Children and adolescents with conduct disorder (CD) and conduct problems (CP) engage in high levels of antisocial and aggressive behaviour, and represent a significant public health burden (Erskine et al.
2014). CP is a heterogeneous category, and one fruitful approach has distinguished between CP children presenting with high (CP/HCU) and low (CP/LCU) levels of callous-unemotional (CU) traits. CU traits index low levels of empathy and guilt, a tendency to use and manipulate others, unconcern about achievement, and flattened emotional responsivity (Essau et al.
2006). Children with CP/HCU represent a particularly severe subgroup within CP (Frick and Viding
2009). Genetic, behavioural, cognitive and functional neuroimaging studies have shown that different vulnerabilities characterise these two subgroups of children with CP. However, little previous structural neuroimaging work has directly compared these groups. The current study used voxel-based morphometry with a large sample (N = 89) to study grey matter (GM) volumes in these two groups relative to both each other and to typically developing (TD) controls.
Evidence suggests that underlying aetiology and neurocognitive processing differ between CP/HCU and CP/LCU. Twin studies have shown that antisocial behaviour is more strongly heritable in children with CP/HCU than CP/LCU (Viding et al.
2005; Viding et al.
2008). Behavioural studies have also shown that CP/HCU is associated with a distinctive information processing profile relative to TD controls, most notably low reactivity to emotional cues (Blair et al.
2001), poor empathy (Jones et al.
2010), impulsivity (Fanti
2013), and poor reversal learning (Budhani and Blair
2005). This profile is similar to that shown by adults with psychopathy (Barry et al.
2000). In contrast, children with CP/LCU show increased emotional reactivity and a profile of reactive aggression and poor emotion regulation, with affective empathy relatively intact (Dadds et al.
2006; De Wied et al.
2012; Eisenberg et al.
2010; Frick et al.
2003; Jones et al.
2010).
Current theories regarding the neurobiology of psychopathic and CU traits suggest that affective and reinforcement learning deficits are underpinned by atypical function in a ventromedial prefrontal cortex-amygdala circuit (Blair et al.
2014) as well as in a more distributed paralimbic network including orbitofrontal cortex (OFC), anterior insula (AI), anterior and posterior cingulate, temporal pole and parahippocampal gyrus (Anderson and Kiehl
2012). Functional magnetic resonance imaging (fMRI) studies in children with CP/HCU have found largely reduced responsiveness across this network relative to TD controls during a variety of emotion processing (Jones et al.
2009; Lockwood et al.
2013; Marsh and Blair
2008; Sebastian et al.
2012) and decision-making (Finger et al.
2011; Marsh et al.
2011) tasks, although some studies have also found increased responses (Cohn et al.
2013). In contrast, fMRI studies which have looked at both CP/HCU and CP/LCU within the same study have found a different pattern of neural response in CP/LCU, for example increased (as opposed to decreased) amygdala response to emotional faces compared with TD controls (Sebastian et al.
2014; Viding et al.
2012).
However, to date, no study has directly compared brain structure in children with CP/HCU, CP/LCU and TD controls. Several structural MRI (sMRI) studies have compared children and adolescents with CP in general against TD controls. The most common technique has been to use voxel-based morphometry (VBM) to explore grey matter volume and/or concentration across the whole brain and within specific regions of interest (ROIs). These studies have found GM reductions in CP relative to TD controls in the amygdala (Cope et al.
2014a; Fairchild et al.
2011; Huebner et al.
2008; Sterzer et al.
2007) AI (Fairchild et al.
2011; Sterzer et al.
2007) OFC (Cope et al.
2014a), and temporal poles (Huebner et al.
2008). Findings in the amygdala in female participants have been similar (Fairchild et al.
2013). Use of alternative metrics such as cortical thickness has also shown reductions in cingulate, prefrontal and insular cortices (Fahim et al.
2011) and in several temporal and parietal regions (Hyatt et al.
2012; Wallace et al.
2014). The study by Wallace and colleagues also found some evidence of reduced gyrification in ventromedial prefrontal cortex and a significant negative correlation between the severity of CU traits and cortical thickness in the right superior temporal cortex. Overall, therefore, most studies show a general pattern of reduced GM across children and young people with CP in brain regions associated with antisocial behaviour and psychopathy, even after controlling for common confounding variables such as IQ, substance misuse and attention deficit hyperactivity disorder (ADHD) symptoms.
Of the sMRI studies discussed above, a subset have explored the contribution of CU or psychopathic traits to variance in grey matter indices across the CP sample. The pattern of findings has been somewhat mixed. For example, in a very large (N = 191) sample of incarcerated male adolescents (Ermer et al.
2013) found negative associations between GM volume and psychopathic traits in OFC extending into temporal poles and parahippocampal cortex, and posterior cingulate cortex. This pattern of results was recently replicated in a female incarcerated sample (Cope et al.
2014b). However, also in female participants, Fairchild et al. (
2013) found that CU traits were positively (rather than negatively) associated with bilateral OFC volumes. Furthermore, some studies have found no associations between brain structure and CU traits; for example Fairchild et al. (
2011) found no relationship between GM volumes and CU traits in males, although CD symptoms were negatively associated with right insula volume and the CD sample size was large (N = 63).
Discussion
The current study used VBM to compare grey matter volumes in subgroups of children with CP and high vs. low levels of CU traits, and TD controls. We used a large sample (N = 89) together with state-of-the-art anatomical registration methods to maximise statistical power and accuracy of data pre-processing. In whole-brain analyses, we found evidence for reduced GM volume in CP/HCU relative to TD controls in the left middle frontal gyrus. Comparing groups in a priori ROIs, we found that reduced bilateral OFC GM volume in the CP group as a whole relative to TDs was largely driven by the CP/HCU group. Directly comparing all three groups, the CP/HCU group showed reduced bilateral OFC volume relative to TD controls, and reduced left OFC volume relative to both TD controls and CP/LCU. In contrast, there were no differences between CP/LCU and TD groups at FWE-corrected levels (although it is worth noting that a small cluster showed reduced right OFC volume in CP/LCU compared with TD controls at
p < 0.001 uncorrected; Table S
1). Continuous analyses within the CP group showed that these group differences were likely driven by CU traits as opposed to levels of CD symptoms. A largely comparable pattern of group differences was also seen in right ACC, again with reduced GM volume in CP/HCU relative to TD controls and no difference between CP/LCU and controls, even at uncorrected levels. However, CP/HCU and CP/LCU groups did not differ from each other in this region, even at uncorrected levels.
Taken together, the pattern of results supports several previous studies which have found GM reduction in children and adolescents with CP relative to TD controls in regions including OFC and ACC (Cope et al.
2014a; Fahim et al.
2011; Fairchild et al.
2011; Huebner et al.
2008). Our study extends these findings to suggest that GM reduction in some of the regions identified in these studies may in fact be primarily attributable to those children in whom CP co-occur with high levels of CU traits, as opposed to being associated with CP in general. This finding is in line with a study by Ermer et al. (
2013) which found negative associations between OFC GM volume and psychopathic traits in a sample of incarcerated male adolescents. However, we extend this work by showing that differences between CP/HCU and TD controls, specifically in left OFC and right ACC, did not characterise the CP/LCU group. Indeed in left OFC, the CP/LCU group showed significantly increased GM volume relative to CP/HCU, i.e. the same pattern as TD controls. These results show GM volume in these two regions (particularly in left OFC), may differentiate these two subgroups of children with CP. Overall, considering the results at corrected and uncorrected levels (Table S
1), the CP/HCU group showed more extensive grey matter reductions than the CP/LCU group when compared to the TD groups as evidenced by additional reductions in grey matter volumes in the ACC (FWE-corrected) and in the anterior insula (uncorrected), among other regions, which were not observed in the CP/LCU group, even at uncorrected levels. These results tentatively suggest that, compared to TD controls, the two CP groups might be characterized by distinct grey matter differences in regions central to decision-making, empathy and emotion regulation. The more widespread reduction in the CP/HCU group might thus partly explain their different behavioural and neurobiological profiles (Frick and Viding
2009).
These results are consistent with evidence suggesting that CP/HCU in childhood and psychopathic traits in adulthood are associated with atypical OFC and ACC function (e.g., (Anderson and Kiehl
2012; Blair
2013). In OFC, Finger et al. (
2011) found reduced fMRI responses in a network of regions including OFC during a reinforcement learning task in youths with CP and elevated psychopathic traits; while Marsh et al. (
2011) found reduced OFC-amygdala connectivity in a similar sample during a moral judgment task. Behavioural work has also shown subtle impairments on OFC-dependent tasks such as reversal learning (e.g., Budhani and Blair
2005). In ACC, a recent study using a partially overlapping sample to that reported here found reduced responses in children with CP when viewing pictures of others in pain (Lockwood et al.
2013); moreover, activity in this region was negatively associated with levels of CU traits. A similar result in ACC was also found by Marsh et al. (2013). While we cannot equate functional hypo-reactivity and reduced GM volume, the current data are consistent with theories suggesting that atypical neural function in regions underlying emotional processing and reinforcement learning contributes to CP/HCU (e.g., Anderson and Kiehl
2012; Blair
2013). Future studies could use multimodal imaging to explore relationships between structural and functional measures in children with CP.
Of the four ROIs, amygdala and AI did not show group differences at FWE-corrected levels (although the CP group as a whole showed reduced right AI volume at
p < 0.001 uncorrected; Table S
1). This was somewhat surprising, since previous studies have found reduced volume of these regions in children and adolescents with CP (Fairchild et al.
2011; Sterzer et al.
2007), while several fMRI studies (including three based on a subset of the participants included in the current study) have found evidence for amygdala and/or AI hypoactivity during emotional processing in CP/HCU (Jones et al.
2009; Lockwood et al.
2013; Marsh et al.
2008; Sebastian et al.
2012; Viding et al.
2012). There is therefore strong evidence across imaging modalities for atypical amygdala and AI function in this group. However, not all sMRI studies in children with CP have found GM reductions in these regions. For example De Brito et al. (
2009) and Fahim et al. (
2011) did not find any group difference in amygdala volume, while Huebner et al. (
2008) did not observe reduced volume of the AI. Additionally, in the largest VBM study to date in this area (Ermer et al.
2013), only weak relationships were found between amygdala volume and psychopathic traits, while no relationships were observed between AI GM volume and psychopathic traits. Atypical amygdala and AI function therefore appear to be more robustly associated with CU traits than does atypical structure in these regions.
At the whole brain level, reduced GM volume was also seen in left middle frontal gyrus in CP/HCU relative to controls. While this region was not hypothesised to show group differences a priori, the result survived FWE-correction across the whole brain and so may represent an additional marker for CP/HCU. Studies reporting activation within 6 mm of the peak within this cluster (based on Neurosynth location data,
http://www.neurosynth.org/locations) have typically implicated this region in higher cognitive processes such as strategy use (Bor and Owen
2007), and context-dependent episodic retrieval (King et al.
2005). However, it has also been shown to contribute to emotional processes, for example dynamic (vs. static) emotional face perception (Trautmann et al.
2009). While it is too early to conclude that this region is implicated in the pathophysiology of CP/HCU, our finding could nonetheless be useful in motivating further investigations of this region’s structure and function in this group.
The pattern of GM reduction reported in the current study is not in line with our previous finding of increased GM concentration in OFC and ACC in CP/HCU (De Brito et al.
2009). In our view, the most likely explanation is the differing ages of the samples. The current sample were considerably older (mean age = 14 years, 0 months; age range: 10.2–16.9), whilst De Brito et al. (
2009) studied a younger group (mean age = 11 years, 7 months; age range: 10.0–13.3). These previous results were interpreted as reflecting delayed cortical maturation in the CP/HCU sample relative to the typically observed pattern of GM reduction with age (Gogtay et al.
2004; Shaw et al.
2008). Delayed cortical maturation is common to several developmental disorders (Shaw et al.
2010). However, developmental trajectories may look different at later points in the lifespan; for example in children with ADHD, (Shaw et al.
2012) found a delay in the age at which childhood increase in cortical thickness gives way to cortical thinning, similar to the delayed reduction in GM concentration seen in De Brito et al. (
2009). However, a longitudinal study including adults with ADHD symptoms found that by adulthood, symptom severity was associated with reduced GM thickness (Shaw et al.
2013). The relatively young age of participants in De Brito et al. (
2009) may therefore contribute to the differing pattern of results seen relative to the current study. However, longitudinal investigation of GM trajectories in CP is required to test this hypothesis directly.
While the current study has several methodological strengths, including a large sample size and the use of age- and gender-specific templates together with DARTEL registration, it is also worth noting some limitations. First, as with all previous sMRI studies of CP, given the cross-sectional design we are unable to infer whether neural differences are a cause or a consequence of the group differences observed. Second, VBM provides a composite measure of surface area and cortical thickness, and cannot provide a fine-grained analysis of specific GM metrics that may be driving the observed group differences (Raznahan et al.
2011). Further, given evidence that antisocial behaviour is more strongly heritable in children with CP/HCU than CP/LCU (Viding et al.
2005; Viding et al.
2008) and that surface area and cortical thickness are highly heritable, yet genetically unrelated (Panizzon et al.
2009), future studies should directly compare youths with CP/HCU and CP/LCU using those metrics. Additionally, while our groups did not differ significantly in age, IQ, ethnicity, handedness and SES, they differed on several comorbid psychopathology variables including ADHD, depression, anxiety and alcohol use. Therefore it could be argued that the group differences obtained are not specific to CP/HCU, but reflect a more severe profile of general psychopathology and a lower IQ. A related argument is that group differences resulted from differences in the severity of conduct disorder symptoms rather than CU traits. However, we think these alternative explanations are unlikely. Results from symptom and IQ covariate analyses were very similar to the main findings (see footnote Table S
1). Importantly, while CP/HCU and CP/LCU groups differed on both CU traits and CD symptoms (which are typically modestly correlated, and were correlated at
r = 0.53 in the current sample), only CU traits correlated negatively with left OFC volume. When the unique contribution of CU traits and CD symptoms to differences in left OFC was examined (after controlling for the other variable), the negative relationship with CU traits was strengthened, while the trend-level contribution of CD symptoms was reduced to near zero. Moreover, CU traits significantly improved the ability of our regression model to predict left OFC volume relative to CD symptoms alone. We took the decision to recruit a representative sample of children with CP as opposed to recruiting a fully matched sample which would likely have been unrepresentative in unpredictable ways. It is also worth noting that there is a strong theoretical basis to the idea that CU traits are a contributing explanatory factor for more severe conduct problem symptoms (Frick and Viding
2009); hence the focus on CU traits in the current study.
Overall, we replicate and extend previous studies showing a reduction in grey matter volume in children with CP in OFC and ACC: key regions of interest associated with emotional processing and reinforcement learning. Reductions in left OFC and right ACC were restricted to a subgroup of children with CP characterised by high levels of CU traits: reduced GM volume here therefore seems to characterise the CP/HCU, but not the children with CP/LCU, who also exhibit conduct disturbance but have differing genetic and neurocognitive vulnerabilities. To our knowledge, no previous study has compared CP/HCU and CP/LCU groups directly on tasks tapping the functions of these regions, so this may be a fruitful avenue of research. More generally, the present findings strengthen the case that it is important to take into account levels of CU traits in the diagnosis and treatment of children with CP.