Introduction

Pediatric posttraumatic stress disorder (PTSD) is common, affecting 5% of youth by age 18 years (McLaughlin et al, 2013). Pediatric PTSD is frequently comorbid with other psychiatric illnesses including depression, anxiety, and substance use disorders (Cohen and Scheeringa, 2009). Although significant progress has been made in understanding the neural substrates of adult PTSD, the corresponding pathophysiology in youth remains incompletely understood. Elucidating neural dysfunction in pediatric PTSD, including potential developmental abnormalities, is vital to improving detection and treatment of the disorder.

Meta-analyses of adult PTSD neuroimaging studies suggest abnormal functioning of circuitry underlying emotion appraisal and regulation, including hyperactivation of the amygdala, and hypoactivation of medial prefrontal (mPFC) regulatory regions, including ventromedial (vm)PFC and dorsomedial (dm)PFC, to negative and threat-related stimuli (Etkin and Wager, 2007; Hayes et al, 2012; Patel et al, 2012). The dorsal anterior cingulate cortex (dACC) also shows functional abnormalities in adult PTSD in response to threat-related stimuli, although with mixed findings (Etkin and Wager, 2007; Hayes et al, 2012; Patel et al, 2012). Hippocampal findings have been variable, although hypoactivation has been observed during fear extinction tasks (Milad et al, 2009). Functional connectivity findings in adult PTSD have also been mixed, suggesting both greater (Fonzo et al, 2010; Gilboa et al, 2004; St Jacques et al, 2011) and lower (Stevens et al, 2013) amygdala–mPFC connectivity to negative stimuli. Overall, neural models of adult PTSD posit hyperactivation of fear-promoting regions (amygdala, dACC), and hypoactivation of emotion regulatory areas (vmPFC, dmPFC) to negative stimuli (Pitman et al, 2012).

Relatively few studies have examined functional brain abnormalities during emotion processing in pediatric PTSD. In response to trauma-related imagery, a small study revealed rostral (r)ACC hypoactivation, but no differences in amygdala activation, relative to healthy controls (Yang et al, 2004). Our prior report in this sample using trauma-related imagery also revealed no differences in amygdala activation, but showed dACC hyperactivation and reduced amygdala–mPFC connectivity in pediatric PTSD (Wolf and Herringa, 2016). During face processing, youth with posttraumatic stress symptoms (PTSS) showed hyperactivation of the amygdala and vmPFC, but no differences in dACC activation across neutral and emotional faces (Garrett et al, 2012). Decreased dmPFC activation to fear faces has been reported in female youth with PTSS, suggesting possible sex differences (Crozier et al, 2014). Finally, PTSS were found to be negatively correlated with amygdala–dmPFC/pre-supplementary motor area (SMA) functional connectivity to fear versus neutral faces (Cisler et al, 2013).

These studies provide initial evidence of prefrontal–amygdala dysfunction in pediatric PTSD/PTSS, yet important questions remain. First, it remains unclear whether abnormalities in face processing are emotion specific (eg, to threat), as seen in adult PTSD. Second, little is known regarding functional brain connectivity in pediatric PTSD that can provide important information about network-level dysfunction. Finally, age-related abnormalities in face processing have not been reported in pediatric PTSD, and this may provide important clues about neurodevelopmental divergence. Based on studies in healthy youth, this may be especially true for the development of prefrontal–amygdala (Burghy et al, 2012; Gabard-Durnam et al, 2014; Gee et al, 2013b; Herringa et al, 2013a) and prefrontal–hippocampus function (Herringa et al, 2013a). Furthermore, our prior study in this sample revealed decreased dmPFC activation and amygdala–vmPFC connectivity with age during trauma imagery, suggesting altered neurodevelopment in key emotion processing pathways in pediatric PTSD. Whether similar developmental abnormalities are present during emotional face processing remains unknown.

To address these knowledge gaps, we used functional (f)MRI to examine brain activation and functional connectivity to dynamic emotional faces in a cross-sectional sample of youth with severe PTSD and nontraumatized healthy youth. We predicted that youth with PTSD would display age-related sensitization of the amygdala based on prior studies of childhood adversity (see, eg, Gee et al, 2013a; Swartz et al, 2015; Herringa et al, 2016), and reduced mPFC activation and amygdala–mPFC connectivity relative to healthy youth. Importantly, we predicted that these abnormalities would be specific to threat-related emotion (angry) but would not be present for happy faces.

Materials and methods

Participants

Participant recruitment and assessment have been previously described (Keding and Herringa, 2015; Wolf and Herringa, 2016; Patriat et al, 2016) and are briefly summarized here. A total of 28 medication-free youth with PTSD were recruited from area mental health facilities, whereas 33 nontraumatized healthy youth of comparable age and sex were recruited by local advertisements. Exclusion criteria for youth with PTSD included active suicidality, history of psychotic disorder, bipolar disorder, or OCD; recent (past 4 weeks) substance abuse or dependence; IQ<70; unstable medical condition; recent use of psychotropic medication (past 4 weeks; 6 weeks for fluoxetine); MRI contraindication; and possibility of pregnancy in females. Healthy participants were free of any history of mental illness. Participant data were excluded for movement and early scan termination (n=3 PTSD, 5 healthy), leaving 25 youth with PTSD and 28 healthy youth in the final analyses. Written parental consent and youth assent were obtained for all participants. All procedures were approved by the University of Wisconsin Health Sciences institutional review board.

Clinical and Behavioral Assessments

All participants and their caregivers underwent traumatic events and psychiatric screening by a board-certified, child and adolescent psychiatrist (RJH) with the Kiddie Schedule for Affective Disorders and Schizophrenia (KSADS) (Kaufman et al, 1997). PTSD diagnosis was determined using modified DSM-IV criteria (Cohen et al, 2011) by combination of the KSADS and Clinician-Administered PTSD Scale for Children and Adolescents (CAPS-CA) (Weathers et al, 2001). Additional self/caregiver report of symptoms were obtained for PTSD using the UCLA PTSD Reaction Index (PTSD-RI) (Steinberg et al, 2004), depressive symptoms with the Mood and Feelings Questionnaire (MFQ) (Costello and Angold, 1988), and anxiety symptoms with the Screen for Child Anxiety Related Emotional Disorders (SCARED) (Birmaher et al, 1997). See Supplementary Material for further details on behavioral measures.

Dynamic Face Task

Participants underwent fMRI while completing a dynamic face task (Supplementary Figure S1) that employs implicit cognitive–emotional processing (Almeida et al, 2011; Herringa et al, 2013b). We used an implicit paradigm to ensure that the task could be successfully completed across the entire age range of our sample, as well as both PTSD and healthy youth. Thus, the chances of group- and age-related effects on fMRI are unlikely to be confounded by performance effects, a concern on tasks with increased cognitive load (intrinsic to many explicit paradigms). Participants were asked to identify the color of a semitransparent overlay atop a changing emotional face (angry, happy) or oval shape (control condition) distractor. Each dynamic face changed from neutral to emotion over a 1-s period. Three blocks, each with 12 faces, were presented pseudorandomly for each emotion condition. Emotional face blocks were interspersed by shape blocks, ensuring no two emotion blocks were presented sequentially. See Supplementary Material for further details.

Image Acquisition and Preprocessing

Please see Supplementary Material for details on image acquisition. Image processing was performed using AFNI (Cox, 1996) and FSL (Woolrich et al, 2009). T1 structural images were registered to the MNI152, 2 mm3 template with linear and nonlinear warps (FSL FLIRT and FNIRT). Functional data were slice-time and motion corrected, and aligned to their respective T1 images. The first three volumes of EPI time series were removed because of T1-equilibrium effects, and the transformation matrix used to register the T1-weighted image to MNI space was applied to the functional data. Volume-to-volume displacement (SSD) was estimated from the six rigid body motion registration parameters. Any functional volume with SSD >1 mm and its preceding volume were censored. Participants with ⩾20% of volumes censored were excluded from analysis. Functional data were smoothed using a 6 mm Gaussian kernel. Final voxel size was 2 × 2 × 2 mm.

Statistical Analysis

A first-level, within-subject model was constructed within AFNI 3dDeconvolve. The model included three blocks for each emotion condition, and the shape condition as fixed-effect regressors, modeled with a gamma hemodynamic response function of 1-s duration. Six motion parameters and their derivatives were included as nuisance regressors, along with four polynomial drift terms. General linear tests were used to create emotion-shape contrasts for each emotion block that were then used in group analyses.

A second-level, between-subject model was constructed within AFNI 3dMVM (Chen et al, 2014). A 2 × 2 × 3 mixed design included group (healthy, PTSD) as the between-subject factor, and emotion (angry-shape, happy-shape) and block (1–3) as within-subject factors. Age was included as a covariate and interactions with the primary factors were examined, whereas sex was included as a nuisance covariate. Note that block was modeled to ascertain group differences in habituation/potentiation with repeated stimulus type presentation, as indicated in our prior study (Herringa et al, 2013b). Because we did not find any habituation/potentiation effects by group, we averaged findings across block to facilitate presentation of findings. A priori search regions included bilateral mPFC and amygdala/hippocampus, using masks generated from AFNI standard template. The mPFC mask (Supplementary Figure S2) included the vmPFC (BA 10, 11, 25), dmPFC (BA 6, 8, 9), and ACC (BA 24, 32, 33). Multiple comparison correction was performed using Monte Carlo simulation in AFNI 3dClustSim (version April 2016). At a voxel-wise p=0.01, the corrected α=0.05 cluster threshold was 141 voxels and 32 voxels for the mPFC and amygdala/hippocampus respectively. An additional Bonferroni correction for the use of two a priori search regions was applied (corrected α=0.025). Additional results are reported outside of a priori search regions surviving whole-brain correction (cluster threshold 283 voxels).

Psychophysiological interaction (PPI) analysis was conducted within AFNI using seeds derived from our a priori search regions in the activation analysis. Voxel-wise interaction regressors were created using AFNI 3dSynthesize and 3dTfitter, measuring the correlation between voxels in a given seed region with other voxels for each emotion condition versus shape. Interaction regressors and seed time series were entered as additional regressors in the first-level model from the activation analysis, for each seed. Group-level PPI analyses were conducted as described for the activation analysis. Given the exploratory nature of the PPI analyses, we did not apply additional Bonferroni correction for multiple seed regions to reduce the rate of false-negative results.

Secondary Analyses

Three multiple linear regression models were run on extracted cluster averages in SPSS v. 21 (IBM, Armonk, NY) to examine (1) potential confounds in group differences, (2) their relationships with PTSD, depression, and anxiety symptoms, and (3) their relationships with trauma exposure measures. Models (2) and (3) were performed within the PTSD group only. Group differences in task performance were also assessed within a separate linear model. All analyses were covaried for age and sex.

Given the high rates of comorbid affective and anxiety disorders in our sample, we used a transdiagnostic, dimensional symptom approach to examine the relationship between symptom measures and brain findings within the PTSD group as previously described (Patriat et al, 2016). This analysis is summarized here with additional details in Supplementary Material. Using a principal component analysis (PCA) of PTSD, depressive, and anxiety symptoms from the PTSD-RI, MFQ, and SCARED, we extracted five symptom dimension components explaining 74.8% of the total variance in symptom measures: social aversion, hopelessness, negative affect, hyperarousal, and re-experiencing.

Results

Participant Characteristics and Task Performance

Participant characteristics are displayed in Table 1. The groups did not significantly differ in sex distribution, age, pubertal stage, IQ, or handedness. In addition, there were no group- or age-related differences on task performance. Please see Supplementary Material for further details.

Table 1 Participant Characteristics
Full size table

Regional Brain Activation

Results in a priori search regions are summarized in Table 2. Additional results including whole brain findings are displayed in Supplementary Table 1. Note that trauma exposure measures (number of trauma types, time elapsed since index trauma, index trauma type) were not significant predictors of activation in the following results.

Table 2 Summary of Functional Activation Differences between PTSD and Healthy Youth in a Priori Search Regions of the mPFC and Amygdala/Hippocampus (Corrected α=0.025)
Full size table

Within the amygdala and hippocampus, no PTSD main effects were present. However, a group by age interaction was observed in the right amygdala (extending into the anterior hippocampus; Figure 1a). Here, age was negatively correlated with activation in healthy youth, but positively correlated with activation in PTSD youth across emotion conditions (r=−0.43 and 0.55, respectively). However, amygdala activation was not related to symptom severity in the PTSD group.

Figure 1
figure 1

Regional brain activation abnormalities to dynamic emotional faces in pediatric PTSD. (a) A group by age interaction in right amygdala activation revealed decreases with age in healthy youth, but increased activation with age in PTSD youth across emotion conditions. Age-related findings remained when covaried for PTSD duration and time elapsed since index trauma. (b) Valence-abnormal brain activation to dynamic emotional faces in pediatric PTSD. A group by emotion interaction in the dACC (BA 24, 32) revealed hyperactivation in PTSD youth in the happy-shape contrast relative to healthy youth, with no group differences in the angry-shape contrast. Dorsal ACC activation in the happy-shape contrast was positively associated with social aversion symptoms within the PTSD cohort. All results were covaried for age and sex. Error bars indicate±1 SEM. dACC=dorsal anterior cingulate cortex.

PowerPoint slide

Full size image

Within the mPFC, a PTSD main effect in the right dmPFC (BA 9; Supplementary Figure S3) revealed greater activation, independent of emotion conditions, in PTSD compared with healthy youth. Furthermore, dmPFC activation was positively correlated with negative affect symptoms in the PTSD group (Table 2 and Supplementary Figure S3). In addition, a group by age interaction was present in the left rACC (BA 32; Supplementary Figure S3). Here, age was negatively correlated with activation in healthy youth, but positively correlated with activation in PTSD youth across emotion conditions (r=−0.65 and 0.17, respectively). Rostral ACC activation was also positively correlated with negative affect symptoms in the PTSD group (Table 2 and Supplementary Figure S3). Finally, a group by emotion interaction was observed in bilateral dACC (BA 24/32; Figure 1b). Surprisingly, PTSD youth showed hyperactivation to happy faces relative to healthy youth, with no differences to angry faces. Dorsal ACC activation to happy faces was positively correlated with social aversion symptoms in the PTSD group, with no correlation for angry faces (Figure 1b and Table 2).

Functional Connectivity

PPI results for the amygdala and dACC seeds are presented here given their central role in our hypotheses. PPI results for the rACC and dmPFC seeds are presented in Supplementary Material. All results from a priori search regions are summarized in Table 3. Additional results, including whole-brain findings, are displayed in Supplementary Table 2.

Table 3 Summary of Functional Connectivity Differences between PTSD and Healthy Youth in a Priori Search Regions of the mPFC and Amygdala/Hippocampus (Corrected α=0.025)
Full size table

Right amygdala connectivity

Group by emotion interactions were present with bilateral dmPFC extending into the supplementary motor area (BA 9/8; Figure 2a), and left ventrolateral (vl)PFC/superior temporal gyrus (BA 47/38; Figure 2b). In each case, PTSD youth showed reduced connectivity to angry faces but increased connectivity to happy faces, the reverse pattern of healthy youth. Amygdala–dmPFC and amygdala–vlPFC connectivity to happy faces were positively correlated with hyperarousal symptoms, with trending negative correlations to angry faces, within the PTSD group (Figures 2a and b).

Figure 2
figure 2

Valence-abnormal functional connectivity to dynamic emotional faces in pediatric PTSD. Seeds used in the connectivity analysis were derived from regional activation findings in the mPFC and amygdala/hippocampus. Group by emotion interactions were observed for functional connectivity in prefrontal–amygdala pathways involved in emotion appraisal and regulation. In each case, PTSD youth showed decreased functional connectivity to angry faces, but increased connectivity to happy faces, the reverse pattern of healthy youth. (a) Right amygdala to bilateral dmPFC/SMA that was positively and negatively associated with hyperarousal symptoms in the happy-shape and angry-shape contrasts respectively; (b) right amygdala to left vlPFC/STG that was positively and negatively associated with hyperarousal symptoms in the happy-shape and angry-shape contrasts respectively; and (c) bilateral dACC to bilateral dmPFC that was positively and negatively associated with trauma type load in the happy-shape and angry-shape contrasts respectively. All results were covaried for age and sex. Error bars indicate±1 SEM. dmPFC=dorsomedial prefrontal cortex; SMA=supplementary motor area; dACC=dorsal anterior cingulate cortex; vlPFC=ventromedial prefrontal cortex; STG=superior temporal gyrus.

PowerPoint slide

Full size image

dACC connectivity

A group by emotion interaction was found with bilateral dmPFC (BA 9; Figure 2c). Here, PTSD youth again showed reduced connectivity to angry faces and increased connectivity to happy faces, the reverse pattern of healthy youth. Dorsal ACC–dmPFC connectivity was associated with trauma load: as the number of trauma types endorsed increased, dACC–dmPFC connectivity decreased in angry-shape and increased in happy-shape, reaching significance for ⩾4 trauma types endorsed (Figure 2c).

Confound Analyses

All results remained significant or trending (p<0.1) when adjusted for potentially confounding variables. Please see Supplementary Material for further details.

Discussion

To our knowledge, this is the first study to report age-related and valence-specific functional neural abnormalities during emotional face processing in pediatric PTSD. Consistent with our hypotheses, our study revealed increased amygdala activation with age in pediatric PTSD, suggesting a developmental sensitization of the amygdala to emotional stimuli. Unexpectedly, our study also revealed heightened recruitment of emotion appraisal and regulatory circuits to happy faces, and decreased recruitment to angry faces. These patterns are opposite to that of healthy youth and appear to differ from adult PTSD, where abnormal prefrontal–amygdala function appears specific to negative emotional expressions. Importantly, valence-abnormal prefrontal–amygdala recruitment was associated with social aversion and hyperarousal symptoms. Together, these findings point to abnormal salience and appraisal processing of emotional faces in pediatric PTSD that may directly contribute to dysregulated fear and anxiety in these youth.

Our findings indicate abnormal function and coupling in a network comprising the amygdala, dACC, and dmPFC. Dorso-rostral aspects of the mPFC have been heavily implicated in emotion appraisal and cognitive–emotional conflict regulation (Comte et al, 2016; Etkin et al, 2011; Kalisch et al, 2006; Kalisch and Gerlicher, 2014; Mechias et al, 2010). Notably, these regions contribute to high-level appraisal of emotional stimuli in a context-dependent manner (Kalisch and Gerlicher, 2014; Maren et al, 2013), including threat–safety discrimination (Lissek et al, 2014). Abnormal functioning of this network to threat stimuli has been implicated in pathological states of fear and anxiety (Etkin and Wager, 2007; Wolf and Herringa, 2016), putatively via enhanced amygdala responses (Mechias et al, 2010; Robinson et al, 2012).

Within this framework, we originally hypothesized that pediatric PTSD would be characterized by hyperactivation of emotion processing regions (amygdala, dACC) to threat-related expressions, specifically angry faces. Adult PTSD studies (Dunkley et al, 2016; Felmingham et al, 2010; Fonzo et al, 2013) and studies of childhood trauma exposure (see, eg, Dannlowski et al, 2013; van Harmelen et al, 2013) suggest that this is indeed the case. However, our findings indicate that pediatric PTSD is characterized by relatively greater recruitment of emotion appraisal and conflict regulation circuits to happy faces, namely dACC hyperactivation, and increased amygdala–dmPFC and dACC–dmPFC connectivity. This suggests that youth with PTSD engage additional neural resources for successful task completion during happy faces, akin to processing of angry faces in healthy youth. Interestingly, amygdala activation itself did not show emotion-specific differential activation. Rather, valence-abnormal processing may begin with the dACC, the only node in the frontolimbic network to show emotion-specific activation differences.

The current findings extend those of our previous report in this sample examining prefrontal–amygdala function to threat images (Wolf and Herringa, 2016). In that study, we found dACC hyperactivation, accompanied by decreased amygdala–dmPFC connectivity, to threat images in PTSD youth. In contrast, the present study found dACC hyperactivation, accompanied by increased amygdala–dmPFC connectivity, to happy faces. One intriguing possibility is that youth with PTSD may engage similar implicit appraisal processes to threat images and happy faces. Here, dACC hyperactivation may reflect greater ambiguity of happy expressions with respect to threat. This notion is supported by the relationship between dACC activation and social aversion symptoms. At the same time, youth with PTSD may implicitly engage compensatory circuits, as suggested by increased amygdala–dmPFC coupling to happy faces, to allow successful task completion. On the other hand, the lack of dACC response to angry faces, combined with decreased amygdala–dmPFC coupling, may indicate insufficient recruitment of appraisal and regulatory resources to canonical threat expressions in pediatric PTSD. Together, these neural patterns suggest abnormal implicit threat–safety processing of emotional expressions, consistent with reports of abnormal facial emotion recognition in maltreated youth (Pollak et al, 2000; Pollak and Kistler, 2002; Pollak and Tolley-Schell, 2003).

Similarly, functional connectivity analyses revealed valence-abnormal connectivity between the dACC and dmPFC in pediatric PTSD. Here, youth with PTSD display increased dACC–dmPFC connectivity to happy faces, but decreased connectivity to angry faces. The dmPFC has been implicated in emotion appraisal, including the prediction of others’ intentions based on facial information (Bzdok et al, 2013; Isoda and Noritake, 2013; Rushworth et al, 2013). Functional connectivity between the dACC and dmPFC has also been implicated in the evaluation of others’ mental states (Li et al, 2014). In light of this, increased dACC–dmPFC connectivity to happy faces may reflect increased ambiguity in positive expressions, eliciting greater recruitment of regions involved in emotion appraisal.

Youth with PTSD also show increased amygdala–dmPFC and amygdala–vlPFC connectivity to happy faces, but decreased connectivity to angry faces. The amygdala, dmPFC, and vlPFC are key nodes in a network engaged during the evaluation of trustworthiness of other human faces (Bzdok et al, 2012), and their recruitment increases with cognitive–emotional conflict to emotional faces in healthy adults (Zaki et al, 2010). Furthermore, the dmPFC and vlPFC have been implicated in the regulation of emotional responses through their connections with the amygdala (Kim et al, 2013; Lee et al, 2012). Thus, it appears that pediatric PTSD is characterized by valence-inappropriate recruitment of emotion regulatory pathways while viewing happy and angry faces. Although this may allow for successful task completion, it may also reflect abnormal salience and appraisal processing of emotional faces, a notion supported by the association of these connectivity patterns with hyperarousal symptoms.

We find evidence of age-related abnormalities in amygdala function in pediatric PTSD that may indicate abnormal functional brain development. Consistent with prior studies in typically developing samples (Gee et al, 2013b; Vink et al, 2014), our healthy youth show decreased amygdala activation with age irrespective of face valence. In contrast, youth with PTSD show increased amygdala activation with age. These age-related findings remained in the PTSD group when covarying for illness duration and time elapsed since index trauma, suggesting they are not secondary to these effects. Interestingly, amygdala activation was not related to symptom severity in the PTSD group, suggesting it may reflect a normative sensitization to face stimuli across development that could confer enhanced threat detection. Indeed, multiple studies have shown that childhood adversity and trauma are associated with amygdala hyperactivation by adulthood irrespective of symptoms (see, eg, Gee et al, 2013a; Herringa et al, 2016; Kim et al, 2013; McCrory et al, 2011; Swartz et al, 2015). Thus, although age-related sensitization of the amygdala may be present in youth who develop PTSD, the current study findings suggest that valence-abnormal processing in higher-level cortical regions may ultimately be responsible for the expression of PTSD. Future work, including treatment intervention designs, would be merited to explore these possibilities.

The results presented here show partial overlap with the few existing fMRI studies of pediatric PTSD/PTSS. Adolescents with PTSS show amygdala and mPFC hyperactivation to neutral and emotional faces (Garrett et al, 2012), and this is consistent with our findings of amygdala sensitization with age and dmPFC hyperactivation. In both studies, mPFC activation was positively related to symptom severity. On the other hand, Crozier et al (2014) reported sex differences in dmPFC activation, where maltreated males showed hyperactivation but maltreated females showed hypoactivation to fear faces. Surprisingly, other studies of pediatric PTSS/PTSD have not reported dACC hyperactivation (Crozier et al, 2014; Garrett et al, 2012; Yang et al, 2004) that we have observed in our sample across different emotion tasks (Wolf and Herringa, 2016). One possible reason for this is greater PTSD severity and comorbidity within the current sample, where other studies have recruited youth based on lower symptom thresholds. Finally, Cisler et al (2013) found a negative relationship between PTSD symptom severity and amygdala–dmPFC/pre-SMA connectivity to fearful faces in girls with a history of assault. This is perhaps one of the most consistent findings to date in pediatric PTSD/PTSS, with both the current study and our prior study (Wolf and Herringa, 2016) showing decreased amygdala–dmPFC connectivity to negative emotional stimuli, and further predicting PTSD severity.

The present study details novel findings regarding neural dysfunction in pediatric PTSD. It is not, however, without limitations. First, the lack of a healthy, trauma-exposed comparison group mitigates interpretation of results exclusively in light of PTSD and not trauma exposure per se. Yet, secondary analyses revealed differential relationships with trauma exposure and PTSD, with the benefit of examining these variables within subjects. Second, although our post hoc analyses suggest specificity of neural abnormalities to PTSD, anxiety, and depression symptom dimensions, we cannot determine diagnostic specificity given that PTSD status was the primary inclusion criterion in the clinical group. It is also important to note that PCA components were derived from a relatively small sample of youth and may not generalize to other study populations. Third, network abnormalities for PTSD main effects (with no valence effect) could reflect general face processing abnormalities. In addition, caution should be taken to avoid reverse inference errors with regard to fMRI. Although not confounded by task performance, we cannot definitively conclude whether neural abnormalities reflect compensation, state, or trait effects in our sample given the cross-sectional nature of the study. On the other hand, relating network abnormalities to a third variable, such as symptom severity, does allow for reasonable interpretation of functional abnormalities. Finally, it is possible that stimulus differences account for the apparent neural differences between adult and pediatric PTSD, given that many adult PTSD studies use fear and not angry faces.

In conclusion, the present study offers novel insights into the neural dysfunction associated with emotional face processing in pediatric PTSD. Our findings suggest that pediatric PTSD is characterized by paradoxical, valence-abnormal prefrontal–amygdala recruitment during emotional face processing. These neural patterns suggest that youth with PTSD may require increased appraisal and regulatory resources for happy faces for successful task completion, akin to processing angry faces in healthy youth. These neural abnormalities suggest that positive emotional expressions may carry a greater degree of ambiguity and salience, whereas negative expressions have become reliably predictable for these afflicted youth. Although conceivably adaptive in abusive environments, abnormal face processing may come at the cost of poor or inefficient threat–safety discrimination, requiring heightened neural appraisal and regulatory resources in putatively safe contexts. Future studies would be merited to explore these findings longitudinally and assess whether successful treatment may restore normal functioning and development of this circuitry in pediatric PTSD.

Funding and disclosure

The authors declare no conflict of interest.