Nonlinear Methods for Detecting Timing Relations in Neural Data
One hypothesis concerning the coding principles employed by the brain states that information is represented and transmitted in the simultaneous activation of groups of neurons, commonly termed assemblies. Such assembly activity would be expressed through coherent spatio-temporal patterns of neuronal firing, rather than by the simple rate response of populations of neurons. However, it is difficult to analyze neuronal recordings with respect to the occurrence of such temporal patterns. In current typical experimental setups the number of neurons recorded from simultaneously rarely exceeds ten. Therefore, we are often left with a very limited sample of the neuronal population activity. Even with recent advances in recording techniques that allow simultaneous recordings from up to about 100 electrodes, it is still nearly impossible to analyze the resulting data given current methods due to their complexity and the resulting combinatorial explosion.
Within this context, one approach to circumvent these difficulties is to examine the local field potential (LFP) in parallel. The LFP is a spatially slow-changing extracellularly recorded signal composed of low frequencies up to about 300 Hz. It is thought to reflect primarily synaptic currents in a large area around the recording site. Increased oscillation strength of this signal, as observed in numerous experimental paradigms, is hypothesized to be related to synchronized network activity rather than an overall increase in activity (see, e.g., Elul 1967). However, hitherto the intricate details of the relationship between neuron activity, network dynamics and LFP signal in cortex are little known. A deeper understanding of how activity at the neuron level is connected to LFP oscillations might aid in combining simultaneous LFP and single neuron recordings to characterize the network dynamics.
In this project, we link the activity of single neurons to the global LFP network activity. In particular, we develop novel analysis methods based on phase synchronization analysis of continuous time series (cf., e.g., Rosenblum et al. 1996; Varela et al. 2000). This includes the use of surrogate methods to evaluate and quantify the significance of observed periods of phase locking between spike and LFP time series independent of the underlying spiking statistics. The difference of the proposed methods to existing methods employed in current experimental work, e.g. spike-triggered averaging or spike-field coherence (Fries et al., 1997; Jarvis 2001), is the explicit separation between phase and amplitude of the LFP signal on a spike-by-spike basis. Advantages of the new method over standard methods are systematically analyzed, both theoretical and in terms of calibrations using controlled simulated data.
Cortical neurons, which typically tend to exhibit a sequentially uncorrelated (Poisson) spiking statistics, are unlikely per se to show a fixed phase relationship to certain frequencies of the LFP. Here we ask, whether neurons display an increased degree of phase synchrony with the LFP network activity at certain time points or under certain experimental conditions. Such increased phase locking would indicate that the recorded neuron becomes part of a coordinated network dynamics. Related studies have shown evidence that coincident spiking behavior of two simultaneously recorded neurons is correlated to behavioral events (e.g., Riehle et al 1997). Therefore, it is an interesting question how the behavior-modulated occurrence of such coincident events relates to amplitudes of LFP oscillations, and to the LFP phase. In addition, we extend this analysis to include coincidences which occur above chance level (Unitary Events, Grün et al 1999), and are thus believed to be of computational relevance in information processing. A natural extension of these questions is the relation of higher order correlations within spike data to LFP signals.
Our work is applied within the scope of several collaborations aimed to test these ideas in a several animals and systems. For the development of the method, both in a time-resolved and an averaged manner, we concentrate on recordings from the antennal lobe of the olfactory system of the honeybee (in collaboration with R. Finke and R. Menzel, FU Berlin and BCCN Berlin, Germany), and recordings form the striatum in the basal ganglia of healthy anesthetized rat (in collaboration with A. Sharott and A. Engel, UKE Hamburg, Germany). For the analysis of the relationship between local spike synchrony and LFP, we use data from motor cortex of monkey in an instructed time estimation and time discrimination task, which is known to exhibit temporally localized volleys of synchronized activity related to behavior (in collaboration with S. Roux, Freiburg, and A. Riehle, CNRS Marseille, France).
M. Denker, R. Finke, F. Schaupp, S. Grün, R. Menzel, Neural Correlates of Odor Coding and Learning in the Honeybee Antennal Lobe, J Neurophyiol. (submitted)
M. Denker, S. Roux, H. Lind´en, M. Diesmann, A. Riehle, S. Grün, Reflections of synchronous spiking activity in local field potentials, Online Abstract Viewer/Itinerary Planner. Washington, DC. Society for Neuroscience, 2007, Program No. 793.4.
A. Sharrot, C. Moll, G. Engler, M. Denker, S. Grün, A. Engel, Phase Locking of Striatal Neurons to Population Oscillations in the Anesthetised Rat, Online Abstract Viewer/Itinerary Planner. Washington, DC. Society for Neuroscience, 2005, Program No. 180.11.
M. Denker, M. Timme, S. Roux, A. Riehle, S. Grün, Detecting Transient Temporal Relationships between Spikes and LFP by Phase Analysis, Online Abstract Viewer/Itinerary Planner. Washington, DC. Society for Neuroscience, 2005, Program No. 970.10.
Amplitudes increase for a classic spike-triggered averages composed of all spikes (gray), only spikes coincident with another neuron (blue), and spikes significantly coincident with another neuron (red), respectively. This finding suggests that the occurrence of synchrony between individual neurons is correlated with the overall level of synchrony in the local network (larger LFP amplitudes), and/or that it implies synchrony with a larger pool of neurons (increased spike-LFP locking precision). To disentangle these hypothesis, we employ phase synchronisation methods to seperate amplitude from phase.