While this is probably an oversimplification given the complex dynamics of the brain, there is no reason to think that there would be complex differences in dependencies across SWRs before correct and incorrect trials that would result in illusory significance values for our analyses. We also chose to use the number of cell pairs, individual cells, or trials as the N in our statistical analyses, as is standard in the field. We note here that our results are highly significant and consistent

across individual animals and across tracks. We also carried out a complementary analysis to determine whether we could predict the outcomes on individual trials. Our goal here was to use a measure that allowed us to combine multiple run sessions from multiple

GSK2656157 cost animals together, and as each run session was associated with a different number of recorded place cells, we measured the proportion of possible cell pairs that were active before each trial. We calculated, for each run session, the total number of possible coactive cell pairs, which is (number place cells recorded) × (number place cells recorded − 1)/2. We then determined, for each trial, the number of those cell pairs that were coactive within an SWR preceding that trial and then divided that number by the total to get a proportion. Given that measure for each correct selleck compound or incorrect trial, we then used logistic regression to relate the proportion of coactive cells to the trial outcome (correct or incorrect). The model was estimated

based on half of the total data, subsampled to include an equal number of correct and incorrect trials from each run session. The specific correct and incorrect trials were chosen at random. We then tested the model prediction on the other half of the data, once again subsampled to include an equal number of correct and incorrect trials from each run session. We repeated that estimation and testing process 1,000 times with different sets of correct and incorrect trials to produce a distribution of predictions and PDK4 compared that distribution across performance categories and to chance performance of 50% correct. We also examined the content of individual SWRs. We used our previously developed decoding approach (Karlsson and Frank, 2009) to translate the activity of neurons active during the SWR to a trajectory through space. Briefly, for all SWRs with at least two active place cells, we divided the SWR into 15 ms bins and for each bin used the place fields of neurons active in that bin to derive a probability distribution function over distance from the end of the center arm. For each bin, that pdf represented where we would expect the animal to be on the track given that those cells had fired the observed numbers of spikes. To determine whether a given decoded trajectory was best described as inbound or outbound, we fit a line to samples from the sequence of pdfs plotted versus time.