Signal Processing & Signal Interpretation

Responses of individual neurons to identical repeats of a sensory stimulus are highly variable. However, the brain can processes information and make decisions based on single events, and can thus make sense of the noisy messages of individual neurons by evaluating the activity of populations and by merging information carried by different aspects of neural activity and by different networks. How the brain achieves such stable representation of sensory events even with noisy computing elements is a central, yet unaddressed, question in neuroscience.

Neuroscientists can now record, simultaneously and from the same or different cortical regions, several types of neural signals, each reflecting different and complementary aspects of neural activity at different scales of its organization. Spiking activity captures the output of individual neurons. Local field potentials (LFPs) capture massed synaptic activity and other slow aspects of the activity of large local populations. Recording all these electrophysiological signals as well as the fMRI BOLD signal gives us an unprecedented opportunity to understand how the brain integrates all the information carried, at different spatial and temporal scales, by single neurons and by widespread networks. Yet progress in understanding how neural populations process information has been limited by a lack of analysis methods capable of comparing and merging the different types of information carried by different neural signals, and by the lack of realistic computational models of how integration of information can be achieved. Our aim is to develop the mathematical analysis and mathematical modeling tools to address these questions.

Using information theory, we develop new analytical methods that quantify the amount of information available in a single trial and for each type of neural signal. At the same time, we determine whether different signals carry information about the same or complementary sensory features [1, 2] and how different signals influence each other. The unique advantage of these methods is their objectiveness, because they do not rely on any assumptions about what is “signal” and what is “noise” in neural responses or on what features of the stimulus trigger neural responses.

Applying these tools to LFPs and spikes from monkey primary visual cortex during stimulation with naturalistic movies, we found that the visual information carried by low frequency (<12 Hz) LFPs was completely independent of that carried by spikes or by gamma-range (70 – 110 Hz) LFPs [1]. The information content of spikes was dramatically increased by measuring spike times with respect to the phase of slow LFP fluctuations (Figure). This suggests that cortex multiplexes information: the slow ongoing rhythms provide a “clock” against which precise spike times carry additional independent information.

Interpretation of empirical neural data ultimately requires the development of credible models that explain their origin and their functional meaning. We develop a biophysically credible model of local, interconnected neural populations of inhibitory and excitatory neurons in sensory cortex, and study the dynamics of the model LFP and spikes in response to dynamic stimuli [3]. This model explains most published results on LFP sensory responses, and suggests that gamma-range oscillations are generated by inhibitory-excitatory neural interactions, whereas slow, sensory-evoked LFPs encode slow changes in the sensory stimulus. We are extending these models to include interactions between areas and the effect of neuromodulation. These models provide a strong tool for inferring the underlying network origin of LFP and spike signals measured empirically. Another approach to understanding the origin of LFPs is to control the sensory stimuli to manipulate the size of the activated neural populations [4]. This approach has shown that gamma LFPs originate from populations within 800 μm from the electrode tip.

Most of our projects require a multiple-channel electrode or tetrode recording system with the smallest permissible cross talk between neighboring channels. In addition, local laminar information is necessary in electrode configurations that may be used within the MRI magnet to localize the sources of neural activity reflected in the fMRI BOLD signal. Therefore, considerable effort is being devoted to developing these recording systems.


  1. Belitski, A., A. Gretton, C. Magri C, Y. Murayama, M. A. Montemurro, N. K. Logothetis, S. Panzeri: Low-frequency Local Field Potentials and Spikes in Primary Visual Cortex Convey Independent Visual Information. Journal of Neuroscience 28, 5696–5709 (2008).
  2. Montemurro, M. A., M. J. Rasch, Y. Murayama, N. K. Logothetis, S. Panzeri: Phase-of-firing Visual Stimuli in Coding of Natural Primary Visual Cortex. Current Biology 18, 375–380 (2008). '
  3. Mazzoni, A., S. Panzeri, N. K. Logothetis, N. Brunel: Encoding of Naturalistic Stimuli by Local Field Potential Spectra in Networks of Excitatory and Inhibitory Neurons. PLOS Comput. Biol., 21000239 (2008).
  4. Berens, P., G. A. Keliris, A. S. Ecker, N. K. Logothetis, A. S. Tolias: Comparing the Feature Selectivity of the Gammaband of the Local Field Potential and the Underlying Spiking Activity in Primate Visual Cortex. Frontiers in System Neuroscience 2, 1–11 (2008).
Go to Editor View