(Left column) See Figure 4 legend.
(A–C) See Figure 4 legend, except for well-isolated SUs recorded from mothers. The interspike interval distribution (1-ms bins) is shown as an inset in the overall PSTH panels. The lack of short interspike intervals within the absolute refractory period (1 ms) supports the SU designation. The number of trials collected depended on how long each SU was maintained: 50 trials for SU A, 24 for SU B, and 14 for SU C.

For each site in mothers and naïve females that conveyed significant call discrimination information, the time courses of information about call frequency (three ranges) and call duration (three ranges) were also computed. The maximum (over time) information for both frequency and duration at each site were then determined and plotted. Sites generally conveyed more information about the call frequency rather than the duration (points mostly on the right of the diagonal). Furthermore, the most informative sites in mothers clearly conveyed more frequency information than those from naïve females. Thus, the improvement in call discrimination probably arose from changes in the spectral encoding of these natural calls. The symbols identify results for the corresponding sites in Figure 4.

The contour plot shows the probability that a natural pup calls has a specific combination of frequency and duration, with darker gray corresponding to higher probability (contours at probability densities of 0.19, 0.37, 0.56, 0.75, and 0.93/s/kHz). Only the main pup call cluster near 67 kHz and 59 ms is shown [19]. A grid of nine frequency-duration regions (centered at 67, 72.5 and 78 kHz and 14, 36.5 and 59 ms) laid across both high and low probability calls was used to condition the selection of playback calls on the basis of frequency and duration. Two natural calls from each region were randomly selected.

(A) Histograms show CFs (5-kHz bins) for sites in mothers (n = 83 out of 96 sites had identifiable CFs) and naïve females (n = 79 out of 102 sites). No significant differences were found between the distributions (two-sample KS test, n.s.).
(B) A whisker plot of the tone thresholds is shown. The notched lines indicate medians; the ends of each box mark the upper and lower quartiles; the whiskers denote the most extreme data values within 1.5× the interquartile range; and the crosses mark the outliers. No significant differences were found between the distributions (two-sample KS test, n.s.).

(A) Average time course of detection information in mothers for all sites where this ensemble was presented. Both the measured (magenta) and randomized (gray) information are shown.
(B) Average time course of detection information in naïve females for all sites where this ensemble was presented. Both the measured (sage) and randomized (gray) information are shown. The information peak (above the mean randomized information) for mothers was 1.5× larger than that for naïve females, irrespective of whether all sites or only those with information z-scores exceeding z_c were used.
(C) Histogram of peak detection information times (4-ms bins, relative to stimulus onset) for all sites is presented. The distribution for mothers was not significantly different from that of naïve females (two-sample KS test, n.s.), neither were their medians (two-sided rank sum test, n.s.), nor their means (two-sample t-test, n.s.).
(D) Average time course of discrimination information in mothers for sites with significant information is presented. Both the measured (magenta) and randomized (gray) information are shown.
(E) Average time course of discrimination information in naïve females for sites with significant information is presented. Both the measured (sage) and randomized (gray) information are shown.
(F) Histogram of peak discrimination information times (4-ms bins, relative to stimulus onset), for all sites. The distribution for mothers was not significantly different from that of naïve females (two-sample KS test, n.s.), neither were their medians (two-sided rank sum test, n.s.), nor their means (two-sample t-test, n.s.).

(A) The overall probability that a specific number of spikes was observed during either a “call” or “no call” (spontaneous activity) is shown.
(B) Initial probability for a “call” or “no call” is shown. The diameter of each circle is proportional to the probability. Both stimulus possibilities were considered equally likely before a response was observed.
(C) The conditional probability for a “call” or “no call” given that no spikes were observed is presented. Zero spikes did not substantially improve the uncertainty.
(D–F) Conditional probability for a “call” or “no call” given that one, two, or three spikes was observed, respectively. With each larger spike count, the probability for “call” increased, thus reducing the uncertainty in the stimulus and contributing detection information, weighted by the probability for that spike response (A).
(G) The overall probability that a specific number of spikes was observed during any of the 18 calls is shown.
(H) The initial probability for each call is presented. The diameter of each circle is proportional to the probability. All calls were considered equally likely before observing the response.
(I) Conditional probability for each call given that no spikes were observed. Zero spikes did not substantially improve the uncertainty.
(J–L) The conditional probability for each call is shown given that 1, 2 or 3 spikes were observed, respectively. With each larger spike count, the probability increased that one of the lower frequency calls occurred, thus narrowing the uncertainty in the stimulus, and contributing discrimination information, weighted by the probability for that spike response (G).

(A–H) Detection information for MUs is shown. Each panel corresponds to a site in Figures 4 or 5, as indicated by the labels and symbols. Most sites for both mothers (magenta) and naïve females (sage) exhibited peaks in detection information above the randomized control estimate (gray) during the stimulus presentation period. Note the change in scale (indicated to the right) between different panels.
(I–K) Detection information for SUs is shown. SU A from a mother was notable in that information remained significantly above the randomized control throughout and even after the duration of the stimuli. This corresponded to the period when calls inhibited neural firing relative to the spontaneous background, showing that zero spikes can be informative. Scale is to the right in each panel.
(L–S) Discrimination information for MUs for the same site in each column is presented. Some sites with good detection information (e.g., MUs 533 and 541) showed only weak discrimination information. Note the change in scale (indicated to the right) between different panels.
(T–V) Discrimination information for SUs is shown.

(Left column) Frequency trajectories and amplitude waveforms of the 18 pup calls are presented. All calls were simple whistles with different frequency and amplitude modulations. They increased systematically in duration and frequency.
(A–H) a raster plot of MU responses is shown. Each subpanel includes both the raster of individual spike responses from 12 trials, as well as its corresponding PSTH (2-ms bins). The top panel shows the spontaneous activity. The middle 18 panels show the responses to individual pup calls presented at the start of the horizontal black bar, and continuing for the duration of each call (bar's length indicates the duration of the longest stimulus, 65 ms). The larger, bottom panel shows the overall PSTH to all the pup calls, with a spike rate scale along the right edge of each panel. The labeling at the top of each column indicates the recording site (magenta for mothers and sage for naïve females) and its CF (in parentheses). The symbols are used for identifying each of the sites in subsequent figures. The downward arrow in the overall PSTH plot for panel (C) indicates the time bin used for the illustrations in Figure 6.

(A) The average time course of detection information in mothers, for significant recording sites (see text and Figure 8 legend) is presented. Both the population-averaged actual (magenta) and randomized (gray) information are shown.
(B) The average time course of detection information in naïve females, for all recording sites is presented. Both the actual (sage) and randomized (gray) information are shown. By fitting the average information time courses to the empirical function, a(t − t₀)³ exp(−b(t − t₀)) + c (fits had adjusted R-squares of 0.83 for mothers and 0.98 for naïve females), the peak in the population-averaged information time trace was found to be 1.3× larger in mothers than in naïve females (if actual peaks relative to the randomized baselines are used instead, mothers were 1.9× larger than naïve females).
(C) Peak detection information versus latency for each significant MU site is presented. The y-axis is plotted on a logarithmic scale to clearly separate individual points. The symbols identify results for the corresponding sites in Figure 4. Even at the level of individual recording sites, there was a clear tendency for high information sites in mothers to have shorter latencies. This was less so for naïve females.
(D) Histogram of peak detection information times (4-ms bins, relative to stimulus onset), for sites with significant information is presented. The distribution in mothers (magenta) was shifted to shorter latencies compared to naïve females (sage). The two distributions were significantly different (p ≪ 0.05, two-sample KS test), as were their medians (p ≪ 0.05, two-sided rank sum test) and means (p ≪ 0.05, two-sample t-test). The peak information times for SU A and SU B are indicated at the top of the panels. They coincided with the main peak in the distribution for mothers.
(E) Peak detection information versus CF for each significant MU site is presented. The y-axis is plotted on a logarithmic scale to clearly separate individual points. The symbols identify results for the corresponding sites in Figure 4. There was a significant correlation between peak detection information and CF for both mothers (r = 0.35, 95% confidence interval 0.11–0.55, p ≪ 0.05) and naïve females (r = 0.45, 95% confidence interval 0.23–0.63, p ≪ 0.05).

(A) Average time course of discrimination information in mothers for significant recording sites is presented (see text and Figure 8 legend). Both the population-averaged actual (magenta) and randomized (gray) information are shown.
(B) Average time course of discrimination information in naïve females for all recording sites is presented. Both the actual (sage) and randomized (gray) information are shown. By fitting the average information time courses to the empirical function a(t − t₀)³ exp(−b(t − t₀)) + c (fits had adjusted R-squares of 0.84 for mothers and 0.42 for naïve females), the peak in the population-averaged information time trace was found to be 3.4× larger in mothers than in naïve females (if actual peaks relative to the randomized baselines are used instead, mothers were 2.3× larger than naïve females).
(C) Peak discrimination information versus latency for each significant MU site is presented. The y-axis is plotted on a logarithmic scale to clearly separate individual points. The symbols identify results for the corresponding sites in Figure 4. Even at the level of individual recording sites, there was a clear tendency for high information sites to have shorter latencies, which was most apparent for mothers.
(D) Histogram of peak discrimination information times (4-ms bins, relative to stimulus onset) for sites with significant information is presented. The distribution in mothers (magenta) was shifted to shorter latencies compared to naïve females (sage). The two distributions were significantly different (p < 0.05, two-sample KS test), as were their medians (p ≪ 0.05, two-sided rank sum test), and means (p ≪ 0.05, two-sample t-test). The peak information times for SU A and SU B are indicated at the top of the panels. They coincided with the main peak in the distribution for mothers.
(E) Peak discrimination information versus CF for each significant MU site is presented. The y-axis is plotted on a logarithmic scale to clearly separate individual points. The symbols identify results for the corresponding sites in Figure 4. There was a correlation between peak discrimination information and CF for mothers (r = 0.36, 95% confidence interval 0.04–0.61, p < 0.05), but not for naïve females (n.s.). This correlation was weaker than the case of detection, perhaps reflecting the tendency that even lower CF sites (e.g., around 20 kHz) could have high peak discrimination information (~0.2 bits).

(A) The lines plot the cumulative probability that a site had a peak detection information value equal to or less than the abscissa.
(B) The lines plot the cumulative probability that a site had a peak discrimination information value equal to or less than the abscissa.
The peak (detection or discrimination) information was determined for each MU site as the maximum information within either 5–70 ms after the onset of the calls (solid lines), or within 365–430 ms after the onset (dashed lines). The former represents the peak information during the response to the calls. The latter provides an estimate of the noise in our information calculations since little information was expected during this late period, when the activity had essentially returned to the spontaneous level. The symbols identify the peak information values in response to calls of corresponding sites in Figure 4.
During the interval in response to calls, the distribution of peak information values in mothers clearly differed from naïve females, unlike the case for the late period. This difference appeared at higher peak information values, rather than across all information values. One reason the cumulative distributions coincided at lower information values may be due to noise or bias in our estimation procedure. This would have been more likely for the discrimination case (B), where the cumulative distribution for the late period estimate tracked that of the call period up to ~0.1 bits. This value corresponds to the naively expected bias for 18 stimulus possibilities and four response possibilities [31]. However, realistic simulations (see Materials and Methods) indicate that our estimation procedure is accurate when the true information is higher than ~0.1 bit. Indeed, if our estimation procedure were dominated by a systematic bias, the distribution of information values for the two animal groups would be quite similar across the whole range of values, as was the case for the late period estimate (dashed lines). An alternative explanation for the coincidence of the cumulative distributions at lower values might be that the information improvement between naïve females and mothers only involves sites that already convey some minimum amount of information about the calls. This seems plausible for detection since its systematic bias was negligible.
Nevertheless, to avoid possibly contaminating our comparisons between the two animal groups, we set a significance threshold (thin vertical black lines in each panel) for peak information values of 0.005 bits (detection) and 0.1 bits (discrimination). Sites with peak information estimates above these were considered valid. These thresholds corresponded to 90% of the largest peak detection and discrimination information values measured during the late period.

(A) Frequency trajectory (top) and amplitude waveform (middle) of the pup call (64-kHz median frequency, 50-ms duration, 65 dBSPL) are shown. At the bottom is the raster plot of responses from a MU (white circle) and extracted SU (gold circle) to 24 presentations of the call.
(B) A whisker plot is shown of the average number of spikes (across 24 trials) in progressively larger integration windows, triggered 6 ms after the onset of a call. The notched lines indicate medians; the ends of each box mark the upper and lower quartiles; the whiskers denote the most extreme data values within 1.5× the interquartile range; and the crosses mark the outliers. No significant differences (p > 0.05) between mothers (magenta, n = 86) and naïve females (sage, n = 74) were found in the median (two-sided rank sum test), mean (two-sample t-test), or distribution (two-sample Kolmogorov-Smirnov [KS] test) for spike counts at any window duration. The same color designation for mothers and naïve females is used in all plots.
(C) PSTH (smoothed at 2 ms) is shown segregated by recording sites' CF into three ranges (0–20 kHz, 20–40 kHz, and 40–80 kHz). The black, horizontal bar indicates the playback period. The vertical scale bar equals 50 spikes/s. Fitting the populations' PSTHs to the empirical function, a(t − t₀)³ exp(−b(t − t₀)) + c, the peak times and widths could be extracted (fit with MATLAB cftool, restricted to the time interval from six to 71 ms relative to the stimulus onset, resulting in adjusted R-square values between 0.87 and 0.96). The highest CF group showed an earlier (16.6 ms versus 21.2 ms) and narrower (15.1 ms versus 27.4 ms half max) peak in mothers compared to naïve females, respectively. This was also true of the less strongly driven middle CF group (18.2 ms versus 29.5 ms peak and 20.1 ms versus 28.1 ms width).