Neuro-heuristique

GLOBAL SPATIOTEMPORAL ACTIVITY IN A SIMULATED "CORTICAL" MODEL DETERMINED BY LOCAL KINETICS


The neural networks of vertebrate cerebral cortex form complex spatiotemporal patterns as part of information processing activity. Repeating temporal patterns have been observed in many areas of the central nervous system. Oscillatory and wave-like activity occur consistently in hippocampus, cortex, and thalamo-cortical circuitry. Spatiotemporal patterns may be the main representation of information within the brain.

We have developed an in-house platform for developing and experimenting with the generation properties of such patterns of activity in networks of neuromimes. It has an architecture which provides for run-time loading of new models for the kinetics, connectivity and interaction dynamics. We have designed Nous to provide a fully extensible but efficient and fast simulation environment. The simulator is also capable of being fully distributed with each object running on a separate machine.

With this system it is posible to simulate large-scale models of cortex (100x100 unit sheet or more) of interconnected cells, roughly representing a patch of 1 square mm. The population is composed of two subpopulations of 80% excitatory and 20% inhibitory cells interspersed according to a space-filling pseudo-random distribution. Interconnections are determined randomly according to a Gaussian probability distribution. The strength of the connections from the inhibitory population is set to twice the strength of the excitatory connections. The levels of grey indicate the membrane potential for each neuromime. The yellow shading indicates neuromimes that are above threshold. With the topology held constant, we ran experiments with two different models for local kinetics, FitzHugh-Nagumo and Leaky Integrate-and-Fire (for the former see the quiktime movie).



We compare the models based on local phase plots, global activity levels and the power spectrums of population activities. In addition, we gather spike-train data and compare the results with spike-train data gathered from multiple microelectrode recordings from real brains, performed in our laboratory. While it is clear that the models can not be compared at the local kinetics level, comparisons can be made based on global activity and spike-train data.

Parameter changes at the single neuron level can influence global spatiotemporal pattern generation in large-scale neural networks. To study this influence on functional organization, we develop a very simple integrate-and-fire model for an individual neuron which allows the threshold and post-synaptic potential (PSP) rate of decay to be controlled. By combining this neuron model with a generalized model of layer IV cortical topology we are able to view the influence of varying the threshold and PSP decay on the patterns of activity generated in this network.

Spatiotemporal patterns can constitute an important means of information processing and transport in the central nervous system (Abeles, 1991). These patterns may be observed by electrophysiological recordings using multiple microelectrode devices (Villa and Abeles, 1990; Villa and Fuster, 1992). These experimental techniques are becoming accessible to a larger number of laboratories every year. Oscillatory and wave-like patterns of activity need to be better understood and the principles underlying their generation should be further explored.

Large-scale simulations and parameter space searches were performed using an extremely simplified neuron model (Villa, 1989). This neuron is an integrate-and-fire with a refractory period and PSP decay. Click to see the following figures:

Map of responses in parameter space. Each block represents the activity of 500 ms of simulation for a particular parameter set. The firing threshold theta increases along the x-axis and the decay of the PSPs increases along the y-axis. Along the diagonal from the upper left to the lower right, separating the high activity region with stable blobs from the low activity region, one can see there is increased movement of the activity patches.

We are interested in the ability of neural media to act as a substrate and generator for spatiotemporal patterns. To investigate the influence of particular models of local kinetics on pattern generation, we have developed a simulator that allows us to systematically explore parameter ranges and various neuron models within a fixed topology. This allows us explicitly to identify the effects of local kinetics changes. This approach to searching parameter spaces can be used for other kinetics models, such as the FitzHugh-Nagumo (FitzHugh, 1961) or Hodgkin-Huxley (Hodgkin and Huxley, 1952) models.

Patterns or waves of activity form can be broken up by an input stimulus simulating spontaneous activity, e.g. to 4% of the population chosen at random for a duration of 5-10 ms. The network becomes very quiet. Immediately after the stimulus is removed, a new pattern begins to form. For this reason we would not expect to see these types of blobs and waves of activity in vivo where there is spontaneous background activity.

In another case, when a more focused stimulus is applied, the activity level immediately decreases in the vicinity of the stimulus. Only after the stimulus is removed does a spatiotemporal representation begin to form.

By using the same set of random connections for all neurons of the same type we can observe the collective effect of the uneven random distribution. For these cases activity consistently forms very similar spatial patterns and tends to move in the same direction. This could be further explored by generating random connections with specific uneven spatial distributions.

Actual work incorporates this model of cortical layer IV into a multiple layer thalamo-cortical circuitry to observe the types of patterns generated in a network with intralayer connectivity.