Introduction
Youth externalizing behavior problems such as aggression are considered a global public health problem (World Health Organization, 2015). Among the most robust individual-level predictors of externalizing behaviors in youths are callous-unemotional traits—which include an uncaring disposition and reduced empathy, guilt, and remorse (Frick, 2012). Youths with these traits exhibit elevated aggression (Helseth et al., 2015), delinquency (e.g., theft, vandalism), and risk for conduct problems (Frick, 2012). These traits typically emerge in early childhood, with significant inter-individual variability observed in how they change over time (Fanti et al., 2017; Masi et al., 2018; Muratori et al., 2016). Some neurodevelopmental evidence links both increased callous-unemotional traits and externalizing behaviors to changes in the structure and function of the amygdala (Cardinale et al., 2019; Fanti et al., 2020; Gao et al., 2024; Lozier et al., 2014; Viding et al., 2012). Specifically, in youths with callous-unemotional traits, reduced volume and activation of the amygdala have been linked to reinforcement insensitivity, reduced empathic responsiveness, and heightened proactive aggression (Blair et al., 2018; Cardinale et al., 2019; Lozier et al., 2014). However, despite these findings, mechanisms that contribute to the emergence and persistence of callous-unemotional traits remain poorly understood.
Understanding of the developmental course of callous-unemotional traits is complicated by the fact that children with these traits are often exposed to numerous forms of adversity, such as harsh parenting, maltreatment, and violence, which are known risk factors for externalizing disorders (Afifi et al., 2009; Carliner et al., 2017; Fong et al., 2019). Youths with higher levels of callous-unemotional traits are exposed to more violence even relative to other youths with conduct problems, and this exposure in turn predicts more severe outcomes (Howard et al., 2012; Kahn et al., 2013). Some evidence suggests adverse experiences contribute to heterogeneity among youths with callous-unemotional traits (Jiang et al., 2024; Kimonis, 2023; Robertson et al., 2023; Todorov et al., 2023). Specifically, a primary low-anxiety subtype of callous-unemotional youths may be distinguishable from a secondary high-anxiety subtype characterized by higher levels of adverse experiences (Craig et al., 2021). Cross-sectional research suggests these subtypes may be distinguished by differences in amygdala reactivity to neutral male faces (Fanti et al., 2020) and resting-state connectivity between the left amygdala and left thalamus (Dugré & Potvin, 2023). However, no prior research has tracked how baseline maltreatment in combination with amygdala structural and functional variables predict changes over time in callous-unemotional traits.
The amygdala is a subcortical structure that is critical to coordinating responses to threats, learning in response to punishment, and generating adaptive social responses including prosocial motivation and behavior across species (Adolphs, 2008; Blair, 2008; Marsh, 2018). Given all of these functions are impaired in youths with callous-unemotional traits, atypical structural and functional development of the amygdala has long been thought to underlie the emergence and progression of these traits (Blair, 2008). Accordingly, callous-unemotional traits have been linked to both reduced amygdala volume (Cardinale et al., 2019; Viding et al., 2012), most recently in a recent large ENIGMA consortium study (Gao et al., 2024), as well as reduced amygdala activation in response to social stimuli and fear-relevant cues (e.g., Cardinale et al., 2018; Fanti et al., 2020; Lozier et al., 2014; Viding et al., 2012). Whether and how exposure to maltreatment may mediate or moderate these associations is not yet clear, however.
Specifically, it remains unclear if maltreatment leads to more severe callous-unemotional traits through its effects on amygdala development (Kahn et al., 2013), or if the effects of maltreatment are separate from the developmental effects of the amygdala on callous-unemotional traits. Neuroimaging research has also linked childhood maltreatment to reduced amygdala volume (Mclaughlin et al., 2019). This suggests the possibility that maltreatment could mediate increases in callous-unemotional traits through its effects on the structure of the amygdala. But whereas callous-unemotional traits are typically linked to reduced amygdala responsiveness to threat (Fanti et al., 2020; Lozier et al., 2014; Viding et al., 2012), exposure to childhood trauma has been associated with heightened amygdala responses to threat (Hein & Monk, 2017; McCrory et al., 2017; Mclaughlin et al., 2019). It has thus been proposed that exposure to trauma may moderate the relationship between callous-unemotional traits and amygdala response to fearful expressions (Fanti et al., 2020; Meffert et al., 2018). Consistent with this possibility, youths with callous-unemotional traits and maltreatment exposure (who may have secondary callous-unemotional traits) show increased physiological anxiety and startle (Dackis et al., 2015; Kimonis et al., 2012).
One potential impediment to understanding associations among callous-unemotional traits, maltreatment, and neurodevelopment of the amygdala is neuroimaging studies’ frequent reliance on cross-sectional samples (e.g., Fanti et al., 2020; Jiang et al., 2024; Meffert et al., 2018), which preclude the ability to determine whether, for example, changes in callous-unemotional traits over time can be predicted by prior maltreatment exposure and/or amygdala functional and structural variables at baseline. A second potential impediment is frequent use of exclusively linear models despite evidence of non-linear relationships between callous-unemotional traits and neurodevelopmental outcomes such as fearfulness (Cardinale et al., 2021) and delinquency (Markowitz et al., 2015), such that variation in observed associations across studies could reflect different patterns of associations depending on the severity of callous-unemotional traits in a sample.
To address these gaps, we assessed callous-unemotional traits and amygdala structure and function at two timepoints in youths who varied in exposure to maltreatment. We predicted that, at baseline, callous-unemotional traits would be associated with reduced amygdala volume and activation in response to fearful faces, in line with prior findings. We also predicted that prior maltreatment would in part account for these associations, specifically, that maltreatment exposure would partially mediate the association between callous-unemotional traits and amygdala volume, and would moderate the association between callous-unemotional traits and amygdala activation. Finally, we assessed two competing hypotheses: whether maltreatment exposure and amygdala volume and functional activation independently predicted changes in callous-unemotional traits over time, or whether the association of maltreatment and callous-unemotional traits was mediated by amygdala structure. Following evidence of non-linear associations between callous-unemotional traits and various neurodevelopmental outcomes (Cardinale et al., 2021; Markowitz et al., 2015; Walker et al., 2020), we tested both first and second order (non-linear) polynomial models for each predicted association.
In summary, the overarching goals of the present study were first, to test the prediction that callous-unemotional traits are related to reduced amygdala structure and function, both cross-sectionally and over time. Second, we aimed to test the contribution of maltreatment to those relationships. Our final goal was to assess evidence for two alternate mechanistic pathways: whether maltreatment exposure and amygdala volume and functional activation independently predict changes in callous-unemotional traits, or whether the association of maltreatment with callous-unemotional traits are mediated by amygdala structure.
Methods and Materials
Participants
Participants were 161 children and adolescents ages 8–17 recruited from the Seattle, Washington area. Youth and caregivers were recruited at schools, after-school and prevention programs, adoption programs, food banks, shelters, parenting programs, medical clinics, and the general community in Seattle between January 2015 and June 2017. Recruitment efforts aimed to create a sample with variation in exposure to violence and maltreatment. Therefore, youth were recruited from neighborhoods with high levels of violent crime, clinics that served a predominantly low-SES catchment area, and agencies that work with families who have been victims of violence (e.g., domestic violence shelters, programs for parents mandated to receive intervention by Child Protective Services). Exposure to maltreatment and other inclusion and exclusion criteria were assessed during the first study visit, including exposure to physical or sexual abuse or direct witnessing of domestic violence. For all participants, exclusion criteria included IQ < 75 (measured using the WASI–II; Wechsler, 2011), presence of a pervasive developmental disorder, active psychotic symptoms or mania, active substance misuse and other safety concerns (e.g., imminent suicidality), or failure to complete relevant screening measures during the first visit. Participants who completed the magnetic resonance imaging (MRI) reported no contradictions to scanning (i.e., braces, claustrophobia). Due to attrition and 11 participants who were excluded from further MRI analysis due to excess head motion or failure to respond to an attention check, 117 youth completed the 18-month follow-up visit. All guardians provided written informed consent and youth provided written assent prior to testing. All procedures were approved by the University of Washington Institutional Review Board. Clinical and demographic information for all study participants is reported in Table 1.
Table 1
Demographic information for the MRI only sample
MRI sample (n = 161) | ||||||
|---|---|---|---|---|---|---|
Mean | S.D. | Min | Max | Skewness | Kurtosis | |
Demographic variables | ||||||
Age, M(SD) | 12.64 | 2.67 | 8 | 17.25 | 0.1 | 1.72 |
IQ | 110.9 | 14.94 | 75 | 145 | -0.15 | 2.43 |
Gender, % | ||||||
Male | 52.17%, | |||||
Female | 47.83% | |||||
Race/ethnicity, % | ||||||
Asian, non-Hispanic | 10.76% | |||||
African American, non-Hispanic | 24.05% | |||||
White, Hispanic | 9.49% | |||||
White, non-Hispanic | 43.67% | |||||
Other | 12.03% | |||||
Clinical Measures | ||||||
Callous-Unemotional Traits | 0 0.41 | 0.29 | 0 | 1.3 | 0.79 | 0.25 |
Internalizing (T-score) | 56.83 | 12.19 | 33 | 87 | 0.24 | 2.6 |
Externalizing (T-score) | 53.36 | 11.7 | 33 | 81 | 0.22 | 2.38 |
Maltreatment History, % With History | 54.04% | |||||
Brain Volume Estimates | ||||||
L amygdala | 1848.85 | 290.66 | 1046 | 2607.9 | 0.36 | 3.12 |
R amygdala | 2006.29 | 311.81 | 1464.4 | 3035.4 | 0.63 | 3.33 |
Data from this sample related to risk for general psychopathology in children exposed to violence have been previously published (Weissman et al., 2020). No analyses related to callous-unemotional traits have been previously published.
Procedures
Participants completed study procedures across two time points. In the initial visit, measures of exposure to maltreatment and callous-unemotional traits were administered and baseline structural and functional MRI data were collected (n = 161). Participants returned for a follow-up visit approximately 18 months later and completed measures related to maltreatment and psychopathology (n = 117).
Maltreatment Exposure. Following our previous work (Weissman et al., 2020), maltreatment history was assessed using a multi-informant, multi method approach. Both parent and child reports were used to determine exposure to maltreatment as a function of experiencing interpersonal violence, including physical or sexual abuse. Witnessing domestic violence and experiencing emotional abuse were assessed by child report only, as caregivers are generally not considered valid reporters on emotional abuse, as emotional abuse is assessed in relation to caregiver behaviors. Thus, it is standard to assess these experiences from the child’s perspective only. Additionally, we were primarily concerned with domestic violence events that were directly observed by the child participant that would constitute a traumatic event for the child. Details of maltreatment measures and scoring procedures can be found in the online supplemental materials. A total of 100 youth (38.6%) experienced physical or sexual abuse, 82 youth (31.7%) experienced emotional abuse, and 99 youth (38.2%) witnessed domestic violence. Youth were dichotomously classified as either having or never having experiencing maltreatment (Herrenkohl et al., 2019).
Callous-unemotional traits. Following approaches used in recent work (Hawes et al., 2020a, b), callous-unemotional traits were calculated using items drawn from the Child Behavior Checklist (CBCL) (Achenbach & Edelbrock, 1983), and Youth Self-Report (YSR) (Achenbach, 1987), that index core features of callous-unemotional traits. Items were included that correspond to callous (e.g., ‘mean to others’, ‘lack of guilt’), uncaring (e.g., ‘doesn’t finish things’, ‘poor schoolwork’) and unemotional (e.g., ‘fearful’) constructs (Table 2) (Frick, 2004). Responses to all items were collected using a 3-point scale: 2 = “very often true”, 1 = “somewhat true”, 0 = “not true”. Because the CBCL and YSR do not contain identical items, a max scoring approach could not be used (Frick, 2004). Item selection is further discussed in the online supplemental materials.
Table 2
Factor loadings based on an exploratory factor analysis with items from the YSR and CBCL
Callous- unemotional | Reporter | |
|---|---|---|
CBCL | ||
4. Fails to finish things he or she starts | 0.61 | − 0.28 |
16. Cruelty, bullying, or meanness to others | 0.45 | − 0.47 |
26. Doesn’t feel guilty after misbehaving | 0.64 | − 0.37 |
61. Poor schoolwork | 0.66 | − 0.09 |
94. Teases a lot | 0.60 | − 0.50 |
YSR | ||
4. I fail to finish things that I start | 0.46 | 0.30 |
16. I am mean to others | 0.53 | 0.53 |
26. I don’t feel guilty after doing something I shouldn’t | 0.37 | 0.36 |
61. My school work is poor | 0.44 | 0.34 |
94. I tease a lot | 0.40 | 0.68 |
To produce a unidimensional callous-unemotional scale, inter-item correlations were calculated and items with mean inter-item correlation values between r = 0.15–0.5 were retained (Clark & Watson, 1995). Selected items were entered into an exploratory factor analysis and visual inspection of the scree plot indicated the presence of two factors with first and second eigenvalues of λ = 2.74 and λ = 1.75 respectively (Fig. 1). Further inspection of factor structure using Horn’s parallel analysis confirmed both factors had eigenvalues greater than the simulated data, indicating these eigenvalues scored above chance (Hayton et al., 2004; Horn, 1965). Investigation of factor loadings indicated the first factor loadings were capturing callous-unemotional traits and the second were capturing the survey reporter (i.e. parent or youth), thus the scale was treated as unidimensional (Table 2).
Fig. 1
Parallel analysis of the callous-unemotional items. Simulated data mean eigenvalues from 100 random iterations are plotted in blue. Adjusted eigenvalues accounting for the sample size are plotted in black, unadjusted eigenvalues are plotted in red. All eigenvalues are from an unrotated factor analysis
The final scale consisted of 10 items. Internal consistency of items was good (α = 0.70) and average interitem correlations were acceptable (Clark & Watson, 1995). Scores were calculated using the same items for participants with complete item data at follow-up (N = 183) and internal consistency was again good (α = 0.74). Final items were averaged to create a callous-unemotional score. External validity was supported by robust bivariate relationships between callous-unemotional scores and conduct problems scores (r = 0.67, df = 232, p < 0.001) (after removing callous-unemotional items from this scale), oppositional defiant scale scores (r = 0.62, df = 232, p < 0.001), and attention deficit scores (r = 0.62, df = 232, p < 0.001) (again, after removing callous-unemotional items from this scale). Additional details on item selection and scale validation are reported in Methods in the online supplement.
Callous-unemotional scores at baseline (M = 0.42, SD = 0.31, skewness = 0.9, kurtosis = 3.59) (follow-up: M = 0.39, SD = 0.31, skewness = 0.9, kurtosis = 3.17) translated to mean untransformed baseline scores of 4.2 (SD = 3.1) 21% of the maximum possible value (POMP). This value is only slightly below the empirically derived POMP scores that Kemp and colleagues (2023) suggest as a cutoff between low and high scorers in community samples using the ICU (29–33%). The mean CBCL-Externalizing t-score for youths in the upper half of our callous-unemotional distribution was 68.6, with t-scores of 65–69 typically considered borderline and 70 being the clinical cutoff. Together these data suggest our sample includes sufficient “high scorers” to test our hypotheses. Because callous-unemotional scores deviated from normalcy, scores were log transformed (baseline: M = 0.33, SD = 0.21, skewness = 0.4, kurtosis = 2.62; follow-up: M = 0.31, SD = 0.21, skewness = 0.46, kurtosis = 2.37).
Imaging
fMRI task. During fMRI scanning participants completed two runs of a task in which they viewed fearful, calm, or scrambled facial expressions (Tottenham et al., 2002). Scrambled facial expressions were used rather than calm expressions because of concerns that children exposed to violence may interpret even these expressions as more threatening (Lozier et al., 2014; Pollak et al., 2000). Each run contained nine 18-second blocks. Each block was comprised of 36 trials (300 ms) of a single emotion type displayed by 36 different actors who varied in gender and race. Each image was followed by a 200ms fixation. The order of emotion blocks was pseudo-randomized so that no emotion block occurred twice in a row and blocks were separated by three inter-trial interval blocks. A visual attention check occurred once per run where participants pressed a button when a cartoon character was displayed. See findings from Weissman et al., 2020 (Supplementary Fig. 1) for the basic effects associated with this task.
Image acquisition. Using a 3T Phillips Achieva scanner with a 32-channel coil in the University of Washington Integrated Brain Imaging Center we collected high-resolution anatomical images (176 axial slices; 2530ms repetition time; 3.5ms echo time; 1mm3 voxels; 7° flip angle; 256 × 256 mm field of view). An echo-planar imaging (EPI) sequence (37 3 mm slices parallel to the AC-PC line; 2s repetition time; 25ms echo time; 79° flip angle; 0.6 mm inner slice gap; 76 × 74 matrix; 3mm3 voxels; 224 × 224 mm field of view) was used to collect Functional T2*-weighted images. The first 4 images at the start of each scan were discarded. All functional volumes were registered to the subject’s anatomical grid collected during each session via the high-resolution T1-weighted anatomical scan and preprocessed (see online supplement).
Left and right amygdala regions of interest (ROIs) were created using the Harvard Oxford subcortical probabilistic structural atlas (50% threshold). Mean scores were calculated for each contrast of interest in both ROIs in subject anatomical space. We extracted the contrast indexing responses to fearful versus scrambled facial expressions (Weissman et al., 2020). Information on the relationship between amygdala volume and amygdala response to fearful facial expression are reported in the online supplement (Supplementary Table 3).
Data Analysis
We first used a linear regression analysis to examine associations between callous-unemotional traits and maltreatment history and baseline amygdala volume and activation. Baseline amygdala volume and functional BOLD activation were regressed on mean-centered total log-transformed callous-unemotional traits scores. Exposure to maltreatment was included as a predictor variable in relevant analyses. All analyses included gender and age as covariates (Falcón et al., 2021; Frick & Kemp, 2021) and a second-order polynomial was used in all models to estimate non-linear relationships. We next examined longitudinal relationships using residualized change models to assess whether changes in callous-unemotional traits across an 18-month period can be predicted from baseline functional or structural imaging data and/or maltreatment exposure. Models included age and gender covariates as well as baseline callous-unemotional trait scores, such that all other beta coefficients reflect changes in callous-unemotional trait scores from baseline. All analyses are reported, and values are not corrected for multiple comparisons across independent regression analysis. Correlations between study variables are presented in Supplemental Table 1.
All data were analyzed in Python using the statsmodel package (version 0.13.5) except for mediation analyses which were conducted in R using the mediation package (version 4.5.0).
Results
Baseline Associations among Amygdala Volume, Callous-Unemotional Traits, and Maltreatment
Supporting our prediction that callous-unemotional traits are associated with reduced amygdala volume, we found higher baseline callous-unemotional traits were associated with linear reductions in volume in both left and right amygdala. Each 1% increase in callous-unemotional traits corresponded to a 0.49 mm3 reduction in both left and right amygdala volume (Table 3). We also found that in left amygdala a second-order polynomial better fit the relationship between callous-unemotional traits and amygdala volume, indicating that the negative relationship between callous-unemotional traits and left amygdala volume was strongest at low levels of severity but its slope decreases as severity increases (Table 3; Fig. 2). In right amygdala, the linear association between volume and callous-unemotional traits remained the best-fitting model. We also found a linear bivariate association between callous-unemotional traits and maltreatment exposure at baseline (r = 0.35, p < 0.01) (Supplementary Table 1).
Table 3
Models predicting amygdala volume using callous-unemotional traits (n = 148)
Left amygdala | Right amygdala | |||||
|---|---|---|---|---|---|---|
b | s.e. | p | b | s.e. | p | |
Amygdala Volume | ||||||
Callous-Unemotional Traits | ||||||
Callous-Unemotional Traits | -48.71 | 22.57 | 0.033 | -48.84 | 23.36 | 0.038 |
Gender | -200.92 | 45.71 | < 0.001 | -273 | 47.32 | < 0.001 |
Age | 52.22 | 22.66 | 0.023 | 45.73 | 23.46 | 0.053 |
Second Order Polynomial | ||||||
Callous-Unemotional Traits | -63.79 | 23.0 | 0.006 | -57.3 | 24.16 | 0.019 |
Gender | -183.91 | 45.44 | < 0.001 | -263.46 | 47.75 | < 0.001 |
Age | 51.03 | 22.27 | 0.023 | 45.05 | 23.40 | 0.056 |
Callous-Unemotional Traits^2 | 46.97 | 19.0 | 0.015 | 26.35 | 19.94 | 0.188 |
Callous-Unemotional Traits and Maltreatment Exposure | ||||||
Callous-Unemotional Traits | -8.92 | 26.73 | 0.739 | -15.66 | 27.91 | 0.576 |
Maltreatment Exposure | -141.1 | 53.27 | 0.009 | -117.68 | 55.61 | 0.036 |
Gender | -179.61 | 45.5 | < 0.001 | -255.23 | 47.51 | < 0.001 |
Age | 53.42 | 22.21 | 0.017 | 46.73 | 23.18 | 0.046 |
Second Order Polynomial | ||||||
Callous-Unemotional Traits | -24.44 | 27.01 | 0.367 | -24.27 | 28.63 | 0.398 |
Maltreatment Exposure | -138.35 | 52.35 | 0.009 | -116.15 | 55.5 | 0.038 |
Gender | -163.39 | 45.19 | < 0.001 | -246.23 | 47.91 | < 0.001 |
Age | 52.23 | 21.83 | 0.018 | 46.06 | 23.14 | 0.048 |
Callous-Unemotional Traits^2 | 45.92 | 18.60 | 0.015 | 25.47 | 19.72 | 0.198 |
Fig. 2
Marginal means plots excluding bins with less than 10 participants
In partial support of our second prediction that maltreatment exposure accounts for the association between callous-unemotional traits and amygdala volume, we found that maltreatment exposure was also associated with lower volume in both left and right amygdala. Notably, when maltreatment and callous-unemotional traits were modeled simultaneously, the linear relationship between callous-unemotional traits and both right and left amygdala volume was eliminated. However, the non-linear relationship between callous-unemotional traits and left (but not right amygdala) volume persisted, with stronger associations again observed between amygdala volume and callous-unemotional traits at lower severity levels (Table 3; Fig. 2).
Baseline Associations Between Amygdala Activation, Callous-Unemotional Traits, and Maltreatment
In support of the prediction that callous-unemotional traits are associated with reduced amygdala activation we found, at baseline, increased callous-unemotional traits were associated with linear reductions in right, but not left, amygdala activation when viewing fearful expressions. A non-linear relationship between callous-unemotional traits and right amygdala response to fearful expressions was again observed, with stronger associations between activation and callous-unemotional traits observed at lower severity levels (Table 4; Fig. 2). No non-linear relationship was observed for left amygdala activation.
Table 4
Models predicting amygdala volume using amygdala response to fearful expressions (contrast: fear > scrambled faces) (n=147)
Left amygdala | Right amygdala | |||||
|---|---|---|---|---|---|---|
b | s.e. | p | b | s.e. | p | |
Amygdala Response to Fearful Expressions | ||||||
Callous-Unemotional Traits | ||||||
Callous-Unemotional Traits | -0.15 | 0.09 | 0.098 | -0.18 | 0.09 | 0.044 |
Gender | 0.07 | 0.18 | 0.70 | -0.02 | 0.18 | 0.915 |
Age | 0.06 | 0.09 | 0.493 | 0.10 | 0.09 | 0.247 |
Second Order Polynomial | ||||||
Callous-Unemotional Traits | -0.17 | 0.09 | 0.06 | -0.23 | 0.09 | 0.011 |
Gender | 0.10 | 0.18 | 0.586 | 0.04 | 0.18 | 0.82 |
Age | 0.06 | 0.09 | 0.518 | 0.10 | 0.09 | 0.27 |
Callous-Unemotional Traits^2 | 0.09 | 0.08 | 0.25 | 0.18 | 0.08 | 0.019 |
Callous-Unemotional Traits and Maltreatment Exposure | ||||||
Callous-Unemotional Traits | -0.55 | 0.55 | 0.312 | -0.57 | 0.55 | 0.298 |
Maltreatment Exposure | -0.14 | 0.21 | 0.525 | -0.24 | 0.21 | 0.262 |
Gender | 0.09 | 0.18 | 0.618 | 0.02 | 0.18 | 0.913 |
Age | 0.06 | 0.09 | 0.48 | 0.11 | 0.09 | 0.232 |
Second Order Polynomial | ||||||
Callous-Unemotional Traits | -0.55 | 0.55 | 0.312 | -0.85 | 0.55 | 0.126 |
Maltreatment Exposure | -0.14 | 0.21 | 0.525 | -0.24 | 0.21 | 0.265 |
Gender | 0.09 | 0.18 | 0.618 | 0.08 | 0.18 | 0.667 |
Age | 0.06 | 0.09 | 0.505 | 0.10 | 0.09 | 0.256 |
Callous-Unemotional Traits^2 | 0.09 | 0.08 | 0.253 | 0.18 | 0.08 | 0.02 |
By contrast, no relationship was observed at baseline between maltreatment exposure and left or right amygdala activation. Unlike our structural findings, when maltreatment was included in our models, neither maltreatment nor callous-unemotional traits showed linear associations with responses to fearful expressions in left or right amygdala. However, in these models a non-linear relationship again emerged between amygdala activation and callous-unemotional traits in right but not left amygdala (Table 4; Fig. 2).
We also conducted exploratory analyses in an effort to approximately replicate prior findings (Meffert et al., 2018) that trauma exposure moderates the relationship between callous-unemotional traits and amygdala responses to fearful expressions. These analyses found no support for the prediction that maltreatment exposure moderates the association between callous-unemotional traits and activation in left (b = − 0.13, p = 0.22) or right (b = − 0.05, p = 0.70) amygdala. (Supplementary Table 4).
Predicting Changes in Callous-Unemotional Traits
After establishing associations between both callous-unemotional traits and maltreatment exposure and amygdala outcomes at baseline, we next used residualized change models to test our prediction that both baseline amygdala structure and function would predict changes in callous-unemotional traits across an 18-month period. Results indicated that lower left (but not right) amygdala volume at baseline is associated with linear increases in callous-unemotional traits over time (Table 5). When considering non-linear relationships, we found reduced left (but not right) amygdala volume at baseline again predicted greater increases in callous-unemotional traits (Table 5; Fig. 2).
Table 5
Residualized change models of changes in callous-unemotional traits across an 18-month period associated with baseline amygdala structure and function, and maltreatment exposure
Left amygdala | Right amygdala | |||||
|---|---|---|---|---|---|---|
b | s.e. | p | b | s.e. | p | |
Residualized Change Models | ||||||
Callous-Unemotional Traits | ||||||
Amygdala Volume | -8.9E− 5 | 5.0E− 5 | 0.077 | -6.7E− 5 | 4.5E− 5 | 0.141 |
Gender | -0.03 | 0.03 | 0.241 | -0.03 | 0.03 | 0.247 |
Age | 0.01 | 0.01 | 0.568 | 0.01 | 0.01 | 0.626 |
Baseline Callous-Unemotional Traits | 0.69 | 0.07 | < 0.001 | 0.69 | 0.07 | < 0.001 |
Second Order Polynomial | ||||||
Amygdala Volume | 0.001 | 0.00 | 0.026 | 3.0E− 4 | 0.00 | 0.448 |
Gender | -0.02 | 0.03 | 0.505 | -0.03 | 0.03 | 0.262 |
Age (in years) | 0.01 | 0.01 | 0.546 | 0.01 | 0.01 | 0.582 |
Baseline Callous-Unemotional Traits | 0.71 | 0.07 | < 0.001 | 0.69 | 0.07 | < 0.001 |
Amygdala Volume^2 | -3.0E− 7 | 1.2E− 7 | 0.015 | -9.2E− 8 | 9.9E− 8 | 0.355 |
Callous-Unemotional Traits and Maltreatment Exposure | ||||||
Amygdala Volume | -5.7E− 5 | 5.2E− 5 | 0.271 | -4.5E− 5 | 4.6E− 5 | 0.323 |
Gender | -0.04 | 0.03 | 0.203 | -0.04 | 0.03 | 0.198 |
Age (in years) | 0.01 | 0.01 | 0.672 | 0.01 | 0.01 | 0.71 |
Baseline Callous-Unemotional Traits | 0.60 | 0.08 | < 0.001 | 0.59 | 0.08 | < 0.001 |
Maltreatment Exposure | 0.07 | 0.03 | 0.041 | 0.07 | 0.03 | 0.027 |
Second Order Polynomial | ||||||
Amygdala Volume | 0.001 | 0.00 | 0.029 | 4.0E− 4 | 0.00 | 0.277 |
Gender | -0.02 | 0.03 | 0.434 | -0.04 | 0.03 | 0.21 |
Age | 0.01 | 0.01 | 0.643 | 0.01 | 0.01 | 0.654 |
Baseline Callous-Unemotional Traits | 0.63 | 0.08 | < 0.001 | 0.59 | 0.08 | < 0.001 |
Maltreatment Exposure | 0.06 | 0.03 | 0.055 | 0.08 | 0.03 | 0.02 |
Amygdala Volume^2 | -2.8E− 7 | 1.2E− 7 | 0.02 | -1.2E− 7 | 9.7E− 8 | 0.229 |
We next included maltreatment exposure in our models and found that maltreatment exposure was also associated with increases in callous-unemotional traits at follow-up. When the non-linear relationship between baseline amygdala volume and changes in callous-unemotional traits was modeled while also controlling for baseline maltreatment we found reduced left (but not right) amygdala volume at baseline predicted increases in callous-unemotional traits at follow-up (Table 5; Fig. 2). However, in this model maltreatment did not predict callous-unemotional traits at follow-up.
We next employed a bootstrap-mediation analysis with 1000 simulations to assess if the relationship between maltreatment and increased callous-unemotional traits at follow-up was mediated by left amygdala volume. Age, gender, and baseline callous-unemotional trait scores were included as covariates. Results indicated that left amygdala volume did not mediate the relationship between maltreatment exposure and changes in callous-unemotional traits (bias-corrected 95% CI, − 0.008 to 0.03).
Analyses including amygdala activation parameters found neither linear nor non-linear bivariate associations between baseline functional activation and changes in callous-unemotional traits (see online supplement, Supplementary Table 3).
Discussion
We investigated cross-section and longitudinal associations of callous-unemotional traits in a diverse sample of children and adolescents with high levels of maltreatment exposure. To our knowledge, this is the first study to simultaneously consider interrelationships among four variables known to be robustly associated with externalizing behavior in children and adolescents: callous-unemotional traits, amygdala structure, amygdala response to fearful faces, and maltreatment exposure. We find for the first time that maltreatment and amygdala structure and activation variables are associated with callous-unemotional traits at baseline, and in combination, can longitudinally predict changes in callous-unemotional traits over the course of 18 months. Baseline amygdala volume and maltreatment exposure were most strongly related to increased callous-unemotional traits 18 months later. Importantly, however, the association between maltreatment and increases in callous-unemotional traits was not mediated by amygdala structure or function. Instead, we found support for the alternate hypothesis that maltreatment exposure and amygdala volume and functional activation independently predicted changes in callous-unemotional traits. We identified these effects using a large single-site sample. These findings may be consequential for understanding the etiology of externalizing disorders, which are the most prevalent mental health concern among school-aged children in North America, with 6–17% meeting criteria for a disruptive behavior disorder (Danielson et al., 2021).
As we hypothesized, and consistent with prior work, callous-unemotional traits were linearly associated with lower amygdala volume (Cardinale et al., 2019; Gao et al., 2024) and activation in response to fearful expressions (Lozier et al., 2014; Viding et al., 2012) at baseline—although this is the first study to confirm both findings in a single sample. Our findings also yielded several novel insights. One is that non-linear models may better capture the relationship between callous-unemotional traits and both structural and functional outcomes, suggesting that the relationship between callous-unemotional traits and neurodevelopmental outcomes varies as a function of symptom severity. Thus, variation in amygdala structure and functional activation may not usefully discriminate among children with higher levels of callous-unemotional traits. In addition, we found support for our second hypothesis, which was that linear associations between callous-unemotional traits and both the structure and function of the amygdala were eliminated by including maltreatment as a covariate. Non-linear relationships persisted, however, which largely reflected the strongest associations between callous-unemotional traits and reduced amygdala volume and activation at lower levels of callous-unemotional traits (Howard et al., 2012).
These findings have implications for understanding the etiology of callous-unemotional traits. We focused on the amygdala in our analyses following robust prior evidence linking atypical development in this structure to callous-unemotional traits, maltreatment exposure, and externalizing behavior (Lozier et al., 2014; McCrory et al., 2017; Mclaughlin et al., 2019; Viding et al., 2012). Our results confirm prior findings that maltreatment is associated with reduced amygdala volume and predicts worsening callous-unemotional traits over time. We also found that smaller amygdala volume at baseline predicts worsening callous-unemotional traits over time. However, we did not find these two risk factors to be related. This suggests that simply reducing maltreatment exposure is unlikely, on its own, to eliminate risk of callous-unemotional traits—although it would likely reduce the severity of these traits, particularly secondary variants. Amygdala volume and functional activation patterns are at least partly heritable (Lewis et al., 2014; Mufford et al., 2024; Quarto et al., 2023), thus, genetically influenced atypical amygdala development from an early age may contribute to the fearless temperament and low social affiliation that are early-emerging signs of callous-unemotional traits (Fanti et al., 2023; Paz et al., 2024). This is consistent with evidence that amygdala plays a dual role in supporting both fear acquisition and learning (Adolphs, 2008; Blair, 2008) and empathic and prosocial responding (Chang et al., 2015; Marsh et al., 2018; Rhoads et al., 2023). In both humans and non-human mammals, the amygdala specifically supports empathic responding to others’ distress, possibly by supporting the coordination of internal representations of fear in response to others’ nonverbal cues, consistent with simulation theories of empathy (Olsson et al., 2007). In other words, increased amygdala activation in response to others’ fear may represent an empathic response, possibly supported by increased volume. This interpretation is supported by the fact that increased amygdala volume and functional activation in response to fear are associated with improved empathic accuracy for these cues and with real-world prosocial motivation and behavior (Marsh et al., 2014).
Why our findings related to amygdala volume were more robust than those related to functional activation is unclear. Amygdala activation may be inherently more heterogeneous in its relationship to callous-unemotional traits, although our moderation analysis as a function of maltreatment did not identify any moderating effect of this variable. It may also reflect measurement differences, with gray matter volume being potentially more robust to measurement error than measures of functional activation due in part to known sources of error in the amygdala BOLD signal (Boubela et al., 2015). This may help explain some reviews of amygdala functional activation finding inconsistent associations with psychopathy (e.g., Deming et al., 2022), although the fact that this review collapsed across classes of evoking stimuli may also be the cause of the putatively null findings, as amygdala activation anomalies in samples with callous-unemotional traits are more likely to emerge in response to fear-evoking stimuli than stimuli related to anger, disgust, or some other emotion categories (e.g., Cardinale et al., 2018; Deming et al., 2020; Lozier et al., 2014).
Observed non-linear effects may help reconcile some inconsistencies in prior findings (Dotterer et al., 2017; Meffert et al., 2018). Prior work comparing extreme samples—for example, typically developing youths versus those recruited for conduct problems and callous-unemotional traits—find reduced amygdala activation and volume in callous-unemotional youths (Dotterer et al., 2020) consistent with either a linear or a non-linear effect. By contrast, samples that also include participants with conduct problems and low callous-unemotional traits (whose scores are nonetheless higher than those of developing children) generally yield results consistent with non-linear effects (Lozier et al., 2014; Viding et al., 2012) as do studies that consider the interaction of callous-unemotional traits and trauma (Meffert et al., 2018), although the specific patterns vary. And at least one sample from the community that was nonetheless very high-risk (drawn from the Fragile Families and Child Well-Being sample, children from which exhibit elevated conduct problems (Osborne, 2007; Waldfogel et al., 2010) examining the linear correspondence between callous-unemotional traits and amygdala responses found no effect. (Another community sample study had such uniformly low average callous-unemotional trait scores (M = 2.1, SD = 1.6, range 0–6 on a 10-point scale) and also used combined fear/anger stimuli such that comparisons may be difficult (Dotterer et al., 2017).
We note two things about these patterns: First, they are overall consistent with our observed non-linear findings in a sample that spans low, moderate, and high callous-unemotional traits, although our findings are the first to indicate that callous-unemotional traits are more strongly associated with amygdala volume and activation at lower levels of these traits, with youths having moderately versus very high levels of these traits appearing more similar. Second, to our knowledge no prior imaging study in callous-unemotional youths has explicitly examined linear and non-linear effects that may help resolve apparent inconsistencies in the existing literature, which we believe increases the value of our approach. Only one prior study to our knowledge has tested for and found non-linear effects of amygdala function or structure as a function of any psychopathy-related measure (PCL-R traits in adult male offenders) (Schiffer et al., 2011). Although we cannot over-extrapolate from their findings given major differences in the samples, their findings support the importance of considering non-linear effects.
These results should be considered in light of certain limitations. We used an index of callous-unemotional traits derived from the parent-report version of the CBCL and its youth-report analogue, the YSL (Frick, 2004), an increasingly common approach for secondary analyses of datasets in which more targeted inventories like the Inventory of Callous-Unemotional Traits (ICU) were not included, such as the Adolescent Brain Cognitive Development dataset (Hawes et al., 2020a, b). This can result in stronger correlations between callous-unemotional traits and externalizing behaviors in part because these items are drawn from the same scale (Willoughby et al., 2011). When rigorously validated, as was the case here, this approach permits the assessment of callous-unemotional traits in larger and more diverse samples of participants, including those not originally recruited for research on callous-unemotional traits. One limitation of this approach is that items related to unemotionality other than guilt-proneness are poorly covered by the CBCL and YSR, and we found items indexing fearfulness did not converge with the remaining items and thus were not included in our scale (similarly, the ICU’s unemotionality subscale shows relatively poor convergence with the callous and uncaring subscales) (Cardinale & Marsh, 2020). However, we were able to leverage this method to consider the joint effects of callous-unemotional traits and maltreatment, a risk factor for which this sample was recruited and screened. Using our measure, we replicated prior findings related to associations between callous-unemotional traits and both amygdala structure and function and extended them by considering longitudinal outcomes and the moderating role of maltreatment (Meffert et al., 2018). This approach, however, precludes precise comparisons of callous-unemotional trait severity between our sample and other recent studies. Without ICU scores we cannot definitively ascertain how our sample would compare to other samples using this or other similar validated measures of callous-unemotional traits, or what proportion of our sample would be classified as having clinically significant callous-unemotional traits. Because callous-unemotional trait scores were log-transformed in order to conduct parametric analyses, the shape of non-linear relationships should be interpreted with caution because log-transformation pulls extreme scores closer to the mean. High co-occurrence of maltreatment types in our sample also precluded analysis of potential differences between them; prior work finds neglect may increase risk, therefore, future work exploring callous-unemotional traits should focus on this variable (Mclaughlin, 2019). Finally, investigating the influence of maltreatment and callous-unemotional traits on other brain regions or in response to other emotional expressions was beyond the scope of this research.
Despite these limitations, our study found new evidence of non-linear and longitudinal relationships between callous-unemotional traits and neurodevelopment of the amygdala, a finding with important implications for risk assessment and treatment. Children with callous-unemotional traits may be at highest risk for increasing symptom severity who have lower amygdala volume at baseline and who have been exposed to maltreatment. These findings illustrate the importance of early intervention particularly in youth exposed to maltreatment who exhibit callous-unemotional traits to prevent further worsening of callous-unemotional traits and externalizing behaviors.
Declarations
Ethical Approval
All procedures were approved by the University of Washington Institutional Review Board.
Conflict of Interest
The authors report no financial relationships with commercial interests.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, 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 you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. 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-nc-nd/4.0/.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.