Predictive Mouse Ultrasonic Vocalization Sequences: Uncovering Behavioral Significance, Auditory Cortex Neuronal Preferences, and Social-Experience-Driven Plasticity

Mice produce context-dependent syllable sequences

We developed a protocol for mouse vocalization production in three different contexts., referred to as Alone, Separated, and Together (Fig. 1A). When lone naive adult male mice [n = 4, postnatal day (P)56–P90] with no previous social exposure to a female were placed in a cage within a sound chamber (Alone, 5–10 min), they emitted no vocalizations (Fig. 1A, top left). Following introduction of a naive female mouse (P56–P90) into the cage, separated from the male by a mesh (Fig. 1A, middle row), vocalizations were produced by the male (Separated, 5–10 min). Finally, the mesh was removed, and the female and male were exposed to each other, and vocalizations were recorded (Together, 5–10 min). We repeated the process with the same pair of mice for at least 5 d until the male mouse emitted vocalizations in the Alone condition (see above, Materials and Methods). Ultrasonic frequency (UF) vocalization recordings (20–120 kHz) made from the above stage onward for the following 5-7 d in four pairs of mice were further analyzed and grouped into the three contexts above.

Vocalization sequences were automatically annotated by detecting ultra-frequency syllables based on energy threshold (see above, Materials and Methods), and syllables were categorized into five types (see above, Materials and Methods; Fig. 1B; Agarwalla et al., 2020), based on their spectrograms with pitch jumps as the distinguishing feature (Holy and Guo, 2005). The syllable types were, Noisy (N); Single (S) frequency contour; Jump (J) type, with a single jump in frequency; Harmonics (H); and Others (O) consisting of multiple Jumps. The relative percentage of the different kinds of syllables produced by male mice in each context (Fig. 1C) showed distinct differences. The overall difference in the distributions of syllables in each pair of the contexts was quantified using KLD (see above, Materials and Methods; Cover and Thomas, 2006; Bandyopadhyay and Young, 2004; Agarwalla et al., 2020). All pairs of distributions were significantly different at 95% CI (see above, Materials and Methods). However, the Together condition showed the most distinct distribution having high distances from the others, with the Alone and Separated cases showing a similar nature. Despite differences in distributions, order of syllables could be random in the three contexts. To probe whether the syllables are random or nonrandom, we analyzed syllable-to-syllable transition probability matrices as done in multiple studies (Holy and Guo, 2005; Chabout et al., 2015, 2016) (see above, Materials and Methods) with two successive syllables (Fig. 1D) or disyllables. We found distinct differences, quantified with KLD at 95% CI (Fig. 1D, right) between the different contexts suggesting differences in structure of syllable order.

To probe higher-order structure in syllable sequences, analyses as above with three or more successive syllables require large amounts of data for reliable estimation of transition probabilities. Thus, to analyze structure in the order of syllables emitted (Holy and Guo, 2005; Grimsley et al., 2011; Chabout et al., 2015; Agarwalla et al., 2020), the entire period of vocalization recordings of each session was divided into bouts based on the ISS distribution. Silence intervals >250 ms in duration were used to mark the end of a bout of vocalizations for Alone (80.7 ± 79.2 ms), Separated (99.4 ± 81.4 ms), and Together (88.9 ± 80.6 ms). ISS distributions based on merging the distributions from all contexts had a mean +2 * SD of (90 + 2 * 80) = 250 ms (Fig. 1E). The syllable following the immediately preceding end of a bout was considered as the start of the next bout, and successive syllables in a bout were considered as a vocalization sequence of syllables. Thus, bouts of vocalizations could have only one or many syllables (observed up to 30 syllables; Fig. 1F; total 948 Alone, 3320 Separated, and 6656 Together bouts recorded), with bouts of length 1 being dominant, independent of the context. Bouts of vocalization with three or more syllables are referred to as sequences (see above, Materials and Methods) hereafter. With S_i denoting the ith syllable of a bout, the MI between S₁ and S_i, denoted I(S₁;S_i), was computed, debiased, and tested for significance based on CI (see above, Materials and Methods). Significant MI between syllables at different positions in a bout shows dependence between the syllables or predictability, the basis of structure in sequences (Shannon, 1951; Agarwalla et al., 2020). We found that in the three contexts significant dependence existed between the first and other syllables in bouts of vocalizations (Fig. 1G). The dependence was strongest and longest in the Together context and weakest and shortest in the Alone condition. Thus, the three contexts had different degrees of dependence in syllables of sequences. Similar results were observed for I(S_j;S_j+k), j = 1,2,3, and k = 1,2 (Fig. 1H), showing dependence exists within the sequence beyond that on the first syllable. Thus, context-dependent sequences are produced by mice with degrees of structure varying with context.

To further understand what order of syllables, if any, were involved in providing such dependence in the sequences, we found the high probable sequences produced in the three contexts (Fig. 2B, sequence 13–20; see above, Materials and Methods). The spectrogram of the syllables used for the generation of the stimulus is shown in Figure 2A. The eight sequences obtained are NSeqs. The NSeqs or their subsequences (from the starting syllable with at least three syllables) constituted 36% (Together) and 24% (Separated) of all the recorded sequences of length three or more in the two contexts. The percentages of such sequences occurring by chance based on the Together and Separated syllable probability distributions (Fig. 1C) were only 11 and 8%. The NSeqs were also 8 of the 10 sequences with the highest accumulated surprise (Itti and Baldi, 2009; see above, Materials and Methods) showing that they were the most informative among the repertoire of syllable sequences produced in the three contexts. As expected from the results of MI (Fig. 1G), the longest NSeqs were in the Together condition, and Together and Separated conditions had distinct sets of NSeqs indicating context-dependent specific sequence production in social encounters of male and female mice. The only high-probable sequence in the Alone context was also present in the Separated context and is possibly a search sequence before finding a female mouse. However, all the NSeqs were present in the Together condition, and thus during the Together condition the female was exposed to all the NSeqs. To study the importance of the NSeqs, if any, both from a behavioral perspective and coding perspective, as control, 12 other sequences were designed. Four randomly ordered sequences were created from the probability distributions of syllables in each context (Fig. 1C). The above sequences, each with seven syllables, the length of the longest NSeqs, are considered RSeqs (Fig. 2B, sequence 1–12; see above, Materials and Methods).

Naive adult female mice prefer NSeqs over RSeqs

Although NSeqs were obtained from vocalizations produced during social exposure of males to females, the sequences need not be behaviorally relevant to naive females. We tested the relevance of the specific sequences obtained with a two-sided free access/choice test (Musolf et al., 2015; Chabout et al., 2015; see above, Materials and Methods). Two sets of experiments were conducted, one in which 8 of the 12 RSeqs were truncated to match the number of syllables in NSeqs (Fig. 2B, shaded RSeq, same length) and the other in which all RSeqs had seven syllables as in the longest NSeqs. The latter case was used to provide the maximum possibilities of syllable-to-syllable transitions that could be NSeqs occurring. The above allowed us to probe the relevance of the full sequence as opposed to NSeq transitions or disyllables occurring randomly. Naive adult female mice (females with no prior exposure to adult male mice; P56–P90, n = 8, two outliers, same length; n = 7, 1 outlier, 7 length) were placed in an open cage for four distinct sessions (S1–S4) with sounds presented in S2 and S4 (see above, Materials and Methods). Sequences from RSeq were played back from one corner and those from NSeqs from the opposite corner, alternately every 5 s, in S2. RSeq and NSeq respective sides were switched in S4 to remove any side bias (Fig. 2C, top row). Female mice had significantly higher preference (p  <  0.001, both cases) for NSeqs independent of which side NSeqs were played back, both for equal length and length 7 sequences (Fig. 2C, bottom row, quantified by dividing the cage floor into three equal parts denoting NSeq, Neutral, and RSeq; see above, Materials and Methods). Thus, the extracted NSeq sequences are relevant to naive female mice and are preferred over the designed RSeq.

Higher c-fos expression in the mouse ACX for the NSeqs over RSeqs

Because the theoretically derived informative sequences have behavioral relevance for naive adult female mice over random sequences (Fig. 2D), we further investigated the relative degree of effect of NSeqs over RSeqs at the neuronal level, particularly in the ACX. For the above, we used c-fos immunohistochemistry (see above, Materials and Methods) as a marker for immediate early gene expression and hence neuronal activity (Clayton, 2000). The presence of c-fos in a neuron is one of the methods to identify the specific neurons that participate in a given function without losing the ability to precisely locate such neurons (Clayton, 2000). c-fos Expression has been known to mark synaptic plasticity occurring in neurons because of novel experience, memory trace formation, and learning, among other such phenomena (Liu et al., 2012; Moreno et al., 2018; Minatohara et al., 2016). Three different sound environments were used to expose three groups of female mice (n = 5 per group) age between P56 and P90 (see above, Materials and Methods). First, the control condition, in the absence of auditory stimulus, started with 60 min of habituation followed by 15 min of a silent period, which is the same as the duration of stimulus presentation for the two other conditions, and ended with an additional 60 min as a consolidation period (Fig. 3A). In the second and third conditions the 15 min silent period in control condition was replaced by RSeq and NSeq presentations (Fig. 3A), keeping other things identical to control conditions. Coronal sections (Fig. 3B) across the rostrocaudal extent of ACX were used to mark the boundaries of ACX (Paxinos and Franklin, 2019). The neuronal activation of the different regions involved for the three different contexts (Fig. 3B, left) were quantified with the number of c-fos-positive neurons per unit area (Fig. 3B, right column). To further assess how much of these effects are because of general arousal or specific auditory cortical processing, we considered primary (V1) and secondary (V2L) visual cortex as control brain regions. As shown in the bar plot (Fig. 3C, top row), the results of the stereological analysis (see above, Materials and Methods) revealed no differential expression in any of the conditions for visual cortex regions V1 and V2L. A higher number of c-fos+ neurons were present in the NSeq condition compared with RSeq and control conditions in the ACX. The overall observation did not alter when considering the different subregions of ACX separately, namely, Au1, AuD, and AuV (Paxinos and Franklin, 2019; Fig. 3C, middle and bottom). Thus, our observations indicate that exposure to NSeqs has very different effects in terms of expression of c-fos which can indicate salience of NSeqs over RSeqs or significant behavioral relevance of NSeqs over RSeqs or initiation of synaptic plasticity in ACX with NSeq exposure. Results of behavioral relevance (Fig. 2D) together with those of c-fos expression (Fig. 3) warrant investigation of coding of NSeqs as opposed to RSeqs. As NSeqs and Rseqs are from the same distribution, the question of whether presence of predictiveness for the NSeqs over RSeqs is important to understand as it may relate to perceptual learning of the specific sequences.

• Download figure
• Open in new tab
• Download powerpoint

Naive adult female mice show higher activation of c-fos+ cells for NSeqs over RSeqs. A, Schematic depicting the three stages followed to investigate activation of c-fos+ cells. Throughout the experiment up to killing, the mouse remained in a sound-attenuating chamber; Stage 1, 60 min of habituation; Stage 2, 15 min of exposure to auditory stimulus (RSeqs or NSeqs or silence for Ctrl); Stage 3, 60 min of consolidation in silence. B, Representative coronal brain sections for c-fos expression from different groups, Ctrl (top), Ran (middle), and Nat (bottom), with ACX subregions Au1, AuV, AuD, and visual cortex areas (V1 and V2L as control) demarcated with dotted lines. Right, Enlarged images show sampled Au1 regions with c-fos+ cells (top, white arrow, red), corresponding DAPI-stained nuclei identification (middle, white arrows, blue), and the overlay of the two (bottom). C, Bar plots show average count of c-fos+ cells/mm² in three conditions, Ctrl, Ran, and Nat, for V1. V2L, and total ACX, subregions of ACX (Au1, AuD, AuV). Quantitative differences of c-fos+ cells for each of the cortical subregions are marked with statistical significance at 5% significance level using a one-way ANOVA; *p < 0.05, **p < 0.01, ***p < 0.001; NS, Not significant.

Single units in Au1 code single syllables and disyllables differentially for NSeqs and RSeqs

To investigate coding of behaviorally relevant NSeqs of vocalizations relative to RSeqs, we first performed extracellular single-unit recordings from layer 2/3 (200–350 μm from the surface) of mouse Au1 using multielectrode arrays (Sharma and Bandyopadhyay, 2020; Srivastava and Bandyopadhyay, 2020). Three groups of mice were used, passively listening awake naive females (Awk_F; 328 units from 6 animals, chronic recordings, over 10–14 d), anesthetized naive females (Anes_F; 266 units from 12 animals), and naive anesthetized males (Anes_M; 195 units from 8 animals).

Typically, five (between four and six) repetitions of each stimulus of RSeqs (1–12) and NSeqs (13–20) were presented in pseudorandom order (Fig. 4A, left) and single-unit spiking (Fig. 4A, inset) with 500 ms baseline was recorded. The stimuli varied in length based on the component syllables, from 0.389 s to 1.233 s. Peristimulus histograms (PSTHs) were constructed, and responses locked to single tokens were observed throughout the RSeqs and NSeqs (Fig. 4A, right).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Differential coding of single syllables in Au1 for NSeqs and RSeqs. A, Representative dot raster plot of single-unit spiking responses to NSeqs and RSeqs presented in pseudorandom order (right, spike shape). Smoothed PSTHs (binning of 10 ms) of the same unit for each stimulus sequence is shown with stimulus start (tall vertical line) followed by lines marking start and end of each subsequent syllable. B, Schematic for calculation of responses to common syllables within the sequences in NSeq and RSeq, not considering the syllables in the starting position. Different color bars represent different syllable types, and the width of color bars is indicative of syllable duration. The block dots are the spikes corresponding to each iteration (REP1, REP2…REPN). The mean responses of the syllables with the same color (indicated by colored space on the schematic of spikes over n repetitions) were calculated for RSeqs (Baseline, left) and NSeqs (Baseline, right) for each of the different common syllable types. C, Scatter plots show comparison of mean response rates for all common first syllables in NSeqs and RSeqs in three groups of mice, Awk_F, Anes_F, and Anes_M. D, Scatter plots show comparison of mean response rates of all common first (identified by solid symbols (S, large circle; H, square; O, triangle; syllable types in NSeq and RSeq in 3 groups of mice, Awk_F, Anes_F and Anes_M). E, Scatter plots comparing mean response rates of common syllables (identified by solid symbols (S, large circle; J, small circle; H, square; O, triangle) in NSeqs and RSeqs, excluding occurrence in the first position for the same groups in C. F, Profile of changes in the normalized response strength for each position of the sequences NSeq (blue) and RSeq (red) over time, normalized by response to the syllable in the starting position. None of the profiles at any position showed any significant difference (one-way ANOVA). Each dot in the scatter plots (C, D, E) represents an individual neuron; *p < 0.05, **p < 0.01, ***p < 0.001; NS, Not significant.

Syllables S, J, H, and O and disyllables or transitions S–S, J–J, H–H, and O–O were present in both RSeqs and NSeqs, excluding the first syllable of any sequence. The first syllable was excluded as its occurrence does not contain any sequence information (Fig. 4B). Thus, as a control, in all three groups of the average rate responses of animals of each syllable type when occurring in the first position were compared (First syllable after the dashed vertical line, Fig. 4B, see above, Materials and Methods). As expected, none of the responses of the syllable in the first position showed any difference between NSeqs and RSeqs (Fig. 4C,D, for each syllable). Next, we compared the single-unit rate responses of syllables, present in both RSeq and NSeqs (Fig. 4E; see above, Materials and Methods). Syllables J and O, across all three groups of animals, elicited higher average rates in Au1 single units when occurring in NSeqs than when they occurred in RSeqs (paired t test, p  <  0.01 in all cases, except syllable O in Anes_F had p  <  0.05; Fig. 4E). However, syllable H across all the animal groups evoked lower average rates for NSeqs than for RSeqs (paired t test, p  <  0.01 in all cases). Syllable S produced significantly higher rates when in NSeqs (paired t test p  <  0.01) compared with when in RSeqs for Awk_F group, and response strength of S was not significantly different in NSeqs and RSeqs in the other two groups of animals.

Thus, the single syllables were encoded differentially in a NSeq context than in an RSeq context, specifically a higher response was generally elicited by syllables in NSeqs than RSeqs, except for the H syllable. We rule out the possibility that the above differences could be an effect of stimulus-specific adaptation (Carbajal and Malmierca, 2018; Srivastava and Bandyopadhyay, 2020; Mehra et al., 2022b). It may appear that because of repetitions of syllables in NSeqs (e.g., H in NSeq 15 and NSeq 16) and absence of such long repeats of H in RSeqs could potentially explain the lower response rates to H in NSeqs compared with RSeqs (Fig. 4E). However, repetitions of up to four successive syllables are present for S, J, and O as well as H in NSeq with such repeated occurrences of them being absent in RSeqs. But unlike H, the S, J, and O syllables evoked higher rates in NSeqs than RSeqs. Thus, it is unlikely that adaptation can explain the observed context-dependent responses to tokens. To more conclusively rule out adaptation as the cause of such differences, we compared the average adaptation profiles (Fig. 4F) in NSeqs and RSeqs by normalizing each single-unit mean response to the first token as one, which was not significantly different between NSeqs and RSeqs for any token (Fig. 4C,D). No significant differences were found in the adaptation profile based on average token-wise responses over NSeqs and RSeqs across animal groups.

The above results suggest that predictiveness binding the syllables is likely an important determinant of the difference in responses of individual syllables in NSeqs and RSeqs. We thus considered coding of the common disyllables in NSeqs and RSeqs, considering only the response to the second syllable in the transition (Fig. 5A; see above, Materials and Methods). Comparison of rate responses to transitions in NSeqs and RSeqs showed similar results as single syllables, with S–S, J–J, and O–O transitions generally showing stronger responses when present in NSeqs than when present in RSeqs across all animal groups (except S–S in Anes_F and Anes_M; Fig. 5B–D, right). As with responses to the single syllable H, the transition H–H had higher responses in the RSeq context compared with the NSeq context across all groups of animals (Fig. 5B–D, right). Representative example PSTHs are shown for the groups under consideration (Fig. 5B–D, left). Syllable transitions of other kinds (Fig. 1D) were not common to both NSeqs and RSeqs. All the other kinds of disyllables in NSeqs were present as the first transition, which were excluded, as the first syllable did not have context information (Fig. 4C,D). However, the other disyllables occurring as the first transition in NSeqs (S–J, S–H, S–O, and O–H) were significant in transition matrices (Fig. 1D). Thus, we compared responses of the above transitions based on the response to the second token of the transition, between NSeqs and RSeqs (Fig. 5E–G). In the above remaining cases also, we generally find higher responses to the transitions occurring in NSeqs compared with those occurring in RSeqs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Differential coding of disyllables in Au1 of mice for NSeqs and RSeqs. A, Schematic for calculation of mean rate responses to transitions/disyllables based on response to the second component, excluding the first transition. Each color bar represents a different syllable type; the width is representative of syllable duration. The black dots stand for the spikes correspond to each iteration (REP1, REP2…REPN). Right, The common transitions. Left, The color lines used for showing the transitions, namely, green, blue, and cyan, have been correspondingly used for denoting the spikes over iterations for those transitions. Mean spike rates were computed for each of the same colored areas indicated in the schematic of the spikes over N repetitions for RSeqs (Baseline, left) and NSeqs (Baseline, right) for each of the common transitions. B–D, Representative example PSTHs (vertical gray bar indicates start of a sequence, red line along the x-axis represents the stimulus duration) and scatter plots comparing mean response rates of common disyllables in NSeqs and RSeqs for the three groups, Awk_F (B), Anes_F (C), and Anes_M (D). E–G, Scatter plot of comparison between mean rate responses to first transition in NSeqs (S–J, S–H, S–O, and O–H) and the same transition present at any position in RSeqs based on response to the second component, excluding the first transition for the groups AwK_F (E), Anes_F (F), and Anes_M (G); *p < 0.05, **p < 0.01, ***p < 0.001; NS, Not significant.

Social exposure of adult female to adult male alters coding of entire NSeq sequences but not their component syllables in female A1

Our results show that naive males and females both show the same results of differential coding of single syllables in the two sets of stimuli, RSeqs and NSeqs (Fig. 5). Further, the NSeq stimuli used are preferred by naive female mice compared with RSeqs (Fig. 2C). Social exposure between adult male and female mice (>P56) is a significant event constituting the NSeq USVs produced (Egnor and Seagraves, 2016; Chabout et al., 2015; Yang et al., 2013) (Fig. 1, 2). Further H syllables and sequence of Hs were a major component of the USVs especially in Together context (Fig. 1C,D). Thus, we hypothesize that the context-specific coding of syllables and disyllables would change, especially for syllable H and disyllabic H–H, both of which showed lower responses for NSeqs.

We first tested the above hypothesis with single-unit recordings from layer 2/3 of Au1 of experienced female mice (Fig. 6A; see above, Materials and Methods) with a different number of days of exposure to male mice (1–5 d of exposure, 421 units from 8 animals, Anes_F-Aft_Expo). Because the differential coding observed for syllables and disyllables was the same in Awk_F and Anes_F (Figs. 4E, 5B–D) except for S, we considered rate responses in anesthetized mice. First, as a control for context independence, we confirm that responses to all first syllables did not change between NSeq and RSeq sequences (Fig. 6B,C, each syllable type). Example PSTH for a unit for all the sequences are shown in Figure 6D for an experienced female mouse. Contrary to our hypothesis, we found that experienced females had identical differential representation of single syllables (Fig. 6E) and disyllables (Fig. 6F,G) as the naive females (Figs. 4E, 5B–D) before exposure (Awk_F-Bef_Expo and Anes_F-Bef_Expo), except S and S–S in the latter case. However, selectively stronger responses to S in NSeq sequences in Awk_F-Bef_Expo and Anes_F-Aft_Expo compared with that in RSeq, suggests that the exposure likely had little effect on context-specific coding of single syllables and transitions. As in the other groups, adaptation could not explain the context-dependent coding of single syllables in the experienced females (Fig. 6H).

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Coding of syllables in Au1 remains unaltered after social experience. A, Schematic of social exposure protocol of female mice with male mice over days. B, Scatter plot for comparison of mean response rates to the common first syllables in NSeqs and RSeqs, following exposure in anesthetized females (Anes_F-Aft_Expo). C, Scatter plots show comparison of mean response rates of all common (solid symbols, S, large circle; H, square; O, triangle) first syllable types in NSeqs and RSeqs for Anes_F-Aft_Expo. D, Representative PSTH for an experienced female (vertical gray bar represents the onset of the stimulus, the red line along the x-axis represents the duration of the sequence). E, F, Scatter plot to compare mean rate responses to common syllables (E) and disyllables (F) for anesthetized females (Anes_F-Aft_Expo) as in Figure 4E and Figure 5B–D, respectively. G, Scatter plot of mean rate responses to first transition in NSeqs (S–J, S–H, S–O, and O–H) based on response to the second component, excluding the first transition. H, Profile of changes in the normalized response strength for each position of the sequences NSeq (blue) and RSeq (red) over time with respect to the syllable in the starting position; not significant (NS) across position, one-way ANOVA; *p < 0.05, **p < 0.01, ***p < 0.001; NS.

We observed that the NSeqs as a whole may have the potential to carry information that is behaviorally relevant in the male–female exposure compared with its components. Further, sequences in NSeqs appeared as the most informative set of sequences in the exposure event over multiple days. Thus, we made a rate comparison for monosyllables, disyllables, and sequence selectivity of individual neurons (see above, Materials and Methods) and compared mean selectivity to RSeqs and NSeqs in the population of neurons in each group (Fig. 7). No difference in neuronal selectivity was observed for monosyllables and disyllables among the different groups under consideration (Fig. 7A,B). However, differential encoding was well captured at the sequence level (Fig. 7C). Interestingly, both Awk_F-Bef-Expo and Anes_F-Bef-Expo showed higher selectivity to NSeqs than RSeqs (paired t test, p < 0.001) with no difference between the two groups of females (one-way, ANOVA, p > 0.05, NSeqs; p > 0.05, RSeqs), reiterating the lack of difference in such selectivity in the awake and anaesthetized conditions. Anes_M-Bef-Expo also showed increased selectivity to NSeqs compared with RSeqs, and surprisingly, had higher selectivity for NSeqs than naive females (one-way ANOVA, p < 0.001, AwF-Bef-Expo; p < 0.001, Anes_F-Aft_Expo). The Anes_F-Aft_Expo group of mice showed the highest difference between selectivity to NSeqs and RSeqs. Interestingly, females after exposure had lower selectivity to RSeqs compared with all other groups (one-way ANOVA p < 0.001, Awk_F-Bef-Expo; p < 0.001, Anes_F-Bef-Expo; p < 0.001, Anes_M-Bef-Expo). Also, the selectivity to NSeqs of Anes_F-Aft_Expo was similar to that of Anes_M-Bef-Expo (one-way ANOVA, p > 0.05) but significantly higher than naive females (one-way ANOVA, p < 0.001, Anes_F-Bef-Expo; p < 0.001, Awk_F-Bef-Expo).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Plasticity observed in single-unit selectivity to entire the NSeq and not its components. A–C, Comparison of mean overall selectivity in NSeqs to that in RSeqs and comparisons across groups before exposure (Awk_F-Bef-Expo, Anes_F-Bef-Expo, and Anes_M-Bef-Expo) and after exposure (Anes_F-Aft_Expo), monosyllables (A), disyllables (B), and overall sequences (C). D, The mean selectivity to NSeqs and RSeqs in anesthetized females before exposure (Anes_F-Bef_Expo) is compared with selectivity to NSeqs and RSeqs over days of exposure and within days of exposure (Anes_F-Aft_Expo); *p < 0.05, **p < 0.01, ***p < 0.001; NS, Not significant.

To further test that the increased difference in selectivity in experienced female mice was because of the exposure to male mice, we considered the effect of multiple days of exposure (Day1, Day2, Day3 and Day > 3; Fig. 7D). First we observe that on every time point considered, selectivity to NSeqs was significantly higher than selectivity to RSeq during the exposure period (Fig. 7D; paired t test, p < 0.001; Day1, p < 0.001; Day2, p < 0.001; Day > 3, p < 0.001). The effect of exposure over days shows that the selectivity to NSeqs in Anes_F-Aft_Expo was significantly higher than that in Anes_F-Bef_Expo on all the days (one-way ANOVA, p < 0.001, Day1; p < 0.001, Day2; p < 0.001, Day3, p < 0.001, Day > 3). Moreover, we found that selectivity to NSeq increased significantly over the first two days (one-way ANOVA, Day1 > Bef_Expo, p < 0.001; Day2 > Day1, p < 0.001). Following Day2, there was no significant change in selectivity to NSeq sequences and was saturated. On the other hand, selectivity to RSeq was unchanged on Day1 (one-way ANOVA p > 0.05) and decreased on Day2 (one-way ANOVA, p < 0.001) and then remained unchanged over further days (one-way ANOVA, p > 0.05, Day2/Day3; p > 0.05, Day3/Day > 3) while being significantly less than that of naive females and that of Anes_F-Aft_Expo Day1 (one-way ANOVA, p > 0.05). These observations are clearly consistent with the idea that exposure or experience-dependent plasticity occurs to increase selectivity to the entire sequence but not its component syllables and disyllables. As responses of RSeqs and NSeqs are from the same neurons collected in pseudorandom order, it suggests that neurons with higher selectivity to NSeq also lose selectivity to RSeqs over days to maximize differences of the relevant sequences from any others. Thus, although the Aft_Expo Day2 group of mice was never exposed to the RSeq, we find a decrease in selectivity to RSeq. The difference in selectivity to NSeqs and RSeqs saturates after Day2 and it is explained by the fraction of units in layer 2/3 of A1 that have higher selectivity to NSeqs than to RSeqs (χ² test, p > 0.01). Fraction of units with higher selectivity for NSeq than RSeq significantly increases on Day2 from Day1 (χ² test, p < 0.05) coincident with the decrease in mean selectivity of RSeqs.

SOM-positive interneurons and Thy-1 excitatory neurons exhibit different experience dependent plasticity in NSeq selectivity

To elucidate the mechanisms underlying the observed plasticity, we hypothesized the involvement of inhibitory interneurons as observed in multiple studies (Schreiner and Polley, 2014). Our single-unit recordings showed mostly regular spiking, and hence extracting putative fast-spiking inhibitory neurons was not possible. Our final conclusions above are based on responses to entire sequences (0.389–1.232 s) and not fine time scale activity information. Thus Ca²⁺-dependent fluorescence imaging was performed in naive and experienced female mice in the awake state through a cranial window (see above, Materials and Methods). We used three types of mice to image activity of Thy-1-positive EXNs (naive n = 3360 neurons, experienced n = 2853 neurons from n = 3 mice, strain #024339, The Jackson Laboratory) and genetically identified individual inhibitory neurons (SOM, naive n = 517 neurons, experienced n = 460 neurons from n = 4 mice, strain #013044 crossed with strain #024105, and PV-positive, naive n = 395 neurons, experienced n = 438 neurons from n = 4 mice, strain #08069 crossed with strain #024105, all from The Jackson Laboratory).

We first performed wide-field Ca²⁺ imaging (Mehra et al., 2022a) (see above, Materials and Methods) to identify the location of Au1 with UF (Stiebler et al., 1997; Bandyopadhyay et al., 2010) and other auditory areas (Issa et al., 2014; Fig. 8A). Following identification of Au1 (see above, Materials and Methods), fine scale two-photon imaging (see above, Materials and Methods) with single-neuron resolution was performed by restricting individual ROIs (120–150 × 300–350 μm) within Au1. Single ROIs (Fig. 8B; two to three ROIs per day) were imaged in each group of mice, female before exposure (Fig. 8B, left, Thy1-GCamp; Fig. 8B, middle, SOM-GCamp; Fig. 8B, right, PV-GCamp) and after exposure. Chronic recordings in this case allowed collecting data from the same animals over different days of exposure (5 d); however, the same neurons could not always be tracked reliably over days. As expected from the literature, we obtained data from 100 ± 20 neurons in each ROI of Thy-GCamp mice and 6 ± 4 and 8 ± 4 neurons in each ROI from SOM-GCamp and PV-GCamp mice, respectively. Within an ROI, many neurons produced significant responses (mean df/f compared with mean baseline df/f, t test, p < 0.05; see above, Materials and Methods) to one or more stimuli in all the cases (Fig. 8C).

• Download figure
• Open in new tab
• Download powerpoint

Figure 8.

Responses to sequences obtained with two-photon Ca²⁺ imaging of Thy1, SOM, and PV neurons. A, Representative examples of tonotopy in Au1 and other auditory areas in mouse Au1 obtained in all three groups of mice (Thy-1-GCamp, SOM-GCamp, and PV-GCamp, respectively, in three columns, separated by dashed lines) are shown. White box marks the area shown in B. B, Sample two-photon image of an ROI in A1 in each of the three groups of mice. C, Average df/f plots obtained with two-photon imaging, in response to each of the 20 stimuli for two cells in each ROI (Cell A1, B1, C1 and Cell A2, B2, C2, respectively). D, Left, Bar graphs show percentage of single units (blue) with significant rate responses to each stimulus (1–20; Fig. 2B) in Awk_F-Bef-Expo and that of single Thy-1-positive EXNs (brown) with significant responses in Ca²⁺ in EXN-Bef-Expo. Right, Bar graphs show (same color representation as left) percentage of neurons responding to number of stimuli either 1, 2, or all 20 of the stimuli; the stimuli identity (id) doesn't matter.

Distinct differences in response behavior to sequences existed in the population of neurons between single units (primarily EXNs) and Thy-1 neuron Ca²⁺ responses. In general, responses were sparser with two-photon imaging and underlie some of the discrepancies observed between imaging and extracellular recordings (Bandyopadhyay et al., 2010; Siegle et al., 2021; Rothschild et al., 2010). Fractions of neurons responding to each of the 20 sequences (Fig. 2B) in the awake state show lower fractions with imaging compared with single-unit recordings in the awake state (Fig. 8D). Similarly, the fraction of neurons (above two groups) responding to the number stimuli (independent of the identities of the stimuli) showed that the single-unit population responded less selectively than the population of neurons observed with imaging (mean 2.9 and 5.6 stimuli, respectively, with Ca²⁺ imaging and single units; Fig. 8D; χ² test, p < 0.001). Thus, direct comparisons with single-unit data may not be possible when using Ca²⁺-based responses.

We considered the selectivity of the three different types of neurons to NSeqs and RSeqs and the effect of exposure in two ways. First, we considered the relative selectivity to NSeqs and RSeqs by grouping the neuronal populations according to how many of the RSeq and NSeq neurons of each type respond simultaneously. We first observe that before exposure, Au1 single Thy-1 EXNs and SOM respond to subsets of both NSeqs and RSeqs similarly, with neurons responding to few or none of RSeqs and also responding to few or none of NSeq sequences. Neurons responding more for RSeqs are more likely to respond to more of the NSeqs (Fig. 9A). However, with exposure, EXNs reduce the number of NSeqs they respond to (Fig. 9A, histograms). The same is true of SOM but to a lesser degree. PVs, on the other hand (Fig. 9A, right), show a similar degree of responding simultaneously to NSeqs and RSeqs before and after exposure. Thus, it is likely that PV neurons are less involved in the observed exposure-based change in selectivity to NSeq sequences. We quantified the compactness and its change before and after exposure in the three groups of neurons by the bootstrap (n = 1000) mean of the number of nonzero elements in the matrix representations in Figure 9, A–C. EXNs and SOM showed significant changes (EX-Bef_Expo, 63.3 and EX-Aft_Expo, 33.2; SOM-Bef_Expo, 56.8 and SOM-Aft_Expo, 38.7; nonoverlapping 95% confidence intervals) expected as above, whereas the PV neuron population did not show any change (PV-Bef_Expo, 24.3 and PV-Aft_Expo, 21.1; n.s.).

• Download figure
• Open in new tab
• Download powerpoint

Figure 9.

Differential effects of social-experience-driven plasticity in EXNs and SOM and not in PV. A–C, Population data with two-photon imaging in Thy1-GCamp (A), SOM-GCamp (B), and PV-GCamp (C) mice, with two matrix plots for before (left) and after (right) exposure. The rows in each matrix represent the percentage of neurons in the group and condition that respond to none (0), one, two, all (8) of the NSeq (x-axis) and of the neurons that respond to none (0), one, two, all (12) of the RSeq (y-axis). Right, Marginal distributions show the number of neurons responding to the different number of RSeqs. Distribution at the bottom of each matrix plot shows the average of the rows. Comparison of average selectivity to NSeqs and RSeqs in each condition Bef_Expo and Aft_Expo and between the conditions for each neuronal type (D) Thy-1; (E) SOM; (F) PV; *p < 0.05, **p < 0.01, ***p < 0.001; NS, Not significant.

Next, we compared mean selectivity to NSeqs and RSeqs, based on responses quantified with df/f (see above, Materials and Methods) before and after exposure in Thy-1-GCamp female mice. We found that EXNs had increased selectivity for NSeqs compared with RSeqs (one-way ANOVA, p < 0.001) following exposure. Thus, the observed experience-dependent plasticity of the entire NSeqs observed with single units was also observed with Ca²⁺ responses, relative to responses to RSeq (Fig. 9B, left). However, there was an overall decrease in response selectivity to both RSeqs and NSeqs not observed in single units. The above deviation from single-unit population behavior seen with Ca²⁺ imaging is because of the inherent differences in the two as stated previously (Fig. 8D). Thus, with the Ca²⁺ imaging data we consider changes with respect to RSeq in each condition. Unlike EXNs, SOM had significantly less selectivity for RSeq before exposure, which is abolished following exposure (Fig. 9B, middle; one-way ANOVA, p < 0.001). As with Thy-1 EXNs, exposure caused SOM to also have decreased selectivity to both NSeqs and RSeqs. On the contrary, PV neither showed a difference in selectivity to NSeqs compared with RSeqs both before and after exposure (paired t test, p > 0.05) nor did their overall selectivity vary with exposure (Fig. 9B, right, one-way ANOVA, p > 0.05, NSeqs; p > 0.05, RSeqs). Thus, among the two inhibitory neuron types tested, we hypothesize SOM to be involved in the observed experience-dependent plasticity. Also, overall SOM had the highest selectivity both before and after exposure and thus is likely capable of mediating changes to specific stimuli.

Optogenetic silencing of SOM combined with sequence presentation induces plasticity in selectivity to sequences but not sequence components

To test the hypothesis of the involvement of SOM in the above experience-dependent plasticity of entire sequences, we performed experiments in naive anesthetized female mice by deactivating the SOM neurons (Fig. 10A, schematic. As Awk_F-Bef_Expo and Anes_F-Bef-Expo mice did not show any difference in sequence selectivity between them both for NSeqs and RSeqs, use of anesthetized mice is justified. The mice used (P56–P90) expressed ArchT-EGFP specifically in SOM neurons, obtained by crossbreeding strain #021188 (Ai40D) and strain #013044 (SST-IRES-Cre; both from The Jackson Laboratory; Fig. 10A). First, we decided on the power level of the 589 nm laser for ArchT-EGFP activation to silence SOM. We used the highest power level at which the spontaneous activity after light off recovered to initial spontaneous activity before light on while producing an average increase in spontaneous rate with light on in the prestimulus period (n = 6 mice, 3 female, 3 male; n = 51 units, noise at five intensities, 55–95 db SPL; see above, Materials and Methods; Fig. 10B). There was no significant difference in spontaneous activity between before light on and after light off periods (Fig. 10C; paired t test, p > 0.05, bottom), with increased light-on spontaneous activity (Fig. 10B, green box, C, top and middle).

• Download figure
• Open in new tab
• Download powerpoint

Figure 10.

Reversible silencing of SOM paired with sequence presentation mimics plasticity in sequence selectivity without altering syllable selectivity. A, Schematic for optogenetic silencing of SOM using a laser of 589 nm (middle). Recordings were made in Au1 using a multielectrode array with an optical fiber in a mouse model expressing ArchT-EGFP in only SOM neurons. Right, Representative image of brain sections collected postrecording is shown with some SOM-ArchT-EGFP-positive neurons marked with white arrows. B, Determining laser power. For all such experiments the laser was turned on 100 ms prestimulus onset and turned off 100 ms stimulus offset (orange shading). Spontaneous or nonauditory-driven activity in three periods were used, OFF1, ON and OFF2, each 100 ms long and were as depicted. PSTH of a neuron is shown below the schematic. For sequences stimulus onset and offset were onset of the first syllable and offset of the last syllable of the sequence. C, Histograms of modulation of spontaneous activity of all cases show significant modulation of spontaneous spiking by light (left and middle, mean, red arrow), and the histogram to the right shows comparisons of spontaneous activity before and after light on. D, Representative example of a neuron with optogenetics for the sequences. E, Scatter plot to compare single syllable mean responses in NSeqs and RSeqs during SOM silencing paired with sequences (Anes_F-Opto) and after the period of pairing (Anes_F-Aft_Opto) as in Figures 4E and 6E. F, G, Scatter plots comparing mean response rates of common disyllables excluding occurrence in the first position for the Anes_F-Opto and Anes_F-Aft_Opto groups. Similar plot as in Figure 5C, with Anes_F-Bef_Expo and Anes_F-Aft_Expo (from Fig. 6F). H, Comparison of mean overall selectivity in NSeqs to that in RSeqs and comparisons across groups similar to Figure 7, C and D; *p < 0.05, **p < 0.01, ***p < 0.001; NS, Not significant.

We performed optogenetic silencing of SOM at 30 mW output power over a period encompassing presentation of the 20 sequences (both NSeqs and RSeqs, 100 ms before to 100 ms after the stimulus as in Fig. 10B; see above, Materials and Methods) for a total 100 presentations (five for each NSeq and RSeq) to test our hypothesis of involvement of SOM neurons (n = 5 female mice, n = 90 single units). Representative example of the PSTH of a neuron with optogenetics is shown in Figure 10D. We found that the relationship of responses of single syllables for NSeqs and RSeqs during SOM silencing (Anes_F-Opto; Fig. 10E, left) was similar to Anes_F-Bef_Expo (Fig. 10E, right) except the reduced responses of H in NSeqs compared with RSeqs were now the same (Fig. 4E, middle). We also probed for longer-term plastic changes induced by the pairing of SOM turning off (effectively disinhibiting EXNs to which they primarily project; Pi et al., 2013) with simultaneous sound stimulus (NSeqs and RSeqs) presentations. We found that the comparative responses to each of the single syllables in NSeqs and RSeqs following the above pairing (Anes_F-Aft_Opto; see above, Materials and Methods) were largely the same as Anes_F-Bef-Expo, Awk_F-Bef-Expo, and Anes_F-Aft_Expo (Figs. 4E, 6E, 10E). No distinct switch in preference between NSeqs and RSeqs was observed. Similar observations were made with transitions common to NSeqs and RSeqs in female mice before and after exposure (Fig. 5C; Bef_Expo; Fig. 6F, Aft_Expo) and during and after optogenetic silencing of SOM (Fig. 10F,G), with no distinct switching between NSeqs and RSeqs. We then tested the idea of plasticity of selectivity in sequences. The above experiments could also be considered to be pairing of sound stimuli and silencing of SOM neurons. For both groups, Anes_F-Opto and Anes_F-Aft_Opto, there was a significant increase in selectivity to NSeqs compared with the Anes_F-Bef-Expo group and was the same as that of the Anes_F-Aft_Expo group of mice (Fig. 10H, one-way ANOVA, p < 0.001, p < 0.001, p < 0.001). The mean selectivity to RSeq remained the same as the Anes_F-Bef-Expo group. In the subgroup of units from Anes_F-Bef_Expo mice in which SOM-silencing-based pairing was performed, the mean selectivity for NSeqs was the same as Anes_F-Bef_Expo; however, there was a small reduction in selectivity to RSeqs in the same population (median 4% lower), indicating an effect of rise in selectivity to RSeqs during and after pairing. Thus, SOM inhibition during sequence presentation is capable of inducing rapid plastic changes in coding of entire sequences without changing coding of the component syllables and disyllables. We thus hypothesize that during the activity-dependent plasticity occurring in naive females, VIP-mediated inhibition of SOM could underlie the plasticity in sequence selectivity.