Detection and Removal of Hyper-synchronous Artifacts in Massively Parallel Spike Recordings

Contemporary electrophysiology experiments often involve massively parallel recordings of neuronal activity using multi-electrode arrays. While researchers have been aware of artifacts arising fromelectric cross-talk between channels in setups for such recordings, systematic and quantitative assessment of the effects of those artifacts on the data quality has never been reported. Here we present, based on examination of electrophysiology recordings from multiple laboratories, that multi-electrode recordings of spiking activity commonly contain extremely precise (at the data sampling resolution) spike coincidences far above the chance level. The recordings analyzed here were obtained from two rhesus macaques (Macaca mulatta; one male, one female) and one male mouse (C57BL/6J).We derive, throughmodeling of the electric cross-talk, a systematic relation between the amount of such hyper-synchronous events (HSEs) in channel pairs and the correlation between the raw signals of those channels in the multi-unit activity frequency range (500-7500 Hz). We show that whitening the band-pass filtered raw signals removes the above chance HSEs; strongly suggesting they originate from linear mixing of signals. Whitening should therefore be performed prior to spike sorting and any further analysis of precise spike correlation, otherwise analysis results may be considerably affected.Significance statement Artifacts are widespread in electrophysiological recordings. To mitigate their impact, these artifacts need to be detected and they should be removed from the data without impacting the quality of the data. This work presents measures to identify and quantify the amount of artifacts within a multichannel recording by evaluating the occurrence of hyper-synchronous events i.e., spikes that are synchronous on a sub-millisecond time scale, and further introduces zero-phase component analysis (ZCA) as a method to remove these artifacts from the data. Thus, we recommend to use ZCA as a general preprocessing for electrophysiological recordings.