Action prediction beyond the mirror neurons: novel fMRI evidence compared to a meta-analysis

Acting in a social environment requires predicting other people’s actions and their outcomes. This can be done based on contextual and movement cues. Here, we focused on the latter and explored the neurophysiology of action prediction processes when they are only based on postural information. We investigated whether the same brain regions previously reported by studies requiring action prediction during the observation of complex everyday-life or sport actions1,2 are recruited when participants have to predict the conclusion of overlearned simple actions, like grasping and pointing. We performed an event-related functional magnetic resonance (fMRI) experiment requiring 32 participants (17 females, age-range 19-27) to observe implied-motion pictures showing mid-flight grasping or pointing actions performed towards a cube-shaped object. Participants watched the implied-motion picture (200 ms, ‘observation phase’) followed by a mask (scramble version of the picture, duration 2000 ms) and by a final picture that correctly concluded the mid-fight action in only 50% of the cases. The participants responded whether the final picture matched with the mid-flight one (‘response phase’). See Figure 1a. The participants also performed a control color discrimination task, which was treated as a perceptually-matched baseline in the analyses. We analyzed which brain activations in the ‘observation phase’ (i.e., in response to the mid-flight action) predicted the participants’ behavioral performance in the ‘response phase’ of the experimental task (as compared to baseline). We also compared our fMRI results with those of a meta-analysis we performed using the ALE method3 on 15 fMRI studies (318 subjects, 256 foci) requiring motor predictions on more complex every-day life or sport actions. Overall, the participants showed brain responses that overlapped with those shown by the results of the meta-analysis (cluster-level FWE-corrected, as assessed with 1000 permutations), including a wide bilateral network of cortical areas comprising dorsal frontal and posterior parietal areas, as well as occipito-temporal regions including the posterior part of the superior temporal sulcus, all included in the so-called Action Observation ‘mirror’ Network4, and the pre-supplementary motor cortex. See Figure 1b. While predictions of sport-related actions also bilaterally activated the opercular cortices involved in action monitoring, this was not the case for pointing and grasping actions. Importantly, the individual performance at the action prediction task (indexed by inverse efficiency scores, i.e., the Reaction Times / Accuracy ratio) predicted the activity of two different brain regions (Figure 1c): a whole-brain regression analysis showed that better performance was associated with stronger occipital activations, whereas worse performance correlated with stronger activation of the right inferior frontal gyrus. These results were confirmed by a median-split multivariate pattern analysis that served as control: after dividing the sample in ‘good’ and ‘bad’ predictors groups, based on the behavioral performance, a support vector machine classifier could correctly classify from each participant’s activation patterns of the observation phase whether the participant was included in the ‘good’ vs. ‘bad’ predictors group (balanced accuracy = 84%, p < .001). Our results indicate that similar brain regions are responsible for motor predictions during the observation of both complex and simple actions: motor predictions based on postural information and regarding easy everyday-life actions, like grasping and pointing, require the whole action-observation ‘predictive’ network anyway, with the exclusion of opercular cortices. Moreover, we found interindividual differences in how participants approach this task, with a performance advantage for those who apply a more visually-based strategy involving more posterior brain regions.