Layer specific sharpening of frequency tuning by selective attention in primary auditory cortex

Abstract

Recent electrophysiological and neuroimaging studies provide converging evidence that attending to sounds increases the response selectivity of neuronal ensembles even at the first cortical stage of auditory stimulus processing in primary auditory cortex (A1). This is achieved by enhancement of responses in the regions that process attended frequency content, and by suppression of responses in the surrounding regions. The goals of our study were to define the extent to which A1 neuronal ensembles are involved in this process, determine its effect on the frequency tuning of A1 neuronal ensembles, and examine the involvement of the different cortical layers. To accomplish these, we analyzed laminar profiles of synaptic activity and action potentials recorded in A1 of macaques performing a rhythmic intermodal selective attention task. We found that the frequency tuning of neuronal ensembles was sharpened due to both increased gain at the preferentially processed or best frequency and increased response suppression at all other frequencies when auditory stimuli were attended. Our results suggest that these effects are due to a frequency-specific counterphase entrainment of ongoing delta oscillations, which predictively orchestrates opposite sign excitability changes across all of A1. This results in a net suppressive effect due to the large proportion of neuronal ensembles that do not specifically process the attended frequency content. Furthermore, analysis of laminar activation profiles revealed that although attention-related suppressive effects predominate the responses of supragranular neuronal ensembles, response enhancement is dominant in the granular and infragranular layers, providing evidence for layer-specific cortical operations in attentive stimulus processing.

Keywords: auditory; current source density; entrainment; macaque; oscillations; selective attention.

Figure 1.

Representative laminar CSD and MUA profiles in response to AA and IA BF and off-BF streams in A1. A, Schematic of a linear array multielectrode positioned in primary auditory cortex (A1). Cortical layers are indicated by numbers. To the right are CSD and concomitant MUA response profiles related to the AA condition BF (8 kHz) and an off-BF (2 kHz) tone stream. B, The CSD and MUA profiles evoked by the same tone streams as in A, but while they were in the IA condition. Note that there is a response enhancement for the BF tone in the AA versus IA condition, whereas in the case of the off-BF tone there appears to be a response-suppression. C, The frequency tuning curves created using cross-laminar (averaged across all layers) MUA amplitudes in response to 14 different frequency tones presented in blocks during both attentional conditions (AA, red; IA, blue) for this particular A1 site. Arrows mark the frequencies of BF and off-BF tone streams.

Figure 2.

The effect of attention on the MUA response and frequency tuning. A, Pooled cross-laminar MUA responses to BF (top) and off-BF (middle) tone streams during AA (red traces) and IA (blue traces) conditions. The off-BF responses shown here were selected based on largest MUA response amplitude difference between attention conditions in each experiment. p Value graph (bottom) displays the result of AA versus IA statistical comparison for each time point (Wilcoxon signed rank test). Even though the sign of the attention effect on response amplitudes is different, the largest significant effect occurs around the same time from response onset (10 ms) to ∼40 ms poststimulus (marked with dotted green vertical lines) for both BF and off-BF responses. B, The frequency tuning curves display tuning curves pooled across all experiments in the AA and AI conditions. For each site (n = 39) frequency tuning curves were created from cross-laminar MUA responses to streams of different frequency tones in the 10–40 ms poststimulus timeframe. The tuning curves were then normalized to the value of the ignored BF MUA response measure, and shifted to align the BF of all sites (n = 39) in the same position (BF in the graph). Asterisks indicate significantly different MUA amplitudes between the two conditions (Wilcoxon signed rank test). C, Boxplots show the pooled amplification indices (AA–IA BF-related MUA response amplitudes, top) and suppression indices (ΣAA-ΣIA of all off-BF related MUA response amplitudes, bottom). The indices indicate opposite sign attentional modulation (enhancement vs suppression for BF vs off-BF response related measures), both of which are significant (see Results). D, Supragranular, granular, and infragranular MUA responses associated with BF and selected (same as in A) off-BF stimuli averaged across all experiments. The 10–40 ms time interval is marked with dotted green vertical lines. E, Same modulation indices as in C but separately for each layer.

Figure 3.

Delta entrainment to different frequency tone streams. A, Overlaid frequency tuning curves created from cross-laminar MUA responses to AA and IA stimulus streams from a representative experiment where the BF of the site was 4 kHz. Inset, Layer-specific MUA tuning curves in AA versus IA conditions from same experiment. Asterisks denote significant differences (Wilcoxon rank sum, p < 0.01) between attention conditions: red asterisks indicate that responses to a given frequency tone were significantly larger, whereas blue asterisks denote that responses were significantly smaller when the tone stream was attended. Note the gradual transition from more suppressive to more enhancing effects of attention from supragranular to infragranular layers. Bottom, Red histograms show the supragranular delta frequency (1.6 Hz, which matches the repetition rate of tones) CSD phase distribution across single trials at time of stimulus onset, related to a subset of AA tone streams (shown by arrows), black vertical lines denote the mean phase. Blue histograms show the delta-phase distribution across single trials related to the same tone streams but in the IA condition. B, Red trace shows supragranular delta ITC values for each AA tone stream in the same experiment. The purple dotted line denotes the average value above which ITC can be considered significantly nonrandom. All 14 frequency-tone streams resulted in significantly biased delta phase distribution at this specific A1 site (100% significant delta ITC). C, Graph shows the phase opposition index for the same experiment. Phase opposition was defined as a mean phase that differs at least a half-π radians from the BF-tone related delta phase (shown by green oval), and thus falls within the opposite half of the delta oscillatory cycle. Mean delta phases associated with all off-BF tone streams are shown by red ovals. Only one other tone stream resulted in a mean delta phase that fell within a half-π (shown by the dotted blue lines) of the BF stream-related phase. In this specific case the phase opposition index was 86%.

Figure 4.

Pooled delta phase measures. A, Pooled mean supragranular delta oscillatory phases related to all 14 AA tone streams in all of the 39 A1 sites (n = 14 × 39 = 546). Purple bars show mean delta phases related to AA BF tone streams, while the blue bars display the delta phase distribution related to all other AA tone streams. B, Boxplots show percentage of tone streams across all experiments, which resulted in significant supragranular delta ITC in the AA (red) and IA (blue) conditions. C, Boxplot shows percentage of AA tone streams in each experiment which entrained supragranular delta oscillations to a phase opposite to the BF stream related phase (calculated as in Fig. 3C). D, Boxplots show the pooled suppression indices (Fig. 2C) of A1 sites that showed greater, equal to or lesser phase opposition than the median phase opposition across all experiments (79%). The bracket indicates significant difference (Wilcoxon rank sum, p < 0.05).

Figure 5.

Layer-specific sharpening of frequency tuning and delta entrainment. A, Tuning curves in the AA (red) and IA (blue) conditions based on layer-specific MUA responses to different frequency tone streams averaged in the 10–40 ms poststimulus time interval. Note that as indicated by the blue asterisks, supragranular responses to AA tones that do not match the BF of this representative recording site (5.6 kHz) are generally significantly suppressed (Wilcoxon rank sum, p < 0.01) compared with responses to the same tones during the IA condition. In contrast, in the granular and infragranular layers of this site while there was no significant suppression of responses to AA off-BF tone streams, responses to the AA BF tone stream were significantly enhanced (indicated by red asterisks). B, Laminar delta amplitude profiles from three representative sites from AA off-BF tone trial blocks. Blue arrows indicate that the BF tone related initial CSD response in the laminar location of a given delta peak is a source (Fig. 1 shows a representative BF response profile), whereas red arrows denote laminar locations of BF related initial sinks. The peaks associated with BF related sinks represent active, whereas peaks associated with BF-related sources represent passive current in the supragranular and infragranular layers. C, Red and blue histograms show the distribution of single-trial delta phases related to 14 different frequency AA tone streams in a supragranular (B, red, marked by s2) and infragranular location (B, blue, marked by i1), from same experiment as A. Black lines denote mean phases. Note the opposition of supragranular delta phase distributions related to BF versus off-BF tone streams (similar to Fig. 3A) and the matching phase opposition in the infragranular layers. D, Red and blue squares show the mean phases of stimulus related delta oscillations in the supragranular and infragranular layers respectively (again from same experiments as in A), which appear largely the same within trial blocks. This is nicely illustrated by the distribution of the supragranular–infragranular delta phase differences related to different frequency tones (histogram on the right), which show a significantly nonrandom distribution (Rayleigh p < 0.0001), with a mean of −0.49 radians. E, The histogram displays the distribution of mean supra–infragranular phase differences across all experiments (n = 39). The black line indicates the mean (−0.37 radians), the Rayleigh p value signals a significantly nonrandom phase-difference distribution.

Figure 6.

Modulation of prestimulus excitability. A, The first column of color maps shows representative laminar MUA profiles related to the AA BF tone stream (top) and related to all AA tone streams (averaged), including the BF (bottom). Color maps in the middle and to the right show gamma band activity (25–55 Hz) amplitude profiles extracted from CSD and LFP, respectively. Traces on the bottom display the time course of MUA and gamma range activity averaged across all layers. Dotted vertical lines denote the immediate prestimulus (2) and interstimulus (1) timeframes used to calculate the modulation indices in B. B, Boxplots display the pooled difference of immediate prestimulus (−150—30 ms, marked by 2) and interstimulus (−300 to −150 ms, marked by 1), MUA and gamma band activity. Note that in the case of BF stream related prestimulus activity, MUA and gamma are up-modulated toward the timing of attended stimuli, in the net activity related to all streams these measures indexing the excitability of the local neuronal ensemble are down-modulated. Inset p values indicate the probability that the pooled measures are not significantly different from zero (Wilcoxon signed rank).