, 2007 and Lundqvist et al., 2010) and ensure irregular low-rate CHIR-99021 cost firing (Brunel and Wang, 2003 and Lundqvist et al., 2010). To illustrate the local origin of gamma, the basket cells were demonstrated to fire with a slight delay in time when compared to the pyramidal cells (Fig. 5A), which created the descending slope of the gamma cycle. This preferred phase of firing was not seen (in fact, tests for uniform circular distribution could not be rejected) when the same analysis was performed with the use of spiking activity of basket cells in the neighboring hypercolumns. We

were also interested in the mechanism underlying the relatively broad gamma peak, seen in Fig. 2C and D, when compared to the other distinct frequency components in the power spectrum. For this purpose, we examined the temporal evolution

of the rhythm over the attractor activation period. We found that the dominant gamma frequency was higher at the burst onset and then gradually slowed down (black and red curves, respectively, in Fig. 5B) due to adaptation. We have previously reported that the modularization into distinct hypercolumns with local feedback inhibition significantly increases the stability of persistent oscillatory activity (Lundqvist et al., 2010) through desynchronization of excitatory inputs. Here, we aimed to study how this desynchronization affected long-range coherence in the gamma band. To this end, we applied a moving window approach over entire trials in both simulation paradigms and examined the Cyclopamine average coherence. As expected, strong coherence was found locally within each hypercolumn (Jacobs et al., 2007 and Sirota et al., 2008) whereas between different hypercolumns over distances up clonidine to 500 μm the coherence was significantly weaker (Fig. 5C) in agreement with experimental findings (Sirota et al., 2008). Further, we increased the conductance of long-range excitation, forming the cell assemblies, to gain insight into the effect of long-range excitatory connections on the synchronization of distant minicolumns. As a result, a considerable

difference between the lowest and the highest level of long-range excitation tested in our simulations was observed (Fig. 5D, shown only for memory pattern completion) with higher coherence for lower excitation. This was likely due to the shorter attractor dwell times for low excitation configurations. When CaNMDACaNMDA influx rate was reduced in such a manner that attractors were stationary (see Experimental procedures), coherence dropped overall by ~0.1 (dotted line in Fig. 5D). In order to examine whether coherence accounted only for amplitude covariance, we estimated local and global phase locking in the gamma band. As expected, we found locally strong phase locking close to 1 (Fig. 5E), with essentially all firing events occurring at a specific phase of the gamma cycle (Fig. 8A), as observed experimentally (Jacobs et al., 2007 and Sirota et al., 2008).