Adult-like neural representation of species-specific songs in the auditory forebrain of zebra finch nestlings

Abstract

Encoding of conspecific signals during development can reinforce species barriers as well as set the stage for learning and production of species-typical vocalizations. In altricial songbirds, the development of the auditory system is not complete at hatching, so it is unknown the degree to which recently hatched young can process auditory signals like birdsong. We measured in vivo extracellular responses to song stimuli in a zebra finch (Taeniopygia guttata) secondary auditory forebrain region, the caudomedial nidopallium (NCM). We recorded from three age groups between 13 days post-hatch and adult to identify possible shifts in stimulus encoding that occur before the opening of the sensitive period of song motor learning. We did not find differences in putative cell type composition, firing rate, response strength, and selectivity across ages. Across ages narrow-spiking units had higher firing rates, response strength, accuracy, and trial-by-trial reliability along with lower selectivity than broad-spiking units. In addition, we showed that stimulus-specific adaptation, a characteristic of adult NCM, was also present in nestlings and fledglings. These results indicate that most features of secondary auditory processing are already adult-like shortly after hatching. Furthermore, we showed that selectivity for species-specific stimuli is similar across all ages, with the greatest fidelity in temporal coding in response to conspecific song and domesticated Bengalese finch song, and reduced fidelity in response to owl finch song, a more ecologically relevant heterospecific, and white noise. Our study provides the first evidence that the electrophysiological properties of higher-order auditory neurons are already mature in nestling songbirds.

1 |. INTRODUCTION

Altricial sensory systems are not fully developed at birth. It is often assumed that altricial young are oblivious to the complexity of sensory signals in the world around them. Complex signal processing in high-order nuclei depends on prior sensation and processing in afferent regions that may still be developing. For example, orientation selectivity in layer 2/3 of the mouse primary visual cortex does not fully mature until it receives input from the thalamorecipient layer 4 (Hooks & Chen, 2020). Similarly, hearing sensitivity at specific sound pressure level thresholds is not fully mature until three weeks after hatching in songbirds (Amin et al., 2007; Brittan-Powell & Dooling, 2004), thus adult-like processing in the auditory forebrain before ~20 days post-hatch (dph) is unexpected. It is, therefore, reasonable to predict that telencephalic responses to conspecific and heterospecific vocalizations are fairly indistinguishable, at least until fledging, in altricial songbirds.

Studies on vocal learning and sensory responses in songbirds have established that the sensitive period for tutor song acquisition opens well after hatching (25 dph in zebra finches; Roper & Zann, 2006). The potential effects of auditory stimulation earlier in nestlings and embryos have only recently been examined. In contrast to the idea that auditory processing is limited at these stages, analyses of behavior and metabolic rate show that nestlings of multiple species can discriminate between species-specific vocalizations (McFarlane et al., 2016; Nelson & Marler, 1993; Shizuka, 2014; Whaling et al., 1997) and likewise, can incorporate learned elements into their begging calls (Langmore et al., 2008; Madden & Davies, 2006). Furthermore, embryo heart rate responses also exhibit discrimination among songs (Colombelli-Négrel et al., 2014; Kleindorfer et al., 2018) and embryos can learn call elements that they then produce post-hatch (Colombelli-Négrel et al., 2012). This evidence suggests that complex stimulus processing in high-level auditory regions of altricial songbirds could be mature much earlier than previously thought.

Our current knowledge of neural responses in very young songbirds is minimal. Extant neurophysiological studies in young birds show that auditory brainstem responses to sound stimulation do not reach adult-level sensitivity until about 20 dph (Amin et al., 2007; Brittan-Powell & Dooling, 2004) and that ZENK expression increases significantly in response to conspecific song over silence levels in embryo and nestling auditory forebrain (Rivera et al., 2019). To our knowledge, no previous study has performed in vivo extracellular recordings in the auditory forebrain of birds younger than 20 dph. Investigation of the forebrain of nestlings is crucial to understanding how processing and categorization of species-specific signals are accomplished in recently hatched songbirds.

To fill in this gap, we recorded responses to species-specific songs in the caudomedial nidopallium (NCM) of nestling, fledgling, and adult zebra finches (Taeniopygia guttata). NCM is a secondary auditory forebrain region characterized by novel song discrimination (Gentner et al., 2004; Lynch et al., 2017, 2018; Terpstra et al., 2006) and stimulus-specific adaptation (SSA) that persists longer in response to conspecific than heterospecific songs (Chew et al., 1995). NCM responses to song are also robust in juveniles at least starting by 20 dph (Miller-Sims & Bottjer, 2014; Stripling et al., 2001; Vahaba et al., 2017). NCM is a logical starting point to assess the encoding of complex auditory stimuli, including species-specific discrimination, in embryos and nestlings.

Our goals were twofold: 1. Characterize electrophysiological properties in NCM of young birds before the onset of the sensory period, and 2. Compare juvenile and adult electrophysiological responses to conspecific and heterospecific song playback in NCM. Following evidence that nestling sensory systems are not yet mature, we hypothesized that there are differences in electrophysiological properties of NCM neurons across ages. In particular, we predicted that firing rates, stimulus responsiveness, and selectivity would be overall lower in young birds, and that stimulus-specific adaptation and species-specific responses would be absent in nestlings. On the other hand, data showing behavioral and physiological discrimination of songs in embryos and nestlings favor the alternative hypothesis that electrophysiological responses in NCM are already adult-like during early development. This hypothesis predicts that firing rates, stimulus responsiveness, selectivity, stimulus-specific adaptation, and species-specific responses would be similar across these age groups.

2 |. MATERIALS AND METHODS

2.1 |. Subjects

Our study sample included nestling (N = 2 males, 2 females, 1 undetermined sex; n = 49 units), fledgling (N = 2 males, 1 undetermined sex; n = 27 units), and adult (N = 2 males, 2 females; n = 75 units) zebra finches from the breeding colony at the University of Massachusetts Amherst (housed on a 14:10 light-dark cycle). The adults used in this experiment were housed with a group of same-sex individuals and had visual and auditory access to breeding flight aviaries. We removed nestlings between 13 and 19 dph and fledglings between 20 and 24 dph (collectively hereafter “juveniles”) from mixed-sex breeding aviaries, while still in the nest or just after fledging. We found one “fledgling” subject still in the nest on the day of experimentation but included it with the fledgling group due to age. We chose to separate the juvenile age groups at 20 dph because most birds fledge by this age (Trillmich et al., 2016), this is the age where auditory brainstem audiograms become adult-like (Amin et al., 2007; Brittan-Powell & Dooling, 2004), and to more easily compare our results to previous studies including birds only ≥20 dph (Miller-Sims & Bottjer, 2014; Stripling et al., 2001). All animal procedures were approved by the Institutional Animal Care and Use Committee at the University of Massachusetts Amherst.

2.2 |. Surgery

The general surgical procedure was adapted from previous studies (Krentzel et al., 2018; Lee et al., 2018; Vahaba et al., 2017). We fasted birds for 30 min prior to surgery. At 10 min prior to surgery, we administered an oral dose of 15 μl of meloxicam to adults or 7.5 μl to juveniles and measured the bird’s mass using an electronic balance. At the start of surgery, birds were anesthetized with 2% isoflurane in 2 L/min O₂, wrapped in a cloth blanket, and fixed to a custom stereotaxic apparatus (Herb Adams Engineering) with a heating pad at a 45° head angle. Birds were maintained at 1% isoflurane in 2 L/min O₂ and 36°C throughout the surgery. Scalp feathers were removed, a 15 μl subcutaneous injection of 2% lidocaine was administered under the scalp, and an incision was made to expose the skull. Using the posterior edge of the bifurcation of the midsagittal sinus as a zero-point reference, we scored the skull above the left and right NCM at coordinates 1.1 mm rostral and 0.7 mm lateral using a scalpel. Small craniotomies were made at these locations by removing skull tissue and resecting the dura to expose the brain surface. We then fixed a metal head post at the base of the bird’s beak with dental cement to stabilize the head during recordings. Following previous experimental protocols in our lab, adult craniotomies were sealed with Kwik-Cast and subjects were allowed to recover for 24–48 hr in an individual cage before electrophysiology. Because the juveniles use in this study were still reliant on parental feeding, the same long-term recovery was not feasible. Juveniles were instead allowed to recover from anesthesia for about 10 min, then were taken directly to the recording chamber.

2.3 |. Electrophysiology and auditory playback

Extracellular single-unit recording was performed in vivo in awake-restrained birds as in previous studies (Miller-Sims & Bottjer, 2014; Schneider & Woolley, 2013; Vahaba et al., 2020). Subjects were brought into an anechoic chamber and fixed to a stereotaxic apparatus (Herb Adams Engineering) on an air table at a 45° head angle. A single 500 kΩ tungsten electrode (A-M Systems) was lowered into the brain at the craniotomy sites. We recorded from both hemispheres in all birds except one juvenile. Recordings were made between 1.1 and 2 mm ventral to the brain surface at sites with spontaneous bursting activity and selectivity for song versus broadband noise that is characteristic of NCM. Units were obtained from 1 to 2 penetrations per hemisphere and at multiple sites per penetration, separated on the dorsal-ventral axis by at least 100 microns. We recorded from at least eight sites for most birds. Recordings were amplified, bandpass filtered (300 to 5,000 Hz; A-M Systems), and digitized at 20 kHz (Micro 1401, Spike2 software; Cambridge Electronic Design). Recording sessions lasted a maximum of 4 hr.

We created two unique sets of auditory stimuli, each consisting of five sounds: two zebra finch songs, one heterospecific song from the confamilial and allopatric Bengalese finch (or society finch, Lonchura striata), one song from the heterospecific congeneric and sympatric owl finch (or double-barred finch, Taeniopygia bichenovii), and white noise (Figure 1a–c). We measured spectral characteristics of our experimental stimuli (Table 1) in Sound Analysis Pro (Tchernichovski et al., 2000) using an 11.29 ms window advancing at 2 ms. We began each trial with either Stimulus Set A or B, then switched after 5–10 trials to reduce habituation to song stimuli over the course of the recording session. This strategy was maintained across all birds in the dataset with one exception. This single subject (15 dph) was only presented with four (n = 14 units, no owl finch song) or three (n = 2 units, no owl finch or Bengalese finch song) stimuli. All stimuli were between 1.5 and 2.8 s in duration and were played from a speaker placed 45 cm from the bird’s head at a peak amplitude of ~70 dB at 1m. At each recording site, each playback trial consisted of 15 repetitions of each of the 5 stimuli in random order with an interstimulus interval of 10 ± 2 s, thus each trial lasted approximately 15 min. After recordings, birds were sacrificed via rapid decapitation and their brains were fixed in 30% sucrose-PBS solution and stored at −80°C for the later verification of electrode placement.

FIGURE 1.

Example spectrograms (left) and amplitude spectral density (right) of experimental stimuli. Zebra finch (a), Bengalese finch (b), and owl finch (c)

TABLE 1.

Spectral averages ± SE for the song stimuli of each species used in this experiment

	Zebra finch	Bengalese finch	Owl finch
Duration (s)	1.29 ± 0.47	2.35 ± 0.81	2.49 ± 0.11
Mean frequency (Hz)	2,778 ± 344	2,621 ± 681	2,810 ± 165
Peak frequency (Hz)	2,685 ± 406	2,586 ± 764	2,804 ± 184
Goodness of pitch	665 ± 95	547 ± 78	324 ± 1
Frequency modulation (rad)	46.6 ± 2.1	43.3 ± 0.4	37.0 ± 4.3
Amplitude modulation (1/t)	−0.015 ± 0.001	−0.019 ± 0.010	−0.002 ± 0.002
Wiener entropy	−1.61 ± 0.13	−2.18 ± 0.13	−1.89 ± 0.53

2.4 |. Juvenile sex determination

We determined the sex of juvenile birds from blood samples taken at the time of sacrifice. We extracted DNA using a QIAamp DNA Mini Kit (Qiagen Cat. No. 51304) and amplified it with PCR using primers for homologous regions (P2: YTKCCAAGRATGAGAAACTG and P8: TCTGCATCACTAAAKCCTTT) of the chromo-helicase DNA-binding gene found on the sex chromosomes of most birds (Griffiths et al., 1998). We used gel electrophoresis to separate amplified DNA segments alongside a positive male and female control on a 2% agarose gel. Sex was determined based on the number of bands, males showing one band and females showing two. We were unable to determine the sex of two individuals used in this study due to low DNA yield from samples.

2.5 |. Spike sorting and data analysis

We used Spike2 (Version 7.04) to isolate units in each recording. We first created templates using the default length of −0.48–1.145 ms and setting the amplitude thresholds from −0.6 to 2 times the noise band for individual recordings. We set the minimum percent of points in the template to 65%. We retained distinct units that did not substantially overlap other clusters in principle components space (Figure 2a); none of these units had an interspike interval violation within the 1 ms bin. All units used in the final analysis had a J3 statistic > 2 (mean ± SE = 3.74 ± 0.23), also indicating well-isolated single units (Friend et al., 2015). We measured peak to peak duration (ms; depolarization peak to peak of after-hyperpolarization) for each unit.

FIGURE 2.

NCM of young zebra finches is highly responsive to song before the sensory period of song learning. (a) PCA clustering (left) and waveform templates (right) of two representative single units from the same recording site in a 20 dph subject (BS top, black; NS bottom, gray). (b) Conspecific song stimulus waveform (top), raster plot (middle), and peristimulus time histogram (bottom) showing the activity of a single unit from nestlings (13 and 15 dph), a fledgling (24 dph), and an adult zebra finch. Example NS units are shown on the left and example BS units are shown on the right

Next, we calculated spontaneous (baseline) firing rate (number of spikes during 500 ms before stimulus onset divided by 0.5 s) and stimulus-evoked firing rate (number of spikes during stimulus divided by stimulus duration in s). As the last test to ensure the appropriateness of our single units used in the analysis, we also assessed whether units were auditory-responsive. To do this, we used paired Wilcoxon signed-rank tests to compare spontaneous firing rates during the 500 ms preceding each stimulus to evoked firing rates during the full duration of each stimulus. Units that did not have a significant change (p < .05) in firing rate in response to at least one of the five stimuli were removed from the analysis.

We used a normalized Z-score transformation to account for differences in spontaneous firing rate across neurons when assessing response strength to specific stimuli. As in previous studies (e.g., Lee et al., 2018; Vahaba et al., 2017), we calculated Z-scores based on the following equation:

$$Z-\mathit{\text{score}}=\frac{\overline{S}-\overline{B}}{\sqrt{var\left(S\right)+var\left(B\right)-2\mathit{\text{covar}}\left(S,B\right)}}$$

where S is the number of spikes during the stimulus and B is the number of spikes during the baseline period 500 ms before stimulus onset. Firing rates and Z-scores are averages of a unit’s response over 15 presentations of each stimulus. In a few cases, we removed individual stimulus presentations that occurred during significant movement artifacts. Movement was especially an issue with juvenile birds because the skull was not completely hardened at the ages tested, thus reducing the stabilizing effectiveness of the head post. All analyses were based on at least seven repetitions of a stimulus.

Neurons within the NCM of young songbirds may respond equally to all stimuli if they are not yet experientially tuned, or they may be highly selective owing to innate predispositions for conspecific song or experiential learning. In order to compare the selectivity of auditory-evoked units between adults and juveniles, we utilized a selectivity index similar to that used by Schneider and Woolley (2013). Selectivity was calculated using the formula 1-(N/5), where N is the number of stimuli that the unit responded to with a firing rate significantly above or below baseline (Wilcoxon signed-rank tests as above) out of the five stimuli to which it was exposed. We adjusted the formula as appropriate for units from one bird (n = 16) that were only presented with three or four stimuli. Selectivity values range between 0 and 0.8, with a value of 0 indicating that the unit’s response was statistically above baseline to all stimuli and a value of 0.8 indicating that the unit’s response was statistically above baseline only to a specific stimulus. We acknowledge that the criteria we used to define an auditory unit may affect the distribution of our selectivity index, although we could make the same conclusion for this variable even with a more stringent threshold of p < .001 for change in firing rate in response to a stimulus. Additionally, selectivity was negatively correlated with Z-score (see Results for stats), indicating that weakly firing units may have met our auditory criteria simply by chance. We did not see this as a major limitation here given that only 8/151 units included in our analysis had an SI = 0.8 (only “selective” to one stimulus), and that these eight units were spread almost evenly across the nestling and adult age groups.

Stimulus-specific adaptation (SSA) is a defining characteristic of adult NCM in songbirds. In order to explore this phenomenon in juveniles in relation to adults, we analyzed the subset of units that we recorded during the first trial with Stimulus Set 1 or Stimulus Set 2 for each subject (n = 31). This ensured that we chose units that were responding to the stimuli for the first time while we were recording from them. For each individual presentation of a stimulus, we first subtracted the baseline firing rate during the 500 ms before the stimulus from the firing rate during the duration of the stimulus. We then iteratively subtracted the firing rate during each of the 15 presentations of the same stimulus to the same unit from the firing rate during the first presentation and multiplied the values by 100. This resulted in a list of 15 responses for each stimulus-unit combination represented as a percentage of the response on the first presentation of the stimulus (Miller-Sims & Bottjer, 2014). We used linear regression to calculate the slope of these normalized firing rates across the multiple presentations of the stimulus. For each unit, we chose to analyze only the minimum slope, instead of slopes to all stimuli, to account for differences in selectivity among units. We used only these unit-stimulus combinations with the minimum slope to additionally assess the magnitude of change in response strength over multiple presentations of the same stimulus. To do this, we determined the median response within the first five stimulus presentations and also within the last five stimulus presentations, then used these values to calculate a percent difference score similar to Miller-Sims and Bottjer (2014).

We designated units with a peak to peak duration above 0.4 ms (broad-spiking, hereafter “BS”) as putative projection neurons and units < 0.4 ms (narrow-spiking, hereafter “NS”) as putative inhibitory interneurons similar to previous studies (Ono et al., 2016; Schneider & Woolley, 2013; Yanagihara & Yazaki-Sugiyama, 2016). This cutoff separated units significantly by both spontaneous and stimulus-evoked firing rate (see statistics reported in Results). We also found a third unit type that appeared inverted in Spike2 and had very large peak to peak durations. Similar units have been observed in extracellular recordings during other studies (e.g., Gold et al., 2009), although our understanding of the origin of these units remains limited. In our study, these units were still auditory-responsive and most similar to BS units in terms of peak to peak duration, spontaneous firing rate, and response behaviors, so we combined inverted and BS units for all analyses as in previous studies (e.g., Krentzel et al., 2018).

2.6 |. Temporal coding properties

In order to assess neuron response consistency, while taking into account the variability of spike timing across presentations of the same stimulus, we fed spike trains into a custom pattern classifier in Python (Lee et al., 2018; Vahaba et al., 2017). We compared the firing responses of each unit to each of the five stimuli. For each unit, its spike train during one of the presentations of each stimulus was pseudorandomly chosen as the template. Spike trains for all remaining stimulus presentations were then iteratively compared to this template. After repeating the process 1,000 times, this produced a matrix showing the similarity of spike timing between the predicted (template) and actual responses. The R_corr method (Caras et al., 2015) was then used to calculate an accuracy value for each unit to each stimulus, indicating the proportion of cases where the stimulus presentation with the highest correlation to the template was from the same stimulus (i.e., CON1 spike trains were most similar to a CON1 template). See Vahaba et al. (2017) for further details on the pattern classifier.

Accuracy scores range from 0 to 1: a score of 0 indicating complete dissimilarity and 1 indicating identical spike train patterns. They are a representation of how well the spike pattern predicts the stimulus. In order to account for the 16 units that were only presented with three or four different stimuli instead of five, which results in a different random accuracy value if spike patterns do not predict stimulus at all (0.2 if 5 stimuli, 0.25 if 4 stimuli, and 0.33 if 3 stimuli), we normalized accuracy score by subtracting random chance accuracy on a unit-by-unit basis. Thus, the reported accuracy values indicate the proportion of cases where spike trains correctly predicted the stimulus relative to random chance (random chance represented by accuracy = 0).

As an additional analysis of temporal coding properties, we created a custom trial-by-trial correlation similar to Schreiber et al. (2003) and Bottjer et al. (2019). We compared individual pairs of spike trains from repeated presentations of the same stimulus. Reliability, or R_corr, assessed the average similarity of all presentations of the same stimulus to the same unit, ranging between 0 and 1. Spike trains were evaluated for the trial-by-trial correlation of spike timing in response to stimulus playback across trials. A custom Python algorithm compared responses to stimuli using the R_corr metric (Schreiber et al., 2003) after smoothing spike trains with a Gaussian filter (16-ms standard deviation). While the pattern classifier above describes how clearly spike trains distinguish among different stimuli, R_corr estimates the temporal reliability of trial-by-trial responses to stimuli for individual NCM neurons.

2.7 |. Statistical analysis

We tested the competing hypotheses that either electrophysiological responses in NCM are different across ages or that electrophysiological responses in NCM are already adult-like in early development. All statistical analyses were conducted in R (R Core Team, 2018). We needed to account for the non-independence of points in our hierarchical dataset because we recorded from multiple neurons within the same individual. To do this, we used a cluster bootstrapping statistical approach similar to that described in (Saravanan et al., 2020). This approach reduces Type I error rate compared to traditional methods (e.g., Student’s t-tests) but retains statistical power with low sample sizes or small effect sizes (Huang, 2018).

We performed analysis using the means of the responses of each unit to each stimulus. Using the rms package (Harrell, 2019), we created a separate global model for response variables (spontaneous firing rate, stimulus-evoked firing rate, Z-score, selectivity, and accuracy) including stimulus species, age, hemisphere, unit type, and age*stimulus and age*unit type interactions as fixed effects. We then conducted an information-theoretic model selection process in MuMIn using AIC values (Barton, 2019) to determine which fixed effects to retain in the final model. With the reduced model, we performed the hierarchical bootstrap using the subject ID as a cluster variable and age as a grouping variable with 1,000 replications to calculate parameter estimates and standard errors. Post hoc tests following bootstrapping were conducted with the emmeans package (Lenth, 2020), which uses the Tukey method to correct p-values for multiple tests.

Sex was not included in the above statistical analyses to allow the inclusion of units from the two juveniles that we were not able to reliably classify. The effect of sex and the interaction between age and sex were tested separately for each response variable using the same statistical approach as above. For these analyses, we combined all juveniles together to account for the absence of data from female fledglings.

To assess the rate and magnitude of stimulus-specific adaptation in juveniles and adults, we used Kruskal-Wallis rank-sum tests to compare adaptation slopes and percent differences between ages. We combined all juveniles for these analyses to account for the small sample size and because nestling and fledglings had similar responses. We also conducted paired Wilcoxon rank-sum tests to compare median responses between the first five and last five stimulus presentations separately for juveniles and adults to determine if adaptation occurred in both age groups. All values reported in the Results section represent mean ± SE.

3 |. RESULTS

3.1 |. Single unit activity in NCM is overall similar across age groups

We found that subjects of all ages were highly responsive to auditory stimuli (Figure 2b). In all three age groups, we found distinct unit types in similar proportions corresponding to putative inhibitory interneurons (NS units; n = 99 in nestlings, 55 in fledglings, and 170 in adults ) and excitatory neurons (BS units; n = 128 in nestlings, 80 in fledglings, and 205 in adults) that could be separated based on waveform and response behavior as described in previous studies (Calabrese & Woolley, 2015; Miller-Sims & Bottjer, 2014; Ono et al., 2016; Yanagihara & Yazaki-Sugiyama, 2016). Response characteristics of these unit types were largely consistent for firing rates, Z-scores, accuracy, and selectivity regardless of age.

Our final model for spontaneous firing rate included age, hemisphere, unit type, and an age*unit type interaction as fixed effects and subject ID as the cluster variable. Only age, unit type, and subject ID were included for stimulus-evoked firing rate. Combining all subjects, NS units had higher spontaneous (BS: 4.74 ± 0.49 Hz; NS: 10.04 ± 1.18 Hz; Z = −3.24, p < .01; Figure 3a) and stimulus-evoked (BS: 11.41 ± 1.04 Hz; NS: 25.20 ± 2.34 Hz; Z = −5.64, p < .001; Figure 3b) firing rates than BS units, but the difference in spontaneous firing rates between unit types within age groups was highest in adults (nestling: Z = −1.44, p = .15; fledgling: Z = −0.88, p = .4; adult: Z = −3.07, p < .01); Figure 3a). There were no differences in firing rates between age groups, although firing rates tended to increase with age (Figure 3a,b). Our confidence that we would be able to detect age-dependent differences is driven by the observation that model parameter estimates (differences between means) were 1.4-fold lower between age groups than within age group unit type differences for spontaneous firing rates and 4.4-fold lower for stimulus-evoked firing rates.

FIGURE 3.

Spontaneous and auditory-evoked response firing activity is similar across ages in zebra finches. NS units have higher spontaneous (a) and stimulus-evoked firing rates (b) than BS units when combining all ages, although this difference in unit types is not significant for spontaneous firing rates within juveniles. (c) NS units have higher Z-scores than BS units both overall and within ages. There is an interaction of age and sex on Z-score, with differential responses across sexes in juveniles (nestlings and fledglings combined) but not adults (d). Units from individuals of undetermined sex are not included in (d). Colored dots represent mean responses from each single unit. Black shapes and error bars show the group mean ± SE. ***p < .001; **p < .01; *p < .05; ^#p < .1

In order to compare responses to auditory stimuli, while accounting for differences in firing rate across units, we analyzed auditory Z-scores. The fixed effects of age, unit type, and the age*unit type interaction were included in the final model, along with the cluster variable of subject ID. NS units had higher Z-scores than BS units both overall (Z = −4.30, p < .001) and within groups (fledglings: Z ratio = −5.55, p < .001; adults: Z ratio = −3.70, p < .001; Figure 3c) but only marginally within nestlings (Z ratio = −1.93, p = .05). There were no differences in Z-scores across development, although Z-scores were superficially lowest in adults (Figure 3c). Model parameter estimates were 2.3-fold lower between age groups than within age group differences among BS and NS units. We found only one difference in Z-scores among age groups. There was a weak age*sex interaction (p = .08) where male juveniles (nestlings and fledglings combined) tended to have higher Z-scores than female juveniles but adults showed no sex difference (Figure 3d).

We assessed the selectivity of single units by calculating a proportion of stimuli that induced a significant change in spike activity between baseline and stimulus periods. In addition to the subject ID cluster variable, the model selection process retained only unit type and hemisphere as fixed effects in the final model. Thus, there were no differences in selectivity among age groups. There was a clear unit type effect where BS units were more selective than NS units (Z = 2.83, p < .01; Figure 4a). Owing to these characteristic response properties of BS and NS units, there was a negative correlation between selectivity and Z-score, but this relationship was also noted within unit types (BS linear regression R² = 0.40, p < .001; NS linear regression R² = 0.35, p < .001). We also found that the left hemisphere was more selective than the right hemisphere across all subjects (Z = −2.63, p < .01; Figure 4b). This hemisphere difference may be driven in part by the curious fact that we only recorded from BS units in the left hemisphere of fledglings, and BS units have high selectivity. However, the hemisphere pattern is also shown within nestlings and adults where this unit type recording bias did not exist.

FIGURE 4.

Zebra finches show similar patterns of selectivity and stimulus-specific adaptation across all ages. (a) BS units are overall more selective than NS units, and this pattern is shown in all age groups. (b) Selectivity is lateralized. Regardless of age, left hemisphere units are more selective than right hemisphere units. Size of colored circles in (a) and (b) represent the number of units with the same X- and Y-value. Black shapes and error bars show the group mean ± SE. Minimum adaptation slopes (c) and the percent difference in response between the first five and last five stimulus presentations (d) are not different across ages. Values in (c) and (d) indicate mean ± SE. Responses to the first five stimulus presentations were significantly different than responses to the last five stimulus presentations in both juveniles (nestlings and fledglings combined) and adults. Statistical tests for (c) and (d) did not include stimulus, but stimulus is shown here for qualitative interpretation. No juvenile units adapted most strongly to the Bengalese finch stimulus. **p < .01

Stimulus-specific adaptation occurred at all ages in the small subset of units analyzed. There was no difference between juvenile and adult slopes (Kruskal-Wallis chi-squared = 0.9, df = 1, p = .3), and slopes for both age groups were statistically different from zero (juveniles: V = 9, p < .01; adults: V = 8, p < .001; Figure 4c). For these same unit-stimulus combinations, the median response during the first five stimulus presentations was higher than the median response during the last five stimulus presentations in both juveniles (V = 118, p < .01) and adults (V = 117, p < .001). The percent difference between the response to the first five and last five stimuli was not different across age groups (Kruskal-Wallis chi-squared = 0.51, df = 1, p = .5; Figure 4d). Although we did not test for differences in stimulus-specific adaptation across stimuli due to the small sample size, we note that, qualitatively, responses in both juveniles and adults decreased more with repeated presentations of white noise and owl finch song than zebra finch and Bengalese finch song (Figure 4c,d).

Classification accuracy also showed an effect of unit type. Accuracy was higher for NS units across groups (Z = −3.47, p < .001; Figure 5a), but within an age group unit types differed only in adults (nestlings: p > .1; fledglings: p = .06; adults: Z ratio = −5.06, p < .001). There were age*unit type interactions, but post hoc analyses showed that the only pairwise difference was between fledgling BS units and adult BS units (Z ratio = −2.83, p = .01). We also noted differences across ages and sexes in classification accuracy. There was a marginal age*sex interaction effect on accuracy (p = .06) where accuracy tended to be higher in males within juveniles but higher in females within adults (Figure 5b). Notably, accuracy was also higher in fledglings than adults (Z ratio = −2.68, p < .02; Figure 5c), mainly driven by a difference between the accuracy of BS units between those age groups. There was a strong tendency for units with higher firing rates to also have higher accuracy scores (linear regression R² = 0.45, p < .001), but that does not fully explain the age differences in accuracy shown here because adults and fledglings have stimulus-evoked firing rates that are statistically indistinguishable. On the whole, the coding accuracy of units in nestlings and fledglings were on par with, or in some cases exceeding, that observed in adults.

FIGURE 5.

Accuracy is higher in fledgling birds than in adults, but spike response patterns are related to the species of playback stimuli across all ages. (a) NS units have higher accuracy scores than BS units when all ages are combined, but this difference is only shown in adults when comparing within ages separately. (b) Higher accuracy scores are associated with males within juveniles (nestlings and fledglings combined) but females within adults, although these differences within age groups are not statistically significant. (c) Classification accuracy is higher in fledglings than in adults. (d) Across all ages, accuracy is higher in response to conspecific and Bengalese finch song, intermediate in response to owl finch song, and lowest in response to white noise. (e) Trial-by-trial reliability (R_corr) of temporal spike precision is higher in NS units than BS units, with the largest difference among unit types in adults. Colored dots represent mean responses from each single unit. Black shapes and error bars show the group mean ± SE. Dashed lines in all plots represent chance level classification accuracy (1/number of stimuli, normalized to 0). ***p < .001; **p < .01; *p < .05; ^#p < .1

Our second measure of spike timing precision showed a similar pattern. Age, hemisphere, unit type, and age*unit type were included in the final model with the subject ID cluster variable. NS units had higher trial-by-trial reliability than BS units overall (Z = −4.97, p < .001; Figure 5e), but this difference was most pronounced in adults (nestlings: p = .4; fledglings: p = .2; adults: Z ratio = −3.16, p < .01). Reliability tended to increase with age, although post hoc analysis of age*unit type interactions revealed that the main difference among groups is between nestling and adult NS units (Z ratio = 2.89, p = .01).

3.2 |. NCM responses to species-specific song show the same pattern regardless of age

Stimulus-evoked firing rates, Z-scores, and trial-by-trial reliability did not show significant discrimination of stimuli at any age since stimulus was not retained in these models as a fixed effect. However, classification accuracy was strongly related to the species of the playback stimuli. Accuracy was highest and equivalent to both conspecific and Bengalese finch song, intermediate to owl finch song, and lowest for white noise (conspecific-Bengalese: p > .05; conspecific-owl: Z ratio = 4.32, p < .001; conspecific-white noise: Z ratio = 9.29, p < .001; Bengalese-owl: Z ratio = 2.55, p = .05; Bengalese-white noise: Z ratio = 6.67, p < .001; owl-white noise: Z ratio = 3.24, p < .01; Figure 5d). This pattern of responses was apparent at all ages and was unrelated to the stimulus-evoked firing rate (Figure 3b). Therefore, units in all age groups exhibited slightly higher coding accuracy for songs than white noise, consistent with patterns observed in other electrophysiological studies of adult zebra finches.

4 |. DISCUSSION

This study demonstrates that electrophysiological responses to auditory stimuli are adult-like in nestlings well before the sensitive period of motor song learning opens, and remarkably before the auditory system has completed development. We found BS and NS units in nestling NCM that behave like those in adults. NS units had higher firing rates, Z-scores, and trial-by-trial temporal firing reliability but lower selectivity than BS units across age groups. The NCM in all ages exhibited stimulus-specific adaptation, while classification accuracy was slightly higher in fledglings than adults. Furthermore, classification accuracy was significantly related to stimulus across groups, with nestlings and fledglings showing the same pattern of coding accuracy for songs versus white noise as adults. This is the first study to explore electrophysiological responses to auditory stimuli in the nestling forebrain.

4.1 |. State of secondary auditory processing in young songbirds

Our results, in combination with previous studies, indicate that very few shifts occur in NCM encoding across development. Processing in NCM is already well-established before the earliest point that extracellular recordings become possible for head-fixed recording methods. We observed that spike waveform type predicted electrophysiological properties across ages, and our results match previously reported differences between BS and NS units with the exception of an opposite pattern in spike timing reliability to that found by Bottjer et al. (2019). We found that NCM firing rates and Z-scores remain similar from nestling to adult, although there may be an uptick in spontaneous firing rate and a decrease in Z-score during the sensorimotor stage (Vahaba et al., 2017), and habituation and selectivity for different stimulus categories change from juvenile to adult (Miller-Sims & Bottjer, 2014; Stripling et al., 2001). Nevertheless, selectivity for natural auditory stimuli, such as songs, is a feature that is present in NCM throughout development (e.g., Chew et al., 1995; Stripling et al., 2001; Vahaba et al., 2017). Even embryos and nestlings show a nonsignificant trend toward conspecific selectivity in immediate early gene ZENK expression in the auditory forebrain (Rivera et al., 2019).

In recently hatched songbirds, prior studies set up the expectation that neural responses to auditory stimuli are still developing and that auditory experience has no effect. Hearing sensitivity is not adult-like until around 20 dph (Amin et al., 2007; Brittan-Powell & Dooling, 2004), habituation rates change across development (Miller-Sims & Bottjer, 2014; Stripling et al., 2001), ZENK expression in auditory regions is not selective for conspecific song until 30 dph (Jin & Clayton, 1997; Stripling et al., 2001), and individuals do not memorize songs, for production or during operant conditioning, before the sensitive period opens at ~25 dph (Braaten, 2010; Roper & Zann, 2006). However, there are a growing number of studies indicating that prenatal acoustic experience can have postnatal effects (Katsis et al., 2018; Mariette & Buchanan, 2016) and that embryos and nestlings already have the ability to habituate to and discriminate among specific songs (current study; Colombelli-Négrel et al., 2014; Kleindorfer et al., 2018) and even to learn and produce certain call elements (Colombelli-Négrel et al., 2012; Langmore et al., 2008; Madden & Davies, 2006). For example, cross-fostered superb fairy-wren chick begging calls closely match the incubation calls of their foster mothers at hatching, presumably as a mechanism to detect unsimilar brood parasitic chicks (Colombelli-Négrel et al., 2012). Therefore, while it is true that the nestling auditory system either is not completely developed and/or is lacking the experience that shapes some neural mechanisms in adults, these limitations do not preclude complex auditory processing and online recognition memory in the auditory forebrain of very young songbirds.

At the same time, while behaviors and NCM responses to song seem to be largely mature quite early in development, this may not the case for all auditory forebrain regions. For example, Amin et al. (2007) showed that auditory responsiveness in Field L and CLM and selectivity for natural sounds in Field L were reduced in 35 dph zebra finches relative to adults. Given these developmental differences at earlier processing levels, it is curious that we found that NCM does not show the same differences. Of course, auditory representation within these regions could be independent (for example, we note the opposite relationship between selectivity and Z-score to that found by Amin et al. (2007)). Perhaps the developmental mismatch between studies can also be explained by differences in age, stimuli, and anesthetized versus awake recording procedures across studies. Particularly, while we did not test any individuals during the sensitive period of song learning, several studies show changes in NCM properties between ~20–60 dph (Jin & Clayton, 1997; Kudo et al., 2020; Miller-Sims & Bottjer, 2014; Stripling et al., 2001; Vahaba et al., 2017). This indicates that the sensitive period (including ages tested by Amin et al. (2007)), owing to its importance for tutor song acquisition, represents a unique time of increased circuit tuning and high plasticity within the auditory system, while perhaps neural properties both before and after this period (ages tested in the current study) are largely similar and unchanging.

There has been little exploration into the timing of structural and mechanistic changes that occur in the brain of juvenile songbirds, but evidence in precocial chickens shows that the auditory system develops relatively early. Development of the chicken (hatch on embryonic day E21) auditory system starts by E7 in the ear (Cohen & Fermin, 1978), and behavioral responses to auditory stimuli are apparent by E16 (Jackson & Rubel, 1978). Songbirds are altricial, and developmental processes in altricial species lag behind those in precocial species (Kubke & Carr, 2000). Still, sparse data in songbirds suggest at least that the auditory forebrain is largely developed before the sensitive period—cell density in NCM of 20 and 30 dph zebra finches is similar to, or even greater than, that in adults (Stripling et al., 2001; Vahaba et al., 2020)—and is mature before the song system. In song motor nuclei, neurogenesis, neural responses, and N-methyl-D-aspartate receptor (NMDAR) expression change significantly during the sensitive period into adulthood (Burek et al., 1991; Nordeen & Nordeen, 2004; Whaling et al., 1997). Since song learning and production necessarily require input to motor regions from the auditory forebrain, the developmental timeline follows expectation in this case.

In our data, we noted an interaction effect of age and sex on Z-score and classification accuracy, suggesting that auditory processing may occur differently in males and females across development. While some sex differences in firing frequency and cell-type-specific responsiveness have been reported in adult NCM (Krentzel et al., 2018), most studies do not report sex differences in the NCM of juveniles. Bailey and Wade (2003), however, noted that auditory forebrain ZENK expression showed selectivity for conspecific over heterospecific stimuli in 30 dph males, but 30 dph females show conspecific selectivity only in a different immediate early gene, FOS. This sex difference is no longer present at 45 dph (Bailey & Wade, 2005). Kudo et al. (2020) also report that isolation from tutors also has differential effects on developmental changes in the proportion of burst-type neurons in NCM of males and females. Perhaps the reason for the lack of evidence for sex differences in NCM is that many studies do not address its role through development. Therefore, age and sex comparisons could be informative in future studies.

We found age differences in our two measures of temporal coding. Trial-by-trial reliability tended to increase with age, a pattern that was more pronounced in NS than BS units (putative inhibitory interneurons and excitatory neurons respectively; Calabrese & Woolley, 2015; Miller-Sims & Bottjer, 2014; Ono et al., 2016; Yanagihara & Yazaki-Sugiyama, 2016). This suggests that increased reliability of temporal coding during development and auditory experience may be achieved through changes in inhibition rather than excitation. Classification accuracy was also higher in fledglings than adults. Higher classification accuracy in sensory-aged birds may be related to the memorization of tutor songs. Vahaba et al. (2017) found that classification accuracy was higher during the sensory stage than the sensorimotor stage, and our results indicate that individuals just entering the sensory stage (fledglings) have higher accuracy scores than adults. Thus, this peak in fidelity of the single-unit spike patterns just before and during the sensory stage coincides with the timing of when a tutor song memory is forming. NCM has been suggested as a potential location of tutor song memory (Bolhuis et al., 2000; Phan et al., 2006; Terpstra et al., 2004), and NCM neurons become highly selective for tutor song after only a few days of experience (Yanagihara & Yazaki-Sugiyama, 2016). Future experiments should work to determine if tutor experience increases classification accuracy or, conversely, if a developmental shift in accuracy enhances the formation of tutor song memory in NCM.

Our data do not distinguish innate from experience-shaped mechanisms. Our birds were at least 2 weeks old before testing and we did not control pre-testing auditory experience. All birds were raised in flight aviaries with many adult conspecific males and females, thus auditory exposure both in ovo and immediately after hatching may have already begun tuning neural circuits. Future studies that follow response properties and tuning of individual auditory forebrain neurons through development while controlling auditory experience will help elucidate which processes are in place at hatching and which mature over time or within certain experiential contexts.

Our study does not address the issue of hearing sensitivity in nestlings, although we do not believe that this presents a problem for the current dataset. First, while Amin et al. (2007) concluded that responses in the brainstem were adult-like at 20 dph but not 10 dph, they found no difference in auditory forebrain responses of 35 dph birds to stimuli of 70 dB and 85 dB. Therefore, the sound level is important for very young finches in brainstem processing. Even so, the sound levels that nestlings experienced during this study are squarely within the natural experience range. Male zebra finch songs are generally between ~70–90 dB (Brumm, 2009; Zollinger et al., 2011) and can reach 109 dB when birds are singing in high levels of background noise (Zollinger et al., 2011). Wild zebra finch males sing 70% of undirected song within 1 m of the nest during the nesting period (Dunn & Zann, 2010), thus wild nestlings are, indeed, exposed to loud songs at close range, comparable to our experimental conditions. That said, we did not thoroughly characterize the rate-level functions for NCM neurons and how they differ across ages. Since NS and BS neurons in adults clearly have differences in level coding (Bottjer et al., 2019), it will be important in the future to analyze the way forebrain neurons respond to varying sound amplitudes across development.

4.2 |. Auditory responses to species-specific songs

Zebra finches across all ages in our study showed higher classification accuracy to zebra finch and Bengalese finch song than owl finch song, but higher accuracy for all songs over white noise. There was also a trend for faster habituation to owl finch song and white noise. This finding is in alignment with previous studies showing that NCM is selective for conspecific song (Louder et al., 2016; Mello et al., 1992; Stripling et al., 2001). We did not find differences in firing rate or Z-scores in response to species-specific stimuli, in accordance with Stripling et al. (1997), instead the stimulus category was distinguished here by the consistency of firing patterns to individual stimuli.

We find it interesting that neural responses did not distinguish zebra finch and Bengalese finch song but did respond differentially to owl finch song. The fact that this same pattern occurred at all ages and no subjects had prior experience with heterospecific songs suggests that this response pattern is either innate, as species recognition appears to be in birds that do not learn vocalizations (e.g., Long et al., 2001; Wheatcroft & Qvarnström, 2017), or due to similarity of spectral features in zebra and Bengalese finch songs. While songs of the two species are within a similar frequency range, they differ in many features including the distribution of syllable duration, repetition of syllables (Araki et al., 2016), and spectrotemporal modulation (Woolley et al., 2010). This suggests that temporal and frequency related similarities in species-specific songs do not explain our results.

Therefore, we suggest that zebra finches are born with the ability to distinguish between conspecific and owl finch songs but cannot separate conspecific and Bengalese songs, at least at the level of NCM. Perhaps this is the result of the difference in ecological relevance of the two heterospecific species. Owl finches are more closely related to zebra finches and their range partially overlaps the zebra finch range in the wild (Payne, 2020a), unlike the range of the white-rumped munia, the Bengalese finch’s wild counterpart (Payne, 2020b). Selective pressure in wild populations should, therefore, favor the ability to distinguish owl finch song to avoid hybridization and territorial dispute, but there would be no benefit to recognizing Bengalese song. Remnants of this ecological pressure on wild ancestors may have shaped the response patterns seen in our captive zebra finches.

Another possibility is that selective pressure to maintain species recognition is relaxed in captive populations. For example, captive Bengalese finches have more complex songs than their wild counterparts (Honda & Okanoya, 1999) and are more likely to learn notes from heterospecific tutors (Takahasi & Okanoya, 2010). Suzuki et al. (2014) suggest that this could have occurred due to the lack of natural selection on captive finches to learn and maintain species-specific characteristics in their songs. It is possible that relaxed selection on domesticated zebra finches has similarly resulted in reduced discrimination of Bengalese finches, and zebra finch discrimination of owl finch song may simply be changing at a slower rate.

5 |. CONCLUSION

The processing, encoding, and learning of complex auditory signals occur in songbirds before hearing sensitivity is mature and before the sensitive period of motor song learning begins. Here, we provide one of the first snapshots into the progression of neural mechanisms in pre-sensory auditory forebrain. Our finding—that electrophysiological properties and responses to species-specific songs in NCM are already adult-like in nestlings—further reinforces recent evidence showing that songbirds can perceive and utilize information in auditory signals very early on. Further exploration of neural and behavioral responses to song in young birds will further inform how we frame perceptual learning, vocal learning, and the sensitive period in songbirds.