Children sleep and antisocial behavior: Differential association of sleep with aggression and rule-breaking

Abstract

There is a strong relationship between sleep and behavioral problems. These findings are often interpreted via environmental explanations, such that poor sleep directly exacerbates or causes symptoms of aggression and behavior problems. However, there are other possible explanations, such that the genes predicting poor sleep also predict aggression or rule-breaking. The current study sought to elucidate the origin of this relationship. The sample was composed of 1,030 twin pairs (426 monozygotic and 604 dizygotic). The sample was 51.3% male with a mean age of 8.06 years (range 6-11.96; SD=1.45). Aggression, rule-breaking and sleep were assessed through the Child Behavior Checklist (CBCL). We fitted bivariate Cholesky genetic models to the data, decomposing the variance within, and the covariance among, aggression, rule-breaking, and sleep functioning into their genetic and environmental components. Genetic correlations between all sleep variables and aggression were significant and moderate to large in magnitude, but mostly small and non-significant between sleep and rule-breaking. We did not find evidence of a causal or environmental relationship between the majority of sleep variables and aggression, but rather clear evidence of genetic pleiotropy. However, the pattern of associations between rule-breaking and sleep measures was less consistent. Aggression and rule-breaking appear to be differentially associated with sleep.

Introduction

Sleep quality is critical for health and cognitive functioning, predicting obesity, cardiovascular diseases and psychopathology, among other things (Buysse et al., 2008; Cappuccio et al., 2008; Medic, Wille, & Hemels, 2017). Sleep quality may be especially important in the case of children, since sleep is critical for brain development and memory processes (Diekelmann, Wilhelm, & Born, 2009; Picchioni, Reith, Nadel, & Smith, 2014). Consistent with this, a number of studies have reported that fewer hours of sleep and sleep problems are associated with reduced academic performance, poor school adjustment, psychopathology, and behavioral disinhibition (Dahl & Lewin, 2002; Dimitriou, Le Cornu Knight, & Milton, 2015; Williams, Nicholson, Walker, & Berthelsen, 2016). Additionally, there is a robust relationship between child sleep problems and increased aggression (AGG) (Gregory & O'Connor, 2002) and behavioral problems (Hoedlmoser, Kloesch, Wiater, & Schabus, 2010). Moreover, this association appears to persist over time. Sivertsen et al. (2015) found that short sleep duration and frequent nocturnal awakenings at 18 months are significant predictors for the development of emotional and behavioral problems at 5 years.

Such findings are often interpreted to indicate a causal pathway, such that sleep problems exacerbate or give rise to symptoms of AGG and behavior problems. However, this is only one of the possible explanations for the association. Another possibility, which is just as consistent with published phenotypic associations, is that poor sleep and antisocial behavior are associated via shared genes. This possibility is bolstered by prior studies demonstrating an influence of genetic factors on sleep duration, with heritability estimates ranging from .31 to .71 (Brescianini et al., 2011; Gregory, Rijsdijk, & Eley, 2006). Given the extensive data documenting genetic influences on antisocial behavior (Burt, 2009), it is thus entirely possible that the association between antisocial behavior and sleep is at least partially genetic in origin. There are several neural mechanisms that could underlie a genetically-mediated relationship. For example, sleep problems may stem in part from cortical functioning (e.g., poor functioning of prefrontal cortex), which has a well-documented link with antisocial behavior (Hyde, Shaw, & Hariri, 2013). Alternately, antisocial behavior may be related to serotonin levels which are important for the wake-sleep cycle, or to the functioning of the hypothalamic-pituitary-adrenal axis, where sleep has an inhibitory effect (Kamphuis, Meerlo, Koolhaas, & Lancel, 2012). To date, however no joint analyses using a genetically informative sample have yet been conducted. As such, this possibility remains untested.

The current study aims to do just this, evaluating the genetic and environmental origins of the association between different sleep features and childhood antisocial behavior for the very first time. In doing so, we also attended to the heterogeneity within the broader construct of antisocial behavior. As reviewed at length in Burt (2012), there is now substantial evidence that AGG (i.e., bullying, assault, fighting) and non-aggressive rule-breaking (RB) (i.e., theft, vandalism, lying) are distinguishable but related dimensions of antisocial behavior (Frick et al., 1993; Loeber & Schmaling, 1985; J. Tackett, Krueger, Sawyer, & Graetz, 2003; J. L. Tackett, Krueger, Iacono, & McGue, 2005), characterized by distinct etiologies, developmental trajectories, and demographic and phenotypic correlates. For example, AGG demonstrates higher rank-order stability than does RB (Stanger, Achenbach, & Verhulst, 1997; Tremblay, 2003) and appears to be a particularly heritable form of antisocial behavior (65% of the variance in AGG is genetic in origin, as compared to 48% in RB (Burt, 2009). Moreover, the genetic influences on AGG and RB are at least partially distinct from one another (Burt, 2013). Genetic influences on AGG also appear to remain stable across mid-childhood and adolescence while genetic influences on RB increase substantially during adolescence (Burt & Klump, 2009; Burt & Neiderhiser, 2009). AGG is also more strongly associated with the tendency to experience negative emotions (Burt & Donnellan, 2008; Burt & Larson, 2007) and low empathy (Cohen & Strayer, 1996; Pardini, Lochman, & Frick, 2003). RB, by contrast, is more strongly associated with impulsivity (Burt & Donnellan, 2008). Finally, joint analyses have indicated that while AGG is associated with negative parenting via both genetic and environmental factors, the association between RB and negative parenting is entirely environmental in origin (Burt, 2009).

Given the many developmental and etiologic differences between the aggressive and RB dimensions of antisocial behavior, it is quite possible that the two dimensions are also differentially associated with sleep functioning at an etiologic level. The current study aimed to evaluate and contrast, in a large sample of twins in middle childhood, the phenotypic and etiologic associations of AGG and RB with multiple measures of sleep.

Materials and Methods

Participants

The sample was composed of 2,060 twin children in 1,030 families recruited as part of the Twin Study of Behavioral and Emotional Development in Children (TBED-C), a study within the Michigan State University Twin Registry (MSUTR) (Burt & Klump, 2013; Klump & Burt, 2006). The TBED-C project contains two independent sub-samples. The first sub-sample is population-based, and is composed of 1,056 twins belonging to 528 families recruited from across lower Michigan. The second, ‘at-risk’ sub-sample is composed of 1,004 twins from 502 families residing in modestly to severely disadvantaged neighborhoods in the same recruitment area. To be eligible for the TBED-C, neither twin could have a physical or cognitive impairment (e.g., a significant developmental delay) that would interfere with their in-person assessment. Ethics approval was provided by the Michigan State University IRB. Parent(s) provided an informed consent for both their children and themselves. The twins provided informed assent.

The Department of Vital Records in the Michigan Department of Health and Human Services (formerly, the Michigan Department of Community Health) identified twins in our age-range either directly from birth records or via the Michigan Twins Project, a large-scale population-based registry of twins in lower Michigan that were themselves recruited via birth records. The Michigan Bureau of Integration, Information, and Planning Services database was used to locate family addresses within 120 miles of East Lansing, MI through parent drivers’ license information. Pre-made recruitment packets were then mailed to parents on our behalf. A reply postcard was included for parents to indicate their interest in participating. Interested families were contacted directly by project staff. Parents who did not respond to the first mailing were sent additional mailings approximately one month apart until either a reply was received or up to four letters had been mailed.

The participation rate for the population-based and at-risk sub-samples were 62% and 56%, respectively, which are similar to or better than other twin registries that use anonymous recruitment mailing (Baker, Barton, & Raine, 2002; Hay, McStephen, Levy, & Pearsall-Jones, 2002). Both sub-samples were representative of recruited families (as assessed via a brief questionnaire screen administered to nearly 85% of non-participating families) (for details, please see Burt & Klump, 2013).

The mean age of the overall sample was 8.06 years (range 6-11.96; SD=1.45). Parent-reports of twin ethnicity were primarily White non-Hispanic (81.7%), followed by African-American (9.5%), Native American (1.1%), Asian (0.8%), Latinx (0.7%) and other (6.2%). The twins were 51.3% male. Twin zygosity was established using a standard 5-item questionnaire that assesses within-pair physical similarity and is over 95% accurate when compared to DNA (Peeters, Van Gestel, Vlietinck, Derom, & Derom, 1998). 426 pairs were monozygotic (MZ) and 604 pairs were dizygotic (DZ).

Measures

Mothers completed the Child Behavior Checklist (CBCL; Achenbach & Rescorla, 2001) separately for each twin. Mothers were asked to rate the extent to which a series of statements described each of their children’s behavior over the past six months using a three-point scale (0=never to 2=often/mostly true). We utilized the well-known aggressive (e.g., ‘bullied’, ‘cruel’, ‘gets in fights’, ‘attacks people’; 18 items; α =.88) and rule-breaking (e.g., ‘cheat or lie’, ‘breaks rules’, ‘no guilt’, ‘steals’; 17 items; α =.65) scales. Higher scores reflect endorsement of more aggressive and/or RB behaviors. Consistent with recommendations in the manual (Achenbach & Rescorla, 2001), analyses were conducted on the raw scale scores.

Children’s sleep functioning was assessed using the individual CBCL sleep items: ‘Sleep less than most kids’; ‘Nightmares’; ‘Overtired without good reason’; ‘Sleep more than most kids during day and/or night’; ‘Talks or walks in sleep’; and ‘Trouble sleeping’ (0=never, 1=somewhat or sometimes true, 2=often/mostly true). Since response option 2 was infrequently endorsed (ranging from 0.5 to 3.3%), we collapsed the three possible responses into a single category for each item, effectively dichotomizing our sleep variables (0=Not true; 1= True). CBCL sleep items have shown convergent validity with validated sleep measures and sleep disorder diagnoses and may be useful individually when assessing specific facets of sleep functioning, especially items such as ‘trouble sleeping’ (Becker, Ramsey, & Byars, 2015; Gregory et al., 2011). Moreover, these set of items have been often used with this purpose with similar age samples (Gregory & O'Connor, 2002; Pirinen, Kolho, Simola, Ashorn, & Aronen, 2010).

Statistical analyses

Behavioral genetic analyses make use of the difference in the proportion of genes shared between reared-together siblings: MZ twins (who share 100% of their genetic material) and DZ twins (who share an average of 50% of their segregating genetic material). Utilizing these differences, the variance within observed behaviors or characteristics (i.e., phenotypes) can be partitioned into different components, additive genetic, non-additive genetic, shared environment, and non-shared environment plus measurement error. The additive genetic component (a²) is the effect of individual genes summed over loci, and acts to increase twin correlations relative to the proportion of genes shared. The shared environment (c²) is that part of the environment common to siblings that acts to make them similar to each other, regardless of the proportion of genes shared. Non-additive genetic influences (d²) capture interactive effects between genes, including genetic dominance and, possibly, epistasis. The non-shared environment (e²) encompasses environmental factors (and measurement error) differentiating twins within a pair. It is not possible to estimate C and D at the same time because they are negatively correlated. The pattern of twin correlations indicates the most suitable model: ACE or ADE. More information on genetically-informative studies is provided elsewhere (Knopik, Neiderhiser, DeFries, & Plomin, 2016).

Full univariate ACE/ADE models were first fitted to the data for each of the two quantitative variables (AGG and RB) and the six dichotomous variables (sleep variables), so as to evaluate the baseline etiology of each trait. AE, CE and E nested submodels were also evaluated. The fit of the different models and submodels was compared using the likelihood-ratio chi-square test. The difference in minus two times the log-likelihood (−2LL) between two models has a χ2 distribution with the degrees of freedom (df) equaling the difference in df between the two models. Model fit was also evaluated using Akaike’s information criterion (AIC; Akaike, 1987). A lower AIC value would indicate a better fit to the data.

We then fitted bivariate correlated factors models between AGG or RB and each of the sleep variables, respectively. Fitting correlated factors models (see Figure 1) with Open Mx allowed us to examine genetic and environmental contributions to both individual variance and to sources of covariance among aggressive behavior or RB behavior, and each of the sleep variables. Specifically, the correlated factors model parses the phenotypic variance of each variable, and the covariances between pairs of phenotypes, into that which is due to genetic, shared environmental, and non-shared environmental factors. The covariances can then be standardized on their respective variances to produce genetic, shared environmental and non-shared environmental correlations between each pair of phenotypes. These statistics reveal the extent to which a specific effect (e.g., the additive genetic effect) on one variable is correlated with the same effect on another variable. In these ways, the correlated factors models allowed us to make inferences regarding the sources of co-occurrence between each variable of sleep quality and each of the two forms of antisocial behavior.

Fig 1.

Bivariate correlated factors model

Log transformations (log10 X + 1) were performed for AGG and RB in order to adjust for positive skew. All the analyses were run in R v3.3.2 (R Core Team, 2016), and models were fitted using the raw data with the OpenMx package v2.8.3 (Neale et al., 2016).

Results

Total scores on the AGG scale ranged from 0-31 (Mean=4.53; SD=4.92), while scores for RB ranged from 0-14 (Mean= 1.51; SD=1.92). As shown in Table 1 and Figure 2, those children with sleep features lying outside of the norm evidenced notably higher mean levels of both AGG and RB as compared to children with adequate sleep. All the comparisons were significant and Cohen’s d ranged from 0.28 to 0.77 for AGG and 0.25 to 0.64 for RB.

Table 1:

Means comparison in aggression and rule breaking for each sleep variable

		No	Yes	P	Cohen’s d
Aggression
	Sleep less than most kids	4.18	8.11	<0.001	0.63
	Nightmares	3.99	6.29	<0.001	0.42
	Overtired	4.29	9.45	<0.001	0.77
	Sleep more than most kids	4.44	6.24	0.01	0.30
	Talks or walks	4.26	5.80	<0.001	0.28
	Trouble sleeping	4.13	7.44	<0.001	0.57
Rule-Breaking
	Sleep less than most kids	1.39	2.74	<0.001	0.56
	Nightmares	1.31	2.17	<0.001	0.39
	Overtired	1.44	3.11	<0.001	0.64
	Sleep more than most kids	1.46	2.61	<0.001	0.40
	Talks or walks	1.42	1.95	<0.001	0.25
	Trouble sleeping	1.41	2.30	<0.001	0.39

Fig 2.

Difference (z-scores) in aggression and rule-breaking behavior between children with problematic and adequate sleep.

* Z-score difference: mean z-score of the problematic sleep group minus mean z score of the adequate sleep group.

Univariate analysis

The MZ and DZ intraclass correlations for AGG, RB, and sleep variables are presented in Table 2, together with the results of univariate ACE/ADE models. As seen in the table, the AE model provided the best fit to the AGG data and most of the sleep variables. For RB, by contrast, the ACE model provided the best fit, consistent with prior studies pointing to a clear contribution of the shared environment to RB in particular (Burt, 2009). Finally, for the variable ‘sleep more than other kids’ the best fit was provided by an ADE model. All phenotypes were highly heritable: .65 for AGG, .53 for RB, and ranging from .62 to .89 for sleep variables.

Table 2:

Univariate analyses

			Standardized Estimates (95% CI)
Model	rMZ	rDZ	a2	c2 /d2	e2	df	−2LL	AIC	Δχ2	Δdf	p
Aggression	0.63	0.35
ACE			0.59 (0.43, 0.69)	0.05 (0, 0.19)	0.36 (0.31, 0.41)	2028	6430.12	2374.12	-	-	-
AE			0.65 (0.60,0.69)	-	0.35 (0.31, 0.40)	2029	6430.58	2372.58	0.47	1	0.49
CE			-	0.47 (0.42, 0.51)	0.53 (0.49, 0.58)	2029	6477.73	2419.73	47.62	1	<0.001
E			-	-	1	2030	6726.16	2666.16	295.58	1	<0.001
Rule-breaking	0.68	0.46
ACE			0.53 (0.39, 0.67)	0.18 (0.05, 0.30)	0.29 (0.25, 0.33)	2028	4998.57	942.57	-	-	-
AE			0.72 (0.68, 0.76)	-	0.28 (0.24, 0.32)	2029	5006.04	948.04	7.47	1	<0.001
CE			-	0.55 (0.50, 0.59)	0.45 (0.41, 0.50)	2029	5050.96	992.96	52.39	1	<0.001
E			-	-	1	2030	5413.41	1353.41	407.37	1	<0.001
Sleep less than most kids	0.82	0.40
ADE			0.71 (0, 0.93)	0.15 (0 ,0.90)	0.14 (0.06, 0.28)	2026	1136.91	−2915.10	-	-	-
AE			0.85 (0.72,0.93)	-	0.15 (0.07,0.28)	2027	1137.08	−2916.92	0.17	1	0.68
E			-	-	1	2028	1206.84	−2849.16	69.76	1	<0.001
Nightmares	0.70	0.47
ACE			0.47 (0.14, 0.78),	0.23 (0, 0.49)	0.30 (0.20, 0.42)	2029	2088.30	−1969.70
AE			0.73 (0.63, 0.82)	*	0.27 (0.18, 0.37)	2030	2090.91	−1969.09	2.61	1	0.11
CE			*	0.57 (0.48, 0.65)	0.43 (0.35, 0.52)	2030	2095.95	−1964.05	7.66	1	<0.001
E			*	*	1 (1, 1)	2031	2211.56	−1850.44	120.65	1	<0.001
Overtired	0.80	0.43
ACE			0.82 (0.23, 0.94)	0.01 (0, 0.47)	0.17 (0.06, 0.38)	2028	724.39	−3331.61
AE			0.83 (0.64, 0.94)	*	0.17 (0.06, 36)	2029	724.40	−3333.60	<0.001	1	0.96
CE			*	0.59 (0.41, 0.73)	0.41 (0.27, 0.59)	2029	731.51	−3326.49	7.12	1	0.007
E			*	*	1 (1, 1)	2030	765.88	−3294.12	41.49	1	<0.001
Sleep more than most kids	0.89	0.02
ADE			0 (0, 0)	0.89 (0.22, 0.96)	0.11 (0.03, 0.27)	2027	722.77	−3331.23
AE			0.86 (0.65, 0.96)	*	0.14 (0.04, 0.35)	2028	728.60	−3327.40	5.83	1	0.02
E			*	*	1 (1, 1)	2029	776.68	−3291.22	38.17	1	<0.001
Talks or walks	0.73	0.33
ADE			0.61 (0, 0.81)	0.11 (0, 0.78)	0.28 (0.18, 0.40)	2028	1793.16	−2262.84
AE			0.72 (0.60, 0.81)	*	0.28 (0.19, 0.40)	2029	1793.26	−2264.74	0.10	1	0.75
E			*	*	1 (1, 1)	2030	1889.14	−2170.86	95.88	1	<0.001
Trouble sleeping	0.65	0.19
ADE			0.06 (0, 0.09)	0.62 (0, 0.82)	0.31 (0.18, 0.49)	2027	1450.80	−2603.20
AE			0.62 (0.44, 0.77)	*	0.38 (0.23, 0.56)	2028	1453.30	−2602.70	2.51	1	0.11
E			*	*	1 (1, 1)	2029	1490.63	−2567.37	37.33	1	<0.001

a², d², c², e², proportions of variance explained by additive genetic, non-additive genetic, shared and unique environmental effects, respectively; −2LL, negative 2 log-likelihood; AIC, Akaike’s information criterion; CI, confidence interval; df, degrees of freedom; Δχ², model fit statistic: difference in −2LL of two nested models; Δdf, difference in the number of parameters between the models; p, p-value of the chi-square-test; rDZ, dizygotic twin correlations; rMZ, monozygotic twin correlations. Bold text indicates best fitting models

Multivariate analyses

Results pointed to a strong genetic correlation between AGG and RB (rG=.76; 95% CI: 0.67, 0.84), albeit one that was significantly lower than unity (the 95% CIs did not overlap with 1.0). Such results indicate that, while AGG and RB share many of their genetic influences, some are dimension-specific. Phenotypic, genetic and environmental correlations obtained through the twelve bivariate models between AGG/RB and the six sleep measures are presented in Table 3. Interestingly, there was evidence that the associations of sleep with AGG and RB have different etiologies.

Table 3:

Phenotypic, genetic and environmental correlations between aggression, rule-breaking behavior and sleep measures from the bivariate analyses

Phenotypes (model)	Phenotypic correlation	rG	rC	rE
Aggression with
Sleep less than most kids (AE)	0.38	0.64 (0.50, 0.80)	*	−0.27 (−0.51, 0.01
Nightmares (AE)	0.28	0.40 (0.30, 0.51)	*	0.02 (−0.14, 0.18)
Overtired (AE)	0.42	0.58 (0.42, 0.76)	*	0.02 (−0.29, 0.36)
Sleep more than most kids (AE)	0.17	0.42 (0.24, 0.63)	*	−0.47 (−0.70, −0.16)
Talks or walks (AE)	0.15	0.27 (0.15, 0.38)	*	−0.10 (−0.27, 0.08)
Trouble sleeping (AE)	0.34	0.56 (0.41, 0.74)	*	−0.01 (−0.19, 0.19)
Rule-Breaking with
Sleep less than most kids (ACE)	0.32	0.24 (−0.07, 0.49)	1 (0.58, 1)	−0.03 (−0.30, 0.25)
Nightmares (ACE)	0.27	0.06 (−0.27, 0.42)	0.99 (0.37, 1)	0.09 (−0.08, 0.26)
Overtired (ACE)	0.36	0.13 (−0.54, 0.44)	1 (0.45, 1)	0.28 (−0.04, 0.60)
Sleep more than most kids (AE)	0.23	0.40 (0.25, 0.57)	*	−0.30 (−0.57, 0.02)
Talks or walks (ACE)	0.17	0.01 (−0.31, 0.27)	1 (0.36, 1)	−0.01 (−0.18, 0.18)
Trouble sleeping (ACE)	0.23	0.46 (0.18, 0.84)	1 (−1, 1)	−0.25 (−0.44, −0.05)

rC: shared environment correlation; rE: non-shared environmental correlation: rG: genetic correlation. Bold text indicates significant results.

Aggression and sleep:

AE models provided the best fit in all cases. These results were expected since C was not significant for AGG or any of the sleep variables. The phenotypic correlations ranged from 0.15 (‘talks or walks’) to 0.42 (‘overtired’). All of the genetic correlations between the sleep variables and AGG were significant, ranging from 0.27 to 0.64. The stronger genetic association was found with ‘sleep less than most kids’ (0.64, 95% CI 0.50,0.80), followed by ‘overtired’ (0.58, 95% CI 0.42,0.76). The weakest one was found with ‘talks or walks’ (0.27, 95% CI 0.15,0.38), although still significant. By contrast, the non-shared environmental correlations were not significant except for ‘sleep more than most kids’ where a negative correlation was found (−0.47, 95% CI −0.70,−0.16).

Rule-Breaking and sleep:

ACE models fitted best to the data in this case, except for ‘overtired’ where an AE model was selected. Genetic correlations were mostly low and non-significant. Only ‘sleep more than most kids’ (0.40, 95% CI 0.25,0.57) and ‘trouble sleeping’ (0.46, 95% CI 0.18,0.84) showed a genetic overlap with RB. Shared environmental correlations were very high, but not relevant, given the lack of detectable levels of C for sleep variables. Among the environmental correlations, only ‘trouble sleeping’ showed a significant association (−0.25, 95% CI −0.44,−0.05) with RB.

Discussion

The main objective of this article was to investigate the relationship between different measures of sleep and antisocial behavior in children. We also delved into the possibility of differential relationships between the sleep variables and the two primary dimensions of antisocial behavior. Our results confirm that there is a close relationship between sleep and antisocial behavior in children. All measures of sleep showed a significant phenotypic association with both AGG and RB, such that children who experience more problematic sleep were reportedly engaging in significantly higher AGG and RB. The largest effect size was specifically observed for ‘overtired’ and ‘sleep less than most kids’, a pattern that held for both for AGG and RB.

Our results also supported the presence of common genetic pathway(s) between aggression and poor sleep in children, as well as the presence of differential etiologic overlap for AGG and RB, respectively. Results pointed to significant genetic overlap between sleep problems and AGG, for all six indices of sleep, indicating an important genetic overlap between all dimensions of children sleep and AGG. In comparison, only two variables (‘trouble sleeping’ and ‘sleep more than most kids’) evidenced significant genetic correlations with RB. Put differently, the association between sleep and AGG appears to be largely genetic in origin, while that between RB and sleep does not. Moreover, our finding of strong rA and non-significant rE between most of the sleep indices and AGG is not consistent with a causal association between sleep and AGG (in which poor sleep increases AGG), but rather with genetic pleiotropy, in which a given set of genes is associated with more than one outcome.

That said, two of the CBCL sleep items appeared to behave somewhat differently. There was a significant genetic correlation between ‘trouble sleeping’ and both dimensions of antisocial behavior, but only RB showed also a significant environmental correlation. Something similar occurred for ‘sleep more than most kids’, although in this case the dimension that correlated environmentally was AGG. These results can be taken as further evidence of differences in the kind of relationship between sleep and AGG or RB. They could also be interpreted as possible indication of some causality between sleep and antisocial behavior, whereby the environmental factors affecting those specific aspects of sleep may shape antisocial behavior. Critically, however, the direction of these non-shared environmental correlations is in the opposite direction as would be consistent with a causal explanation, undercutting that interpretation.

To our knowledge, this is the first study to investigate the relationship between sleep functioning and AGG and RB from a genetically informed point of view, which makes comparison with other studies difficult. That said, the children in our sample who exhibit altered sleeping features had higher mean levels in AGG and RB than those children with typical sleep functioning. These results are consistent with previous studies in which sleep problems were related with an increased probability of physical and verbal AGG, hostility and anger (Randler & Vollmer, 2013), even in longitudinal studies (Gregory & O'Connor, 2002; Sivertsen et al., 2015). Our findings were also consistent with prior studies demonstrating that AGG and RB are distinguishable (if strongly correlated) dimensions of antisocial behavior (Frick et al., 1993; Loeber & Schmaling, 1985; J. Tackett et al., 2003; J. L. Tackett et al., 2005), The best fitting model for AGG specified additive genetic and unique environmental contributions, while that for RB included a common environmental component as well. As has been discussed at length in prior work (e.g., Burt, 2009, 2012), such findings bolster prior conclusions that AGG and RB evidence meaningful etiologic differences, despite the fact that both constitute a form of antisocial behavior. The current findings also further these conclusions, in that they also highlight differences between AGG and RB in the origin of their association with sleep problems.

Our univariate analyses similarly demonstrated that our heritability values are similar to those of previous studies, especially for RB and AGG (Chen, Yu, Zhang, Li, & McGue, 2015; Porsch et al., 2016). For sleep duration, our estimate of genetic influence was on the higher end of the wide span of heritability estimates for children found in the literature, where values from .31 to .71 have been reported (Brescianini et al., 2011; Gregory et al., 2006). These differences could be due to various methodological characteristics of the study. Age, for example, could be a relevant factor since evidence to date indicates that genetic influences are relative low in the toddler years (Brescianini et al., 2011), and increase up to .71 in school-aged children (Gregory et al., 2006) (the latter of which is on par with our estimates).

This study has several strengths. First, we had a large sample, providing us with ample statistical power for our analyses. Moreover, the use of twins allowed us to disentangle the genetic and environmental influences to the studied phenotypes. Our use of a set of bivariate models allowed us to investigate genetic and environmental influences on each phenotype, and on the overlap between pairs of phenotypes. Moreover, we used a set of sleep variables which gives us a more complete picture of child sleep than using only one, such as duration. On the other hand, our study has several limitations. The method to evaluate sleep was based on single questions with three response options, an overly simplistic scale that may have minimized the ascertained variation in sleep problems Furthermore, all of our measures are mother-reported which may influence the associations amount them. However, sleep questions from the CBCL have been previously used and demonstrated convergent validity with validated scales and sleep disorders diagnoses (Becker et al., 2015). Actually, the item ‘trouble sleeping’ is strongly correlated with sleep latency measured both through diary and actigraphy (Gregory et al., 2011). In addition, sleep patterns of twins and singletons could show differences: for example twins usually share the same room and noise from one could wake the other (Gregory, Eley, O'Connor, & Plomin, 2004). Available work, however, has indicated that there are no differences between singletons and twins in their sleep patterns (Bartlet & Witoonchart, 2003), and, in any case, this question would not affect our results since this problem would affect twins equally, regardless of their zygosity. Finally, these findings are specific to non-clinical populations and should not be generalized to specific clinical populations; e.g., pediatric bipolar disorder (Hernandez, Marangoni, Grant, Estrada, & Faedda, 2017).

Conclusion

The current results strongly suggest that common genes demonstrate pleiotropic effects over sleep and AGG. Put another way, poor sleep does not appear to cause aggressive behavior, but instead appears to be a reflection of genes common between them. Such findings suggest that treating sleep dysfunctions may not cause a reduction in AGG, although more experimental work is needed before firm conclusion can be drawn. This provocative conclusion contradicts previous suggestion in the scientific literature, although it is worth noting that such arguments are based on limited evidence coming from research on clinical populations and focused on sleep problems rather than on the normal range of sleeping habits (Kamphuis et al., 2012).

What sort of mechanisms may drive this genetic pleiotropy? There are a number of possibilities, including serotonergic activity and/or hypothalamic-pituitary-adrenal axis functioning. For example, serotonergic activity is independently related to both aggressive behavior (Booij et al., 2010; Duke, Bègue, Bell, & Eisenlohr-Moul, 2013) and sleep (Monti, 2011), and is likely genetic influenced. The common genetic influences on sleep and AGG could thus reflect genetic influencing of the serotonergic system. Similarly, alterations to the hypothalamic-pituitary-adrenal axis system have been linked to aggressive behavior and sleep through the hormonal responses under its control (Kamphuis et al., 2012). Such findings point to an amalgam of factors related to both sleep and antisocial behavior that may increase the vulnerability to impulsive AGG.

However, a different structure of associations was found for RB. Most genetic correlations between sleep features and RB were negligible and the pattern was inconsistent as compared to the case of AGG. These results underline the importance of taking into account the different features of sleep in order to study their specific relationship with the different dimensions of antisocial behavior.

In sum, we can conclude that there is a strong phenotypic relationship between all sleep variables measured and antisocial behavior in children. However, the pattern of associations between them and AGG and RB, respectively, appear to be different. This finding lends additional support to the conceptualization of AGG and RB as different, if overlapping, dimensions of antisocial behavior (Burt, 2012). Our results also reinforce the need to identify the specific genetic mechanisms underlying the association between poor sleep and aggressive behavior. Adequately powered quantitative genetic and molecular studies could shed light over this relationship by pointing to the common physiological pathways underpinning these phenotypes. Additionally, future research should directly evaluate whether particular environmental variables modulate the sleep-aggression association. Such information would be not only of relevance to our understanding of these behavioral processes, but to the design of new treatments and intervention procedures for the different dimensions of antisocial behavior.