Tracing 'driver' versus 'modulator' information flow throughout large-scale, task-related neural circuitry

PRIMARY OBJECTIVE: To determine the relative uses of neural action potential ('spike') data versus local field potentials (LFPs) for modeling information flow through complex brain networks. HYPOTHESIS: The common use of LFP data, which are continuous and therefore more mathematically suited for spectral information-flow modeling techniques such as Granger causality analysis, can lead to spurious inferences about whether a given brain area 'drives' the spiking in a downstream area. EXPERIMENT: We recorded spikes and LFPs from the forelimb motor cortex (M1) and the magnocellular red nucleus (mRN), which receives axon collaterals from M1 projection cells onto its distal dendrites, but not onto its perisomatic regions, as rats performed a skilled reaching task. RESULTS AND IMPLICATIONS: As predicted, Granger causality analysis on the LFPs-which are mainly composed of vector-summed dendritic currents-produced results that if conventionally interpreted would suggest that the M1 cells drove spike firing in the mRN, whereas analyses of spiking in the two recorded regions revealed no significant correlations. These results suggest that mathematical models of information flow should treat the sampled dendritic activity as more likely to reflect intrinsic dendritic and input-related processing in neural networks, whereas spikes are more likely to provide information about the output of neural network processing.