    Willemsen, G. et al. The Netherlands Twin Register biobank: A resource for genetic epidemiological studies. Twin Res. Hum. Genet. 13, 231–245. https://doi.org/10.1375/twin.13.3.231 (2010).

    Article  PubMed  Google Scholar 

  46. 46.

    Sirota, M. et al. Effect of genome and environment on metabolic and inflammatory profiles. PLoS ONE 10, e0120898. https://doi.org/10.1371/journal.pone.0120898 (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  47. 47.

    Auro, K. et al. A metabolic view on menopause and ageing. Nat. Commun. 5, 4708. https://doi.org/10.1038/ncomms5708 (2014).

    ADS  CAS  Article  PubMed  Google Scholar 

  48. 48.

    Williams, R. L. A note on robust variance estimation for cluster-correlated data. Biometrics 56, 645–646 (2000).

    CAS  Article  Google Scholar 

  49. 49.

    Heidari, S., Babor, T. F., De Castro, P., Tort, S. & Curno, M. Sex and Gender Equity in Research: Rationale for the SAGER guidelines and recommended use. Res. Integr. Peer Rev. 1, 2. https://doi.org/10.1186/s41073-016-0007-6 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  50. 50.

    Dunson, D. B. Commentary: Practical advantages of Bayesian analysis of epidemiologic data. Am. J. Epidemiol. 153, 1222–1226. https://doi.org/10.1093/aje/153.12.1222 (2001).

    CAS  Article  PubMed  Google Scholar 

  51. 51.

    Pulkkinen, L. In The Cambridge Handbook of Violent Behavior and Aggression 2nd edn (eds Flannery, D. J. et al.) 31–51 (Cambridge University Press, Cambridge, 2018).

    Chapter  Google Scholar 

  52. 52.

    Achenbach, T. M., Ivanova, M. Y., Rescorla, L. A., Turner, L. V. & Althoff, R. R. Internalizing/externalizing problems: Review and recommendations for clinical and research applications. J. Am. Acad. Child Adolesc. Psychiatry 55, 647–656. https://doi.org/10.1016/j.jaac.2016.05.012 (2016).

    Article  PubMed  Google Scholar 

  53. 53.

    Caspi, A. et al. The p factor: One general psychopathology factor in the structure of psychiatric disorders?. Clin. Psychol. Sci. 2, 119–137. https://doi.org/10.1177/2167702613497473 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

  54. 54.

    Carroll, J. E. et al. A urinary marker of oxidative stress covaries positively with hostility among midlife community volunteers. Psychosom. Med. 72, 273–280. https://doi.org/10.1097/PSY.0b013e3181d0d72b (2010).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  55. 55.

    Newman, J. C. & Verdin, E. beta-hydroxybutyrate: Much more than a metabolite. Diabetes Res. Clin. Pract. 106, 173–181. https://doi.org/10.1016/j.diabres.2014.08.009 (2014).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  56. 56.

    Guzman, M. & Blazquez, C. Ketone body synthesis in the brain: Possible neuroprotective effects. Prostaglandins Leukot. Essent. Fatty Acids 70, 287–292. https://doi.org/10.1016/j.plefa.2003.05.001 (2004).

    CAS  Article  PubMed  Google Scholar 

  57. 57.

    Sleiman, S. F. et al. Exercise promotes the expression of brain derived neurotrophic factor (BDNF) through the action of the ketone body beta-hydroxybutyrate. Elife https://doi.org/10.7554/eLife.15092 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  58. 58.

    Ilchibaeva, T. V. et al. Brain-derived neurotrophic factor (BDNF) and its precursor (proBDNF) in genetically defined fear-induced aggression. Behav. Brain Res. 290, 45–50. https://doi.org/10.1016/j.bbr.2015.04.041 (2015).

    CAS  Article  PubMed  Google Scholar 

  59. 59.

    Seo, D., Patrick, C. J. & Kennealy, P. J. Role of serotonin and dopamine system interactions in the neurobiology of impulsive aggression and its comorbidity with other clinical disorders. Aggress. Violent Behav. 13, 383–395. https://doi.org/10.1016/j.avb.2008.06.003 (2008).

    Article  PubMed  PubMed Central  Google Scholar 

  60. 60.

    Cullingford, T. E. The ketogenic diet; fatty acids, fatty acid-activated receptors and neurological disorders. Prostaglandins Leukot. Essent. Fatty Acids 70, 253–264. https://doi.org/10.1016/j.plefa.2003.09.008 (2004).

    CAS  Article  PubMed  Google Scholar 

  61. 61.

    Bostock, E. C., Kirkby, K. C. & Taylor, B. V. The current status of the ketogenic diet in psychiatry. Front. Psychiatry 8, 43. https://doi.org/10.3389/fpsyt.2017.00043 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

  62. 62.

    Lee, D. C. et al. Dietary supplementation with the ketogenic diet metabolite beta-hydroxybutyrate ameliorates post-TBI aggression in young-adult male Drosophila. Front. Neurosci. 13, 1140. https://doi.org/10.3389/fnins.2019.01140 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  63. 63.

    Huesmann, L. R., Eron, L. D., Lefkowitz, M. M. & Walder, L. O. Stability of aggression over time and generations. Dev. Psychol. 20, 1120–1134 (1984).

    Article  Google Scholar 

  64. 64.

    Tremblay, R. The development of aggressive behaviour during childhood: What have we learned in the past century?. Int. J. Behav. Dev. 24, 124–141 (2000).

    Article  Google Scholar 

  65. 65.

    Porsch, R. M. et al. Longitudinal heritability of childhood aggression. Am. J. Med. Genet. B Neuropsychiatr. Genet. 171, 697–707. https://doi.org/10.1002/ajmg.b.32420 (2016).

    Article  PubMed  Google Scholar 

  66. 66.

    Kettunen, J. et al. Genome-wide association study identifies multiple loci influencing human serum metabolite levels. Nat. Genet. 44, 269–276. https://doi.org/10.1038/ng.1073 (2012).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

FinnTwin12 wishes to thank all participating twins, their parents and teachers. Data collection in FinnTwin12 was supported by NIAAA12502, AA-00145, AA-08315 (to RJR) and the Academy of Finland (Grants 100499, 205585, 118555, 141054 and 264146 to JK). JK has been supported by the Academy of Finland (Grants 308248 and 312073). EV has been supported by the Academy Research Fellow funding from the Academy of Finland (Grant 314639). TK was supported by the Academy of Finland (Grant 1309119). This project has also received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement no. 874724. NTR warmly thanks all twins and family members for their participation. NTR data collection and analyses were supported by the Netherlands Organization for Scientific Research: Netherlands Twin Registry Repository: researching the interplay between genome and environment (480-15-001/674); the European Union Seventh Framework Program (FP7/2007-2013): ACTION Consortium (Aggression in Children: Unravelling gene-environment interplay to inform Treatment and InterventiON strategies; grant number 602768) and Biobanking and Biomolecular Resources Research Infrastructure Netherlands (BBMRI-NL: 184.021.007 and 184.033.111). Additionally, we wish to thank Teemu Palviainen M.Sc. for his statistical help, and Dr. Mariëlle Zondervan-Zwijnenburg for her advice regarding Bayesian modelling.

Author information

Affiliations

Authors

Contributions

A.M.W., D.I.B., and J.K. made substantial contributions to the conception/design of the study; D.I.B., J.K., L.P., M.B., and R.J.R. made substantial contributions to the acquisition of the data; A.M.W., E.V., J.K., L.H.B., R.P., A.B., L.L., and T.K. made substantial contributions to the analysis of the data; A.M.W., D.I.B., E.V., F.A.H., J.K., L.H.B., R.P., A.B., L.L., and T.K. made substantial contributions to the interpretations of the data; A.M.W. drafted the manuscript; and all authors made substantial contributions to the revision of the manuscript draft.

Corresponding author

Correspondence to A. M. Whipp.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher's note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Whipp, A.M., Vuoksimaa, E., Korhonen, T. et al. Ketone body 3-hydroxybutyrate as a biomarker of aggression. Sci Rep 11, 5813 (2021). https://doi.org/10.1038/s41598-021-84635-6

Download citation

Further reading

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Search

Quick links