Synaptic input statistics tune the variability and reproducibility of neuronal responses

Publisher Website
Google Scholar
Abstract
Synaptic waveforms, constructed from excitatory and inhibitory presynaptic Poisson trains, are presented to living and computational neurons. We review how the average output of a neuron e.g., the firing rate is set by the difference between excitatory and inhibitory event rates while neuronal variability is set by their sum. We distinguish neuronal variability from reproducibility. Variability quantifies how much an output measure is expected to vary; for example, the interspike interval coefficient of variation quantifies the typical range of interspike intervals. Reproducibility quantifies the similarity of neuronal outputs in response to repeated presentations of identical stimuli. Al- though variability and reproducibility are conceptually distinct, we show that, for ideal current source synapses, reproducibility is defined entirely by variability. For physiologically realistic conductance-based synapses, however, reproducibility is distinct from variability and average out- put, set by the Poisson rate and the degree of synchrony within the synaptic waveform. © 2006 American Institute of Physics. DOI: 10.1063/1.2209427 Cortical neurons receive thousands of excitatory and in- hibitory synaptic inputs per second, each constituting a small conductance change coupled with a synaptic rever- sal potential. According to some encoding scheme that is still a subject of study, neurons convert their synaptic inputs into trains of action potentials. In this work, we review the relationships between standard synaptic input parameters and the neuronal outputs they produce. We focus on three measures of output: average output re- sponses (membrane potential or spike rate), the temporal variability of responses, and the reproducibility of re- sponses to repeated presentations of "frozen noise" in- puts. We contrast these measures in response to experi- mentally convenient current-based inputs and more realistic conductance-based inputs. With current-based inputs, we find that the mean firing rate and variability can be tuned independently with simple input structures, but that reliability is not independently tunable. In con- trast, conductance-based inputs enable a richer output space in which spike rate, variability, and reliability can be independently tuned using simple synaptic inputs.