This process gave us τ summarizing the similarity in activity between all pairs of trial blocks. Resampling methods were used to confirm whether a τ for two blocks was unusually low. That is, we obtained an empirical distribution of τ obtained under the null hypothesis that the pattern of activity drug discovery for two blocks was the same. The τ computed between two blocks was considered to have changed if it was lower than 99% of the τ values obtained for the empirical distribution under the null hypothesis. Using this approach, we could determine whether

a neuron changed its firing pattern from block to block. To test whether a neuron rescaled its delay activity when the delay was doubled, the same approach was taken, but the PSTH for the longer delay used time bins whose duration was also doubled. Further details on this analysis are provided in the Supplemental Experimental Procedures. To assess the effect of time on firing rate related to the rat’s position, we generated spatial firing rate maps for the delay zone as 1 × 1 cm bins, and calculated occupancy-normalized firing rates. To assess firing rates

related to head direction, we assigned each head direction observation to 1 of Nutlin-3a chemical structure 60, nonoverlapping 6° bins and calculated occupancy-normalized firing rates for each bin. Speed firing rate plots were based on computations of the difference in the X-Y position for successive frames, assigned to 1 of 30 speed bins that spanned 0–30 cm/s, and occupancy-normalized firing rates were calculated for each bin. ANOVAs were performed on trial-by-trial, unfiltered firing rates for each 1 s segment of the delay. We used only those bins whose firing rate could be estimated in all of the 1 s segments across trials, allowing

an ANOVA with factors time and bin to test whether time modulated neural activity. Further details on this method are provided in the Supplemental Experimental Procedures. Analysis of LFP frequency as a function of time used the multi-taper Digestive enzyme functions written for MATLAB that are freely available as part of the Chronux toolbox (Mitra and Bokil, 2008; http://www.chronux.org). For the delay the trial-averaged multi-tapered spectrum was determined (mtspectrumc.m) using a window size of 1 s that started at the beginning of the delay and was slid across time using 100 ms increments. For the object and odor periods, a window size of 1.2 s was time locked to the beginning of the either period and slid with one 100 ms increment. The trial-averaged spectrum was computed separately depending on the object that was presented. For a given tetrode, in order to test whether θ (i.e., 4–12 Hz) power differed depending on the object presented during each trial period, a trial-average spectrogram was generated using the same parameters as above except that the frequency range was confined to 4–12 Hz. Further details of the ANOVAs performed on the LFPs are provided in the Supplemental Experimental Procedures.