Emotional descriptions increase accidental harm punishment and its cortico-limbic signatures during moral judgment in autism

Individuals with autism spectrum disorder (ASD) present difficulties in integrating mental state information in complex moral tasks. Yet, ASD research has not examined whether this process is influenced by emotions, let alone while capturing its neural bases. We investigated how language-induced emotions modulate intent-based moral judgment in ASD. In a fMRI task, 30 adults with ASD and 27 neurotypical controls read vignettes whose protagonists commit harm either accidentally or intentionally, and then decided how much punishment the protagonist deserved. Emotional content was manipulated across scenarios through the use of graphic language (designed to trigger arousing negative responses) vs. plain (just-the-facts, emotionless) language. Off-line functional connectivity correlates of task performance were also analyzed. In ASD, emotional (graphic) descriptions amplified punishment ratings of accidental harms, associated with increased activity in fronto-temporo-limbic, precentral, and postcentral/supramarginal regions (critical for emotional and empathic processes), and reduced connectivity among the orbitofrontal cortex and the angular gyrus (involved in mentalizing). Language manipulation did not influence intentional harm processing in ASD. In conclusion, in arousing and ambiguous social situations that lack intentionality clues (i.e. graphic accidental harm scenarios), individuals with ASD would misuse their emotional responses as the main source of information to guide their moral decisions. Conversely, in face of explicit harmful intentions, they would be able to compensate their socioemotional alterations and assign punishment through non-emotional pathways. Despite limitations, such as the small sample size and low ecological validity of the task, results of the present study proved reliable and have relevant theoretical and translational implications.


Sample size estimation
Given our statistical design (2x2x2 mixed ANOVA, within-between interaction), a power analysis run on G*Power 3.1 1 showed that a sample size of 48 participants was required to achieve an effect size of 0.25 (as a previous study with this paradigm 2 ) with α = 0.05, and a power of 80%. A post-hoc power estimation confirmed that this assumption was met, since our analyses were performed on 57 participants, yielding a power of 0.87.

The Montreal Cognitive Assessment
The Montreal Cognitive Assessment (MoCA) 3 is a brief screening tool that includes fourteen subtests to tap the following domains: attention, visuospatial, visuoconstructional and executive functions, language, naming, memory, abstraction, and orientation. Its maximum score is 30 (cut-off: 26), with higher scores indicating better performance.

The INECO Frontal Screening
The INECO Frontal Screening (IFS) battery 4 includes eight subtests to assess three executive functions: response inhibition and set shifting, working memory, and abstraction capacity. The IFS maximum score is 30 (cut-off: 25), with higher scores indicating better performance.

Task-related fMRI data
While participants performed the experimental task, we acquired GRE-EPI volumes in sequential ascent, parallel to the anterior-posterior commissures, covering the whole-brain except the cerebellum. The following parameters were used: TR = 2000 ms; TE = 50 ms; flip angle = 90°; nº of slices = 20; matrix dimension = 64 × 64; voxel size in plane = 3.75 × 3.75 mm; slice thickness = 5 mm; interslice gap = 0 mm; sequence duration = 25 min; total number of volumes = 750. The run began with five dummy volumes (subsequently discarded from analyses) to allow for equilibration effects. After the fifth volume, the task was automatically triggered.
Before preprocessing, for each participant, we segmented the volumes corresponding to the task from the complete fMRI series (750 volumes in total). The remaining volumes were discarded. The ASD group took an average of 555.66 volumes (SD = 117) to perform the task, and the NT group an average of 579.85 (SD = 83.74). The difference was not statistically significant [t(52) = -0.90, p = 0.3], revealing that both groups required comparable time to complete the task. In addition, reaction times during the reading phase were similar between groups [MASD = 37.27 s, SDASD = 14.41; MNT = 37.88 s, SDNT = 6.77; t(42) = 0.2, p = 0.83].
As recommended by SPM12, and done in recent related works 5,6 , preprocessing steps included manual reorientation of each participant's scan series to the anterior-posterior commissures, slice timing correction using the middle slice of each volume as reference scan, realignment and unwrapping to correct for movement artefacts, co-registration with each participant's T1 structural image (after tissue segmentation), normalization to the MNI space employing the echo-planar imaging (EPI) template with an isotropic 2 mm voxel size to correct for inter-subject spatial variability, and smoothing using an 8 mm Gaussian kernel (full-width at half-maximum) to improve the signal-to-noise ratio.

Resting-state fMRI data
We obtained resting-state fMRI recordings from 56 participants. GRE-EPI volumes were acquired in a sequentially ascending order, parallel to the anterior-posterior commissures, covering the whole brain except the cerebellum. The following parameters were used: TR = 2000 ms; TE = 50 ms; flip angle = 90°; nº of slices = 22; matrix dimension = 64 × 64; voxel size in plane = 3.75 × 3.75 mm; slice thickness = 5 mm; interslice gap = 0 mm; sequence duration = 7 min; total number of volumes = 210. Participants were requested to lie still with their eyes open (to prevent them falling asleep 7 ) and not to think about anything in particular.
Before preprocessing, the first 10 volumes of each participant's series were discarded to ensure that magnetization achieved a steady state. Then, following previous procedures 8- 13 volumes were manually reoriented the to the anterior-posterior commissures, slice-timing corrected (using the middle slice of each volume as the reference scan), realigned to the first scan of the session to correct head movement (SPM functions, called by DPARSF), normalized to the MNI space using the EPI template from SPM, smoothed using an 8 mm full-width-at-half-maximum isotropic Gaussian kernel (SPM functions, called by DPARSF), and filtered (0.01-0.08 Hz). Six motion parameters, CFS, and WM signals were regressed out to reduce the effect of motion and physiological artifacts such as cardiac and respiration effects (REST V1.7 toolbox, called by DPARSF). Motion parameters were estimated during realignment. CFS and WM masks were derived from the tissue segmentation of each participant's T1 scan in native space using SPM12 (after co-registration of each participant's structural image with the functional image).

fMRI analysis of the reading phase
Given that text processing might differ between ASD and NT people, as a complementary analysis, we re-run the fMRI ANOVA using the reading phase of each participant as input instead of the 'decision phase'. At the first level, contrast images were calculated for the accidental > intentional harm contrast by applying linear weights to the parameter estimates. At the second-level group analysis, we performed a between-subject ANOVA (SPM module) with language and group as factors. No interaction results survived when controlling for multiple comparisons at the cluster level (p < 0.05, corrected for multiple comparisons using AlphaSim, k ≥ 203). Thus, post-hoc group comparisons were not performed.     Data are presented as mean (SD). n indicates the sample size for each group corresponding to each condition after outlier removal (e.g., In the ASD group, 15 participants read GL descriptions and 15 participants read PL descriptions, as language is a between-subject factor. In the accidental harm condition, one outlier datapoint was removed in each language condition, resulting in n = 14 subjects per group). ASD: Autism spectrum disorder; GL: graphic language; NT: neurotypical; PL: plain language.