Plasticity of temporal pattern codes for vocalization stimuli in primary auditory cortex

Abstract

It has been suggested that "call-selective" neurons may play an important role in the encoding of vocalizations in primary auditory cortex (A1). For example, marmoset A1 neurons often respond more vigorously to natural than to time-reversed twitter calls, although the spectral energy distribution in the natural and time-reversed signals is the same. Neurons recorded in cat A1, in contrast, showed no such selectivity for natural marmoset calls. To investigate whether call selectivity in A1 can arise purely as a result of auditory experience, we recorded responses to marmoset calls in A1 of naive ferrets, as well as in ferrets that had been trained to recognize these natural marmoset calls. We found that training did not induce call selectivity for the trained vocalizations in A1. However, although ferret A1 neurons were not call selective, they efficiently represented the vocalizations through temporal pattern codes, and trained animals recognized marmoset twitters with a high degree of accuracy. These temporal patterns needed to be analyzed at timescales of 10-50 ms to ensure efficient decoding. Training led to a substantial increase in the amount of information transmitted by these temporal discharge patterns, but the fundamental nature of the temporal pattern code remained unaltered. These results emphasize the importance of temporal discharge patterns and cast doubt on the functional significance of call-selective neurons in the processing of animal communication sounds at the level of A1.

Figure 1.

Spectrograms of the natural sound recordings used as stimuli for this study. The color scale saturates over 70 dB.

Figure 2.

Mean response rate evoked by natural (R_Nat) plotted against response to time-reversed (R_Rev) twitter call stimulus. Responses for first, second, and third exemplar of the twitter stimuli are shown in A, B, and C respectively.

Figure 3.

A, Raster plot display of responses from one unit to the second twitter call and its time-reversed counterpart. Each dot indicates the timing of one action potential, and each row of dots gives the response to a single stimulus presentation. The thick black line underneath the raster plots indicates the duration of the stimulus. B, Responses from a different unit to the same stimuli.

Figure 4.

A, B, Responses of two different units. Each individual response is plotted as a grayscale histogram, i.e., the number of spikes in each 20 ms bin is given in a grayscale from white (0 spikes/bin) to black (5 spikes/bin). Individual responses to each of the 20 repeats of each stimulus are shown. Responses are grouped by stimulus class, and the corresponding stimulus is indicated by the label to the right (1, 2, 3 indicates first, second, and third twitter call stimulus from Fig. 1, whereas 1 R, 2R, 3R indicates the corresponding time-reversed counterpart). C, D, “Assignment” or “confusion” matrices illustrating the performance of our pattern classifier algorithm in decoding the spike patterns shown in A and B, respectively. The grayscale indicates the proportion of the 20 responses to the stimulus class indicated on the ordinate that was attributed by the algorithm to the stimulus class indicated on the abscissa.

Figure 5.

A, MI between stimulus and response as estimated from the performance of the classifier algorithm at different temporal resolutions for the two units shown in Figure 3, A (solid line) and B (dotted line). MI values above the hatched horizontal line at y = 0.1 are deemed statistically significant at α = 0.05. B, Waterfall plot of MI as a function of temporal resolution for all 142 units from the untrained ferrets in this study. The units are ranked by maximum MI. C, Normalized MI as a function of temporal resolution for units with maximal MI values >0.5 bits/response are plotted in black. The light gray line shows the mean of these normalized MI functions. D, Waterfall plot of MI at a temporal resolution of 10 ms as a function of the length of the response period analyzed. The data are from the same units, ranked in the same order, as in B.

Figure 6.

Diversity of response patterns. Responses to twitter 1 and twitter 1 reversed for 47 different A1 units shown in raster plot format. Responses are shown for every third unit in our dataset, arranged in ascending order of maximum MI, starting with the responses of the least informative unit at the bottom. Alternating dark and light gray dots and dividing lines are used to visually offset responses from different units. Above the raster plots, the temporal waveforms of the corresponding acoustic stimuli are shown. The arrows on the left mark the responses of the units ranked 142 of 142 (top) and 109 of 142 (12 rows lower down) in maximal MI.

Figure 7.

A, Histogram showing a typical example of the behavioral performance after 2 months of training in a Go/No-go paradigm in which marmoset twitter calls served as a Go stimulus. The length of the black and gray horizontal bars gives the number of No-go and Go responses, respectively, to each of the stimuli indicated along the y-axis. In this run, the animal made only two inappropriate No-go responses (1 each to twitters 2 and 3) and only four inappropriate Go responses (1 to the katydid and 3 to the cricket call). B, Reaction times, relative to stimulus onset, for Go and No-go trials, respectively, from the training session shown in A. C, Learning curves plotting performance (percentage of correct responses) against days from the start of training.

Figure 8.

Mean response rate evoked by natural (R_Nat) plotted against response to time-reversed (R_Rev) twitter call stimulus in ferrets trained to recognize marmoset twitter calls. Responses for first, second, and third exemplar of the twitter stimuli are shown in A, B, and C, respectively.

Figure 9.

A, Waterfall plot of MI as a function of temporal resolution for all 501 units from the trained ferrets in this study. B, Grayscale map showing the difference between trained (A) and untrained (Fig. 5B) animals. White contour lines are drawn at z = 0, 0.1, 0.2, and 0.3 bits.