Gamma power reductions accompany stimulus-specific representations of dynamic events.

Neural representations of specific stimuli rely on activity patterns in distributed neural assemblies [1-4]. According to one influential view, these assemblies are characterized by synchronized gamma-band activity (GBA) [5-11] that reflects stimulus-specific representations [12-14]. However, recent studies have shown that GBA is closely correlated with the overall amount of cellular activity and may be detrimental for precise representations of specific stimuli [15, 16]. Until now, the role of GBA for the formation of dynamically changing representations has been unknown. Here, we applied representational similarity analysis (RSA) [17] to intracranial electroencephalogram (iEEG) data from ten presurgical epilepsy patients to identify stimulus-specific neural representations. Patients first learned and then retrieved their paths through virtual houses. Dynamic representations were identified by the rapidly changing distributions of frequency-specific global (spatial) activity patterns across the brain. We found that GBA patterns during successful (but not unsuccessful) retrieval of one sequence were more similar to activity during encoding of that same sequence compared to other sequences. The contribution of individual electrodes to these global representations was correlated with local similarity in individual electrodes (i.e., with RSA across time). Moreover, time-resolved RSA values were negatively correlated with the magnitude of iEEG gamma power: RSA values were higher at time points when gamma power was reduced. Both global and local representations relied on a small proportion of electrodes. These results show that behaviorally relevant neural representations of specific dynamically changing stimuli can be tracked by iEEG recordings and that they are associated with reductions of gamma power.