Temporal association cortex plays a key role in auditory-driven maternal plasticity

Summary

Mother-infant bonding develops rapidly following parturition and is accompanied by changes in sensory perception and behavior. Here, we studied how ultrasonic vocalizations (USVs) are represented in the brain of mothers. Using a mouse line that allows temporally controlled genetic access to active neurons, we found that the temporal association cortex (TeA) in mothers exhibits robust USV responses. Rabies tracing from USV-responsive neurons revealed extensive subcortical and cortical inputs into TeA. A particularly dominant cortical source of inputs is primary auditory cortex (A1), suggesting strong A1-to-TeA connectivity. Chemogenetic silencing of USV-responsive neurons in TeA impaired auditory-driven maternal preference in a pup-retrieval assay. Furthermore, dense extracellular recordings from awake mice revealed changes of both single neurons’ and population responses to USVs in TeA; improving discriminability of pup calls in mothers as compared to naïve females. These data indicate that TeA plays a key role in encoding and perceiving pup cries during motherhood.

INTRODUCTION

Social interactions among animals are vital for survival and fitness of the species. Social communication cues exploit the full breadth of the senses. Visual gestures, modulation of speech and sensitivity of touch all carry useful information during inter-animal interaction. The sensitivity to these sensory cues is critical because they are used to interpret the social context and drive specific behavioral responses. In addition, an individual’s response, both its perception and action, relies strongly on past experience and internal physiological state, which change due to past and present social interactions (Burgess et al., 2018; Carcea and Froemke, 2013). Studying the neural circuits underlying perception and how these change with experience is an entry point for understanding social engagement (Chen and Hong, 2018).

Parenting is a complex and important set of social interactions required for ensuring survival of offspring. Parenting involves physiological- and experience-dependent changes, both of which prepare the animal for better caregiving towards its offspring (Dulac et al., 2014). Maternal behaviors and the neural circuits driving them have been studied for decades (Numan and Insel, 2003). Several subcortical regions and cell types therein have been causally linked to parental behaviors (Autry et al., 2019; Fang et al., 2018; Kohl et al., 2018; Wu et al., 2014). Here, we set out to study how the cortex, an important site of experience dependent plasticity (Feldman, 2009; Rothschild and Mizrahi, 2015), processes salient sensory cues in motherhood. We focused on auditory circuits processing pup cries.

In mice, vocalizations are important cues used by pups to communicate with their parents. In particular, ultrasonic vocalizations (USVs) have been shown to convey distress and drive the mothers to retrieve the vocalizing pup back to the nest (Ehret, 2005; Noirot, 1972). Like other sounds, pup calls are processed by the auditory system. The primary auditory cortex (A1), that has been shown to undergo plasticity following motherhood (Cohen et al., 2011; Galindo-Leon et al., 2009; Liu et al., 2006; Liu and Schreiner, 2007; Marlin et al., 2015; Shepard et al., 2016; Tasaka et al., 2018). However, it is not known how maternal plasticity affects processing in higher cortical regions and to what extent other cortical regions play a role in maternal behavior. Therefore, we asked whether and how are pup cries encoded along the auditory cortical hierarchy, focusing on USVs in mothers.

To study cortical circuits involved in processing pup USVs, we combined mouse genetics, monosynaptic rabies tracing, and chemogenetic silencing of neurons during behavior. We found that a neural circuit from A1 to the temporal association cortex (TeA) is intimately involved in processing USVs, and that this circuit supports maternal preference to pup calls. Using electrophysiology in awake mice we measured single neuron responses in A1 and TeA, describing how USVs are encoded in those regions. We propose that the A1-to-TeA circuit plays an important role in maternal recognition of auditory cues such as pup cries.

RESULTS

TeA is activated by pup USVs

To determine which cortical regions respond to pup ultrasonic vocalizations (USVs) in mothers we utilized ‘targeted recombination in active populations’ (TRAP) that allows permanent genetic tagging by a specific experience ((Allen et al., 2017; DeNardo et al., 2019), see Methods for more details). We used a previously calibrated TRAP driver system with mice expressing histone-BFP as a reporter and tTA2 for additional manipulation (Figure 1A) (Tasaka et al., 2018). Playback sounds of recorded pup USVs were used to induce tagging of USV-responsive neurons, whole brains were sliced, and their cortices were analyzed for expression (Figure 1B; USV-TRAP). Control mice received no sound stimuli but were otherwise identical (Figure 1B; NS-TRAP). Focusing on cortical induction, we found a 2.6 r 0.3 and 3.1 r 0.2 fold increase in A1 and TeA of mothers stimulated with USVs as compared to no-sound controls, respectively (Figure 1C–1D; NS-TRAP, N = 6 mice; USV-TRAP, N = 6 mice; values of absolute density are shown in Figure S1A). Overrepresentation of USV-TRAP cells was concentrated in the central-posterior region of A1 and TeA (Figure S1B). There were no interhemispheric differences in the number of USV-TRAP cells (Figure S1C). However, given that left asymmetry was reported in other measures by other studies (Levy et al., 2019; Marlin et al., 2015), we focused our measurements (below) on the left hemisphere. USVs recruited TRAPed neurons in several other, putative downstream, brain regions such as ventral and lateral orbitofrontal cortex (Figure S1D–S1F, VO and LO, Figure S1D–S1E; NS-TRAP, N = 6 mice; USV-TRAP, N = 9 mice; data-set of TRAP1 x TB mice). Importantly, as compared to all other cortical regions tested, USV stimuli tagged relatively more neurons in TeA of mothers as compared to naives (Figure S1E–S1F). Additionally, we tested other pup sounds like wriggling calls (WC), which elicit maternal responses that are distinct from USV-mediated behaviors (Ehret, 2005). Induction rate by WC was similar between mothers and naives in all tested regions (Figure S1F). Based on these data we hypothesized that TeA may be an important brain region for processing USVs in mothers.

Figure 1. USVs recruit neurons in TeA.

(A) Schematic of the TRAP x TB system. (B) The experimental protocol used for TRAPing in mothers. Top - no stimulation (NS-TRAP); Bottom- USV stimulation (USV-TRAP). (C) Representative fluorescent micrographs of coronal brain slices stained for TRAPed cells (Myc). Slices are from a region containing A1 and TeA, corresponding to Bregma – 2.92 mm in the Brain Atlas (Paxinos and Franklin, 2004). Scale bar, 500 Pm. (D) Quantification of the fold induction of TRAPed cells in A1 and TeA as compared to S1 (relative density), normalized to the ‘No Stim (NS-TRAP)’ condition (mean ± SEM; NS-TRAP, N = 6 mice; USV-TRAP, N = 6 mice; **, p < 0.01, Mann-Whitney U-test with Bonferroni correction, all statistical tests and results are listed in Table S3).

The auditory TeA is located just ventral to the secondary auditory cortex and dorsal to the rhinal fissure (Paxinos and Franklin, 2004). Knowledge about the connectivity or function of TeA in any species is scarce. In contrast, A1 has been extensively studied in numerous species including mice (Budinger and Scheich, 2009; Rothschild and Mizrahi, 2015; Theunissen and Elie, 2014). We, therefore, studied the anatomy and physiology of TeA with reference to the well-studied A1.

The long-range presynaptic landscape of TeA

To reveal the anatomical presynaptic input of USV-responsive neurons in TeA vs. A1, we combined monosynaptic trans-synaptic rabies tracing with TRAP (TRAP-rabies; TRAP1 and TRAP2 were used for tracing from A1 and TeA, respectively). We sub-cloned a new version of an optimized rabies glycoprotein oG (Figure 2A, AAV-CAG-FLEx-oG) (Kim et al., 2016), and used it together with a TVA expressing virus 3–4 weeks before TRAPing (Figure 2B; TVA^66T and TVA were used for tracing from A1 and TeA, respectively). We then TRAPed injected animals with USVs and one week after TRAPing, injected pseudotyped G-deleted rabies virus into the exact same site as the AAV injections (either into A1 or TeA, Figure 2B–2C). We sacrificed the animals for histology five days later, verified in all mice that starter cells were predominantly restricted to the injected region (either A1 or TeA) and mapped the numbers and exact locations of GFP expressing cells across the whole brain (Figure 2D–2H). This allowed us to visualize the presynaptic landscape onto USV-TRAP cells in A1 or TeA.

Figure 2. TeA has a particularly rich presynaptic landscape.

(A) Overview of the TRAP-rabies monosynaptic retrograde tracing components. TRAP1 and TVA^66T were used for tracing from A1 and TRAP2 and TVA were used for tracing from TeA. (B) The experimental protocol for TRAP-rabies. (C) Schematic of the TRAP-rabies experiments from two different targets – A1 (left) and TeA (right). Yellow cells indicate the location of starter cells. Green cells indicate the monosynaptic inputs to the starter cells. (D) Representative fluorescent micrographs from the injection sites in A1 (left) or TeA (right). Zoomed-in micrographs (middle) show starter cells indicated by white arrows. Scale bar, 200 Pm. (E and F) Representative micrographs from select input regions into A1 (E) and TeA (F). Scale bar, 200 Pm. (G and H) A schematic map of selected-long range monosynaptic inputs into A1 (G) and TeA (H) (A1, N = 4 mice; TeA, N = 5 mice). Not all but regions received more than 0.2 % of total input fraction are shown (see text and Table S1 and S2 for full lists and abbreviations of regions). The colors indicate the proportion of each region out of the total inputs. Injection sites are shown by a red circle. The values of inputs into A1 from MGv and TeA from A1 are indicated separately as their value exceeds the color bar scale.

Qualitatively, the presynaptic landscape onto neurons in A1 was consistent with previous literature (Budinger and Scheich, 2009; Nelson and Mooney, 2016; Nelson et al., 2013). More specifically, inputs were detected in local circuits within auditory cortex and in 37 distant brain locations (Table S1, USV-TRAP). The majority of long range inputs to A1 were from the ventral part of the medial geniculate body (MGv) of the thalamus. Following the thalamus, the majority of inputs came from contralateral A1, TeA, basal forebrain (BF) and neighboring cortices (Figure 2E and 2G; N = 4 mice; for the full list see Table S1).

The input landscape of TeA was different than A1. Tracing the long-range inputs revealed ~100 brain sites sending inputs to TeA, which were both qualitatively and quantitatively distinct from the inputs into A1 (Figure 2F and 2H; N = 5 mice; for the full list see Table S2). At a first approximation, the tracing data from TeA are consistent with traditional tracing studies showing inputs from medial geniculate body, lateral amygdala (LA), hippocampal CA1, and several sensory cortices (Arszovszki et al., 2014; Doron and Ledoux, 2000; Vaudano et al., 1991; Zingg et al., 2014). A closer analysis of our rabies tracing data reveals new qualitative insights, as well as quantitative evaluation of the long-range input distributions. 45.3 % of the inputs into TeA arise directly from A1 (Figure 2H), most of which are from layer 5 neurons (Figure S2A, red). These results suggest that TeA is, foremost, an auditory processing station directly downstream of A1. Although TeA still receives a major fraction of its inputs from the auditory cortex, its inputs are far more diverse as compared to A1, particularly from subcortical regions. Monosynaptic inputs to TeA arise from a variety of cortical and sub-cortical areas (the top part of the list of inputs are shown in Figure 2H). Based on this input-landscape, TeA could be an integrative site of principally auditory information with other (presumably contextual) information arriving from other cortical and subcortical brain regions.

USV-responsive neurons in TeA but not A1 receive more long-range inputs relative to neighboring neurons

We next exploited TRAP-rabies to test whether the USV-responsive neurons in TeA that were TRAPed have a unique input signature. To do so, we traced USV-TRAP neurons from TeA and compared the input landscape to similar tracing from two other groups of mice - NS-TRAP and WC-TRAP. We verified that the number of starter cells and their targeting into TeA were similar between the groups (Figure 3A–3C; NS-TRAP, N = 4 mice; USV-TRAP, N = 5 mice; WC-TRAP, N = 5 mice), as was the layer distribution of starter cells in TeA (most of which were in layer 2/3 and layer 5; Figure S2B).

Figure 3. USV responsive neurons in TeA receive more long range inputs.

(A) Representative micrographs from TeA of TRAP-rabies injected mice. Left: NS-TRAP, Middle: USV-TRAP, Right: WC-TRAP. Scale bar, 200 Pm. (B-C) The total number of starter cells from NS-, USV-, and WC-TRAP mice are not different (B) (NS-TRAP, N = 4 mice; USV-TRAP, N = 5 mice; WC-TRAP, N = 5 mice; ns, not significant, Mann-Whitney U test), and also spread equally in the TeA and adjacent regions (C) (ns, not significant, Mann-Whitney U test). (D) Convergence index (CI) from all regions projecting into TeA for the NS-, USV-, and WC-TRAP groups. USV-TRAP had larger CI than NS- and WC-TRAP (*, p < 0.05, Mann-Whitney U test with Bonferroni correction). (E) CI from A1 into TeA for indicated groups. USV-TRAP had the larger number of inputs from A1 to TeA than NS- and WC-TRAP (*, p < 0.05, Mann-Whitney U test with Bonferroni correction). (F) A differential input map from TeA (left) and A1 (right) comparing USV-TRAP neurons versus the inputs of NS-TRAP neurons. Color indicates the d-prime evaluated by comparing the CI of the indicated regions comparing NS- and USV-TRAP animals. White colored regions show roughly similar CIs (d’ < 1). d-prime was calculated as $\text{d}’=\frac{\left|{\mathit{\mu }}_{\mathit{U}\mathit{S}\mathit{V}-\mathit{T}\mathit{R}\mathit{A}\mathit{P}}-{\mathit{\mu }}_{\mathit{N}\mathit{S}-\mathit{T}\mathit{R}\mathit{A}\mathit{P}}\right|}{\sqrt{\frac{\mathbf{1}}{\mathbf{2}}\left({\mathit{\sigma }}_{\mathit{U}\mathit{S}\mathit{V}-\mathit{T}\mathit{R}\mathit{A}\mathit{P}}{}^{\mathbf{2}}+{\mathit{\sigma }}_{\mathit{N}\mathit{S}-\mathit{T}\mathit{R}\mathit{A}\mathit{P}}{}^{\mathbf{2}}\right)}}$. See Table S2 for abbreviations of indicated regions.

To quantify the long-range inputs onto TRAPed neurons we calculated a Convergence Index (CI) defined as the number of input neurons in a brain region per starter cell. USV-TRAP cells had significantly higher CI values as compared to NS-TRAP or WC-TRAP neurons when all input regions were considered (Figure 3D). Given that A1 is the major input region to TeA and known to respond to USVs, we hypothesized it will be a major source of input to the USV-TRAP cells. Total inputs from A1 to USV-TRAP neurons was significantly higher as compared to the NS-TRAP and WC-TRAP groups (>3 fold increase on average; Figure 3E). We assumed that the differential distribution of inputs to TeA may arise from other regions beyond A1. However, due to the larger number of brain sites that were identified and the multiple groups in the experimental design, none withstood significance after statistical correction for multiple comparisons of all regions (Figure S3A shows CI values of selected regions). To describe the relative magnitude of differences in CI between regions, we calculated d’ values from each brain region separately (using $\text{d}’=\frac{\left|{\mu }_{USV-TRAP}-{\mu }_{NS-TRAP}\right|}{\sqrt{\frac{1}{2}\left({\sigma }_{USV-TRAP}{}^2+{\sigma }_{NS-TRAP}{}^2\right)}}$; Figures 3F and S3B show the top part of the list that is different for USV-TRAP vs. NS-TRAP inputs; for the full lists see Table S2). Based on this analysis we conclude that differential inputs onto USV-TRAP neurons could potentially arise from cortical and sub-cortical regions alike (Figure 3F). Notably, differences between USV- and WC-TRAP input landscapes were largely similar to those between USV- and NS-TRAP (compare Figure 3F (TeA) and Figure S3B). Only few regions showed potential differences between WC- and NS-TRAP (Figure S3C). These results suggest that the USV responsive cells receives unique functional connectivity as compared to neighboring neurons.

Importantly, we conducted a similar experiment from starter cells in A1, comparing USV-TRAP to NS-TRAP mouse groups. In A1, we found no differences between the CI of USV-TRAP neurons and NS-TRAP neurons from any brain region (Figure 3F - A1; Figure S3D–S3H). Thus, USV-responsive neurons in TeA, but not A1, receive a particularly rich set of presynaptic inputs from long range sources.

USV-responsive neurons in TeA receive preferential input from USV-responsive neurons in A1

The combination of TRAP and rabies allows us to test functional connectivity among brain areas. We next asked whether neurons in A1 and TeA form like-to-like connections. This question would be a long-range connectivity equivalent to the well-documented like-to-like functional connectivity observed in local cortical circuits (Ko et al., 2011; Yoshimura et al., 2005). To answer this question, we analyzed TRAP-tracing data from four groups of mice – NS-TRAP, USV-TRAP, WC-TRAP, and an additional experimental group “NBN-TRAP”. The NBN (Narrow Band Noise, Figure S4A) group was added as another control to USVs. Specifically, the NBN control tests connectivity among neurons responding to sounds in the spectral range of USVs, but using a sound that lacks behavioral saliency.

We TRAPed and traced neurons from TeA using different sounds and then counted how many of the input cells from A1 are also TRAPed. We calculated the likelihood of connection between like-to-like neurons in A1 and TeA (Figure 4, NS-TRAP, N = 4 mice; USV-TRAP, N = 5 mice; WC-TRAP, N = 5 mice; NBN-TRAP, N = 4 mice). Specifically, we measured the density of TRAP-only cells, input-only cells, TRAP-input cells (double-labeled cells). We estimated total cell density in A1 as 109,730 cells/mm³ (Herculano-Houzel et al., 2013; Keller et al., 2018). Using these values, we computed the probability of finding a double-labeled neuron under assumption of independence. We used this probability to estimate the expected number of double-labeled neurons as follows:

$$\begin{gathered} (Expected\hspace{0.17em}number\hspace{0.17em}of\hspace{0.17em}double-labeled\hspace{0.17em}cells=\left[Cell\hspace{0.17em}density\hspace{0.17em}in\hspace{0.17em}A1\right]\hspace{0.17em}\times \\ \left[Volume\hspace{0.17em}of\hspace{0.17em}ROI\right]\times \mathrm{Pr}\left[TRAP\hspace{0.17em}cells\right]\times \mathrm{Pr}\left[Input\hspace{0.17em}cells\right]). \end{gathered}$$

Then, we compared the observed number of double-labeled neurons with the expected number from a Poisson distribution for each group (Figure 4C, black curve, λ = expected number of double− labeled cells). The observed number of double-labeled neurons was consistent with the assumption of independence in the NS-TRAP and NBNgroups (Figure 4C, NS-TRAP [blue, p = 0.448] and NBN-TRAP [cyan, p = 0.912], extreme upper tail probability computed by the Poisson cumulative distribution function). On the other hand, the number of double-labeled neurons was significantly larger than expected in the USV-TRAP and WC-TRAP group (Figure 4C, USV-TRAP [red, p = 0.0015] and WC-TRAP [green, p = 0.038], extreme upper tail probability computed by the Poisson cumulative distribution function). These results show that neurons responding to noise bursts (i.e. NBN-TRAP) or randomly TRAPed populations (i.e. NS-TRAP) are not preferentially connected but neurons responding to natural sounds are. Thus, USV-responsive neurons in A1 are preferentially connected to USV-responsive neurons in TeA, forming a functional subnetwork from A1-to-TeA.

Figure 4. USV responsive neurons in TeA receive preferential input from USV-responsive neurons in A1.

(A) Schematic of the analysis used to quantify the number of presynaptic cells onto TeA TRAPed neurons that are also TRAPed in A1 (yellow cells). (B) Representative micrographs from the two TRAP-rabies experiments, injected in TeA and analyzed in A1. In A1, TRAP cells are in red (stained with anti-Myc) and green cells are presynaptic to TeA. Scale bars, 200 Pm. (C) Quantitative analysis of functional connectivity between A1 and TeA in the four experimental groups. Black curve shows the Poisson distribution estimated from the expected number of double-labeled cells as lambda. The dotted color line and an arrow indicate the observed number of double-labeled cells for each group (NS-TRAP, n = 12 cells; USV-TRAP, n = 132 cells; WC-TRAP, n = 74 cells; NBN-TRAP, n = 5 cells). USV- and WC-TRAP but not NS- and NBN-TRAP neurons in A1 have significantly higher probability to connect to TRAPed neurons in TeA (NS-TRAP, N = 4 mice; USV-TRAP, N = 5 mice; WC-TRAP, N = 5 mice; NBN-TRAP, N = 4 mice; *, p < 0.05, **, p < 0.01, extreme upper tail probability computed by a Poisson cumulative distribution function).

USV-TRAP neurons are causally related to USV preference behavior by mothers

We next asked to what extent do USV-responsive neurons in TeA participate in perceiving maternally-salient information like pup USVs? To examine whether USV-TRAP cells in TeA are causally involved in an auditory-driven maternal behavior, we designed a behavioral paradigm based on a two-alternative forced-choice preference task of pup retrieval (Figure 5A). In short, each mother was allowed to retrieve live pups from two chambers connected to her home-cage. After retrieving three pups consecutively (one from each chamber and one from the home-cage), the mother was tested on a USV-preference test as follows. In one chamber we played back USVs as the salient stimulus, and in the other chamber we played NBN as the non-salient stimulus. Notably, each “syllable” in the NBN encompassed a similar frequency band as the USVs (50–75 kHz) and the full-length stimulus had identical temporal and amplitude envelope properties (Figure S4A). Pup USVs and NBN were played simultaneously, one in each chamber, and their location changed pseudo-randomly on different trials. Pups were not present in the chambers during test trials. The first chamber that the mother decided to enter was scored as her choice on that trial. Each mother was tested on 8–14 trials (see Supplemental Movie 1 for one example trial).

Figure 5. TeA is causally related to auditory driven maternal preference.

(A) Schematic of the behavioral test. (B) Overview of the TRAP-chemogenetics components. (C) Experimental protocol for TRAP-chemogenetics to test auditory-driven maternal behavior. (D) Representative micrographs from DREADD injected mice bilaterally into TeA. Left: left hemisphere; Right: right hemisphere. Scale bar, 200 Pm. (E) Quantification of the maternal preference to a chamber playing USVs over NBN. Chance level is 50% (Control, N = 12 mice; USV-TeA, N = 13 mice; WC-TeA, N = 11 mice; UPTTeA, N = 9 mice; **, p < 0.01; ns, not significant, Mann-Whitney U test with Bonferroni correction).

To test whether USV-TRAP neurons in TeA had a role in this behavior we combined TRAP and chemogenetics (TRAP-chemogenetics). We injected a Cre-dependent inhibitory DREADD encoding virus (AAV9-hSyn-DIO-hM4Di-mCherry) into TeA of TRAP mice bilaterally 3–4 weeks before TRAPing (Figure 5B–5D). Control mice were not injected with any virus before TRAPing. We injected CNO into both control and experimental groups 30 minutes prior to the behavioral test. Since mothers quickly learned that pups were not present during test trials and in order to compare all mothers at the same postnatal period, we compared different mice in different groups (i.e. paired analysis was not possible; Figure S4B). In a subset of mice (N = 4 mice TRAPed for USVs in A1, see below), we verified that CNO induced a decrease in the firing rates of the USV-responsive neurons (Figure S5).

Control mothers showed a preference to USVs over NBN as they entered the chamber that was playing USVs on 72.7% r 2.7% of the trials (Figure 5E, blue data points, N = 12 mice). Chemogenetic silencing USV-TRAP cells in TeA decreased the maternal preference to USVs to 59.9% r 3.2% of the trials (Figure 5E, red data points, N = 13 mice), suggesting that USV-TRAP neurons in mothers have a causal role in this task. To evaluate the extent of specificity of the USV responsive TRAPed cells in this task, we TRAPed two additional groups of mice with different sounds. A WC-TRAP group was used as control to sounds known to induce other maternal behaviors but encompass a lower spectral range. In addition, a group of mice was TRAPed with ultrasonic pure tones (UPT, see Figure S4A). UPTs are a series of ultrasonic pure tones that are in the spectral range of USVs but lack the rich spectrotemporal dynamics of the syllables. We used UPT instead of the NBN described above, because NBNs were used as the actual reference stimulus in the behavioral assay itself.

Mice in which WC-TRAP neurons were silenced behaved similarly to the controls (Figure 5E, green data points, N = 11 mice). Note that the WC-TRAP group was higher but, nevertheless, not significantly different when compared to silencing the USV-TRAP neurons (Figure 5E, p = 0.07). This suggests that WC-TRAP neurons have a weak contribution to the USV-driven effect seen in mothers. This may be expected since a fraction of WC-TRAP neurons are known to respond to USVs as well (Maor et al., 2016; Tasaka et al., 2018). Silencing UPT-TRAP neurons did not affect the maternal behavior (Figure 5E, cyan data points, N = 9 mice). To rule out the possibility that the behavioral effect is merely a reflection of the number of DREADD expressing cells, we tested for correlations between the number of DREADD expressing cells and behavioral effects. We found no such correlations across the data and in any of the experimental groups (Figure S6A). Finally, since viral injections are inherently leaky, we verified that most DREADD expressing neurons were predominantly in TeA (Figure S6B–S6C). Nevertheless, we still cannot rule out a small contribution of USV-TRAP neurons in other brain regions like AuV and Ect to the behavioral effect (Figure S6B–S6C).

Since USV responsive neurons in A1 are expected to drive USV neurons in TeA (Fig. 4), we also TRAPed and chemogenetically silenced neurons in A1 that respond to USVs (Figure S6D–S6G). Indeed, silencing USV-TRAP neurons in A1 of mothers decreased the behavioral performance of mothers to levels similar to silencing USV-TRAP neurons in TeA (Fig. S6E). Taken together, these results show that USV-responsive cells in TeA and also A1 are causally related to maternal preference towards pup USVs.

Single unit responses to USVs in A1 and TeA following motherhood

To this end our experiments used TRAP and its variants as the main method suggesting that TeA is involved in encoding USVs. To assess this general conclusion using an independent and unbiased method, we next used blind electrophysiological recordings in TeA with reference to A1. We recorded spiking activity simultaneously in A1 and TeA using the high-density microelectrode arrays, Neuropixels probes (Figure 6A and S7A–S7D) (Jun et al., 2017). All recordings were carried out in awake, head restrained, animals. To validate the recording location, after the experiments we reconstructed the trajectories of the probes using DiI or DiO-coated electrodes, and annotated the exact regions we recorded from in high resolution (Figure S7A–S7C) (Shamash et al., 2018). Here, we only describe responses from well-isolated single units (SU) in response to USVs and for two control sounds in the same frequency range. We used both NBNs and UPTs as controls because NBN is the actual reference sound we played during behavior, and UPTs is a sound in the same ultrasonic range.

Figure 6. Distinct plasticity of USV coding at single cell level in A1 and TeA.

(A) Schematic of the Neuropixels probe trajectory. Scale bar, 1 mm. (B-C) Evoked vs spontaneous spike rates of single neurons (mean ± SEM) in response to USV, NBN, or UPT in A1 (B) and TeA (C) (naïve-A1, n = 238 cells, mother-A1, n = 182 cells, naïve-TeA, n = 69 cells, mother-TeA, n = 132 cells; *, p < 0.05; **, p < 0.01; ***, p < 0.001; ns, not significant, Mann-Whitney U-test with Bonferroni correction). (D) Raster and PSTHs from three neurons, which show different preference indices (Top/middle: raster plots in response to USV/NBN; Bottom: PSTHs). 3 different sound intensities were presented 20 times for each and thus each raster shows 60 trials in total. The gray bars indicate the position of syllables in the sound. (E-F) Plots of the preference index of USV over NBN (E) and USV over UPT (F) in A1 and TeA (*, p < 0.05, **, p < 0.01; ns, not significant, Mann-Whitney U-test).

We recorded from 5 mothers (13 probe penetrations) and 5 naïve females (11 probe penetrations), for a total of 661 SUs, out of which 40 SUs did not show response to any sound and were not included in the analysis (see Methods). Thus, we included 621 SUs in our dataset (Naïve: A1, n = 238 SU and TeA, n = 69 SU; Mothers: A1, n = 182 SU and TeA, n = 132 SU). The penetration sites were similar between naives and mothers as shown by the full reconstruction of all probes (Figure S7B). Further, the distribution of best frequencies (BF) in response to pure tones between naives and mothers was similar (Figure S7E). Response latencies to pure tones was higher in TeA consistent with the hierarchal relationship suggested by the rabies tracing (Figure S7F). As compared to A1 in naïve mice, the mean spontaneous firing rates were higher in mothers, and mean evoked responses were higher for NBN and UPT (Figure 6B). In TeA, the mean spontaneous firing rates were higher in mothers, and the mean evoked responses were higher only in response to USVs (Figure 6C).

To test whether SU’s respond preferentially to USVs versus NBN or UPT we calculated a Preference Index (PI) for each neuron as the difference between its firing rate to USV versus either NBN or UPT. For example, the PI of USV vs NBN was calculated as $PI=\frac{\text{firing}\hspace{0.17em}{\text{rate}}^{USV}-\text{firing}\hspace{0.17em}{\text{rate}}^{NBN}}{\text{firing}\hspace{0.17em}{\text{rate}}^{USV}+\text{firing}\hspace{0.17em}{\text{rate}}^{NBN}}]$. Neurons with a positive PI prefer USV, a negative PI prefer NBN, and neurons with PI ≈ 0 respond equally to the two stimuli (see Figure 6D for three examples). In A1 of naïve mice, SUs had an innate preference to USVs over NBN, which was slightly decreased in mothers (Figure 6E, top). In TeA, the transition to motherhood was accompanied by an opposite effect – on average, SUs did not prefer one stimulus over another in naives but shifted to preferring USVs over NBNs in mothers (Figure 6E, bottom). Similar changes were found when comparing neuronal responses to USV vs UPT (Figure 6F). Thus, neurons in A1 of mothers seem to generalize similar sounds (i.e. PIs shift towards 0) and in TeA a small, yet significant, discrimination develops towards USVs (i.e. PI’s shift towards positive values).

Population responses in TeA improve USV discrimination in mothers

Cortical coding is likely carried out by populations of neurons. Thus, we analyzed the information from multiple simultaneously recorded neurons (Figure 7A). We first calculated pairwise PSTH correlations (a measure similar to signal correlation, see Methods and (Maor et al., 2016)), and pairwise noise correlations (NC) (Rothschild et al., 2010). PSTH correlations describe the average similarity among neurons in responses to USVs, and pairwise NCs describe trial-to-trial variability around the mean, which has been hypothesized to reflect shared inputs (Averbeck et al., 2006). Pairwise correlations between naives and mothers differed only in TeA (Figure 7B–7C). PSTH correlation in TeA of mothers decreased, becoming uncorrelated, as compared to naives (Figure 7B; red curve is shifted to the left and not different from shuffled data). The distribution of pairwise NC did not differ in A1 but completely changed in TeA (Figure 7C). NC in TeA of naives had a wide distribution, with both low and high values. After the transition to motherhood, these positive and negative correlations decreased towards zero (Figure 7C; compare slope of the blue and red curves). Both of these results suggest that population responses in TeA become sparser – i.e., representations diverge and trial to trial correlations shift towards zero.

Figure 7. Distinct plasticity of population coding in A1 and TeA.

(A) Raster plots from one Neuropixels probe recording simultaneously A1 and TeA of a mother (Cell#1–19; A1, Cell#20–33; TeA). (B) Cumulative plots of pairwise PSTH correlation within regions (‘A1-A1’ naives, n = 3302 pairs [r = 0.10 ± 0.20]; mothers, 1613 pairs [r = 0.09 ± 0.21]. ‘TeA-TeA’ naives, n = 409 pairs [r = 0.17 ± 0.21]; mothers, 844 pairs [r = 0.12 ± 0.19]). Dotted lines indicate shuffled PSTHs (**, p < 0.01; ns, not significant, Kolmogorov-Smirnov test). All values listed here are [mean ± SD]. (C) Cumulative plots of pairwise noise correlation (A1-A1: naives, n = 3302 pairs [r = 0.03 ± 0.08]; mothers: 1613 pairs [r = 0.04 ± 0.08]. TeA-TeA: naives, n = 409 pairs [r = 0.04 ± 0.33]; mothers, 844 pairs [r = 0.04 ± 0.21]). Dotted lines indicate shuffled PSTHs (***, p < 0.001; ns, not significant, Kolmogorov-Smirnov test). All values listed here are [mean ± SD]. (D) Trajectories of PCA components from representative animals. (E) Mean eucledian distance of PCA components between USV and NBN (red, mothers; blue, naives). (F) Cumulative distance of PCA components between USV and NBN. (G) Classification performance of a SVM decoder trained by the dataset of A1 neurons. The decoder was tested for its accuracy to differentiate USV from NBN or UPT. The decoder was trained with first five syllables. The graphs indicate the mean value of accuracy across 1000 iterations (‘USV vs NBN’ mothers, 88.0 %; naives, 85.5 %, ‘USV vs UPT’ mothers, 88.6 %; naives, 91.3 %) (H) Same as (G) but in TeA (‘USV vs NBN’ mothers, 73.9 %; naives, 65.3 %, ‘USV vs UPT’ mothers, 89.8 %; naives, 61.7 %).

To analyze our data beyond pairs, we calculated response-vectors from all neurons that were recorded simultaneously in each mouse. Population responses from one naive female mouse and one mother are shown as three dimensional principal component analyses in Figure 7D, plotting responses to USV and NBN stimuli in A1 and TeA. To quantify discriminability between the sounds, we calculated Euclidean distances from the full n-dimensional space of the responses, where n is the number of neurons in each mouse. The mean distance between USV and NBN was increased in mothers, and significantly so in TeA (Figure 7E–7F). Lastly, we evaluated the information contained in the population by calculating classification accuracy of a decoder (using support vector machine). Decoding USVs from other stimuli in A1 was similarly high in naives and mothers (Figure 7G). Decoding USVs from the population of neurons in TeA improved in mothers when compared to both NBN and UPT (Figure 7H). Together, these results show a physiological signature that favors more efficient coding of USVs as compared to similar sounds in TeA of mothers.

DISCUSSION

Temporal Association cortex (TeA) is to a large extent an uncharted brain region. TeA is located below the somatosensory, auditory, and visual cortices, and subdivided to these three different sensory modalities according to its position along the rostro-caudal axis (Ramesh et al., 2018; Yamashita et al., 2018; Zingg et al., 2014). Only few studies focused on TeA in any modality. Imaging neurons in the visual TeA suggest that they form a mix of networks encoding either low level visual features or associative outcomes like predicted value (Ramesh et al., 2018). Here, we focused on the auditory TeA (TeA, in short), and studied its involvement in encoding USVs by testing natural sounds that are salient to mothers. TeA showed strong activation when mothers were exposed to USVs, as assessed by mouse genetic tools (TRAP). Combining TRAP with other methods like rabies tracing and chemogenetics showed that USV-TRAP neurons in TeA have a wide connectivity landscape and that these neurons are causally related to auditory-driven maternal behavior. Independent of TRAP, our electrophysiological recording show that neurons in TeA of mothers change in the way they encode USVs, promoting fine discrimination from similar sounds. Our work suggests that TeA plays a role in encoding pup cries during motherhood.

Functional anatomy of TeA

The anatomy of TeA has been studied using classic anterograde and retrograde labeling, revealing that TeA receives inputs from nearly the entire neocortex and projects back to nearly the entire neocortex (Zingg et al., 2014). We found that TeA receives direct inputs from diverse regions across the brain, and quantified these connection probabilities. Our study supports the notion that TeA is a higher-order auditory cortex because 45% of its long-range inputs arise from A1 and it also receives a high number of non-lemniscal thalamic inputs (Figure 2F–2H) (Doron and Ledoux, 2000; Lee, 2015; Romanski and LeDoux, 1993; Shi and Cassell, 1997; Vaudano et al., 1991). Yet, TeA also receives a large number of non-auditory inputs from almost all other sensory cortices as well as cognitive related regions such as the amygdala, orbitofrontal cortex, and hippocampus (Figure 2H, Table S2). Given this diverse input landscape, its reasonable to argue that TeA is not merely a high-order auditory cortex but rather a site that integrates sounds with other information. Since the input sources to TeA are involved in experience dependent plasticity and internal/emotional states (Tye, 2018), it is well positioned to integrate auditory information with past experience, emotional state, internal state and perhaps other sensory information; all of which are central to parenting.

TRAP-tracing allowed us to reveal that neurons in TeA activated by USVs receive higher numbers of long-range inputs as compared to their neighboring neurons responsive to other sounds (such as WC-TRAP) or random cells (NS-TRAP; Figure 3). The nature of these anatomical results (i.e. identification of >100 brain sites) and the relatively low number of absolute inputs from each region using rabies makes it difficult to pinpoint statistical differences from specific input sources. Nevertheless, at the top of the list of putative differential inputs we detected areas like basal forebrain, high order auditory thalamus, lateral amygdala, and hippocampus (Figure 3F and S3A, left graph). Future experiments will be needed to assess the contribution of each of these suspected pathways in isolation.

Neuromodulation is increasingly recognized as fundamental in coding of social information in the cortex, including during motherhood (Brunton and Russell, 2008; Froemke, 2015; Valtcheva and Froemke, 2018). For example, the cholinergic neurons of the basal forebrain have been shown to shape receptive field plasticity in A1 (Froemke, 2015; Froemke et al., 2007; Nelson and Mooney, 2016). Thus, how basal forebrain shapes USV responses in motherhood will be particularly interesting. Raphe nuclei too could be suspected to modulate USVs in motherhood (Figure 3F). Indeed, disruption of serotonergic neurons in raphe nuclei are known to cause impairment of maternal behaviors such as pup retrieval (Alenina et al., 2009; Lerch-Haner et al., 2008; Pawluski et al., 2019).

Other neuromodulatory systems such as dopamine, noradrenaline, and oxytocin have been reported as key to drive parental behaviors (Fang et al., 2018; Hansen et al., 1991; Marlin et al., 2015; Numan and Smith, 1984; Thomas and Palmiter, 1997). Unfortunately, not all neuromodulatory input sources are detected by rabies tracing because they are likely communicated by volumetric axonal release rather than by chemical synapses. For instance, we found very few or no input cells onto A1 from noradrenergic in locus coeruleus or oxytocinergic neurons in paraventricular hypothalamic nucleus, which are known to send inputs into A1 (Marlin et al., 2015; Schwarz et al., 2015). Particularly, oxytocin from the paraventricular hypothalamic nucleus has been shown to increase the response reliability to USV in A1 of mothers by coordinating excitatory/inhibitory balance (Marlin et al., 2015). The role of neuromodulation and its impact on TeA remains an open topic to explore.

The role of TeA in processing complex sounds

TeA was initially studied in the context of acquisition and fear memory using lesions. Although the lesion studies were non-specific to TeA they nevertheless highlighted its potential role in fear conditioning (Romanski and LeDoux, 1992). The first in vivo recordings from TeA were also in the context of auditory fear conditioning (Quirk et al., 1997). By comparing responses of neurons in TeA to those in the amygdala, TeA responses were found to be conditioned later and slower, suggesting that TeA encodes mnemonic or attentional information of the fear memory. Recently work using optogenetics, supported the notion that TeA (and AuV) are not particularly critical for processing complex auditory information per se but rather causally involved in memory formation (Dalmay et al., 2019). Thus, TeA’s central involvement in processing USVs could be a substrate for supporting the memory of pup cries by the parents.

Other evidence from recordings in TeA highlights its role simply as a high order auditory cortex. Recordings from TeA in anesthetized rats (also called suprarhinal auditory field) showed long onset latency responses, tolerance to distortion of vocalizations stimuli, implying its involvement in higher cognitive processing such as categorization (Carruthers et al., 2015; Polley et al., 2007). The sparser and decorrelated representation of USVs that we observe in TeA of mothers (Figure 7) has, at least potentially, several advantages for representing high-order information such as vocalizations (Olshausen and Field, 2004). For instance, sparse representation can be effective in pattern separation (Barlow, 1972; Olshausen and Field, 2004), or improving discrimination of complex sounds (Clemens et al., 2011), which indeed we observe in TeA. One computation to consider is the extent of discrimination vs generalization mothers have to perform. On the one hand, mothers may benefit from generalizing to any pup call, thereby increasing their detectability regardless of the fine details they heard (Tasaka et al. 2018). On the other hand, they could benefit from the ability to better discriminate their own pup’s call from other pups. Theoretical considerations have argued that the level of sparse coding, especially for neurons with mixed selectivity such as expected from TeA neurons, could balance the tradeoff between discrimination and generalization (Barak et al., 2013). Although both A1 and TeA were equally important to maternal choice towards pup calls (Figure 5 and S6), their distinct electrophysiological signature in mothers versus naives suggest that they perform different functions. Manipulating either region alone does not necessarily tap onto the same physiological mechanism that contributes to the final choice made by the mother. Although our experiments did not reveal any differences in the behavioral phenotype of silencing A1 versus silencing TeA (Figure 5 and S6), we cannot rule out that different mechanisms were perturbed in the different experiments as the connectivity of A1 and TeA are not equal. Further experiments will be required to tease apart the relevant TeA circuits and isolate whether the A1-to-TeA connection is the central pathway contributing to this behavior.

Methodological consideration

We used TRAP as a basis for several of our discoveries. While TRAP has numerous advantages, it is still noisy in some aspects. TRAP has poor temporal resolution due to the multiple and often slow time scales of the molecules and reagents that drive recombination in TRAPed neurons (e.g. c-Fos, 4-OHT; (DeNardo and Luo, 2017). Since the TRAP signal depends on transcription, there is an inherent gap of at least several days between tagging and manipulation. Further, neural networks are physiologically plastic, so not all TRAPed neurons are expected to maintain similar function when assayed days apart. New methodologies to read-from and immediately manipulate neurons as the animal behaves (Carrillo-Reid et al., 2019; Jennings et al., 2019) could be instrumental in teasing out fine functional correlation between USV-responsive neurons to maternal behavior. Nevertheless, the combination of TRAP with chemogenetics enabled the manipulation of several thousands of functionally tagged neurons in a relatively homogenous manner. Intersectional tools such as TRAP-rabies or TRAP-chemogenetics are valuable tools to elucidate the anatomy and physiology in any new brain region, as we demonstrate here for TeA.

STAR Methods

RESOURCE AVAILABILITY

Lead Contact

Further information about resources, reagents used, and requests for code should be directed to and will be fulfilled by the Lead Contact, Adi Mizrahi (Mizrahi.adi@mail.huji.ac.il).

Materials Availability

This study did not generate new unique reagents.

Data and Code Availability

The datasets/code generated during this study are available from the Lead Contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animals

All experimental procedures were approved by the Hebrew University Animal Care and Use Committee. Mice were kept in 12-hour light-dark cycle with access to food and water ad lib. TRAP1 (Fos-CreER^T2 [ref. (Guenthner et al., 2013)]) was obtained from the Jackson laboratories (Background strain C57BL/6). TRAP2 (Fos^2A-iCreER; background strain C57BL/6) and TB (background strain FVB) mice were generated as detailed in references (Allen et al., 2017; DeNardo et al., 2019; Tasaka et al., 2018). We used the following mouse strains: TRAP1;TB double heterozygous female mice (F1 hybrid of C57BL/6 and FVB strain, 8–15 weeks old) and TRAP2;TB double heterozygous female mice (F1 hybrid of C57BL/6 and FVB strain, 8–15 weeks old. TRAP1;TB mice were used for unbiased screening of regions activated by USVs (Figure S1D–F; same mice that we used for Tasaka et al. 2018) and rabies tracing from A1 (Figure 2 and Figure S3D–S3H). We used TRAP2;TB mice for all other experiments.

METHOD DETAILS

DNA Constructs

pAAV-CAG-FLEx-oG was constructed as described in a previous study (Vinograd et al., 2019). Briefly, oG was amplified by PCR from pAAV-EF1a-DIO-oG (Addgene Plasmid #74290; RRID: Addgene_74290; a gift from Edward Callaway) (Kim et al., 2016) and then subcloned into pAAV-CAG-FLEx-RG (Addgene Plasmid #48333; RRID: Addgene_48333) (Miyamichi et al., 2013), digested with SalI and AscI. pAAV-CAG-FLEx- TC^66T and pAAV-CAG-FLEx-TC were gifts from Liqun Luo (Addgene Plasmid #48331 and #48332; RRID: Addgene_48331 and Addgene_48332) (Miyamichi et al., 2013). pAAV-hSyn-DIO-hM4D(Gi)-mCherry was a gift from Bryan Roth (Addgene plasmid #50475; http://n2t.net/addgene:50475; RRID:Addgene_50475)

Viral Procedure

AAV vectors containing CAG-FLEx-TC (2 × 10¹³ genomic copies per ml), CAG-FLEx-TC^66T (2 × 10¹² genomic copies per ml), CAG-FLEx-oG (1 × 10¹² genomic copies per ml), and hSyn-DIO-hM4D(Gi)-mCherry (9 × 10¹² genomic copies per ml) were produced by ELSC vector core facility (EVCF). For trans-synaptic tracing from TRAP cells, 0.1 Pl of mixture of AAV2-CAG-FLEx-TC or AAV2-CAG-FLEx-TC^66T and AAV2-CAG-FLEx-oG was stereotaxically injected into the left auditory cortex (coordinates relative to Bregma: anterior 2.5mm, lateral 4.2mm, depth 1.85mm at 20 degrees tilt from a vertical position) or into left TeA (anterior 2.7mm from Bregma, lateral 1.4 mm from the curvature of the bone at the boundary between the muscle and dorsal part of skull at the lateral edge) by using Nanoject 2 (Drummond Scientific). AAV2-CAG-FLEx-TC and AAV2-CAG-FLEx-TC^66T were used for rabies tracing from A1 and TeA, respectively. EnvA-Pseudotyped Rabies’G (2 × 10¹¹ infectious particles per ml) was produced following the established protocol (Osakada and Callaway, 2013; Wickersham et al., 2007). For the behavioral assay in Figure 5, 0.2 Pl of AAV9-hSyn-DIO-hM4D(Gi)-mCherry was injected bilaterally to the auditory cortex or TeA (coordinates were the same as above for both auditory cortex and TeA).

Drug Preparation

4-hydroxytamoxifen (Sigma-Aldrich, Cat#H6278) was dissolved to 20 mg/mL in ethanol by shaking at 37 °C for 15 min, then aliquoted and stored at −20 °C for up to a month. Before use, 4-OHT was re-dissolved in ethanol by shaking at 37 °C for 15 min. Corn oil (Sigma-Aldrich) was added to give a final concentration of 5 mg/mL, and the ethanol was evaporated by vacuum under centrifugation. The final 5 mg/mL 4-OHT solutions were stored at 4 °C before use (for no more than 24 hours). All injections were made intraperitoneally (i.p.). 50 mg/kg 4-OHT was delivered for all TRAP experiments except for Figure S1D–F. In Figure S1D–F, 15 or 25 mg/kg 4-OHT were delivered for the experimental groups of naives or mothers, respectively.

Auditory Stimulation for TRAP

Mice were moved to a new cage more than 12 hours before injection of 4-OHT. Then the cage was placed in a sound proof-chamber (IAC Acoustics). Mothers were always cohoused with their pups during TRAP procedure. Sound stimulation and injection of 4-OHT were conducted in a sound proof-chamber. The cage was moved back to our animal facility more than 12 hours after injection of 4-OHT. During TRAP procedure, we rarely saw pup retrieval behaviors because the pups stayed mostly in their nest. The TRAPing sound was the dominant sound in the soundscape and the sound energy was nearly identical between the groups of naïves and mothers. This suggests that any other sounds that animals produce in the cage during the TRAPing session is negligible. Additionally, TRAPing sounds did not induce any obvious observable behaviors. TRAPing is thus described as playing sounds to an awake passively-listening animal in its home cage. Sound stimuli were custom-generated in MATLAB (MathWorks) and delivered by a free field speaker (EC1, Tucker-Davis Technologies) placed above the home cage. USVs, ultrasonic pure tones (UPT), Narrow-Band-Noise (NBN) or WCs were delivered for 1 hr (total of 900 repetition of 3 seconds duration and ISI of 1 second). USVs (from the F2 hybrid of C57BL/6 and FVB strain) and WCs (from the C57BL/6 strain) were recorded with a one-quarter inch microphone (Brüel & Kjær) from P4–P5 pups. Additional details can be found in our previous paper (Tasaka et al., 2018). USVs contained 10 syllables. To synthesize NBN, we produced a narrow band noise at 50–75 kHz band along the time window for each syllable in USVs. The amplitude of the NBN was matched to fit the recorded USVs. To synthesize UPT, we generated 10 pure tones whose frequencies range is between 50 and 75 kHz in the log space. The sequence of 10 pure tones were randomized and aligned along the time window for each syllable in USVs. The amplitude was fit to the maximum of USVs.

Histology

Mice were given an overdose of Pental and were perfused transcardially with phosphate buffered saline (PBS) followed by 4% paraformaldehyde (PFA) in PBS. Brains were post-fixed for 12–24 hours in 4% PFA in PBS and then cryoprotected for >24 hrs in 30% sucrose in PBS. Coronal slices were made using a freezing microtome (Leica SM 2000R) and preserved in PBS. Free floating slices were then incubated in the following solutions with gentle agitation at room temperature: 2 hrs in blocking solution (5% heat inactivated goat serum, 0.4% Triton-X100 in PBS); Overnight at room temperature in primary antibody 1:1000 mouse anti-Myc (Santa Cruz, Cat #sc-40, RRID: AB_627268) in blocking solution; 2–3 hrs in secondary antibody 1:500 goat anti-mouse-IgG Cy3-conjugated or goat anti-mouse-IgG Alexa647-conjugated (Jackson ImmunoResearch, Cat #115–165-166 [Cy3], RRID: AB_2338692 and #115–605-166 [Alexa647], RRID: AB_2338914) in blocking solution; 15 minutes in 2.5 Pg/mL of DAPI (Santa Cruz, Cat #sc-3598) in PBS. Sections were mounted on slides and cover-slipped with mounting media (Vectashield H-1000). Secondary antibodies were diluted at the final concentration 1:500 from 50% glycerol stocks. Sections were imaged using an Olympus IX-81 epifluorescent microscope with a 4x and 10x objective lens (0.16 and 0.3 NA; Olympus). Images were processed in Photoshop and adjusted for contrast and brightness in each channel.

Counting of TRAPed cells

For counting the number of TRAPed cells (Figure 1 and S1A–S1C), we imaged every 50 Pm 12 consecutive coronal sections through most of the rostral-caudal extent of A1 and TeA. For internal control, we imaged every 50 Pm 10 consecutive coronal sections of the forelimb region of S1. Since we found no inter-hemispheric differences between left and right of A1 and TeA (Tasaka et al., 2018), we selected only the right hemisphere of A1 and TeA from each mouse for quantitative analysis. One of the left or right hemisphere of S1 was chosen for quantitative analysis for internal control. The density of TRAPed cells in both regions were calculated from the sum of the number of TRAPed cells and the volume of the regions for each slice across all sections for each animal (Figure S1A–S1C). We normalized the density of TRAPed cells in A1 and TeA to the density of TRAPed cells in S1 for each mouse. The experimental group of ‘NS (No Stim) -TRAP’ was set to 1 (Figure 1D).

For counting the number of TRAPed cells in downstream regions of A1 throughout the brain (Figure S1D–S1F), 40 Pm coronal slices from the entire brain were immune-stained as described above. We imaged consecutively 12 coronal sections through most of the rostral-caudal extent of A1, TeA, and the forelimb region of S1, 6 coronal sections through prefrontal cortex including prelimbic (PrL) and orbitofrontal cortex (medial [MO], ventral [VO], lateral [LO]). We counted cells for the right A1, right TeA, left or right S1 (either right or left, chosen randomly), left and right prefrontal cortex (data from left and right were merged) for each mouse. We normalized the density of TRAPed cells in each region to the density of TRAPed cells in S1 for each mouse. The experimental group of ‘NS-TRAP’ was set to 1.

Rabies transsynaptic tracing from TRAP cells

At first we injected AAVs (TC^66T into A1, and TC into TeA) and animals were mated several days after recovery. We waited for animals to be lactating mothers more than three weeks from AAV injection. Then we performed TRAP at pup age P2–4. A week after TRAP, we injected RV’G-GFP. We sacrificed animals 5 days after rabies injection.

For quantification of the rabies tracing data, we imaged consecutively 50 Pm coronal slices along the whole brain with 4x (for long range input cells detection) and 10x (for starter cells detection) objectives. Then, we counted starter and input cells manually using a custom-written MATLAB code. We manually registered the locations and layer distribution of starter cells and input cells using a custom-written MATLAB code. The starter cells were carefully determined by their overlapped expression of TVA-mCherry and GFP. Convergence Indices were calculated by division of the number of input cells over the number of starter cells. For regional registration, boundaries were based on the Paxinos and Franklin atlas (Paxinos and Franklin, 2004). To avoid potential bias due to possible local leakage of TVA (Miyamichi et al., 2013) and given the high density of GFP labeling at the injection site, we limited our quantitative analysis to long range inputs, and excluded the two immediate neighboring cortical regions, Ect and AuV.

For counting input⁺TRAP⁺ cells, we stained consecutively 50 Pm coronal slices with anti-Myc primary antibody and anti-mouse IgG Alexa-647 conjugated secondary antibody. Sections were additionally stained with DAPI. We counted the number of input⁺TRAP⁻/input⁻TRAP⁺/input⁺TRAP⁺ cells manually using a custom-written MATLAB code. To normalize across animals, we measured the volume of the counted regions in A1 of each animal. We counted all slices which contained starter cells in TeA. In Figure 4C, we calculated the expected number of input⁺TRAP⁺ cells for each group under the assumption that becoming a TRAP or an input cells are independent events as follows:

$$\begin{gathered} (Expected\hspace{0.17em}number\hspace{0.17em}of\hspace{0.17em}double-labeled\hspace{0.17em}cells=\left[cell\hspace{0.17em}density\hspace{0.17em}in\hspace{0.17em}A1\right]\hspace{0.17em}\times \\ \left[volume\hspace{0.17em}of\hspace{0.17em}ROI\right]\times \mathrm{Pr}\left[TRAP\hspace{0.17em}cells\right]\times \mathrm{Pr}\left[Input\hspace{0.17em}cells\right]. \end{gathered}$$

We estimated the cell density in A1 as 109,730 cells/mm³ (Herculano-Houzel et al., 2013; Keller et al., 2018). The probability of being TRAP or input cells was calculated as the following equation:

$$\begin{gathered} (\mathrm{Pr}\left[TRAP\right]=\frac{\left[observed\hspace{0.17em}number\hspace{0.17em}of\hspace{0.17em}TRAPed\hspace{0.17em}cells\right]}{\left[cell\hspace{0.17em}density\hspace{0.17em}in\hspace{0.17em}A1\right]\times \left[volume\hspace{0.17em}of\hspace{0.17em}ROI\right]},\mathrm{Pr}\left[Input\right]= \\ \frac{\left[observed\hspace{0.17em}number\hspace{0.17em}of\hspace{0.17em}input\hspace{0.17em}cells\right]}{\left[cell\hspace{0.17em}density\hspace{0.17em}in\hspace{0.17em}A1\right]\times \left[volume\hspace{0.17em}of\hspace{0.17em}ROI\right]}). \end{gathered}$$

We set the Poisson distribution with the sum of the expected number of double-labeled cells as lambda. Then, we calculated the complement of Poisson cumulative distribution function for the observed number of double-labeled cells in each group to estimate the upper tail probability.

Auditory driven maternal behavior - Two-alternative forced choice task

Animals were placed in a custom built Y-shape test cage (see Figure 5A, Supplemental movie 1) with standard wood chip bedding. Animals had full access to food and water during the entire period. In all groups, one week after TRAPing, we habituated animals in this cage for more than 12 hours before the test. The nest was located in the center of the back wall of the main cage (Figure 5A). In an effort to maintain the location of the nest, we transferred the nest built in old the cage and monitored whether it stayed in the same place. We initiated a trial by three consecutive pup retrievals. We placed one pup in left chamber and one pup in right chamber. After the two pups were retrieved, we placed a third pup in the center to reduce positional bias. Immediately after the retrieval of the third pup, we started to play sounds (USVs or NBN) from the speakers positioned towards the left and right chambers. USVs or NBN were played from either the left or right speaker pseudo-randomly. The stimuli played until the animal entered one room. We determine a full body entrance into one of chambers as a choice. When animals stopped approaching the speakers, we terminated the test. Otherwise, we repeated the test for up to 14 trials. 120 uL of CNO (0.5 mg/mL, Sigma; C0832) was injected with i.p. 30 min before the test. To show that silencing USV-TRAP neurons does not affect the intact perception of ultrasonic sounds, we trained TRAPed mice to discriminate between 50 kHz and 75 kHz pure tones. Mice were TRAPed as described above using USV’s (USV-TRAP) and then trained in an automated behavioral training system (see (Maor et al., 2020) for details). Following habituation, mice were trained on a go/no-go task design to lick in response to a target tone (a series of six 75 kHz pure tones, duration 100ms) in order to receive water reward and withhold licking in response to a non-target tone (a series of six 50 kHz pure tones, duration 100ms). Once mice reached a stable discrimination performance, they were injected with saline (day 11) and CNO (day 13). Their discriminability index (d’) was then calculated based on the probability of licking in response to the target and non-target tones. CNO injection did not induce any significant change in discrimination, as d’ values were similar between conditions and higher than the threshold for discrimination (mouse 1: d’ = 1.4 and 1.4 after saline and CNO application respectively; mouse 2: d’ = 1.5 and 1.2 after saline and CNO application respectively). These experiments (not shown) suggest that USV-TRAP mice are not impaired in general processing of pure tones in the ultrasound range.

For counting the number of DREADD (hM4Di-mCherry) expressing cells in Figure S6, we imaged consecutively 40 Pm slices with a 10x objective lens using an Olympus IX-81 epifluorescent microscope. We manually counted mCherry positive cells from every four section of both hemispheres in total 4 or 5 slices for each using custom-written MATLAB code.

Extracellular recordings using Neuropixels probe

For awake recording, we implanted a custom-made metal bar on the skull and performed small craniotomy on the left hemisphere 3–4 days prior to recording day (coordinates relative to Bregma: anterior 2.5mm, lateral 4.2mm [same coordinate as when we injected viruses into A1]). The craniotomy was protected by the wall of dental cement and covered by silicone elastomer (WPI; Kwik-Cast cat#KWIK-CAST). After the recovery, animals were head-fixed about 30 min to habituate to the recording setup 1–2 days before recording. On the day of the recording, animals were headfixed and the craniotomy was exposed. Then, a Neuropixels probe (imec, phase 3A) was inserted from the craniotomy and lowered down to a 3 mm depth with a 20 degree tilt from a vertical position. The depth of a probe was monitored by the manipulator (Scientifica PathchStar Micromanipulator). We painted the probe with a dye (DiI [Invitrogen cat#V22885] or DiO [Invitrogen cat#V22886]), to allow us to reconstruct the penetration sites in high resolution. To annotate the brain regions according to the probe position, the exact trajectories were reconstructed from consecutive coronal slices using an open source software (Figure S7) (Shamash et al., 2018).

We acquired all recordings using Neuropixels phase 3A probes (imec), a commercially available FPGA board (KC705, Xilinx) and a base-station connector (imec). Acquisition was performed in external reference mode (Ag/AgCl wire positioned on the skull) at a sampling rate of 30 kHz, with action potential band filtered between 0.3–10 kHz. action potential band gain was set to 500. All recordings from the same animal and position were concatenated, and automatically spike sorted using the ‘Kilosort’ open-source software (UCL; https://github.com/cortex-lab/KiloSort). Following automatic sorting, manual sorting was performed using the ‘Phy’ GUI (UCL; https://github.com/kwikteam/phy-contrib). During manual sorting, spike clusters were merged based on assessment of waveform similarity and the appearance of drift patterns. Finally, each spike cluster was assessed in criteria of waveform size, waveform consistency and the presence of short-latency inter-spike-intervals (ISIs). If and only if a cluster was satisfactory on all accounts, it was tagged as a single-unit (SU) corresponding to a single neuron, and was included in the analysis. Figure S7B and S7C were generated using the software Allen CCF developed by UCL cortex-lab based on the Allen brain atlas (see https://github.com/cortex-lab/allenCCF). Figure S7D was generated using the Neuropixels-utils software kit, developed by Dan O’shea (see https://github.com/djoshea/neuropixel-utils).

Auditory stimuli for electrophysiological recording with awake animals

For recordings from awake animals, USVs, NBN, and UTP were presented 20 times at 3 attenuation levels (0, 15, 30 dB attenuation). Sound attenuation was controlled by a programmable attenuator and delivered by the speaker driver (PA5 and ED1, Tucker-Davis Technologies). The highest intensity (0 dB attenuation) was adjusted to be the same as the sound intensity of the stimuli played during TRAPing. Inter stimulus interval was set to 1 second. We also collected data in response to pure tones (3–80 kHz, 30 frequencies) and another pup calls. Our analysis here focused only on responses to USV, NBN, and UPT. For recordings from anesthetized animals, USVs and NBN were presented 13–18 times at 3 attenuation levels (0, 15, 30 dB attenuation). The highest intensity (0 dB attenuation) was adjusted to be the same as the sound intensity of the stimuli played during TRAPing. Inter stimulus interval was set to 1 second. We also collected data in response to pure tones (380 kHz, 30 frequencies) and several pup calls. Our analysis here focused only on responses to USV and NBN.

Data Analyses - electrophysiology

We performed all data analyses and statistics using custom-written code in MATLAB. For Neuropixels recording data, we extract spiking activity using kilosort as described above. Spike times were then assigned to the local peaks of supra-threshold segments and rounded to the nearest millisecond. For each cell, we obtained a peri-stimulus time histogram (PSTH, binned at 1 ms). Evoked firing rates were extracted based on time windows of 100ms or the same size as the length of the syllable, because several syllables have a longer time window than 100ms. The values presented for evoked firing rate of each cell is the mean value of the evoked firing rate to all the syllables. The spontaneous firing rate of the cell was calculated based on the average of all 800 ms preceding each natural sounds stimulus presentation (corresponding to −1 to −0.2 sec in the PSTH plots in the figures). We determined significant response to each syllable by a Mann-Whitney U test of the firing rates based on the time window of the full-width-half-maximum (FWHM) of the PSTH compared to the cell’s spontaneous firing rate. The cells which had no significant response to all calls were excluded from our dataset (4, 5, 14 and 17 cells were excluded from A1-naive, A1-mother, TeA-naive and TeA-mother, respectively). Preference index in Figure 6 was calculated from firing rate (FR) in response to two stimuli for each cell using the following equation

$$\text{Preference}\hspace{0.17em}\text{Index}=\frac{F{R}_{USV}-F{R}_{NBN\hspace{0.17em}or\hspace{0.17em}UPT}}{F{R}_{USV}+F{R}_{NBN\hspace{0.17em}or\hspace{0.17em}UPT}}$$

Pairwise PSTH correlations in Figure 7B were calculated as Pearson correlations between PSTHs matrices, whose spike count was binned at 20 ms.

We calculated trial-to-trial noise correlation between cells within same animals recorded simultaneously (Figure 7C). We took a 200 ms time window following the syllables’ onset for calculating the single-trial responses of a neuron. Vectors containing the number of spike counts were binned at 40 ms and concatenated with all syllables. We subtracted the mean firing rate of the neuron to each sound stimulus from each single-trial response of a neuron, resulting in a vector of fluctuations around the mean responses to the different stimuli. We then used correlation coefficients between pairs of such vectors as estimates of the noise correlation between two neurons.

For best frequencies (BF) and latency analyses in Figure S7E and S7F, we first chose the subset of single units (SUs) showing significant onset excitatory responses. To identify onset responses, for each unit we looked at the 60 msec time window within a 150 msec interval beginning at stimulus onset in which the unit’s firing rate was maximal. We compared the firing rates in different trials during the chosen time window to firing rates within a 60 msec time window of maximal spontaneous firing rate (chosen from the 100 msec time window preceding each stimulus presentation). All SUs that exhibited a significant onset response (t-test, p < 0.05) were considered for the following analyses. To extract the CF of each unit, we first extracted the unit’s frequency response area (FRA). This was done by calculating the firing rate in response to each pair of (frequency × attenuation) during the unit’s evoked time window, and subtracting from it the spontaneous firing rate. The frequency which evoked the maximal mean firing rate across all attenuations was defined as the unit’s BF.

To calculate response latencies, we looked for the first 1 msec time bin in which the unit’s PSTH was 2 SDs larger than its baseline response. PSTH was calculated as the mean response across all trials (from 100 msec before stimulus onsets to 600 msec following), smoothed using a 13 msec time window. The baseline response and its SD were calculated as the mean and SD of the PSTH during the 100 msec before stimulus onset. The first time bin was of stimulus onset + 5 msec (as a baseline latency to avoid false positives immediately after stimulus onset), in which the response exceeded the baseline+2SDs was considered as the unit’s latency. Units which did not exhibit a response >baseline+2SD were excluded from this analysis.

Population data analysis

In order to estimate the ability of a network to discriminate between USVs and NBN, we started by calculating pairwise Euclidean distances and d primes for each group of simultaneously recorded units. Prior to the calculation, raster plots were binned into 50 msec bins. For each group of simultaneously recorded units, the response difference vector was calculated for each time bin according to:

$$\Delta \text{resp}\left(\text{t}\right)=\mu \left(USVresp\left(t\right)\right)-\mu \left(NBNresp\left(t\right)\right)$$

Where μ is the mean response of the population for the given stimulus, and is a (1 x n) vector, where n is the number of units in a given recording. t refers to a specific time bin. The Euclidean distance between trajectories was calculated as the norm of the difference vector divided by the square root of the number of units contributing to the distance measure (in order to normalize the distance).

$$\text{dist}\left(\text{t}\right)=\frac{\left|\Delta \text{resp}\left(\text{t}\right)\right|}{\sqrt{n}}$$

To calculate the mean distance for a given time bin across all the population of recorded units, we created a weighted average of the distances, scaled by the number of contributing units in each recording

$$\mu \left(dist,t\right)=\frac{{\sum }_{i=1}^{N}dis{t}_i\left(t\right)\cdot n_i}{{\sum }_{i=1}^N{n}_i}$$

Where N is the total number of recordings, and dist_i and n_i are the trajectory distance and number of units in a given recording. Cumulative distance was calculated by calculating the area under the curve up to bin t.

PCA analysis was used to demonstrate the divergence of responses to USV’s and NBN. For every population of simultaneously recorded units a time series of all trials for both stimuli was constructed to give a (n x (n-trials * time-bins-per-trial)) matrix (where n is the number of units in the population, and the number of trials is 80 as we use 40 trials from each stimulus types). The first 3 PCs were calculated for the entire time series, and their coefficients for every time bin were extracted. The mean trajectory in PCA space for each stimulus type was plotted on top of a 3D axis (Figure 7D).

Decoder Analysis

Classification of call identity using population activity was performed using a Support Vector Machine (SVM) classifier with a linear kernel. The input to a SVM consisted of the spike count of each neuron in the 100 ms or as the same length of the syllable (i.e. if the syllable has a longer duration than 100 ms) following syllable onset. Thus, a response vector to the stimulus is composed of each neuron’s spike count to each of the first syllables of a stimulus up to some number of syllables. In each run, 60% of the data (12 trials from the lowest sound intensity) were used for training and 40% of the trials (8 trials from the lowest sound intensity) were used as the test set for decoding accuracy. Each classifier was iterated 1000 times and a mean accuracy was calculated. Different combinations of trials were chosen randomly for each run. We increased the number of syllables cumulatively from 1 up to 10 for each sound stimulus to measure the accuracy at an increasing number of syllables. In Figure 7G–7H, we show the accuracy at 5 syllables because the decoder accuracy reached plateau around 5 syllables. We ran 4 classifiers – 2 different populations (A1 and TeA), comparing a USV stimulus against NBN and UPT.

Extracellular recordings from anesthetized animals during silencing neuronal activity

We performed extracellular recordings in anesthetized animals while silencing TRAPed cells in A1 by DREADDs (Figure S5). Those animals were anesthetized with an i.p. injection of ketamine and medetomidine (80 mg/kg and 0.65 mg/kg, respectively) and a subcutaneous injection of Carprofen (0.004 mg/g). The depth of anesthesia was assessed by monitoring the pinch withdrawal reflex and ketamine/medetomidine was added to maintain it. The animal’s rectal temperature was continuously monitored and maintained at 36°C ± 1°C. For recording, a custom made metal pin was glued to the skull using dental cement, and connected to a custom stage to allow precise positioning of the head relative to the speaker (facing the right ear). A small craniotomy was performed on the left hemisphere (coordinates relative to Bregma: anterior 2.5 mm, lateral 4.2 mm). Then, a Neuropixels probe was inserted from this craniotomy and lowered down to a 3 mm depth with a 20 degree tilt from a vertical position. We painted the probe with a dye (DiI, [Invitrogen cat#V22885] or DiO [Abcam #ab189809]), to allow us to reconstruct the penetration sites in high resolution. To silence neural activity by DREADDs, 120 uL of CNO (0.5 mg/mL, Sigma; C0832) was injected with i.p. after the first delivery of sound protocol (pre CNO session). 30 min after the injection of CNO, the same sound protocol was delivered (post CNO session). Spike sorting and data analysis were performed in the same way as described above.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical tests were performed using MATLAB. All tests were two-tailed. The sample size and statistical tests used are indicated in the figure legends. A full report of P-values and performed statistical tests were summarized in Table S3. To correct multiple comparisons, we used Bonferroni correction for less than five simultaneous comparisons and Benjamini-Hochberg correction for more than five simultaneous comparisons. Benjamini-Hochberg correction was based on the codes at MATLAB central file exchange (David Groppe (2020). fdr_bh (https://www.mathworks.com/matlabcentral/fileexchange/27418-fdr_bh), MATLAB Central File Exchange). Criteria for statistical significance is set at P < 0.05.

Supplementary Material

1

Video S1. Auditory driven maternal behavior, related to Figure 5

Table S3. All results from statistical tests performed in this study, related Figure 1, 3, 4, 5, 6 and 7.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-Myc	Santa Cruz	Cat #sc-40 RRID: AB_627268
anti-mouse donkey antibody conjugated with Cy3	Jackson ImmunoResearch	Cat #115-165-166 RRID: AB_2338692
anti-mouse donkey antibody conjugated with Alexa 647	Jackson ImmunoResearch	Cat #115-605-166 RRID: AB_2338914
Bacterial and Virus Strains
AAV2-CAG-FLEx-TCB	ELSC vector core facility	N/A
AAV2-CAG-FLEx-TC66T	ELSC vector core facility	N/A
AAV2-CAG-FLEx-oG	ELSC vector core facility	N/A
AAV9-hSyn-DIO-hM4Di-mCherry	ELSC vector core facility	N/A
EnvA-pseudotyped, G-deleted Rabies-eGFP	This study	N/A
Chemicals, Peptides, and Recombinant Proteins
DAPI	Santa Cruz	Cat #sc-3598
Clozapine N-oxide (CNO)	Sigma-Aldrich	Cat #C0832
Mounting Medium	Vectashield	Cat #H1000
4-OHT	Sigma-Aldrich	Cat #H6278
Corn oil	Sigma-Aldrich	Cat #C8267
DiI	Invitrogen	cat#V22885
DiO	Abcam Invitrogen	cat#ab189809 cat#V22886
Experimental Models: Cell Lines
B7GG	Osakada and Callaway, 2013; Wickersham et al., 2007	N/A
BHK-EnvA	Osakada and Callaway, 2013; Wickersham et al., 2007	N/A
HEK293T-TVA	Osakada and Callaway, 2013; Wickersham et al., 2007	N/A
Experimental Models: Organisms/Strains
Mouse: FosCreER (TRAP)	The Jackson Laboratory	JAX: 021882
Mouse: Fos2A-iCreER (TRAP2)	The Jackson Laboratory	JAX: 030323
Mouse: Igs2tm5(CAG-tTA2,-TagBFP)Luo (TB)	The Jackson Laboratory (Tasaka et al., 2018)	JAX: 031776
Recombinant DNA
pAAV-CAG-FLEx-TCB	Miyamichi et al., 2013	RRID:Addgene_483 31
pAAV-CAG-FLEx-TC66T	Miyamichi et al., 2013	RRID:Addgene_483 32
pAAV-CAG-FLEx-oG	This study; Vinograd et al., 2019	N/A
pAAV-hSyn-DIO-hM4Di-mCherry	A gift from Bryan Roth	RRID:Addgene_504 75
pAAV-EF1D-DIO-oG	A gift from Edward Callaway (Kim et al., 2016)	RRID: Addgene_74290
pAAV-CAG-FLEx-RG	Miyamichi et al., 2013	RRID: Addgene_48333
Software and Algorithms
MATLAB	MathWorks	http://www.mathworks.com/
Photoshop	Adobe	N/A
Illustrator	Adobe	N/A
Kilosort	UCL	https://github.com/cortex-lab/KiloSort
Phy	UCL	https://github.com/kwikteam/phy-contrib
Allen Institute’s Common Coordinate Framework (CCF)	Allen Institute for Brain Science	https://github.com/cortex-lab/allenCCF
Neuropixels-utils software kit	Dr. Daniel J O’shea	https://github.com/djoshea/neuropixel-utils
Other
Nanoject 2	Drummond Scientific	Cat # 3-000-204
Programmable attenuator	TDT	PA5
Speaker driver	TDT	ED1
Speaker	TDT	EC1