Because spike-field coherence depends weakly on mean firing rate even for Poisson spike trains (Curtis et al., 2009), as well as on LFP amplitude, we determined coherence significance levels using a bootstrap. For each spike train, the order of interspike intervals (ISIs) was randomized 500 times to generate a set of shuffled spike trains which preserved both the cell's mean firing rate and its ISI distribution. Spike-field coherence was computed for all shuffled spike trains to yield a distribution of coherence values expected if there were no relationship between spike times and the LFP. Significance of the observed coherence for the (unshuffled, observed) spike train was then computed from the z-score against this distribution. We used freely available, cross-platform distributed computing software (Condor2) with MATLAB to speed up this computationally intensive task.
