hide
Free keywords:
-
Abstract:
Human electrical imaging studies have shown that slow oscillatory (theta-band) activity in auditory cortex entrains to the rhythmic structure of naturalistic sounds such as speech or animal vocalizations (Luo & Poeppel, Neuron 2007). Assuming that slow oscillations reflect changes in the excitation-inhibition balance of local networks this suggests that as a consequence of the entrainment and the locking of spikes to slow oscillations auditory cortical neurons should sample the environment in a rhythmic fashion (Giraud & Poeppel, Nat Neurosci 2012). We here confirm this hypothesis using auditory cortical responses recorded in the alert non-human primate.
We recorded single neuron and local field potential responses to long sequences of naturalistic sounds in macaque auditory cortex (Kayser et al Neuron 2009). We then used methods of single-trial decoding and spectro-temporal receptive field (STRF) mapping to study how neural coding varies as a function of oscillatory phase (derived from theta-band 4-8Hz field potentials). Neurons fired more during one half of the oscillatory cycle than during the other, resulting in systematic variations of firing rate and stimulus information with oscillation cycle (both by about 20%). However, the coding capacity, defined as information per spike was roughly constant across the cycle. STRFs mapped separately for spikes in different phase quadrants differed both quantitatively and qualitatively. During the phase range of higher firing STRFs yielded better prediction quality (best vs. worst quadrant: 0.35 vs. 0.2 median cross-validation correlation) and exhibited stronger excitation-inhibition ratios (medians 1.3 vs. 1) and signal-to-noise ratios (medians 20 vs. 12) than during the remaining cycle (even when corrected for changes in overall spike count). This suggests that the degree to which linear models can account for neural response selectivity varies systematically during the oscillatory cycle. The time epoch of greatest selectivity (time lag of peak in STRF) also systematically varied along the cycle rather than being constant. Lags of STRFs derived using only spikes occurring during later phase quadrants became progressively longer, resulting in a clustering of the peak sensitivity of all STRFs in time. Spikes occurring e.g. during the fourth phase quadrant were evoked not much later as spikes occurring during the phase quadrant, resulting a clustered and periodic input sampling of auditory cortical neurons.
Our findings demonstrate that the quality of linear STRFs of auditory cortical neurons depends on phase of slow and stimulus driven oscillatory network activity, with linear models performing much better during certain phases of the slow rhythm. These observations may have important implications for the interpretation of STRFs. In addition, our results directly confirm the hypothesis that auditory neurons do not unfold incoming stimuli linearly in time; rather they periodically sample the environment based on the auditory cortical theta rhythm. These findings have important implications for the understanding of auditory coding and perception that both seem to operate in a periodic fashion.
