There is a growing interest in large-scale network simulations that capture the cellular heterogeneity and variation observed in real cortical tissue, as more and more experimental information becomes available on the role of cellular diversity in cortical dynamics and computation. Both developing (fitting) and simulating such a diversity of cell models which capture the spiking behavior of their real-cell counterparts in sufficient detail can be computationally quite demanding. It may therefore be desirable to have single cell models which, on the one hand, are simple enough to allow for fast fitting to experimental data and relatively short simulation times, but which on the other hand are still physiologically highly valid in the sense that they can reproduce and predict a number of spiking characteristics of the real cells. In preparation for a physiologically detailed prefrontal cortex (PFC) network model, we characterized in vitro in adult rodent PFC >200 pyramidal and inter-neurons from various layers with standard electrophysiological protocols. For the purpose of translating these data efficiently into single cell models, we derived closed-form expressions for onset and steady-state f-I curves (firing rate over step current) for a simplified version of the adaptive exponential (AdEx) and for the adaptive leaky (aLIF) integrate-&-fire model. These expressions allowed for a fast and completely automatized procedure which tightly fits these cell models to their experimental counterparts. Although the model neurons have been fitted solely on standard experimental f-I and sub-rheobase V-I curves, it is shown that they still can predict the spiking behavior (timing and rate) of their experimental counterparts to in-vivo-like fluctuating current inputs very well within the bounds of physiological reliability.
This approach was then used to translate our pool of adult rodent PFC slice recordings into a variety of physiological cell types and distributions of cellular parameters for the construction of a large-scale realistic PFC network model from which first results will be shown. This approach may therefore offer a promising tool for developing physiologically valid yet still simple and relatively fast to simulate neuron and network models.