Social and monetary reward processing in autism spectrum disorders

Twenty-one right-handed Caucasian ASD (mean age, 17.64 (3.45) years; age range, 13.58 to 25.91 years) and 21 right-handed Caucasian control participants (mean age, 17.00 ± 3.37 years; age range, 12.04 to 25.66 years) were included in the analyses. ASD participants were recruited through an associated genetics research program, clinical services, schools, and advocacy groups. Controls were recruited through schools, the university, and volunteer websites. Ethical approval was obtained from St. James’s Hospital/AMNCH (ref: 2010/09/07) and the Linn Dara CAMHS Ethics Committees (ref: 2010/12/07). Written informed consents/assents were obtained from all participants and their parents (where under 18 years of age).

Exclusion criteria included a Full Scale IQ (FSIQ) <70, known psychiatric, neurological, or genetic disorders, a history of a loss of consciousness for >5 min and those currently taking psychoactive medication. Four subjects in the ASD group had a secondary diagnosis of Attention Deficit Disorder (ADD) or Attention Deficit Hyperactivity Disorder (ADHD). Controls were excluded if they had a first-degree relative with ASD or scored >50 on the Social Responsiveness Scale (SRS)[29] or >10 on the Social Communication Questionnaire (SCQ)[30]. The adult prepublication version of the SRS was used with permission in cases 18 years or older[31]. All participants had normal, or corrected to normal, vision.

ASD diagnosis was confirmed using the Autism Diagnostic Observation Schedule (ADOS)[32] and the Autism Diagnostic Interview Revised (ADI-R)[33]. All ASD participants met criteria for autism on the ADI-R. Twelve participants met criteria for autism and nine met criteria for ASD on the ADOS. Clinical consensus diagnosis was established using DSM-IVTR criteria and expert clinician (LG).

FSIQ was measured using the four-subtest version of the Wechsler Abbreviated Scale of Intelligence (WASI;[34]) or the Wechsler Intelligence scale for Children-Fourth Edition (WISC-IV;[35]). Performance IQ (PIQ) score was based on the Matrix Reasoning and Block Design subtests and Verbal IQ (VIQ) score on the Vocabulary and Similarities subtests.

Figure1 illustrates the adapted versions of the MID[27] and the SID[13]. In both tasks participants had to respond as quickly as possible to a trigger (white square) while it remained on screen. The amount of time the participant had to respond to the trigger depended on the number of correct or incorrect prior responses (see below). Trigger cues were preceded by an instruction cue signaling the level of potential reward. For ‘reward’ trials a circle denoted that participants would be rewarded if they responded quickly enough (n per task = 60) while for ‘no reward’ trials a triangle denoted that the participant would not receive a reward, regardless of whether or not they responded quickly enough to the trigger (n = 30). Reward magnitude varied on two levels indicated by the number of horizontal lines on a cue stimulus. In the MID the levels of monetary reward were €0.20 (n = 30, preceded by a cue depicting a circle with one horizontal line) and €1.00 (n = 30, preceded by a cue showing a circle with two horizontal lines). Success was acknowledged by showing a picture of a coin with the money earned on that trial. In the case of a ‘no reward’ trial, or when participants did not respond to the trigger quickly enough, they were shown a coin stimulus of the same size and luminance but with no features. SID instruction cues were identical to MID instruction cue except in color. Feedback was a female face from the NimStim set of Facial Expressions[36] with a happy facial expression at two levels of intensity (small smile and larger smile), as used in previous studies of social reward[16]. This face stimulus was presented as it was rated as the most pleasant and attractive of the Caucasian faces in the NimStim set by a sample of 20 male participants (see supplementary material) and was used previously in a study of social reward in children[37]. Unlike the original SID task, which used 22 different faces, a single female face was used to remove novelty as a confounding difference between tasks. Two levels of social reward, rather than three (as in the original SID task), were used to reduce task duration. The ‘no reward’ facial stimulus was the same face graphically dysmorphed, with facial features eliminated but size and luminance retained.

Each task consisted of 90 trials (two 45 trial runs (TR) each lasting 9 min) presented in a counterbalanced order across participants. Each trial lasted 12 s (six TRs). A variable delay was introduced between the instruction cue and trigger (1,492 to 6,848 ms), and trigger and feedback (1,417 to 6,569 ms) to ensure that BOLD activity time-locked to the instruction cue was specific to reward anticipation and uncontaminated by the subsequent response or feedback. Similarly, activity at the time of feedback was specific to reward receipt and uncontaminated by reward anticipation or motor responses[38, 39]. This variable delay was achieved by randomly varying the onset time of instruction cues, triggers and feedback across the first two TRs (0 to 4 s), second two TRs (4 to 8 s) and last two TRs (8 to 12 s), respectively, from trial to trial. Cues and feedback were each presented for 1,000 ms. As in previous MID studies, the duration of the trigger was adjusted to maintain an accuracy rate for approximately two-thirds of trials. Response periods were reduced by 30 ms after each correct response, and increased by 90 ms when participants failed to respond within the given time frame. Manipulations of the response period were separate for each reward level given that RTs are known to be faster for higher levels of reward[11, 27, 40]. An upper limit was imposed, such that trigger duration could not exceed more than 500 ms.

Participants maintained focus on the cross hair in the center of the screen throughout the fMRI sessions. They were instructed to respond quickly to the trigger using a button in their right hand. For the MID they were told that they could ‘win’ real money up to a value of €30. All subjects were given €25 at the end of the experiment, regardless of their performance. For the SID they were informed that success would be acknowledged by a smiling face on the screen. Practice versions of each task (consisting of 30 trials) were performed to familiarize participants with the experiments prior to scanning.

MRI data were collected on a Philips 3 T Achieva MRI Scanner at the Centre for Advanced Medical Imaging (CAMI), St. James’s Hospital, Dublin. A high-resolution 3D T₁-weighted MPRAGE image was acquired for each participant (FOV, 256×256×160 mm³; TR, 8.5 ms; TE, 3.9 ms; total acquisition time, 7.3 mins; voxel size, 1×1×1 mm³). Two hundred and eighty functional images were acquired for each run using a T₂* weighted gradient echo sequence to visualize changes in the BOLD signal (TR, 2,000 ms; TE, 28 ms; flip angle, 90°; FOV, 256×256 mm²; voxel size, 3×3×3.5 mm³; slice gap, 0.35 mm; 38 slices; slice order scan order: ascending; total acquisition time, 9.3 min). Presentation® software (Version 14.4,http://​www.​neurobs.​com) was used for stimulus presentation. Subjects lay supine and stimuli were projected onto a screen behind the subject and viewed in a mirror above the subject’s face.

Behavioral data were analyzed using SPSSv16. Two sample t-tests were used to examine group differences in age, IQ measures, SRS, and SCQ scores. Mixed model (between/within subjects) ANOVAs were used to examine accuracy and reaction time (RT) data. Pearson’s correlations were conducted to examine the relationship between the BOLD response and SRS score and RT. Correlations between BOLD response and ADOS/ADI scores were calculated using Spearman’s rho, as ADOS/ADI scores are ranked/ordinal. Correlations were corrected for multiple comparisons using Bonferroni correction.

fMRI analysis was carried out in SPM8 (http://​www.​fil.​ion.​ucl.​ac.​uk/​spm) in Matlab 2009a (MathWorks Inc., UK). Before preprocessing, the origin was set to the anterior commisure for both T₁-weighted and EPI images. Slice-timing correction was then applied to the data, given the recent evidence that this approach is superior to flexible modeling strategies in correcting for differences in image acquisition time between slices[41]. The images were then realigned to correct for motion artefacts and co-registered to the skull stripped T₁-weighted image. Subjects (ASD, n = 6; controls, n = 4; additional to the 21 cases/controls presented here) were excluded for excessive head motion during scanning (that is, movements >3 mm). Normalization to standard stereotaxic space (Montreal Neurological Institute; MNI) was performed using the ICBM EPI template and the unified segmentation approach[42]. The data were then re-sliced to a voxel size of 2×2×2 mm³. Finally, the images were smoothed using a 5-mm full-width-half-maximum (FWHM) Gaussian kernel to conform to assumptions of statistical inference using Gaussian Random Field Theory[43, 44].

Nine event types were modeled at the first level for each task: anticipation/cue (‘no reward’, ‘small reward’, ‘large reward’, ‘error’), feedback (‘no reward’, ‘small reward’, ‘large reward’, ‘error’), and ‘trigger’. ‘Cue error’ and ‘feedback error’ comprised reward trials on which participants failed to respond within the given time frame. Nine regressors were created by convolving a delta function of event onset times for each event with the canonical hemodynamic response function (HRF). Given that slice time correction was used, micro-time onset was set to the middle temporal slice. Covariates of no interest included the six head motion parameters.

Key anatomical regions within the reward system (striatum, amygdala, vmPFC, OFC, and ACC) were defined a priori for small volume correction to correct for multiple comparisons at the family wise error rate (FWE; P <0.05). Masks for each of these regions were generated in FSL (http://​www.​fmrib.​ox.​ac.​uk/​fsl/​) using the Harvard Oxford cortical and subcortical atlases (http://​www.​cma.​mgh.​harvard.​edu/​). The caudate nucleus, putamen, and nucleus accumbens were combined into a striatal mask (one for each hemisphere) using the image calculator in SPM8. All masks were thresholded at >20% probability. Percent signal change in significant activations was calculated using the Anatomy Toolbox[45] in SPM8.

Reaction time values are shown in Figure2. A mixed model two-by-two-by-three ANOVA (between-subjects factor: group; within subjects factors: reward type and reward magnitude) revealed a significant effect of reward magnitude (F (1.61, 64.33) = 47.49, P <0.0001; faster responses to ‘reward’ compared to ‘no reward’) and a significant interaction between group and reward magnitude, (F (1.61, 64.33) = 4.70, P = 0.018). Pair-wise comparisons to examine the main effect of reward magnitude indicated a significant decrease in RT between ‘no reward’ and ‘small reward’ levels (t(41) = 8.660, P <0.0001) as well as ‘no reward’ and ‘large reward’ levels (t(41) = 6.112, P <0.001) but no difference in RT between the ‘small’ and ‘large’ rewards (t(41) = −1.592, P = 0.119). Difference scores (RT ‘small reward’ - RT ‘no reward’; RT ‘large reward’ - RT ‘no reward’; RT ‘large reward’ - RT ‘small reward’) were calculated to examine the group by magnitude interaction. These indicated that the ASD group showed less of a difference in RT between ‘no reward’ and ‘small reward’ (t(40) = −2.337, P = 0.025)) and between ‘no reward’ and ‘large reward’ than the control group (t(40) = −2.434, P = 0.020) but not between ‘large reward’ and ‘small reward’ (t(40) = −0.809, P = 0.424). There was no significant effect of group, group by reward type interaction, magnitude by reward type interaction, or group by reward type by magnitude interaction.

As discussed above mean accuracy was maintained for all conditions by adjusting the duration of the trigger from trial to trial (see Methods). However, a significant effect of reward magnitude was observed (F (1.65, 66.17) = 21.15, P <0.0001, that is, greater accuracy for ‘reward’ compared to ‘no reward’), using a mixed model two-by-two-by-three ANOVA (between-subjects factor: group; within subjects factors: reward type and reward magnitude). Pair-wise comparisons indicated a significant increase in accuracy between ‘no reward’ and ‘small reward’ levels (t(41) = 5.76, P <0.0001) as well as ‘no reward’ and ‘large reward’ levels (t(41) = 4.52, P <0.0001) but no difference between the ‘small’ and ‘large’ rewards (t(41) = −1.00, P = 0.323). No other significant effects were observed.

The DS is involved in the reinforcement of action[6], playing a fundamental role in goal-directed action through the selection of appropriate goals based on the evaluation of action outcomes[12]. Actor-critic models have informed understanding of striatal function, by positing that the ventral striatum (VS) predicts future rewards (‘the critic’) whereas the DS maintains information about the rewarding outcome of actions (‘the actor’)[48]. In line with this model, it has been found that the VS supports stimulus-reward learning whereas the DS is necessary for stimulus–response-reward learning[47]. The DS plays an important role in updating the reward value of chosen actions to guide subsequent behavior and maximize reward consumption[49, 50]. Representations of chosen actions can be used to aid learning or to modulate movements to reflect the value of the action, for example, by modulating RT[50].

In this study, the ASD group had difficulty modulating their RT according to reward level. The ASD group also showed reduced activation compared to controls in the DS for social rewards. Increased BOLD response in the DS to social rewards was associated with faster responses to social rewards in both groups. This suggests that participants with ASD may have difficulty in using social reinforcement to update reward representations and guide subsequent behavior. DS activation to monetary rewards was not associated with faster responses, implying that the region may be more important for social reward processing. Accumulating evidence implicates the DS in processing of complex social rewards such as trust[14], mutual social co-operation[20], receiving positive feedback about one’s personality[17], and altruistic punishment[19, 21]. Though the task used in the present study was a simple social reward task, the results further implicate the DS in social reward processing and suggest a deficit in ASD evidenced by deactivation to social rewards in this region.