Datensatz

Zusammenfassung

Spiking activity provides information about the outputs of neurons. Recently, there has been increased interest in the study of local field potentials (LFPs), partly due to their correlation with fMRI BOLD measurements [1], to the possibility of studying local inputs [2] and as a tool to assess neuronal synchrony [3]. The LFP is operationally defined by low-pass filtering (100 Hz) the extracellular recordings, and its precise biophysical origin of remains only poorly understood. Recently, Rasch and colleagues used a SVM algorithm to infer the spiking activity at a given site from the LFPs [4]. To further understand the relationship between spikes and LFPs, we asked whether we could predict the detailed timecourse of the LFP based solely on the spiking activity of units recorded from the same electrode or nearby electrodes. We used a Wiener-Kolmogorov approach to derive the optimum linear filter that estimates the LFPs [5, 6]. We considered electrophysiological recordings in the macaque lateral geniculate nucleus and primary visual cortex during spontaneous activity (86 electrodes, 7 monkeys) [4]. We found that it is possible to predict LFPs from V1 solely using spike trains from single electrodes in that area. The mean correlation coefficient (r) between the predictions and the actual LFP varied between 0.23 and 0.65. We found that the estimations were highly significant (p < 10, based on generating a Poisson spike train with the same rate and re-estimating the filters). In contrast, trying to predict LGN LFPs resulted in a performance hardly above chance level. The reconstruction filter was closely related to the spike-triggered average of the LFPs. A causal filter that used only the spikes occurring before the actual time of the LFP yielded a higher error (p < 0.01) than a filter that used only the spikes occurring after. It was possible to predict LFPs in V1 from spike trains recorded in LGN (r = 0.3 to 0.7). The algorithm performed at chance level when trying to predict LFPs in LGN from spikes in V1. In sum, these results support the notion that LFPs represent the input and local processing while spikes represent the output and suggest that a linear convolution can account for a large fraction of the timecourse of the LFP. We have observed similar results in recordings from macaque monkey inferior temporal cortex and the human temporal lobe, suggesting that there may be a universal relationship between spikes and LFPs.