Statistics of Neuronal Identification with Open- and Closed-Loop Measures of Intrinsic Excitability

Abstract

In complex nervous systems patterns of neuronal activity and measures of intrinsic neuronal excitability are often used as criteria for identifying and/or classifying neurons. We asked how well identification of neurons by conventional measures of intrinsic excitability compares with a measure of neuronal excitability derived from a neuron’s behavior in a dynamic clamp constructed two-cell network. We used four cell types from the crab stomatogastric ganglion: the pyloric dilator, lateral pyloric, gastric mill, and dorsal gastric neurons. Each neuron was evaluated for six conventional measures of intrinsic excitability (intrinsic properties, IPs). Additionally, each neuron was coupled by reciprocal inhibitory synapses made with the dynamic clamp to a Morris–Lecar model neuron and the resulting network was assayed for four measures of network activity (network activity properties, NAPs). We searched for linear combinations of IPs that correlated with each NAP, and combinations of NAPs that correlated with each IP. In the process we developed a method to correct for multiple correlations while searching for correlating features. When properly controlled for multiple correlations, four of the IPs were correlated with NAPs, and all four NAPs were correlated with IPs. Neurons were classified into cell types by training a linear classifier on sets of properties, or using k-medoids clustering. The IPs were modestly successful in classifying the neurons, and the NAPs were more successful. Combining the two measures did better than either measure alone, but not well enough to classify neurons with perfect accuracy, thus reiterating that electrophysiological measures of single-cell properties alone are not sufficient for reliable cell identification.