アイテム詳細

要旨

Modern multi-channel recording devices offer the opportunity to untangle the coordination mechanisms of neural activity at a mesoscopic level. However, this requires novel multivariate analysis techniques that link the high-dimensional data with biophysically realistic models of network activity. We develop a methodology to generalize the analysis of spike-field coupling (SFC) to the multi-channel setting, while allowing interpretability of such coupling based on neural field models of network activity. SFC characterizes the relation between the spiking activity of individual neurons and field activity, reflecting mesoscopic network dynamics, and typically measured by Local Field Potentials (LFP). Although SFC approaches have been extensively applied, they were focused on quantifying coupling one pair at a time (between one unit and one LFP channel), thereby remaining largely blind to the population-level interactions that generate them. As a consequence, the rich spatiotemporal patterns of spikes and LFPs remained largely under-exploited, despite the insights into the underlying circuit dynamics they may provide.

We introduce "Generalized Phase Locking Analysis (GPLA)" that summarizes the spike-field coupling from multi-channel data by three interpretable quantities: LFP and spike vectors represent the dominant spatial distribution and relative phase shifts of neuronal ensemble and field activity in the recorded structure, while the generalized Phase Locking Value quantifies the strength of the coupling between these patterns. We demonstrate the capability and interpretability of GPLA with various biophysically realistic simulations. For instance, we demonstrate that GPLA phase shift properties reflect the local recurrent connectivity of the underlying microcircuits. Finally, combining neural field simulations and GPLA applied on neural data recorded from monkey PFC supports long-range excitatory horizontal connections and strong local recurrent inhibition as a plausible connectivity scheme of the recorded structure.

Furthermore, we address one of the challenges of statistical analysis of parallel spike trains [1] by taking a step beyond bootstrapping methods. We exploit novel statistical frameworks [2], namely the random matrix theory, and provide an analytical significance test. This is of paramount importance given the increasing dimensionality of modern recording techniques, such as Neuropixel [3], that need computationally efficient statistical test
.