Spectrotemporal receptive fields are useful for determining properties such as an auditory neuron's characteristic frequency, response latency, and spectral and temporal modulation preferences. Above all, spectrotemporal receptive fields represent the specific stimulus features that elicit spiking activity. Unfortunately, most classic receptive field techniques, such as the spike-triggered average (STA), are only capable of capturing single stimulus features, and recent studies have shown that auditory cortical neurons are simultaneously influenced by multiple stimulus features.
In this study, maximally informative dimensions (MID) analysis, a multi-feature receptive field technique, was applied to neurons in the auditory thalamus and cortex. Specifically, thalamic neurons from the ventral division of the medial geniculate body (MGBv) and cortical neurons from the anterior auditory field (AAF) and primary auditory cortex (A1) were examined. Our results showed that whereas most A1 neurons displayed significant dual-feature processing characteristics, two-feature MID models failed to yield improved response predictions over one-feature MID models for most neurons in MGBv and AAF.
In neurons that did show evidence of dual-feature processing, the spectrotemporal properties of each of the MID features were examined. Although the first MID feature component (MID1) was highly similar to the STA, the second MID feature component (MID2) captured stimulus processing not represented in single-feature models. MID2 features generally occurred at longer latencies and integrated stimulus information over a broader spectral range.
Interestingly, however, the two-feature MID model was rarely able to capture the entirety of a neuron's predictable response. We hypothesized that this was because auditory neurons use temporal coding that cannot be captured by receptive field techniques that assume independent spiking. To test this hypothesis, we examined the encoding properties of the simplest temporal pattern: the interspike interval (ISI). The findings demonstrated that pairs of spikes with short ISIs constituted a source of highly reliable, easily decoded stimulus information that contributed disproportionately to the stimulus encoding of A1 neurons.
Taken together, these results provide evidence of increasing multi-dimensional processing as information ascends the auditory pathway and illuminate the limitations of independent-spiking analysis techniques in sensory areas that display significant temporal coding.