Abstract
Imitation plays a key role in the acquisition of speech and cultural behaviors. Studies suggest that social interaction facilitates imitative learning, indicating that neural circuits involved in social behaviors can also influence the process of imitation. Vocal imitation in juvenile songbirds serves as a valuable model to investigate this idea. Here, we explore the mechanisms of tutor–pupil social interaction and selective song learning in male zebra finches, with a particular focus on the amygdala, which can regulate social behaviors via its processing of values and emotions in mammals. When sequentially exposed to two tutors, normal pupils selectively learned song from the tutor who sang longer but less frequently. When hearing songs, pupils preferentially approached the selected tutor. Excitotoxic lesions of the amygdala increased pupils’ social motivation toward tutors yet diminished their song-responsive approach, especially to the selected tutor. Whereas the pupils with amygdala lesions retained their ability to imitate song, the tutor selection became more unpredictable with diminished preference for a specific tutor. Neuronal tracing confirmed that the zebra finch amygdala is connected to the circuits involved in social functions but lacks direct connections to those critical for song control and learning. These results suggest that the amygdala regulates social motivation and tutor selection in juvenile zebra finches, highlighting its role in imitative learning.
Significance Statement
Social interaction plays a critical role in imitation, particularly in the acquisition of speech and cultural behaviors like dance and song. Although studies in rodents indicate the involvement of the amygdala in social behaviors and social learning, the mechanisms coordinating social behaviors and imitative learning remain poorly understood. Vocal imitation in juvenile songbirds is an ideal model to investigate such mechanisms. Here, we report that lesioning the amygdala in juvenile zebra finches increases overall social motivation but disrupts selective song learning and their preferential approach to the selected tutor. These findings provide new insights into the social function of the amygdala in imitation.
Introduction
Imitation is a form of efficient learning that enables precise transmission of behaviors among individuals. In humans, imitation facilitates the acquisition of speech and cultural behaviors such as dancing and singing (Heyes, 2023). Imitation often accompanies social interaction between a teacher and a pupil. Social interaction plays a crucial role in infant language acquisition (Kuhl, 2007; De Felice et al., 2023). For example, interaction with a live adult teacher enhances phoneme and word learning (Kuhl et al., 2003; Yu and Smith, 2012; Bosseler et al., 2024). Given the social facilitation of imitation, neural circuits involved in social behaviors should play a significant role in the process of imitation, yet the evidence for such mechanisms is currently scarce.
Songbirds are an ideal model for dissecting the mechanisms linking social behaviors and imitation. Songbirds learn a complex vocal sequence “song” through vocal imitation. Juvenile pupils first memorize the songs of adult tutors (sensory learning) and practice singing to match their own vocalization to the memorized song (sensorimotor learning). Live, face-to-face interactions with tutors lead to better imitation, compared with passive listening of song playback, suggesting social facilitation of sensory learning (Baptista and Petrinovich, 1986; Chen et al., 2016). Importantly, pupils do not indiscriminately copy all the exposed songs but selectively learn from a preferred tutor (Clayton, 1987; Rodríguez-Saltos et al., 2023). Studies suggest that pupils’ attention and memory-driven motivation toward the tutor correlate with their performance in future imitation (Chen et al., 2016; Liu et al., 2021).
This kind of social facilitation of imitation suggests that the neural circuits controlling social behaviors can influence song imitation. A candidate for such social circuits is the amygdala, which can regulate social behaviors and learning via its processing of values and emotions in mammals (Adolphs, 2010; Janak and Tye, 2015; Raam and Hong, 2021). In the avian brain, a forebrain structure, the nucleus taeniae of the amygdala (TnA), also known as the medial ventral arcopallium (AMV), has been proposed as the avian equivalent of the amygdala (Reiner et al., 2004; O’Connell and Hofmann, 2011; Mello et al., 2019). The adult songbird TnA is responsive to song playbacks (Earp and Maney, 2012; Lampen et al., 2014; Fujii et al., 2016) and involved in social behaviors (Cheng et al., 1999; Ikebuchi et al., 2009; Svec et al., 2009; Schubloom and Woolley, 2016), suggesting that the TnA could modulate social behaviors in response to tutor songs. However, the function of the TnA in juvenile pupils during song imitation remains unexplored.
Here, we analyze whole-day social interactions between pupil and tutor zebra finches across days of live tutoring sessions and how the TnA regulates the pupils’ social behaviors during sensory learning. Our data suggest that pupils with excitotoxic lesions in the TnA exhibit stronger motivation to approach tutors during sensory learning. Whereas TnA lesions did not compromise the ability to imitate song per se, the lesions reduced the pupils’ song-responsive approach to the preferred tutor, disrupting the selective song learning observed in normal pupils. These results suggest the social role of the amygdala in modulating selective imitation, a critical process for refining cultural behaviors.
Materials and Methods
Animals
Male zebra finches (Taeniopygia guttata) bred in the lab colony at Waseda University were used. For live tutoring experiments, we used 15 juveniles and five adults (>110 d post hatch; dph) as song tutors. Of the 15 juveniles, eight birds were used for the TnA lesion (ibotenic acid injections), and seven birds were for the control condition (six received saline injections; one remained intact). Juveniles were raised by both parents so that they were adequately fed for normal physical and behavioral development. They were isolated from their parents and siblings at ∼30 dph (mean ± SD; control, 33.1 ± 2.9 dph; lesion, 34.0 ± 2.1 dph), received surgery at ∼40 dph (mean ± SD; control, 40.0 ± 2.1 dph; lesion, 38.4 ± 1.0 dph). At the start of live tutoring with the first tutor, they were ∼50 dph (mean ± SD; control, 47.0 ± 1.6 dph; lesion, 47.9 ± 3.6 dph). For neuronal tracing of the TnA, four juveniles and four adults were used (mean age at the time of injection; juvenile, 50.8 dph; range, 38–61 dph; adult, 162 dph; range, 121–225 dph). Another five birds were used for tracer injections into the song nucleus HVC (n = 3 for AAV injection; ages at the time of injection, 40, 84, 95 dph; n = 1 for dextran injection; age at the time of injection, 48 dph) and the robust nucleus of the arcopallium (RA; n = 1 for dextran injection; age at the time of injection, 48 dph). Birds were kept under a 14/10 h light/dark cycle with food and water provided ad libitum. All experimental procedures were in accordance with the JSPS guidelines and approved by the Waseda University Animal Experiment Committee.
Live tutoring and behavioral recording
Each juvenile was sequentially tutored by two different adult males for 3 d each with a 1 d interval between the 3 d sessions (Fig. 1A). The tutors were genetically unrelated and unfamiliar to the pupils prior to tutoring. A pupil–tutor pair was kept in a tutoring cage (size 250 × 180 × 210 mm, W × D × H) placed in a sound attenuation box. A grid partition at the center of the cage physically separated a pupil and a tutor (Fig. 1B), but they could still interact visually and acoustically. For acclimation, the pupil was moved to the cage at least 2 d before the first tutoring session. The birds’ behavior was continuously recorded at 1 fps with a web camera (BSW200MBK, Buffalo) throughout the 14 h light period for each day. The camera was mounted on the ceiling to capture the entire area of the tutoring cage. The video recording started from the day before the first tutoring session and continued throughout all the tutoring sessions (Fig. 1A). Sounds were also continuously recorded with a microphone (PRO35, Audio-Technica) and sampled at 44.1 kHz with an audio interface (8pre, MOTU). Vocalizations longer than 20 ms were automatically detected and saved with Sound Analysis Pro (SAP2011; Tchernichovski et al., 2000). After the tutoring experiment, pupils were individually housed in a sound attenuation box until 90 dph for the analysis of song learning. The pupils were not exposed to other birds including females during this period to prevent possible social facilitation of sensorimotor learning (Carouso-Peck and Goldstein, 2019; Bistere et al., 2024). Although we retained the pupils with their father until ∼30 dph, the effect of song experience before the tutoring experiment was minimal, as the pupil–father song similarity (see below, Song analysis) was significantly lower than the pupil–tutor song similarity of the selected tutor and comparable with that of other (nonselected) tutors in both control (n = 7; effect of tutor, F(2,12) = 14.38; p = 0.0007; selected vs other, t(6) = 4.49; p = 0.0042; selected vs father, t(6) = 5.72; p = 0.0037; other vs father, t(6) = 0.55; p = 0.604; one-way ANOVA) and TnA-lesioned birds (n = 8; effect of tutor, F(2,14) = 11.86; p = 0.001; selected vs other, t(7) = 4.15; p = 0.0043; selected vs father, t(7) = 5.68; p = 0.0022; other vs father, t(7) = 0.55; p = 0.602; one-way ANOVA).
Behavioral analysis
The positions of tutor and pupil were extracted from each video frame by a custom MATLAB code with Image Processing Toolbox. Images were first converted to grayscale images and were cropped so that only the floor area of the tutoring cage was included for analysis. Each grayscale image was then normalized by the median brightness of that image. The contrast was enhanced by clipping values below/above 20% of the input range to the minimum/maximum values. Next, the images were Gaussian-filtered to minimize the high spatial frequency noises such as small seeds on the floor. The filtered images were binarized with a brightness threshold. After edge detection, objects with area and circularity exceeding thresholds were defined as a pupil and a tutor. The thresholds of brightness and area were optimized for each pupil–tutor pair. The centroid of each bird was used as their positions in the following analyses.
To examine the effect of tutor presence on the pupils’ position, the shortest distance (mm) from the pupil to the center partition was calculated (Fig. 1B). As a more direct proxy of pupil's approach to tutors, “pupil distance” was defined as the distance (mm) from the pupil's position to the point where the center partition meets the line connecting the pupil and the tutor (Fig. 1E). “Tutor distance” was similarly defined as the distance from the tutor's position to the point where the center partition meets the line connecting the pupil and the tutor. To consider whether pupils were motivated to actively approach tutors, pupils' behavioral states were categorized into “active” or “inactive” based on a threshold for locomotion (“active” if >10 mm/s for >12 s, otherwise “inactive”). The pupil distance in the active/inactive states was measured (Figs. 1E, 5C). The percentage of active phase duration to the entire duration of 3 d session was also used to analyze the birds’ activity level. Additionally, we noted that pupils often showed continuous, side-to-side hopping in front of the center partition under the presence of a tutor. This “pacing” behavior was defined as a large (>20 mm/s), continuous (>6 s) movement along the plane of the partition, observed within 75 mm of the partition (Fig. 1F). We measured the duration that pupils engaged in “pacing” and reported the percentage of pacing time relative to the total duration of the “active” state, taking the variation in locomotor activity into account (Figs. 1F, 5D).
To analyze the pupils’ response to a singing tutor, we extracted song bouts where a tutor sang for >6 s. Silent intervals between song motifs (for definition, see below, Song analysis) shorter than 1 s in total were allowed to be included in a bout. The 10 s time windows preceding song bout onsets and following song bout offsets were defined as “pre-singing” and “post-singing” phases, respectively. The duration of time windows was determined based on the observation that the change in the pupil's position typically occurred within these periods. We analyzed pupils’ movement and approach in response to tutor's song bouts. The amount of pupils’ movement was defined as the distance traveled per second (Figs. 2A, 6A, “pupil movement (mm/s)”). The song-responsive approach was analyzed using the pupil distance (Figs. 2B,C, 6C–E). The values of pupil movement or pupil distance were averaged across the time window of each phase (pre-singing, singing, and post-singing), and their mean across all song bouts that occurred in a 3 d session was calculated for each pupil–tutor pair. To compare the degree of approach between the selected and other tutors or between the control and lesion groups, we used the change in pupil distance from the pre- to the post-singing phases [Δpupil distance (mm) in Figs. 3B, 6E, 7B].
Song analysis
To evaluate the outcome of song imitation, songs of pupils were recorded at 90 dph. Songs of tutors were recorded prior to the live tutoring experiment. All songs in this study were collected as undirected singing, produced in isolation from other birds. We used the same sound recording system as the live tutoring experiment. Zebra finch songs have a unit of structure called motif, which is a stereotyped sequence composed of several syllables. Based on a careful inspection into the spectrograms, a most-frequently produced, typical song motif was first determined for each bird. From a full-day recording, 30 different renditions of the typical motif were extracted for each bird by an experimenter who was not blind to the experimental conditions. The extracted motifs were high-pass filtered (>0.5 kHz) to reduce low-frequency background noise. The percentage song similarity (SAP %similarity) ranging from 0 (totally different sounds) to 100 (completely the same sounds) was calculated between each pupil–tutor pair using the similarity measurements function in SAP2011 (Tchernichovski et al., 2000) with default parameters optimized for zebra finches, using the modes of asymmetric (tutor song as template) and time course. The median percentage song similarity of 30 pairs of renditions was used as a representative value of each pupil–tutor pair. This analysis was also used to calculate the song similarity for each pupil–father pair.
To analyze the tutors’ singing patterns, all tutor song bouts longer than 3 s were extracted from each 3 d tutoring session. With these song bouts, the median duration (s) of song bouts (Figs. 3D, 7D) and the total number of song bouts produced during a 3 d session (Figs. 3E, 7E) were analyzed for each pupil–tutor pair.
Stereotaxic surgery
Birds were anesthetized with isoflurane inhalation and placed on a custom stereotaxic apparatus with ear and beak bars. A heating pad was used to maintain their body temperature during surgery. After removing feathers on the head, local anesthesia 1% (wt/vol) lidocaine hydrochloride was applied onto the scalp before making an incision. The TnA and song nuclei (HVC or RA) were targeted using a stereotaxic coordinate from the bifurcation of the midsagittal sinus (TnA, 1.1 mm anterior, 1.7 mm lateral, and 3.6 mm ventral, head angle 47.5°; HVC, 0.0 mm anterior, 2.4 mm lateral, and 0.5 mm ventral, head angle 20°; RA, 1.5 mm posterior, 2.4 mm lateral, and 2.1 mm ventral, head angle 5°). A craniotomy was made above the injection site on each hemisphere, and reagents or virus was injected through a glass micropipette using a pressure injection system (Nanoject III, Drummond Scientific). After injections, the craniotomies were covered with their skull bone and/or a silicone elastomer (Kwik-Sil, World Precision Instruments), and the incision was closed with a veterinary adhesive (Vetbond, 3M). Birds were returned to their isolation cage and monitored to ensure recovery.
Lesion and quantification
About a week prior to the live tutoring, 1% (wt/vol) ibotenic acid (ab120041, Abcam) dissolved in saline (0.9% NaCl), pH 8.4–8.5, was bilaterally injected into the TnA (70–100 nl per hemisphere) of juvenile birds (n = 8). The total volume was injected into a single coordinate using a glass micropipette, with 7–10 injections of 10 nl at 30–50 s intervals. Lesions did not always cover the full volume of the TnA, especially on the rostrocaudal axis (Fig. 4), but we chose to inject into a single coordinate rather than multiple sites to avoid extensive spread of the injectant to surrounding areas of the TnA. In six juveniles in the control group, the same volume of saline was bilaterally injected. After the song recording at 90 dph, birds were transcardially perfused for collecting fixed brain samples. In the zebra finch brain, the TnA is a nucleus located in the medioventral corner of the arcopallium, spanning >1 mm on the rostral–caudal axis. Its anatomical and cytoarchitectonic features are elaborated as the AMV in Mello et al. (2019), demonstrating that the TnA starts rostrally as a horizontally elongated thin band, becomes thicker and oval on the more caudal planes (Figs. 4, 9A–C, middle, right), and then terminates as a small round area at the level of RA (Fig. 9A–C, left). To cover the lesions ranging over the volume of this nucleus, we sampled eight coronal sections separated by 80 μm each other and containing the thickest middle part of the TnA from each hemisphere of ibotenic acid- or saline-injected birds. Area measurements and cell count analyses were conducted with Fiji ImageJ with the following procedures. Nissl-stained images acquired through a 4× objective lens were used for area measurements. The remaining TnA area with intact cells was measured for each sample and summed across samples from both hemispheres (16 sections). The total TnA intact area of each bird was normalized by the mean total area of six saline-injected birds (Fig. 4D). Cell count was conducted with the images acquired through a 20×objective lens to examine the cell density in the regions defined as intact/lesioned in the 4× images. In four saline-injected birds and four ibotenic acid-injected birds, ROIs of 0.04 mm2 within the area of TnA (the putative area in case of lesioned samples) were extracted. The normalized grayscale images of ROIs were top hat-filtered (with 1.5 μm radius), binarized with a threshold determined by the “Moments” method in the Fiji auto threshold function, processed with watershed separation. The number of objects larger than 72 μm2 was then counted using “analyze particle” function. For each bird, the mean count of eight ROIs (four from each hemisphere) was used as a representative value (Fig. 4E).
Neuronal tracing
For the neuronal tracing experiment, an adeno-associated virus encoding green fluorescent protein AAV9-CAG-EGFP (#37825-AAV9, Addgene) was injected into the bilateral/unilateral TnA with a volume of 400 nl per hemisphere (except one hemisphere that received 330 nl). Data of TnA tracing were collected from 13 hemispheres of eight birds in total (bilateral injection in three adults and two juveniles, unilateral injection in one adult and two juveniles). The birds were perfused after >1 month of incubation (mean ± SD, 44.8 ± 7.8 d). AAV9-CAG-EGFP was also injected into bilateral HVC with the same volume (400 nl per hemisphere) in three birds with an incubation period of 23–25 d. Another neuronal tracer dextran Alexa Fluor 488 (D22910, Thermo Fisher Scientific) was injected into bilateral HVC (150 nl per hemisphere) or RA (100 or 150 nl per hemisphere) in one juvenile for each nucleus. These birds were perfused 5 d after the surgery. To unambiguously locate and identify the structures labeled with the tracer, we performed immunohistochemistry against tyrosine hydroxylase (TH) or parvalbumin (PV) on subsets of the brain sections.
Histology
Birds were deeply anesthetized with intramuscular injections of a mixture of medetomidine, midazolam, and butorphanol and transcardially perfused with >15 ml of PBS and 4% (wt/vol) paraformaldehyde (PFA) in PBS. After overnight fixation in 4% PFA in PBS, the brain was cryoprotected with 30% (wt/vol) sucrose in PBS for >12 h. Coronal sections (40 μm thickness) were prepared using a freezing microtome (REM-710, Yamato-Koki) and stored in PBS. For immunohistochemical staining, the sections were washed twice in PBS, permeabilized with 0.3% Tween-20 in PBS (PBST) for 1 h, blocked with 10% Blocking One Hist (06349-64, Nacalai Tesque) in PBST (blocking solution) for 1 h, and incubated overnight at 4°C in the blocking solution containing mouse anti-PV primary antibody (1:500; MAB1572, Sigma-Aldrich) or rabbit anti-TH primary antibody (1:500; AB152, Sigma-Aldrich). The sections were washed three times in PBST and incubated with donkey anti-mouse or anti-rabbit secondary antibody conjugated with Alexa Fluor (1:500; 715-545-150, or 711-585-152, Jackson ImmunoResearch Laboratories) in PBST for 1.5 h at room temperature and then washed three times in PBS. To quantify the lesioned area, the brain sections were washed and permeabilized in PBST three times (5 min each), incubated with fluorescent Nissl stain solution (1:500; N21480, Invitrogen) in PBST for 20 min, and then washed in PBST for 10 min and in PBS overnight. After staining, sections were coverslipped with mounting medium (Fluoromount-G, Invitrogen) and imaged with a fluorescent microscope (BZ-X810, Keyence).
Statistical analysis
The effect of tutor presence on the pupils’ position (distance to the partition) and the pacing time (%; Figs. 1D,F, 5B) and the effect of phase before and after tutor song bouts (pre-singing, singing, and post-singing) on the pupil distance and movement (Figs. 2A,C, 6A,D) were tested by one-way ANOVA. To test the effect of active/inactive states and the tutoring order (first/second) on the pupil distance (Fig. 1E) or tutor distance (provided in the text) in the control group and the effect of these activity states and the control/lesion conditions on the pupil distance (Fig. 5C), we used two-way ANOVA. All ANOVAs were performed with R, and Schaffer's method was used to adjust p values in the post hoc multiple comparisons (adjusted p values are reported in the manuscript). For comparing the pacing time (%; Fig. 5D), song-responsive pupil movement and approach (Δpupil distance; Fig. 6B,E), the tutor distance, the pupils’/tutors’ activity level, and the amount of pupil singing (provided in the text) between the control and lesion groups, Mann–Whitney U test was used. For comparing the song-responsive approach (Δpupil distance) (Figs. 3B, 7B), song similarity (Figs. 3C, 7C), and tutors’ singing pattern (Figs. 3D,E, 7D,E) between the selected and other tutors within the control/lesion condition, the Wilcoxon signed-rank test was used. These nonparametric tests were performed by statistical modules in a Python-based package SciPy. The statistical results are reported within the legends of corresponding figures, including the number of samples and the types of the statistical test. Bar or point plots with error bars in the figures show mean and 95% confidence intervals unless otherwise specified. In all kinds of plots, asterisks indicate p < 0.05. All plots and heatmaps were generated using a Python data visualization library Seaborn.
Results
Pupils normally exhibit social motivation toward tutors
To explore the process of selective song learning in male juvenile zebra finches, two adult male tutors unfamiliar to the pupil were sequentially introduced into the cage, each for 3 d (Fig. 1A). A grid partition at the center of the cage physically separated the pupil and tutor during the 3 d tutoring session, but the pupil could socially interact with the tutor across the partition (Fig. 1B). Following the introduction of a tutor into the cage, pupils increased the time spent in proximity to the tutor's side (Fig. 1C,D), approaching significantly closer to the center partition when the tutor was present. To quantify the pupils’ approach to the tutor, a more direct measure “pupil distance” was calculated (Fig. 1E). The pupil distance was shorter when pupils engaged in active movement, indicating their motivation to approach the tutor. Notably, pupils frequently engaged in pacing behavior, repetitively moving back and forth in front of the center partition when the tutor was present (Fig. 1F). Pupils’ pacing in front of the partition was significantly more frequent when the tutor was present, accounting for ∼30% of active movement, which indicates that the frequent pacing behavior reflects pupils’ strong motivation to approach the tutor. We did not notice any difference in the pupils’ social motivation between the first and second tutors. The “tutor distance” was similar for the first and second tutors, irrespective of whether pupils were active or not (effect of tutor order, F(1,12) = 0.03; p = 0.8607; effect of state, F(1,12) = 80.20; p < 0.0001; interaction, F(2,12) = 0.16; p = 0.6952; two-way ANOVA). To summarize, our whole-day behavioral analysis highlights the pupils’ strong social motivation toward tutors during live tutoring.
Social motivation of control pupils during live tutoring. A, Experimental time course (control group). B, Schematic illustration of a tutoring cage. A grid partition at the center (dashed line) separated the pupil and tutor. A blue rectangle marks the cage area shown in C. C, Heatmaps showing the probability of a representative control pupil's positions before and during the first and second tutoring sessions, calculated for each of the 10 by 10 grids in the cage area (averaged across days for each condition; no tutor, first and second tutor). The partition (tutor) was on the left side (black arrows). D, Mean distance from the pupil to the partition (control, n = 7; effect of tutor, F(2,12) = 30.78; p < 0.0001; no tutor vs first tutor, t(6) = 6.15; p = 0.0025; no tutor vs second tutor, t(6) = 5.39; p = 0.0025; first tutor vs second tutor, t(6) = 1.97; p = 0.097; 1-way ANOVA). E, Illustration of “pupil distance” (left) and comparison of the pupil distance (mm) between active/inactive states (right). A single data point indicates one pupil–tutor pair (n = 14 pairs; effect of active state, F(1,6) = 16.72; p = 0.0064; effect of tutor order, F(1,6) = 5.39; p = 0.0594; interaction, F(1,6) = 1.16; p = 0.3234; 2-way ANOVA). Dark circle and light triangle symbols indicate first and second tutor, respectively. F, Illustration of pupil's “pacing behavior” (left) and the percentage of time spent pacing relative to the total time the pupils were active (right; n = 7; F(2,12) = 29.54; p < 0.0001; no tutor vs first tutor, t(6) = 6.82; p = 0.0015; no tutor vs second tutor, t(6) = 5.24; p = 0.0019; first tutor vs second tutor, t(6) = 0.81; p = 0.4509; 1-way ANOVA).
Selective song learning in normal pupils
When the tutor was singing, pupils reduced their movement, suggesting their attentive listening to the tutor (Fig. 2A). At the same time, pupils tended to approach the tutor after hearing the tutor sing (Fig. 2B). The pupil distance significantly decreased from pre-singing to singing, as well as from singing to post-singing phases (Fig. 2C). This song-responsive approach of pupils toward the tutor predicted their eventual performance in song learning. The pupils exhibit a bias to learn from one of the two tutors (Fig. 3A), with more frequent song-responsive approaches to the selected tutor (Fig. 3B). The tutoring order did not affect the selective song learning (Fig. 3C). Remarkably, the same tutor (K076) was always preferred by the pupils when available, resulting in higher learning outcomes compared with other tutors. Preferred tutors tended to sing longer, as reflected in their significantly longer median song bout durations (Fig. 3D). In contrast, the preferred tutors sang fewer songs, as indicated by their significantly smaller number of song bouts across tutoring sessions (Fig. 3E). These results suggest that concentrated tutoring with longer songs may be more effective than more frequent tutoring with shorter songs for facilitating song imitation in juvenile zebra finches.
Responses of control pupils to a singing tutor. A, Change in pupil movement (mm/s) in response to tutor singing (control n = 14 pairs; effect of phase, F(2,26) = 6.3; p = 0.0056; pre vs singing, t(13) = 2.32; p = 0.0374; singing vs post, t(13) = 3.62; p = 0.0094; pre vs post, t(13) = 0.92; p = 0.3759; 1-way ANOVA). B, Pupil's approach in response to tutor singing. Representative data from a control bird. A heatmap shows the pupil distance (mm) aligned to the onsets of tutor song bouts (20 bouts of one tutor in the order of occurrence; time bin, 1 s). Offsets are indicated with yellow ticks. In the bottom graphs, gray thin lines show Δpupil distance (pupil distance relative to the pre-singing phase) of the same 20 song bouts as in the heatmap, and the bold blue lines indicate their mean. The offset-aligned data are shown in the right panel. The values are smoothed by taking a moving average of 6 s window for a visualization purpose. C, Change in pupil distance (mm) in response to tutor singing (n = 14; effect of phase, F(2,26) = 39.28; p < 0.0001; pre vs singing, t(13) = 3.31; p = 0.0056; singing vs post, t(13) = 6.64; p < 0.001; pre vs post, t(13) = 7.96; p < 0.0001; 1-way ANOVA). In A,C, each gray thin line indicates one pupil–tutor pair. Point plots indicate the mean ± 95% confidence intervals.
Song learning in control pupils. A, Representative song spectrograms of a control pupil (90 dph) and of two adult tutors for this pupil. B, Song-responsive approach (negative Δpupil distance) to the selected tutor was stronger than that to the other tutor in the post-singing phase (n = 7; signed rank = 1.0; p = 0.0313; Wilcoxon signed-rank test). C, Song similarity to the first and second tutors (signed rank = 9.0; p = 0.469; Wilcoxon signed-rank test). D, Song similarity plotted against the median of song bout duration, which was longer in the selected than in other tutors (signed rank = 1.0; p = 0.03125; Wilcoxon signed-rank test). E, Song similarity plotted against the total number of song bouts across 3 d, which was smaller in the selected than in other tutors (signed rank = 0.0; p = 0.01563; Wilcoxon signed-rank test). In B–E, each line binds two data points (2 different tutors) from a single pupil. In D and E, tutor IDs are indicated by different symbols.
Exocytotoxic lesioning of the TnA
Next, we examined the role of the TnA in social motivation toward the tutor and selective song learning in pupils using the same sequential tutoring paradigm (Fig. 1A). Prior to the live tutoring sessions, ibotenic acid was bilaterally injected into the TnA (Fig. 4A–C). We could unambiguously delineate the TnA with immunohistochemistry against PV (Braun et al., 1985), in combination with the high density of Nissl-stained cells in the TnA (Mello et al., 2019). Injections of ibotenic acid reduced the intact area in the TnA to approximately half, on average (0.492 ± 0.254, mean ± SD) compared with saline injections (Fig. 4D). With higher-magnification images, we confirmed that the lesioned areas contained almost no intact cells (Fig. 4E).
Confirmation of TnA lesions. A–C, Representative images of the TnA of a saline-injected and an ibotenic acid-injected bird. Scale bars indicate 100 μm. A, PV-labeled images acquired with a 4× objective lens. B, Nissl-stained images acquired with a 4× objective lens. C, Magnified images (20×) of marked areas in B. D, Comparison of the intact area within the TnA, normalized to the mean TnA areas in the saline condition. E, The number of intact cells per 0.04 mm2 ROI in the intact or lesioned areas.
TnA-lesioned pupils increased social motivation toward tutors
Comparable to the control pupils (Fig. 1), TnA-lesioned pupils exhibited the social motivation to approach the tutor (Fig. 5A), as indicated by significantly shorter distance from the center partition when the tutor was present (Fig. 5B). The approach to the tutors was more frequent when the pupils engaged in active movement (Fig. 5C). Notably, however, the pupil distance was significantly shorter in the TnA-lesioned pupils compared with the control pupils, especially when the pupils were active. Moreover, the TnA-lesioned pupils exhibited pacing behavior in front of the center partition more frequently than the control pupils during tutoring (Fig. 5D). These results suggest that TnA lesions heightened pupils’ social motivation toward the tutors. The activity level of TnA-lesioned pupils was similar to that of the control pupils (control, 33.47 ± 11.34%; lesion, 37.72 ± 13.97% (mean ± SD); U = 88.0; p = 0.3286; Mann–Whitney U test). We did not notice differences in tutor's behavior, including the tutor distance (control, 37.23 ± 10.57 mm; lesion, 34.04 ± 7.12 mm; U = 102; p = 0.6929; Mann–Whitney U test) and activity level (control, 55.35 ± 5.79%; lesion, 56.27 ± 6.48%; U = 151; p = 0.1095; Mann–Whitney U test). Therefore, the increased social motivation of TnA-lesioned pupils toward tutors cannot be explained by disrupted motor vigor in the pupils or altered social behaviors of the tutors.
Alteration of social motivation in TnA-lesioned pupils. A, Heatmaps of the positions of a representative TnA-lesioned pupil mapped onto the cage area, as in Figure 1C. B, Mean distance from the pupil to the partition (lesion n = 8; F(2,14) = 85.42; p < 0.0001; no tutor vs first tutor, t(7) = 4.81; p = 0.0059; no tutor vs second tutor, t(7) = 4.73; p = 0.0059; first tutor vs second tutor, t(7) = 2.58; p = 0.0363; 1-way ANOVA). C, Mean pupil distance (mm) in active and inactive states compared between the control and lesion groups. A single data point indicates one pupil–tutor pair (control n = 14; lesion n = 16; effect of state, F(1,28) = 111.91; p < 0.0001; effect of lesion, F(1,28) = 4.81; p = 0.0367; interaction of state and lesion, F(1,28) = 14.40; p = 0.0007; effect of lesion in active state, F(1,28) = 14.65; p = 0.0007; effect of lesion in inactive state, F(1,28) = 0.31; p = 0.5824; effect of states in control, F(1,13) = 19.36; p = 0.0007; effect of states in lesion, F(1,15) = 122.88; p < 0.0001; 2-way ANOVA). D, Time of pupil's pacing behavior relative to the total time the pupils were active (U = 33.0; p = 0.0011; Mann–Whitney U test).
Disrupted song responses and selective song learning in TnA-lesioned pupils
When the tutor was singing, TnA-lesioned pupils did not clearly reduce their movement, suggesting their disrupted modulation of the attention to the singing tutor (Fig. 6A,B). TnA-lesioned pupils did exhibit song-responsive approach toward tutors, as the mean pupil distance was significantly shorter in the post-singing phase (Fig. 6C), compared with that of the pre-singing and singing phases (Fig. 6D). However, their song-responsive approach was not as consistent as the control group (see Fig. 2B,C for comparison), as evident in smaller decrease in the pupil distance from the pre- to post-singing phase in TnA-lesioned pupils (Fig. 6E).
Responses of TnA-lesioned pupils to a singing tutor. A, Change in pupil movement (mm/s) in response to tutor singing (lesion n = 16 pairs; effect of phase, F(2,30) = 1.07; p = 0.3552; 1-way ANOVA). B, Comparison of the pupil movement during the tutor singing phase between the control and lesion groups (control n = 14; lesion n = 16; U = 86.0; p = 0.2891; Mann–Whitney U test). C, Pupil's approach in response to tutor songs, as in Figure 2B. Representative data from a TnA-lesioned bird. Data from 20 song bouts are shown. D, Change in pupil distance (mm) in response to tutor singing (n = 16; effect of phase, F(2,30) = 6.33; p = 0.0051; pre vs singing, t(15) = 0.05; p = 0.9586; singing vs post, t(15) = 3.15; p = 0.0200; pre vs post, t(15) = 2.60; p = 0.0200; 1-way ANOVA). E, Song-responsive approach (negative Δpupil distance) was stronger in control pupils than TnA-lesioned pupils in the post-singing phase (U = 56.0; p = 0.0211; Mann–Whitney U test). In A and D, each gray thin line indicates an individual pupil–tutor pair. Point plots indicate the mean ± 95% confidence intervals.
This diminished song-responsive approach accompanied altered patterns of song learning in TnA-lesioned pupils (Fig. 7). First, their song-responsive approach did not predict the song learning outcome, as shown by the similar decrease in the pupil distance from the pre- to post-singing phase for the selected and other tutors (Fig. 7B). Second, the median song bout duration (Fig. 7D) and the number of song bouts (Fig. 7E) were not significantly different between the selected and other tutors, indicating that TnA-lesioned pupils were not sensitive to a concentrated, intensive tutoring pattern. As a result, the normally preferred tutor was not always chosen by the TnA-lesioned pupils for imitation, even when they were tutored by the same pairs of tutors as the control pupils.
Song learning in TnA-lesioned pupils. A, Representative song spectrograms of a TnA-lesioned pupil (90 dph) and two adult tutors for this pupil. B, Song-responsive approach (negative Δpupil distance) to the selected tutor was not significantly different from that to the other tutor in the post-singing phase (n = 8; signed rank = 16.0; p = 0.8438; Wilcoxon signed-rank test). C, Song similarity to the first and second tutors (signed rank = 7.5; p = 0.195; Wilcoxon signed-rank test). D, Song similarity plotted against the median of song bout duration, which showed no significant difference between the selected and other tutors (signed rank = 17.0; p = 0.9453; Wilcoxon signed-rank test). E, Song similarity plotted against the total number of song bouts across 3 d, which showed no significant difference between the selected and other tutors (signed rank = 16.0; p = 0.8438; Wilcoxon signed-rank test). In B–E, each line binds two data points (2 different tutors) from a single pupil. In D and E, tutor IDs are indicated by different symbols.
To be noted, the TnA-lesioned pupils still exhibited a bias to learn from one of the two tutors, with the learning performance comparable with that of the control pupils [song similarity to the selected tutor, control 62.6 ± 12.2; lesion 69.0 ± 12.9 (mean ± SD); U = 23.0; p = 0.602; Mann–Whitney U test]. Like control pupils, their tutor choice was not influenced by order (Fig. 7C). Moreover, their amount of singing (song bouts at ∼90 dph) was also comparable with that of the control pupils [control 1,057 ± 589; lesion 1,031 ± 309 (mean ± SD); U = 26.0; p = 0.866; Mann–Whitney U test], indicating that TnA lesions had little impact on the ability to imitate song and the amount of practice. These results suggest that TnA lesions specifically disrupt the association between social motivation and song learning, leading to unpredictable learning patterns.
Neural circuits connected with the TnA
To explore the neural circuits linking the social motivation toward the tutor and song learning in zebra finches, AAV9-CAG-EGFP was injected into the TnA. The retrograde transfection with AAV9 has been reported in zebra finches (Roberts et al., 2017; Chen et al., 2019; Sánchez-Valpuesta et al., 2019). Indeed, injections of AAV9-CAG-EGFP into HVC, with an incubation period of 3–4 weeks, retrogradely labeled EGFP-positive cell bodies in its upstream cortical (nucleus interface of the nidopallium, NIf) and subcortical (nucleus uvaeformis, Uva) nuclei, demonstrating its effectiveness for retrograde tracing (Fig. 8).
Retrograde tracing with AAV9. A, EGFP expression localized in the injection site HVC. EGFP-positive cell bodies were observed in the thalamic nucleus Uva (B) and the cortical nucleus NIf (C). Red rectangles in the schematized images indicate the fields of view of fluorescent images (middle, PV; bottom, EGFP). Scale bars, 100 μm.
After injecting AAV9-CAG-EGFP into the TnA, the neurons expressing EGFP were largely confined within the TnA in all juvenile birds (n = 4), overlapping with the region marked by a high density of PV-positive neurons (Fig. 9). However, we also noted that leaked viruses sparsely transfected cells beyond the boundary of the TnA, particularly in areas dorsal to the TnA.
Neuronal tracer injection and immunohistochemical labeling of the TnA. A, Schematized images of coronal planes around a typical injection site arranged from a caudal (left) to rostral section (right). Red rectangles highlight the fields of view shown in the microscopic images in B and C. B,C, Coronal sections of EGFP-injected hemispheres (320 μm caudal/rostral to one another) labeled with PV. White arrow heads indicate medioventral edge of the TnA (B). Scale bars, 100 μm.
The TnA in zebra finches had broad connections with the neural circuits implicated in social behaviors, similar to other avian species (Cheng et al., 1999) and rodents (Knapska et al., 2007). TnA densely projected to the septum (Fig. 10A), especially within the medial septal nuclei. TnA also projected to the hypothalamic area, including the ventromedial and lateral hypothalamus (Fig. 10B), with putative passing fibers observed in the dorsal part of the hypothalamus near the paraventricular nucleus and the anterior hypothalamus. In addition, there were sparse but reciprocal connections between the TnA and the hippocampus (Fig. 10C,D). Axon terminals were usually distributed in the medial part of hippocampus, just lateral to the midplane, but broadly observed along the caudal–rostral axis.
TnA neuronal connections. A, Septum. B, Hypothalamus. C, Hippocampus (rostral). D, Hippocampus (caudal). White arrow heads indicate the cell bodies labeled with EGFP. E, A caudal part of the medial striatum (top, EGFP; bottom, TH). MSt, medial striatum; lv, lateral ventricle; TSM, septomesencephalic tract. F, A rostral part of the medial striatum. G, Midbrain ICo (top, EGFP; bottom, PV). ICo, intercollicular nucleus; DM, dorsomedial part of the ICo; MLd, dorsal part of the lateral mesencephalic nucleus. In E and G, a white square in the bottom panel (TH or PV) corresponds to the field of view of the middle panel (EGFP). In all panels, scale bars indicate 100 μm, and small red rectangles in the schematic drawings show the fields of view.
TnA neurons also projected to the medial striatum, especially to the ventral edge of lateral ventricle from a caudal part near the lateral bed nucleus of the stria terminalis (Fig. 10E) to a rostral part near the nucleus accumbens (Fig. 10F). In the midbrain periaqueductal gray (PAG), relatively sparse fibers were observed around the intercollicular nucleus (ICo) (Fig. 10G), which is located medial to the dorsomedial nucleus (DM) and the dorsal part of the lateral mesencephalic nucleus, while the dopaminergic neurons expressing TH in the PAG involved in song imitation (Tanaka et al., 2018) were devoid of their projections. Similarly, no projections were found in the ventral tegmental area (VTA), the substantia nigra pars compacta (SNc), and the locus ceruleus (LC). To be noted, the VTA/SNc receive projections from the ventral part of the intermediate arcopallium, AIv (Mandelblat-Cerf et al., 2014; Chen et al., 2019), highlighting the specificity of our viral transfection in the TnA without significant leakage into nearby AIv.
We did not notice any connections between the TnA and the forebrain song nuclei. Injection of AAV9-CAG-EGFP or an anterograde tracer dextran Alexa Fluor 488 into HVC did not label axon terminals in the TnA, confirming that there is no direct connection between HVC and the TnA. While a small number of RA neurons expressed EGFP (Fig. 9C, left), due to the leakage of viruses from the injection pipette, injection of anterograde tracer dextran Alexa Fluor 488 into the RA did not label axon terminals in the TnA. At least a part of the EGFP-positive axon terminals found in the nucleus DM after injection of AAV9-CAG-EGFP into the TnA may originate from the RA, although the expression in RA neurons was minimal with no retrogradely labeled cell bodies in the lateral portion of the magnocellular nucleus of the anterior nidopallium (LMAN). No EGFP-positive cell bodies or axon terminals were found in LMAN and Area X after injection of AAV9-CAG-EGFP into the TnA.
These projection patterns largely align with those in European starlings (Cheng et al., 1999), while we could not find the previously reported projections from HVC and the shell part of the auditory thalamic nucleus ovoidalis to the TnA. We did not notice any differences in the connectivity of the TnA between juvenile and adult zebra finches (four adults, four juveniles, 13 hemispheres).
Discussion
In this study, we investigated the function of the avian amygdala in social interactions during imitative learning in juvenile zebra finches. Lesions in the TnA in pupils increased the social motivation toward the tutor, diminished the song-responsive approach to the tutor, and disrupted the selective song learning from normally preferred tutors. It has long been recognized that the amygdala damage causes indiscriminate approach toward social and nonsocial stimuli in primates (Adolphs, 2010). It is also well established that the amygdala plays a pivotal role in social behaviors and cognition (Adolphs, 2010; Gothard, 2020), including social preference (Zinn et al., 2016; Hu et al., 2021; Djerdjaj et al., 2022) and social learning such as observational fear conditioning (Twining et al., 2017; Allsop et al., 2018). Our findings of increased but less selective responses to tutors in TnA-lesioned juvenile zebra finches are in line with the previous studies of the mammalian amygdala and provide additional insights into its role in imitative learning.
The similarity of the TnA to the medial amygdala and basolateral amygdala in mammals has been extensively discussed, based on anatomical (Zeier and Karten, 1971; Cheng et al., 1987, 1999), molecular (Vicario et al., 2017; Mello et al., 2019; Medina et al., 2023), immunohistochemical (Balthazart et al., 1992; Yamamoto et al., 2005; Montagnese et al., 2015), and functional perspectives (Thompson et al., 1998; Absil et al., 2002; Ikebuchi et al., 2009). Our observation of the dense distribution of PV-positive neurons in the TnA, compared with the surrounding arcopallium, further supports its similarity to the basolateral amygdala, as a high density of PV-positive neurons characterizes the basolateral amygdala in rodents (Kemppainen and Pitkänen, 2000; Keshavarzi et al., 2014). The basolateral amygdala plays an important role in social motivation and interaction (Ferri et al., 2016; Paine et al., 2017; Schönfeld et al., 2020; Ferrara et al., 2021) via its projections to the hippocampus, septum, and hypothalamus (Felix-Ortiz and Tye, 2014; Zinn et al., 2016; Hu et al., 2021; Rodriguez et al., 2023). Our finding that the TnA projects these circuits and regulates social motivation in zebra finches underscores its similarity to the basolateral amygdala in mammals.
Our results suggest that social motivation is regulated by the TnA and its downstream circuits (Székely and Krebs, 1996; Wild, 2017). Social motivation is possibly regulated by altering emotional processing such as fear and anxiety, as a huge body of studies in mammals supports the crucial role of the amygdala in emotional processing (LeDoux, 2000; Phelps and LeDoux, 2005). In our study, TnA lesions may have reduced anxiety and allowed the pupils to approach tutors more frequently. However, reduced anxiety would not fully capture the effects of TnA lesions because TnA lesions did not increase but rather decreased the pupil's approach to singing tutors (Fig. 6E). Moreover, TnA lesions disrupted the pupils’ selection of normally preferred tutors as their imitation model (Fig. 7B,D,E), which may not be simply explained by reduced anxiety.
However, the intricate relationships between emotional processing, social behavior, and imitative learning remain poorly understood. Behavioral studies hint at potential connections between emotions and imitation. In songbirds, tutor's aggression toward pupils, such as pecking and chasing, may facilitate pupils’ imitation (Clayton, 1987), though such facilitatory effects appear to vary depending on experimental paradigms (Williams, 1990; Akçay et al., 2014). On the other hand, neuroscience research has begun to uncover that various neuronal populations process distinct emotional valences, regulating social behaviors (O’Neill et al., 2018; Pignatelli and Beyeler, 2019; Raam and Hong, 2021). Since our lesion method cannot segregate neuronal subpopulations within the TnA, precise manipulation of specific neuronal subpopulations and pathways, in combination with behavioral assays to separate emotional responses, would be necessary in future studies to clarify the complex roles of the avian amygdala in emotional processing, such as fear (Saint-Dizier et al., 2009), as well as social and sexual behaviors (Ikebuchi et al., 2009; Svec et al., 2009; Schubloom and Woolley, 2016).
The lack of direct projections from the TnA to the song system aligns with the retained imitation performance in TnA-lesioned pupils. This also suggests that the selective song learning in control pupils is indirectly controlled by the social circuits downstream of the TnA. For example, the septum has been reported to regulate anxiety-like behavior and facilitate flocking (Kelly et al., 2011). Also, studies suggest that the ventromedial hypothalamus regulates song-related social responses (Maney et al., 2008; Schubloom and Woolley, 2016; Spool et al., 2024). These downstream targets of the TnA may coordinate to regulate social motivation in response to singing tutors during imitation.
It seems unlikely that the TnA and its downstream circuits are the only players linking social behaviors and song imitation. Indeed, prior studies have suggested that neuromodulatory systems in the midbrain (i.e., PAG and LC) induce plastic changes in auditory and motor cortical circuits during pupil–tutor social interaction, driving song imitation (Tanaka et al., 2018; Yanagihara and Yazaki-Sugiyama, 2019; Chen and Sakata, 2021; Katic et al., 2022). The TnA would plausibly work in parallel with these neural circuits in the juvenile brain to control song imitation.
Our data suggest that more intensive (longer song bout duration) but less excessive (small number of song bouts) tutoring is effective for imitation by normal pupils. Such sensitivity to singing patterns was diminished by TnA lesions, although TnA-lesioned pupils could still choose one of the two tutors for imitation. Further experiments would be necessary to precisely dissociate the factors that drive their choice, such as social quality, acoustic experience, and motor constraints. Using other songbird species like Bengalese finches capable of learning more complex songs from multiple tutors would be informative (Soma et al., 2009; Takahasi et al., 2010). Elucidating the factors that shape selective song learning in songbirds would expand our understanding of the neural mechanisms underlying imitation, a fundamental process driving cultural behaviors.
Footnotes
This work was supported by Ministry of Education, Culture, Sports, Science and Technology/Japan Society for the Promotion of Science KAKENHI Grant Numbers JP22KJ2905 (to T.G.F.), JP21H00963, JP22K18657, JP24H00179, and JP24H02199 (to M.T.) and Waseda University Grant for Special Research Projects 2022C-369, 2023C-350, and 2024C-348 (to M.T.). We thank NA Hessler, Y Osuka, and M Hara for their comments on earlier versions of this manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Masashi Tanaka at masashitanaka{at}waseda.jp.
