Figure 1
Trial structure of the Incongruent Facial Emotion task. At the beginning of the task, the following instructions were displayed: “When instructed, either frown or smile. The face on the screen will then make a facial expression”. Each trial had the following structure. Participants fixated on a fixation dot at the centre of the screen (1000 ms). An instruction to smile or frown appeared at the centre of the screen (500 ms). Participants were told to remember the instruction but not to smile or frown yet. There was a variable delay (400–11,600 ms) between this instruction and the trigger to start smiling or frowning. The variable delay allowed the fMRI activity associated with making facial expressions to be dissociated from the fMRI activity to viewing the instruction. Delays were calculated using the optseq2 tool to optimize statistical power (Dale, 1999). The trigger was the reappearance of a fixation dot lasting 400 ms. Participants were instructed to smile or frown at the appearance of the trigger. Participants held the expression until the face was replaced by a fixation dot indicating the end of the trial. After an additional delay (750 ms), participants viewed the stimulus’ reaction (1750 ms), a gradual transformation of the neutral face to either a happy (panel a, condition Hh) or angry (panel b, condition Ha) expression (35 frames × 50 ms per frame). At the conclusion of the trial (the start of the next fixation period), participants were asked to return to a neutral facial expression.
Figure 2
Analysis of single-trial voluntary facial expression time series in a sample participant (participant 020). a) Exemplar time series for the first principal components (PC) of smile and frown responses. PCs were estimated in each participant separately. The first PC corresponds to the unique combination of facial action units that are most activated in that participant. Vertical lines show triggers indicating that the participant should smile (purple) or frown (red). b) A single participant’s trial-wise facial responses to smile-instruction trials. Time zero represents the trial’s trigger. Vertical grey lines demarcate trial segments. Black line represents the median response across all trials. c) Smile amplitude in the clinical and control groups. Each dot represents a single participant. Boxplots show data quartiles.
Figure 3
Clusters with significant activity when participants’ voluntary smile was followed by an incongruent angry stimulus. Left: right anterior insula, MNI coordinates (37.0, 19.2, 1.0). Right: right supplementary motor cortex, MNI coordinates (5.2, 7.5, 54.0).
Figure 4
Dynamic causal models with inter-areal connectivity (black arrows), direct task inputs (orange), and bilinear modulation of inter-areal connectivity by task (purple). Blue arrows indicate the effect of illness severity on bilinear modulation terms. Regular arrowheads indicate excitatory connections. Flat arrowheads indicate inhibitory connections. Network nodes comprised right fusiform (FUS), right anterior insula (AI), and right supplementary motor cortex (SMC). a) Full model. Bilinear connections were tested for significance. Circular arrowheads indicate that prior to model reduction, the valence of connections (excitatory, inhibitory) is unknown. b) Reduced model with illness severity (CGI) as a covariate. Participants with greater illness severity had greater disinhibition of AI and SMC by incongruent stimuli.