Neural bases of Theory of Mind in children with autism spectrum disorders and children with conduct problems and callous-unemotional traits

Individuals with autism spectrum disorders (ASD) have difficulty understanding other minds (Theory of Mind; ToM), with atypical processing evident at both behavioural and neural levels. Individuals with conduct problems and high levels of callous-unemotional (CU) traits (CP/HCU) exhibit reduced responsiveness to others' emotions and difficulties interacting with others, but nonetheless perform normally in experimental tests of ToM. The present study aimed to examine the neural underpinnings of ToM in children (aged 10–16) with ASD (N = 16), CP/HCU (N = 16) and typically developing (TD) controls (N = 16) using a non-verbal cartoon vignette task. Whilst individuals with ASD were predicted to show reduced fMRI responses across regions involved in ToM processing, CP/HCU individuals were predicted to show no differences compared with TD controls. The analyses indicated that neural responses did not differ between TD and CP/HCU groups during ToM. TD and CP/HCU children exhibited significantly greater medial prefrontal cortex responses during ToM than did the ASD group. Within the ASD group, responses in medial prefrontal cortex and right temporoparietal junction (TPJ) correlated with symptom severity as measured by the Autism Diagnostic Observation Schedule (ADOS). Findings suggest that although both ASD and CP/HCU are characterized by social difficulties, only children with ASD display atypical neural processing associated with ToM.

Whilst behavioural evidence suggests that ToM is impaired in ASD but intact in CP/HCU, no study has directly compared the neural basis of ToM across these groups. Indeed, only one imaging study has directly compared ASD and anti-social traits, determining differential contributions to structural brain development in children, with only ASD traits associated with cortical thinning in superior temporal regions recruited during performance (Wallace, Shaw, Lee, Clasen, Raznahan, Lenroot, Martin & Giedd, 2012). Imaging methods can pick up subtle differences not always detectable at the cognitive or behavioural levels (e.g. Carter, Williams, Minshew & Lehman, 2012;Kana, Keller, Cherkassky, Minshew & Just, 2009), and as such provide a clearer picture of whether atypical processing associated with ToM is limited to ASD.
Here we used a cartoon-based vignette task (Sebastian, Fontaine, Bird, Blakemore, De Brito, McCrory & Viding, 2012b;Sebastian et al., 2012a) to explore the neural bases of ToM in ASD, CP/HCU and TD children. We compared responses during cartoons requiring ToM (understanding intentions) versus physical causality (PC) (understanding cause and effect without ToM demands). The CP/HCU and TD groups reported here contributed data to a previous paper (Sebastian et al., 2012a), which focused on a third, affective condition, tapping understanding of emotions within an intentional, narrative context. In the previous study, the focus was on the CP/ HCU group and contrasting affective versus ToM conditions to examine affective processing in children with conduct problems. The current study focuses on the comparison of ASD and CP/HCU groups in ToM processing relative to the PC control condition. Since affective processing in an intentional context involves the integration of ToM and empathy-related processes (Shamay-Tsoory, 2011), and hence requires intact ToM, the affective condition was not analysed in the current study, since interpretation of any observed differences would be problematic given likely ToM deficits in the ASD group. Previous use of the present vignette task (Sebastian et al., 2012a(Sebastian et al., , 2012b, comparing ToM versus PC conditions in typically developing populations has yielded reliable and replicable responses in the 'mentalizing network' comprising the pSTS extending to the TPJ, precuneus, temporal poles, and mPFC, in line with other studies using nonverbal cartoon stimuli (Brunet, Sarfati, Hardy-Bayle & Decety, 2000;Carter et al., 2012;Gallagher, Happ e, Brunswick, Fletcher, Frith & Frith, 2000;Kana, Libero, Hu, Deshpande & Colburn, 2014;V€ ollm, Taylor, Richardson, Corcoran, Stirling, McKie, Deakin & Elliott, 2006).
Three fMRI studies examining ToM in ASD have been conducted with child samples. One study, using static cartoon stimuli in children aged 7 to 16 with and without ASD, reported significantly reduced activation in mPFC, STS, left temporal pole, and precuneus during ToM (Carter et al., 2012). The two other studies have reported similar results, including negative correlations between autistic social symptoms and superior temporal responses (Wang et al., 2006), and between social responsiveness and medial PFC activation (Wang, Lee, Sigman & Dapretto, 2007).
This study aimed to test whether neural processing associated with ToM is abnormal in ASD in contrast to CP/HCU. Despite social deficits in both groups, they are rarely directly compared in the existing literature, and have never been compared with fMRI. To our knowledge, ours is also the first fMRI study to examine the neural correlates of ToM in children with CP/HCU, and the first to extend use of a non-verbal cartoon vignette task to a sample of children with ASD. Use of nonverbal vignettes to assay ToM is an advantage because individuals with ASD are reported to show impairments in deriving meaning from verbal stimuli (e.g. Randi, Newman & Grigorenko, 2010). In line with previous studies, we predicted reduced activation across the 'mentalizing network' in individuals with ASD compared to both CP/HCU and TD control groups. In particular, atypical responses were predicted in mPFC, pSTS/TPJ, and temporal poles in ASD. By contrast, based on cognitive-experimental evidence, we predicted intact responses in the CP/HCU group.
The ASD group were identified through their previous participation in research studies at King's College London or UCL. They were originally recruited from the community via schools and parent groups, and had a clinical diagnosis of autism or Asperger syndrome. They were assessed using the Autism Diagnostic Observational Schedule (Lord, Risi, Lambrecht, Cook, Leventhal, DiLavore, Pickles & Rutter, 2000) and, for nine participants, a full developmental interview (either the Autism Diagnostic Interview -Revised (ADI-R) or the 3di; Table  S1) (Lord, Rutter & Couteur, 1994;Skuse, Warrington, Bishop, Chowdhury, Lau, Mandy & Place, 2004). For the remainder, developmental history was assessed using the Social Communication Questionnaire (based on the ADI; Eaves, Wingert, Ho & Mickelson, 2006). No cutoff for conduct problems or CU traits was imposed in the ASD group, but all ASD participants scored below the median for CU traits reported by Sebastian et al. (2012a), and ASD and TD groups did not differ on CP symptoms (from the Child and Adolescent Symptom Inventory-4R -Conduct Disorder subscale, CASI-CD; Gadow & Sprafkin, 2009). In line with previous studies (e.g. Bird, Silani, Brindley, White, Frith & Singer, 2010), participants who did not meet full criteria on all ASD assessment measures but had a clinical diagnosis were retained in the analyses. This allowed us to recruit sufficient numbers of high-functioning participants who could tolerate scanning. No participants were currently taking psychoactive medication, except for occasional melatonin (N = 1), which was not used for 24 hours prior to scanning. Participants with CP/HCU scored in the clinically relevant range on the CASI-CD subscale (> 2 at 10-14 years; > 5 at 15-16 years), and comprised the top 50% of the sample from Sebastian et al. (2012a) in terms of scores on the Inventory of Callous-Unemotional traits (ICU; Essau, Sasagawa & Frick, 2006). TD controls all scored below the clinical threshold for conduct problems and below 45 on the ICU. Parent/ guardian data screening for psychiatric and neurological conditions, general psychopathology (including conduct problems and CU traits), and demographic data (including parent-defined ethnicity, handedness and socioeconomic status, based on National Statistics Socioeconomic Classification NS-SEC coding; www.ons.gov. uk) was available for all participants and is displayed in Table 1. To ensure that case groups were representative of ASD and CP/HCU, co-occurring symptoms of generalized anxiety disorder, major depression, substance abuse, or ADHD did not result in exclusion, but were measured using the CASI and their effects explored in the analyses. All participants completed the Wechsler Abbreviated Scale of Intelligence (2-subtest version; Wechsler, 1999), the Alcohol Use Disorders Identification Test (AUDIT; Saunders, Aasland, Babor, de la Fuente & Grant, 1993), and the Drug Use Disorders Identification Test (DUDIT; Berman, Bergman, Palmstierna & Schlyter, 2005).
Nineteen children with ASD were scanned. Data from three ASD participants were excluded due to excessive motion, leaving a final ASD sample of 16 to be compared with the 16 TD and 16 CP/HCU participants described above. Written informed parental consent and written assent from participants was obtained. Groups were matched on age, IQ, SES, gender and handedness (see Table 1).

Experimental task
The task involved 30 cartoons, 10 each for ToM, physical causality (PC) and Affective ToM conditions. As noted in the introduction, the present study focuses on ToM and PC conditions only. Each cartoon was silent and static, and depicted two people in everyday scenarios (e.g. pouring a drink; going for a walk).
The task was structured as in Sebastian et al. (2012aSebastian et al. ( , 2012b. In total, 30 cartoons (10 of each condition) were used, presented in sets of six, with a 15-second fixation period between sets. The six cartoons in each set included two cartoons from each condition, which were always yoked together. The order in which the ToM, Affective ToM and PC cartoon pairs in each set were presented was randomized for each participant. Each cartoon exemplar was presented once only.
Each cartoon involved four sequential frames. The first screen (3 seconds) displayed 'What happens next?' This was followed by three sequentially presented story frames, each presented for 2 seconds. The final screen, displayed for 5 seconds, showed a choice of two possible endings for the cartoon. During this time participants made their choice using a keypad. The inter-stimulus interval was 1 second, so each trial lasted 15 seconds in total.
For ToM cartoons, selecting the correct ending required understanding behaviour based on intentions (e.g. using an umbrella to help reach a door handle). PC cartoons required an understanding of cause and effect reasoning (e.g. understanding that a hat cannot blow against the wind). Affective ToM cartoons involved understanding behaviour based on emotion.

fMRI data acquisition
Images were acquired using a Siemens Avanto 1.5-T MRI scanner. These included a 5.5-minute T1 weighted structural scan, and 184 multislice T2-weighted echo planar volumes with blood oxygenation level-dependent contrast, taken during 1 9-minute run of the cartoon task. Acquisition parameters were: 35 2 mm slices with a 1 mm gap; echo-time = 50 milliseconds; repetition time = 2975 milliseconds; slice tilt = À30 (T > C): flip angle = 90 o ; field of view = 192 mm; matrix size = 64 9 64. Fieldmaps were also obtained and used to adjust functional scans for deformations due to magnetic field in-homogeneities during pre-processing.
fMRI data analysis SPM8 (http://www.fil.ion.ucl.ac.uk/spm) running in MATLAB R2007b was used to analyse imaging data. Five volumes were removed from the beginning of the sequence and two from the end, to allow for T1 equilibration and the final 'Thank you' screen. Voxel displacement maps were created from the fieldmaps for each participant, and used during the realign and unwarp stage of pre-processing. Images were normalized using warps created from segmented structural scans which had been co-registered to the functional data, written with a voxel size of 2 9 2 9 2 mm, and smoothed using an 8-mm Gaussian kernel. Images showing visible motion-related distortions were removed and interpolated using adjacent scans to prevent distortion of the between-subjects mask. Interpolated scans were then regressed out in the first-level design matrix. Movement artefacts were detected in 17 participants (TD: N = 3, ASD: N = 7, CP/HCU: N = 7), and always constituted less than 10% of each subject's data.
The first-level design matrix deconstructed the time series into segments corresponding to each of the three cartoon types (11 seconds each), periods of fixation (15 seconds) or instructions (3 s) and inter-stimulus intervals (1 s). The regressors were modelled as box-car functions and convolved with a canonical hemodynamic response function. Realignment parameters and interpolated scans were also included as regressors. The contrast of interest (ToM > PC) was estimated in each participant. Contrast images were then used in secondlevel analyses; with group (TD; CP/HCU; ASD) as the between-subjects factor in a one-way ANOVA.
Main effects for ToM > PC were thresholded at p < .05 family-wise error (FWE) corrected at peak level. For interactions with group, a priori regions of interest (bilateral medial/ventromedial prefrontal cortex, temporal poles and pSTS/TPJ) were defined using the aal atlas in the WFU Pickatlas toolbox for SPM (Maldjian, Laurienti, Kraft & Burdette, 2003;Tzourio-Mazoyer, Landeau, Papathanassiou, Crivello, Etard, Delcroix, Mazoyer & Joliot, 2002). Results within these ROIs were thresholded at p < .05, small-volume FWE-corrected at the peak level within each ROI. The Marsbar toolbox (Brett, Jean-Luc, Valabregue & Poline, 2002) was used to extract mean responses across significant clusters within ROIs for plotting purposes, and for conducting correlational analyses within the ASD group. An exploratory whole brain analysis was also conducted, with results reported at p < .001, k > 5, uncorrected (see Table S3).

ToM > PC: main effects
Regions surviving whole brain FWE-correction at p < .05 for ToM > PC are detailed in Table 3. Significant clusters were detected in the bilateral TPJ, extending to occipito-temporal cortex, bilateral temporal poles, left parahippocampal gyrus and posterior cingulate cortex.
No responses were observed in the medial prefrontal cortex (mPFC) when data from all three groups were included, although at an uncorrected threshold of p < .001, k > 5, a cluster was seen in ventromedial PFC (MNI: x = 4, y = 50, z = À14, k = 15; Figure 1a). When main effects were re-run with only TD and CP/ HCU groups, a cluster in the anterior rostral mPFC (MNI: x = 6, y = 54, z = 14; k = 5) survived FWEcorrection at p < .05 (Figure 1b). Reverse contrasts are included in Table S2.

ToM > PC: interactions with group
We first examined whether TD and CP/HCU responses differed within ROIs and across the whole brain for ToM > PC. As predicted, no differences between groups survived small-volume FWE-correction within ROIs, and only one small cluster (middle frontal gyrus, k = 6) was significant in the whole brain at a liberal threshold of p < .001, k > 5 (Table S3). Because no regions, bar this one area, showed group differences at corrected or liberal thresholds for either TD > CP/HCU or CP/HCU > TD, we compared ASD against TD and CP/HCU groups combined (set up in contrasts as 0.5 (TD) 0.5 (CP/HCU) and À1 (ASD)). Within a priori ROIs, three clusters in medial/ventromedial PFC showed a Group 9 Condition interaction in the predicted direction (TD = CP/HCU > ASD for ToM > PC) at p < .05, small-volume FWE-corrected (Table 4). No other ROIs yielded significant results.
Parameter estimates were extracted by averaging across voxels in the clusters surviving small-volume FWE-correction using Marsbar. Figure 2 illustrates that neural responses of children with ASD were significantly reduced compared with TD and CP/HCU children. Results for group comparisons were very similar when symptom scores which significantly differed between groups (ADHD, depression and anxiety) or IQ were included as covariates.

Correlational analyses in the ASD group
To explore whether differences in functional response for the clusters exhibiting Group 9 Condition interactions were also associated with severity of symptoms within the ASD group, we investigated the relationship between parameter estimates across clusters for ToM > PC and Table 3 Regions showing a main effect at p < .05 with FWE correction at the peak level for ToM > PC. Abbreviations: TPJ = temporoparietal junction; BA = Brodmann area; k = cluster size; ext. = extending to. Where more than one BA is shown, the peak voxel falls in the first BA but the cluster extends to the others listed  combined social and communication subscales of the ADOS. An inverse correlation was observed between mPFC response (cluster C) and social and communication subscales combined (r = À.69, p = .004; Figure 3). This result survives correction for multiple comparisons across the three correlations performed. Results for the other mPFC clusters were in the predicted direction, but not significant. Bilateral TPJ was also of a priori interest because of its relevance to ToM (e.g. Saxe & Kanwisher, 2003), and previous reports of correlations between functional response during ToM and ASD symptoms (Kana et al., 2009;Lombardo et al., 2011). Marsbar was used to extract mean contrast estimates within the ASD group for right and left TPJ clusters showing a main effect of ToM > PC at whole brain FWE-corrected levels across the whole sample. There was no significant correlation between ADOS social and communication subscales combined; however, the social subscale was inversely correlated with rTPJ functional response (r = À.57, p = .027; Figure 4). Cook's distance and leverage values were within acceptable limits.

Discussion
This study is the first to compare the neural basis of ToM processing in two groups of children who present with marked social deficitsthose with ASD and those with CP/HCU. Compared to CP/HCU children, children with ASD exhibited reduced activation in the medial prefrontal cortex during ToM processing. The same  pattern was observed when ASD children were compared to TD controls. There were no differences across ROIs between TD and CP/HCU groups. These findings indicate that whilst individuals with ASD show atypical neural processing, CP/HCU do not have a functional neural impairment during ToM. This is consistent with neurocognitive models that identify a core deficit in ASD as 'knowing' about others' mental states (Baron-Cohen et al., 1985). By contrast, this process appears spared in CP/HCU children who are instead characterized by not caring about others' feelings (Sebastian et al., 2012a;Jones et al., 2011;Blair, 2005). Reduced mPFC responses in ASD compared to TD and CP/HCU children are in line with previous studies in both adult (Castelli et al., 2002;Happ e et al., 1996;Kana et al., 2009;Kennedy & Courchesne, 2008;Murdaugh et al., 2012;Watanabe et al., 2012) and developmental (Carter et al., 2012;Wang et al., 2007) ASD samples, even when, as in the present study, behavioural responses do not differ. The mPFC is considered of central importance in ToM. Amodio and Frith (2006) propose that this region could be implicated in our ability to reason about other minds in the abstract, and integrate knowledge about their attributes with ongoing processing of intentions. In the ROI analysis, all clusters that survived correction were in the left hemisphere. This was surprising given that neuropsychological studies have implicated the right hemisphere in ToM (e.g. Brownell, Griffin, Winner, Friedman & Happ e, 2000). However, lowering the threshold revealed that the mPFC cluster exhibiting reduced responses in ASD vs. TD and CP/ HCU children did extend to the right hemisphere.
Extracted parameter estimates from mPFC illustrate that whilst TD and CP/HCU groups exhibited an increase in functional response during ToM relative to PC, the ASD group displayed a relatively reduced response. Though most studies have reported group differences in ASD reflecting a lack of differential response for ToM compared to baseline, one previous study also reported a relative deactivation for this contrast (in the RTPJ when making mentalistic judgements about the Queen's views; Lombardo et al., 2011). Other studies have not provided parameter estimates for contrasts, making it difficult to explore relative deactivation in ASD. Differences in the paradigms used (e.g. whether or not the PC condition involved agents) could also contribute to differences in results across studies (Castelli et al., 2002). One possible explanation for relative reductions in functional response for ToM may be that the presence of people in the logically engaging PC condition provoked more social processing than the ToM condition in the ASD group. Alternatively, mPFC responses could reflect domain general computation selectively engaged in ASD when processing cause and effect related to physical events.
Within the ASD group, autistic symptoms correlated significantly with functional response for ToM > PC in the most ventral of three clusters exhibiting a Group 9 Condition interaction in the medial prefrontal cortex. As socio-communicative impairment increased, participants' mPFC response for ToM > PC decreased, in line with similar reports in previous studies (e.g. Wang et al., 2007;Watanabe et al., 2012). This finding is consistent with group-level differences, and suggests that severity of autistic symptoms is related to degree of atypicality of social processing in mPFCalthough causal direction cannot be assumed.
A further a priori region of interest was the TPJ, strongly implicated in ToM (Gweon, Dodell-Feder, Bedny & Saxe, 2012;Saxe & Kanwisher, 2003) and showing reduced responses during ToM in ASD compared with controls in previous studies (Kana et al., 2009;Lombardo et al., 2011). Consistent with these reports, RTPJ response for ToM > PC showed a negative relationship with ADOS social symptoms. However, no RTPJ clusters exhibited a Group 9 Condition interaction at small-volume FWE thresholds, or even at a liberal threshold (p < .001, k > 5; Table S3), in contrast to previous studies in adults (Lombardo et al., 2011).
In conclusion, ASD appears characterized by attenuated mPFC activation during ToM processing. By contrast, CP/HCU children show typical patterns of neural response during ToM processing, comparable to that seen in controls. Given the small sample size, these results should be considered preliminary until replicated in a larger population. Future studies could also explore differences in functional or effective connectivity across the social brain network in these groups. Given the need for methods to differentiate ASD from other clinical populations, including CP/HCU, these findings suggest that sensitive indicators of ToM could assist in differentiating these groups in the clinic.