Callous traits in children with and without conduct problems predict reduced connectivity when viewing harm to others

The presence of elevated callous unemotional (CU) traits seems to designate a distinct group of children and adolescents with serious conduct problems. However, the extent to which CU traits impact the aversive reaction to harm is still a contentious issue. Here, we examined the effective connectivity seeded in the anterior insula and anterior cingulate cortex in a large number of children (N = 123, age 9–11, 60 females) with various levels of conduct disorder (CD) symptoms in response to visual stimuli depicting other people being physically injured. Perceiving others being harmed was associated with increased hemodynamic activity in the left amygdala and right temporoparietal junction (rTPJ). Children with higher callous traits showed less functional connectivity seeded in anterior cingulate with left amygdala and anterior insula. Conversely, CD symptoms were positively related to connectivity of insula with rTPJ. Overall, these results suggest that callousness is marked by the disruption of widespread cortical networks responsible for detecting and appropriately responding to important environmental cues, such as the distress of others.

Scientific RepoRts | 6:20216 | DOI: 10.1038/srep20216 harm aversion 10 . The pain of others is an evolutionarily important cue that serves to capture an individual's attention and, in healthy adults, motivate prosocial behavior [21][22][23] . Appropriately detecting and responding to such cues relies on the recruitment of a more general "salience network", which integrates cognitive, affective, and physiological state information to appropriately respond to motivationally relevant events across the domains of nociception, negative affect, and cognitive control 24,25 . Within this neural network, the aINS and dACC have been argued to serve complementary input and output functions, respectively 26 . Thus, in the context of pain perception, aINS and dACC (sometimes called frontoinsula and anterior midcingulate cortices) coordinate widespread cortical and subcortical activity to determine the optimal response to internal and external events 24,25,27 .
Importantly, several studies of negative affect and harm processing in children have demonstrated associations between the hemodynamic activity within the aINS and dACC and both antisocial behavior and CU traits. In fact, a twin study proposed gray matter volume in dACC as an endophenotype for psychopathic traits 28 . One functional MRI study of boys aged 9-15 found that aggressive behavior scores were negatively correlated with neural activity in dACC when viewing negatively valenced images 29 . Another study with adolescents found that conduct disorder (CD) symptoms are associated with greater activity in insula, dACC, striatum, and amygdala for stimuli depicting accidental harm compared to neutral matched stimuli 16 . More recently, an fMRI study reported reduced activity in the insula and ACC, as well as superior frontal gyrus and amygdala to stimuli depicting physical harm in adolescents with CD or Oppositional Defiant Disorder (ODD) but only when they imagined the pain happening to another person 19 . Moreover, the interpersonal/affective dimension of the Psychopathy Checklist-Revised was negatively correlated with signal in the ACC. In a larger study of young males (age 10-16) using similar stimuli, those youth exhibiting conduct problems showed reduced activity in the aINS, ACC, and inferior frontal gyrus 18 . Importantly, CU traits were significantly related to decreased neural activity in the aINS and ACC, while CP symptoms were linked to greater activity in ACC.
Conduct problems and CU traits have also been shown to independently influence social cognition, though sometimes in opposing fashion. For instance, during affective compared to cognitive theory of mind (i.e. judging how someone will feel compared to judging what someone believes or wants), young males with conduct problems showed reduced activity in right amygdala and right aINS 30 . Moreover, as previously observed in ACC response during pain processing 18 , CU traits and CP symptoms exerted suppressor effects on right amygdala response, with higher CP symptoms independently associated with increased activity, but higher CU traits independently related to decreased activity. Characterizing brain activity in terms of functionally segregated regions does not reveal anything about how different brain regions communicate with each other. Whereas standard contrast analyses create a ''snapshot'' of regional brain activity in response to a task or condition, functional connectivity analyses can identify patterns of communication between regions that contrast analyses may not detect 14,31,32 , and how callous unemotional traits might modulate such networks 33 .
To our knowledge, no study to date has examined how CU traits and CP symptoms impact the functional connectivity of the aINS and dACC (two critical nodes for empathy and harm aversion) when children view stimuli depicting other people being harmed. To fill this gap in the literature, we used functional MRI in a diverse sample of young children (age 9-11) while they engaged in a well-established pain perception paradigm with functional MRI 12,34,35 . Given their central role in saliency processing, dACC and aINS were chosen as seeds for a functional connectivity analysis. We hypothesized that children high in callous traits, irrespective of their conduct disorder symptoms, would be less sensitive to the pain of others, and thus, show reduced functional connectivity within the salience network. Additionally, this stimulus set includes both intentional and accidental harmful actions. This allowed us to examine the influence of intentionality on the neural encoding of harm. Children with high callous traits were expected to incorporate less of the intentionality information, and thus show reduced neural differentiation between intentional and accidental harmful actions in rTPJ, a region that plays a pivotal role in the extraction of intentionality [36][37][38] .
Across all participants, there was significantly greater hemodynamic activity in the Harm condition within the left amygdala ROI (Fig. 1). The ACC ROI showed a trend towards the same effect (FDR q = 0.061). Conversely, rTPJ showed significantly greater activity during No Harm. Both ACC and raINS seeds demonstrated significant increased coupling with rTPJ/pSTS (Fig. 2). Activity within the ACC ROI was significantly greater for Intentional Harm, while rTPJ/pSTS response was greater for Accidental Harm (Fig. 1d). Activity in ACC was negatively related to aversiveness ratings (partial r = − 0.32, FDR q = 0.017) and positively related to pain perception scores Scientific RepoRts | 6:20216 | DOI: 10.1038/srep20216 (partial r = 0.29, FDR q = 0.046). Postscan ratings were not significantly related to any other measure of hemodynamic activity or functional connectivity. CU traits were assessed by parental report on the Inventory of Callous and Unemotional Traits (ICU), which includes separate subscales for callous, unemotional, and uncaring traits 39 . Standardized beta weights for each ICU subscores and CD symptoms (age, gender, race-ethnicity, and maternal education were also included as nuisance regressors) are reported in Table 1. After controlling for multiple comparisons, ICU-Callous scores were not significantly related to activity within any of the ROIs for the Harm versus No Harm contrast. ICU-Callous scores were negatively correlated with ACC-seeded connectivity with rTPJ and raINS (Fig. 2). For the PPI seeded in raINS, the rTPJ response was negatively related to callousness, but positively associated with number of CD symptoms. Symptom number was also related to significantly increased connectivity between raINS and right amygdala (Fig. 3). Follow-up analyses revealed no significant interactions with gender for either the whole-brain or PPIs.

Discussion
The current study demonstrates the specific and independent influences of CU traits and CD symptoms on neuronal coupling when viewing others being harmed in a diverse sample of preadolescent children. As expected, when viewing harm to others, compared to visually similar control stimuli without harm, children showed greater activity in a host of regions, including bilateral amygdala, hippocampus, insula, MCC, and brainstem ( Fig. 1). Similarly, the perception of harm was also associated with increased functional connectivity from ACC with rTPJ, as well as connectivity from raINS to rTPJ (Fig. 2). However, these patterns of connectivity were significantly impacted by children's callous traits and, to a lesser extent, CD symptoms (Figs 2 and 3).
Neuronal activity in left amygdala was significantly greater for stimuli depicting harm (Fig. 1). Interestingly, we did not observe a significant negative relationship between callousness or CD symptoms and amygdala response (Table 1). This is a surprising finding since many studies investigating children with CD and callous unemotional traits 19,30 (but see 12,16 ), as well as studies of psychopathy in adults (see 40 for a review) have reported an abnormal amygdala hemodynamic response to negative information. In our study CU traits are unrelated to amygdala response when witnessing harm to others. This may suggest that harm aversion is processed at an early stage in the amygdala, as documented by intra-cranial deep electrodes recordings 41 as well as source localization with high-density EEG 42 with the same stimuli, but, as indicated by the functional connectivity analysis, this   information is not coupled with the dACC and aINS, which play pivotal roles in the interoception necessary for the construction of conscious affective experiences 43 .
For the PPI seeded in dACC, children with higher ICU-Callous scores showed reduced effective connectivity with right aINS and left amygdala (Fig. 2) when viewing depictions of others being harmed. Given their proposed function for detecting and responding to motivationally relevant information, especially in the context of negative associations between ICU-Callous scores and postscan ratings, the decreased connectivity between dACC and aINS suggests that individuals with high CU traits encode the pain of others as less salient than individuals with low CU traits. This fits with previous accounts of CU traits, as well as psychopathic traits 3,14,44 . Such an interpretation is also consistent with the observed reductions in left amygdala connectivity, since the amygdala plays an important role in motivation 45 .
The observed negative influence of callousness on amygdala-dACC connectivity is in line with models of psychopathy which posit amygdala disruption as a core feature 46,47 . A similar effect was recently reported between trait coldheartedness and amygdala-dACC coupling when healthy adults watched violence 14 . Thus, our findings suggest that childhood callousness and adult coldheartedness may both reflect specific disrupted communication between amygdala and dACC in the context of harm processing.
When right aINS was used as a functional seed, CD symptoms were related to increased neuronal coupling with TPJ/pSTS (Fig. 3). Similar stimuli have been used previously in incarcerated populations with mixed results. One study found that high levels of psychopathic traits were related to increased coupling seeded in right aINS with right TPJ/pSTS when participants imagined the action was happening to them 11 . In a different study, individuals with higher psychopathy scores demonstrated decreased functional connectivity seeded in rTPJ/pSTS with right aINS when viewing Harm compared to No Harm in a decision-making context 48 . Interestingly, previous of children with conduct problems have found suppressor effects between callousness and CP symptoms in the contexts of pain perception and cognitive and affective theory of mind 18,30 . Thus, future studies should continue to examine independent contributions of callousness and CP symptoms, and work to further distinguish between the two.
Unlike a previous study of callous traits in preteens 18 , at the whole-brain level, callousness scores were not significantly related to response in dACC or aINS. These differences could arise from the social nature of our stimuli (rather than simply hands and feet) or because we tested a slightly younger population (mean age of 10.5, compared to 13.7). However, an intriguing possibility is that the influence of callousness on neural functioning is first detectable in disrupted neuronal connectivity, and only later manifests as whole-brain differences in BOLD signal. Thus, future research will benefit from longitudinally measuring both neuronal responses as well as patterns of functional connectivity in order to better characterize how CU traits impact neural networks across development.
Overall, this study was designed to assess the influence of CU traits on neural function while children viewed others being harmed in a variety of social situations. After controlling for age, gender, CD symptoms, and maternal education, callousness scores uniquely predicted increased hemodynamic activity in right amygdala. Across all children, viewing harm was associated with increased effective connectivity between ACC and rTPJ, and between raINS and rTPJ. Moreover, callousness predicted reduced neuronal coupling seeded in dACC with left amygdala and right aINS.
Together, these data provide evidence in support of the hypothesis that children with high levels of callous traits do not respond appropriately to the harm of others because they experience a widespread disconnection in the cortical and subcortical networks which underlie important socio-emotional behaviors, such as empathy. Further investigation will be needed to better clarify the specific contexts in which CU traits are associated with network disruption, and whether or not children with high levels of CU traits might benefit from therapies designed to improve the function and integrative capacities of these networks.

Methods
Participants. Recruitment was designed to obtain a diverse sample of 123 children (10.55 ± 0.99 years; 60 females; 69 African Americans, collection stopped after obtaining 60 children of each gender). Children were excluded from the final analysis for missing data (n = 7) or excessive movement (n = 9), leaving a final sample of 106 children (see Table 2 for demographic information). Flyers advertising for well-behaved children were placed in pediatric well-visit waiting rooms, and flyers calling for children with behavioral problems were placed in outpatient child mental health clinics. Telephone screening interviews were used to recruit children into high or low risk for meeting DSM-IV diagnostic criteria for CD. The DISC Predictive Scale (DPS) for CD was administered to parents, then children. The DPS consists of 8 "stem questions" from the Diagnostic Interview Schedule for Children (DISC-IV) CD module 49 and predicts a full diagnosis of CD with high specificity and sensitivity 50 . Inclusion in the high-risk stratum was based on parent alone endorsing 2 or more DPS items, child along endorsing 3 or more items, or the parent and child collectively endorsing 3 or more separate items. The low-risk stratum contained only children for whom neither parent nor child endorsed any CD items. Equal numbers of high risk and low risk girls and boys were recruited. Exclusion criteria included head trauma resulting in loss of consciousness exceeding 15 minutes and presence of pervasive developmental disorder. On the day of scanning, trained interviewers administered the full DISC-IV 49 , including the module for CD symptoms within the past 12 months, to the primary caregiver and the child in separate rooms. Parents also completed the Inventory of Callous and Unemotional Traits 39 . Parents provided written informed consent, and children gave informed assent. All study procedures were approved by the Institutional Review Board at the University of Chicago. All methods were performed in accordance with the approved guidelines.
Stimuli and task. Prior to MRI scanning, participants were acclimated to study procedures in a mock scanner. This included lying in the mock scanner and practicing holding still while watching a documentary. Additionally, participants listened to recordings of scanner pulse sequences, and watched example stimuli from each condition (that were not used during the actual MRI scanning session).
Once in the magnet, participants viewed dynamic visual stimuli depicting either one or two individuals, with one person either being harmed (Harm), or not (No Harm). In the two person scenarios, harm was inflicted either intentionally or accidentally. Thus, there were five possible stimuli categories. Scenarios consisted of three pictures shown sequentially to suggest visual motion (1000 ms, 200 ms, 1000 ms). Participants viewed stimuli in alternating blocks of fixation cross (17.8 s duration) and task (20 s duration). The 20 task blocks were presented in pseudorandom order and consisted of six scenarios of the same type interspersed with a fixation cross (1134 ms). Stimuli presentation was controlled with E-Prime 1.0 (Psychology Software Tools, Inc., Pittsburgh, PA) and viewed via a back-projection system. Passive viewing was used to reduce cognitive load. Eye tracking was collected to ensure that children remained awake and attentive to the stimuli.
Postscan ratings. After the scanning session, participants viewed 25% of the images again and answered questions using a visual analog scale (VAS). The VAS responses were coded so that they ranged from 0 to 100. Questions assessed harm perception ("How painful was it for the person who was hurt?") and personal aversive response ("How upset are you when you watch this?"). Scores were averaged across each stimulus type, then across stimulus types to obtain average painfulness and distress ratings for Harm and No Harm images. Finally, change scores were calculated by subtracting ratings in the No Harm scenarios from ratings of Harm scenarios. Image processing and analysis. MRI data were processed with the FMRIB Software Library (FSL) version 5 51 . After removal of skull and other non-brain voxels, EPI images were realigned with MCFLIRT, high-pass filtered, and smoothed (6 mm FWHM). For normalization, functional images were first registered to each individual participant's structural scan with boundary-based registration 52 , then registered to the standard MNI152 template via linear transformation with 12 degrees of freedom.
General linear modeling was used for statistical analysis as implemented in FEAT. Because the goal of this study was to examine neural processing associated with the perception of harm across a range of situations, stimuli were collapsed into Harm and No Harm categories. These two broad categories were modeled separately in the GLM, beginning at the onset of the first picture in each block, and continuing until the end of the last picture of the block. Motion parameters were entered as nuisance regressors, and volumes with abnormally large motion artifacts were de-weighted in the design matrix. All participants had fewer than 16% percent of volumes with motion contamination (average 36.6 ± 12.6 out of 450 volumes). At the second level, a second GLM was used, with ICU subscores, CD symptoms, age, gender, race-ethnicity, and maternal education entered as covariates. Completion of high school was used to assess maternal education, because it is reliably associated with tested child intelligence [53][54][55] . Number of CD symptoms was used, rather than a dichotomous classification scheme, because dimensional analyses are better suited for hypothesis testing 56,57 and more biologically valid for mental disorders 56 . Using FSL defaults, clusters at the whole-brain level were defined with a z threshold of 2.3 then corrected according to Gaussian Random Field theory for a cluster threshold of p < 0.05.
Based on previous studies of socioemotional processing, some using the same stimuli, including a recent meta-analysis of morality, empathy, and theory of mind 35,58,59 , regions of interest were chosen for the right TPJ/ posterior superior temporal sulcus (TPJ/pSTS; x = 62, y = −54, z = 16), ACC (x = 0, y = 36, z = 20), raINS (x = 30, y = 18, z = −12), left amygdala (x = −22, y = −2, z = −24), and right amygdala (x = 16, y = −4, z = − 24). For each ROI, masks were generated by placing a sphere with radius 6-mm centered at each coordinate. The TPJ/pSTS ROI, taken from a "moral cognition" meta-analysis 59 , was chosen because this ROI's hemodynamic response and effective connectivity have been previously shown to be responsive to harmful scenarios in healthy adults 60 , and are influenced by levels of psychopathic personality traits in inmates, particularly when passively viewing harmful scenes 48 . The ACC and raINS ROIs we selected both for their role as core nodes of the salience network and previous work demonstrating their causal influence on network activity during socioemotional processing 58 . Finally, bilateral amygdala ROIs were selected because the amygdala plays a critical role in harm perception in childhood 35 and is reliably linked to variation in callous traits in adolescent males 30 and psychopathic personality traits in adult inmates 11,12 .
Functional connectivity was assessed using a psychophysiological interaction (PPI). For the ACC and raINS seeds, mean activity within the ROI was extracted and used as the physiological regressor (Harm-NoHarm served as the psychological regressor). Again, at the second level, age, gender, race-ethnicity, and maternal education were entered as covariates of no interest. Previous work suggests that CU traits and CD symptoms often demonstrate suppressor effects 18,30 , so we sought to identify the independent influences of CU traits and CD symptoms with multiple regression analyses for each ROI. In addition to ICU scores and CD symptoms, age, gender, race-ethnicity, and maternal education were entered as nuisance regressors. Participants who demonstrated signal change or connectivity values more than three standard deviations away from the full sample mean in any analysis were removed as outliers (n = 14). Finally, the graphically sharpened method for False Discovery Rate (FDR) was used to correct for multiple comparisons 61 . Across all analyses, two-sided tests were used and significance was set at α = 0.05.