Exploring parameter space in detailed single neuron models: simulations of the mitral and granule cells of the olfactory bulb

Abstract

1. Detailed compartmental computer simulations of single mitral and granule cells of the vertebrate olfactory bulb were constructed using previously published geometric data. Electrophysiological properties were determined by comparing model output to previously published experimental data, mainly current-clamp recordings. 2. The passive electrical properties of each model were explored by comparing model output with intracellular potential data from hyperpolarizing current injection experiments. The results suggest that membrane resistivity in both cells is nonuniform, with somatas having a substantially lower resistivity than the dendrites. 3. The active properties of these cells were explored by incorporating active ion channels into modeled compartments. On the basis of evidence from the literature, the mitral cell model included six channel types: fast sodium, fast delayed rectifier (Kfast), slow delayed rectifier (K), transient outward potassium current (KA), voltage- and calcium-dependent potassium current (KCa), and L-type calcium current. The granule cell model included four channel types: rat brain sodium, K, KA, and the non-inactivating muscarinic potassium current (KM). Modeled channels were based on the Hodgkin-Huxley formalism. 4. Representative kinetics for each of the channel classes above were obtained from the literature. The experimentally unknown spatial distributions of each included channel were obtained by systematic parameter searches. These were conducted in two ways: large-scale simulation series, in which each parameter was varied in turn, and an adaptation of a multidimensional conjugate gradient method. In each case, the simulated results were compared wtih experimental data using a curve-matching function evaluating mean squared differences of several aspects of the simulated and experimental voltage waveforms. 5. Systematic parameter variations revealed a single distinct region of parameter space in which the mitral cell model best fit the data. This region of parameter space was also very robust to parameter variations. Specifically, optimum performance was obtained when calcium and slow K channels were concentrated in the glomeruli, with a lower density in the soma and proximal secondary dendrites. The distribution of sodium and fast potassium channels, on the other hand, was highest at the soma and axon, with a much lighter distribution throughout the secondary dendrites. The KA and KCa channels were also concentrated near the soma. 6. The parameter search of the granule cell model was much less restrained by experimental data. Several parameter regimes were found that gave a good match to the data.(ABSTRACT TRUNCATED AT 400 WORDS)

Variability, compensation, and modulation in neurons and circuits

“…1A). This illustrates an important general principle, namely, that similar electrophysiological behavior can arise from widely disparate sets of underlying conductances (4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15). Fig.…”

Intermediate intrinsic diversity enhances neural population coding

“…The statistical modeling approach that we have used accurately captures the neuronal properties determining spiking and avoids overfitting. It also avoids making specific but difficult-to-verify claims about channel densities or properties that can arise from underconstrained Hodgkin-Huxley models (36). We show that diverse populations offer the advantages of more efficient encoding (defined in terms of information per cell or information per spike) and more robust coding of different kinds of stimuli, such as stimuli with wide ranges of spectral properties.…”

Using Evolutionary Algorithms to Search for Control Parameters in a Nonlinear Partial Differential Equation

“…A statistical proof called the Schema Theorem guarantees improvement of the search over time (Holland 1975). EAs have not previously been applied to the field of compartmental neuronal modeling, although reduced dimensional parameter searches using systematic (Bhalla and Bower 1993) or stochastic (Foster et al 1993) search methods have been reported. Would these or methods other than an EA perform better?…”