Deafening drives cell type-specific changes to dendritic spines in a sensorimotor nucleus important to learned vocalizations

Summary

Hearing loss prevents vocal learning and causes learned vocalizations to deteriorate, but how vocalization-related auditory feedback acts on neural circuits that control vocalization remains poorly understood. We deafened adult zebra finches, which rely on auditory feedback to maintain their learned songs, to test the hypothesis that deafening modifies synapses on neurons in a sensorimotor nucleus important to song production. Longitudinal in vivo imaging revealed that deafening selectively decreased the size and stability of dendritic spines on neurons that provide input to a striatothalamic pathway important to audition-dependent vocal plasticity, and changes in spine size preceded and predicted subsequent vocal degradation. Moreover, electrophysiological recordings from these neurons showed that structural changes were accompanied by functional weakening of both excitatory and inhibitory synapses, increased intrinsic excitability, and changes in spontaneous action potential output. These findings shed light on where and how auditory feedback acts within sensorimotor circuits to shape learned vocalizations.

Introduction

Auditory feedback is critical for learning and maintaining complex motor skills ranging from musical performance to speech. For example, hearing loss prevents speech learning in children and degrades speech in adults (Petitto, 1993; Waldstein, 1990). Despite the central role of auditory feedback in these vocal processes, where and how auditory feedback acts on neural circuits for vocal learning and maintenance remain poorly understood.

Birdsong provides a highly tractable system for understanding the neural mechanisms by which auditory feedback affects vocalization. Juvenile songbirds use auditory feedback to learn their songs (Konishi, 1965; Marler, 1970; Scharff and Nottebohm, 1991), and the adults of some species use feedback to maintain stable songs (Leonardo and Konishi, 1999; Nordeen and Nordeen, 1992). Indeed, deafening adult zebra finches, which sing highly stereotyped songs, causes the spectral and temporal features of their songs to degrade over days to weeks (Horita et al., 1998; Lombardino and Nottebohm, 2000; Nordeen and Nordeen, 1992). Thus, the stereotypy of adult song enables precise quantification of the vocal effects of deafening, and the delayed effects of deafening on adult song allow for resulting neural changes to be temporally correlated with either initial removal of feedback or later changes to the vocal pattern.

Songbirds are also advantageous for studying how feedback affects vocalization because the neural circuits for singing (i.e., the song system) are well described. Although feedback must act on the song system to affect vocalization, the sites of this sensorimotor interaction remain unclear. One potential site of interaction is the song system nucleus HVC, which is necessary for singing (Nottebohm et al., 1976), contains neurons that exhibit precisely time- locked activity when the bird is producing or passively listening to song (Fujimoto et al., 2011; Hahnloser et al., 2002: Prather et al., 2008), and is the only song system nucleus known to receive direction projections from auditory areas (Bauer et al., 2008; Cardin et al., 2005; Coleman et al., 2004; Vates et al., 1996). Moreover, HVC includes two projection neuron (PN) types, one (HVC_RA) that innervates the song premotor nucleus RA, and another (HVC_X) that provides auditory and song motor-related input to the anterior forebrain pathway (AFP), a striatothalamic circuit necessary for audition-dependent vocal plasticity in adult zebra finches (Andalman and Fee, 2009; Brainard and Doupe, 2000; Williams and Mehta, 1999). These features advance HVC PNs, and particularly HVC_X neurons, as sites where feedback-related information could access circuitry important for song learning and maintenance.

Despite the attractiveness of this idea, disrupting feedback with singing-triggered noise, which can drive song degradation over hours or days, fails to alter the singing-related activity of HVC_X neurons, at least over tens of minutes (Kozhevnikov and Fee, 2007; Prather et al., 2008). Interestingly, feedback perturbation in zebra finches immediately alters the singing-related activity of neurons that supply direct or indirect auditory input to HVC (Keller and Hahnloser, 2009; Lei and Mooney, 2010) and, in Bengalese finches, can acutely suppress the singing-related activity of putative HVC interneurons (Sakata and Brainard, 2008), some of which may provide inhibitory input to HVC_X neurons (Mooney and Prather, 2005). These different findings could be reconciled by a model in which HVC_X neurons accumulate feedback information slowly (hours to days) and where feedback-driven changes in these cells first appear as a subtle modification of synaptic input, rather than changes in action potential output. Testing this model requires a way of longitudinally monitoring synapses on identified neurons before and after manipulation of auditory feedback changes song output, a goal currently impractical to achieve using electrophysiological methods.

In vivo, multi-photon imaging of fluorescently labeled neurons can resolve individual dendritic spines, which are postsynaptic components of excitatory synapses in the vertebrate bra in (DeRobertis and Bennett, 1955; Palay, 1956), and this method has been used in a variety of longitudinal studies to measure experience-dependent changes to synapses (for reviews, see Alvarez and Sabatini, 2007, and Holtmaat and Svoboda, 2009). Recently, this method has also been used to show that auditory experience of a vocal model stabilizes and strengthens HVC dendritic spines in juvenile songbirds over a period of days (Roberts et al., 2010), advancing it as a suitable method for detecting relatively slow feedback-related changes to synapses in the HVC of adult songbirds. Here, we used longitudinal in vivo two-photon imaging of dendritic spines in deafened adult zebra finches to test the idea that synapses on HVC PNs are sensitive to changes in auditory feedback.

Results

Deafening causes a cell type-specific decrease in the size of dendritic spines in HVC

To label and identify HVC projection neurons for in vivo imaging, a GFP-lentivirus was injected into HVC, and differently-colored retrograde tracers were injected into the two downstream targets of HVC, the striatal region Area X and the song premotor nucleus RA, in young adult male zebra finches (Figure 1A, Supplemental figure 1A; 80 to 150 days post-hatch (dph), mean age was 97 ± 5 days, all reported errors are s.e.m. unless otherwise noted). Birds were maintained on a reverse day-night cycle and imaging sessions were conducted during the birds’ subjective nighttimes, to minimize interference with singing behavior (2 sessions per night separated by a 2h interval). Images were obtained through a cranial window and collection of imaging data was restricted to neurons with dendritic spines, because both populations of HVC PNs are spinous (Mooney, 2000). Neurons were identified as either HVC_X or HVC_RA cells by the presence of blue or red retrograde label or, in the absence of retrograde label, by the measurement of soma size, which differed significantly for the two PN types (Figure 1A, Supplemental figure 1B). After collecting 1–2 nights of baseline imaging data, birds were deafened by bilateral removal of the cochleae, and data collection was continued as long as possible (13 birds were imaged for an average of 7.2 ± 4.1 nights post-deafening).

Figure 1. Deafening causes a cell type-specific decrease in the size of dendritic spines in HVC_X neurons.

A) Examples of a lentivirally-labeled HVC_RA neuron (top, arrowhead) and HVC_X neuron (bottom), double-labeled with red and blue retrograde tracers, respectively. Scale bars, 20 μm. B) Representative images showing spine size measurements (see Experimental Procedures) from an HVC_X neuron before and after deafening. Spines tend to increase in size over 24h intervals prior to deafening (size index > 1) and decrease in size following deafening (size index < 1). Scale bar, 5 μm. Note that image contrast has been enhanced for figure presentation; adjustments to brightness and contrast were not made to images used for data analysis. Although the size of all clearly visible spines was scored, for clarity, the scores for only a few representative spines are indicated in this and subsequent figures. C) Deafening causes a cell type-specific decrease in spine size index in HVC_X neurons (p = 0.03, Wilcoxon signed-ranks test) but not in HVC_RA neurons (p = 0.67). Numbers in parentheses indicate the total number of each PN type analyzed. See also Supplemental figure 1.

An initial assessment was made of the effects of deafening on the size of spines in HVC PNs, because changes in spine size often correlate with functional changes in synaptic strength (Kopec et al., 2006; Matsuzaki et al., 2004; Okamoto et al., 2004; Roberts et al., 2010; Zhou et al., 2004). Size measurements were made from spines that were maintained across two nights of imaging (over a 24h interval), and a size index was calculated for each measured spine (time 24 size/time 0 size), with values greater than 1 indicating an increase in size and values less than 1 indicating a decrease in size. Prior to deafening, spines in HVC_X neurons tended to increase slightly in size, while spines in HVC_RA neurons tended not to change in size over 24h (size index = 1.07 ± 0.03 for HVC_X neurons: 106 spines, 10 cells, 9 birds; size index = 1.00 ± 0.02 for HVC_RA neurons: 94 spines, 9 cells, 8 birds; p = 0.05 for difference between cell types, Mann Whitney U test). Interestingly, comparing spine size measurements made in a subset of these cells during the first 24h time window to those obtained in the last 24h time window following deafening (7–8 nights post-deafening on average) revealed that spine size index decreased significantly following deafening in HVC_X but not HVC_RA neurons (example images in Figure 1B; group data in Figure 1C; HVC_X: average of 10.8 ± 0.3 spines scored per cell in each 24h comparison, total of 152 spines from 7 cells in 6 birds, p = 0.03, Wilcoxon signed-ranks test; HVC_RA: average of 11.3 ± 0.4 spines scored per cell in each 24h comparison, total of 146 spinesfrom 8 cells in 6 birds, p = 0.67). Thus, deafening causes a cell type-specific decrease in the size of spines of HVC_X neurons.

Decreases in HVC_X neuron spine size precede and predict vocal changes following deafening

Establishing when these structural changes occur relative to deafening-induced song degradation depends on detecting initially subtle vocal changes following deafening. To this end, we analyzed two spectral features, Wiener entropy and entropy variance (EV), of each syllable in a bird’s song over time (see Experimental Procedures). These parameters respectively measure the uniformity of a sound’s power spectrum and intra-syllabic transitions from tonal to broadband sounds (Tchernichovski et al., 2000) and were chosen because they remain stable in hearing adults (Supplemental figure 2A), change in predictable directions following deafening (Supplemental figure 2B), and were found to be the earliest spectral features that changed following deafening (data not shown). This analysis detected subtle but significant effects of deafening on syllable spectral features in nearly all birds (18/19) within the first four days that they sang following deafening, with ~ 50% (10/19) of birds showing significant degradation over the first day of singing after deafening (Figure 2A–C). Notably, the changes we detected occur days to weeks earlier than those reported in previous studies (Brainard and Doupe, 2000; Horita et al., 2008; Lombardino and Nottebohm, 2000; Nordeen and Nordeen, 1992), demonstrating that this analysis method is extremely sensitive to early deafening-induced song changes. The number of days to onset of this subtle degradation was not predicted by the age of the bird (Figure 2C), although older birds sang a larger number of motifs before their songs degraded (Figure 2D; see Supplemental Experimental Procedures). Finally, the effects of deafening on syllable sequencing occurred later than spectral changes in all birds (data not shown; see Experimental Procedures), indicating that measurement of spectral features serves as the most reliable early marker of deafening-induced song degradation.

Figure 2. Analysis of song spectral features reveals that significant song degradation occurs within several days after deafening.

A) Example of syllable-by-syllable analysis carried out for a bird whose song contains four syllables, A–D. Although syllable entropy and EV were both analyzed, only EV is shown for clarity. Song behavior is shown for two pre-deafening days (−2 and 0) and post-deafening days 1, 2, 4, and 9. Red arrowhead indicates the first syllable to undergo significant spectral degradation (syllable C, shown in red, defined as onset of degradation using one-way ANOVA). B) Representative spectrograms of the same song analyzed in A. Syllable C (red bars) underwent degradation starting at 2d post-deafening. EV values are included below each rendition of the syllable. Frequency range (y axis), 0.4–9 kHz, scale bar (white bar, bottom spectrogram), 150 ms. C) The onset of song degradation (day of post-deafening song) was not related to the age at which birds were deafened (p = 0.82, linear regression). D) Older birds sang a larger number of motifs before their songs underwent significant degradation (p < 0.01, linear regression). These data come from a subset of the birds analyzed in C (8/19). See also Supplemental figure 2.

The onset of song degradation estimated in this manner was used to temporally align in vivo imaging data collected from different birds. To facilitate comparison between HVC_X and HVC_RA neurons and take into account different pre-deafening values of spine size index, each cell’s last pre-deafening size index value was used to normalize its subsequent size index values (Supplemental Figure 3A, left and middle panels), and these normalized values were pooled separately for the two cell types (Figure 3A; Supplemental Figure 3A, right panel). Interestingly, these pooled comparisons revealed that spine size index of HVC_X neurons decreased prior to the onset of song degradation, whereas spine size index of HVC_RA neurons did not change before or after songs began to degrade (Figure 3A; HVC_X: average of 11.2 ± 0.4 spines scored per 24h comparison, total of 495 spines from 7 neurons in 6 birds; HVC_RA: average of 11.0 ± 0.3 spines scored per 24h comparison, total of 428 spines from 8 neurons in 6 birds, time > 0 is post-degradation). Although we also attempted to assess whether changes in HVC_X neuron spine size occurred prior to the onset of song degradation on a bird-by-bird basis, size index data from individual neurons were noisy (Supplemental Figure 3A), and decreases in size index were rarely significantly different from baseline for individual cells. In summary, deafening causes a cell type-specific decrease in the size of spines in HVC_X neurons that on average precedes the onset of song degradation.

Figure 3. Deafening-induced decreases in spine size in HVC_X neurons precede song degradation.

A) Deafening causes a cell type-specific decrease in HVC_X spine size index that occurs prior to the onset of song degradation (time > 0 is post-degradation). Data are grouped into 4 time bins (2 bins before onset of song degradation and 2 bins after), the average time of deafening is just prior to night −2, and asterisks indicate a significant difference (p < 0.05, two-factor ANOVA) between the two PN types for a particular time bin. Numbers in parentheses indicate the total number of each PN type analyzed. B) Spines from HVC_X neurons and HVC_RA neurons from longitudinally-imaged, age-matched control birds do not exhibit decreases in size index. See also Supplemental figure 3.

The finding that deafening-induced decreases in HVC_X neuron spine size precede the onset of song degradation raises the possibility that spine size changes are predictive of subsequent changes in vocal behavior. To test this idea, we calculated the correlation between post-deafening HVC_X spine size index measurements and the amount of song degradation that occurred on the following day of singing (“day +1,” measured as % change from baseline entropy or EV of the first syllable to degrade). This comparison revealed a significant positive correlation, indicating that larger decreases in spine size index preceded more severe song degradation (Figure 4A, R = 0.57, p < 0.001, linear regression). In contrast, no significant correlations were observed between spine size index and behavior when similar comparisons were made to song behavior recorded on the day preceding the 24h size index measurement (day −1), the intervening day between the two nights of imaging (day 0), or greater than day +1 (Figure 4B). These findings support the idea that decreases in HVC_X neuron spine size index predict subsequent behavioral change with a ~12h time lag, rather than accompanying or following vocal changes.

Figure 4. Decreases in HVC_X spine size index predict the severity of song degradation on the subsequent day of singing.

A) Normalized spine size index measurements from HVC_X neurons were compared to the amount of song degradation on the following day of singing (day +1). The amount of song degradation on each day was defined for each bird by calculating the percentage decrease from the baseline value of entropy or EV for the syllable that degraded first. This comparison reveals that larger decreases in spine size index predict a larger magnitude of song degradation on the subsequent day (p < 0.001, linear regression). B) Comparisons were made between spine size index measurements from HVC_X neurons and the amount of song degradation on days preceding, intervening, or following the 24h size index measurement, and correlation coefficients for those comparisons are plotted. Only comparison of spine size changes to the following day’s song degradation yielded a significant relationship.

Various control measurements ensured that decreases in HVC_X neuron spine size index were unrelated to imaging methodology. First, decreases in HVC_X spine size index were not due to effects of longitudinal imaging, because HVC_X spine size index never underwent a significant decrease in longitudinally-imaged, age-matched hearing birds (Figure 3B; control HVC_X: average of 9.5 ± 0.3 spines scored per 24h comparison, total of 95 spines from 4 cells in 4 birds; control HVC_RA: average of 9.6 ± 0.5 spines scored per 24h comparison, total of 77 spines from 3 cells in 3 birds). Second, decreases in HVC_X size index were unrelated to variable sampling of dendritic branches over time (Supplemental figure 3B), ruling out the possibility that the spatial variability in spine sampling could account for decreases in HVC_X neuron spine size index. Finally, spine size decreased in slightly more than half of the individual spines (20/35) that were tracked for multiple nights following deafening (average of 6.6 ± 0.5 nights), indicating that decreases in size index were also unrelated to variable sampling of individual dendritic spines (Supplemental figure 3C). Interestingly, the change in size for individual spines was negatively and significantly correlated with their intial, pre-deafening size, suggesting that deafening preferentially weakens stronger excitatory synapses (Supplemental figure 3C; R = −0.44, p < 0.01, linear regression). A similar relationship was not observed for individual spines tracked from longitudinally-imaged HVC_X neurons in hearing birds (i.e., a smaller proportion of tracked spines decreased in size, and there was no relationship between initial spine size and subsequent change in size, R = −0.06, p = 0.81, data not shown). These various measurements are consistent with the idea that deafening selectively weakens synapses on HVC neurons that innervate a striatothalamic circuit necessary for audition-dependent vocal plasticity.

Deafening causes a cell type-specific decrease in spine stability

Because spine stability is a structural correlate of synaptic strength (De Roo et al., 2008; Engert and Bonhoeffer, 1999; Hofer et al., 2009; Maletic-Savatic et al., 1999; Nägerl et al., 2004) that can change in concert with spine size (Roberts et al., 2010), we also examined whether deafening destabilizes spines in HVC. Stable spines were defined as those that were maintained over a 2h interval (within night, see Experimental Procedures). Spine stability was relatively high in both cell types prior to deafening (HVC_X: 92.0 ± 1.6% spines stable over 2h, average of 56 ± 6 spines scored per 2h comparison, total of 731 spines from 13 cells in 8 birds; HVC_RA: 93.9 ± 1.0% spines stable over 2h, average of 79 ± 13 spines scored per 2h comparison, total of 789 spines from 10 cells in 7 birds; p = 0.33 for difference between PN types, Mann Whitney U test). However, tracking spine stability before and after deafening revealed that spine stability decreased in HVC_X but not HVC_RA neurons (Figure 5A–B; HVC_X: average of 55 ± 6 spines scored per 2h comparison, total of 3562 spines from 14 cells in 9 birds; HVC_RA: average of 63 ± 6 spines scored per 2h comparison, total of 3217 spines from 12 cells in 8 birds). This destabilization reflected increases in spine gain and loss (Figure 5C, both measures tended to increase, albeit non-significantly), consistent with our observation that deafening did not affect spine density in HVC_X neurons (data not shown). In contrast to the more rapid effects of deafening on spine size, however, deafening destabilized spines only after the onset of song degradation (Figure 5B).

Figure 5. Deafening causes a cell type-specific decrease in spine stability that follows the onset of song degradation.

A) Representative images showing spine stability measurements from an HVC_X neuron before and after deafening. ‘S’ indicates spines that were stable over 2h, ‘+’ indicates spines that were gained, and ‘−’ indicates spines that were lost. Although the stability of all clearly visible spines was scored, for clarity, the scores for only a few representative spines are indicated in this figure. Scale bar, 5 μm. B) Deafening causes a cell type-specific decrease in spine stability in HVC_X neurons that occurs after the onset of song degradation (time > 0 is post-degradation). Data are grouped into 4 time bins, the average time of deafening is just prior to night −2, and asterisks indicate a significant difference (p < 0.05, two-factor ANOVA) between the two PN types for a particular time bin. Numbers in parentheses indicate the total number of each PN type analyzed. C) Spine gain and spine loss both tend to increase in HVC_X neurons following deafening. D) Spines from HVC_X and HVC_RA neurons in longitudinally-imaged, age-matched controls do not exhibit decreases in stability. See also Supplemental figure 4.

Decreases in spine stability were not attributable to effects of longitudinal imaging, because HVC_X neurons from longitudinally-imaged, age-matched hearing birds never underwent a significant decrease in spine stability (Figure 5D; control HVC_X: average of 74 ± 13 spines scored per cell in each 2h comparison, total of 1964 spines from 6 cells in 4 birds; control HVC_RA: average of 51 ± 7 spines scored per cell in each 2h comparison, total of 1168 spines from 7 cells in 4 birds). Further, although there was a slight negative relationship between the variability of dendritic sampling and levels of spine stabilty (i.e., post-deafening measurements including dendritic segments that were not scored on the pre-deafening, baseline night tended to have lower stability values), subsequent resampling of the data to include only post-deafening measurements in which > 50% of the dendritic segments sampled were the same as those sampled in the baseline measurement did not support the idea that variability in spatial sampling accounts for decreased spine stability in HVC_X neurons (Supplementary Figure 4A–B). Thus, deafening decreases HVC_X spine size and stability, which are two structural correlates of synaptic weakening (Nägerl et al., 2004; Okamoto et al., 2004; Zhou et al., 2004), but these structural changes differ in when they first appear relative to the onset of song degradation.

We also conducted a series of additional measurements to ensure that the effects of deafening on spine size and stability in HVC_X neurons were not due to decreased levels of singing following deafening. First, in one bird that did not sing for the first week following deafening, a single HVC_X neuron that we imaged failed to undergo decreases in spine size and stability (Supplemental figure 4C). Thus, even a marked decrease in singing rate was not sufficient to decrease HVC_X neuron spine size and stability. Second, the correlation between HVC_X neuron spine size index measurements from each bird and the total number of motifs sung during the intervening day of behavior revealed a small, non-significant negative correlation (i.e., following deafening, HVC_X neuron spine size index was lower on days that birds sang more; data not shown). Third, we observed that some birds did not sing for the first 1–2 days following the deafening surgery. However, the mean spine size index of HVC_X neurons measured across post-deafening days when birds didn’t sing was not significantly different from the mean spine size index measured across baseline, pre-deafening 24h periods (mean HVC_X size index across post-deafening days without song, 1.03 ± 0.04; mean size index during pre-deafening baseline, 1.07 ± 0.03; p = 0.59, Mann Whitney U test). Additionally, alignment of the HVC_X spine size index measurements with the first day of post-deafening song (as opposed to alignment with the onset of song degradation, as shown in Figure 3A) revealed that decreases in HVC_X spine size index did not occur until after singing resumed following deafening (data not shown). Finally, longitudinally-imaged birds frequently exhibited decreased singing rates following the windowing surgery. Although this decrease was shorter-lived and reduced in magnitude as compared to birds that were also deafened (data not shown), HVC_X neurons imaged in these birds never showed decreases in spine size or stability (Figure 3B, Figure 5D). Thus, decreased singing rates cannot account for the structural changes to HVC_X neuron dendritic spines following deafening.

Deafening functionally weakens synapses on HVC_X neurons

To determine whether deafening-induced structural changes to HVC_X dendritic spines reflect functional changes in the strength of excitatory synapses on these neurons, sharp intracellular current-clamp recordings were made from HVC_X neurons in anesthetized adult male zebra finches several days after deafening, within the time range when structural changes to HVC_X dendritic spines were observed (16 HVC_X cells from 5 birds, mean age of 97 dph, ranging from 88–114 dph, recorded on average at 2.8 ± 0.8 days post-deafening). Similar recordings were also carried out in a second group of age-matched, hearing control birds (22 HVC_X cells from 14 birds, mean age of 105 dph, ranging from 88–143 dph). Tonic hyperpolarizing current injected into the impaled cell facilitated measurement of depolarizing postsynaptic potentials (dPSPs) without contamination from action potentials (Figure 6A, example traces shown in top panel, V_m = −87.5 ± 2.1 mV in deafened group, −85.8 ± 1.9 in control group, p = 0.78 for difference between groups, Mann Whitney U test). Deafening significantly decreased the amplitude but not the frequency of spontaneous dPSPs recorded in HVC_X neurons (amplitude: Figure 6A, lower left; p < 0.0001, KS test; frequency: lower right, p = 0.30, Mann Whitney U test). The mean decrease in median dPSP amplitude (16.4%) was comparable to the mean decrease in HVC_X spine size index observed between 1 and 4 nights post-deafening (11.2%), consistent with the interpretation that deafening weakens excitatory synapses on HVC_X neurons. However, because cells were hyperpolarized to values near or beyond the reversal potential for inhibitory synaptic currents (~−85 mV; evidenced here by the lack of hyperpolarizing PSPs), the dPSPs we measured likely represent a mix of excitatory and inhibitory synaptic events, and thus one or both types of synapses may be affected by deafening.

Figure 6. Deafening decreases the amplitude of spontaneous synaptic activity and alters spontaneous action potential activity in HVC_X neurons.

A) Top: example traces of depolarizing postsynaptic potentials (dPSPs) from HVC_X neurons recorded in current-clamp configuration in anesthetized hearing control and deafened birds. Deafening drives a significant decrease in the amplitude of spontaneous dPSPs (bottom left, p < 0.0001, KS test) but has no significant effect on dPSP frequency (bottom right, p = 0.30, Mann Whitney U test). B) Top: example traces of spontaneous action potential activity from HVC_X neurons recorded in anesthetized hearing control and deafened birds. Deafening drives a significant decrease in ISI duration (bottom left, p < 0.0001, KS test) but has no significant effect on mean spontaneous action potential frequency (bottom left, p = 0.25, Mann Whitney U test). Synaptic activity was measured while injecting tonic hyperpolarizing current into the recorded cell; action potential activity was measured without current injection. See also Supplemental figure 5.

To resolve the effects of deafening on excitatory and inhibitory synapses on HVC_X neurons, we made visualized whole-cell voltage-clamp recordings from retrogradely-labeled HVC_X neurons in brain slices prepared from 50–60 dph male zebra finches (Figure 7A; mean age was 53 ± 0.6 dph, 3 +/− 0.0 days post-deafening; 6 HVC_X neurons from 3 control birds; 6 HVC_X from 2 deafened birds; younger animals were used to ensure viable recordings). Recordings were made in pharmacological conditions that blocked voltage-gated sodium and potassium currents and at two different holding potentials (−70 mV and 0 mV) to isolate spontaneous miniature excitatory and inhibitory postsynaptic currents (mEPSCs and mIPSCs, respectively, see Experimental Procedures). These measurements revealed that deafening decreased the amplitude of both mEPSCs and mIPSCs (Figure 7; KS test: p < 0.01 for mEPSCs, p < 0.0001 for mIPSCs; Mann Whitney U test: p = 0.02 for mEPSCs, mean decrease in median value was ~8%; p < 0.0001 for mIPSCs, mean decrease in median value was 7%). In contrast, deafening had no effect on the frequency of mEPSCs or mIPSCs (data not shown). These data from brain slices closely parallel those from in vivo current-clamp recordings, providing further evidence that synapses on HVC_X neurons are weakened but not lost following deafening. Furthermore, the decrease in mEPSC amplitude following deafening was more pronounced for larger events (Figure 7B, left), consistent with our observation that larger spines were more likely to decrease in size following deafening (Supplemental figure 3C). These findings further support the idea that deafening weakens excitatory synapses on HVC_X neurons and also reveal an effect of deafening on inhibitory synapses on these cells.

Figure 7. Deafening decreases the amplitude of both excitatory and inhibitory miniature synaptic currents in HVC_X neurons.

A) Left: Confocal images of a retrogradely-labeled HVC_X neuron that was recorded in whole-cell voltage-clamp configuration in vitro and filled with Alexa 488 (green). Arrowheads in the lower image indicate retrogradely-labeled HVC_X neurons (red); retrograde labeling underwent bleaching over the course of the experiment. Scale bar, 20 μm. Right: Representative mIPSCs and mEPSCs are shown for HVC_X neurons recorded in brain slices from control and deafened birds. B) Deafening significantly decreases the amplitude of mEPSCs (left) and mIPSCs (right) in HVC_X neurons (KS test: p < 0.01 for mEPSCs, p < 0.0001 for mIPSCs). Numbers in parentheses indicate the total number of HVC_X neurons analyzed in each experimental group.

In other systems, neurons have been shown to homeostatically modulate their intrinsic membrane properties and excitability in response to diminished synaptic input and sensory deafferentation (for reviews, see Turrigiano and Nelson, 2004, and Walmsley et al., 2006). To assess whether intrinsic properties of HVC_X neurons change following deafening, sharp intracellular current-clamp recordings were made from HVC_X neurons in brain slices prepared at one week post-deafening, when both structural and functional synaptic changes were evident. An additional set of recordings was conducted in brain slices prepared from age-matched, hearing control birds (33 HVC_X neurons recorded in slices from 5 deafened birds, 93–98 dph, and 30 HVC_X neurons recorded in slices from 4 control birds, 89–97 dph). Families of negative and positive currents were injected into neurons, and the resulting changes in membrane potential were used to calculate various intrinsic properties (see Experimental Procedures). These measurements revealed that HVC_X neurons from deafened birds exhibited several significant changes associated with increased excitability, including increased action potential frequency in response to current injection, decreased action potential duration, decreased afterhyperpolarization time-to-peak, and increased resting membrane potential (Supplemental figure 5). However, deafening did not affect the input resistance of HVC_X neurons (64.5 ±4.7 MΩ for control HVC_X, 75.0 ± 5.1 MΩ for deafened HVC_X, p = 0.17, Mann Whitney U test), a finding that, along with the lack of any effect of deafening on spine density, dPSP frequency, and mEPSC and mIPSC frequencies, suggests that deafening does not significantly reduce the number of synapses on these cells.

Changes in intrinsic excitability could potentially translate changes in synaptic strength to changes in action potential output. Consistent with this idea, sharp intracellular current-clamp recordings made in anesthetized male zebra finches revealed a significant decrease in inter-spike intervals (ISIs) in HVC_X neurons and a trend toward increased mean spontaneous action potential firing rates (Figure 6B; ISIs, lower left, p < 0.0001, KS test; mean spike rates, lower right, p = 0.25, Mann Whitney U test; 25 HVC_X cells from 15 hearing control birds, 18 HVC_X from 5 deafened birds). In summary, we observed structural changes to dendritic spines, functional weakening of excitatory and inhibitory synapses, increased intrinsic excitability, and alterations of the spontaneous action potential output of HVC_X neurons following deafening. Taken together, these structural and functional changes indicate that the synapses onto and the action potential output of HVC_X neurons are sensitive to deafening.

Discussion

This study shows that deafening modifies synapses on HVC neurons that provide input to a striatothalamic pathway important to audition-dependent vocal plasticity. Longitudinal, in vivo imaging of dendritic spines revealed that deafening induces two structural correlates of synaptic weakening in HVC_X neurons, namely decreased spine size and stability. In contrast, deafening has no effect on spines on HVC_RA neurons, the other HVC PN type. A sensitive method of behavioral analysis determined that spine shrinkage precedes deafening-induced vocal change and that the magnitude of these structural changes could be used to predict the severity of subsequent song degradation. Importantly, spine changes could not be attributed to the effects of longitudinal imaging, imaging methodology, or decreased singing rate following deafening. In vivo sharp electrode current-clamp recordings and in vitro whole-cell voltage-clamp recordings demonstrated that deafening weakens excitatory and inhibitory synapses in this same cell type over a similar time course. Finally, these structural and functional changes to synapses are accompanied by increased intrinsic excitability and alterations to the spontaneous action potential output of HVC_X neurons.

There are three important aspects to the current study. First, while previous studies in songbirds (Keller and Hahnloser, 2009; Lei and Mooney, 2010) and marmosets (Eliades and Wang, 2003, 2005, and 2008) have detected neurons in auditory areas that are sensitive to perturbation of vocalization-related auditory feedback, this study provides the first observation of sensitivity to deafening in any song sensorimotor area and, to our knowledge, offers the first description of the synaptic effects of hearing loss in sensorimotor neurons important to vocal control. Second, deafening selectively affects spines on HVC_X neurons, and changes in spine size preceded and predicted song degradation, implicating HVC_X neurons in the processing of auditory feedback-related information. Third, structural changes to dendritic spines were accompanied by functional changes in synaptic strength, intrinsic excitability, and spontaneous action potential output of HVC_X neurons, raising the possibility that deafening ultimately affects the singing-related action potential output of these cells. Taken together, these findings indicate that HVC_X neurons are sensitive to deafening and implicate the input stage to the AFP in the processing of feedback-related information.

Synapses on HVC_X neurons are sensitive to deafening

While previous studies employing chronic electrophysiological recordings failed to detect feedback-driven changes in the singing-related activity of HVC_X neurons (Kozhevnikov and Fee, 2007; Prather et al., 2008), in vivo imaging permitted detection of subtle changes in HVC_X dendritic spines within the first few days of deafening. By tracking dendritic spines on single HVC neurons, it was possible to characterize the cell type-specificity and time course of deafening-induced changes in synapses with a degree of precision that would be difficult to achieve using electrophysiological methods. However, these findings do not exclude the possibility that other cells and synapses within HVC are also affected by deafening. Indeed, a previous study in Bengalese finches found that the action potential output of putative HVC interneurons changes subtly and rapidly (~20 msec) following acute feedback perturbation (Sakata and Brainard, 2008), raising the possibility that synapses on this cell type are sensitive to deafening in zebra finches. Further, although the structural and functional measurements performed here support the idea that deafening weakens excitatory and inhibitory synapses on HVC_X neurons, inhibitory synapses may also change on HVC_RA neurons following deafening. Finally, our findings support the idea that deafening-induced changes propagate into HVC but do not exclude the possibility that other song system neurons are sensitive to feedback perturbation. Regardless of these additional possibilities, the fact that structural changes to dendritic spines occurred only in HVC_X neurons supports the idea that synapses on this cell type are especially sensitive to deafening.

Deafening reduces the size of spines in HVC_X neurons prior to song degradation

The finding that deafening-induced changes in HVC_X spine size index occur prior to the onset of song degradation links the observed changes in HVC spines to perturbation of auditory feedback rather than to changes in vocal output. Notably, the method of behavioral analysis employed here allowed us to detect changes to song days to weeks earlier than in prior deafening studies (Brainard and Doupe, 2000; Horita et al., 2008; Lombardino and Nottebohm, 2000; Nordeen and Nordeen, 1992), indicating that this analysis is sensitive to early changes to song. Additionally, the finding that deafening drives subtle song degradation within several days in older birds contrasts with an earlier study (Lombardino and Nottebohm, 2000) and further supports the idea that the analysis used here is sensitive to early deafening-induced changes to song. Finally, following deafening, decreases in HVC_X neuron spine size were predictive of subsequent song degradation, supporting the idea that these structural changes were driven by altered auditory experience, rather than degradation of vocal performance.

Central effects of deafening in other systems

Although prior studies had not resolved synaptic level consequences of deafening in sensorimotor areas important to vocal control, previous studies in both humans and animal models indicate that hearing loss alters synaptic transmission in the auditory cortex. For example, imaging studies in humans reveal that deaf or hearing-impaired subjects exhibit larger ratios of gray to white matter in auditory cortical areas (Emmorey et al., 2003; Kim et al., 2009; Shibata, 2007; Smith et al., 2011), suggesting that deafening reduces myelinated axonal connections in the auditory cortex. Additionally, auditory cortical neurons in deafened animals display increased levels of spontaneous activity and excitability (Noreña and Eggermont, 2003; Seki and Eggermont, 2003; Kotak et al., 2005) and decreased amplitudes of spontaneous and evoked inhibitory currents (Kotak et al., 2008), consistent with the idea that hearing loss alters the balance of excitation and inhibition in the auditory cortex (for a review, see Sanes and Bao, 2009).

Notably, this study is the first demonstration that deafening alters synaptic strength and intrinsic excitability within a sensorimotor area important to learned vocal control, providing a framework to begin to understand how changes in auditory feedback drive changes in vocal output. More broadly, damage to the basal ganglia in adult humans can impair speech prosody, articulation, and comprehension (Damasio et al., 1982). In this light, the current finding that deafening drives changes to dendritic spines in HVC_X neurons prior to the onset of song degradation suggests that altered auditory feedback permeates relatively quickly into the song sensorimotor network. Likewise, the effects of hearing loss in humans on sensorimotor neurons that control speech learning and production, including corticostriatal circuits, may be relatively rapid, even when speech deterioration is delayed in onset.

Comparisons to other forms of sensory deprivation

The deafening-induced spine changes in HVC observed here share many similarities with the effects of sensory deprivation on spine dynamics in other sensory domains. In the mouse somatosensory system, whisker trimming decreases the stability of dendritic spines in barrel cortex, by driving the loss of spines that were previously stable and stabilizing newly-formed spines (Holtmaat et al., 2006; Trachtenberg et al., 2002). In the visual system, focal lesions of the retina can dramatically decrease levels of spine stability, leading to an almost complete replacement of dendritic spines in the deafferented region of cortex (Keck et al., 2008). Additionally, previous studies in barrel cortex found that whisker trimming has more pronounced effects on large, stable spines (Holtmaat et al., 2006; Zuo et al., 2005), similar to our finding that larger spines on HVC_X neurons were more likely to shrink following deafening. Thus, the decreases in spine size and stability in HVC_X neurons observed following deafening support the idea that increased spine dynamics leading to synaptic reorganization is an effect of sensory deprivation that extends to sensorimotor as well as sensory brain regions.

Direct versus indirect effects of auditory feedback

Although the current set of experiments cannot resolve the identity of the excitatory synapses on HVC_X neurons that reorganize following deafening, several scenarios could account for the observed structural and functional changes to this cell type following deafening. First, excitatory synaptic inputs from auditory areas may relay feedback-related information selectively to HVC_X neurons, and silencing these inputs by deafening could trigger changes to HVC_X dendritic spines. One major source of auditory input to HVC is the sensorimotor nucleus interfacialis (NIf) (Cardin and Schmidt, 2004; Coleman and Mooney, 2004). However, NIf lesions do not trigger song degradation in adult zebra finches (Cardin et al., 2005), and do not block song degradation driven by vocal nerve cut (Roy and Mooney, 2009), a process that is thought to result from distorted auditory feedback (Williams and McKibben, 1992). Additionally, strong and selective auditory responses persist in HVC following NIf lesions, indicating that HVC receives an alternate source of auditory information (Roy and Mooney, 2009). Interestingly, the caudal mesopallium (CM), a secondary auditory telencephalic area, supplies an independent source of auditory drive to HVC and contains neurons whose singing-related activity is sensitive in real-time to feedback perturbation (Bauer et al., 2008; Keller and Hahnloser, 2009). Although these findings hint that CM could convey auditory feedback information to HVC, a causal role for CM in feedback-dependent song degradation remains to be established, and the cell type-specificity of its projections to HVC await description. Thus, investigating the circuit basis for the cell type-selectivity of deafening-induced changes in HVC, with a particular focus on CM, is an important future direction.

Another possibility is that HVC interneurons, which may act as acute sensors of auditory feedback (Sakata and Brainard, 2008), alter their singing-related activity immediately upon deafening and, via their inhibitory connections with HVC_X cells, indirectly drive changes to excitatory synapses. In fact, whisker plucking in rodents can drive sprouting of inhibitory inputs from deprived to non-deprived regions of barrel cortex, followed by reciprocal sprouting of excitatory inputs from non-deprived to deprived areas, consistent with such a sequential process (Marik et al., 2010). Similarly, retinal lesions drive a decrease in the density of inhibitory boutons in the visual cortex (Keck et al., 2011) that precedes increases in spine dynamics of excitatory cortical cells (Keck et al., 2008). The idea that deafening, like other forms of sensory deprivation, could induce rapid alterations in inhibition followed by slower changes in excitatory synapses on HVC_X neurons is especially appealing given that acute feedback perturbation alters the singing-related action potential output of putative interneurons in HVC of Bengalese finches (Sakata and Brainard, 2008). Although we have demonstrated that deafening alters the strength of both excitatory and inhibitory synapses on HVC_X neurons, a full test of these ideas would require assessing the relative timing of the effects of feedback perturbation on excitatory and inhibitory inputs to HVC_X neurons and investigating whether experimentally manipulating levels of inhibition can modify excitatory synapses on HVC_X neurons.

Finally, given the insensitivity of HVC_X singing-related activity to feedback perturbation over short timescales, it remains plausible that HVC does not receive a direct feedback signal. In this scenario, feedback information would be acutely processed by areas upstream of HVC and transformed into a modulatory signal that acts more slowly to affect excitatory and inhibitory synapses on HVC_X cells. Regardless of whether the selective remodeling of dendritic spines on HVC_X neurons following deafening is driven by direct or indirect mechanisms, the current findings implicate this cell type in the processing or implementation of auditory feedback.

Relevance of structural and functional changes in HVC_X neurons to vocal output

A major remaining issue is whether the structural and functional effects of deafening on HVC_X neurons affect singing. The current findings show that deafening alters synapses on HVC_X neurons while also increasing their intrinsic excitability, providing at least two ways that deafening could affect the singing-related action potential activity of these cells. First, deafening-induced weakening of synapses that are active during singing may diminish or alter the singing-related action potential output of HVC_X neurons. Second, synapses on HVC_X neurons that are active during singing and unaffected by deafening could act on these more intrinsically excitable cells to increase their singing-related action potential output. Indeed, the finding that deafening alters the spontaneous action potential output of HVC_X neurons in anesthetized birds hints that the output of this cell type may also be altered during singing. Although the exact role of HVC_X neuron output in singing awaits full elucidation, a recent study in Bengalese finches implicates the singing-related action potential activity of these cells in the encoding of syllable sequences (Fujimoto et al., 2011), a song feature that is sensitive to feedback perturbation, including deafening (Sakata and Brainard, 2006; Woolley and Rubel, 1997). Ultimately, long-term measurements of the effects of deafening on the singing-related activity of HVC_X neurons await the development of in vivo functional imaging techniques or improvements to single unit recording methods.

Experimental Procedures

All procedures were in accordance with a protocol approved by the Duke University Institutional Animal Care and Use Committee. See also Supplemental experimental procedures.

Deafening

Male zebra finches (85 to 150 dph) were anesthetized by isoflurane inhalation (2%) and deafened by bilateral cochlear removal. Complete removal of each cochlea was confirmed by visual inspection under a microscope.

Quantification of deafening-induced song degradation

Standard parametric and nonparametric statistical analyses were used for all comparisons (alpha = 0.05); reported errors are s.e.m. unless otherwise noted. Undirected song was recorded continuously starting at least 2d before deafening until at least 1 week post-deafening. To assess song degradation, spectral features of song were quantified by measuring the Wiener entropy and entropy variance (EV) of each syllable in a bird’s song using Sound Analysis Pro (Tchernichovski et al., 2000). Thirty examples of each syllable were measured on each day of song, and values from two pre-deafening days were pooled to obtain a baseline distribution of entropy and EV values for each syllable. The onset of song degradation for each bird was defined as the day on which the distribution of values for either the entropy or EV of any syllable differed significantly from the baseline distribution and remained significantly different on all subsequent days (one-way ANOVA). Syllable sequence changes were measured using sequence consistency (adapted from Scharff and Nottebohm, 1991). The onset of temporal change to song was defined as the first day on which the mean sequence consistency was less than the lower bound of the 95% confidence interval for the mean sequence consistency measured on the last pre-deafening day.

Fluorescent labeling of HVC neurons

Birds were anesthetized with isoflurane inhalation (2%) and placed in a stereotaxic apparatus. Injection sites were localized using stereotaxic coordinates and multi-unit neural recordings. A glass pipette attached to a Nanoject II (Drummond Scientific) was used to deliver GFP-lentivirus (eGFP expressed under the control of the Rous Sarcoma Virus LTR (FRGW)) to HVC or neuronal retrograde tracers to Area X and RA (lentivirus: 32.2 nL/injection, total injection volume ~ 1 μL; tracers, Fast Blue or Alexa-Fluor 594 conjugated dextran amine: 32.2 nL/injection, total injection volume of 64–160 nL). Lentivirus was injected two weeks before imaging, and tracers were injected 4–7 days before imaging.

In vivo two-photon imaging

Birds were placed on a reverse day-night cycle one week before the first imaging session to minimize effects of imaging on their daytime behavior and were imaged longitudinally starting 1–2 nights prior to deafening. On the first night of imaging, birds were anesthetized with isoflurane inhalation (2%) and placed in a stereotaxic apparatus. A headpost was affixed to the skull using dental acrylic, and bilateral craniotomies 1–2 mm² were made over HVC. The dura was excised, and a custom-cut coverslip (No. 1 thickness) was placed over the pial surface and sealed in with dental acrylic. Birds were placed on a custom stage under a Zeiss Laser Scanning Two-Photon Microscope 510. Only GFP-labeled neurons within a field of retrogradely-labeled neurons were classified as HVC neurons and imaged. Dendritic segments of identified HVC neurons were imaged twice nightly at 2h intervals at high resolution (1024 × 1024 pixels, 76 × 76 μm² image size, 3.2 μs/pixel, averaging 2 sample per pixel with 1 μm z-steps, using a 40x/0.8NA Zeiss IR-Archoplan immersion objective).

Image analysis

Three-dimensional image stacks were smoothed using a Gaussian filter (ImageJ); brightness and contrast adjustments were not made for data analysis, although images were contrast-enhanced for figure presentation. Dendritic segments to be analyzed were selected and identified in image stacks collected either 2h or 24h apart. Spine size (measured across nights, 24h interval), was calculated by measuring the integrated optical density of each spine head; these values were background-subtracted and normalized to the mean brightness of the adjacent dendritic shaft. Change in size for a single spine across 24h (spine size index) was calculated as (time 24 size)/(time 0 size). Spine stability for each cell was calculated as the percentage of spines that were maintained (as opposed to spines that were lost or gained) within night (2h interval).

In vivo and in vitro sharp intracellular current-clamp recordings

Sharp intracellular recordings were made in vitro and in vivo from HVC neurons, identified based on their intrinsic electrophysiological properties (Mooney, 2000; Mooney and Prather, 2005). Electrode impedances were 80–150 MΩ when filled with 2M KAc. Recordings were amplified, low-pass filtered at 3 kHz, and digitized at 10 kHz. For in vivo recordings, birds were anesthetized with diazepam (50 μl, 2.5 mg/mL). Mean spontaneous firing rates and inter-spike intervals (ISIs) were measured from recordings of spontaneous activity, and the frequency and amplitude of depolarizing postsynaptic potentials (dPSPs) were measured during tonic injection of hyperpolarizing current, from median filtered traces using custom event detection software (Matlab, K. Hamaguchi). To compare distributions of dPSP amplitude and ISIs between deafened and control neurons, the HVC_X neuron from each group with the smallest number of events was identified. The same number of data points was randomly sampled from all other neurons in that group, so each neuron in the group contributed the same number of dPSPs or ISIs to the group distribution. Sharp intracellular recordings were also made from electrophysiologically-identified HVC_X neurons in 400 μm-thick sagittal brain slices. Negative and positive current pulses were injected into impaled neurons, and resulting membrane potential changes were used to calculate a number of intrinsic membrane properties (Matlab, K. Tschida).

In vitro whole-cell voltage-clamp recordings

Visualized whole-cell voltage-clamp recordings were carried out in retrogradely-labeled HVC_X neurons in tissue from 50–60 dph birds. 300 μm-thick sagittal brain slices were stored briefly at 35° C and then allowed to cool to room temperature over 45 minutes prior to recording. Electrodes had resistances of 2–6 MΩ and were filled with pipette solutions containing (mM): 5 QX-314, 2 ATP, 0.3 GTP, 10 phosphocreatine, 0.2 EGTA, 2 MgCl₂, 5 NaCl, 10 HEPES, 120 cesium methanesulfonate, 0.1 Alexa 488. During recording, slices were superfused with ACSF containing 1 μM TTX. Membrane potential was clamped at −70 mV for measurements of spontaneous mEPSCs and at 0 mV for measurements of mIPSCs. Analyses of mEPSC and mIPSC amplitude were carried out using pCLAMP 10 (Molecular Devices), and data from each HVC_X neuron were randomly sampled so that each HVC_X neuron within the deafened and control groups contributed the same number of PSCs to the group distribution.