The influence of artificial synchronous spikes in multi-electrode recordings on correlation analysis

Many modern electrophysiological experiments focus on collecting an increasing amount of data by recording at high time resolution with large numbers of electrodes simultaneously. Therefore, commercial full package recording and data acquisition systems are often used, that provide the experimentalist with a compact data set, thus hiding the complexity of the system and any intermediate preprocessing steps. This amounts to increasingly convoluted data that finally enter analysis to elucidate mechanisms of neuronal processes. We investigated precise spike correlations in massively parallel cortical single-unit spike data recorded by a multi-electrode array implanted in motor cortex of awake behaving monkeys. We detected spurious synchrony of high order on a fine time resolution (30kHz), occurring with varying numbers of contributing spikes, including events of few synchronous spikes only. The occurrence probabilities of such 'synchrofacts' are typically higher than expected based on the firing rates, whereas the surplus of spikes is only visible in the population histogram using bin sizes in the range of the sampling rate. A biological origin of synchrofacts is unlikely, as cortical neurons seem not to be able to synchronize on such a shorttime scale that is about 2 orders of magnitude smaller than the spike duration. Despite their high precision, a substantial fraction of synchrofacts is yet imprecise enough to evade classification as artifacts by various algorithms applied during preprocessing steps. We investigated the raw signal (at 30kHz sampling, wide band) at the time of occurrence of each individual synchrofact and found that some synchrofacts are reflected by small, spike-like, coherent deflections. However, other synchrofacts could not be related to such obvious coherent deviations in many channels, but were identified to relate to coherent signalfluctuations just above high pass cutoff frequency for spike extraction. We demonstrate how the observed artifacts, though small in absolute number, may have a large impact on a population correlation analysis and cross-electrode spike LFP correlations (spike-triggered averages). Therefore, we investigate possible origins of the artifacts in our recording set-up and suggest approaches for artifact removal