Report Influence of Uncertainty and Surprise on Human Corticospinal Excitability during Preparation for Action

Institute of Cognitive NeuroscienceUniversity College LondonLondon WC1N3BGUnited KingdomSummaryActions are guided by prior sensory information [1–10],which is inherently uncertain. However, how the motor sys-tem is sculpted by trial-by-trial content of current sensoryinformation remainslargelyunexplored.Previous work sug-gests that conditional probabilities, learned under a particu-larcontext,canbeusedpreemptivelytoinfluencetheoutputofthemotorsystem[11–14].Totestthisweusedtranscranialmagnetic stimulation (TMS) to read out corticospinal excit-ability (CSE) during preparation for action in an instructeddelay task [15, 16]. We systematically varied the uncertaintyabout an impending action by changing the validity of theinstructive visual cue. We used two information-theoreticquantities to predict changes in CSE, prior to action, ona trial-by-trial basis: entropy (average uncertainty) and sur-prise (the stimulus-bound information conveyed by a visualcue) [17–19]. Our data show that during preparation for ac-tion, human CSE varies according to the entropy and sur-prise conveyed by visual events guiding action. CSE in-creases on trials with low entropy about the impendingaction and low surprise conveyed by an event. Commensu-rate effects were observed in reaction times. We suggestthatmotoroutputisbiasedaccordingtocontextualprobabil-ities that are represented dynamically in the brain.Results and DiscussionA fundamentalfeatureof humanmovementis thatanticipatoryknowledge of an impending action improves the speed andaccuracy of responses. For example, reaction times (RTs)are faster when visual information indicates in advance whichaction we will have to make [11–14]. Sensory information pro-videsusefulcuesforguidingactions,whichmaybeprobabilis-tic in nature. Learning the relative probabilities of impendingactions may enable the nervous system to prepare motor out-put prior to an event.Sensory cues that predict action enable a gradual build-upof preparatory activity in premotor and motor cortex prior toaction [4–10, 20]. This build-up is reflected by specific excit-ability changes in corticospinal projections [15, 16]. Indeed,a growing number of studies demonstrate quantifiable effectsof preparatory sensory information on the peripheral motorsystem [21, 22] and the spinal cord [21]. This implies that thepredictive aspects of sensory information are learned andrep-resented explicitly in the brain and that these representationsinfluence action preparation at several levels [15, 16, 23–25].In the current study we asked how corticospinal excitability(CSE) changes when subjects prepare an action based on vi-sual cues (Figure 1A) under changing degrees of uncertaintyassociated with an impending action. Understanding howthebrainusesthepredictabilityofeventstoinformpreparationfor action requires models of how this predictability is learnedand represented over time rather than how they change onaverage. We, therefore, measured CSE prior to overt actionby measuring muscular responses to stimulation of the motorcortex using transcranial magnetic stimulation (TMS) (see theExperimental Procedures and the Supplemental ExperimentalProcedures available online). We used established computa-tional models to examine how the motor system might encodethe probability of future events for action preparation.Nai¨ve, healthy participants prepared one of two actions(thumb or little-finger flexion) in an instructed delay task. Inmost trials a visual cue (CS) validly predicted a subsequentimperative stimulus (IS); on invalid trials, it was followed bythe alternative imperative stimulus (Figure 1A). The proportionof valid cues (i.e., the predictability of required actions) wasvariedsystematicallyacrossexperimentalblocks.Cueandim-perative stimuli were sampled in each block from distributionscontaining 85%–15%, 70%–30%, or 55%–45% of valid-invalidtrials. By using TMS, we probed effector-specific changes inCSE after the CS but before the IS signaled the required action(see the Experimental Procedures). An initial (conventionalANOVA) analysis of RTs revealed that subjects, on average,responded faster in blocks containing more valid trials (blocktype,F